Commit | Line | Data |
---|---|---|
3bd1e081 MD |
1 | /* |
2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> | |
3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
00e2e675 | 4 | * 2012 - David Goulet <dgoulet@efficios.com> |
3bd1e081 | 5 | * |
d14d33bf AM |
6 | * This program is free software; you can redistribute it and/or modify |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
3bd1e081 | 9 | * |
d14d33bf AM |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
3bd1e081 | 14 | * |
d14d33bf AM |
15 | * You should have received a copy of the GNU General Public License along |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
3bd1e081 MD |
18 | */ |
19 | ||
6c1c0768 | 20 | #define _LGPL_SOURCE |
3bd1e081 | 21 | #include <assert.h> |
3bd1e081 MD |
22 | #include <poll.h> |
23 | #include <pthread.h> | |
24 | #include <stdlib.h> | |
25 | #include <string.h> | |
26 | #include <sys/mman.h> | |
27 | #include <sys/socket.h> | |
28 | #include <sys/types.h> | |
29 | #include <unistd.h> | |
77c7c900 | 30 | #include <inttypes.h> |
331744e3 | 31 | #include <signal.h> |
3bd1e081 | 32 | |
51a9e1c7 | 33 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 34 | #include <common/common.h> |
fb3a43a9 DG |
35 | #include <common/utils.h> |
36 | #include <common/compat/poll.h> | |
f263b7fd | 37 | #include <common/compat/endian.h> |
309167d2 | 38 | #include <common/index/index.h> |
10a8a223 | 39 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 40 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
41 | #include <common/sessiond-comm/sessiond-comm.h> |
42 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 43 | #include <common/relayd/relayd.h> |
10a8a223 | 44 | #include <common/ust-consumer/ust-consumer.h> |
c8fea79c JR |
45 | #include <common/consumer/consumer-timer.h> |
46 | #include <common/consumer/consumer.h> | |
47 | #include <common/consumer/consumer-stream.h> | |
48 | #include <common/consumer/consumer-testpoint.h> | |
49 | #include <common/align.h> | |
5feafd41 | 50 | #include <common/consumer/consumer-metadata-cache.h> |
3bd1e081 MD |
51 | |
52 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
53 | .stream_count = 0, |
54 | .need_update = 1, | |
55 | .type = LTTNG_CONSUMER_UNKNOWN, | |
56 | }; | |
57 | ||
d8ef542d MD |
58 | enum consumer_channel_action { |
59 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 60 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
61 | CONSUMER_CHANNEL_QUIT, |
62 | }; | |
63 | ||
64 | struct consumer_channel_msg { | |
65 | enum consumer_channel_action action; | |
a0cbdd2e MD |
66 | struct lttng_consumer_channel *chan; /* add */ |
67 | uint64_t key; /* del */ | |
d8ef542d MD |
68 | }; |
69 | ||
19246bd6 | 70 | /* Flag used to temporarily pause data consumption from testpoints. */ |
cf0bcb51 JG |
71 | int data_consumption_paused; |
72 | ||
3bd1e081 MD |
73 | /* |
74 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
75 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
76 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
77 | * polling threads. | |
78 | */ | |
f7389fbb | 79 | int consumer_quit; |
3bd1e081 | 80 | |
43c34bc3 | 81 | /* |
43c34bc3 DG |
82 | * Global hash table containing respectively metadata and data streams. The |
83 | * stream element in this ht should only be updated by the metadata poll thread | |
84 | * for the metadata and the data poll thread for the data. | |
85 | */ | |
40dc48e0 DG |
86 | static struct lttng_ht *metadata_ht; |
87 | static struct lttng_ht *data_ht; | |
43c34bc3 | 88 | |
acdb9057 DG |
89 | /* |
90 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
91 | * global state has changed so we just send back the thread in a poll wait | |
92 | * call. | |
93 | */ | |
94 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
95 | { | |
96 | struct lttng_consumer_stream *null_stream = NULL; | |
97 | ||
98 | assert(pipe); | |
99 | ||
100 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
101 | } | |
102 | ||
5c635c72 MD |
103 | static void notify_health_quit_pipe(int *pipe) |
104 | { | |
6cd525e8 | 105 | ssize_t ret; |
5c635c72 | 106 | |
6cd525e8 MD |
107 | ret = lttng_write(pipe[1], "4", 1); |
108 | if (ret < 1) { | |
5c635c72 MD |
109 | PERROR("write consumer health quit"); |
110 | } | |
111 | } | |
112 | ||
d8ef542d MD |
113 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
114 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 115 | uint64_t key, |
d8ef542d MD |
116 | enum consumer_channel_action action) |
117 | { | |
118 | struct consumer_channel_msg msg; | |
6cd525e8 | 119 | ssize_t ret; |
d8ef542d | 120 | |
e56251fc DG |
121 | memset(&msg, 0, sizeof(msg)); |
122 | ||
d8ef542d MD |
123 | msg.action = action; |
124 | msg.chan = chan; | |
f21dae48 | 125 | msg.key = key; |
6cd525e8 MD |
126 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
127 | if (ret < sizeof(msg)) { | |
128 | PERROR("notify_channel_pipe write error"); | |
129 | } | |
d8ef542d MD |
130 | } |
131 | ||
a0cbdd2e MD |
132 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
133 | uint64_t key) | |
134 | { | |
135 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
136 | } | |
137 | ||
d8ef542d MD |
138 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
139 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 140 | uint64_t *key, |
d8ef542d MD |
141 | enum consumer_channel_action *action) |
142 | { | |
143 | struct consumer_channel_msg msg; | |
6cd525e8 | 144 | ssize_t ret; |
d8ef542d | 145 | |
6cd525e8 MD |
146 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
147 | if (ret < sizeof(msg)) { | |
148 | ret = -1; | |
149 | goto error; | |
d8ef542d | 150 | } |
6cd525e8 MD |
151 | *action = msg.action; |
152 | *chan = msg.chan; | |
153 | *key = msg.key; | |
154 | error: | |
155 | return (int) ret; | |
d8ef542d MD |
156 | } |
157 | ||
212d67a2 DG |
158 | /* |
159 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
160 | * visible | |
161 | */ | |
162 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
163 | { | |
164 | struct lttng_consumer_stream *stream, *stmp; | |
165 | ||
166 | assert(channel); | |
167 | ||
168 | /* Delete streams that might have been left in the stream list. */ | |
169 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
170 | send_node) { | |
171 | cds_list_del(&stream->send_node); | |
172 | /* | |
173 | * Once a stream is added to this list, the buffers were created so we | |
174 | * have a guarantee that this call will succeed. Setting the monitor | |
175 | * mode to 0 so we don't lock nor try to delete the stream from the | |
176 | * global hash table. | |
177 | */ | |
178 | stream->monitor = 0; | |
179 | consumer_stream_destroy(stream, NULL); | |
180 | } | |
181 | } | |
182 | ||
3bd1e081 MD |
183 | /* |
184 | * Find a stream. The consumer_data.lock must be locked during this | |
185 | * call. | |
186 | */ | |
d88aee68 | 187 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 188 | struct lttng_ht *ht) |
3bd1e081 | 189 | { |
e4421fec | 190 | struct lttng_ht_iter iter; |
d88aee68 | 191 | struct lttng_ht_node_u64 *node; |
e4421fec | 192 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 193 | |
8389e4f8 DG |
194 | assert(ht); |
195 | ||
d88aee68 DG |
196 | /* -1ULL keys are lookup failures */ |
197 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 198 | return NULL; |
7a57cf92 | 199 | } |
e4421fec | 200 | |
6065ceec DG |
201 | rcu_read_lock(); |
202 | ||
d88aee68 DG |
203 | lttng_ht_lookup(ht, &key, &iter); |
204 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
205 | if (node != NULL) { |
206 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 207 | } |
e4421fec | 208 | |
6065ceec DG |
209 | rcu_read_unlock(); |
210 | ||
e4421fec | 211 | return stream; |
3bd1e081 MD |
212 | } |
213 | ||
da009f2c | 214 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
215 | { |
216 | struct lttng_consumer_stream *stream; | |
217 | ||
04253271 | 218 | rcu_read_lock(); |
ffe60014 | 219 | stream = find_stream(key, ht); |
04253271 | 220 | if (stream) { |
da009f2c | 221 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
222 | /* |
223 | * We don't want the lookup to match, but we still need | |
224 | * to iterate on this stream when iterating over the hash table. Just | |
225 | * change the node key. | |
226 | */ | |
da009f2c | 227 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
228 | } |
229 | rcu_read_unlock(); | |
7ad0a0cb MD |
230 | } |
231 | ||
d56db448 DG |
232 | /* |
233 | * Return a channel object for the given key. | |
234 | * | |
235 | * RCU read side lock MUST be acquired before calling this function and | |
236 | * protects the channel ptr. | |
237 | */ | |
d88aee68 | 238 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 239 | { |
e4421fec | 240 | struct lttng_ht_iter iter; |
d88aee68 | 241 | struct lttng_ht_node_u64 *node; |
e4421fec | 242 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 243 | |
d88aee68 DG |
244 | /* -1ULL keys are lookup failures */ |
245 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 246 | return NULL; |
7a57cf92 | 247 | } |
e4421fec | 248 | |
d88aee68 DG |
249 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
250 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
251 | if (node != NULL) { |
252 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 253 | } |
e4421fec DG |
254 | |
255 | return channel; | |
3bd1e081 MD |
256 | } |
257 | ||
b5a6470f DG |
258 | /* |
259 | * There is a possibility that the consumer does not have enough time between | |
260 | * the close of the channel on the session daemon and the cleanup in here thus | |
261 | * once we have a channel add with an existing key, we know for sure that this | |
262 | * channel will eventually get cleaned up by all streams being closed. | |
263 | * | |
264 | * This function just nullifies the already existing channel key. | |
265 | */ | |
266 | static void steal_channel_key(uint64_t key) | |
267 | { | |
268 | struct lttng_consumer_channel *channel; | |
269 | ||
270 | rcu_read_lock(); | |
271 | channel = consumer_find_channel(key); | |
272 | if (channel) { | |
273 | channel->key = (uint64_t) -1ULL; | |
274 | /* | |
275 | * We don't want the lookup to match, but we still need to iterate on | |
276 | * this channel when iterating over the hash table. Just change the | |
277 | * node key. | |
278 | */ | |
279 | channel->node.key = (uint64_t) -1ULL; | |
280 | } | |
281 | rcu_read_unlock(); | |
282 | } | |
283 | ||
ffe60014 | 284 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 285 | { |
d88aee68 DG |
286 | struct lttng_ht_node_u64 *node = |
287 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
288 | struct lttng_consumer_channel *channel = |
289 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 290 | |
b83e03c4 MD |
291 | switch (consumer_data.type) { |
292 | case LTTNG_CONSUMER_KERNEL: | |
293 | break; | |
294 | case LTTNG_CONSUMER32_UST: | |
295 | case LTTNG_CONSUMER64_UST: | |
296 | lttng_ustconsumer_free_channel(channel); | |
297 | break; | |
298 | default: | |
299 | ERR("Unknown consumer_data type"); | |
300 | abort(); | |
301 | } | |
ffe60014 | 302 | free(channel); |
702b1ea4 MD |
303 | } |
304 | ||
00e2e675 DG |
305 | /* |
306 | * RCU protected relayd socket pair free. | |
307 | */ | |
ffe60014 | 308 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 309 | { |
d88aee68 DG |
310 | struct lttng_ht_node_u64 *node = |
311 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
312 | struct consumer_relayd_sock_pair *relayd = |
313 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
314 | ||
8994307f DG |
315 | /* |
316 | * Close all sockets. This is done in the call RCU since we don't want the | |
317 | * socket fds to be reassigned thus potentially creating bad state of the | |
318 | * relayd object. | |
319 | * | |
320 | * We do not have to lock the control socket mutex here since at this stage | |
321 | * there is no one referencing to this relayd object. | |
322 | */ | |
323 | (void) relayd_close(&relayd->control_sock); | |
324 | (void) relayd_close(&relayd->data_sock); | |
325 | ||
7f3df516 | 326 | pthread_mutex_destroy(&relayd->ctrl_sock_mutex); |
00e2e675 DG |
327 | free(relayd); |
328 | } | |
329 | ||
330 | /* | |
331 | * Destroy and free relayd socket pair object. | |
00e2e675 | 332 | */ |
51230d70 | 333 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
334 | { |
335 | int ret; | |
336 | struct lttng_ht_iter iter; | |
337 | ||
173af62f DG |
338 | if (relayd == NULL) { |
339 | return; | |
340 | } | |
341 | ||
00e2e675 DG |
342 | DBG("Consumer destroy and close relayd socket pair"); |
343 | ||
344 | iter.iter.node = &relayd->node.node; | |
345 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 346 | if (ret != 0) { |
8994307f | 347 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
348 | return; |
349 | } | |
00e2e675 | 350 | |
00e2e675 | 351 | /* RCU free() call */ |
ffe60014 DG |
352 | call_rcu(&relayd->node.head, free_relayd_rcu); |
353 | } | |
354 | ||
355 | /* | |
356 | * Remove a channel from the global list protected by a mutex. This function is | |
357 | * also responsible for freeing its data structures. | |
358 | */ | |
359 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
360 | { | |
361 | int ret; | |
362 | struct lttng_ht_iter iter; | |
363 | ||
d88aee68 | 364 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
365 | |
366 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 367 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 368 | |
212d67a2 DG |
369 | /* Destroy streams that might have been left in the stream list. */ |
370 | clean_channel_stream_list(channel); | |
51e762e5 | 371 | |
d3e2ba59 JD |
372 | if (channel->live_timer_enabled == 1) { |
373 | consumer_timer_live_stop(channel); | |
374 | } | |
e9404c27 JG |
375 | if (channel->monitor_timer_enabled == 1) { |
376 | consumer_timer_monitor_stop(channel); | |
377 | } | |
d3e2ba59 | 378 | |
ffe60014 DG |
379 | switch (consumer_data.type) { |
380 | case LTTNG_CONSUMER_KERNEL: | |
381 | break; | |
382 | case LTTNG_CONSUMER32_UST: | |
383 | case LTTNG_CONSUMER64_UST: | |
384 | lttng_ustconsumer_del_channel(channel); | |
385 | break; | |
386 | default: | |
387 | ERR("Unknown consumer_data type"); | |
388 | assert(0); | |
389 | goto end; | |
390 | } | |
391 | ||
392 | rcu_read_lock(); | |
393 | iter.iter.node = &channel->node.node; | |
394 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
395 | assert(!ret); | |
396 | rcu_read_unlock(); | |
397 | ||
398 | call_rcu(&channel->node.head, free_channel_rcu); | |
399 | end: | |
a9838785 | 400 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 401 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
402 | } |
403 | ||
228b5bf7 DG |
404 | /* |
405 | * Iterate over the relayd hash table and destroy each element. Finally, | |
406 | * destroy the whole hash table. | |
407 | */ | |
408 | static void cleanup_relayd_ht(void) | |
409 | { | |
410 | struct lttng_ht_iter iter; | |
411 | struct consumer_relayd_sock_pair *relayd; | |
412 | ||
413 | rcu_read_lock(); | |
414 | ||
415 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
416 | node.node) { | |
51230d70 | 417 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
418 | } |
419 | ||
228b5bf7 | 420 | rcu_read_unlock(); |
36b588ed MD |
421 | |
422 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
423 | } |
424 | ||
8994307f DG |
425 | /* |
426 | * Update the end point status of all streams having the given network sequence | |
427 | * index (relayd index). | |
428 | * | |
429 | * It's atomically set without having the stream mutex locked which is fine | |
430 | * because we handle the write/read race with a pipe wakeup for each thread. | |
431 | */ | |
da009f2c | 432 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
433 | enum consumer_endpoint_status status) |
434 | { | |
435 | struct lttng_ht_iter iter; | |
436 | struct lttng_consumer_stream *stream; | |
437 | ||
da009f2c | 438 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
439 | |
440 | rcu_read_lock(); | |
441 | ||
442 | /* Let's begin with metadata */ | |
443 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
444 | if (stream->net_seq_idx == net_seq_idx) { | |
445 | uatomic_set(&stream->endpoint_status, status); | |
446 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
447 | } | |
448 | } | |
449 | ||
450 | /* Follow up by the data streams */ | |
451 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
452 | if (stream->net_seq_idx == net_seq_idx) { | |
453 | uatomic_set(&stream->endpoint_status, status); | |
454 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
455 | } | |
456 | } | |
457 | rcu_read_unlock(); | |
458 | } | |
459 | ||
460 | /* | |
461 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
462 | * destroying the object meaning removing it from the relayd hash table, | |
463 | * closing the sockets and freeing the memory in a RCU call. | |
464 | * | |
465 | * If a local data context is available, notify the threads that the streams' | |
466 | * state have changed. | |
467 | */ | |
4c5691ba | 468 | void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd) |
8994307f | 469 | { |
da009f2c | 470 | uint64_t netidx; |
8994307f DG |
471 | |
472 | assert(relayd); | |
473 | ||
4c5691ba | 474 | DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx); |
9617607b | 475 | |
8994307f DG |
476 | /* Save the net sequence index before destroying the object */ |
477 | netidx = relayd->net_seq_idx; | |
478 | ||
479 | /* | |
480 | * Delete the relayd from the relayd hash table, close the sockets and free | |
