src/common/consumer-stream.c

   1 /*
   2  * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
   3  *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  * Copyright (C) 2013 - David Goulet <dgoulet@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 _GNU_SOURCE
  21 #include <assert.h>
  22 #include <inttypes.h>
  23 #include <sys/mman.h>
  24 #include <unistd.h>
  25
  26 #include <common/common.h>
  27 #include <common/index/index.h>
  28 #include <common/kernel-consumer/kernel-consumer.h>
  29 #include <common/relayd/relayd.h>
  30 #include <common/ust-consumer/ust-consumer.h>
  31
  32 #include "consumer-stream.h"
  33
  34 /*
  35  * RCU call to free stream. MUST only be used with call_rcu().
  36  */
  37 static void free_stream_rcu(struct rcu_head *head)
  38 {
  39         struct lttng_ht_node_u64 *node =
  40                 caa_container_of(head, struct lttng_ht_node_u64, head);
  41         struct lttng_consumer_stream *stream =
  42                 caa_container_of(node, struct lttng_consumer_stream, node);
  43
  44         pthread_mutex_destroy(&stream->lock);
  45         free(stream);
  46 }
  47
  48 /*
  49  * Close stream on the relayd side. This call can destroy a relayd if the
  50  * conditions are met.
  51  *
  52  * A RCU read side lock MUST be acquired if the relayd object was looked up in
  53  * a hash table before calling this.
  54  */
  55 void consumer_stream_relayd_close(struct lttng_consumer_stream *stream,
  56                 struct consumer_relayd_sock_pair *relayd)
  57 {
  58         int ret;
  59
  60         assert(stream);
  61         assert(relayd);
  62
  63         if (stream->sent_to_relayd) {
  64                 uatomic_dec(&relayd->refcount);
  65                 assert(uatomic_read(&relayd->refcount) >= 0);
  66         }
  67
  68         /* Closing streams requires to lock the control socket. */
  69         pthread_mutex_lock(&relayd->ctrl_sock_mutex);
  70         ret = relayd_send_close_stream(&relayd->control_sock,
  71                         stream->relayd_stream_id,
  72                         stream->next_net_seq_num - 1);
  73         pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
  74         if (ret < 0) {
  75                 DBG("Unable to close stream on the relayd. Continuing");
  76                 /*
  77                  * Continue here. There is nothing we can do for the relayd.
  78                  * Chances are that the relayd has closed the socket so we just
  79                  * continue cleaning up.
  80                  */
  81         }
  82
  83         /* Both conditions are met, we destroy the relayd. */
  84         if (uatomic_read(&relayd->refcount) == 0 &&
  85                         uatomic_read(&relayd->destroy_flag)) {
  86                 consumer_destroy_relayd(relayd);
  87         }
  88         stream->net_seq_idx = (uint64_t) -1ULL;
  89         stream->sent_to_relayd = 0;
  90 }
  91
  92 /*
  93  * Close stream's file descriptors and, if needed, close stream also on the
  94  * relayd side.
  95  *
  96  * The consumer data lock MUST be acquired.
  97  * The stream lock MUST be acquired.
  98  */
  99 void consumer_stream_close(struct lttng_consumer_stream *stream)
 100 {
 101         int ret;
 102         struct consumer_relayd_sock_pair *relayd;
 103
 104         assert(stream);
 105
 106         switch (consumer_data.type) {
 107         case LTTNG_CONSUMER_KERNEL:
 108                 if (stream->mmap_base != NULL) {
 109                         ret = munmap(stream->mmap_base, stream->mmap_len);
 110                         if (ret != 0) {
 111                                 PERROR("munmap");
 112                         }
 113                 }
 114
 115                 if (stream->wait_fd >= 0) {
 116                         ret = close(stream->wait_fd);
 117                         if (ret) {
 118                                 PERROR("close");
 119                         }
 120                         stream->wait_fd = -1;
 121                 }
 122                 break;
 123         case LTTNG_CONSUMER32_UST:
 124         case LTTNG_CONSUMER64_UST:
 125         {
 126                 /*
 127                  * Special case for the metadata since the wait fd is an internal pipe
 128                  * polled in the metadata thread.
 129                  */
 130                 if (stream->metadata_flag && stream->chan->monitor) {
 131                         int rpipe = stream->ust_metadata_poll_pipe[0];
 132
 133                         /*
 134                          * This will stop the channel timer if one and close the write side
 135                          * of the metadata poll pipe.
