Commit | Line | Data |
---|---|---|
3bd1e081 MD |
1 | /* |
2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> | |
3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
b3530820 | 4 | * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com> |
3bd1e081 | 5 | * |
d14d33bf AM |
6 | * This program is free software; you can redistribute it and/or modify |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
3bd1e081 MD |
9 | * |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
d14d33bf AM |
15 | * You should have received a copy of the GNU General Public License along |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
3bd1e081 MD |
18 | */ |
19 | ||
b770aa7f | 20 | #include "common/buffer-view.h" |
7775df52 | 21 | #include <stdint.h> |
6c1c0768 | 22 | #define _LGPL_SOURCE |
3bd1e081 | 23 | #include <assert.h> |
3bd1e081 MD |
24 | #include <poll.h> |
25 | #include <pthread.h> | |
26 | #include <stdlib.h> | |
27 | #include <string.h> | |
28 | #include <sys/mman.h> | |
29 | #include <sys/socket.h> | |
30 | #include <sys/types.h> | |
77c7c900 | 31 | #include <inttypes.h> |
3bd1e081 | 32 | #include <unistd.h> |
dbb5dfe6 | 33 | #include <sys/stat.h> |
3bd1e081 | 34 | |
51a9e1c7 | 35 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 36 | #include <common/common.h> |
10a8a223 | 37 | #include <common/kernel-ctl/kernel-ctl.h> |
10a8a223 | 38 | #include <common/sessiond-comm/sessiond-comm.h> |
00e2e675 | 39 | #include <common/sessiond-comm/relayd.h> |
dbb5dfe6 | 40 | #include <common/compat/fcntl.h> |
f263b7fd | 41 | #include <common/compat/endian.h> |
acdb9057 | 42 | #include <common/pipe.h> |
00e2e675 | 43 | #include <common/relayd/relayd.h> |
fe4477ee | 44 | #include <common/utils.h> |
c8fea79c | 45 | #include <common/consumer/consumer-stream.h> |
309167d2 | 46 | #include <common/index/index.h> |
c8fea79c | 47 | #include <common/consumer/consumer-timer.h> |
e5148e25 | 48 | #include <common/optional.h> |
0857097f | 49 | |
10a8a223 | 50 | #include "kernel-consumer.h" |
3bd1e081 MD |
51 | |
52 | extern struct lttng_consumer_global_data consumer_data; | |
53 | extern int consumer_poll_timeout; | |
3bd1e081 | 54 | |
3bd1e081 MD |
55 | /* |
56 | * Take a snapshot for a specific fd | |
57 | * | |
58 | * Returns 0 on success, < 0 on error | |
59 | */ | |
ffe60014 | 60 | int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
61 | { |
62 | int ret = 0; | |
63 | int infd = stream->wait_fd; | |
64 | ||
65 | ret = kernctl_snapshot(infd); | |
d2d2f190 JD |
66 | /* |
67 | * -EAGAIN is not an error, it just means that there is no data to | |
68 | * be read. | |
69 | */ | |
70 | if (ret != 0 && ret != -EAGAIN) { | |
5a510c9f | 71 | PERROR("Getting sub-buffer snapshot."); |
3bd1e081 MD |
72 | } |
73 | ||
74 | return ret; | |
75 | } | |
76 | ||
e9404c27 JG |
77 | /* |
78 | * Sample consumed and produced positions for a specific fd. | |
79 | * | |
80 | * Returns 0 on success, < 0 on error. | |
81 | */ | |
82 | int lttng_kconsumer_sample_snapshot_positions( | |
83 | struct lttng_consumer_stream *stream) | |
84 | { | |
85 | assert(stream); | |
86 | ||
87 | return kernctl_snapshot_sample_positions(stream->wait_fd); | |
88 | } | |
89 | ||
3bd1e081 MD |
90 | /* |
91 | * Get the produced position | |
92 | * | |
93 | * Returns 0 on success, < 0 on error | |
94 | */ | |
ffe60014 | 95 | int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
96 | unsigned long *pos) |
97 | { | |
98 | int ret; | |
99 | int infd = stream->wait_fd; | |
100 | ||
101 | ret = kernctl_snapshot_get_produced(infd, pos); | |
102 | if (ret != 0) { | |
5a510c9f | 103 | PERROR("kernctl_snapshot_get_produced"); |
3bd1e081 MD |
104 | } |
105 | ||
106 | return ret; | |
107 | } | |
108 | ||
07b86b52 JD |
109 | /* |
110 | * Get the consumerd position | |
111 | * | |
112 | * Returns 0 on success, < 0 on error | |
113 | */ | |
114 | int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, | |
115 | unsigned long *pos) | |
116 | { | |
117 | int ret; | |
118 | int infd = stream->wait_fd; | |
119 | ||
120 | ret = kernctl_snapshot_get_consumed(infd, pos); | |
121 | if (ret != 0) { | |
5a510c9f | 122 | PERROR("kernctl_snapshot_get_consumed"); |
07b86b52 JD |
123 | } |
124 | ||
125 | return ret; | |
126 | } | |
127 | ||
7775df52 JG |
128 | static |
129 | int get_current_subbuf_addr(struct lttng_consumer_stream *stream, | |
130 | const char **addr) | |
131 | { | |
132 | int ret; | |
133 | unsigned long mmap_offset; | |
134 | const char *mmap_base = stream->mmap_base; | |
135 | ||
136 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); | |
137 | if (ret < 0) { | |
138 | PERROR("Failed to get mmap read offset"); | |
139 | goto error; | |
140 | } | |
141 | ||
142 | *addr = mmap_base + mmap_offset; | |
143 | error: | |
144 | return ret; | |
145 | } | |
146 | ||
07b86b52 JD |
147 | /* |
148 | * Take a snapshot of all the stream of a channel | |
b0226bd4 | 149 | * RCU read-side lock must be held across this function to ensure existence of |
e5148e25 | 150 | * channel. The channel lock must be held by the caller. |
07b86b52 JD |
151 | * |
152 | * Returns 0 on success, < 0 on error | |
153 | */ | |
f0f048c9 JG |
154 | static int lttng_kconsumer_snapshot_channel( |
155 | struct lttng_consumer_channel *channel, | |
156 | uint64_t key, char *path, uint64_t relayd_id, | |
157 | uint64_t nb_packets_per_stream, | |
5c786ded | 158 | struct lttng_consumer_local_data *ctx) |
07b86b52 JD |
159 | { |
160 | int ret; | |
07b86b52 JD |
161 | struct lttng_consumer_stream *stream; |
162 | ||
6a00837f | 163 | DBG("Kernel consumer snapshot channel %" PRIu64, key); |
07b86b52 JD |
164 | |
165 | rcu_read_lock(); | |
166 | ||
07b86b52 JD |
167 | /* Splice is not supported yet for channel snapshot. */ |
168 | if (channel->output != CONSUMER_CHANNEL_MMAP) { | |
7318a78f JG |
169 | ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot", |
170 | channel->name); | |
07b86b52 JD |
171 | ret = -1; |
172 | goto end; | |
173 | } | |
174 | ||
10a50311 | 175 | cds_list_for_each_entry(stream, &channel->streams.head, send_node) { |
923333cd | 176 | unsigned long consumed_pos, produced_pos; |
9ce5646a MD |
177 | |
178 | health_code_update(); | |
179 | ||
07b86b52 JD |
180 | /* |
181 | * Lock stream because we are about to change its state. | |
182 | */ | |
183 | pthread_mutex_lock(&stream->lock); | |
184 | ||
e5148e25 JG |
185 | assert(channel->trace_chunk); |
186 | if (!lttng_trace_chunk_get(channel->trace_chunk)) { | |
187 | /* | |
188 | * Can't happen barring an internal error as the channel | |
189 | * holds a reference to the trace chunk. | |
190 | */ | |
191 | ERR("Failed to acquire reference to channel's trace chunk"); | |
192 | ret = -1; | |
193 | goto end_unlock; | |
194 | } | |
195 | assert(!stream->trace_chunk); | |
196 | stream->trace_chunk = channel->trace_chunk; | |
197 | ||
29decac3 DG |
198 | /* |
199 | * Assign the received relayd ID so we can use it for streaming. The streams | |
200 | * are not visible to anyone so this is OK to change it. | |
201 | */ | |
07b86b52 JD |
202 | stream->net_seq_idx = relayd_id; |
203 | channel->relayd_id = relayd_id; | |
204 | if (relayd_id != (uint64_t) -1ULL) { | |
10a50311 | 205 | ret = consumer_send_relayd_stream(stream, path); |
07b86b52 JD |
206 | if (ret < 0) { |
207 | ERR("sending stream to relayd"); | |
208 | goto end_unlock; | |
209 | } | |
07b86b52 | 210 | } else { |
e5148e25 JG |
211 | ret = consumer_stream_create_output_files(stream, |
212 | false); | |
07b86b52 | 213 | if (ret < 0) { |
07b86b52 JD |
214 | goto end_unlock; |
215 | } | |
e5148e25 JG |
216 | DBG("Kernel consumer snapshot stream (%" PRIu64 ")", |
217 | stream->key); | |
07b86b52 JD |
218 | } |
219 | ||
f22dd891 | 220 | ret = kernctl_buffer_flush_empty(stream->wait_fd); |
07b86b52 | 221 | if (ret < 0) { |
f22dd891 MD |
222 | /* |
223 | * Doing a buffer flush which does not take into | |
224 | * account empty packets. This is not perfect | |
225 | * for stream intersection, but required as a | |
226 | * fall-back when "flush_empty" is not | |
227 | * implemented by lttng-modules. | |
228 | */ | |
229 | ret = kernctl_buffer_flush(stream->wait_fd); | |
230 | if (ret < 0) { | |
231 | ERR("Failed to flush kernel stream"); | |
232 | goto end_unlock; | |
233 | } | |
07b86b52 JD |
234 | goto end_unlock; |
235 | } | |
236 | ||
237 | ret = lttng_kconsumer_take_snapshot(stream); | |
238 | if (ret < 0) { | |
239 | ERR("Taking kernel snapshot"); | |
240 | goto end_unlock; | |
241 | } | |
242 | ||
243 | ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos); | |
244 | if (ret < 0) { | |
245 | ERR("Produced kernel snapshot position"); | |
246 | goto end_unlock; | |
247 | } | |
248 | ||
249 | ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos); | |
250 | if (ret < 0) { | |
251 | ERR("Consumerd kernel snapshot position"); | |
252 | goto end_unlock; | |
253 | } | |
254 | ||
d07ceecd MD |
255 | consumed_pos = consumer_get_consume_start_pos(consumed_pos, |
256 | produced_pos, nb_packets_per_stream, | |
257 | stream->max_sb_size); | |
5c786ded | 258 | |
0fdaf1ed | 259 | while ((long) (consumed_pos - produced_pos) < 0) { |
07b86b52 JD |
260 | ssize_t read_len; |
261 | unsigned long len, padded_len; | |
7775df52 | 262 | const char *subbuf_addr; |
