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         bool has_ref = false;
  36
  37         pthread_mutex_lock(&stream->reflock);
  38         if (stream->ref.refcount != 0) {
  39                 has_ref = true;
  40                 urcu_ref_get(&stream->ref);
  41         }
  42         pthread_mutex_unlock(&stream->reflock);
  43
  44         return has_ref;
  45 }
  46
  47 /*
  48  * Get stream from stream id from the streams hash table. Return stream
  49  * if found else NULL. A stream reference is taken when a stream is
  50  * returned. stream_put() must be called on that stream.
  51  */
  52 struct relay_stream *stream_get_by_id(uint64_t stream_id)
  53 {
  54         struct lttng_ht_node_u64 *node;
  55         struct lttng_ht_iter iter;
  56         struct relay_stream *stream = NULL;
  57
  58         rcu_read_lock();
  59         lttng_ht_lookup(relay_streams_ht, &stream_id, &iter);
  60         node = lttng_ht_iter_get_node_u64(&iter);
  61         if (!node) {
  62                 DBG("Relay stream %" PRIu64 " not found", stream_id);
  63                 goto end;
  64         }
  65         stream = caa_container_of(node, struct relay_stream, node);
  66         if (!stream_get(stream)) {
  67                 stream = NULL;
  68         }
  69 end:
  70         rcu_read_unlock();
  71         return stream;
  72 }
  73
  74 /*
  75  * We keep ownership of path_name and channel_name.
  76  */
  77 struct relay_stream *stream_create(struct ctf_trace *trace,
  78         uint64_t stream_handle, char *path_name,
  79         char *channel_name, uint64_t tracefile_size,
  80         uint64_t tracefile_count)
  81 {
  82         int ret;
  83         struct relay_stream *stream = NULL;
  84         struct relay_session *session = trace->session;
  85
  86         stream = zmalloc(sizeof(struct relay_stream));
  87         if (stream == NULL) {
  88                 PERROR("relay stream zmalloc");
  89                 ret = -1;
  90                 goto error_no_alloc;
  91         }
  92
  93         stream->stream_handle = stream_handle;
  94         stream->prev_seq = -1ULL;
  95         stream->last_net_seq_num = -1ULL;
  96         stream->ctf_stream_id = -1ULL;
  97         stream->tracefile_size = tracefile_size;
  98         stream->tracefile_count = tracefile_count;
  99         stream->path_name = path_name;
 100         stream->channel_name = channel_name;
 101         lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
 102         pthread_mutex_init(&stream->lock, NULL);
 103         pthread_mutex_init(&stream->reflock, NULL);
 104         urcu_ref_init(&stream->ref);
 105         ctf_trace_get(trace);
 106         stream->trace = trace;
 107
 108         stream->indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
 109         if (!stream->indexes_ht) {
 110                 ERR("Cannot created indexes_ht");
 111                 ret = -1;
 112                 goto end;
 113         }
 114
 115         ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG,
 116                         -1, -1);
 117         if (ret < 0) {
 118                 ERR("relay creating output directory");
 119                 goto end;
 120         }
 121
 122         /*
 123          * No need to use run_as API here because whatever we receive,
 124          * the relayd uses its own credentials for the stream files.
 125          */
 126         stream->stream_fd = stream_fd_create(stream->path_name,
 127                         stream->channel_name, stream->tracefile_size, 0, NULL);
 128         if (!stream->stream_fd) {
 129                 if (close(ret)) {
 130                         PERROR("Error closing file %d", ret);
 131                 }
 132                 ret = -1;
 133                 goto end;
 134         }
 135         stream->tfa = tracefile_array_create(stream->tracefile_count);
 136         if (!stream->tfa) {
 137                 ret = -1;
 138                 goto end;
 139         }
 140         if (stream->tracefile_size) {
 141                 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
 142         } else {
 143                 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
 144         }
 145
 146         if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, LTTNG_NAME_MAX)) {
 147                 stream->is_metadata = 1;
 148         }
 149
 150         stream->in_recv_list = true;
 151
 152         /*
 153          * Add the stream in the recv list of the session. Once the end stream
 154          * message is received, all session streams are published.
