src/bin/lttng-relayd/live.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
   7  * it under the terms of the GNU General Public License, version 2 only,
   8  * as 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
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19
  20 #define _GNU_SOURCE
  21 #include <getopt.h>
  22 #include <grp.h>
  23 #include <limits.h>
  24 #include <pthread.h>
  25 #include <signal.h>
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include <sys/mman.h>
  30 #include <sys/mount.h>
  31 #include <sys/resource.h>
  32 #include <sys/socket.h>
  33 #include <sys/stat.h>
  34 #include <sys/types.h>
  35 #include <sys/wait.h>
  36 #include <inttypes.h>
  37 #include <urcu/futex.h>
  38 #include <urcu/uatomic.h>
  39 #include <urcu/rculist.h>
  40 #include <unistd.h>
  41 #include <fcntl.h>
  42 #include <config.h>
  43
  44 #include <lttng/lttng.h>
  45 #include <common/common.h>
  46 #include <common/compat/poll.h>
  47 #include <common/compat/socket.h>
  48 #include <common/compat/endian.h>
  49 #include <common/defaults.h>
  50 #include <common/futex.h>
  51 #include <common/index/index.h>
  52 #include <common/sessiond-comm/sessiond-comm.h>
  53 #include <common/sessiond-comm/inet.h>
  54 #include <common/sessiond-comm/relayd.h>
  55 #include <common/uri.h>
  56 #include <common/utils.h>
  57
  58 #include "cmd.h"
  59 #include "live.h"
  60 #include "lttng-relayd.h"
  61 #include "utils.h"
  62 #include "health-relayd.h"
  63 #include "testpoint.h"
  64 #include "viewer-stream.h"
  65 #include "stream.h"
  66 #include "session.h"
  67 #include "ctf-trace.h"
  68 #include "connection.h"
  69 #include "viewer-session.h"
  70
  71 #define SESSION_BUF_DEFAULT_COUNT       16
  72
  73 static struct lttng_uri *live_uri;
  74
  75 /*
  76  * This pipe is used to inform the worker thread that a command is queued and
  77  * ready to be processed.
  78  */
  79 static int live_conn_pipe[2] = { -1, -1 };
  80
  81 /* Shared between threads */
  82 static int live_dispatch_thread_exit;
  83
  84 static pthread_t live_listener_thread;
  85 static pthread_t live_dispatcher_thread;
  86 static pthread_t live_worker_thread;
  87
  88 /*
  89  * Relay command queue.
  90  *
  91  * The live_thread_listener and live_thread_dispatcher communicate with this
  92  * queue.
  93  */
  94 static struct relay_conn_queue viewer_conn_queue;
  95
  96 static uint64_t last_relay_viewer_session_id;
  97 static pthread_mutex_t last_relay_viewer_session_id_lock =
  98                 PTHREAD_MUTEX_INITIALIZER;
  99
 100 /*
 101  * Cleanup the daemon
 102  */
 103 static
 104 void cleanup_relayd_live(void)
 105 {
 106         DBG("Cleaning up");
 107
 108         free(live_uri);
 109 }
 110
 111 /*
 112  * Receive a request buffer using a given socket, destination allocated buffer
 113  * of length size.
 114  *
 115  * Return the size of the received message or else a negative value on error
 116  * with errno being set by recvmsg() syscall.
 117  */
 118 static
 119 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 120 {
 121         ssize_t ret;
 122
 123         ret = sock->ops->recvmsg(sock, buf, size, 0);
 124         if (ret < 0 || ret != size) {
 125                 if (ret == 0) {
 126                         /* Orderly shutdown. Not necessary to print an error. */
 127                         DBG("Socket %d did an orderly shutdown", sock->fd);
 128                 } else {
 129                         ERR("Relay failed to receive request.");
 130                 }
 131                 ret = -1;
 132         }
 133
 134         return ret;
 135 }
 136
 137 /*
 138  * Send a response buffer using a given socket, source allocated buffer of
 139  * length size.
 140  *
 141  * Return the size of the sent message or else a negative value on error with
 142  * errno being set by sendmsg() syscall.
 143  */
 144 static
 145 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 146 {
 147         ssize_t ret;
 148
 149         ret = sock->ops->sendmsg(sock, buf, size, 0);
 150         if (ret < 0) {
 151                 ERR("Relayd failed to send response.");
 152         }
 153
 154         return ret;
 155 }
 156
 157 /*
 158  * Atomically check if new streams got added in one of the sessions attached
 159  * and reset the flag to 0.
 160  *
 161  * Returns 1 if new streams got added, 0 if nothing changed, a negative value
 162  * on error.
 163  */
 164 static
 165 int check_new_streams(struct relay_connection *conn)
 166 {
 167         struct relay_session *session;
 168         unsigned long current_val;
 169         int ret = 0;
 170
 171         if (!conn->viewer_session) {
 172                 goto end;
 173         }
 174         rcu_read_lock();
 175         cds_list_for_each_entry_rcu(session,
 176                         &conn->viewer_session->session_list,
 177                         viewer_session_node) {
 178                 if (!session_get(session)) {
 179                         continue;
 180                 }
 181                 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
 182                 ret = current_val;
 183                 session_put(session);
 184                 if (ret == 1) {
 185                         goto end;
 186                 }
 187         }
 188 end:
 189         rcu_read_unlock();
 190         return ret;
 191 }
 192
 193 /*
 194  * Send viewer streams to the given socket. The ignore_sent_flag indicates if
 195  * this function should ignore the sent flag or not.
 196  *
 197  * Return 0 on success or else a negative value.
