Commit | Line | Data |
---|---|---|
3bd1e081 MD |
1 | /* |
2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> | |
3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
00e2e675 | 4 | * 2012 - David Goulet <dgoulet@efficios.com> |
3bd1e081 | 5 | * |
d14d33bf AM |
6 | * This program is free software; you can redistribute it and/or modify |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
3bd1e081 | 9 | * |
d14d33bf AM |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
3bd1e081 | 14 | * |
d14d33bf AM |
15 | * You should have received a copy of the GNU General Public License along |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
3bd1e081 MD |
18 | */ |
19 | ||
20 | #define _GNU_SOURCE | |
6c1c0768 | 21 | #define _LGPL_SOURCE |
3bd1e081 | 22 | #include <assert.h> |
3bd1e081 MD |
23 | #include <poll.h> |
24 | #include <pthread.h> | |
25 | #include <stdlib.h> | |
26 | #include <string.h> | |
27 | #include <sys/mman.h> | |
28 | #include <sys/socket.h> | |
29 | #include <sys/types.h> | |
30 | #include <unistd.h> | |
77c7c900 | 31 | #include <inttypes.h> |
331744e3 | 32 | #include <signal.h> |
3bd1e081 | 33 | |
51a9e1c7 | 34 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 35 | #include <common/common.h> |
fb3a43a9 DG |
36 | #include <common/utils.h> |
37 | #include <common/compat/poll.h> | |
f263b7fd | 38 | #include <common/compat/endian.h> |
309167d2 | 39 | #include <common/index/index.h> |
10a8a223 | 40 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 41 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
42 | #include <common/sessiond-comm/sessiond-comm.h> |
43 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 44 | #include <common/relayd/relayd.h> |
10a8a223 | 45 | #include <common/ust-consumer/ust-consumer.h> |
d3e2ba59 | 46 | #include <common/consumer-timer.h> |
10a8a223 DG |
47 | |
48 | #include "consumer.h" | |
1d1a276c | 49 | #include "consumer-stream.h" |
2d57de81 | 50 | #include "consumer-testpoint.h" |
d07ceecd | 51 | #include "align.h" |
3bd1e081 MD |
52 | |
53 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
54 | .stream_count = 0, |
55 | .need_update = 1, | |
56 | .type = LTTNG_CONSUMER_UNKNOWN, | |
57 | }; | |
58 | ||
d8ef542d MD |
59 | enum consumer_channel_action { |
60 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 61 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
62 | CONSUMER_CHANNEL_QUIT, |
63 | }; | |
64 | ||
65 | struct consumer_channel_msg { | |
66 | enum consumer_channel_action action; | |
a0cbdd2e MD |
67 | struct lttng_consumer_channel *chan; /* add */ |
68 | uint64_t key; /* del */ | |
d8ef542d MD |
69 | }; |
70 | ||
3bd1e081 MD |
71 | /* |
72 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
73 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
74 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
75 | * polling threads. | |
76 | */ | |
a98dae5f | 77 | volatile int consumer_quit; |
3bd1e081 | 78 | |
43c34bc3 | 79 | /* |
43c34bc3 DG |
80 | * Global hash table containing respectively metadata and data streams. The |
81 | * stream element in this ht should only be updated by the metadata poll thread | |
82 | * for the metadata and the data poll thread for the data. | |
83 | */ | |
40dc48e0 DG |
84 | static struct lttng_ht *metadata_ht; |
85 | static struct lttng_ht *data_ht; | |
43c34bc3 | 86 | |
acdb9057 DG |
87 | /* |
88 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
89 | * global state has changed so we just send back the thread in a poll wait | |
90 | * call. | |
91 | */ | |
92 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
93 | { | |
94 | struct lttng_consumer_stream *null_stream = NULL; | |
95 | ||
96 | assert(pipe); | |
97 | ||
98 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
99 | } | |
100 | ||
5c635c72 MD |
101 | static void notify_health_quit_pipe(int *pipe) |
102 | { | |
6cd525e8 | 103 | ssize_t ret; |
5c635c72 | 104 | |
6cd525e8 MD |
105 | ret = lttng_write(pipe[1], "4", 1); |
106 | if (ret < 1) { | |
5c635c72 MD |
107 | PERROR("write consumer health quit"); |
108 | } | |
109 | } | |
110 | ||
d8ef542d MD |
111 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
112 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 113 | uint64_t key, |
d8ef542d MD |
114 | enum consumer_channel_action action) |
115 | { | |
116 | struct consumer_channel_msg msg; | |
6cd525e8 | 117 | ssize_t ret; |
d8ef542d | 118 | |
e56251fc DG |
119 | memset(&msg, 0, sizeof(msg)); |
120 | ||
d8ef542d MD |
121 | msg.action = action; |
122 | msg.chan = chan; | |
f21dae48 | 123 | msg.key = key; |
6cd525e8 MD |
124 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
125 | if (ret < sizeof(msg)) { | |
126 | PERROR("notify_channel_pipe write error"); | |
127 | } | |
d8ef542d MD |
128 | } |
129 | ||
a0cbdd2e MD |
130 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
131 | uint64_t key) | |
132 | { | |
133 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
134 | } | |
135 | ||
d8ef542d MD |
136 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
137 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 138 | uint64_t *key, |
d8ef542d MD |
139 | enum consumer_channel_action *action) |
140 | { | |
141 | struct consumer_channel_msg msg; | |
6cd525e8 | 142 | ssize_t ret; |
d8ef542d | 143 | |
6cd525e8 MD |
144 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
145 | if (ret < sizeof(msg)) { | |
146 | ret = -1; | |
147 | goto error; | |
d8ef542d | 148 | } |
6cd525e8 MD |
149 | *action = msg.action; |
150 | *chan = msg.chan; | |
151 | *key = msg.key; | |
152 | error: | |
153 | return (int) ret; | |
d8ef542d MD |
154 | } |
155 | ||
212d67a2 DG |
156 | /* |
157 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
158 | * visible | |
159 | */ | |
160 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
161 | { | |
162 | struct lttng_consumer_stream *stream, *stmp; | |
163 | ||
164 | assert(channel); | |
165 | ||
166 | /* Delete streams that might have been left in the stream list. */ | |
167 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
168 | send_node) { | |
169 | cds_list_del(&stream->send_node); | |
170 | /* | |
171 | * Once a stream is added to this list, the buffers were created so we | |
172 | * have a guarantee that this call will succeed. Setting the monitor | |
173 | * mode to 0 so we don't lock nor try to delete the stream from the | |
174 | * global hash table. | |
175 | */ | |
176 | stream->monitor = 0; | |
177 | consumer_stream_destroy(stream, NULL); | |
178 | } | |
179 | } | |
180 | ||
3bd1e081 MD |
181 | /* |
182 | * Find a stream. The consumer_data.lock must be locked during this | |
183 | * call. | |
184 | */ | |
d88aee68 | 185 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 186 | struct lttng_ht *ht) |
3bd1e081 | 187 | { |
e4421fec | 188 | struct lttng_ht_iter iter; |
d88aee68 | 189 | struct lttng_ht_node_u64 *node; |
e4421fec | 190 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 191 | |
8389e4f8 DG |
192 | assert(ht); |
193 | ||
d88aee68 DG |
194 | /* -1ULL keys are lookup failures */ |
195 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 196 | return NULL; |
7a57cf92 | 197 | } |
e4421fec | 198 | |
6065ceec DG |
199 | rcu_read_lock(); |
200 | ||
d88aee68 DG |
201 | lttng_ht_lookup(ht, &key, &iter); |
202 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
203 | if (node != NULL) { |
204 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 205 | } |
e4421fec | 206 | |
6065ceec DG |
207 | rcu_read_unlock(); |
208 | ||
e4421fec | 209 | return stream; |
3bd1e081 MD |
210 | } |
211 | ||
da009f2c | 212 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
213 | { |
214 | struct lttng_consumer_stream *stream; | |
215 | ||
04253271 | 216 | rcu_read_lock(); |
ffe60014 | 217 | stream = find_stream(key, ht); |
04253271 | 218 | if (stream) { |
da009f2c | 219 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
220 | /* |
221 | * We don't want the lookup to match, but we still need | |
222 | * to iterate on this stream when iterating over the hash table. Just | |
223 | * change the node key. | |
224 | */ | |
da009f2c | 225 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
226 | } |
227 | rcu_read_unlock(); | |
7ad0a0cb MD |
228 | } |
229 | ||
d56db448 DG |
230 | /* |
231 | * Return a channel object for the given key. | |
232 | * | |
233 | * RCU read side lock MUST be acquired before calling this function and | |
234 | * protects the channel ptr. | |
235 | */ | |
d88aee68 | 236 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 237 | { |
e4421fec | 238 | struct lttng_ht_iter iter; |
d88aee68 | 239 | struct lttng_ht_node_u64 *node; |
e4421fec | 240 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 241 | |
d88aee68 DG |
242 | /* -1ULL keys are lookup failures */ |
243 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 244 | return NULL; |
7a57cf92 | 245 | } |
e4421fec | 246 | |
d88aee68 DG |
247 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
248 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
249 | if (node != NULL) { |
250 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 251 | } |
e4421fec DG |
252 | |
253 | return channel; | |
3bd1e081 MD |
254 | } |
255 | ||
b5a6470f DG |
256 | /* |
257 | * There is a possibility that the consumer does not have enough time between | |
258 | * the close of the channel on the session daemon and the cleanup in here thus | |
259 | * once we have a channel add with an existing key, we know for sure that this | |
260 | * channel will eventually get cleaned up by all streams being closed. | |
261 | * | |
262 | * This function just nullifies the already existing channel key. | |
263 | */ | |
264 | static void steal_channel_key(uint64_t key) | |
265 | { | |
266 | struct lttng_consumer_channel *channel; | |
267 | ||
268 | rcu_read_lock(); | |
269 | channel = consumer_find_channel(key); | |
270 | if (channel) { | |
271 | channel->key = (uint64_t) -1ULL; | |
272 | /* | |
273 | * We don't want the lookup to match, but we still need to iterate on | |
274 | * this channel when iterating over the hash table. Just change the | |
275 | * node key. | |
276 | */ | |
277 | channel->node.key = (uint64_t) -1ULL; | |
278 | } | |
279 | rcu_read_unlock(); | |
280 | } | |
281 | ||
ffe60014 | 282 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 283 | { |
d88aee68 DG |
284 | struct lttng_ht_node_u64 *node = |
285 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
286 | struct lttng_consumer_channel *channel = |
287 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 288 | |
11544fe5 MD |
289 | switch (consumer_data.type) { |
290 | case LTTNG_CONSUMER_KERNEL: | |
291 | break; | |
292 | case LTTNG_CONSUMER32_UST: | |
293 | case LTTNG_CONSUMER64_UST: | |
294 | lttng_ustconsumer_free_channel(channel); | |
295 | break; | |
296 | default: | |
297 | ERR("Unknown consumer_data type"); | |
298 | abort(); | |
299 | } | |
ffe60014 | 300 | free(channel); |
702b1ea4 MD |
301 | } |
302 | ||
00e2e675 DG |
303 | /* |
304 | * RCU protected relayd socket pair free. | |
305 | */ | |
ffe60014 | 306 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 307 | { |
d88aee68 DG |
308 | struct lttng_ht_node_u64 *node = |
309 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
310 | struct consumer_relayd_sock_pair *relayd = |
311 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
312 | ||
8994307f DG |
313 | /* |
314 | * Close all sockets. This is done in the call RCU since we don't want the | |
315 | * socket fds to be reassigned thus potentially creating bad state of the | |
316 | * relayd object. | |
317 | * | |
318 | * We do not have to lock the control socket mutex here since at this stage | |
319 | * there is no one referencing to this relayd object. | |
320 | */ | |
321 | (void) relayd_close(&relayd->control_sock); | |
322 | (void) relayd_close(&relayd->data_sock); | |
323 | ||
00e2e675 DG |
324 | free(relayd); |
325 | } | |
326 | ||
327 | /* | |
328 | * Destroy and free relayd socket pair object. | |
00e2e675 | 329 | */ |
51230d70 | 330 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
331 | { |
332 | int ret; | |
333 | struct lttng_ht_iter iter; | |
334 | ||
173af62f DG |
335 | if (relayd == NULL) { |
336 | return; | |
337 | } | |
338 | ||
00e2e675 DG |
339 | DBG("Consumer destroy and close relayd socket pair"); |
340 | ||
341 | iter.iter.node = &relayd->node.node; | |
342 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 343 | if (ret != 0) { |
8994307f | 344 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
345 | return; |
346 | } | |
00e2e675 | 347 | |
00e2e675 | 348 | /* RCU free() call */ |
ffe60014 DG |
349 | call_rcu(&relayd->node.head, free_relayd_rcu); |
350 | } | |
351 | ||
352 | /* | |
353 | * Remove a channel from the global list protected by a mutex. This function is | |
354 | * also responsible for freeing its data structures. | |
355 | */ | |
356 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
357 | { | |
358 | int ret; | |
359 | struct lttng_ht_iter iter; | |
360 | ||
d88aee68 | 361 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
362 | |
363 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 364 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 365 | |
212d67a2 DG |
366 | /* Destroy streams that might have been left in the stream list. */ |
367 | clean_channel_stream_list(channel); | |
51e762e5 | 368 | |
d3e2ba59 JD |
369 | if (channel->live_timer_enabled == 1) { |
370 | consumer_timer_live_stop(channel); | |
371 | } | |
372 | ||
ffe60014 DG |
373 | switch (consumer_data.type) { |
374 | case LTTNG_CONSUMER_KERNEL: | |
375 | break; | |
376 | case LTTNG_CONSUMER32_UST: | |
377 | case LTTNG_CONSUMER64_UST: | |
378 | lttng_ustconsumer_del_channel(channel); | |
379 | break; | |
380 | default: | |
381 | ERR("Unknown consumer_data type"); | |
382 | assert(0); | |
383 | goto end; | |
384 | } | |
385 | ||
386 | rcu_read_lock(); | |
387 | iter.iter.node = &channel->node.node; | |
388 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
389 | assert(!ret); | |
390 | rcu_read_unlock(); | |
391 | ||
392 | call_rcu(&channel->node.head, free_channel_rcu); | |
393 | end: | |
a9838785 | 394 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 395 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
396 | } |
397 | ||
228b5bf7 DG |
398 | /* |
399 | * Iterate over the relayd hash table and destroy each element. Finally, | |
400 | * destroy the whole hash table. | |
401 | */ | |
402 | static void cleanup_relayd_ht(void) | |
403 | { | |
404 | struct lttng_ht_iter iter; | |
405 | struct consumer_relayd_sock_pair *relayd; | |
406 | ||
407 | rcu_read_lock(); | |
408 | ||
409 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
410 | node.node) { | |
51230d70 | 411 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
412 | } |
413 | ||
228b5bf7 | 414 | rcu_read_unlock(); |
36b588ed MD |
415 | |
416 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
417 | } |
418 | ||
8994307f DG |
419 | /* |
420 | * Update the end point status of all streams having the given network sequence | |
421 | * index (relayd index). | |
422 | * | |
423 | * It's atomically set without having the stream mutex locked which is fine | |
424 | * because we handle the write/read race with a pipe wakeup for each thread. | |
425 | */ | |
da009f2c | 426 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
427 | enum consumer_endpoint_status status) |
428 | { | |
429 | struct lttng_ht_iter iter; | |
430 | struct lttng_consumer_stream *stream; | |
431 | ||
da009f2c | 432 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
433 | |
434 | rcu_read_lock(); | |
435 | ||
436 | /* Let's begin with metadata */ | |
437 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
438 | if (stream->net_seq_idx == net_seq_idx) { | |
439 | uatomic_set(&stream->endpoint_status, status); | |
440 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
441 | } | |
442 | } | |
443 | ||
444 | /* Follow up by the data streams */ | |
445 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
446 | if (stream->net_seq_idx == net_seq_idx) { | |
447 | uatomic_set(&stream->endpoint_status, status); | |
448 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
449 | } | |
450 | } | |
451 | rcu_read_unlock(); | |
452 | } | |
453 | ||
454 | /* | |
455 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
456 | * destroying the object meaning removing it from the relayd hash table, | |
457 | * closing the sockets and freeing the memory in a RCU call. | |
458 | * | |
459 | * If a local data context is available, notify the threads that the streams' | |
460 | * state have changed. | |
461 | */ | |
462 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
463 | struct lttng_consumer_local_data *ctx) | |
464 | { | |
da009f2c | 465 | uint64_t netidx; |
8994307f DG |
466 | |
467 | assert(relayd); | |
468 | ||
9617607b DG |
469 | DBG("Cleaning up relayd sockets"); |
470 | ||
8994307f DG |
471 | /* Save the net sequence index before destroying the object */ |
472 | netidx = relayd->net_seq_idx; | |
473 | ||
474 | /* | |
475 | * Delete the relayd from the relayd hash table, close the sockets and free | |
476 | * the object in a RCU call. | |
477 | */ | |