481 | * the object in a RCU call. | |
482 | */ | |
51230d70 | 483 | consumer_destroy_relayd(relayd); |
8994307f DG |
484 | |
485 | /* Set inactive endpoint to all streams */ | |
486 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
487 | ||
488 | /* | |
489 | * With a local data context, notify the threads that the streams' state | |
490 | * have changed. The write() action on the pipe acts as an "implicit" | |
491 | * memory barrier ordering the updates of the end point status from the | |
492 | * read of this status which happens AFTER receiving this notify. | |
493 | */ | |
4c5691ba JR |
494 | notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe); |
495 | notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe); | |
8994307f DG |
496 | } |
497 | ||
a6ba4fe1 DG |
498 | /* |
499 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
500 | * reaches zero. | |
501 | * | |
502 | * RCU read side lock MUST be aquired before calling this function. | |
503 | */ | |
504 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
505 | { | |
506 | assert(relayd); | |
507 | ||
508 | /* Set destroy flag for this object */ | |
509 | uatomic_set(&relayd->destroy_flag, 1); | |
510 | ||
511 | /* Destroy the relayd if refcount is 0 */ | |
512 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 513 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
514 | } |
515 | } | |
516 | ||
3bd1e081 | 517 | /* |
1d1a276c DG |
518 | * Completly destroy stream from every visiable data structure and the given |
519 | * hash table if one. | |
520 | * | |
521 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 522 | */ |
e316aad5 DG |
523 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
524 | struct lttng_ht *ht) | |
3bd1e081 | 525 | { |
1d1a276c | 526 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
527 | } |
528 | ||
5ab66908 MD |
529 | /* |
530 | * XXX naming of del vs destroy is all mixed up. | |
531 | */ | |
532 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
533 | { | |
534 | consumer_stream_destroy(stream, data_ht); | |
535 | } | |
536 | ||
537 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
538 | { | |
539 | consumer_stream_destroy(stream, metadata_ht); | |
540 | } | |
541 | ||
d88aee68 DG |
542 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
543 | uint64_t stream_key, | |
3bd1e081 | 544 | enum lttng_consumer_stream_state state, |
ffe60014 | 545 | const char *channel_name, |
6df2e2c9 | 546 | uid_t uid, |
00e2e675 | 547 | gid_t gid, |
57a269f2 | 548 | uint64_t relayd_id, |
53632229 | 549 | uint64_t session_id, |
ffe60014 DG |
550 | int cpu, |
551 | int *alloc_ret, | |
4891ece8 DG |
552 | enum consumer_channel_type type, |
553 | unsigned int monitor) | |
3bd1e081 | 554 | { |
ffe60014 | 555 | int ret; |
3bd1e081 | 556 | struct lttng_consumer_stream *stream; |
3bd1e081 | 557 | |
effcf122 | 558 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 559 | if (stream == NULL) { |
7a57cf92 | 560 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 561 | ret = -ENOMEM; |
7a57cf92 | 562 | goto end; |
3bd1e081 | 563 | } |
7a57cf92 | 564 | |
d56db448 DG |
565 | rcu_read_lock(); |
566 | ||
3bd1e081 | 567 | stream->key = stream_key; |
3bd1e081 MD |
568 | stream->out_fd = -1; |
569 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 570 | stream->output_written = 0; |
3bd1e081 | 571 | stream->state = state; |
6df2e2c9 MD |
572 | stream->uid = uid; |
573 | stream->gid = gid; | |
ffe60014 | 574 | stream->net_seq_idx = relayd_id; |
53632229 | 575 | stream->session_id = session_id; |
4891ece8 | 576 | stream->monitor = monitor; |
774d490c | 577 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
f8f3885c | 578 | stream->index_file = NULL; |
fb83fe64 | 579 | stream->last_sequence_number = -1ULL; |
53632229 | 580 | pthread_mutex_init(&stream->lock, NULL); |
c585821b | 581 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
58b1f425 | 582 | |
ffe60014 DG |
583 | /* If channel is the metadata, flag this stream as metadata. */ |
584 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
585 | stream->metadata_flag = 1; | |
586 | /* Metadata is flat out. */ | |
587 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
94d49140 JD |
588 | /* Live rendez-vous point. */ |
589 | pthread_cond_init(&stream->metadata_rdv, NULL); | |
590 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); | |
58b1f425 | 591 | } else { |
ffe60014 DG |
592 | /* Format stream name to <channel_name>_<cpu_number> */ |
593 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
594 | channel_name, cpu); | |
595 | if (ret < 0) { | |
596 | PERROR("snprintf stream name"); | |
597 | goto error; | |
598 | } | |
58b1f425 | 599 | } |
c30aaa51 | 600 | |
ffe60014 | 601 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 602 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 603 | |
d8ef542d MD |
604 | /* Init node per channel id key */ |
605 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
606 | ||
53632229 | 607 | /* Init session id node with the stream session id */ |
d88aee68 | 608 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 609 | |
07b86b52 JD |
610 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
611 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
612 | stream->name, stream->key, channel_key, | |
613 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
614 | |
615 | rcu_read_unlock(); | |
3bd1e081 | 616 | return stream; |
c80048c6 MD |
617 | |
618 | error: | |
d56db448 | 619 | rcu_read_unlock(); |
c80048c6 | 620 | free(stream); |
7a57cf92 | 621 | end: |
ffe60014 DG |
622 | if (alloc_ret) { |
623 | *alloc_ret = ret; | |
624 | } | |
c80048c6 | 625 | return NULL; |
3bd1e081 MD |
626 | } |
627 | ||
628 | /* | |
629 | * Add a stream to the global list protected by a mutex. | |
630 | */ | |
5ab66908 | 631 | int consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 632 | { |
5ab66908 | 633 | struct lttng_ht *ht = data_ht; |
3bd1e081 MD |
634 | int ret = 0; |
635 | ||
e316aad5 | 636 | assert(stream); |
43c34bc3 | 637 | assert(ht); |
c77fc10a | 638 | |
d88aee68 | 639 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
640 | |
641 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 642 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 643 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 644 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 645 | rcu_read_lock(); |
e316aad5 | 646 | |
43c34bc3 | 647 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 648 | steal_stream_key(stream->key, ht); |
43c34bc3 | 649 | |
d88aee68 | 650 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 651 | |
d8ef542d MD |
652 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
653 | &stream->node_channel_id); | |
654 | ||
ca22feea DG |
655 | /* |
656 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
657 | * the key since the HT does not use it and we allow to add redundant keys | |
658 | * into this table. | |
659 | */ | |
d88aee68 | 660 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 661 | |
e316aad5 | 662 | /* |
ffe60014 DG |
663 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
664 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
665 | * causes the count to become 0 also causes a stream to be added. The |
666 | * channel deletion will thus be triggered by the following removal of this | |
667 | * stream. | |
668 | */ | |
ffe60014 | 669 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
670 | /* Increment refcount before decrementing nb_init_stream_left */ |
671 | cmm_smp_wmb(); | |
ffe60014 | 672 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
673 | } |
674 | ||
675 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
676 | consumer_data.stream_count++; |
677 | consumer_data.need_update = 1; | |
678 | ||
e316aad5 | 679 | rcu_read_unlock(); |
2e818a6a | 680 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 681 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 682 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 683 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 684 | |
3bd1e081 MD |
685 | return ret; |
686 | } | |
687 | ||
5ab66908 MD |
688 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
689 | { | |
690 | consumer_del_stream(stream, data_ht); | |
691 | } | |
692 | ||
00e2e675 | 693 | /* |
3f8e211f DG |
694 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
695 | * be acquired before calling this. | |
00e2e675 | 696 | */ |
d09e1200 | 697 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
698 | { |
699 | int ret = 0; | |
d88aee68 | 700 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
701 | struct lttng_ht_iter iter; |
702 | ||
ffe60014 | 703 | assert(relayd); |
00e2e675 | 704 | |
00e2e675 | 705 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
706 | &relayd->net_seq_idx, &iter); |
707 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 708 | if (node != NULL) { |
00e2e675 DG |
709 | goto end; |
710 | } | |
d88aee68 | 711 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 712 | |
00e2e675 DG |
713 | end: |
714 | return ret; | |
715 | } | |
716 | ||
717 | /* | |
718 | * Allocate and return a consumer relayd socket. | |
719 | */ | |
027a694f | 720 | static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( |
da009f2c | 721 | uint64_t net_seq_idx) |
00e2e675 DG |
722 | { |
723 | struct consumer_relayd_sock_pair *obj = NULL; | |
724 | ||
da009f2c MD |
725 | /* net sequence index of -1 is a failure */ |
726 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
727 | goto error; |
728 | } | |
729 | ||
730 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
731 | if (obj == NULL) { | |
732 | PERROR("zmalloc relayd sock"); | |
733 | goto error; | |
734 | } | |
735 | ||
736 | obj->net_seq_idx = net_seq_idx; | |
737 | obj->refcount = 0; | |
173af62f | 738 | obj->destroy_flag = 0; |
f96e4545 MD |
739 | obj->control_sock.sock.fd = -1; |
740 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 741 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
742 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
743 | ||
744 | error: | |
745 | return obj; | |
746 | } | |
747 | ||
748 | /* | |
749 | * Find a relayd socket pair in the global consumer data. | |
750 | * | |
751 | * Return the object if found else NULL. | |
b0b335c8 MD |
752 | * RCU read-side lock must be held across this call and while using the |
753 | * returned object. | |
00e2e675 | 754 | */ |
d88aee68 | 755 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
756 | { |
757 | struct lttng_ht_iter iter; | |
d88aee68 | 758 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
759 | struct consumer_relayd_sock_pair *relayd = NULL; |
760 | ||
761 | /* Negative keys are lookup failures */ | |
d88aee68 | 762 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
763 | goto error; |
764 | } | |
765 | ||
d88aee68 | 766 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 767 | &iter); |
d88aee68 | 768 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
769 | if (node != NULL) { |
770 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
771 | } | |
772 | ||
00e2e675 DG |
773 | error: |
774 | return relayd; | |
775 | } | |
776 | ||
10a50311 JD |
777 | /* |
778 | * Find a relayd and send the stream | |
779 | * | |
780 | * Returns 0 on success, < 0 on error | |
781 | */ | |
782 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
783 | char *path) | |
784 | { | |
785 | int ret = 0; | |
786 | struct consumer_relayd_sock_pair *relayd; | |
787 | ||
788 | assert(stream); | |
789 | assert(stream->net_seq_idx != -1ULL); | |
790 | assert(path); | |
791 | ||
792 | /* The stream is not metadata. Get relayd reference if exists. */ | |
793 | rcu_read_lock(); | |
794 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
795 | if (relayd != NULL) { | |
796 | /* Add stream on the relayd */ | |
797 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
798 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
799 | path, &stream->relayd_stream_id, | |
800 | stream->chan->tracefile_size, stream->chan->tracefile_count); | |
801 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
802 | if (ret < 0) { | |
4c5691ba JR |
803 | ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
804 | lttng_consumer_cleanup_relayd(relayd); | |
10a50311 JD |
805 | goto end; |
806 | } | |
1c20f0e2 | 807 | |
10a50311 | 808 | uatomic_inc(&relayd->refcount); |
d01178b6 | 809 | stream->sent_to_relayd = 1; |
10a50311 JD |
810 | } else { |
811 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
812 | stream->key, stream->net_seq_idx); | |
813 | ret = -1; | |
814 | goto end; | |
815 | } | |
816 | ||
817 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
818 | stream->name, stream->key, stream->net_seq_idx); | |
819 | ||
820 | end: | |
821 | rcu_read_unlock(); | |
822 | return ret; | |
823 | } | |
824 | ||
a4baae1b JD |
825 | /* |
826 | * Find a relayd and send the streams sent message | |
827 | * | |
828 | * Returns 0 on success, < 0 on error | |
829 | */ | |
830 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
831 | { | |
832 | int ret = 0; | |
833 | struct consumer_relayd_sock_pair *relayd; | |
834 | ||
835 | assert(net_seq_idx != -1ULL); | |
836 | ||
837 | /* The stream is not metadata. Get relayd reference if exists. */ | |
838 | rcu_read_lock(); | |
839 | relayd = consumer_find_relayd(net_seq_idx); | |
840 | if (relayd != NULL) { | |
841 | /* Add stream on the relayd */ | |
842 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
843 | ret = relayd_streams_sent(&relayd->control_sock); | |
844 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
845 | if (ret < 0) { | |
4c5691ba JR |
846 | ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
847 | lttng_consumer_cleanup_relayd(relayd); | |
a4baae1b JD |
848 | goto end; |
849 | } | |
850 | } else { | |
851 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
852 | net_seq_idx); | |
853 | ret = -1; | |
854 | goto end; | |
855 | } | |
856 | ||
857 | ret = 0; | |
858 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
859 | ||
860 | end: | |
861 | rcu_read_unlock(); | |
862 | return ret; | |
863 | } | |
864 | ||
10a50311 JD |
865 | /* |
866 | * Find a relayd and close the stream | |
867 | */ | |
868 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
869 | { | |
870 | struct consumer_relayd_sock_pair *relayd; | |
871 | ||
872 | /* The stream is not metadata. Get relayd reference if exists. */ | |
873 | rcu_read_lock(); | |
874 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
875 | if (relayd) { | |
876 | consumer_stream_relayd_close(stream, relayd); | |
877 | } | |
878 | rcu_read_unlock(); | |
879 | } | |
880 | ||
00e2e675 DG |
881 | /* |
882 | * Handle stream for relayd transmission if the stream applies for network | |
883 | * streaming where the net sequence index is set. | |
884 | * | |
885 | * Return destination file descriptor or negative value on error. | |
886 | */ | |
6197aea7 | 887 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
888 | size_t data_size, unsigned long padding, |
889 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
890 | { |
891 | int outfd = -1, ret; | |
00e2e675 DG |
892 | struct lttcomm_relayd_data_hdr data_hdr; |
893 | ||
894 | /* Safety net */ | |
895 | assert(stream); | |
6197aea7 | 896 | assert(relayd); |
00e2e675 DG |
897 | |
898 | /* Reset data header */ | |
899 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
900 | ||
00e2e675 DG |
901 | if (stream->metadata_flag) { |
902 | /* Caller MUST acquire the relayd control socket lock */ | |
903 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
904 | if (ret < 0) { | |
905 | goto error; | |
906 | } | |
907 | ||
908 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 909 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
910 | } else { |
911 | /* Set header with stream information */ | |
912 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
913 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 914 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
915 | /* |
916 | * Note that net_seq_num below is assigned with the *current* value of | |
917 | * next_net_seq_num and only after that the next_net_seq_num will be | |
918 | * increment. This is why when issuing a command on the relayd using | |
919 | * this next value, 1 should always be substracted in order to compare | |
920 | * the last seen sequence number on the relayd side to the last sent. | |
921 | */ | |
3604f373 | 922 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
923 | /* Other fields are zeroed previously */ |
924 | ||
925 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
926 | sizeof(data_hdr)); | |
927 | if (ret < 0) { | |
928 | goto error; | |
929 | } | |
930 | ||
3604f373 DG |
931 | ++stream->next_net_seq_num; |
932 | ||
00e2e675 | 933 | /* Set to go on data socket */ |
6151a90f | 934 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
935 | } |
936 | ||
937 | error: | |
938 | return outfd; | |
939 | } | |
940 | ||
3bd1e081 | 941 | /* |
ffe60014 DG |
942 | * Allocate and return a new lttng_consumer_channel object using the given key |
943 | * to initialize the hash table node. | |
944 | * | |
945 | * On error, return NULL. | |
3bd1e081 | 946 | */ |
886224ff | 947 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
948 | uint64_t session_id, |
949 | const char *pathname, | |
950 | const char *name, | |
951 | uid_t uid, | |
952 | gid_t gid, | |
57a269f2 | 953 | uint64_t relayd_id, |
1624d5b7 JD |
954 | enum lttng_event_output output, |
955 | uint64_t tracefile_size, | |
2bba9e53 | 956 | uint64_t tracefile_count, |
