Fix: multiple consumer locking problems
[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
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.
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 *
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.
17 */
18
19 #define _GNU_SOURCE
20 #include <assert.h>
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>
28 #include <unistd.h>
29 #include <sys/stat.h>
30
31 #include <common/common.h>
32 #include <common/kernel-ctl/kernel-ctl.h>
33 #include <common/sessiond-comm/sessiond-comm.h>
34 #include <common/sessiond-comm/relayd.h>
35 #include <common/compat/fcntl.h>
36 #include <common/relayd/relayd.h>
37
38 #include "kernel-consumer.h"
39
40 extern struct lttng_consumer_global_data consumer_data;
41 extern int consumer_poll_timeout;
42 extern volatile int consumer_quit;
43
44 /*
45 * Take a snapshot for a specific fd
46 *
47 * Returns 0 on success, < 0 on error
48 */
49 int lttng_kconsumer_take_snapshot(struct lttng_consumer_local_data *ctx,
50 struct lttng_consumer_stream *stream)
51 {
52 int ret = 0;
53 int infd = stream->wait_fd;
54
55 ret = kernctl_snapshot(infd);
56 if (ret != 0) {
57 errno = -ret;
58 perror("Getting sub-buffer snapshot.");
59 }
60
61 return ret;
62 }
63
64 /*
65 * Get the produced position
66 *
67 * Returns 0 on success, < 0 on error
68 */
69 int lttng_kconsumer_get_produced_snapshot(
70 struct lttng_consumer_local_data *ctx,
71 struct lttng_consumer_stream *stream,
72 unsigned long *pos)
73 {
74 int ret;
75 int infd = stream->wait_fd;
76
77 ret = kernctl_snapshot_get_produced(infd, pos);
78 if (ret != 0) {
79 errno = -ret;
80 perror("kernctl_snapshot_get_produced");
81 }
82
83 return ret;
84 }
85
86 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
87 int sock, struct pollfd *consumer_sockpoll)
88 {
89 ssize_t ret;
90 struct lttcomm_consumer_msg msg;
91
92 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
93 if (ret != sizeof(msg)) {
94 lttng_consumer_send_error(ctx, CONSUMERD_ERROR_RECV_CMD);
95 return ret;
96 }
97 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
98 return -ENOENT;
99 }
100
101 /* relayd needs RCU read-side protection */
102 rcu_read_lock();
103
104 switch (msg.cmd_type) {
105 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
106 {
107 int fd;
108 struct consumer_relayd_sock_pair *relayd;
109
110 DBG("Consumer adding relayd socket");
111
112 /* Get relayd reference if exists. */
113 relayd = consumer_find_relayd(msg.u.relayd_sock.net_index);
114 if (relayd == NULL) {
115 /* Not found. Allocate one. */
116 relayd = consumer_allocate_relayd_sock_pair(
117 msg.u.relayd_sock.net_index);
118 if (relayd == NULL) {
119 lttng_consumer_send_error(ctx, CONSUMERD_OUTFD_ERROR);
120 goto end_nosignal;
121 }
122 }
123
124 /* Poll on consumer socket. */
125 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
126 rcu_read_unlock();
127 return -EINTR;
128 }
129
130 /* Get relayd socket from session daemon */
131 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
132 if (ret != sizeof(fd)) {
133 lttng_consumer_send_error(ctx, CONSUMERD_ERROR_RECV_FD);
134 goto end_nosignal;
135 }
136
137 /* Copy socket information and received FD */
138 switch (msg.u.relayd_sock.type) {
139 case LTTNG_STREAM_CONTROL:
140 /* Copy received lttcomm socket */
141 lttcomm_copy_sock(&relayd->control_sock, &msg.u.relayd_sock.sock);
142
143 ret = lttcomm_create_sock(&relayd->control_sock);
144 if (ret < 0) {
145 goto end_nosignal;
146 }
147
148 /* Close the created socket fd which is useless */
149 close(relayd->control_sock.fd);
150
151 /* Assign new file descriptor */
152 relayd->control_sock.fd = fd;
153 break;
154 case LTTNG_STREAM_DATA:
155 /* Copy received lttcomm socket */
156 lttcomm_copy_sock(&relayd->data_sock, &msg.u.relayd_sock.sock);
157 ret = lttcomm_create_sock(&relayd->data_sock);
158 if (ret < 0) {
159 goto end_nosignal;
160 }
161
162 /* Close the created socket fd which is useless */
163 close(relayd->data_sock.fd);
164
165 /* Assign new file descriptor */
166 relayd->data_sock.fd = fd;
167 break;
168 default:
169 ERR("Unknown relayd socket type");
170 goto end_nosignal;
171 }
172
173 DBG("Consumer %s socket created successfully with net idx %d (fd: %d)",
174 msg.u.relayd_sock.type == LTTNG_STREAM_CONTROL ? "control" : "data",
175 relayd->net_seq_idx, fd);
176
177 /*
178 * Add relayd socket pair to consumer data hashtable. If object already
179 * exists or on error, the function gracefully returns.