 198  */
 199 static
 200 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
 201                 struct relay_session *session, unsigned int ignore_sent_flag)
 202 {
 203         ssize_t ret;
 204         struct lttng_viewer_stream send_stream;
 205         struct lttng_ht_iter iter;
 206         struct relay_viewer_stream *vstream;
 207
 208         rcu_read_lock();
 209
 210         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
 211                         stream_n.node) {
 212                 struct ctf_trace *ctf_trace;
 213
 214                 health_code_update();
 215
 216                 if (!viewer_stream_get(vstream)) {
 217                         continue;
 218                 }
 219
 220                 pthread_mutex_lock(&vstream->stream->lock);
 221                 /* Ignore if not the same session. */
 222                 if (vstream->stream->trace->session->id != session->id ||
 223                                 (!ignore_sent_flag && vstream->sent_flag)) {
 224                         pthread_mutex_unlock(&vstream->stream->lock);
 225                         viewer_stream_put(vstream);
 226                         continue;
 227                 }
 228
 229                 ctf_trace = vstream->stream->trace;
 230                 send_stream.id = htobe64(vstream->stream->stream_handle);
 231                 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
 232                 send_stream.metadata_flag = htobe32(
 233                                 vstream->stream->is_metadata);
 234                 if (lttng_strncpy(send_stream.path_name, vstream->path_name,
 235                                 sizeof(send_stream.path_name))) {
 236                         pthread_mutex_unlock(&vstream->stream->lock);
 237                         viewer_stream_put(vstream);
 238                         ret = -1;       /* Error. */
 239                         goto end_unlock;
 240                 }
 241                 if (lttng_strncpy(send_stream.channel_name,
 242                                 vstream->channel_name,
 243                                 sizeof(send_stream.channel_name))) {
 244                         pthread_mutex_unlock(&vstream->stream->lock);
 245                         viewer_stream_put(vstream);
 246                         ret = -1;       /* Error. */
 247                         goto end_unlock;
 248                 }
 249
 250                 DBG("Sending stream %" PRIu64 " to viewer",
 251                                 vstream->stream->stream_handle);
 252                 vstream->sent_flag = 1;
 253                 pthread_mutex_unlock(&vstream->stream->lock);
 254
 255                 ret = send_response(sock, &send_stream, sizeof(send_stream));
 256                 viewer_stream_put(vstream);
 257                 if (ret < 0) {
 258                         goto end_unlock;
 259                 }
 260         }
 261
 262         ret = 0;
 263
 264 end_unlock:
 265         rcu_read_unlock();
 266         return ret;
 267 }
 268
 269 /*
 270  * Create every viewer stream possible for the given session with the seek
 271  * type. Three counters *can* be return which are in order the total amount of
 272  * viewer stream of the session, the number of unsent stream and the number of
 273  * stream created. Those counters can be NULL and thus will be ignored.
 274  *
 275  * Return 0 on success or else a negative value.
 276  */
 277 static
 278 int make_viewer_streams(struct relay_session *session,
 279                 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
 280                 uint32_t *nb_created, bool *closed)
 281 {
 282         int ret;
 283         struct lttng_ht_iter iter;
 284         struct ctf_trace *ctf_trace;
 285
 286         assert(session);
 287
 288         /*
 289          * Hold the session lock to ensure that we see either none or
 290          * all initial streams for a session, but no intermediate state.
 291          */
 292         pthread_mutex_lock(&session->lock);
 293
 294         if (session->connection_closed) {
 295                 *closed = true;
 296         }
 297
 298         /*
 299          * Create viewer streams for relay streams that are ready to be
 300          * used for a the given session id only.
 301          */
 302         rcu_read_lock();
 303         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 304                         node.node) {
 305                 struct relay_stream *stream;
 306
 307                 health_code_update();
 308
 309                 if (!ctf_trace_get(ctf_trace)) {
 310                         continue;
 311                 }
 312
 313                 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
 314                         struct relay_viewer_stream *vstream;
 315
 316                         if (!stream_get(stream)) {
 317                                 continue;
 318                         }
 319                         /*
 320                          * stream published is protected by the session lock.
 321                          */
 322                         if (!stream->published) {
 323                                 goto next;
 324                         }
 325                         vstream = viewer_stream_get_by_id(stream->stream_handle);
 326                         if (!vstream) {
 327                                 vstream = viewer_stream_create(stream, seek_t);
 328                                 if (!vstream) {
 329                                         ret = -1;
 330                                         ctf_trace_put(ctf_trace);
 331                                         stream_put(stream);
 332                                         goto error_unlock;
 333                                 }
 334
 335                                 if (nb_created) {
 336                                         /* Update number of created stream counter. */
 337                                         (*nb_created)++;
 338                                 }
 339                                 /*
 340                                  * Ensure a self-reference is preserved even
 341                                  * after we have put our local reference.
 342                                  */
 343                                 viewer_stream_get(vstream);
 344                         } else {
 345                                 if (!vstream->sent_flag && nb_unsent) {
 346                                         /* Update number of unsent stream counter. */
 347                                         (*nb_unsent)++;
 348                                 }
 349                         }
 350                         /* Update number of total stream counter. */
 351                         if (nb_total) {
 352                                 if (stream->is_metadata) {
 353                                         if (!stream->closed ||
 354                                                         stream->metadata_received > vstream->metadata_sent) {
 355                                                 (*nb_total)++;
 356                                         }
 357                                 } else {
 358                                         if (!stream->closed ||
 359                                                 !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
 360
 361                                                 (*nb_total)++;
 362                                         }
 363                                 }
 364                         }
 365                         /* Put local reference. */
 366                         viewer_stream_put(vstream);
 367                 next:
 368                         stream_put(stream);
 369                 }
 370                 ctf_trace_put(ctf_trace);
 371         }
 372
 373         ret = 0;
 374
 375 error_unlock:
 376         rcu_read_unlock();
 377         pthread_mutex_unlock(&session->lock);
 378         return ret;
 379 }
 380
 381 int relayd_live_stop(void)
 382 {
 383         /* Stop dispatch thread */
 384         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 385         futex_nto1_wake(&viewer_conn_queue.futex);
 386         return 0;
 387 }
 388
 389 /*
 390  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 391  */
 392 static
 393 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 394 {
 395         int ret;
 396
 397         if (events == NULL || size == 0) {
 398                 ret = -1;
 399                 goto error;
 400         }
 401
 402         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 403         if (ret < 0) {
 404                 goto error;
 405         }
 406
 407         /* Add quit pipe */
 408         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 409         if (ret < 0) {
 410                 goto error;
 411         }
 412
 413         return 0;
 414
 415 error:
 416         return ret;
 417 }
 418
 419 /*
 420  * Check if the thread quit pipe was triggered.
 421  *
 422  * Return 1 if it was triggered else 0;
 423  */
 424 static
 425 int check_thread_quit_pipe(int fd, uint32_t events)
 426 {
 427         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 428                 return 1;
 429         }
 430
 431         return 0;
 432 }
 433
 434 /*
 435  * Create and init socket from uri.