1950109e | 957 | uint64_t session_id_per_pid, |
ecc48a90 | 958 | unsigned int monitor, |
d7ba1388 | 959 | unsigned int live_timer_interval, |
3d071855 | 960 | const char *root_shm_path, |
d7ba1388 | 961 | const char *shm_path) |
3bd1e081 MD |
962 | { |
963 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 964 | |
276b26d1 | 965 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 966 | if (channel == NULL) { |
7a57cf92 | 967 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
968 | goto end; |
969 | } | |
ffe60014 DG |
970 | |
971 | channel->key = key; | |
3bd1e081 | 972 | channel->refcount = 0; |
ffe60014 | 973 | channel->session_id = session_id; |
1950109e | 974 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
975 | channel->uid = uid; |
976 | channel->gid = gid; | |
977 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
978 | channel->tracefile_size = tracefile_size; |
979 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 980 | channel->monitor = monitor; |
ecc48a90 | 981 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 982 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 983 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 984 | |
0c759fc9 DG |
985 | switch (output) { |
986 | case LTTNG_EVENT_SPLICE: | |
987 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
988 | break; | |
989 | case LTTNG_EVENT_MMAP: | |
990 | channel->output = CONSUMER_CHANNEL_MMAP; | |
991 | break; | |
992 | default: | |
993 | assert(0); | |
994 | free(channel); | |
995 | channel = NULL; | |
996 | goto end; | |
997 | } | |
998 | ||
07b86b52 JD |
999 | /* |
1000 | * In monitor mode, the streams associated with the channel will be put in | |
1001 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
1002 | * here meaning that the channel itself has streams that are referenced. | |
1003 | * | |
1004 | * On a channel deletion, once the channel is no longer visible, the | |
1005 | * refcount is decremented and checked for a zero value to delete it. With | |
1006 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
1007 | */ | |
1008 | if (!channel->monitor) { | |
1009 | channel->refcount = 1; | |
1010 | } | |
1011 | ||
ffe60014 DG |
1012 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1013 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1014 | ||
1015 | strncpy(channel->name, name, sizeof(channel->name)); | |
1016 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1017 | ||
3d071855 MD |
1018 | if (root_shm_path) { |
1019 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1020 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1021 | } | |
d7ba1388 MD |
1022 | if (shm_path) { |
1023 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1024 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1025 | } | |
1026 | ||
d88aee68 | 1027 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
1028 | |
1029 | channel->wait_fd = -1; | |
1030 | ||
ffe60014 DG |
1031 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1032 | ||
62a7b8ed | 1033 | DBG("Allocated channel (key %" PRIu64 ")", channel->key); |
3bd1e081 | 1034 | |
3bd1e081 MD |
1035 | end: |
1036 | return channel; | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1041 | * |
b5a6470f | 1042 | * Always return 0 indicating success. |
3bd1e081 | 1043 | */ |
d8ef542d MD |
1044 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1045 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1046 | { |
3bd1e081 | 1047 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1048 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1049 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1050 | |
b5a6470f DG |
1051 | /* |
1052 | * This gives us a guarantee that the channel we are about to add to the | |
1053 | * channel hash table will be unique. See this function comment on the why | |
1054 | * we need to steel the channel key at this stage. | |
1055 | */ | |
1056 | steal_channel_key(channel->key); | |
c77fc10a | 1057 | |
b5a6470f | 1058 | rcu_read_lock(); |
d88aee68 | 1059 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
6065ceec | 1060 | rcu_read_unlock(); |
b5a6470f | 1061 | |
ec6ea7d0 | 1062 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1063 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1064 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1065 | |
b5a6470f | 1066 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1067 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1068 | } |
b5a6470f DG |
1069 | |
1070 | return 0; | |
3bd1e081 MD |
1071 | } |
1072 | ||
1073 | /* | |
1074 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1075 | * doing a lookup in the linked list and concurrency issues when writing is | |
1076 | * needed. Called with consumer_data.lock held. | |
1077 | * | |
1078 | * Returns the number of fds in the structures. | |
1079 | */ | |
ffe60014 DG |
1080 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1081 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
2afa177e | 1082 | struct lttng_ht *ht, int *nb_inactive_fd) |
3bd1e081 | 1083 | { |
3bd1e081 | 1084 | int i = 0; |
e4421fec DG |
1085 | struct lttng_ht_iter iter; |
1086 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1087 | |
ffe60014 DG |
1088 | assert(ctx); |
1089 | assert(ht); | |
1090 | assert(pollfd); | |
1091 | assert(local_stream); | |
1092 | ||
3bd1e081 | 1093 | DBG("Updating poll fd array"); |
2afa177e | 1094 | *nb_inactive_fd = 0; |
481d6c57 | 1095 | rcu_read_lock(); |
43c34bc3 | 1096 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1097 | /* |
1098 | * Only active streams with an active end point can be added to the | |
1099 | * poll set and local stream storage of the thread. | |
1100 | * | |
1101 | * There is a potential race here for endpoint_status to be updated | |
1102 | * just after the check. However, this is OK since the stream(s) will | |
1103 | * be deleted once the thread is notified that the end point state has | |
1104 | * changed where this function will be called back again. | |
2afa177e JD |
1105 | * |
1106 | * We track the number of inactive FDs because they still need to be | |
1107 | * closed by the polling thread after a wakeup on the data_pipe or | |
1108 | * metadata_pipe. | |
8994307f DG |
1109 | */ |
1110 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1111 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
2afa177e | 1112 | (*nb_inactive_fd)++; |
3bd1e081 MD |
1113 | continue; |
1114 | } | |
7972aab2 DG |
1115 | /* |
1116 | * This clobbers way too much the debug output. Uncomment that if you | |
1117 | * need it for debugging purposes. | |
1118 | * | |
1119 | * DBG("Active FD %d", stream->wait_fd); | |
1120 | */ | |
e4421fec | 1121 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1122 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1123 | local_stream[i] = stream; |
3bd1e081 MD |
1124 | i++; |
1125 | } | |
481d6c57 | 1126 | rcu_read_unlock(); |
3bd1e081 MD |
1127 | |
1128 | /* | |
50f8ae69 | 1129 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1130 | * increment i so nb_fd is the number of real FD. |
1131 | */ | |
acdb9057 | 1132 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1133 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1134 | |
1135 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1136 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1137 | return i; |
1138 | } | |
1139 | ||
1140 | /* | |
84382d49 MD |
1141 | * Poll on the should_quit pipe and the command socket return -1 on |
1142 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1143 | */ |
1144 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1145 | { | |
1146 | int num_rdy; | |
1147 | ||
88f2b785 | 1148 | restart: |
3bd1e081 MD |
1149 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1150 | if (num_rdy == -1) { | |
88f2b785 MD |
1151 | /* |
1152 | * Restart interrupted system call. | |
1153 | */ | |
1154 | if (errno == EINTR) { | |
1155 | goto restart; | |
1156 | } | |
7a57cf92 | 1157 | PERROR("Poll error"); |
84382d49 | 1158 | return -1; |
3bd1e081 | 1159 | } |
509bb1cf | 1160 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1161 | DBG("consumer_should_quit wake up"); |
84382d49 | 1162 | return 1; |
3bd1e081 MD |
1163 | } |
1164 | return 0; | |
3bd1e081 MD |
1165 | } |
1166 | ||
1167 | /* | |
1168 | * Set the error socket. | |
1169 | */ | |
ffe60014 DG |
1170 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1171 | int sock) | |
3bd1e081 MD |
1172 | { |
1173 | ctx->consumer_error_socket = sock; | |
1174 | } | |
1175 | ||
1176 | /* | |
1177 | * Set the command socket path. | |
1178 | */ | |
3bd1e081 MD |
1179 | void lttng_consumer_set_command_sock_path( |
1180 | struct lttng_consumer_local_data *ctx, char *sock) | |
1181 | { | |
1182 | ctx->consumer_command_sock_path = sock; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * Send return code to the session daemon. | |
1187 | * If the socket is not defined, we return 0, it is not a fatal error | |
1188 | */ | |
ffe60014 | 1189 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1190 | { |
1191 | if (ctx->consumer_error_socket > 0) { | |
1192 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1193 | sizeof(enum lttcomm_sessiond_command)); | |
1194 | } | |
1195 | ||
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | /* | |
228b5bf7 DG |
1200 | * Close all the tracefiles and stream fds and MUST be called when all |
1201 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1202 | */ |
1203 | void lttng_consumer_cleanup(void) | |
1204 | { | |
e4421fec | 1205 | struct lttng_ht_iter iter; |
ffe60014 | 1206 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1207 | |
1208 | rcu_read_lock(); | |
3bd1e081 | 1209 | |
ffe60014 DG |
1210 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1211 | node.node) { | |
702b1ea4 | 1212 | consumer_del_channel(channel); |
3bd1e081 | 1213 | } |
6065ceec DG |
1214 | |
1215 | rcu_read_unlock(); | |
d6ce1df2 | 1216 | |
d6ce1df2 | 1217 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1218 | |
1219 | cleanup_relayd_ht(); | |
1220 | ||
d8ef542d MD |
1221 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1222 | ||
228b5bf7 DG |
1223 | /* |
1224 | * This HT contains streams that are freed by either the metadata thread or | |
1225 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1226 | * it. | |
1227 | */ | |
1228 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1229 | } |
1230 | ||
1231 | /* | |
1232 | * Called from signal handler. | |
1233 | */ | |
1234 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1235 | { | |
6cd525e8 MD |
1236 | ssize_t ret; |
1237 | ||
f7389fbb | 1238 | CMM_STORE_SHARED(consumer_quit, 1); |
6cd525e8 MD |
1239 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1240 | if (ret < 1) { | |
7a57cf92 | 1241 | PERROR("write consumer quit"); |
3bd1e081 | 1242 | } |
ab1027f4 DG |
1243 | |
1244 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1245 | } |
1246 | ||
5199ffc4 JG |
1247 | |
1248 | /* | |
1249 | * Flush pending writes to trace output disk file. | |
1250 | */ | |
1251 | static | |
00e2e675 DG |
1252 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1253 | off_t orig_offset) | |
3bd1e081 | 1254 | { |
c7a78aab | 1255 | int ret; |
3bd1e081 MD |
1256 | int outfd = stream->out_fd; |
1257 | ||
1258 | /* | |
1259 | * This does a blocking write-and-wait on any page that belongs to the | |
1260 | * subbuffer prior to the one we just wrote. | |
1261 | * Don't care about error values, as these are just hints and ways to | |
1262 | * limit the amount of page cache used. | |
1263 | */ | |
ffe60014 | 1264 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1265 | return; |
1266 | } | |
ffe60014 DG |
1267 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1268 | stream->max_sb_size, | |
3bd1e081 MD |
1269 | SYNC_FILE_RANGE_WAIT_BEFORE |
1270 | | SYNC_FILE_RANGE_WRITE | |
1271 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1272 | /* | |
1273 | * Give hints to the kernel about how we access the file: | |
1274 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1275 | * we write it. | |
1276 | * | |
1277 | * We need to call fadvise again after the file grows because the | |
1278 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1279 | * file. | |
1280 | * | |
1281 | * Call fadvise _after_ having waited for the page writeback to | |
1282 | * complete because the dirty page writeback semantic is not well | |
1283 | * defined. So it can be expected to lead to lower throughput in | |
1284 | * streaming. | |
1285 | */ | |
c7a78aab | 1286 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1287 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
a0d0e127 | 1288 | if (ret && ret != -ENOSYS) { |
a74a5f4a JG |
1289 | errno = ret; |
1290 | PERROR("posix_fadvise on fd %i", outfd); | |
c7a78aab | 1291 | } |
3bd1e081 MD |
1292 | } |
1293 | ||
1294 | /* | |
1295 | * Initialise the necessary environnement : | |
1296 | * - create a new context | |
1297 | * - create the poll_pipe | |
1298 | * - create the should_quit pipe (for signal handler) | |
1299 | * - create the thread pipe (for splice) | |
1300 | * | |
1301 | * Takes a function pointer as argument, this function is called when data is | |
1302 | * available on a buffer. This function is responsible to do the | |
1303 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1304 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1305 | * | |
1306 | * Returns a pointer to the new context or NULL on error. | |
1307 | */ | |
1308 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1309 | enum lttng_consumer_type type, | |
4078b776 | 1310 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1311 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1312 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1313 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1314 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1315 | { |
d8ef542d | 1316 | int ret; |
3bd1e081 MD |
1317 | struct lttng_consumer_local_data *ctx; |
1318 | ||
1319 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1320 | consumer_data.type == type); | |
1321 | consumer_data.type = type; | |
1322 | ||
effcf122 | 1323 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1324 | if (ctx == NULL) { |
7a57cf92 | 1325 | PERROR("allocating context"); |
3bd1e081 MD |
1326 | goto error; |
1327 | } | |
1328 | ||
1329 | ctx->consumer_error_socket = -1; | |
331744e3 | 1330 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1331 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1332 | /* assign the callbacks */ |
1333 | ctx->on_buffer_ready = buffer_ready; | |
1334 | ctx->on_recv_channel = recv_channel; | |
1335 | ctx->on_recv_stream = recv_stream; | |
1336 | ctx->on_update_stream = update_stream; | |
1337 | ||
acdb9057 DG |
1338 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1339 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1340 | goto error_poll_pipe; |
1341 | } | |
1342 | ||
02b3d176 DG |
1343 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1344 | if (!ctx->consumer_wakeup_pipe) { | |
1345 | goto error_wakeup_pipe; | |
1346 | } | |
1347 | ||
3bd1e081 MD |
1348 | ret = pipe(ctx->consumer_should_quit); |
1349 | if (ret < 0) { | |
7a57cf92 | 1350 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1351 | goto error_quit_pipe; |
1352 | } | |
1353 | ||
d8ef542d MD |
1354 | ret = pipe(ctx->consumer_channel_pipe); |
1355 | if (ret < 0) { | |
1356 | PERROR("Error creating channel pipe"); | |
1357 | goto error_channel_pipe; | |
1358 | } | |
1359 | ||
13886d2d DG |
1360 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1361 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1362 | goto error_metadata_pipe; |
1363 | } | |
3bd1e081 | 1364 | |
e9404c27 JG |
1365 | ctx->channel_monitor_pipe = -1; |
1366 | ||
fb3a43a9 | 1367 | return ctx; |
3bd1e081 | 1368 | |
fb3a43a9 | 1369 | error_metadata_pipe: |
d8ef542d MD |
1370 | utils_close_pipe(ctx->consumer_channel_pipe); |
1371 | error_channel_pipe: | |
d8ef542d | 1372 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1373 | error_quit_pipe: |
02b3d176 DG |
1374 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1375 | error_wakeup_pipe: | |
acdb9057 | 1376 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1377 | error_poll_pipe: |
1378 | free(ctx); | |
1379 | error: | |
1380 | return NULL; | |
1381 | } | |
1382 | ||
282dadbc MD |
1383 | /* |
1384 | * Iterate over all streams of the hashtable and free them properly. | |
1385 | */ | |
1386 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1387 | { | |
1388 | struct lttng_ht_iter iter; | |
1389 | struct lttng_consumer_stream *stream; | |
1390 | ||
1391 | if (ht == NULL) { | |
1392 | return; | |
1393 | } | |
1394 | ||
1395 | rcu_read_lock(); | |
1396 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1397 | /* | |
1398 | * Ignore return value since we are currently cleaning up so any error | |
1399 | * can't be handled. | |
1400 | */ | |
1401 | (void) consumer_del_stream(stream, ht); | |
1402 | } | |
1403 | rcu_read_unlock(); | |
1404 | ||
1405 | lttng_ht_destroy(ht); | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * Iterate over all streams of the metadata hashtable and free them | |
1410 | * properly. | |
1411 | */ | |
1412 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1413 | { | |
1414 | struct lttng_ht_iter iter; | |
1415 | struct lttng_consumer_stream *stream; | |
1416 | ||
1417 | if (ht == NULL) { | |
1418 | return; | |
1419 | } | |
1420 | ||
1421 | rcu_read_lock(); | |
1422 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1423 | /* | |
