Commit | Line | Data |
---|---|---|
3bd1e081 | 1 | /* |
90c106c6 | 2 | * Copyright (C) 2011 EfficiOS Inc. |
ab5be9fa MJ |
3 | * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
4 | * Copyright (C) 2012 David Goulet <dgoulet@efficios.com> | |
3bd1e081 | 5 | * |
ab5be9fa | 6 | * SPDX-License-Identifier: GPL-2.0-only |
3bd1e081 | 7 | * |
3bd1e081 MD |
8 | */ |
9 | ||
6f9449c2 | 10 | #include "common/index/ctf-index.h" |
fe7bf564 | 11 | #include <stdint.h> |
6c1c0768 | 12 | #define _LGPL_SOURCE |
3bd1e081 | 13 | #include <assert.h> |
3bd1e081 MD |
14 | #include <poll.h> |
15 | #include <pthread.h> | |
16 | #include <stdlib.h> | |
17 | #include <string.h> | |
18 | #include <sys/mman.h> | |
19 | #include <sys/socket.h> | |
20 | #include <sys/types.h> | |
21 | #include <unistd.h> | |
77c7c900 | 22 | #include <inttypes.h> |
331744e3 | 23 | #include <signal.h> |
3bd1e081 | 24 | |
51a9e1c7 | 25 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 26 | #include <common/common.h> |
fb3a43a9 | 27 | #include <common/utils.h> |
d2956687 | 28 | #include <common/time.h> |
fb3a43a9 | 29 | #include <common/compat/poll.h> |
f263b7fd | 30 | #include <common/compat/endian.h> |
309167d2 | 31 | #include <common/index/index.h> |
10a8a223 | 32 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 33 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
34 | #include <common/sessiond-comm/sessiond-comm.h> |
35 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 36 | #include <common/relayd/relayd.h> |
10a8a223 | 37 | #include <common/ust-consumer/ust-consumer.h> |
c8fea79c JR |
38 | #include <common/consumer/consumer-timer.h> |
39 | #include <common/consumer/consumer.h> | |
40 | #include <common/consumer/consumer-stream.h> | |
41 | #include <common/consumer/consumer-testpoint.h> | |
42 | #include <common/align.h> | |
5feafd41 | 43 | #include <common/consumer/consumer-metadata-cache.h> |
d2956687 JG |
44 | #include <common/trace-chunk.h> |
45 | #include <common/trace-chunk-registry.h> | |
46 | #include <common/string-utils/format.h> | |
c35f9726 | 47 | #include <common/dynamic-array.h> |
3bd1e081 | 48 | |
fa29bfbf SM |
49 | struct lttng_consumer_global_data the_consumer_data = { |
50 | .stream_count = 0, | |
51 | .need_update = 1, | |
52 | .type = LTTNG_CONSUMER_UNKNOWN, | |
3bd1e081 MD |
53 | }; |
54 | ||
d8ef542d MD |
55 | enum consumer_channel_action { |
56 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 57 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
58 | CONSUMER_CHANNEL_QUIT, |
59 | }; | |
60 | ||
61 | struct consumer_channel_msg { | |
62 | enum consumer_channel_action action; | |
a0cbdd2e MD |
63 | struct lttng_consumer_channel *chan; /* add */ |
64 | uint64_t key; /* del */ | |
d8ef542d MD |
65 | }; |
66 | ||
80957876 | 67 | /* Flag used to temporarily pause data consumption from testpoints. */ |
cf0bcb51 JG |
68 | int data_consumption_paused; |
69 | ||
3bd1e081 MD |
70 | /* |
71 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
72 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
73 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
74 | * polling threads. | |
75 | */ | |
10211f5c | 76 | int consumer_quit; |
3bd1e081 | 77 | |
43c34bc3 | 78 | /* |
43c34bc3 DG |
79 | * Global hash table containing respectively metadata and data streams. The |
80 | * stream element in this ht should only be updated by the metadata poll thread | |
81 | * for the metadata and the data poll thread for the data. | |
82 | */ | |
40dc48e0 DG |
83 | static struct lttng_ht *metadata_ht; |
84 | static struct lttng_ht *data_ht; | |
43c34bc3 | 85 | |
5da88b0f MD |
86 | static const char *get_consumer_domain(void) |
87 | { | |
fa29bfbf | 88 | switch (the_consumer_data.type) { |
5da88b0f MD |
89 | case LTTNG_CONSUMER_KERNEL: |
90 | return DEFAULT_KERNEL_TRACE_DIR; | |
91 | case LTTNG_CONSUMER64_UST: | |
92 | /* Fall-through. */ | |
93 | case LTTNG_CONSUMER32_UST: | |
94 | return DEFAULT_UST_TRACE_DIR; | |
95 | default: | |
96 | abort(); | |
97 | } | |
98 | } | |
99 | ||
acdb9057 DG |
100 | /* |
101 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
102 | * global state has changed so we just send back the thread in a poll wait | |
103 | * call. | |
104 | */ | |
105 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
106 | { | |
107 | struct lttng_consumer_stream *null_stream = NULL; | |
108 | ||
109 | assert(pipe); | |
110 | ||
111 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
112 | } | |
113 | ||
5c635c72 MD |
114 | static void notify_health_quit_pipe(int *pipe) |
115 | { | |
6cd525e8 | 116 | ssize_t ret; |
5c635c72 | 117 | |
6cd525e8 MD |
118 | ret = lttng_write(pipe[1], "4", 1); |
119 | if (ret < 1) { | |
5c635c72 MD |
120 | PERROR("write consumer health quit"); |
121 | } | |
122 | } | |
123 | ||
d8ef542d MD |
124 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
125 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 126 | uint64_t key, |
d8ef542d MD |
127 | enum consumer_channel_action action) |
128 | { | |
129 | struct consumer_channel_msg msg; | |
6cd525e8 | 130 | ssize_t ret; |
d8ef542d | 131 | |
e56251fc DG |
132 | memset(&msg, 0, sizeof(msg)); |
133 | ||
d8ef542d MD |
134 | msg.action = action; |
135 | msg.chan = chan; | |
f21dae48 | 136 | msg.key = key; |
6cd525e8 MD |
137 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
138 | if (ret < sizeof(msg)) { | |
139 | PERROR("notify_channel_pipe write error"); | |
140 | } | |
d8ef542d MD |
141 | } |
142 | ||
a0cbdd2e MD |
143 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
144 | uint64_t key) | |
145 | { | |
146 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
147 | } | |
148 | ||
d8ef542d MD |
149 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
150 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 151 | uint64_t *key, |
d8ef542d MD |
152 | enum consumer_channel_action *action) |
153 | { | |
154 | struct consumer_channel_msg msg; | |
6cd525e8 | 155 | ssize_t ret; |
d8ef542d | 156 | |
6cd525e8 MD |
157 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
158 | if (ret < sizeof(msg)) { | |
159 | ret = -1; | |
160 | goto error; | |
d8ef542d | 161 | } |
6cd525e8 MD |
162 | *action = msg.action; |
163 | *chan = msg.chan; | |
164 | *key = msg.key; | |
165 | error: | |
166 | return (int) ret; | |
d8ef542d MD |
167 | } |
168 | ||
212d67a2 DG |
169 | /* |
170 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
171 | * visible | |
172 | */ | |
173 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
174 | { | |
175 | struct lttng_consumer_stream *stream, *stmp; | |
176 | ||
177 | assert(channel); | |
178 | ||
179 | /* Delete streams that might have been left in the stream list. */ | |
180 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
181 | send_node) { | |
212d67a2 DG |
182 | /* |
183 | * Once a stream is added to this list, the buffers were created so we | |
184 | * have a guarantee that this call will succeed. Setting the monitor | |
185 | * mode to 0 so we don't lock nor try to delete the stream from the | |
186 | * global hash table. | |
187 | */ | |
188 | stream->monitor = 0; | |
189 | consumer_stream_destroy(stream, NULL); | |
190 | } | |
191 | } | |
192 | ||
3bd1e081 MD |
193 | /* |
194 | * Find a stream. The consumer_data.lock must be locked during this | |
195 | * call. | |
196 | */ | |
d88aee68 | 197 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 198 | struct lttng_ht *ht) |
3bd1e081 | 199 | { |
e4421fec | 200 | struct lttng_ht_iter iter; |
d88aee68 | 201 | struct lttng_ht_node_u64 *node; |
e4421fec | 202 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 203 | |
8389e4f8 DG |
204 | assert(ht); |
205 | ||
d88aee68 DG |
206 | /* -1ULL keys are lookup failures */ |
207 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 208 | return NULL; |
7a57cf92 | 209 | } |
e4421fec | 210 | |
6065ceec DG |
211 | rcu_read_lock(); |
212 | ||
d88aee68 DG |
213 | lttng_ht_lookup(ht, &key, &iter); |
214 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
215 | if (node != NULL) { |
216 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 217 | } |
e4421fec | 218 | |
6065ceec DG |
219 | rcu_read_unlock(); |
220 | ||
e4421fec | 221 | return stream; |
3bd1e081 MD |
222 | } |
223 | ||
da009f2c | 224 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
225 | { |
226 | struct lttng_consumer_stream *stream; | |
227 | ||
04253271 | 228 | rcu_read_lock(); |
ffe60014 | 229 | stream = find_stream(key, ht); |
04253271 | 230 | if (stream) { |
da009f2c | 231 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
232 | /* |
233 | * We don't want the lookup to match, but we still need | |
234 | * to iterate on this stream when iterating over the hash table. Just | |
235 | * change the node key. | |
236 | */ | |
da009f2c | 237 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
238 | } |
239 | rcu_read_unlock(); | |
7ad0a0cb MD |
240 | } |
241 | ||
d56db448 DG |
242 | /* |
243 | * Return a channel object for the given key. | |
244 | * | |
245 | * RCU read side lock MUST be acquired before calling this function and | |
246 | * protects the channel ptr. | |
247 | */ | |
d88aee68 | 248 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 249 | { |
e4421fec | 250 | struct lttng_ht_iter iter; |
d88aee68 | 251 | struct lttng_ht_node_u64 *node; |
e4421fec | 252 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 253 | |
d88aee68 DG |
254 | /* -1ULL keys are lookup failures */ |
255 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 256 | return NULL; |
7a57cf92 | 257 | } |
e4421fec | 258 | |
fa29bfbf | 259 | lttng_ht_lookup(the_consumer_data.channel_ht, &key, &iter); |
d88aee68 | 260 | node = lttng_ht_iter_get_node_u64(&iter); |
e4421fec DG |
261 | if (node != NULL) { |
262 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 263 | } |
e4421fec DG |
264 | |
265 | return channel; | |
3bd1e081 MD |
266 | } |
267 | ||
b5a6470f DG |
268 | /* |
269 | * There is a possibility that the consumer does not have enough time between | |
270 | * the close of the channel on the session daemon and the cleanup in here thus | |
271 | * once we have a channel add with an existing key, we know for sure that this | |
272 | * channel will eventually get cleaned up by all streams being closed. | |
273 | * | |
274 | * This function just nullifies the already existing channel key. | |
275 | */ | |
276 | static void steal_channel_key(uint64_t key) | |
277 | { | |
278 | struct lttng_consumer_channel *channel; | |
279 | ||
280 | rcu_read_lock(); | |
281 | channel = consumer_find_channel(key); | |
282 | if (channel) { | |
283 | channel->key = (uint64_t) -1ULL; | |
284 | /* | |
285 | * We don't want the lookup to match, but we still need to iterate on | |
286 | * this channel when iterating over the hash table. Just change the | |
287 | * node key. | |
288 | */ | |
289 | channel->node.key = (uint64_t) -1ULL; | |
290 | } | |
291 | rcu_read_unlock(); | |
292 | } | |
293 | ||
ffe60014 | 294 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 295 | { |
d88aee68 DG |
296 | struct lttng_ht_node_u64 *node = |
297 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
298 | struct lttng_consumer_channel *channel = |
299 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 300 | |
fa29bfbf | 301 | switch (the_consumer_data.type) { |
b83e03c4 MD |
302 | case LTTNG_CONSUMER_KERNEL: |
303 | break; | |
304 | case LTTNG_CONSUMER32_UST: | |
305 | case LTTNG_CONSUMER64_UST: | |
306 | lttng_ustconsumer_free_channel(channel); | |
307 | break; | |
308 | default: | |
309 | ERR("Unknown consumer_data type"); | |
310 | abort(); | |
311 | } | |
ffe60014 | 312 | free(channel); |
702b1ea4 MD |
313 | } |
314 | ||
00e2e675 DG |
315 | /* |
316 | * RCU protected relayd socket pair free. | |
317 | */ | |
ffe60014 | 318 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 319 | { |
d88aee68 DG |
320 | struct lttng_ht_node_u64 *node = |
321 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
322 | struct consumer_relayd_sock_pair *relayd = |
323 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
324 | ||
8994307f DG |
325 | /* |
326 | * Close all sockets. This is done in the call RCU since we don't want the | |
327 | * socket fds to be reassigned thus potentially creating bad state of the | |
328 | * relayd object. | |
329 | * | |
330 | * We do not have to lock the control socket mutex here since at this stage | |
331 | * there is no one referencing to this relayd object. | |
332 | */ | |
333 | (void) relayd_close(&relayd->control_sock); | |
334 | (void) relayd_close(&relayd->data_sock); | |
335 | ||
3a84e2f3 | 336 | pthread_mutex_destroy(&relayd->ctrl_sock_mutex); |
00e2e675 DG |
337 | free(relayd); |
338 | } | |
339 | ||
340 | /* | |
341 | * Destroy and free relayd socket pair object. | |
00e2e675 | 342 | */ |
51230d70 | 343 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
344 | { |
345 | int ret; | |
346 | struct lttng_ht_iter iter; | |
347 | ||
173af62f DG |
348 | if (relayd == NULL) { |
349 | return; | |
350 | } | |
351 | ||
00e2e675 DG |
352 | DBG("Consumer destroy and close relayd socket pair"); |
353 | ||
354 | iter.iter.node = &relayd->node.node; | |
fa29bfbf | 355 | ret = lttng_ht_del(the_consumer_data.relayd_ht, &iter); |
173af62f | 356 | if (ret != 0) { |
8994307f | 357 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
358 | return; |
359 | } | |
00e2e675 | 360 | |
00e2e675 | 361 | /* RCU free() call */ |
ffe60014 DG |
362 | call_rcu(&relayd->node.head, free_relayd_rcu); |
363 | } | |
364 | ||
365 | /* | |
366 | * Remove a channel from the global list protected by a mutex. This function is | |
367 | * also responsible for freeing its data structures. | |
368 | */ | |
369 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
370 | { | |
ffe60014 DG |
371 | struct lttng_ht_iter iter; |
372 | ||
d88aee68 | 373 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 | 374 | |
fa29bfbf | 375 | pthread_mutex_lock(&the_consumer_data.lock); |
a9838785 | 376 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 377 | |
212d67a2 DG |
378 | /* Destroy streams that might have been left in the stream list. */ |
379 | clean_channel_stream_list(channel); | |
51e762e5 | 380 | |
d3e2ba59 JD |
381 | if (channel->live_timer_enabled == 1) { |
382 | consumer_timer_live_stop(channel); | |
383 | } | |
e9404c27 JG |
384 | if (channel->monitor_timer_enabled == 1) { |
385 | consumer_timer_monitor_stop(channel); | |
386 | } | |
d3e2ba59 | 387 | |
fa29bfbf | 388 | switch (the_consumer_data.type) { |
ffe60014 DG |
389 | case LTTNG_CONSUMER_KERNEL: |
390 | break; | |
391 | case LTTNG_CONSUMER32_UST: | |
392 | case LTTNG_CONSUMER64_UST: | |
393 | lttng_ustconsumer_del_channel(channel); | |
394 | break; | |
395 | default: | |
396 | ERR("Unknown consumer_data type"); | |
397 | assert(0); | |
398 | goto end; | |
399 | } | |
400 | ||
d2956687 JG |
401 | lttng_trace_chunk_put(channel->trace_chunk); |
402 | channel->trace_chunk = NULL; | |
5c3892a6 | 403 | |
d2956687 JG |
404 | if (channel->is_published) { |
405 | int ret; | |
406 | ||
407 | rcu_read_lock(); | |
408 | iter.iter.node = &channel->node.node; | |
fa29bfbf | 409 | ret = lttng_ht_del(the_consumer_data.channel_ht, &iter); |
d2956687 | 410 | assert(!ret); |
ffe60014 | 411 | |
d2956687 | 412 | iter.iter.node = &channel->channels_by_session_id_ht_node.node; |
fa29bfbf | 413 | ret = lttng_ht_del(the_consumer_data.channels_by_session_id_ht, |
d2956687 JG |
414 | &iter); |
415 | assert(!ret); | |
416 | rcu_read_unlock(); | |
417 | } | |
418 | ||
b6921a17 JG |
419 | channel->is_deleted = true; |
420 | call_rcu(&channel->node.head, free_channel_rcu); | |
ffe60014 | 421 | end: |
a9838785 | 422 | pthread_mutex_unlock(&channel->lock); |
fa29bfbf | 423 | pthread_mutex_unlock(&the_consumer_data.lock); |
00e2e675 DG |
424 | } |
425 | ||
228b5bf7 DG |
426 | /* |
427 | * Iterate over the relayd hash table and destroy each element. Finally, | |
428 | * destroy the whole hash table. | |
429 | */ | |
430 | static void cleanup_relayd_ht(void) | |
431 | { | |
432 | struct lttng_ht_iter iter; | |
433 | struct consumer_relayd_sock_pair *relayd; | |
434 | ||
435 | rcu_read_lock(); | |
436 | ||
fa29bfbf SM |
437 | cds_lfht_for_each_entry(the_consumer_data.relayd_ht->ht, &iter.iter, |
438 | relayd, node.node) { | |
51230d70 | 439 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
440 | } |
441 | ||
228b5bf7 | 442 | rcu_read_unlock(); |
36b588ed | 443 | |
fa29bfbf | 444 | lttng_ht_destroy(the_consumer_data.relayd_ht); |
228b5bf7 DG |
445 | } |
446 | ||
8994307f DG |
447 | /* |
448 | * Update the end point status of all streams having the given network sequence | |
449 | * index (relayd index). | |
450 | * | |
451 | * It's atomically set without having the stream mutex locked which is fine | |
452 | * because we handle the write/read race with a pipe wakeup for each thread. | |
453 | */ | |
da009f2c | 454 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
455 | enum consumer_endpoint_status status) |
456 | { | |
457 | struct lttng_ht_iter iter; | |
458 | struct lttng_consumer_stream *stream; | |
459 | ||
da009f2c | 460 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
461 | |
462 | rcu_read_lock(); | |
463 | ||
464 | /* Let's begin with metadata */ | |
465 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
466 | if (stream->net_seq_idx == net_seq_idx) { | |
467 | uatomic_set(&stream->endpoint_status, status); | |
95671f53 JG |
468 | lttng_wait_queue_wake_all(&stream->chan->metadata_pushed_wait_queue); |
469 | ||
8994307f DG |
470 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); |
471 | } | |
472 | } | |
473 | ||
474 | /* Follow up by the data streams */ | |
475 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
476 | if (stream->net_seq_idx == net_seq_idx) { | |
477 | uatomic_set(&stream->endpoint_status, status); | |
478 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
479 | } | |
480 | } | |
481 | rcu_read_unlock(); | |
482 | } | |
483 | ||
484 | /* | |
485 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
486 | * destroying the object meaning removing it from the relayd hash table, | |
487 | * closing the sockets and freeing the memory in a RCU call. | |
488 | * | |
489 | * If a local data context is available, notify the threads that the streams' | |
490 | * state have changed. | |
491 | */ | |
9276e5c8 | 492 | void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd) |
8994307f | 493 | { |
da009f2c | 494 | uint64_t netidx; |
8994307f DG |
495 | |
496 | assert(relayd); | |
497 | ||
9276e5c8 | 498 | DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx); |
9617607b | 499 | |
8994307f DG |
500 | /* Save the net sequence index before destroying the object */ |
501 | netidx = relayd->net_seq_idx; | |
502 | ||
503 | /* | |
504 | * Delete the relayd from the relayd hash table, close the sockets and free | |
505 | * the object in a RCU call. | |
506 | */ | |
51230d70 | 507 | consumer_destroy_relayd(relayd); |
8994307f DG |
508 | |
509 | /* Set inactive endpoint to all streams */ | |
510 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
511 | ||
512 | /* | |
513 | * With a local data context, notify the threads that the streams' state | |
514 | * have changed. The write() action on the pipe acts as an "implicit" | |
515 | * memory barrier ordering the updates of the end point status from the | |
516 | * read of this status which happens AFTER receiving this notify. | |
517 | */ | |
9276e5c8 JR |
518 | notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe); |
519 | notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe); | |
8994307f DG |
520 | } |
521 | ||
a6ba4fe1 DG |
522 | /* |
523 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
524 | * reaches zero. | |
525 | * | |
526 | * RCU read side lock MUST be aquired before calling this function. | |
527 | */ | |
528 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
529 | { | |
530 | assert(relayd); | |
531 | ||
532 | /* Set destroy flag for this object */ | |
533 | uatomic_set(&relayd->destroy_flag, 1); | |
534 | ||
535 | /* Destroy the relayd if refcount is 0 */ | |
536 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 537 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
538 | } |
539 | } | |
540 | ||
3bd1e081 | 541 | /* |
1d1a276c DG |
542 | * Completly destroy stream from every visiable data structure and the given |
543 | * hash table if one. | |
544 | * | |
545 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 546 | */ |
e316aad5 DG |
547 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
548 | struct lttng_ht *ht) | |
3bd1e081 | 549 | { |
1d1a276c | 550 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
551 | } |
552 | ||
5ab66908 MD |
553 | /* |
554 | * XXX naming of del vs destroy is all mixed up. | |
555 | */ | |
556 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
557 | { | |
558 | consumer_stream_destroy(stream, data_ht); | |
559 | } | |
560 | ||
561 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
562 | { | |
563 | consumer_stream_destroy(stream, metadata_ht); | |
564 | } | |
565 | ||
d9a2e16e JD |
566 | void consumer_stream_update_channel_attributes( |
567 | struct lttng_consumer_stream *stream, | |
568 | struct lttng_consumer_channel *channel) | |
569 | { | |
570 | stream->channel_read_only_attributes.tracefile_size = | |
571 | channel->tracefile_size; | |
d9a2e16e JD |
572 | } |
573 | ||
3bd1e081 MD |
574 | /* |
575 | * Add a stream to the global list protected by a mutex. | |
576 | */ | |
66d583dc | 577 | void consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 578 | { |
5ab66908 | 579 | struct lttng_ht *ht = data_ht; |
3bd1e081 | 580 | |
e316aad5 | 581 | assert(stream); |
43c34bc3 | 582 | assert(ht); |
c77fc10a | 583 | |
d88aee68 | 584 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 | 585 | |
fa29bfbf | 586 | pthread_mutex_lock(&the_consumer_data.lock); |
a9838785 | 587 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 588 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 589 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 590 | rcu_read_lock(); |
e316aad5 | 591 | |
43c34bc3 | 592 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 593 | steal_stream_key(stream->key, ht); |
43c34bc3 | 594 | |
d88aee68 | 595 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 596 | |
fa29bfbf | 597 | lttng_ht_add_u64(the_consumer_data.stream_per_chan_id_ht, |
d8ef542d MD |
598 | &stream->node_channel_id); |
599 | ||
ca22feea DG |
600 | /* |
601 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
602 | * the key since the HT does not use it and we allow to add redundant keys | |
603 | * into this table. | |
604 | */ | |
fa29bfbf SM |
605 | lttng_ht_add_u64(the_consumer_data.stream_list_ht, |
606 | &stream->node_session_id); | |
ca22feea | 607 | |
e316aad5 | 608 | /* |
ffe60014 DG |
609 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
610 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
611 | * causes the count to become 0 also causes a stream to be added. The |
612 | * channel deletion will thus be triggered by the following removal of this | |
613 | * stream. | |
614 | */ | |
ffe60014 | 615 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
616 | /* Increment refcount before decrementing nb_init_stream_left */ |
617 | cmm_smp_wmb(); | |
ffe60014 | 618 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
619 | } |
620 | ||
621 | /* Update consumer data once the node is inserted. */ | |
fa29bfbf SM |
622 | the_consumer_data.stream_count++; |
623 | the_consumer_data.need_update = 1; | |
3bd1e081 | 624 | |
e316aad5 | 625 | rcu_read_unlock(); |
2e818a6a | 626 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 627 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 628 | pthread_mutex_unlock(&stream->chan->lock); |
fa29bfbf | 629 | pthread_mutex_unlock(&the_consumer_data.lock); |
3bd1e081 MD |
630 | } |
631 | ||
00e2e675 | 632 | /* |
3f8e211f DG |
633 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
634 | * be acquired before calling this. | |
00e2e675 | 635 | */ |
d09e1200 | 636 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
637 | { |
638 | int ret = 0; | |
d88aee68 | 639 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
640 | struct lttng_ht_iter iter; |
641 | ||
ffe60014 | 642 | assert(relayd); |
00e2e675 | 643 | |
fa29bfbf SM |
644 | lttng_ht_lookup(the_consumer_data.relayd_ht, &relayd->net_seq_idx, |
645 | &iter); | |
d88aee68 | 646 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 | 647 | if (node != NULL) { |
00e2e675 DG |
648 | goto end; |
649 | } | |
fa29bfbf | 650 | lttng_ht_add_unique_u64(the_consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 651 | |
00e2e675 DG |
652 | end: |
653 | return ret; | |
654 | } | |
655 | ||
656 | /* | |
657 | * Allocate and return a consumer relayd socket. | |
658 | */ | |
027a694f | 659 | static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( |
da009f2c | 660 | uint64_t net_seq_idx) |
00e2e675 DG |
661 | { |
662 | struct consumer_relayd_sock_pair *obj = NULL; | |
663 | ||
da009f2c MD |
664 | /* net sequence index of -1 is a failure */ |
665 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
666 | goto error; |
667 | } | |
668 | ||
669 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