 136                          */
 137                         lttng_ustconsumer_close_metadata(stream->chan);
 138                         if (rpipe >= 0) {
 139                                 ret = close(rpipe);
 140                                 if (ret < 0) {
 141                                         PERROR("closing metadata pipe read side");
 142                                 }
 143                                 stream->ust_metadata_poll_pipe[0] = -1;
 144                         }
 145                 }
 146                 break;
 147         }
 148         default:
 149                 ERR("Unknown consumer_data type");
 150                 assert(0);
 151         }
 152
 153         /* Close output fd. Could be a socket or local file at this point. */
 154         if (stream->out_fd >= 0) {
 155                 ret = close(stream->out_fd);
 156                 if (ret) {
 157                         PERROR("close");
 158                 }
 159                 stream->out_fd = -1;
 160         }
 161
 162         if (stream->index_fd >= 0) {
 163                 ret = close(stream->index_fd);
 164                 if (ret) {
 165                         PERROR("close stream index_fd");
 166                 }
 167                 stream->index_fd = -1;
 168         }
 169
 170         /* Check and cleanup relayd if needed. */
 171         rcu_read_lock();
 172         relayd = consumer_find_relayd(stream->net_seq_idx);
 173         if (relayd != NULL) {
 174                 consumer_stream_relayd_close(stream, relayd);
 175         }
 176         rcu_read_unlock();
 177 }
 178
 179 /*
 180  * Delete the stream from all possible hash tables.
 181  *
 182  * The consumer data lock MUST be acquired.
 183  * The stream lock MUST be acquired.
 184  */
 185 void consumer_stream_delete(struct lttng_consumer_stream *stream,
 186                 struct lttng_ht *ht)
 187 {
 188         int ret;
 189         struct lttng_ht_iter iter;
 190
 191         assert(stream);
 192         /* Should NEVER be called not in monitor mode. */
 193         assert(stream->chan->monitor);
 194
 195         rcu_read_lock();
 196
 197         if (ht) {
 198                 iter.iter.node = &stream->node.node;
 199                 ret = lttng_ht_del(ht, &iter);
 200                 assert(!ret);
 201         }
 202
 203         /* Delete from stream per channel ID hash table. */
 204         iter.iter.node = &stream->node_channel_id.node;
 205         /*
 206          * The returned value is of no importance. Even if the node is NOT in the
 207          * hash table, we continue since we may have been called by a code path
 208          * that did not add the stream to a (all) hash table. Same goes for the
 209          * next call ht del call.
 210          */
 211         (void) lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter);
 212
 213         /* Delete from the global stream list. */
 214         iter.iter.node = &stream->node_session_id.node;
 215         /* See the previous ht del on why we ignore the returned value. */
 216         (void) lttng_ht_del(consumer_data.stream_list_ht, &iter);
 217
 218         rcu_read_unlock();
 219
 220         if (!stream->metadata_flag) {
 221                 /* Decrement the stream count of the global consumer data. */
 222                 assert(consumer_data.stream_count > 0);
 223                 consumer_data.stream_count--;
 224         }
 225 }
 226
 227 /*
 228  * Free the given stream within a RCU call.
 229  */
 230 void consumer_stream_free(struct lttng_consumer_stream *stream)
 231 {
 232         assert(stream);
 233
 234         call_rcu(&stream->node.head, free_stream_rcu);
 235 }
 236
 237 /*
 238  * Destroy the stream's buffers of the tracer.
 239  */
 240 void consumer_stream_destroy_buffers(struct lttng_consumer_stream *stream)
 241 {
 242         assert(stream);
 243
 244         switch (consumer_data.type) {
 245         case LTTNG_CONSUMER_KERNEL:
 246                 break;
 247         case LTTNG_CONSUMER32_UST:
 248         case LTTNG_CONSUMER64_UST:
 249                 lttng_ustconsumer_del_stream(stream);
 250                 break;
 251         default:
 252                 ERR("Unknown consumer_data type");
 253                 assert(0);
 254         }
 255 }
 256
 257 /*
 258  * Destroy and close a already created stream.
 259  */
 260 static void destroy_close_stream(struct lttng_consumer_stream *stream)
 261 {
 262         assert(stream);
 263
 264         DBG("Consumer stream destroy monitored key: %" PRIu64, stream->key);
 265
 266         /* Destroy tracer buffers of the stream. */
 267         consumer_stream_destroy_buffers(stream);
 268         /* Close down everything including the relayd if one. */
 269         consumer_stream_close(stream);
 270 }
 271
 272 /*
 273  * Decrement the stream's channel refcount and if down to 0, return the channel
 274  * pointer so it can be destroyed by the caller or NULL if not.