51230d70 | 478 | consumer_destroy_relayd(relayd); |
8994307f DG |
479 | |
480 | /* Set inactive endpoint to all streams */ | |
481 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
482 | ||
483 | /* | |
484 | * With a local data context, notify the threads that the streams' state | |
485 | * have changed. The write() action on the pipe acts as an "implicit" | |
486 | * memory barrier ordering the updates of the end point status from the | |
487 | * read of this status which happens AFTER receiving this notify. | |
488 | */ | |
489 | if (ctx) { | |
acdb9057 | 490 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 491 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
492 | } |
493 | } | |
494 | ||
a6ba4fe1 DG |
495 | /* |
496 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
497 | * reaches zero. | |
498 | * | |
499 | * RCU read side lock MUST be aquired before calling this function. | |
500 | */ | |
501 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
502 | { | |
503 | assert(relayd); | |
504 | ||
505 | /* Set destroy flag for this object */ | |
506 | uatomic_set(&relayd->destroy_flag, 1); | |
507 | ||
508 | /* Destroy the relayd if refcount is 0 */ | |
509 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 510 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
511 | } |
512 | } | |
513 | ||
3bd1e081 | 514 | /* |
1d1a276c DG |
515 | * Completly destroy stream from every visiable data structure and the given |
516 | * hash table if one. | |
517 | * | |
518 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 519 | */ |
e316aad5 DG |
520 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
521 | struct lttng_ht *ht) | |
3bd1e081 | 522 | { |
1d1a276c | 523 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
524 | } |
525 | ||
5ab66908 MD |
526 | /* |
527 | * XXX naming of del vs destroy is all mixed up. | |
528 | */ | |
529 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
530 | { | |
531 | consumer_stream_destroy(stream, data_ht); | |
532 | } | |
533 | ||
534 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
535 | { | |
536 | consumer_stream_destroy(stream, metadata_ht); | |
537 | } | |
538 | ||
d88aee68 DG |
539 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
540 | uint64_t stream_key, | |
3bd1e081 | 541 | enum lttng_consumer_stream_state state, |
ffe60014 | 542 | const char *channel_name, |
6df2e2c9 | 543 | uid_t uid, |
00e2e675 | 544 | gid_t gid, |
57a269f2 | 545 | uint64_t relayd_id, |
53632229 | 546 | uint64_t session_id, |
ffe60014 DG |
547 | int cpu, |
548 | int *alloc_ret, | |
4891ece8 DG |
549 | enum consumer_channel_type type, |
550 | unsigned int monitor) | |
3bd1e081 | 551 | { |
ffe60014 | 552 | int ret; |
3bd1e081 | 553 | struct lttng_consumer_stream *stream; |
3bd1e081 | 554 | |
effcf122 | 555 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 556 | if (stream == NULL) { |
7a57cf92 | 557 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 558 | ret = -ENOMEM; |
7a57cf92 | 559 | goto end; |
3bd1e081 | 560 | } |
7a57cf92 | 561 | |
d56db448 DG |
562 | rcu_read_lock(); |
563 | ||
3bd1e081 | 564 | stream->key = stream_key; |
3bd1e081 MD |
565 | stream->out_fd = -1; |
566 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 567 | stream->output_written = 0; |
3bd1e081 | 568 | stream->state = state; |
6df2e2c9 MD |
569 | stream->uid = uid; |
570 | stream->gid = gid; | |
ffe60014 | 571 | stream->net_seq_idx = relayd_id; |
53632229 | 572 | stream->session_id = session_id; |
4891ece8 | 573 | stream->monitor = monitor; |
774d490c | 574 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
309167d2 | 575 | stream->index_fd = -1; |
53632229 | 576 | pthread_mutex_init(&stream->lock, NULL); |
288bdb30 | 577 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
58b1f425 | 578 | |
ffe60014 DG |
579 | /* If channel is the metadata, flag this stream as metadata. */ |
580 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
581 | stream->metadata_flag = 1; | |
582 | /* Metadata is flat out. */ | |
583 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
94d49140 JD |
584 | /* Live rendez-vous point. */ |
585 | pthread_cond_init(&stream->metadata_rdv, NULL); | |
586 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); | |
58b1f425 | 587 | } else { |
ffe60014 DG |
588 | /* Format stream name to <channel_name>_<cpu_number> */ |
589 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
590 | channel_name, cpu); | |
591 | if (ret < 0) { | |
592 | PERROR("snprintf stream name"); | |
593 | goto error; | |
594 | } | |
58b1f425 | 595 | } |
c30aaa51 | 596 | |
ffe60014 | 597 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 598 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 599 | |
d8ef542d MD |
600 | /* Init node per channel id key */ |
601 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
602 | ||
53632229 | 603 | /* Init session id node with the stream session id */ |
d88aee68 | 604 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 605 | |
07b86b52 JD |
606 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
607 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
608 | stream->name, stream->key, channel_key, | |
609 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
610 | |
611 | rcu_read_unlock(); | |
3bd1e081 | 612 | return stream; |
c80048c6 MD |
613 | |
614 | error: | |
d56db448 | 615 | rcu_read_unlock(); |
c80048c6 | 616 | free(stream); |
7a57cf92 | 617 | end: |
ffe60014 DG |
618 | if (alloc_ret) { |
619 | *alloc_ret = ret; | |
620 | } | |
c80048c6 | 621 | return NULL; |
3bd1e081 MD |
622 | } |
623 | ||
624 | /* | |
625 | * Add a stream to the global list protected by a mutex. | |
626 | */ | |
5ab66908 | 627 | int consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 628 | { |
5ab66908 | 629 | struct lttng_ht *ht = data_ht; |
3bd1e081 MD |
630 | int ret = 0; |
631 | ||
e316aad5 | 632 | assert(stream); |
43c34bc3 | 633 | assert(ht); |
c77fc10a | 634 | |
d88aee68 | 635 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
636 | |
637 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 638 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 639 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 640 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 641 | rcu_read_lock(); |
e316aad5 | 642 | |
43c34bc3 | 643 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 644 | steal_stream_key(stream->key, ht); |
43c34bc3 | 645 | |
d88aee68 | 646 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 647 | |
d8ef542d MD |
648 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
649 | &stream->node_channel_id); | |
650 | ||
ca22feea DG |
651 | /* |
652 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
653 | * the key since the HT does not use it and we allow to add redundant keys | |
654 | * into this table. | |
655 | */ | |
d88aee68 | 656 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 657 | |
e316aad5 | 658 | /* |
ffe60014 DG |
659 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
660 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
661 | * causes the count to become 0 also causes a stream to be added. The |
662 | * channel deletion will thus be triggered by the following removal of this | |
663 | * stream. | |
664 | */ | |
ffe60014 | 665 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
666 | /* Increment refcount before decrementing nb_init_stream_left */ |
667 | cmm_smp_wmb(); | |
ffe60014 | 668 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
669 | } |
670 | ||
671 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
672 | consumer_data.stream_count++; |
673 | consumer_data.need_update = 1; | |
674 | ||
e316aad5 | 675 | rcu_read_unlock(); |
2e818a6a | 676 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 677 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 678 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 679 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 680 | |
3bd1e081 MD |
681 | return ret; |
682 | } | |
683 | ||
5ab66908 MD |
684 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
685 | { | |
686 | consumer_del_stream(stream, data_ht); | |
687 | } | |
688 | ||
00e2e675 | 689 | /* |
3f8e211f DG |
690 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
691 | * be acquired before calling this. | |
00e2e675 | 692 | */ |
d09e1200 | 693 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
694 | { |
695 | int ret = 0; | |
d88aee68 | 696 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
697 | struct lttng_ht_iter iter; |
698 | ||
ffe60014 | 699 | assert(relayd); |
00e2e675 | 700 | |
00e2e675 | 701 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
702 | &relayd->net_seq_idx, &iter); |
703 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 704 | if (node != NULL) { |
00e2e675 DG |
705 | goto end; |
706 | } | |
d88aee68 | 707 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 708 | |
00e2e675 DG |
709 | end: |
710 | return ret; | |
711 | } | |
712 | ||
713 | /* | |
714 | * Allocate and return a consumer relayd socket. | |
715 | */ | |
716 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
da009f2c | 717 | uint64_t net_seq_idx) |
00e2e675 DG |
718 | { |
719 | struct consumer_relayd_sock_pair *obj = NULL; | |
720 | ||
da009f2c MD |
721 | /* net sequence index of -1 is a failure */ |
722 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
723 | goto error; |
724 | } | |
725 | ||
726 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
727 | if (obj == NULL) { | |
728 | PERROR("zmalloc relayd sock"); | |
729 | goto error; | |
730 | } | |
731 | ||
732 | obj->net_seq_idx = net_seq_idx; | |
733 | obj->refcount = 0; | |
173af62f | 734 | obj->destroy_flag = 0; |
f96e4545 MD |
735 | obj->control_sock.sock.fd = -1; |
736 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 737 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
738 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
739 | ||
740 | error: | |
741 | return obj; | |
742 | } | |
743 | ||
744 | /* | |
745 | * Find a relayd socket pair in the global consumer data. | |
746 | * | |
747 | * Return the object if found else NULL. | |
b0b335c8 MD |
748 | * RCU read-side lock must be held across this call and while using the |
749 | * returned object. | |
00e2e675 | 750 | */ |
d88aee68 | 751 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
752 | { |
753 | struct lttng_ht_iter iter; | |
d88aee68 | 754 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
755 | struct consumer_relayd_sock_pair *relayd = NULL; |
756 | ||
757 | /* Negative keys are lookup failures */ | |
d88aee68 | 758 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
759 | goto error; |
760 | } | |
761 | ||
d88aee68 | 762 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 763 | &iter); |
d88aee68 | 764 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
765 | if (node != NULL) { |
766 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
767 | } | |
768 | ||
00e2e675 DG |
769 | error: |
770 | return relayd; | |
771 | } | |
772 | ||
10a50311 JD |
773 | /* |
774 | * Find a relayd and send the stream | |
775 | * | |
776 | * Returns 0 on success, < 0 on error | |
777 | */ | |
778 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
779 | char *path) | |
780 | { | |
781 | int ret = 0; | |
782 | struct consumer_relayd_sock_pair *relayd; | |
783 | ||
784 | assert(stream); | |
785 | assert(stream->net_seq_idx != -1ULL); | |
786 | assert(path); | |
787 | ||
788 | /* The stream is not metadata. Get relayd reference if exists. */ | |
789 | rcu_read_lock(); | |
790 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
791 | if (relayd != NULL) { | |
792 | /* Add stream on the relayd */ | |
793 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
794 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
795 | path, &stream->relayd_stream_id, | |
796 | stream->chan->tracefile_size, stream->chan->tracefile_count); | |
797 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
798 | if (ret < 0) { | |
799 | goto end; | |
800 | } | |
1c20f0e2 | 801 | |
10a50311 | 802 | uatomic_inc(&relayd->refcount); |
d01178b6 | 803 | stream->sent_to_relayd = 1; |
10a50311 JD |
804 | } else { |
805 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
806 | stream->key, stream->net_seq_idx); | |
807 | ret = -1; | |
808 | goto end; | |
809 | } | |
810 | ||
811 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
812 | stream->name, stream->key, stream->net_seq_idx); | |
813 | ||
814 | end: | |
815 | rcu_read_unlock(); | |
816 | return ret; | |
817 | } | |
818 | ||
a4baae1b JD |
819 | /* |
820 | * Find a relayd and send the streams sent message | |
821 | * | |
822 | * Returns 0 on success, < 0 on error | |
823 | */ | |
824 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
825 | { | |
826 | int ret = 0; | |
827 | struct consumer_relayd_sock_pair *relayd; | |
828 | ||
829 | assert(net_seq_idx != -1ULL); | |
830 | ||
831 | /* The stream is not metadata. Get relayd reference if exists. */ | |
832 | rcu_read_lock(); | |
833 | relayd = consumer_find_relayd(net_seq_idx); | |
834 | if (relayd != NULL) { | |
835 | /* Add stream on the relayd */ | |
836 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
837 | ret = relayd_streams_sent(&relayd->control_sock); | |
838 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
839 | if (ret < 0) { | |
840 | goto end; | |
841 | } | |
842 | } else { | |
843 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
844 | net_seq_idx); | |
845 | ret = -1; | |
846 | goto end; | |
847 | } | |
848 | ||
849 | ret = 0; | |
850 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
851 | ||
852 | end: | |
853 | rcu_read_unlock(); | |
854 | return ret; | |
855 | } | |
856 | ||
10a50311 JD |
857 | /* |
858 | * Find a relayd and close the stream | |
859 | */ | |
860 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
861 | { | |
862 | struct consumer_relayd_sock_pair *relayd; | |
863 | ||
864 | /* The stream is not metadata. Get relayd reference if exists. */ | |
865 | rcu_read_lock(); | |
866 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
867 | if (relayd) { | |
868 | consumer_stream_relayd_close(stream, relayd); | |
869 | } | |
870 | rcu_read_unlock(); | |
871 | } | |
872 | ||
00e2e675 DG |
873 | /* |
874 | * Handle stream for relayd transmission if the stream applies for network | |
875 | * streaming where the net sequence index is set. | |
876 | * | |
877 | * Return destination file descriptor or negative value on error. | |
878 | */ | |
6197aea7 | 879 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
880 | size_t data_size, unsigned long padding, |
881 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
882 | { |
883 | int outfd = -1, ret; | |
00e2e675 DG |
884 | struct lttcomm_relayd_data_hdr data_hdr; |
885 | ||
886 | /* Safety net */ | |
887 | assert(stream); | |
6197aea7 | 888 | assert(relayd); |
00e2e675 DG |
889 | |
890 | /* Reset data header */ | |
891 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
892 | ||
00e2e675 DG |
893 | if (stream->metadata_flag) { |
894 | /* Caller MUST acquire the relayd control socket lock */ | |
895 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
896 | if (ret < 0) { | |
897 | goto error; | |
898 | } | |
899 | ||
900 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 901 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
902 | } else { |
903 | /* Set header with stream information */ | |
904 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
905 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 906 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
907 | /* |
908 | * Note that net_seq_num below is assigned with the *current* value of | |
909 | * next_net_seq_num and only after that the next_net_seq_num will be | |
910 | * increment. This is why when issuing a command on the relayd using | |
911 | * this next value, 1 should always be substracted in order to compare | |
912 | * the last seen sequence number on the relayd side to the last sent. | |
913 | */ | |
3604f373 | 914 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
915 | /* Other fields are zeroed previously */ |
916 | ||
917 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
918 | sizeof(data_hdr)); | |
919 | if (ret < 0) { | |
920 | goto error; | |
921 | } | |
922 | ||
3604f373 DG |
923 | ++stream->next_net_seq_num; |
924 | ||
00e2e675 | 925 | /* Set to go on data socket */ |
6151a90f | 926 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
927 | } |
928 | ||
929 | error: | |
930 | return outfd; | |
931 | } | |
932 | ||
3bd1e081 | 933 | /* |
ffe60014 DG |
934 | * Allocate and return a new lttng_consumer_channel object using the given key |
935 | * to initialize the hash table node. | |
936 | * | |
937 | * On error, return NULL. | |
3bd1e081 | 938 | */ |
886224ff | 939 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
940 | uint64_t session_id, |
941 | const char *pathname, | |
942 | const char *name, | |
943 | uid_t uid, | |
944 | gid_t gid, | |
57a269f2 | 945 | uint64_t relayd_id, |
1624d5b7 JD |
946 | enum lttng_event_output output, |
947 | uint64_t tracefile_size, | |
2bba9e53 | 948 | uint64_t tracefile_count, |
1950109e | 949 | uint64_t session_id_per_pid, |
