Commit | Line | Data |
---|---|---|
3bd1e081 MD |
1 | /* |
2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> | |
3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
00e2e675 | 4 | * 2012 - David Goulet <dgoulet@efficios.com> |
3bd1e081 | 5 | * |
d14d33bf AM |
6 | * This program is free software; you can redistribute it and/or modify |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
3bd1e081 | 9 | * |
d14d33bf AM |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
3bd1e081 | 14 | * |
d14d33bf AM |
15 | * You should have received a copy of the GNU General Public License along |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
3bd1e081 MD |
18 | */ |
19 | ||
20 | #define _GNU_SOURCE | |
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 | |
990570ed | 33 | #include <common/common.h> |
fb3a43a9 DG |
34 | #include <common/utils.h> |
35 | #include <common/compat/poll.h> | |
10a8a223 | 36 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 37 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
38 | #include <common/sessiond-comm/sessiond-comm.h> |
39 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 40 | #include <common/relayd/relayd.h> |
10a8a223 DG |
41 | #include <common/ust-consumer/ust-consumer.h> |
42 | ||
43 | #include "consumer.h" | |
1d1a276c | 44 | #include "consumer-stream.h" |
3bd1e081 MD |
45 | |
46 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
47 | .stream_count = 0, |
48 | .need_update = 1, | |
49 | .type = LTTNG_CONSUMER_UNKNOWN, | |
50 | }; | |
51 | ||
d8ef542d MD |
52 | enum consumer_channel_action { |
53 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 54 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
55 | CONSUMER_CHANNEL_QUIT, |
56 | }; | |
57 | ||
58 | struct consumer_channel_msg { | |
59 | enum consumer_channel_action action; | |
a0cbdd2e MD |
60 | struct lttng_consumer_channel *chan; /* add */ |
61 | uint64_t key; /* del */ | |
d8ef542d MD |
62 | }; |
63 | ||
3bd1e081 MD |
64 | /* |
65 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
66 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
67 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
68 | * polling threads. | |
69 | */ | |
a98dae5f | 70 | volatile int consumer_quit; |
3bd1e081 | 71 | |
43c34bc3 | 72 | /* |
43c34bc3 DG |
73 | * Global hash table containing respectively metadata and data streams. The |
74 | * stream element in this ht should only be updated by the metadata poll thread | |
75 | * for the metadata and the data poll thread for the data. | |
76 | */ | |
40dc48e0 DG |
77 | static struct lttng_ht *metadata_ht; |
78 | static struct lttng_ht *data_ht; | |
43c34bc3 | 79 | |
acdb9057 DG |
80 | /* |
81 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
82 | * global state has changed so we just send back the thread in a poll wait | |
83 | * call. | |
84 | */ | |
85 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
86 | { | |
87 | struct lttng_consumer_stream *null_stream = NULL; | |
88 | ||
89 | assert(pipe); | |
90 | ||
91 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
92 | } | |
93 | ||
d8ef542d MD |
94 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
95 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 96 | uint64_t key, |
d8ef542d MD |
97 | enum consumer_channel_action action) |
98 | { | |
99 | struct consumer_channel_msg msg; | |
100 | int ret; | |
101 | ||
e56251fc DG |
102 | memset(&msg, 0, sizeof(msg)); |
103 | ||
d8ef542d MD |
104 | msg.action = action; |
105 | msg.chan = chan; | |
f21dae48 | 106 | msg.key = key; |
d8ef542d MD |
107 | do { |
108 | ret = write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); | |
109 | } while (ret < 0 && errno == EINTR); | |
110 | } | |
111 | ||
a0cbdd2e MD |
112 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
113 | uint64_t key) | |
114 | { | |
115 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
116 | } | |
117 | ||
d8ef542d MD |
118 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
119 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 120 | uint64_t *key, |
d8ef542d MD |
121 | enum consumer_channel_action *action) |
122 | { | |
123 | struct consumer_channel_msg msg; | |
124 | int ret; | |
125 | ||
126 | do { | |
127 | ret = read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); | |
128 | } while (ret < 0 && errno == EINTR); | |
129 | if (ret > 0) { | |
130 | *action = msg.action; | |
131 | *chan = msg.chan; | |
a0cbdd2e | 132 | *key = msg.key; |
d8ef542d MD |
133 | } |
134 | return ret; | |
135 | } | |
136 | ||
3bd1e081 MD |
137 | /* |
138 | * Find a stream. The consumer_data.lock must be locked during this | |
139 | * call. | |
140 | */ | |
d88aee68 | 141 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 142 | struct lttng_ht *ht) |
3bd1e081 | 143 | { |
e4421fec | 144 | struct lttng_ht_iter iter; |
d88aee68 | 145 | struct lttng_ht_node_u64 *node; |
e4421fec | 146 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 147 | |
8389e4f8 DG |
148 | assert(ht); |
149 | ||
d88aee68 DG |
150 | /* -1ULL keys are lookup failures */ |
151 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 152 | return NULL; |
7a57cf92 | 153 | } |
e4421fec | 154 | |
6065ceec DG |
155 | rcu_read_lock(); |
156 | ||
d88aee68 DG |
157 | lttng_ht_lookup(ht, &key, &iter); |
158 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
159 | if (node != NULL) { |
160 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 161 | } |
e4421fec | 162 | |
6065ceec DG |
163 | rcu_read_unlock(); |
164 | ||
e4421fec | 165 | return stream; |
3bd1e081 MD |
166 | } |
167 | ||
da009f2c | 168 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
169 | { |
170 | struct lttng_consumer_stream *stream; | |
171 | ||
04253271 | 172 | rcu_read_lock(); |
ffe60014 | 173 | stream = find_stream(key, ht); |
04253271 | 174 | if (stream) { |
da009f2c | 175 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
176 | /* |
177 | * We don't want the lookup to match, but we still need | |
178 | * to iterate on this stream when iterating over the hash table. Just | |
179 | * change the node key. | |
180 | */ | |
da009f2c | 181 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
182 | } |
183 | rcu_read_unlock(); | |
7ad0a0cb MD |
184 | } |
185 | ||
d56db448 DG |
186 | /* |
187 | * Return a channel object for the given key. | |
188 | * | |
189 | * RCU read side lock MUST be acquired before calling this function and | |
190 | * protects the channel ptr. | |
191 | */ | |
d88aee68 | 192 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 193 | { |
e4421fec | 194 | struct lttng_ht_iter iter; |
d88aee68 | 195 | struct lttng_ht_node_u64 *node; |
e4421fec | 196 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 197 | |
d88aee68 DG |
198 | /* -1ULL keys are lookup failures */ |
199 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 200 | return NULL; |
7a57cf92 | 201 | } |
e4421fec | 202 | |
d88aee68 DG |
203 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
204 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
205 | if (node != NULL) { |
206 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 207 | } |
e4421fec DG |
208 | |
209 | return channel; | |
3bd1e081 MD |
210 | } |
211 | ||
ffe60014 | 212 | static void free_stream_rcu(struct rcu_head *head) |
7ad0a0cb | 213 | { |
d88aee68 DG |
214 | struct lttng_ht_node_u64 *node = |
215 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
216 | struct lttng_consumer_stream *stream = |
217 | caa_container_of(node, struct lttng_consumer_stream, node); | |
7ad0a0cb | 218 | |
ffe60014 | 219 | free(stream); |
7ad0a0cb MD |
220 | } |
221 | ||
ffe60014 | 222 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 223 | { |
d88aee68 DG |
224 | struct lttng_ht_node_u64 *node = |
225 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
226 | struct lttng_consumer_channel *channel = |
227 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 228 | |
ffe60014 | 229 | free(channel); |
702b1ea4 MD |
230 | } |
231 | ||
00e2e675 DG |
232 | /* |
233 | * RCU protected relayd socket pair free. | |
234 | */ | |
ffe60014 | 235 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 236 | { |
d88aee68 DG |
237 | struct lttng_ht_node_u64 *node = |
238 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
239 | struct consumer_relayd_sock_pair *relayd = |
240 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
241 | ||
8994307f DG |
242 | /* |
243 | * Close all sockets. This is done in the call RCU since we don't want the | |
244 | * socket fds to be reassigned thus potentially creating bad state of the | |
245 | * relayd object. | |
246 | * | |
247 | * We do not have to lock the control socket mutex here since at this stage | |
248 | * there is no one referencing to this relayd object. | |
249 | */ | |
250 | (void) relayd_close(&relayd->control_sock); | |
251 | (void) relayd_close(&relayd->data_sock); | |
252 | ||
00e2e675 DG |
253 | free(relayd); |
254 | } | |
255 | ||
256 | /* | |
257 | * Destroy and free relayd socket pair object. | |
00e2e675 | 258 | */ |
51230d70 | 259 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
260 | { |
261 | int ret; | |
262 | struct lttng_ht_iter iter; | |
263 | ||
173af62f DG |
264 | if (relayd == NULL) { |
265 | return; | |
266 | } | |
267 | ||
00e2e675 DG |
268 | DBG("Consumer destroy and close relayd socket pair"); |
269 | ||
270 | iter.iter.node = &relayd->node.node; | |
271 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 272 | if (ret != 0) { |
8994307f | 273 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
274 | return; |
275 | } | |
00e2e675 | 276 | |
00e2e675 | 277 | /* RCU free() call */ |
ffe60014 DG |
278 | call_rcu(&relayd->node.head, free_relayd_rcu); |
279 | } | |
280 | ||
281 | /* | |
282 | * Remove a channel from the global list protected by a mutex. This function is | |
283 | * also responsible for freeing its data structures. | |
284 | */ | |
285 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
286 | { | |
287 | int ret; | |
288 | struct lttng_ht_iter iter; | |
f2a444f1 | 289 | struct lttng_consumer_stream *stream, *stmp; |
ffe60014 | 290 | |
d88aee68 | 291 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
292 | |
293 | pthread_mutex_lock(&consumer_data.lock); | |
294 | ||
51e762e5 JD |
295 | /* Delete streams that might have been left in the stream list. */ |
296 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
297 | send_node) { | |
298 | cds_list_del(&stream->send_node); | |
299 | /* | |
300 | * Once a stream is added to this list, the buffers were created so | |
301 | * we have a guarantee that this call will succeed. | |
302 | */ | |
303 | consumer_stream_destroy(stream, NULL); | |
304 | } | |
305 | ||
ffe60014 DG |
306 | switch (consumer_data.type) { |
307 | case LTTNG_CONSUMER_KERNEL: | |
308 | break; | |
309 | case LTTNG_CONSUMER32_UST: | |
310 | case LTTNG_CONSUMER64_UST: | |
311 | lttng_ustconsumer_del_channel(channel); | |
312 | break; | |
313 | default: | |
314 | ERR("Unknown consumer_data type"); | |
315 | assert(0); | |
316 | goto end; | |
317 | } | |
318 | ||
319 | rcu_read_lock(); | |
320 | iter.iter.node = &channel->node.node; | |
321 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
322 | assert(!ret); | |
323 | rcu_read_unlock(); | |
324 | ||
325 | call_rcu(&channel->node.head, free_channel_rcu); | |
326 | end: | |
327 | pthread_mutex_unlock(&consumer_data.lock); | |
00e2e675 DG |
328 | } |
329 | ||
228b5bf7 DG |
330 | /* |
331 | * Iterate over the relayd hash table and destroy each element. Finally, | |
332 | * destroy the whole hash table. | |
333 | */ | |
334 | static void cleanup_relayd_ht(void) | |
335 | { | |
336 | struct lttng_ht_iter iter; | |
337 | struct consumer_relayd_sock_pair *relayd; | |
338 | ||
339 | rcu_read_lock(); | |
340 | ||
341 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
342 | node.node) { | |
51230d70 | 343 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
344 | } |
345 | ||
228b5bf7 | 346 | rcu_read_unlock(); |
36b588ed MD |
347 | |
348 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
349 | } |
350 | ||
8994307f DG |
351 | /* |
352 | * Update the end point status of all streams having the given network sequence | |
353 | * index (relayd index). | |
354 | * | |
355 | * It's atomically set without having the stream mutex locked which is fine | |
356 | * because we handle the write/read race with a pipe wakeup for each thread. | |
357 | */ | |
da009f2c | 358 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
359 | enum consumer_endpoint_status status) |
360 | { | |
361 | struct lttng_ht_iter iter; | |
362 | struct lttng_consumer_stream *stream; | |
363 | ||
da009f2c | 364 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
365 | |
366 | rcu_read_lock(); | |
367 | ||
368 | /* Let's begin with metadata */ | |
369 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
370 | if (stream->net_seq_idx == net_seq_idx) { | |
371 | uatomic_set(&stream->endpoint_status, status); | |
372 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
373 | } | |
374 | } | |
375 | ||
376 | /* Follow up by the data streams */ | |
377 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
378 | if (stream->net_seq_idx == net_seq_idx) { | |
379 | uatomic_set(&stream->endpoint_status, status); | |
380 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
381 | } | |
382 | } | |
383 | rcu_read_unlock(); | |
384 | } | |
385 | ||
386 | /* | |
387 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
388 | * destroying the object meaning removing it from the relayd hash table, | |
389 | * closing the sockets and freeing the memory in a RCU call. | |
390 | * | |
391 | * If a local data context is available, notify the threads that the streams' | |
392 | * state have changed. | |
393 | */ | |
394 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
395 | struct lttng_consumer_local_data *ctx) | |
396 | { | |
da009f2c | 397 | uint64_t netidx; |
8994307f DG |
398 | |
399 | assert(relayd); | |
400 | ||
9617607b DG |
401 | DBG("Cleaning up relayd sockets"); |
402 | ||
8994307f DG |
403 | /* Save the net sequence index before destroying the object */ |
404 | netidx = relayd->net_seq_idx; | |
405 | ||
406 | /* | |
407 | * Delete the relayd from the relayd hash table, close the sockets and free | |
408 | * the object in a RCU call. | |
409 | */ | |
51230d70 | 410 | consumer_destroy_relayd(relayd); |
8994307f DG |
411 | |
412 | /* Set inactive endpoint to all streams */ | |
413 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
414 | ||
415 | /* | |
416 | * With a local data context, notify the threads that the streams' state | |
417 | * have changed. The write() action on the pipe acts as an "implicit" | |
418 | * memory barrier ordering the updates of the end point status from the | |
419 | * read of this status which happens AFTER receiving this notify. | |
420 | */ | |
421 | if (ctx) { | |
acdb9057 | 422 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 423 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
424 | } |
425 | } | |
426 | ||
a6ba4fe1 DG |
427 | /* |
428 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
429 | * reaches zero. | |
430 | * | |
431 | * RCU read side lock MUST be aquired before calling this function. | |
432 | */ | |
433 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
434 | { | |
435 | assert(relayd); | |
436 | ||
437 | /* Set destroy flag for this object */ | |
438 | uatomic_set(&relayd->destroy_flag, 1); | |
439 | ||
440 | /* Destroy the relayd if refcount is 0 */ | |
441 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 442 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
443 | } |
444 | } | |
445 | ||
3bd1e081 | 446 | /* |
1d1a276c DG |
447 | * Completly destroy stream from every visiable data structure and the given |
448 | * hash table if one. | |
449 | * | |
450 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 451 | */ |
e316aad5 DG |
452 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
453 | struct lttng_ht *ht) | |
