Commit | Line | Data |
---|---|---|
3bd1e081 | 1 | /* |
ab5be9fa MJ |
2 | * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca> |
3 | * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
4 | * Copyright (C) 2017 Jérémie Galarneau <jeremie.galarneau@efficios.com> | |
3bd1e081 | 5 | * |
ab5be9fa | 6 | * SPDX-License-Identifier: GPL-2.0-only |
3bd1e081 | 7 | * |
3bd1e081 MD |
8 | */ |
9 | ||
276cd1d7 | 10 | #include <stdint.h> |
6c1c0768 | 11 | #define _LGPL_SOURCE |
3bd1e081 | 12 | #include <assert.h> |
3bd1e081 MD |
13 | #include <poll.h> |
14 | #include <pthread.h> | |
15 | #include <stdlib.h> | |
16 | #include <string.h> | |
17 | #include <sys/mman.h> | |
18 | #include <sys/socket.h> | |
19 | #include <sys/types.h> | |
77c7c900 | 20 | #include <inttypes.h> |
3bd1e081 | 21 | #include <unistd.h> |
dbb5dfe6 | 22 | #include <sys/stat.h> |
3bd1e081 | 23 | |
51a9e1c7 | 24 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 25 | #include <common/common.h> |
10a8a223 | 26 | #include <common/kernel-ctl/kernel-ctl.h> |
10a8a223 | 27 | #include <common/sessiond-comm/sessiond-comm.h> |
00e2e675 | 28 | #include <common/sessiond-comm/relayd.h> |
dbb5dfe6 | 29 | #include <common/compat/fcntl.h> |
f263b7fd | 30 | #include <common/compat/endian.h> |
acdb9057 | 31 | #include <common/pipe.h> |
00e2e675 | 32 | #include <common/relayd/relayd.h> |
fe4477ee | 33 | #include <common/utils.h> |
c8fea79c | 34 | #include <common/consumer/consumer-stream.h> |
309167d2 | 35 | #include <common/index/index.h> |
c8fea79c | 36 | #include <common/consumer/consumer-timer.h> |
d2956687 | 37 | #include <common/optional.h> |
0857097f | 38 | |
10a8a223 | 39 | #include "kernel-consumer.h" |
3bd1e081 MD |
40 | |
41 | extern struct lttng_consumer_global_data consumer_data; | |
42 | extern int consumer_poll_timeout; | |
3bd1e081 | 43 | |
3bd1e081 MD |
44 | /* |
45 | * Take a snapshot for a specific fd | |
46 | * | |
47 | * Returns 0 on success, < 0 on error | |
48 | */ | |
ffe60014 | 49 | int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
50 | { |
51 | int ret = 0; | |
52 | int infd = stream->wait_fd; | |
53 | ||
54 | ret = kernctl_snapshot(infd); | |
d2d2f190 JD |
55 | /* |
56 | * -EAGAIN is not an error, it just means that there is no data to | |
57 | * be read. | |
58 | */ | |
59 | if (ret != 0 && ret != -EAGAIN) { | |
5a510c9f | 60 | PERROR("Getting sub-buffer snapshot."); |
3bd1e081 MD |
61 | } |
62 | ||
63 | return ret; | |
64 | } | |
65 | ||
e9404c27 JG |
66 | /* |
67 | * Sample consumed and produced positions for a specific fd. | |
68 | * | |
69 | * Returns 0 on success, < 0 on error. | |
70 | */ | |
71 | int lttng_kconsumer_sample_snapshot_positions( | |
72 | struct lttng_consumer_stream *stream) | |
73 | { | |
74 | assert(stream); | |
75 | ||
76 | return kernctl_snapshot_sample_positions(stream->wait_fd); | |
77 | } | |
78 | ||
3bd1e081 MD |
79 | /* |
80 | * Get the produced position | |
81 | * | |
82 | * Returns 0 on success, < 0 on error | |
83 | */ | |
ffe60014 | 84 | int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
85 | unsigned long *pos) |
86 | { | |
87 | int ret; | |
88 | int infd = stream->wait_fd; | |
89 | ||
90 | ret = kernctl_snapshot_get_produced(infd, pos); | |
91 | if (ret != 0) { | |
5a510c9f | 92 | PERROR("kernctl_snapshot_get_produced"); |
3bd1e081 MD |
93 | } |
94 | ||
95 | return ret; | |
96 | } | |
97 | ||
07b86b52 JD |
98 | /* |
99 | * Get the consumerd position | |
100 | * | |
101 | * Returns 0 on success, < 0 on error | |
102 | */ | |
103 | int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, | |
104 | unsigned long *pos) | |
105 | { | |
106 | int ret; | |
107 | int infd = stream->wait_fd; | |
108 | ||
109 | ret = kernctl_snapshot_get_consumed(infd, pos); | |
110 | if (ret != 0) { | |
5a510c9f | 111 | PERROR("kernctl_snapshot_get_consumed"); |
07b86b52 JD |
112 | } |
113 | ||
114 | return ret; | |
115 | } | |
116 | ||
276cd1d7 JG |
117 | static |
118 | int get_current_subbuf_addr(struct lttng_consumer_stream *stream, | |
119 | const char **addr) | |
120 | { | |
121 | int ret; | |
122 | unsigned long mmap_offset; | |
123 | const char *mmap_base = stream->mmap_base; | |
124 | ||
125 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); | |
126 | if (ret < 0) { | |
127 | PERROR("Failed to get mmap read offset"); | |
128 | goto error; | |
129 | } | |
130 | ||
131 | *addr = mmap_base + mmap_offset; | |
132 | error: | |
133 | return ret; | |
134 | } | |
135 | ||
07b86b52 JD |
136 | /* |
137 | * Take a snapshot of all the stream of a channel | |
3eb928aa | 138 | * RCU read-side lock must be held across this function to ensure existence of |
d2956687 | 139 | * channel. The channel lock must be held by the caller. |
07b86b52 JD |
140 | * |
141 | * Returns 0 on success, < 0 on error | |
142 | */ | |
f72bb42f JG |
143 | static int lttng_kconsumer_snapshot_channel( |
144 | struct lttng_consumer_channel *channel, | |
145 | uint64_t key, char *path, uint64_t relayd_id, | |
146 | uint64_t nb_packets_per_stream, | |
5c786ded | 147 | struct lttng_consumer_local_data *ctx) |
07b86b52 JD |
148 | { |
149 | int ret; | |
07b86b52 JD |
150 | struct lttng_consumer_stream *stream; |
151 | ||
6a00837f | 152 | DBG("Kernel consumer snapshot channel %" PRIu64, key); |
07b86b52 JD |
153 | |
154 | rcu_read_lock(); | |
155 | ||
07b86b52 JD |
156 | /* Splice is not supported yet for channel snapshot. */ |
157 | if (channel->output != CONSUMER_CHANNEL_MMAP) { | |
9381314c JG |
158 | ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot", |
159 | channel->name); | |
07b86b52 JD |
160 | ret = -1; |
161 | goto end; | |
162 | } | |
163 | ||
10a50311 | 164 | cds_list_for_each_entry(stream, &channel->streams.head, send_node) { |
923333cd | 165 | unsigned long consumed_pos, produced_pos; |
9ce5646a MD |
166 | |
167 | health_code_update(); | |
168 | ||
07b86b52 JD |
169 | /* |
170 | * Lock stream because we are about to change its state. | |
171 | */ | |
172 | pthread_mutex_lock(&stream->lock); | |
173 | ||
d2956687 JG |
174 | assert(channel->trace_chunk); |
175 | if (!lttng_trace_chunk_get(channel->trace_chunk)) { | |
176 | /* | |
177 | * Can't happen barring an internal error as the channel | |
178 | * holds a reference to the trace chunk. | |
179 | */ | |
180 | ERR("Failed to acquire reference to channel's trace chunk"); | |
181 | ret = -1; | |
182 | goto end_unlock; | |
183 | } | |
184 | assert(!stream->trace_chunk); | |
185 | stream->trace_chunk = channel->trace_chunk; | |
186 | ||
29decac3 DG |
187 | /* |
188 | * Assign the received relayd ID so we can use it for streaming. The streams | |
189 | * are not visible to anyone so this is OK to change it. | |
190 | */ | |
07b86b52 JD |
191 | stream->net_seq_idx = relayd_id; |
192 | channel->relayd_id = relayd_id; | |
193 | if (relayd_id != (uint64_t) -1ULL) { | |
10a50311 | 194 | ret = consumer_send_relayd_stream(stream, path); |
07b86b52 JD |
195 | if (ret < 0) { |
196 | ERR("sending stream to relayd"); | |
197 | goto end_unlock; | |
198 | } | |
07b86b52 | 199 | } else { |
d2956687 JG |
200 | ret = consumer_stream_create_output_files(stream, |
201 | false); | |
07b86b52 | 202 | if (ret < 0) { |
07b86b52 JD |
203 | goto end_unlock; |
204 | } | |
d2956687 JG |
205 | DBG("Kernel consumer snapshot stream (%" PRIu64 ")", |
206 | stream->key); | |
07b86b52 JD |
207 | } |
208 | ||
f22dd891 | 209 | ret = kernctl_buffer_flush_empty(stream->wait_fd); |
07b86b52 | 210 | if (ret < 0) { |
f22dd891 MD |
211 | /* |
212 | * Doing a buffer flush which does not take into | |
213 | * account empty packets. This is not perfect | |
214 | * for stream intersection, but required as a | |
215 | * fall-back when "flush_empty" is not | |
216 | * implemented by lttng-modules. | |
217 | */ | |
218 | ret = kernctl_buffer_flush(stream->wait_fd); | |
219 | if (ret < 0) { | |
220 | ERR("Failed to flush kernel stream"); | |
221 | goto end_unlock; | |
222 | } | |
07b86b52 JD |
223 | goto end_unlock; |
224 | } | |
225 | ||
226 | ret = lttng_kconsumer_take_snapshot(stream); | |
227 | if (ret < 0) { | |
228 | ERR("Taking kernel snapshot"); | |
229 | goto end_unlock; | |
230 | } | |
231 | ||
232 | ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos); | |
233 | if (ret < 0) { | |
234 | ERR("Produced kernel snapshot position"); | |
235 | goto end_unlock; | |
236 | } | |
237 | ||
238 | ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos); | |
239 | if (ret < 0) { | |
240 | ERR("Consumerd kernel snapshot position"); | |
241 | goto end_unlock; | |
242 | } | |
243 | ||
d07ceecd MD |
244 | consumed_pos = consumer_get_consume_start_pos(consumed_pos, |
245 | produced_pos, nb_packets_per_stream, | |
246 | stream->max_sb_size); | |
5c786ded | 247 | |
9377d830 | 248 | while ((long) (consumed_pos - produced_pos) < 0) { |
07b86b52 JD |
249 | ssize_t read_len; |
250 | unsigned long len, padded_len; | |
276cd1d7 | 251 | const char *subbuf_addr; |
07b86b52 | 252 | |
9ce5646a MD |
253 | health_code_update(); |
254 | ||
07b86b52 JD |
255 | DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos); |
256 | ||
257 | ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos); | |
258 | if (ret < 0) { | |
32af2c95 | 259 | if (ret != -EAGAIN) { |
07b86b52 JD |