 155          */
 156         pthread_mutex_lock(&session->recv_list_lock);
 157         cds_list_add_rcu(&stream->recv_node, &session->recv_list);
 158         session->stream_count++;
 159         pthread_mutex_unlock(&session->recv_list_lock);
 160
 161         /*
 162          * Both in the ctf_trace object and the global stream ht since the data
 163          * side of the relayd does not have the concept of session.
 164          */
 165         lttng_ht_add_unique_u64(relay_streams_ht, &stream->node);
 166         stream->in_stream_ht = true;
 167
 168         DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
 169                         stream->stream_handle);
 170         ret = 0;
 171
 172 end:
 173         if (ret) {
 174                 if (stream->stream_fd) {
 175                         stream_fd_put(stream->stream_fd);
 176                         stream->stream_fd = NULL;
 177                 }
 178                 stream_put(stream);
 179                 stream = NULL;
 180         }
 181         return stream;
 182
 183 error_no_alloc:
 184         /*
 185          * path_name and channel_name need to be freed explicitly here
 186          * because we cannot rely on stream_put().
 187          */
 188         free(path_name);
 189         free(channel_name);
 190         return NULL;
 191 }
 192
 193 /*
 194  * Called with the session lock held.
 195  */
 196 void stream_publish(struct relay_stream *stream)
 197 {
 198         struct relay_session *session;
 199
 200         pthread_mutex_lock(&stream->lock);
 201         if (stream->published) {
 202                 goto unlock;
 203         }
 204
 205         session = stream->trace->session;
 206
 207         pthread_mutex_lock(&session->recv_list_lock);
 208         if (stream->in_recv_list) {
 209                 cds_list_del_rcu(&stream->recv_node);
 210                 stream->in_recv_list = false;
 211         }
 212         pthread_mutex_unlock(&session->recv_list_lock);
 213
 214         pthread_mutex_lock(&stream->trace->stream_list_lock);
 215         cds_list_add_rcu(&stream->stream_node, &stream->trace->stream_list);
 216         pthread_mutex_unlock(&stream->trace->stream_list_lock);
 217
 218         stream->published = true;
 219 unlock:
 220         pthread_mutex_unlock(&stream->lock);
 221 }
 222
 223 /*
 224  * Stream must be protected by holding the stream lock or by virtue of being
 225  * called from stream_destroy, in which case it is guaranteed to be accessed
 226  * from a single thread by the reflock.
 227  */
 228 static void stream_unpublish(struct relay_stream *stream)
 229 {
 230         if (stream->in_stream_ht) {
 231                 struct lttng_ht_iter iter;
 232                 int ret;
 233
 234                 iter.iter.node = &stream->node.node;
 235                 ret = lttng_ht_del(relay_streams_ht, &iter);
 236                 assert(!ret);
 237                 stream->in_stream_ht = false;
 238         }
 239         if (stream->published) {
 240                 pthread_mutex_lock(&stream->trace->stream_list_lock);
 241                 cds_list_del_rcu(&stream->stream_node);
 242                 pthread_mutex_unlock(&stream->trace->stream_list_lock);
 243                 stream->published = false;
 244         }
 245 }
 246
 247 static void stream_destroy(struct relay_stream *stream)
 248 {
 249         if (stream->indexes_ht) {
 250                 /*
 251                  * Calling lttng_ht_destroy in call_rcu worker thread so
 252                  * we don't hold the RCU read-side lock while calling
 253                  * it.
 254                  */
 255                 lttng_ht_destroy(stream->indexes_ht);
 256         }
 257         if (stream->tfa) {
 258                 tracefile_array_destroy(stream->tfa);
 259         }
 260         free(stream->path_name);
 261         free(stream->channel_name);
 262         free(stream);
 263 }
 264
 265 static void stream_destroy_rcu(struct rcu_head *rcu_head)
 266 {
 267         struct relay_stream *stream =
 268                 caa_container_of(rcu_head, struct relay_stream, rcu_node);
 269
 270         stream_destroy(stream);
 271 }
 272
 273 /*
 274  * No need to take stream->lock since this is only called on the final
 275  * stream_put which ensures that a single thread may act on the stream.