3bd1e081 | 454 | { |
1d1a276c | 455 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
456 | } |
457 | ||
d88aee68 DG |
458 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
459 | uint64_t stream_key, | |
3bd1e081 | 460 | enum lttng_consumer_stream_state state, |
ffe60014 | 461 | const char *channel_name, |
6df2e2c9 | 462 | uid_t uid, |
00e2e675 | 463 | gid_t gid, |
57a269f2 | 464 | uint64_t relayd_id, |
53632229 | 465 | uint64_t session_id, |
ffe60014 DG |
466 | int cpu, |
467 | int *alloc_ret, | |
4891ece8 DG |
468 | enum consumer_channel_type type, |
469 | unsigned int monitor) | |
3bd1e081 | 470 | { |
ffe60014 | 471 | int ret; |
3bd1e081 | 472 | struct lttng_consumer_stream *stream; |
3bd1e081 | 473 | |
effcf122 | 474 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 475 | if (stream == NULL) { |
7a57cf92 | 476 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 477 | ret = -ENOMEM; |
7a57cf92 | 478 | goto end; |
3bd1e081 | 479 | } |
7a57cf92 | 480 | |
d56db448 DG |
481 | rcu_read_lock(); |
482 | ||
3bd1e081 | 483 | stream->key = stream_key; |
3bd1e081 MD |
484 | stream->out_fd = -1; |
485 | stream->out_fd_offset = 0; | |
486 | stream->state = state; | |
6df2e2c9 MD |
487 | stream->uid = uid; |
488 | stream->gid = gid; | |
ffe60014 | 489 | stream->net_seq_idx = relayd_id; |
53632229 | 490 | stream->session_id = session_id; |
4891ece8 | 491 | stream->monitor = monitor; |
53632229 | 492 | pthread_mutex_init(&stream->lock, NULL); |
58b1f425 | 493 | |
ffe60014 DG |
494 | /* If channel is the metadata, flag this stream as metadata. */ |
495 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
496 | stream->metadata_flag = 1; | |
497 | /* Metadata is flat out. */ | |
498 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
58b1f425 | 499 | } else { |
ffe60014 DG |
500 | /* Format stream name to <channel_name>_<cpu_number> */ |
501 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
502 | channel_name, cpu); | |
503 | if (ret < 0) { | |
504 | PERROR("snprintf stream name"); | |
505 | goto error; | |
506 | } | |
58b1f425 | 507 | } |
c30aaa51 | 508 | |
ffe60014 | 509 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 510 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 511 | |
d8ef542d MD |
512 | /* Init node per channel id key */ |
513 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
514 | ||
53632229 | 515 | /* Init session id node with the stream session id */ |
d88aee68 | 516 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 517 | |
07b86b52 JD |
518 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
519 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
520 | stream->name, stream->key, channel_key, | |
521 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
522 | |
523 | rcu_read_unlock(); | |
3bd1e081 | 524 | return stream; |
c80048c6 MD |
525 | |
526 | error: | |
d56db448 | 527 | rcu_read_unlock(); |
c80048c6 | 528 | free(stream); |
7a57cf92 | 529 | end: |
ffe60014 DG |
530 | if (alloc_ret) { |
531 | *alloc_ret = ret; | |
532 | } | |
c80048c6 | 533 | return NULL; |
3bd1e081 MD |
534 | } |
535 | ||
536 | /* | |
537 | * Add a stream to the global list protected by a mutex. | |
538 | */ | |
ffe60014 | 539 | static int add_stream(struct lttng_consumer_stream *stream, |
43c34bc3 | 540 | struct lttng_ht *ht) |
3bd1e081 MD |
541 | { |
542 | int ret = 0; | |
00e2e675 | 543 | struct consumer_relayd_sock_pair *relayd; |
3bd1e081 | 544 | |
e316aad5 | 545 | assert(stream); |
43c34bc3 | 546 | assert(ht); |
c77fc10a | 547 | |
d88aee68 | 548 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
549 | |
550 | pthread_mutex_lock(&consumer_data.lock); | |
2e818a6a | 551 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 552 | rcu_read_lock(); |
e316aad5 | 553 | |
43c34bc3 | 554 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 555 | steal_stream_key(stream->key, ht); |
43c34bc3 | 556 | |
d88aee68 | 557 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 558 | |
d8ef542d MD |
559 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
560 | &stream->node_channel_id); | |
561 | ||
ca22feea DG |
562 | /* |
563 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
564 | * the key since the HT does not use it and we allow to add redundant keys | |
565 | * into this table. | |
566 | */ | |
d88aee68 | 567 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 568 | |
00e2e675 DG |
569 | /* Check and cleanup relayd */ |
570 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
571 | if (relayd != NULL) { | |
b0b335c8 | 572 | uatomic_inc(&relayd->refcount); |
00e2e675 DG |
573 | } |
574 | ||
e316aad5 | 575 | /* |
ffe60014 DG |
576 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
577 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
578 | * causes the count to become 0 also causes a stream to be added. The |
579 | * channel deletion will thus be triggered by the following removal of this | |
580 | * stream. | |
581 | */ | |
ffe60014 | 582 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
583 | /* Increment refcount before decrementing nb_init_stream_left */ |
584 | cmm_smp_wmb(); | |
ffe60014 | 585 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
586 | } |
587 | ||
588 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
589 | consumer_data.stream_count++; |
590 | consumer_data.need_update = 1; | |
591 | ||
e316aad5 | 592 | rcu_read_unlock(); |
2e818a6a | 593 | pthread_mutex_unlock(&stream->lock); |
3bd1e081 | 594 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 595 | |
3bd1e081 MD |
596 | return ret; |
597 | } | |
598 | ||
00e2e675 | 599 | /* |
3f8e211f DG |
600 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
601 | * be acquired before calling this. | |
00e2e675 | 602 | */ |
d09e1200 | 603 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
604 | { |
605 | int ret = 0; | |
d88aee68 | 606 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
607 | struct lttng_ht_iter iter; |
608 | ||
ffe60014 | 609 | assert(relayd); |
00e2e675 | 610 | |
00e2e675 | 611 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
612 | &relayd->net_seq_idx, &iter); |
613 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 614 | if (node != NULL) { |
00e2e675 DG |
615 | goto end; |
616 | } | |
d88aee68 | 617 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 618 | |
00e2e675 DG |
619 | end: |
620 | return ret; | |
621 | } | |
622 | ||
623 | /* | |
624 | * Allocate and return a consumer relayd socket. | |
625 | */ | |
626 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
da009f2c | 627 | uint64_t net_seq_idx) |
00e2e675 DG |
628 | { |
629 | struct consumer_relayd_sock_pair *obj = NULL; | |
630 | ||
da009f2c MD |
631 | /* net sequence index of -1 is a failure */ |
632 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
633 | goto error; |
634 | } | |
635 | ||
636 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
637 | if (obj == NULL) { | |
638 | PERROR("zmalloc relayd sock"); | |
639 | goto error; | |
640 | } | |
641 | ||
642 | obj->net_seq_idx = net_seq_idx; | |
643 | obj->refcount = 0; | |
173af62f | 644 | obj->destroy_flag = 0; |
f96e4545 MD |
645 | obj->control_sock.sock.fd = -1; |
646 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 647 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
648 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
649 | ||
650 | error: | |
651 | return obj; | |
652 | } | |
653 | ||
654 | /* | |
655 | * Find a relayd socket pair in the global consumer data. | |
656 | * | |
657 | * Return the object if found else NULL. | |
b0b335c8 MD |
658 | * RCU read-side lock must be held across this call and while using the |
659 | * returned object. | |
00e2e675 | 660 | */ |
d88aee68 | 661 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
662 | { |
663 | struct lttng_ht_iter iter; | |
d88aee68 | 664 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
665 | struct consumer_relayd_sock_pair *relayd = NULL; |
666 | ||
667 | /* Negative keys are lookup failures */ | |
d88aee68 | 668 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
669 | goto error; |
670 | } | |
671 | ||
d88aee68 | 672 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 673 | &iter); |
d88aee68 | 674 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
675 | if (node != NULL) { |
676 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
677 | } | |
678 | ||
00e2e675 DG |
679 | error: |
680 | return relayd; | |
681 | } | |
682 | ||
10a50311 JD |
683 | /* |
684 | * Find a relayd and send the stream | |
685 | * | |
686 | * Returns 0 on success, < 0 on error | |
687 | */ | |
688 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
689 | char *path) | |
690 | { | |
691 | int ret = 0; | |
692 | struct consumer_relayd_sock_pair *relayd; | |
693 | ||
694 | assert(stream); | |
695 | assert(stream->net_seq_idx != -1ULL); | |
696 | assert(path); | |
697 | ||
698 | /* The stream is not metadata. Get relayd reference if exists. */ | |
699 | rcu_read_lock(); | |
700 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
701 | if (relayd != NULL) { | |
702 | /* Add stream on the relayd */ | |
703 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
704 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
705 | path, &stream->relayd_stream_id, | |
706 | stream->chan->tracefile_size, stream->chan->tracefile_count); | |
707 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
708 | if (ret < 0) { | |
709 | goto end; | |
710 | } | |
711 | uatomic_inc(&relayd->refcount); | |
712 | } else { | |
713 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
714 | stream->key, stream->net_seq_idx); | |
715 | ret = -1; | |
716 | goto end; | |
717 | } | |
718 | ||
719 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
720 | stream->name, stream->key, stream->net_seq_idx); | |
721 | ||
722 | end: | |
723 | rcu_read_unlock(); | |
724 | return ret; | |
725 | } | |
726 | ||
727 | /* | |
728 | * Find a relayd and close the stream | |
729 | */ | |
730 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
731 | { | |
732 | struct consumer_relayd_sock_pair *relayd; | |
733 | ||
734 | /* The stream is not metadata. Get relayd reference if exists. */ | |
735 | rcu_read_lock(); | |
736 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
737 | if (relayd) { | |
738 | consumer_stream_relayd_close(stream, relayd); | |
739 | } | |
740 | rcu_read_unlock(); | |
741 | } | |
742 | ||
00e2e675 DG |
743 | /* |
744 | * Handle stream for relayd transmission if the stream applies for network | |
745 | * streaming where the net sequence index is set. | |
746 | * | |
747 | * Return destination file descriptor or negative value on error. | |
748 | */ | |
6197aea7 | 749 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
750 | size_t data_size, unsigned long padding, |
751 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
752 | { |
753 | int outfd = -1, ret; | |
00e2e675 DG |
754 | struct lttcomm_relayd_data_hdr data_hdr; |
755 | ||
756 | /* Safety net */ | |
757 | assert(stream); | |
6197aea7 | 758 | assert(relayd); |
00e2e675 DG |
759 | |
760 | /* Reset data header */ | |
761 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
762 | ||
00e2e675 DG |
763 | if (stream->metadata_flag) { |
764 | /* Caller MUST acquire the relayd control socket lock */ | |
765 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
766 | if (ret < 0) { | |
767 | goto error; | |
768 | } | |
769 | ||
770 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 771 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
772 | } else { |
773 | /* Set header with stream information */ | |
774 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
775 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 776 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
777 | /* |
778 | * Note that net_seq_num below is assigned with the *current* value of | |
779 | * next_net_seq_num and only after that the next_net_seq_num will be | |
780 | * increment. This is why when issuing a command on the relayd using | |
781 | * this next value, 1 should always be substracted in order to compare | |
782 | * the last seen sequence number on the relayd side to the last sent. | |
783 | */ | |
3604f373 | 784 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
785 | /* Other fields are zeroed previously */ |
786 | ||
787 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
788 | sizeof(data_hdr)); | |
789 | if (ret < 0) { | |
790 | goto error; | |
791 | } | |
792 | ||
3604f373 DG |
793 | ++stream->next_net_seq_num; |
794 | ||
00e2e675 | 795 | /* Set to go on data socket */ |
6151a90f | 796 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
797 | } |
798 | ||
799 | error: | |
800 | return outfd; | |
801 | } | |
802 | ||
3bd1e081 | 803 | /* |
ffe60014 DG |
804 | * Allocate and return a new lttng_consumer_channel object using the given key |
805 | * to initialize the hash table node. | |
806 | * | |
807 | * On error, return NULL. | |
3bd1e081 | 808 | */ |
886224ff | 809 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
810 | uint64_t session_id, |
811 | const char *pathname, | |
812 | const char *name, | |
813 | uid_t uid, | |
814 | gid_t gid, | |
57a269f2 | 815 | uint64_t relayd_id, |
1624d5b7 JD |
816 | enum lttng_event_output output, |
817 | uint64_t tracefile_size, | |
2bba9e53 | 818 | uint64_t tracefile_count, |
1950109e | 819 | uint64_t session_id_per_pid, |
2bba9e53 | 820 | unsigned int monitor) |
3bd1e081 MD |
821 | { |
822 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 823 | |
276b26d1 | 824 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 825 | if (channel == NULL) { |
7a57cf92 | 826 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
827 | goto end; |
828 | } | |
ffe60014 DG |
829 | |
830 | channel->key = key; | |
3bd1e081 | 831 | channel->refcount = 0; |
ffe60014 | 832 | channel->session_id = session_id; |
1950109e | 833 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
834 | channel->uid = uid; |
835 | channel->gid = gid; | |
836 | channel->relayd_id = relayd_id; | |
837 | channel->output = output; | |
1624d5b7 JD |
838 | channel->tracefile_size = tracefile_size; |
839 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 840 | channel->monitor = monitor; |
ffe60014 | 841 | |
07b86b52 JD |
842 | /* |
843 | * In monitor mode, the streams associated with the channel will be put in | |
844 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
845 | * here meaning that the channel itself has streams that are referenced. | |
846 | * | |
847 | * On a channel deletion, once the channel is no longer visible, the | |
848 | * refcount is decremented and checked for a zero value to delete it. With | |
849 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
850 | */ | |
851 | if (!channel->monitor) { | |
852 | channel->refcount = 1; | |
853 | } | |
854 | ||
ffe60014 DG |
855 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
856 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
857 | ||
858 | strncpy(channel->name, name, sizeof(channel->name)); | |
859 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
860 | ||
d88aee68 | 861 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
862 | |
863 | channel->wait_fd = -1; | |
864 | ||
ffe60014 DG |
865 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
866 | ||
d88aee68 | 867 | DBG("Allocated channel (key %" PRIu64 ")", channel->key) |
3bd1e081 | 868 | |
3bd1e081 MD |
869 | end: |
870 | return channel; | |
871 | } | |
872 | ||
873 | /* | |
874 | * Add a channel to the global list protected by a mutex. | |
821fffb2 DG |
875 | * |
876 | * On success 0 is returned else a negative value. | |
3bd1e081 | 877 | */ |
d8ef542d MD |
878 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
879 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 880 | { |
ffe60014 | 881 | int ret = 0; |
d88aee68 | 882 | struct lttng_ht_node_u64 *node; |
c77fc10a DG |
883 | struct lttng_ht_iter iter; |
884 | ||
3bd1e081 | 885 | pthread_mutex_lock(&consumer_data.lock); |