260 | PERROR("kernctl_get_subbuf snapshot"); |
261 | goto end_unlock; | |
262 | } | |
263 | DBG("Kernel consumer get subbuf failed. Skipping it."); | |
264 | consumed_pos += stream->max_sb_size; | |
ddc93ee4 | 265 | stream->chan->lost_packets++; |
07b86b52 JD |
266 | continue; |
267 | } | |
268 | ||
269 | ret = kernctl_get_subbuf_size(stream->wait_fd, &len); | |
270 | if (ret < 0) { | |
271 | ERR("Snapshot kernctl_get_subbuf_size"); | |
29decac3 | 272 | goto error_put_subbuf; |
07b86b52 JD |
273 | } |
274 | ||
275 | ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len); | |
276 | if (ret < 0) { | |
277 | ERR("Snapshot kernctl_get_padded_subbuf_size"); | |
29decac3 | 278 | goto error_put_subbuf; |
07b86b52 JD |
279 | } |
280 | ||
276cd1d7 JG |
281 | ret = get_current_subbuf_addr(stream, &subbuf_addr); |
282 | if (ret) { | |
283 | goto error_put_subbuf; | |
284 | } | |
285 | ||
286 | read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, | |
287 | stream, subbuf_addr, len, | |
309167d2 | 288 | padded_len - len, NULL); |
07b86b52 | 289 | /* |
29decac3 DG |
290 | * We write the padded len in local tracefiles but the data len |
291 | * when using a relay. Display the error but continue processing | |
292 | * to try to release the subbuffer. | |
07b86b52 JD |
293 | */ |
294 | if (relayd_id != (uint64_t) -1ULL) { | |
295 | if (read_len != len) { | |
296 | ERR("Error sending to the relay (ret: %zd != len: %lu)", | |
297 | read_len, len); | |
298 | } | |
299 | } else { | |
300 | if (read_len != padded_len) { | |
301 | ERR("Error writing to tracefile (ret: %zd != len: %lu)", | |
302 | read_len, padded_len); | |
303 | } | |
304 | } | |
305 | ||
306 | ret = kernctl_put_subbuf(stream->wait_fd); | |
307 | if (ret < 0) { | |
308 | ERR("Snapshot kernctl_put_subbuf"); | |
309 | goto end_unlock; | |
310 | } | |
311 | consumed_pos += stream->max_sb_size; | |
312 | } | |
313 | ||
314 | if (relayd_id == (uint64_t) -1ULL) { | |
fdf9986c MD |
315 | if (stream->out_fd >= 0) { |
316 | ret = close(stream->out_fd); | |
317 | if (ret < 0) { | |
318 | PERROR("Kernel consumer snapshot close out_fd"); | |
319 | goto end_unlock; | |
320 | } | |
321 | stream->out_fd = -1; | |
07b86b52 | 322 | } |
07b86b52 JD |
323 | } else { |
324 | close_relayd_stream(stream); | |
325 | stream->net_seq_idx = (uint64_t) -1ULL; | |
326 | } | |
d2956687 JG |
327 | lttng_trace_chunk_put(stream->trace_chunk); |
328 | stream->trace_chunk = NULL; | |
07b86b52 JD |
329 | pthread_mutex_unlock(&stream->lock); |
330 | } | |
331 | ||
332 | /* All good! */ | |
333 | ret = 0; | |
334 | goto end; | |
335 | ||
29decac3 DG |
336 | error_put_subbuf: |
337 | ret = kernctl_put_subbuf(stream->wait_fd); | |
338 | if (ret < 0) { | |
339 | ERR("Snapshot kernctl_put_subbuf error path"); | |
340 | } | |
07b86b52 JD |
341 | end_unlock: |
342 | pthread_mutex_unlock(&stream->lock); | |
343 | end: | |
344 | rcu_read_unlock(); | |
345 | return ret; | |
346 | } | |
347 | ||
348 | /* | |
349 | * Read the whole metadata available for a snapshot. | |
3eb928aa | 350 | * RCU read-side lock must be held across this function to ensure existence of |
d2956687 | 351 | * metadata_channel. The channel lock must be held by the caller. |
07b86b52 JD |
352 | * |
353 | * Returns 0 on success, < 0 on error | |
354 | */ | |
d2956687 JG |
355 | static int lttng_kconsumer_snapshot_metadata( |
356 | struct lttng_consumer_channel *metadata_channel, | |
3eb928aa MD |
357 | uint64_t key, char *path, uint64_t relayd_id, |
358 | struct lttng_consumer_local_data *ctx) | |
07b86b52 | 359 | { |
d771f832 DG |
360 | int ret, use_relayd = 0; |
361 | ssize_t ret_read; | |
07b86b52 | 362 | struct lttng_consumer_stream *metadata_stream; |
d771f832 DG |
363 | |
364 | assert(ctx); | |
07b86b52 JD |
365 | |
366 | DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s", | |
367 | key, path); | |
368 | ||
369 | rcu_read_lock(); | |
370 | ||
07b86b52 JD |
371 | metadata_stream = metadata_channel->metadata_stream; |
372 | assert(metadata_stream); | |
d2956687 | 373 | |
fa27abe8 | 374 | pthread_mutex_lock(&metadata_stream->lock); |
d2956687 JG |
375 | assert(metadata_channel->trace_chunk); |
376 | assert(metadata_stream->trace_chunk); | |
07b86b52 | 377 | |
d771f832 | 378 | /* Flag once that we have a valid relayd for the stream. */ |
e2039c7a | 379 | if (relayd_id != (uint64_t) -1ULL) { |
d771f832 DG |
380 | use_relayd = 1; |
381 | } | |
382 | ||
383 | if (use_relayd) { | |
10a50311 | 384 | ret = consumer_send_relayd_stream(metadata_stream, path); |
e2039c7a | 385 | if (ret < 0) { |
fa27abe8 | 386 | goto error_snapshot; |
e2039c7a | 387 | } |
e2039c7a | 388 | } else { |
d2956687 JG |
389 | ret = consumer_stream_create_output_files(metadata_stream, |
390 | false); | |
e2039c7a | 391 | if (ret < 0) { |
fa27abe8 | 392 | goto error_snapshot; |
e2039c7a | 393 | } |
07b86b52 | 394 | } |
07b86b52 | 395 | |
d771f832 | 396 | do { |
9ce5646a MD |
397 | health_code_update(); |
398 | ||
d2956687 | 399 | ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx); |
d771f832 | 400 | if (ret_read < 0) { |
56591bac | 401 | if (ret_read != -EAGAIN) { |
6a00837f | 402 | ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)", |
d771f832 | 403 | ret_read); |
fa27abe8 JG |
404 | ret = ret_read; |
405 | goto error_snapshot; | |
07b86b52 | 406 | } |
d771f832 | 407 | /* ret_read is negative at this point so we will exit the loop. */ |
07b86b52 JD |
408 | continue; |
409 | } | |
d771f832 | 410 | } while (ret_read >= 0); |
07b86b52 | 411 | |
d771f832 DG |
412 | if (use_relayd) { |
413 | close_relayd_stream(metadata_stream); | |
414 | metadata_stream->net_seq_idx = (uint64_t) -1ULL; | |
415 | } else { | |
fdf9986c MD |
416 | if (metadata_stream->out_fd >= 0) { |
417 | ret = close(metadata_stream->out_fd); | |
418 | if (ret < 0) { | |
419 | PERROR("Kernel consumer snapshot metadata close out_fd"); | |
420 | /* | |
421 | * Don't go on error here since the snapshot was successful at this | |
422 | * point but somehow the close failed. | |
423 | */ | |
424 | } | |
425 | metadata_stream->out_fd = -1; | |
d2956687 JG |
426 | lttng_trace_chunk_put(metadata_stream->trace_chunk); |
427 | metadata_stream->trace_chunk = NULL; | |
e2039c7a | 428 | } |
e2039c7a JD |
429 | } |
430 | ||
07b86b52 | 431 | ret = 0; |
fa27abe8 JG |
432 | error_snapshot: |
433 | pthread_mutex_unlock(&metadata_stream->lock); | |
cf53a8a6 JD |
434 | cds_list_del(&metadata_stream->send_node); |
435 | consumer_stream_destroy(metadata_stream, NULL); | |
436 | metadata_channel->metadata_stream = NULL; | |
07b86b52 JD |
437 | rcu_read_unlock(); |
438 | return ret; | |
439 | } | |
440 | ||
1803a064 MD |
441 | /* |
442 | * Receive command from session daemon and process it. | |
443 | * | |
444 | * Return 1 on success else a negative value or 0. | |
445 | */ | |
3bd1e081 MD |
446 | int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
447 | int sock, struct pollfd *consumer_sockpoll) | |
448 | { | |
449 | ssize_t ret; | |
0c759fc9 | 450 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
3bd1e081 MD |
451 | struct lttcomm_consumer_msg msg; |
452 | ||
9ce5646a MD |
453 | health_code_update(); |
454 | ||
3bd1e081 MD |
455 | ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); |
456 | if (ret != sizeof(msg)) { | |
1803a064 | 457 | if (ret > 0) { |
c6857fcf | 458 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); |
1803a064 MD |
459 | ret = -1; |
460 | } | |
3bd1e081 MD |
461 | return ret; |
462 | } | |
9ce5646a MD |
463 | |
464 | health_code_update(); | |
465 | ||
84382d49 MD |
466 | /* Deprecated command */ |
467 | assert(msg.cmd_type != LTTNG_CONSUMER_STOP); | |
3bd1e081 | 468 | |
9ce5646a MD |
469 | health_code_update(); |
470 | ||
b0b335c8 MD |
471 | /* relayd needs RCU read-side protection */ |
472 | rcu_read_lock(); | |
473 | ||
3bd1e081 | 474 | switch (msg.cmd_type) { |
00e2e675 DG |
475 | case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: |
476 | { | |
f50f23d9 | 477 | /* Session daemon status message are handled in the following call. */ |
2527bf85 | 478 | consumer_add_relayd_socket(msg.u.relayd_sock.net_index, |
7735ef9e | 479 | msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, |
d3e2ba59 | 480 | &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id, |
2527bf85 | 481 | msg.u.relayd_sock.relayd_session_id); |
00e2e675 DG |
482 | goto end_nosignal; |
483 | } | |
3bd1e081 MD |
484 | case LTTNG_CONSUMER_ADD_CHANNEL: |
485 | { | |
486 | struct lttng_consumer_channel *new_channel; | |
e43c41c5 | 487 | int ret_recv; |
d2956687 | 488 | const uint64_t chunk_id = msg.u.channel.chunk_id.value; |
3bd1e081 | 489 | |
9ce5646a MD |
490 | health_code_update(); |
491 | ||
f50f23d9 DG |
492 | /* First send a status message before receiving the fds. */ |
493 | ret = consumer_send_status_msg(sock, ret_code); | |
494 | if (ret < 0) { | |
495 | /* Somehow, the session daemon is not responding anymore. */ | |
1803a064 | 496 | goto error_fatal; |
f50f23d9 | 497 | } |
9ce5646a MD |
498 | |
499 | health_code_update(); | |
500 | ||
d88aee68 | 501 | DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key); |
3bd1e081 | 502 | new_channel = consumer_allocate_channel(msg.u.channel.channel_key, |
d2956687 JG |