180 */
181 consumer_add_relayd(relayd);
182
183 goto end_nosignal;
184 }
185 case LTTNG_CONSUMER_ADD_CHANNEL:
186 {
187 struct lttng_consumer_channel *new_channel;
188
189 DBG("consumer_add_channel %d", msg.u.channel.channel_key);
190 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
191 -1, -1,
192 msg.u.channel.mmap_len,
193 msg.u.channel.max_sb_size);
194 if (new_channel == NULL) {
195 lttng_consumer_send_error(ctx, CONSUMERD_OUTFD_ERROR);
196 goto end_nosignal;
197 }
198 if (ctx->on_recv_channel != NULL) {
199 ret = ctx->on_recv_channel(new_channel);
200 if (ret == 0) {
201 consumer_add_channel(new_channel);
202 } else if (ret < 0) {
203 goto end_nosignal;
204 }
205 } else {
206 consumer_add_channel(new_channel);
207 }
208 goto end_nosignal;
209 }
210 case LTTNG_CONSUMER_ADD_STREAM:
211 {
212 int fd;
213 struct consumer_relayd_sock_pair *relayd = NULL;
214 struct lttng_consumer_stream *new_stream;
215
216 /* block */
217 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
218 rcu_read_unlock();
219 return -EINTR;
220 }
221
222 /* Get stream file descriptor from socket */
223 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
224 if (ret != sizeof(fd)) {
225 lttng_consumer_send_error(ctx, CONSUMERD_ERROR_RECV_FD);
226 rcu_read_unlock();
227 return ret;
228 }
229
230 new_stream = consumer_allocate_stream(msg.u.stream.channel_key,
231 msg.u.stream.stream_key,
232 fd, fd,
233 msg.u.stream.state,
234 msg.u.stream.mmap_len,
235 msg.u.stream.output,
236 msg.u.stream.path_name,
237 msg.u.stream.uid,
238 msg.u.stream.gid,
239 msg.u.stream.net_index,
240 msg.u.stream.metadata_flag);
241 if (new_stream == NULL) {
242 lttng_consumer_send_error(ctx, CONSUMERD_OUTFD_ERROR);
243 goto end_nosignal;
244 }
245
246 /* The stream is not metadata. Get relayd reference if exists. */
247 relayd = consumer_find_relayd(msg.u.stream.net_index);
248 if (relayd != NULL) {
249 /* Add stream on the relayd */
250 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
251 ret = relayd_add_stream(&relayd->control_sock,
252 msg.u.stream.name, msg.u.stream.path_name,
253 &new_stream->relayd_stream_id);
254 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
255 if (ret < 0) {
256 goto end_nosignal;
257 }
258 } else if (msg.u.stream.net_index != -1) {
259 ERR("Network sequence index %d unknown. Not adding stream.",
260 msg.u.stream.net_index);
261 free(new_stream);
262 goto end_nosignal;
263 }
264
265 if (ctx->on_recv_stream != NULL) {
266 ret = ctx->on_recv_stream(new_stream);
267 if (ret == 0) {
268 consumer_add_stream(new_stream);
269 } else if (ret < 0) {
270 goto end_nosignal;
271 }
272 } else {
273 consumer_add_stream(new_stream);
274 }
275
276 DBG("Kernel consumer_add_stream (%d)", fd);
277 break;
278 }
279 case LTTNG_CONSUMER_UPDATE_STREAM:
280 {
281 rcu_read_unlock();
282 return -ENOSYS;
283 }
284 case LTTNG_CONSUMER_DESTROY_RELAYD:
285 {
286 struct consumer_relayd_sock_pair *relayd;
287
288 DBG("Kernel consumer destroying relayd %zu",
289 msg.u.destroy_relayd.net_seq_idx);
290
291 /* Get relayd reference if exists. */
292 relayd = consumer_find_relayd(msg.u.destroy_relayd.net_seq_idx);
293 if (relayd == NULL) {
294 ERR("Unable to find relayd %zu", msg.u.destroy_relayd.net_seq_idx);
295 goto end_nosignal;
296 }
297
298 /* Set destroy flag for this object */
299 uatomic_set(&relayd->destroy_flag, 1);
300
301 /* Destroy the relayd if refcount is 0 else set the destroy flag. */
302 if (uatomic_read(&relayd->refcount) == 0) {
303 consumer_destroy_relayd(relayd);
304 }
305 goto end_nosignal;
306 }
307 default:
308 goto end_nosignal;
309 }
310
311 /*
312 * Wake-up the other end by writing a null byte in the pipe (non-blocking).