6065ceec | 886 | rcu_read_lock(); |
c77fc10a | 887 | |
7972aab2 | 888 | lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter); |
d88aee68 | 889 | node = lttng_ht_iter_get_node_u64(&iter); |
c77fc10a DG |
890 | if (node != NULL) { |
891 | /* Channel already exist. Ignore the insertion */ | |
d88aee68 DG |
892 | ERR("Consumer add channel key %" PRIu64 " already exists!", |
893 | channel->key); | |
821fffb2 | 894 | ret = -EEXIST; |
c77fc10a DG |
895 | goto end; |
896 | } | |
897 | ||
d88aee68 | 898 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
899 | |
900 | end: | |
6065ceec | 901 | rcu_read_unlock(); |
3bd1e081 | 902 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 903 | |
d8ef542d | 904 | if (!ret && channel->wait_fd != -1 && |
10a50311 | 905 | channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 906 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 907 | } |
ffe60014 | 908 | return ret; |
3bd1e081 MD |
909 | } |
910 | ||
911 | /* | |
912 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
913 | * doing a lookup in the linked list and concurrency issues when writing is | |
914 | * needed. Called with consumer_data.lock held. | |
915 | * | |
916 | * Returns the number of fds in the structures. | |
917 | */ | |
ffe60014 DG |
918 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
919 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
920 | struct lttng_ht *ht) | |
3bd1e081 | 921 | { |
3bd1e081 | 922 | int i = 0; |
e4421fec DG |
923 | struct lttng_ht_iter iter; |
924 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 925 | |
ffe60014 DG |
926 | assert(ctx); |
927 | assert(ht); | |
928 | assert(pollfd); | |
929 | assert(local_stream); | |
930 | ||
3bd1e081 | 931 | DBG("Updating poll fd array"); |
481d6c57 | 932 | rcu_read_lock(); |
43c34bc3 | 933 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
934 | /* |
935 | * Only active streams with an active end point can be added to the | |
936 | * poll set and local stream storage of the thread. | |
937 | * | |
938 | * There is a potential race here for endpoint_status to be updated | |
939 | * just after the check. However, this is OK since the stream(s) will | |
940 | * be deleted once the thread is notified that the end point state has | |
941 | * changed where this function will be called back again. | |
942 | */ | |
943 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 944 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
945 | continue; |
946 | } | |
7972aab2 DG |
947 | /* |
948 | * This clobbers way too much the debug output. Uncomment that if you | |
949 | * need it for debugging purposes. | |
950 | * | |
951 | * DBG("Active FD %d", stream->wait_fd); | |
952 | */ | |
e4421fec | 953 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 954 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 955 | local_stream[i] = stream; |
3bd1e081 MD |
956 | i++; |
957 | } | |
481d6c57 | 958 | rcu_read_unlock(); |
3bd1e081 MD |
959 | |
960 | /* | |
50f8ae69 | 961 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
962 | * increment i so nb_fd is the number of real FD. |
963 | */ | |
acdb9057 | 964 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 965 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
966 | return i; |
967 | } | |
968 | ||
969 | /* | |
970 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
971 | * should exit, 0 if data is available on the command socket | |
972 | */ | |
973 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
974 | { | |
975 | int num_rdy; | |
976 | ||
88f2b785 | 977 | restart: |
3bd1e081 MD |
978 | num_rdy = poll(consumer_sockpoll, 2, -1); |
979 | if (num_rdy == -1) { | |
88f2b785 MD |
980 | /* |
981 | * Restart interrupted system call. | |
982 | */ | |
983 | if (errno == EINTR) { | |
984 | goto restart; | |
985 | } | |
7a57cf92 | 986 | PERROR("Poll error"); |
3bd1e081 MD |
987 | goto exit; |
988 | } | |
509bb1cf | 989 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
990 | DBG("consumer_should_quit wake up"); |
991 | goto exit; | |
992 | } | |
993 | return 0; | |
994 | ||
995 | exit: | |
996 | return -1; | |
997 | } | |
998 | ||
999 | /* | |
1000 | * Set the error socket. | |
1001 | */ | |
ffe60014 DG |
1002 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1003 | int sock) | |
3bd1e081 MD |
1004 | { |
1005 | ctx->consumer_error_socket = sock; | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * Set the command socket path. | |
1010 | */ | |
3bd1e081 MD |
1011 | void lttng_consumer_set_command_sock_path( |
1012 | struct lttng_consumer_local_data *ctx, char *sock) | |
1013 | { | |
1014 | ctx->consumer_command_sock_path = sock; | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * Send return code to the session daemon. | |
1019 | * If the socket is not defined, we return 0, it is not a fatal error | |
1020 | */ | |
ffe60014 | 1021 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1022 | { |
1023 | if (ctx->consumer_error_socket > 0) { | |
1024 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1025 | sizeof(enum lttcomm_sessiond_command)); | |
1026 | } | |
1027 | ||
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | /* | |
228b5bf7 DG |
1032 | * Close all the tracefiles and stream fds and MUST be called when all |
1033 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1034 | */ |
1035 | void lttng_consumer_cleanup(void) | |
1036 | { | |
e4421fec | 1037 | struct lttng_ht_iter iter; |
ffe60014 | 1038 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1039 | |
1040 | rcu_read_lock(); | |
3bd1e081 | 1041 | |
ffe60014 DG |
1042 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1043 | node.node) { | |
702b1ea4 | 1044 | consumer_del_channel(channel); |
3bd1e081 | 1045 | } |
6065ceec DG |
1046 | |
1047 | rcu_read_unlock(); | |
d6ce1df2 | 1048 | |
d6ce1df2 | 1049 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1050 | |
1051 | cleanup_relayd_ht(); | |
1052 | ||
d8ef542d MD |
1053 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1054 | ||
228b5bf7 DG |
1055 | /* |
1056 | * This HT contains streams that are freed by either the metadata thread or | |
1057 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1058 | * it. | |
1059 | */ | |
1060 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1061 | } |
1062 | ||
1063 | /* | |
1064 | * Called from signal handler. | |
1065 | */ | |
1066 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1067 | { | |
1068 | int ret; | |
1069 | consumer_quit = 1; | |
6f94560a MD |
1070 | do { |
1071 | ret = write(ctx->consumer_should_quit[1], "4", 1); | |
1072 | } while (ret < 0 && errno == EINTR); | |
4cec016f | 1073 | if (ret < 0 || ret != 1) { |
7a57cf92 | 1074 | PERROR("write consumer quit"); |
3bd1e081 | 1075 | } |
ab1027f4 DG |
1076 | |
1077 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1078 | } |
1079 | ||
00e2e675 DG |
1080 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1081 | off_t orig_offset) | |
3bd1e081 MD |
1082 | { |
1083 | int outfd = stream->out_fd; | |
1084 | ||
1085 | /* | |
1086 | * This does a blocking write-and-wait on any page that belongs to the | |
1087 | * subbuffer prior to the one we just wrote. | |
1088 | * Don't care about error values, as these are just hints and ways to | |
1089 | * limit the amount of page cache used. | |
1090 | */ | |
ffe60014 | 1091 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1092 | return; |
1093 | } | |
ffe60014 DG |
1094 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1095 | stream->max_sb_size, | |
3bd1e081 MD |
1096 | SYNC_FILE_RANGE_WAIT_BEFORE |
1097 | | SYNC_FILE_RANGE_WRITE | |
1098 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1099 | /* | |
1100 | * Give hints to the kernel about how we access the file: | |
1101 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1102 | * we write it. | |
1103 | * | |
1104 | * We need to call fadvise again after the file grows because the | |
1105 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1106 | * file. | |
1107 | * | |
1108 | * Call fadvise _after_ having waited for the page writeback to | |
1109 | * complete because the dirty page writeback semantic is not well | |
1110 | * defined. So it can be expected to lead to lower throughput in | |
1111 | * streaming. | |
1112 | */ | |
ffe60014 DG |
1113 | posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
1114 | stream->max_sb_size, POSIX_FADV_DONTNEED); | |
3bd1e081 MD |
1115 | } |
1116 | ||
1117 | /* | |
1118 | * Initialise the necessary environnement : | |
1119 | * - create a new context | |
1120 | * - create the poll_pipe | |
1121 | * - create the should_quit pipe (for signal handler) | |
1122 | * - create the thread pipe (for splice) | |
1123 | * | |
1124 | * Takes a function pointer as argument, this function is called when data is | |
1125 | * available on a buffer. This function is responsible to do the | |
1126 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1127 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1128 | * | |
1129 | * Returns a pointer to the new context or NULL on error. | |
1130 | */ | |
1131 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1132 | enum lttng_consumer_type type, | |
4078b776 | 1133 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1134 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1135 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1136 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
1137 | int (*update_stream)(int stream_key, uint32_t state)) | |
1138 | { | |
d8ef542d | 1139 | int ret; |
3bd1e081 MD |
1140 | struct lttng_consumer_local_data *ctx; |
1141 | ||
1142 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1143 | consumer_data.type == type); | |
1144 | consumer_data.type = type; | |
1145 | ||
effcf122 | 1146 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1147 | if (ctx == NULL) { |
7a57cf92 | 1148 | PERROR("allocating context"); |
3bd1e081 MD |
1149 | goto error; |
1150 | } | |
1151 | ||
1152 | ctx->consumer_error_socket = -1; | |
331744e3 | 1153 | ctx->consumer_metadata_socket = -1; |
3bd1e081 MD |
1154 | /* assign the callbacks */ |
1155 | ctx->on_buffer_ready = buffer_ready; | |
1156 | ctx->on_recv_channel = recv_channel; | |
1157 | ctx->on_recv_stream = recv_stream; | |
1158 | ctx->on_update_stream = update_stream; | |
1159 | ||
acdb9057 DG |
1160 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1161 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1162 | goto error_poll_pipe; |
1163 | } | |
1164 | ||
1165 | ret = pipe(ctx->consumer_should_quit); | |
1166 | if (ret < 0) { | |
7a57cf92 | 1167 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1168 | goto error_quit_pipe; |
1169 | } | |
1170 | ||
1171 | ret = pipe(ctx->consumer_thread_pipe); | |
1172 | if (ret < 0) { | |
7a57cf92 | 1173 | PERROR("Error creating thread pipe"); |
3bd1e081 MD |
1174 | goto error_thread_pipe; |
1175 | } | |
1176 | ||
d8ef542d MD |
1177 | ret = pipe(ctx->consumer_channel_pipe); |
1178 | if (ret < 0) { | |
1179 | PERROR("Error creating channel pipe"); | |
1180 | goto error_channel_pipe; | |
1181 | } | |
1182 | ||
13886d2d DG |
1183 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1184 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1185 | goto error_metadata_pipe; |
1186 | } | |
3bd1e081 | 1187 | |
fb3a43a9 DG |
1188 | ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe); |
1189 | if (ret < 0) { | |
1190 | goto error_splice_pipe; | |
1191 | } | |
1192 | ||
1193 | return ctx; | |
3bd1e081 | 1194 | |
fb3a43a9 | 1195 | error_splice_pipe: |
13886d2d | 1196 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
fb3a43a9 | 1197 | error_metadata_pipe: |
d8ef542d MD |
1198 | utils_close_pipe(ctx->consumer_channel_pipe); |
1199 | error_channel_pipe: | |
fb3a43a9 | 1200 | utils_close_pipe(ctx->consumer_thread_pipe); |
3bd1e081 | 1201 | error_thread_pipe: |
d8ef542d | 1202 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1203 | error_quit_pipe: |
acdb9057 | 1204 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1205 | error_poll_pipe: |
1206 | free(ctx); | |
1207 | error: | |
1208 | return NULL; | |
1209 | } | |
1210 | ||
1211 | /* | |
1212 | * Close all fds associated with the instance and free the context. | |
1213 | */ | |
1214 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1215 | { | |
4c462e79 MD |
1216 | int ret; |
1217 | ||
ab1027f4 DG |
1218 | DBG("Consumer destroying it. Closing everything."); |
1219 | ||
4c462e79 MD |
1220 | ret = close(ctx->consumer_error_socket); |
1221 | if (ret) { | |
1222 | PERROR("close"); | |
1223 | } | |
331744e3 JD |
1224 | ret = close(ctx->consumer_metadata_socket); |
1225 | if (ret) { | |
1226 | PERROR("close"); | |
1227 | } | |
d8ef542d MD |
1228 | utils_close_pipe(ctx->consumer_thread_pipe); |
1229 | utils_close_pipe(ctx->consumer_channel_pipe); | |
acdb9057 | 1230 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1231 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
d8ef542d | 1232 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 DG |
1233 | utils_close_pipe(ctx->consumer_splice_metadata_pipe); |
1234 | ||
3bd1e081 MD |
1235 | unlink(ctx->consumer_command_sock_path); |
1236 | free(ctx); | |
1237 | } | |
1238 | ||
6197aea7 DG |
1239 | /* |
1240 | * Write the metadata stream id on the specified file descriptor. | |
1241 | */ | |
1242 | static int write_relayd_metadata_id(int fd, | |
1243 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1244 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 DG |
1245 | { |
1246 | int ret; | |
1d4dfdef | 1247 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1248 | |
1d4dfdef DG |
1249 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1250 | hdr.padding_size = htobe32(padding); | |
6197aea7 | 1251 | do { |
1d4dfdef | 1252 | ret = write(fd, (void *) &hdr, sizeof(hdr)); |
6197aea7 | 1253 | } while (ret < 0 && errno == EINTR); |
4cec016f | 1254 | if (ret < 0 || ret != sizeof(hdr)) { |
d7b75ec8 DG |
1255 | /* |
1256 | * This error means that the fd's end is closed so ignore the perror | |
1257 | * not to clubber the error output since this can happen in a normal | |
1258 | * code path. | |
1259 | */ | |
1260 | if (errno != EPIPE) { | |
1261 | PERROR("write metadata stream id"); | |
1262 | } | |
1263 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1264 | /* |
1265 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1266 | * handle writting the missing part so report that as an error and | |
1267 | * don't lie to the caller. | |
1268 | */ | |
1269 | ret = -1; | |
6197aea7 DG |
1270 | goto end; |
1271 | } | |
1d4dfdef DG |
1272 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1273 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1274 | |
1275 | end: | |
1276 | return ret; | |
1277 | } | |
1278 | ||
3bd1e081 | 1279 | /* |
09e26845 DG |
1280 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1281 | * core function for writing trace buffers to either the local filesystem or | |
1282 | * the network. | |
1283 | * | |
79d4ffb7 DG |
1284 | * It must be called with the stream lock held. |
1285 | * | |
09e26845 | 1286 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1287 | * |
1288 | * Returns the number of bytes written | |
1289 | */ | |
4078b776 | 1290 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1291 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1292 | struct lttng_consumer_stream *stream, unsigned long len, |
1293 | unsigned long padding) | |
3bd1e081 | 1294 | { |
f02e1e8a | 1295 | unsigned long mmap_offset; |
ffe60014 | 1296 | void *mmap_base; |
f02e1e8a DG |
1297 | ssize_t ret = 0, written = 0; |
1298 | off_t orig_offset = stream->out_fd_offset; | |
1299 | /* Default is on the disk */ | |
1300 | int outfd = stream->out_fd; | |
f02e1e8a | 1301 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1302 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1303 | |
1304 | /* RCU lock for the relayd pointer */ | |
1305 | rcu_read_lock(); | |
1306 | ||
