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