relayd: keep track of prev_index_seq in relayd_stream
[lttng-tools.git] / src / bin / lttng-relayd / stream.c
1 /*
2 * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License, version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <common/common.h>
22 #include <common/utils.h>
23 #include <common/defaults.h>
24 #include <urcu/rculist.h>
25 #include <sys/stat.h>
26
27 #include "lttng-relayd.h"
28 #include "index.h"
29 #include "stream.h"
30 #include "viewer-stream.h"
31
32 /* Should be called with RCU read-side lock held. */
33 bool stream_get(struct relay_stream *stream)
34 {
35 return urcu_ref_get_unless_zero(&stream->ref);
36 }
37
38 /*
39 * Get stream from stream id from the streams hash table. Return stream
40 * if found else NULL. A stream reference is taken when a stream is
41 * returned. stream_put() must be called on that stream.
42 */
43 struct relay_stream *stream_get_by_id(uint64_t stream_id)
44 {
45 struct lttng_ht_node_u64 *node;
46 struct lttng_ht_iter iter;
47 struct relay_stream *stream = NULL;
48
49 rcu_read_lock();
50 lttng_ht_lookup(relay_streams_ht, &stream_id, &iter);
51 node = lttng_ht_iter_get_node_u64(&iter);
52 if (!node) {
53 DBG("Relay stream %" PRIu64 " not found", stream_id);
54 goto end;
55 }
56 stream = caa_container_of(node, struct relay_stream, node);
57 if (!stream_get(stream)) {
58 stream = NULL;
59 }
60 end:
61 rcu_read_unlock();
62 return stream;
63 }
64
65 /*
66 * We keep ownership of path_name and channel_name.
67 */
68 struct relay_stream *stream_create(struct ctf_trace *trace,
69 uint64_t stream_handle, char *path_name,
70 char *channel_name, uint64_t tracefile_size,
71 uint64_t tracefile_count,
72 const struct relay_stream_chunk_id *chunk_id)
73 {
74 int ret;
75 struct relay_stream *stream = NULL;
76 struct relay_session *session = trace->session;
77
78 stream = zmalloc(sizeof(struct relay_stream));
79 if (stream == NULL) {
80 PERROR("relay stream zmalloc");
81 goto error_no_alloc;
82 }
83
84 stream->stream_handle = stream_handle;
85 stream->prev_seq = -1ULL;
86 stream->prev_index_seq = -1ULL;
87 stream->last_net_seq_num = -1ULL;
88 stream->ctf_stream_id = -1ULL;
89 stream->tracefile_size = tracefile_size;
90 stream->tracefile_count = tracefile_count;
91 stream->path_name = path_name;
92 stream->channel_name = channel_name;
93 stream->rotate_at_seq_num = -1ULL;
94 lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
95 pthread_mutex_init(&stream->lock, NULL);
96 urcu_ref_init(&stream->ref);
97 ctf_trace_get(trace);
98 stream->trace = trace;
99 stream->current_chunk_id = *chunk_id;
100
101 stream->indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
102 if (!stream->indexes_ht) {
103 ERR("Cannot created indexes_ht");
104 ret = -1;
105 goto end;
106 }
107
108 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG,
109 -1, -1);
110 if (ret < 0) {
111 ERR("relay creating output directory");
112 goto end;
113 }
114
115 /*
116 * No need to use run_as API here because whatever we receive,
117 * the relayd uses its own credentials for the stream files.
118 */
119 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
120 stream->tracefile_size, 0, -1, -1, NULL);
121 if (ret < 0) {
122 ERR("Create output file");
123 goto end;
124 }
125 stream->stream_fd = stream_fd_create(ret);
126 if (!stream->stream_fd) {
127 if (close(ret)) {
128 PERROR("Error closing file %d", ret);
129 }
130 ret = -1;
131 goto end;
132 }
133 stream->tfa = tracefile_array_create(stream->tracefile_count);
134 if (!stream->tfa) {
135 ret = -1;
136 goto end;
137 }
138 if (stream->tracefile_size) {
139 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
140 } else {
141 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
142 }
143
144 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, LTTNG_NAME_MAX)) {
145 stream->is_metadata = 1;
146 }
147
148 stream->in_recv_list = true;
149
150 /*
151 * Add the stream in the recv list of the session. Once the end stream
152 * message is received, all session streams are published.
153 */
154 pthread_mutex_lock(&session->recv_list_lock);
155 cds_list_add_rcu(&stream->recv_node, &session->recv_list);
156 session->stream_count++;
157 pthread_mutex_unlock(&session->recv_list_lock);
158
159 /*
160 * Both in the ctf_trace object and the global stream ht since the data
161 * side of the relayd does not have the concept of session.