1307 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1308 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1309 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1310 | if (relayd == NULL) { | |
1311 | goto end; | |
1312 | } | |
1313 | } | |
1314 | ||
1315 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1316 | switch (consumer_data.type) { |
1317 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1318 | mmap_base = stream->mmap_base; |
f02e1e8a DG |
1319 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
1320 | break; | |
7753dea8 MD |
1321 | case LTTNG_CONSUMER32_UST: |
1322 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1323 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1324 | if (!mmap_base) { | |
1325 | ERR("read mmap get mmap base for stream %s", stream->name); | |
1326 | written = -1; | |
1327 | goto end; | |
1328 | } | |
1329 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
331744e3 | 1330 | |
f02e1e8a | 1331 | break; |
3bd1e081 MD |
1332 | default: |
1333 | ERR("Unknown consumer_data type"); | |
1334 | assert(0); | |
1335 | } | |
f02e1e8a DG |
1336 | if (ret != 0) { |
1337 | errno = -ret; | |
1338 | PERROR("tracer ctl get_mmap_read_offset"); | |
1339 | written = ret; | |
1340 | goto end; | |
1341 | } | |
b9182dd9 | 1342 | |
f02e1e8a DG |
1343 | /* Handle stream on the relayd if the output is on the network */ |
1344 | if (relayd) { | |
1345 | unsigned long netlen = len; | |
1346 | ||
1347 | /* | |
1348 | * Lock the control socket for the complete duration of the function | |
1349 | * since from this point on we will use the socket. | |
1350 | */ | |
1351 | if (stream->metadata_flag) { | |
1352 | /* Metadata requires the control socket. */ | |
1353 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1354 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1355 | } |
1356 | ||
1d4dfdef | 1357 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
f02e1e8a DG |
1358 | if (ret >= 0) { |
1359 | /* Use the returned socket. */ | |
1360 | outfd = ret; | |
1361 | ||
1362 | /* Write metadata stream id before payload */ | |
1363 | if (stream->metadata_flag) { | |
1d4dfdef | 1364 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); |
f02e1e8a | 1365 | if (ret < 0) { |
f02e1e8a | 1366 | written = ret; |
8994307f DG |
1367 | /* Socket operation failed. We consider the relayd dead */ |
1368 | if (ret == -EPIPE || ret == -EINVAL) { | |
1369 | relayd_hang_up = 1; | |
1370 | goto write_error; | |
1371 | } | |
f02e1e8a DG |
1372 | goto end; |
1373 | } | |
f02e1e8a | 1374 | } |
8994307f DG |
1375 | } else { |
1376 | /* Socket operation failed. We consider the relayd dead */ | |
1377 | if (ret == -EPIPE || ret == -EINVAL) { | |
1378 | relayd_hang_up = 1; | |
1379 | goto write_error; | |
1380 | } | |
1381 | /* Else, use the default set before which is the filesystem. */ | |
f02e1e8a | 1382 | } |
1d4dfdef DG |
1383 | } else { |
1384 | /* No streaming, we have to set the len with the full padding */ | |
1385 | len += padding; | |
1624d5b7 JD |
1386 | |
1387 | /* | |
1388 | * Check if we need to change the tracefile before writing the packet. | |
1389 | */ | |
1390 | if (stream->chan->tracefile_size > 0 && | |
1391 | (stream->tracefile_size_current + len) > | |
1392 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1393 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1394 | stream->name, stream->chan->tracefile_size, | |
1395 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
1396 | stream->out_fd, &(stream->tracefile_count_current)); | |
1624d5b7 JD |
1397 | if (ret < 0) { |
1398 | ERR("Rotating output file"); | |
1399 | goto end; | |
1400 | } | |
fe4477ee | 1401 | outfd = stream->out_fd = ret; |
a6976990 DG |
1402 | /* Reset current size because we just perform a rotation. */ |
1403 | stream->tracefile_size_current = 0; | |
1624d5b7 JD |
1404 | } |
1405 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1406 | } |
1407 | ||
1408 | while (len > 0) { | |
1409 | do { | |
ffe60014 | 1410 | ret = write(outfd, mmap_base + mmap_offset, len); |
f02e1e8a | 1411 | } while (ret < 0 && errno == EINTR); |
1d4dfdef | 1412 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); |
f02e1e8a | 1413 | if (ret < 0) { |
c5c45efa DG |
1414 | /* |
1415 | * This is possible if the fd is closed on the other side (outfd) | |
1416 | * or any write problem. It can be verbose a bit for a normal | |
1417 | * execution if for instance the relayd is stopped abruptly. This | |
1418 | * can happen so set this to a DBG statement. | |
1419 | */ | |
1420 | DBG("Error in file write mmap"); | |
f02e1e8a DG |
1421 | if (written == 0) { |
1422 | written = ret; | |
1423 | } | |
8994307f DG |
1424 | /* Socket operation failed. We consider the relayd dead */ |
1425 | if (errno == EPIPE || errno == EINVAL) { | |
1426 | relayd_hang_up = 1; | |
1427 | goto write_error; | |
1428 | } | |
f02e1e8a DG |
1429 | goto end; |
1430 | } else if (ret > len) { | |
77c7c900 | 1431 | PERROR("Error in file write (ret %zd > len %lu)", ret, len); |
f02e1e8a DG |
1432 | written += ret; |
1433 | goto end; | |
1434 | } else { | |
1435 | len -= ret; | |
1436 | mmap_offset += ret; | |
1437 | } | |
f02e1e8a DG |
1438 | |
1439 | /* This call is useless on a socket so better save a syscall. */ | |
1440 | if (!relayd) { | |
1441 | /* This won't block, but will start writeout asynchronously */ | |
1442 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret, | |
1443 | SYNC_FILE_RANGE_WRITE); | |
1444 | stream->out_fd_offset += ret; | |
1445 | } | |
1446 | written += ret; | |
1447 | } | |
1448 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1449 | ||
8994307f DG |
1450 | write_error: |
1451 | /* | |
1452 | * This is a special case that the relayd has closed its socket. Let's | |
1453 | * cleanup the relayd object and all associated streams. | |
1454 | */ | |
1455 | if (relayd && relayd_hang_up) { | |
1456 | cleanup_relayd(relayd, ctx); | |
1457 | } | |
1458 | ||
f02e1e8a DG |
1459 | end: |
1460 | /* Unlock only if ctrl socket used */ | |
1461 | if (relayd && stream->metadata_flag) { | |
1462 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1463 | } | |
1464 | ||
1465 | rcu_read_unlock(); | |
1466 | return written; | |
3bd1e081 MD |
1467 | } |
1468 | ||
1469 | /* | |
1470 | * Splice the data from the ring buffer to the tracefile. | |
1471 | * | |
79d4ffb7 DG |
1472 | * It must be called with the stream lock held. |
1473 | * | |
3bd1e081 MD |
1474 | * Returns the number of bytes spliced. |
1475 | */ | |
4078b776 | 1476 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1477 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1478 | struct lttng_consumer_stream *stream, unsigned long len, |
1479 | unsigned long padding) | |
3bd1e081 | 1480 | { |
f02e1e8a DG |
1481 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1482 | loff_t offset = 0; | |
1483 | off_t orig_offset = stream->out_fd_offset; | |
1484 | int fd = stream->wait_fd; | |
1485 | /* Default is on the disk */ | |
1486 | int outfd = stream->out_fd; | |
f02e1e8a | 1487 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1488 | int *splice_pipe; |
8994307f | 1489 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1490 | |
3bd1e081 MD |
1491 | switch (consumer_data.type) { |
1492 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1493 | break; |
7753dea8 MD |
1494 | case LTTNG_CONSUMER32_UST: |
1495 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1496 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1497 | return -ENOSYS; |
1498 | default: | |
1499 | ERR("Unknown consumer_data type"); | |
1500 | assert(0); | |
3bd1e081 MD |
1501 | } |
1502 | ||
f02e1e8a DG |
1503 | /* RCU lock for the relayd pointer */ |
1504 | rcu_read_lock(); | |
1505 | ||
1506 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1507 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1508 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1509 | if (relayd == NULL) { | |
1510 | goto end; | |
1511 | } | |
1512 | } | |
1513 | ||
fb3a43a9 DG |
1514 | /* |
1515 | * Choose right pipe for splice. Metadata and trace data are handled by | |
1516 | * different threads hence the use of two pipes in order not to race or | |
1517 | * corrupt the written data. | |
1518 | */ | |
1519 | if (stream->metadata_flag) { | |
1520 | splice_pipe = ctx->consumer_splice_metadata_pipe; | |
1521 | } else { | |
1522 | splice_pipe = ctx->consumer_thread_pipe; | |
1523 | } | |
1524 | ||
f02e1e8a | 1525 | /* Write metadata stream id before payload */ |
1d4dfdef DG |
1526 | if (relayd) { |
1527 | int total_len = len; | |
f02e1e8a | 1528 | |
1d4dfdef DG |
1529 | if (stream->metadata_flag) { |
1530 | /* | |
1531 | * Lock the control socket for the complete duration of the function | |
1532 | * since from this point on we will use the socket. | |
1533 | */ | |
1534 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1535 | ||
1536 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1537 | padding); | |
1538 | if (ret < 0) { | |
1539 | written = ret; | |
8994307f DG |
1540 | /* Socket operation failed. We consider the relayd dead */ |
1541 | if (ret == -EBADF) { | |
1542 | WARN("Remote relayd disconnected. Stopping"); | |
1543 | relayd_hang_up = 1; | |
1544 | goto write_error; | |
1545 | } | |
1d4dfdef DG |
1546 | goto end; |
1547 | } | |
1548 | ||
1549 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1550 | } | |
1551 | ||
1552 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
1553 | if (ret >= 0) { | |
1554 | /* Use the returned socket. */ | |
1555 | outfd = ret; | |
1556 | } else { | |
8994307f DG |
1557 | /* Socket operation failed. We consider the relayd dead */ |
1558 | if (ret == -EBADF) { | |
1559 | WARN("Remote relayd disconnected. Stopping"); | |
1560 | relayd_hang_up = 1; | |
1561 | goto write_error; | |
1562 | } | |
f02e1e8a DG |
1563 | goto end; |
1564 | } | |
1d4dfdef DG |
1565 | } else { |
1566 | /* No streaming, we have to set the len with the full padding */ | |
1567 | len += padding; | |
1624d5b7 JD |
1568 | |
1569 | /* | |
1570 | * Check if we need to change the tracefile before writing the packet. | |
1571 | */ | |
1572 | if (stream->chan->tracefile_size > 0 && | |
1573 | (stream->tracefile_size_current + len) > | |
1574 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1575 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1576 | stream->name, stream->chan->tracefile_size, | |
1577 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
1578 | stream->out_fd, &(stream->tracefile_count_current)); | |
1624d5b7 JD |
1579 | if (ret < 0) { |
1580 | ERR("Rotating output file"); | |
1581 | goto end; | |
1582 | } | |
fe4477ee | 1583 | outfd = stream->out_fd = ret; |
a6976990 DG |
1584 | /* Reset current size because we just perform a rotation. */ |
1585 | stream->tracefile_size_current = 0; | |
1624d5b7 JD |
1586 | } |
1587 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1588 | } |
1589 | ||
1590 | while (len > 0) { | |
1d4dfdef DG |
1591 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1592 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1593 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1594 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1595 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1596 | if (ret_splice < 0) { | |
1597 | PERROR("Error in relay splice"); | |
1598 | if (written == 0) { | |
1599 | written = ret_splice; | |
1600 | } | |
1601 | ret = errno; | |
1602 | goto splice_error; | |
1603 | } | |
1604 | ||
1605 | /* Handle stream on the relayd if the output is on the network */ | |
1606 | if (relayd) { | |
1607 | if (stream->metadata_flag) { | |
1d4dfdef DG |
1608 | size_t metadata_payload_size = |
1609 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1610 | ||
f02e1e8a | 1611 | /* Update counter to fit the spliced data */ |
1d4dfdef DG |
1612 | ret_splice += metadata_payload_size; |
1613 | len += metadata_payload_size; | |
f02e1e8a DG |
1614 | /* |
1615 | * We do this so the return value can match the len passed as | |
1616 | * argument to this function. | |
1617 | */ | |
1d4dfdef | 1618 | written -= metadata_payload_size; |
f02e1e8a DG |
1619 | } |
1620 | } | |
1621 | ||
1622 | /* Splice data out */ | |
fb3a43a9 | 1623 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1624 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
1d4dfdef | 1625 | DBG("Consumer splice pipe to file, ret %zd", ret_splice); |
f02e1e8a DG |
1626 | if (ret_splice < 0) { |
1627 | PERROR("Error in file splice"); | |
1628 | if (written == 0) { | |
1629 | written = ret_splice; | |
1630 | } | |
8994307f | 1631 | /* Socket operation failed. We consider the relayd dead */ |
00c8752b | 1632 | if (errno == EBADF || errno == EPIPE) { |
8994307f DG |
1633 | WARN("Remote relayd disconnected. Stopping"); |
1634 | relayd_hang_up = 1; | |
1635 | goto write_error; | |
1636 | } | |
f02e1e8a DG |
1637 | ret = errno; |
1638 | goto splice_error; | |
1639 | } else if (ret_splice > len) { | |
1640 | errno = EINVAL; | |
1641 | PERROR("Wrote more data than requested %zd (len: %lu)", | |
1642 | ret_splice, len); | |
1643 | written += ret_splice; | |
1644 | ret = errno; | |
1645 | goto splice_error; | |
1646 | } | |
1647 | len -= ret_splice; | |
1648 | ||
1649 | /* This call is useless on a socket so better save a syscall. */ | |
1650 | if (!relayd) { | |
1651 | /* This won't block, but will start writeout asynchronously */ | |
1652 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1653 | SYNC_FILE_RANGE_WRITE); | |
1654 | stream->out_fd_offset += ret_splice; | |
1655 | } | |
1656 | written += ret_splice; | |
1657 | } | |
1658 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1659 | ||
1660 | ret = ret_splice; | |
1661 | ||
1662 | goto end; | |
1663 | ||
8994307f DG |
1664 | write_error: |
1665 | /* | |
1666 | * This is a special case that the relayd has closed its socket. Let's | |
1667 | * cleanup the relayd object and all associated streams. | |
1668 | */ | |
1669 | if (relayd && relayd_hang_up) { | |
1670 | cleanup_relayd(relayd, ctx); | |
1671 | /* Skip splice error so the consumer does not fail */ | |
1672 | goto end; | |
1673 | } | |
1674 | ||
f02e1e8a DG |
1675 | splice_error: |
1676 | /* send the appropriate error description to sessiond */ | |
1677 | switch (ret) { | |
f02e1e8a | 1678 | case EINVAL: |
f73fabfd | 1679 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1680 | break; |
1681 | case ENOMEM: | |
f73fabfd | 1682 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1683 | break; |
1684 | case ESPIPE: | |
f73fabfd | 1685 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1686 | break; |
1687 | } | |
1688 | ||
1689 | end: | |
1690 | if (relayd && stream->metadata_flag) { | |
1691 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1692 | } | |
1693 | ||
1694 | rcu_read_unlock(); | |
1695 | return written; | |
3bd1e081 MD |
1696 | } |
1697 | ||
1698 | /* | |
1699 | * Take a snapshot for a specific fd | |
1700 | * | |
1701 | * Returns 0 on success, < 0 on error | |
1702 | */ | |
ffe60014 | 1703 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1704 | { |
1705 | switch (consumer_data.type) { | |
1706 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1707 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1708 | case LTTNG_CONSUMER32_UST: |
1709 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1710 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1711 | default: |
1712 | ERR("Unknown consumer_data type"); | |
1713 | assert(0); | |
1714 | return -ENOSYS; | |
1715 | } | |
3bd1e081 MD |
1716 | } |
1717 | ||
1718 | /* | |
1719 | * Get the produced position | |
1720 | * | |
1721 | * Returns 0 on success, < 0 on error | |
1722 | */ | |
ffe60014 | 1723 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1724 | unsigned long *pos) |
1725 | { | |
1726 | switch (consumer_data.type) { | |
1727 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1728 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1729 | case LTTNG_CONSUMER32_UST: |