503 | msg.u.channel.session_id, |
504 | msg.u.channel.chunk_id.is_set ? | |
505 | &chunk_id : NULL, | |
506 | msg.u.channel.pathname, | |
507 | msg.u.channel.name, | |
1624d5b7 JD |
508 | msg.u.channel.relayd_id, msg.u.channel.output, |
509 | msg.u.channel.tracefile_size, | |
1950109e | 510 | msg.u.channel.tracefile_count, 0, |
ecc48a90 | 511 | msg.u.channel.monitor, |
d7ba1388 | 512 | msg.u.channel.live_timer_interval, |
3d071855 | 513 | NULL, NULL); |
3bd1e081 | 514 | if (new_channel == NULL) { |
f73fabfd | 515 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
3bd1e081 MD |
516 | goto end_nosignal; |
517 | } | |
ffe60014 | 518 | new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams; |
95a1109b JD |
519 | switch (msg.u.channel.output) { |
520 | case LTTNG_EVENT_SPLICE: | |
521 | new_channel->output = CONSUMER_CHANNEL_SPLICE; | |
522 | break; | |
523 | case LTTNG_EVENT_MMAP: | |
524 | new_channel->output = CONSUMER_CHANNEL_MMAP; | |
525 | break; | |
526 | default: | |
527 | ERR("Channel output unknown %d", msg.u.channel.output); | |
528 | goto end_nosignal; | |
529 | } | |
ffe60014 DG |
530 | |
531 | /* Translate and save channel type. */ | |
532 | switch (msg.u.channel.type) { | |
533 | case CONSUMER_CHANNEL_TYPE_DATA: | |
534 | case CONSUMER_CHANNEL_TYPE_METADATA: | |
535 | new_channel->type = msg.u.channel.type; | |
536 | break; | |
537 | default: | |
538 | assert(0); | |
539 | goto end_nosignal; | |
540 | }; | |
541 | ||
9ce5646a MD |
542 | health_code_update(); |
543 | ||
3bd1e081 | 544 | if (ctx->on_recv_channel != NULL) { |
e43c41c5 JD |
545 | ret_recv = ctx->on_recv_channel(new_channel); |
546 | if (ret_recv == 0) { | |
547 | ret = consumer_add_channel(new_channel, ctx); | |
548 | } else if (ret_recv < 0) { | |
3bd1e081 MD |
549 | goto end_nosignal; |
550 | } | |
551 | } else { | |
e43c41c5 | 552 | ret = consumer_add_channel(new_channel, ctx); |
3bd1e081 | 553 | } |
e9404c27 JG |
554 | if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) { |
555 | int monitor_start_ret; | |
556 | ||
557 | DBG("Consumer starting monitor timer"); | |
94d49140 JD |
558 | consumer_timer_live_start(new_channel, |
559 | msg.u.channel.live_timer_interval); | |
e9404c27 JG |
560 | monitor_start_ret = consumer_timer_monitor_start( |
561 | new_channel, | |
562 | msg.u.channel.monitor_timer_interval); | |
563 | if (monitor_start_ret < 0) { | |
564 | ERR("Starting channel monitoring timer failed"); | |
565 | goto end_nosignal; | |
566 | } | |
567 | ||
94d49140 | 568 | } |
e43c41c5 | 569 | |
9ce5646a MD |
570 | health_code_update(); |
571 | ||
e43c41c5 | 572 | /* If we received an error in add_channel, we need to report it. */ |
821fffb2 | 573 | if (ret < 0) { |
1803a064 MD |
574 | ret = consumer_send_status_msg(sock, ret); |
575 | if (ret < 0) { | |
576 | goto error_fatal; | |
577 | } | |
e43c41c5 JD |
578 | goto end_nosignal; |
579 | } | |
580 | ||
3bd1e081 MD |
581 | goto end_nosignal; |
582 | } | |
583 | case LTTNG_CONSUMER_ADD_STREAM: | |
584 | { | |
dae10966 DG |
585 | int fd; |
586 | struct lttng_pipe *stream_pipe; | |
00e2e675 | 587 | struct lttng_consumer_stream *new_stream; |
ffe60014 | 588 | struct lttng_consumer_channel *channel; |
c80048c6 | 589 | int alloc_ret = 0; |
3bd1e081 | 590 | |
ffe60014 DG |
591 | /* |
592 | * Get stream's channel reference. Needed when adding the stream to the | |
593 | * global hash table. | |
594 | */ | |
595 | channel = consumer_find_channel(msg.u.stream.channel_key); | |
596 | if (!channel) { | |
597 | /* | |
598 | * We could not find the channel. Can happen if cpu hotplug | |
599 | * happens while tearing down. | |
600 | */ | |
d88aee68 | 601 | ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key); |
e462382a | 602 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; |
ffe60014 DG |
603 | } |
604 | ||
9ce5646a MD |
605 | health_code_update(); |
606 | ||
f50f23d9 DG |
607 | /* First send a status message before receiving the fds. */ |
608 | ret = consumer_send_status_msg(sock, ret_code); | |
1803a064 | 609 | if (ret < 0) { |
d771f832 | 610 | /* Somehow, the session daemon is not responding anymore. */ |
c5c7998f | 611 | goto error_add_stream_fatal; |
1803a064 | 612 | } |
9ce5646a MD |
613 | |
614 | health_code_update(); | |
615 | ||
0c759fc9 | 616 | if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) { |
d771f832 | 617 | /* Channel was not found. */ |
c5c7998f | 618 | goto error_add_stream_nosignal; |
f50f23d9 DG |
619 | } |
620 | ||
d771f832 | 621 | /* Blocking call */ |
9ce5646a MD |
622 | health_poll_entry(); |
623 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
624 | health_poll_exit(); | |
84382d49 | 625 | if (ret) { |
c5c7998f | 626 | goto error_add_stream_fatal; |
3bd1e081 | 627 | } |
00e2e675 | 628 | |
9ce5646a MD |
629 | health_code_update(); |
630 | ||
00e2e675 | 631 | /* Get stream file descriptor from socket */ |
f2fc6720 MD |
632 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); |
633 | if (ret != sizeof(fd)) { | |
f73fabfd | 634 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
c5c7998f | 635 | goto end; |
3bd1e081 | 636 | } |
3bd1e081 | 637 | |
9ce5646a MD |
638 | health_code_update(); |
639 | ||
f50f23d9 DG |
640 | /* |
641 | * Send status code to session daemon only if the recv works. If the | |
642 | * above recv() failed, the session daemon is notified through the | |
643 | * error socket and the teardown is eventually done. | |
644 | */ | |
645 | ret = consumer_send_status_msg(sock, ret_code); | |
646 | if (ret < 0) { | |
647 | /* Somehow, the session daemon is not responding anymore. */ | |
c5c7998f | 648 | goto error_add_stream_nosignal; |
f50f23d9 DG |
649 | } |
650 | ||
9ce5646a MD |
651 | health_code_update(); |
652 | ||
d2956687 | 653 | pthread_mutex_lock(&channel->lock); |
ffe60014 DG |
654 | new_stream = consumer_allocate_stream(channel->key, |
655 | fd, | |
ffe60014 | 656 | channel->name, |
ffe60014 DG |
657 | channel->relayd_id, |
658 | channel->session_id, | |
d2956687 | 659 | channel->trace_chunk, |
ffe60014 DG |
660 | msg.u.stream.cpu, |
661 | &alloc_ret, | |
4891ece8 | 662 | channel->type, |
d2956687 | 663 | channel->monitor); |
3bd1e081 | 664 | if (new_stream == NULL) { |
c80048c6 MD |
665 | switch (alloc_ret) { |
666 | case -ENOMEM: | |
667 | case -EINVAL: | |
668 | default: | |
669 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); | |
670 | break; | |
c80048c6 | 671 | } |
d2956687 | 672 | pthread_mutex_unlock(&channel->lock); |
c5c7998f | 673 | goto error_add_stream_nosignal; |
3bd1e081 | 674 | } |
d771f832 | 675 | |
ffe60014 DG |
676 | new_stream->chan = channel; |
677 | new_stream->wait_fd = fd; | |
d05185fa JG |
678 | ret = kernctl_get_max_subbuf_size(new_stream->wait_fd, |
679 | &new_stream->max_sb_size); | |
680 | if (ret < 0) { | |
681 | pthread_mutex_unlock(&channel->lock); | |
682 | ERR("Failed to get kernel maximal subbuffer size"); | |
c5c7998f | 683 | goto error_add_stream_nosignal; |
d05185fa JG |
684 | } |
685 | ||
d9a2e16e JD |
686 | consumer_stream_update_channel_attributes(new_stream, |
687 | channel); | |
07b86b52 JD |
688 | switch (channel->output) { |
689 | case CONSUMER_CHANNEL_SPLICE: | |
690 | new_stream->output = LTTNG_EVENT_SPLICE; | |
a2361a61 JD |
691 | ret = utils_create_pipe(new_stream->splice_pipe); |
692 | if (ret < 0) { | |
d2956687 | 693 | pthread_mutex_unlock(&channel->lock); |
c5c7998f | 694 | goto error_add_stream_nosignal; |
a2361a61 | 695 | } |
07b86b52 JD |
696 | break; |
697 | case CONSUMER_CHANNEL_MMAP: | |
698 | new_stream->output = LTTNG_EVENT_MMAP; | |
699 | break; | |
700 | default: | |
701 | ERR("Stream output unknown %d", channel->output); | |
d2956687 | 702 | pthread_mutex_unlock(&channel->lock); |
c5c7998f | 703 | goto error_add_stream_nosignal; |
07b86b52 | 704 | } |
00e2e675 | 705 | |
a0c83db9 DG |
706 | /* |
707 | * We've just assigned the channel to the stream so increment the | |
07b86b52 JD |
708 | * refcount right now. We don't need to increment the refcount for |
709 | * streams in no monitor because we handle manually the cleanup of | |
710 | * those. It is very important to make sure there is NO prior | |
711 | * consumer_del_stream() calls or else the refcount will be unbalanced. | |
a0c83db9 | 712 | */ |
07b86b52 JD |
713 | if (channel->monitor) { |
714 | uatomic_inc(&new_stream->chan->refcount); | |
715 | } | |
9d9353f9 | 716 | |
fb3a43a9 DG |
717 | /* |
718 | * The buffer flush is done on the session daemon side for the kernel | |
719 | * so no need for the stream "hangup_flush_done" variable to be | |
720 | * tracked. This is important for a kernel stream since we don't rely | |
721 | * on the flush state of the stream to read data. It's not the case for | |
722 | * user space tracing. | |
723 | */ | |
724 | new_stream->hangup_flush_done = 0; | |
725 | ||
9ce5646a MD |
726 | health_code_update(); |
727 | ||
d2956687 | 728 | pthread_mutex_lock(&new_stream->lock); |
633d0084 DG |
729 | if (ctx->on_recv_stream) { |
730 | ret = ctx->on_recv_stream(new_stream); | |
731 | if (ret < 0) { | |
d2956687 JG |
732 | pthread_mutex_unlock(&new_stream->lock); |
733 | pthread_mutex_unlock(&channel->lock); | |
d771f832 | 734 | consumer_stream_free(new_stream); |
c5c7998f | 735 | goto error_add_stream_nosignal; |