b770aa7f | 263 | struct lttng_buffer_view subbuf_view; |
07b86b52 | 264 | |
9ce5646a | 265 | health_code_update(); |
07b86b52 JD |
266 | DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos); |
267 | ||
268 | ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos); | |
269 | if (ret < 0) { | |
32af2c95 | 270 | if (ret != -EAGAIN) { |
07b86b52 JD |
271 | PERROR("kernctl_get_subbuf snapshot"); |
272 | goto end_unlock; | |
273 | } | |
274 | DBG("Kernel consumer get subbuf failed. Skipping it."); | |
275 | consumed_pos += stream->max_sb_size; | |
ddc93ee4 | 276 | stream->chan->lost_packets++; |
07b86b52 JD |
277 | continue; |
278 | } | |
279 | ||
280 | ret = kernctl_get_subbuf_size(stream->wait_fd, &len); | |
281 | if (ret < 0) { | |
282 | ERR("Snapshot kernctl_get_subbuf_size"); | |
29decac3 | 283 | goto error_put_subbuf; |
07b86b52 JD |
284 | } |
285 | ||
286 | ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len); | |
287 | if (ret < 0) { | |
288 | ERR("Snapshot kernctl_get_padded_subbuf_size"); | |
29decac3 | 289 | goto error_put_subbuf; |
07b86b52 JD |
290 | } |
291 | ||
7775df52 JG |
292 | ret = get_current_subbuf_addr(stream, &subbuf_addr); |
293 | if (ret) { | |
294 | goto error_put_subbuf; | |
295 | } | |
296 | ||
b770aa7f JG |
297 | subbuf_view = lttng_buffer_view_init( |
298 | subbuf_addr, 0, padded_len); | |
7775df52 | 299 | read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, |
b770aa7f | 300 | stream, &subbuf_view, |
309167d2 | 301 | padded_len - len, NULL); |
07b86b52 | 302 | /* |
29decac3 DG |
303 | * We write the padded len in local tracefiles but the data len |
304 | * when using a relay. Display the error but continue processing | |
305 | * to try to release the subbuffer. | |
07b86b52 JD |
306 | */ |
307 | if (relayd_id != (uint64_t) -1ULL) { | |
308 | if (read_len != len) { | |
309 | ERR("Error sending to the relay (ret: %zd != len: %lu)", | |
310 | read_len, len); | |
311 | } | |
312 | } else { | |
313 | if (read_len != padded_len) { | |
314 | ERR("Error writing to tracefile (ret: %zd != len: %lu)", | |
315 | read_len, padded_len); | |
316 | } | |
317 | } | |
318 | ||
319 | ret = kernctl_put_subbuf(stream->wait_fd); | |
320 | if (ret < 0) { | |
321 | ERR("Snapshot kernctl_put_subbuf"); | |
322 | goto end_unlock; | |
323 | } | |
324 | consumed_pos += stream->max_sb_size; | |
325 | } | |
326 | ||
327 | if (relayd_id == (uint64_t) -1ULL) { | |
fdf9986c MD |
328 | if (stream->out_fd >= 0) { |
329 | ret = close(stream->out_fd); | |
330 | if (ret < 0) { | |
331 | PERROR("Kernel consumer snapshot close out_fd"); | |
332 | goto end_unlock; | |
333 | } | |
334 | stream->out_fd = -1; | |
07b86b52 | 335 | } |
07b86b52 JD |
336 | } else { |
337 | close_relayd_stream(stream); | |
338 | stream->net_seq_idx = (uint64_t) -1ULL; | |
339 | } | |
e5148e25 JG |
340 | lttng_trace_chunk_put(stream->trace_chunk); |
341 | stream->trace_chunk = NULL; | |
07b86b52 JD |
342 | pthread_mutex_unlock(&stream->lock); |
343 | } | |
344 | ||
345 | /* All good! */ | |
346 | ret = 0; | |
347 | goto end; | |
348 | ||
29decac3 DG |
349 | error_put_subbuf: |
350 | ret = kernctl_put_subbuf(stream->wait_fd); | |
351 | if (ret < 0) { | |
352 | ERR("Snapshot kernctl_put_subbuf error path"); | |
353 | } | |
07b86b52 JD |
354 | end_unlock: |
355 | pthread_mutex_unlock(&stream->lock); | |
356 | end: | |
357 | rcu_read_unlock(); | |
358 | return ret; | |
359 | } | |
360 | ||
361 | /* | |
362 | * Read the whole metadata available for a snapshot. | |
b0226bd4 | 363 | * RCU read-side lock must be held across this function to ensure existence of |
e5148e25 | 364 | * metadata_channel. The channel lock must be held by the caller. |
07b86b52 JD |
365 | * |
366 | * Returns 0 on success, < 0 on error | |
367 | */ | |
e5148e25 JG |
368 | static int lttng_kconsumer_snapshot_metadata( |
369 | struct lttng_consumer_channel *metadata_channel, | |
b0226bd4 MD |
370 | uint64_t key, char *path, uint64_t relayd_id, |
371 | struct lttng_consumer_local_data *ctx) | |
07b86b52 | 372 | { |
d771f832 DG |
373 | int ret, use_relayd = 0; |
374 | ssize_t ret_read; | |
07b86b52 | 375 | struct lttng_consumer_stream *metadata_stream; |
d771f832 DG |
376 | |
377 | assert(ctx); | |
07b86b52 JD |
378 | |
379 | DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s", | |
380 | key, path); | |
381 | ||
382 | rcu_read_lock(); | |
383 | ||
07b86b52 JD |
384 | metadata_stream = metadata_channel->metadata_stream; |
385 | assert(metadata_stream); | |
e5148e25 | 386 | |
c55fe3e3 | 387 | pthread_mutex_lock(&metadata_stream->lock); |
e5148e25 JG |
388 | assert(metadata_channel->trace_chunk); |
389 | assert(metadata_stream->trace_chunk); | |
07b86b52 | 390 | |
d771f832 | 391 | /* Flag once that we have a valid relayd for the stream. */ |
e2039c7a | 392 | if (relayd_id != (uint64_t) -1ULL) { |
d771f832 DG |
393 | use_relayd = 1; |
394 | } | |
395 | ||
396 | if (use_relayd) { | |
10a50311 | 397 | ret = consumer_send_relayd_stream(metadata_stream, path); |
e2039c7a | 398 | if (ret < 0) { |
c55fe3e3 | 399 | goto error_snapshot; |
e2039c7a | 400 | } |
e2039c7a | 401 | } else { |
e5148e25 JG |
402 | ret = consumer_stream_create_output_files(metadata_stream, |
403 | false); | |
e2039c7a | 404 | if (ret < 0) { |
c55fe3e3 | 405 | goto error_snapshot; |
e2039c7a | 406 | } |
07b86b52 | 407 | } |
07b86b52 | 408 | |
d771f832 | 409 | do { |
9ce5646a MD |
410 | health_code_update(); |
411 | ||
e5148e25 | 412 | ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx); |
d771f832 | 413 | if (ret_read < 0) { |
56591bac | 414 | if (ret_read != -EAGAIN) { |
6a00837f | 415 | ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)", |
d771f832 | 416 | ret_read); |
c55fe3e3 JG |
417 | ret = ret_read; |
418 | goto error_snapshot; | |
07b86b52 | 419 | } |
d771f832 | 420 | /* ret_read is negative at this point so we will exit the loop. */ |
07b86b52 JD |
421 | continue; |
422 | } | |
d771f832 | 423 | } while (ret_read >= 0); |
07b86b52 | 424 | |
d771f832 DG |
425 | if (use_relayd) { |
426 | close_relayd_stream(metadata_stream); | |
427 | metadata_stream->net_seq_idx = (uint64_t) -1ULL; | |
428 | } else { | |
fdf9986c MD |
429 | if (metadata_stream->out_fd >= 0) { |
430 | ret = close(metadata_stream->out_fd); | |
431 | if (ret < 0) { | |
432 | PERROR("Kernel consumer snapshot metadata close out_fd"); | |
433 | /* | |
434 | * Don't go on error here since the snapshot was successful at this | |
435 | * point but somehow the close failed. | |
436 | */ | |
437 | } | |
438 | metadata_stream->out_fd = -1; | |
e5148e25 JG |
439 | lttng_trace_chunk_put(metadata_stream->trace_chunk); |
440 | metadata_stream->trace_chunk = NULL; | |
e2039c7a | 441 | } |
e2039c7a JD |
442 | } |
443 | ||
07b86b52 | 444 | ret = 0; |
c55fe3e3 JG |
445 | error_snapshot: |
446 | pthread_mutex_unlock(&metadata_stream->lock); | |
cf53a8a6 JD |
447 | cds_list_del(&metadata_stream->send_node); |
448 | consumer_stream_destroy(metadata_stream, NULL); | |
449 | metadata_channel->metadata_stream = NULL; | |
07b86b52 JD |
450 | rcu_read_unlock(); |
451 | return ret; | |
452 | } | |
453 | ||
1803a064 MD |
454 | /* |
455 | * Receive command from session daemon and process it. | |
456 | * | |
457 | * Return 1 on success else a negative value or 0. | |
458 | */ | |
3bd1e081 MD |
459 | int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
460 | int sock, struct pollfd *consumer_sockpoll) | |
461 | { | |
462 | ssize_t ret; | |
0c759fc9 | 463 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
3bd1e081 MD |
464 | struct lttcomm_consumer_msg msg; |
465 | ||
9ce5646a MD |
466 | health_code_update(); |
467 | ||
3bd1e081 MD |
468 | ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); |
469 | if (ret != sizeof(msg)) { | |
1803a064 | 470 | if (ret > 0) { |
c6857fcf | 471 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); |
1803a064 MD |
472 | ret = -1; |
473 | } | |
3bd1e081 MD |
474 | return ret; |
475 | } | |
9ce5646a MD |
476 | |
477 | health_code_update(); | |
478 | ||
84382d49 MD |
479 | /* Deprecated command */ |
480 | assert(msg.cmd_type != LTTNG_CONSUMER_STOP); | |
3bd1e081 | 481 | |
9ce5646a MD |
482 | health_code_update(); |
483 | ||
b0b335c8 MD |
484 | /* relayd needs RCU read-side protection */ |
485 | rcu_read_lock(); | |
486 | ||
3bd1e081 | 487 | switch (msg.cmd_type) { |
00e2e675 DG |
488 | case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: |
489 | { | |
f50f23d9 | 490 | /* Session daemon status message are handled in the following call. */ |
2527bf85 | 491 | consumer_add_relayd_socket(msg.u.relayd_sock.net_index, |
7735ef9e | 492 | msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, |
d3e2ba59 | 493 | &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id, |
2527bf85 | 494 | msg.u.relayd_sock.relayd_session_id); |
00e2e675 DG |
495 | goto end_nosignal; |
496 | } | |
3bd1e081 MD |
497 | case LTTNG_CONSUMER_ADD_CHANNEL: |
498 | { | |
499 | struct lttng_consumer_channel *new_channel; | |
e43c41c5 | 500 | int ret_recv; |
e5148e25 | 501 | const uint64_t chunk_id = msg.u.channel.chunk_id.value; |
3bd1e081 | 502 | |
9ce5646a MD |
503 | health_code_update(); |
504 | ||
f50f23d9 DG |
505 | /* First send a status message before receiving the fds. */ |