162 */
163 lttng_ht_add_unique_u64(relay_streams_ht, &stream->node);
164 stream->in_stream_ht = true;
165
166 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
167 stream->stream_handle);
168 ret = 0;
169
170 end:
171 if (ret) {
172 if (stream->stream_fd) {
173 stream_fd_put(stream->stream_fd);
174 stream->stream_fd = NULL;
175 }
176 stream_put(stream);
177 stream = NULL;
178 }
179 return stream;
180
181 error_no_alloc:
182 /*
183 * path_name and channel_name need to be freed explicitly here
184 * because we cannot rely on stream_put().
185 */
186 free(path_name);
187 free(channel_name);
188 return NULL;
189 }
190
191 /*
192 * Called with the session lock held.
193 */
194 void stream_publish(struct relay_stream *stream)
195 {
196 struct relay_session *session;
197
198 pthread_mutex_lock(&stream->lock);
199 if (stream->published) {
200 goto unlock;
201 }
202
203 session = stream->trace->session;
204
205 pthread_mutex_lock(&session->recv_list_lock);
206 if (stream->in_recv_list) {
207 cds_list_del_rcu(&stream->recv_node);
208 stream->in_recv_list = false;
209 }
210 pthread_mutex_unlock(&session->recv_list_lock);
211
212 pthread_mutex_lock(&stream->trace->stream_list_lock);
213 cds_list_add_rcu(&stream->stream_node, &stream->trace->stream_list);
214 pthread_mutex_unlock(&stream->trace->stream_list_lock);
215
216 stream->published = true;
217 unlock:
218 pthread_mutex_unlock(&stream->lock);
219 }
220
221 /*
222 * Stream must be protected by holding the stream lock or by virtue of being
223 * called from stream_destroy.
224 */
225 static void stream_unpublish(struct relay_stream *stream)
226 {
227 if (stream->in_stream_ht) {
228 struct lttng_ht_iter iter;
229 int ret;
230
231 iter.iter.node = &stream->node.node;
232 ret = lttng_ht_del(relay_streams_ht, &iter);
233 assert(!ret);
234 stream->in_stream_ht = false;
235 }
236 if (stream->published) {
237 pthread_mutex_lock(&stream->trace->stream_list_lock);
238 cds_list_del_rcu(&stream->stream_node);
239 pthread_mutex_unlock(&stream->trace->stream_list_lock);
240 stream->published = false;
241 }
242 }
243
244 static void stream_destroy(struct relay_stream *stream)
245 {
246 if (stream->indexes_ht) {
247 /*
248 * Calling lttng_ht_destroy in call_rcu worker thread so
249 * we don't hold the RCU read-side lock while calling
250 * it.
251 */
252 lttng_ht_destroy(stream->indexes_ht);
253 }
254 if (stream->tfa) {
255 tracefile_array_destroy(stream->tfa);
256 }
257 free(stream->path_name);
258 free(stream->channel_name);
259 free(stream);
260 }
261
262 static void stream_destroy_rcu(struct rcu_head *rcu_head)
263 {
264 struct relay_stream *stream =
265 caa_container_of(rcu_head, struct relay_stream, rcu_node);
266
267 stream_destroy(stream);
268 }
269
270 /*
271 * No need to take stream->lock since this is only called on the final
272 * stream_put which ensures that a single thread may act on the stream.
273 */
274 static void stream_release(struct urcu_ref *ref)
275 {
276 struct relay_stream *stream =
277 caa_container_of(ref, struct relay_stream, ref);
278 struct relay_session *session;
279
280 session = stream->trace->session;
281
282 DBG("Releasing stream id %" PRIu64, stream->stream_handle);
283
284 pthread_mutex_lock(&session->recv_list_lock);
285 session->stream_count--;
286 if (stream->in_recv_list) {
287 cds_list_del_rcu(&stream->recv_node);
288 stream->in_recv_list = false;
289 }
290 pthread_mutex_unlock(&session->recv_list_lock);
291
292 stream_unpublish(stream);
293
294 if (stream->stream_fd) {
295 stream_fd_put(stream->stream_fd);
296 stream->stream_fd = NULL;
297 }
298 if (stream->index_file) {
299 lttng_index_file_put(stream->index_file);
300 stream->index_file = NULL;
301 }
302 if (stream->trace) {
303 ctf_trace_put(stream->trace);
304 stream->trace = NULL;
305 }
306
307 call_rcu(&stream->rcu_node, stream_destroy_rcu);
308 }
309
310 void stream_put(struct relay_stream *stream)
311 {
312 DBG("stream put for stream id %" PRIu64, stream->stream_handle);
313 rcu_read_lock();
314 assert(stream->ref.refcount != 0);
315 /*
316 * Wait until we have processed all the stream packets before
317 * actually putting our last stream reference.