fb3a43a9 | 736 | } |
633d0084 | 737 | } |
9ce5646a MD |
738 | health_code_update(); |
739 | ||
07b86b52 JD |
740 | if (new_stream->metadata_flag) { |
741 | channel->metadata_stream = new_stream; | |
742 | } | |
743 | ||
2bba9e53 DG |
744 | /* Do not monitor this stream. */ |
745 | if (!channel->monitor) { | |
5eecee74 | 746 | DBG("Kernel consumer add stream %s in no monitor mode with " |
6dc3064a | 747 | "relayd id %" PRIu64, new_stream->name, |
5eecee74 | 748 | new_stream->net_seq_idx); |
10a50311 | 749 | cds_list_add(&new_stream->send_node, &channel->streams.head); |
d2956687 JG |
750 | pthread_mutex_unlock(&new_stream->lock); |
751 | pthread_mutex_unlock(&channel->lock); | |
c5c7998f | 752 | goto end_add_stream; |
6dc3064a DG |
753 | } |
754 | ||
e1b71bdc DG |
755 | /* Send stream to relayd if the stream has an ID. */ |
756 | if (new_stream->net_seq_idx != (uint64_t) -1ULL) { | |
194ee077 DG |
757 | ret = consumer_send_relayd_stream(new_stream, |
758 | new_stream->chan->pathname); | |
e1b71bdc | 759 | if (ret < 0) { |
d2956687 JG |
760 | pthread_mutex_unlock(&new_stream->lock); |
761 | pthread_mutex_unlock(&channel->lock); | |
e1b71bdc | 762 | consumer_stream_free(new_stream); |
c5c7998f | 763 | goto error_add_stream_nosignal; |
e1b71bdc | 764 | } |
001b7e62 MD |
765 | |
766 | /* | |
767 | * If adding an extra stream to an already | |
768 | * existing channel (e.g. cpu hotplug), we need | |
769 | * to send the "streams_sent" command to relayd. | |
770 | */ | |
771 | if (channel->streams_sent_to_relayd) { | |
772 | ret = consumer_send_relayd_streams_sent( | |
773 | new_stream->net_seq_idx); | |
774 | if (ret < 0) { | |
d2956687 JG |
775 | pthread_mutex_unlock(&new_stream->lock); |
776 | pthread_mutex_unlock(&channel->lock); | |
c5c7998f | 777 | goto error_add_stream_nosignal; |
001b7e62 MD |
778 | } |
779 | } | |
e2039c7a | 780 | } |
d2956687 JG |
781 | pthread_mutex_unlock(&new_stream->lock); |
782 | pthread_mutex_unlock(&channel->lock); | |
e2039c7a | 783 | |
50f8ae69 | 784 | /* Get the right pipe where the stream will be sent. */ |
633d0084 | 785 | if (new_stream->metadata_flag) { |
66d583dc | 786 | consumer_add_metadata_stream(new_stream); |
dae10966 | 787 | stream_pipe = ctx->consumer_metadata_pipe; |
3bd1e081 | 788 | } else { |
66d583dc | 789 | consumer_add_data_stream(new_stream); |
dae10966 | 790 | stream_pipe = ctx->consumer_data_pipe; |
50f8ae69 DG |
791 | } |
792 | ||
66d583dc | 793 | /* Visible to other threads */ |
5ab66908 MD |
794 | new_stream->globally_visible = 1; |
795 | ||
9ce5646a MD |
796 | health_code_update(); |
797 | ||
dae10966 | 798 | ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream)); |
50f8ae69 | 799 | if (ret < 0) { |
dae10966 | 800 | ERR("Consumer write %s stream to pipe %d", |
50f8ae69 | 801 | new_stream->metadata_flag ? "metadata" : "data", |
dae10966 | 802 | lttng_pipe_get_writefd(stream_pipe)); |
5ab66908 MD |
803 | if (new_stream->metadata_flag) { |
804 | consumer_del_stream_for_metadata(new_stream); | |
805 | } else { | |
806 | consumer_del_stream_for_data(new_stream); | |
807 | } | |
c5c7998f | 808 | goto error_add_stream_nosignal; |
3bd1e081 | 809 | } |
00e2e675 | 810 | |
02d02e31 JD |
811 | DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64, |
812 | new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id); | |
c5c7998f | 813 | end_add_stream: |
3bd1e081 | 814 | break; |
c5c7998f JG |
815 | error_add_stream_nosignal: |
816 | goto end_nosignal; | |
817 | error_add_stream_fatal: | |
818 | goto error_fatal; | |
3bd1e081 | 819 | } |
a4baae1b JD |
820 | case LTTNG_CONSUMER_STREAMS_SENT: |
821 | { | |
822 | struct lttng_consumer_channel *channel; | |
823 | ||
824 | /* | |
825 | * Get stream's channel reference. Needed when adding the stream to the | |
826 | * global hash table. | |
827 | */ | |
828 | channel = consumer_find_channel(msg.u.sent_streams.channel_key); | |
829 | if (!channel) { | |
830 | /* | |
831 | * We could not find the channel. Can happen if cpu hotplug | |
832 | * happens while tearing down. | |
833 | */ | |
834 | ERR("Unable to find channel key %" PRIu64, | |
835 | msg.u.sent_streams.channel_key); | |
e462382a | 836 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; |
a4baae1b JD |
837 | } |
838 | ||
839 | health_code_update(); | |
840 | ||
841 | /* | |
842 | * Send status code to session daemon. | |
843 | */ | |
844 | ret = consumer_send_status_msg(sock, ret_code); | |
f261ad0a | 845 | if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) { |
a4baae1b | 846 | /* Somehow, the session daemon is not responding anymore. */ |
80d5a658 | 847 | goto error_streams_sent_nosignal; |
a4baae1b JD |
848 | } |
849 | ||
850 | health_code_update(); | |
851 | ||
852 | /* | |
853 | * We should not send this message if we don't monitor the | |
854 | * streams in this channel. | |
855 | */ | |
856 | if (!channel->monitor) { | |
80d5a658 | 857 | goto end_error_streams_sent; |
a4baae1b JD |
858 | } |
859 | ||
860 | health_code_update(); | |
861 | /* Send stream to relayd if the stream has an ID. */ | |
862 | if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) { | |
863 | ret = consumer_send_relayd_streams_sent( | |
864 | msg.u.sent_streams.net_seq_idx); | |
865 | if (ret < 0) { | |
80d5a658 | 866 | goto error_streams_sent_nosignal; |
a4baae1b | 867 | } |
001b7e62 | 868 | channel->streams_sent_to_relayd = true; |
a4baae1b | 869 | } |
80d5a658 | 870 | end_error_streams_sent: |
a4baae1b | 871 | break; |
80d5a658 JG |
872 | error_streams_sent_nosignal: |
873 | goto end_nosignal; | |
a4baae1b | 874 | } |
3bd1e081 MD |
875 | case LTTNG_CONSUMER_UPDATE_STREAM: |
876 | { | |
3f8e211f DG |
877 | rcu_read_unlock(); |
878 | return -ENOSYS; | |
879 | } | |
880 | case LTTNG_CONSUMER_DESTROY_RELAYD: | |
881 | { | |
a6ba4fe1 | 882 | uint64_t index = msg.u.destroy_relayd.net_seq_idx; |
3f8e211f DG |
883 | struct consumer_relayd_sock_pair *relayd; |
884 | ||
a6ba4fe1 | 885 | DBG("Kernel consumer destroying relayd %" PRIu64, index); |
3f8e211f DG |
886 | |
887 | /* Get relayd reference if exists. */ | |
a6ba4fe1 | 888 | relayd = consumer_find_relayd(index); |
3f8e211f | 889 | if (relayd == NULL) { |
3448e266 | 890 | DBG("Unable to find relayd %" PRIu64, index); |
e462382a | 891 | ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; |
3bd1e081 | 892 | } |
3f8e211f | 893 | |
a6ba4fe1 DG |
894 | /* |
895 | * Each relayd socket pair has a refcount of stream attached to it | |
896 | * which tells if the relayd is still active or not depending on the | |
897 | * refcount value. | |
898 | * | |
899 | * This will set the destroy flag of the relayd object and destroy it | |
900 | * if the refcount reaches zero when called. | |
901 | * | |
902 | * The destroy can happen either here or when a stream fd hangs up. | |
903 | */ | |
f50f23d9 DG |
904 | if (relayd) { |
905 | consumer_flag_relayd_for_destroy(relayd); | |
906 | } | |
907 | ||
9ce5646a MD |
908 | health_code_update(); |
909 | ||
f50f23d9 DG |
910 | ret = consumer_send_status_msg(sock, ret_code); |
911 | if (ret < 0) { | |
912 | /* Somehow, the session daemon is not responding anymore. */ | |
1803a064 | 913 | goto error_fatal; |
f50f23d9 | 914 | } |
3f8e211f | 915 | |
3f8e211f | 916 | goto end_nosignal; |
3bd1e081 | 917 | } |
6d805429 | 918 | case LTTNG_CONSUMER_DATA_PENDING: |
53632229 | 919 | { |
c8f59ee5 | 920 | int32_t ret; |
6d805429 | 921 | uint64_t id = msg.u.data_pending.session_id; |
c8f59ee5 | 922 | |
6d805429 | 923 | DBG("Kernel consumer data pending command for id %" PRIu64, id); |
c8f59ee5 | 924 | |
6d805429 | 925 | ret = consumer_data_pending(id); |
c8f59ee5 | 926 | |
9ce5646a MD |
927 | health_code_update(); |
928 | ||
c8f59ee5 DG |
929 | /* Send back returned value to session daemon */ |
930 | ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret)); | |
931 | if (ret < 0) { | |
6d805429 | 932 | PERROR("send data pending ret code"); |
1803a064 | 933 | goto error_fatal; |
c8f59ee5 | 934 | } |
f50f23d9 DG |
935 | |
936 | /* | |
937 | * No need to send back a status message since the data pending | |
938 | * returned value is the response. | |
939 | */ | |
c8f59ee5 | 940 | break; |
53632229 | 941 | } |
6dc3064a DG |
942 | case LTTNG_CONSUMER_SNAPSHOT_CHANNEL: |
943 | { | |
3eb928aa MD |
944 | struct lttng_consumer_channel *channel; |
945 | uint64_t key = msg.u.snapshot_channel.key; | |
946 | ||
947 | channel = consumer_find_channel(key); | |
948 | if (!channel) { | |
949 | ERR("Channel %" PRIu64 " not found", key); | |
950 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; | |
07b86b52 | 951 | } else { |
d2956687 | 952 | pthread_mutex_lock(&channel->lock); |
3eb928aa MD |
953 | if (msg.u.snapshot_channel.metadata == 1) { |
954 | ret = lttng_kconsumer_snapshot_metadata(channel, key, | |
955 | msg.u.snapshot_channel.pathname, | |
956 | msg.u.snapshot_channel.relayd_id, ctx); | |
957 | if (ret < 0) { | |
958 | ERR("Snapshot metadata failed"); | |
959 | ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; | |
960 | } | |
961 | } else { | |
962 | ret = lttng_kconsumer_snapshot_channel(channel, key, | |
963 | msg.u.snapshot_channel.pathname, | |
964 | msg.u.snapshot_channel.relayd_id, | |