1424 | * Ignore return value since we are currently cleaning up so any error | |
1425 | * can't be handled. | |
1426 | */ | |
1427 | (void) consumer_del_metadata_stream(stream, ht); | |
1428 | } | |
1429 | rcu_read_unlock(); | |
1430 | ||
1431 | lttng_ht_destroy(ht); | |
1432 | } | |
1433 | ||
3bd1e081 MD |
1434 | /* |
1435 | * Close all fds associated with the instance and free the context. | |
1436 | */ | |
1437 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1438 | { | |
4c462e79 MD |
1439 | int ret; |
1440 | ||
ab1027f4 DG |
1441 | DBG("Consumer destroying it. Closing everything."); |
1442 | ||
4f2e75b9 DG |
1443 | if (!ctx) { |
1444 | return; | |
1445 | } | |
1446 | ||
282dadbc MD |
1447 | destroy_data_stream_ht(data_ht); |
1448 | destroy_metadata_stream_ht(metadata_ht); | |
1449 | ||
4c462e79 MD |
1450 | ret = close(ctx->consumer_error_socket); |
1451 | if (ret) { | |
1452 | PERROR("close"); | |
1453 | } | |
331744e3 JD |
1454 | ret = close(ctx->consumer_metadata_socket); |
1455 | if (ret) { | |
1456 | PERROR("close"); | |
1457 | } | |
d8ef542d | 1458 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1459 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1460 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1461 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1462 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1463 | |
3bd1e081 MD |
1464 | unlink(ctx->consumer_command_sock_path); |
1465 | free(ctx); | |
1466 | } | |
1467 | ||
6197aea7 DG |
1468 | /* |
1469 | * Write the metadata stream id on the specified file descriptor. | |
1470 | */ | |
1471 | static int write_relayd_metadata_id(int fd, | |
1472 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1473 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 | 1474 | { |
6cd525e8 | 1475 | ssize_t ret; |
1d4dfdef | 1476 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1477 | |
1d4dfdef DG |
1478 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1479 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1480 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1481 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1482 | /* |
6f04ed72 | 1483 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1484 | * not to clubber the error output since this can happen in a normal |
1485 | * code path. | |
1486 | */ | |
1487 | if (errno != EPIPE) { | |
1488 | PERROR("write metadata stream id"); | |
1489 | } | |
1490 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1491 | /* |
1492 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1493 | * handle writting the missing part so report that as an error and | |
1494 | * don't lie to the caller. | |
1495 | */ | |
1496 | ret = -1; | |
6197aea7 DG |
1497 | goto end; |
1498 | } | |
1d4dfdef DG |
1499 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1500 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1501 | |
1502 | end: | |
6cd525e8 | 1503 | return (int) ret; |
6197aea7 DG |
1504 | } |
1505 | ||
3bd1e081 | 1506 | /* |
09e26845 DG |
1507 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1508 | * core function for writing trace buffers to either the local filesystem or | |
1509 | * the network. | |
1510 | * | |
79d4ffb7 DG |
1511 | * It must be called with the stream lock held. |
1512 | * | |
09e26845 | 1513 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1514 | * |
1515 | * Returns the number of bytes written | |
1516 | */ | |
4078b776 | 1517 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1518 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1519 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1520 | unsigned long padding, |
50adc264 | 1521 | struct ctf_packet_index *index) |
3bd1e081 | 1522 | { |
f02e1e8a | 1523 | unsigned long mmap_offset; |
ffe60014 | 1524 | void *mmap_base; |
994ab360 | 1525 | ssize_t ret = 0; |
f02e1e8a DG |
1526 | off_t orig_offset = stream->out_fd_offset; |
1527 | /* Default is on the disk */ | |
1528 | int outfd = stream->out_fd; | |
f02e1e8a | 1529 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1530 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1531 | |
1532 | /* RCU lock for the relayd pointer */ | |
1533 | rcu_read_lock(); | |
1534 | ||
1535 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1536 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1537 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1538 | if (relayd == NULL) { | |
56591bac | 1539 | ret = -EPIPE; |
f02e1e8a DG |
1540 | goto end; |
1541 | } | |
1542 | } | |
1543 | ||
1544 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1545 | switch (consumer_data.type) { |
1546 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1547 | mmap_base = stream->mmap_base; |
f02e1e8a | 1548 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
994ab360 | 1549 | if (ret < 0) { |
56591bac | 1550 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1551 | goto end; |
1552 | } | |
f02e1e8a | 1553 | break; |
7753dea8 MD |
1554 | case LTTNG_CONSUMER32_UST: |
1555 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1556 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1557 | if (!mmap_base) { | |
1558 | ERR("read mmap get mmap base for stream %s", stream->name); | |
994ab360 | 1559 | ret = -EPERM; |
ffe60014 DG |
1560 | goto end; |
1561 | } | |
1562 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1563 | if (ret != 0) { |
1564 | PERROR("tracer ctl get_mmap_read_offset"); | |
994ab360 | 1565 | ret = -EINVAL; |
56591bac MD |
1566 | goto end; |
1567 | } | |
f02e1e8a | 1568 | break; |
3bd1e081 MD |
1569 | default: |
1570 | ERR("Unknown consumer_data type"); | |
1571 | assert(0); | |
1572 | } | |
b9182dd9 | 1573 | |
f02e1e8a DG |
1574 | /* Handle stream on the relayd if the output is on the network */ |
1575 | if (relayd) { | |
1576 | unsigned long netlen = len; | |
1577 | ||
1578 | /* | |
1579 | * Lock the control socket for the complete duration of the function | |
1580 | * since from this point on we will use the socket. | |
1581 | */ | |
1582 | if (stream->metadata_flag) { | |
1583 | /* Metadata requires the control socket. */ | |
1584 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
93ec662e JD |
1585 | if (stream->reset_metadata_flag) { |
1586 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1587 | stream->relayd_stream_id, | |
1588 | stream->metadata_version); | |
1589 | if (ret < 0) { | |
1590 | relayd_hang_up = 1; | |
1591 | goto write_error; | |
1592 | } | |
1593 | stream->reset_metadata_flag = 0; | |
1594 | } | |
1d4dfdef | 1595 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1596 | } |
1597 | ||
1d4dfdef | 1598 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1599 | if (ret < 0) { |
1600 | relayd_hang_up = 1; | |
1601 | goto write_error; | |
1602 | } | |
1603 | /* Use the returned socket. */ | |
1604 | outfd = ret; | |
f02e1e8a | 1605 | |
994ab360 DG |
1606 | /* Write metadata stream id before payload */ |
1607 | if (stream->metadata_flag) { | |
1608 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); | |
1609 | if (ret < 0) { | |
8994307f DG |
1610 | relayd_hang_up = 1; |
1611 | goto write_error; | |
1612 | } | |
f02e1e8a | 1613 | } |
1d4dfdef DG |
1614 | } else { |
1615 | /* No streaming, we have to set the len with the full padding */ | |
1616 | len += padding; | |
1624d5b7 | 1617 | |
93ec662e JD |
1618 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1619 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1620 | if (ret < 0) { | |
1621 | ERR("Reset metadata file"); | |
1622 | goto end; | |
1623 | } | |
1624 | stream->reset_metadata_flag = 0; | |
1625 | } | |
1626 | ||
1624d5b7 JD |
1627 | /* |
1628 | * Check if we need to change the tracefile before writing the packet. | |
1629 | */ | |
1630 | if (stream->chan->tracefile_size > 0 && | |
1631 | (stream->tracefile_size_current + len) > | |
1632 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1633 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1634 | stream->name, stream->chan->tracefile_size, | |
1635 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1636 | stream->out_fd, &(stream->tracefile_count_current), |
1637 | &stream->out_fd); | |
1624d5b7 JD |
1638 | if (ret < 0) { |
1639 | ERR("Rotating output file"); | |
1640 | goto end; | |
1641 | } | |
309167d2 JD |
1642 | outfd = stream->out_fd; |
1643 | ||
f8f3885c MD |
1644 | if (stream->index_file) { |
1645 | lttng_index_file_put(stream->index_file); | |
1646 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1647 | stream->name, stream->uid, stream->gid, |
1648 | stream->chan->tracefile_size, | |
f8f3885c MD |
1649 | stream->tracefile_count_current, |
1650 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1651 | if (!stream->index_file) { | |
309167d2 JD |
1652 | goto end; |
1653 | } | |
309167d2 JD |
1654 | } |
1655 | ||
a6976990 DG |
1656 | /* Reset current size because we just perform a rotation. */ |
1657 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1658 | stream->out_fd_offset = 0; |
1659 | orig_offset = 0; | |
1624d5b7 JD |
1660 | } |
1661 | stream->tracefile_size_current += len; | |
309167d2 JD |
1662 | if (index) { |
1663 | index->offset = htobe64(stream->out_fd_offset); | |
1664 | } | |
f02e1e8a DG |
1665 | } |
1666 | ||
d02b8372 DG |
1667 | /* |
1668 | * This call guarantee that len or less is returned. It's impossible to | |
1669 | * receive a ret value that is bigger than len. | |
1670 | */ | |
1671 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1672 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1673 | if (ret < 0 || ((size_t) ret != len)) { | |
1674 | /* | |
1675 | * Report error to caller if nothing was written else at least send the | |
1676 | * amount written. | |
1677 | */ | |
1678 | if (ret < 0) { | |
994ab360 | 1679 | ret = -errno; |
f02e1e8a | 1680 | } |
994ab360 | 1681 | relayd_hang_up = 1; |
f02e1e8a | 1682 | |
d02b8372 | 1683 | /* Socket operation failed. We consider the relayd dead */ |
994ab360 | 1684 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
d02b8372 DG |
1685 | /* |
1686 | * This is possible if the fd is closed on the other side | |
1687 | * (outfd) or any write problem. It can be verbose a bit for a | |
1688 | * normal execution if for instance the relayd is stopped | |
1689 | * abruptly. This can happen so set this to a DBG statement. | |
1690 | */ | |
1691 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1692 | } else { |
1693 | /* Unhandled error, print it and stop function right now. */ | |
1694 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1695 | } |
994ab360 | 1696 | goto write_error; |
d02b8372 DG |
1697 | } |
1698 | stream->output_written += ret; | |
d02b8372 DG |
1699 | |
1700 | /* This call is useless on a socket so better save a syscall. */ | |
1701 | if (!relayd) { | |
1702 | /* This won't block, but will start writeout asynchronously */ | |
1703 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1704 | SYNC_FILE_RANGE_WRITE); | |
1705 | stream->out_fd_offset += len; | |
f5dbe415 | 1706 | lttng_consumer_sync_trace_file(stream, orig_offset); |
f02e1e8a | 1707 | } |
f02e1e8a | 1708 | |
8994307f DG |
1709 | write_error: |
1710 | /* | |
1711 | * This is a special case that the relayd has closed its socket. Let's | |
1712 | * cleanup the relayd object and all associated streams. | |
1713 | */ | |
1714 | if (relayd && relayd_hang_up) { | |
4c5691ba JR |
1715 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1716 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1717 | } |
1718 | ||
f02e1e8a DG |
1719 | end: |
1720 | /* Unlock only if ctrl socket used */ | |
1721 | if (relayd && stream->metadata_flag) { | |
1722 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1723 | } | |
1724 | ||
1725 | rcu_read_unlock(); | |
994ab360 | 1726 | return ret; |
3bd1e081 MD |
1727 | } |
1728 | ||
1729 | /* | |
1730 | * Splice the data from the ring buffer to the tracefile. | |
1731 | * | |
79d4ffb7 DG |
1732 | * It must be called with the stream lock held. |
1733 | * | |
3bd1e081 MD |
1734 | * Returns the number of bytes spliced. |
1735 | */ | |
4078b776 | 1736 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1737 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1738 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1739 | unsigned long padding, |
50adc264 | 1740 | struct ctf_packet_index *index) |
3bd1e081 | 1741 | { |
f02e1e8a DG |
1742 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1743 | loff_t offset = 0; | |
1744 | off_t orig_offset = stream->out_fd_offset; | |
1745 | int fd = stream->wait_fd; | |
1746 | /* Default is on the disk */ | |
1747 | int outfd = stream->out_fd; | |
f02e1e8a | 1748 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1749 | int *splice_pipe; |
8994307f | 1750 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1751 | |
3bd1e081 MD |
1752 | switch (consumer_data.type) { |
1753 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1754 | break; |
7753dea8 MD |
1755 | case LTTNG_CONSUMER32_UST: |
1756 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1757 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1758 | return -ENOSYS; |
1759 | default: | |
1760 | ERR("Unknown consumer_data type"); | |
1761 | assert(0); | |
3bd1e081 MD |
1762 | } |
1763 | ||
f02e1e8a DG |
1764 | /* RCU lock for the relayd pointer */ |
1765 | rcu_read_lock(); | |
1766 | ||
1767 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1768 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1769 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1770 | if (relayd == NULL) { | |
ad0b0d23 | 1771 | written = -ret; |
f02e1e8a DG |
1772 | goto end; |
1773 | } | |
1774 | } | |
a2361a61 | 1775 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1776 | |
f02e1e8a | 1777 | /* Write metadata stream id before payload */ |
1d4dfdef | 1778 | if (relayd) { |
ad0b0d23 | 1779 | unsigned long total_len = len; |
f02e1e8a | 1780 | |
1d4dfdef DG |
1781 | if (stream->metadata_flag) { |
1782 | /* | |
1783 | * Lock the control socket for the complete duration of the function | |
1784 | * since from this point on we will use the socket. | |
1785 | */ | |
1786 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1787 | ||
93ec662e JD |
1788 | if (stream->reset_metadata_flag) { |
1789 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1790 | stream->relayd_stream_id, | |
1791 | stream->metadata_version); | |
1792 | if (ret < 0) { | |
1793 | relayd_hang_up = 1; | |
1794 | goto write_error; | |
1795 | } | |
1796 | stream->reset_metadata_flag = 0; | |
1797 | } | |
1d4dfdef DG |
1798 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, |
1799 | padding); | |
1800 | if (ret < 0) { | |
1801 | written = ret; | |
ad0b0d23 DG |
1802 | relayd_hang_up = 1; |
1803 | goto write_error; | |
1d4dfdef DG |
1804 | } |
1805 | ||
1806 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1807 | } | |
1808 | ||
1809 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1810 | if (ret < 0) { |
1811 | written = ret; | |
1812 | relayd_hang_up = 1; | |
1813 | goto write_error; | |
f02e1e8a | 1814 | } |
ad0b0d23 DG |
1815 | /* Use the returned socket. */ |
1816 | outfd = ret; | |
1d4dfdef DG |
1817 | } else { |
1818 | /* No streaming, we have to set the len with the full padding */ | |
1819 | len += padding; | |
1624d5b7 | 1820 | |
93ec662e JD |
1821 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1822 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1823 | if (ret < 0) { | |
1824 | ERR("Reset metadata file"); | |
1825 | goto end; | |
1826 | } | |
1827 | stream->reset_metadata_flag = 0; | |
1828 | } | |
1624d5b7 JD |
1829 | /* |
1830 | * Check if we need to change the tracefile before writing the packet. | |
1831 | */ | |
1832 | if (stream->chan->tracefile_size > 0 && | |
1833 | (stream->tracefile_size_current + len) > | |
1834 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1835 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1836 | stream->name, stream->chan->tracefile_size, | |
1837 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1838 | stream->out_fd, &(stream->tracefile_count_current), |
1839 | &stream->out_fd); | |
1624d5b7 | 1840 | if (ret < 0) { |
ad0b0d23 | 1841 | written = ret; |
1624d5b7 JD |
1842 | ERR("Rotating output file"); |
1843 | goto end; | |
1844 | } | |
309167d2 JD |
1845 | outfd = stream->out_fd; |
1846 | ||
f8f3885c MD |
1847 | if (stream->index_file) { |
1848 | lttng_index_file_put(stream->index_file); | |
1849 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1850 | stream->name, stream->uid, stream->gid, |
1851 | stream->chan->tracefile_size, | |
f8f3885c MD |
1852 | stream->tracefile_count_current, |
1853 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1854 | if (!stream->index_file) { | |
309167d2 JD |
1855 | goto end; |
1856 | } | |
309167d2 JD |
1857 | } |
1858 | ||
a6976990 DG |
1859 | /* Reset current size because we just perform a rotation. */ |
1860 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1861 | stream->out_fd_offset = 0; |
1862 | orig_offset = 0; | |
1624d5b7 JD |
1863 | } |
1864 | stream->tracefile_size_current += len; | |
309167d2 | 1865 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1866 | } |
1867 | ||
1868 | while (len > 0) { | |
1d4dfdef DG |
1869 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1870 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1871 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1872 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1873 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1874 | if (ret_splice < 0) { | |
d02b8372 | 1875 | ret = errno; |