ecc48a90 | 950 | unsigned int monitor, |
d7ba1388 | 951 | unsigned int live_timer_interval, |
3d071855 | 952 | const char *root_shm_path, |
d7ba1388 | 953 | const char *shm_path) |
3bd1e081 MD |
954 | { |
955 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 956 | |
276b26d1 | 957 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 958 | if (channel == NULL) { |
7a57cf92 | 959 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
960 | goto end; |
961 | } | |
ffe60014 DG |
962 | |
963 | channel->key = key; | |
3bd1e081 | 964 | channel->refcount = 0; |
ffe60014 | 965 | channel->session_id = session_id; |
1950109e | 966 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
967 | channel->uid = uid; |
968 | channel->gid = gid; | |
969 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
970 | channel->tracefile_size = tracefile_size; |
971 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 972 | channel->monitor = monitor; |
ecc48a90 | 973 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 974 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 975 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 976 | |
0c759fc9 DG |
977 | switch (output) { |
978 | case LTTNG_EVENT_SPLICE: | |
979 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
980 | break; | |
981 | case LTTNG_EVENT_MMAP: | |
982 | channel->output = CONSUMER_CHANNEL_MMAP; | |
983 | break; | |
984 | default: | |
985 | assert(0); | |
986 | free(channel); | |
987 | channel = NULL; | |
988 | goto end; | |
989 | } | |
990 | ||
07b86b52 JD |
991 | /* |
992 | * In monitor mode, the streams associated with the channel will be put in | |
993 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
994 | * here meaning that the channel itself has streams that are referenced. | |
995 | * | |
996 | * On a channel deletion, once the channel is no longer visible, the | |
997 | * refcount is decremented and checked for a zero value to delete it. With | |
998 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
999 | */ | |
1000 | if (!channel->monitor) { | |
1001 | channel->refcount = 1; | |
1002 | } | |
1003 | ||
ffe60014 DG |
1004 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1005 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1006 | ||
1007 | strncpy(channel->name, name, sizeof(channel->name)); | |
1008 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1009 | ||
3d071855 MD |
1010 | if (root_shm_path) { |
1011 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1012 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1013 | } | |
d7ba1388 MD |
1014 | if (shm_path) { |
1015 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1016 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1017 | } | |
1018 | ||
d88aee68 | 1019 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
1020 | |
1021 | channel->wait_fd = -1; | |
1022 | ||
ffe60014 DG |
1023 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1024 | ||
d88aee68 | 1025 | DBG("Allocated channel (key %" PRIu64 ")", channel->key) |
3bd1e081 | 1026 | |
3bd1e081 MD |
1027 | end: |
1028 | return channel; | |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1033 | * |
b5a6470f | 1034 | * Always return 0 indicating success. |
3bd1e081 | 1035 | */ |
d8ef542d MD |
1036 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1037 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1038 | { |
3bd1e081 | 1039 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1040 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1041 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1042 | |
b5a6470f DG |
1043 | /* |
1044 | * This gives us a guarantee that the channel we are about to add to the | |
1045 | * channel hash table will be unique. See this function comment on the why | |
1046 | * we need to steel the channel key at this stage. | |
1047 | */ | |
1048 | steal_channel_key(channel->key); | |
c77fc10a | 1049 | |
b5a6470f | 1050 | rcu_read_lock(); |
d88aee68 | 1051 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
6065ceec | 1052 | rcu_read_unlock(); |
b5a6470f | 1053 | |
ec6ea7d0 | 1054 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1055 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1056 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1057 | |
b5a6470f | 1058 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1059 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1060 | } |
b5a6470f DG |
1061 | |
1062 | return 0; | |
3bd1e081 MD |
1063 | } |
1064 | ||
1065 | /* | |
1066 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1067 | * doing a lookup in the linked list and concurrency issues when writing is | |
1068 | * needed. Called with consumer_data.lock held. | |
1069 | * | |
1070 | * Returns the number of fds in the structures. | |
1071 | */ | |
ffe60014 DG |
1072 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1073 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
1074 | struct lttng_ht *ht) | |
3bd1e081 | 1075 | { |
3bd1e081 | 1076 | int i = 0; |
e4421fec DG |
1077 | struct lttng_ht_iter iter; |
1078 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1079 | |
ffe60014 DG |
1080 | assert(ctx); |
1081 | assert(ht); | |
1082 | assert(pollfd); | |
1083 | assert(local_stream); | |
1084 | ||
3bd1e081 | 1085 | DBG("Updating poll fd array"); |
481d6c57 | 1086 | rcu_read_lock(); |
43c34bc3 | 1087 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1088 | /* |
1089 | * Only active streams with an active end point can be added to the | |
1090 | * poll set and local stream storage of the thread. | |
1091 | * | |
1092 | * There is a potential race here for endpoint_status to be updated | |
1093 | * just after the check. However, this is OK since the stream(s) will | |
1094 | * be deleted once the thread is notified that the end point state has | |
1095 | * changed where this function will be called back again. | |
1096 | */ | |
1097 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1098 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
1099 | continue; |
1100 | } | |
7972aab2 DG |
1101 | /* |
1102 | * This clobbers way too much the debug output. Uncomment that if you | |
1103 | * need it for debugging purposes. | |
1104 | * | |
1105 | * DBG("Active FD %d", stream->wait_fd); | |
1106 | */ | |
e4421fec | 1107 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1108 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1109 | local_stream[i] = stream; |
3bd1e081 MD |
1110 | i++; |
1111 | } | |
481d6c57 | 1112 | rcu_read_unlock(); |
3bd1e081 MD |
1113 | |
1114 | /* | |
50f8ae69 | 1115 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1116 | * increment i so nb_fd is the number of real FD. |
1117 | */ | |
acdb9057 | 1118 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1119 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1120 | |
1121 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1122 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1123 | return i; |
1124 | } | |
1125 | ||
1126 | /* | |
84382d49 MD |
1127 | * Poll on the should_quit pipe and the command socket return -1 on |
1128 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1129 | */ |
1130 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1131 | { | |
1132 | int num_rdy; | |
1133 | ||
88f2b785 | 1134 | restart: |
3bd1e081 MD |
1135 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1136 | if (num_rdy == -1) { | |
88f2b785 MD |
1137 | /* |
1138 | * Restart interrupted system call. | |
1139 | */ | |
1140 | if (errno == EINTR) { | |
1141 | goto restart; | |
1142 | } | |
7a57cf92 | 1143 | PERROR("Poll error"); |
84382d49 | 1144 | return -1; |
3bd1e081 | 1145 | } |
509bb1cf | 1146 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1147 | DBG("consumer_should_quit wake up"); |
84382d49 | 1148 | return 1; |
3bd1e081 MD |
1149 | } |
1150 | return 0; | |
3bd1e081 MD |
1151 | } |
1152 | ||
1153 | /* | |
1154 | * Set the error socket. | |
1155 | */ | |
ffe60014 DG |
1156 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1157 | int sock) | |
3bd1e081 MD |
1158 | { |
1159 | ctx->consumer_error_socket = sock; | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Set the command socket path. | |
1164 | */ | |
3bd1e081 MD |
1165 | void lttng_consumer_set_command_sock_path( |
1166 | struct lttng_consumer_local_data *ctx, char *sock) | |
1167 | { | |
1168 | ctx->consumer_command_sock_path = sock; | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * Send return code to the session daemon. | |
1173 | * If the socket is not defined, we return 0, it is not a fatal error | |
1174 | */ | |
ffe60014 | 1175 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1176 | { |
1177 | if (ctx->consumer_error_socket > 0) { | |
1178 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1179 | sizeof(enum lttcomm_sessiond_command)); | |
1180 | } | |
1181 | ||
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | /* | |
228b5bf7 DG |
1186 | * Close all the tracefiles and stream fds and MUST be called when all |
1187 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1188 | */ |
1189 | void lttng_consumer_cleanup(void) | |
1190 | { | |
e4421fec | 1191 | struct lttng_ht_iter iter; |
ffe60014 | 1192 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1193 | |
1194 | rcu_read_lock(); | |
3bd1e081 | 1195 | |
ffe60014 DG |
1196 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1197 | node.node) { | |
702b1ea4 | 1198 | consumer_del_channel(channel); |
3bd1e081 | 1199 | } |
6065ceec DG |
1200 | |
1201 | rcu_read_unlock(); | |
d6ce1df2 | 1202 | |
d6ce1df2 | 1203 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1204 | |
1205 | cleanup_relayd_ht(); | |
1206 | ||
d8ef542d MD |
1207 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1208 | ||
228b5bf7 DG |
1209 | /* |
1210 | * This HT contains streams that are freed by either the metadata thread or | |
1211 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1212 | * it. | |
1213 | */ | |
1214 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1215 | } |
1216 | ||
1217 | /* | |
1218 | * Called from signal handler. | |
1219 | */ | |
1220 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1221 | { | |
6cd525e8 MD |
1222 | ssize_t ret; |
1223 | ||
3bd1e081 | 1224 | consumer_quit = 1; |
6cd525e8 MD |
1225 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1226 | if (ret < 1) { | |
7a57cf92 | 1227 | PERROR("write consumer quit"); |
3bd1e081 | 1228 | } |
ab1027f4 DG |
1229 | |
1230 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1231 | } |
1232 | ||
00e2e675 DG |
1233 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1234 | off_t orig_offset) | |
3bd1e081 | 1235 | { |
e2f6bedf | 1236 | int ret; |
3bd1e081 MD |
1237 | int outfd = stream->out_fd; |
1238 | ||
1239 | /* | |
1240 | * This does a blocking write-and-wait on any page that belongs to the | |
1241 | * subbuffer prior to the one we just wrote. | |
1242 | * Don't care about error values, as these are just hints and ways to | |
1243 | * limit the amount of page cache used. | |
1244 | */ | |
ffe60014 | 1245 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1246 | return; |
1247 | } | |
ffe60014 DG |
1248 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1249 | stream->max_sb_size, | |
3bd1e081 MD |
1250 | SYNC_FILE_RANGE_WAIT_BEFORE |
1251 | | SYNC_FILE_RANGE_WRITE | |
1252 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1253 | /* | |
1254 | * Give hints to the kernel about how we access the file: | |
1255 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1256 | * we write it. | |
1257 | * | |
1258 | * We need to call fadvise again after the file grows because the | |
1259 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1260 | * file. | |
1261 | * | |
1262 | * Call fadvise _after_ having waited for the page writeback to | |
1263 | * complete because the dirty page writeback semantic is not well | |
1264 | * defined. So it can be expected to lead to lower throughput in | |
1265 | * streaming. | |
1266 | */ | |
e2f6bedf | 1267 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1268 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
e2f6bedf JG |
1269 | if (ret) { |
1270 | WARN("posix_fadvise() error (%i)", ret); | |
1271 | } | |
3bd1e081 MD |
1272 | } |
1273 | ||
1274 | /* | |
1275 | * Initialise the necessary environnement : | |
1276 | * - create a new context | |
1277 | * - create the poll_pipe | |
1278 | * - create the should_quit pipe (for signal handler) | |
1279 | * - create the thread pipe (for splice) | |
1280 | * | |
1281 | * Takes a function pointer as argument, this function is called when data is | |
1282 | * available on a buffer. This function is responsible to do the | |
1283 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1284 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1285 | * | |
1286 | * Returns a pointer to the new context or NULL on error. | |
1287 | */ | |
1288 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1289 | enum lttng_consumer_type type, | |
4078b776 | 1290 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1291 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1292 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1293 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1294 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1295 | { |
d8ef542d | 1296 | int ret; |
3bd1e081 MD |
1297 | struct lttng_consumer_local_data *ctx; |
1298 | ||
1299 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1300 | consumer_data.type == type); | |
1301 | consumer_data.type = type; | |
1302 | ||
effcf122 | 1303 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1304 | if (ctx == NULL) { |
7a57cf92 | 1305 | PERROR("allocating context"); |
3bd1e081 MD |
1306 | goto error; |
1307 | } | |
1308 | ||
1309 | ctx->consumer_error_socket = -1; | |
331744e3 | 1310 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1311 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1312 | /* assign the callbacks */ |
1313 | ctx->on_buffer_ready = buffer_ready; | |
1314 | ctx->on_recv_channel = recv_channel; | |
1315 | ctx->on_recv_stream = recv_stream; | |
1316 | ctx->on_update_stream = update_stream; | |
1317 | ||
acdb9057 DG |
1318 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1319 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1320 | goto error_poll_pipe; |
1321 | } | |
1322 | ||
02b3d176 DG |
1323 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1324 | if (!ctx->consumer_wakeup_pipe) { | |
1325 | goto error_wakeup_pipe; | |
1326 | } | |
1327 | ||
3bd1e081 MD |
1328 | ret = pipe(ctx->consumer_should_quit); |
1329 | if (ret < 0) { | |
7a57cf92 | 1330 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1331 | goto error_quit_pipe; |
1332 | } | |
1333 | ||
d8ef542d MD |
1334 | ret = pipe(ctx->consumer_channel_pipe); |
1335 | if (ret < 0) { | |
1336 | PERROR("Error creating channel pipe"); | |
1337 | goto error_channel_pipe; | |
1338 | } | |
1339 | ||
13886d2d DG |
1340 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1341 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1342 | goto error_metadata_pipe; |
1343 | } | |
3bd1e081 | 1344 | |
fb3a43a9 | 1345 | return ctx; |
3bd1e081 | 1346 | |
fb3a43a9 | 1347 | error_metadata_pipe: |
d8ef542d MD |
1348 | utils_close_pipe(ctx->consumer_channel_pipe); |
1349 | error_channel_pipe: | |
d8ef542d | 1350 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1351 | error_quit_pipe: |
02b3d176 DG |
1352 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1353 | error_wakeup_pipe: | |
acdb9057 | 1354 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1355 | error_poll_pipe: |
1356 | free(ctx); | |
1357 | error: | |
1358 | return NULL; | |
1359 | } | |
1360 | ||
282dadbc MD |
1361 | /* |
1362 | * Iterate over all streams of the hashtable and free them properly. | |
1363 | */ | |
1364 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1365 | { | |
1366 | struct lttng_ht_iter iter; | |
1367 | struct lttng_consumer_stream *stream; | |
1368 | ||
1369 | if (ht == NULL) { | |
1370 | return; | |
1371 | } | |
1372 | ||
1373 | rcu_read_lock(); | |
1374 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1375 | /* | |
1376 | * Ignore return value since we are currently cleaning up so any error | |
1377 | * can't be handled. | |
1378 | */ | |
1379 | (void) consumer_del_stream(stream, ht); | |
1380 | } | |
1381 | rcu_read_unlock(); | |
1382 | ||
1383 | lttng_ht_destroy(ht); | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * Iterate over all streams of the metadata hashtable and free them | |
1388 | * properly. | |
1389 | */ | |
1390 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1391 | { | |
1392 | struct lttng_ht_iter iter; | |
1393 | struct lttng_consumer_stream *stream; | |
1394 | ||
1395 | if (ht == NULL) { | |
1396 | return; | |
1397 | } | |
1398 | ||
1399 | rcu_read_lock(); | |
1400 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1401 | /* | |
1402 | * Ignore return value since we are currently cleaning up so any error | |
1403 | * can't be handled. | |
1404 | */ | |
1405 | (void) consumer_del_metadata_stream(stream, ht); | |
1406 | } | |
1407 | rcu_read_unlock(); | |
1408 | ||
1409 | lttng_ht_destroy(ht); | |