670 | if (obj == NULL) { | |
671 | PERROR("zmalloc relayd sock"); | |
672 | goto error; | |
673 | } | |
674 | ||
675 | obj->net_seq_idx = net_seq_idx; | |
676 | obj->refcount = 0; | |
173af62f | 677 | obj->destroy_flag = 0; |
f96e4545 MD |
678 | obj->control_sock.sock.fd = -1; |
679 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 680 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
681 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
682 | ||
683 | error: | |
684 | return obj; | |
685 | } | |
686 | ||
687 | /* | |
688 | * Find a relayd socket pair in the global consumer data. | |
689 | * | |
690 | * Return the object if found else NULL. | |
b0b335c8 MD |
691 | * RCU read-side lock must be held across this call and while using the |
692 | * returned object. | |
00e2e675 | 693 | */ |
d88aee68 | 694 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
695 | { |
696 | struct lttng_ht_iter iter; | |
d88aee68 | 697 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
698 | struct consumer_relayd_sock_pair *relayd = NULL; |
699 | ||
700 | /* Negative keys are lookup failures */ | |
d88aee68 | 701 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
702 | goto error; |
703 | } | |
704 | ||
fa29bfbf | 705 | lttng_ht_lookup(the_consumer_data.relayd_ht, &key, &iter); |
d88aee68 | 706 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
707 | if (node != NULL) { |
708 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
709 | } | |
710 | ||
00e2e675 DG |
711 | error: |
712 | return relayd; | |
713 | } | |
714 | ||
10a50311 JD |
715 | /* |
716 | * Find a relayd and send the stream | |
717 | * | |
718 | * Returns 0 on success, < 0 on error | |
719 | */ | |
720 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
721 | char *path) | |
722 | { | |
723 | int ret = 0; | |
724 | struct consumer_relayd_sock_pair *relayd; | |
725 | ||
726 | assert(stream); | |
727 | assert(stream->net_seq_idx != -1ULL); | |
728 | assert(path); | |
729 | ||
730 | /* The stream is not metadata. Get relayd reference if exists. */ | |
731 | rcu_read_lock(); | |
732 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
733 | if (relayd != NULL) { | |
734 | /* Add stream on the relayd */ | |
735 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
736 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
5da88b0f | 737 | get_consumer_domain(), path, &stream->relayd_stream_id, |
d2956687 JG |
738 | stream->chan->tracefile_size, |
739 | stream->chan->tracefile_count, | |
740 | stream->trace_chunk); | |
10a50311 JD |
741 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
742 | if (ret < 0) { | |
9276e5c8 JR |
743 | ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
744 | lttng_consumer_cleanup_relayd(relayd); | |
10a50311 JD |
745 | goto end; |
746 | } | |
1c20f0e2 | 747 | |
10a50311 | 748 | uatomic_inc(&relayd->refcount); |
d01178b6 | 749 | stream->sent_to_relayd = 1; |
10a50311 JD |
750 | } else { |
751 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
752 | stream->key, stream->net_seq_idx); | |
753 | ret = -1; | |
754 | goto end; | |
755 | } | |
756 | ||
757 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
758 | stream->name, stream->key, stream->net_seq_idx); | |
759 | ||
760 | end: | |
761 | rcu_read_unlock(); | |
762 | return ret; | |
763 | } | |
764 | ||
a4baae1b JD |
765 | /* |
766 | * Find a relayd and send the streams sent message | |
767 | * | |
768 | * Returns 0 on success, < 0 on error | |
769 | */ | |
770 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
771 | { | |
772 | int ret = 0; | |
773 | struct consumer_relayd_sock_pair *relayd; | |
774 | ||
775 | assert(net_seq_idx != -1ULL); | |
776 | ||
777 | /* The stream is not metadata. Get relayd reference if exists. */ | |
778 | rcu_read_lock(); | |
779 | relayd = consumer_find_relayd(net_seq_idx); | |
780 | if (relayd != NULL) { | |
781 | /* Add stream on the relayd */ | |
782 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
783 | ret = relayd_streams_sent(&relayd->control_sock); | |
784 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
785 | if (ret < 0) { | |
9276e5c8 JR |
786 | ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
787 | lttng_consumer_cleanup_relayd(relayd); | |
a4baae1b JD |
788 | goto end; |
789 | } | |
790 | } else { | |
791 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
792 | net_seq_idx); | |
793 | ret = -1; | |
794 | goto end; | |
795 | } | |
796 | ||
797 | ret = 0; | |
798 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
799 | ||
800 | end: | |
801 | rcu_read_unlock(); | |
802 | return ret; | |
803 | } | |
804 | ||
10a50311 JD |
805 | /* |
806 | * Find a relayd and close the stream | |
807 | */ | |
808 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
809 | { | |
810 | struct consumer_relayd_sock_pair *relayd; | |
811 | ||
812 | /* The stream is not metadata. Get relayd reference if exists. */ | |
813 | rcu_read_lock(); | |
814 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
815 | if (relayd) { | |
816 | consumer_stream_relayd_close(stream, relayd); | |
817 | } | |
818 | rcu_read_unlock(); | |
819 | } | |
820 | ||
00e2e675 DG |
821 | /* |
822 | * Handle stream for relayd transmission if the stream applies for network | |
823 | * streaming where the net sequence index is set. | |
824 | * | |
825 | * Return destination file descriptor or negative value on error. | |
826 | */ | |
6197aea7 | 827 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
828 | size_t data_size, unsigned long padding, |
829 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
830 | { |
831 | int outfd = -1, ret; | |
00e2e675 DG |
832 | struct lttcomm_relayd_data_hdr data_hdr; |
833 | ||
834 | /* Safety net */ | |
835 | assert(stream); | |
6197aea7 | 836 | assert(relayd); |
00e2e675 DG |
837 | |
838 | /* Reset data header */ | |
839 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
840 | ||
00e2e675 DG |
841 | if (stream->metadata_flag) { |
842 | /* Caller MUST acquire the relayd control socket lock */ | |
843 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
844 | if (ret < 0) { | |
845 | goto error; | |
846 | } | |
847 | ||
848 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 849 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
850 | } else { |
851 | /* Set header with stream information */ | |
852 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
853 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 854 | data_hdr.padding_size = htobe32(padding); |
c35f9726 | 855 | |
39df6d9f DG |
856 | /* |
857 | * Note that net_seq_num below is assigned with the *current* value of | |
858 | * next_net_seq_num and only after that the next_net_seq_num will be | |
859 | * increment. This is why when issuing a command on the relayd using | |
860 | * this next value, 1 should always be substracted in order to compare | |
861 | * the last seen sequence number on the relayd side to the last sent. | |
862 | */ | |
3604f373 | 863 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
864 | /* Other fields are zeroed previously */ |
865 | ||
866 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
867 | sizeof(data_hdr)); | |
868 | if (ret < 0) { | |
869 | goto error; | |
870 | } | |
871 | ||
3604f373 DG |
872 | ++stream->next_net_seq_num; |
873 | ||
00e2e675 | 874 | /* Set to go on data socket */ |
6151a90f | 875 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
876 | } |
877 | ||
878 | error: | |
879 | return outfd; | |
880 | } | |
881 | ||
b1316da1 JG |
882 | /* |
883 | * Write a character on the metadata poll pipe to wake the metadata thread. | |
884 | * Returns 0 on success, -1 on error. | |
885 | */ | |
886 | int consumer_metadata_wakeup_pipe(const struct lttng_consumer_channel *channel) | |
887 | { | |
888 | int ret = 0; | |
889 | ||
890 | DBG("Waking up metadata poll thread (writing to pipe): channel name = '%s'", | |
891 | channel->name); | |
892 | if (channel->monitor && channel->metadata_stream) { | |
893 | const char dummy = 'c'; | |
894 | const ssize_t write_ret = lttng_write( | |
895 | channel->metadata_stream->ust_metadata_poll_pipe[1], | |
896 | &dummy, 1); | |
897 | ||
898 | if (write_ret < 1) { | |
899 | if (errno == EWOULDBLOCK) { | |
900 | /* | |
901 | * This is fine, the metadata poll thread | |
902 | * is having a hard time keeping-up, but | |
903 | * it will eventually wake-up and consume | |
904 | * the available data. | |
905 | */ | |
906 | ret = 0; | |
907 | } else { | |
908 | PERROR("Failed to write to UST metadata pipe while attempting to wake-up the metadata poll thread"); | |
909 | ret = -1; | |
910 | goto end; | |
911 | } | |
912 | } | |
913 | } | |
914 | ||
915 | end: | |
916 | return ret; | |
917 | } | |
918 | ||
d2956687 JG |
919 | /* |
920 | * Trigger a dump of the metadata content. Following/during the succesful | |
921 | * completion of this call, the metadata poll thread will start receiving | |
922 | * metadata packets to consume. | |
923 | * | |
924 | * The caller must hold the channel and stream locks. | |
925 | */ | |
926 | static | |
927 | int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream) | |
928 | { | |
929 | int ret; | |
930 | ||
931 | ASSERT_LOCKED(stream->chan->lock); | |
932 | ASSERT_LOCKED(stream->lock); | |
933 | assert(stream->metadata_flag); | |
934 | assert(stream->chan->trace_chunk); | |
935 | ||
fa29bfbf | 936 | switch (the_consumer_data.type) { |
d2956687 JG |
937 | case LTTNG_CONSUMER_KERNEL: |
938 | /* | |
939 | * Reset the position of what has been read from the | |
940 | * metadata cache to 0 so we can dump it again. | |
941 | */ | |
942 | ret = kernctl_metadata_cache_dump(stream->wait_fd); | |
943 | break; | |
944 | case LTTNG_CONSUMER32_UST: | |
945 | case LTTNG_CONSUMER64_UST: | |
946 | /* | |
947 | * Reset the position pushed from the metadata cache so it | |
948 | * will write from the beginning on the next push. | |
949 | */ | |
950 | stream->ust_metadata_pushed = 0; | |
951 | ret = consumer_metadata_wakeup_pipe(stream->chan); | |
952 | break; | |
953 | default: | |
954 | ERR("Unknown consumer_data type"); | |
955 | abort(); | |
956 | } | |
957 | if (ret < 0) { | |
958 | ERR("Failed to dump the metadata cache"); | |
959 | } | |
960 | return ret; | |
961 | } | |
962 | ||
963 | static | |
964 | int lttng_consumer_channel_set_trace_chunk( | |
965 | struct lttng_consumer_channel *channel, | |
966 | struct lttng_trace_chunk *new_trace_chunk) | |
967 | { | |
d2956687 | 968 | pthread_mutex_lock(&channel->lock); |
b6921a17 JG |
969 | if (channel->is_deleted) { |
970 | /* | |
971 | * The channel has been logically deleted and should no longer | |
972 | * be used. It has released its reference to its current trace | |
973 | * chunk and should not acquire a new one. | |
974 | * | |
975 | * Return success as there is nothing for the caller to do. | |
976 | */ | |
977 | goto end; | |
978 | } | |
d2956687 JG |
979 | |
980 | /* | |
981 | * The acquisition of the reference cannot fail (barring | |
982 | * a severe internal error) since a reference to the published | |
983 | * chunk is already held by the caller. | |
984 | */ | |
985 | if (new_trace_chunk) { | |
986 | const bool acquired_reference = lttng_trace_chunk_get( | |
987 | new_trace_chunk); | |
988 | ||
989 | assert(acquired_reference); | |
990 | } | |
991 | ||
992 | lttng_trace_chunk_put(channel->trace_chunk); | |
993 | channel->trace_chunk = new_trace_chunk; | |
d2956687 JG |
994 | end: |
995 | pthread_mutex_unlock(&channel->lock); | |
ce1aa6fe | 996 | return 0; |
d2956687 JG |
997 | } |
998 | ||
3bd1e081 | 999 | /* |
ffe60014 DG |
1000 | * Allocate and return a new lttng_consumer_channel object using the given key |
1001 | * to initialize the hash table node. | |
1002 | * | |
1003 | * On error, return NULL. | |
3bd1e081 | 1004 | */ |
886224ff | 1005 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 | 1006 | uint64_t session_id, |
d2956687 | 1007 | const uint64_t *chunk_id, |
ffe60014 DG |
1008 | const char *pathname, |
1009 | const char *name, | |
57a269f2 | 1010 | uint64_t relayd_id, |
1624d5b7 JD |
1011 | enum lttng_event_output output, |
1012 | uint64_t tracefile_size, | |
2bba9e53 | 1013 | uint64_t tracefile_count, |
1950109e | 1014 | uint64_t session_id_per_pid, |
ecc48a90 | 1015 | unsigned int monitor, |
d7ba1388 | 1016 | unsigned int live_timer_interval, |
a2814ea7 | 1017 | bool is_in_live_session, |
3d071855 | 1018 | const char *root_shm_path, |
d7ba1388 | 1019 | const char *shm_path) |
3bd1e081 | 1020 | { |
d2956687 JG |
1021 | struct lttng_consumer_channel *channel = NULL; |
1022 | struct lttng_trace_chunk *trace_chunk = NULL; | |
1023 | ||
1024 | if (chunk_id) { | |
1025 | trace_chunk = lttng_trace_chunk_registry_find_chunk( | |
fa29bfbf | 1026 | the_consumer_data.chunk_registry, session_id, |
d2956687 JG |
1027 | *chunk_id); |
1028 | if (!trace_chunk) { | |
1029 | ERR("Failed to find trace chunk reference during creation of channel"); | |
1030 | goto end; | |
1031 | } | |
1032 | } | |
3bd1e081 | 1033 | |
276b26d1 | 1034 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 1035 | if (channel == NULL) { |
7a57cf92 | 1036 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
1037 | goto end; |
1038 | } | |
ffe60014 DG |
1039 | |
1040 | channel->key = key; | |
3bd1e081 | 1041 | channel->refcount = 0; |
ffe60014 | 1042 | channel->session_id = session_id; |
1950109e | 1043 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 | 1044 | channel->relayd_id = relayd_id; |
1624d5b7 JD |
1045 | channel->tracefile_size = tracefile_size; |
1046 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 1047 | channel->monitor = monitor; |
ecc48a90 | 1048 | channel->live_timer_interval = live_timer_interval; |
a2814ea7 | 1049 | channel->is_live = is_in_live_session; |
a9838785 | 1050 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 1051 | pthread_mutex_init(&channel->timer_lock, NULL); |
95671f53 | 1052 | lttng_wait_queue_init(&channel->metadata_pushed_wait_queue); |
ffe60014 | 1053 | |
0c759fc9 DG |
1054 | switch (output) { |
1055 | case LTTNG_EVENT_SPLICE: | |
1056 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
1057 | break; | |
1058 | case LTTNG_EVENT_MMAP: | |
1059 | channel->output = CONSUMER_CHANNEL_MMAP; | |
1060 | break; | |
1061 | default: | |
1062 | assert(0); | |
1063 | free(channel); | |
1064 | channel = NULL; | |
1065 | goto end; | |
1066 | } | |
1067 | ||
07b86b52 JD |
1068 | /* |
1069 | * In monitor mode, the streams associated with the channel will be put in | |
1070 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
1071 | * here meaning that the channel itself has streams that are referenced. | |
1072 | * | |
1073 | * On a channel deletion, once the channel is no longer visible, the | |
1074 | * refcount is decremented and checked for a zero value to delete it. With | |
1075 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
1076 | */ | |
1077 | if (!channel->monitor) { | |
1078 | channel->refcount = 1; | |
1079 | } | |
1080 | ||
ffe60014 DG |
1081 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1082 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1083 | ||
1084 | strncpy(channel->name, name, sizeof(channel->name)); | |
1085 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1086 | ||
3d071855 MD |
1087 | if (root_shm_path) { |
1088 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1089 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1090 | } | |
d7ba1388 MD |
1091 | if (shm_path) { |
1092 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1093 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1094 | } | |
1095 | ||
d88aee68 | 1096 | lttng_ht_node_init_u64(&channel->node, channel->key); |
5c3892a6 JG |
1097 | lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node, |
1098 | channel->session_id); | |
d8ef542d MD |
1099 | |
1100 | channel->wait_fd = -1; | |
ffe60014 DG |
1101 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1102 | ||
d2956687 JG |
1103 | if (trace_chunk) { |
1104 | int ret = lttng_consumer_channel_set_trace_chunk(channel, | |
1105 | trace_chunk); | |
1106 | if (ret) { | |
1107 | goto error; | |
1108 | } | |
1109 | } | |
1110 | ||
62a7b8ed | 1111 | DBG("Allocated channel (key %" PRIu64 ")", channel->key); |
3bd1e081 | 1112 | |
3bd1e081 | 1113 | end: |
d2956687 | 1114 | lttng_trace_chunk_put(trace_chunk); |
3bd1e081 | 1115 | return channel; |
d2956687 JG |
1116 | error: |
1117 | consumer_del_channel(channel); | |
1118 | channel = NULL; | |
1119 | goto end; | |
3bd1e081 MD |
1120 | } |
1121 | ||
1122 | /* | |
1123 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1124 | * |
b5a6470f | 1125 | * Always return 0 indicating success. |
3bd1e081 | 1126 | */ |
d8ef542d MD |
1127 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1128 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1129 | { |
fa29bfbf | 1130 | pthread_mutex_lock(&the_consumer_data.lock); |
a9838785 | 1131 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1132 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1133 | |
b5a6470f DG |
1134 | /* |
1135 | * This gives us a guarantee that the channel we are about to add to the | |
1136 | * channel hash table will be unique. See this function comment on the why | |
1137 | * we need to steel the channel key at this stage. | |
1138 | */ | |
1139 | steal_channel_key(channel->key); | |
c77fc10a | 1140 | |
b5a6470f | 1141 | rcu_read_lock(); |
fa29bfbf SM |
1142 | lttng_ht_add_unique_u64(the_consumer_data.channel_ht, &channel->node); |
1143 | lttng_ht_add_u64(the_consumer_data.channels_by_session_id_ht, | |
5c3892a6 | 1144 | &channel->channels_by_session_id_ht_node); |
6065ceec | 1145 | rcu_read_unlock(); |
d2956687 | 1146 | channel->is_published = true; |
b5a6470f | 1147 | |
ec6ea7d0 | 1148 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1149 | pthread_mutex_unlock(&channel->lock); |
fa29bfbf | 1150 | pthread_mutex_unlock(&the_consumer_data.lock); |
702b1ea4 | 1151 | |
b5a6470f | 1152 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1153 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1154 | } |
b5a6470f DG |
1155 | |
1156 | return 0; | |
3bd1e081 MD |
1157 | } |
1158 | ||
1159 | /* | |
1160 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1161 | * doing a lookup in the linked list and concurrency issues when writing is | |
1162 | * needed. Called with consumer_data.lock held. | |
1163 | * | |
1164 | * Returns the number of fds in the structures. | |
1165 | */ | |
ffe60014 DG |
1166 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1167 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
9a2fcf78 | 1168 | struct lttng_ht *ht, int *nb_inactive_fd) |
3bd1e081 | 1169 | { |
3bd1e081 | 1170 | int i = 0; |
e4421fec DG |
1171 | struct lttng_ht_iter iter; |
1172 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1173 | |
ffe60014 DG |
1174 | assert(ctx); |
1175 | assert(ht); | |
1176 | assert(pollfd); | |
1177 | assert(local_stream); | |
1178 | ||
3bd1e081 | 1179 | DBG("Updating poll fd array"); |
9a2fcf78 | 1180 | *nb_inactive_fd = 0; |
481d6c57 | 1181 | rcu_read_lock(); |
43c34bc3 | 1182 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1183 | /* |
1184 | * Only active streams with an active end point can be added to the | |
1185 | * poll set and local stream storage of the thread. | |
1186 | * | |
1187 | * There is a potential race here for endpoint_status to be updated | |
1188 | * just after the check. However, this is OK since the stream(s) will | |
1189 | * be deleted once the thread is notified that the end point state has | |
1190 | * changed where this function will be called back again. | |
9a2fcf78 JD |
1191 | * |
1192 | * We track the number of inactive FDs because they still need to be | |
1193 | * closed by the polling thread after a wakeup on the data_pipe or | |
1194 | * metadata_pipe. | |
8994307f | 1195 | */ |
d2956687 | 1196 | if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
9a2fcf78 | 1197 | (*nb_inactive_fd)++; |
3bd1e081 MD |
1198 | continue; |
1199 | } | |
7972aab2 DG |
1200 | /* |
1201 | * This clobbers way too much the debug output. Uncomment that if you | |
1202 | * need it for debugging purposes. | |
7972aab2 | 1203 | */ |
e4421fec | 1204 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1205 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1206 | local_stream[i] = stream; |
3bd1e081 MD |
1207 | i++; |
1208 | } | |
481d6c57 | 1209 | rcu_read_unlock(); |
3bd1e081 MD |
1210 | |
1211 | /* | |
50f8ae69 | 1212 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1213 | * increment i so nb_fd is the number of real FD. |
1214 | */ | |
acdb9057 | 1215 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1216 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1217 | |
1218 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1219 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1220 | return i; |
1221 | } | |
1222 | ||
1223 | /* | |
84382d49 MD |
1224 | * Poll on the should_quit pipe and the command socket return -1 on |
1225 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1226 | */ |
1227 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1228 | { | |
1229 | int num_rdy; | |
1230 | ||
88f2b785 | 1231 | restart: |
3bd1e081 MD |
1232 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1233 | if (num_rdy == -1) { | |
88f2b785 MD |
1234 | /* |
1235 | * Restart interrupted system call. | |
1236 | */ | |
1237 | if (errno == EINTR) { | |
1238 | goto restart; | |
1239 | } | |
7a57cf92 | 1240 | PERROR("Poll error"); |
84382d49 | 1241 | return -1; |
3bd1e081 | 1242 | } |
509bb1cf | 1243 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1244 | DBG("consumer_should_quit wake up"); |
84382d49 | 1245 | return 1; |
3bd1e081 MD |
1246 | } |
1247 | return 0; | |
3bd1e081 MD |
1248 | } |
1249 | ||
1250 | /* | |
1251 | * Set the error socket. | |
1252 | */ | |
ffe60014 DG |
1253 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1254 | int sock) | |
3bd1e081 MD |
1255 | { |
1256 | ctx->consumer_error_socket = sock; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Set the command socket path. | |
1261 | */ | |
3bd1e081 MD |
1262 | void lttng_consumer_set_command_sock_path( |
1263 | struct lttng_consumer_local_data *ctx, char *sock) | |
1264 | { | |
1265 | ctx->consumer_command_sock_path = sock; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * Send return code to the session daemon. | |
1270 | * If the socket is not defined, we return 0, it is not a fatal error | |
1271 | */ | |
a73d9496 JG |
1272 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, |
1273 | enum lttcomm_return_code error_code) | |
3bd1e081 MD |
1274 | { |
1275 | if (ctx->consumer_error_socket > 0) { | |
a73d9496 JG |
1276 | const int32_t comm_code = (int32_t) error_code; |
1277 | ||
1278 | return lttcomm_send_unix_sock( | |
1279 | ctx->consumer_error_socket, &comm_code, sizeof(comm_code)); | |
3bd1e081 MD |
1280 | } |
1281 | ||
1282 | return 0; | |
1283 | } | |
1284 | ||
1285 | /* | |
228b5bf7 DG |
1286 | * Close all the tracefiles and stream fds and MUST be called when all |
1287 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1288 | */ |
1289 | void lttng_consumer_cleanup(void) | |
1290 | { | |
e4421fec | 1291 | struct lttng_ht_iter iter; |
ffe60014 | 1292 | struct lttng_consumer_channel *channel; |
e10aec8f | 1293 | unsigned int trace_chunks_left; |
6065ceec DG |
1294 | |
1295 | rcu_read_lock(); | |
3bd1e081 | 1296 | |
fa29bfbf SM |
1297 | cds_lfht_for_each_entry(the_consumer_data.channel_ht->ht, &iter.iter, |
1298 | channel, node.node) { | |
702b1ea4 | 1299 | consumer_del_channel(channel); |
3bd1e081 | 1300 | } |
6065ceec DG |
1301 | |
1302 | rcu_read_unlock(); | |
d6ce1df2 | 1303 | |
fa29bfbf SM |
1304 | lttng_ht_destroy(the_consumer_data.channel_ht); |
1305 | lttng_ht_destroy(the_consumer_data.channels_by_session_id_ht); | |
228b5bf7 DG |
1306 | |
1307 | cleanup_relayd_ht(); | |
1308 | ||
fa29bfbf | 1309 | lttng_ht_destroy(the_consumer_data.stream_per_chan_id_ht); |
d8ef542d | 1310 | |
228b5bf7 DG |
1311 | /* |
1312 | * This HT contains streams that are freed by either the metadata thread or | |
1313 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1314 | * it. | |
1315 | */ | |
fa29bfbf | 1316 | lttng_ht_destroy(the_consumer_data.stream_list_ht); |
28cc88f3 | 1317 | |
e10aec8f MD |
1318 | /* |
1319 | * Trace chunks in the registry may still exist if the session | |
1320 | * daemon has encountered an internal error and could not | |
1321 | * tear down its sessions and/or trace chunks properly. | |
1322 | * | |
1323 | * Release the session daemon's implicit reference to any remaining | |
1324 | * trace chunk and print an error if any trace chunk was found. Note | |
1325 | * that there are _no_ legitimate cases for trace chunks to be left, | |
1326 | * it is a leak. However, it can happen following a crash of the | |
1327 | * session daemon and not emptying the registry would cause an assertion | |
1328 | * to hit. | |
1329 | */ | |
1330 | trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk( | |
fa29bfbf | 1331 | the_consumer_data.chunk_registry); |
e10aec8f MD |
1332 | if (trace_chunks_left) { |
1333 | ERR("%u trace chunks are leaked by lttng-consumerd. " | |
1334 | "This can be caused by an internal error of the session daemon.", | |
