Commit | Line | Data |
---|---|---|
3bd1e081 MD |
1 | /* |
2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> | |
3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
4 | * | |
d14d33bf AM |
5 | * This program is free software; you can redistribute it and/or modify |
6 | * it under the terms of the GNU General Public License, version 2 only, | |
7 | * as published by the Free Software Foundation. | |
3bd1e081 MD |
8 | * |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
d14d33bf AM |
14 | * You should have received a copy of the GNU General Public License along |
15 | * with this program; if not, write to the Free Software Foundation, Inc., | |
16 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
3bd1e081 MD |
17 | */ |
18 | ||
19 | #define _GNU_SOURCE | |
20 | #include <assert.h> | |
3bd1e081 MD |
21 | #include <poll.h> |
22 | #include <pthread.h> | |
23 | #include <stdlib.h> | |
24 | #include <string.h> | |
25 | #include <sys/mman.h> | |
26 | #include <sys/socket.h> | |
27 | #include <sys/types.h> | |
77c7c900 | 28 | #include <inttypes.h> |
3bd1e081 | 29 | #include <unistd.h> |
dbb5dfe6 | 30 | #include <sys/stat.h> |
3bd1e081 | 31 | |
990570ed | 32 | #include <common/common.h> |
10a8a223 | 33 | #include <common/kernel-ctl/kernel-ctl.h> |
10a8a223 | 34 | #include <common/sessiond-comm/sessiond-comm.h> |
00e2e675 | 35 | #include <common/sessiond-comm/relayd.h> |
dbb5dfe6 | 36 | #include <common/compat/fcntl.h> |
00e2e675 | 37 | #include <common/relayd/relayd.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; | |
43 | extern volatile int consumer_quit; | |
44 | ||
3bd1e081 MD |
45 | /* |
46 | * Take a snapshot for a specific fd | |
47 | * | |
48 | * Returns 0 on success, < 0 on error | |
49 | */ | |
50 | int lttng_kconsumer_take_snapshot(struct lttng_consumer_local_data *ctx, | |
51 | struct lttng_consumer_stream *stream) | |
52 | { | |
53 | int ret = 0; | |
54 | int infd = stream->wait_fd; | |
55 | ||
56 | ret = kernctl_snapshot(infd); | |
57 | if (ret != 0) { | |
87dc6a9c | 58 | errno = -ret; |
3bd1e081 MD |
59 | perror("Getting sub-buffer snapshot."); |
60 | } | |
61 | ||
62 | return ret; | |
63 | } | |
64 | ||
65 | /* | |
66 | * Get the produced position | |
67 | * | |
68 | * Returns 0 on success, < 0 on error | |
69 | */ | |
70 | int lttng_kconsumer_get_produced_snapshot( | |
71 | struct lttng_consumer_local_data *ctx, | |
72 | struct lttng_consumer_stream *stream, | |
73 | unsigned long *pos) | |
74 | { | |
75 | int ret; | |
76 | int infd = stream->wait_fd; | |
77 | ||
78 | ret = kernctl_snapshot_get_produced(infd, pos); | |
79 | if (ret != 0) { | |
87dc6a9c | 80 | errno = -ret; |
3bd1e081 MD |
81 | perror("kernctl_snapshot_get_produced"); |
82 | } | |
83 | ||
84 | return ret; | |
85 | } | |
86 | ||
87 | int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
88 | int sock, struct pollfd *consumer_sockpoll) | |
89 | { | |
90 | ssize_t ret; | |
91 | struct lttcomm_consumer_msg msg; | |
92 | ||
93 | ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); | |
94 | if (ret != sizeof(msg)) { | |
f73fabfd | 95 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); |
3bd1e081 MD |
96 | return ret; |
97 | } | |
98 | if (msg.cmd_type == LTTNG_CONSUMER_STOP) { | |
99 | return -ENOENT; | |
100 | } | |
101 | ||
b0b335c8 MD |
102 | /* relayd needs RCU read-side protection */ |
103 | rcu_read_lock(); | |
104 | ||
3bd1e081 | 105 | switch (msg.cmd_type) { |
00e2e675 DG |
106 | case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: |
107 | { | |
7735ef9e DG |
108 | ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index, |
109 | msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, | |