1410 | } | |
1411 | ||
3bd1e081 MD |
1412 | /* |
1413 | * Close all fds associated with the instance and free the context. | |
1414 | */ | |
1415 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1416 | { | |
4c462e79 MD |
1417 | int ret; |
1418 | ||
ab1027f4 DG |
1419 | DBG("Consumer destroying it. Closing everything."); |
1420 | ||
4f2e75b9 DG |
1421 | if (!ctx) { |
1422 | return; | |
1423 | } | |
1424 | ||
282dadbc MD |
1425 | destroy_data_stream_ht(data_ht); |
1426 | destroy_metadata_stream_ht(metadata_ht); | |
1427 | ||
4c462e79 MD |
1428 | ret = close(ctx->consumer_error_socket); |
1429 | if (ret) { | |
1430 | PERROR("close"); | |
1431 | } | |
331744e3 JD |
1432 | ret = close(ctx->consumer_metadata_socket); |
1433 | if (ret) { | |
1434 | PERROR("close"); | |
1435 | } | |
d8ef542d | 1436 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1437 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1438 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1439 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1440 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1441 | |
3bd1e081 MD |
1442 | unlink(ctx->consumer_command_sock_path); |
1443 | free(ctx); | |
1444 | } | |
1445 | ||
6197aea7 DG |
1446 | /* |
1447 | * Write the metadata stream id on the specified file descriptor. | |
1448 | */ | |
1449 | static int write_relayd_metadata_id(int fd, | |
1450 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1451 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 | 1452 | { |
6cd525e8 | 1453 | ssize_t ret; |
1d4dfdef | 1454 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1455 | |
1d4dfdef DG |
1456 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1457 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1458 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1459 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1460 | /* |
6f04ed72 | 1461 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1462 | * not to clubber the error output since this can happen in a normal |
1463 | * code path. | |
1464 | */ | |
1465 | if (errno != EPIPE) { | |
1466 | PERROR("write metadata stream id"); | |
1467 | } | |
1468 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1469 | /* |
1470 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1471 | * handle writting the missing part so report that as an error and | |
1472 | * don't lie to the caller. | |
1473 | */ | |
1474 | ret = -1; | |
6197aea7 DG |
1475 | goto end; |
1476 | } | |
1d4dfdef DG |
1477 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1478 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1479 | |
1480 | end: | |
6cd525e8 | 1481 | return (int) ret; |
6197aea7 DG |
1482 | } |
1483 | ||
3bd1e081 | 1484 | /* |
09e26845 DG |
1485 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1486 | * core function for writing trace buffers to either the local filesystem or | |
1487 | * the network. | |
1488 | * | |
79d4ffb7 DG |
1489 | * It must be called with the stream lock held. |
1490 | * | |
09e26845 | 1491 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1492 | * |
1493 | * Returns the number of bytes written | |
1494 | */ | |
4078b776 | 1495 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1496 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1497 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1498 | unsigned long padding, |
50adc264 | 1499 | struct ctf_packet_index *index) |
3bd1e081 | 1500 | { |
f02e1e8a | 1501 | unsigned long mmap_offset; |
ffe60014 | 1502 | void *mmap_base; |
994ab360 | 1503 | ssize_t ret = 0; |
f02e1e8a DG |
1504 | off_t orig_offset = stream->out_fd_offset; |
1505 | /* Default is on the disk */ | |
1506 | int outfd = stream->out_fd; | |
f02e1e8a | 1507 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1508 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1509 | |
1510 | /* RCU lock for the relayd pointer */ | |
1511 | rcu_read_lock(); | |
1512 | ||
1513 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1514 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1515 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1516 | if (relayd == NULL) { | |
56591bac | 1517 | ret = -EPIPE; |
f02e1e8a DG |
1518 | goto end; |
1519 | } | |
1520 | } | |
1521 | ||
1522 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1523 | switch (consumer_data.type) { |
1524 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1525 | mmap_base = stream->mmap_base; |
f02e1e8a | 1526 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
994ab360 DG |
1527 | if (ret < 0) { |
1528 | ret = -errno; | |
56591bac | 1529 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1530 | goto end; |
1531 | } | |
f02e1e8a | 1532 | break; |
7753dea8 MD |
1533 | case LTTNG_CONSUMER32_UST: |
1534 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1535 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1536 | if (!mmap_base) { | |
1537 | ERR("read mmap get mmap base for stream %s", stream->name); | |
994ab360 | 1538 | ret = -EPERM; |
ffe60014 DG |
1539 | goto end; |
1540 | } | |
1541 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1542 | if (ret != 0) { |
1543 | PERROR("tracer ctl get_mmap_read_offset"); | |
994ab360 | 1544 | ret = -EINVAL; |
56591bac MD |
1545 | goto end; |
1546 | } | |
f02e1e8a | 1547 | break; |
3bd1e081 MD |
1548 | default: |
1549 | ERR("Unknown consumer_data type"); | |
1550 | assert(0); | |
1551 | } | |
b9182dd9 | 1552 | |
f02e1e8a DG |
1553 | /* Handle stream on the relayd if the output is on the network */ |
1554 | if (relayd) { | |
1555 | unsigned long netlen = len; | |
1556 | ||
1557 | /* | |
1558 | * Lock the control socket for the complete duration of the function | |
1559 | * since from this point on we will use the socket. | |
1560 | */ | |
1561 | if (stream->metadata_flag) { | |
1562 | /* Metadata requires the control socket. */ | |
1563 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1564 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1565 | } |
1566 | ||
1d4dfdef | 1567 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1568 | if (ret < 0) { |
1569 | relayd_hang_up = 1; | |
1570 | goto write_error; | |
1571 | } | |
1572 | /* Use the returned socket. */ | |
1573 | outfd = ret; | |
f02e1e8a | 1574 | |
994ab360 DG |
1575 | /* Write metadata stream id before payload */ |
1576 | if (stream->metadata_flag) { | |
1577 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); | |
1578 | if (ret < 0) { | |
8994307f DG |
1579 | relayd_hang_up = 1; |
1580 | goto write_error; | |
1581 | } | |
f02e1e8a | 1582 | } |
1d4dfdef DG |
1583 | } else { |
1584 | /* No streaming, we have to set the len with the full padding */ | |
1585 | len += padding; | |
1624d5b7 JD |
1586 | |
1587 | /* | |
1588 | * Check if we need to change the tracefile before writing the packet. | |
1589 | */ | |
1590 | if (stream->chan->tracefile_size > 0 && | |
1591 | (stream->tracefile_size_current + len) > | |
1592 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1593 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1594 | stream->name, stream->chan->tracefile_size, | |
1595 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1596 | stream->out_fd, &(stream->tracefile_count_current), |
1597 | &stream->out_fd); | |
1624d5b7 JD |
1598 | if (ret < 0) { |
1599 | ERR("Rotating output file"); | |
1600 | goto end; | |
1601 | } | |
309167d2 JD |
1602 | outfd = stream->out_fd; |
1603 | ||
1604 | if (stream->index_fd >= 0) { | |
778d7cda JG |
1605 | ret = close(stream->index_fd); |
1606 | if (ret < 0) { | |
1607 | PERROR("Closing index"); | |
1608 | goto end; | |
1609 | } | |
1610 | stream->index_fd = -1; | |
309167d2 JD |
1611 | ret = index_create_file(stream->chan->pathname, |
1612 | stream->name, stream->uid, stream->gid, | |
1613 | stream->chan->tracefile_size, | |
1614 | stream->tracefile_count_current); | |
1615 | if (ret < 0) { | |
1616 | goto end; | |
1617 | } | |
1618 | stream->index_fd = ret; | |
1619 | } | |
1620 | ||
a6976990 DG |
1621 | /* Reset current size because we just perform a rotation. */ |
1622 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1623 | stream->out_fd_offset = 0; |
1624 | orig_offset = 0; | |
1624d5b7 JD |
1625 | } |
1626 | stream->tracefile_size_current += len; | |
309167d2 JD |
1627 | if (index) { |
1628 | index->offset = htobe64(stream->out_fd_offset); | |
1629 | } | |
f02e1e8a DG |
1630 | } |
1631 | ||
d02b8372 DG |
1632 | /* |
1633 | * This call guarantee that len or less is returned. It's impossible to | |
1634 | * receive a ret value that is bigger than len. | |
1635 | */ | |
1636 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1637 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1638 | if (ret < 0 || ((size_t) ret != len)) { | |
1639 | /* | |
1640 | * Report error to caller if nothing was written else at least send the | |
1641 | * amount written. | |
1642 | */ | |
1643 | if (ret < 0) { | |
994ab360 | 1644 | ret = -errno; |
f02e1e8a | 1645 | } |
994ab360 | 1646 | relayd_hang_up = 1; |
f02e1e8a | 1647 | |
d02b8372 | 1648 | /* Socket operation failed. We consider the relayd dead */ |
994ab360 | 1649 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
d02b8372 DG |
1650 | /* |
1651 | * This is possible if the fd is closed on the other side | |
1652 | * (outfd) or any write problem. It can be verbose a bit for a | |
1653 | * normal execution if for instance the relayd is stopped | |
1654 | * abruptly. This can happen so set this to a DBG statement. | |
1655 | */ | |
1656 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1657 | } else { |
1658 | /* Unhandled error, print it and stop function right now. */ | |
1659 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1660 | } |
994ab360 | 1661 | goto write_error; |
d02b8372 DG |
1662 | } |
1663 | stream->output_written += ret; | |
d02b8372 DG |
1664 | |
1665 | /* This call is useless on a socket so better save a syscall. */ | |
1666 | if (!relayd) { | |
1667 | /* This won't block, but will start writeout asynchronously */ | |
1668 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1669 | SYNC_FILE_RANGE_WRITE); | |
1670 | stream->out_fd_offset += len; | |
f02e1e8a DG |
1671 | } |
1672 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1673 | ||
8994307f DG |
1674 | write_error: |
1675 | /* | |
1676 | * This is a special case that the relayd has closed its socket. Let's | |
1677 | * cleanup the relayd object and all associated streams. | |
1678 | */ | |
1679 | if (relayd && relayd_hang_up) { | |
1680 | cleanup_relayd(relayd, ctx); | |
1681 | } | |
1682 | ||
f02e1e8a DG |
1683 | end: |
1684 | /* Unlock only if ctrl socket used */ | |
1685 | if (relayd && stream->metadata_flag) { | |
1686 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1687 | } | |
1688 | ||
1689 | rcu_read_unlock(); | |
994ab360 | 1690 | return ret; |
3bd1e081 MD |
1691 | } |
1692 | ||
1693 | /* | |
1694 | * Splice the data from the ring buffer to the tracefile. | |
1695 | * | |
79d4ffb7 DG |
1696 | * It must be called with the stream lock held. |
1697 | * | |
3bd1e081 MD |
1698 | * Returns the number of bytes spliced. |
1699 | */ | |
4078b776 | 1700 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1701 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1702 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1703 | unsigned long padding, |
50adc264 | 1704 | struct ctf_packet_index *index) |
3bd1e081 | 1705 | { |
f02e1e8a DG |
1706 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1707 | loff_t offset = 0; | |
1708 | off_t orig_offset = stream->out_fd_offset; | |
1709 | int fd = stream->wait_fd; | |
1710 | /* Default is on the disk */ | |
1711 | int outfd = stream->out_fd; | |
f02e1e8a | 1712 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1713 | int *splice_pipe; |
8994307f | 1714 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1715 | |
3bd1e081 MD |
1716 | switch (consumer_data.type) { |
1717 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1718 | break; |
7753dea8 MD |
1719 | case LTTNG_CONSUMER32_UST: |
1720 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1721 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1722 | return -ENOSYS; |
1723 | default: | |
1724 | ERR("Unknown consumer_data type"); | |
1725 | assert(0); | |
3bd1e081 MD |
1726 | } |
1727 | ||
f02e1e8a DG |
1728 | /* RCU lock for the relayd pointer */ |
1729 | rcu_read_lock(); | |
1730 | ||
1731 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1732 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1733 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1734 | if (relayd == NULL) { | |
ad0b0d23 | 1735 | written = -ret; |
f02e1e8a DG |
1736 | goto end; |
1737 | } | |
1738 | } | |
a2361a61 | 1739 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1740 | |
f02e1e8a | 1741 | /* Write metadata stream id before payload */ |
1d4dfdef | 1742 | if (relayd) { |
ad0b0d23 | 1743 | unsigned long total_len = len; |
f02e1e8a | 1744 | |
1d4dfdef DG |
1745 | if (stream->metadata_flag) { |
1746 | /* | |
1747 | * Lock the control socket for the complete duration of the function | |
1748 | * since from this point on we will use the socket. | |
1749 | */ | |
1750 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1751 | ||
1752 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1753 | padding); | |
1754 | if (ret < 0) { | |
1755 | written = ret; | |
ad0b0d23 DG |
1756 | relayd_hang_up = 1; |
1757 | goto write_error; | |
1d4dfdef DG |
1758 | } |
1759 | ||
1760 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1761 | } | |
1762 | ||
1763 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1764 | if (ret < 0) { |
1765 | written = ret; | |
1766 | relayd_hang_up = 1; | |
1767 | goto write_error; | |
f02e1e8a | 1768 | } |
ad0b0d23 DG |
1769 | /* Use the returned socket. */ |
1770 | outfd = ret; | |
1d4dfdef DG |
1771 | } else { |
1772 | /* No streaming, we have to set the len with the full padding */ | |
1773 | len += padding; | |
1624d5b7 JD |
1774 | |
1775 | /* | |
1776 | * Check if we need to change the tracefile before writing the packet. | |
1777 | */ | |
1778 | if (stream->chan->tracefile_size > 0 && | |
1779 | (stream->tracefile_size_current + len) > | |
1780 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1781 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1782 | stream->name, stream->chan->tracefile_size, | |
1783 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1784 | stream->out_fd, &(stream->tracefile_count_current), |
1785 | &stream->out_fd); | |
1624d5b7 | 1786 | if (ret < 0) { |
ad0b0d23 | 1787 | written = ret; |
1624d5b7 JD |
1788 | ERR("Rotating output file"); |
1789 | goto end; | |
1790 | } | |
309167d2 JD |
1791 | outfd = stream->out_fd; |
1792 | ||
1793 | if (stream->index_fd >= 0) { | |
ad772158 JD |
1794 | ret = close(stream->index_fd); |
1795 | if (ret < 0) { | |
1796 | PERROR("Closing index"); | |
1797 | goto end; | |
1798 | } | |
1799 | stream->index_fd = -1; | |
309167d2 JD |
1800 | ret = index_create_file(stream->chan->pathname, |
1801 | stream->name, stream->uid, stream->gid, | |
1802 | stream->chan->tracefile_size, | |
1803 | stream->tracefile_count_current); | |
1804 | if (ret < 0) { | |
ad0b0d23 | 1805 | written = ret; |
309167d2 JD |
1806 | goto end; |
1807 | } | |
1808 | stream->index_fd = ret; | |
1809 | } | |
1810 | ||
a6976990 DG |
1811 | /* Reset current size because we just perform a rotation. */ |
1812 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1813 | stream->out_fd_offset = 0; |
1814 | orig_offset = 0; | |
1624d5b7 JD |
1815 | } |
1816 | stream->tracefile_size_current += len; | |
309167d2 | 1817 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1818 | } |
1819 | ||
1820 | while (len > 0) { | |
1d4dfdef DG |
1821 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1822 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1823 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1824 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1825 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1826 | if (ret_splice < 0) { | |
d02b8372 | 1827 | ret = errno; |
ad0b0d23 | 1828 | written = -ret; |
d02b8372 | 1829 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1830 | goto splice_error; |
1831 | } | |
1832 | ||
1833 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1834 | if (relayd && stream->metadata_flag) { |
1835 | size_t metadata_payload_size = | |
1836 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1837 | ||
1838 | /* Update counter to fit the spliced data */ | |
1839 | ret_splice += metadata_payload_size; | |
1840 | len += metadata_payload_size; | |
1841 | /* | |
1842 | * We do this so the return value can match the len passed as | |
1843 | * argument to this function. | |
1844 | */ | |
1845 | written -= metadata_payload_size; | |
f02e1e8a DG |
1846 | } |
1847 | ||
1848 | /* Splice data out */ | |
fb3a43a9 | 1849 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1850 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1851 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1852 | outfd, ret_splice); | |
f02e1e8a | 1853 | if (ret_splice < 0) { |
d02b8372 | 1854 | ret = errno; |
ad0b0d23 DG |
1855 | written = -ret; |