ad0b0d23 | 1876 | written = -ret; |
d02b8372 | 1877 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1878 | goto splice_error; |
1879 | } | |
1880 | ||
1881 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1882 | if (relayd && stream->metadata_flag) { |
1883 | size_t metadata_payload_size = | |
1884 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1885 | ||
1886 | /* Update counter to fit the spliced data */ | |
1887 | ret_splice += metadata_payload_size; | |
1888 | len += metadata_payload_size; | |
1889 | /* | |
1890 | * We do this so the return value can match the len passed as | |
1891 | * argument to this function. | |
1892 | */ | |
1893 | written -= metadata_payload_size; | |
f02e1e8a DG |
1894 | } |
1895 | ||
1896 | /* Splice data out */ | |
fb3a43a9 | 1897 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1898 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1899 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1900 | outfd, ret_splice); | |
f02e1e8a | 1901 | if (ret_splice < 0) { |
d02b8372 | 1902 | ret = errno; |
ad0b0d23 DG |
1903 | written = -ret; |
1904 | relayd_hang_up = 1; | |
1905 | goto write_error; | |
f02e1e8a | 1906 | } else if (ret_splice > len) { |
d02b8372 DG |
1907 | /* |
1908 | * We don't expect this code path to be executed but you never know | |
1909 | * so this is an extra protection agains a buggy splice(). | |
1910 | */ | |
f02e1e8a | 1911 | ret = errno; |
ad0b0d23 | 1912 | written += ret_splice; |
d02b8372 DG |
1913 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1914 | len); | |
f02e1e8a | 1915 | goto splice_error; |
d02b8372 DG |
1916 | } else { |
1917 | /* All good, update current len and continue. */ | |
1918 | len -= ret_splice; | |
f02e1e8a | 1919 | } |
f02e1e8a DG |
1920 | |
1921 | /* This call is useless on a socket so better save a syscall. */ | |
1922 | if (!relayd) { | |
1923 | /* This won't block, but will start writeout asynchronously */ | |
1924 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1925 | SYNC_FILE_RANGE_WRITE); | |
1926 | stream->out_fd_offset += ret_splice; | |
1927 | } | |
e5d1a9b3 | 1928 | stream->output_written += ret_splice; |
f02e1e8a DG |
1929 | written += ret_splice; |
1930 | } | |
f5dbe415 JG |
1931 | if (!relayd) { |
1932 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1933 | } | |
f02e1e8a DG |
1934 | goto end; |
1935 | ||
8994307f DG |
1936 | write_error: |
1937 | /* | |
1938 | * This is a special case that the relayd has closed its socket. Let's | |
1939 | * cleanup the relayd object and all associated streams. | |
1940 | */ | |
1941 | if (relayd && relayd_hang_up) { | |
4c5691ba JR |
1942 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1943 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1944 | /* Skip splice error so the consumer does not fail */ |
1945 | goto end; | |
1946 | } | |
1947 | ||
f02e1e8a DG |
1948 | splice_error: |
1949 | /* send the appropriate error description to sessiond */ | |
1950 | switch (ret) { | |
f02e1e8a | 1951 | case EINVAL: |
f73fabfd | 1952 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1953 | break; |
1954 | case ENOMEM: | |
f73fabfd | 1955 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1956 | break; |
1957 | case ESPIPE: | |
f73fabfd | 1958 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1959 | break; |
1960 | } | |
1961 | ||
1962 | end: | |
1963 | if (relayd && stream->metadata_flag) { | |
1964 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1965 | } | |
1966 | ||
1967 | rcu_read_unlock(); | |
1968 | return written; | |
3bd1e081 MD |
1969 | } |
1970 | ||
1971 | /* | |
1972 | * Take a snapshot for a specific fd | |
1973 | * | |
1974 | * Returns 0 on success, < 0 on error | |
1975 | */ | |
ffe60014 | 1976 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1977 | { |
1978 | switch (consumer_data.type) { | |
1979 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1980 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1981 | case LTTNG_CONSUMER32_UST: |
1982 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1983 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1984 | default: |
1985 | ERR("Unknown consumer_data type"); | |
1986 | assert(0); | |
1987 | return -ENOSYS; | |
1988 | } | |
3bd1e081 MD |
1989 | } |
1990 | ||
1991 | /* | |
1992 | * Get the produced position | |
1993 | * | |
1994 | * Returns 0 on success, < 0 on error | |
1995 | */ | |
ffe60014 | 1996 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1997 | unsigned long *pos) |
1998 | { | |
1999 | switch (consumer_data.type) { | |
2000 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 2001 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
2002 | case LTTNG_CONSUMER32_UST: |
2003 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2004 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
2005 | default: |
2006 | ERR("Unknown consumer_data type"); | |
2007 | assert(0); | |
2008 | return -ENOSYS; | |
2009 | } | |
2010 | } | |
2011 | ||
2012 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
2013 | int sock, struct pollfd *consumer_sockpoll) | |
2014 | { | |
2015 | switch (consumer_data.type) { | |
2016 | case LTTNG_CONSUMER_KERNEL: | |
2017 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
2018 | case LTTNG_CONSUMER32_UST: |
2019 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
2020 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
2021 | default: | |
2022 | ERR("Unknown consumer_data type"); | |
2023 | assert(0); | |
2024 | return -ENOSYS; | |
2025 | } | |
2026 | } | |
2027 | ||
6d574024 | 2028 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
2029 | { |
2030 | switch (consumer_data.type) { | |
2031 | case LTTNG_CONSUMER_KERNEL: | |
2032 | /* | |
2033 | * The Kernel consumer has a different metadata scheme so we don't | |
2034 | * close anything because the stream will be closed by the session | |
2035 | * daemon. | |
2036 | */ | |
2037 | break; | |
2038 | case LTTNG_CONSUMER32_UST: | |
2039 | case LTTNG_CONSUMER64_UST: | |
2040 | /* | |
2041 | * Close all metadata streams. The metadata hash table is passed and | |
2042 | * this call iterates over it by closing all wakeup fd. This is safe | |
2043 | * because at this point we are sure that the metadata producer is | |
2044 | * either dead or blocked. | |
2045 | */ | |
6d574024 | 2046 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
2047 | break; |
2048 | default: | |
2049 | ERR("Unknown consumer_data type"); | |
2050 | assert(0); | |
2051 | } | |
2052 | } | |
2053 | ||
fb3a43a9 DG |
2054 | /* |
2055 | * Clean up a metadata stream and free its memory. | |
2056 | */ | |
e316aad5 DG |
2057 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
2058 | struct lttng_ht *ht) | |
fb3a43a9 | 2059 | { |
e316aad5 | 2060 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
2061 | |
2062 | assert(stream); | |
2063 | /* | |
2064 | * This call should NEVER receive regular stream. It must always be | |
2065 | * metadata stream and this is crucial for data structure synchronization. | |
2066 | */ | |
2067 | assert(stream->metadata_flag); | |
2068 | ||
e316aad5 DG |
2069 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2070 | ||
74251bb8 | 2071 | pthread_mutex_lock(&consumer_data.lock); |
fb549e7a | 2072 | pthread_mutex_lock(&stream->chan->lock); |
3dad2c0f | 2073 | pthread_mutex_lock(&stream->lock); |
081424af JG |
2074 | if (stream->chan->metadata_cache) { |
2075 | /* Only applicable to userspace consumers. */ | |
2076 | pthread_mutex_lock(&stream->chan->metadata_cache->lock); | |
2077 | } | |
8994307f | 2078 | |
6d574024 DG |
2079 | /* Remove any reference to that stream. */ |
2080 | consumer_stream_delete(stream, ht); | |
ca22feea | 2081 | |
6d574024 DG |
2082 | /* Close down everything including the relayd if one. */ |
2083 | consumer_stream_close(stream); | |
2084 | /* Destroy tracer buffers of the stream. */ | |
2085 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2086 | |
2087 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2088 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2089 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2090 | /* Go for channel deletion! */ |
e316aad5 | 2091 | free_chan = stream->chan; |
fb3a43a9 DG |
2092 | } |
2093 | ||
73811ecc DG |
2094 | /* |
2095 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2096 | * channel lock MUST be acquired before being able to check for a NULL |
2097 | * pointer value. | |
73811ecc DG |
2098 | */ |
2099 | stream->chan->metadata_stream = NULL; | |
2100 | ||
081424af JG |
2101 | if (stream->chan->metadata_cache) { |
2102 | pthread_mutex_unlock(&stream->chan->metadata_cache->lock); | |
2103 | } | |
3dad2c0f | 2104 | pthread_mutex_unlock(&stream->lock); |
fb549e7a | 2105 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2106 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2107 | |
2108 | if (free_chan) { | |
2109 | consumer_del_channel(free_chan); | |
2110 | } | |
2111 | ||
6d574024 | 2112 | consumer_stream_free(stream); |
fb3a43a9 DG |
2113 | } |
2114 | ||
2115 | /* | |
2116 | * Action done with the metadata stream when adding it to the consumer internal | |
2117 | * data structures to handle it. | |
2118 | */ | |
5ab66908 | 2119 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2120 | { |
5ab66908 | 2121 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2122 | int ret = 0; |
76082088 | 2123 | struct lttng_ht_iter iter; |
d88aee68 | 2124 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2125 | |
e316aad5 DG |
2126 | assert(stream); |
2127 | assert(ht); | |
2128 | ||
d88aee68 | 2129 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2130 | |
2131 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2132 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2133 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2134 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2135 | |
e316aad5 DG |
2136 | /* |
2137 | * From here, refcounts are updated so be _careful_ when returning an error | |
2138 | * after this point. | |
2139 | */ | |
2140 | ||
fb3a43a9 | 2141 | rcu_read_lock(); |
76082088 DG |
2142 | |
2143 | /* | |
2144 | * Lookup the stream just to make sure it does not exist in our internal | |
2145 | * state. This should NEVER happen. | |
2146 | */ | |
d88aee68 DG |
2147 | lttng_ht_lookup(ht, &stream->key, &iter); |
2148 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2149 | assert(!node); |
2150 | ||
e316aad5 | 2151 | /* |
ffe60014 DG |
2152 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2153 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2154 | * causes the count to become 0 also causes a stream to be added. The |
2155 | * channel deletion will thus be triggered by the following removal of this | |
2156 | * stream. | |
2157 | */ | |
ffe60014 | 2158 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2159 | /* Increment refcount before decrementing nb_init_stream_left */ |
2160 | cmm_smp_wmb(); | |
ffe60014 | 2161 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2162 | } |
2163 | ||
d88aee68 | 2164 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2165 | |
08eea918 | 2166 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
d8ef542d MD |
2167 | &stream->node_channel_id); |
2168 | ||
ca22feea DG |
2169 | /* |
2170 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2171 | * the key since the HT does not use it and we allow to add redundant keys | |
2172 | * into this table. | |
2173 | */ | |
d88aee68 | 2174 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2175 | |
fb3a43a9 | 2176 | rcu_read_unlock(); |
e316aad5 | 2177 | |
2e818a6a | 2178 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2179 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2180 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2181 | pthread_mutex_unlock(&consumer_data.lock); |
2182 | return ret; | |
fb3a43a9 DG |
2183 | } |
2184 | ||
8994307f DG |
2185 | /* |
2186 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2187 | */ | |
2188 | static void validate_endpoint_status_data_stream(void) | |
2189 | { | |
2190 | struct lttng_ht_iter iter; | |
2191 | struct lttng_consumer_stream *stream; | |
2192 | ||
2193 | DBG("Consumer delete flagged data stream"); | |
2194 | ||
2195 | rcu_read_lock(); | |
2196 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2197 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2198 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2199 | continue; |
2200 | } | |
2201 | /* Delete it right now */ | |
2202 | consumer_del_stream(stream, data_ht); | |
2203 | } | |
2204 | rcu_read_unlock(); | |
2205 | } | |
2206 | ||
2207 | /* | |
2208 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2209 | */ | |
2210 | static void validate_endpoint_status_metadata_stream( | |
2211 | struct lttng_poll_event *pollset) | |
2212 | { | |
2213 | struct lttng_ht_iter iter; | |
2214 | struct lttng_consumer_stream *stream; | |
2215 | ||
2216 | DBG("Consumer delete flagged metadata stream"); | |
2217 | ||
2218 | assert(pollset); | |
2219 | ||
2220 | rcu_read_lock(); | |
2221 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2222 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2223 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2224 | continue; |
2225 | } | |
2226 | /* | |
2227 | * Remove from pollset so the metadata thread can continue without | |
2228 | * blocking on a deleted stream. | |
2229 | */ | |
2230 | lttng_poll_del(pollset, stream->wait_fd); | |
2231 | ||
2232 | /* Delete it right now */ | |
2233 | consumer_del_metadata_stream(stream, metadata_ht); | |
2234 | } | |
2235 | rcu_read_unlock(); | |
2236 | } | |
2237 | ||
fb3a43a9 DG |
2238 | /* |
2239 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2240 | * network. | |
2241 | */ | |
7d980def | 2242 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2243 | { |
1fc79fb4 | 2244 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2245 | uint32_t revents, nb_fd; |
e316aad5 | 2246 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2247 | struct lttng_ht_iter iter; |
d88aee68 | 2248 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2249 | struct lttng_poll_event events; |
2250 | struct lttng_consumer_local_data *ctx = data; | |
2251 | ssize_t len; | |
2252 | ||
2253 | rcu_register_thread(); | |
2254 | ||
1fc79fb4 MD |
2255 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2256 | ||
2d57de81 MD |
2257 | if (testpoint(consumerd_thread_metadata)) { |
2258 | goto error_testpoint; | |
2259 | } | |
2260 | ||
9ce5646a MD |
2261 | health_code_update(); |
2262 | ||
fb3a43a9 DG |
2263 | DBG("Thread metadata poll started"); |
2264 | ||
fb3a43a9 DG |
2265 | /* Size is set to 1 for the consumer_metadata pipe */ |
2266 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2267 | if (ret < 0) { | |
2268 | ERR("Poll set creation failed"); | |
d8ef542d | 2269 | goto end_poll; |
fb3a43a9 DG |
2270 | } |
2271 | ||
13886d2d DG |
2272 | ret = lttng_poll_add(&events, |
2273 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2274 | if (ret < 0) { |
2275 | goto end; | |
2276 | } | |
2277 | ||
2278 | /* Main loop */ | |
2279 | DBG("Metadata main loop started"); | |
2280 | ||
2281 | while (1) { | |
fb3a43a9 | 2282 | restart: |
7fa2082e | 2283 | health_code_update(); |
9ce5646a | 2284 | health_poll_entry(); |
7fa2082e | 2285 | DBG("Metadata poll wait"); |
fb3a43a9 | 2286 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2287 | DBG("Metadata poll return from wait with %d fd(s)", |
2288 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2289 | health_poll_exit(); |
40063ead | 2290 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2291 | if (ret < 0) { |
2292 | if (errno == EINTR) { | |
40063ead | 2293 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2294 | goto restart; |
2295 | } | |
d9607cd7 MD |
2296 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2297 | err = 0; /* All is OK */ | |
2298 | } | |
2299 | goto end; | |
fb3a43a9 DG |
2300 | } |
2301 | ||
0d9c5d77 DG |
2302 | nb_fd = ret; |
2303 | ||
e316aad5 | 2304 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2305 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2306 | health_code_update(); |
2307 | ||
fb3a43a9 DG |
2308 | revents = LTTNG_POLL_GETEV(&events, i); |
2309 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2310 | ||
fd20dac9 MD |
2311 | if (!revents) { |
2312 | /* No activity for this FD (poll implementation). */ | |
2313 | continue; | |
2314 | } | |
2315 | ||
13886d2d | 2316 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
03e43155 | 2317 | if (revents & LPOLLIN) { |
13886d2d DG |
2318 | ssize_t pipe_len; |
2319 | ||
2320 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2321 | &stream, sizeof(stream)); | |
6cd525e8 | 2322 | if (pipe_len < sizeof(stream)) { |
03e43155 MD |
2323 | if (pipe_len < 0) { |
2324 | PERROR("read metadata stream"); | |
2325 | } | |
fb3a43a9 | 2326 | /* |
03e43155 MD |
2327 | * Remove the pipe from the poll set and continue the loop |
2328 | * since their might be data to consume. | |
fb3a43a9 | 2329 | */ |
03e43155 MD |
2330 | lttng_poll_del(&events, |
2331 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2332 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2333 | continue; |
2334 | } | |
2335 | ||
8994307f DG |
2336 | /* A NULL stream means that the state has changed. */ |
2337 | if (stream == NULL) { | |
2338 | /* Check for deleted streams. */ | |
2339 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2340 | goto restart; |
8994307f DG |
2341 | } |
2342 | ||
fb3a43a9 DG |
2343 | DBG("Adding metadata stream %d to poll set", |