 276  *
 277  * At that point, the object is also protected by the reflock which
 278  * guarantees that no other thread may share ownership of this stream.
 279  */
 280 static void stream_release(struct urcu_ref *ref)
 281 {
 282         struct relay_stream *stream =
 283                 caa_container_of(ref, struct relay_stream, ref);
 284         struct relay_session *session;
 285
 286         session = stream->trace->session;
 287
 288         DBG("Releasing stream id %" PRIu64, stream->stream_handle);
 289
 290         pthread_mutex_lock(&session->recv_list_lock);
 291         session->stream_count--;
 292         if (stream->in_recv_list) {
 293                 cds_list_del_rcu(&stream->recv_node);
 294                 stream->in_recv_list = false;
 295         }
 296         pthread_mutex_unlock(&session->recv_list_lock);
 297
 298         stream_unpublish(stream);
 299
 300         if (stream->stream_fd) {
 301                 stream_fd_put(stream->stream_fd);
 302                 stream->stream_fd = NULL;
 303         }
 304         if (stream->index_file) {
 305                 lttng_index_file_put(stream->index_file);
 306                 stream->index_file = NULL;
 307         }
 308         if (stream->trace) {
 309                 ctf_trace_put(stream->trace);
 310                 stream->trace = NULL;
 311         }
 312
 313         call_rcu(&stream->rcu_node, stream_destroy_rcu);
 314 }
 315
 316 void stream_put(struct relay_stream *stream)
 317 {
 318         DBG("stream put for stream id %" PRIu64, stream->stream_handle);
 319         /*
 320          * Ensure existence of stream->reflock for stream unlock.
 321          */
 322         rcu_read_lock();
 323         /*
 324          * Stream reflock ensures that concurrent test and update of
 325          * stream ref is atomic.
 326          */
 327         pthread_mutex_lock(&stream->reflock);
 328         assert(stream->ref.refcount != 0);
 329         /*
 330          * Wait until we have processed all the stream packets before
 331          * actually putting our last stream reference.
 332          */
 333         DBG("stream put stream id %" PRIu64 " refcount %d",
 334                         stream->stream_handle,
 335                         (int) stream->ref.refcount);
 336         urcu_ref_put(&stream->ref, stream_release);
 337         pthread_mutex_unlock(&stream->reflock);
 338         rcu_read_unlock();
 339 }
 340
 341 void try_stream_close(struct relay_stream *stream)
 342 {
 343         bool session_aborted;
 344         struct relay_session *session = stream->trace->session;
 345
 346         DBG("Trying to close stream %" PRIu64, stream->stream_handle);
 347
 348         pthread_mutex_lock(&session->lock);
 349         session_aborted = session->aborted;
 350         pthread_mutex_unlock(&session->lock);
 351
 352         pthread_mutex_lock(&stream->lock);
 353         /*
 354          * Can be called concurently by connection close and reception of last
 355          * pending data.
 356          */
 357         if (stream->closed) {
 358                 pthread_mutex_unlock(&stream->lock);
 359                 DBG("closing stream %" PRIu64 " aborted since it is already marked as closed", stream->stream_handle);
 360                 return;
 361         }
 362
 363         stream->close_requested = true;
 364
 365         if (stream->last_net_seq_num == -1ULL) {
 366                 /*
 367                  * Handle connection close without explicit stream close
 368                  * command.
 369                  *
 370                  * We can be clever about indexes partially received in
 371                  * cases where we received the data socket part, but not
 372                  * the control socket part: since we're currently closing
 373                  * the stream on behalf of the control socket, we *know*
 374                  * there won't be any more control information for this
 375                  * socket. Therefore, we can destroy all indexes for
 376                  * which we have received only the file descriptor (from
 377                  * data socket). This takes care of consumerd crashes
 378                  * between sending the data and control information for
 379                  * a packet. Since those are sent in that order, we take
 380                  * care of consumerd crashes.