965 | msg.u.snapshot_channel.nb_packets_per_stream, | |
966 | ctx); | |
967 | if (ret < 0) { | |
968 | ERR("Snapshot channel failed"); | |
969 | ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; | |
970 | } | |
07b86b52 | 971 | } |
d2956687 | 972 | pthread_mutex_unlock(&channel->lock); |
07b86b52 | 973 | } |
9ce5646a MD |
974 | health_code_update(); |
975 | ||
6dc3064a DG |
976 | ret = consumer_send_status_msg(sock, ret_code); |
977 | if (ret < 0) { | |
978 | /* Somehow, the session daemon is not responding anymore. */ | |
979 | goto end_nosignal; | |
980 | } | |
981 | break; | |
982 | } | |
07b86b52 JD |
983 | case LTTNG_CONSUMER_DESTROY_CHANNEL: |
984 | { | |
985 | uint64_t key = msg.u.destroy_channel.key; | |
986 | struct lttng_consumer_channel *channel; | |
987 | ||
988 | channel = consumer_find_channel(key); | |
989 | if (!channel) { | |
990 | ERR("Kernel consumer destroy channel %" PRIu64 " not found", key); | |
e462382a | 991 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; |
07b86b52 JD |
992 | } |
993 | ||
9ce5646a MD |
994 | health_code_update(); |
995 | ||
07b86b52 JD |
996 | ret = consumer_send_status_msg(sock, ret_code); |
997 | if (ret < 0) { | |
998 | /* Somehow, the session daemon is not responding anymore. */ | |
a9d36096 | 999 | goto end_destroy_channel; |
07b86b52 JD |
1000 | } |
1001 | ||
9ce5646a MD |
1002 | health_code_update(); |
1003 | ||
15dc512a DG |
1004 | /* Stop right now if no channel was found. */ |
1005 | if (!channel) { | |
a9d36096 | 1006 | goto end_destroy_channel; |
15dc512a DG |
1007 | } |
1008 | ||
07b86b52 JD |
1009 | /* |
1010 | * This command should ONLY be issued for channel with streams set in | |
1011 | * no monitor mode. | |
1012 | */ | |
1013 | assert(!channel->monitor); | |
1014 | ||
1015 | /* | |
1016 | * The refcount should ALWAYS be 0 in the case of a channel in no | |
1017 | * monitor mode. | |
1018 | */ | |
1019 | assert(!uatomic_sub_return(&channel->refcount, 1)); | |
1020 | ||
1021 | consumer_del_channel(channel); | |
a9d36096 | 1022 | end_destroy_channel: |
07b86b52 JD |
1023 | goto end_nosignal; |
1024 | } | |
fb83fe64 JD |
1025 | case LTTNG_CONSUMER_DISCARDED_EVENTS: |
1026 | { | |
66ab32be JD |
1027 | ssize_t ret; |
1028 | uint64_t count; | |
fb83fe64 JD |
1029 | struct lttng_consumer_channel *channel; |
1030 | uint64_t id = msg.u.discarded_events.session_id; | |
1031 | uint64_t key = msg.u.discarded_events.channel_key; | |
1032 | ||
e5742757 MD |
1033 | DBG("Kernel consumer discarded events command for session id %" |
1034 | PRIu64 ", channel key %" PRIu64, id, key); | |
1035 | ||
fb83fe64 JD |
1036 | channel = consumer_find_channel(key); |
1037 | if (!channel) { | |
1038 | ERR("Kernel consumer discarded events channel %" | |
1039 | PRIu64 " not found", key); | |
66ab32be | 1040 | count = 0; |
e5742757 | 1041 | } else { |
66ab32be | 1042 | count = channel->discarded_events; |
fb83fe64 JD |
1043 | } |
1044 | ||
fb83fe64 JD |
1045 | health_code_update(); |
1046 | ||
1047 | /* Send back returned value to session daemon */ | |
66ab32be | 1048 | ret = lttcomm_send_unix_sock(sock, &count, sizeof(count)); |
fb83fe64 JD |
1049 | if (ret < 0) { |
1050 | PERROR("send discarded events"); | |
1051 | goto error_fatal; | |
1052 | } | |
1053 | ||
1054 | break; | |
1055 | } | |
1056 | case LTTNG_CONSUMER_LOST_PACKETS: | |
1057 | { | |
66ab32be JD |
1058 | ssize_t ret; |
1059 | uint64_t count; | |
fb83fe64 JD |
1060 | struct lttng_consumer_channel *channel; |
1061 | uint64_t id = msg.u.lost_packets.session_id; | |
1062 | uint64_t key = msg.u.lost_packets.channel_key; | |
1063 | ||
e5742757 MD |
1064 | DBG("Kernel consumer lost packets command for session id %" |
1065 | PRIu64 ", channel key %" PRIu64, id, key); | |
1066 | ||
fb83fe64 JD |
1067 | channel = consumer_find_channel(key); |
1068 | if (!channel) { | |
1069 | ERR("Kernel consumer lost packets channel %" | |
1070 | PRIu64 " not found", key); | |
66ab32be | 1071 | count = 0; |
e5742757 | 1072 | } else { |
66ab32be | 1073 | count = channel->lost_packets; |
fb83fe64 JD |
1074 | } |
1075 | ||
fb83fe64 JD |
1076 | health_code_update(); |
1077 | ||
1078 | /* Send back returned value to session daemon */ | |
66ab32be | 1079 | ret = lttcomm_send_unix_sock(sock, &count, sizeof(count)); |
fb83fe64 JD |
1080 | if (ret < 0) { |
1081 | PERROR("send lost packets"); | |
1082 | goto error_fatal; | |
1083 | } | |
1084 | ||
1085 | break; | |
1086 | } | |
b3530820 JG |
1087 | case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE: |
1088 | { | |
1089 | int channel_monitor_pipe; | |
1090 | ||
1091 | ret_code = LTTCOMM_CONSUMERD_SUCCESS; | |
1092 | /* Successfully received the command's type. */ | |
1093 | ret = consumer_send_status_msg(sock, ret_code); | |
1094 | if (ret < 0) { | |
1095 | goto error_fatal; | |
1096 | } | |
1097 | ||
1098 | ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe, | |
1099 | 1); | |
1100 | if (ret != sizeof(channel_monitor_pipe)) { | |
1101 | ERR("Failed to receive channel monitor pipe"); | |
1102 | goto error_fatal; | |
1103 | } | |
1104 | ||
1105 | DBG("Received channel monitor pipe (%d)", channel_monitor_pipe); | |
1106 | ret = consumer_timer_thread_set_channel_monitor_pipe( | |
1107 | channel_monitor_pipe); | |
1108 | if (!ret) { | |
1109 | int flags; | |
1110 | ||
1111 | ret_code = LTTCOMM_CONSUMERD_SUCCESS; | |
1112 | /* Set the pipe as non-blocking. */ | |
1113 | ret = fcntl(channel_monitor_pipe, F_GETFL, 0); | |
1114 | if (ret == -1) { | |
1115 | PERROR("fcntl get flags of the channel monitoring pipe"); | |
1116 | goto error_fatal; | |
1117 | } | |
1118 | flags = ret; | |
1119 | ||
1120 | ret = fcntl(channel_monitor_pipe, F_SETFL, | |
1121 | flags | O_NONBLOCK); | |
1122 | if (ret == -1) { | |
1123 | PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe"); | |
1124 | goto error_fatal; | |
1125 | } | |
1126 | DBG("Channel monitor pipe set as non-blocking"); | |
1127 | } else { | |
1128 | ret_code = LTTCOMM_CONSUMERD_ALREADY_SET; | |
1129 | } | |
1130 | ret = consumer_send_status_msg(sock, ret_code); | |
1131 | if (ret < 0) { | |
1132 | goto error_fatal; | |
1133 | } | |
1134 | break; | |
1135 | } | |
b99a8d42 JD |
1136 | case LTTNG_CONSUMER_ROTATE_CHANNEL: |
1137 | { | |
92b7a7f8 MD |
1138 | struct lttng_consumer_channel *channel; |
1139 | uint64_t key = msg.u.rotate_channel.key; | |
b99a8d42 | 1140 | |
92b7a7f8 | 1141 | DBG("Consumer rotate channel %" PRIu64, key); |
b99a8d42 | 1142 | |
92b7a7f8 MD |
1143 | channel = consumer_find_channel(key); |
1144 | if (!channel) { | |
1145 | ERR("Channel %" PRIu64 " not found", key); | |
1146 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; | |
1147 | } else { | |
1148 | /* | |
1149 | * Sample the rotate position of all the streams in this channel. | |
1150 | */ | |
1151 | ret = lttng_consumer_rotate_channel(channel, key, | |
92b7a7f8 MD |
1152 | msg.u.rotate_channel.relayd_id, |
1153 | msg.u.rotate_channel.metadata, | |
92b7a7f8 MD |
1154 | ctx); |
1155 | if (ret < 0) { | |
1156 | ERR("Rotate channel failed"); | |
1157 | ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL; | |
1158 | } | |
b99a8d42 | 1159 | |
92b7a7f8 MD |
1160 | health_code_update(); |
1161 | } | |
b99a8d42 JD |
1162 | ret = consumer_send_status_msg(sock, ret_code); |
1163 | if (ret < 0) { | |
1164 | /* Somehow, the session daemon is not responding anymore. */ | |
713bdd26 | 1165 | goto error_rotate_channel; |
b99a8d42 | 1166 | } |
92b7a7f8 MD |
1167 | if (channel) { |
1168 | /* Rotate the streams that are ready right now. */ | |
1169 | ret = lttng_consumer_rotate_ready_streams( | |
1170 | channel, key, ctx); | |
1171 | if (ret < 0) { | |
1172 | ERR("Rotate ready streams failed"); | |
1173 | } | |
b99a8d42 | 1174 | } |
b99a8d42 | 1175 | break; |
713bdd26 JG |
1176 | error_rotate_channel: |
1177 | goto end_nosignal; | |
b99a8d42 | 1178 | } |
5f3aff8b MD |
1179 | case LTTNG_CONSUMER_CLEAR_CHANNEL: |
1180 | { | |
1181 | struct lttng_consumer_channel *channel; | |
1182 | uint64_t key = msg.u.clear_channel.key; | |
1183 | ||
1184 | channel = consumer_find_channel(key); | |
1185 | if (!channel) { | |
1186 | DBG("Channel %" PRIu64 " not found", key); | |
1187 | ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; | |
1188 | } else { | |
1189 | ret = lttng_consumer_clear_channel(channel); | |
1190 | if (ret) { | |
1191 | ERR("Clear channel failed"); | |
1192 | ret_code = ret; | |
1193 | } | |
1194 | ||
1195 | health_code_update(); | |
1196 | } | |
1197 | ret = consumer_send_status_msg(sock, ret_code); | |
1198 | if (ret < 0) { | |
1199 | /* Somehow, the session daemon is not responding anymore. */ | |
1200 | goto end_nosignal; | |
1201 | } | |
1202 | ||
1203 | break; | |
1204 | } | |
d2956687 | 1205 | case LTTNG_CONSUMER_INIT: |
00fb02ac | 1206 | { |
d2956687 JG |
1207 | ret_code = lttng_consumer_init_command(ctx, |
1208 | msg.u.init.sessiond_uuid); | |
00fb02ac | 1209 | health_code_update(); |
00fb02ac JD |
1210 | ret = consumer_send_status_msg(sock, ret_code); |
1211 | if (ret < 0) { | |
1212 | /* Somehow, the session daemon is not responding anymore. */ | |
1213 | goto end_nosignal; | |
1214 | } | |
1215 | break; | |
1216 | } | |
d2956687 | 1217 | case LTTNG_CONSUMER_CREATE_TRACE_CHUNK: |
d88744a4 | 1218 | { |
