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