1335 | trace_chunks_left); | |
1336 | } | |
1337 | /* Run all callbacks freeing each chunk. */ | |
1338 | rcu_barrier(); | |
fa29bfbf | 1339 | lttng_trace_chunk_registry_destroy(the_consumer_data.chunk_registry); |
3bd1e081 MD |
1340 | } |
1341 | ||
1342 | /* | |
1343 | * Called from signal handler. | |
1344 | */ | |
1345 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1346 | { | |
6cd525e8 MD |
1347 | ssize_t ret; |
1348 | ||
10211f5c | 1349 | CMM_STORE_SHARED(consumer_quit, 1); |
6cd525e8 MD |
1350 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1351 | if (ret < 1) { | |
7a57cf92 | 1352 | PERROR("write consumer quit"); |
3bd1e081 | 1353 | } |
ab1027f4 DG |
1354 | |
1355 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1356 | } |
1357 | ||
5199ffc4 JG |
1358 | |
1359 | /* | |
1360 | * Flush pending writes to trace output disk file. | |
1361 | */ | |
1362 | static | |
00e2e675 DG |
1363 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1364 | off_t orig_offset) | |
3bd1e081 | 1365 | { |
c7a78aab | 1366 | int ret; |
3bd1e081 MD |
1367 | int outfd = stream->out_fd; |
1368 | ||
1369 | /* | |
1370 | * This does a blocking write-and-wait on any page that belongs to the | |
1371 | * subbuffer prior to the one we just wrote. | |
1372 | * Don't care about error values, as these are just hints and ways to | |
1373 | * limit the amount of page cache used. | |
1374 | */ | |
ffe60014 | 1375 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1376 | return; |
1377 | } | |
ffe60014 DG |
1378 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1379 | stream->max_sb_size, | |
3bd1e081 MD |
1380 | SYNC_FILE_RANGE_WAIT_BEFORE |
1381 | | SYNC_FILE_RANGE_WRITE | |
1382 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1383 | /* | |
1384 | * Give hints to the kernel about how we access the file: | |
1385 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1386 | * we write it. | |
1387 | * | |
1388 | * We need to call fadvise again after the file grows because the | |
1389 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1390 | * file. | |
1391 | * | |
1392 | * Call fadvise _after_ having waited for the page writeback to | |
1393 | * complete because the dirty page writeback semantic is not well | |
1394 | * defined. So it can be expected to lead to lower throughput in | |
1395 | * streaming. | |
1396 | */ | |
c7a78aab | 1397 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1398 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
a0d0e127 | 1399 | if (ret && ret != -ENOSYS) { |
a74a5f4a JG |
1400 | errno = ret; |
1401 | PERROR("posix_fadvise on fd %i", outfd); | |
c7a78aab | 1402 | } |
3bd1e081 MD |
1403 | } |
1404 | ||
1405 | /* | |
1406 | * Initialise the necessary environnement : | |
1407 | * - create a new context | |
1408 | * - create the poll_pipe | |
1409 | * - create the should_quit pipe (for signal handler) | |
1410 | * - create the thread pipe (for splice) | |
1411 | * | |
1412 | * Takes a function pointer as argument, this function is called when data is | |
1413 | * available on a buffer. This function is responsible to do the | |
1414 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1415 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1416 | * | |
1417 | * Returns a pointer to the new context or NULL on error. | |
1418 | */ | |
1419 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1420 | enum lttng_consumer_type type, | |
4078b776 | 1421 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
6f9449c2 | 1422 | struct lttng_consumer_local_data *ctx, bool locked_by_caller), |
3bd1e081 MD |
1423 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1424 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1425 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1426 | { |
d8ef542d | 1427 | int ret; |
3bd1e081 MD |
1428 | struct lttng_consumer_local_data *ctx; |
1429 | ||
fa29bfbf SM |
1430 | assert(the_consumer_data.type == LTTNG_CONSUMER_UNKNOWN || |
1431 | the_consumer_data.type == type); | |
1432 | the_consumer_data.type = type; | |
3bd1e081 | 1433 | |
effcf122 | 1434 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1435 | if (ctx == NULL) { |
7a57cf92 | 1436 | PERROR("allocating context"); |
3bd1e081 MD |
1437 | goto error; |
1438 | } | |
1439 | ||
1440 | ctx->consumer_error_socket = -1; | |
331744e3 | 1441 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1442 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1443 | /* assign the callbacks */ |
1444 | ctx->on_buffer_ready = buffer_ready; | |
1445 | ctx->on_recv_channel = recv_channel; | |
1446 | ctx->on_recv_stream = recv_stream; | |
1447 | ctx->on_update_stream = update_stream; | |
1448 | ||
acdb9057 DG |
1449 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1450 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1451 | goto error_poll_pipe; |
1452 | } | |
1453 | ||
02b3d176 DG |
1454 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1455 | if (!ctx->consumer_wakeup_pipe) { | |
1456 | goto error_wakeup_pipe; | |
1457 | } | |
1458 | ||
3bd1e081 MD |
1459 | ret = pipe(ctx->consumer_should_quit); |
1460 | if (ret < 0) { | |
7a57cf92 | 1461 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1462 | goto error_quit_pipe; |
1463 | } | |
1464 | ||
d8ef542d MD |
1465 | ret = pipe(ctx->consumer_channel_pipe); |
1466 | if (ret < 0) { | |
1467 | PERROR("Error creating channel pipe"); | |
1468 | goto error_channel_pipe; | |
1469 | } | |
1470 | ||
13886d2d DG |
1471 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1472 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1473 | goto error_metadata_pipe; |
1474 | } | |
3bd1e081 | 1475 | |
e9404c27 JG |
1476 | ctx->channel_monitor_pipe = -1; |
1477 | ||
fb3a43a9 | 1478 | return ctx; |
3bd1e081 | 1479 | |
fb3a43a9 | 1480 | error_metadata_pipe: |
d8ef542d MD |
1481 | utils_close_pipe(ctx->consumer_channel_pipe); |
1482 | error_channel_pipe: | |
d8ef542d | 1483 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1484 | error_quit_pipe: |
02b3d176 DG |
1485 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1486 | error_wakeup_pipe: | |
acdb9057 | 1487 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1488 | error_poll_pipe: |
1489 | free(ctx); | |
1490 | error: | |
1491 | return NULL; | |
1492 | } | |
1493 | ||
282dadbc MD |
1494 | /* |
1495 | * Iterate over all streams of the hashtable and free them properly. | |
1496 | */ | |
1497 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1498 | { | |
1499 | struct lttng_ht_iter iter; | |
1500 | struct lttng_consumer_stream *stream; | |
1501 | ||
1502 | if (ht == NULL) { | |
1503 | return; | |
1504 | } | |
1505 | ||
1506 | rcu_read_lock(); | |
1507 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1508 | /* | |
1509 | * Ignore return value since we are currently cleaning up so any error | |
1510 | * can't be handled. | |
1511 | */ | |
1512 | (void) consumer_del_stream(stream, ht); | |
1513 | } | |
1514 | rcu_read_unlock(); | |
1515 | ||
1516 | lttng_ht_destroy(ht); | |
1517 | } | |
1518 | ||
1519 | /* | |
1520 | * Iterate over all streams of the metadata hashtable and free them | |
1521 | * properly. | |
1522 | */ | |
1523 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1524 | { | |
1525 | struct lttng_ht_iter iter; | |
1526 | struct lttng_consumer_stream *stream; | |
1527 | ||
1528 | if (ht == NULL) { | |
1529 | return; | |
1530 | } | |
1531 | ||
1532 | rcu_read_lock(); | |
1533 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1534 | /* | |
1535 | * Ignore return value since we are currently cleaning up so any error | |
1536 | * can't be handled. | |
1537 | */ | |
1538 | (void) consumer_del_metadata_stream(stream, ht); | |
1539 | } | |
1540 | rcu_read_unlock(); | |
1541 | ||
1542 | lttng_ht_destroy(ht); | |
1543 | } | |
1544 | ||
3bd1e081 MD |
1545 | /* |
1546 | * Close all fds associated with the instance and free the context. | |
1547 | */ | |
1548 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1549 | { | |
4c462e79 MD |
1550 | int ret; |
1551 | ||
ab1027f4 DG |
1552 | DBG("Consumer destroying it. Closing everything."); |
1553 | ||
4f2e75b9 DG |
1554 | if (!ctx) { |
1555 | return; | |
1556 | } | |
1557 | ||
282dadbc MD |
1558 | destroy_data_stream_ht(data_ht); |
1559 | destroy_metadata_stream_ht(metadata_ht); | |
1560 | ||
4c462e79 MD |
1561 | ret = close(ctx->consumer_error_socket); |
1562 | if (ret) { | |
1563 | PERROR("close"); | |
1564 | } | |
331744e3 JD |
1565 | ret = close(ctx->consumer_metadata_socket); |
1566 | if (ret) { | |
1567 | PERROR("close"); | |
1568 | } | |
d8ef542d | 1569 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1570 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1571 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1572 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1573 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1574 | |
3bd1e081 MD |
1575 | unlink(ctx->consumer_command_sock_path); |
1576 | free(ctx); | |
1577 | } | |
1578 | ||
6197aea7 DG |
1579 | /* |
1580 | * Write the metadata stream id on the specified file descriptor. | |
1581 | */ | |
1582 | static int write_relayd_metadata_id(int fd, | |
1583 | struct lttng_consumer_stream *stream, | |
239f61af | 1584 | unsigned long padding) |
6197aea7 | 1585 | { |
6cd525e8 | 1586 | ssize_t ret; |
1d4dfdef | 1587 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1588 | |
1d4dfdef DG |
1589 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1590 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1591 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1592 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1593 | /* |
6f04ed72 | 1594 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1595 | * not to clubber the error output since this can happen in a normal |
1596 | * code path. | |
1597 | */ | |
1598 | if (errno != EPIPE) { | |
1599 | PERROR("write metadata stream id"); | |
1600 | } | |
1601 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1602 | /* |
1603 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1604 | * handle writting the missing part so report that as an error and | |
1605 | * don't lie to the caller. | |
1606 | */ | |
1607 | ret = -1; | |
6197aea7 DG |
1608 | goto end; |
1609 | } | |
1d4dfdef DG |
1610 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1611 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1612 | |
1613 | end: | |
6cd525e8 | 1614 | return (int) ret; |
6197aea7 DG |
1615 | } |
1616 | ||
3bd1e081 | 1617 | /* |
09e26845 DG |
1618 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1619 | * core function for writing trace buffers to either the local filesystem or | |
1620 | * the network. | |
1621 | * | |
d2956687 | 1622 | * It must be called with the stream and the channel lock held. |
79d4ffb7 | 1623 | * |
09e26845 | 1624 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1625 | * |
1626 | * Returns the number of bytes written | |
1627 | */ | |
4078b776 | 1628 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
128708c3 | 1629 | struct lttng_consumer_stream *stream, |
fd424d99 | 1630 | const struct lttng_buffer_view *buffer, |
6f9449c2 | 1631 | unsigned long padding) |
3bd1e081 | 1632 | { |
994ab360 | 1633 | ssize_t ret = 0; |
f02e1e8a DG |
1634 | off_t orig_offset = stream->out_fd_offset; |
1635 | /* Default is on the disk */ | |
1636 | int outfd = stream->out_fd; | |
f02e1e8a | 1637 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1638 | unsigned int relayd_hang_up = 0; |
fd424d99 JG |
1639 | const size_t subbuf_content_size = buffer->size - padding; |
1640 | size_t write_len; | |
f02e1e8a DG |
1641 | |
1642 | /* RCU lock for the relayd pointer */ | |
1643 | rcu_read_lock(); | |
7fd975c5 | 1644 | assert(stream->net_seq_idx != (uint64_t) -1ULL || |
948411cd | 1645 | stream->trace_chunk); |
d2956687 | 1646 | |
f02e1e8a | 1647 | /* Flag that the current stream if set for network streaming. */ |
da009f2c | 1648 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1649 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1650 | if (relayd == NULL) { | |
56591bac | 1651 | ret = -EPIPE; |
f02e1e8a DG |
1652 | goto end; |
1653 | } | |
1654 | } | |
1655 | ||
f02e1e8a DG |
1656 | /* Handle stream on the relayd if the output is on the network */ |
1657 | if (relayd) { | |
fd424d99 | 1658 | unsigned long netlen = subbuf_content_size; |
f02e1e8a DG |
1659 | |
1660 | /* | |
1661 | * Lock the control socket for the complete duration of the function | |
1662 | * since from this point on we will use the socket. | |
1663 | */ | |
1664 | if (stream->metadata_flag) { | |
1665 | /* Metadata requires the control socket. */ | |
1666 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
93ec662e JD |
1667 | if (stream->reset_metadata_flag) { |
1668 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1669 | stream->relayd_stream_id, | |
1670 | stream->metadata_version); | |
1671 | if (ret < 0) { | |
1672 | relayd_hang_up = 1; | |
1673 | goto write_error; | |
1674 | } | |
1675 | stream->reset_metadata_flag = 0; | |
1676 | } | |
1d4dfdef | 1677 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1678 | } |
1679 | ||
1d4dfdef | 1680 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1681 | if (ret < 0) { |
1682 | relayd_hang_up = 1; | |
1683 | goto write_error; | |
1684 | } | |
1685 | /* Use the returned socket. */ | |
1686 | outfd = ret; | |
f02e1e8a | 1687 | |
994ab360 DG |
1688 | /* Write metadata stream id before payload */ |
1689 | if (stream->metadata_flag) { | |
239f61af | 1690 | ret = write_relayd_metadata_id(outfd, stream, padding); |
994ab360 | 1691 | if (ret < 0) { |
8994307f DG |
1692 | relayd_hang_up = 1; |
1693 | goto write_error; | |
1694 | } | |
f02e1e8a | 1695 | } |
1624d5b7 | 1696 | |
fd424d99 JG |
1697 | write_len = subbuf_content_size; |
1698 | } else { | |
1699 | /* No streaming; we have to write the full padding. */ | |
93ec662e JD |
1700 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1701 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1702 | if (ret < 0) { | |
1703 | ERR("Reset metadata file"); | |
1704 | goto end; | |
1705 | } | |
1706 | stream->reset_metadata_flag = 0; | |
1707 | } | |
1708 | ||
1624d5b7 JD |
1709 | /* |
1710 | * Check if we need to change the tracefile before writing the packet. | |
1711 | */ | |
1712 | if (stream->chan->tracefile_size > 0 && | |
fd424d99 | 1713 | (stream->tracefile_size_current + buffer->size) > |
1624d5b7 | 1714 | stream->chan->tracefile_size) { |
d2956687 JG |
1715 | ret = consumer_stream_rotate_output_files(stream); |
1716 | if (ret) { | |
1624d5b7 JD |
1717 | goto end; |
1718 | } | |
309167d2 | 1719 | outfd = stream->out_fd; |
a1ae300f | 1720 | orig_offset = 0; |
1624d5b7 | 1721 | } |
fd424d99 | 1722 | stream->tracefile_size_current += buffer->size; |
fd424d99 | 1723 | write_len = buffer->size; |
f02e1e8a DG |
1724 | } |
1725 | ||
d02b8372 DG |
1726 | /* |
1727 | * This call guarantee that len or less is returned. It's impossible to | |
1728 | * receive a ret value that is bigger than len. | |
1729 | */ | |
fd424d99 | 1730 | ret = lttng_write(outfd, buffer->data, write_len); |
e2d1190b | 1731 | DBG("Consumer mmap write() ret %zd (len %zu)", ret, write_len); |
fd424d99 | 1732 | if (ret < 0 || ((size_t) ret != write_len)) { |
d02b8372 DG |
1733 | /* |
1734 | * Report error to caller if nothing was written else at least send the | |
1735 | * amount written. | |
1736 | */ | |
1737 | if (ret < 0) { | |
994ab360 | 1738 | ret = -errno; |
f02e1e8a | 1739 | } |
994ab360 | 1740 | relayd_hang_up = 1; |
f02e1e8a | 1741 | |
d02b8372 | 1742 | /* Socket operation failed. We consider the relayd dead */ |
fcf0f774 | 1743 | if (errno == EPIPE) { |
d02b8372 DG |
1744 | /* |
1745 | * This is possible if the fd is closed on the other side | |
1746 | * (outfd) or any write problem. It can be verbose a bit for a | |
1747 | * normal execution if for instance the relayd is stopped | |
1748 | * abruptly. This can happen so set this to a DBG statement. | |
1749 | */ | |
1750 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1751 | } else { |
1752 | /* Unhandled error, print it and stop function right now. */ | |
fd424d99 JG |
1753 | PERROR("Error in write mmap (ret %zd != write_len %zu)", ret, |
1754 | write_len); | |
f02e1e8a | 1755 | } |
994ab360 | 1756 | goto write_error; |
d02b8372 DG |
1757 | } |
1758 | stream->output_written += ret; | |
d02b8372 DG |
1759 | |
1760 | /* This call is useless on a socket so better save a syscall. */ | |
1761 | if (!relayd) { | |
1762 | /* This won't block, but will start writeout asynchronously */ | |
fd424d99 | 1763 | lttng_sync_file_range(outfd, stream->out_fd_offset, write_len, |
d02b8372 | 1764 | SYNC_FILE_RANGE_WRITE); |
fd424d99 | 1765 | stream->out_fd_offset += write_len; |
f5dbe415 | 1766 | lttng_consumer_sync_trace_file(stream, orig_offset); |
f02e1e8a | 1767 | } |
f02e1e8a | 1768 | |
8994307f DG |
1769 | write_error: |
1770 | /* | |
1771 | * This is a special case that the relayd has closed its socket. Let's | |
1772 | * cleanup the relayd object and all associated streams. | |
1773 | */ | |
1774 | if (relayd && relayd_hang_up) { | |
9276e5c8 JR |
1775 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1776 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1777 | } |
1778 | ||
f02e1e8a DG |
1779 | end: |
1780 | /* Unlock only if ctrl socket used */ | |
1781 | if (relayd && stream->metadata_flag) { | |
1782 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1783 | } | |
1784 | ||
1785 | rcu_read_unlock(); | |
994ab360 | 1786 | return ret; |
3bd1e081 MD |
1787 | } |
1788 | ||
1789 | /* | |
1790 | * Splice the data from the ring buffer to the tracefile. | |
1791 | * | |
79d4ffb7 DG |
1792 | * It must be called with the stream lock held. |
1793 | * | |
3bd1e081 MD |
1794 | * Returns the number of bytes spliced. |
1795 | */ | |
4078b776 | 1796 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1797 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1798 | struct lttng_consumer_stream *stream, unsigned long len, |
6f9449c2 | 1799 | unsigned long padding) |
3bd1e081 | 1800 | { |
f02e1e8a DG |
1801 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1802 | loff_t offset = 0; | |
1803 | off_t orig_offset = stream->out_fd_offset; | |
1804 | int fd = stream->wait_fd; | |
1805 | /* Default is on the disk */ | |
1806 | int outfd = stream->out_fd; | |
f02e1e8a | 1807 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1808 | int *splice_pipe; |
8994307f | 1809 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1810 | |
fa29bfbf | 1811 | switch (the_consumer_data.type) { |
3bd1e081 | 1812 | case LTTNG_CONSUMER_KERNEL: |
f02e1e8a | 1813 | break; |
7753dea8 MD |
1814 | case LTTNG_CONSUMER32_UST: |
1815 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1816 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1817 | return -ENOSYS; |
1818 | default: | |
1819 | ERR("Unknown consumer_data type"); | |
1820 | assert(0); | |
3bd1e081 MD |
1821 | } |
1822 | ||
f02e1e8a DG |
1823 | /* RCU lock for the relayd pointer */ |
1824 | rcu_read_lock(); | |
1825 | ||
1826 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1827 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1828 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1829 | if (relayd == NULL) { | |
ad0b0d23 | 1830 | written = -ret; |
f02e1e8a DG |
1831 | goto end; |
1832 | } | |
1833 | } | |
a2361a61 | 1834 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1835 | |
f02e1e8a | 1836 | /* Write metadata stream id before payload */ |
1d4dfdef | 1837 | if (relayd) { |
ad0b0d23 | 1838 | unsigned long total_len = len; |
f02e1e8a | 1839 | |
1d4dfdef DG |
1840 | if (stream->metadata_flag) { |
1841 | /* | |
1842 | * Lock the control socket for the complete duration of the function | |
1843 | * since from this point on we will use the socket. | |
1844 | */ | |
1845 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1846 | ||
93ec662e JD |
1847 | if (stream->reset_metadata_flag) { |
1848 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1849 | stream->relayd_stream_id, | |
1850 | stream->metadata_version); | |
1851 | if (ret < 0) { | |
1852 | relayd_hang_up = 1; | |
1853 | goto write_error; | |
1854 | } | |
1855 | stream->reset_metadata_flag = 0; | |
1856 | } | |
239f61af | 1857 | ret = write_relayd_metadata_id(splice_pipe[1], stream, |
1d4dfdef DG |
1858 | padding); |
1859 | if (ret < 0) { | |
1860 | written = ret; | |
ad0b0d23 DG |
1861 | relayd_hang_up = 1; |
1862 | goto write_error; | |
1d4dfdef DG |
1863 | } |
1864 | ||
1865 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1866 | } | |
1867 | ||
1868 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1869 | if (ret < 0) { |
1870 | written = ret; | |
1871 | relayd_hang_up = 1; | |
1872 | goto write_error; | |
f02e1e8a | 1873 | } |
ad0b0d23 DG |
1874 | /* Use the returned socket. */ |
1875 | outfd = ret; | |
1d4dfdef DG |
1876 | } else { |
1877 | /* No streaming, we have to set the len with the full padding */ | |
1878 | len += padding; | |
1624d5b7 | 1879 | |
93ec662e JD |
1880 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1881 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1882 | if (ret < 0) { | |
1883 | ERR("Reset metadata file"); | |
1884 | goto end; | |
1885 | } | |
1886 | stream->reset_metadata_flag = 0; | |
1887 | } | |
1624d5b7 JD |
1888 | /* |
1889 | * Check if we need to change the tracefile before writing the packet. | |
1890 | */ | |
1891 | if (stream->chan->tracefile_size > 0 && | |
1892 | (stream->tracefile_size_current + len) > | |
1893 | stream->chan->tracefile_size) { | |
d2956687 | 1894 | ret = consumer_stream_rotate_output_files(stream); |
1624d5b7 | 1895 | if (ret < 0) { |
ad0b0d23 | 1896 | written = ret; |
1624d5b7 JD |
1897 | goto end; |
1898 | } | |
309167d2 | 1899 | outfd = stream->out_fd; |
a1ae300f | 1900 | orig_offset = 0; |
1624d5b7 JD |
1901 | } |
1902 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1903 | } |
1904 | ||
1905 | while (len > 0) { | |
1d4dfdef DG |
1906 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1907 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1908 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1909 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1910 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1911 | if (ret_splice < 0) { | |
d02b8372 | 1912 | ret = errno; |
ad0b0d23 | 1913 | written = -ret; |
d02b8372 | 1914 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1915 | goto splice_error; |
1916 | } | |
1917 | ||
1918 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1919 | if (relayd && stream->metadata_flag) { |
1920 | size_t metadata_payload_size = | |
1921 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1922 | ||
1923 | /* Update counter to fit the spliced data */ | |
1924 | ret_splice += metadata_payload_size; | |
1925 | len += metadata_payload_size; | |
1926 | /* | |
1927 | * We do this so the return value can match the len passed as | |
1928 | * argument to this function. | |
1929 | */ | |
1930 | written -= metadata_payload_size; | |
f02e1e8a DG |
1931 | } |
1932 | ||
1933 | /* Splice data out */ | |
fb3a43a9 | 1934 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1935 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1936 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1937 | outfd, ret_splice); | |
f02e1e8a | 1938 | if (ret_splice < 0) { |
d02b8372 | 1939 | ret = errno; |
ad0b0d23 DG |
1940 | written = -ret; |
1941 | relayd_hang_up = 1; | |
1942 | goto write_error; | |
f02e1e8a | 1943 | } else if (ret_splice > len) { |
d02b8372 DG |
1944 | /* |
1945 | * We don't expect this code path to be executed but you never know | |
1946 | * so this is an extra protection agains a buggy splice(). | |
1947 | */ | |
f02e1e8a | 1948 | ret = errno; |
ad0b0d23 | 1949 | written += ret_splice; |
d02b8372 DG |
1950 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1951 | len); | |
f02e1e8a | 1952 | goto splice_error; |
d02b8372 DG |
1953 | } else { |
1954 | /* All good, update current len and continue. */ | |
1955 | len -= ret_splice; | |
f02e1e8a | 1956 | } |
f02e1e8a DG |
1957 | |
1958 | /* This call is useless on a socket so better save a syscall. */ | |
1959 | if (!relayd) { | |
1960 | /* This won't block, but will start writeout asynchronously */ | |
1961 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1962 | SYNC_FILE_RANGE_WRITE); | |
1963 | stream->out_fd_offset += ret_splice; | |
1964 | } | |
e5d1a9b3 | 1965 | stream->output_written += ret_splice; |
f02e1e8a DG |
1966 | written += ret_splice; |
1967 | } | |
f5dbe415 JG |
1968 | if (!relayd) { |
1969 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1970 | } | |
f02e1e8a DG |
1971 | goto end; |
1972 | ||
8994307f DG |
1973 | write_error: |
1974 | /* | |
1975 | * This is a special case that the relayd has closed its socket. Let's | |
1976 | * cleanup the relayd object and all associated streams. | |
1977 | */ | |
1978 | if (relayd && relayd_hang_up) { | |
9276e5c8 JR |
1979 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1980 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1981 | /* Skip splice error so the consumer does not fail */ |
1982 | goto end; | |
1983 | } | |
1984 | ||
f02e1e8a DG |
1985 | splice_error: |