110 | &msg.u.relayd_sock.sock); | |
00e2e675 DG |
111 | goto end_nosignal; |
112 | } | |
3bd1e081 MD |
113 | case LTTNG_CONSUMER_ADD_CHANNEL: |
114 | { | |
115 | struct lttng_consumer_channel *new_channel; | |
116 | ||
117 | DBG("consumer_add_channel %d", msg.u.channel.channel_key); | |
118 | new_channel = consumer_allocate_channel(msg.u.channel.channel_key, | |
119 | -1, -1, | |
120 | msg.u.channel.mmap_len, | |
c30aaa51 MD |
121 | msg.u.channel.max_sb_size, |
122 | msg.u.channel.nb_init_streams); | |
3bd1e081 | 123 | if (new_channel == NULL) { |
f73fabfd | 124 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
3bd1e081 MD |
125 | goto end_nosignal; |
126 | } | |
127 | if (ctx->on_recv_channel != NULL) { | |
128 | ret = ctx->on_recv_channel(new_channel); | |
129 | if (ret == 0) { | |
130 | consumer_add_channel(new_channel); | |
131 | } else if (ret < 0) { | |
132 | goto end_nosignal; | |
133 | } | |
134 | } else { | |
135 | consumer_add_channel(new_channel); | |
136 | } | |
137 | goto end_nosignal; | |
138 | } | |
139 | case LTTNG_CONSUMER_ADD_STREAM: | |
140 | { | |
50f8ae69 | 141 | int fd, stream_pipe; |
00e2e675 DG |
142 | struct consumer_relayd_sock_pair *relayd = NULL; |
143 | struct lttng_consumer_stream *new_stream; | |
c80048c6 | 144 | int alloc_ret = 0; |
3bd1e081 MD |
145 | |
146 | /* block */ | |
147 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
3f8e211f | 148 | rcu_read_unlock(); |
3bd1e081 MD |
149 | return -EINTR; |
150 | } | |
00e2e675 DG |
151 | |
152 | /* Get stream file descriptor from socket */ | |
f2fc6720 MD |
153 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); |
154 | if (ret != sizeof(fd)) { | |
f73fabfd | 155 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
3f8e211f | 156 | rcu_read_unlock(); |
3bd1e081 MD |
157 | return ret; |
158 | } | |
3bd1e081 | 159 | |
3bd1e081 MD |
160 | new_stream = consumer_allocate_stream(msg.u.stream.channel_key, |
161 | msg.u.stream.stream_key, | |
f2fc6720 | 162 | fd, fd, |
3bd1e081 MD |
163 | msg.u.stream.state, |
164 | msg.u.stream.mmap_len, | |
165 | msg.u.stream.output, | |
6df2e2c9 MD |
166 | msg.u.stream.path_name, |
167 | msg.u.stream.uid, | |
00e2e675 DG |
168 | msg.u.stream.gid, |
169 | msg.u.stream.net_index, | |
c80048c6 | 170 | msg.u.stream.metadata_flag, |
53632229 | 171 | msg.u.stream.session_id, |
c80048c6 | 172 | &alloc_ret); |
3bd1e081 | 173 | if (new_stream == NULL) { |
c80048c6 MD |
174 | switch (alloc_ret) { |
175 | case -ENOMEM: | |
176 | case -EINVAL: | |
177 | default: | |
178 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); | |
179 | break; | |
180 | case -ENOENT: | |
181 | /* | |
182 | * We could not find the channel. Can happen if cpu hotplug | |
183 | * happens while tearing down. | |
184 | */ | |
185 | DBG3("Could not find channel"); | |
186 | break; | |
187 | } | |
3f8e211f | 188 | goto end_nosignal; |
3bd1e081 | 189 | } |
00e2e675 | 190 | |
fb3a43a9 DG |
191 | /* |
192 | * The buffer flush is done on the session daemon side for the kernel | |
193 | * so no need for the stream "hangup_flush_done" variable to be | |
194 | * tracked. This is important for a kernel stream since we don't rely | |
195 | * on the flush state of the stream to read data. It's not the case for | |
196 | * user space tracing. | |
197 | */ | |
198 | new_stream->hangup_flush_done = 0; | |
199 | ||
00e2e675 DG |
200 | /* The stream is not metadata. Get relayd reference if exists. */ |
201 | relayd = consumer_find_relayd(msg.u.stream.net_index); | |
202 | if (relayd != NULL) { | |
203 | /* Add stream on the relayd */ | |
204 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
205 | ret = relayd_add_stream(&relayd->control_sock, | |