d2956687 | 1219 | const struct lttng_credentials credentials = { |
e5add6d0 JG |
1220 | .uid = msg.u.create_trace_chunk.credentials.value.uid, |
1221 | .gid = msg.u.create_trace_chunk.credentials.value.gid, | |
d2956687 JG |
1222 | }; |
1223 | const bool is_local_trace = | |
1224 | !msg.u.create_trace_chunk.relayd_id.is_set; | |
1225 | const uint64_t relayd_id = | |
1226 | msg.u.create_trace_chunk.relayd_id.value; | |
1227 | const char *chunk_override_name = | |
1228 | *msg.u.create_trace_chunk.override_name ? | |
1229 | msg.u.create_trace_chunk.override_name : | |
1230 | NULL; | |
cbf53d23 | 1231 | struct lttng_directory_handle *chunk_directory_handle = NULL; |
d88744a4 | 1232 | |
d2956687 JG |
1233 | /* |
1234 | * The session daemon will only provide a chunk directory file | |
1235 | * descriptor for local traces. | |
1236 | */ | |
1237 | if (is_local_trace) { | |
1238 | int chunk_dirfd; | |
19990ed5 | 1239 | |
d2956687 JG |
1240 | /* Acnowledge the reception of the command. */ |
1241 | ret = consumer_send_status_msg(sock, | |
1242 | LTTCOMM_CONSUMERD_SUCCESS); | |
1243 | if (ret < 0) { | |
1244 | /* Somehow, the session daemon is not responding anymore. */ | |
1245 | goto end_nosignal; | |
1246 | } | |
92816cc3 | 1247 | |
d2956687 JG |
1248 | ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1); |
1249 | if (ret != sizeof(chunk_dirfd)) { | |
1250 | ERR("Failed to receive trace chunk directory file descriptor"); | |
1251 | goto error_fatal; | |
1252 | } | |
92816cc3 | 1253 | |
d2956687 JG |
1254 | DBG("Received trace chunk directory fd (%d)", |
1255 | chunk_dirfd); | |
cbf53d23 | 1256 | chunk_directory_handle = lttng_directory_handle_create_from_dirfd( |
d2956687 | 1257 | chunk_dirfd); |
cbf53d23 | 1258 | if (!chunk_directory_handle) { |
d2956687 JG |
1259 | ERR("Failed to initialize chunk directory handle from directory file descriptor"); |
1260 | if (close(chunk_dirfd)) { | |
1261 | PERROR("Failed to close chunk directory file descriptor"); | |
1262 | } | |
1263 | goto error_fatal; | |
1264 | } | |
92816cc3 JG |
1265 | } |
1266 | ||
d2956687 JG |
1267 | ret_code = lttng_consumer_create_trace_chunk( |
1268 | !is_local_trace ? &relayd_id : NULL, | |
1269 | msg.u.create_trace_chunk.session_id, | |
1270 | msg.u.create_trace_chunk.chunk_id, | |
e5add6d0 JG |
1271 | (time_t) msg.u.create_trace_chunk |
1272 | .creation_timestamp, | |
d2956687 | 1273 | chunk_override_name, |
e5add6d0 JG |
1274 | msg.u.create_trace_chunk.credentials.is_set ? |
1275 | &credentials : | |
1276 | NULL, | |
cbf53d23 JG |
1277 | chunk_directory_handle); |
1278 | lttng_directory_handle_put(chunk_directory_handle); | |
d2956687 | 1279 | goto end_msg_sessiond; |
d88744a4 | 1280 | } |
d2956687 | 1281 | case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK: |
a1ae2ea5 | 1282 | { |
bbc4768c JG |
1283 | enum lttng_trace_chunk_command_type close_command = |
1284 | msg.u.close_trace_chunk.close_command.value; | |
d2956687 JG |
1285 | const uint64_t relayd_id = |
1286 | msg.u.close_trace_chunk.relayd_id.value; | |
ecd1a12f MD |
1287 | struct lttcomm_consumer_close_trace_chunk_reply reply; |
1288 | char path[LTTNG_PATH_MAX]; | |
d2956687 JG |
1289 | |
1290 | ret_code = lttng_consumer_close_trace_chunk( | |
1291 | msg.u.close_trace_chunk.relayd_id.is_set ? | |
bbc4768c JG |
1292 | &relayd_id : |
1293 | NULL, | |
d2956687 JG |
1294 | msg.u.close_trace_chunk.session_id, |
1295 | msg.u.close_trace_chunk.chunk_id, | |
bbc4768c JG |
1296 | (time_t) msg.u.close_trace_chunk.close_timestamp, |
1297 | msg.u.close_trace_chunk.close_command.is_set ? | |
1298 | &close_command : | |
ecd1a12f MD |
1299 | NULL, path); |
1300 | reply.ret_code = ret_code; | |
1301 | reply.path_length = strlen(path) + 1; | |
1302 | ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply)); | |
1303 | if (ret != sizeof(reply)) { | |
1304 | goto error_fatal; | |
1305 | } | |
1306 | ret = lttcomm_send_unix_sock(sock, path, reply.path_length); | |
1307 | if (ret != reply.path_length) { | |
1308 | goto error_fatal; | |
1309 | } | |
1310 | goto end_nosignal; | |
3654ed19 | 1311 | } |
d2956687 | 1312 | case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS: |
3654ed19 | 1313 | { |
d2956687 JG |
1314 | const uint64_t relayd_id = |
1315 | msg.u.trace_chunk_exists.relayd_id.value; | |
1316 | ||
1317 | ret_code = lttng_consumer_trace_chunk_exists( | |
1318 | msg.u.trace_chunk_exists.relayd_id.is_set ? | |
1319 | &relayd_id : NULL, | |
1320 | msg.u.trace_chunk_exists.session_id, | |
1321 | msg.u.trace_chunk_exists.chunk_id); | |
1322 | goto end_msg_sessiond; | |
a1ae2ea5 | 1323 | } |
3bd1e081 | 1324 | default: |
3f8e211f | 1325 | goto end_nosignal; |
3bd1e081 | 1326 | } |
3f8e211f | 1327 | |
3bd1e081 | 1328 | end_nosignal: |
4cbc1a04 DG |
1329 | /* |
1330 | * Return 1 to indicate success since the 0 value can be a socket | |
1331 | * shutdown during the recv() or send() call. | |
1332 | */ | |
c5c7998f JG |
1333 | ret = 1; |
1334 | goto end; | |
1335 | error_fatal: | |
1336 | /* This will issue a consumer stop. */ | |
1337 | ret = -1; | |
1338 | goto end; | |
d2956687 JG |
1339 | end_msg_sessiond: |
1340 | /* | |
1341 | * The returned value here is not useful since either way we'll return 1 to | |
1342 | * the caller because the session daemon socket management is done | |
1343 | * elsewhere. Returning a negative code or 0 will shutdown the consumer. | |
1344 | */ | |
1345 | ret = consumer_send_status_msg(sock, ret_code); | |
1346 | if (ret < 0) { | |
1347 | goto error_fatal; | |
1348 | } | |
c5c7998f JG |
1349 | ret = 1; |
1350 | end: | |
d2956687 | 1351 | health_code_update(); |
1803a064 | 1352 | rcu_read_unlock(); |
c5c7998f | 1353 | return ret; |
3bd1e081 | 1354 | } |
d41f73b7 | 1355 | |
309167d2 JD |
1356 | /* |
1357 | * Populate index values of a kernel stream. Values are set in big endian order. | |
1358 | * | |
1359 | * Return 0 on success or else a negative value. | |
1360 | */ | |
50adc264 | 1361 | static int get_index_values(struct ctf_packet_index *index, int infd) |
309167d2 JD |
1362 | { |
1363 | int ret; | |
85d1db9f JG |
1364 | uint64_t packet_size, content_size, timestamp_begin, timestamp_end, |
1365 | events_discarded, stream_id, stream_instance_id, | |
1366 | packet_seq_num; | |
309167d2 | 1367 | |
85d1db9f | 1368 | ret = kernctl_get_timestamp_begin(infd, ×tamp_begin); |
309167d2 JD |
1369 | if (ret < 0) { |
1370 | PERROR("kernctl_get_timestamp_begin"); | |
1371 | goto error; | |
1372 | } | |
309167d2 | 1373 | |
85d1db9f | 1374 | ret = kernctl_get_timestamp_end(infd, ×tamp_end); |
309167d2 JD |
1375 | if (ret < 0) { |
1376 | PERROR("kernctl_get_timestamp_end"); | |
1377 | goto error; | |
1378 | } | |
309167d2 | 1379 | |
85d1db9f | 1380 | ret = kernctl_get_events_discarded(infd, &events_discarded); |
309167d2 JD |
1381 | if (ret < 0) { |
1382 | PERROR("kernctl_get_events_discarded"); | |
1383 | goto error; | |
1384 | } | |
309167d2 | 1385 | |
85d1db9f | 1386 | ret = kernctl_get_content_size(infd, &content_size); |
309167d2 JD |
1387 | if (ret < 0) { |
1388 | PERROR("kernctl_get_content_size"); | |
1389 | goto error; | |
1390 | } | |
309167d2 | 1391 | |
85d1db9f | 1392 | ret = kernctl_get_packet_size(infd, &packet_size); |
309167d2 JD |
1393 | if (ret < 0) { |
1394 | PERROR("kernctl_get_packet_size"); | |
1395 | goto error; | |
1396 | } | |
309167d2 | 1397 | |
85d1db9f | 1398 | ret = kernctl_get_stream_id(infd, &stream_id); |
309167d2 JD |
1399 | if (ret < 0) { |
1400 | PERROR("kernctl_get_stream_id"); | |
1401 | goto error; | |
1402 | } | |
309167d2 | 1403 | |
85d1db9f | 1404 | ret = kernctl_get_instance_id(infd, &stream_instance_id); |
234cd636 | 1405 | if (ret < 0) { |
f0b03c22 MD |
1406 | if (ret == -ENOTTY) { |
1407 | /* Command not implemented by lttng-modules. */ | |
85d1db9f | 1408 | stream_instance_id = -1ULL; |
f0b03c22 MD |
1409 | } else { |
1410 | PERROR("kernctl_get_instance_id"); | |
1411 | goto error; | |
1412 | } | |
234cd636 | 1413 | } |
234cd636 | 1414 | |
85d1db9f | 1415 | ret = kernctl_get_sequence_number(infd, &packet_seq_num); |
234cd636 | 1416 | if (ret < 0) { |
f0b03c22 MD |
1417 | if (ret == -ENOTTY) { |
1418 | /* Command not implemented by lttng-modules. */ | |
85d1db9f | 1419 | packet_seq_num = -1ULL; |
f0b03c22 MD |
1420 | ret = 0; |
1421 | } else { | |
1422 | PERROR("kernctl_get_sequence_number"); | |
1423 | goto error; | |
1424 | } | |
234cd636 JD |
1425 | } |
1426 | index->packet_seq_num = htobe64(index->packet_seq_num); | |
1427 | ||
85d1db9f JG |
1428 | *index = (typeof(*index)) { |
1429 | .offset = index->offset, | |
1430 | .packet_size = htobe64(packet_size), | |
1431 | .content_size = htobe64(content_size), | |
1432 | .timestamp_begin = htobe64(timestamp_begin), | |
1433 | .timestamp_end = htobe64(timestamp_end), | |
1434 | .events_discarded = htobe64(events_discarded), | |
1435 | .stream_id = htobe64(stream_id), | |
1436 | .stream_instance_id = htobe64(stream_instance_id), | |
1437 | .packet_seq_num = htobe64(packet_seq_num), | |
1438 | }; | |
1439 | ||
309167d2 JD |
1440 | error: |
1441 | return ret; | |
1442 | } | |
94d49140 JD |
1443 | /* |
1444 | * Sync metadata meaning request them to the session daemon and snapshot to the | |
1445 | * metadata thread can consumer them. | |