1986 | /* send the appropriate error description to sessiond */ | |
1987 | switch (ret) { | |
f02e1e8a | 1988 | case EINVAL: |
f73fabfd | 1989 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1990 | break; |
1991 | case ENOMEM: | |
f73fabfd | 1992 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1993 | break; |
1994 | case ESPIPE: | |
f73fabfd | 1995 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1996 | break; |
1997 | } | |
1998 | ||
1999 | end: | |
2000 | if (relayd && stream->metadata_flag) { | |
2001 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
2002 | } | |
2003 | ||
2004 | rcu_read_unlock(); | |
2005 | return written; | |
3bd1e081 MD |
2006 | } |
2007 | ||
15055ce5 JD |
2008 | /* |
2009 | * Sample the snapshot positions for a specific fd | |
2010 | * | |
2011 | * Returns 0 on success, < 0 on error | |
2012 | */ | |
2013 | int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream) | |
2014 | { | |
fa29bfbf | 2015 | switch (the_consumer_data.type) { |
15055ce5 JD |
2016 | case LTTNG_CONSUMER_KERNEL: |
2017 | return lttng_kconsumer_sample_snapshot_positions(stream); | |
2018 | case LTTNG_CONSUMER32_UST: | |
2019 | case LTTNG_CONSUMER64_UST: | |
2020 | return lttng_ustconsumer_sample_snapshot_positions(stream); | |
2021 | default: | |
2022 | ERR("Unknown consumer_data type"); | |
2023 | assert(0); | |
2024 | return -ENOSYS; | |
2025 | } | |
2026 | } | |
3bd1e081 MD |
2027 | /* |
2028 | * Take a snapshot for a specific fd | |
2029 | * | |
2030 | * Returns 0 on success, < 0 on error | |
2031 | */ | |
ffe60014 | 2032 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 | 2033 | { |
fa29bfbf | 2034 | switch (the_consumer_data.type) { |
3bd1e081 | 2035 | case LTTNG_CONSUMER_KERNEL: |
ffe60014 | 2036 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
2037 | case LTTNG_CONSUMER32_UST: |
2038 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2039 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
2040 | default: |
2041 | ERR("Unknown consumer_data type"); | |
2042 | assert(0); | |
2043 | return -ENOSYS; | |
2044 | } | |
3bd1e081 MD |
2045 | } |
2046 | ||
2047 | /* | |
2048 | * Get the produced position | |
2049 | * | |
2050 | * Returns 0 on success, < 0 on error | |
2051 | */ | |
ffe60014 | 2052 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
2053 | unsigned long *pos) |
2054 | { | |
fa29bfbf | 2055 | switch (the_consumer_data.type) { |
3bd1e081 | 2056 | case LTTNG_CONSUMER_KERNEL: |
ffe60014 | 2057 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
2058 | case LTTNG_CONSUMER32_UST: |
2059 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2060 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
2061 | default: |
2062 | ERR("Unknown consumer_data type"); | |
2063 | assert(0); | |
2064 | return -ENOSYS; | |
2065 | } | |
2066 | } | |
2067 | ||
15055ce5 JD |
2068 | /* |
2069 | * Get the consumed position (free-running counter position in bytes). | |
2070 | * | |
2071 | * Returns 0 on success, < 0 on error | |
2072 | */ | |
2073 | int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, | |
2074 | unsigned long *pos) | |
2075 | { | |
fa29bfbf | 2076 | switch (the_consumer_data.type) { |
15055ce5 JD |
2077 | case LTTNG_CONSUMER_KERNEL: |
2078 | return lttng_kconsumer_get_consumed_snapshot(stream, pos); | |
2079 | case LTTNG_CONSUMER32_UST: | |
2080 | case LTTNG_CONSUMER64_UST: | |
2081 | return lttng_ustconsumer_get_consumed_snapshot(stream, pos); | |
2082 | default: | |
2083 | ERR("Unknown consumer_data type"); | |
2084 | assert(0); | |
2085 | return -ENOSYS; | |
2086 | } | |
2087 | } | |
2088 | ||
3bd1e081 MD |
2089 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
2090 | int sock, struct pollfd *consumer_sockpoll) | |
2091 | { | |
fa29bfbf | 2092 | switch (the_consumer_data.type) { |
3bd1e081 MD |
2093 | case LTTNG_CONSUMER_KERNEL: |
2094 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
2095 | case LTTNG_CONSUMER32_UST: |
2096 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
2097 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
2098 | default: | |
2099 | ERR("Unknown consumer_data type"); | |
2100 | assert(0); | |
2101 | return -ENOSYS; | |
2102 | } | |
2103 | } | |
2104 | ||
1f8d1c14 | 2105 | static |
6d574024 | 2106 | void lttng_consumer_close_all_metadata(void) |
d88aee68 | 2107 | { |
fa29bfbf | 2108 | switch (the_consumer_data.type) { |
d88aee68 DG |
2109 | case LTTNG_CONSUMER_KERNEL: |
2110 | /* | |
2111 | * The Kernel consumer has a different metadata scheme so we don't | |
2112 | * close anything because the stream will be closed by the session | |
2113 | * daemon. | |
2114 | */ | |
2115 | break; | |
2116 | case LTTNG_CONSUMER32_UST: | |
2117 | case LTTNG_CONSUMER64_UST: | |
2118 | /* | |
2119 | * Close all metadata streams. The metadata hash table is passed and | |
2120 | * this call iterates over it by closing all wakeup fd. This is safe | |
2121 | * because at this point we are sure that the metadata producer is | |
2122 | * either dead or blocked. | |
2123 | */ | |
6d574024 | 2124 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
2125 | break; |
2126 | default: | |
2127 | ERR("Unknown consumer_data type"); | |
2128 | assert(0); | |
2129 | } | |
2130 | } | |
2131 | ||
fb3a43a9 DG |
2132 | /* |
2133 | * Clean up a metadata stream and free its memory. | |
2134 | */ | |
e316aad5 DG |
2135 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
2136 | struct lttng_ht *ht) | |
fb3a43a9 | 2137 | { |
a6ef8ee6 JG |
2138 | struct lttng_consumer_channel *channel = NULL; |
2139 | bool free_channel = false; | |
fb3a43a9 DG |
2140 | |
2141 | assert(stream); | |
2142 | /* | |
2143 | * This call should NEVER receive regular stream. It must always be | |
2144 | * metadata stream and this is crucial for data structure synchronization. | |
2145 | */ | |
2146 | assert(stream->metadata_flag); | |
2147 | ||
e316aad5 DG |
2148 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2149 | ||
fa29bfbf | 2150 | pthread_mutex_lock(&the_consumer_data.lock); |
a6ef8ee6 JG |
2151 | /* |
2152 | * Note that this assumes that a stream's channel is never changed and | |
2153 | * that the stream's lock doesn't need to be taken to sample its | |
2154 | * channel. | |
2155 | */ | |
2156 | channel = stream->chan; | |
2157 | pthread_mutex_lock(&channel->lock); | |
3dad2c0f | 2158 | pthread_mutex_lock(&stream->lock); |
a6ef8ee6 | 2159 | if (channel->metadata_cache) { |
081424af | 2160 | /* Only applicable to userspace consumers. */ |
a6ef8ee6 | 2161 | pthread_mutex_lock(&channel->metadata_cache->lock); |
081424af | 2162 | } |
8994307f | 2163 | |
6d574024 DG |
2164 | /* Remove any reference to that stream. */ |
2165 | consumer_stream_delete(stream, ht); | |
ca22feea | 2166 | |
6d574024 DG |
2167 | /* Close down everything including the relayd if one. */ |
2168 | consumer_stream_close(stream); | |
2169 | /* Destroy tracer buffers of the stream. */ | |
2170 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2171 | |
2172 | /* Atomically decrement channel refcount since other threads can use it. */ | |
a6ef8ee6 JG |
2173 | if (!uatomic_sub_return(&channel->refcount, 1) |
2174 | && !uatomic_read(&channel->nb_init_stream_left)) { | |
c30aaa51 | 2175 | /* Go for channel deletion! */ |
a6ef8ee6 | 2176 | free_channel = true; |
fb3a43a9 | 2177 | } |
a6ef8ee6 | 2178 | stream->chan = NULL; |
fb3a43a9 | 2179 | |
73811ecc DG |
2180 | /* |
2181 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2182 | * channel lock MUST be acquired before being able to check for a NULL |
2183 | * pointer value. | |
73811ecc | 2184 | */ |
a6ef8ee6 | 2185 | channel->metadata_stream = NULL; |
95671f53 | 2186 | lttng_wait_queue_wake_all(&channel->metadata_pushed_wait_queue); |
73811ecc | 2187 | |
a6ef8ee6 JG |
2188 | if (channel->metadata_cache) { |
2189 | pthread_mutex_unlock(&channel->metadata_cache->lock); | |
081424af | 2190 | } |
3dad2c0f | 2191 | pthread_mutex_unlock(&stream->lock); |
a6ef8ee6 | 2192 | pthread_mutex_unlock(&channel->lock); |
fa29bfbf | 2193 | pthread_mutex_unlock(&the_consumer_data.lock); |
e316aad5 | 2194 | |
a6ef8ee6 JG |
2195 | if (free_channel) { |
2196 | consumer_del_channel(channel); | |
e316aad5 DG |
2197 | } |
2198 | ||
d2956687 JG |
2199 | lttng_trace_chunk_put(stream->trace_chunk); |
2200 | stream->trace_chunk = NULL; | |
6d574024 | 2201 | consumer_stream_free(stream); |
fb3a43a9 DG |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * Action done with the metadata stream when adding it to the consumer internal | |
2206 | * data structures to handle it. | |
2207 | */ | |
66d583dc | 2208 | void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2209 | { |
5ab66908 | 2210 | struct lttng_ht *ht = metadata_ht; |
76082088 | 2211 | struct lttng_ht_iter iter; |
d88aee68 | 2212 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2213 | |
e316aad5 DG |
2214 | assert(stream); |
2215 | assert(ht); | |
2216 | ||
d88aee68 | 2217 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 | 2218 | |
fa29bfbf | 2219 | pthread_mutex_lock(&the_consumer_data.lock); |
a9838785 | 2220 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2221 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2222 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2223 | |
e316aad5 DG |
2224 | /* |
2225 | * From here, refcounts are updated so be _careful_ when returning an error | |
2226 | * after this point. | |
2227 | */ | |
2228 | ||
fb3a43a9 | 2229 | rcu_read_lock(); |
76082088 DG |
2230 | |
2231 | /* | |
2232 | * Lookup the stream just to make sure it does not exist in our internal | |
2233 | * state. This should NEVER happen. | |
2234 | */ | |
d88aee68 DG |
2235 | lttng_ht_lookup(ht, &stream->key, &iter); |
2236 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2237 | assert(!node); |
2238 | ||
e316aad5 | 2239 | /* |
ffe60014 DG |
2240 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2241 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2242 | * causes the count to become 0 also causes a stream to be added. The |
2243 | * channel deletion will thus be triggered by the following removal of this | |
2244 | * stream. | |
2245 | */ | |
ffe60014 | 2246 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2247 | /* Increment refcount before decrementing nb_init_stream_left */ |
2248 | cmm_smp_wmb(); | |
ffe60014 | 2249 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2250 | } |
2251 | ||
d88aee68 | 2252 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2253 | |
fa29bfbf SM |
2254 | lttng_ht_add_u64(the_consumer_data.stream_per_chan_id_ht, |
2255 | &stream->node_channel_id); | |
d8ef542d | 2256 | |
ca22feea DG |
2257 | /* |
2258 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2259 | * the key since the HT does not use it and we allow to add redundant keys | |
2260 | * into this table. | |
2261 | */ | |
fa29bfbf SM |
2262 | lttng_ht_add_u64(the_consumer_data.stream_list_ht, |
2263 | &stream->node_session_id); | |
ca22feea | 2264 | |
fb3a43a9 | 2265 | rcu_read_unlock(); |
e316aad5 | 2266 | |
2e818a6a | 2267 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2268 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2269 | pthread_mutex_unlock(&stream->chan->timer_lock); |
fa29bfbf | 2270 | pthread_mutex_unlock(&the_consumer_data.lock); |
fb3a43a9 DG |
2271 | } |
2272 | ||
8994307f DG |
2273 | /* |
2274 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2275 | */ | |
2276 | static void validate_endpoint_status_data_stream(void) | |
2277 | { | |
2278 | struct lttng_ht_iter iter; | |
2279 | struct lttng_consumer_stream *stream; | |
2280 | ||
2281 | DBG("Consumer delete flagged data stream"); | |
2282 | ||
2283 | rcu_read_lock(); | |
2284 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2285 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2286 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2287 | continue; |
2288 | } | |
2289 | /* Delete it right now */ | |
2290 | consumer_del_stream(stream, data_ht); | |
2291 | } | |
2292 | rcu_read_unlock(); | |
2293 | } | |
2294 | ||
2295 | /* | |
2296 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2297 | */ | |
2298 | static void validate_endpoint_status_metadata_stream( | |
2299 | struct lttng_poll_event *pollset) | |
2300 | { | |
2301 | struct lttng_ht_iter iter; | |
2302 | struct lttng_consumer_stream *stream; | |
2303 | ||
2304 | DBG("Consumer delete flagged metadata stream"); | |
2305 | ||
2306 | assert(pollset); | |
2307 | ||
2308 | rcu_read_lock(); | |
2309 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2310 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2311 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2312 | continue; |
2313 | } | |
2314 | /* | |
2315 | * Remove from pollset so the metadata thread can continue without | |
2316 | * blocking on a deleted stream. | |
2317 | */ | |
2318 | lttng_poll_del(pollset, stream->wait_fd); | |
2319 | ||
2320 | /* Delete it right now */ | |
2321 | consumer_del_metadata_stream(stream, metadata_ht); | |
2322 | } | |
2323 | rcu_read_unlock(); | |
2324 | } | |
2325 | ||
fb3a43a9 DG |
2326 | /* |
2327 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2328 | * network. | |
2329 | */ | |
7d980def | 2330 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2331 | { |
1fc79fb4 | 2332 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2333 | uint32_t revents, nb_fd; |
e316aad5 | 2334 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2335 | struct lttng_ht_iter iter; |
d88aee68 | 2336 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2337 | struct lttng_poll_event events; |
2338 | struct lttng_consumer_local_data *ctx = data; | |
2339 | ssize_t len; | |
2340 | ||
2341 | rcu_register_thread(); | |
2342 | ||
1fc79fb4 MD |
2343 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2344 | ||
2d57de81 MD |
2345 | if (testpoint(consumerd_thread_metadata)) { |
2346 | goto error_testpoint; | |
2347 | } | |
2348 | ||
9ce5646a MD |
2349 | health_code_update(); |
2350 | ||
fb3a43a9 DG |
2351 | DBG("Thread metadata poll started"); |
2352 | ||
fb3a43a9 DG |
2353 | /* Size is set to 1 for the consumer_metadata pipe */ |
2354 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2355 | if (ret < 0) { | |
2356 | ERR("Poll set creation failed"); | |
d8ef542d | 2357 | goto end_poll; |
fb3a43a9 DG |
2358 | } |
2359 | ||
13886d2d DG |
2360 | ret = lttng_poll_add(&events, |
2361 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2362 | if (ret < 0) { |
2363 | goto end; | |
2364 | } | |
2365 | ||
2366 | /* Main loop */ | |
2367 | DBG("Metadata main loop started"); | |
2368 | ||
2369 | while (1) { | |
fb3a43a9 | 2370 | restart: |
7fa2082e | 2371 | health_code_update(); |
9ce5646a | 2372 | health_poll_entry(); |
7fa2082e | 2373 | DBG("Metadata poll wait"); |
fb3a43a9 | 2374 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2375 | DBG("Metadata poll return from wait with %d fd(s)", |
2376 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2377 | health_poll_exit(); |
40063ead | 2378 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2379 | if (ret < 0) { |
2380 | if (errno == EINTR) { | |
40063ead | 2381 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2382 | goto restart; |
2383 | } | |
d9607cd7 MD |
2384 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2385 | err = 0; /* All is OK */ | |
2386 | } | |
2387 | goto end; | |
fb3a43a9 DG |
2388 | } |
2389 | ||
0d9c5d77 DG |
2390 | nb_fd = ret; |
2391 | ||
e316aad5 | 2392 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2393 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2394 | health_code_update(); |
2395 | ||
fb3a43a9 DG |
2396 | revents = LTTNG_POLL_GETEV(&events, i); |
2397 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2398 | ||
13886d2d | 2399 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
03e43155 | 2400 | if (revents & LPOLLIN) { |
13886d2d DG |
2401 | ssize_t pipe_len; |
2402 | ||
2403 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2404 | &stream, sizeof(stream)); | |
6cd525e8 | 2405 | if (pipe_len < sizeof(stream)) { |
03e43155 MD |
2406 | if (pipe_len < 0) { |
2407 | PERROR("read metadata stream"); | |
2408 | } | |
fb3a43a9 | 2409 | /* |
03e43155 MD |
2410 | * Remove the pipe from the poll set and continue the loop |
2411 | * since their might be data to consume. | |
fb3a43a9 | 2412 | */ |
03e43155 MD |
2413 | lttng_poll_del(&events, |
2414 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2415 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2416 | continue; |
2417 | } | |
2418 | ||
8994307f DG |
2419 | /* A NULL stream means that the state has changed. */ |
2420 | if (stream == NULL) { | |
2421 | /* Check for deleted streams. */ | |
2422 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2423 | goto restart; |
8994307f DG |
2424 | } |
2425 | ||
fb3a43a9 DG |
2426 | DBG("Adding metadata stream %d to poll set", |
2427 | stream->wait_fd); | |
2428 | ||
fb3a43a9 DG |
2429 | /* Add metadata stream to the global poll events list */ |
2430 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2431 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
03e43155 MD |
2432 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2433 | DBG("Metadata thread pipe hung up"); | |
2434 | /* | |
2435 | * Remove the pipe from the poll set and continue the loop | |
2436 | * since their might be data to consume. | |
2437 | */ | |
2438 | lttng_poll_del(&events, | |
2439 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2440 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2441 | continue; | |
2442 | } else { | |
2443 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2444 | goto end; | |
fb3a43a9 DG |
2445 | } |
2446 | ||
e316aad5 | 2447 | /* Handle other stream */ |
fb3a43a9 DG |
2448 | continue; |
2449 | } | |
2450 | ||
d09e1200 | 2451 | rcu_read_lock(); |
d88aee68 DG |
2452 | { |
2453 | uint64_t tmp_id = (uint64_t) pollfd; | |
2454 | ||
2455 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2456 | } | |
2457 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2458 | assert(node); |
fb3a43a9 DG |
2459 | |
2460 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2461 | node); |
fb3a43a9 | 2462 | |
03e43155 MD |
2463 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2464 | /* Get the data out of the metadata file descriptor */ | |
2465 | DBG("Metadata available on fd %d", pollfd); | |
2466 | assert(stream->wait_fd == pollfd); | |
2467 | ||
2468 | do { | |
2469 | health_code_update(); | |
2470 | ||
6f9449c2 | 2471 | len = ctx->on_buffer_ready(stream, ctx, false); |
03e43155 MD |
2472 | /* |
2473 | * We don't check the return value here since if we get | |
83f4233d | 2474 | * a negative len, it means an error occurred thus we |
03e43155 MD |
2475 | * simply remove it from the poll set and free the |
2476 | * stream. | |
2477 | */ | |
2478 | } while (len > 0); | |
2479 | ||
2480 | /* It's ok to have an unavailable sub-buffer */ | |
2481 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2482 | /* Clean up stream from consumer and free it. */ | |
2483 | lttng_poll_del(&events, stream->wait_fd); | |
2484 | consumer_del_metadata_stream(stream, metadata_ht); | |
2485 | } | |
2486 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2487 | DBG("Metadata fd %d is hup|err.", pollfd); |
fa29bfbf SM |
2488 | if (!stream->hangup_flush_done && |
2489 | (the_consumer_data.type == LTTNG_CONSUMER32_UST || | |
2490 | the_consumer_data.type == | |
2491 | LTTNG_CONSUMER64_UST)) { | |
fb3a43a9 DG |
2492 | DBG("Attempting to flush and consume the UST buffers"); |
2493 | lttng_ustconsumer_on_stream_hangup(stream); | |
2494 | ||
2495 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2496 | do { |
9ce5646a MD |
2497 | health_code_update(); |
2498 | ||
6f9449c2 | 2499 | len = ctx->on_buffer_ready(stream, ctx, false); |
4bb94b75 DG |
2500 | /* |
2501 | * We don't check the return value here since if we get | |
83f4233d | 2502 | * a negative len, it means an error occurred thus we |
4bb94b75 DG |
2503 | * simply remove it from the poll set and free the |
2504 | * stream. | |
2505 | */ | |
2506 | } while (len > 0); | |
fb3a43a9 DG |
2507 | } |
2508 | ||
fb3a43a9 | 2509 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2510 | /* |
2511 | * This call update the channel states, closes file descriptors | |
2512 | * and securely free the stream. | |
2513 | */ | |
2514 | consumer_del_metadata_stream(stream, metadata_ht); | |
03e43155 MD |
2515 | } else { |
2516 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
6f2f1a70 | 2517 | rcu_read_unlock(); |
03e43155 | 2518 | goto end; |
fb3a43a9 | 2519 | } |
e316aad5 | 2520 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2521 | rcu_read_unlock(); |
fb3a43a9 DG |
2522 | } |
2523 | } | |
2524 | ||
1fc79fb4 MD |
2525 | /* All is OK */ |
2526 | err = 0; | |
fb3a43a9 DG |
2527 | end: |
2528 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2529 | |
d8ef542d MD |
2530 | lttng_poll_clean(&events); |
2531 | end_poll: | |
2d57de81 | 2532 | error_testpoint: |
1fc79fb4 MD |
2533 | if (err) { |
2534 | health_error(); | |
2535 | ERR("Health error occurred in %s", __func__); | |
2536 | } | |
2537 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2538 | rcu_unregister_thread(); |
2539 | return NULL; | |
2540 | } | |
2541 | ||
3bd1e081 | 2542 | /* |
e4421fec | 2543 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2544 | * it to tracefile if necessary. |
2545 | */ | |
7d980def | 2546 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2547 | { |
1fc79fb4 | 2548 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2549 | struct pollfd *pollfd = NULL; |
2550 | /* local view of the streams */ | |
c869f647 | 2551 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 | 2552 | /* local view of consumer_data.fds_count */ |
8bdcc002 JG |
2553 | int nb_fd = 0; |
2554 | /* 2 for the consumer_data_pipe and wake up pipe */ | |
2555 | const int nb_pipes_fd = 2; | |
9a2fcf78 JD |
2556 | /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */ |
2557 | int nb_inactive_fd = 0; | |
3bd1e081 | 2558 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2559 | ssize_t len; |
3bd1e081 | 2560 | |
e7b994a3 DG |
2561 | rcu_register_thread(); |
2562 | ||
1fc79fb4 MD |
2563 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2564 | ||
2d57de81 MD |
2565 | if (testpoint(consumerd_thread_data)) { |
2566 | goto error_testpoint; | |
2567 | } | |
2568 | ||
9ce5646a MD |
2569 | health_code_update(); |
2570 | ||
4df6c8cb MD |
2571 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2572 | if (local_stream == NULL) { | |
2573 | PERROR("local_stream malloc"); | |
2574 | goto end; | |
2575 | } | |
3bd1e081 MD |
2576 | |
2577 | while (1) { | |
9ce5646a MD |
2578 | health_code_update(); |
2579 | ||
3bd1e081 MD |
2580 | high_prio = 0; |
2581 | num_hup = 0; | |
2582 | ||
2583 | /* | |
e4421fec | 2584 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2585 | * local array as well |
2586 | */ | |
fa29bfbf SM |
2587 | pthread_mutex_lock(&the_consumer_data.lock); |
2588 | if (the_consumer_data.need_update) { | |
0e428499 DG |
2589 | free(pollfd); |
2590 | pollfd = NULL; | |
2591 | ||
2592 | free(local_stream); | |
2593 | local_stream = NULL; | |
3bd1e081 | 2594 | |
8bdcc002 | 2595 | /* Allocate for all fds */ |
fa29bfbf SM |
2596 | pollfd = zmalloc((the_consumer_data.stream_count + |
2597 | nb_pipes_fd) * | |
2598 | sizeof(struct pollfd)); | |
3bd1e081 | 2599 | if (pollfd == NULL) { |
7a57cf92 | 2600 | PERROR("pollfd malloc"); |
fa29bfbf | 2601 | pthread_mutex_unlock(&the_consumer_data.lock); |
3bd1e081 MD |
2602 | goto end; |
2603 | } | |
2604 | ||
fa29bfbf SM |
2605 | local_stream = zmalloc((the_consumer_data.stream_count + |
2606 | nb_pipes_fd) * | |
747f8642 | 2607 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2608 | if (local_stream == NULL) { |
7a57cf92 | 2609 | PERROR("local_stream malloc"); |
fa29bfbf | 2610 | pthread_mutex_unlock(&the_consumer_data.lock); |
3bd1e081 MD |
2611 | goto end; |
2612 | } | |
ffe60014 | 2613 | ret = update_poll_array(ctx, &pollfd, local_stream, |
9a2fcf78 | 2614 | data_ht, &nb_inactive_fd); |
3bd1e081 MD |
2615 | if (ret < 0) { |
2616 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2617 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
fa29bfbf | 2618 | pthread_mutex_unlock(&the_consumer_data.lock); |
3bd1e081 MD |
2619 | goto end; |
2620 | } | |
2621 | nb_fd = ret; | |
fa29bfbf | 2622 | the_consumer_data.need_update = 0; |
3bd1e081 | 2623 | } |
fa29bfbf | 2624 | pthread_mutex_unlock(&the_consumer_data.lock); |
3bd1e081 | 2625 | |
4078b776 | 2626 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
9a2fcf78 JD |
2627 | if (nb_fd == 0 && nb_inactive_fd == 0 && |
2628 | CMM_LOAD_SHARED(consumer_quit) == 1) { | |
1fc79fb4 | 2629 | err = 0; /* All is OK */ |
4078b776 MD |
2630 | goto end; |
2631 | } | |
3bd1e081 | 2632 | /* poll on the array of fds */ |
88f2b785 | 2633 | restart: |
261de637 | 2634 | DBG("polling on %d fd", nb_fd + nb_pipes_fd); |
cf0bcb51 JG |
2635 | if (testpoint(consumerd_thread_data_poll)) { |
2636 | goto end; | |
2637 | } | |
9ce5646a | 2638 | health_poll_entry(); |
261de637 | 2639 | num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1); |
9ce5646a | 2640 | health_poll_exit(); |
3bd1e081 MD |
2641 | DBG("poll num_rdy : %d", num_rdy); |