 275  */
 276 static struct lttng_consumer_channel *unref_channel(
 277                 struct lttng_consumer_stream *stream)
 278 {
 279         struct lttng_consumer_channel *free_chan = NULL;
 280
 281         assert(stream);
 282         assert(stream->chan);
 283
 284         /* Update refcount of channel and see if we need to destroy it. */
 285         if (!uatomic_sub_return(&stream->chan->refcount, 1)
 286                         && !uatomic_read(&stream->chan->nb_init_stream_left)) {
 287                 free_chan = stream->chan;
 288         }
 289
 290         return free_chan;
 291 }
 292
 293 /*
 294  * Destroy a stream completely. This will delete, close and free the stream.
 295  * Once return, the stream is NO longer usable. Its channel may get destroyed
 296  * if conditions are met for a monitored stream.
 297  *
 298  * This MUST be called WITHOUT the consumer data and stream lock acquired if
 299  * the stream is in _monitor_ mode else it does not matter.
 300  */
 301 void consumer_stream_destroy(struct lttng_consumer_stream *stream,
 302                 struct lttng_ht *ht)
 303 {
 304         assert(stream);
 305
 306         /* Stream is in monitor mode. */
 307         if (stream->monitor) {
 308                 struct lttng_consumer_channel *free_chan = NULL;
 309
 310                 /*
 311                  * This means that the stream was successfully removed from the streams
 312                  * list of the channel and sent to the right thread managing this
 313                  * stream thus being globally visible.
 314                  */
 315                 if (stream->globally_visible) {
 316                         pthread_mutex_lock(&consumer_data.lock);
 317                         pthread_mutex_lock(&stream->chan->lock);
 318                         pthread_mutex_lock(&stream->lock);
 319                         /* Remove every reference of the stream in the consumer. */
 320                         consumer_stream_delete(stream, ht);
 321
 322                         destroy_close_stream(stream);
 323
 324                         /* Update channel's refcount of the stream. */
 325                         free_chan = unref_channel(stream);
 326
 327                         /* Indicates that the consumer data state MUST be updated after this. */
 328                         consumer_data.need_update = 1;
 329
 330                         pthread_mutex_unlock(&stream->lock);
 331                         pthread_mutex_unlock(&stream->chan->lock);
 332                         pthread_mutex_unlock(&consumer_data.lock);
 333                 } else {
 334                         /*
 335                          * If the stream is not visible globally, this needs to be done
 336                          * outside of the consumer data lock section.
 337                          */
 338                         free_chan = unref_channel(stream);
 339                 }
 340
 341                 if (free_chan) {
 342                         consumer_del_channel(free_chan);
 343                 }
 344         } else {
 345                 destroy_close_stream(stream);
 346         }
 347
 348         /* Free stream within a RCU call. */
 349         consumer_stream_free(stream);
 350 }
 351
 352 /*
 353  * Write index of a specific stream either on the relayd or local disk.
 354  *
 355  * Return 0 on success or else a negative value.
 356  */
 357 int consumer_stream_write_index(struct lttng_consumer_stream *stream,
 358                 struct ctf_packet_index *index)
 359 {
 360         int ret;
 361         struct consumer_relayd_sock_pair *relayd;
 362
 363         assert(stream);
 364         assert(index);
 365
 366         rcu_read_lock();
 367         relayd = consumer_find_relayd(stream->net_seq_idx);
 368         if (relayd) {
 369                 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
 370                 ret = relayd_send_index(&relayd->control_sock, index,
 371                                 stream->relayd_stream_id, stream->next_net_seq_num - 1);
 372                 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 373         } else {
 374                 ssize_t size_ret;
 375
 376                 size_ret = index_write(stream->index_fd, index,
 377                                 sizeof(struct ctf_packet_index));
 378                 if (size_ret < sizeof(struct ctf_packet_index)) {
 379                         ret = -1;
 380                 } else {
 381                         ret = 0;
 382                 }
 383         }
 384         if (ret < 0) {
 385                 goto error;
 386         }
 387
 388 error:
 389         rcu_read_unlock();
 390         return ret;
 391 }
 392
 393 /*
 394  * Actually do the metadata sync using the given metadata stream.
 395  *
 396  * Return 0 on success else a negative value. ENODATA can be returned also
 397  * indicating that there is no metadata available for that stream.
 398  */
 399 static int do_sync_metadata(struct lttng_consumer_stream *metadata,
 400                 struct lttng_consumer_local_data *ctx)
 401 {
 402         int ret;
 403
 404         assert(metadata);
 405         assert(metadata->metadata_flag);
 406         assert(ctx);
 407
 408         /*
 409          * In UST, since we have to write the metadata from the cache packet
 410          * by packet, we might need to start this procedure multiple times
 411          * until all the metadata from the cache has been extracted.
 412          */
 413         do {
 414                 /*
 415                  * Steps :
 416                  * - Lock the metadata stream
 417                  * - Check if metadata stream node was deleted before locking.
 418                  *   - if yes, release and return success
 419                  * - Check if new metadata is ready (flush + snapshot pos)
 420                  * - If nothing : release and return.