 436  */
 437 static
 438 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 439 {
 440         int ret;
 441         struct lttcomm_sock *sock = NULL;
 442
 443         sock = lttcomm_alloc_sock_from_uri(uri);
 444         if (sock == NULL) {
 445                 ERR("Allocating socket");
 446                 goto error;
 447         }
 448
 449         ret = lttcomm_create_sock(sock);
 450         if (ret < 0) {
 451                 goto error;
 452         }
 453         DBG("Listening on sock %d for live", sock->fd);
 454
 455         ret = sock->ops->bind(sock);
 456         if (ret < 0) {
 457                 goto error;
 458         }
 459
 460         ret = sock->ops->listen(sock, -1);
 461         if (ret < 0) {
 462                 goto error;
 463
 464         }
 465
 466         return sock;
 467
 468 error:
 469         if (sock) {
 470                 lttcomm_destroy_sock(sock);
 471         }
 472         return NULL;
 473 }
 474
 475 /*
 476  * This thread manages the listening for new connections on the network
 477  */
 478 static
 479 void *thread_listener(void *data)
 480 {
 481         int i, ret, pollfd, err = -1;
 482         uint32_t revents, nb_fd;
 483         struct lttng_poll_event events;
 484         struct lttcomm_sock *live_control_sock;
 485
 486         DBG("[thread] Relay live listener started");
 487
 488         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 489
 490         health_code_update();
 491
 492         live_control_sock = init_socket(live_uri);
 493         if (!live_control_sock) {
 494                 goto error_sock_control;
 495         }
 496
 497         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 498         ret = create_thread_poll_set(&events, 2);
 499         if (ret < 0) {
 500                 goto error_create_poll;
 501         }
 502
 503         /* Add the control socket */
 504         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 505         if (ret < 0) {
 506                 goto error_poll_add;
 507         }
 508
 509         lttng_relay_notify_ready();
 510
 511         if (testpoint(relayd_thread_live_listener)) {
 512                 goto error_testpoint;
 513         }
 514
 515         while (1) {
 516                 health_code_update();
 517
 518                 DBG("Listener accepting live viewers connections");
 519
 520 restart:
 521                 health_poll_entry();
 522                 ret = lttng_poll_wait(&events, -1);
 523                 health_poll_exit();
 524                 if (ret < 0) {
 525                         /*
 526                          * Restart interrupted system call.
 527                          */
 528                         if (errno == EINTR) {
 529                                 goto restart;
 530                         }
 531                         goto error;
 532                 }
 533                 nb_fd = ret;
 534
 535                 DBG("Relay new viewer connection received");
 536                 for (i = 0; i < nb_fd; i++) {
 537                         health_code_update();
 538
 539                         /* Fetch once the poll data */
 540                         revents = LTTNG_POLL_GETEV(&events, i);
 541                         pollfd = LTTNG_POLL_GETFD(&events, i);
 542
 543                         if (!revents) {
 544                                 /* No activity for this FD (poll implementation). */
 545                                 continue;
 546                         }
 547
 548                         /* Thread quit pipe has been closed. Killing thread. */
 549                         ret = check_thread_quit_pipe(pollfd, revents);
 550                         if (ret) {
 551                                 err = 0;
 552                                 goto exit;
 553                         }
 554
 555                         if (revents & LPOLLIN) {
 556                                 /*
 557                                  * A new connection is requested, therefore a
 558                                  * viewer connection is allocated in this
 559                                  * thread, enqueued to a global queue and
 560                                  * dequeued (and freed) in the worker thread.
 561                                  */
 562                                 int val = 1;
 563                                 struct relay_connection *new_conn;
 564                                 struct lttcomm_sock *newsock;
 565
 566                                 newsock = live_control_sock->ops->accept(live_control_sock);
 567                                 if (!newsock) {
 568                                         PERROR("accepting control sock");
 569                                         goto error;
 570                                 }
 571                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 572
 573                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 574                                                 sizeof(val));
 575                                 if (ret < 0) {
 576                                         PERROR("setsockopt inet");
 577                                         lttcomm_destroy_sock(newsock);
 578                                         goto error;
 579                                 }
 580                                 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
 581                                 if (!new_conn) {
 582                                         lttcomm_destroy_sock(newsock);
 583                                         goto error;
 584                                 }
 585                                 /* Ownership assumed by the connection. */
 586                                 newsock = NULL;
 587
 588                                 /* Enqueue request for the dispatcher thread. */
 589                                 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
 590                                                  &new_conn->qnode);
 591
 592                                 /*
 593                                  * Wake the dispatch queue futex.
 594                                  * Implicit memory barrier with the
 595                                  * exchange in cds_wfcq_enqueue.
 596                                  */
 597                                 futex_nto1_wake(&viewer_conn_queue.futex);
 598                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 599                                 ERR("socket poll error");
 600                                 goto error;
 601                         } else {
 602                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
 603                                 goto error;
 604                         }
 605                 }
 606         }
 607
 608 exit:
 609 error:
 610 error_poll_add:
 611 error_testpoint:
 612         lttng_poll_clean(&events);
 613 error_create_poll:
 614         if (live_control_sock->fd >= 0) {
 615                 ret = live_control_sock->ops->close(live_control_sock);
 616                 if (ret) {
 617                         PERROR("close");
 618                 }
 619         }
 620         lttcomm_destroy_sock(live_control_sock);
 621 error_sock_control:
 622         if (err) {
 623                 health_error();
 624                 DBG("Live viewer listener thread exited with error");
 625         }
 626         health_unregister(health_relayd);
 627         DBG("Live viewer listener thread cleanup complete");
 628         if (lttng_relay_stop_threads()) {
 629                 ERR("Error stopping threads");
 630         }
 631         return NULL;
 632 }
 633
 634 /*
 635  * This thread manages the dispatching of the requests to worker threads
 636  */
 637 static
 638 void *thread_dispatcher(void *data)
 639 {
 640         int err = -1;
 641         ssize_t ret;
 642         struct cds_wfcq_node *node;
 643         struct relay_connection *conn = NULL;
 644
 645         DBG("[thread] Live viewer relay dispatcher started");
 646
 647         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 648
 649         if (testpoint(relayd_thread_live_dispatcher)) {
 650                 goto error_testpoint;
 651         }
 652
 653         health_code_update();
 654
 655         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 656                 health_code_update();
 657
 658                 /* Atomically prepare the queue futex */
 659                 futex_nto1_prepare(&viewer_conn_queue.futex);
 660
 661                 do {
 662                         health_code_update();
 663
 664                         /* Dequeue commands */
 665                         node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
 666                                                          &viewer_conn_queue.tail);
 667                         if (node == NULL) {
 668                                 DBG("Woken up but nothing in the live-viewer "
 669                                                 "relay command queue");
 670                                 /* Continue thread execution */
 671                                 break;
 672                         }
 673                         conn = caa_container_of(node, struct relay_connection, qnode);
 674                         DBG("Dispatching viewer request waiting on sock %d",
 675                                         conn->sock->fd);
 676
 677                         /*
 678                          * Inform worker thread of the new request. This
 679                          * call is blocking so we can be assured that
 680                          * the data will be read at some point in time
 681                          * or wait to the end of the world :)
 682                          */
 683                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 684                         if (ret < 0) {
 685                                 PERROR("write conn pipe");
 686                                 connection_put(conn);
 687                                 goto error;
 688                         }
 689                 } while (node != NULL);
 690
 691                 /* Futex wait on queue. Blocking call on futex() */
 692                 health_poll_entry();
 693                 futex_nto1_wait(&viewer_conn_queue.futex);
 694                 health_poll_exit();
 695         }
 696
 697         /* Normal exit, no error */
 698         err = 0;
 699
 700 error:
 701 error_testpoint:
 702         if (err) {
 703                 health_error();
 704                 ERR("Health error occurred in %s", __func__);
 705         }
 706         health_unregister(health_relayd);
 707         DBG("Live viewer dispatch thread dying");
 708         if (lttng_relay_stop_threads()) {
 709                 ERR("Error stopping threads");
 710         }
 711         return NULL;
 712 }
 713
 714 /*
 715  * Establish connection with the viewer and check the versions.