2642 | if (num_rdy == -1) { | |
88f2b785 MD |
2643 | /* |
2644 | * Restart interrupted system call. | |
2645 | */ | |
2646 | if (errno == EINTR) { | |
2647 | goto restart; | |
2648 | } | |
7a57cf92 | 2649 | PERROR("Poll error"); |
f73fabfd | 2650 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2651 | goto end; |
2652 | } else if (num_rdy == 0) { | |
2653 | DBG("Polling thread timed out"); | |
2654 | goto end; | |
2655 | } | |
2656 | ||
80957876 JG |
2657 | if (caa_unlikely(data_consumption_paused)) { |
2658 | DBG("Data consumption paused, sleeping..."); | |
2659 | sleep(1); | |
2660 | goto restart; | |
2661 | } | |
2662 | ||
3bd1e081 | 2663 | /* |
50f8ae69 | 2664 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2665 | * beginning of the loop to update the array. We want to prioritize |
2666 | * array update over low-priority reads. | |
3bd1e081 | 2667 | */ |
509bb1cf | 2668 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2669 | ssize_t pipe_readlen; |
04fdd819 | 2670 | |
50f8ae69 | 2671 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2672 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2673 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2674 | if (pipe_readlen < sizeof(new_stream)) { |
2675 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2676 | /* Continue so we can at least handle the current stream(s). */ |
2677 | continue; | |
2678 | } | |
c869f647 DG |
2679 | |
2680 | /* | |
2681 | * If the stream is NULL, just ignore it. It's also possible that | |
2682 | * the sessiond poll thread changed the consumer_quit state and is | |
2683 | * waking us up to test it. | |
2684 | */ | |
2685 | if (new_stream == NULL) { | |
8994307f | 2686 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2687 | continue; |
2688 | } | |
2689 | ||
c869f647 | 2690 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2691 | continue; |
2692 | } | |
2693 | ||
02b3d176 DG |
2694 | /* Handle wakeup pipe. */ |
2695 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2696 | char dummy; | |
2697 | ssize_t pipe_readlen; | |
2698 | ||
2699 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2700 | sizeof(dummy)); | |
2701 | if (pipe_readlen < 0) { | |
2702 | PERROR("Consumer data wakeup pipe"); | |
2703 | } | |
2704 | /* We've been awakened to handle stream(s). */ | |
2705 | ctx->has_wakeup = 0; | |
2706 | } | |
2707 | ||
3bd1e081 MD |
2708 | /* Take care of high priority channels first. */ |
2709 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2710 | health_code_update(); |
2711 | ||
9617607b DG |
2712 | if (local_stream[i] == NULL) { |
2713 | continue; | |
2714 | } | |
fb3a43a9 | 2715 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2716 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2717 | high_prio = 1; | |
6f9449c2 | 2718 | len = ctx->on_buffer_ready(local_stream[i], ctx, false); |
d41f73b7 | 2719 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2720 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2721 | /* Clean the stream and free it. */ |
2722 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2723 | local_stream[i] = NULL; |
4078b776 MD |
2724 | } else if (len > 0) { |
2725 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2726 | } |
3bd1e081 MD |
2727 | } |
2728 | } | |
2729 | ||
4078b776 MD |
2730 | /* |
2731 | * If we read high prio channel in this loop, try again | |
2732 | * for more high prio data. | |
2733 | */ | |
2734 | if (high_prio) { | |
3bd1e081 MD |
2735 | continue; |
2736 | } | |
2737 | ||
2738 | /* Take care of low priority channels. */ | |
4078b776 | 2739 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2740 | health_code_update(); |
2741 | ||
9617607b DG |
2742 | if (local_stream[i] == NULL) { |
2743 | continue; | |
2744 | } | |
4078b776 | 2745 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2746 | local_stream[i]->hangup_flush_done || |
2747 | local_stream[i]->has_data) { | |
4078b776 | 2748 | DBG("Normal read on fd %d", pollfd[i].fd); |
6f9449c2 | 2749 | len = ctx->on_buffer_ready(local_stream[i], ctx, false); |
4078b776 | 2750 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2751 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2752 | /* Clean the stream and free it. */ |
2753 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2754 | local_stream[i] = NULL; |
4078b776 MD |
2755 | } else if (len > 0) { |
2756 | local_stream[i]->data_read = 1; | |
2757 | } | |
2758 | } | |
2759 | } | |
2760 | ||
2761 | /* Handle hangup and errors */ | |
2762 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2763 | health_code_update(); |
2764 | ||
9617607b DG |
2765 | if (local_stream[i] == NULL) { |
2766 | continue; | |
2767 | } | |
4078b776 MD |
2768 | if (!local_stream[i]->hangup_flush_done |
2769 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
fa29bfbf SM |
2770 | && (the_consumer_data.type == LTTNG_CONSUMER32_UST |
2771 | || the_consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
4078b776 | 2772 | DBG("fd %d is hup|err|nval. Attempting flush and read.", |
9617607b | 2773 | pollfd[i].fd); |
4078b776 MD |
2774 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2775 | /* Attempt read again, for the data we just flushed. */ | |
2776 | local_stream[i]->data_read = 1; | |
2777 | } | |
2778 | /* | |
2779 | * If the poll flag is HUP/ERR/NVAL and we have | |
2780 | * read no data in this pass, we can remove the | |
2781 | * stream from its hash table. | |
2782 | */ | |
2783 | if ((pollfd[i].revents & POLLHUP)) { | |
2784 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2785 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2786 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2787 | local_stream[i] = NULL; |
4078b776 MD |
2788 | num_hup++; |
2789 | } | |
2790 | } else if (pollfd[i].revents & POLLERR) { | |
2791 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2792 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2793 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2794 | local_stream[i] = NULL; |
4078b776 MD |
2795 | num_hup++; |
2796 | } | |
2797 | } else if (pollfd[i].revents & POLLNVAL) { | |
2798 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2799 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2800 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2801 | local_stream[i] = NULL; |
4078b776 | 2802 | num_hup++; |
3bd1e081 MD |
2803 | } |
2804 | } | |
9617607b DG |
2805 | if (local_stream[i] != NULL) { |
2806 | local_stream[i]->data_read = 0; | |
2807 | } | |
3bd1e081 MD |
2808 | } |
2809 | } | |
1fc79fb4 MD |
2810 | /* All is OK */ |
2811 | err = 0; | |
3bd1e081 MD |
2812 | end: |
2813 | DBG("polling thread exiting"); | |
0e428499 DG |
2814 | free(pollfd); |
2815 | free(local_stream); | |
fb3a43a9 DG |
2816 | |
2817 | /* | |
2818 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2819 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2820 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2821 | * not return and could create a endless wait period if the pipe is the | |
2822 | * only tracked fd in the poll set. The thread will take care of closing | |
2823 | * the read side. | |
fb3a43a9 | 2824 | */ |
13886d2d | 2825 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2826 | |
2d57de81 | 2827 | error_testpoint: |
1fc79fb4 MD |
2828 | if (err) { |
2829 | health_error(); | |
2830 | ERR("Health error occurred in %s", __func__); | |
2831 | } | |
2832 | health_unregister(health_consumerd); | |
2833 | ||
e7b994a3 | 2834 | rcu_unregister_thread(); |
3bd1e081 MD |
2835 | return NULL; |
2836 | } | |
2837 | ||
d8ef542d MD |
2838 | /* |
2839 | * Close wake-up end of each stream belonging to the channel. This will | |
2840 | * allow the poll() on the stream read-side to detect when the | |
2841 | * write-side (application) finally closes them. | |
2842 | */ | |
2843 | static | |
2844 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2845 | { | |
2846 | struct lttng_ht *ht; | |
2847 | struct lttng_consumer_stream *stream; | |
2848 | struct lttng_ht_iter iter; | |
2849 | ||
fa29bfbf | 2850 | ht = the_consumer_data.stream_per_chan_id_ht; |
d8ef542d MD |
2851 | |
2852 | rcu_read_lock(); | |
2853 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2854 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2855 | ht->match_fct, &channel->key, | |
2856 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2857 | /* |
2858 | * Protect against teardown with mutex. | |
2859 | */ | |
2860 | pthread_mutex_lock(&stream->lock); | |
2861 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2862 | goto next; | |
2863 | } | |
fa29bfbf | 2864 | switch (the_consumer_data.type) { |
d8ef542d MD |
2865 | case LTTNG_CONSUMER_KERNEL: |
2866 | break; | |
2867 | case LTTNG_CONSUMER32_UST: | |
2868 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2869 | if (stream->metadata_flag) { |
2870 | /* Safe and protected by the stream lock. */ | |
2871 | lttng_ustconsumer_close_metadata(stream->chan); | |
2872 | } else { | |
2873 | /* | |
2874 | * Note: a mutex is taken internally within | |
2875 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2876 | * use from concurrent close. | |
2877 | */ | |
2878 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2879 | } | |
d8ef542d MD |
2880 | break; |
2881 | default: | |
2882 | ERR("Unknown consumer_data type"); | |
2883 | assert(0); | |
2884 | } | |
f2ad556d MD |
2885 | next: |
2886 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2887 | } |
2888 | rcu_read_unlock(); | |
2889 | } | |
2890 | ||
2891 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2892 | { | |
2893 | struct lttng_ht_iter iter; | |
2894 | struct lttng_consumer_channel *channel; | |
2895 | int ret; | |
2896 | ||
2897 | if (ht == NULL) { | |
2898 | return; | |
2899 | } | |
2900 | ||
2901 | rcu_read_lock(); | |
2902 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2903 | ret = lttng_ht_del(ht, &iter); | |
2904 | assert(ret != 0); | |
2905 | } | |
2906 | rcu_read_unlock(); | |
2907 | ||
2908 | lttng_ht_destroy(ht); | |
2909 | } | |
2910 | ||
2911 | /* | |
2912 | * This thread polls the channel fds to detect when they are being | |
2913 | * closed. It closes all related streams if the channel is detected as | |
2914 | * closed. It is currently only used as a shim layer for UST because the | |
2915 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2916 | * periodical flush. | |
2917 | */ | |
2918 | void *consumer_thread_channel_poll(void *data) | |
2919 | { | |
1fc79fb4 | 2920 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2921 | uint32_t revents, nb_fd; |
2922 | struct lttng_consumer_channel *chan = NULL; | |
2923 | struct lttng_ht_iter iter; | |
2924 | struct lttng_ht_node_u64 *node; | |
2925 | struct lttng_poll_event events; | |
2926 | struct lttng_consumer_local_data *ctx = data; | |
2927 | struct lttng_ht *channel_ht; | |
2928 | ||
2929 | rcu_register_thread(); | |
2930 | ||
1fc79fb4 MD |
2931 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2932 | ||
2d57de81 MD |
2933 | if (testpoint(consumerd_thread_channel)) { |
2934 | goto error_testpoint; | |
2935 | } | |
2936 | ||
9ce5646a MD |
2937 | health_code_update(); |
2938 | ||
d8ef542d MD |
2939 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2940 | if (!channel_ht) { | |
2941 | /* ENOMEM at this point. Better to bail out. */ | |
2942 | goto end_ht; | |
2943 | } | |
2944 | ||
2945 | DBG("Thread channel poll started"); | |
2946 | ||
2947 | /* Size is set to 1 for the consumer_channel pipe */ | |
2948 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2949 | if (ret < 0) { | |
2950 | ERR("Poll set creation failed"); | |
2951 | goto end_poll; | |
2952 | } | |
2953 | ||
2954 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2955 | if (ret < 0) { | |
2956 | goto end; | |
2957 | } | |
2958 | ||
2959 | /* Main loop */ | |
2960 | DBG("Channel main loop started"); | |
2961 | ||
2962 | while (1) { | |
d8ef542d | 2963 | restart: |
7fa2082e MD |
2964 | health_code_update(); |
2965 | DBG("Channel poll wait"); | |
9ce5646a | 2966 | health_poll_entry(); |
d8ef542d | 2967 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2968 | DBG("Channel poll return from wait with %d fd(s)", |
2969 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2970 | health_poll_exit(); |
40063ead | 2971 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2972 | if (ret < 0) { |
2973 | if (errno == EINTR) { | |
40063ead | 2974 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2975 | goto restart; |
2976 | } | |
d9607cd7 MD |
2977 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2978 | err = 0; /* All is OK */ | |
2979 | } | |
d8ef542d MD |
2980 | goto end; |
2981 | } | |
2982 | ||
2983 | nb_fd = ret; | |
2984 | ||
2985 | /* From here, the event is a channel wait fd */ | |
2986 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2987 | health_code_update(); |
2988 | ||
d8ef542d MD |
2989 | revents = LTTNG_POLL_GETEV(&events, i); |
2990 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2991 | ||
d8ef542d | 2992 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
03e43155 | 2993 | if (revents & LPOLLIN) { |
d8ef542d | 2994 | enum consumer_channel_action action; |
a0cbdd2e | 2995 | uint64_t key; |
d8ef542d | 2996 | |
a0cbdd2e | 2997 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2998 | if (ret <= 0) { |
03e43155 MD |
2999 | if (ret < 0) { |
3000 | ERR("Error reading channel pipe"); | |
3001 | } | |
3002 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
3003 | continue; |
3004 | } | |
3005 | ||
3006 | switch (action) { | |
3007 | case CONSUMER_CHANNEL_ADD: | |
3008 | DBG("Adding channel %d to poll set", | |
3009 | chan->wait_fd); | |
3010 | ||
3011 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
3012 | chan->wait_fd); | |
c7260a81 | 3013 | rcu_read_lock(); |
d8ef542d MD |
3014 | lttng_ht_add_unique_u64(channel_ht, |
3015 | &chan->wait_fd_node); | |
c7260a81 | 3016 | rcu_read_unlock(); |
d8ef542d MD |
3017 | /* Add channel to the global poll events list */ |
3018 | lttng_poll_add(&events, chan->wait_fd, | |
03e43155 | 3019 | LPOLLERR | LPOLLHUP); |
d8ef542d | 3020 | break; |
a0cbdd2e MD |
3021 | case CONSUMER_CHANNEL_DEL: |
3022 | { | |
b4a650f3 DG |
3023 | /* |
3024 | * This command should never be called if the channel | |
3025 | * has streams monitored by either the data or metadata | |
3026 | * thread. The consumer only notify this thread with a | |
3027 | * channel del. command if it receives a destroy | |
3028 | * channel command from the session daemon that send it | |
3029 | * if a command prior to the GET_CHANNEL failed. | |
3030 | */ | |
3031 | ||
c7260a81 | 3032 | rcu_read_lock(); |
a0cbdd2e MD |
3033 | chan = consumer_find_channel(key); |
3034 | if (!chan) { | |
c7260a81 | 3035 | rcu_read_unlock(); |
a0cbdd2e MD |
3036 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
3037 | break; | |
3038 | } | |
3039 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 3040 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
3041 | ret = lttng_ht_del(channel_ht, &iter); |
3042 | assert(ret == 0); | |
a0cbdd2e | 3043 | |
fa29bfbf | 3044 | switch (the_consumer_data.type) { |
f2a444f1 DG |
3045 | case LTTNG_CONSUMER_KERNEL: |
3046 | break; | |
3047 | case LTTNG_CONSUMER32_UST: | |
3048 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
3049 | health_code_update(); |
3050 | /* Destroy streams that might have been left in the stream list. */ | |
3051 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
3052 | break; |
3053 | default: | |
3054 | ERR("Unknown consumer_data type"); | |
3055 | assert(0); | |
3056 | } | |
3057 | ||
a0cbdd2e MD |
3058 | /* |
3059 | * Release our own refcount. Force channel deletion even if | |
3060 | * streams were not initialized. | |
3061 | */ | |
3062 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
3063 | consumer_del_channel(chan); | |
3064 | } | |
c7260a81 | 3065 | rcu_read_unlock(); |
a0cbdd2e MD |
3066 | goto restart; |
3067 | } | |
d8ef542d MD |
3068 | case CONSUMER_CHANNEL_QUIT: |
3069 | /* | |
3070 | * Remove the pipe from the poll set and continue the loop | |
3071 | * since their might be data to consume. | |
3072 | */ | |
3073 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3074 | continue; | |
3075 | default: | |
3076 | ERR("Unknown action"); | |
3077 | break; | |
3078 | } | |
03e43155 MD |
3079 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
3080 | DBG("Channel thread pipe hung up"); | |
3081 | /* | |
3082 | * Remove the pipe from the poll set and continue the loop | |
3083 | * since their might be data to consume. | |
3084 | */ | |
3085 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3086 | continue; | |
3087 | } else { | |
3088 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3089 | goto end; | |
d8ef542d MD |
3090 | } |
3091 | ||
3092 | /* Handle other stream */ | |
3093 | continue; | |
3094 | } | |
3095 | ||
3096 | rcu_read_lock(); | |
3097 | { | |
3098 | uint64_t tmp_id = (uint64_t) pollfd; | |
3099 | ||
3100 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
3101 | } | |
3102 | node = lttng_ht_iter_get_node_u64(&iter); | |
3103 | assert(node); | |
3104 | ||
3105 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
3106 | wait_fd_node); | |
3107 | ||
3108 | /* Check for error event */ | |
3109 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
3110 | DBG("Channel fd %d is hup|err.", pollfd); | |
3111 | ||
3112 | lttng_poll_del(&events, chan->wait_fd); | |
3113 | ret = lttng_ht_del(channel_ht, &iter); | |
3114 | assert(ret == 0); | |
b4a650f3 DG |
3115 | |
3116 | /* | |
3117 | * This will close the wait fd for each stream associated to | |
3118 | * this channel AND monitored by the data/metadata thread thus | |
3119 | * will be clean by the right thread. | |
3120 | */ | |
d8ef542d | 3121 | consumer_close_channel_streams(chan); |
f2ad556d MD |
3122 | |
3123 | /* Release our own refcount */ | |
3124 | if (!uatomic_sub_return(&chan->refcount, 1) | |
3125 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
3126 | consumer_del_channel(chan); | |
3127 | } | |
03e43155 MD |
3128 | } else { |
3129 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3130 | rcu_read_unlock(); | |
3131 | goto end; | |
d8ef542d MD |
3132 | } |
3133 | ||
3134 | /* Release RCU lock for the channel looked up */ | |
3135 | rcu_read_unlock(); | |
3136 | } | |
3137 | } | |
3138 | ||
1fc79fb4 MD |
3139 | /* All is OK */ |
3140 | err = 0; | |
d8ef542d MD |
3141 | end: |
3142 | lttng_poll_clean(&events); | |
3143 | end_poll: | |
3144 | destroy_channel_ht(channel_ht); | |
3145 | end_ht: | |
2d57de81 | 3146 | error_testpoint: |
d8ef542d | 3147 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
3148 | if (err) { |
3149 | health_error(); | |
3150 | ERR("Health error occurred in %s", __func__); | |
3151 | } | |
3152 | health_unregister(health_consumerd); | |
d8ef542d MD |
3153 | rcu_unregister_thread(); |
3154 | return NULL; | |
3155 | } | |
3156 | ||
331744e3 JD |
3157 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
3158 | struct pollfd *sockpoll, int client_socket) | |
3159 | { | |
3160 | int ret; | |
3161 | ||
3162 | assert(ctx); | |
3163 | assert(sockpoll); | |
3164 | ||
84382d49 MD |
3165 | ret = lttng_consumer_poll_socket(sockpoll); |
3166 | if (ret) { | |
331744e3 JD |
3167 | goto error; |
3168 | } | |
3169 | DBG("Metadata connection on client_socket"); | |
3170 | ||
3171 | /* Blocking call, waiting for transmission */ | |
3172 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3173 | if (ctx->consumer_metadata_socket < 0) { | |
3174 | WARN("On accept metadata"); | |
3175 | ret = -1; | |
3176 | goto error; | |
3177 | } | |
3178 | ret = 0; | |
3179 | ||
3180 | error: | |
3181 | return ret; | |
3182 | } | |
3183 | ||
3bd1e081 MD |
3184 | /* |
3185 | * This thread listens on the consumerd socket and receives the file | |
3186 | * descriptors from the session daemon. | |
3187 | */ | |
7d980def | 3188 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3189 | { |
1fc79fb4 | 3190 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3191 | /* |
3192 | * structure to poll for incoming data on communication socket avoids | |
3193 | * making blocking sockets. | |
3194 | */ | |
3195 | struct pollfd consumer_sockpoll[2]; | |
3196 | struct lttng_consumer_local_data *ctx = data; | |
3197 | ||
e7b994a3 DG |
3198 | rcu_register_thread(); |
3199 | ||
1fc79fb4 MD |
3200 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3201 | ||
2d57de81 MD |
3202 | if (testpoint(consumerd_thread_sessiond)) { |
3203 | goto error_testpoint; | |
3204 | } | |
3205 | ||
9ce5646a MD |
3206 | health_code_update(); |
3207 | ||
3bd1e081 MD |
3208 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3209 | unlink(ctx->consumer_command_sock_path); | |
3210 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3211 | if (client_socket < 0) { | |
3212 | ERR("Cannot create command socket"); | |
3213 | goto end; | |
3214 | } | |
3215 | ||
3216 | ret = lttcomm_listen_unix_sock(client_socket); | |
3217 | if (ret < 0) { | |
3218 | goto end; | |
3219 | } | |
3220 | ||
32258573 | 3221 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3222 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3223 | /* return < 0 on error, but == 0 is not fatal */ |
3224 | if (ret < 0) { | |
32258573 | 3225 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3226 | goto end; |
3227 | } | |
3228 | ||
3bd1e081 MD |
3229 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3230 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3231 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3232 | consumer_sockpoll[1].fd = client_socket; | |
3233 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3234 | ||
84382d49 MD |
3235 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3236 | if (ret) { | |
3237 | if (ret > 0) { | |
3238 | /* should exit */ | |
3239 | err = 0; | |
3240 | } | |
3bd1e081 MD |
3241 | goto end; |
3242 | } | |
3243 | DBG("Connection on client_socket"); | |
3244 | ||
3245 | /* Blocking call, waiting for transmission */ | |
3246 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3247 | if (sock < 0) { |
3bd1e081 MD |
3248 | WARN("On accept"); |
3249 | goto end; | |
3250 | } | |
3bd1e081 | 3251 | |
331744e3 JD |
3252 | /* |
3253 | * Setup metadata socket which is the second socket connection on the | |
3254 | * command unix socket. | |
3255 | */ | |
3256 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3257 | if (ret) { |
3258 | if (ret > 0) { | |
3259 | /* should exit */ | |
3260 | err = 0; | |
3261 | } | |
331744e3 JD |
3262 | goto end; |
3263 | } | |
3264 | ||
d96f09c6 DG |
3265 | /* This socket is not useful anymore. */ |
3266 | ret = close(client_socket); | |
3267 | if (ret < 0) { | |
3268 | PERROR("close client_socket"); | |
3269 | } | |
3270 | client_socket = -1; | |
3271 | ||
3bd1e081 MD |
3272 | /* update the polling structure to poll on the established socket */ |
3273 | consumer_sockpoll[1].fd = sock; | |
3274 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3275 | ||
3276 | while (1) { | |
9ce5646a MD |
3277 | health_code_update(); |
3278 | ||
3279 | health_poll_entry(); | |
3280 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3281 | health_poll_exit(); | |
84382d49 MD |
3282 | if (ret) { |
3283 | if (ret > 0) { | |
3284 | /* should exit */ | |
3285 | err = 0; | |
3286 | } | |
3bd1e081 MD |
3287 | goto end; |
3288 | } | |
3289 | DBG("Incoming command on sock"); | |
3290 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3291 | if (ret <= 0) { |
3292 | /* | |
3293 | * This could simply be a session daemon quitting. Don't output | |
3294 | * ERR() here. | |
3295 | */ | |
3296 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3297 | err = 0; |
3bd1e081 MD |
3298 | goto end; |
3299 | } | |
10211f5c | 3300 | if (CMM_LOAD_SHARED(consumer_quit)) { |
3bd1e081 | 3301 | DBG("consumer_thread_receive_fds received quit from signal"); |
1fc79fb4 | 3302 | err = 0; /* All is OK */ |
3bd1e081 MD |
3303 | goto end; |
3304 | } | |
a1ed855a | 3305 | DBG("Received command on sock"); |
3bd1e081 | 3306 | } |
1fc79fb4 MD |
3307 | /* All is OK */ |
3308 | err = 0; | |
3309 | ||
3bd1e081 | 3310 | end: |
ffe60014 | 3311 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3312 | |
d88aee68 DG |
3313 | /* |
3314 | * Close metadata streams since the producer is the session daemon which | |
3315 | * just died. | |
3316 | * | |
3317 | * NOTE: for now, this only applies to the UST tracer. | |
3318 | */ | |
6d574024 | 3319 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3320 | |
3bd1e081 MD |
3321 | /* |
3322 | * when all fds have hung up, the polling thread | |
3323 | * can exit cleanly | |
3324 | */ | |
10211f5c | 3325 | CMM_STORE_SHARED(consumer_quit, 1); |
3bd1e081 | 3326 | |
04fdd819 | 3327 | /* |
c869f647 | 3328 | * Notify the data poll thread to poll back again and test the |
8994307f | 3329 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3330 | */ |
acdb9057 | 3331 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3332 | |
a0cbdd2e | 3333 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3334 | |
5c635c72 MD |
3335 | notify_health_quit_pipe(health_quit_pipe); |
3336 | ||
d96f09c6 DG |
3337 | /* Cleaning up possibly open sockets. */ |
3338 | if (sock >= 0) { | |