506 | ret = consumer_send_status_msg(sock, ret_code); | |
507 | if (ret < 0) { | |
508 | /* Somehow, the session daemon is not responding anymore. */ | |
1803a064 | 509 | goto error_fatal; |
f50f23d9 | 510 | } |
9ce5646a MD |
511 | |
512 | health_code_update(); | |
513 | ||
d88aee68 | 514 | DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key); |
3bd1e081 | 515 | new_channel = consumer_allocate_channel(msg.u.channel.channel_key, |
e5148e25 JG |
516 | msg.u.channel.session_id, |
517 | msg.u.channel.chunk_id.is_set ? | |
518 | &chunk_id : NULL, | |
519 | msg.u.channel.pathname, | |
520 | msg.u.channel.name, | |
1624d5b7 JD |
521 | msg.u.channel.relayd_id, msg.u.channel.output, |
522 | msg.u.channel.tracefile_size, | |
1950109e | 523 | msg.u.channel.tracefile_count, 0, |
ecc48a90 | 524 | msg.u.channel.monitor, |
d7ba1388 | 525 | msg.u.channel.live_timer_interval, |
3d071855 | 526 | NULL, NULL); |
3bd1e081 | 527 | if (new_channel == NULL) { |
f73fabfd | 528 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
3bd1e081 MD |
529 | goto end_nosignal; |
530 | } | |
ffe60014 | 531 | new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams; |
95a1109b JD |
532 | switch (msg.u.channel.output) { |
533 | case LTTNG_EVENT_SPLICE: | |
534 | new_channel->output = CONSUMER_CHANNEL_SPLICE; | |
535 | break; | |
536 | case LTTNG_EVENT_MMAP: | |
537 | new_channel->output = CONSUMER_CHANNEL_MMAP; | |
538 | break; | |
539 | default: | |
540 | ERR("Channel output unknown %d", msg.u.channel.output); | |
541 | goto end_nosignal; | |
542 | } | |
ffe60014 DG |
543 | |
544 | /* Translate and save channel type. */ | |
545 | switch (msg.u.channel.type) { | |
546 | case CONSUMER_CHANNEL_TYPE_DATA: | |
547 | case CONSUMER_CHANNEL_TYPE_METADATA: | |
548 | new_channel->type = msg.u.channel.type; | |
549 | break; | |
550 | default: | |
551 | assert(0); | |
552 | goto end_nosignal; | |
553 | }; | |
554 | ||
9ce5646a MD |
555 | health_code_update(); |
556 | ||
3bd1e081 | 557 | if (ctx->on_recv_channel != NULL) { |
e43c41c5 JD |
558 | ret_recv = ctx->on_recv_channel(new_channel); |
559 | if (ret_recv == 0) { | |
560 | ret = consumer_add_channel(new_channel, ctx); | |
561 | } else if (ret_recv < 0) { | |
3bd1e081 MD |
562 | goto end_nosignal; |
563 | } | |
564 | } else { | |
e43c41c5 | 565 | ret = consumer_add_channel(new_channel, ctx); |
3bd1e081 | 566 | } |
e9404c27 JG |
567 | if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) { |
568 | int monitor_start_ret; | |
569 | ||
570 | DBG("Consumer starting monitor timer"); | |
94d49140 JD |
571 | consumer_timer_live_start(new_channel, |
572 | msg.u.channel.live_timer_interval); | |
e9404c27 JG |
573 | monitor_start_ret = consumer_timer_monitor_start( |
574 | new_channel, | |
575 | msg.u.channel.monitor_timer_interval); | |
576 | if (monitor_start_ret < 0) { | |
577 | ERR("Starting channel monitoring timer failed"); | |
578 | goto end_nosignal; | |
579 | } | |
580 | ||
94d49140 | 581 | } |
e43c41c5 | 582 | |
9ce5646a MD |
583 | health_code_update(); |
584 | ||
e43c41c5 | 585 | /* If we received an error in add_channel, we need to report it. */ |
821fffb2 | 586 | if (ret < 0) { |
1803a064 MD |
587 | ret = consumer_send_status_msg(sock, ret); |
588 | if (ret < 0) { | |
589 | goto error_fatal; | |
590 | } | |
e43c41c5 JD |
591 | goto end_nosignal; |
592 | } | |
593 | ||
3bd1e081 MD |
594 | goto end_nosignal; |
595 | } | |
596 | case LTTNG_CONSUMER_ADD_STREAM: | |
597 | { | |
dae10966 DG |
598 | int fd; |
599 | struct lttng_pipe *stream_pipe; | |
00e2e675 | 600 | struct lttng_consumer_stream *new_stream; |
ffe60014 | 601 | struct lttng_consumer_channel *channel; |
c80048c6 | 602 | int alloc_ret = 0; |
3bd1e081 | 603 | |
ffe60014 DG |
604 | /* |
605 | * Get stream's channel reference. Needed when adding the stream to the | |
606 | * global hash table. | |
607 | */ | |
608 | channel = consumer_find_channel(msg.u.stream.channel_key); | |
609 | if (!channel) { | |
610 | /* | |
611 | * We could not find the channel. Can happen if cpu hotplug | |
612 | * happens while tearing down. | |
613 | */ | |
d88aee68 | 614 | ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key); |
e462382a | 615 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; |
ffe60014 DG |
616 | } |
617 | ||
9ce5646a MD |
618 | health_code_update(); |
619 | ||
f50f23d9 DG |
620 | /* First send a status message before receiving the fds. */ |
621 | ret = consumer_send_status_msg(sock, ret_code); | |
1803a064 | 622 | if (ret < 0) { |
d771f832 | 623 | /* Somehow, the session daemon is not responding anymore. */ |
e0e7757d | 624 | goto error_add_stream_fatal; |
1803a064 | 625 | } |
9ce5646a MD |
626 | |
627 | health_code_update(); | |
628 | ||
0c759fc9 | 629 | if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) { |
d771f832 | 630 | /* Channel was not found. */ |
e0e7757d | 631 | goto error_add_stream_nosignal; |
f50f23d9 DG |
632 | } |
633 | ||
d771f832 | 634 | /* Blocking call */ |
9ce5646a MD |
635 | health_poll_entry(); |
636 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
637 | health_poll_exit(); | |
84382d49 | 638 | if (ret) { |
e0e7757d | 639 | goto error_add_stream_fatal; |
3bd1e081 | 640 | } |
00e2e675 | 641 | |
9ce5646a MD |
642 | health_code_update(); |
643 | ||
00e2e675 | 644 | /* Get stream file descriptor from socket */ |
f2fc6720 MD |
645 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); |
646 | if (ret != sizeof(fd)) { | |
f73fabfd | 647 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
e0e7757d | 648 | goto end; |
3bd1e081 | 649 | } |
3bd1e081 | 650 | |
9ce5646a MD |
651 | health_code_update(); |
652 | ||
f50f23d9 DG |
653 | /* |
654 | * Send status code to session daemon only if the recv works. If the | |
655 | * above recv() failed, the session daemon is notified through the | |
656 | * error socket and the teardown is eventually done. | |
657 | */ | |
658 | ret = consumer_send_status_msg(sock, ret_code); | |
659 | if (ret < 0) { | |
660 | /* Somehow, the session daemon is not responding anymore. */ | |
e0e7757d | 661 | goto error_add_stream_nosignal; |
f50f23d9 DG |
662 | } |
663 | ||
9ce5646a MD |
664 | health_code_update(); |
665 | ||
e5148e25 | 666 | pthread_mutex_lock(&channel->lock); |
c0c85e4f JG |
667 | new_stream = consumer_allocate_stream( |
668 | channel, | |
669 | channel->key, | |
ffe60014 | 670 | fd, |
ffe60014 | 671 | channel->name, |
ffe60014 DG |
672 | channel->relayd_id, |
673 | channel->session_id, | |
e5148e25 | 674 | channel->trace_chunk, |
ffe60014 DG |
675 | msg.u.stream.cpu, |
676 | &alloc_ret, | |
4891ece8 | 677 | channel->type, |
e5148e25 | 678 | channel->monitor); |
3bd1e081 | 679 | if (new_stream == NULL) { |
c80048c6 MD |
680 | switch (alloc_ret) { |
681 | case -ENOMEM: | |
682 | case -EINVAL: | |
683 | default: | |
684 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); | |
685 | break; | |
c80048c6 | 686 | } |
e5148e25 | 687 | pthread_mutex_unlock(&channel->lock); |
e0e7757d | 688 | goto error_add_stream_nosignal; |
3bd1e081 | 689 | } |
d771f832 | 690 | |
ffe60014 | 691 | new_stream->wait_fd = fd; |
2c42aa5c JG |
692 | ret = kernctl_get_max_subbuf_size(new_stream->wait_fd, |
693 | &new_stream->max_sb_size); | |
694 | if (ret < 0) { | |
695 | pthread_mutex_unlock(&channel->lock); | |
696 | ERR("Failed to get kernel maximal subbuffer size"); | |
e0e7757d | 697 | goto error_add_stream_nosignal; |
2c42aa5c JG |
698 | } |
699 | ||
d9a2e16e JD |
700 | consumer_stream_update_channel_attributes(new_stream, |
701 | channel); | |
07b86b52 JD |
702 | switch (channel->output) { |
703 | case CONSUMER_CHANNEL_SPLICE: | |
704 | new_stream->output = LTTNG_EVENT_SPLICE; | |
a2361a61 JD |
705 | ret = utils_create_pipe(new_stream->splice_pipe); |
706 | if (ret < 0) { | |
e5148e25 | 707 | pthread_mutex_unlock(&channel->lock); |
e0e7757d | 708 | goto error_add_stream_nosignal; |
a2361a61 | 709 | } |
07b86b52 JD |
710 | break; |
711 | case CONSUMER_CHANNEL_MMAP: | |
712 | new_stream->output = LTTNG_EVENT_MMAP; | |
713 | break; | |
714 | default: | |
715 | ERR("Stream output unknown %d", channel->output); | |
e5148e25 | 716 | pthread_mutex_unlock(&channel->lock); |
e0e7757d | 717 | goto error_add_stream_nosignal; |
07b86b52 | 718 | } |
00e2e675 | 719 | |
a0c83db9 DG |
720 | /* |
721 | * We've just assigned the channel to the stream so increment the | |
07b86b52 JD |
722 | * refcount right now. We don't need to increment the refcount for |
723 | * streams in no monitor because we handle manually the cleanup of | |
724 | * those. It is very important to make sure there is NO prior | |
725 | * consumer_del_stream() calls or else the refcount will be unbalanced. | |
a0c83db9 | 726 | */ |
07b86b52 JD |
727 | if (channel->monitor) { |
728 | uatomic_inc(&new_stream->chan->refcount); | |
729 | } | |
9d9353f9 | 730 | |
fb3a43a9 DG |
731 | /* |
732 | * The buffer flush is done on the session daemon side for the kernel | |
733 | * so no need for the stream "hangup_flush_done" variable to be | |
734 | * tracked. This is important for a kernel stream since we don't rely | |
735 | * on the flush state of the stream to read data. It's not the case for | |
736 | * user space tracing. | |
737 | */ | |