206 | msg.u.stream.name, msg.u.stream.path_name, | |
207 | &new_stream->relayd_stream_id); | |
208 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
209 | if (ret < 0) { | |
e316aad5 | 210 | consumer_del_stream(new_stream, NULL); |
3f8e211f | 211 | goto end_nosignal; |
00e2e675 DG |
212 | } |
213 | } else if (msg.u.stream.net_index != -1) { | |
214 | ERR("Network sequence index %d unknown. Not adding stream.", | |
215 | msg.u.stream.net_index); | |
e316aad5 | 216 | consumer_del_stream(new_stream, NULL); |
3f8e211f | 217 | goto end_nosignal; |
00e2e675 DG |
218 | } |
219 | ||
633d0084 DG |
220 | if (ctx->on_recv_stream) { |
221 | ret = ctx->on_recv_stream(new_stream); | |
222 | if (ret < 0) { | |
e316aad5 | 223 | consumer_del_stream(new_stream, NULL); |
633d0084 | 224 | goto end_nosignal; |
fb3a43a9 | 225 | } |
633d0084 | 226 | } |
fb3a43a9 | 227 | |
50f8ae69 | 228 | /* Get the right pipe where the stream will be sent. */ |
633d0084 | 229 | if (new_stream->metadata_flag) { |
50f8ae69 | 230 | stream_pipe = ctx->consumer_metadata_pipe[1]; |
3bd1e081 | 231 | } else { |
50f8ae69 DG |
232 | stream_pipe = ctx->consumer_data_pipe[1]; |
233 | } | |
234 | ||
235 | do { | |
236 | ret = write(stream_pipe, &new_stream, sizeof(new_stream)); | |
237 | } while (ret < 0 && errno == EINTR); | |
238 | if (ret < 0) { | |
239 | PERROR("Consumer write %s stream to pipe %d", | |
240 | new_stream->metadata_flag ? "metadata" : "data", | |
241 | stream_pipe); | |
242 | consumer_del_stream(new_stream, NULL); | |
243 | goto end_nosignal; | |
3bd1e081 | 244 | } |
00e2e675 | 245 | |
50f8ae69 DG |
246 | DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64, |
247 | msg.u.stream.path_name, fd, new_stream->relayd_stream_id); | |
3bd1e081 MD |
248 | break; |
249 | } | |
250 | case LTTNG_CONSUMER_UPDATE_STREAM: | |
251 | { | |
3f8e211f DG |
252 | rcu_read_unlock(); |
253 | return -ENOSYS; | |
254 | } | |
255 | case LTTNG_CONSUMER_DESTROY_RELAYD: | |
256 | { | |
a6ba4fe1 | 257 | uint64_t index = msg.u.destroy_relayd.net_seq_idx; |
3f8e211f DG |
258 | struct consumer_relayd_sock_pair *relayd; |
259 | ||
a6ba4fe1 | 260 | DBG("Kernel consumer destroying relayd %" PRIu64, index); |
3f8e211f DG |
261 | |
262 | /* Get relayd reference if exists. */ | |
a6ba4fe1 | 263 | relayd = consumer_find_relayd(index); |
3f8e211f | 264 | if (relayd == NULL) { |
a6ba4fe1 | 265 | ERR("Unable to find relayd %" PRIu64, index); |
3f8e211f | 266 | goto end_nosignal; |
3bd1e081 | 267 | } |
3f8e211f | 268 | |
a6ba4fe1 DG |
269 | /* |
270 | * Each relayd socket pair has a refcount of stream attached to it | |
271 | * which tells if the relayd is still active or not depending on the | |
272 | * refcount value. | |
273 | * | |
274 | * This will set the destroy flag of the relayd object and destroy it | |
275 | * if the refcount reaches zero when called. | |
276 | * | |
277 | * The destroy can happen either here or when a stream fd hangs up. | |
278 | */ | |
279 | consumer_flag_relayd_for_destroy(relayd); | |
3f8e211f | 280 | |
3f8e211f | 281 | goto end_nosignal; |
3bd1e081 | 282 | } |
53632229 DG |
283 | case LTTNG_CONSUMER_DATA_AVAILABLE: |
284 | { | |
285 | rcu_read_unlock(); | |
286 | return -ENOSYS; | |
287 | } | |
3bd1e081 | 288 | default: |
3f8e211f | 289 | goto end_nosignal; |
3bd1e081 | 290 | } |
3f8e211f | 291 | |
3bd1e081 | 292 | end_nosignal: |
b0b335c8 | 293 | rcu_read_unlock(); |
4cbc1a04 DG |
294 | |
295 | /* | |
296 | * Return 1 to indicate success since the 0 value can be a socket | |
297 | * shutdown during the recv() or send() call. | |
298 | */ | |
299 | return 1; | |
3bd1e081 | 300 | } |
d41f73b7 MD |
301 | |
302 | /* | |
303 | * Consume data on a file descriptor and write it on a trace file. | |