318 */
319 DBG("stream put stream id %" PRIu64 " refcount %d",
320 stream->stream_handle,
321 (int) stream->ref.refcount);
322 urcu_ref_put(&stream->ref, stream_release);
323 rcu_read_unlock();
324 }
325
326 void try_stream_close(struct relay_stream *stream)
327 {
328 bool session_aborted;
329 struct relay_session *session = stream->trace->session;
330
331 DBG("Trying to close stream %" PRIu64, stream->stream_handle);
332
333 pthread_mutex_lock(&session->lock);
334 session_aborted = session->aborted;
335 pthread_mutex_unlock(&session->lock);
336
337 pthread_mutex_lock(&stream->lock);
338 /*
339 * Can be called concurently by connection close and reception of last
340 * pending data.
341 */
342 if (stream->closed) {
343 pthread_mutex_unlock(&stream->lock);
344 DBG("closing stream %" PRIu64 " aborted since it is already marked as closed", stream->stream_handle);
345 return;
346 }
347
348 stream->close_requested = true;
349
350 if (stream->last_net_seq_num == -1ULL) {
351 /*
352 * Handle connection close without explicit stream close
353 * command.
354 *
355 * We can be clever about indexes partially received in
356 * cases where we received the data socket part, but not
357 * the control socket part: since we're currently closing
358 * the stream on behalf of the control socket, we *know*
359 * there won't be any more control information for this
360 * socket. Therefore, we can destroy all indexes for
361 * which we have received only the file descriptor (from
362 * data socket). This takes care of consumerd crashes
363 * between sending the data and control information for
364 * a packet. Since those are sent in that order, we take
365 * care of consumerd crashes.
366 */
367 DBG("relay_index_close_partial_fd");
368 relay_index_close_partial_fd(stream);
369 /*
370 * Use the highest net_seq_num we currently have pending
371 * As end of stream indicator. Leave last_net_seq_num
372 * at -1ULL if we cannot find any index.
373 */
374 stream->last_net_seq_num = relay_index_find_last(stream);
375 DBG("Updating stream->last_net_seq_num to %" PRIu64, stream->last_net_seq_num);
376 /* Fall-through into the next check. */
377 }
378
379 if (stream->last_net_seq_num != -1ULL &&
380 ((int64_t) (stream->prev_seq - stream->last_net_seq_num)) < 0
381 && !session_aborted) {
382 /*
383 * Don't close since we still have data pending. This
384 * handles cases where an explicit close command has
385 * been received for this stream, and cases where the
386 * connection has been closed, and we are awaiting for
387 * index information from the data socket. It is
388 * therefore expected that all the index fd information
389 * we need has already been received on the control
390 * socket. Matching index information from data socket
391 * should be Expected Soon(TM).
392 *
393 * TODO: We should implement a timer to garbage collect
394 * streams after a timeout to be resilient against a
395 * consumerd implementation that would not match this
396 * expected behavior.
397 */
398 pthread_mutex_unlock(&stream->lock);
399 DBG("closing stream %" PRIu64 " aborted since it still has data pending", stream->stream_handle);
400 return;
401 }
402 /*
403 * We received all the indexes we can expect.
404 */
405 stream_unpublish(stream);
406 stream->closed = true;
407 /* Relay indexes are only used by the "consumer/sessiond" end. */
408 relay_index_close_all(stream);
409 pthread_mutex_unlock(&stream->lock);
410 DBG("Succeeded in closing stream %" PRIu64, stream->stream_handle);
411 stream_put(stream);
412 }
413
414 static void print_stream_indexes(struct relay_stream *stream)
415 {
416 struct lttng_ht_iter iter;
417 struct relay_index *index;
418
419 rcu_read_lock();
420 cds_lfht_for_each_entry(stream->indexes_ht->ht, &iter.iter, index,
421 index_n.node) {
422 DBG("index %p net_seq_num %" PRIu64 " refcount %ld"
423 " stream %" PRIu64 " trace %" PRIu64
424 " session %" PRIu64,
425 index,
426 index->index_n.key,
427 stream->ref.refcount,
428 index->stream->stream_handle,
429 index->stream->trace->id,
430 index->stream->trace->session->id);
431 }
432 rcu_read_unlock();
433 }
434
435 void print_relay_streams(void)
436 {
437 struct lttng_ht_iter iter;
438 struct relay_stream *stream;
439
440 if (!relay_streams_ht) {
441 return;
442 }
443
444 rcu_read_lock();
445 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
446 node.node) {
447 if (!stream_get(stream)) {
448 continue;
449 }
450 DBG("stream %p refcount %ld stream %" PRIu64 " trace %" PRIu64
451 " session %" PRIu64,
452 stream,
453 stream->ref.refcount,
454 stream->stream_handle,
455 stream->trace->id,
456 stream->trace->session->id);
457 print_stream_indexes(stream);
458 stream_put(stream);
459 }
460 rcu_read_unlock();
461 }
This page took 0.038646 seconds and 5 git commands to generate.