3339 | ret = close(sock); | |
3340 | if (ret < 0) { | |
3341 | PERROR("close sock sessiond poll"); | |
3342 | } | |
3343 | } | |
3344 | if (client_socket >= 0) { | |
38476d24 | 3345 | ret = close(client_socket); |
d96f09c6 DG |
3346 | if (ret < 0) { |
3347 | PERROR("close client_socket sessiond poll"); | |
3348 | } | |
3349 | } | |
3350 | ||
2d57de81 | 3351 | error_testpoint: |
1fc79fb4 MD |
3352 | if (err) { |
3353 | health_error(); | |
3354 | ERR("Health error occurred in %s", __func__); | |
3355 | } | |
3356 | health_unregister(health_consumerd); | |
3357 | ||
e7b994a3 | 3358 | rcu_unregister_thread(); |
3bd1e081 MD |
3359 | return NULL; |
3360 | } | |
d41f73b7 | 3361 | |
503fefca JG |
3362 | static int post_consume(struct lttng_consumer_stream *stream, |
3363 | const struct stream_subbuffer *subbuffer, | |
3364 | struct lttng_consumer_local_data *ctx) | |
f96af312 | 3365 | { |
503fefca | 3366 | size_t i; |
f96af312 | 3367 | int ret = 0; |
503fefca JG |
3368 | const size_t count = lttng_dynamic_array_get_count( |
3369 | &stream->read_subbuffer_ops.post_consume_cbs); | |
f96af312 | 3370 | |
503fefca JG |
3371 | for (i = 0; i < count; i++) { |
3372 | const post_consume_cb op = *(post_consume_cb *) lttng_dynamic_array_get_element( | |
3373 | &stream->read_subbuffer_ops.post_consume_cbs, | |
3374 | i); | |
3375 | ||
3376 | ret = op(stream, subbuffer, ctx); | |
3377 | if (ret) { | |
3378 | goto end; | |
f96af312 | 3379 | } |
f96af312 | 3380 | } |
f96af312 JG |
3381 | end: |
3382 | return ret; | |
3383 | } | |
3384 | ||
4078b776 | 3385 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
6f9449c2 JG |
3386 | struct lttng_consumer_local_data *ctx, |
3387 | bool locked_by_caller) | |
d41f73b7 | 3388 | { |
12bddd1d | 3389 | ssize_t ret, written_bytes = 0; |
23d56598 | 3390 | int rotation_ret; |
6f9449c2 | 3391 | struct stream_subbuffer subbuffer = {}; |
823a3926 | 3392 | enum get_next_subbuffer_status get_next_status; |
74251bb8 | 3393 | |
6f9449c2 JG |
3394 | if (!locked_by_caller) { |
3395 | stream->read_subbuffer_ops.lock(stream); | |
a48ac75a JR |
3396 | } else { |
3397 | stream->read_subbuffer_ops.assert_locked(stream); | |
6f9449c2 JG |
3398 | } |
3399 | ||
3400 | if (stream->read_subbuffer_ops.on_wake_up) { | |
3401 | ret = stream->read_subbuffer_ops.on_wake_up(stream); | |
3402 | if (ret) { | |
3403 | goto end; | |
3404 | } | |
94d49140 | 3405 | } |
74251bb8 | 3406 | |
23d56598 JG |
3407 | /* |
3408 | * If the stream was flagged to be ready for rotation before we extract | |
3409 | * the next packet, rotate it now. | |
3410 | */ | |
3411 | if (stream->rotate_ready) { | |
3412 | DBG("Rotate stream before consuming data"); | |
3413 | ret = lttng_consumer_rotate_stream(ctx, stream); | |
3414 | if (ret < 0) { | |
3415 | ERR("Stream rotation error before consuming data"); | |
3416 | goto end; | |
3417 | } | |
3418 | } | |
3419 | ||
823a3926 JG |
3420 | get_next_status = stream->read_subbuffer_ops.get_next_subbuffer( |
3421 | stream, &subbuffer); | |
3422 | switch (get_next_status) { | |
3423 | case GET_NEXT_SUBBUFFER_STATUS_OK: | |
3424 | break; | |
3425 | case GET_NEXT_SUBBUFFER_STATUS_NO_DATA: | |
3426 | /* Not an error. */ | |
3427 | ret = 0; | |
3428 | goto sleep_stream; | |
3429 | case GET_NEXT_SUBBUFFER_STATUS_ERROR: | |
3430 | ret = -1; | |
6f9449c2 | 3431 | goto end; |
823a3926 JG |
3432 | default: |
3433 | abort(); | |
d41f73b7 | 3434 | } |
74251bb8 | 3435 | |
6f9449c2 JG |
3436 | ret = stream->read_subbuffer_ops.pre_consume_subbuffer( |
3437 | stream, &subbuffer); | |
3438 | if (ret) { | |
3439 | goto error_put_subbuf; | |
3440 | } | |
3441 | ||
3442 | written_bytes = stream->read_subbuffer_ops.consume_subbuffer( | |
3443 | ctx, stream, &subbuffer); | |
514775d9 FD |
3444 | if (written_bytes <= 0) { |
3445 | ERR("Error consuming subbuffer: (%zd)", written_bytes); | |
3446 | ret = (int) written_bytes; | |
3447 | goto error_put_subbuf; | |
6f9449c2 JG |
3448 | } |
3449 | ||
3450 | ret = stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer); | |
3451 | if (ret) { | |
23d56598 JG |
3452 | goto end; |
3453 | } | |
3454 | ||
503fefca JG |
3455 | ret = post_consume(stream, &subbuffer, ctx); |
3456 | if (ret) { | |
3457 | goto end; | |
6f9449c2 JG |
3458 | } |
3459 | ||
23d56598 JG |
3460 | /* |
3461 | * After extracting the packet, we check if the stream is now ready to | |
3462 | * be rotated and perform the action immediately. | |
3463 | * | |
3464 | * Don't overwrite `ret` as callers expect the number of bytes | |
3465 | * consumed to be returned on success. | |
3466 | */ | |
3467 | rotation_ret = lttng_consumer_stream_is_rotate_ready(stream); | |
3468 | if (rotation_ret == 1) { | |
3469 | rotation_ret = lttng_consumer_rotate_stream(ctx, stream); | |
3470 | if (rotation_ret < 0) { | |
3471 | ret = rotation_ret; | |
3472 | ERR("Stream rotation error after consuming data"); | |
3473 | goto end; | |
3474 | } | |
503fefca | 3475 | |
23d56598 JG |
3476 | } else if (rotation_ret < 0) { |
3477 | ret = rotation_ret; | |
3478 | ERR("Failed to check if stream was ready to rotate after consuming data"); | |
3479 | goto end; | |
3480 | } | |
3481 | ||
82e72193 | 3482 | sleep_stream: |
6f9449c2 JG |
3483 | if (stream->read_subbuffer_ops.on_sleep) { |
3484 | stream->read_subbuffer_ops.on_sleep(stream, ctx); | |
3485 | } | |
3486 | ||
3487 | ret = written_bytes; | |
23d56598 | 3488 | end: |
6f9449c2 JG |
3489 | if (!locked_by_caller) { |
3490 | stream->read_subbuffer_ops.unlock(stream); | |
94d49140 | 3491 | } |
6f9449c2 | 3492 | |
74251bb8 | 3493 | return ret; |
6f9449c2 JG |
3494 | error_put_subbuf: |
3495 | (void) stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer); | |
3496 | goto end; | |
d41f73b7 MD |
3497 | } |
3498 | ||
3499 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3500 | { | |
fa29bfbf | 3501 | switch (the_consumer_data.type) { |
d41f73b7 MD |
3502 | case LTTNG_CONSUMER_KERNEL: |
3503 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3504 | case LTTNG_CONSUMER32_UST: |
3505 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3506 | return lttng_ustconsumer_on_recv_stream(stream); |
3507 | default: | |
3508 | ERR("Unknown consumer_data type"); | |
3509 | assert(0); | |
3510 | return -ENOSYS; | |
3511 | } | |
3512 | } | |
e4421fec DG |
3513 | |
3514 | /* | |
3515 | * Allocate and set consumer data hash tables. | |
3516 | */ | |
282dadbc | 3517 | int lttng_consumer_init(void) |
e4421fec | 3518 | { |
fa29bfbf SM |
3519 | the_consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3520 | if (!the_consumer_data.channel_ht) { | |
282dadbc MD |
3521 | goto error; |
3522 | } | |
3523 | ||
fa29bfbf | 3524 | the_consumer_data.channels_by_session_id_ht = |
5c3892a6 | 3525 | lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
fa29bfbf | 3526 | if (!the_consumer_data.channels_by_session_id_ht) { |
5c3892a6 JG |
3527 | goto error; |
3528 | } | |
3529 | ||
fa29bfbf SM |
3530 | the_consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3531 | if (!the_consumer_data.relayd_ht) { | |
282dadbc MD |
3532 | goto error; |
3533 | } | |
3534 | ||
fa29bfbf SM |
3535 | the_consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3536 | if (!the_consumer_data.stream_list_ht) { | |
282dadbc MD |
3537 | goto error; |
3538 | } | |
3539 | ||
fa29bfbf SM |
3540 | the_consumer_data.stream_per_chan_id_ht = |
3541 | lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3542 | if (!the_consumer_data.stream_per_chan_id_ht) { | |
282dadbc MD |
3543 | goto error; |
3544 | } | |
3545 | ||
3546 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3547 | if (!data_ht) { | |
3548 | goto error; | |
3549 | } | |
3550 | ||
3551 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3552 | if (!metadata_ht) { | |
3553 | goto error; | |
3554 | } | |
3555 | ||
fa29bfbf SM |
3556 | the_consumer_data.chunk_registry = lttng_trace_chunk_registry_create(); |
3557 | if (!the_consumer_data.chunk_registry) { | |
28cc88f3 JG |
3558 | goto error; |
3559 | } | |
3560 | ||
282dadbc MD |
3561 | return 0; |
3562 | ||
3563 | error: | |
3564 | return -1; | |
e4421fec | 3565 | } |
7735ef9e DG |
3566 | |
3567 | /* | |
3568 | * Process the ADD_RELAYD command receive by a consumer. | |
3569 | * | |
3570 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3571 | * The caller MUST acquire a RCU read side lock before calling it. | |
3572 | */ | |
fe7bf564 JR |
3573 | void consumer_add_relayd_socket(uint64_t net_seq_idx, |
3574 | int sock_type, | |
3575 | struct lttng_consumer_local_data *ctx, | |
3576 | int sock, | |
6151a90f | 3577 | struct pollfd *consumer_sockpoll, |
fe7bf564 JR |
3578 | uint64_t sessiond_id, |
3579 | uint64_t relayd_session_id, | |
3580 | uint32_t relayd_version_major, | |
3581 | uint32_t relayd_version_minor, | |
3582 | enum lttcomm_sock_proto relayd_socket_protocol) | |
7735ef9e | 3583 | { |
cd2b09ed | 3584 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3585 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3586 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3587 | |
6151a90f | 3588 | assert(ctx); |
6151a90f | 3589 | |
da009f2c | 3590 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3591 | |
3592 | /* Get relayd reference if exists. */ | |
3593 | relayd = consumer_find_relayd(net_seq_idx); | |
3594 | if (relayd == NULL) { | |
da009f2c | 3595 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3596 | /* Not found. Allocate one. */ |
3597 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3598 | if (relayd == NULL) { | |
618a6a28 MD |
3599 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3600 | goto error; | |
0d08d75e | 3601 | } else { |
30319bcb | 3602 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3603 | relayd_created = 1; |
7735ef9e | 3604 | } |
0d08d75e DG |
3605 | |
3606 | /* | |
3607 | * This code path MUST continue to the consumer send status message to | |
3608 | * we can notify the session daemon and continue our work without | |
3609 | * killing everything. | |
3610 | */ | |
da009f2c MD |
3611 | } else { |
3612 | /* | |
3613 | * relayd key should never be found for control socket. | |
3614 | */ | |
3615 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3616 | } |
3617 | ||
3618 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3619 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3620 | if (ret < 0) { |
0d08d75e | 3621 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3622 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3623 | goto error_nosignal; | |
7735ef9e DG |
3624 | } |
3625 | ||
3626 | /* Poll on consumer socket. */ | |
84382d49 MD |
3627 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3628 | if (ret) { | |
3629 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3630 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
618a6a28 | 3631 | goto error_nosignal; |
7735ef9e DG |
3632 | } |
3633 | ||
3634 | /* Get relayd socket from session daemon */ | |
3635 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3636 | if (ret != sizeof(fd)) { | |
4028eeb9 | 3637 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3638 | |
3639 | /* | |
3640 | * Failing to receive FDs might indicate a major problem such as | |
3641 | * reaching a fd limit during the receive where the kernel returns a | |
3642 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3643 | * don't take any chances and stop everything. | |
3644 | * | |
3645 | * XXX: Feature request #558 will fix that and avoid this possible | |
3646 | * issue when reaching the fd limit. | |
3647 | */ | |
3648 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3649 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3650 | goto error; |
3651 | } | |
3652 | ||
7735ef9e DG |
3653 | /* Copy socket information and received FD */ |
3654 | switch (sock_type) { | |
3655 | case LTTNG_STREAM_CONTROL: | |
3656 | /* Copy received lttcomm socket */ | |
fe7bf564 JR |
3657 | ret = lttcomm_populate_sock_from_open_socket( |
3658 | &relayd->control_sock.sock, fd, | |
3659 | relayd_socket_protocol); | |
7735ef9e | 3660 | |
6151a90f | 3661 | /* Assign version values. */ |
fe7bf564 JR |
3662 | relayd->control_sock.major = relayd_version_major; |
3663 | relayd->control_sock.minor = relayd_version_minor; | |
c5b6f4f0 | 3664 | |
d3e2ba59 | 3665 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3666 | |
7735ef9e DG |
3667 | break; |
3668 | case LTTNG_STREAM_DATA: | |
3669 | /* Copy received lttcomm socket */ | |
fe7bf564 JR |
3670 | ret = lttcomm_populate_sock_from_open_socket( |
3671 | &relayd->data_sock.sock, fd, | |
3672 | relayd_socket_protocol); | |
6151a90f | 3673 | /* Assign version values. */ |
fe7bf564 JR |
3674 | relayd->data_sock.major = relayd_version_major; |
3675 | relayd->data_sock.minor = relayd_version_minor; | |
7735ef9e DG |
3676 | break; |
3677 | default: | |
3678 | ERR("Unknown relayd socket type (%d)", sock_type); | |
618a6a28 | 3679 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3680 | goto error; |
3681 | } | |
3682 | ||
fe7bf564 JR |
3683 | if (ret < 0) { |
3684 | ret_code = LTTCOMM_CONSUMERD_FATAL; | |
3685 | goto error; | |
3686 | } | |
3687 | ||
d88aee68 | 3688 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3689 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3690 | relayd->net_seq_idx, fd); | |
39d9954c FD |
3691 | /* |
3692 | * We gave the ownership of the fd to the relayd structure. Set the | |
3693 | * fd to -1 so we don't call close() on it in the error path below. | |
3694 | */ | |
3695 | fd = -1; | |
7735ef9e | 3696 | |
618a6a28 MD |
3697 | /* We successfully added the socket. Send status back. */ |
3698 | ret = consumer_send_status_msg(sock, ret_code); | |
3699 | if (ret < 0) { | |
3700 | /* Somehow, the session daemon is not responding anymore. */ | |
3701 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3702 | goto error_nosignal; | |
3703 | } | |
3704 | ||
7735ef9e DG |
3705 | /* |
3706 | * Add relayd socket pair to consumer data hashtable. If object already | |
3707 | * exists or on error, the function gracefully returns. | |
3708 | */ | |
9276e5c8 | 3709 | relayd->ctx = ctx; |
d09e1200 | 3710 | add_relayd(relayd); |
7735ef9e DG |
3711 | |
3712 | /* All good! */ | |
2527bf85 | 3713 | return; |
7735ef9e DG |
3714 | |
3715 | error: | |
618a6a28 MD |
3716 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3717 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3718 | } | |
3719 | ||
3720 | error_nosignal: | |
4028eeb9 DG |
3721 | /* Close received socket if valid. */ |
3722 | if (fd >= 0) { | |
3723 | if (close(fd)) { | |
3724 | PERROR("close received socket"); | |
3725 | } | |
3726 | } | |
cd2b09ed DG |
3727 | |
3728 | if (relayd_created) { | |
cd2b09ed DG |
3729 | free(relayd); |
3730 | } | |
7735ef9e | 3731 | } |
ca22feea | 3732 | |
f7079f67 DG |
3733 | /* |
3734 | * Search for a relayd associated to the session id and return the reference. | |
3735 | * | |
3736 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3737 | * until the relayd object is no longer necessary. | |
3738 | */ | |
3739 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3740 | { | |
3741 | struct lttng_ht_iter iter; | |
f7079f67 | 3742 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3743 | |
3744 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
fa29bfbf SM |
3745 | cds_lfht_for_each_entry(the_consumer_data.relayd_ht->ht, &iter.iter, |
3746 | relayd, node.node) { | |
18261bd1 DG |
3747 | /* |
3748 | * Check by sessiond id which is unique here where the relayd session | |
3749 | * id might not be when having multiple relayd. | |
3750 | */ | |
3751 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3752 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3753 | goto found; |
f7079f67 DG |
3754 | } |
3755 | } | |
3756 | ||
18261bd1 DG |
3757 | return NULL; |
3758 | ||
3759 | found: | |
f7079f67 DG |
3760 | return relayd; |
3761 | } | |
3762 | ||
ca22feea DG |
3763 | /* |
3764 | * Check if for a given session id there is still data needed to be extract | |
3765 | * from the buffers. | |
3766 | * | |
6d805429 | 3767 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3768 | */ |
6d805429 | 3769 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3770 | { |
3771 | int ret; | |
3772 | struct lttng_ht_iter iter; | |
3773 | struct lttng_ht *ht; | |
3774 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3775 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3776 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3777 | |
6d805429 | 3778 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3779 | |
6f6eda74 | 3780 | rcu_read_lock(); |
fa29bfbf | 3781 | pthread_mutex_lock(&the_consumer_data.lock); |
ca22feea | 3782 | |
fa29bfbf | 3783 | switch (the_consumer_data.type) { |
ca22feea | 3784 | case LTTNG_CONSUMER_KERNEL: |
6d805429 | 3785 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3786 | break; |
3787 | case LTTNG_CONSUMER32_UST: | |
3788 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3789 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3790 | break; |
3791 | default: | |
3792 | ERR("Unknown consumer data type"); | |
3793 | assert(0); | |
3794 | } | |
3795 | ||
3796 | /* Ease our life a bit */ | |
fa29bfbf | 3797 | ht = the_consumer_data.stream_list_ht; |
ca22feea | 3798 | |
c8f59ee5 | 3799 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3800 | ht->hash_fct(&id, lttng_ht_seed), |
3801 | ht->match_fct, &id, | |
ca22feea | 3802 | &iter.iter, stream, node_session_id.node) { |
bb586a6e | 3803 | pthread_mutex_lock(&stream->lock); |
ca22feea | 3804 | |
4e9a4686 DG |
3805 | /* |
3806 | * A removed node from the hash table indicates that the stream has | |
3807 | * been deleted thus having a guarantee that the buffers are closed | |
3808 | * on the consumer side. However, data can still be transmitted | |
3809 | * over the network so don't skip the relayd check. | |
3810 | */ | |
3811 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3812 | if (!ret) { | |
3813 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3814 | ret = data_pending(stream); |
3815 | if (ret == 1) { | |
4e9a4686 | 3816 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3817 | goto data_pending; |
4e9a4686 DG |
3818 | } |
3819 | } | |
3820 | ||
d9f0c7c7 JR |
3821 | pthread_mutex_unlock(&stream->lock); |
3822 | } | |
3823 | ||
3824 | relayd = find_relayd_by_session_id(id); | |
3825 | if (relayd) { | |
3826 | unsigned int is_data_inflight = 0; | |
3827 | ||
3828 | /* Send init command for data pending. */ | |
3829 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3830 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3831 | relayd->relayd_session_id); | |
3832 | if (ret < 0) { | |
3833 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3834 | /* Communication error thus the relayd so no data pending. */ | |
3835 | goto data_not_pending; | |
3836 | } | |
3837 | ||
3838 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
3839 | ht->hash_fct(&id, lttng_ht_seed), | |
3840 | ht->match_fct, &id, | |
3841 | &iter.iter, stream, node_session_id.node) { | |
c8f59ee5 | 3842 | if (stream->metadata_flag) { |
ad7051c0 DG |
3843 | ret = relayd_quiescent_control(&relayd->control_sock, |
3844 | stream->relayd_stream_id); | |
c8f59ee5 | 3845 | } else { |
6d805429 | 3846 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3847 | stream->relayd_stream_id, |
3848 | stream->next_net_seq_num - 1); | |
c8f59ee5 | 3849 | } |
d9f0c7c7 JR |
3850 | |
3851 | if (ret == 1) { | |
3852 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3853 | goto data_pending; | |
3854 | } else if (ret < 0) { | |
9276e5c8 JR |
3855 | ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
3856 | lttng_consumer_cleanup_relayd(relayd); | |
3857 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
9276e5c8 JR |
3858 | goto data_not_pending; |
3859 | } | |
c8f59ee5 | 3860 | } |
f7079f67 | 3861 | |
d9f0c7c7 | 3862 | /* Send end command for data pending. */ |
f7079f67 DG |
3863 | ret = relayd_end_data_pending(&relayd->control_sock, |
3864 | relayd->relayd_session_id, &is_data_inflight); | |
3865 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3866 | if (ret < 0) { |
9276e5c8 JR |
3867 | ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
3868 | lttng_consumer_cleanup_relayd(relayd); | |
f7079f67 DG |
3869 | goto data_not_pending; |
3870 | } | |
bdd88757 DG |
3871 | if (is_data_inflight) { |
3872 | goto data_pending; | |
3873 | } | |
f7079f67 DG |
3874 | } |
3875 | ||
ca22feea | 3876 | /* |
f7079f67 DG |
3877 | * Finding _no_ node in the hash table and no inflight data means that the |
3878 | * stream(s) have been removed thus data is guaranteed to be available for | |
3879 | * analysis from the trace files. | |
ca22feea DG |
3880 | */ |
3881 | ||
f7079f67 | 3882 | data_not_pending: |
ca22feea | 3883 | /* Data is available to be read by a viewer. */ |
fa29bfbf | 3884 | pthread_mutex_unlock(&the_consumer_data.lock); |
c8f59ee5 | 3885 | rcu_read_unlock(); |
6d805429 | 3886 | return 0; |
ca22feea | 3887 | |
f7079f67 | 3888 | data_pending: |
ca22feea | 3889 | /* Data is still being extracted from buffers. */ |
fa29bfbf | 3890 | pthread_mutex_unlock(&the_consumer_data.lock); |
c8f59ee5 | 3891 | rcu_read_unlock(); |
6d805429 | 3892 | return 1; |
ca22feea | 3893 | } |
f50f23d9 DG |
3894 | |
3895 | /* | |
3896 | * Send a ret code status message to the sessiond daemon. | |
3897 | * | |
3898 | * Return the sendmsg() return value. | |
3899 | */ | |
3900 | int consumer_send_status_msg(int sock, int ret_code) | |
3901 | { | |
3902 | struct lttcomm_consumer_status_msg msg; | |
3903 | ||
53efb85a | 3904 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3905 | msg.ret_code = ret_code; |
3906 | ||
3907 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3908 | } | |
ffe60014 DG |
3909 | |
3910 | /* | |
3911 | * Send a channel status message to the sessiond daemon. | |
3912 | * | |
3913 | * Return the sendmsg() return value. | |
3914 | */ | |
3915 | int consumer_send_status_channel(int sock, | |
3916 | struct lttng_consumer_channel *channel) | |
3917 | { | |
3918 | struct lttcomm_consumer_status_channel msg; | |
3919 | ||
3920 | assert(sock >= 0); | |
3921 | ||
53efb85a | 3922 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3923 | if (!channel) { |
0c759fc9 | 3924 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3925 | } else { |
0c759fc9 | 3926 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3927 | msg.key = channel->key; |
3928 | msg.stream_count = channel->streams.count; | |
3929 | } | |
3930 | ||
3931 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3932 | } | |
5c786ded | 3933 | |
d07ceecd MD |
3934 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3935 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3936 | uint64_t max_sb_size) | |
5c786ded | 3937 | { |
d07ceecd | 3938 | unsigned long start_pos; |
5c786ded | 3939 | |
d07ceecd MD |
3940 | if (!nb_packets_per_stream) { |
3941 | return consumed_pos; /* Grab everything */ | |
3942 | } | |
3943 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3944 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3945 | if ((long) (start_pos - consumed_pos) < 0) { | |
3946 | return consumed_pos; /* Grab everything */ | |
3947 | } | |
3948 | return start_pos; | |
5c786ded | 3949 | } |
a1ae2ea5 | 3950 | |
c1dcb8bb JG |
3951 | /* Stream lock must be held by the caller. */ |
3952 | static int sample_stream_positions(struct lttng_consumer_stream *stream, | |
3953 | unsigned long *produced, unsigned long *consumed) | |
3954 | { | |
3955 | int ret; | |
3956 | ||
3957 | ASSERT_LOCKED(stream->lock); | |
3958 | ||
3959 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
3960 | if (ret < 0) { | |
3961 | ERR("Failed to sample snapshot positions"); | |
3962 | goto end; | |
3963 | } | |
3964 | ||
3965 | ret = lttng_consumer_get_produced_snapshot(stream, produced); | |
3966 | if (ret < 0) { | |
3967 | ERR("Failed to sample produced position"); | |
3968 | goto end; | |
3969 | } | |
3970 | ||
3971 | ret = lttng_consumer_get_consumed_snapshot(stream, consumed); | |
3972 | if (ret < 0) { | |
3973 | ERR("Failed to sample consumed position"); | |
3974 | goto end; | |
3975 | } | |
3976 | ||
3977 | end: | |
3978 | return ret; | |
3979 | } | |
3980 | ||
b99a8d42 JD |
3981 | /* |
3982 | * Sample the rotate position for all the streams of a channel. If a stream | |
3983 | * is already at the rotate position (produced == consumed), we flag it as | |
3984 | * ready for rotation. The rotation of ready streams occurs after we have | |
3985 | * replied to the session daemon that we have finished sampling the positions. | |
92b7a7f8 | 3986 | * Must be called with RCU read-side lock held to ensure existence of channel. |
b99a8d42 JD |
3987 | * |
3988 | * Returns 0 on success, < 0 on error | |
3989 | */ | |
92b7a7f8 | 3990 | int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, |
d2956687 | 3991 | uint64_t key, uint64_t relayd_id, uint32_t metadata, |
b99a8d42 JD |
3992 | struct lttng_consumer_local_data *ctx) |
3993 | { | |
3994 | int ret; | |
b99a8d42 JD |
3995 | struct lttng_consumer_stream *stream; |
3996 | struct lttng_ht_iter iter; | |
fa29bfbf | 3997 | struct lttng_ht *ht = the_consumer_data.stream_per_chan_id_ht; |
c35f9726 JG |
3998 | struct lttng_dynamic_array stream_rotation_positions; |