1730 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1731 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1732 | default: |
1733 | ERR("Unknown consumer_data type"); | |
1734 | assert(0); | |
1735 | return -ENOSYS; | |
1736 | } | |
1737 | } | |
1738 | ||
1739 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1740 | int sock, struct pollfd *consumer_sockpoll) | |
1741 | { | |
1742 | switch (consumer_data.type) { | |
1743 | case LTTNG_CONSUMER_KERNEL: | |
1744 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1745 | case LTTNG_CONSUMER32_UST: |
1746 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1747 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1748 | default: | |
1749 | ERR("Unknown consumer_data type"); | |
1750 | assert(0); | |
1751 | return -ENOSYS; | |
1752 | } | |
1753 | } | |
1754 | ||
43c34bc3 DG |
1755 | /* |
1756 | * Iterate over all streams of the hashtable and free them properly. | |
1757 | * | |
1758 | * WARNING: *MUST* be used with data stream only. | |
1759 | */ | |
1760 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1761 | { | |
43c34bc3 DG |
1762 | struct lttng_ht_iter iter; |
1763 | struct lttng_consumer_stream *stream; | |
1764 | ||
1765 | if (ht == NULL) { | |
1766 | return; | |
1767 | } | |
1768 | ||
1769 | rcu_read_lock(); | |
1770 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
5c540210 DG |
1771 | /* |
1772 | * Ignore return value since we are currently cleaning up so any error | |
1773 | * can't be handled. | |
1774 | */ | |
1775 | (void) consumer_del_stream(stream, ht); | |
43c34bc3 DG |
1776 | } |
1777 | rcu_read_unlock(); | |
1778 | ||
1779 | lttng_ht_destroy(ht); | |
1780 | } | |
1781 | ||
fb3a43a9 | 1782 | /* |
f724d81e | 1783 | * Iterate over all streams of the hashtable and free them properly. |
e316aad5 DG |
1784 | * |
1785 | * XXX: Should not be only for metadata stream or else use an other name. | |
fb3a43a9 DG |
1786 | */ |
1787 | static void destroy_stream_ht(struct lttng_ht *ht) | |
1788 | { | |
fb3a43a9 DG |
1789 | struct lttng_ht_iter iter; |
1790 | struct lttng_consumer_stream *stream; | |
1791 | ||
1792 | if (ht == NULL) { | |
1793 | return; | |
1794 | } | |
1795 | ||
d09e1200 | 1796 | rcu_read_lock(); |
58b1f425 | 1797 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
5c540210 DG |
1798 | /* |
1799 | * Ignore return value since we are currently cleaning up so any error | |
1800 | * can't be handled. | |
1801 | */ | |
1802 | (void) consumer_del_metadata_stream(stream, ht); | |
fb3a43a9 | 1803 | } |
d09e1200 | 1804 | rcu_read_unlock(); |
fb3a43a9 DG |
1805 | |
1806 | lttng_ht_destroy(ht); | |
1807 | } | |
1808 | ||
d88aee68 DG |
1809 | void lttng_consumer_close_metadata(void) |
1810 | { | |
1811 | switch (consumer_data.type) { | |
1812 | case LTTNG_CONSUMER_KERNEL: | |
1813 | /* | |
1814 | * The Kernel consumer has a different metadata scheme so we don't | |
1815 | * close anything because the stream will be closed by the session | |
1816 | * daemon. | |
1817 | */ | |
1818 | break; | |
1819 | case LTTNG_CONSUMER32_UST: | |
1820 | case LTTNG_CONSUMER64_UST: | |
1821 | /* | |
1822 | * Close all metadata streams. The metadata hash table is passed and | |
1823 | * this call iterates over it by closing all wakeup fd. This is safe | |
1824 | * because at this point we are sure that the metadata producer is | |
1825 | * either dead or blocked. | |
1826 | */ | |
1827 | lttng_ustconsumer_close_metadata(metadata_ht); | |
1828 | break; | |
1829 | default: | |
1830 | ERR("Unknown consumer_data type"); | |
1831 | assert(0); | |
1832 | } | |
1833 | } | |
1834 | ||
fb3a43a9 DG |
1835 | /* |
1836 | * Clean up a metadata stream and free its memory. | |
1837 | */ | |
e316aad5 DG |
1838 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
1839 | struct lttng_ht *ht) | |
fb3a43a9 DG |
1840 | { |
1841 | int ret; | |
e316aad5 DG |
1842 | struct lttng_ht_iter iter; |
1843 | struct lttng_consumer_channel *free_chan = NULL; | |
fb3a43a9 DG |
1844 | struct consumer_relayd_sock_pair *relayd; |
1845 | ||
1846 | assert(stream); | |
1847 | /* | |
1848 | * This call should NEVER receive regular stream. It must always be | |
1849 | * metadata stream and this is crucial for data structure synchronization. | |
1850 | */ | |
1851 | assert(stream->metadata_flag); | |
1852 | ||
e316aad5 DG |
1853 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
1854 | ||
1855 | if (ht == NULL) { | |
1856 | /* Means the stream was allocated but not successfully added */ | |
ffe60014 | 1857 | goto free_stream_rcu; |
e316aad5 DG |
1858 | } |
1859 | ||
74251bb8 | 1860 | pthread_mutex_lock(&consumer_data.lock); |
8994307f DG |
1861 | pthread_mutex_lock(&stream->lock); |
1862 | ||
fb3a43a9 DG |
1863 | switch (consumer_data.type) { |
1864 | case LTTNG_CONSUMER_KERNEL: | |
1865 | if (stream->mmap_base != NULL) { | |
1866 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
1867 | if (ret != 0) { | |
1868 | PERROR("munmap metadata stream"); | |
1869 | } | |
1870 | } | |
4c95e622 JD |
1871 | if (stream->wait_fd >= 0) { |
1872 | ret = close(stream->wait_fd); | |
1873 | if (ret < 0) { | |
1874 | PERROR("close kernel metadata wait_fd"); | |
1875 | } | |
1876 | } | |
fb3a43a9 DG |
1877 | break; |
1878 | case LTTNG_CONSUMER32_UST: | |
1879 | case LTTNG_CONSUMER64_UST: | |
1880 | lttng_ustconsumer_del_stream(stream); | |
1881 | break; | |
1882 | default: | |
1883 | ERR("Unknown consumer_data type"); | |
1884 | assert(0); | |
e316aad5 | 1885 | goto end; |
fb3a43a9 | 1886 | } |
fb3a43a9 | 1887 | |
c869f647 | 1888 | rcu_read_lock(); |
58b1f425 | 1889 | iter.iter.node = &stream->node.node; |
c869f647 DG |
1890 | ret = lttng_ht_del(ht, &iter); |
1891 | assert(!ret); | |
ca22feea | 1892 | |
d8ef542d MD |
1893 | iter.iter.node = &stream->node_channel_id.node; |
1894 | ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); | |
1895 | assert(!ret); | |
1896 | ||
ca22feea DG |
1897 | iter.iter.node = &stream->node_session_id.node; |
1898 | ret = lttng_ht_del(consumer_data.stream_list_ht, &iter); | |
1899 | assert(!ret); | |
c869f647 DG |
1900 | rcu_read_unlock(); |
1901 | ||
fb3a43a9 DG |
1902 | if (stream->out_fd >= 0) { |
1903 | ret = close(stream->out_fd); | |
1904 | if (ret) { | |
1905 | PERROR("close"); | |
1906 | } | |
1907 | } | |
1908 | ||
fb3a43a9 DG |
1909 | /* Check and cleanup relayd */ |
1910 | rcu_read_lock(); | |
1911 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1912 | if (relayd != NULL) { | |
1913 | uatomic_dec(&relayd->refcount); | |
1914 | assert(uatomic_read(&relayd->refcount) >= 0); | |
1915 | ||
1916 | /* Closing streams requires to lock the control socket. */ | |
1917 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1918 | ret = relayd_send_close_stream(&relayd->control_sock, | |
1919 | stream->relayd_stream_id, stream->next_net_seq_num - 1); | |
1920 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1921 | if (ret < 0) { | |
1922 | DBG("Unable to close stream on the relayd. Continuing"); | |
1923 | /* | |
1924 | * Continue here. There is nothing we can do for the relayd. | |
1925 | * Chances are that the relayd has closed the socket so we just | |
1926 | * continue cleaning up. | |
1927 | */ | |
1928 | } | |
1929 | ||
1930 | /* Both conditions are met, we destroy the relayd. */ | |
1931 | if (uatomic_read(&relayd->refcount) == 0 && | |
1932 | uatomic_read(&relayd->destroy_flag)) { | |
51230d70 | 1933 | consumer_destroy_relayd(relayd); |
fb3a43a9 DG |
1934 | } |
1935 | } | |
1936 | rcu_read_unlock(); | |
1937 | ||
1938 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 1939 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 1940 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 1941 | /* Go for channel deletion! */ |
e316aad5 | 1942 | free_chan = stream->chan; |
fb3a43a9 DG |
1943 | } |
1944 | ||
e316aad5 | 1945 | end: |
73811ecc DG |
1946 | /* |
1947 | * Nullify the stream reference so it is not used after deletion. The | |
1948 | * consumer data lock MUST be acquired before being able to check for a | |
1949 | * NULL pointer value. | |
1950 | */ | |
1951 | stream->chan->metadata_stream = NULL; | |
1952 | ||
8994307f | 1953 | pthread_mutex_unlock(&stream->lock); |
74251bb8 | 1954 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
1955 | |
1956 | if (free_chan) { | |
1957 | consumer_del_channel(free_chan); | |
1958 | } | |
1959 | ||
ffe60014 DG |
1960 | free_stream_rcu: |
1961 | call_rcu(&stream->node.head, free_stream_rcu); | |
fb3a43a9 DG |
1962 | } |
1963 | ||
1964 | /* | |
1965 | * Action done with the metadata stream when adding it to the consumer internal | |
1966 | * data structures to handle it. | |
1967 | */ | |
ffe60014 | 1968 | static int add_metadata_stream(struct lttng_consumer_stream *stream, |
e316aad5 | 1969 | struct lttng_ht *ht) |
fb3a43a9 | 1970 | { |
e316aad5 | 1971 | int ret = 0; |
fb3a43a9 | 1972 | struct consumer_relayd_sock_pair *relayd; |
76082088 | 1973 | struct lttng_ht_iter iter; |
d88aee68 | 1974 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 1975 | |
e316aad5 DG |
1976 | assert(stream); |
1977 | assert(ht); | |
1978 | ||
d88aee68 | 1979 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
1980 | |
1981 | pthread_mutex_lock(&consumer_data.lock); | |
2e818a6a | 1982 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 1983 | |
e316aad5 DG |
1984 | /* |
1985 | * From here, refcounts are updated so be _careful_ when returning an error | |
1986 | * after this point. | |
1987 | */ | |
1988 | ||
fb3a43a9 | 1989 | rcu_read_lock(); |
76082088 DG |
1990 | |
1991 | /* | |
1992 | * Lookup the stream just to make sure it does not exist in our internal | |
1993 | * state. This should NEVER happen. | |
1994 | */ | |
d88aee68 DG |
1995 | lttng_ht_lookup(ht, &stream->key, &iter); |
1996 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
1997 | assert(!node); |
1998 | ||
e316aad5 | 1999 | /* Find relayd and, if one is found, increment refcount. */ |
fb3a43a9 DG |
2000 | relayd = consumer_find_relayd(stream->net_seq_idx); |
2001 | if (relayd != NULL) { | |
2002 | uatomic_inc(&relayd->refcount); | |
2003 | } | |
e316aad5 | 2004 | |
e316aad5 | 2005 | /* |
ffe60014 DG |
2006 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2007 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2008 | * causes the count to become 0 also causes a stream to be added. The |
2009 | * channel deletion will thus be triggered by the following removal of this | |
2010 | * stream. | |
2011 | */ | |
ffe60014 | 2012 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2013 | /* Increment refcount before decrementing nb_init_stream_left */ |
2014 | cmm_smp_wmb(); | |
ffe60014 | 2015 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2016 | } |
2017 | ||
d88aee68 | 2018 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2019 | |
d8ef542d MD |
2020 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2021 | &stream->node_channel_id); | |
2022 | ||
ca22feea DG |
2023 | /* |
2024 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2025 | * the key since the HT does not use it and we allow to add redundant keys | |
2026 | * into this table. | |
2027 | */ | |
d88aee68 | 2028 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2029 | |
fb3a43a9 | 2030 | rcu_read_unlock(); |
e316aad5 | 2031 | |
2e818a6a | 2032 | pthread_mutex_unlock(&stream->lock); |
e316aad5 DG |
2033 | pthread_mutex_unlock(&consumer_data.lock); |
2034 | return ret; | |
fb3a43a9 DG |
2035 | } |
2036 | ||
8994307f DG |
2037 | /* |
2038 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2039 | */ | |
2040 | static void validate_endpoint_status_data_stream(void) | |
2041 | { | |
2042 | struct lttng_ht_iter iter; | |
2043 | struct lttng_consumer_stream *stream; | |
2044 | ||
2045 | DBG("Consumer delete flagged data stream"); | |
2046 | ||
2047 | rcu_read_lock(); | |
2048 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2049 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2050 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2051 | continue; |
2052 | } | |
2053 | /* Delete it right now */ | |
2054 | consumer_del_stream(stream, data_ht); | |
2055 | } | |
2056 | rcu_read_unlock(); | |
2057 | } | |
2058 | ||
2059 | /* | |
2060 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2061 | */ | |
2062 | static void validate_endpoint_status_metadata_stream( | |
2063 | struct lttng_poll_event *pollset) | |
2064 | { | |
2065 | struct lttng_ht_iter iter; | |
2066 | struct lttng_consumer_stream *stream; | |
2067 | ||
2068 | DBG("Consumer delete flagged metadata stream"); | |
2069 | ||
2070 | assert(pollset); | |
2071 | ||
2072 | rcu_read_lock(); | |
2073 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2074 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2075 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2076 | continue; |
2077 | } | |
2078 | /* | |
2079 | * Remove from pollset so the metadata thread can continue without | |
2080 | * blocking on a deleted stream. | |
2081 | */ | |
2082 | lttng_poll_del(pollset, stream->wait_fd); | |
2083 | ||
2084 | /* Delete it right now */ | |
2085 | consumer_del_metadata_stream(stream, metadata_ht); | |
2086 | } | |
2087 | rcu_read_unlock(); | |
2088 | } | |
2089 | ||
fb3a43a9 DG |
2090 | /* |
2091 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2092 | * network. | |
2093 | */ | |
7d980def | 2094 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 DG |
2095 | { |
2096 | int ret, i, pollfd; | |
2097 | uint32_t revents, nb_fd; | |
e316aad5 | 2098 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2099 | struct lttng_ht_iter iter; |
d88aee68 | 2100 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2101 | struct lttng_poll_event events; |
2102 | struct lttng_consumer_local_data *ctx = data; | |
2103 | ssize_t len; | |
2104 | ||
2105 | rcu_register_thread(); | |
2106 | ||
d88aee68 | 2107 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
04bb2b64 DG |
2108 | if (!metadata_ht) { |
2109 | /* ENOMEM at this point. Better to bail out. */ | |
d8ef542d | 2110 | goto end_ht; |
04bb2b64 DG |
2111 | } |
2112 | ||
fb3a43a9 DG |
2113 | DBG("Thread metadata poll started"); |
2114 | ||
fb3a43a9 DG |
2115 | /* Size is set to 1 for the consumer_metadata pipe */ |
2116 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2117 | if (ret < 0) { | |
2118 | ERR("Poll set creation failed"); | |
d8ef542d | 2119 | goto end_poll; |
fb3a43a9 DG |
2120 | } |
2121 | ||
13886d2d DG |
2122 | ret = lttng_poll_add(&events, |
2123 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2124 | if (ret < 0) { |
2125 | goto end; | |
2126 | } | |
2127 | ||
2128 | /* Main loop */ | |
2129 | DBG("Metadata main loop started"); | |
2130 | ||
2131 | while (1) { | |
fb3a43a9 | 2132 | /* Only the metadata pipe is set */ |
d21b0d71 | 2133 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { |
fb3a43a9 DG |
2134 | goto end; |
2135 | } | |
2136 | ||
2137 | restart: | |
d21b0d71 | 2138 | DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); |
fb3a43a9 DG |
2139 | ret = lttng_poll_wait(&events, -1); |
2140 | DBG("Metadata event catched in thread"); | |
2141 | if (ret < 0) { | |
2142 | if (errno == EINTR) { | |
e316aad5 | 2143 | ERR("Poll EINTR catched"); |
fb3a43a9 DG |
2144 | goto restart; |
2145 | } | |
2146 | goto error; | |
2147 | } | |
2148 | ||
0d9c5d77 DG |
2149 | nb_fd = ret; |
2150 | ||
e316aad5 | 2151 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 DG |
2152 | for (i = 0; i < nb_fd; i++) { |
2153 | revents = LTTNG_POLL_GETEV(&events, i); | |
2154 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2155 | ||
e316aad5 DG |
2156 | /* Just don't waste time if no returned events for the fd */ |
2157 | if (!revents) { | |
2158 | continue; | |
2159 | } | |
2160 | ||
13886d2d | 2161 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
4adabd61 | 2162 | if (revents & (LPOLLERR | LPOLLHUP )) { |
fb3a43a9 DG |
2163 | DBG("Metadata thread pipe hung up"); |
2164 | /* | |
2165 | * Remove the pipe from the poll set and continue the loop | |
2166 | * since their might be data to consume. | |
