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