3999 | uint64_t next_chunk_id, stream_count = 0; | |
4000 | enum lttng_trace_chunk_status chunk_status; | |
4001 | const bool is_local_trace = relayd_id == -1ULL; | |
4002 | struct consumer_relayd_sock_pair *relayd = NULL; | |
4003 | bool rotating_to_new_chunk = true; | |
b32703d6 JG |
4004 | /* Array of `struct lttng_consumer_stream *` */ |
4005 | struct lttng_dynamic_pointer_array streams_packet_to_open; | |
4006 | size_t stream_idx; | |
b99a8d42 JD |
4007 | |
4008 | DBG("Consumer sample rotate position for channel %" PRIu64, key); | |
4009 | ||
c35f9726 JG |
4010 | lttng_dynamic_array_init(&stream_rotation_positions, |
4011 | sizeof(struct relayd_stream_rotation_position), NULL); | |
b32703d6 | 4012 | lttng_dynamic_pointer_array_init(&streams_packet_to_open, NULL); |
c35f9726 | 4013 | |
b99a8d42 JD |
4014 | rcu_read_lock(); |
4015 | ||
b99a8d42 | 4016 | pthread_mutex_lock(&channel->lock); |
c35f9726 JG |
4017 | assert(channel->trace_chunk); |
4018 | chunk_status = lttng_trace_chunk_get_id(channel->trace_chunk, | |
4019 | &next_chunk_id); | |
4020 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4021 | ret = -1; | |
4022 | goto end_unlock_channel; | |
4023 | } | |
b99a8d42 JD |
4024 | |
4025 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
4026 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
4027 | ht->match_fct, &channel->key, &iter.iter, | |
4028 | stream, node_channel_id.node) { | |
a40a503f | 4029 | unsigned long produced_pos = 0, consumed_pos = 0; |
b99a8d42 JD |
4030 | |
4031 | health_code_update(); | |
4032 | ||
4033 | /* | |
4034 | * Lock stream because we are about to change its state. | |
4035 | */ | |
4036 | pthread_mutex_lock(&stream->lock); | |
4037 | ||
c35f9726 JG |
4038 | if (stream->trace_chunk == stream->chan->trace_chunk) { |
4039 | rotating_to_new_chunk = false; | |
4040 | } | |
4041 | ||
a40a503f | 4042 | /* |
c1dcb8bb | 4043 | * Do not flush a packet when rotating from a NULL trace |
a9dde553 | 4044 | * chunk. The stream has no means to output data, and the prior |
c1dcb8bb JG |
4045 | * rotation which rotated to NULL performed that side-effect |
4046 | * already. No new data can be produced when a stream has no | |
4047 | * associated trace chunk (e.g. a stop followed by a rotate). | |
a40a503f | 4048 | */ |
a9dde553 | 4049 | if (stream->trace_chunk) { |
c1dcb8bb JG |
4050 | bool flush_active; |
4051 | ||
4052 | if (stream->metadata_flag) { | |
4053 | /* | |
4054 | * Don't produce an empty metadata packet, | |
4055 | * simply close the current one. | |
4056 | * | |
4057 | * Metadata is regenerated on every trace chunk | |
4058 | * switch; there is no concern that no data was | |
4059 | * produced. | |
4060 | */ | |
4061 | flush_active = true; | |
4062 | } else { | |
4063 | /* | |
4064 | * Only flush an empty packet if the "packet | |
4065 | * open" could not be performed on transition | |
4066 | * to a new trace chunk and no packets were | |
4067 | * consumed within the chunk's lifetime. | |
4068 | */ | |
4069 | if (stream->opened_packet_in_current_trace_chunk) { | |
4070 | flush_active = true; | |
4071 | } else { | |
4072 | /* | |
4073 | * Stream could have been full at the | |
4074 | * time of rotation, but then have had | |
4075 | * no activity at all. | |
4076 | * | |
4077 | * It is important to flush a packet | |
4078 | * to prevent 0-length files from being | |
4079 | * produced as most viewers choke on | |
4080 | * them. | |
4081 | * | |
4082 | * Unfortunately viewers will not be | |
4083 | * able to know that tracing was active | |
4084 | * for this stream during this trace | |
4085 | * chunk's lifetime. | |
4086 | */ | |
4087 | ret = sample_stream_positions(stream, &produced_pos, &consumed_pos); | |
4088 | if (ret) { | |
4089 | goto end_unlock_stream; | |
4090 | } | |
4091 | ||
4092 | /* | |
4093 | * Don't flush an empty packet if data | |
4094 | * was produced; it will be consumed | |
4095 | * before the rotation completes. | |
4096 | */ | |
4097 | flush_active = produced_pos != consumed_pos; | |
4098 | if (!flush_active) { | |
c1dcb8bb JG |
4099 | const char *trace_chunk_name; |
4100 | uint64_t trace_chunk_id; | |
4101 | ||
4102 | chunk_status = lttng_trace_chunk_get_name( | |
4103 | stream->trace_chunk, | |
4104 | &trace_chunk_name, | |
4105 | NULL); | |
4106 | if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NONE) { | |
4107 | trace_chunk_name = "none"; | |
4108 | } | |
4109 | ||
4110 | /* | |
4111 | * Consumer trace chunks are | |
4112 | * never anonymous. | |
4113 | */ | |
4114 | chunk_status = lttng_trace_chunk_get_id( | |
4115 | stream->trace_chunk, | |
4116 | &trace_chunk_id); | |
4117 | assert(chunk_status == | |
4118 | LTTNG_TRACE_CHUNK_STATUS_OK); | |
4119 | ||
4120 | DBG("Unable to open packet for stream during trace chunk's lifetime. " | |
4121 | "Flushing an empty packet to prevent an empty file from being created: " | |
4122 | "stream id = %" PRIu64 ", trace chunk name = `%s`, trace chunk id = %" PRIu64, | |
4123 | stream->key, trace_chunk_name, trace_chunk_id); | |
4124 | } | |
4125 | } | |
4126 | } | |
4127 | ||
a9dde553 | 4128 | /* |
c1dcb8bb JG |
4129 | * Close the current packet before sampling the |
4130 | * ring buffer positions. | |
a9dde553 | 4131 | */ |
c1dcb8bb | 4132 | ret = consumer_stream_flush_buffer(stream, flush_active); |
a9dde553 MD |
4133 | if (ret < 0) { |
4134 | ERR("Failed to flush stream %" PRIu64 " during channel rotation", | |
4135 | stream->key); | |
4136 | goto end_unlock_stream; | |
4137 | } | |
b99a8d42 JD |
4138 | } |
4139 | ||
a40a503f MD |
4140 | ret = lttng_consumer_take_snapshot(stream); |
4141 | if (ret < 0 && ret != -ENODATA && ret != -EAGAIN) { | |
4142 | ERR("Failed to sample snapshot position during channel rotation"); | |
b99a8d42 JD |
4143 | goto end_unlock_stream; |
4144 | } | |
a40a503f MD |
4145 | if (!ret) { |
4146 | ret = lttng_consumer_get_produced_snapshot(stream, | |
4147 | &produced_pos); | |
4148 | if (ret < 0) { | |
4149 | ERR("Failed to sample produced position during channel rotation"); | |
4150 | goto end_unlock_stream; | |
4151 | } | |
b99a8d42 | 4152 | |
a40a503f MD |
4153 | ret = lttng_consumer_get_consumed_snapshot(stream, |
4154 | &consumed_pos); | |
4155 | if (ret < 0) { | |
4156 | ERR("Failed to sample consumed position during channel rotation"); | |
4157 | goto end_unlock_stream; | |
4158 | } | |
4159 | } | |
4160 | /* | |
4161 | * Align produced position on the start-of-packet boundary of the first | |
4162 | * packet going into the next trace chunk. | |
4163 | */ | |
4164 | produced_pos = ALIGN_FLOOR(produced_pos, stream->max_sb_size); | |
4165 | if (consumed_pos == produced_pos) { | |
f8528c7a MD |
4166 | DBG("Set rotate ready for stream %" PRIu64 " produced = %lu consumed = %lu", |
4167 | stream->key, produced_pos, consumed_pos); | |
b99a8d42 | 4168 | stream->rotate_ready = true; |
f8528c7a MD |
4169 | } else { |
4170 | DBG("Different consumed and produced positions " | |
4171 | "for stream %" PRIu64 " produced = %lu consumed = %lu", | |
4172 | stream->key, produced_pos, consumed_pos); | |
b99a8d42 | 4173 | } |
633d0182 | 4174 | /* |
a40a503f MD |
4175 | * The rotation position is based on the packet_seq_num of the |
4176 | * packet following the last packet that was consumed for this | |
4177 | * stream, incremented by the offset between produced and | |
4178 | * consumed positions. This rotation position is a lower bound | |
4179 | * (inclusive) at which the next trace chunk starts. Since it | |
4180 | * is a lower bound, it is OK if the packet_seq_num does not | |
4181 | * correspond exactly to the same packet identified by the | |
4182 | * consumed_pos, which can happen in overwrite mode. | |
633d0182 | 4183 | */ |
a40a503f MD |
4184 | if (stream->sequence_number_unavailable) { |
4185 | /* | |
4186 | * Rotation should never be performed on a session which | |
4187 | * interacts with a pre-2.8 lttng-modules, which does | |
4188 | * not implement packet sequence number. | |
4189 | */ | |
4190 | ERR("Failure to rotate stream %" PRIu64 ": sequence number unavailable", | |
b99a8d42 | 4191 | stream->key); |
a40a503f | 4192 | ret = -1; |
b99a8d42 JD |
4193 | goto end_unlock_stream; |
4194 | } | |
a40a503f MD |
4195 | stream->rotate_position = stream->last_sequence_number + 1 + |
4196 | ((produced_pos - consumed_pos) / stream->max_sb_size); | |
f8528c7a MD |
4197 | DBG("Set rotation position for stream %" PRIu64 " at position %" PRIu64, |
4198 | stream->key, stream->rotate_position); | |
b99a8d42 | 4199 | |
c35f9726 | 4200 | if (!is_local_trace) { |
633d0182 JG |
4201 | /* |
4202 | * The relay daemon control protocol expects a rotation | |
4203 | * position as "the sequence number of the first packet | |
a40a503f | 4204 | * _after_ the current trace chunk". |
633d0182 | 4205 | */ |
c35f9726 JG |
4206 | const struct relayd_stream_rotation_position position = { |
4207 | .stream_id = stream->relayd_stream_id, | |
a40a503f | 4208 | .rotate_at_seq_num = stream->rotate_position, |
c35f9726 JG |
4209 | }; |
4210 | ||
4211 | ret = lttng_dynamic_array_add_element( | |
4212 | &stream_rotation_positions, | |
4213 | &position); | |
4214 | if (ret) { | |
4215 | ERR("Failed to allocate stream rotation position"); | |
4216 | goto end_unlock_stream; | |
4217 | } | |
4218 | stream_count++; | |
4219 | } | |
f96af312 JG |
4220 | |
4221 | stream->opened_packet_in_current_trace_chunk = false; | |
4222 | ||
4223 | if (rotating_to_new_chunk && !stream->metadata_flag) { | |
4224 | /* | |
4225 | * Attempt to flush an empty packet as close to the | |
4226 | * rotation point as possible. In the event where a | |
4227 | * stream remains inactive after the rotation point, | |
4228 | * this ensures that the new trace chunk has a | |
4229 | * beginning timestamp set at the begining of the | |
4230 | * trace chunk instead of only creating an empty | |
4231 | * packet when the trace chunk is stopped. | |
4232 | * | |
4233 | * This indicates to the viewers that the stream | |
4234 | * was being recorded, but more importantly it | |
4235 | * allows viewers to determine a useable trace | |
4236 | * intersection. | |
4237 | * | |
4238 | * This presents a problem in the case where the | |
4239 | * ring-buffer is completely full. | |
4240 | * | |
4241 | * Consider the following scenario: | |
4242 | * - The consumption of data is slow (slow network, | |
4243 | * for instance), | |
4244 | * - The ring buffer is full, | |
4245 | * - A rotation is initiated, | |
4246 | * - The flush below does nothing (no space left to | |
4247 | * open a new packet), | |
4248 | * - The other streams rotate very soon, and new | |
4249 | * data is produced in the new chunk, | |
4250 | * - This stream completes its rotation long after the | |
4251 | * rotation was initiated | |
4252 | * - The session is stopped before any event can be | |
4253 | * produced in this stream's buffers. | |
4254 | * | |
4255 | * The resulting trace chunk will have a single packet | |
4256 | * temporaly at the end of the trace chunk for this | |
4257 | * stream making the stream intersection more narrow | |
4258 | * than it should be. | |
4259 | * | |
4260 | * To work-around this, an empty flush is performed | |
4261 | * after the first consumption of a packet during a | |
4262 | * rotation if open_packet fails. The idea is that | |
4263 | * consuming a packet frees enough space to switch | |
4264 | * packets in this scenario and allows the tracer to | |
4265 | * "stamp" the beginning of the new trace chunk at the | |
4266 | * earliest possible point. | |
b32703d6 JG |
4267 | * |
4268 | * The packet open is performed after the channel | |
4269 | * rotation to ensure that no attempt to open a packet | |
4270 | * is performed in a stream that has no active trace | |
4271 | * chunk. | |
f96af312 | 4272 | */ |
b32703d6 JG |
4273 | ret = lttng_dynamic_pointer_array_add_pointer( |
4274 | &streams_packet_to_open, stream); | |
4275 | if (ret) { | |
4276 | PERROR("Failed to add a stream pointer to array of streams in which to open a packet"); | |
f96af312 JG |
4277 | ret = -1; |
4278 | goto end_unlock_stream; | |
f96af312 JG |
4279 | } |
4280 | } | |
4281 | ||
b99a8d42 JD |
4282 | pthread_mutex_unlock(&stream->lock); |
4283 | } | |
c35f9726 | 4284 | stream = NULL; |
b99a8d42 | 4285 | |
b32703d6 JG |
4286 | if (!is_local_trace) { |
4287 | relayd = consumer_find_relayd(relayd_id); | |
4288 | if (!relayd) { | |
4289 | ERR("Failed to find relayd %" PRIu64, relayd_id); | |
4290 | ret = -1; | |
4291 | goto end_unlock_channel; | |
4292 | } | |
c35f9726 | 4293 | |
b32703d6 JG |
4294 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
4295 | ret = relayd_rotate_streams(&relayd->control_sock, stream_count, | |
4296 | rotating_to_new_chunk ? &next_chunk_id : NULL, | |
4297 | (const struct relayd_stream_rotation_position *) | |
4298 | stream_rotation_positions.buffer | |
4299 | .data); | |
4300 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
4301 | if (ret < 0) { | |
4302 | ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64, | |
4303 | relayd->net_seq_idx); | |
4304 | lttng_consumer_cleanup_relayd(relayd); | |
4305 | goto end_unlock_channel; | |
4306 | } | |
c35f9726 JG |
4307 | } |
4308 | ||
b32703d6 JG |
4309 | for (stream_idx = 0; |
4310 | stream_idx < lttng_dynamic_pointer_array_get_count( | |
4311 | &streams_packet_to_open); | |
4312 | stream_idx++) { | |
4313 | enum consumer_stream_open_packet_status status; | |
4314 | ||
4315 | stream = lttng_dynamic_pointer_array_get_pointer( | |
4316 | &streams_packet_to_open, stream_idx); | |
4317 | ||
4318 | pthread_mutex_lock(&stream->lock); | |
4319 | status = consumer_stream_open_packet(stream); | |
4320 | pthread_mutex_unlock(&stream->lock); | |
4321 | switch (status) { | |
4322 | case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED: | |
4323 | DBG("Opened a packet after a rotation: stream id = %" PRIu64 | |
4324 | ", channel name = %s, session id = %" PRIu64, | |
4325 | stream->key, stream->chan->name, | |
4326 | stream->chan->session_id); | |
4327 | break; | |
4328 | case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE: | |
4329 | /* | |
4330 | * Can't open a packet as there is no space left | |
4331 | * in the buffer. A new packet will be opened | |
4332 | * once one has been consumed. | |
4333 | */ | |
4334 | DBG("No space left to open a packet after a rotation: stream id = %" PRIu64 | |
4335 | ", channel name = %s, session id = %" PRIu64, | |
4336 | stream->key, stream->chan->name, | |
4337 | stream->chan->session_id); | |
4338 | break; | |
4339 | case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR: | |
4340 | /* Logged by callee. */ | |
4341 | ret = -1; | |
7a86c13d | 4342 | goto end_unlock_channel; |
b32703d6 JG |
4343 | default: |
4344 | abort(); | |
4345 | } | |
c35f9726 JG |
4346 | } |
4347 | ||
b32703d6 | 4348 | pthread_mutex_unlock(&channel->lock); |
b99a8d42 JD |
4349 | ret = 0; |
4350 | goto end; | |
4351 | ||
4352 | end_unlock_stream: | |
4353 | pthread_mutex_unlock(&stream->lock); | |
c35f9726 | 4354 | end_unlock_channel: |
b99a8d42 JD |
4355 | pthread_mutex_unlock(&channel->lock); |
4356 | end: | |
4357 | rcu_read_unlock(); | |
c35f9726 | 4358 | lttng_dynamic_array_reset(&stream_rotation_positions); |
b32703d6 | 4359 | lttng_dynamic_pointer_array_reset(&streams_packet_to_open); |
b99a8d42 JD |
4360 | return ret; |
4361 | } | |
4362 | ||
5f3aff8b MD |
4363 | static |
4364 | int consumer_clear_buffer(struct lttng_consumer_stream *stream) | |
4365 | { | |
4366 | int ret = 0; | |
4367 | unsigned long consumed_pos_before, consumed_pos_after; | |
4368 | ||
4369 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
4370 | if (ret < 0) { | |
4371 | ERR("Taking snapshot positions"); | |
4372 | goto end; | |
4373 | } | |
4374 | ||
4375 | ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_before); | |
4376 | if (ret < 0) { | |
4377 | ERR("Consumed snapshot position"); | |
4378 | goto end; | |
4379 | } | |
4380 | ||
fa29bfbf | 4381 | switch (the_consumer_data.type) { |
5f3aff8b MD |
4382 | case LTTNG_CONSUMER_KERNEL: |
4383 | ret = kernctl_buffer_clear(stream->wait_fd); | |
4384 | if (ret < 0) { | |
96393977 | 4385 | ERR("Failed to clear kernel stream (ret = %d)", ret); |
5f3aff8b MD |
4386 | goto end; |
4387 | } | |
4388 | break; | |
4389 | case LTTNG_CONSUMER32_UST: | |
4390 | case LTTNG_CONSUMER64_UST: | |
d85707b0 MD |
4391 | ret = lttng_ustconsumer_clear_buffer(stream); |
4392 | if (ret < 0) { | |
4393 | ERR("Failed to clear ust stream (ret = %d)", ret); | |
4394 | goto end; | |
4395 | } | |
5f3aff8b MD |
4396 | break; |
4397 | default: | |
4398 | ERR("Unknown consumer_data type"); | |
4399 | abort(); | |
4400 | } | |
4401 | ||
4402 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
4403 | if (ret < 0) { | |
4404 | ERR("Taking snapshot positions"); | |
4405 | goto end; | |
4406 | } | |
4407 | ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_after); | |
4408 | if (ret < 0) { | |
4409 | ERR("Consumed snapshot position"); | |
4410 | goto end; | |
4411 | } | |
4412 | DBG("clear: before: %lu after: %lu", consumed_pos_before, consumed_pos_after); | |
4413 | end: | |
4414 | return ret; | |
4415 | } | |
4416 | ||
4417 | static | |
4418 | int consumer_clear_stream(struct lttng_consumer_stream *stream) | |
4419 | { | |
4420 | int ret; | |
4421 | ||
503fefca | 4422 | ret = consumer_stream_flush_buffer(stream, 1); |
5f3aff8b MD |
4423 | if (ret < 0) { |
4424 | ERR("Failed to flush stream %" PRIu64 " during channel clear", | |
4425 | stream->key); | |
4426 | ret = LTTCOMM_CONSUMERD_FATAL; | |
4427 | goto error; | |
4428 | } | |
4429 | ||
4430 | ret = consumer_clear_buffer(stream); | |
4431 | if (ret < 0) { | |
4432 | ERR("Failed to clear stream %" PRIu64 " during channel clear", | |
4433 | stream->key); | |
4434 | ret = LTTCOMM_CONSUMERD_FATAL; | |
4435 | goto error; | |
4436 | } | |
4437 | ||
4438 | ret = LTTCOMM_CONSUMERD_SUCCESS; | |
4439 | error: | |
4440 | return ret; | |
4441 | } | |
4442 | ||
4443 | static | |
4444 | int consumer_clear_unmonitored_channel(struct lttng_consumer_channel *channel) | |
4445 | { | |
4446 | int ret; | |
4447 | struct lttng_consumer_stream *stream; | |
4448 | ||
4449 | rcu_read_lock(); | |
4450 | pthread_mutex_lock(&channel->lock); | |
4451 | cds_list_for_each_entry(stream, &channel->streams.head, send_node) { | |
4452 | health_code_update(); | |
4453 | pthread_mutex_lock(&stream->lock); | |
4454 | ret = consumer_clear_stream(stream); | |
4455 | if (ret) { | |
4456 | goto error_unlock; | |
4457 | } | |
4458 | pthread_mutex_unlock(&stream->lock); | |
4459 | } | |
4460 | pthread_mutex_unlock(&channel->lock); | |
4461 | rcu_read_unlock(); | |
4462 | return 0; | |
4463 | ||
4464 | error_unlock: | |
4465 | pthread_mutex_unlock(&stream->lock); | |
4466 | pthread_mutex_unlock(&channel->lock); | |
4467 | rcu_read_unlock(); | |
5f3aff8b MD |
4468 | return ret; |
4469 | } | |
4470 | ||
02d02e31 JD |
4471 | /* |
4472 | * Check if a stream is ready to be rotated after extracting it. | |
4473 | * | |
4474 | * Return 1 if it is ready for rotation, 0 if it is not, a negative value on | |
4475 | * error. Stream lock must be held. | |
4476 | */ | |
4477 | int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream) | |
4478 | { | |
f8528c7a MD |
4479 | DBG("Check is rotate ready for stream %" PRIu64 |
4480 | " ready %u rotate_position %" PRIu64 | |
4481 | " last_sequence_number %" PRIu64, | |
4482 | stream->key, stream->rotate_ready, | |
4483 | stream->rotate_position, stream->last_sequence_number); | |
02d02e31 | 4484 | if (stream->rotate_ready) { |
a40a503f | 4485 | return 1; |
02d02e31 JD |
4486 | } |
4487 | ||
4488 | /* | |
a40a503f MD |
4489 | * If packet seq num is unavailable, it means we are interacting |
4490 | * with a pre-2.8 lttng-modules which does not implement the | |
4491 | * sequence number. Rotation should never be used by sessiond in this | |
4492 | * scenario. | |
02d02e31 | 4493 | */ |
a40a503f MD |
4494 | if (stream->sequence_number_unavailable) { |
4495 | ERR("Internal error: rotation used on stream %" PRIu64 | |
4496 | " with unavailable sequence number", | |
4497 | stream->key); | |
4498 | return -1; | |
02d02e31 JD |
4499 | } |
4500 | ||
a40a503f MD |
4501 | if (stream->rotate_position == -1ULL || |
4502 | stream->last_sequence_number == -1ULL) { | |
4503 | return 0; | |
02d02e31 JD |
4504 | } |
4505 | ||
a40a503f MD |
4506 | /* |
4507 | * Rotate position not reached yet. The stream rotate position is | |
4508 | * the position of the next packet belonging to the next trace chunk, | |
4509 | * but consumerd considers rotation ready when reaching the last | |
4510 | * packet of the current chunk, hence the "rotate_position - 1". | |
4511 | */ | |
f8528c7a MD |
4512 | |
4513 | DBG("Check is rotate ready for stream %" PRIu64 | |
4514 | " last_sequence_number %" PRIu64 | |
4515 | " rotate_position %" PRIu64, | |
4516 | stream->key, stream->last_sequence_number, | |
4517 | stream->rotate_position); | |
a40a503f MD |
4518 | if (stream->last_sequence_number >= stream->rotate_position - 1) { |
4519 | return 1; | |
02d02e31 | 4520 | } |
02d02e31 | 4521 | |
a40a503f | 4522 | return 0; |
02d02e31 JD |
4523 | } |
4524 | ||
d73bf3d7 JD |
4525 | /* |
4526 | * Reset the state for a stream after a rotation occurred. | |
4527 | */ | |
4528 | void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream) | |
4529 | { | |
f8528c7a MD |
4530 | DBG("lttng_consumer_reset_stream_rotate_state for stream %" PRIu64, |
4531 | stream->key); | |
a40a503f | 4532 | stream->rotate_position = -1ULL; |
d73bf3d7 JD |
4533 | stream->rotate_ready = false; |
4534 | } | |
4535 | ||
4536 | /* | |
4537 | * Perform the rotation a local stream file. | |
4538 | */ | |
d2956687 | 4539 | static |
d73bf3d7 JD |
4540 | int rotate_local_stream(struct lttng_consumer_local_data *ctx, |
4541 | struct lttng_consumer_stream *stream) | |
4542 | { | |
d2956687 | 4543 | int ret = 0; |
d73bf3d7 | 4544 | |
d2956687 | 4545 | DBG("Rotate local stream: stream key %" PRIu64 ", channel key %" PRIu64, |
d73bf3d7 | 4546 | stream->key, |
d2956687 | 4547 | stream->chan->key); |
d73bf3d7 | 4548 | stream->tracefile_size_current = 0; |
d2956687 | 4549 | stream->tracefile_count_current = 0; |
d73bf3d7 | 4550 | |
d2956687 JG |
4551 | if (stream->out_fd >= 0) { |
4552 | ret = close(stream->out_fd); | |
4553 | if (ret) { | |
4554 | PERROR("Failed to close stream out_fd of channel \"%s\"", | |
4555 | stream->chan->name); | |
4556 | } | |
4557 | stream->out_fd = -1; | |
4558 | } | |
d73bf3d7 | 4559 | |
d2956687 | 4560 | if (stream->index_file) { |
d73bf3d7 | 4561 | lttng_index_file_put(stream->index_file); |
d2956687 | 4562 | stream->index_file = NULL; |
d73bf3d7 JD |
4563 | } |
4564 | ||
d2956687 JG |
4565 | if (!stream->trace_chunk) { |
4566 | goto end; | |
4567 | } | |
d73bf3d7 | 4568 | |
d2956687 | 4569 | ret = consumer_stream_create_output_files(stream, true); |
d73bf3d7 JD |
4570 | end: |
4571 | return ret; | |
d73bf3d7 JD |
4572 | } |
4573 | ||
d73bf3d7 JD |
4574 | /* |
4575 | * Performs the stream rotation for the rotate session feature if needed. | |
d2956687 | 4576 | * It must be called with the channel and stream locks held. |
d73bf3d7 JD |
4577 | * |
4578 | * Return 0 on success, a negative number of error. | |
4579 | */ | |
4580 | int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx, | |
d2956687 | 4581 | struct lttng_consumer_stream *stream) |
d73bf3d7 JD |
4582 | { |
4583 | int ret; | |
4584 | ||
4585 | DBG("Consumer rotate stream %" PRIu64, stream->key); | |
4586 | ||
d2956687 JG |
4587 | /* |
4588 | * Update the stream's 'current' chunk to the session's (channel) | |
4589 | * now-current chunk. | |
4590 | */ | |
4591 | lttng_trace_chunk_put(stream->trace_chunk); | |
4592 | if (stream->chan->trace_chunk == stream->trace_chunk) { | |
4593 | /* | |
4594 | * A channel can be rotated and not have a "next" chunk | |
4595 | * to transition to. In that case, the channel's "current chunk" | |
4596 | * has not been closed yet, but it has not been updated to | |
4597 | * a "next" trace chunk either. Hence, the stream, like its | |
4598 | * parent channel, becomes part of no chunk and can't output | |
4599 | * anything until a new trace chunk is created. | |
4600 | */ | |
4601 | stream->trace_chunk = NULL; | |
4602 | } else if (stream->chan->trace_chunk && | |
4603 | !lttng_trace_chunk_get(stream->chan->trace_chunk)) { | |
4604 | ERR("Failed to acquire a reference to channel's trace chunk during stream rotation"); | |
4605 | ret = -1; | |
4606 | goto error; | |
4607 | } else { | |
4608 | /* | |
4609 | * Update the stream's trace chunk to its parent channel's | |
4610 | * current trace chunk. | |
4611 | */ | |
4612 | stream->trace_chunk = stream->chan->trace_chunk; | |
4613 | } | |
4614 | ||
c35f9726 | 4615 | if (stream->net_seq_idx == (uint64_t) -1ULL) { |
d73bf3d7 | 4616 | ret = rotate_local_stream(ctx, stream); |
c35f9726 JG |
4617 | if (ret < 0) { |
4618 | ERR("Failed to rotate stream, ret = %i", ret); | |
4619 | goto error; | |
4620 | } | |
d73bf3d7 JD |
4621 | } |
4622 | ||
d2956687 JG |
4623 | if (stream->metadata_flag && stream->trace_chunk) { |
4624 | /* | |
4625 | * If the stream has transitioned to a new trace | |
4626 | * chunk, the metadata should be re-dumped to the | |