2344 | stream->wait_fd); | |
2345 | ||
fb3a43a9 DG |
2346 | /* Add metadata stream to the global poll events list */ |
2347 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2348 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
03e43155 MD |
2349 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2350 | DBG("Metadata thread pipe hung up"); | |
2351 | /* | |
2352 | * Remove the pipe from the poll set and continue the loop | |
2353 | * since their might be data to consume. | |
2354 | */ | |
2355 | lttng_poll_del(&events, | |
2356 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2357 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2358 | continue; | |
2359 | } else { | |
2360 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2361 | goto end; | |
fb3a43a9 DG |
2362 | } |
2363 | ||
e316aad5 | 2364 | /* Handle other stream */ |
fb3a43a9 DG |
2365 | continue; |
2366 | } | |
2367 | ||
d09e1200 | 2368 | rcu_read_lock(); |
d88aee68 DG |
2369 | { |
2370 | uint64_t tmp_id = (uint64_t) pollfd; | |
2371 | ||
2372 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2373 | } | |
2374 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2375 | assert(node); |
fb3a43a9 DG |
2376 | |
2377 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2378 | node); |
fb3a43a9 | 2379 | |
03e43155 MD |
2380 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2381 | /* Get the data out of the metadata file descriptor */ | |
2382 | DBG("Metadata available on fd %d", pollfd); | |
2383 | assert(stream->wait_fd == pollfd); | |
2384 | ||
2385 | do { | |
2386 | health_code_update(); | |
2387 | ||
2388 | len = ctx->on_buffer_ready(stream, ctx); | |
2389 | /* | |
2390 | * We don't check the return value here since if we get | |
83f4233d | 2391 | * a negative len, it means an error occurred thus we |
03e43155 MD |
2392 | * simply remove it from the poll set and free the |
2393 | * stream. | |
2394 | */ | |
2395 | } while (len > 0); | |
2396 | ||
2397 | /* It's ok to have an unavailable sub-buffer */ | |
2398 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2399 | /* Clean up stream from consumer and free it. */ | |
2400 | lttng_poll_del(&events, stream->wait_fd); | |
2401 | consumer_del_metadata_stream(stream, metadata_ht); | |
2402 | } | |
2403 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2404 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2405 | if (!stream->hangup_flush_done |
2406 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2407 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2408 | DBG("Attempting to flush and consume the UST buffers"); | |
2409 | lttng_ustconsumer_on_stream_hangup(stream); | |
2410 | ||
2411 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2412 | do { |
9ce5646a MD |
2413 | health_code_update(); |
2414 | ||
4bb94b75 DG |
2415 | len = ctx->on_buffer_ready(stream, ctx); |
2416 | /* | |
2417 | * We don't check the return value here since if we get | |
83f4233d | 2418 | * a negative len, it means an error occurred thus we |
4bb94b75 DG |
2419 | * simply remove it from the poll set and free the |
2420 | * stream. | |
2421 | */ | |
2422 | } while (len > 0); | |
fb3a43a9 DG |
2423 | } |
2424 | ||
fb3a43a9 | 2425 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2426 | /* |
2427 | * This call update the channel states, closes file descriptors | |
2428 | * and securely free the stream. | |
2429 | */ | |
2430 | consumer_del_metadata_stream(stream, metadata_ht); | |
03e43155 MD |
2431 | } else { |
2432 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
6f2f1a70 | 2433 | rcu_read_unlock(); |
03e43155 | 2434 | goto end; |
fb3a43a9 | 2435 | } |
e316aad5 | 2436 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2437 | rcu_read_unlock(); |
fb3a43a9 DG |
2438 | } |
2439 | } | |
2440 | ||
1fc79fb4 MD |
2441 | /* All is OK */ |
2442 | err = 0; | |
fb3a43a9 DG |
2443 | end: |
2444 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2445 | |
d8ef542d MD |
2446 | lttng_poll_clean(&events); |
2447 | end_poll: | |
2d57de81 | 2448 | error_testpoint: |
1fc79fb4 MD |
2449 | if (err) { |
2450 | health_error(); | |
2451 | ERR("Health error occurred in %s", __func__); | |
2452 | } | |
2453 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2454 | rcu_unregister_thread(); |
2455 | return NULL; | |
2456 | } | |
2457 | ||
3bd1e081 | 2458 | /* |
e4421fec | 2459 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2460 | * it to tracefile if necessary. |
2461 | */ | |
7d980def | 2462 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2463 | { |
1fc79fb4 | 2464 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2465 | struct pollfd *pollfd = NULL; |
2466 | /* local view of the streams */ | |
c869f647 | 2467 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2468 | /* local view of consumer_data.fds_count */ |
2469 | int nb_fd = 0; | |
2afa177e JD |
2470 | /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */ |
2471 | int nb_inactive_fd = 0; | |
3bd1e081 | 2472 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2473 | ssize_t len; |
3bd1e081 | 2474 | |
e7b994a3 DG |
2475 | rcu_register_thread(); |
2476 | ||
1fc79fb4 MD |
2477 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2478 | ||
2d57de81 MD |
2479 | if (testpoint(consumerd_thread_data)) { |
2480 | goto error_testpoint; | |
2481 | } | |
2482 | ||
9ce5646a MD |
2483 | health_code_update(); |
2484 | ||
4df6c8cb MD |
2485 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2486 | if (local_stream == NULL) { | |
2487 | PERROR("local_stream malloc"); | |
2488 | goto end; | |
2489 | } | |
3bd1e081 MD |
2490 | |
2491 | while (1) { | |
9ce5646a MD |
2492 | health_code_update(); |
2493 | ||
3bd1e081 MD |
2494 | high_prio = 0; |
2495 | num_hup = 0; | |
2496 | ||
2497 | /* | |
e4421fec | 2498 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2499 | * local array as well |
2500 | */ | |
2501 | pthread_mutex_lock(&consumer_data.lock); | |
2502 | if (consumer_data.need_update) { | |
0e428499 DG |
2503 | free(pollfd); |
2504 | pollfd = NULL; | |
2505 | ||
2506 | free(local_stream); | |
2507 | local_stream = NULL; | |
3bd1e081 | 2508 | |
02b3d176 DG |
2509 | /* |
2510 | * Allocate for all fds +1 for the consumer_data_pipe and +1 for | |
2511 | * wake up pipe. | |
2512 | */ | |
2513 | pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd)); | |
3bd1e081 | 2514 | if (pollfd == NULL) { |
7a57cf92 | 2515 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2516 | pthread_mutex_unlock(&consumer_data.lock); |
2517 | goto end; | |
2518 | } | |
2519 | ||
02b3d176 | 2520 | local_stream = zmalloc((consumer_data.stream_count + 2) * |
747f8642 | 2521 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2522 | if (local_stream == NULL) { |
7a57cf92 | 2523 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2524 | pthread_mutex_unlock(&consumer_data.lock); |
2525 | goto end; | |
2526 | } | |
ffe60014 | 2527 | ret = update_poll_array(ctx, &pollfd, local_stream, |
2afa177e | 2528 | data_ht, &nb_inactive_fd); |
3bd1e081 MD |
2529 | if (ret < 0) { |
2530 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2531 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2532 | pthread_mutex_unlock(&consumer_data.lock); |
2533 | goto end; | |
2534 | } | |
2535 | nb_fd = ret; | |
2536 | consumer_data.need_update = 0; | |
2537 | } | |
2538 | pthread_mutex_unlock(&consumer_data.lock); | |
2539 | ||
4078b776 | 2540 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2afa177e JD |
2541 | if (nb_fd == 0 && nb_inactive_fd == 0 && |
2542 | CMM_LOAD_SHARED(consumer_quit) == 1) { | |
1fc79fb4 | 2543 | err = 0; /* All is OK */ |
4078b776 MD |
2544 | goto end; |
2545 | } | |
3bd1e081 | 2546 | /* poll on the array of fds */ |
88f2b785 | 2547 | restart: |
02b3d176 | 2548 | DBG("polling on %d fd", nb_fd + 2); |
cf0bcb51 JG |
2549 | if (testpoint(consumerd_thread_data_poll)) { |
2550 | goto end; | |
2551 | } | |
9ce5646a | 2552 | health_poll_entry(); |
02b3d176 | 2553 | num_rdy = poll(pollfd, nb_fd + 2, -1); |
9ce5646a | 2554 | health_poll_exit(); |
3bd1e081 MD |
2555 | DBG("poll num_rdy : %d", num_rdy); |
2556 | if (num_rdy == -1) { | |
88f2b785 MD |
2557 | /* |
2558 | * Restart interrupted system call. | |
2559 | */ | |
2560 | if (errno == EINTR) { | |
2561 | goto restart; | |
2562 | } | |
7a57cf92 | 2563 | PERROR("Poll error"); |
f73fabfd | 2564 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2565 | goto end; |
2566 | } else if (num_rdy == 0) { | |
2567 | DBG("Polling thread timed out"); | |
2568 | goto end; | |
2569 | } | |
2570 | ||
19246bd6 JG |
2571 | if (caa_unlikely(data_consumption_paused)) { |
2572 | DBG("Data consumption paused, sleeping..."); | |
2573 | sleep(1); | |
2574 | goto restart; | |
2575 | } | |
2576 | ||
3bd1e081 | 2577 | /* |
50f8ae69 | 2578 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2579 | * beginning of the loop to update the array. We want to prioritize |
2580 | * array update over low-priority reads. | |
3bd1e081 | 2581 | */ |
509bb1cf | 2582 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2583 | ssize_t pipe_readlen; |
04fdd819 | 2584 | |
50f8ae69 | 2585 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2586 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2587 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2588 | if (pipe_readlen < sizeof(new_stream)) { |
2589 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2590 | /* Continue so we can at least handle the current stream(s). */ |
2591 | continue; | |
2592 | } | |
c869f647 DG |
2593 | |
2594 | /* | |
2595 | * If the stream is NULL, just ignore it. It's also possible that | |
2596 | * the sessiond poll thread changed the consumer_quit state and is | |
2597 | * waking us up to test it. | |
2598 | */ | |
2599 | if (new_stream == NULL) { | |
8994307f | 2600 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2601 | continue; |
2602 | } | |
2603 | ||
c869f647 | 2604 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2605 | continue; |
2606 | } | |
2607 | ||
02b3d176 DG |
2608 | /* Handle wakeup pipe. */ |
2609 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2610 | char dummy; | |
2611 | ssize_t pipe_readlen; | |
2612 | ||
2613 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2614 | sizeof(dummy)); | |
2615 | if (pipe_readlen < 0) { | |
2616 | PERROR("Consumer data wakeup pipe"); | |
2617 | } | |
2618 | /* We've been awakened to handle stream(s). */ | |
2619 | ctx->has_wakeup = 0; | |
2620 | } | |
2621 | ||
3bd1e081 MD |
2622 | /* Take care of high priority channels first. */ |
2623 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2624 | health_code_update(); |
2625 | ||
9617607b DG |
2626 | if (local_stream[i] == NULL) { |
2627 | continue; | |
2628 | } | |
fb3a43a9 | 2629 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2630 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2631 | high_prio = 1; | |
4078b776 | 2632 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2633 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2634 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2635 | /* Clean the stream and free it. */ |
2636 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2637 | local_stream[i] = NULL; |
4078b776 MD |
2638 | } else if (len > 0) { |
2639 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2640 | } |
3bd1e081 MD |
2641 | } |
2642 | } | |
2643 | ||
4078b776 MD |
2644 | /* |
2645 | * If we read high prio channel in this loop, try again | |
2646 | * for more high prio data. | |
2647 | */ | |
2648 | if (high_prio) { | |
3bd1e081 MD |
2649 | continue; |
2650 | } | |
2651 | ||
2652 | /* Take care of low priority channels. */ | |
4078b776 | 2653 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2654 | health_code_update(); |
2655 | ||
9617607b DG |
2656 | if (local_stream[i] == NULL) { |
2657 | continue; | |
2658 | } | |
4078b776 | 2659 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2660 | local_stream[i]->hangup_flush_done || |
2661 | local_stream[i]->has_data) { | |
4078b776 MD |
2662 | DBG("Normal read on fd %d", pollfd[i].fd); |
2663 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2664 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2665 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2666 | /* Clean the stream and free it. */ |
2667 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2668 | local_stream[i] = NULL; |
4078b776 MD |
2669 | } else if (len > 0) { |
2670 | local_stream[i]->data_read = 1; | |
2671 | } | |
2672 | } | |
2673 | } | |
2674 | ||
2675 | /* Handle hangup and errors */ | |
2676 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2677 | health_code_update(); |
2678 | ||
9617607b DG |
2679 | if (local_stream[i] == NULL) { |
2680 | continue; | |
2681 | } | |
4078b776 MD |
2682 | if (!local_stream[i]->hangup_flush_done |
2683 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2684 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2685 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2686 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2687 | pollfd[i].fd); |
4078b776 MD |
2688 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2689 | /* Attempt read again, for the data we just flushed. */ | |
2690 | local_stream[i]->data_read = 1; | |
2691 | } | |
2692 | /* | |
2693 | * If the poll flag is HUP/ERR/NVAL and we have | |
2694 | * read no data in this pass, we can remove the | |
2695 | * stream from its hash table. | |
2696 | */ | |
2697 | if ((pollfd[i].revents & POLLHUP)) { | |
2698 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2699 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2700 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2701 | local_stream[i] = NULL; |
4078b776 MD |
2702 | num_hup++; |
2703 | } | |
2704 | } else if (pollfd[i].revents & POLLERR) { | |
2705 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2706 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2707 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2708 | local_stream[i] = NULL; |
4078b776 MD |
2709 | num_hup++; |
2710 | } | |
2711 | } else if (pollfd[i].revents & POLLNVAL) { | |
2712 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2713 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2714 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2715 | local_stream[i] = NULL; |
4078b776 | 2716 | num_hup++; |
3bd1e081 MD |
2717 | } |
2718 | } | |
9617607b DG |
2719 | if (local_stream[i] != NULL) { |
2720 | local_stream[i]->data_read = 0; | |
2721 | } | |
3bd1e081 MD |
2722 | } |
2723 | } | |
1fc79fb4 MD |
2724 | /* All is OK */ |
2725 | err = 0; | |
3bd1e081 MD |
2726 | end: |
2727 | DBG("polling thread exiting"); | |
0e428499 DG |
2728 | free(pollfd); |
2729 | free(local_stream); | |
fb3a43a9 DG |
2730 | |
2731 | /* | |
2732 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2733 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2734 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2735 | * not return and could create a endless wait period if the pipe is the | |
2736 | * only tracked fd in the poll set. The thread will take care of closing | |
2737 | * the read side. | |
fb3a43a9 | 2738 | */ |
13886d2d | 2739 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2740 | |
2d57de81 | 2741 | error_testpoint: |
1fc79fb4 MD |
2742 | if (err) { |
2743 | health_error(); | |
2744 | ERR("Health error occurred in %s", __func__); | |
2745 | } | |
2746 | health_unregister(health_consumerd); | |
2747 | ||
e7b994a3 | 2748 | rcu_unregister_thread(); |
3bd1e081 MD |
2749 | return NULL; |
2750 | } | |
2751 | ||
d8ef542d MD |
2752 | /* |
2753 | * Close wake-up end of each stream belonging to the channel. This will | |
2754 | * allow the poll() on the stream read-side to detect when the | |
2755 | * write-side (application) finally closes them. | |
2756 | */ | |
2757 | static | |
2758 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2759 | { | |
2760 | struct lttng_ht *ht; | |
2761 | struct lttng_consumer_stream *stream; | |
2762 | struct lttng_ht_iter iter; | |
2763 | ||
2764 | ht = consumer_data.stream_per_chan_id_ht; | |
2765 | ||
2766 | rcu_read_lock(); | |
2767 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2768 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2769 | ht->match_fct, &channel->key, | |
2770 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2771 | /* |
2772 | * Protect against teardown with mutex. | |
2773 | */ | |
2774 | pthread_mutex_lock(&stream->lock); | |
2775 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2776 | goto next; | |
2777 | } | |
d8ef542d MD |
2778 | switch (consumer_data.type) { |
2779 | case LTTNG_CONSUMER_KERNEL: | |
2780 | break; | |
2781 | case LTTNG_CONSUMER32_UST: | |
2782 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2783 | if (stream->metadata_flag) { |
2784 | /* Safe and protected by the stream lock. */ | |
2785 | lttng_ustconsumer_close_metadata(stream->chan); | |
2786 | } else { | |
2787 | /* | |
2788 | * Note: a mutex is taken internally within | |
2789 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2790 | * use from concurrent close. | |
2791 | */ | |
2792 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2793 | } | |
d8ef542d MD |
2794 | break; |
2795 | default: | |
2796 | ERR("Unknown consumer_data type"); | |
2797 | assert(0); | |
2798 | } | |
f2ad556d MD |
2799 | next: |
2800 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2801 | } |
2802 | rcu_read_unlock(); | |
2803 | } | |
2804 | ||
2805 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2806 | { | |
2807 | struct lttng_ht_iter iter; | |