 381                  */
 382                 DBG("relay_index_close_partial_fd");
 383                 relay_index_close_partial_fd(stream);
 384                 /*
 385                  * Use the highest net_seq_num we currently have pending
 386                  * As end of stream indicator.  Leave last_net_seq_num
 387                  * at -1ULL if we cannot find any index.
 388                  */
 389                 stream->last_net_seq_num = relay_index_find_last(stream);
 390                 DBG("Updating stream->last_net_seq_num to %" PRIu64, stream->last_net_seq_num);
 391                 /* Fall-through into the next check. */
 392         }
 393
 394         if (stream->last_net_seq_num != -1ULL &&
 395                         ((int64_t) (stream->prev_seq - stream->last_net_seq_num)) < 0
 396                         && !session_aborted) {
 397                 /*
 398                  * Don't close since we still have data pending. This
 399                  * handles cases where an explicit close command has
 400                  * been received for this stream, and cases where the
 401                  * connection has been closed, and we are awaiting for
 402                  * index information from the data socket. It is
 403                  * therefore expected that all the index fd information
 404                  * we need has already been received on the control
 405                  * socket. Matching index information from data socket
 406                  * should be Expected Soon(TM).
 407                  *
 408                  * TODO: We should implement a timer to garbage collect
 409                  * streams after a timeout to be resilient against a
 410                  * consumerd implementation that would not match this
 411                  * expected behavior.
 412                  */
 413                 pthread_mutex_unlock(&stream->lock);
 414                 DBG("closing stream %" PRIu64 " aborted since it still has data pending", stream->stream_handle);
 415                 return;
 416         }
 417         /*
 418          * We received all the indexes we can expect.
 419          */
 420         stream_unpublish(stream);
 421         stream->closed = true;
 422         /* Relay indexes are only used by the "consumer/sessiond" end. */
 423         relay_index_close_all(stream);
 424         pthread_mutex_unlock(&stream->lock);
 425         DBG("Succeeded in closing stream %" PRIu64, stream->stream_handle);
 426         stream_put(stream);
 427 }
 428
 429 static void print_stream_indexes(struct relay_stream *stream)
 430 {
 431         struct lttng_ht_iter iter;
 432         struct relay_index *index;
 433
 434         rcu_read_lock();
 435         cds_lfht_for_each_entry(stream->indexes_ht->ht, &iter.iter, index,
 436                         index_n.node) {
 437                 DBG("index %p net_seq_num %" PRIu64 " refcount %ld"
 438                                 " stream %" PRIu64 " trace %" PRIu64
 439                                 " session %" PRIu64,
 440                                 index,
 441                                 index->index_n.key,
 442                                 stream->ref.refcount,
 443                                 index->stream->stream_handle,
 444                                 index->stream->trace->id,
 445                                 index->stream->trace->session->id);
 446         }
 447         rcu_read_unlock();
 448 }
 449
 450 void print_relay_streams(void)
 451 {
 452         struct lttng_ht_iter iter;
 453         struct relay_stream *stream;
 454
 455         if (!relay_streams_ht) {
 456                 return;
 457         }
 458
 459         rcu_read_lock();
 460         cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
 461                         node.node) {
 462                 if (!stream_get(stream)) {
 463                         continue;
 464                 }
 465                 DBG("stream %p refcount %ld stream %" PRIu64 " trace %" PRIu64
 466                                 " session %" PRIu64,
 467                                 stream,
 468                                 stream->ref.refcount,
 469                                 stream->stream_handle,
 470                                 stream->trace->id,
 471                                 stream->trace->session->id);
 472                 print_stream_indexes(stream);
 473                 stream_put(stream);
 474         }
 475         rcu_read_unlock();
 476 }