 716  *
 717  * Return 0 on success or else negative value.
 718  */
 719 static
 720 int viewer_connect(struct relay_connection *conn)
 721 {
 722         int ret;
 723         struct lttng_viewer_connect reply, msg;
 724
 725         conn->version_check_done = 1;
 726
 727         health_code_update();
 728
 729         DBG("Viewer is establishing a connection to the relayd.");
 730
 731         ret = recv_request(conn->sock, &msg, sizeof(msg));
 732         if (ret < 0) {
 733                 goto end;
 734         }
 735
 736         health_code_update();
 737
 738         memset(&reply, 0, sizeof(reply));
 739         reply.major = RELAYD_VERSION_COMM_MAJOR;
 740         reply.minor = RELAYD_VERSION_COMM_MINOR;
 741
 742         /* Major versions must be the same */
 743         if (reply.major != be32toh(msg.major)) {
 744                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 745                                 reply.major, be32toh(msg.major));
 746                 ret = -1;
 747                 goto end;
 748         }
 749
 750         conn->major = reply.major;
 751         /* We adapt to the lowest compatible version */
 752         if (reply.minor <= be32toh(msg.minor)) {
 753                 conn->minor = reply.minor;
 754         } else {
 755                 conn->minor = be32toh(msg.minor);
 756         }
 757
 758         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 759                 conn->type = RELAY_VIEWER_COMMAND;
 760         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 761                 conn->type = RELAY_VIEWER_NOTIFICATION;
 762         } else {
 763                 ERR("Unknown connection type : %u", be32toh(msg.type));
 764                 ret = -1;
 765                 goto end;
 766         }
 767
 768         reply.major = htobe32(reply.major);
 769         reply.minor = htobe32(reply.minor);
 770         if (conn->type == RELAY_VIEWER_COMMAND) {
 771                 /*
 772                  * Increment outside of htobe64 macro, because the argument can
 773                  * be used more than once within the macro, and thus the
 774                  * operation may be undefined.
 775                  */
 776                 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
 777                 last_relay_viewer_session_id++;
 778                 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
 779                 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
 780         }
 781
 782         health_code_update();
 783
 784         ret = send_response(conn->sock, &reply, sizeof(reply));
 785         if (ret < 0) {
 786                 goto end;
 787         }
 788
 789         health_code_update();
 790
 791         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 792         ret = 0;
 793
 794 end:
 795         return ret;
 796 }
 797
 798 /*
 799  * Send the viewer the list of current sessions.
 800  * We need to create a copy of the hash table content because otherwise
 801  * we cannot assume the number of entries stays the same between getting
 802  * the number of HT elements and iteration over the HT.
 803  *
 804  * Return 0 on success or else a negative value.
 805  */
 806 static
 807 int viewer_list_sessions(struct relay_connection *conn)
 808 {
 809         int ret;
 810         struct lttng_viewer_list_sessions session_list;
 811         struct lttng_ht_iter iter;
 812         struct relay_session *session;
 813         struct lttng_viewer_session *send_session_buf = NULL;
 814         uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
 815         uint32_t count = 0;
 816
 817         DBG("List sessions received");
 818
 819         send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
 820         if (!send_session_buf) {
 821                 return -1;
 822         }
 823
 824         rcu_read_lock();
 825         cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
 826                         session_n.node) {
 827                 struct lttng_viewer_session *send_session;
 828
 829                 health_code_update();
 830
 831                 if (count >= buf_count) {
 832                         struct lttng_viewer_session *newbuf;
 833                         uint32_t new_buf_count = buf_count << 1;
 834
 835                         newbuf = realloc(send_session_buf,
 836                                 new_buf_count * sizeof(*send_session_buf));
 837                         if (!newbuf) {
 838                                 ret = -1;
 839                                 rcu_read_unlock();
 840                                 goto end_free;
 841                         }
 842                         send_session_buf = newbuf;
 843                         buf_count = new_buf_count;
 844                 }
 845                 send_session = &send_session_buf[count];
 846                 if (lttng_strncpy(send_session->session_name,
 847                                 session->session_name,
 848                                 sizeof(send_session->session_name))) {
 849                         ret = -1;
 850                         rcu_read_unlock();
 851                         goto end_free;
 852                 }
 853                 if (lttng_strncpy(send_session->hostname, session->hostname,
 854                                 sizeof(send_session->hostname))) {
 855                         ret = -1;
 856                         rcu_read_unlock();
 857                         goto end_free;
 858                 }
 859                 send_session->id = htobe64(session->id);
 860                 send_session->live_timer = htobe32(session->live_timer);
 861                 if (session->viewer_attached) {
 862                         send_session->clients = htobe32(1);
 863                 } else {
 864                         send_session->clients = htobe32(0);
 865                 }
 866                 send_session->streams = htobe32(session->stream_count);
 867                 count++;
 868         }
 869         rcu_read_unlock();
 870
 871         session_list.sessions_count = htobe32(count);
 872
 873         health_code_update();
 874
 875         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 876         if (ret < 0) {
 877                 goto end_free;
 878         }
 879
 880         health_code_update();
 881
 882         ret = send_response(conn->sock, send_session_buf,
 883                         count * sizeof(*send_session_buf));
 884         if (ret < 0) {
 885                 goto end_free;
 886         }
 887         health_code_update();
 888
 889         ret = 0;
 890 end_free:
 891         free(send_session_buf);
 892         return ret;
 893 }
 894
 895 /*
 896  * Send the viewer the list of current streams.