738 | new_stream->hangup_flush_done = 0; | |
739 | ||
9ce5646a MD |
740 | health_code_update(); |
741 | ||
e5148e25 | 742 | pthread_mutex_lock(&new_stream->lock); |
633d0084 DG |
743 | if (ctx->on_recv_stream) { |
744 | ret = ctx->on_recv_stream(new_stream); | |
745 | if (ret < 0) { | |
e5148e25 JG |
746 | pthread_mutex_unlock(&new_stream->lock); |
747 | pthread_mutex_unlock(&channel->lock); | |
d771f832 | 748 | consumer_stream_free(new_stream); |
e0e7757d | 749 | goto error_add_stream_nosignal; |
fb3a43a9 | 750 | } |
633d0084 | 751 | } |
9ce5646a MD |
752 | health_code_update(); |
753 | ||
07b86b52 JD |
754 | if (new_stream->metadata_flag) { |
755 | channel->metadata_stream = new_stream; | |
756 | } | |
757 | ||
2bba9e53 DG |
758 | /* Do not monitor this stream. */ |
759 | if (!channel->monitor) { | |
5eecee74 | 760 | DBG("Kernel consumer add stream %s in no monitor mode with " |
6dc3064a | 761 | "relayd id %" PRIu64, new_stream->name, |
5eecee74 | 762 | new_stream->net_seq_idx); |
10a50311 | 763 | cds_list_add(&new_stream->send_node, &channel->streams.head); |
e5148e25 JG |
764 | pthread_mutex_unlock(&new_stream->lock); |
765 | pthread_mutex_unlock(&channel->lock); | |
e0e7757d | 766 | goto end_add_stream; |
6dc3064a DG |
767 | } |
768 | ||
e1b71bdc DG |
769 | /* Send stream to relayd if the stream has an ID. */ |
770 | if (new_stream->net_seq_idx != (uint64_t) -1ULL) { | |
194ee077 DG |
771 | ret = consumer_send_relayd_stream(new_stream, |
772 | new_stream->chan->pathname); | |
e1b71bdc | 773 | if (ret < 0) { |
e5148e25 JG |
774 | pthread_mutex_unlock(&new_stream->lock); |
775 | pthread_mutex_unlock(&channel->lock); | |
e1b71bdc | 776 | consumer_stream_free(new_stream); |
e0e7757d | 777 | goto error_add_stream_nosignal; |
e1b71bdc | 778 | } |
001b7e62 MD |
779 | |
780 | /* | |
781 | * If adding an extra stream to an already | |
782 | * existing channel (e.g. cpu hotplug), we need | |
783 | * to send the "streams_sent" command to relayd. | |
784 | */ | |
785 | if (channel->streams_sent_to_relayd) { | |
786 | ret = consumer_send_relayd_streams_sent( | |
787 | new_stream->net_seq_idx); | |
788 | if (ret < 0) { | |
e5148e25 JG |
789 | pthread_mutex_unlock(&new_stream->lock); |
790 | pthread_mutex_unlock(&channel->lock); | |
e0e7757d | 791 | goto error_add_stream_nosignal; |
001b7e62 MD |
792 | } |
793 | } | |
e2039c7a | 794 | } |
e5148e25 JG |
795 | pthread_mutex_unlock(&new_stream->lock); |
796 | pthread_mutex_unlock(&channel->lock); | |
e2039c7a | 797 | |
50f8ae69 | 798 | /* Get the right pipe where the stream will be sent. */ |
633d0084 | 799 | if (new_stream->metadata_flag) { |
66d583dc | 800 | consumer_add_metadata_stream(new_stream); |
dae10966 | 801 | stream_pipe = ctx->consumer_metadata_pipe; |
3bd1e081 | 802 | } else { |
66d583dc | 803 | consumer_add_data_stream(new_stream); |
dae10966 | 804 | stream_pipe = ctx->consumer_data_pipe; |
50f8ae69 DG |
805 | } |
806 | ||
66d583dc | 807 | /* Visible to other threads */ |
5ab66908 MD |
808 | new_stream->globally_visible = 1; |
809 | ||
9ce5646a MD |
810 | health_code_update(); |
811 | ||
dae10966 | 812 | ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream)); |
50f8ae69 | 813 | if (ret < 0) { |
dae10966 | 814 | ERR("Consumer write %s stream to pipe %d", |
50f8ae69 | 815 | new_stream->metadata_flag ? "metadata" : "data", |
dae10966 | 816 | lttng_pipe_get_writefd(stream_pipe)); |
5ab66908 MD |
817 | if (new_stream->metadata_flag) { |
818 | consumer_del_stream_for_metadata(new_stream); | |
819 | } else { | |
820 | consumer_del_stream_for_data(new_stream); | |
821 | } | |
e0e7757d | 822 | goto error_add_stream_nosignal; |
3bd1e081 | 823 | } |
00e2e675 | 824 | |
02d02e31 JD |
825 | DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64, |
826 | new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id); | |
e0e7757d | 827 | end_add_stream: |
3bd1e081 | 828 | break; |
e0e7757d JG |
829 | error_add_stream_nosignal: |
830 | goto end_nosignal; | |
831 | error_add_stream_fatal: | |
832 | goto error_fatal; | |
3bd1e081 | 833 | } |
a4baae1b JD |
834 | case LTTNG_CONSUMER_STREAMS_SENT: |
835 | { | |
836 | struct lttng_consumer_channel *channel; | |
837 | ||
838 | /* | |
839 | * Get stream's channel reference. Needed when adding the stream to the | |
840 | * global hash table. | |
841 | */ | |
842 | channel = consumer_find_channel(msg.u.sent_streams.channel_key); | |
843 | if (!channel) { | |
844 | /* | |
845 | * We could not find the channel. Can happen if cpu hotplug | |
846 | * happens while tearing down. | |
847 | */ | |
848 | ERR("Unable to find channel key %" PRIu64, | |
849 | msg.u.sent_streams.channel_key); | |
e462382a | 850 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; |
a4baae1b JD |
851 | } |
852 | ||
853 | health_code_update(); | |
854 | ||
855 | /* | |
856 | * Send status code to session daemon. | |
857 | */ | |
858 | ret = consumer_send_status_msg(sock, ret_code); | |
f261ad0a | 859 | if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) { |
a4baae1b | 860 | /* Somehow, the session daemon is not responding anymore. */ |
6ec577e9 | 861 | goto error_streams_sent_nosignal; |
a4baae1b JD |
862 | } |
863 | ||
864 | health_code_update(); | |
865 | ||
866 | /* | |
867 | * We should not send this message if we don't monitor the | |
868 | * streams in this channel. | |
869 | */ | |
870 | if (!channel->monitor) { | |
6ec577e9 | 871 | goto end_error_streams_sent; |
a4baae1b JD |
872 | } |
873 | ||
874 | health_code_update(); | |
875 | /* Send stream to relayd if the stream has an ID. */ | |
876 | if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) { | |
877 | ret = consumer_send_relayd_streams_sent( | |
878 | msg.u.sent_streams.net_seq_idx); | |
879 | if (ret < 0) { | |
6ec577e9 | 880 | goto error_streams_sent_nosignal; |
a4baae1b | 881 | } |
001b7e62 | 882 | channel->streams_sent_to_relayd = true; |
a4baae1b | 883 | } |
6ec577e9 | 884 | end_error_streams_sent: |
a4baae1b | 885 | break; |
6ec577e9 JG |
886 | error_streams_sent_nosignal: |
887 | goto end_nosignal; | |
a4baae1b | 888 | } |
3bd1e081 MD |
889 | case LTTNG_CONSUMER_UPDATE_STREAM: |
890 | { | |
3f8e211f DG |
891 | rcu_read_unlock(); |
892 | return -ENOSYS; | |
893 | } | |
894 | case LTTNG_CONSUMER_DESTROY_RELAYD: | |
895 | { | |
a6ba4fe1 | 896 | uint64_t index = msg.u.destroy_relayd.net_seq_idx; |
3f8e211f DG |
897 | struct consumer_relayd_sock_pair *relayd; |
898 | ||
a6ba4fe1 | 899 | DBG("Kernel consumer destroying relayd %" PRIu64, index); |
3f8e211f DG |
900 | |
901 | /* Get relayd reference if exists. */ | |
a6ba4fe1 | 902 | relayd = consumer_find_relayd(index); |
3f8e211f | 903 | if (relayd == NULL) { |
3448e266 | 904 | DBG("Unable to find relayd %" PRIu64, index); |
e462382a | 905 | ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; |
3bd1e081 | 906 | } |
3f8e211f | 907 | |
a6ba4fe1 DG |
908 | /* |
909 | * Each relayd socket pair has a refcount of stream attached to it | |
910 | * which tells if the relayd is still active or not depending on the | |
911 | * refcount value. | |
912 | * | |
913 | * This will set the destroy flag of the relayd object and destroy it | |
914 | * if the refcount reaches zero when called. | |
915 | * | |
916 | * The destroy can happen either here or when a stream fd hangs up. | |
917 | */ | |
f50f23d9 DG |
918 | if (relayd) { |
919 | consumer_flag_relayd_for_destroy(relayd); | |
920 | } | |
921 | ||
9ce5646a MD |
922 | health_code_update(); |
923 | ||
f50f23d9 DG |
924 | ret = consumer_send_status_msg(sock, ret_code); |
925 | if (ret < 0) { | |
926 | /* Somehow, the session daemon is not responding anymore. */ | |
1803a064 | 927 | goto error_fatal; |
f50f23d9 | 928 | } |
3f8e211f | 929 | |
3f8e211f | 930 | goto end_nosignal; |
3bd1e081 | 931 | } |
6d805429 | 932 | case LTTNG_CONSUMER_DATA_PENDING: |
53632229 | 933 | { |
c8f59ee5 | 934 | int32_t ret; |
6d805429 | 935 | uint64_t id = msg.u.data_pending.session_id; |
c8f59ee5 | 936 | |
6d805429 | 937 | DBG("Kernel consumer data pending command for id %" PRIu64, id); |
c8f59ee5 | 938 | |
6d805429 | 939 | ret = consumer_data_pending(id); |
c8f59ee5 | 940 | |
9ce5646a MD |
941 | health_code_update(); |
942 | ||
c8f59ee5 DG |
943 | /* Send back returned value to session daemon */ |
944 | ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret)); | |
945 | if (ret < 0) { | |
6d805429 | 946 | PERROR("send data pending ret code"); |
1803a064 | 947 | goto error_fatal; |
c8f59ee5 | 948 | } |
f50f23d9 DG |
949 | |
950 | /* | |
951 | * No need to send back a status message since the data pending | |
952 | * returned value is the response. | |
953 | */ | |
c8f59ee5 | 954 | break; |
53632229 | 955 | } |
6dc3064a DG |
956 | case LTTNG_CONSUMER_SNAPSHOT_CHANNEL: |
957 | { | |
b0226bd4 MD |
958 | struct lttng_consumer_channel *channel; |
959 | uint64_t key = msg.u.snapshot_channel.key; | |
960 | ||
961 | channel = consumer_find_channel(key); | |
962 | if (!channel) { | |
963 | ERR("Channel %" PRIu64 " not found", key); | |
964 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; | |
07b86b52 | 965 | } else { |
e5148e25 | 966 | pthread_mutex_lock(&channel->lock); |
b0226bd4 MD |
967 | if (msg.u.snapshot_channel.metadata == 1) { |