304 | */ | |
4078b776 | 305 | ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
306 | struct lttng_consumer_local_data *ctx) |
307 | { | |
1d4dfdef | 308 | unsigned long len, subbuf_size, padding; |
d41f73b7 | 309 | int err; |
4078b776 | 310 | ssize_t ret = 0; |
d41f73b7 MD |
311 | int infd = stream->wait_fd; |
312 | ||
313 | DBG("In read_subbuffer (infd : %d)", infd); | |
314 | /* Get the next subbuffer */ | |
315 | err = kernctl_get_next_subbuf(infd); | |
316 | if (err != 0) { | |
1d4dfdef | 317 | ret = err; |
d41f73b7 MD |
318 | /* |
319 | * This is a debug message even for single-threaded consumer, | |
320 | * because poll() have more relaxed criterions than get subbuf, | |
321 | * so get_subbuf may fail for short race windows where poll() | |
322 | * would issue wakeups. | |
323 | */ | |
324 | DBG("Reserving sub buffer failed (everything is normal, " | |
325 | "it is due to concurrency)"); | |
326 | goto end; | |
327 | } | |
328 | ||
1d4dfdef DG |
329 | /* Get the full subbuffer size including padding */ |
330 | err = kernctl_get_padded_subbuf_size(infd, &len); | |
331 | if (err != 0) { | |
332 | errno = -err; | |
333 | perror("Getting sub-buffer len failed."); | |
334 | ret = err; | |
335 | goto end; | |
336 | } | |
337 | ||
d41f73b7 | 338 | switch (stream->output) { |
1d4dfdef | 339 | case LTTNG_EVENT_SPLICE: |
d41f73b7 | 340 | |
1d4dfdef DG |
341 | /* |
342 | * XXX: The lttng-modules splice "actor" does not handle copying | |
343 | * partial pages hence only using the subbuffer size without the | |
344 | * padding makes the splice fail. | |
345 | */ | |
346 | subbuf_size = len; | |
347 | padding = 0; | |
348 | ||
349 | /* splice the subbuffer to the tracefile */ | |
91dfef6e DG |
350 | ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size, |
351 | padding); | |
352 | /* | |
353 | * XXX: Splice does not support network streaming so the return value | |
354 | * is simply checked against subbuf_size and not like the mmap() op. | |
355 | */ | |
1d4dfdef DG |
356 | if (ret != subbuf_size) { |
357 | /* | |
358 | * display the error but continue processing to try | |
359 | * to release the subbuffer | |
360 | */ | |
361 | ERR("Error splicing to tracefile (ret: %zd != len: %lu)", | |
362 | ret, subbuf_size); | |
363 | } | |
364 | break; | |
365 | case LTTNG_EVENT_MMAP: | |
366 | /* Get subbuffer size without padding */ | |
367 | err = kernctl_get_subbuf_size(infd, &subbuf_size); | |
368 | if (err != 0) { | |
369 | errno = -err; | |
370 | perror("Getting sub-buffer len failed."); | |
371 | ret = err; | |
372 | goto end; | |
373 | } | |
47e81c02 | 374 | |
1d4dfdef DG |
375 | /* Make sure the tracer is not gone mad on us! */ |
376 | assert(len >= subbuf_size); | |
377 | ||
378 | padding = len - subbuf_size; | |
379 | ||
380 | /* write the subbuffer to the tracefile */ | |
91dfef6e DG |
381 | ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, |
382 | padding); | |
383 | /* | |
384 | * The mmap operation should write subbuf_size amount of data when | |
385 | * network streaming or the full padding (len) size when we are _not_ | |
386 | * streaming. | |
387 | */ | |
388 | if ((ret != subbuf_size && stream->net_seq_idx != -1) || | |
389 | (ret != len && stream->net_seq_idx == -1)) { | |
1d4dfdef | 390 | /* |
91dfef6e DG |
391 | * Display the error but continue processing to try to release the |
392 | * subbuffer | |
1d4dfdef | 393 | */ |
91dfef6e DG |
394 | ERR("Error writing to tracefile " |
395 | "(ret: %zd != len: %lu != subbuf_size: %lu)", | |
396 | ret, len, subbuf_size); | |
1d4dfdef DG |
397 | } |
398 | break; | |
399 | default: | |
400 | ERR("Unknown output method"); | |