 897  */
 898 static
 899 int viewer_get_new_streams(struct relay_connection *conn)
 900 {
 901         int ret, send_streams = 0;
 902         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
 903         struct lttng_viewer_new_streams_request request;
 904         struct lttng_viewer_new_streams_response response;
 905         struct relay_session *session;
 906         uint64_t session_id;
 907         bool closed = false;
 908
 909         assert(conn);
 910
 911         DBG("Get new streams received");
 912
 913         health_code_update();
 914
 915         /* Receive the request from the connected client. */
 916         ret = recv_request(conn->sock, &request, sizeof(request));
 917         if (ret < 0) {
 918                 goto error;
 919         }
 920         session_id = be64toh(request.session_id);
 921
 922         health_code_update();
 923
 924         memset(&response, 0, sizeof(response));
 925
 926         session = session_get_by_id(session_id);
 927         if (!session) {
 928                 DBG("Relay session %" PRIu64 " not found", session_id);
 929                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 930                 goto send_reply;
 931         }
 932
 933         if (!viewer_session_is_attached(conn->viewer_session, session)) {
 934                 send_streams = 0;
 935                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 936                 goto send_reply;
 937         }
 938
 939         send_streams = 1;
 940         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 941
 942         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
 943                         &nb_created, &closed);
 944         if (ret < 0) {
 945                 goto end_put_session;
 946         }
 947         /* Only send back the newly created streams with the unsent ones. */
 948         nb_streams = nb_created + nb_unsent;
 949         response.streams_count = htobe32(nb_streams);
 950
 951         /*
 952          * If the session is closed, HUP when there are no more streams
 953          * with data.
 954          */
 955         if (closed && nb_total == 0) {
 956                 send_streams = 0;
 957                 response.streams_count = 0;
 958                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 959                 goto send_reply;
 960         }
 961
 962 send_reply:
 963         health_code_update();
 964         ret = send_response(conn->sock, &response, sizeof(response));
 965         if (ret < 0) {
 966                 goto end_put_session;
 967         }
 968         health_code_update();
 969
 970         /*
 971          * Unknown or empty session, just return gracefully, the viewer
 972          * knows what is happening.
 973          */
 974         if (!send_streams || !nb_streams) {
 975                 ret = 0;
 976                 goto end_put_session;
 977         }
 978
 979         /*
 980          * Send stream and *DON'T* ignore the sent flag so every viewer
 981          * streams that were not sent from that point will be sent to
 982          * the viewer.
 983          */
 984         ret = send_viewer_streams(conn->sock, session, 0);
 985         if (ret < 0) {
 986                 goto end_put_session;
 987         }
 988
 989 end_put_session:
 990         if (session) {
 991                 session_put(session);
 992         }
 993 error:
 994         return ret;
 995 }
 996
 997 /*
 998  * Send the viewer the list of current sessions.
 999  */
1000 static
1001 int viewer_attach_session(struct relay_connection *conn)
1002 {
1003         int send_streams = 0;
1004         ssize_t ret;
1005         uint32_t nb_streams = 0;
1006         enum lttng_viewer_seek seek_type;
1007         struct lttng_viewer_attach_session_request request;
1008         struct lttng_viewer_attach_session_response response;
1009         struct relay_session *session = NULL;
1010         bool closed = false;
1011
1012         assert(conn);
1013
1014         health_code_update();
1015
1016         /* Receive the request from the connected client. */
1017         ret = recv_request(conn->sock, &request, sizeof(request));
1018         if (ret < 0) {
1019                 goto error;
1020         }
1021
1022         health_code_update();
1023
1024         memset(&response, 0, sizeof(response));
1025
1026         if (!conn->viewer_session) {
1027                 DBG("Client trying to attach before creating a live viewer session");
1028                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1029                 goto send_reply;
1030         }
1031
1032         session = session_get_by_id(be64toh(request.session_id));
1033         if (!session) {
1034                 DBG("Relay session %" PRIu64 " not found",
1035                                 be64toh(request.session_id));
1036                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1037                 goto send_reply;
1038         }
1039         DBG("Attach session ID %" PRIu64 " received",
1040                 be64toh(request.session_id));
1041
1042         if (session->live_timer == 0) {
1043                 DBG("Not live session");
1044                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1045                 goto send_reply;
1046         }
1047
1048         send_streams = 1;
1049         ret = viewer_session_attach(conn->viewer_session, session);
1050         if (ret) {
1051                 DBG("Already a viewer attached");
1052                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1053                 goto send_reply;
1054         }
1055
1056         switch (be32toh(request.seek)) {
1057         case LTTNG_VIEWER_SEEK_BEGINNING:
1058         case LTTNG_VIEWER_SEEK_LAST:
1059                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1060                 seek_type = be32toh(request.seek);
1061                 break;
1062         default:
1063                 ERR("Wrong seek parameter");
1064                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1065                 send_streams = 0;
1066                 goto send_reply;
1067         }
1068
1069         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1070                         NULL, &closed);
1071         if (ret < 0) {
1072                 goto end_put_session;
1073         }
1074         response.streams_count = htobe32(nb_streams);
1075
1076         /*
1077          * If the session is closed when the viewer is attaching, it
1078          * means some of the streams may have been concurrently removed,
1079          * so we don't allow the viewer to attach, even if there are
1080          * streams available.
1081          */
1082         if (closed) {
1083                 send_streams = 0;
1084                 response.streams_count = 0;
1085                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1086                 goto send_reply;
1087         }
1088
1089 send_reply:
1090         health_code_update();
1091         ret = send_response(conn->sock, &response, sizeof(response));
1092         if (ret < 0) {
1093                 goto end_put_session;
1094         }
1095         health_code_update();
1096
1097         /*
1098          * Unknown or empty session, just return gracefully, the viewer
1099          * knows what is happening.