1856 | relayd_hang_up = 1; | |
1857 | goto write_error; | |
f02e1e8a | 1858 | } else if (ret_splice > len) { |
d02b8372 DG |
1859 | /* |
1860 | * We don't expect this code path to be executed but you never know | |
1861 | * so this is an extra protection agains a buggy splice(). | |
1862 | */ | |
f02e1e8a | 1863 | ret = errno; |
ad0b0d23 | 1864 | written += ret_splice; |
d02b8372 DG |
1865 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1866 | len); | |
f02e1e8a | 1867 | goto splice_error; |
d02b8372 DG |
1868 | } else { |
1869 | /* All good, update current len and continue. */ | |
1870 | len -= ret_splice; | |
f02e1e8a | 1871 | } |
f02e1e8a DG |
1872 | |
1873 | /* This call is useless on a socket so better save a syscall. */ | |
1874 | if (!relayd) { | |
1875 | /* This won't block, but will start writeout asynchronously */ | |
1876 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1877 | SYNC_FILE_RANGE_WRITE); | |
1878 | stream->out_fd_offset += ret_splice; | |
1879 | } | |
e5d1a9b3 | 1880 | stream->output_written += ret_splice; |
f02e1e8a DG |
1881 | written += ret_splice; |
1882 | } | |
1883 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
f02e1e8a DG |
1884 | goto end; |
1885 | ||
8994307f DG |
1886 | write_error: |
1887 | /* | |
1888 | * This is a special case that the relayd has closed its socket. Let's | |
1889 | * cleanup the relayd object and all associated streams. | |
1890 | */ | |
1891 | if (relayd && relayd_hang_up) { | |
1892 | cleanup_relayd(relayd, ctx); | |
1893 | /* Skip splice error so the consumer does not fail */ | |
1894 | goto end; | |
1895 | } | |
1896 | ||
f02e1e8a DG |
1897 | splice_error: |
1898 | /* send the appropriate error description to sessiond */ | |
1899 | switch (ret) { | |
f02e1e8a | 1900 | case EINVAL: |
f73fabfd | 1901 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1902 | break; |
1903 | case ENOMEM: | |
f73fabfd | 1904 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1905 | break; |
1906 | case ESPIPE: | |
f73fabfd | 1907 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1908 | break; |
1909 | } | |
1910 | ||
1911 | end: | |
1912 | if (relayd && stream->metadata_flag) { | |
1913 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1914 | } | |
1915 | ||
1916 | rcu_read_unlock(); | |
1917 | return written; | |
3bd1e081 MD |
1918 | } |
1919 | ||
1920 | /* | |
1921 | * Take a snapshot for a specific fd | |
1922 | * | |
1923 | * Returns 0 on success, < 0 on error | |
1924 | */ | |
ffe60014 | 1925 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1926 | { |
1927 | switch (consumer_data.type) { | |
1928 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1929 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1930 | case LTTNG_CONSUMER32_UST: |
1931 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1932 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1933 | default: |
1934 | ERR("Unknown consumer_data type"); | |
1935 | assert(0); | |
1936 | return -ENOSYS; | |
1937 | } | |
3bd1e081 MD |
1938 | } |
1939 | ||
1940 | /* | |
1941 | * Get the produced position | |
1942 | * | |
1943 | * Returns 0 on success, < 0 on error | |
1944 | */ | |
ffe60014 | 1945 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1946 | unsigned long *pos) |
1947 | { | |
1948 | switch (consumer_data.type) { | |
1949 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1950 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1951 | case LTTNG_CONSUMER32_UST: |
1952 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1953 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1954 | default: |
1955 | ERR("Unknown consumer_data type"); | |
1956 | assert(0); | |
1957 | return -ENOSYS; | |
1958 | } | |
1959 | } | |
1960 | ||
1961 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1962 | int sock, struct pollfd *consumer_sockpoll) | |
1963 | { | |
1964 | switch (consumer_data.type) { | |
1965 | case LTTNG_CONSUMER_KERNEL: | |
1966 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1967 | case LTTNG_CONSUMER32_UST: |
1968 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1969 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1970 | default: | |
1971 | ERR("Unknown consumer_data type"); | |
1972 | assert(0); | |
1973 | return -ENOSYS; | |
1974 | } | |
1975 | } | |
1976 | ||
6d574024 | 1977 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
1978 | { |
1979 | switch (consumer_data.type) { | |
1980 | case LTTNG_CONSUMER_KERNEL: | |
1981 | /* | |
1982 | * The Kernel consumer has a different metadata scheme so we don't | |
1983 | * close anything because the stream will be closed by the session | |
1984 | * daemon. | |
1985 | */ | |
1986 | break; | |
1987 | case LTTNG_CONSUMER32_UST: | |
1988 | case LTTNG_CONSUMER64_UST: | |
1989 | /* | |
1990 | * Close all metadata streams. The metadata hash table is passed and | |
1991 | * this call iterates over it by closing all wakeup fd. This is safe | |
1992 | * because at this point we are sure that the metadata producer is | |
1993 | * either dead or blocked. | |
1994 | */ | |
6d574024 | 1995 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
1996 | break; |
1997 | default: | |
1998 | ERR("Unknown consumer_data type"); | |
1999 | assert(0); | |
2000 | } | |
2001 | } | |
2002 | ||
fb3a43a9 DG |
2003 | /* |
2004 | * Clean up a metadata stream and free its memory. | |
2005 | */ | |
e316aad5 DG |
2006 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
2007 | struct lttng_ht *ht) | |
fb3a43a9 | 2008 | { |
e316aad5 | 2009 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
2010 | |
2011 | assert(stream); | |
2012 | /* | |
2013 | * This call should NEVER receive regular stream. It must always be | |
2014 | * metadata stream and this is crucial for data structure synchronization. | |
2015 | */ | |
2016 | assert(stream->metadata_flag); | |
2017 | ||
e316aad5 DG |
2018 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2019 | ||
74251bb8 | 2020 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 2021 | pthread_mutex_lock(&stream->chan->lock); |
8994307f DG |
2022 | pthread_mutex_lock(&stream->lock); |
2023 | ||
6d574024 DG |
2024 | /* Remove any reference to that stream. */ |
2025 | consumer_stream_delete(stream, ht); | |
ca22feea | 2026 | |
6d574024 DG |
2027 | /* Close down everything including the relayd if one. */ |
2028 | consumer_stream_close(stream); | |
2029 | /* Destroy tracer buffers of the stream. */ | |
2030 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2031 | |
2032 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2033 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2034 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2035 | /* Go for channel deletion! */ |
e316aad5 | 2036 | free_chan = stream->chan; |
fb3a43a9 DG |
2037 | } |
2038 | ||
73811ecc DG |
2039 | /* |
2040 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2041 | * channel lock MUST be acquired before being able to check for a NULL |
2042 | * pointer value. | |
73811ecc DG |
2043 | */ |
2044 | stream->chan->metadata_stream = NULL; | |
2045 | ||
8994307f | 2046 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2047 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2048 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2049 | |
2050 | if (free_chan) { | |
2051 | consumer_del_channel(free_chan); | |
2052 | } | |
2053 | ||
6d574024 | 2054 | consumer_stream_free(stream); |
fb3a43a9 DG |
2055 | } |
2056 | ||
2057 | /* | |
2058 | * Action done with the metadata stream when adding it to the consumer internal | |
2059 | * data structures to handle it. | |
2060 | */ | |
5ab66908 | 2061 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2062 | { |
5ab66908 | 2063 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2064 | int ret = 0; |
76082088 | 2065 | struct lttng_ht_iter iter; |
d88aee68 | 2066 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2067 | |
e316aad5 DG |
2068 | assert(stream); |
2069 | assert(ht); | |
2070 | ||
d88aee68 | 2071 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2072 | |
2073 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2074 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2075 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2076 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2077 | |
e316aad5 DG |
2078 | /* |
2079 | * From here, refcounts are updated so be _careful_ when returning an error | |
2080 | * after this point. | |
2081 | */ | |
2082 | ||
fb3a43a9 | 2083 | rcu_read_lock(); |
76082088 DG |
2084 | |
2085 | /* | |
2086 | * Lookup the stream just to make sure it does not exist in our internal | |
2087 | * state. This should NEVER happen. | |
2088 | */ | |
d88aee68 DG |
2089 | lttng_ht_lookup(ht, &stream->key, &iter); |
2090 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2091 | assert(!node); |
2092 | ||
e316aad5 | 2093 | /* |
ffe60014 DG |
2094 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2095 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2096 | * causes the count to become 0 also causes a stream to be added. The |
2097 | * channel deletion will thus be triggered by the following removal of this | |
2098 | * stream. | |
2099 | */ | |
ffe60014 | 2100 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2101 | /* Increment refcount before decrementing nb_init_stream_left */ |
2102 | cmm_smp_wmb(); | |
ffe60014 | 2103 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2104 | } |
2105 | ||
d88aee68 | 2106 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2107 | |
d8ef542d MD |
2108 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2109 | &stream->node_channel_id); | |
2110 | ||
ca22feea DG |
2111 | /* |
2112 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2113 | * the key since the HT does not use it and we allow to add redundant keys | |
2114 | * into this table. | |
2115 | */ | |
d88aee68 | 2116 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2117 | |
fb3a43a9 | 2118 | rcu_read_unlock(); |
e316aad5 | 2119 | |
2e818a6a | 2120 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2121 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2122 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2123 | pthread_mutex_unlock(&consumer_data.lock); |
2124 | return ret; | |
fb3a43a9 DG |
2125 | } |
2126 | ||
8994307f DG |
2127 | /* |
2128 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2129 | */ | |
2130 | static void validate_endpoint_status_data_stream(void) | |
2131 | { | |
2132 | struct lttng_ht_iter iter; | |
2133 | struct lttng_consumer_stream *stream; | |
2134 | ||
2135 | DBG("Consumer delete flagged data stream"); | |
2136 | ||
2137 | rcu_read_lock(); | |
2138 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2139 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2140 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2141 | continue; |
2142 | } | |
2143 | /* Delete it right now */ | |
2144 | consumer_del_stream(stream, data_ht); | |
2145 | } | |
2146 | rcu_read_unlock(); | |
2147 | } | |
2148 | ||
2149 | /* | |
2150 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2151 | */ | |
2152 | static void validate_endpoint_status_metadata_stream( | |
2153 | struct lttng_poll_event *pollset) | |
2154 | { | |
2155 | struct lttng_ht_iter iter; | |
2156 | struct lttng_consumer_stream *stream; | |
2157 | ||
2158 | DBG("Consumer delete flagged metadata stream"); | |
2159 | ||
2160 | assert(pollset); | |
2161 | ||
2162 | rcu_read_lock(); | |
2163 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2164 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2165 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2166 | continue; |
2167 | } | |
2168 | /* | |
2169 | * Remove from pollset so the metadata thread can continue without | |
2170 | * blocking on a deleted stream. | |
2171 | */ | |
2172 | lttng_poll_del(pollset, stream->wait_fd); | |
2173 | ||
2174 | /* Delete it right now */ | |
2175 | consumer_del_metadata_stream(stream, metadata_ht); | |
2176 | } | |
2177 | rcu_read_unlock(); | |
2178 | } | |
2179 | ||
fb3a43a9 DG |
2180 | /* |
2181 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2182 | * network. | |
2183 | */ | |
7d980def | 2184 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2185 | { |
1fc79fb4 | 2186 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2187 | uint32_t revents, nb_fd; |
e316aad5 | 2188 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2189 | struct lttng_ht_iter iter; |
d88aee68 | 2190 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2191 | struct lttng_poll_event events; |
2192 | struct lttng_consumer_local_data *ctx = data; | |
2193 | ssize_t len; | |
2194 | ||
2195 | rcu_register_thread(); | |
2196 | ||
1fc79fb4 MD |
2197 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2198 | ||
2d57de81 MD |
2199 | if (testpoint(consumerd_thread_metadata)) { |
2200 | goto error_testpoint; | |
2201 | } | |
2202 | ||
9ce5646a MD |
2203 | health_code_update(); |
2204 | ||
fb3a43a9 DG |
2205 | DBG("Thread metadata poll started"); |
2206 | ||
fb3a43a9 DG |
2207 | /* Size is set to 1 for the consumer_metadata pipe */ |
2208 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2209 | if (ret < 0) { | |
2210 | ERR("Poll set creation failed"); | |
d8ef542d | 2211 | goto end_poll; |
fb3a43a9 DG |
2212 | } |
2213 | ||
13886d2d DG |
2214 | ret = lttng_poll_add(&events, |
2215 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2216 | if (ret < 0) { |
2217 | goto end; | |
2218 | } | |
2219 | ||
2220 | /* Main loop */ | |
2221 | DBG("Metadata main loop started"); | |
2222 | ||
2223 | while (1) { | |
fb3a43a9 | 2224 | restart: |
7fa2082e | 2225 | health_code_update(); |
9ce5646a | 2226 | health_poll_entry(); |
7fa2082e | 2227 | DBG("Metadata poll wait"); |
fb3a43a9 | 2228 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2229 | DBG("Metadata poll return from wait with %d fd(s)", |
2230 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2231 | health_poll_exit(); |
707341a7 | 2232 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2233 | if (ret < 0) { |
2234 | if (errno == EINTR) { | |
707341a7 | 2235 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2236 | goto restart; |
2237 | } | |
d9607cd7 MD |
2238 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2239 | err = 0; /* All is OK */ | |
2240 | } | |
2241 | goto end; | |
fb3a43a9 DG |
2242 | } |
2243 | ||
0d9c5d77 DG |
2244 | nb_fd = ret; |
2245 | ||
e316aad5 | 2246 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2247 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2248 | health_code_update(); |
2249 | ||
fb3a43a9 DG |
2250 | revents = LTTNG_POLL_GETEV(&events, i); |
2251 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2252 | ||
fd20dac9 MD |
2253 | if (!revents) { |
2254 | /* No activity for this FD (poll implementation). */ | |
2255 | continue; | |
2256 | } | |
2257 | ||
13886d2d | 2258 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
92d6debb | 2259 | if (revents & LPOLLIN) { |
13886d2d DG |
2260 | ssize_t pipe_len; |
2261 | ||
2262 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2263 | &stream, sizeof(stream)); | |
6cd525e8 | 2264 | if (pipe_len < sizeof(stream)) { |
92d6debb MD |
2265 | if (pipe_len < 0) { |
2266 | PERROR("read metadata stream"); | |
2267 | } | |
fb3a43a9 | 2268 | /* |
92d6debb MD |
2269 | * Remove the pipe from the poll set and continue the loop |
2270 | * since their might be data to consume. | |
fb3a43a9 | 2271 | */ |
92d6debb MD |
2272 | lttng_poll_del(&events, |
2273 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2274 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2275 | continue; |
2276 | } | |
2277 | ||
8994307f DG |
2278 | /* A NULL stream means that the state has changed. */ |
2279 | if (stream == NULL) { | |
2280 | /* Check for deleted streams. */ | |
2281 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2282 | goto restart; |
8994307f DG |
2283 | } |
2284 | ||
fb3a43a9 DG |
2285 | DBG("Adding metadata stream %d to poll set", |
2286 | stream->wait_fd); | |
2287 | ||
fb3a43a9 DG |
2288 | /* Add metadata stream to the global poll events list */ |
2289 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2290 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
92d6debb MD |
2291 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2292 | DBG("Metadata thread pipe hung up"); | |
2293 | /* | |
2294 | * Remove the pipe from the poll set and continue the loop | |
2295 | * since their might be data to consume. | |
2296 | */ | |
2297 | lttng_poll_del(&events, | |
2298 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2299 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2300 | continue; | |
2301 | } else { | |
2302 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2303 | goto end; | |
fb3a43a9 DG |
2304 | } |
2305 | ||
e316aad5 | 2306 | /* Handle other stream */ |
fb3a43a9 DG |
2307 | continue; |
2308 | } | |
2309 | ||
d09e1200 | 2310 | rcu_read_lock(); |
d88aee68 DG |
2311 | { |
2312 | uint64_t tmp_id = (uint64_t) pollfd; | |
2313 | ||
2314 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2315 | } | |
2316 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2317 | assert(node); |
fb3a43a9 DG |
2318 | |