401 | ret = -1; | |
d41f73b7 MD |
402 | } |
403 | ||
404 | err = kernctl_put_next_subbuf(infd); | |
405 | if (err != 0) { | |
21073eaa | 406 | errno = -err; |
d41f73b7 MD |
407 | if (errno == EFAULT) { |
408 | perror("Error in unreserving sub buffer\n"); | |
409 | } else if (errno == EIO) { | |
410 | /* Should never happen with newer LTTng versions */ | |
411 | perror("Reader has been pushed by the writer, last sub-buffer corrupted."); | |
412 | } | |
21073eaa DG |
413 | |
414 | ret = -err; | |
d41f73b7 MD |
415 | goto end; |
416 | } | |
417 | ||
418 | end: | |
419 | return ret; | |
420 | } | |
421 | ||
422 | int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
423 | { | |
424 | int ret; | |
425 | ||
426 | /* Opening the tracefile in write mode */ | |
00e2e675 | 427 | if (strlen(stream->path_name) > 0 && stream->net_seq_idx == -1) { |
e11d277b | 428 | ret = run_as_open(stream->path_name, |
60b6c79c MD |
429 | O_WRONLY|O_CREAT|O_TRUNC, |
430 | S_IRWXU|S_IRWXG|S_IRWXO, | |
431 | stream->uid, stream->gid); | |
d41f73b7 MD |
432 | if (ret < 0) { |
433 | ERR("Opening %s", stream->path_name); | |
434 | perror("open"); | |
435 | goto error; | |
436 | } | |
437 | stream->out_fd = ret; | |
438 | } | |
439 | ||
440 | if (stream->output == LTTNG_EVENT_MMAP) { | |
441 | /* get the len of the mmap region */ | |
442 | unsigned long mmap_len; | |
443 | ||
444 | ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); | |
445 | if (ret != 0) { | |
87dc6a9c | 446 | errno = -ret; |
d41f73b7 MD |
447 | perror("kernctl_get_mmap_len"); |
448 | goto error_close_fd; | |
449 | } | |
450 | stream->mmap_len = (size_t) mmap_len; | |
451 | ||
452 | stream->mmap_base = mmap(NULL, stream->mmap_len, | |
453 | PROT_READ, MAP_PRIVATE, stream->wait_fd, 0); | |
454 | if (stream->mmap_base == MAP_FAILED) { | |
455 | perror("Error mmaping"); | |
456 | ret = -1; | |
457 | goto error_close_fd; | |
458 | } | |
459 | } | |
460 | ||
461 | /* we return 0 to let the library handle the FD internally */ | |
462 | return 0; | |
463 | ||
464 | error_close_fd: | |
465 | { | |
466 | int err; | |
467 | ||
468 | err = close(stream->out_fd); | |
469 | assert(!err); | |
470 | } | |
471 | error: | |
472 | return ret; | |
473 | } | |
474 | ||
ca22feea DG |
475 | /* |
476 | * Check if data is still being extracted from the buffers for a specific | |
477 | * stream. Consumer data lock MUST be acquired before calling this function. | |
478 | * | |
479 | * Return 0 if the traced data are still getting read else 1 meaning that the | |
480 | * data is available for trace viewer reading. | |
481 | */ | |
482 | int lttng_kconsumer_data_available(struct lttng_consumer_stream *stream) | |
483 | { | |
484 | int ret; | |
485 | ||
486 | assert(stream); | |
487 | ||
488 | /* | |
489 | * Try to lock the stream mutex. On failure, we know that the stream is | |
490 | * being used else where hence there is data still being extracted. | |
491 | */ | |
492 | ret = pthread_mutex_trylock(&stream->lock); | |
493 | if (ret == EBUSY) { | |
494 | goto data_not_available; | |
495 | } | |
496 | /* The stream is now locked so we can do our ustctl calls */ | |
497 | ||
498 | ret = kernctl_get_next_subbuf(stream->wait_fd); | |
499 | if (ret == 0) { | |
500 | /* There is still data so let's put back this subbuffer. */ | |
501 | ret = kernctl_put_subbuf(stream->wait_fd); | |
502 | assert(ret == 0); | |
503 | pthread_mutex_unlock(&stream->lock); | |
504 | goto data_not_available; | |
505 | } | |
506 | ||
507 | /* Data is available to be read for this stream. */ | |
508 | pthread_mutex_unlock(&stream->lock); | |
509 | return 1; | |
510 | ||
511 | data_not_available: | |
512 | return 0; | |
513 | } |