1100          */
1101         if (!send_streams || !nb_streams) {
1102                 ret = 0;
1103                 goto end_put_session;
1104         }
1105
1106         /* Send stream and ignore the sent flag. */
1107         ret = send_viewer_streams(conn->sock, session, 1);
1108         if (ret < 0) {
1109                 goto end_put_session;
1110         }
1111
1112 end_put_session:
1113         if (session) {
1114                 session_put(session);
1115         }
1116 error:
1117         return ret;
1118 }
1119
1120 /*
1121  * Open the index file if needed for the given vstream.
1122  *
1123  * If an index file is successfully opened, the vstream index_fd set with
1124  * it.
1125  *
1126  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1127  *
1128  * Called with rstream lock held.
1129  */
1130 static int try_open_index(struct relay_viewer_stream *vstream,
1131                 struct relay_stream *rstream)
1132 {
1133         int ret = 0;
1134
1135         if (vstream->index_fd) {
1136                 goto end;
1137         }
1138
1139         /*
1140          * First time, we open the index file and at least one index is ready.
1141          */
1142         if (rstream->index_received_seqcount == 0) {
1143                 ret = -ENOENT;
1144                 goto end;
1145         }
1146         ret = index_open(vstream->path_name, vstream->channel_name,
1147                         vstream->stream->tracefile_count,
1148                         vstream->current_tracefile_id);
1149         if (ret >= 0) {
1150                 vstream->index_fd = stream_fd_create(ret);
1151                 if (!vstream->index_fd) {
1152                         if (close(ret)) {
1153                                 PERROR("close");
1154                         }
1155                         ret = -1;
1156                 } else {
1157                         ret = 0;
1158                 }
1159                 goto end;
1160         }
1161
1162 end:
1163         return ret;
1164 }
1165
1166 /*
1167  * Check the status of the index for the given stream. This function
1168  * updates the index structure if needed and can put (close) the vstream
1169  * in the HUP situation.
1170  *
1171  * Return 0 means that we can proceed with the index. A value of 1 means
1172  * that the index has been updated and is ready to be sent to the
1173  * client. A negative value indicates an error that can't be handled.
1174  *
1175  * Called with rstream lock held.
1176  */
1177 static int check_index_status(struct relay_viewer_stream *vstream,
1178                 struct relay_stream *rstream, struct ctf_trace *trace,
1179                 struct lttng_viewer_index *index)
1180 {
1181         int ret;
1182
1183         if (trace->session->connection_closed
1184                         && rstream->index_received_seqcount
1185                                 == vstream->index_sent_seqcount) {
1186                 /* Last index sent and session connection is closed. */
1187                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1188                 goto hup;
1189         } else if (rstream->beacon_ts_end != -1ULL &&
1190                         rstream->index_received_seqcount
1191                                 == vstream->index_sent_seqcount) {
1192                 /*
1193                  * We've received a synchronization beacon and the last index
1194                  * available has been sent, the index for now is inactive.
1195                  *
1196                  * In this case, we have received a beacon which allows us to
1197                  * inform the client of a time interval during which we can
1198                  * guarantee that there are no events to read (and never will
1199                  * be).
1200                  */
1201                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1202                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1203                 index->stream_id = htobe64(rstream->ctf_stream_id);
1204                 goto index_ready;
1205         } else if (rstream->index_received_seqcount
1206                         == vstream->index_sent_seqcount) {
1207                 /*
1208                  * This checks whether received == sent seqcount. In
1209                  * this case, we have not received a beacon. Therefore,
1210                  * we can only ask the client to retry later.
1211                  */
1212                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1213                 goto index_ready;
1214         } else if (!tracefile_array_seq_in_file(rstream->tfa,
1215                         vstream->current_tracefile_id,
1216                         vstream->index_sent_seqcount)) {
1217                 /*
1218                  * The next index we want to send cannot be read either
1219                  * because we need to perform a rotation, or due to
1220                  * the producer having overwritten its trace file.
1221                  */
1222                 DBG("Viewer stream %" PRIu64 " rotation",
1223                                 vstream->stream->stream_handle);
1224                 ret = viewer_stream_rotate(vstream);
1225                 if (ret < 0) {
1226                         goto end;
1227                 } else if (ret == 1) {
1228                         /* EOF across entire stream. */
1229                         index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1230                         goto hup;
1231                 }
1232                 /*
1233                  * If we have been pushed due to overwrite, it
1234                  * necessarily means there is data that can be read in
1235                  * the stream. If we rotated because we reached the end
1236                  * of a tracefile, it means the following tracefile
1237                  * needs to contain at least one index, else we would
1238                  * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1239                  * viewer. The updated index_sent_seqcount needs to
1240                  * point to a readable index entry now.
1241                  *
1242                  * In the case where we "rotate" on a single file, we
1243                  * can end up in a case where the requested index is
1244                  * still unavailable.
1245                  */
1246                 if (rstream->tracefile_count == 1 &&
1247                                 !tracefile_array_seq_in_file(
1248                                         rstream->tfa,
1249                                         vstream->current_tracefile_id,
1250                                         vstream->index_sent_seqcount)) {
1251                         index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1252                         goto index_ready;
1253                 }
1254                 assert(tracefile_array_seq_in_file(rstream->tfa,
1255                                 vstream->current_tracefile_id,
1256                                 vstream->index_sent_seqcount));
1257         }
1258         /* ret == 0 means successful so we continue. */
1259         ret = 0;
1260 end:
1261         return ret;
1262
1263 hup:
1264         viewer_stream_put(vstream);
1265 index_ready:
1266         return 1;
1267 }
1268
1269 /*
1270  * Send the next index for a stream.
1271  *
1272  * Return 0 on success or else a negative value.