2167 | */ | |
13886d2d DG |
2168 | lttng_poll_del(&events, |
2169 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2170 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2171 | continue; |
2172 | } else if (revents & LPOLLIN) { | |
13886d2d DG |
2173 | ssize_t pipe_len; |
2174 | ||
2175 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2176 | &stream, sizeof(stream)); | |
2177 | if (pipe_len < 0) { | |
2178 | ERR("read metadata stream, ret: %ld", pipe_len); | |
fb3a43a9 | 2179 | /* |
13886d2d | 2180 | * Continue here to handle the rest of the streams. |
fb3a43a9 DG |
2181 | */ |
2182 | continue; | |
2183 | } | |
2184 | ||
8994307f DG |
2185 | /* A NULL stream means that the state has changed. */ |
2186 | if (stream == NULL) { | |
2187 | /* Check for deleted streams. */ | |
2188 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2189 | goto restart; |
8994307f DG |
2190 | } |
2191 | ||
fb3a43a9 DG |
2192 | DBG("Adding metadata stream %d to poll set", |
2193 | stream->wait_fd); | |
2194 | ||
ffe60014 | 2195 | ret = add_metadata_stream(stream, metadata_ht); |
e316aad5 DG |
2196 | if (ret) { |
2197 | ERR("Unable to add metadata stream"); | |
2198 | /* Stream was not setup properly. Continuing. */ | |
2199 | consumer_del_metadata_stream(stream, NULL); | |
2200 | continue; | |
2201 | } | |
fb3a43a9 DG |
2202 | |
2203 | /* Add metadata stream to the global poll events list */ | |
2204 | lttng_poll_add(&events, stream->wait_fd, | |
2205 | LPOLLIN | LPOLLPRI); | |
fb3a43a9 DG |
2206 | } |
2207 | ||
e316aad5 | 2208 | /* Handle other stream */ |
fb3a43a9 DG |
2209 | continue; |
2210 | } | |
2211 | ||
d09e1200 | 2212 | rcu_read_lock(); |
d88aee68 DG |
2213 | { |
2214 | uint64_t tmp_id = (uint64_t) pollfd; | |
2215 | ||
2216 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2217 | } | |
2218 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2219 | assert(node); |
fb3a43a9 DG |
2220 | |
2221 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2222 | node); |
fb3a43a9 | 2223 | |
e316aad5 | 2224 | /* Check for error event */ |
4adabd61 | 2225 | if (revents & (LPOLLERR | LPOLLHUP)) { |
e316aad5 | 2226 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2227 | if (!stream->hangup_flush_done |
2228 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2229 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2230 | DBG("Attempting to flush and consume the UST buffers"); | |
2231 | lttng_ustconsumer_on_stream_hangup(stream); | |
2232 | ||
2233 | /* We just flushed the stream now read it. */ | |
4bb94b75 DG |
2234 | do { |
2235 | len = ctx->on_buffer_ready(stream, ctx); | |
2236 | /* | |
2237 | * We don't check the return value here since if we get | |
2238 | * a negative len, it means an error occured thus we | |
2239 | * simply remove it from the poll set and free the | |
2240 | * stream. | |
2241 | */ | |
2242 | } while (len > 0); | |
fb3a43a9 DG |
2243 | } |
2244 | ||
fb3a43a9 | 2245 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2246 | /* |
2247 | * This call update the channel states, closes file descriptors | |
2248 | * and securely free the stream. | |
2249 | */ | |
2250 | consumer_del_metadata_stream(stream, metadata_ht); | |
2251 | } else if (revents & (LPOLLIN | LPOLLPRI)) { | |
2252 | /* Get the data out of the metadata file descriptor */ | |
2253 | DBG("Metadata available on fd %d", pollfd); | |
2254 | assert(stream->wait_fd == pollfd); | |
2255 | ||
2256 | len = ctx->on_buffer_ready(stream, ctx); | |
2257 | /* It's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2258 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2259 | /* Clean up stream from consumer and free it. */ |
2260 | lttng_poll_del(&events, stream->wait_fd); | |
2261 | consumer_del_metadata_stream(stream, metadata_ht); | |
e316aad5 DG |
2262 | } else if (len > 0) { |
2263 | stream->data_read = 1; | |
2264 | } | |
fb3a43a9 | 2265 | } |
e316aad5 DG |
2266 | |
2267 | /* Release RCU lock for the stream looked up */ | |
d09e1200 | 2268 | rcu_read_unlock(); |
fb3a43a9 DG |
2269 | } |
2270 | } | |
2271 | ||
2272 | error: | |
2273 | end: | |
2274 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2275 | |
d8ef542d MD |
2276 | lttng_poll_clean(&events); |
2277 | end_poll: | |
04bb2b64 | 2278 | destroy_stream_ht(metadata_ht); |
d8ef542d | 2279 | end_ht: |
fb3a43a9 DG |
2280 | rcu_unregister_thread(); |
2281 | return NULL; | |
2282 | } | |
2283 | ||
3bd1e081 | 2284 | /* |
e4421fec | 2285 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2286 | * it to tracefile if necessary. |
2287 | */ | |
7d980def | 2288 | void *consumer_thread_data_poll(void *data) |
3bd1e081 MD |
2289 | { |
2290 | int num_rdy, num_hup, high_prio, ret, i; | |
2291 | struct pollfd *pollfd = NULL; | |
2292 | /* local view of the streams */ | |
c869f647 | 2293 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2294 | /* local view of consumer_data.fds_count */ |
2295 | int nb_fd = 0; | |
3bd1e081 | 2296 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2297 | ssize_t len; |
3bd1e081 | 2298 | |
e7b994a3 DG |
2299 | rcu_register_thread(); |
2300 | ||
d88aee68 | 2301 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
43c34bc3 | 2302 | if (data_ht == NULL) { |
04bb2b64 | 2303 | /* ENOMEM at this point. Better to bail out. */ |
43c34bc3 DG |
2304 | goto end; |
2305 | } | |
2306 | ||
effcf122 | 2307 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream)); |
3bd1e081 MD |
2308 | |
2309 | while (1) { | |
2310 | high_prio = 0; | |
2311 | num_hup = 0; | |
2312 | ||
2313 | /* | |
e4421fec | 2314 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2315 | * local array as well |
2316 | */ | |
2317 | pthread_mutex_lock(&consumer_data.lock); | |
2318 | if (consumer_data.need_update) { | |
0e428499 DG |
2319 | free(pollfd); |
2320 | pollfd = NULL; | |
2321 | ||
2322 | free(local_stream); | |
2323 | local_stream = NULL; | |
3bd1e081 | 2324 | |
50f8ae69 | 2325 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2326 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 | 2327 | if (pollfd == NULL) { |
7a57cf92 | 2328 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2329 | pthread_mutex_unlock(&consumer_data.lock); |
2330 | goto end; | |
2331 | } | |
2332 | ||
50f8ae69 | 2333 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2334 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
747f8642 | 2335 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2336 | if (local_stream == NULL) { |
7a57cf92 | 2337 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2338 | pthread_mutex_unlock(&consumer_data.lock); |
2339 | goto end; | |
2340 | } | |
ffe60014 | 2341 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2342 | data_ht); |
3bd1e081 MD |
2343 | if (ret < 0) { |
2344 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2345 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2346 | pthread_mutex_unlock(&consumer_data.lock); |
2347 | goto end; | |
2348 | } | |
2349 | nb_fd = ret; | |
2350 | consumer_data.need_update = 0; | |
2351 | } | |
2352 | pthread_mutex_unlock(&consumer_data.lock); | |
2353 | ||
4078b776 MD |
2354 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2355 | if (nb_fd == 0 && consumer_quit == 1) { | |
2356 | goto end; | |
2357 | } | |
3bd1e081 | 2358 | /* poll on the array of fds */ |
88f2b785 | 2359 | restart: |
3bd1e081 | 2360 | DBG("polling on %d fd", nb_fd + 1); |
cb365c03 | 2361 | num_rdy = poll(pollfd, nb_fd + 1, -1); |
3bd1e081 MD |
2362 | DBG("poll num_rdy : %d", num_rdy); |
2363 | if (num_rdy == -1) { | |
88f2b785 MD |
2364 | /* |
2365 | * Restart interrupted system call. | |
2366 | */ | |
2367 | if (errno == EINTR) { | |
2368 | goto restart; | |
2369 | } | |
7a57cf92 | 2370 | PERROR("Poll error"); |
f73fabfd | 2371 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2372 | goto end; |
2373 | } else if (num_rdy == 0) { | |
2374 | DBG("Polling thread timed out"); | |
2375 | goto end; | |
2376 | } | |
2377 | ||
3bd1e081 | 2378 | /* |
50f8ae69 | 2379 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2380 | * beginning of the loop to update the array. We want to prioritize |
2381 | * array update over low-priority reads. | |
3bd1e081 | 2382 | */ |
509bb1cf | 2383 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2384 | ssize_t pipe_readlen; |
04fdd819 | 2385 | |
50f8ae69 | 2386 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2387 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2388 | &new_stream, sizeof(new_stream)); | |
23f5f35d | 2389 | if (pipe_readlen < 0) { |
acdb9057 | 2390 | ERR("Consumer data pipe ret %ld", pipe_readlen); |
23f5f35d DG |
2391 | /* Continue so we can at least handle the current stream(s). */ |
2392 | continue; | |
2393 | } | |
c869f647 DG |
2394 | |
2395 | /* | |
2396 | * If the stream is NULL, just ignore it. It's also possible that | |
2397 | * the sessiond poll thread changed the consumer_quit state and is | |
2398 | * waking us up to test it. | |
2399 | */ | |
2400 | if (new_stream == NULL) { | |
8994307f | 2401 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2402 | continue; |
2403 | } | |
2404 | ||
ffe60014 | 2405 | ret = add_stream(new_stream, data_ht); |
c869f647 | 2406 | if (ret) { |
d88aee68 | 2407 | ERR("Consumer add stream %" PRIu64 " failed. Continuing", |
c869f647 DG |
2408 | new_stream->key); |
2409 | /* | |
2410 | * At this point, if the add_stream fails, it is not in the | |
2411 | * hash table thus passing the NULL value here. | |
2412 | */ | |
2413 | consumer_del_stream(new_stream, NULL); | |
2414 | } | |
2415 | ||
2416 | /* Continue to update the local streams and handle prio ones */ | |
3bd1e081 MD |
2417 | continue; |
2418 | } | |
2419 | ||
2420 | /* Take care of high priority channels first. */ | |
2421 | for (i = 0; i < nb_fd; i++) { | |
9617607b DG |
2422 | if (local_stream[i] == NULL) { |
2423 | continue; | |
2424 | } | |
fb3a43a9 | 2425 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2426 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2427 | high_prio = 1; | |
4078b776 | 2428 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2429 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2430 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2431 | /* Clean the stream and free it. */ |
2432 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2433 | local_stream[i] = NULL; |
4078b776 MD |
2434 | } else if (len > 0) { |
2435 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2436 | } |
3bd1e081 MD |
2437 | } |
2438 | } | |
2439 | ||
4078b776 MD |
2440 | /* |
2441 | * If we read high prio channel in this loop, try again | |
2442 | * for more high prio data. | |
2443 | */ | |
2444 | if (high_prio) { | |
3bd1e081 MD |
2445 | continue; |
2446 | } | |
2447 | ||
2448 | /* Take care of low priority channels. */ | |
4078b776 | 2449 | for (i = 0; i < nb_fd; i++) { |
9617607b DG |
2450 | if (local_stream[i] == NULL) { |
2451 | continue; | |
2452 | } | |
4078b776 MD |
2453 | if ((pollfd[i].revents & POLLIN) || |
2454 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
2455 | DBG("Normal read on fd %d", pollfd[i].fd); |
2456 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2457 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2458 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2459 | /* Clean the stream and free it. */ |
2460 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2461 | local_stream[i] = NULL; |
4078b776 MD |
2462 | } else if (len > 0) { |
2463 | local_stream[i]->data_read = 1; | |
2464 | } | |
2465 | } | |
2466 | } | |
2467 | ||
2468 | /* Handle hangup and errors */ | |
2469 | for (i = 0; i < nb_fd; i++) { | |
9617607b DG |
2470 | if (local_stream[i] == NULL) { |
2471 | continue; | |
2472 | } | |
4078b776 MD |
2473 | if (!local_stream[i]->hangup_flush_done |
2474 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2475 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2476 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2477 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2478 | pollfd[i].fd); |
4078b776 MD |
2479 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2480 | /* Attempt read again, for the data we just flushed. */ | |
2481 | local_stream[i]->data_read = 1; | |
2482 | } | |
2483 | /* | |
2484 | * If the poll flag is HUP/ERR/NVAL and we have | |
2485 | * read no data in this pass, we can remove the | |
2486 | * stream from its hash table. | |
2487 | */ | |
2488 | if ((pollfd[i].revents & POLLHUP)) { | |
2489 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2490 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2491 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2492 | local_stream[i] = NULL; |
4078b776 MD |
2493 | num_hup++; |
2494 | } | |
2495 | } else if (pollfd[i].revents & POLLERR) { | |
2496 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2497 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2498 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2499 | local_stream[i] = NULL; |
4078b776 MD |
2500 | num_hup++; |
2501 | } | |
2502 | } else if (pollfd[i].revents & POLLNVAL) { | |
2503 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2504 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2505 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2506 | local_stream[i] = NULL; |
4078b776 | 2507 | num_hup++; |
3bd1e081 MD |
2508 | } |
2509 | } | |
9617607b DG |
2510 | if (local_stream[i] != NULL) { |
2511 | local_stream[i]->data_read = 0; | |
2512 | } | |
3bd1e081 MD |
2513 | } |
2514 | } | |
2515 | end: | |
2516 | DBG("polling thread exiting"); | |
0e428499 DG |
2517 | free(pollfd); |
2518 | free(local_stream); | |
fb3a43a9 DG |
2519 | |
2520 | /* | |
2521 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2522 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2523 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2524 | * not return and could create a endless wait period if the pipe is the | |
2525 | * only tracked fd in the poll set. The thread will take care of closing | |
2526 | * the read side. | |
fb3a43a9 | 2527 | */ |
13886d2d | 2528 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2529 | |
04bb2b64 | 2530 | destroy_data_stream_ht(data_ht); |
43c34bc3 | 2531 | |
e7b994a3 | 2532 | rcu_unregister_thread(); |
3bd1e081 MD |
2533 | return NULL; |
2534 | } | |
2535 | ||
d8ef542d MD |
2536 | /* |
2537 | * Close wake-up end of each stream belonging to the channel. This will | |
2538 | * allow the poll() on the stream read-side to detect when the | |
2539 | * write-side (application) finally closes them. | |
2540 | */ | |
2541 | static | |
2542 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2543 | { | |
2544 | struct lttng_ht *ht; | |
2545 | struct lttng_consumer_stream *stream; | |
2546 | struct lttng_ht_iter iter; | |
2547 | ||
2548 | ht = consumer_data.stream_per_chan_id_ht; | |
2549 | ||
2550 | rcu_read_lock(); | |
2551 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2552 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2553 | ht->match_fct, &channel->key, | |
2554 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2555 | /* |
2556 | * Protect against teardown with mutex. | |
2557 | */ | |
2558 | pthread_mutex_lock(&stream->lock); | |
2559 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2560 | goto next; | |
2561 | } | |
d8ef542d MD |
2562 | switch (consumer_data.type) { |
2563 | case LTTNG_CONSUMER_KERNEL: | |
2564 | break; | |
2565 | case LTTNG_CONSUMER32_UST: | |
2566 | case LTTNG_CONSUMER64_UST: | |
2567 | /* | |
2568 | * Note: a mutex is taken internally within | |
2569 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2570 | * use from concurrent close. | |
2571 | */ | |
2572 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2573 | break; | |
2574 | default: | |
2575 | ERR("Unknown consumer_data type"); | |
2576 | assert(0); | |
2577 | } | |
f2ad556d MD |
2578 | next: |
2579 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2580 | } |
2581 | rcu_read_unlock(); | |
2582 | } | |
2583 | ||