313 * Important note: Because writing into the pipe is non-blocking (and
314 * therefore we allow dropping wakeup data, as long as there is wakeup data
315 * present in the pipe buffer to wake up the other end), the other end
316 * should perform the following sequence for waiting:
317 *
318 * 1) empty the pipe (reads).
319 * 2) perform update operation.
320 * 3) wait on the pipe (poll).
321 */
322 do {
323 ret = write(ctx->consumer_poll_pipe[1], "", 1);
324 } while (ret < 0 && errno == EINTR);
325 end_nosignal:
326 rcu_read_unlock();
327 return 0;
328 }
329
330 /*
331 * Consume data on a file descriptor and write it on a trace file.
332 */
333 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
334 struct lttng_consumer_local_data *ctx)
335 {
336 unsigned long len;
337 int err;
338 ssize_t ret = 0;
339 int infd = stream->wait_fd;
340
341 DBG("In read_subbuffer (infd : %d)", infd);
342 /* Get the next subbuffer */
343 err = kernctl_get_next_subbuf(infd);
344 if (err != 0) {
345 /*
346 * This is a debug message even for single-threaded consumer,
347 * because poll() have more relaxed criterions than get subbuf,
348 * so get_subbuf may fail for short race windows where poll()
349 * would issue wakeups.
350 */
351 DBG("Reserving sub buffer failed (everything is normal, "
352 "it is due to concurrency)");
353 goto end;
354 }
355
356 switch (stream->output) {
357 case LTTNG_EVENT_SPLICE:
358 /* read the whole subbuffer */
359 err = kernctl_get_padded_subbuf_size(infd, &len);
360 if (err != 0) {
361 errno = -ret;
362 perror("Getting sub-buffer len failed.");
363 goto end;
364 }
365
366 /* splice the subbuffer to the tracefile */
367 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, len);
368 if (ret != len) {
369 /*
370 * display the error but continue processing to try
371 * to release the subbuffer
372 */
373 ERR("Error splicing to tracefile (ret: %ld != len: %ld)",
374 ret, len);
375 }
376
377 break;
378 case LTTNG_EVENT_MMAP:
379 /* read the used subbuffer size */
380 err = kernctl_get_padded_subbuf_size(infd, &len);
381 if (err != 0) {
382 errno = -ret;
383 perror("Getting sub-buffer len failed.");
384 goto end;
385 }
386 /* write the subbuffer to the tracefile */
387 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len);
388 if (ret != len) {
389 /*
390 * display the error but continue processing to try
391 * to release the subbuffer
392 */
393 ERR("Error writing to tracefile");
394 }
395 break;
396 default:
397 ERR("Unknown output method");
398 ret = -1;
399 }
400
401 err = kernctl_put_next_subbuf(infd);
402 if (err != 0) {
403 errno = -ret;
404 if (errno == EFAULT) {
405 perror("Error in unreserving sub buffer\n");
406 } else if (errno == EIO) {
407 /* Should never happen with newer LTTng versions */
408 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
409 }
410 goto end;
411 }
412
413 end:
414 return ret;
415 }
416
417 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
418 {
419 int ret;
420
421 /* Opening the tracefile in write mode */
422 if (strlen(stream->path_name) > 0 && stream->net_seq_idx == -1) {
423 ret = run_as_open(stream->path_name,
424 O_WRONLY|O_CREAT|O_TRUNC,
425 S_IRWXU|S_IRWXG|S_IRWXO,
426 stream->uid, stream->gid);
427 if (ret < 0) {
428 ERR("Opening %s", stream->path_name);
429 perror("open");
430 goto error;
431 }
432 stream->out_fd = ret;
433 }
434
435 if (stream->output == LTTNG_EVENT_MMAP) {
436 /* get the len of the mmap region */
437 unsigned long mmap_len;
438
439 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
440 if (ret != 0) {
441 errno = -ret;
442 perror("kernctl_get_mmap_len");
443 goto error_close_fd;
444 }
445 stream->mmap_len = (size_t) mmap_len;
446
447 stream->mmap_base = mmap(NULL, stream->mmap_len,
448 PROT_READ, MAP_PRIVATE, stream->wait_fd, 0);
449 if (stream->mmap_base == MAP_FAILED) {
450 perror("Error mmaping");
451 ret = -1;
452 goto error_close_fd;
453 }
454 }
455
456 /* we return 0 to let the library handle the FD internally */
457 return 0;
458
459 error_close_fd:
460 {
461 int err;
462
463 err = close(stream->out_fd);
464 assert(!err);
465 }
466 error:
467 return ret;
468 }
469
This page took 0.038309 seconds and 4 git commands to generate.