1273  */
1274 static
1275 int viewer_get_next_index(struct relay_connection *conn)
1276 {
1277         int ret;
1278         ssize_t read_ret;
1279         struct lttng_viewer_get_next_index request_index;
1280         struct lttng_viewer_index viewer_index;
1281         struct ctf_packet_index packet_index;
1282         struct relay_viewer_stream *vstream = NULL;
1283         struct relay_stream *rstream = NULL;
1284         struct ctf_trace *ctf_trace = NULL;
1285         struct relay_viewer_stream *metadata_viewer_stream = NULL;
1286
1287         assert(conn);
1288
1289         DBG("Viewer get next index");
1290
1291         memset(&viewer_index, 0, sizeof(viewer_index));
1292         health_code_update();
1293
1294         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1295         if (ret < 0) {
1296                 goto end;
1297         }
1298         health_code_update();
1299
1300         vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1301         if (!vstream) {
1302                 DBG("Client requested index of unknown stream id %" PRIu64,
1303                                 be64toh(request_index.stream_id));
1304                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1305                 goto send_reply;
1306         }
1307
1308         /* Use back. ref. Protected by refcounts. */
1309         rstream = vstream->stream;
1310         ctf_trace = rstream->trace;
1311
1312         /* metadata_viewer_stream may be NULL. */
1313         metadata_viewer_stream =
1314                         ctf_trace_get_viewer_metadata_stream(ctf_trace);
1315
1316         pthread_mutex_lock(&rstream->lock);
1317
1318         /*
1319          * The viewer should not ask for index on metadata stream.
1320          */
1321         if (rstream->is_metadata) {
1322                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1323                 goto send_reply;
1324         }
1325
1326         /* Try to open an index if one is needed for that stream. */
1327         ret = try_open_index(vstream, rstream);
1328         if (ret < 0) {
1329                 if (ret == -ENOENT) {
1330                         /*
1331                          * The index is created only when the first data
1332                          * packet arrives, it might not be ready at the
1333                          * beginning of the session
1334                          */
1335                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1336                 } else {
1337                         /* Unhandled error. */
1338                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1339                 }
1340                 goto send_reply;
1341         }
1342
1343         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1344         if (ret < 0) {
1345                 goto error_put;
1346         } else if (ret == 1) {
1347                 /*
1348                  * We have no index to send and check_index_status has populated
1349                  * viewer_index's status.
1350                  */
1351                 goto send_reply;
1352         }
1353         /* At this point, ret is 0 thus we will be able to read the index. */
1354         assert(!ret);
1355
1356         /*
1357          * vstream->stream_fd may be NULL if it has been closed by
1358          * tracefile rotation, or if we are at the beginning of the
1359          * stream. We open the data stream file here to protect against
1360          * overwrite caused by tracefile rotation (in association with
1361          * unlink performed before overwrite).
1362          */
1363         if (!vstream->stream_fd) {
1364                 char fullpath[PATH_MAX];
1365
1366                 if (vstream->stream->tracefile_count > 0) {
1367                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1368                                         vstream->path_name,
1369                                         vstream->channel_name,
1370                                         vstream->current_tracefile_id);
1371                 } else {
1372                         ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1373                                         vstream->path_name,
1374                                         vstream->channel_name);
1375                 }
1376                 if (ret < 0) {
1377                         goto error_put;
1378                 }
1379                 ret = open(fullpath, O_RDONLY);
1380                 if (ret < 0) {
1381                         PERROR("Relay opening trace file");
1382                         goto error_put;
1383                 }
1384                 vstream->stream_fd = stream_fd_create(ret);
1385                 if (!vstream->stream_fd) {
1386                         if (close(ret)) {
1387                                 PERROR("close");
1388                         }
1389                         goto error_put;
1390                 }
1391         }
1392
1393         ret = check_new_streams(conn);
1394         if (ret < 0) {
1395                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1396                 goto send_reply;
1397         } else if (ret == 1) {
1398                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1399         }
1400
1401         read_ret = lttng_read(vstream->index_fd->fd, &packet_index,
1402                         sizeof(packet_index));
1403         if (read_ret < sizeof(packet_index)) {
1404                 ERR("Relay reading index file %d returned %zd",
1405                         vstream->index_fd->fd, read_ret);
1406                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1407                 goto send_reply;
1408         } else {
1409                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1410                 vstream->index_sent_seqcount++;
1411         }
1412
1413         /*
1414          * Indexes are stored in big endian, no need to switch before sending.
1415          */
1416         DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1417                 rstream->stream_handle,
1418                 be64toh(packet_index.offset));
1419         viewer_index.offset = packet_index.offset;
1420         viewer_index.packet_size = packet_index.packet_size;
1421         viewer_index.content_size = packet_index.content_size;
1422         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1423         viewer_index.timestamp_end = packet_index.timestamp_end;
1424         viewer_index.events_discarded = packet_index.events_discarded;
1425         viewer_index.stream_id = packet_index.stream_id;
1426
1427 send_reply:
1428         if (rstream) {
1429                 pthread_mutex_unlock(&rstream->lock);
1430         }
1431
1432         if (metadata_viewer_stream) {
1433                 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1434                 DBG("get next index metadata check: recv %" PRIu64
1435                                 " sent %" PRIu64,
1436                         metadata_viewer_stream->stream->metadata_received,
1437                         metadata_viewer_stream->metadata_sent);
1438                 if (!metadata_viewer_stream->stream->metadata_received ||
1439                                 metadata_viewer_stream->stream->metadata_received >
1440                                         metadata_viewer_stream->metadata_sent) {
1441                         viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1442                 }
1443                 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1444         }
1445
1446         viewer_index.flags = htobe32(viewer_index.flags);
1447         health_code_update();
1448
1449         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1450         if (ret < 0) {
1451                 goto end;
1452         }
1453         health_code_update();
1454
1455         if (vstream) {
1456                 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1457                                 vstream->index_sent_seqcount,
1458                                 vstream->stream->stream_handle);
1459         }
1460 end:
1461         if (metadata_viewer_stream) {
1462                 viewer_stream_put(metadata_viewer_stream);
1463         }
1464         if (vstream) {
1465                 viewer_stream_put(vstream);
1466         }
1467         return ret;
1468
1469 error_put:
1470         pthread_mutex_unlock(&rstream->lock);
1471         if (metadata_viewer_stream) {
1472                 viewer_stream_put(metadata_viewer_stream);
1473         }
1474         viewer_stream_put(vstream);
1475         return ret;
1476 }
1477
1478 /*
1479  * Send the next index for a stream
1480  *
1481  * Return 0 on success or else a negative value.