2584 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2585 | { | |
2586 | struct lttng_ht_iter iter; | |
2587 | struct lttng_consumer_channel *channel; | |
2588 | int ret; | |
2589 | ||
2590 | if (ht == NULL) { | |
2591 | return; | |
2592 | } | |
2593 | ||
2594 | rcu_read_lock(); | |
2595 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2596 | ret = lttng_ht_del(ht, &iter); | |
2597 | assert(ret != 0); | |
2598 | } | |
2599 | rcu_read_unlock(); | |
2600 | ||
2601 | lttng_ht_destroy(ht); | |
2602 | } | |
2603 | ||
2604 | /* | |
2605 | * This thread polls the channel fds to detect when they are being | |
2606 | * closed. It closes all related streams if the channel is detected as | |
2607 | * closed. It is currently only used as a shim layer for UST because the | |
2608 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2609 | * periodical flush. | |
2610 | */ | |
2611 | void *consumer_thread_channel_poll(void *data) | |
2612 | { | |
2613 | int ret, i, pollfd; | |
2614 | uint32_t revents, nb_fd; | |
2615 | struct lttng_consumer_channel *chan = NULL; | |
2616 | struct lttng_ht_iter iter; | |
2617 | struct lttng_ht_node_u64 *node; | |
2618 | struct lttng_poll_event events; | |
2619 | struct lttng_consumer_local_data *ctx = data; | |
2620 | struct lttng_ht *channel_ht; | |
2621 | ||
2622 | rcu_register_thread(); | |
2623 | ||
2624 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
2625 | if (!channel_ht) { | |
2626 | /* ENOMEM at this point. Better to bail out. */ | |
2627 | goto end_ht; | |
2628 | } | |
2629 | ||
2630 | DBG("Thread channel poll started"); | |
2631 | ||
2632 | /* Size is set to 1 for the consumer_channel pipe */ | |
2633 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2634 | if (ret < 0) { | |
2635 | ERR("Poll set creation failed"); | |
2636 | goto end_poll; | |
2637 | } | |
2638 | ||
2639 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2640 | if (ret < 0) { | |
2641 | goto end; | |
2642 | } | |
2643 | ||
2644 | /* Main loop */ | |
2645 | DBG("Channel main loop started"); | |
2646 | ||
2647 | while (1) { | |
2648 | /* Only the channel pipe is set */ | |
2649 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { | |
2650 | goto end; | |
2651 | } | |
2652 | ||
2653 | restart: | |
2654 | DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); | |
2655 | ret = lttng_poll_wait(&events, -1); | |
2656 | DBG("Channel event catched in thread"); | |
2657 | if (ret < 0) { | |
2658 | if (errno == EINTR) { | |
2659 | ERR("Poll EINTR catched"); | |
2660 | goto restart; | |
2661 | } | |
2662 | goto end; | |
2663 | } | |
2664 | ||
2665 | nb_fd = ret; | |
2666 | ||
2667 | /* From here, the event is a channel wait fd */ | |
2668 | for (i = 0; i < nb_fd; i++) { | |
2669 | revents = LTTNG_POLL_GETEV(&events, i); | |
2670 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2671 | ||
2672 | /* Just don't waste time if no returned events for the fd */ | |
2673 | if (!revents) { | |
2674 | continue; | |
2675 | } | |
2676 | if (pollfd == ctx->consumer_channel_pipe[0]) { | |
2677 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2678 | DBG("Channel thread pipe hung up"); | |
2679 | /* | |
2680 | * Remove the pipe from the poll set and continue the loop | |
2681 | * since their might be data to consume. | |
2682 | */ | |
2683 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2684 | continue; | |
2685 | } else if (revents & LPOLLIN) { | |
2686 | enum consumer_channel_action action; | |
a0cbdd2e | 2687 | uint64_t key; |
d8ef542d | 2688 | |
a0cbdd2e | 2689 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d MD |
2690 | if (ret <= 0) { |
2691 | ERR("Error reading channel pipe"); | |
2692 | continue; | |
2693 | } | |
2694 | ||
2695 | switch (action) { | |
2696 | case CONSUMER_CHANNEL_ADD: | |
2697 | DBG("Adding channel %d to poll set", | |
2698 | chan->wait_fd); | |
2699 | ||
2700 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2701 | chan->wait_fd); | |
c7260a81 | 2702 | rcu_read_lock(); |
d8ef542d MD |
2703 | lttng_ht_add_unique_u64(channel_ht, |
2704 | &chan->wait_fd_node); | |
c7260a81 | 2705 | rcu_read_unlock(); |
d8ef542d MD |
2706 | /* Add channel to the global poll events list */ |
2707 | lttng_poll_add(&events, chan->wait_fd, | |
2708 | LPOLLIN | LPOLLPRI); | |
2709 | break; | |
a0cbdd2e MD |
2710 | case CONSUMER_CHANNEL_DEL: |
2711 | { | |
f2a444f1 DG |
2712 | struct lttng_consumer_stream *stream, *stmp; |
2713 | ||
c7260a81 | 2714 | rcu_read_lock(); |
a0cbdd2e MD |
2715 | chan = consumer_find_channel(key); |
2716 | if (!chan) { | |
c7260a81 | 2717 | rcu_read_unlock(); |
a0cbdd2e MD |
2718 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2719 | break; | |
2720 | } | |
2721 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2722 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2723 | ret = lttng_ht_del(channel_ht, &iter); |
2724 | assert(ret == 0); | |
2725 | consumer_close_channel_streams(chan); | |
2726 | ||
f2a444f1 DG |
2727 | switch (consumer_data.type) { |
2728 | case LTTNG_CONSUMER_KERNEL: | |
2729 | break; | |
2730 | case LTTNG_CONSUMER32_UST: | |
2731 | case LTTNG_CONSUMER64_UST: | |
2732 | /* Delete streams that might have been left in the stream list. */ | |
2733 | cds_list_for_each_entry_safe(stream, stmp, &chan->streams.head, | |
2734 | send_node) { | |
2735 | cds_list_del(&stream->send_node); | |
2736 | lttng_ustconsumer_del_stream(stream); | |
2737 | uatomic_sub(&stream->chan->refcount, 1); | |
2738 | assert(&chan->refcount); | |
2739 | free(stream); | |
2740 | } | |
2741 | break; | |
2742 | default: | |
2743 | ERR("Unknown consumer_data type"); | |
2744 | assert(0); | |
2745 | } | |
2746 | ||
a0cbdd2e MD |
2747 | /* |
2748 | * Release our own refcount. Force channel deletion even if | |
2749 | * streams were not initialized. | |
2750 | */ | |
2751 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2752 | consumer_del_channel(chan); | |
2753 | } | |
c7260a81 | 2754 | rcu_read_unlock(); |
a0cbdd2e MD |
2755 | goto restart; |
2756 | } | |
d8ef542d MD |
2757 | case CONSUMER_CHANNEL_QUIT: |
2758 | /* | |
2759 | * Remove the pipe from the poll set and continue the loop | |
2760 | * since their might be data to consume. | |
2761 | */ | |
2762 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2763 | continue; | |
2764 | default: | |
2765 | ERR("Unknown action"); | |
2766 | break; | |
2767 | } | |
2768 | } | |
2769 | ||
2770 | /* Handle other stream */ | |
2771 | continue; | |
2772 | } | |
2773 | ||
2774 | rcu_read_lock(); | |
2775 | { | |
2776 | uint64_t tmp_id = (uint64_t) pollfd; | |
2777 | ||
2778 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2779 | } | |
2780 | node = lttng_ht_iter_get_node_u64(&iter); | |
2781 | assert(node); | |
2782 | ||
2783 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2784 | wait_fd_node); | |
2785 | ||
2786 | /* Check for error event */ | |
2787 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2788 | DBG("Channel fd %d is hup|err.", pollfd); | |
2789 | ||
2790 | lttng_poll_del(&events, chan->wait_fd); | |
2791 | ret = lttng_ht_del(channel_ht, &iter); | |
2792 | assert(ret == 0); | |
2793 | consumer_close_channel_streams(chan); | |
f2ad556d MD |
2794 | |
2795 | /* Release our own refcount */ | |
2796 | if (!uatomic_sub_return(&chan->refcount, 1) | |
2797 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
2798 | consumer_del_channel(chan); | |
2799 | } | |
d8ef542d MD |
2800 | } |
2801 | ||
2802 | /* Release RCU lock for the channel looked up */ | |
2803 | rcu_read_unlock(); | |
2804 | } | |
2805 | } | |
2806 | ||
2807 | end: | |
2808 | lttng_poll_clean(&events); | |
2809 | end_poll: | |
2810 | destroy_channel_ht(channel_ht); | |
2811 | end_ht: | |
2812 | DBG("Channel poll thread exiting"); | |
2813 | rcu_unregister_thread(); | |
2814 | return NULL; | |
2815 | } | |
2816 | ||
331744e3 JD |
2817 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
2818 | struct pollfd *sockpoll, int client_socket) | |
2819 | { | |
2820 | int ret; | |
2821 | ||
2822 | assert(ctx); | |
2823 | assert(sockpoll); | |
2824 | ||
2825 | if (lttng_consumer_poll_socket(sockpoll) < 0) { | |
2826 | ret = -1; | |
2827 | goto error; | |
2828 | } | |
2829 | DBG("Metadata connection on client_socket"); | |
2830 | ||
2831 | /* Blocking call, waiting for transmission */ | |
2832 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
2833 | if (ctx->consumer_metadata_socket < 0) { | |
2834 | WARN("On accept metadata"); | |
2835 | ret = -1; | |
2836 | goto error; | |
2837 | } | |
2838 | ret = 0; | |
2839 | ||
2840 | error: | |
2841 | return ret; | |
2842 | } | |
2843 | ||
3bd1e081 MD |
2844 | /* |
2845 | * This thread listens on the consumerd socket and receives the file | |
2846 | * descriptors from the session daemon. | |
2847 | */ | |
7d980def | 2848 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 2849 | { |
d96f09c6 | 2850 | int sock = -1, client_socket, ret; |
3bd1e081 MD |
2851 | /* |
2852 | * structure to poll for incoming data on communication socket avoids | |
2853 | * making blocking sockets. | |
2854 | */ | |
2855 | struct pollfd consumer_sockpoll[2]; | |
2856 | struct lttng_consumer_local_data *ctx = data; | |
2857 | ||
e7b994a3 DG |
2858 | rcu_register_thread(); |
2859 | ||
3bd1e081 MD |
2860 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
2861 | unlink(ctx->consumer_command_sock_path); | |
2862 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
2863 | if (client_socket < 0) { | |
2864 | ERR("Cannot create command socket"); | |
2865 | goto end; | |
2866 | } | |
2867 | ||
2868 | ret = lttcomm_listen_unix_sock(client_socket); | |
2869 | if (ret < 0) { | |
2870 | goto end; | |
2871 | } | |
2872 | ||
32258573 | 2873 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 2874 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
2875 | /* return < 0 on error, but == 0 is not fatal */ |
2876 | if (ret < 0) { | |
32258573 | 2877 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
2878 | goto end; |
2879 | } | |
2880 | ||
3bd1e081 MD |
2881 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
2882 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
2883 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
2884 | consumer_sockpoll[1].fd = client_socket; | |
2885 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2886 | ||
2887 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2888 | goto end; | |
2889 | } | |
2890 | DBG("Connection on client_socket"); | |
2891 | ||
2892 | /* Blocking call, waiting for transmission */ | |
2893 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 2894 | if (sock < 0) { |
3bd1e081 MD |
2895 | WARN("On accept"); |
2896 | goto end; | |
2897 | } | |
3bd1e081 | 2898 | |
331744e3 JD |
2899 | /* |
2900 | * Setup metadata socket which is the second socket connection on the | |
2901 | * command unix socket. | |
2902 | */ | |
2903 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
2904 | if (ret < 0) { | |
2905 | goto end; | |
2906 | } | |
2907 | ||
d96f09c6 DG |
2908 | /* This socket is not useful anymore. */ |
2909 | ret = close(client_socket); | |
2910 | if (ret < 0) { | |
2911 | PERROR("close client_socket"); | |
2912 | } | |
2913 | client_socket = -1; | |
2914 | ||
3bd1e081 MD |
2915 | /* update the polling structure to poll on the established socket */ |
2916 | consumer_sockpoll[1].fd = sock; | |
2917 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2918 | ||
2919 | while (1) { | |
2920 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2921 | goto end; | |
2922 | } | |
2923 | DBG("Incoming command on sock"); | |
2924 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
2925 | if (ret == -ENOENT) { | |
2926 | DBG("Received STOP command"); | |
2927 | goto end; | |
2928 | } | |
4cbc1a04 DG |
2929 | if (ret <= 0) { |
2930 | /* | |
2931 | * This could simply be a session daemon quitting. Don't output | |
2932 | * ERR() here. | |
2933 | */ | |
2934 | DBG("Communication interrupted on command socket"); | |
3bd1e081 MD |
2935 | goto end; |
2936 | } | |
2937 | if (consumer_quit) { | |
2938 | DBG("consumer_thread_receive_fds received quit from signal"); | |
2939 | goto end; | |
2940 | } | |
ffe60014 | 2941 | DBG("received command on sock"); |
3bd1e081 MD |
2942 | } |
2943 | end: | |
ffe60014 | 2944 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 2945 | |
d88aee68 DG |
2946 | /* |
2947 | * Close metadata streams since the producer is the session daemon which | |
2948 | * just died. | |
2949 | * | |
2950 | * NOTE: for now, this only applies to the UST tracer. | |
2951 | */ | |
2952 | lttng_consumer_close_metadata(); | |
2953 | ||
3bd1e081 MD |
2954 | /* |
2955 | * when all fds have hung up, the polling thread | |
2956 | * can exit cleanly | |
2957 | */ | |
2958 | consumer_quit = 1; | |
2959 | ||
04fdd819 | 2960 | /* |
c869f647 | 2961 | * Notify the data poll thread to poll back again and test the |
8994307f | 2962 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 2963 | */ |
acdb9057 | 2964 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 2965 | |
a0cbdd2e | 2966 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 2967 | |
d96f09c6 DG |
2968 | /* Cleaning up possibly open sockets. */ |
2969 | if (sock >= 0) { | |
2970 | ret = close(sock); | |
2971 | if (ret < 0) { | |
2972 | PERROR("close sock sessiond poll"); | |
2973 | } | |
2974 | } | |
2975 | if (client_socket >= 0) { | |
38476d24 | 2976 | ret = close(client_socket); |
d96f09c6 DG |
2977 | if (ret < 0) { |
2978 | PERROR("close client_socket sessiond poll"); | |
2979 | } | |
2980 | } | |
2981 | ||
e7b994a3 | 2982 | rcu_unregister_thread(); |
3bd1e081 MD |
2983 | return NULL; |
2984 | } | |
d41f73b7 | 2985 | |
4078b776 | 2986 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
2987 | struct lttng_consumer_local_data *ctx) |
2988 | { | |
74251bb8 DG |
2989 | ssize_t ret; |
2990 | ||
2991 | pthread_mutex_lock(&stream->lock); | |
2992 | ||
d41f73b7 MD |
2993 | switch (consumer_data.type) { |
2994 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
2995 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
2996 | break; | |
7753dea8 MD |
2997 | case LTTNG_CONSUMER32_UST: |
2998 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
2999 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3000 | break; | |
d41f73b7 MD |
3001 | default: |
3002 | ERR("Unknown consumer_data type"); | |
3003 | assert(0); | |
74251bb8 DG |
3004 | ret = -ENOSYS; |
3005 | break; | |
d41f73b7 | 3006 | } |
74251bb8 DG |
3007 | |
3008 | pthread_mutex_unlock(&stream->lock); | |
3009 | return ret; | |
d41f73b7 MD |
3010 | } |
3011 | ||
3012 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3013 | { | |
3014 | switch (consumer_data.type) { | |
3015 | case LTTNG_CONSUMER_KERNEL: | |
3016 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3017 | case LTTNG_CONSUMER32_UST: |
3018 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3019 | return lttng_ustconsumer_on_recv_stream(stream); |
3020 | default: | |
3021 | ERR("Unknown consumer_data type"); | |
3022 | assert(0); | |
3023 | return -ENOSYS; | |
3024 | } | |
3025 | } | |
e4421fec DG |
3026 | |
3027 | /* | |
3028 | * Allocate and set consumer data hash tables. | |
3029 | */ | |
3030 | void lttng_consumer_init(void) | |
3031 | { | |
d88aee68 DG |
3032 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3033 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3034 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
d8ef542d | 3035 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
e4421fec | 3036 | } |
7735ef9e DG |
3037 | |
3038 | /* | |
3039 | * Process the ADD_RELAYD command receive by a consumer. | |
3040 | * | |
3041 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3042 | * The caller MUST acquire a RCU read side lock before calling it. | |
3043 | */ | |
da009f2c | 3044 | int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3045 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f JD |
3046 | struct pollfd *consumer_sockpoll, |
3047 | struct lttcomm_relayd_sock *relayd_sock, unsigned int sessiond_id) | |