 421                  * - Lock the metadata_rdv_lock
 422                  * - Unlock the metadata stream
 423                  * - cond_wait on metadata_rdv to wait the wakeup from the
 424                  *   metadata thread
 425                  * - Unlock the metadata_rdv_lock
 426                  */
 427                 pthread_mutex_lock(&metadata->lock);
 428
 429                 /*
 430                  * There is a possibility that we were able to acquire a reference on the
 431                  * stream from the RCU hash table but between then and now, the node might
 432                  * have been deleted just before the lock is acquired. Thus, after locking,
 433                  * we make sure the metadata node has not been deleted which means that the
 434                  * buffers are closed.
 435                  *
 436                  * In that case, there is no need to sync the metadata hence returning a
 437                  * success return code.
 438                  */
 439                 ret = cds_lfht_is_node_deleted(&metadata->node.node);
 440                 if (ret) {
 441                         ret = 0;
 442                         goto end_unlock_mutex;
 443                 }
 444
 445                 switch (ctx->type) {
 446                 case LTTNG_CONSUMER_KERNEL:
 447                         /*
 448                          * Empty the metadata cache and flush the current stream.
 449                          */
 450                         ret = lttng_kconsumer_sync_metadata(metadata);
 451                         break;
 452                 case LTTNG_CONSUMER32_UST:
 453                 case LTTNG_CONSUMER64_UST:
 454                         /*
 455                          * Ask the sessiond if we have new metadata waiting and update the
 456                          * consumer metadata cache.
 457                          */
 458                         ret = lttng_ustconsumer_sync_metadata(ctx, metadata);
 459                         break;
 460                 default:
 461                         assert(0);
 462                         ret = -1;
 463                         break;
 464                 }
 465                 /*
 466                  * Error or no new metadata, we exit here.
 467                  */
 468                 if (ret <= 0 || ret == ENODATA) {
 469                         goto end_unlock_mutex;
 470                 }
 471
 472                 /*
 473                  * At this point, new metadata have been flushed, so we wait on the
 474                  * rendez-vous point for the metadata thread to wake us up when it
 475                  * finishes consuming the metadata and continue execution.
 476                  */
 477
 478                 pthread_mutex_lock(&metadata->metadata_rdv_lock);
 479
 480                 /*
 481                  * Release metadata stream lock so the metadata thread can process it.
 482                  */
 483                 pthread_mutex_unlock(&metadata->lock);
 484
 485                 /*
 486                  * Wait on the rendez-vous point. Once woken up, it means the metadata was
 487                  * consumed and thus synchronization is achieved.
 488                  */
 489                 pthread_cond_wait(&metadata->metadata_rdv, &metadata->metadata_rdv_lock);
 490                 pthread_mutex_unlock(&metadata->metadata_rdv_lock);
 491         } while (ret == EAGAIN);
 492
 493         /* Success */
 494         return 0;
 495
 496 end_unlock_mutex:
 497         pthread_mutex_unlock(&metadata->lock);
 498         return ret;
 499 }
 500
 501 /*
 502  * Synchronize the metadata using a given session ID. A successful acquisition
 503  * of a metadata stream will trigger a request to the session daemon and a
 504  * snapshot so the metadata thread can consume it.
 505  *
 506  * This function call is a rendez-vous point between the metadata thread and
 507  * the data thread.
 508  *
 509  * Return 0 on success or else a negative value.
 510  */
 511 int consumer_stream_sync_metadata(struct lttng_consumer_local_data *ctx,
 512                 uint64_t session_id)
 513 {
 514         int ret;
 515         struct lttng_consumer_stream *stream = NULL;
 516         struct lttng_ht_iter iter;
 517         struct lttng_ht *ht;
 518
 519         assert(ctx);
 520
 521         /* Ease our life a bit. */
 522         ht = consumer_data.stream_list_ht;
 523
 524         rcu_read_lock();
 525
 526         /* Search the metadata associated with the session id of the given stream. */
 527
 528         cds_lfht_for_each_entry_duplicate(ht->ht,
 529                         ht->hash_fct(&session_id, lttng_ht_seed), ht->match_fct,
 530                         &session_id, &iter.iter, stream, node_session_id.node) {
 531                 if (!stream->metadata_flag) {
 532                         continue;
 533                 }
 534
 535                 ret = do_sync_metadata(stream, ctx);
 536                 if (ret < 0) {
 537                         goto end;
 538                 }
 539         }
 540
 541         /*
 542          * Force return code to 0 (success) since ret might be ENODATA for instance
 543          * which is not an error but rather that we should come back.
 544          */
 545         ret = 0;
 546
 547 end:
 548         rcu_read_unlock();
 549         return ret;
 550 }