2808 | struct lttng_consumer_channel *channel; | |
2809 | int ret; | |
2810 | ||
2811 | if (ht == NULL) { | |
2812 | return; | |
2813 | } | |
2814 | ||
2815 | rcu_read_lock(); | |
2816 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2817 | ret = lttng_ht_del(ht, &iter); | |
2818 | assert(ret != 0); | |
2819 | } | |
2820 | rcu_read_unlock(); | |
2821 | ||
2822 | lttng_ht_destroy(ht); | |
2823 | } | |
2824 | ||
2825 | /* | |
2826 | * This thread polls the channel fds to detect when they are being | |
2827 | * closed. It closes all related streams if the channel is detected as | |
2828 | * closed. It is currently only used as a shim layer for UST because the | |
2829 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2830 | * periodical flush. | |
2831 | */ | |
2832 | void *consumer_thread_channel_poll(void *data) | |
2833 | { | |
1fc79fb4 | 2834 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2835 | uint32_t revents, nb_fd; |
2836 | struct lttng_consumer_channel *chan = NULL; | |
2837 | struct lttng_ht_iter iter; | |
2838 | struct lttng_ht_node_u64 *node; | |
2839 | struct lttng_poll_event events; | |
2840 | struct lttng_consumer_local_data *ctx = data; | |
2841 | struct lttng_ht *channel_ht; | |
2842 | ||
2843 | rcu_register_thread(); | |
2844 | ||
1fc79fb4 MD |
2845 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2846 | ||
2d57de81 MD |
2847 | if (testpoint(consumerd_thread_channel)) { |
2848 | goto error_testpoint; | |
2849 | } | |
2850 | ||
9ce5646a MD |
2851 | health_code_update(); |
2852 | ||
d8ef542d MD |
2853 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2854 | if (!channel_ht) { | |
2855 | /* ENOMEM at this point. Better to bail out. */ | |
2856 | goto end_ht; | |
2857 | } | |
2858 | ||
2859 | DBG("Thread channel poll started"); | |
2860 | ||
2861 | /* Size is set to 1 for the consumer_channel pipe */ | |
2862 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2863 | if (ret < 0) { | |
2864 | ERR("Poll set creation failed"); | |
2865 | goto end_poll; | |
2866 | } | |
2867 | ||
2868 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2869 | if (ret < 0) { | |
2870 | goto end; | |
2871 | } | |
2872 | ||
2873 | /* Main loop */ | |
2874 | DBG("Channel main loop started"); | |
2875 | ||
2876 | while (1) { | |
d8ef542d | 2877 | restart: |
7fa2082e MD |
2878 | health_code_update(); |
2879 | DBG("Channel poll wait"); | |
9ce5646a | 2880 | health_poll_entry(); |
d8ef542d | 2881 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2882 | DBG("Channel poll return from wait with %d fd(s)", |
2883 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2884 | health_poll_exit(); |
40063ead | 2885 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2886 | if (ret < 0) { |
2887 | if (errno == EINTR) { | |
40063ead | 2888 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2889 | goto restart; |
2890 | } | |
d9607cd7 MD |
2891 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2892 | err = 0; /* All is OK */ | |
2893 | } | |
d8ef542d MD |
2894 | goto end; |
2895 | } | |
2896 | ||
2897 | nb_fd = ret; | |
2898 | ||
2899 | /* From here, the event is a channel wait fd */ | |
2900 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2901 | health_code_update(); |
2902 | ||
d8ef542d MD |
2903 | revents = LTTNG_POLL_GETEV(&events, i); |
2904 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2905 | ||
d8ef542d | 2906 | if (!revents) { |
fd20dac9 | 2907 | /* No activity for this FD (poll implementation). */ |
d8ef542d MD |
2908 | continue; |
2909 | } | |
fd20dac9 | 2910 | |
d8ef542d | 2911 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
03e43155 | 2912 | if (revents & LPOLLIN) { |
d8ef542d | 2913 | enum consumer_channel_action action; |
a0cbdd2e | 2914 | uint64_t key; |
d8ef542d | 2915 | |
a0cbdd2e | 2916 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2917 | if (ret <= 0) { |
03e43155 MD |
2918 | if (ret < 0) { |
2919 | ERR("Error reading channel pipe"); | |
2920 | } | |
2921 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
2922 | continue; |
2923 | } | |
2924 | ||
2925 | switch (action) { | |
2926 | case CONSUMER_CHANNEL_ADD: | |
2927 | DBG("Adding channel %d to poll set", | |
2928 | chan->wait_fd); | |
2929 | ||
2930 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2931 | chan->wait_fd); | |
c7260a81 | 2932 | rcu_read_lock(); |
d8ef542d MD |
2933 | lttng_ht_add_unique_u64(channel_ht, |
2934 | &chan->wait_fd_node); | |
c7260a81 | 2935 | rcu_read_unlock(); |
d8ef542d MD |
2936 | /* Add channel to the global poll events list */ |
2937 | lttng_poll_add(&events, chan->wait_fd, | |
03e43155 | 2938 | LPOLLERR | LPOLLHUP); |
d8ef542d | 2939 | break; |
a0cbdd2e MD |
2940 | case CONSUMER_CHANNEL_DEL: |
2941 | { | |
b4a650f3 DG |
2942 | /* |
2943 | * This command should never be called if the channel | |
2944 | * has streams monitored by either the data or metadata | |
2945 | * thread. The consumer only notify this thread with a | |
2946 | * channel del. command if it receives a destroy | |
2947 | * channel command from the session daemon that send it | |
2948 | * if a command prior to the GET_CHANNEL failed. | |
2949 | */ | |
2950 | ||
c7260a81 | 2951 | rcu_read_lock(); |
a0cbdd2e MD |
2952 | chan = consumer_find_channel(key); |
2953 | if (!chan) { | |
c7260a81 | 2954 | rcu_read_unlock(); |
a0cbdd2e MD |
2955 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2956 | break; | |
2957 | } | |
2958 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2959 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2960 | ret = lttng_ht_del(channel_ht, &iter); |
2961 | assert(ret == 0); | |
a0cbdd2e | 2962 | |
f2a444f1 DG |
2963 | switch (consumer_data.type) { |
2964 | case LTTNG_CONSUMER_KERNEL: | |
2965 | break; | |
2966 | case LTTNG_CONSUMER32_UST: | |
2967 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
2968 | health_code_update(); |
2969 | /* Destroy streams that might have been left in the stream list. */ | |
2970 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
2971 | break; |
2972 | default: | |
2973 | ERR("Unknown consumer_data type"); | |
2974 | assert(0); | |
2975 | } | |
2976 | ||
a0cbdd2e MD |
2977 | /* |
2978 | * Release our own refcount. Force channel deletion even if | |
2979 | * streams were not initialized. | |
2980 | */ | |
2981 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2982 | consumer_del_channel(chan); | |
2983 | } | |
c7260a81 | 2984 | rcu_read_unlock(); |
a0cbdd2e MD |
2985 | goto restart; |
2986 | } | |
d8ef542d MD |
2987 | case CONSUMER_CHANNEL_QUIT: |
2988 | /* | |
2989 | * Remove the pipe from the poll set and continue the loop | |
2990 | * since their might be data to consume. | |
2991 | */ | |
2992 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2993 | continue; | |
2994 | default: | |
2995 | ERR("Unknown action"); | |
2996 | break; | |
2997 | } | |
03e43155 MD |
2998 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2999 | DBG("Channel thread pipe hung up"); | |
3000 | /* | |
3001 | * Remove the pipe from the poll set and continue the loop | |
3002 | * since their might be data to consume. | |
3003 | */ | |
3004 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3005 | continue; | |
3006 | } else { | |
3007 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3008 | goto end; | |
d8ef542d MD |
3009 | } |
3010 | ||
3011 | /* Handle other stream */ | |
3012 | continue; | |
3013 | } | |
3014 | ||
3015 | rcu_read_lock(); | |
3016 | { | |
3017 | uint64_t tmp_id = (uint64_t) pollfd; | |
3018 | ||
3019 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
3020 | } | |
3021 | node = lttng_ht_iter_get_node_u64(&iter); | |
3022 | assert(node); | |
3023 | ||
3024 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
3025 | wait_fd_node); | |
3026 | ||
3027 | /* Check for error event */ | |
3028 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
3029 | DBG("Channel fd %d is hup|err.", pollfd); | |
3030 | ||
3031 | lttng_poll_del(&events, chan->wait_fd); | |
3032 | ret = lttng_ht_del(channel_ht, &iter); | |
3033 | assert(ret == 0); | |
b4a650f3 DG |
3034 | |
3035 | /* | |
3036 | * This will close the wait fd for each stream associated to | |
3037 | * this channel AND monitored by the data/metadata thread thus | |
3038 | * will be clean by the right thread. | |
3039 | */ | |
d8ef542d | 3040 | consumer_close_channel_streams(chan); |
f2ad556d MD |
3041 | |
3042 | /* Release our own refcount */ | |
3043 | if (!uatomic_sub_return(&chan->refcount, 1) | |
3044 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
3045 | consumer_del_channel(chan); | |
3046 | } | |
03e43155 MD |
3047 | } else { |
3048 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3049 | rcu_read_unlock(); | |
3050 | goto end; | |
d8ef542d MD |
3051 | } |
3052 | ||
3053 | /* Release RCU lock for the channel looked up */ | |
3054 | rcu_read_unlock(); | |
3055 | } | |
3056 | } | |
3057 | ||
1fc79fb4 MD |
3058 | /* All is OK */ |
3059 | err = 0; | |
d8ef542d MD |
3060 | end: |
3061 | lttng_poll_clean(&events); | |
3062 | end_poll: | |
3063 | destroy_channel_ht(channel_ht); | |
3064 | end_ht: | |
2d57de81 | 3065 | error_testpoint: |
d8ef542d | 3066 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
3067 | if (err) { |
3068 | health_error(); | |
3069 | ERR("Health error occurred in %s", __func__); | |
3070 | } | |
3071 | health_unregister(health_consumerd); | |
d8ef542d MD |
3072 | rcu_unregister_thread(); |
3073 | return NULL; | |
3074 | } | |
3075 | ||
331744e3 JD |
3076 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
3077 | struct pollfd *sockpoll, int client_socket) | |
3078 | { | |
3079 | int ret; | |
3080 | ||
3081 | assert(ctx); | |
3082 | assert(sockpoll); | |
3083 | ||
84382d49 MD |
3084 | ret = lttng_consumer_poll_socket(sockpoll); |
3085 | if (ret) { | |
331744e3 JD |
3086 | goto error; |
3087 | } | |
3088 | DBG("Metadata connection on client_socket"); | |
3089 | ||
3090 | /* Blocking call, waiting for transmission */ | |
3091 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3092 | if (ctx->consumer_metadata_socket < 0) { | |
3093 | WARN("On accept metadata"); | |
3094 | ret = -1; | |
3095 | goto error; | |
3096 | } | |
3097 | ret = 0; | |
3098 | ||
3099 | error: | |
3100 | return ret; | |
3101 | } | |
3102 | ||
3bd1e081 MD |
3103 | /* |
3104 | * This thread listens on the consumerd socket and receives the file | |
3105 | * descriptors from the session daemon. | |
3106 | */ | |
7d980def | 3107 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3108 | { |
1fc79fb4 | 3109 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3110 | /* |
3111 | * structure to poll for incoming data on communication socket avoids | |
3112 | * making blocking sockets. | |
3113 | */ | |
3114 | struct pollfd consumer_sockpoll[2]; | |
3115 | struct lttng_consumer_local_data *ctx = data; | |
3116 | ||
e7b994a3 DG |
3117 | rcu_register_thread(); |
3118 | ||
1fc79fb4 MD |
3119 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3120 | ||
2d57de81 MD |
3121 | if (testpoint(consumerd_thread_sessiond)) { |
3122 | goto error_testpoint; | |
3123 | } | |
3124 | ||
9ce5646a MD |
3125 | health_code_update(); |
3126 | ||
3bd1e081 MD |
3127 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3128 | unlink(ctx->consumer_command_sock_path); | |
3129 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3130 | if (client_socket < 0) { | |
3131 | ERR("Cannot create command socket"); | |
3132 | goto end; | |
3133 | } | |
3134 | ||
3135 | ret = lttcomm_listen_unix_sock(client_socket); | |
3136 | if (ret < 0) { | |
3137 | goto end; | |
3138 | } | |
3139 | ||
32258573 | 3140 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3141 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3142 | /* return < 0 on error, but == 0 is not fatal */ |
3143 | if (ret < 0) { | |
32258573 | 3144 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3145 | goto end; |
3146 | } | |
3147 | ||
3bd1e081 MD |
3148 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3149 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3150 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3151 | consumer_sockpoll[1].fd = client_socket; | |
3152 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3153 | ||
84382d49 MD |
3154 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3155 | if (ret) { | |
3156 | if (ret > 0) { | |
3157 | /* should exit */ | |
3158 | err = 0; | |
3159 | } | |
3bd1e081 MD |
3160 | goto end; |
3161 | } | |
3162 | DBG("Connection on client_socket"); | |
3163 | ||
3164 | /* Blocking call, waiting for transmission */ | |
3165 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3166 | if (sock < 0) { |
3bd1e081 MD |
3167 | WARN("On accept"); |
3168 | goto end; | |
3169 | } | |
3bd1e081 | 3170 | |
331744e3 JD |
3171 | /* |
3172 | * Setup metadata socket which is the second socket connection on the | |
3173 | * command unix socket. | |
3174 | */ | |
3175 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3176 | if (ret) { |
3177 | if (ret > 0) { | |
3178 | /* should exit */ | |
3179 | err = 0; | |
3180 | } | |
331744e3 JD |
3181 | goto end; |
3182 | } | |
3183 | ||
d96f09c6 DG |
3184 | /* This socket is not useful anymore. */ |
3185 | ret = close(client_socket); | |
3186 | if (ret < 0) { | |
3187 | PERROR("close client_socket"); | |
3188 | } | |
3189 | client_socket = -1; | |
3190 | ||
3bd1e081 MD |
3191 | /* update the polling structure to poll on the established socket */ |
3192 | consumer_sockpoll[1].fd = sock; | |
3193 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3194 | ||
3195 | while (1) { | |
9ce5646a MD |
3196 | health_code_update(); |
3197 | ||
3198 | health_poll_entry(); | |
3199 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3200 | health_poll_exit(); | |
84382d49 MD |
3201 | if (ret) { |
3202 | if (ret > 0) { | |
3203 | /* should exit */ | |
3204 | err = 0; | |
3205 | } | |
3bd1e081 MD |
3206 | goto end; |
3207 | } | |
3208 | DBG("Incoming command on sock"); | |
3209 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3210 | if (ret <= 0) { |
3211 | /* | |
3212 | * This could simply be a session daemon quitting. Don't output | |
3213 | * ERR() here. | |
3214 | */ | |
3215 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3216 | err = 0; |
3bd1e081 MD |
3217 | goto end; |
3218 | } | |
f7389fbb | 3219 | if (CMM_LOAD_SHARED(consumer_quit)) { |
3bd1e081 | 3220 | DBG("consumer_thread_receive_fds received quit from signal"); |
1fc79fb4 | 3221 | err = 0; /* All is OK */ |
3bd1e081 MD |
3222 | goto end; |
3223 | } | |
ffe60014 | 3224 | DBG("received command on sock"); |
3bd1e081 | 3225 | } |
1fc79fb4 MD |
3226 | /* All is OK */ |
3227 | err = 0; | |
3228 | ||
3bd1e081 | 3229 | end: |
ffe60014 | 3230 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3231 | |
d88aee68 DG |
3232 | /* |
3233 | * Close metadata streams since the producer is the session daemon which | |
3234 | * just died. | |
3235 | * | |
3236 | * NOTE: for now, this only applies to the UST tracer. | |
3237 | */ | |
6d574024 | 3238 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3239 | |
3bd1e081 MD |
3240 | /* |
3241 | * when all fds have hung up, the polling thread | |
3242 | * can exit cleanly | |
3243 | */ | |
f7389fbb | 3244 | CMM_STORE_SHARED(consumer_quit, 1); |
3bd1e081 | 3245 | |
04fdd819 | 3246 | /* |
c869f647 | 3247 | * Notify the data poll thread to poll back again and test the |
8994307f | 3248 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3249 | */ |
acdb9057 | 3250 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3251 | |
a0cbdd2e | 3252 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3253 | |
5c635c72 MD |
3254 | notify_health_quit_pipe(health_quit_pipe); |
3255 | ||
d96f09c6 DG |
3256 | /* Cleaning up possibly open sockets. */ |
3257 | if (sock >= 0) { | |
3258 | ret = close(sock); | |
3259 | if (ret < 0) { | |
3260 | PERROR("close sock sessiond poll"); | |
3261 | } | |
3262 | } | |
3263 | if (client_socket >= 0) { | |
38476d24 | 3264 | ret = close(client_socket); |
d96f09c6 DG |
3265 | if (ret < 0) { |
3266 | PERROR("close client_socket sessiond poll"); | |
3267 | } | |
3268 | } | |
3269 | ||
2d57de81 | 3270 | error_testpoint: |
1fc79fb4 MD |
3271 | if (err) { |
3272 | health_error(); | |
3273 | ERR("Health error occurred in %s", __func__); | |
3274 | } | |
3275 | health_unregister(health_consumerd); | |
3276 | ||
e7b994a3 | 3277 | rcu_unregister_thread(); |
3bd1e081 MD |
3278 | return NULL; |
3279 | } | |
d41f73b7 | 3280 | |
4078b776 | 3281 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3282 | struct lttng_consumer_local_data *ctx) |
3283 | { | |
74251bb8 DG |
3284 | ssize_t ret; |
3285 | ||
3286 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3287 | if (stream->metadata_flag) { |
3288 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3289 | } | |
74251bb8 | 3290 | |
d41f73b7 MD |
3291 | switch (consumer_data.type) { |
3292 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3293 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3294 | break; | |
7753dea8 MD |
3295 | case LTTNG_CONSUMER32_UST: |
3296 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3297 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3298 | break; | |
d41f73b7 MD |
3299 | default: |
3300 | ERR("Unknown consumer_data type"); | |
3301 | assert(0); | |
74251bb8 DG |
3302 | ret = -ENOSYS; |
3303 | break; | |
d41f73b7 | 3304 | } |
74251bb8 | 3305 | |
94d49140 JD |
3306 | if (stream->metadata_flag) { |
3307 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3308 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3309 | } | |