2319 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2320 | node); |
fb3a43a9 | 2321 | |
92d6debb MD |
2322 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2323 | /* Get the data out of the metadata file descriptor */ | |
2324 | DBG("Metadata available on fd %d", pollfd); | |
2325 | assert(stream->wait_fd == pollfd); | |
2326 | ||
2327 | do { | |
2328 | health_code_update(); | |
2329 | ||
2330 | len = ctx->on_buffer_ready(stream, ctx); | |
2331 | /* | |
2332 | * We don't check the return value here since if we get | |
2333 | * a negative len, it means an error occured thus we | |
2334 | * simply remove it from the poll set and free the | |
2335 | * stream. | |
2336 | */ | |
2337 | } while (len > 0); | |
2338 | ||
2339 | /* It's ok to have an unavailable sub-buffer */ | |
2340 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2341 | /* Clean up stream from consumer and free it. */ | |
2342 | lttng_poll_del(&events, stream->wait_fd); | |
2343 | consumer_del_metadata_stream(stream, metadata_ht); | |
2344 | } | |
2345 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2346 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2347 | if (!stream->hangup_flush_done |
2348 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2349 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2350 | DBG("Attempting to flush and consume the UST buffers"); | |
2351 | lttng_ustconsumer_on_stream_hangup(stream); | |
2352 | ||
2353 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2354 | do { |
9ce5646a MD |
2355 | health_code_update(); |
2356 | ||
4bb94b75 DG |
2357 | len = ctx->on_buffer_ready(stream, ctx); |
2358 | /* | |
2359 | * We don't check the return value here since if we get | |
2360 | * a negative len, it means an error occured thus we | |
2361 | * simply remove it from the poll set and free the | |
2362 | * stream. | |
2363 | */ | |
2364 | } while (len > 0); | |
fb3a43a9 DG |
2365 | } |
2366 | ||
fb3a43a9 | 2367 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2368 | /* |
2369 | * This call update the channel states, closes file descriptors | |
2370 | * and securely free the stream. | |
2371 | */ | |
2372 | consumer_del_metadata_stream(stream, metadata_ht); | |
92d6debb MD |
2373 | } else { |
2374 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
19cee80d | 2375 | rcu_read_unlock(); |
92d6debb | 2376 | goto end; |
fb3a43a9 | 2377 | } |
e316aad5 | 2378 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2379 | rcu_read_unlock(); |
fb3a43a9 DG |
2380 | } |
2381 | } | |
2382 | ||
1fc79fb4 MD |
2383 | /* All is OK */ |
2384 | err = 0; | |
fb3a43a9 DG |
2385 | end: |
2386 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2387 | |
d8ef542d MD |
2388 | lttng_poll_clean(&events); |
2389 | end_poll: | |
2d57de81 | 2390 | error_testpoint: |
1fc79fb4 MD |
2391 | if (err) { |
2392 | health_error(); | |
2393 | ERR("Health error occurred in %s", __func__); | |
2394 | } | |
2395 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2396 | rcu_unregister_thread(); |
2397 | return NULL; | |
2398 | } | |
2399 | ||
3bd1e081 | 2400 | /* |
e4421fec | 2401 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2402 | * it to tracefile if necessary. |
2403 | */ | |
7d980def | 2404 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2405 | { |
1fc79fb4 | 2406 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2407 | struct pollfd *pollfd = NULL; |
2408 | /* local view of the streams */ | |
c869f647 | 2409 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2410 | /* local view of consumer_data.fds_count */ |
2411 | int nb_fd = 0; | |
3bd1e081 | 2412 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2413 | ssize_t len; |
3bd1e081 | 2414 | |
e7b994a3 DG |
2415 | rcu_register_thread(); |
2416 | ||
1fc79fb4 MD |
2417 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2418 | ||
2d57de81 MD |
2419 | if (testpoint(consumerd_thread_data)) { |
2420 | goto error_testpoint; | |
2421 | } | |
2422 | ||
9ce5646a MD |
2423 | health_code_update(); |
2424 | ||
4df6c8cb MD |
2425 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2426 | if (local_stream == NULL) { | |
2427 | PERROR("local_stream malloc"); | |
2428 | goto end; | |
2429 | } | |
3bd1e081 MD |
2430 | |
2431 | while (1) { | |
9ce5646a MD |
2432 | health_code_update(); |
2433 | ||
3bd1e081 MD |
2434 | high_prio = 0; |
2435 | num_hup = 0; | |
2436 | ||
2437 | /* | |
e4421fec | 2438 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2439 | * local array as well |
2440 | */ | |
2441 | pthread_mutex_lock(&consumer_data.lock); | |
2442 | if (consumer_data.need_update) { | |
0e428499 DG |
2443 | free(pollfd); |
2444 | pollfd = NULL; | |
2445 | ||
2446 | free(local_stream); | |
2447 | local_stream = NULL; | |
3bd1e081 | 2448 | |
02b3d176 DG |
2449 | /* |
2450 | * Allocate for all fds +1 for the consumer_data_pipe and +1 for | |
2451 | * wake up pipe. | |
2452 | */ | |
2453 | pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd)); | |
3bd1e081 | 2454 | if (pollfd == NULL) { |
7a57cf92 | 2455 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2456 | pthread_mutex_unlock(&consumer_data.lock); |
2457 | goto end; | |
2458 | } | |
2459 | ||
02b3d176 | 2460 | local_stream = zmalloc((consumer_data.stream_count + 2) * |
747f8642 | 2461 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2462 | if (local_stream == NULL) { |
7a57cf92 | 2463 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2464 | pthread_mutex_unlock(&consumer_data.lock); |
2465 | goto end; | |
2466 | } | |
ffe60014 | 2467 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2468 | data_ht); |
3bd1e081 MD |
2469 | if (ret < 0) { |
2470 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2471 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2472 | pthread_mutex_unlock(&consumer_data.lock); |
2473 | goto end; | |
2474 | } | |
2475 | nb_fd = ret; | |
2476 | consumer_data.need_update = 0; | |
2477 | } | |
2478 | pthread_mutex_unlock(&consumer_data.lock); | |
2479 | ||
4078b776 MD |
2480 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2481 | if (nb_fd == 0 && consumer_quit == 1) { | |
1fc79fb4 | 2482 | err = 0; /* All is OK */ |
4078b776 MD |
2483 | goto end; |
2484 | } | |
3bd1e081 | 2485 | /* poll on the array of fds */ |
88f2b785 | 2486 | restart: |
02b3d176 | 2487 | DBG("polling on %d fd", nb_fd + 2); |
9ce5646a | 2488 | health_poll_entry(); |
02b3d176 | 2489 | num_rdy = poll(pollfd, nb_fd + 2, -1); |
9ce5646a | 2490 | health_poll_exit(); |
3bd1e081 MD |
2491 | DBG("poll num_rdy : %d", num_rdy); |
2492 | if (num_rdy == -1) { | |
88f2b785 MD |
2493 | /* |
2494 | * Restart interrupted system call. | |
2495 | */ | |
2496 | if (errno == EINTR) { | |
2497 | goto restart; | |
2498 | } | |
7a57cf92 | 2499 | PERROR("Poll error"); |
f73fabfd | 2500 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2501 | goto end; |
2502 | } else if (num_rdy == 0) { | |
2503 | DBG("Polling thread timed out"); | |
2504 | goto end; | |
2505 | } | |
2506 | ||
3bd1e081 | 2507 | /* |
50f8ae69 | 2508 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2509 | * beginning of the loop to update the array. We want to prioritize |
2510 | * array update over low-priority reads. | |
3bd1e081 | 2511 | */ |
509bb1cf | 2512 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2513 | ssize_t pipe_readlen; |
04fdd819 | 2514 | |
50f8ae69 | 2515 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2516 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2517 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2518 | if (pipe_readlen < sizeof(new_stream)) { |
2519 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2520 | /* Continue so we can at least handle the current stream(s). */ |
2521 | continue; | |
2522 | } | |
c869f647 DG |
2523 | |
2524 | /* | |
2525 | * If the stream is NULL, just ignore it. It's also possible that | |
2526 | * the sessiond poll thread changed the consumer_quit state and is | |
2527 | * waking us up to test it. | |
2528 | */ | |
2529 | if (new_stream == NULL) { | |
8994307f | 2530 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2531 | continue; |
2532 | } | |
2533 | ||
c869f647 | 2534 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2535 | continue; |
2536 | } | |
2537 | ||
02b3d176 DG |
2538 | /* Handle wakeup pipe. */ |
2539 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2540 | char dummy; | |
2541 | ssize_t pipe_readlen; | |
2542 | ||
2543 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2544 | sizeof(dummy)); | |
2545 | if (pipe_readlen < 0) { | |
2546 | PERROR("Consumer data wakeup pipe"); | |
2547 | } | |
2548 | /* We've been awakened to handle stream(s). */ | |
2549 | ctx->has_wakeup = 0; | |
2550 | } | |
2551 | ||
3bd1e081 MD |
2552 | /* Take care of high priority channels first. */ |
2553 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2554 | health_code_update(); |
2555 | ||
9617607b DG |
2556 | if (local_stream[i] == NULL) { |
2557 | continue; | |
2558 | } | |
fb3a43a9 | 2559 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2560 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2561 | high_prio = 1; | |
4078b776 | 2562 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2563 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2564 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2565 | /* Clean the stream and free it. */ |
2566 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2567 | local_stream[i] = NULL; |
4078b776 MD |
2568 | } else if (len > 0) { |
2569 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2570 | } |
3bd1e081 MD |
2571 | } |
2572 | } | |
2573 | ||
4078b776 MD |
2574 | /* |
2575 | * If we read high prio channel in this loop, try again | |
2576 | * for more high prio data. | |
2577 | */ | |
2578 | if (high_prio) { | |
3bd1e081 MD |
2579 | continue; |
2580 | } | |
2581 | ||
2582 | /* Take care of low priority channels. */ | |
4078b776 | 2583 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2584 | health_code_update(); |
2585 | ||
9617607b DG |
2586 | if (local_stream[i] == NULL) { |
2587 | continue; | |
2588 | } | |
4078b776 | 2589 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2590 | local_stream[i]->hangup_flush_done || |
2591 | local_stream[i]->has_data) { | |
4078b776 MD |
2592 | DBG("Normal read on fd %d", pollfd[i].fd); |
2593 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2594 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2595 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2596 | /* Clean the stream and free it. */ |
2597 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2598 | local_stream[i] = NULL; |
4078b776 MD |
2599 | } else if (len > 0) { |
2600 | local_stream[i]->data_read = 1; | |
2601 | } | |
2602 | } | |
2603 | } | |
2604 | ||
2605 | /* Handle hangup and errors */ | |
2606 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2607 | health_code_update(); |
2608 | ||
9617607b DG |
2609 | if (local_stream[i] == NULL) { |
2610 | continue; | |
2611 | } | |
4078b776 MD |
2612 | if (!local_stream[i]->hangup_flush_done |
2613 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2614 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2615 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2616 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2617 | pollfd[i].fd); |
4078b776 MD |
2618 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2619 | /* Attempt read again, for the data we just flushed. */ | |
2620 | local_stream[i]->data_read = 1; | |
2621 | } | |
2622 | /* | |
2623 | * If the poll flag is HUP/ERR/NVAL and we have | |
2624 | * read no data in this pass, we can remove the | |
2625 | * stream from its hash table. | |
2626 | */ | |
2627 | if ((pollfd[i].revents & POLLHUP)) { | |
2628 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2629 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2630 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2631 | local_stream[i] = NULL; |
4078b776 MD |
2632 | num_hup++; |
2633 | } | |
2634 | } else if (pollfd[i].revents & POLLERR) { | |
2635 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2636 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2637 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2638 | local_stream[i] = NULL; |
4078b776 MD |
2639 | num_hup++; |
2640 | } | |
2641 | } else if (pollfd[i].revents & POLLNVAL) { | |
2642 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2643 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2644 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2645 | local_stream[i] = NULL; |
4078b776 | 2646 | num_hup++; |
3bd1e081 MD |
2647 | } |
2648 | } | |
9617607b DG |
2649 | if (local_stream[i] != NULL) { |
2650 | local_stream[i]->data_read = 0; | |
2651 | } | |
3bd1e081 MD |
2652 | } |
2653 | } | |
1fc79fb4 MD |
2654 | /* All is OK */ |
2655 | err = 0; | |
3bd1e081 MD |
2656 | end: |
2657 | DBG("polling thread exiting"); | |
0e428499 DG |
2658 | free(pollfd); |
2659 | free(local_stream); | |
fb3a43a9 DG |
2660 | |
2661 | /* | |
2662 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2663 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2664 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2665 | * not return and could create a endless wait period if the pipe is the | |
2666 | * only tracked fd in the poll set. The thread will take care of closing | |
2667 | * the read side. | |
fb3a43a9 | 2668 | */ |
13886d2d | 2669 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2670 | |
2d57de81 | 2671 | error_testpoint: |
1fc79fb4 MD |
2672 | if (err) { |
2673 | health_error(); | |
2674 | ERR("Health error occurred in %s", __func__); | |
2675 | } | |
2676 | health_unregister(health_consumerd); | |
2677 | ||
e7b994a3 | 2678 | rcu_unregister_thread(); |
3bd1e081 MD |
2679 | return NULL; |
2680 | } | |
2681 | ||
d8ef542d MD |
2682 | /* |
2683 | * Close wake-up end of each stream belonging to the channel. This will | |
2684 | * allow the poll() on the stream read-side to detect when the | |
2685 | * write-side (application) finally closes them. | |
2686 | */ | |
2687 | static | |
2688 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2689 | { | |
2690 | struct lttng_ht *ht; | |
2691 | struct lttng_consumer_stream *stream; | |
2692 | struct lttng_ht_iter iter; | |
2693 | ||
2694 | ht = consumer_data.stream_per_chan_id_ht; | |
2695 | ||
2696 | rcu_read_lock(); | |
2697 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2698 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2699 | ht->match_fct, &channel->key, | |
2700 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2701 | /* |
2702 | * Protect against teardown with mutex. | |
2703 | */ | |
2704 | pthread_mutex_lock(&stream->lock); | |
2705 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2706 | goto next; | |
2707 | } | |
d8ef542d MD |
2708 | switch (consumer_data.type) { |
2709 | case LTTNG_CONSUMER_KERNEL: | |
2710 | break; | |
2711 | case LTTNG_CONSUMER32_UST: | |
2712 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2713 | if (stream->metadata_flag) { |
2714 | /* Safe and protected by the stream lock. */ | |
2715 | lttng_ustconsumer_close_metadata(stream->chan); | |
2716 | } else { | |
2717 | /* | |
2718 | * Note: a mutex is taken internally within | |
2719 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2720 | * use from concurrent close. | |
2721 | */ | |
2722 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2723 | } | |
d8ef542d MD |
2724 | break; |
2725 | default: | |
2726 | ERR("Unknown consumer_data type"); | |
2727 | assert(0); | |
2728 | } | |
f2ad556d MD |
2729 | next: |
2730 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2731 | } |
2732 | rcu_read_unlock(); | |
2733 | } | |
2734 | ||
2735 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2736 | { | |
2737 | struct lttng_ht_iter iter; | |
2738 | struct lttng_consumer_channel *channel; | |
2739 | int ret; | |
2740 | ||
2741 | if (ht == NULL) { | |
2742 | return; | |
2743 | } | |
2744 | ||
2745 | rcu_read_lock(); | |
2746 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2747 | ret = lttng_ht_del(ht, &iter); | |
2748 | assert(ret != 0); | |
2749 | } | |
2750 | rcu_read_unlock(); | |
2751 | ||
2752 | lttng_ht_destroy(ht); | |
2753 | } | |
2754 | ||
2755 | /* | |
2756 | * This thread polls the channel fds to detect when they are being | |
2757 | * closed. It closes all related streams if the channel is detected as | |
2758 | * closed. It is currently only used as a shim layer for UST because the | |
2759 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2760 | * periodical flush. | |
2761 | */ | |
2762 | void *consumer_thread_channel_poll(void *data) | |
2763 | { | |
1fc79fb4 | 2764 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2765 | uint32_t revents, nb_fd; |
2766 | struct lttng_consumer_channel *chan = NULL; | |
2767 | struct lttng_ht_iter iter; | |
2768 | struct lttng_ht_node_u64 *node; | |
2769 | struct lttng_poll_event events; | |
2770 | struct lttng_consumer_local_data *ctx = data; | |
2771 | struct lttng_ht *channel_ht; | |
2772 | ||
2773 | rcu_register_thread(); | |
2774 | ||
1fc79fb4 MD |
2775 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2776 | ||
2d57de81 MD |