1446 | * | |
1447 | * Metadata stream lock MUST be acquired. | |
1448 | * | |
1449 | * Return 0 if new metadatda is available, EAGAIN if the metadata stream | |
1450 | * is empty or a negative value on error. | |
1451 | */ | |
1452 | int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata) | |
1453 | { | |
1454 | int ret; | |
1455 | ||
1456 | assert(metadata); | |
1457 | ||
1458 | ret = kernctl_buffer_flush(metadata->wait_fd); | |
1459 | if (ret < 0) { | |
1460 | ERR("Failed to flush kernel stream"); | |
1461 | goto end; | |
1462 | } | |
1463 | ||
1464 | ret = kernctl_snapshot(metadata->wait_fd); | |
1465 | if (ret < 0) { | |
32af2c95 | 1466 | if (ret != -EAGAIN) { |
94d49140 JD |
1467 | ERR("Sync metadata, taking kernel snapshot failed."); |
1468 | goto end; | |
1469 | } | |
1470 | DBG("Sync metadata, no new kernel metadata"); | |
1471 | /* No new metadata, exit. */ | |
1472 | ret = ENODATA; | |
1473 | goto end; | |
1474 | } | |
1475 | ||
1476 | end: | |
1477 | return ret; | |
1478 | } | |
309167d2 | 1479 | |
fb83fe64 JD |
1480 | static |
1481 | int update_stream_stats(struct lttng_consumer_stream *stream) | |
1482 | { | |
1483 | int ret; | |
1484 | uint64_t seq, discarded; | |
1485 | ||
1486 | ret = kernctl_get_sequence_number(stream->wait_fd, &seq); | |
1487 | if (ret < 0) { | |
f0b03c22 MD |
1488 | if (ret == -ENOTTY) { |
1489 | /* Command not implemented by lttng-modules. */ | |
1490 | seq = -1ULL; | |
a40a503f | 1491 | stream->sequence_number_unavailable = true; |
f0b03c22 MD |
1492 | } else { |
1493 | PERROR("kernctl_get_sequence_number"); | |
1494 | goto end; | |
1495 | } | |
fb83fe64 JD |
1496 | } |
1497 | ||
1498 | /* | |
1499 | * Start the sequence when we extract the first packet in case we don't | |
1500 | * start at 0 (for example if a consumer is not connected to the | |
1501 | * session immediately after the beginning). | |
1502 | */ | |
1503 | if (stream->last_sequence_number == -1ULL) { | |
1504 | stream->last_sequence_number = seq; | |
1505 | } else if (seq > stream->last_sequence_number) { | |
1506 | stream->chan->lost_packets += seq - | |
1507 | stream->last_sequence_number - 1; | |
1508 | } else { | |
1509 | /* seq <= last_sequence_number */ | |
1510 | ERR("Sequence number inconsistent : prev = %" PRIu64 | |
1511 | ", current = %" PRIu64, | |
1512 | stream->last_sequence_number, seq); | |
1513 | ret = -1; | |
1514 | goto end; | |
1515 | } | |
1516 | stream->last_sequence_number = seq; | |
1517 | ||
1518 | ret = kernctl_get_events_discarded(stream->wait_fd, &discarded); | |
1519 | if (ret < 0) { | |
1520 | PERROR("kernctl_get_events_discarded"); | |
1521 | goto end; | |
1522 | } | |
1523 | if (discarded < stream->last_discarded_events) { | |
1524 | /* | |
83f4233d MJ |
1525 | * Overflow has occurred. We assume only one wrap-around |
1526 | * has occurred. | |
fb83fe64 JD |
1527 | */ |
1528 | stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) - | |
1529 | stream->last_discarded_events + discarded; | |
1530 | } else { | |
1531 | stream->chan->discarded_events += discarded - | |
1532 | stream->last_discarded_events; | |
1533 | } | |
1534 | stream->last_discarded_events = discarded; | |
1535 | ret = 0; | |
1536 | ||
1537 | end: | |
1538 | return ret; | |
1539 | } | |
1540 | ||
93ec662e JD |
1541 | /* |
1542 | * Check if the local version of the metadata stream matches with the version | |
1543 | * of the metadata stream in the kernel. If it was updated, set the reset flag | |
1544 | * on the stream. | |
1545 | */ | |
1546 | static | |
1547 | int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream) | |
1548 | { | |
1549 | int ret; | |
1550 | uint64_t cur_version; | |
1551 | ||
1552 | ret = kernctl_get_metadata_version(infd, &cur_version); | |
1553 | if (ret < 0) { | |
f0b03c22 MD |
1554 | if (ret == -ENOTTY) { |
1555 | /* | |
1556 | * LTTng-modules does not implement this | |
1557 | * command. | |
1558 | */ | |
1559 | ret = 0; | |
1560 | goto end; | |
1561 | } | |
93ec662e JD |
1562 | ERR("Failed to get the metadata version"); |
1563 | goto end; | |
1564 | } | |
1565 | ||
1566 | if (stream->metadata_version == cur_version) { | |
1567 | ret = 0; | |
1568 | goto end; | |
1569 | } | |
1570 | ||
1571 | DBG("New metadata version detected"); | |
1572 | stream->metadata_version = cur_version; | |
1573 | stream->reset_metadata_flag = 1; | |
1574 | ret = 0; | |
1575 | ||
1576 | end: | |
1577 | return ret; | |
1578 | } | |
1579 | ||
d41f73b7 MD |
1580 | /* |
1581 | * Consume data on a file descriptor and write it on a trace file. | |
d2956687 | 1582 | * The stream and channel locks must be held by the caller. |
d41f73b7 | 1583 | */ |
4078b776 | 1584 | ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d2956687 | 1585 | struct lttng_consumer_local_data *ctx) |
d41f73b7 | 1586 | { |
1d4dfdef | 1587 | unsigned long len, subbuf_size, padding; |
02d02e31 | 1588 | int err, write_index = 1, rotation_ret; |
4078b776 | 1589 | ssize_t ret = 0; |
d41f73b7 | 1590 | int infd = stream->wait_fd; |
d3faab47 | 1591 | struct ctf_packet_index index = {}; |
d41f73b7 MD |
1592 | |
1593 | DBG("In read_subbuffer (infd : %d)", infd); | |
309167d2 | 1594 | |
02d02e31 JD |
1595 | /* |
1596 | * If the stream was flagged to be ready for rotation before we extract the | |
1597 | * next packet, rotate it now. | |
1598 | */ | |
1599 | if (stream->rotate_ready) { | |
1600 | DBG("Rotate stream before extracting data"); | |
d2956687 | 1601 | rotation_ret = lttng_consumer_rotate_stream(ctx, stream); |
02d02e31 JD |
1602 | if (rotation_ret < 0) { |
1603 | ERR("Stream rotation error"); | |
1604 | ret = -1; | |
1605 | goto error; | |
1606 | } | |
1607 | } | |
1608 | ||
d41f73b7 MD |
1609 | /* Get the next subbuffer */ |
1610 | err = kernctl_get_next_subbuf(infd); | |
1611 | if (err != 0) { | |
d41f73b7 MD |
1612 | /* |
1613 | * This is a debug message even for single-threaded consumer, | |
1614 | * because poll() have more relaxed criterions than get subbuf, | |
1615 | * so get_subbuf may fail for short race windows where poll() | |
1616 | * would issue wakeups. | |
1617 | */ | |
1618 | DBG("Reserving sub buffer failed (everything is normal, " | |
1619 | "it is due to concurrency)"); | |
32af2c95 | 1620 | ret = err; |
02d02e31 | 1621 | goto error; |
d41f73b7 MD |
1622 | } |
1623 | ||
1d4dfdef DG |
1624 | /* Get the full subbuffer size including padding */ |
1625 | err = kernctl_get_padded_subbuf_size(infd, &len); | |
1626 | if (err != 0) { | |
5a510c9f | 1627 | PERROR("Getting sub-buffer len failed."); |
8265f19e MD |
1628 | err = kernctl_put_subbuf(infd); |
1629 | if (err != 0) { | |
32af2c95 | 1630 | if (err == -EFAULT) { |
5a510c9f | 1631 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1632 | } else if (err == -EIO) { |
8265f19e | 1633 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1634 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
8265f19e | 1635 | } |
32af2c95 | 1636 | ret = err; |
02d02e31 | 1637 | goto error; |
8265f19e | 1638 | } |
32af2c95 | 1639 | ret = err; |
02d02e31 | 1640 | goto error; |
1d4dfdef DG |
1641 | } |
1642 | ||
1c20f0e2 | 1643 | if (!stream->metadata_flag) { |
309167d2 JD |
1644 | ret = get_index_values(&index, infd); |
1645 | if (ret < 0) { | |
8265f19e MD |
1646 | err = kernctl_put_subbuf(infd); |
1647 | if (err != 0) { | |
32af2c95 | 1648 | if (err == -EFAULT) { |
5a510c9f | 1649 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1650 | } else if (err == -EIO) { |
8265f19e | 1651 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1652 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
8265f19e | 1653 | } |
32af2c95 | 1654 | ret = err; |
02d02e31 | 1655 | goto error; |
8265f19e | 1656 | } |
02d02e31 | 1657 | goto error; |
309167d2 | 1658 | } |
fb83fe64 JD |
1659 | ret = update_stream_stats(stream); |
1660 | if (ret < 0) { | |
7b87473d MD |
1661 | err = kernctl_put_subbuf(infd); |
1662 | if (err != 0) { | |
1663 | if (err == -EFAULT) { | |
1664 | PERROR("Error in unreserving sub buffer\n"); | |
1665 | } else if (err == -EIO) { | |
1666 | /* Should never happen with newer LTTng versions */ | |
1667 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); | |
1668 | } | |
1669 | ret = err; | |
02d02e31 | 1670 | goto error; |
7b87473d | 1671 | } |
02d02e31 | 1672 | goto error; |
fb83fe64 | 1673 | } |
1c20f0e2 JD |
1674 | } else { |
1675 | write_index = 0; | |
93ec662e JD |
1676 | ret = metadata_stream_check_version(infd, stream); |
1677 | if (ret < 0) { | |
7b87473d MD |
1678 | err = kernctl_put_subbuf(infd); |
1679 | if (err != 0) { | |
1680 | if (err == -EFAULT) { | |
1681 | PERROR("Error in unreserving sub buffer\n"); | |
1682 | } else if (err == -EIO) { | |
1683 | /* Should never happen with newer LTTng versions */ | |
1684 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); | |
1685 | } | |
1686 | ret = err; | |
02d02e31 | 1687 | goto error; |
7b87473d | 1688 | } |
02d02e31 | 1689 | goto error; |
93ec662e | 1690 | } |
309167d2 JD |
1691 | } |
1692 | ||
ffe60014 | 1693 | switch (stream->chan->output) { |
07b86b52 | 1694 | case CONSUMER_CHANNEL_SPLICE: |
1d4dfdef DG |
1695 | /* |
1696 | * XXX: The lttng-modules splice "actor" does not handle copying | |