74251bb8 DG |
3310 | pthread_mutex_unlock(&stream->lock); |
3311 | return ret; | |
d41f73b7 MD |
3312 | } |
3313 | ||
3314 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3315 | { | |
3316 | switch (consumer_data.type) { | |
3317 | case LTTNG_CONSUMER_KERNEL: | |
3318 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3319 | case LTTNG_CONSUMER32_UST: |
3320 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3321 | return lttng_ustconsumer_on_recv_stream(stream); |
3322 | default: | |
3323 | ERR("Unknown consumer_data type"); | |
3324 | assert(0); | |
3325 | return -ENOSYS; | |
3326 | } | |
3327 | } | |
e4421fec DG |
3328 | |
3329 | /* | |
3330 | * Allocate and set consumer data hash tables. | |
3331 | */ | |
282dadbc | 3332 | int lttng_consumer_init(void) |
e4421fec | 3333 | { |
d88aee68 | 3334 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3335 | if (!consumer_data.channel_ht) { |
3336 | goto error; | |
3337 | } | |
3338 | ||
d88aee68 | 3339 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3340 | if (!consumer_data.relayd_ht) { |
3341 | goto error; | |
3342 | } | |
3343 | ||
d88aee68 | 3344 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3345 | if (!consumer_data.stream_list_ht) { |
3346 | goto error; | |
3347 | } | |
3348 | ||
d8ef542d | 3349 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3350 | if (!consumer_data.stream_per_chan_id_ht) { |
3351 | goto error; | |
3352 | } | |
3353 | ||
3354 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3355 | if (!data_ht) { | |
3356 | goto error; | |
3357 | } | |
3358 | ||
3359 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3360 | if (!metadata_ht) { | |
3361 | goto error; | |
3362 | } | |
3363 | ||
3364 | return 0; | |
3365 | ||
3366 | error: | |
3367 | return -1; | |
e4421fec | 3368 | } |
7735ef9e DG |
3369 | |
3370 | /* | |
3371 | * Process the ADD_RELAYD command receive by a consumer. | |
3372 | * | |
3373 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3374 | * The caller MUST acquire a RCU read side lock before calling it. | |
3375 | */ | |
3ab4ee7d | 3376 | void consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3377 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3378 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3379 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3380 | uint64_t relayd_session_id) | |
7735ef9e | 3381 | { |
cd2b09ed | 3382 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3383 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3384 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3385 | |
6151a90f JD |
3386 | assert(ctx); |
3387 | assert(relayd_sock); | |
3388 | ||
da009f2c | 3389 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3390 | |
3391 | /* Get relayd reference if exists. */ | |
3392 | relayd = consumer_find_relayd(net_seq_idx); | |
3393 | if (relayd == NULL) { | |
da009f2c | 3394 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3395 | /* Not found. Allocate one. */ |
3396 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3397 | if (relayd == NULL) { | |
618a6a28 MD |
3398 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3399 | goto error; | |
0d08d75e | 3400 | } else { |
30319bcb | 3401 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3402 | relayd_created = 1; |
7735ef9e | 3403 | } |
0d08d75e DG |
3404 | |
3405 | /* | |
3406 | * This code path MUST continue to the consumer send status message to | |
3407 | * we can notify the session daemon and continue our work without | |
3408 | * killing everything. | |
3409 | */ | |
da009f2c MD |
3410 | } else { |
3411 | /* | |
3412 | * relayd key should never be found for control socket. | |
3413 | */ | |
3414 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3415 | } |
3416 | ||
3417 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3418 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3419 | if (ret < 0) { |
0d08d75e | 3420 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3421 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3422 | goto error_nosignal; | |
7735ef9e DG |
3423 | } |
3424 | ||
3425 | /* Poll on consumer socket. */ | |
84382d49 MD |
3426 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3427 | if (ret) { | |
3428 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3429 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
618a6a28 | 3430 | goto error_nosignal; |
7735ef9e DG |
3431 | } |
3432 | ||
3433 | /* Get relayd socket from session daemon */ | |
3434 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3435 | if (ret != sizeof(fd)) { | |
4028eeb9 | 3436 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3437 | |
3438 | /* | |
3439 | * Failing to receive FDs might indicate a major problem such as | |
3440 | * reaching a fd limit during the receive where the kernel returns a | |
3441 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3442 | * don't take any chances and stop everything. | |
3443 | * | |
3444 | * XXX: Feature request #558 will fix that and avoid this possible | |
3445 | * issue when reaching the fd limit. | |
3446 | */ | |
3447 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3448 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3449 | goto error; |
3450 | } | |
3451 | ||
7735ef9e DG |
3452 | /* Copy socket information and received FD */ |
3453 | switch (sock_type) { | |
3454 | case LTTNG_STREAM_CONTROL: | |
3455 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3456 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3457 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3458 | /* Handle create_sock error. */ |
f66c074c | 3459 | if (ret < 0) { |
618a6a28 | 3460 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3461 | goto error; |
f66c074c | 3462 | } |
da009f2c MD |
3463 | /* |
3464 | * Close the socket created internally by | |
3465 | * lttcomm_create_sock, so we can replace it by the one | |
3466 | * received from sessiond. | |
3467 | */ | |
3468 | if (close(relayd->control_sock.sock.fd)) { | |
3469 | PERROR("close"); | |
3470 | } | |
7735ef9e DG |
3471 | |
3472 | /* Assign new file descriptor */ | |
6151a90f | 3473 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3474 | fd = -1; /* For error path */ |
6151a90f JD |
3475 | /* Assign version values. */ |
3476 | relayd->control_sock.major = relayd_sock->major; | |
3477 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3478 | |
d3e2ba59 | 3479 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3480 | |
7735ef9e DG |
3481 | break; |
3482 | case LTTNG_STREAM_DATA: | |
3483 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3484 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3485 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3486 | /* Handle create_sock error. */ |
f66c074c | 3487 | if (ret < 0) { |
618a6a28 | 3488 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3489 | goto error; |
f66c074c | 3490 | } |
da009f2c MD |
3491 | /* |
3492 | * Close the socket created internally by | |
3493 | * lttcomm_create_sock, so we can replace it by the one | |
3494 | * received from sessiond. | |
3495 | */ | |
3496 | if (close(relayd->data_sock.sock.fd)) { | |
3497 | PERROR("close"); | |
3498 | } | |
7735ef9e DG |
3499 | |
3500 | /* Assign new file descriptor */ | |
6151a90f | 3501 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3502 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3503 | /* Assign version values. */ |
3504 | relayd->data_sock.major = relayd_sock->major; | |
3505 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3506 | break; |
3507 | default: | |
3508 | ERR("Unknown relayd socket type (%d)", sock_type); | |
618a6a28 | 3509 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3510 | goto error; |
3511 | } | |
3512 | ||
d88aee68 | 3513 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3514 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3515 | relayd->net_seq_idx, fd); | |
3516 | ||
618a6a28 MD |
3517 | /* We successfully added the socket. Send status back. */ |
3518 | ret = consumer_send_status_msg(sock, ret_code); | |
3519 | if (ret < 0) { | |
3520 | /* Somehow, the session daemon is not responding anymore. */ | |
3521 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3522 | goto error_nosignal; | |
3523 | } | |
3524 | ||
7735ef9e DG |
3525 | /* |
3526 | * Add relayd socket pair to consumer data hashtable. If object already | |
3527 | * exists or on error, the function gracefully returns. | |
3528 | */ | |
4c5691ba | 3529 | relayd->ctx = ctx; |
d09e1200 | 3530 | add_relayd(relayd); |
7735ef9e DG |
3531 | |
3532 | /* All good! */ | |
3ab4ee7d | 3533 | return; |
7735ef9e DG |
3534 | |
3535 | error: | |
618a6a28 MD |
3536 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3537 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3538 | } | |
3539 | ||
3540 | error_nosignal: | |
4028eeb9 DG |
3541 | /* Close received socket if valid. */ |
3542 | if (fd >= 0) { | |
3543 | if (close(fd)) { | |
3544 | PERROR("close received socket"); | |
3545 | } | |
3546 | } | |
cd2b09ed DG |
3547 | |
3548 | if (relayd_created) { | |
cd2b09ed DG |
3549 | free(relayd); |
3550 | } | |
7735ef9e | 3551 | } |
ca22feea | 3552 | |
f7079f67 DG |
3553 | /* |
3554 | * Search for a relayd associated to the session id and return the reference. | |
3555 | * | |
3556 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3557 | * until the relayd object is no longer necessary. | |
3558 | */ | |
3559 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3560 | { | |
3561 | struct lttng_ht_iter iter; | |
f7079f67 | 3562 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3563 | |
3564 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3565 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3566 | node.node) { | |
18261bd1 DG |
3567 | /* |
3568 | * Check by sessiond id which is unique here where the relayd session | |
3569 | * id might not be when having multiple relayd. | |
3570 | */ | |
3571 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3572 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3573 | goto found; |
f7079f67 DG |
3574 | } |
3575 | } | |
3576 | ||
18261bd1 DG |
3577 | return NULL; |
3578 | ||
3579 | found: | |
f7079f67 DG |
3580 | return relayd; |
3581 | } | |
3582 | ||
ca22feea DG |
3583 | /* |
3584 | * Check if for a given session id there is still data needed to be extract | |
3585 | * from the buffers. | |
3586 | * | |
6d805429 | 3587 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3588 | */ |
6d805429 | 3589 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3590 | { |
3591 | int ret; | |
3592 | struct lttng_ht_iter iter; | |
3593 | struct lttng_ht *ht; | |
3594 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3595 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3596 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3597 | |
6d805429 | 3598 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3599 | |
6f6eda74 | 3600 | rcu_read_lock(); |
ca22feea DG |
3601 | pthread_mutex_lock(&consumer_data.lock); |
3602 | ||
3603 | switch (consumer_data.type) { | |
3604 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3605 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3606 | break; |
3607 | case LTTNG_CONSUMER32_UST: | |
3608 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3609 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3610 | break; |
3611 | default: | |
3612 | ERR("Unknown consumer data type"); | |
3613 | assert(0); | |
3614 | } | |
3615 | ||
3616 | /* Ease our life a bit */ | |
3617 | ht = consumer_data.stream_list_ht; | |
3618 | ||
c8f59ee5 | 3619 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3620 | ht->hash_fct(&id, lttng_ht_seed), |
3621 | ht->match_fct, &id, | |
ca22feea | 3622 | &iter.iter, stream, node_session_id.node) { |
e3899a4d | 3623 | pthread_mutex_lock(&stream->lock); |
ca22feea | 3624 | |
4e9a4686 DG |
3625 | /* |
3626 | * A removed node from the hash table indicates that the stream has | |
3627 | * been deleted thus having a guarantee that the buffers are closed | |
3628 | * on the consumer side. However, data can still be transmitted | |
3629 | * over the network so don't skip the relayd check. | |
3630 | */ | |
3631 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3632 | if (!ret) { | |
3633 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3634 | ret = data_pending(stream); |
3635 | if (ret == 1) { | |
4e9a4686 | 3636 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3637 | goto data_pending; |
4e9a4686 DG |
3638 | } |
3639 | } | |
3640 | ||
b7891251 JR |
3641 | pthread_mutex_unlock(&stream->lock); |
3642 | } | |
3643 | ||
3644 | relayd = find_relayd_by_session_id(id); | |
3645 | if (relayd) { | |
3646 | unsigned int is_data_inflight = 0; | |
3647 | ||
3648 | /* Send init command for data pending. */ | |
3649 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3650 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3651 | relayd->relayd_session_id); | |
3652 | if (ret < 0) { | |
3653 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3654 | /* Communication error thus the relayd so no data pending. */ | |
3655 | goto data_not_pending; | |
3656 | } | |
3657 | ||
3658 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
3659 | ht->hash_fct(&id, lttng_ht_seed), | |
3660 | ht->match_fct, &id, | |
3661 | &iter.iter, stream, node_session_id.node) { | |
c8f59ee5 | 3662 | if (stream->metadata_flag) { |
ad7051c0 DG |
3663 | ret = relayd_quiescent_control(&relayd->control_sock, |
3664 | stream->relayd_stream_id); | |
c8f59ee5 | 3665 | } else { |
6d805429 | 3666 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3667 | stream->relayd_stream_id, |
3668 | stream->next_net_seq_num - 1); | |
c8f59ee5 | 3669 | } |
b7891251 JR |
3670 | |
3671 | if (ret == 1) { | |
3672 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3673 | goto data_pending; | |
3674 | } else if (ret < 0) { | |
4c5691ba JR |
3675 | ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
3676 | lttng_consumer_cleanup_relayd(relayd); | |
3677 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
4c5691ba JR |
3678 | goto data_not_pending; |
3679 | } | |
c8f59ee5 | 3680 | } |
f7079f67 | 3681 | |
b7891251 | 3682 | /* Send end command for data pending. */ |
f7079f67 DG |
3683 | ret = relayd_end_data_pending(&relayd->control_sock, |
3684 | relayd->relayd_session_id, &is_data_inflight); | |
3685 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3686 | if (ret < 0) { |
4c5691ba JR |
3687 | ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
3688 | lttng_consumer_cleanup_relayd(relayd); | |
f7079f67 DG |
3689 | goto data_not_pending; |
3690 | } | |
bdd88757 DG |
3691 | if (is_data_inflight) { |
3692 | goto data_pending; | |
3693 | } | |
f7079f67 DG |
3694 | } |
3695 | ||
ca22feea | 3696 | /* |
f7079f67 DG |
3697 | * Finding _no_ node in the hash table and no inflight data means that the |
3698 | * stream(s) have been removed thus data is guaranteed to be available for | |
3699 | * analysis from the trace files. | |
ca22feea DG |
3700 | */ |
3701 | ||
f7079f67 | 3702 | data_not_pending: |
ca22feea DG |
3703 | /* Data is available to be read by a viewer. */ |
3704 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3705 | rcu_read_unlock(); |
6d805429 | 3706 | return 0; |
ca22feea | 3707 | |
f7079f67 | 3708 | data_pending: |
ca22feea DG |
3709 | /* Data is still being extracted from buffers. */ |
3710 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3711 | rcu_read_unlock(); |
6d805429 | 3712 | return 1; |
ca22feea | 3713 | } |
f50f23d9 DG |
3714 | |
3715 | /* | |
3716 | * Send a ret code status message to the sessiond daemon. | |
3717 | * | |
3718 | * Return the sendmsg() return value. | |
3719 | */ | |
3720 | int consumer_send_status_msg(int sock, int ret_code) | |
3721 | { | |
3722 | struct lttcomm_consumer_status_msg msg; | |
3723 | ||
53efb85a | 3724 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3725 | msg.ret_code = ret_code; |
3726 | ||
3727 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3728 | } | |
ffe60014 DG |
3729 | |
3730 | /* | |
3731 | * Send a channel status message to the sessiond daemon. | |
3732 | * | |
3733 | * Return the sendmsg() return value. | |
3734 | */ | |
3735 | int consumer_send_status_channel(int sock, | |
3736 | struct lttng_consumer_channel *channel) | |
3737 | { | |
3738 | struct lttcomm_consumer_status_channel msg; | |
3739 | ||
3740 | assert(sock >= 0); | |
3741 | ||
53efb85a | 3742 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3743 | if (!channel) { |
0c759fc9 | 3744 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3745 | } else { |
0c759fc9 | 3746 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3747 | msg.key = channel->key; |
3748 | msg.stream_count = channel->streams.count; | |
3749 | } | |
3750 | ||
3751 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3752 | } | |
5c786ded | 3753 | |
d07ceecd MD |
3754 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3755 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3756 | uint64_t max_sb_size) | |
5c786ded | 3757 | { |
d07ceecd | 3758 | unsigned long start_pos; |
5c786ded | 3759 | |
d07ceecd MD |
3760 | if (!nb_packets_per_stream) { |
3761 | return consumed_pos; /* Grab everything */ | |
3762 | } | |
3763 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3764 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3765 | if ((long) (start_pos - consumed_pos) < 0) { | |
3766 | return consumed_pos; /* Grab everything */ | |
3767 | } | |
3768 | return start_pos; | |
5c786ded | 3769 | } |