968 | ret = lttng_kconsumer_snapshot_metadata(channel, key, | |
969 | msg.u.snapshot_channel.pathname, | |
970 | msg.u.snapshot_channel.relayd_id, ctx); | |
971 | if (ret < 0) { | |
972 | ERR("Snapshot metadata failed"); | |
973 | ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; | |
974 | } | |
975 | } else { | |
976 | ret = lttng_kconsumer_snapshot_channel(channel, key, | |
977 | msg.u.snapshot_channel.pathname, | |
978 | msg.u.snapshot_channel.relayd_id, | |
979 | msg.u.snapshot_channel.nb_packets_per_stream, | |
980 | ctx); | |
981 | if (ret < 0) { | |
982 | ERR("Snapshot channel failed"); | |
983 | ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; | |
984 | } | |
07b86b52 | 985 | } |
e5148e25 | 986 | pthread_mutex_unlock(&channel->lock); |
07b86b52 | 987 | } |
9ce5646a MD |
988 | health_code_update(); |
989 | ||
6dc3064a DG |
990 | ret = consumer_send_status_msg(sock, ret_code); |
991 | if (ret < 0) { | |
992 | /* Somehow, the session daemon is not responding anymore. */ | |
993 | goto end_nosignal; | |
994 | } | |
995 | break; | |
996 | } | |
07b86b52 JD |
997 | case LTTNG_CONSUMER_DESTROY_CHANNEL: |
998 | { | |
999 | uint64_t key = msg.u.destroy_channel.key; | |
1000 | struct lttng_consumer_channel *channel; | |
1001 | ||
1002 | channel = consumer_find_channel(key); | |
1003 | if (!channel) { | |
1004 | ERR("Kernel consumer destroy channel %" PRIu64 " not found", key); | |
e462382a | 1005 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; |
07b86b52 JD |
1006 | } |
1007 | ||
9ce5646a MD |
1008 | health_code_update(); |
1009 | ||
07b86b52 JD |
1010 | ret = consumer_send_status_msg(sock, ret_code); |
1011 | if (ret < 0) { | |
1012 | /* Somehow, the session daemon is not responding anymore. */ | |
09d3dbc0 | 1013 | goto end_destroy_channel; |
07b86b52 JD |
1014 | } |
1015 | ||
9ce5646a MD |
1016 | health_code_update(); |
1017 | ||
15dc512a DG |
1018 | /* Stop right now if no channel was found. */ |
1019 | if (!channel) { | |
09d3dbc0 | 1020 | goto end_destroy_channel; |
15dc512a DG |
1021 | } |
1022 | ||
07b86b52 JD |
1023 | /* |
1024 | * This command should ONLY be issued for channel with streams set in | |
1025 | * no monitor mode. | |
1026 | */ | |
1027 | assert(!channel->monitor); | |
1028 | ||
1029 | /* | |
1030 | * The refcount should ALWAYS be 0 in the case of a channel in no | |
1031 | * monitor mode. | |
1032 | */ | |
1033 | assert(!uatomic_sub_return(&channel->refcount, 1)); | |
1034 | ||
1035 | consumer_del_channel(channel); | |
09d3dbc0 | 1036 | end_destroy_channel: |
07b86b52 JD |
1037 | goto end_nosignal; |
1038 | } | |
fb83fe64 JD |
1039 | case LTTNG_CONSUMER_DISCARDED_EVENTS: |
1040 | { | |
66ab32be JD |
1041 | ssize_t ret; |
1042 | uint64_t count; | |
fb83fe64 JD |
1043 | struct lttng_consumer_channel *channel; |
1044 | uint64_t id = msg.u.discarded_events.session_id; | |
1045 | uint64_t key = msg.u.discarded_events.channel_key; | |
1046 | ||
e5742757 MD |
1047 | DBG("Kernel consumer discarded events command for session id %" |
1048 | PRIu64 ", channel key %" PRIu64, id, key); | |
1049 | ||
fb83fe64 JD |
1050 | channel = consumer_find_channel(key); |
1051 | if (!channel) { | |
1052 | ERR("Kernel consumer discarded events channel %" | |
1053 | PRIu64 " not found", key); | |
66ab32be | 1054 | count = 0; |
e5742757 | 1055 | } else { |
66ab32be | 1056 | count = channel->discarded_events; |
fb83fe64 JD |
1057 | } |
1058 | ||
fb83fe64 JD |
1059 | health_code_update(); |
1060 | ||
1061 | /* Send back returned value to session daemon */ | |
66ab32be | 1062 | ret = lttcomm_send_unix_sock(sock, &count, sizeof(count)); |
fb83fe64 JD |
1063 | if (ret < 0) { |
1064 | PERROR("send discarded events"); | |
1065 | goto error_fatal; | |
1066 | } | |
1067 | ||
1068 | break; | |
1069 | } | |
1070 | case LTTNG_CONSUMER_LOST_PACKETS: | |
1071 | { | |
66ab32be JD |
1072 | ssize_t ret; |
1073 | uint64_t count; | |
fb83fe64 JD |
1074 | struct lttng_consumer_channel *channel; |
1075 | uint64_t id = msg.u.lost_packets.session_id; | |
1076 | uint64_t key = msg.u.lost_packets.channel_key; | |
1077 | ||
e5742757 MD |
1078 | DBG("Kernel consumer lost packets command for session id %" |
1079 | PRIu64 ", channel key %" PRIu64, id, key); | |
1080 | ||
fb83fe64 JD |
1081 | channel = consumer_find_channel(key); |
1082 | if (!channel) { | |
1083 | ERR("Kernel consumer lost packets channel %" | |
1084 | PRIu64 " not found", key); | |
66ab32be | 1085 | count = 0; |
e5742757 | 1086 | } else { |
66ab32be | 1087 | count = channel->lost_packets; |
fb83fe64 JD |
1088 | } |
1089 | ||
fb83fe64 JD |
1090 | health_code_update(); |
1091 | ||
1092 | /* Send back returned value to session daemon */ | |
66ab32be | 1093 | ret = lttcomm_send_unix_sock(sock, &count, sizeof(count)); |
fb83fe64 JD |
1094 | if (ret < 0) { |
1095 | PERROR("send lost packets"); | |
1096 | goto error_fatal; | |
1097 | } | |
1098 | ||
1099 | break; | |
1100 | } | |
b3530820 JG |
1101 | case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE: |
1102 | { | |
1103 | int channel_monitor_pipe; | |
1104 | ||
1105 | ret_code = LTTCOMM_CONSUMERD_SUCCESS; | |
1106 | /* Successfully received the command's type. */ | |
1107 | ret = consumer_send_status_msg(sock, ret_code); | |
1108 | if (ret < 0) { | |
1109 | goto error_fatal; | |
1110 | } | |
1111 | ||
1112 | ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe, | |
1113 | 1); | |
1114 | if (ret != sizeof(channel_monitor_pipe)) { | |
1115 | ERR("Failed to receive channel monitor pipe"); | |
1116 | goto error_fatal; | |
1117 | } | |
1118 | ||
1119 | DBG("Received channel monitor pipe (%d)", channel_monitor_pipe); | |
1120 | ret = consumer_timer_thread_set_channel_monitor_pipe( | |
1121 | channel_monitor_pipe); | |
1122 | if (!ret) { | |
1123 | int flags; | |
1124 | ||
1125 | ret_code = LTTCOMM_CONSUMERD_SUCCESS; | |
1126 | /* Set the pipe as non-blocking. */ | |
1127 | ret = fcntl(channel_monitor_pipe, F_GETFL, 0); | |
1128 | if (ret == -1) { | |
1129 | PERROR("fcntl get flags of the channel monitoring pipe"); | |
1130 | goto error_fatal; | |
1131 | } | |
1132 | flags = ret; | |
1133 | ||
1134 | ret = fcntl(channel_monitor_pipe, F_SETFL, | |
1135 | flags | O_NONBLOCK); | |
1136 | if (ret == -1) { | |
1137 | PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe"); | |
1138 | goto error_fatal; | |
1139 | } | |
1140 | DBG("Channel monitor pipe set as non-blocking"); | |
1141 | } else { | |
1142 | ret_code = LTTCOMM_CONSUMERD_ALREADY_SET; | |
1143 | } | |
1144 | ret = consumer_send_status_msg(sock, ret_code); | |
1145 | if (ret < 0) { | |
1146 | goto error_fatal; | |
1147 | } | |
1148 | break; | |
1149 | } | |
b99a8d42 JD |
1150 | case LTTNG_CONSUMER_ROTATE_CHANNEL: |
1151 | { | |
e96d66b4 MD |
1152 | struct lttng_consumer_channel *channel; |
1153 | uint64_t key = msg.u.rotate_channel.key; | |
b99a8d42 | 1154 | |
e96d66b4 | 1155 | DBG("Consumer rotate channel %" PRIu64, key); |
b99a8d42 | 1156 | |
e96d66b4 MD |
1157 | channel = consumer_find_channel(key); |
1158 | if (!channel) { | |
1159 | ERR("Channel %" PRIu64 " not found", key); | |
1160 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; | |
1161 | } else { | |
1162 | /* | |
1163 | * Sample the rotate position of all the streams in this channel. | |
1164 | */ | |
1165 | ret = lttng_consumer_rotate_channel(channel, key, | |
e96d66b4 MD |
1166 | msg.u.rotate_channel.relayd_id, |
1167 | msg.u.rotate_channel.metadata, | |
e96d66b4 MD |
1168 | ctx); |
1169 | if (ret < 0) { | |
1170 | ERR("Rotate channel failed"); | |
1171 | ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL; | |
1172 | } | |
b99a8d42 | 1173 | |
e96d66b4 MD |
1174 | health_code_update(); |
1175 | } | |
b99a8d42 JD |
1176 | ret = consumer_send_status_msg(sock, ret_code); |
1177 | if (ret < 0) { | |
1178 | /* Somehow, the session daemon is not responding anymore. */ | |
812e4d26 | 1179 | goto error_rotate_channel; |
b99a8d42 | 1180 | } |
e96d66b4 MD |
1181 | if (channel) { |
1182 | /* Rotate the streams that are ready right now. */ | |
1183 | ret = lttng_consumer_rotate_ready_streams( | |
1184 | channel, key, ctx); | |
1185 | if (ret < 0) { | |
1186 | ERR("Rotate ready streams failed"); | |
1187 | } | |
b99a8d42 | 1188 | } |
b99a8d42 | 1189 | break; |
812e4d26 JG |
1190 | error_rotate_channel: |
1191 | goto end_nosignal; | |
b99a8d42 | 1192 | } |
e5148e25 | 1193 | case LTTNG_CONSUMER_INIT: |
00fb02ac | 1194 | { |
e5148e25 JG |
1195 | ret_code = lttng_consumer_init_command(ctx, |
1196 | msg.u.init.sessiond_uuid); | |
00fb02ac | 1197 | health_code_update(); |
00fb02ac JD |
1198 | ret = consumer_send_status_msg(sock, ret_code); |
1199 | if (ret < 0) { | |
1200 | /* Somehow, the session daemon is not responding anymore. */ | |
1201 | goto end_nosignal; | |
1202 | } | |
1203 | break; | |
1204 | } | |
e5148e25 | 1205 | case LTTNG_CONSUMER_CREATE_TRACE_CHUNK: |
d88744a4 | 1206 | { |
e5148e25 | 1207 | const struct lttng_credentials credentials = { |
0ebdafe0 JG |
1208 | .uid = msg.u.create_trace_chunk.credentials.value.uid, |
1209 | .gid = msg.u.create_trace_chunk.credentials.value.gid, | |
e5148e25 JG |
1210 | }; |
1211 | const bool is_local_trace = | |
1212 | !msg.u.create_trace_chunk.relayd_id.is_set; | |
1213 | const uint64_t relayd_id = | |