2777 | if (testpoint(consumerd_thread_channel)) { |
2778 | goto error_testpoint; | |
2779 | } | |
2780 | ||
9ce5646a MD |
2781 | health_code_update(); |
2782 | ||
d8ef542d MD |
2783 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2784 | if (!channel_ht) { | |
2785 | /* ENOMEM at this point. Better to bail out. */ | |
2786 | goto end_ht; | |
2787 | } | |
2788 | ||
2789 | DBG("Thread channel poll started"); | |
2790 | ||
2791 | /* Size is set to 1 for the consumer_channel pipe */ | |
2792 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2793 | if (ret < 0) { | |
2794 | ERR("Poll set creation failed"); | |
2795 | goto end_poll; | |
2796 | } | |
2797 | ||
2798 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2799 | if (ret < 0) { | |
2800 | goto end; | |
2801 | } | |
2802 | ||
2803 | /* Main loop */ | |
2804 | DBG("Channel main loop started"); | |
2805 | ||
2806 | while (1) { | |
d8ef542d | 2807 | restart: |
7fa2082e MD |
2808 | health_code_update(); |
2809 | DBG("Channel poll wait"); | |
9ce5646a | 2810 | health_poll_entry(); |
d8ef542d | 2811 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2812 | DBG("Channel poll return from wait with %d fd(s)", |
2813 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2814 | health_poll_exit(); |
707341a7 | 2815 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2816 | if (ret < 0) { |
2817 | if (errno == EINTR) { | |
707341a7 | 2818 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2819 | goto restart; |
2820 | } | |
d9607cd7 MD |
2821 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2822 | err = 0; /* All is OK */ | |
2823 | } | |
d8ef542d MD |
2824 | goto end; |
2825 | } | |
2826 | ||
2827 | nb_fd = ret; | |
2828 | ||
2829 | /* From here, the event is a channel wait fd */ | |
2830 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2831 | health_code_update(); |
2832 | ||
d8ef542d MD |
2833 | revents = LTTNG_POLL_GETEV(&events, i); |
2834 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2835 | ||
d8ef542d | 2836 | if (!revents) { |
fd20dac9 | 2837 | /* No activity for this FD (poll implementation). */ |
d8ef542d MD |
2838 | continue; |
2839 | } | |
fd20dac9 | 2840 | |
d8ef542d | 2841 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
92d6debb | 2842 | if (revents & LPOLLIN) { |
d8ef542d | 2843 | enum consumer_channel_action action; |
a0cbdd2e | 2844 | uint64_t key; |
d8ef542d | 2845 | |
a0cbdd2e | 2846 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2847 | if (ret <= 0) { |
92d6debb MD |
2848 | if (ret < 0) { |
2849 | ERR("Error reading channel pipe"); | |
2850 | } | |
2851 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
2852 | continue; |
2853 | } | |
2854 | ||
2855 | switch (action) { | |
2856 | case CONSUMER_CHANNEL_ADD: | |
2857 | DBG("Adding channel %d to poll set", | |
2858 | chan->wait_fd); | |
2859 | ||
2860 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2861 | chan->wait_fd); | |
c7260a81 | 2862 | rcu_read_lock(); |
d8ef542d MD |
2863 | lttng_ht_add_unique_u64(channel_ht, |
2864 | &chan->wait_fd_node); | |
c7260a81 | 2865 | rcu_read_unlock(); |
d8ef542d MD |
2866 | /* Add channel to the global poll events list */ |
2867 | lttng_poll_add(&events, chan->wait_fd, | |
92d6debb | 2868 | LPOLLERR | LPOLLHUP); |
d8ef542d | 2869 | break; |
a0cbdd2e MD |
2870 | case CONSUMER_CHANNEL_DEL: |
2871 | { | |
b4a650f3 DG |
2872 | /* |
2873 | * This command should never be called if the channel | |
2874 | * has streams monitored by either the data or metadata | |
2875 | * thread. The consumer only notify this thread with a | |
2876 | * channel del. command if it receives a destroy | |
2877 | * channel command from the session daemon that send it | |
2878 | * if a command prior to the GET_CHANNEL failed. | |
2879 | */ | |
2880 | ||
c7260a81 | 2881 | rcu_read_lock(); |
a0cbdd2e MD |
2882 | chan = consumer_find_channel(key); |
2883 | if (!chan) { | |
c7260a81 | 2884 | rcu_read_unlock(); |
a0cbdd2e MD |
2885 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2886 | break; | |
2887 | } | |
2888 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2889 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2890 | ret = lttng_ht_del(channel_ht, &iter); |
2891 | assert(ret == 0); | |
a0cbdd2e | 2892 | |
f2a444f1 DG |
2893 | switch (consumer_data.type) { |
2894 | case LTTNG_CONSUMER_KERNEL: | |
2895 | break; | |
2896 | case LTTNG_CONSUMER32_UST: | |
2897 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
2898 | health_code_update(); |
2899 | /* Destroy streams that might have been left in the stream list. */ | |
2900 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
2901 | break; |
2902 | default: | |
2903 | ERR("Unknown consumer_data type"); | |
2904 | assert(0); | |
2905 | } | |
2906 | ||
a0cbdd2e MD |
2907 | /* |
2908 | * Release our own refcount. Force channel deletion even if | |
2909 | * streams were not initialized. | |
2910 | */ | |
2911 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2912 | consumer_del_channel(chan); | |
2913 | } | |
c7260a81 | 2914 | rcu_read_unlock(); |
a0cbdd2e MD |
2915 | goto restart; |
2916 | } | |
d8ef542d MD |
2917 | case CONSUMER_CHANNEL_QUIT: |
2918 | /* | |
2919 | * Remove the pipe from the poll set and continue the loop | |
2920 | * since their might be data to consume. | |
2921 | */ | |
2922 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2923 | continue; | |
2924 | default: | |
2925 | ERR("Unknown action"); | |
2926 | break; | |
2927 | } | |
92d6debb MD |
2928 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2929 | DBG("Channel thread pipe hung up"); | |
2930 | /* | |
2931 | * Remove the pipe from the poll set and continue the loop | |
2932 | * since their might be data to consume. | |
2933 | */ | |
2934 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2935 | continue; | |
2936 | } else { | |
2937 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2938 | goto end; | |
d8ef542d MD |
2939 | } |
2940 | ||
2941 | /* Handle other stream */ | |
2942 | continue; | |
2943 | } | |
2944 | ||
2945 | rcu_read_lock(); | |
2946 | { | |
2947 | uint64_t tmp_id = (uint64_t) pollfd; | |
2948 | ||
2949 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2950 | } | |
2951 | node = lttng_ht_iter_get_node_u64(&iter); | |
2952 | assert(node); | |
2953 | ||
2954 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2955 | wait_fd_node); | |
2956 | ||
2957 | /* Check for error event */ | |
2958 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2959 | DBG("Channel fd %d is hup|err.", pollfd); | |
2960 | ||
2961 | lttng_poll_del(&events, chan->wait_fd); | |
2962 | ret = lttng_ht_del(channel_ht, &iter); | |
2963 | assert(ret == 0); | |
b4a650f3 DG |
2964 | |
2965 | /* | |
2966 | * This will close the wait fd for each stream associated to | |
2967 | * this channel AND monitored by the data/metadata thread thus | |
2968 | * will be clean by the right thread. | |
2969 | */ | |
d8ef542d | 2970 | consumer_close_channel_streams(chan); |
f2ad556d MD |
2971 | |
2972 | /* Release our own refcount */ | |
2973 | if (!uatomic_sub_return(&chan->refcount, 1) | |
2974 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
2975 | consumer_del_channel(chan); | |
2976 | } | |
92d6debb MD |
2977 | } else { |
2978 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2979 | rcu_read_unlock(); | |
2980 | goto end; | |
d8ef542d MD |
2981 | } |
2982 | ||
2983 | /* Release RCU lock for the channel looked up */ | |
2984 | rcu_read_unlock(); | |
2985 | } | |
2986 | } | |
2987 | ||
1fc79fb4 MD |
2988 | /* All is OK */ |
2989 | err = 0; | |
d8ef542d MD |
2990 | end: |
2991 | lttng_poll_clean(&events); | |
2992 | end_poll: | |
2993 | destroy_channel_ht(channel_ht); | |
2994 | end_ht: | |
2d57de81 | 2995 | error_testpoint: |
d8ef542d | 2996 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
2997 | if (err) { |
2998 | health_error(); | |
2999 | ERR("Health error occurred in %s", __func__); | |
3000 | } | |
3001 | health_unregister(health_consumerd); | |
d8ef542d MD |
3002 | rcu_unregister_thread(); |
3003 | return NULL; | |
3004 | } | |
3005 | ||
331744e3 JD |
3006 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
3007 | struct pollfd *sockpoll, int client_socket) | |
3008 | { | |
3009 | int ret; | |
3010 | ||
3011 | assert(ctx); | |
3012 | assert(sockpoll); | |
3013 | ||
84382d49 MD |
3014 | ret = lttng_consumer_poll_socket(sockpoll); |
3015 | if (ret) { | |
331744e3 JD |
3016 | goto error; |
3017 | } | |
3018 | DBG("Metadata connection on client_socket"); | |
3019 | ||
3020 | /* Blocking call, waiting for transmission */ | |
3021 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3022 | if (ctx->consumer_metadata_socket < 0) { | |
3023 | WARN("On accept metadata"); | |
3024 | ret = -1; | |
3025 | goto error; | |
3026 | } | |
3027 | ret = 0; | |
3028 | ||
3029 | error: | |
3030 | return ret; | |
3031 | } | |
3032 | ||
3bd1e081 MD |
3033 | /* |
3034 | * This thread listens on the consumerd socket and receives the file | |
3035 | * descriptors from the session daemon. | |
3036 | */ | |
7d980def | 3037 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3038 | { |
1fc79fb4 | 3039 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3040 | /* |
3041 | * structure to poll for incoming data on communication socket avoids | |
3042 | * making blocking sockets. | |
3043 | */ | |
3044 | struct pollfd consumer_sockpoll[2]; | |
3045 | struct lttng_consumer_local_data *ctx = data; | |
3046 | ||
e7b994a3 DG |
3047 | rcu_register_thread(); |
3048 | ||
1fc79fb4 MD |
3049 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3050 | ||
2d57de81 MD |
3051 | if (testpoint(consumerd_thread_sessiond)) { |
3052 | goto error_testpoint; | |
3053 | } | |
3054 | ||
9ce5646a MD |
3055 | health_code_update(); |
3056 | ||
3bd1e081 MD |
3057 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3058 | unlink(ctx->consumer_command_sock_path); | |
3059 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3060 | if (client_socket < 0) { | |
3061 | ERR("Cannot create command socket"); | |
3062 | goto end; | |
3063 | } | |
3064 | ||
3065 | ret = lttcomm_listen_unix_sock(client_socket); | |
3066 | if (ret < 0) { | |
3067 | goto end; | |
3068 | } | |
3069 | ||
32258573 | 3070 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3071 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3072 | /* return < 0 on error, but == 0 is not fatal */ |
3073 | if (ret < 0) { | |
32258573 | 3074 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3075 | goto end; |
3076 | } | |
3077 | ||
3bd1e081 MD |
3078 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3079 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3080 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3081 | consumer_sockpoll[1].fd = client_socket; | |
3082 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3083 | ||
84382d49 MD |
3084 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3085 | if (ret) { | |
3086 | if (ret > 0) { | |
3087 | /* should exit */ | |
3088 | err = 0; | |
3089 | } | |
3bd1e081 MD |
3090 | goto end; |
3091 | } | |
3092 | DBG("Connection on client_socket"); | |
3093 | ||
3094 | /* Blocking call, waiting for transmission */ | |
3095 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3096 | if (sock < 0) { |
3bd1e081 MD |
3097 | WARN("On accept"); |
3098 | goto end; | |
3099 | } | |
3bd1e081 | 3100 | |
331744e3 JD |
3101 | /* |
3102 | * Setup metadata socket which is the second socket connection on the | |
3103 | * command unix socket. | |
3104 | */ | |
3105 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3106 | if (ret) { |
3107 | if (ret > 0) { | |
3108 | /* should exit */ | |
3109 | err = 0; | |
3110 | } | |
331744e3 JD |
3111 | goto end; |
3112 | } | |
3113 | ||
d96f09c6 DG |
3114 | /* This socket is not useful anymore. */ |
3115 | ret = close(client_socket); | |
3116 | if (ret < 0) { | |
3117 | PERROR("close client_socket"); | |
3118 | } | |
3119 | client_socket = -1; | |
3120 | ||
3bd1e081 MD |
3121 | /* update the polling structure to poll on the established socket */ |
3122 | consumer_sockpoll[1].fd = sock; | |
3123 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3124 | ||
3125 | while (1) { | |
9ce5646a MD |
3126 | health_code_update(); |
3127 | ||
3128 | health_poll_entry(); | |
3129 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3130 | health_poll_exit(); | |
84382d49 MD |
3131 | if (ret) { |
3132 | if (ret > 0) { | |
3133 | /* should exit */ | |
3134 | err = 0; | |
3135 | } | |
3bd1e081 MD |
3136 | goto end; |
3137 | } | |
3138 | DBG("Incoming command on sock"); | |
3139 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3140 | if (ret <= 0) { |
3141 | /* | |
3142 | * This could simply be a session daemon quitting. Don't output | |
3143 | * ERR() here. | |
3144 | */ | |
3145 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3146 | err = 0; |
3bd1e081 MD |
3147 | goto end; |
3148 | } | |
3149 | if (consumer_quit) { | |
3150 | DBG("consumer_thread_receive_fds received quit from signal"); | |
1fc79fb4 | 3151 | err = 0; /* All is OK */ |
3bd1e081 MD |
3152 | goto end; |
3153 | } | |
ffe60014 | 3154 | DBG("received command on sock"); |
3bd1e081 | 3155 | } |
1fc79fb4 MD |
3156 | /* All is OK */ |
3157 | err = 0; | |
3158 | ||
3bd1e081 | 3159 | end: |
ffe60014 | 3160 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3161 | |
d88aee68 DG |
3162 | /* |
3163 | * Close metadata streams since the producer is the session daemon which | |
3164 | * just died. | |
3165 | * | |
3166 | * NOTE: for now, this only applies to the UST tracer. | |
3167 | */ | |
6d574024 | 3168 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3169 | |
3bd1e081 MD |
3170 | /* |
3171 | * when all fds have hung up, the polling thread | |
3172 | * can exit cleanly | |
3173 | */ | |
3174 | consumer_quit = 1; | |
3175 | ||
04fdd819 | 3176 | /* |
c869f647 | 3177 | * Notify the data poll thread to poll back again and test the |
8994307f | 3178 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3179 | */ |
acdb9057 | 3180 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3181 | |
a0cbdd2e | 3182 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3183 | |
5c635c72 MD |
3184 | notify_health_quit_pipe(health_quit_pipe); |
3185 | ||
d96f09c6 DG |
3186 | /* Cleaning up possibly open sockets. */ |
3187 | if (sock >= 0) { | |
3188 | ret = close(sock); | |
3189 | if (ret < 0) { | |
3190 | PERROR("close sock sessiond poll"); | |
3191 | } | |
3192 | } | |
3193 | if (client_socket >= 0) { | |
38476d24 | 3194 | ret = close(client_socket); |
d96f09c6 DG |
3195 | if (ret < 0) { |
3196 | PERROR("close client_socket sessiond poll"); | |
3197 | } | |
3198 | } | |
3199 | ||
2d57de81 | 3200 | error_testpoint: |
1fc79fb4 MD |
3201 | if (err) { |
3202 | health_error(); | |
3203 | ERR("Health error occurred in %s", __func__); | |
3204 | } | |
3205 | health_unregister(health_consumerd); | |
3206 | ||
e7b994a3 | 3207 | rcu_unregister_thread(); |
3bd1e081 MD |
3208 | return NULL; |
3209 | } | |
d41f73b7 | 3210 | |
4078b776 | 3211 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3212 | struct lttng_consumer_local_data *ctx) |
3213 | { | |
74251bb8 DG |
3214 | ssize_t ret; |
3215 | ||
3216 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3217 | if (stream->metadata_flag) { |
3218 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3219 | } | |
74251bb8 | 3220 | |
d41f73b7 MD |
3221 | switch (consumer_data.type) { |
3222 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3223 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3224 | break; | |
7753dea8 MD |
3225 | case LTTNG_CONSUMER32_UST: |
3226 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3227 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3228 | break; | |
d41f73b7 MD |
3229 | default: |
3230 | ERR("Unknown consumer_data type"); | |
3231 | assert(0); | |
74251bb8 DG |
3232 | ret = -ENOSYS; |
3233 | break; | |
d41f73b7 | 3234 | } |
74251bb8 | 3235 | |
94d49140 JD |
3236 | if (stream->metadata_flag) { |
3237 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3238 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3239 | } | |
74251bb8 DG |
3240 | pthread_mutex_unlock(&stream->lock); |
3241 | return ret; | |
d41f73b7 MD |
3242 | } |
3243 | ||
3244 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3245 | { | |
3246 | switch (consumer_data.type) { | |
3247 | case LTTNG_CONSUMER_KERNEL: | |
3248 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3249 | case LTTNG_CONSUMER32_UST: |
3250 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3251 | return lttng_ustconsumer_on_recv_stream(stream); |
3252 | default: | |
3253 | ERR("Unknown consumer_data type"); | |
3254 | assert(0); | |
3255 | return -ENOSYS; | |
3256 | } | |
3257 | } | |
e4421fec DG |
3258 | |
3259 | /* | |
3260 | * Allocate and set consumer data hash tables. | |
3261 | */ | |
282dadbc | 3262 | int lttng_consumer_init(void) |
e4421fec | 3263 | { |
d88aee68 | 3264 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3265 | if (!consumer_data.channel_ht) { |
3266 | goto error; | |
3267 | } | |
3268 | ||
d88aee68 | 3269 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3270 | if (!consumer_data.relayd_ht) { |