7735ef9e | 3048 | { |
cd2b09ed | 3049 | int fd = -1, ret = -1, relayd_created = 0; |
f50f23d9 | 3050 | enum lttng_error_code ret_code = LTTNG_OK; |
d4298c99 | 3051 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3052 | |
6151a90f JD |
3053 | assert(ctx); |
3054 | assert(relayd_sock); | |
3055 | ||
da009f2c | 3056 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3057 | |
3058 | /* Get relayd reference if exists. */ | |
3059 | relayd = consumer_find_relayd(net_seq_idx); | |
3060 | if (relayd == NULL) { | |
da009f2c | 3061 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3062 | /* Not found. Allocate one. */ |
3063 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3064 | if (relayd == NULL) { | |
0d08d75e | 3065 | ret = -ENOMEM; |
618a6a28 MD |
3066 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3067 | goto error; | |
0d08d75e DG |
3068 | } else { |
3069 | relayd->sessiond_session_id = (uint64_t) sessiond_id; | |
3070 | relayd_created = 1; | |
7735ef9e | 3071 | } |
0d08d75e DG |
3072 | |
3073 | /* | |
3074 | * This code path MUST continue to the consumer send status message to | |
3075 | * we can notify the session daemon and continue our work without | |
3076 | * killing everything. | |
3077 | */ | |
da009f2c MD |
3078 | } else { |
3079 | /* | |
3080 | * relayd key should never be found for control socket. | |
3081 | */ | |
3082 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3083 | } |
3084 | ||
3085 | /* First send a status message before receiving the fds. */ | |
618a6a28 MD |
3086 | ret = consumer_send_status_msg(sock, LTTNG_OK); |
3087 | if (ret < 0) { | |
0d08d75e | 3088 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3089 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3090 | goto error_nosignal; | |
7735ef9e DG |
3091 | } |
3092 | ||
3093 | /* Poll on consumer socket. */ | |
3094 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
0d08d75e | 3095 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
7735ef9e | 3096 | ret = -EINTR; |
618a6a28 | 3097 | goto error_nosignal; |
7735ef9e DG |
3098 | } |
3099 | ||
3100 | /* Get relayd socket from session daemon */ | |
3101 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3102 | if (ret != sizeof(fd)) { | |
7735ef9e | 3103 | ret = -1; |
4028eeb9 | 3104 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3105 | |
3106 | /* | |
3107 | * Failing to receive FDs might indicate a major problem such as | |
3108 | * reaching a fd limit during the receive where the kernel returns a | |
3109 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3110 | * don't take any chances and stop everything. | |
3111 | * | |
3112 | * XXX: Feature request #558 will fix that and avoid this possible | |
3113 | * issue when reaching the fd limit. | |
3114 | */ | |
3115 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3116 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3117 | goto error; |
3118 | } | |
3119 | ||
7735ef9e DG |
3120 | /* Copy socket information and received FD */ |
3121 | switch (sock_type) { | |
3122 | case LTTNG_STREAM_CONTROL: | |
3123 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3124 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3125 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3126 | /* Handle create_sock error. */ |
f66c074c | 3127 | if (ret < 0) { |
618a6a28 | 3128 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3129 | goto error; |
f66c074c | 3130 | } |
da009f2c MD |
3131 | /* |
3132 | * Close the socket created internally by | |
3133 | * lttcomm_create_sock, so we can replace it by the one | |
3134 | * received from sessiond. | |
3135 | */ | |
3136 | if (close(relayd->control_sock.sock.fd)) { | |
3137 | PERROR("close"); | |
3138 | } | |
7735ef9e DG |
3139 | |
3140 | /* Assign new file descriptor */ | |
6151a90f | 3141 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3142 | fd = -1; /* For error path */ |
6151a90f JD |
3143 | /* Assign version values. */ |
3144 | relayd->control_sock.major = relayd_sock->major; | |
3145 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 DG |
3146 | |
3147 | /* | |
59e71485 DG |
3148 | * Create a session on the relayd and store the returned id. Lock the |
3149 | * control socket mutex if the relayd was NOT created before. | |
c5b6f4f0 | 3150 | */ |
59e71485 DG |
3151 | if (!relayd_created) { |
3152 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3153 | } | |
c5b6f4f0 | 3154 | ret = relayd_create_session(&relayd->control_sock, |
f7079f67 | 3155 | &relayd->relayd_session_id); |
59e71485 DG |
3156 | if (!relayd_created) { |
3157 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3158 | } | |
c5b6f4f0 | 3159 | if (ret < 0) { |
ffe60014 DG |
3160 | /* |
3161 | * Close all sockets of a relayd object. It will be freed if it was | |
3162 | * created at the error code path or else it will be garbage | |
3163 | * collect. | |
3164 | */ | |
3165 | (void) relayd_close(&relayd->control_sock); | |
3166 | (void) relayd_close(&relayd->data_sock); | |
618a6a28 | 3167 | ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; |
c5b6f4f0 DG |
3168 | goto error; |
3169 | } | |
3170 | ||
7735ef9e DG |
3171 | break; |
3172 | case LTTNG_STREAM_DATA: | |
3173 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3174 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3175 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3176 | /* Handle create_sock error. */ |
f66c074c | 3177 | if (ret < 0) { |
618a6a28 | 3178 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3179 | goto error; |
f66c074c | 3180 | } |
da009f2c MD |
3181 | /* |
3182 | * Close the socket created internally by | |
3183 | * lttcomm_create_sock, so we can replace it by the one | |
3184 | * received from sessiond. | |
3185 | */ | |
3186 | if (close(relayd->data_sock.sock.fd)) { | |
3187 | PERROR("close"); | |
3188 | } | |
7735ef9e DG |
3189 | |
3190 | /* Assign new file descriptor */ | |
6151a90f | 3191 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3192 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3193 | /* Assign version values. */ |
3194 | relayd->data_sock.major = relayd_sock->major; | |
3195 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3196 | break; |
3197 | default: | |
3198 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3199 | ret = -1; |
618a6a28 | 3200 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3201 | goto error; |
3202 | } | |
3203 | ||
d88aee68 | 3204 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3205 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3206 | relayd->net_seq_idx, fd); | |
3207 | ||
618a6a28 MD |
3208 | /* We successfully added the socket. Send status back. */ |
3209 | ret = consumer_send_status_msg(sock, ret_code); | |
3210 | if (ret < 0) { | |
3211 | /* Somehow, the session daemon is not responding anymore. */ | |
3212 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3213 | goto error_nosignal; | |
3214 | } | |
3215 | ||
7735ef9e DG |
3216 | /* |
3217 | * Add relayd socket pair to consumer data hashtable. If object already | |
3218 | * exists or on error, the function gracefully returns. | |
3219 | */ | |
d09e1200 | 3220 | add_relayd(relayd); |
7735ef9e DG |
3221 | |
3222 | /* All good! */ | |
4028eeb9 | 3223 | return 0; |
7735ef9e DG |
3224 | |
3225 | error: | |
618a6a28 MD |
3226 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3227 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3228 | } | |
3229 | ||
3230 | error_nosignal: | |
4028eeb9 DG |
3231 | /* Close received socket if valid. */ |
3232 | if (fd >= 0) { | |
3233 | if (close(fd)) { | |
3234 | PERROR("close received socket"); | |
3235 | } | |
3236 | } | |
cd2b09ed DG |
3237 | |
3238 | if (relayd_created) { | |
cd2b09ed DG |
3239 | free(relayd); |
3240 | } | |
3241 | ||
7735ef9e DG |
3242 | return ret; |
3243 | } | |
ca22feea | 3244 | |
4e9a4686 DG |
3245 | /* |
3246 | * Try to lock the stream mutex. | |
3247 | * | |
3248 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3249 | */ | |
3250 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3251 | { | |
3252 | int ret; | |
3253 | ||
3254 | assert(stream); | |
3255 | ||
3256 | /* | |
3257 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3258 | * being used else where hence there is data still being extracted. | |
3259 | */ | |
3260 | ret = pthread_mutex_trylock(&stream->lock); | |
3261 | if (ret) { | |
3262 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3263 | ret = 0; | |
3264 | goto end; | |
3265 | } | |
3266 | ||
3267 | ret = 1; | |
3268 | ||
3269 | end: | |
3270 | return ret; | |
3271 | } | |
3272 | ||
f7079f67 DG |
3273 | /* |
3274 | * Search for a relayd associated to the session id and return the reference. | |
3275 | * | |
3276 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3277 | * until the relayd object is no longer necessary. | |
3278 | */ | |
3279 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3280 | { | |
3281 | struct lttng_ht_iter iter; | |
f7079f67 | 3282 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3283 | |
3284 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3285 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3286 | node.node) { | |
18261bd1 DG |
3287 | /* |
3288 | * Check by sessiond id which is unique here where the relayd session | |
3289 | * id might not be when having multiple relayd. | |
3290 | */ | |
3291 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3292 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3293 | goto found; |
f7079f67 DG |
3294 | } |
3295 | } | |
3296 | ||
18261bd1 DG |
3297 | return NULL; |
3298 | ||
3299 | found: | |
f7079f67 DG |
3300 | return relayd; |
3301 | } | |
3302 | ||
ca22feea DG |
3303 | /* |
3304 | * Check if for a given session id there is still data needed to be extract | |
3305 | * from the buffers. | |
3306 | * | |
6d805429 | 3307 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3308 | */ |
6d805429 | 3309 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3310 | { |
3311 | int ret; | |
3312 | struct lttng_ht_iter iter; | |
3313 | struct lttng_ht *ht; | |
3314 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3315 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3316 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3317 | |
6d805429 | 3318 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3319 | |
6f6eda74 | 3320 | rcu_read_lock(); |
ca22feea DG |
3321 | pthread_mutex_lock(&consumer_data.lock); |
3322 | ||
3323 | switch (consumer_data.type) { | |
3324 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3325 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3326 | break; |
3327 | case LTTNG_CONSUMER32_UST: | |
3328 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3329 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3330 | break; |
3331 | default: | |
3332 | ERR("Unknown consumer data type"); | |
3333 | assert(0); | |
3334 | } | |
3335 | ||
3336 | /* Ease our life a bit */ | |
3337 | ht = consumer_data.stream_list_ht; | |
3338 | ||
f7079f67 DG |
3339 | relayd = find_relayd_by_session_id(id); |
3340 | if (relayd) { | |
3341 | /* Send init command for data pending. */ | |
3342 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3343 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3344 | relayd->relayd_session_id); | |
3345 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3346 | if (ret < 0) { | |
3347 | /* Communication error thus the relayd so no data pending. */ | |
3348 | goto data_not_pending; | |
3349 | } | |
3350 | } | |
3351 | ||
c8f59ee5 | 3352 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3353 | ht->hash_fct(&id, lttng_ht_seed), |
3354 | ht->match_fct, &id, | |
ca22feea | 3355 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3356 | /* If this call fails, the stream is being used hence data pending. */ |
3357 | ret = stream_try_lock(stream); | |
3358 | if (!ret) { | |
f7079f67 | 3359 | goto data_pending; |
ca22feea | 3360 | } |
ca22feea | 3361 | |
4e9a4686 DG |
3362 | /* |
3363 | * A removed node from the hash table indicates that the stream has | |
3364 | * been deleted thus having a guarantee that the buffers are closed | |
3365 | * on the consumer side. However, data can still be transmitted | |
3366 | * over the network so don't skip the relayd check. | |
3367 | */ | |
3368 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3369 | if (!ret) { | |
3370 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3371 | ret = data_pending(stream); |
3372 | if (ret == 1) { | |
4e9a4686 | 3373 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3374 | goto data_pending; |
4e9a4686 DG |
3375 | } |
3376 | } | |
3377 | ||
3378 | /* Relayd check */ | |
f7079f67 | 3379 | if (relayd) { |
c8f59ee5 DG |
3380 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3381 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3382 | ret = relayd_quiescent_control(&relayd->control_sock, |
3383 | stream->relayd_stream_id); | |
c8f59ee5 | 3384 | } else { |
6d805429 | 3385 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3386 | stream->relayd_stream_id, |
3387 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3388 | } |
3389 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3390 | if (ret == 1) { |
4e9a4686 | 3391 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3392 | goto data_pending; |
c8f59ee5 DG |
3393 | } |
3394 | } | |
4e9a4686 | 3395 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3396 | } |
ca22feea | 3397 | |
f7079f67 DG |
3398 | if (relayd) { |
3399 | unsigned int is_data_inflight = 0; | |
3400 | ||
3401 | /* Send init command for data pending. */ | |
3402 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3403 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3404 | relayd->relayd_session_id, &is_data_inflight); | |
3405 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3406 | if (ret < 0) { |
f7079f67 DG |
3407 | goto data_not_pending; |
3408 | } | |
bdd88757 DG |
3409 | if (is_data_inflight) { |
3410 | goto data_pending; | |
3411 | } | |
f7079f67 DG |
3412 | } |
3413 | ||
ca22feea | 3414 | /* |
f7079f67 DG |
3415 | * Finding _no_ node in the hash table and no inflight data means that the |
3416 | * stream(s) have been removed thus data is guaranteed to be available for | |
3417 | * analysis from the trace files. | |
ca22feea DG |
3418 | */ |
3419 | ||
f7079f67 | 3420 | data_not_pending: |
ca22feea DG |
3421 | /* Data is available to be read by a viewer. */ |
3422 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3423 | rcu_read_unlock(); |
6d805429 | 3424 | return 0; |
ca22feea | 3425 | |
f7079f67 | 3426 | data_pending: |
ca22feea DG |
3427 | /* Data is still being extracted from buffers. */ |
3428 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3429 | rcu_read_unlock(); |
6d805429 | 3430 | return 1; |
ca22feea | 3431 | } |
f50f23d9 DG |
3432 | |
3433 | /* | |
3434 | * Send a ret code status message to the sessiond daemon. | |
3435 | * | |
3436 | * Return the sendmsg() return value. | |
3437 | */ | |
3438 | int consumer_send_status_msg(int sock, int ret_code) | |
3439 | { | |
3440 | struct lttcomm_consumer_status_msg msg; | |
3441 | ||
3442 | msg.ret_code = ret_code; | |
3443 | ||
3444 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3445 | } | |
ffe60014 DG |
3446 | |
3447 | /* | |
3448 | * Send a channel status message to the sessiond daemon. | |
3449 | * | |
3450 | * Return the sendmsg() return value. | |
3451 | */ | |
3452 | int consumer_send_status_channel(int sock, | |
3453 | struct lttng_consumer_channel *channel) | |
3454 | { | |
3455 | struct lttcomm_consumer_status_channel msg; | |
3456 | ||
3457 | assert(sock >= 0); | |
3458 | ||
3459 | if (!channel) { | |
3460 | msg.ret_code = -LTTNG_ERR_UST_CHAN_FAIL; | |
3461 | } else { | |
3462 | msg.ret_code = LTTNG_OK; | |
3463 | msg.key = channel->key; | |
3464 | msg.stream_count = channel->streams.count; | |
3465 | } | |
3466 | ||
3467 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3468 | } |