1214 | msg.u.create_trace_chunk.relayd_id.value; | |
1215 | const char *chunk_override_name = | |
1216 | *msg.u.create_trace_chunk.override_name ? | |
1217 | msg.u.create_trace_chunk.override_name : | |
1218 | NULL; | |
1219 | LTTNG_OPTIONAL(struct lttng_directory_handle) chunk_directory_handle = | |
1220 | LTTNG_OPTIONAL_INIT; | |
d88744a4 | 1221 | |
e5148e25 JG |
1222 | /* |
1223 | * The session daemon will only provide a chunk directory file | |
1224 | * descriptor for local traces. | |
1225 | */ | |
1226 | if (is_local_trace) { | |
1227 | int chunk_dirfd; | |
19990ed5 | 1228 | |
e5148e25 JG |
1229 | /* Acnowledge the reception of the command. */ |
1230 | ret = consumer_send_status_msg(sock, | |
1231 | LTTCOMM_CONSUMERD_SUCCESS); | |
1232 | if (ret < 0) { | |
1233 | /* Somehow, the session daemon is not responding anymore. */ | |
1234 | goto end_nosignal; | |
1235 | } | |
82528808 | 1236 | |
e5148e25 JG |
1237 | ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1); |
1238 | if (ret != sizeof(chunk_dirfd)) { | |
1239 | ERR("Failed to receive trace chunk directory file descriptor"); | |
1240 | goto error_fatal; | |
1241 | } | |
82528808 | 1242 | |
e5148e25 JG |
1243 | DBG("Received trace chunk directory fd (%d)", |
1244 | chunk_dirfd); | |
1245 | ret = lttng_directory_handle_init_from_dirfd( | |
1246 | &chunk_directory_handle.value, | |
1247 | chunk_dirfd); | |
1248 | if (ret) { | |
1249 | ERR("Failed to initialize chunk directory handle from directory file descriptor"); | |
1250 | if (close(chunk_dirfd)) { | |
1251 | PERROR("Failed to close chunk directory file descriptor"); | |
1252 | } | |
1253 | goto error_fatal; | |
1254 | } | |
1255 | chunk_directory_handle.is_set = true; | |
82528808 JG |
1256 | } |
1257 | ||
e5148e25 JG |
1258 | ret_code = lttng_consumer_create_trace_chunk( |
1259 | !is_local_trace ? &relayd_id : NULL, | |
1260 | msg.u.create_trace_chunk.session_id, | |
1261 | msg.u.create_trace_chunk.chunk_id, | |
0ebdafe0 JG |
1262 | (time_t) msg.u.create_trace_chunk |
1263 | .creation_timestamp, | |
e5148e25 | 1264 | chunk_override_name, |
0ebdafe0 JG |
1265 | msg.u.create_trace_chunk.credentials.is_set ? |
1266 | &credentials : | |
1267 | NULL, | |
e5148e25 JG |
1268 | chunk_directory_handle.is_set ? |
1269 | &chunk_directory_handle.value : | |
1270 | NULL); | |
82528808 | 1271 | |
e5148e25 JG |
1272 | if (chunk_directory_handle.is_set) { |
1273 | lttng_directory_handle_fini( | |
1274 | &chunk_directory_handle.value); | |
d88744a4 | 1275 | } |
e5148e25 | 1276 | goto end_msg_sessiond; |
d88744a4 | 1277 | } |
e5148e25 | 1278 | case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK: |
a1ae2ea5 | 1279 | { |
6bbcff33 JG |
1280 | enum lttng_trace_chunk_command_type close_command = |
1281 | msg.u.close_trace_chunk.close_command.value; | |
e5148e25 JG |
1282 | const uint64_t relayd_id = |
1283 | msg.u.close_trace_chunk.relayd_id.value; | |
41b23598 MD |
1284 | struct lttcomm_consumer_close_trace_chunk_reply reply; |
1285 | char path[LTTNG_PATH_MAX]; | |
e5148e25 JG |
1286 | |
1287 | ret_code = lttng_consumer_close_trace_chunk( | |
1288 | msg.u.close_trace_chunk.relayd_id.is_set ? | |
6bbcff33 JG |
1289 | &relayd_id : |
1290 | NULL, | |
e5148e25 JG |
1291 | msg.u.close_trace_chunk.session_id, |
1292 | msg.u.close_trace_chunk.chunk_id, | |
6bbcff33 JG |
1293 | (time_t) msg.u.close_trace_chunk.close_timestamp, |
1294 | msg.u.close_trace_chunk.close_command.is_set ? | |
1295 | &close_command : | |
41b23598 MD |
1296 | NULL, path); |
1297 | reply.ret_code = ret_code; | |
1298 | reply.path_length = strlen(path) + 1; | |
1299 | ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply)); | |
1300 | if (ret != sizeof(reply)) { | |
1301 | goto error_fatal; | |
1302 | } | |
1303 | ret = lttcomm_send_unix_sock(sock, path, reply.path_length); | |
1304 | if (ret != reply.path_length) { | |
1305 | goto error_fatal; | |
1306 | } | |
1307 | goto end_nosignal; | |
fc181d72 | 1308 | } |
e5148e25 | 1309 | case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS: |
fc181d72 | 1310 | { |
e5148e25 JG |
1311 | const uint64_t relayd_id = |
1312 | msg.u.trace_chunk_exists.relayd_id.value; | |
1313 | ||
1314 | ret_code = lttng_consumer_trace_chunk_exists( | |
1315 | msg.u.trace_chunk_exists.relayd_id.is_set ? | |
1316 | &relayd_id : NULL, | |
1317 | msg.u.trace_chunk_exists.session_id, | |
1318 | msg.u.trace_chunk_exists.chunk_id); | |
1319 | goto end_msg_sessiond; | |
a1ae2ea5 | 1320 | } |
3bd1e081 | 1321 | default: |
3f8e211f | 1322 | goto end_nosignal; |
3bd1e081 | 1323 | } |
3f8e211f | 1324 | |
3bd1e081 | 1325 | end_nosignal: |
4cbc1a04 DG |
1326 | /* |
1327 | * Return 1 to indicate success since the 0 value can be a socket | |
1328 | * shutdown during the recv() or send() call. | |
1329 | */ | |
e0e7757d JG |
1330 | ret = 1; |
1331 | goto end; | |
1332 | error_fatal: | |
1333 | /* This will issue a consumer stop. */ | |
1334 | ret = -1; | |
1335 | goto end; | |
e5148e25 JG |
1336 | end_msg_sessiond: |
1337 | /* | |
1338 | * The returned value here is not useful since either way we'll return 1 to | |
1339 | * the caller because the session daemon socket management is done | |
1340 | * elsewhere. Returning a negative code or 0 will shutdown the consumer. | |
1341 | */ | |
1342 | ret = consumer_send_status_msg(sock, ret_code); | |
1343 | if (ret < 0) { | |
1344 | goto error_fatal; | |
1345 | } | |
e0e7757d JG |
1346 | ret = 1; |
1347 | end: | |
e5148e25 | 1348 | health_code_update(); |
1803a064 | 1349 | rcu_read_unlock(); |
e0e7757d | 1350 | return ret; |
3bd1e081 | 1351 | } |
d41f73b7 | 1352 | |
309167d2 JD |
1353 | /* |
1354 | * Populate index values of a kernel stream. Values are set in big endian order. | |
1355 | * | |
1356 | * Return 0 on success or else a negative value. | |
1357 | */ | |
50adc264 | 1358 | static int get_index_values(struct ctf_packet_index *index, int infd) |
309167d2 JD |
1359 | { |
1360 | int ret; | |
641e8873 JG |
1361 | uint64_t packet_size, content_size, timestamp_begin, timestamp_end, |
1362 | events_discarded, stream_id, stream_instance_id, | |
1363 | packet_seq_num; | |
309167d2 | 1364 | |
641e8873 | 1365 | ret = kernctl_get_timestamp_begin(infd, ×tamp_begin); |
309167d2 JD |
1366 | if (ret < 0) { |
1367 | PERROR("kernctl_get_timestamp_begin"); | |
1368 | goto error; | |
1369 | } | |
309167d2 | 1370 | |
641e8873 | 1371 | ret = kernctl_get_timestamp_end(infd, ×tamp_end); |
309167d2 JD |
1372 | if (ret < 0) { |
1373 | PERROR("kernctl_get_timestamp_end"); | |
1374 | goto error; | |
1375 | } | |
309167d2 | 1376 | |
641e8873 | 1377 | ret = kernctl_get_events_discarded(infd, &events_discarded); |
309167d2 JD |
1378 | if (ret < 0) { |
1379 | PERROR("kernctl_get_events_discarded"); | |
1380 | goto error; | |
1381 | } | |
309167d2 | 1382 | |
641e8873 | 1383 | ret = kernctl_get_content_size(infd, &content_size); |
309167d2 JD |
1384 | if (ret < 0) { |
1385 | PERROR("kernctl_get_content_size"); | |
1386 | goto error; | |
1387 | } | |
309167d2 | 1388 | |
641e8873 | 1389 | ret = kernctl_get_packet_size(infd, &packet_size); |
309167d2 JD |
1390 | if (ret < 0) { |
1391 | PERROR("kernctl_get_packet_size"); | |
1392 | goto error; | |
1393 | } | |
309167d2 | 1394 | |
641e8873 | 1395 | ret = kernctl_get_stream_id(infd, &stream_id); |
309167d2 JD |
1396 | if (ret < 0) { |
1397 | PERROR("kernctl_get_stream_id"); | |
1398 | goto error; | |
1399 | } | |
309167d2 | 1400 | |
641e8873 | 1401 | ret = kernctl_get_instance_id(infd, &stream_instance_id); |
234cd636 | 1402 | if (ret < 0) { |
f0b03c22 MD |
1403 | if (ret == -ENOTTY) { |
1404 | /* Command not implemented by lttng-modules. */ | |
641e8873 | 1405 | stream_instance_id = -1ULL; |
f0b03c22 MD |
1406 | } else { |
1407 | PERROR("kernctl_get_instance_id"); | |
1408 | goto error; | |
1409 | } | |
234cd636 | 1410 | } |
234cd636 | 1411 | |
641e8873 | 1412 | ret = kernctl_get_sequence_number(infd, &packet_seq_num); |
234cd636 | 1413 | if (ret < 0) { |
f0b03c22 MD |
1414 | if (ret == -ENOTTY) { |
1415 | /* Command not implemented by lttng-modules. */ | |
641e8873 | 1416 | packet_seq_num = -1ULL; |
f0b03c22 MD |
1417 | ret = 0; |
1418 | } else { | |
1419 | PERROR("kernctl_get_sequence_number"); | |
1420 | goto error; | |
1421 | } | |
234cd636 JD |
1422 | } |
1423 | index->packet_seq_num = htobe64(index->packet_seq_num); | |
1424 | ||
641e8873 JG |
1425 | *index = (typeof(*index)) { |
1426 | .offset = index->offset, | |
1427 | .packet_size = htobe64(packet_size), | |
1428 | .content_size = htobe64(content_size), | |
1429 | .timestamp_begin = htobe64(timestamp_begin), | |
1430 | .timestamp_end = htobe64(timestamp_end), | |
1431 | .events_discarded = htobe64(events_discarded), | |
1432 | .stream_id = htobe64(stream_id), | |
1433 | .stream_instance_id = htobe64(stream_instance_id), | |
1434 | .packet_seq_num = htobe64(packet_seq_num), | |
1435 | }; | |
1436 | ||
309167d2 JD |
1437 | error: |
1438 | return ret; | |
1439 | } | |
94d49140 JD |
1440 | /* |
1441 | * Sync metadata meaning request them to the session daemon and snapshot to the | |
1442 | * metadata thread can consumer them. | |
1443 | * | |
1444 | * Metadata stream lock MUST be acquired. | |
1445 | * | |