3271 | goto error; | |
3272 | } | |
3273 | ||
d88aee68 | 3274 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3275 | if (!consumer_data.stream_list_ht) { |
3276 | goto error; | |
3277 | } | |
3278 | ||
d8ef542d | 3279 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3280 | if (!consumer_data.stream_per_chan_id_ht) { |
3281 | goto error; | |
3282 | } | |
3283 | ||
3284 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3285 | if (!data_ht) { | |
3286 | goto error; | |
3287 | } | |
3288 | ||
3289 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3290 | if (!metadata_ht) { | |
3291 | goto error; | |
3292 | } | |
3293 | ||
3294 | return 0; | |
3295 | ||
3296 | error: | |
3297 | return -1; | |
e4421fec | 3298 | } |
7735ef9e DG |
3299 | |
3300 | /* | |
3301 | * Process the ADD_RELAYD command receive by a consumer. | |
3302 | * | |
3303 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3304 | * The caller MUST acquire a RCU read side lock before calling it. | |
3305 | */ | |
da009f2c | 3306 | int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3307 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3308 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3309 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3310 | uint64_t relayd_session_id) | |
7735ef9e | 3311 | { |
cd2b09ed | 3312 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3313 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3314 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3315 | |
6151a90f JD |
3316 | assert(ctx); |
3317 | assert(relayd_sock); | |
3318 | ||
da009f2c | 3319 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3320 | |
3321 | /* Get relayd reference if exists. */ | |
3322 | relayd = consumer_find_relayd(net_seq_idx); | |
3323 | if (relayd == NULL) { | |
da009f2c | 3324 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3325 | /* Not found. Allocate one. */ |
3326 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3327 | if (relayd == NULL) { | |
0d08d75e | 3328 | ret = -ENOMEM; |
618a6a28 MD |
3329 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3330 | goto error; | |
0d08d75e | 3331 | } else { |
30319bcb | 3332 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3333 | relayd_created = 1; |
7735ef9e | 3334 | } |
0d08d75e DG |
3335 | |
3336 | /* | |
3337 | * This code path MUST continue to the consumer send status message to | |
3338 | * we can notify the session daemon and continue our work without | |
3339 | * killing everything. | |
3340 | */ | |
da009f2c MD |
3341 | } else { |
3342 | /* | |
3343 | * relayd key should never be found for control socket. | |
3344 | */ | |
3345 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3346 | } |
3347 | ||
3348 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3349 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3350 | if (ret < 0) { |
0d08d75e | 3351 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3352 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3353 | goto error_nosignal; | |
7735ef9e DG |
3354 | } |
3355 | ||
3356 | /* Poll on consumer socket. */ | |
84382d49 MD |
3357 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3358 | if (ret) { | |
3359 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3360 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
7735ef9e | 3361 | ret = -EINTR; |
618a6a28 | 3362 | goto error_nosignal; |
7735ef9e DG |
3363 | } |
3364 | ||
3365 | /* Get relayd socket from session daemon */ | |
3366 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3367 | if (ret != sizeof(fd)) { | |
7735ef9e | 3368 | ret = -1; |
4028eeb9 | 3369 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3370 | |
3371 | /* | |
3372 | * Failing to receive FDs might indicate a major problem such as | |
3373 | * reaching a fd limit during the receive where the kernel returns a | |
3374 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3375 | * don't take any chances and stop everything. | |
3376 | * | |
3377 | * XXX: Feature request #558 will fix that and avoid this possible | |
3378 | * issue when reaching the fd limit. | |
3379 | */ | |
3380 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3381 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3382 | goto error; |
3383 | } | |
3384 | ||
7735ef9e DG |
3385 | /* Copy socket information and received FD */ |
3386 | switch (sock_type) { | |
3387 | case LTTNG_STREAM_CONTROL: | |
3388 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3389 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3390 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3391 | /* Handle create_sock error. */ |
f66c074c | 3392 | if (ret < 0) { |
618a6a28 | 3393 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3394 | goto error; |
f66c074c | 3395 | } |
da009f2c MD |
3396 | /* |
3397 | * Close the socket created internally by | |
3398 | * lttcomm_create_sock, so we can replace it by the one | |
3399 | * received from sessiond. | |
3400 | */ | |
3401 | if (close(relayd->control_sock.sock.fd)) { | |
3402 | PERROR("close"); | |
3403 | } | |
7735ef9e DG |
3404 | |
3405 | /* Assign new file descriptor */ | |
6151a90f | 3406 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3407 | fd = -1; /* For error path */ |
6151a90f JD |
3408 | /* Assign version values. */ |
3409 | relayd->control_sock.major = relayd_sock->major; | |
3410 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3411 | |
d3e2ba59 | 3412 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3413 | |
7735ef9e DG |
3414 | break; |
3415 | case LTTNG_STREAM_DATA: | |
3416 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3417 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3418 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3419 | /* Handle create_sock error. */ |
f66c074c | 3420 | if (ret < 0) { |
618a6a28 | 3421 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3422 | goto error; |
f66c074c | 3423 | } |
da009f2c MD |
3424 | /* |
3425 | * Close the socket created internally by | |
3426 | * lttcomm_create_sock, so we can replace it by the one | |
3427 | * received from sessiond. | |
3428 | */ | |
3429 | if (close(relayd->data_sock.sock.fd)) { | |
3430 | PERROR("close"); | |
3431 | } | |
7735ef9e DG |
3432 | |
3433 | /* Assign new file descriptor */ | |
6151a90f | 3434 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3435 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3436 | /* Assign version values. */ |
3437 | relayd->data_sock.major = relayd_sock->major; | |
3438 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3439 | break; |
3440 | default: | |
3441 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3442 | ret = -1; |
618a6a28 | 3443 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3444 | goto error; |
3445 | } | |
3446 | ||
d88aee68 | 3447 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3448 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3449 | relayd->net_seq_idx, fd); | |
3450 | ||
618a6a28 MD |
3451 | /* We successfully added the socket. Send status back. */ |
3452 | ret = consumer_send_status_msg(sock, ret_code); | |
3453 | if (ret < 0) { | |
3454 | /* Somehow, the session daemon is not responding anymore. */ | |
3455 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3456 | goto error_nosignal; | |
3457 | } | |
3458 | ||
7735ef9e DG |
3459 | /* |
3460 | * Add relayd socket pair to consumer data hashtable. If object already | |
3461 | * exists or on error, the function gracefully returns. | |
3462 | */ | |
d09e1200 | 3463 | add_relayd(relayd); |
7735ef9e DG |
3464 | |
3465 | /* All good! */ | |
4028eeb9 | 3466 | return 0; |
7735ef9e DG |
3467 | |
3468 | error: | |
618a6a28 MD |
3469 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3470 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3471 | } | |
3472 | ||
3473 | error_nosignal: | |
4028eeb9 DG |
3474 | /* Close received socket if valid. */ |
3475 | if (fd >= 0) { | |
3476 | if (close(fd)) { | |
3477 | PERROR("close received socket"); | |
3478 | } | |
3479 | } | |
cd2b09ed DG |
3480 | |
3481 | if (relayd_created) { | |
cd2b09ed DG |
3482 | free(relayd); |
3483 | } | |
3484 | ||
7735ef9e DG |
3485 | return ret; |
3486 | } | |
ca22feea | 3487 | |
4e9a4686 DG |
3488 | /* |
3489 | * Try to lock the stream mutex. | |
3490 | * | |
3491 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3492 | */ | |
3493 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3494 | { | |
3495 | int ret; | |
3496 | ||
3497 | assert(stream); | |
3498 | ||
3499 | /* | |
3500 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3501 | * being used else where hence there is data still being extracted. | |
3502 | */ | |
3503 | ret = pthread_mutex_trylock(&stream->lock); | |
3504 | if (ret) { | |
3505 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3506 | ret = 0; | |
3507 | goto end; | |
3508 | } | |
3509 | ||
3510 | ret = 1; | |
3511 | ||
3512 | end: | |
3513 | return ret; | |
3514 | } | |
3515 | ||
f7079f67 DG |
3516 | /* |
3517 | * Search for a relayd associated to the session id and return the reference. | |
3518 | * | |
3519 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3520 | * until the relayd object is no longer necessary. | |
3521 | */ | |
3522 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3523 | { | |
3524 | struct lttng_ht_iter iter; | |
f7079f67 | 3525 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3526 | |
3527 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3528 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3529 | node.node) { | |
18261bd1 DG |
3530 | /* |
3531 | * Check by sessiond id which is unique here where the relayd session | |
3532 | * id might not be when having multiple relayd. | |
3533 | */ | |
3534 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3535 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3536 | goto found; |
f7079f67 DG |
3537 | } |
3538 | } | |
3539 | ||
18261bd1 DG |
3540 | return NULL; |
3541 | ||
3542 | found: | |
f7079f67 DG |
3543 | return relayd; |
3544 | } | |
3545 | ||
ca22feea DG |
3546 | /* |
3547 | * Check if for a given session id there is still data needed to be extract | |
3548 | * from the buffers. | |
3549 | * | |
6d805429 | 3550 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3551 | */ |
6d805429 | 3552 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3553 | { |
3554 | int ret; | |
3555 | struct lttng_ht_iter iter; | |
3556 | struct lttng_ht *ht; | |
3557 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3558 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3559 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3560 | |
6d805429 | 3561 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3562 | |
6f6eda74 | 3563 | rcu_read_lock(); |
ca22feea DG |
3564 | pthread_mutex_lock(&consumer_data.lock); |
3565 | ||
3566 | switch (consumer_data.type) { | |
3567 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3568 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3569 | break; |
3570 | case LTTNG_CONSUMER32_UST: | |
3571 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3572 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3573 | break; |
3574 | default: | |
3575 | ERR("Unknown consumer data type"); | |
3576 | assert(0); | |
3577 | } | |
3578 | ||
3579 | /* Ease our life a bit */ | |
3580 | ht = consumer_data.stream_list_ht; | |
3581 | ||
f7079f67 DG |
3582 | relayd = find_relayd_by_session_id(id); |
3583 | if (relayd) { | |
3584 | /* Send init command for data pending. */ | |
3585 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3586 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3587 | relayd->relayd_session_id); | |
3588 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3589 | if (ret < 0) { | |
3590 | /* Communication error thus the relayd so no data pending. */ | |
3591 | goto data_not_pending; | |
3592 | } | |
3593 | } | |
3594 | ||
c8f59ee5 | 3595 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3596 | ht->hash_fct(&id, lttng_ht_seed), |
3597 | ht->match_fct, &id, | |
ca22feea | 3598 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3599 | /* If this call fails, the stream is being used hence data pending. */ |
3600 | ret = stream_try_lock(stream); | |
3601 | if (!ret) { | |
f7079f67 | 3602 | goto data_pending; |
ca22feea | 3603 | } |
ca22feea | 3604 | |
4e9a4686 DG |
3605 | /* |
3606 | * A removed node from the hash table indicates that the stream has | |
3607 | * been deleted thus having a guarantee that the buffers are closed | |
3608 | * on the consumer side. However, data can still be transmitted | |
3609 | * over the network so don't skip the relayd check. | |
3610 | */ | |
3611 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3612 | if (!ret) { | |
3613 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3614 | ret = data_pending(stream); |
3615 | if (ret == 1) { | |
4e9a4686 | 3616 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3617 | goto data_pending; |
4e9a4686 DG |
3618 | } |
3619 | } | |
3620 | ||
3621 | /* Relayd check */ | |
f7079f67 | 3622 | if (relayd) { |
c8f59ee5 DG |
3623 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3624 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3625 | ret = relayd_quiescent_control(&relayd->control_sock, |
3626 | stream->relayd_stream_id); | |
c8f59ee5 | 3627 | } else { |
6d805429 | 3628 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3629 | stream->relayd_stream_id, |
3630 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3631 | } |
3632 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3633 | if (ret == 1) { |
4e9a4686 | 3634 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3635 | goto data_pending; |
c8f59ee5 DG |
3636 | } |
3637 | } | |
4e9a4686 | 3638 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3639 | } |
ca22feea | 3640 | |
f7079f67 DG |
3641 | if (relayd) { |
3642 | unsigned int is_data_inflight = 0; | |
3643 | ||
3644 | /* Send init command for data pending. */ | |
3645 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3646 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3647 | relayd->relayd_session_id, &is_data_inflight); | |
3648 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3649 | if (ret < 0) { |
f7079f67 DG |
3650 | goto data_not_pending; |
3651 | } | |
bdd88757 DG |
3652 | if (is_data_inflight) { |
3653 | goto data_pending; | |
3654 | } | |
f7079f67 DG |
3655 | } |
3656 | ||
ca22feea | 3657 | /* |
f7079f67 DG |
3658 | * Finding _no_ node in the hash table and no inflight data means that the |
3659 | * stream(s) have been removed thus data is guaranteed to be available for | |
3660 | * analysis from the trace files. | |
ca22feea DG |
3661 | */ |
3662 | ||
f7079f67 | 3663 | data_not_pending: |
ca22feea DG |
3664 | /* Data is available to be read by a viewer. */ |
3665 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3666 | rcu_read_unlock(); |
6d805429 | 3667 | return 0; |
ca22feea | 3668 | |
f7079f67 | 3669 | data_pending: |
ca22feea DG |
3670 | /* Data is still being extracted from buffers. */ |
3671 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3672 | rcu_read_unlock(); |
6d805429 | 3673 | return 1; |
ca22feea | 3674 | } |
f50f23d9 DG |
3675 | |
3676 | /* | |
3677 | * Send a ret code status message to the sessiond daemon. | |
3678 | * | |
3679 | * Return the sendmsg() return value. | |
3680 | */ | |
3681 | int consumer_send_status_msg(int sock, int ret_code) | |
3682 | { | |
3683 | struct lttcomm_consumer_status_msg msg; | |
3684 | ||
53efb85a | 3685 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3686 | msg.ret_code = ret_code; |
3687 | ||
3688 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3689 | } | |
ffe60014 DG |
3690 | |
3691 | /* | |
3692 | * Send a channel status message to the sessiond daemon. | |
3693 | * | |
3694 | * Return the sendmsg() return value. | |
3695 | */ | |
3696 | int consumer_send_status_channel(int sock, | |
3697 | struct lttng_consumer_channel *channel) | |
3698 | { | |
3699 | struct lttcomm_consumer_status_channel msg; | |
3700 | ||
3701 | assert(sock >= 0); | |
3702 | ||
53efb85a | 3703 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3704 | if (!channel) { |
0c759fc9 | 3705 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3706 | } else { |
0c759fc9 | 3707 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3708 | msg.key = channel->key; |
3709 | msg.stream_count = channel->streams.count; | |
3710 | } | |
3711 | ||
3712 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3713 | } | |
5c786ded | 3714 | |
d07ceecd MD |
3715 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3716 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3717 | uint64_t max_sb_size) | |
5c786ded | 3718 | { |
d07ceecd | 3719 | unsigned long start_pos; |
5c786ded | 3720 | |
d07ceecd MD |
3721 | if (!nb_packets_per_stream) { |
3722 | return consumed_pos; /* Grab everything */ | |
3723 | } | |
3724 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3725 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3726 | if ((long) (start_pos - consumed_pos) < 0) { | |
3727 | return consumed_pos; /* Grab everything */ | |
3728 | } | |
3729 | return start_pos; | |
5c786ded | 3730 | } |