4627 | * newest chunk. | |
4628 | * | |
4629 | * However, it is possible for a stream to transition to | |
4630 | * a "no-chunk" state. This can happen if a rotation | |
4631 | * occurs on an inactive session. In such cases, the metadata | |
4632 | * regeneration will happen when the next trace chunk is | |
4633 | * created. | |
4634 | */ | |
4635 | ret = consumer_metadata_stream_dump(stream); | |
4636 | if (ret) { | |
4637 | goto error; | |
d73bf3d7 JD |
4638 | } |
4639 | } | |
4640 | lttng_consumer_reset_stream_rotate_state(stream); | |
4641 | ||
4642 | ret = 0; | |
4643 | ||
4644 | error: | |
4645 | return ret; | |
4646 | } | |
4647 | ||
b99a8d42 JD |
4648 | /* |
4649 | * Rotate all the ready streams now. | |
4650 | * | |
4651 | * This is especially important for low throughput streams that have already | |
4652 | * been consumed, we cannot wait for their next packet to perform the | |
4653 | * rotation. | |
92b7a7f8 MD |
4654 | * Need to be called with RCU read-side lock held to ensure existence of |
4655 | * channel. | |
b99a8d42 JD |
4656 | * |
4657 | * Returns 0 on success, < 0 on error | |
4658 | */ | |
92b7a7f8 MD |
4659 | int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel *channel, |
4660 | uint64_t key, struct lttng_consumer_local_data *ctx) | |
b99a8d42 JD |
4661 | { |
4662 | int ret; | |
b99a8d42 JD |
4663 | struct lttng_consumer_stream *stream; |
4664 | struct lttng_ht_iter iter; | |
fa29bfbf | 4665 | struct lttng_ht *ht = the_consumer_data.stream_per_chan_id_ht; |
b99a8d42 JD |
4666 | |
4667 | rcu_read_lock(); | |
4668 | ||
4669 | DBG("Consumer rotate ready streams in channel %" PRIu64, key); | |
4670 | ||
b99a8d42 JD |
4671 | cds_lfht_for_each_entry_duplicate(ht->ht, |
4672 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
4673 | ht->match_fct, &channel->key, &iter.iter, | |
4674 | stream, node_channel_id.node) { | |
4675 | health_code_update(); | |
4676 | ||
d2956687 | 4677 | pthread_mutex_lock(&stream->chan->lock); |
b99a8d42 JD |
4678 | pthread_mutex_lock(&stream->lock); |
4679 | ||
4680 | if (!stream->rotate_ready) { | |
4681 | pthread_mutex_unlock(&stream->lock); | |
d2956687 | 4682 | pthread_mutex_unlock(&stream->chan->lock); |
b99a8d42 JD |
4683 | continue; |
4684 | } | |
4685 | DBG("Consumer rotate ready stream %" PRIu64, stream->key); | |
4686 | ||
d2956687 | 4687 | ret = lttng_consumer_rotate_stream(ctx, stream); |
b99a8d42 | 4688 | pthread_mutex_unlock(&stream->lock); |
d2956687 | 4689 | pthread_mutex_unlock(&stream->chan->lock); |
b99a8d42 JD |
4690 | if (ret) { |
4691 | goto end; | |
4692 | } | |
4693 | } | |
4694 | ||
4695 | ret = 0; | |
4696 | ||
4697 | end: | |
4698 | rcu_read_unlock(); | |
4699 | return ret; | |
4700 | } | |
4701 | ||
d2956687 JG |
4702 | enum lttcomm_return_code lttng_consumer_init_command( |
4703 | struct lttng_consumer_local_data *ctx, | |
4704 | const lttng_uuid sessiond_uuid) | |
00fb02ac | 4705 | { |
d2956687 | 4706 | enum lttcomm_return_code ret; |
c70636a7 | 4707 | char uuid_str[LTTNG_UUID_STR_LEN]; |
00fb02ac | 4708 | |
d2956687 JG |
4709 | if (ctx->sessiond_uuid.is_set) { |
4710 | ret = LTTCOMM_CONSUMERD_ALREADY_SET; | |
00fb02ac JD |
4711 | goto end; |
4712 | } | |
4713 | ||
d2956687 JG |
4714 | ctx->sessiond_uuid.is_set = true; |
4715 | memcpy(ctx->sessiond_uuid.value, sessiond_uuid, sizeof(lttng_uuid)); | |
4716 | ret = LTTCOMM_CONSUMERD_SUCCESS; | |
4717 | lttng_uuid_to_str(sessiond_uuid, uuid_str); | |
4718 | DBG("Received session daemon UUID: %s", uuid_str); | |
00fb02ac JD |
4719 | end: |
4720 | return ret; | |
4721 | } | |
4722 | ||
d2956687 JG |
4723 | enum lttcomm_return_code lttng_consumer_create_trace_chunk( |
4724 | const uint64_t *relayd_id, uint64_t session_id, | |
4725 | uint64_t chunk_id, | |
4726 | time_t chunk_creation_timestamp, | |
4727 | const char *chunk_override_name, | |
4728 | const struct lttng_credentials *credentials, | |
4729 | struct lttng_directory_handle *chunk_directory_handle) | |
00fb02ac JD |
4730 | { |
4731 | int ret; | |
d2956687 | 4732 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
7ea24db3 | 4733 | struct lttng_trace_chunk *created_chunk = NULL, *published_chunk = NULL; |
d2956687 JG |
4734 | enum lttng_trace_chunk_status chunk_status; |
4735 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; | |
4736 | char creation_timestamp_buffer[ISO8601_STR_LEN]; | |
4737 | const char *relayd_id_str = "(none)"; | |
4738 | const char *creation_timestamp_str; | |
4739 | struct lttng_ht_iter iter; | |
4740 | struct lttng_consumer_channel *channel; | |
92816cc3 | 4741 | |
d2956687 JG |
4742 | if (relayd_id) { |
4743 | /* Only used for logging purposes. */ | |
4744 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), | |
4745 | "%" PRIu64, *relayd_id); | |
4746 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { | |
4747 | relayd_id_str = relayd_id_buffer; | |
4748 | } else { | |
4749 | relayd_id_str = "(formatting error)"; | |
4750 | } | |
d01ef216 | 4751 | } |
d2956687 | 4752 | |
d01ef216 | 4753 | /* Local protocol error. */ |
d2956687 JG |
4754 | assert(chunk_creation_timestamp); |
4755 | ret = time_to_iso8601_str(chunk_creation_timestamp, | |
4756 | creation_timestamp_buffer, | |
4757 | sizeof(creation_timestamp_buffer)); | |
4758 | creation_timestamp_str = !ret ? creation_timestamp_buffer : | |
4759 | "(formatting error)"; | |
4760 | ||
4761 | DBG("Consumer create trace chunk command: relay_id = %s" | |
4762 | ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 | |
4763 | ", chunk_override_name = %s" | |
4764 | ", chunk_creation_timestamp = %s", | |
4765 | relayd_id_str, session_id, chunk_id, | |
4766 | chunk_override_name ? : "(none)", | |
4767 | creation_timestamp_str); | |
92816cc3 JG |
4768 | |
4769 | /* | |
d2956687 JG |
4770 | * The trace chunk registry, as used by the consumer daemon, implicitly |
4771 | * owns the trace chunks. This is only needed in the consumer since | |
4772 | * the consumer has no notion of a session beyond session IDs being | |
4773 | * used to identify other objects. | |
4774 | * | |
4775 | * The lttng_trace_chunk_registry_publish() call below provides a | |
4776 | * reference which is not released; it implicitly becomes the session | |
4777 | * daemon's reference to the chunk in the consumer daemon. | |
4778 | * | |
4779 | * The lifetime of trace chunks in the consumer daemon is managed by | |
4780 | * the session daemon through the LTTNG_CONSUMER_CREATE_TRACE_CHUNK | |
4781 | * and LTTNG_CONSUMER_DESTROY_TRACE_CHUNK commands. | |
92816cc3 | 4782 | */ |
d2956687 | 4783 | created_chunk = lttng_trace_chunk_create(chunk_id, |
a7ceb342 | 4784 | chunk_creation_timestamp, NULL); |
d2956687 JG |
4785 | if (!created_chunk) { |
4786 | ERR("Failed to create trace chunk"); | |
4787 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4788 | goto error; |
d2956687 | 4789 | } |
92816cc3 | 4790 | |
d2956687 JG |
4791 | if (chunk_override_name) { |
4792 | chunk_status = lttng_trace_chunk_override_name(created_chunk, | |
4793 | chunk_override_name); | |
4794 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4795 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4796 | goto error; |
92816cc3 JG |
4797 | } |
4798 | } | |
4799 | ||
d2956687 JG |
4800 | if (chunk_directory_handle) { |
4801 | chunk_status = lttng_trace_chunk_set_credentials(created_chunk, | |
4802 | credentials); | |
4803 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4804 | ERR("Failed to set trace chunk credentials"); | |
4805 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4806 | goto error; |
d2956687 JG |
4807 | } |
4808 | /* | |
4809 | * The consumer daemon has no ownership of the chunk output | |
4810 | * directory. | |
4811 | */ | |
4812 | chunk_status = lttng_trace_chunk_set_as_user(created_chunk, | |
4813 | chunk_directory_handle); | |
cbf53d23 | 4814 | chunk_directory_handle = NULL; |
d2956687 JG |
4815 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
4816 | ERR("Failed to set trace chunk's directory handle"); | |
4817 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4818 | goto error; |
92816cc3 JG |
4819 | } |
4820 | } | |
4821 | ||
d2956687 | 4822 | published_chunk = lttng_trace_chunk_registry_publish_chunk( |
fa29bfbf | 4823 | the_consumer_data.chunk_registry, session_id, |
d2956687 JG |
4824 | created_chunk); |
4825 | lttng_trace_chunk_put(created_chunk); | |
4826 | created_chunk = NULL; | |
4827 | if (!published_chunk) { | |
4828 | ERR("Failed to publish trace chunk"); | |
4829 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4830 | goto error; |
d88744a4 JD |
4831 | } |
4832 | ||
d2956687 | 4833 | rcu_read_lock(); |
fa29bfbf SM |
4834 | cds_lfht_for_each_entry_duplicate( |
4835 | the_consumer_data.channels_by_session_id_ht->ht, | |
4836 | the_consumer_data.channels_by_session_id_ht->hash_fct( | |
d2956687 | 4837 | &session_id, lttng_ht_seed), |
fa29bfbf | 4838 | the_consumer_data.channels_by_session_id_ht->match_fct, |
d2956687 JG |
4839 | &session_id, &iter.iter, channel, |
4840 | channels_by_session_id_ht_node.node) { | |
4841 | ret = lttng_consumer_channel_set_trace_chunk(channel, | |
4842 | published_chunk); | |
4843 | if (ret) { | |
4844 | /* | |
4845 | * Roll-back the creation of this chunk. | |
4846 | * | |
4847 | * This is important since the session daemon will | |
4848 | * assume that the creation of this chunk failed and | |
4849 | * will never ask for it to be closed, resulting | |
4850 | * in a leak and an inconsistent state for some | |
4851 | * channels. | |
4852 | */ | |
4853 | enum lttcomm_return_code close_ret; | |
ecd1a12f | 4854 | char path[LTTNG_PATH_MAX]; |
d2956687 JG |
4855 | |
4856 | DBG("Failed to set new trace chunk on existing channels, rolling back"); | |
4857 | close_ret = lttng_consumer_close_trace_chunk(relayd_id, | |
4858 | session_id, chunk_id, | |
ecd1a12f MD |
4859 | chunk_creation_timestamp, NULL, |
4860 | path); | |
d2956687 JG |
4861 | if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { |
4862 | ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, | |
4863 | session_id, chunk_id); | |
4864 | } | |
a1ae2ea5 | 4865 | |
d2956687 JG |
4866 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
4867 | break; | |
4868 | } | |
a1ae2ea5 JD |
4869 | } |
4870 | ||
e5add6d0 JG |
4871 | if (relayd_id) { |
4872 | struct consumer_relayd_sock_pair *relayd; | |
4873 | ||
4874 | relayd = consumer_find_relayd(*relayd_id); | |
4875 | if (relayd) { | |
4876 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
4877 | ret = relayd_create_trace_chunk( | |
4878 | &relayd->control_sock, published_chunk); | |
4879 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
4880 | } else { | |
4881 | ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, *relayd_id); | |
4882 | } | |
4883 | ||
4884 | if (!relayd || ret) { | |
4885 | enum lttcomm_return_code close_ret; | |
ecd1a12f | 4886 | char path[LTTNG_PATH_MAX]; |
e5add6d0 JG |
4887 | |
4888 | close_ret = lttng_consumer_close_trace_chunk(relayd_id, | |
4889 | session_id, | |
4890 | chunk_id, | |
bbc4768c | 4891 | chunk_creation_timestamp, |
ecd1a12f | 4892 | NULL, path); |
e5add6d0 JG |
4893 | if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { |
4894 | ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, | |
4895 | session_id, | |
4896 | chunk_id); | |
4897 | } | |
4898 | ||
4899 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4900 | goto error_unlock; |
e5add6d0 JG |
4901 | } |
4902 | } | |
7ea24db3 | 4903 | error_unlock: |
e5add6d0 | 4904 | rcu_read_unlock(); |
7ea24db3 | 4905 | error: |
d2956687 JG |
4906 | /* Release the reference returned by the "publish" operation. */ |
4907 | lttng_trace_chunk_put(published_chunk); | |
9bb5f1f8 | 4908 | lttng_trace_chunk_put(created_chunk); |
d2956687 | 4909 | return ret_code; |
a1ae2ea5 JD |
4910 | } |
4911 | ||
d2956687 JG |
4912 | enum lttcomm_return_code lttng_consumer_close_trace_chunk( |
4913 | const uint64_t *relayd_id, uint64_t session_id, | |
bbc4768c | 4914 | uint64_t chunk_id, time_t chunk_close_timestamp, |
ecd1a12f MD |
4915 | const enum lttng_trace_chunk_command_type *close_command, |
4916 | char *path) | |
a1ae2ea5 | 4917 | { |
d2956687 JG |
4918 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
4919 | struct lttng_trace_chunk *chunk; | |
4920 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; | |
4921 | const char *relayd_id_str = "(none)"; | |
bbc4768c | 4922 | const char *close_command_name = "none"; |
d2956687 JG |
4923 | struct lttng_ht_iter iter; |
4924 | struct lttng_consumer_channel *channel; | |
4925 | enum lttng_trace_chunk_status chunk_status; | |
a1ae2ea5 | 4926 | |
d2956687 JG |
4927 | if (relayd_id) { |
4928 | int ret; | |
4929 | ||
4930 | /* Only used for logging purposes. */ | |
4931 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), | |
4932 | "%" PRIu64, *relayd_id); | |
4933 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { | |
4934 | relayd_id_str = relayd_id_buffer; | |
4935 | } else { | |
4936 | relayd_id_str = "(formatting error)"; | |
4937 | } | |
bbc4768c JG |
4938 | } |
4939 | if (close_command) { | |
4940 | close_command_name = lttng_trace_chunk_command_type_get_name( | |
4941 | *close_command); | |
4942 | } | |
d2956687 JG |
4943 | |
4944 | DBG("Consumer close trace chunk command: relayd_id = %s" | |
bbc4768c JG |
4945 | ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 |
4946 | ", close command = %s", | |
4947 | relayd_id_str, session_id, chunk_id, | |
4948 | close_command_name); | |
4949 | ||
d2956687 | 4950 | chunk = lttng_trace_chunk_registry_find_chunk( |
fa29bfbf | 4951 | the_consumer_data.chunk_registry, session_id, chunk_id); |
bbc4768c | 4952 | if (!chunk) { |
d2956687 JG |
4953 | ERR("Failed to find chunk: session_id = %" PRIu64 |
4954 | ", chunk_id = %" PRIu64, | |
4955 | session_id, chunk_id); | |
4956 | ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; | |
a1ae2ea5 JD |
4957 | goto end; |
4958 | } | |
4959 | ||
d2956687 JG |
4960 | chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, |
4961 | chunk_close_timestamp); | |
4962 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4963 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
4964 | goto end; | |
45f1d9a1 | 4965 | } |
bbc4768c JG |
4966 | |
4967 | if (close_command) { | |
4968 | chunk_status = lttng_trace_chunk_set_close_command( | |
4969 | chunk, *close_command); | |
4970 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4971 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
4972 | goto end; | |
4973 | } | |
4974 | } | |
a1ae2ea5 | 4975 | |
d2956687 JG |
4976 | /* |
4977 | * chunk is now invalid to access as we no longer hold a reference to | |
4978 | * it; it is only kept around to compare it (by address) to the | |
4979 | * current chunk found in the session's channels. | |
4980 | */ | |
4981 | rcu_read_lock(); | |
fa29bfbf | 4982 | cds_lfht_for_each_entry(the_consumer_data.channel_ht->ht, &iter.iter, |
d2956687 JG |
4983 | channel, node.node) { |
4984 | int ret; | |
a1ae2ea5 | 4985 | |
d2956687 JG |
4986 | /* |
4987 | * Only change the channel's chunk to NULL if it still | |
4988 | * references the chunk being closed. The channel may | |
4989 | * reference a newer channel in the case of a session | |
4990 | * rotation. When a session rotation occurs, the "next" | |
4991 | * chunk is created before the "current" chunk is closed. | |
4992 | */ | |
4993 | if (channel->trace_chunk != chunk) { | |
4994 | continue; | |
4995 | } | |
4996 | ret = lttng_consumer_channel_set_trace_chunk(channel, NULL); | |
4997 | if (ret) { | |
4998 | /* | |
4999 | * Attempt to close the chunk on as many channels as | |
5000 | * possible. | |
5001 | */ | |
5002 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
5003 | } | |
a1ae2ea5 | 5004 | } |
bbc4768c JG |
5005 | |
5006 | if (relayd_id) { | |
5007 | int ret; | |
5008 | struct consumer_relayd_sock_pair *relayd; | |
5009 | ||
5010 | relayd = consumer_find_relayd(*relayd_id); | |
5011 | if (relayd) { | |
5012 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
5013 | ret = relayd_close_trace_chunk( | |
ecd1a12f MD |
5014 | &relayd->control_sock, chunk, |
5015 | path); | |
bbc4768c JG |
5016 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
5017 | } else { | |
5018 | ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, | |
5019 | *relayd_id); | |
5020 | } | |
5021 | ||
5022 | if (!relayd || ret) { | |
5023 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
5024 | goto error_unlock; | |
5025 | } | |
5026 | } | |
5027 | error_unlock: | |
d2956687 JG |
5028 | rcu_read_unlock(); |
5029 | end: | |
bbc4768c JG |
5030 | /* |
5031 | * Release the reference returned by the "find" operation and | |
5032 | * the session daemon's implicit reference to the chunk. | |
5033 | */ | |
5034 | lttng_trace_chunk_put(chunk); | |
5035 | lttng_trace_chunk_put(chunk); | |
5036 | ||
d2956687 | 5037 | return ret_code; |
a1ae2ea5 | 5038 | } |
3654ed19 | 5039 | |
d2956687 JG |
5040 | enum lttcomm_return_code lttng_consumer_trace_chunk_exists( |
5041 | const uint64_t *relayd_id, uint64_t session_id, | |
5042 | uint64_t chunk_id) | |
3654ed19 | 5043 | { |
c35f9726 | 5044 | int ret; |
d2956687 | 5045 | enum lttcomm_return_code ret_code; |
d2956687 JG |
5046 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; |
5047 | const char *relayd_id_str = "(none)"; | |
c35f9726 JG |
5048 | const bool is_local_trace = !relayd_id; |
5049 | struct consumer_relayd_sock_pair *relayd = NULL; | |
6b584c2e | 5050 | bool chunk_exists_local, chunk_exists_remote; |
d2956687 JG |
5051 | |
5052 | if (relayd_id) { | |
d2956687 JG |
5053 | /* Only used for logging purposes. */ |
5054 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), | |
5055 | "%" PRIu64, *relayd_id); | |
5056 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { | |
5057 | relayd_id_str = relayd_id_buffer; | |
5058 | } else { | |
5059 | relayd_id_str = "(formatting error)"; | |
5060 | } | |
d01ef216 | 5061 | } |
d2956687 JG |
5062 | |
5063 | DBG("Consumer trace chunk exists command: relayd_id = %s" | |
d2956687 | 5064 | ", chunk_id = %" PRIu64, relayd_id_str, |
c35f9726 | 5065 | chunk_id); |
6b584c2e | 5066 | ret = lttng_trace_chunk_registry_chunk_exists( |
fa29bfbf SM |
5067 | the_consumer_data.chunk_registry, session_id, chunk_id, |
5068 | &chunk_exists_local); | |
6b584c2e JG |
5069 | if (ret) { |
5070 | /* Internal error. */ | |
5071 | ERR("Failed to query the existence of a trace chunk"); | |
5072 | ret_code = LTTCOMM_CONSUMERD_FATAL; | |
13e3b280 | 5073 | goto end; |
6b584c2e JG |
5074 | } |
5075 | DBG("Trace chunk %s locally", | |
5076 | chunk_exists_local ? "exists" : "does not exist"); | |
5077 | if (chunk_exists_local) { | |
c35f9726 | 5078 | ret_code = LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL; |
c35f9726 JG |
5079 | goto end; |
5080 | } else if (is_local_trace) { | |
5081 | ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; | |
5082 | goto end; | |
5083 | } | |
5084 | ||
5085 | rcu_read_lock(); | |
5086 | relayd = consumer_find_relayd(*relayd_id); | |
5087 | if (!relayd) { | |
5088 | ERR("Failed to find relayd %" PRIu64, *relayd_id); | |
5089 | ret_code = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; | |
5090 | goto end_rcu_unlock; | |
5091 | } | |
5092 | DBG("Looking up existence of trace chunk on relay daemon"); | |
5093 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
5094 | ret = relayd_trace_chunk_exists(&relayd->control_sock, chunk_id, | |
5095 | &chunk_exists_remote); | |
5096 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
5097 | if (ret < 0) { | |
5098 | ERR("Failed to look-up the existence of trace chunk on relay daemon"); | |
5099 | ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; | |
5100 | goto end_rcu_unlock; | |
5101 | } | |
5102 | ||
5103 | ret_code = chunk_exists_remote ? | |
5104 | LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE : | |
d2956687 | 5105 | LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; |
c35f9726 JG |
5106 | DBG("Trace chunk %s on relay daemon", |
5107 | chunk_exists_remote ? "exists" : "does not exist"); | |
d2956687 | 5108 | |
c35f9726 JG |
5109 | end_rcu_unlock: |
5110 | rcu_read_unlock(); | |
5111 | end: | |
d2956687 | 5112 | return ret_code; |
3654ed19 | 5113 | } |
5f3aff8b MD |
5114 | |
5115 | static | |
5116 | int consumer_clear_monitored_channel(struct lttng_consumer_channel *channel) | |
5117 | { | |
5118 | struct lttng_ht *ht; | |
5119 | struct lttng_consumer_stream *stream; | |
5120 | struct lttng_ht_iter iter; | |
5121 | int ret; | |
5122 | ||
fa29bfbf | 5123 | ht = the_consumer_data.stream_per_chan_id_ht; |
5f3aff8b MD |
5124 | |
5125 | rcu_read_lock(); | |
5126 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
5127 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
5128 | ht->match_fct, &channel->key, | |
5129 | &iter.iter, stream, node_channel_id.node) { | |
5130 | /* | |
5131 | * Protect against teardown with mutex. | |
5132 | */ | |
5133 | pthread_mutex_lock(&stream->lock); | |
5134 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
5135 | goto next; | |
5136 | } | |
5137 | ret = consumer_clear_stream(stream); | |
5138 | if (ret) { | |
5139 | goto error_unlock; | |
5140 | } | |
5141 | next: | |
5142 | pthread_mutex_unlock(&stream->lock); | |
5143 | } | |
5144 | rcu_read_unlock(); | |
5145 | return LTTCOMM_CONSUMERD_SUCCESS; | |
5146 | ||
5147 | error_unlock: | |
5148 | pthread_mutex_unlock(&stream->lock); | |
5149 | rcu_read_unlock(); | |
5150 | return ret; | |
5151 | } | |
5152 | ||
5153 | int lttng_consumer_clear_channel(struct lttng_consumer_channel *channel) | |
5154 | { | |
5155 | int ret; | |
5156 | ||
5157 | DBG("Consumer clear channel %" PRIu64, channel->key); | |
5158 | ||
5159 | if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
5160 | /* | |
5161 | * Nothing to do for the metadata channel/stream. | |
5162 | * Snapshot mechanism already take care of the metadata | |
5163 | * handling/generation, and monitored channels only need to | |
5164 | * have their data stream cleared.. | |
5165 | */ | |
5166 | ret = LTTCOMM_CONSUMERD_SUCCESS; | |
5167 | goto end; | |
5168 | } | |
5169 | ||
5170 | if (!channel->monitor) { | |
5171 | ret = consumer_clear_unmonitored_channel(channel); | |
5172 | } else { | |
5173 | ret = consumer_clear_monitored_channel(channel); | |
5174 | } | |
5175 | end: | |
5176 | return ret; | |
5177 | } | |
04ed9e10 JG |
5178 | |
5179 | enum lttcomm_return_code lttng_consumer_open_channel_packets( | |
5180 | struct lttng_consumer_channel *channel) | |
5181 | { | |
5182 | struct lttng_consumer_stream *stream; | |
5183 | enum lttcomm_return_code ret = LTTCOMM_CONSUMERD_SUCCESS; | |
5184 | ||
5185 | if (channel->metadata_stream) { | |
5186 | ERR("Open channel packets command attempted on a metadata channel"); | |
5187 | ret = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; | |
5188 | goto end; | |
5189 | } | |
5190 | ||
5191 | rcu_read_lock(); | |
5192 | cds_list_for_each_entry(stream, &channel->streams.head, send_node) { | |
503fefca | 5193 | enum consumer_stream_open_packet_status status; |
04ed9e10 JG |
5194 | |
5195 | pthread_mutex_lock(&stream->lock); | |
5196 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
5197 | goto next; | |
5198 | } | |
5199 | ||
503fefca | 5200 | status = consumer_stream_open_packet(stream); |
04ed9e10 | 5201 | switch (status) { |
503fefca | 5202 | case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED: |
04ed9e10 JG |
5203 | DBG("Opened a packet in \"open channel packets\" command: stream id = %" PRIu64 |
5204 | ", channel name = %s, session id = %" PRIu64, | |
5205 | stream->key, stream->chan->name, | |
5206 | stream->chan->session_id); | |
5207 | stream->opened_packet_in_current_trace_chunk = true; | |
5208 | break; | |
503fefca | 5209 | case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE: |
04ed9e10 JG |
5210 | DBG("No space left to open a packet in \"open channel packets\" command: stream id = %" PRIu64 |
5211 | ", channel name = %s, session id = %" PRIu64, | |
5212 | stream->key, stream->chan->name, | |
5213 | stream->chan->session_id); | |
5214 | break; | |
503fefca | 5215 | case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR: |
04ed9e10 JG |
5216 | /* |
5217 | * Only unexpected internal errors can lead to this | |
5218 | * failing. Report an unknown error. | |
5219 | */ | |
5220 | ERR("Failed to flush empty buffer in \"open channel packets\" command: stream id = %" PRIu64 | |
5221 | ", channel id = %" PRIu64 | |
5222 | ", channel name = %s" | |
5223 | ", session id = %" PRIu64, | |
5224 | stream->key, channel->key, | |
5225 | channel->name, channel->session_id); | |
5226 | ret = LTTCOMM_CONSUMERD_UNKNOWN_ERROR; | |
5227 | goto error_unlock; | |
5228 | default: | |
5229 | abort(); | |
5230 | } | |
5231 | ||
5232 | next: | |
5233 | pthread_mutex_unlock(&stream->lock); | |
5234 | } | |
5235 | ||
5236 | end_rcu_unlock: | |
5237 | rcu_read_unlock(); | |
5238 | end: | |
5239 | return ret; | |
5240 | ||
5241 | error_unlock: | |
5242 | pthread_mutex_unlock(&stream->lock); | |
5243 | goto end_rcu_unlock; | |
5244 | } | |
d85707b0 MD |
5245 | |
5246 | void lttng_consumer_sigbus_handle(void *addr) | |
5247 | { | |
5248 | lttng_ustconsumer_sigbus_handle(addr); | |
5249 | } |