1446 | * Return 0 if new metadatda is available, EAGAIN if the metadata stream | |
1447 | * is empty or a negative value on error. | |
1448 | */ | |
1449 | int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata) | |
1450 | { | |
1451 | int ret; | |
1452 | ||
1453 | assert(metadata); | |
1454 | ||
1455 | ret = kernctl_buffer_flush(metadata->wait_fd); | |
1456 | if (ret < 0) { | |
1457 | ERR("Failed to flush kernel stream"); | |
1458 | goto end; | |
1459 | } | |
1460 | ||
1461 | ret = kernctl_snapshot(metadata->wait_fd); | |
1462 | if (ret < 0) { | |
32af2c95 | 1463 | if (ret != -EAGAIN) { |
94d49140 JD |
1464 | ERR("Sync metadata, taking kernel snapshot failed."); |
1465 | goto end; | |
1466 | } | |
1467 | DBG("Sync metadata, no new kernel metadata"); | |
1468 | /* No new metadata, exit. */ | |
1469 | ret = ENODATA; | |
1470 | goto end; | |
1471 | } | |
1472 | ||
1473 | end: | |
1474 | return ret; | |
1475 | } | |
309167d2 | 1476 | |
fb83fe64 JD |
1477 | static |
1478 | int update_stream_stats(struct lttng_consumer_stream *stream) | |
1479 | { | |
1480 | int ret; | |
1481 | uint64_t seq, discarded; | |
1482 | ||
1483 | ret = kernctl_get_sequence_number(stream->wait_fd, &seq); | |
1484 | if (ret < 0) { | |
f0b03c22 MD |
1485 | if (ret == -ENOTTY) { |
1486 | /* Command not implemented by lttng-modules. */ | |
1487 | seq = -1ULL; | |
c8eabe73 | 1488 | stream->sequence_number_unavailable = true; |
f0b03c22 MD |
1489 | } else { |
1490 | PERROR("kernctl_get_sequence_number"); | |
1491 | goto end; | |
1492 | } | |
fb83fe64 JD |
1493 | } |
1494 | ||
1495 | /* | |
1496 | * Start the sequence when we extract the first packet in case we don't | |
1497 | * start at 0 (for example if a consumer is not connected to the | |
1498 | * session immediately after the beginning). | |
1499 | */ | |
1500 | if (stream->last_sequence_number == -1ULL) { | |
1501 | stream->last_sequence_number = seq; | |
1502 | } else if (seq > stream->last_sequence_number) { | |
1503 | stream->chan->lost_packets += seq - | |
1504 | stream->last_sequence_number - 1; | |
1505 | } else { | |
1506 | /* seq <= last_sequence_number */ | |
1507 | ERR("Sequence number inconsistent : prev = %" PRIu64 | |
1508 | ", current = %" PRIu64, | |
1509 | stream->last_sequence_number, seq); | |
1510 | ret = -1; | |
1511 | goto end; | |
1512 | } | |
1513 | stream->last_sequence_number = seq; | |
1514 | ||
1515 | ret = kernctl_get_events_discarded(stream->wait_fd, &discarded); | |
1516 | if (ret < 0) { | |
1517 | PERROR("kernctl_get_events_discarded"); | |
1518 | goto end; | |
1519 | } | |
1520 | if (discarded < stream->last_discarded_events) { | |
1521 | /* | |
83f4233d MJ |
1522 | * Overflow has occurred. We assume only one wrap-around |
1523 | * has occurred. | |
fb83fe64 JD |
1524 | */ |
1525 | stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) - | |
1526 | stream->last_discarded_events + discarded; | |
1527 | } else { | |
1528 | stream->chan->discarded_events += discarded - | |
1529 | stream->last_discarded_events; | |
1530 | } | |
1531 | stream->last_discarded_events = discarded; | |
1532 | ret = 0; | |
1533 | ||
1534 | end: | |
1535 | return ret; | |
1536 | } | |
1537 | ||
93ec662e JD |
1538 | /* |
1539 | * Check if the local version of the metadata stream matches with the version | |
1540 | * of the metadata stream in the kernel. If it was updated, set the reset flag | |
1541 | * on the stream. | |
1542 | */ | |
1543 | static | |
1544 | int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream) | |
1545 | { | |
1546 | int ret; | |
1547 | uint64_t cur_version; | |
1548 | ||
1549 | ret = kernctl_get_metadata_version(infd, &cur_version); | |
1550 | if (ret < 0) { | |
f0b03c22 MD |
1551 | if (ret == -ENOTTY) { |
1552 | /* | |
1553 | * LTTng-modules does not implement this | |
1554 | * command. | |
1555 | */ | |
1556 | ret = 0; | |
1557 | goto end; | |
1558 | } | |
93ec662e JD |
1559 | ERR("Failed to get the metadata version"); |
1560 | goto end; | |
1561 | } | |
1562 | ||
1563 | if (stream->metadata_version == cur_version) { | |
1564 | ret = 0; | |
1565 | goto end; | |
1566 | } | |
1567 | ||
1568 | DBG("New metadata version detected"); | |
1569 | stream->metadata_version = cur_version; | |
1570 | stream->reset_metadata_flag = 1; | |
1571 | ret = 0; | |
1572 | ||
1573 | end: | |
1574 | return ret; | |
1575 | } | |
1576 | ||
d41f73b7 MD |
1577 | /* |
1578 | * Consume data on a file descriptor and write it on a trace file. | |
e5148e25 | 1579 | * The stream and channel locks must be held by the caller. |
d41f73b7 | 1580 | */ |
4078b776 | 1581 | ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, |
e5148e25 | 1582 | struct lttng_consumer_local_data *ctx) |
d41f73b7 | 1583 | { |
1d4dfdef | 1584 | unsigned long len, subbuf_size, padding; |
02d02e31 | 1585 | int err, write_index = 1, rotation_ret; |
4078b776 | 1586 | ssize_t ret = 0; |
d41f73b7 | 1587 | int infd = stream->wait_fd; |
b2d8a465 | 1588 | struct ctf_packet_index index = {}; |
d41f73b7 MD |
1589 | |
1590 | DBG("In read_subbuffer (infd : %d)", infd); | |
309167d2 | 1591 | |
02d02e31 JD |
1592 | /* |
1593 | * If the stream was flagged to be ready for rotation before we extract the | |
1594 | * next packet, rotate it now. | |
1595 | */ | |
1596 | if (stream->rotate_ready) { | |
1597 | DBG("Rotate stream before extracting data"); | |
e5148e25 | 1598 | rotation_ret = lttng_consumer_rotate_stream(ctx, stream); |
02d02e31 JD |
1599 | if (rotation_ret < 0) { |
1600 | ERR("Stream rotation error"); | |
1601 | ret = -1; | |
1602 | goto error; | |
1603 | } | |
1604 | } | |
1605 | ||
d41f73b7 MD |
1606 | /* Get the next subbuffer */ |
1607 | err = kernctl_get_next_subbuf(infd); | |
1608 | if (err != 0) { | |
d41f73b7 MD |
1609 | /* |
1610 | * This is a debug message even for single-threaded consumer, | |
1611 | * because poll() have more relaxed criterions than get subbuf, | |
1612 | * so get_subbuf may fail for short race windows where poll() | |
1613 | * would issue wakeups. | |
1614 | */ | |
1615 | DBG("Reserving sub buffer failed (everything is normal, " | |
1616 | "it is due to concurrency)"); | |
32af2c95 | 1617 | ret = err; |
02d02e31 | 1618 | goto error; |
d41f73b7 MD |
1619 | } |
1620 | ||
1d4dfdef DG |
1621 | /* Get the full subbuffer size including padding */ |
1622 | err = kernctl_get_padded_subbuf_size(infd, &len); | |
1623 | if (err != 0) { | |
5a510c9f | 1624 | PERROR("Getting sub-buffer len failed."); |
8265f19e MD |
1625 | err = kernctl_put_subbuf(infd); |
1626 | if (err != 0) { | |
32af2c95 | 1627 | if (err == -EFAULT) { |
5a510c9f | 1628 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1629 | } else if (err == -EIO) { |
8265f19e | 1630 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1631 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
8265f19e | 1632 | } |
32af2c95 | 1633 | ret = err; |
02d02e31 | 1634 | goto error; |
8265f19e | 1635 | } |
32af2c95 | 1636 | ret = err; |
02d02e31 | 1637 | goto error; |
1d4dfdef DG |
1638 | } |
1639 | ||
1c20f0e2 | 1640 | if (!stream->metadata_flag) { |
309167d2 JD |
1641 | ret = get_index_values(&index, infd); |
1642 | if (ret < 0) { | |
8265f19e MD |
1643 | err = kernctl_put_subbuf(infd); |
1644 | if (err != 0) { | |
32af2c95 | 1645 | if (err == -EFAULT) { |
5a510c9f | 1646 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1647 | } else if (err == -EIO) { |
8265f19e | 1648 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1649 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
8265f19e | 1650 | } |
32af2c95 | 1651 | ret = err; |
02d02e31 | 1652 | goto error; |
8265f19e | 1653 | } |
02d02e31 | 1654 | goto error; |
309167d2 | 1655 | } |
fb83fe64 JD |
1656 | ret = update_stream_stats(stream); |
1657 | if (ret < 0) { | |
7b87473d MD |
1658 | err = kernctl_put_subbuf(infd); |
1659 | if (err != 0) { | |
1660 | if (err == -EFAULT) { | |
1661 | PERROR("Error in unreserving sub buffer\n"); | |
1662 | } else if (err == -EIO) { | |
1663 | /* Should never happen with newer LTTng versions */ | |
1664 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); | |
1665 | } | |
1666 | ret = err; | |
02d02e31 | 1667 | goto error; |
7b87473d | 1668 | } |
02d02e31 | 1669 | goto error; |
fb83fe64 | 1670 | } |
1c20f0e2 JD |
1671 | } else { |
1672 | write_index = 0; | |
93ec662e JD |
1673 | ret = metadata_stream_check_version(infd, stream); |
1674 | if (ret < 0) { | |
7b87473d MD |
1675 | err = kernctl_put_subbuf(infd); |
1676 | if (err != 0) { | |
1677 | if (err == -EFAULT) { | |
1678 | PERROR("Error in unreserving sub buffer\n"); | |
1679 | } else if (err == -EIO) { | |
1680 | /* Should never happen with newer LTTng versions */ | |
1681 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); | |
1682 | } | |
1683 | ret = err; | |
02d02e31 | 1684 | goto error; |
7b87473d | 1685 | } |
02d02e31 | 1686 | goto error; |
93ec662e | 1687 | } |
309167d2 JD |
1688 | } |
1689 | ||
ffe60014 | 1690 | switch (stream->chan->output) { |
07b86b52 | 1691 | case CONSUMER_CHANNEL_SPLICE: |
1d4dfdef DG |
1692 | /* |
1693 | * XXX: The lttng-modules splice "actor" does not handle copying | |
1694 | * partial pages hence only using the subbuffer size without the | |
1695 | * padding makes the splice fail. | |
1696 | */ | |
1697 | subbuf_size = len; | |