1697 | * partial pages hence only using the subbuffer size without the | |
1698 | * padding makes the splice fail. | |
1699 | */ | |
1700 | subbuf_size = len; | |
1701 | padding = 0; | |
1702 | ||
1703 | /* splice the subbuffer to the tracefile */ | |
91dfef6e | 1704 | ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size, |
309167d2 | 1705 | padding, &index); |
91dfef6e DG |
1706 | /* |
1707 | * XXX: Splice does not support network streaming so the return value | |
1708 | * is simply checked against subbuf_size and not like the mmap() op. | |
1709 | */ | |
1d4dfdef DG |
1710 | if (ret != subbuf_size) { |
1711 | /* | |
1712 | * display the error but continue processing to try | |
1713 | * to release the subbuffer | |
1714 | */ | |
1715 | ERR("Error splicing to tracefile (ret: %zd != len: %lu)", | |
1716 | ret, subbuf_size); | |
309167d2 | 1717 | write_index = 0; |
1d4dfdef DG |
1718 | } |
1719 | break; | |
07b86b52 | 1720 | case CONSUMER_CHANNEL_MMAP: |
276cd1d7 JG |
1721 | { |
1722 | const char *subbuf_addr; | |
1723 | ||
1d4dfdef DG |
1724 | /* Get subbuffer size without padding */ |
1725 | err = kernctl_get_subbuf_size(infd, &subbuf_size); | |
1726 | if (err != 0) { | |
5a510c9f | 1727 | PERROR("Getting sub-buffer len failed."); |
8265f19e MD |
1728 | err = kernctl_put_subbuf(infd); |
1729 | if (err != 0) { | |
32af2c95 | 1730 | if (err == -EFAULT) { |
5a510c9f | 1731 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1732 | } else if (err == -EIO) { |
8265f19e | 1733 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1734 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
8265f19e | 1735 | } |
32af2c95 | 1736 | ret = err; |
02d02e31 | 1737 | goto error; |
8265f19e | 1738 | } |
32af2c95 | 1739 | ret = err; |
02d02e31 | 1740 | goto error; |
1d4dfdef | 1741 | } |
47e81c02 | 1742 | |
276cd1d7 JG |
1743 | ret = get_current_subbuf_addr(stream, &subbuf_addr); |
1744 | if (ret) { | |
1745 | goto error_put_subbuf; | |
1746 | } | |
1747 | ||
1d4dfdef DG |
1748 | /* Make sure the tracer is not gone mad on us! */ |
1749 | assert(len >= subbuf_size); | |
1750 | ||
1751 | padding = len - subbuf_size; | |
1752 | ||
1753 | /* write the subbuffer to the tracefile */ | |
276cd1d7 JG |
1754 | ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, |
1755 | subbuf_addr, | |
1756 | subbuf_size, | |
309167d2 | 1757 | padding, &index); |
91dfef6e DG |
1758 | /* |
1759 | * The mmap operation should write subbuf_size amount of data when | |
1760 | * network streaming or the full padding (len) size when we are _not_ | |
1761 | * streaming. | |
1762 | */ | |
d88aee68 DG |
1763 | if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) || |
1764 | (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) { | |
1d4dfdef | 1765 | /* |
91dfef6e | 1766 | * Display the error but continue processing to try to release the |
2336629e DG |
1767 | * subbuffer. This is a DBG statement since this is possible to |
1768 | * happen without being a critical error. | |
1d4dfdef | 1769 | */ |
2336629e | 1770 | DBG("Error writing to tracefile " |
91dfef6e DG |
1771 | "(ret: %zd != len: %lu != subbuf_size: %lu)", |
1772 | ret, len, subbuf_size); | |
309167d2 | 1773 | write_index = 0; |
1d4dfdef DG |
1774 | } |
1775 | break; | |
276cd1d7 | 1776 | } |
1d4dfdef DG |
1777 | default: |
1778 | ERR("Unknown output method"); | |
56591bac | 1779 | ret = -EPERM; |
d41f73b7 | 1780 | } |
276cd1d7 | 1781 | error_put_subbuf: |
d41f73b7 MD |
1782 | err = kernctl_put_next_subbuf(infd); |
1783 | if (err != 0) { | |
32af2c95 | 1784 | if (err == -EFAULT) { |
5a510c9f | 1785 | PERROR("Error in unreserving sub buffer\n"); |
32af2c95 | 1786 | } else if (err == -EIO) { |
d41f73b7 | 1787 | /* Should never happen with newer LTTng versions */ |
5a510c9f | 1788 | PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); |
d41f73b7 | 1789 | } |
32af2c95 | 1790 | ret = err; |
02d02e31 | 1791 | goto error; |
d41f73b7 MD |
1792 | } |
1793 | ||
309167d2 | 1794 | /* Write index if needed. */ |
1c20f0e2 | 1795 | if (!write_index) { |
02d02e31 | 1796 | goto rotate; |
1c20f0e2 JD |
1797 | } |
1798 | ||
94d49140 JD |
1799 | if (stream->chan->live_timer_interval && !stream->metadata_flag) { |
1800 | /* | |
1801 | * In live, block until all the metadata is sent. | |
1802 | */ | |
c585821b MD |
1803 | pthread_mutex_lock(&stream->metadata_timer_lock); |
1804 | assert(!stream->missed_metadata_flush); | |
1805 | stream->waiting_on_metadata = true; | |
1806 | pthread_mutex_unlock(&stream->metadata_timer_lock); | |
1807 | ||
94d49140 | 1808 | err = consumer_stream_sync_metadata(ctx, stream->session_id); |
c585821b MD |
1809 | |
1810 | pthread_mutex_lock(&stream->metadata_timer_lock); | |
1811 | stream->waiting_on_metadata = false; | |
1812 | if (stream->missed_metadata_flush) { | |
1813 | stream->missed_metadata_flush = false; | |
1814 | pthread_mutex_unlock(&stream->metadata_timer_lock); | |
1815 | (void) consumer_flush_kernel_index(stream); | |
1816 | } else { | |
1817 | pthread_mutex_unlock(&stream->metadata_timer_lock); | |
1818 | } | |
94d49140 | 1819 | if (err < 0) { |
02d02e31 | 1820 | goto error; |
94d49140 JD |
1821 | } |
1822 | } | |
1823 | ||
1c20f0e2 JD |
1824 | err = consumer_stream_write_index(stream, &index); |
1825 | if (err < 0) { | |
02d02e31 | 1826 | goto error; |
309167d2 JD |
1827 | } |
1828 | ||
02d02e31 JD |
1829 | rotate: |
1830 | /* | |
1831 | * After extracting the packet, we check if the stream is now ready to be | |
1832 | * rotated and perform the action immediately. | |
1833 | */ | |
1834 | rotation_ret = lttng_consumer_stream_is_rotate_ready(stream); | |
1835 | if (rotation_ret == 1) { | |
d2956687 | 1836 | rotation_ret = lttng_consumer_rotate_stream(ctx, stream); |
02d02e31 JD |
1837 | if (rotation_ret < 0) { |
1838 | ERR("Stream rotation error"); | |
1839 | ret = -1; | |
1840 | goto error; | |
1841 | } | |
1842 | } else if (rotation_ret < 0) { | |
1843 | ERR("Checking if stream is ready to rotate"); | |
1844 | ret = -1; | |
1845 | goto error; | |
1846 | } | |
1847 | ||
1848 | error: | |
d41f73b7 MD |
1849 | return ret; |
1850 | } | |
1851 | ||
1852 | int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
1853 | { | |
1854 | int ret; | |
ffe60014 DG |
1855 | |
1856 | assert(stream); | |
1857 | ||
2bba9e53 | 1858 | /* |
d2956687 JG |
1859 | * Don't create anything if this is set for streaming or if there is |
1860 | * no current trace chunk on the parent channel. | |
2bba9e53 | 1861 | */ |
d2956687 JG |
1862 | if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor && |
1863 | stream->chan->trace_chunk) { | |
1864 | ret = consumer_stream_create_output_files(stream, true); | |
1865 | if (ret) { | |
fe4477ee JD |
1866 | goto error; |
1867 | } | |
ffe60014 | 1868 | } |
d41f73b7 | 1869 | |
d41f73b7 MD |
1870 | if (stream->output == LTTNG_EVENT_MMAP) { |
1871 | /* get the len of the mmap region */ | |
1872 | unsigned long mmap_len; | |
1873 | ||
1874 | ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); | |
1875 | if (ret != 0) { | |
ffe60014 | 1876 | PERROR("kernctl_get_mmap_len"); |
d41f73b7 MD |
1877 | goto error_close_fd; |
1878 | } | |
1879 | stream->mmap_len = (size_t) mmap_len; | |
1880 | ||
ffe60014 DG |
1881 | stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ, |
1882 | MAP_PRIVATE, stream->wait_fd, 0); | |
d41f73b7 | 1883 | if (stream->mmap_base == MAP_FAILED) { |
ffe60014 | 1884 | PERROR("Error mmaping"); |
d41f73b7 MD |
1885 | ret = -1; |
1886 | goto error_close_fd; | |
1887 | } | |
1888 | } | |
1889 | ||
1890 | /* we return 0 to let the library handle the FD internally */ | |
1891 | return 0; | |
1892 | ||
1893 | error_close_fd: | |
2f225ce2 | 1894 | if (stream->out_fd >= 0) { |
d41f73b7 MD |
1895 | int err; |
1896 | ||
1897 | err = close(stream->out_fd); | |
1898 | assert(!err); | |
2f225ce2 | 1899 | stream->out_fd = -1; |
d41f73b7 MD |
1900 | } |
1901 | error: | |
1902 | return ret; | |
1903 | } | |
1904 | ||
ca22feea DG |
1905 | /* |
1906 | * Check if data is still being extracted from the buffers for a specific | |
4e9a4686 DG |
1907 | * stream. Consumer data lock MUST be acquired before calling this function |
1908 | * and the stream lock. | |
ca22feea | 1909 | * |
6d805429 | 1910 | * Return 1 if the traced data are still getting read else 0 meaning that the |
ca22feea DG |
1911 | * data is available for trace viewer reading. |
1912 | */ | |
6d805429 | 1913 | int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream) |
ca22feea DG |
1914 | { |
1915 | int ret; | |
1916 | ||
1917 | assert(stream); | |
1918 | ||
873b9e9a MD |
1919 | if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) { |
1920 | ret = 0; | |
1921 | goto end; | |
1922 | } | |
1923 | ||
ca22feea DG |
1924 | ret = kernctl_get_next_subbuf(stream->wait_fd); |
1925 | if (ret == 0) { | |
1926 | /* There is still data so let's put back this subbuffer. */ | |
1927 | ret = kernctl_put_subbuf(stream->wait_fd); | |
1928 | assert(ret == 0); | |
6d805429 | 1929 | ret = 1; /* Data is pending */ |
4e9a4686 | 1930 | goto end; |
ca22feea DG |
1931 | } |
1932 | ||
6d805429 DG |
1933 | /* Data is NOT pending and ready to be read. */ |
1934 | ret = 0; | |
ca22feea | 1935 | |
6efae65e DG |
1936 | end: |
1937 | return ret; | |
ca22feea | 1938 | } |