1698 | padding = 0; | |
1699 | ||
1700 | /* splice the subbuffer to the tracefile */ | |
91dfef6e | 1701 | ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size, |
309167d2 | 1702 | padding, &index); |
91dfef6e DG |
1703 | /* |
1704 | * XXX: Splice does not support network streaming so the return value | |
1705 | * is simply checked against subbuf_size and not like the mmap() op. | |
1706 | */ | |
1d4dfdef DG |
1707 | if (ret != subbuf_size) { |
1708 | /* | |
1709 | * display the error but continue processing to try | |
1710 | * to release the subbuffer | |
1711 | */ | |
1712 | ERR("Error splicing to tracefile (ret: %zd != len: %lu)", | |
1713 | ret, subbuf_size); | |
309167d2 | 1714 | write_index = 0; |
1d4dfdef DG |
1715 | } |
1716 | break; | |
07b86b52 | 1717 | case CONSUMER_CHANNEL_MMAP: |
7775df52 JG |
1718 | { |
1719 | const char *subbuf_addr; | |
b770aa7f | 1720 | struct lttng_buffer_view subbuf_view; |
7775df52 | 1721 | |
1d4dfdef DG |
1722 | /* Get subbuffer size without padding */ |
1723 | err = kernctl_get_subbuf_size(infd, &subbuf_size); | |
1724 | if (err != 0) { | |
5a510c9f | 1725 | PERROR("Getting sub-buffer len failed."); |
8265f19e MD |
1726 | err = kernctl_put_subbuf(infd); |
1727 | if (err != 0) { | |
32af2c95 | 1728 | if (err == -EFAULT) { |
5a510c9f | 1729 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1730 | } else if (err == -EIO) { |
8265f19e | 1731 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1732 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
8265f19e | 1733 | } |
32af2c95 | 1734 | ret = err; |
02d02e31 | 1735 | goto error; |
8265f19e | 1736 | } |
32af2c95 | 1737 | ret = err; |
02d02e31 | 1738 | goto error; |
1d4dfdef | 1739 | } |
47e81c02 | 1740 | |
7775df52 JG |
1741 | ret = get_current_subbuf_addr(stream, &subbuf_addr); |
1742 | if (ret) { | |
1743 | goto error_put_subbuf; | |
1744 | } | |
1745 | ||
1d4dfdef DG |
1746 | /* Make sure the tracer is not gone mad on us! */ |
1747 | assert(len >= subbuf_size); | |
1748 | ||
1749 | padding = len - subbuf_size; | |
1750 | ||
b770aa7f JG |
1751 | subbuf_view = lttng_buffer_view_init(subbuf_addr, 0, len); |
1752 | ||
1d4dfdef | 1753 | /* write the subbuffer to the tracefile */ |
b770aa7f JG |
1754 | ret = lttng_consumer_on_read_subbuffer_mmap( |
1755 | ctx, stream, &subbuf_view, padding, &index); | |
91dfef6e | 1756 | /* |
b770aa7f JG |
1757 | * The mmap operation should write subbuf_size amount of data |
1758 | * when network streaming or the full padding (len) size when we | |
1759 | * are _not_ streaming. | |
91dfef6e | 1760 | */ |
d88aee68 DG |
1761 | if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) || |
1762 | (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) { | |
1d4dfdef | 1763 | /* |
91dfef6e | 1764 | * Display the error but continue processing to try to release the |
2336629e DG |
1765 | * subbuffer. This is a DBG statement since this is possible to |
1766 | * happen without being a critical error. | |
1d4dfdef | 1767 | */ |
2336629e | 1768 | DBG("Error writing to tracefile " |
91dfef6e DG |
1769 | "(ret: %zd != len: %lu != subbuf_size: %lu)", |
1770 | ret, len, subbuf_size); | |
309167d2 | 1771 | write_index = 0; |
1d4dfdef DG |
1772 | } |
1773 | break; | |
7775df52 | 1774 | } |
1d4dfdef DG |
1775 | default: |
1776 | ERR("Unknown output method"); | |
56591bac | 1777 | ret = -EPERM; |
d41f73b7 | 1778 | } |
7775df52 | 1779 | error_put_subbuf: |
d41f73b7 MD |
1780 | err = kernctl_put_next_subbuf(infd); |
1781 | if (err != 0) { | |
32af2c95 | 1782 | if (err == -EFAULT) { |
5a510c9f | 1783 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1784 | } else if (err == -EIO) { |
d41f73b7 | 1785 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1786 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
d41f73b7 | 1787 | } |
32af2c95 | 1788 | ret = err; |
02d02e31 | 1789 | goto error; |
d41f73b7 MD |
1790 | } |
1791 | ||
309167d2 | 1792 | /* Write index if needed. */ |
1c20f0e2 | 1793 | if (!write_index) { |
02d02e31 | 1794 | goto rotate; |
1c20f0e2 JD |
1795 | } |
1796 | ||
94d49140 JD |
1797 | if (stream->chan->live_timer_interval && !stream->metadata_flag) { |
1798 | /* | |
1799 | * In live, block until all the metadata is sent. | |
1800 | */ | |
c585821b MD |
1801 | pthread_mutex_lock(&stream->metadata_timer_lock); |
1802 | assert(!stream->missed_metadata_flush); | |
1803 | stream->waiting_on_metadata = true; | |
1804 | pthread_mutex_unlock(&stream->metadata_timer_lock); | |
1805 | ||
94d49140 | 1806 | err = consumer_stream_sync_metadata(ctx, stream->session_id); |
c585821b MD |
1807 | |
1808 | pthread_mutex_lock(&stream->metadata_timer_lock); | |
1809 | stream->waiting_on_metadata = false; | |
1810 | if (stream->missed_metadata_flush) { | |
1811 | stream->missed_metadata_flush = false; | |
1812 | pthread_mutex_unlock(&stream->metadata_timer_lock); | |
1813 | (void) consumer_flush_kernel_index(stream); | |
1814 | } else { | |
1815 | pthread_mutex_unlock(&stream->metadata_timer_lock); | |
1816 | } | |
94d49140 | 1817 | if (err < 0) { |
02d02e31 | 1818 | goto error; |
94d49140 JD |
1819 | } |
1820 | } | |
1821 | ||
1c20f0e2 JD |
1822 | err = consumer_stream_write_index(stream, &index); |
1823 | if (err < 0) { | |
02d02e31 | 1824 | goto error; |
309167d2 JD |
1825 | } |
1826 | ||
02d02e31 JD |
1827 | rotate: |
1828 | /* | |
1829 | * After extracting the packet, we check if the stream is now ready to be | |
1830 | * rotated and perform the action immediately. | |
1831 | */ | |
1832 | rotation_ret = lttng_consumer_stream_is_rotate_ready(stream); | |
1833 | if (rotation_ret == 1) { | |
e5148e25 | 1834 | rotation_ret = lttng_consumer_rotate_stream(ctx, stream); |
02d02e31 JD |
1835 | if (rotation_ret < 0) { |
1836 | ERR("Stream rotation error"); | |
1837 | ret = -1; | |
1838 | goto error; | |
1839 | } | |
1840 | } else if (rotation_ret < 0) { | |
1841 | ERR("Checking if stream is ready to rotate"); | |
1842 | ret = -1; | |
1843 | goto error; | |
1844 | } | |
1845 | ||
1846 | error: | |
d41f73b7 MD |
1847 | return ret; |
1848 | } | |
1849 | ||
1850 | int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
1851 | { | |
1852 | int ret; | |
ffe60014 DG |
1853 | |
1854 | assert(stream); | |
1855 | ||
2bba9e53 | 1856 | /* |
e5148e25 JG |
1857 | * Don't create anything if this is set for streaming or if there is |
1858 | * no current trace chunk on the parent channel. | |
2bba9e53 | 1859 | */ |
e5148e25 JG |
1860 | if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor && |
1861 | stream->chan->trace_chunk) { | |
1862 | ret = consumer_stream_create_output_files(stream, true); | |
1863 | if (ret) { | |
fe4477ee JD |
1864 | goto error; |
1865 | } | |
ffe60014 | 1866 | } |
d41f73b7 | 1867 | |
d41f73b7 MD |
1868 | if (stream->output == LTTNG_EVENT_MMAP) { |
1869 | /* get the len of the mmap region */ | |
1870 | unsigned long mmap_len; | |
1871 | ||
1872 | ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); | |
1873 | if (ret != 0) { | |
ffe60014 | 1874 | PERROR("kernctl_get_mmap_len"); |
d41f73b7 MD |
1875 | goto error_close_fd; |
1876 | } | |
1877 | stream->mmap_len = (size_t) mmap_len; | |
1878 | ||
ffe60014 DG |
1879 | stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ, |
1880 | MAP_PRIVATE, stream->wait_fd, 0); | |
d41f73b7 | 1881 | if (stream->mmap_base == MAP_FAILED) { |
ffe60014 | 1882 | PERROR("Error mmaping"); |
d41f73b7 MD |
1883 | ret = -1; |
1884 | goto error_close_fd; | |
1885 | } | |
1886 | } | |
1887 | ||
1888 | /* we return 0 to let the library handle the FD internally */ | |
1889 | return 0; | |
1890 | ||
1891 | error_close_fd: | |
2f225ce2 | 1892 | if (stream->out_fd >= 0) { |
d41f73b7 MD |
1893 | int err; |
1894 | ||
1895 | err = close(stream->out_fd); | |
1896 | assert(!err); | |
2f225ce2 | 1897 | stream->out_fd = -1; |
d41f73b7 MD |
1898 | } |
1899 | error: | |
1900 | return ret; | |
1901 | } | |
1902 | ||
ca22feea DG |
1903 | /* |
1904 | * Check if data is still being extracted from the buffers for a specific | |
4e9a4686 DG |
1905 | * stream. Consumer data lock MUST be acquired before calling this function |
1906 | * and the stream lock. | |
ca22feea | 1907 | * |
6d805429 | 1908 | * Return 1 if the traced data are still getting read else 0 meaning that the |
ca22feea DG |
1909 | * data is available for trace viewer reading. |
1910 | */ | |
6d805429 | 1911 | int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream) |
ca22feea DG |
1912 | { |
1913 | int ret; | |
1914 | ||
1915 | assert(stream); | |
1916 | ||
873b9e9a MD |
1917 | if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) { |
1918 | ret = 0; | |
1919 | goto end; | |
1920 | } | |
1921 | ||
ca22feea DG |
1922 | ret = kernctl_get_next_subbuf(stream->wait_fd); |
1923 | if (ret == 0) { | |
1924 | /* There is still data so let's put back this subbuffer. */ | |
1925 | ret = kernctl_put_subbuf(stream->wait_fd); | |
1926 | assert(ret == 0); | |
6d805429 | 1927 | ret = 1; /* Data is pending */ |
4e9a4686 | 1928 | goto end; |
ca22feea DG |
1929 | } |
1930 | ||
6d805429 DG |
1931 | /* Data is NOT pending and ready to be read. */ |
1932 | ret = 0; | |
ca22feea | 1933 | |
6efae65e DG |
1934 | end: |
1935 | return ret; | |
ca22feea | 1936 | } |