1482  */
1483 static
1484 int viewer_get_packet(struct relay_connection *conn)
1485 {
1486         int ret, send_data = 0;
1487         char *data = NULL;
1488         uint32_t len = 0;
1489         ssize_t read_len;
1490         struct lttng_viewer_get_packet get_packet_info;
1491         struct lttng_viewer_trace_packet reply;
1492         struct relay_viewer_stream *vstream = NULL;
1493
1494         DBG2("Relay get data packet");
1495
1496         health_code_update();
1497
1498         ret = recv_request(conn->sock, &get_packet_info,
1499                         sizeof(get_packet_info));
1500         if (ret < 0) {
1501                 goto end;
1502         }
1503         health_code_update();
1504
1505         /* From this point on, the error label can be reached. */
1506         memset(&reply, 0, sizeof(reply));
1507
1508         vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1509         if (!vstream) {
1510                 DBG("Client requested packet of unknown stream id %" PRIu64,
1511                                 be64toh(get_packet_info.stream_id));
1512                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1513                 goto send_reply_nolock;
1514         }
1515
1516         pthread_mutex_lock(&vstream->stream->lock);
1517
1518         len = be32toh(get_packet_info.len);
1519         data = zmalloc(len);
1520         if (!data) {
1521                 PERROR("relay data zmalloc");
1522                 goto error;
1523         }
1524
1525         ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1526                         SEEK_SET);
1527         if (ret < 0) {
1528                 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1529                         be64toh(get_packet_info.offset));
1530                 goto error;
1531         }
1532         read_len = lttng_read(vstream->stream_fd->fd, data, len);
1533         if (read_len < len) {
1534                 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1535                                 vstream->stream_fd->fd,
1536                                 be64toh(get_packet_info.offset));
1537                 goto error;
1538         }
1539         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1540         reply.len = htobe32(len);
1541         send_data = 1;
1542         goto send_reply;
1543
1544 error:
1545         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1546
1547 send_reply:
1548         if (vstream) {
1549                 pthread_mutex_unlock(&vstream->stream->lock);
1550         }
1551 send_reply_nolock:
1552         reply.flags = htobe32(reply.flags);
1553
1554         health_code_update();
1555
1556         ret = send_response(conn->sock, &reply, sizeof(reply));
1557         if (ret < 0) {
1558                 goto end_free;
1559         }
1560         health_code_update();
1561
1562         if (send_data) {
1563                 health_code_update();
1564                 ret = send_response(conn->sock, data, len);
1565                 if (ret < 0) {
1566                         goto end_free;
1567                 }
1568                 health_code_update();
1569         }
1570
1571         DBG("Sent %u bytes for stream %" PRIu64, len,
1572                         be64toh(get_packet_info.stream_id));
1573
1574 end_free:
1575         free(data);
1576 end:
1577         if (vstream) {
1578                 viewer_stream_put(vstream);
1579         }
1580         return ret;
1581 }
1582
1583 /*
1584  * Send the session's metadata
1585  *
1586  * Return 0 on success else a negative value.
1587  */
1588 static
1589 int viewer_get_metadata(struct relay_connection *conn)
1590 {
1591         int ret = 0;
1592         ssize_t read_len;
1593         uint64_t len = 0;
1594         char *data = NULL;
1595         struct lttng_viewer_get_metadata request;
1596         struct lttng_viewer_metadata_packet reply;
1597         struct relay_viewer_stream *vstream = NULL;
1598
1599         assert(conn);
1600
1601         DBG("Relay get metadata");
1602
1603         health_code_update();
1604
1605         ret = recv_request(conn->sock, &request, sizeof(request));
1606         if (ret < 0) {
1607                 goto end;
1608         }
1609         health_code_update();
1610
1611         memset(&reply, 0, sizeof(reply));
1612
1613         vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1614         if (!vstream) {
1615                 /*
1616                  * The metadata stream can be closed by a CLOSE command
1617                  * just before we attach. It can also be closed by
1618                  * per-pid tracing during tracing. Therefore, it is
1619                  * possible that we cannot find this viewer stream.
1620                  * Reply back to the client with an error if we cannot
1621                  * find it.
1622                  */
1623                 DBG("Client requested metadata of unknown stream id %" PRIu64,
1624                                 be64toh(request.stream_id));
1625                 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1626                 goto send_reply;
1627         }
1628         pthread_mutex_lock(&vstream->stream->lock);
1629         if (!vstream->stream->is_metadata) {
1630                 ERR("Invalid metadata stream");
1631                 goto error;
1632         }
1633
1634         assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1635
1636         len = vstream->stream->metadata_received - vstream->metadata_sent;
1637         if (len == 0) {
1638                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1639                 goto send_reply;
1640         }
1641
1642         /* first time, we open the metadata file */
1643         if (!vstream->stream_fd) {
1644                 char fullpath[PATH_MAX];
1645
1646                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1647                                 vstream->channel_name);
1648                 if (ret < 0) {
1649                         goto error;
1650                 }
1651                 ret = open(fullpath, O_RDONLY);
1652                 if (ret < 0) {
1653                         PERROR("Relay opening metadata file");
1654                         goto error;
1655                 }
1656                 vstream->stream_fd = stream_fd_create(ret);
1657                 if (!vstream->stream_fd) {
1658                         if (close(ret)) {
1659                                 PERROR("close");
1660                         }
1661                         goto error;
1662                 }
1663         }
1664
1665         reply.len = htobe64(len);
1666         data = zmalloc(len);
1667         if (!data) {
1668                 PERROR("viewer metadata zmalloc");
1669                 goto error;
1670         }
1671
1672         read_len = lttng_read(vstream->stream_fd->fd, data, len);
1673         if (read_len < len) {
1674                 PERROR("Relay reading metadata file");
1675                 goto error;
1676         }
1677         vstream->metadata_sent += read_len;
1678         if (vstream->metadata_sent == vstream->stream->metadata_received
1679                         && vstream->stream->closed) {
1680                 /* Release ownership for the viewer metadata stream. */
1681                 viewer_stream_put(vstream);
1682         }
1683
1684         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1685
1686         goto send_reply;
1687
1688 error:
1689         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1690
1691 send_reply:
1692         health_code_update();
1693         if (vstream) {
1694                 pthread_mutex_unlock(&vstream->stream->lock);
1695         }
1696         ret = send_response(conn->sock, &reply, sizeof(reply));
1697         if (ret < 0) {
1698                 goto end_free;
1699         }
1700         health_code_update();
1701
1702         if (len > 0) {
1703                 ret = send_response(conn->sock, data, len);
1704                 if (ret < 0) {
1705                         goto end_free;
1706                 }
1707         }
1708
1709         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1710                         be64toh(request.stream_id));
1711
1712         DBG("Metadata sent");
1713
1714 end_free:
1715         free(data);
1716 end:
1717         if (vstream) {
1718                 viewer_stream_put(vstream);
1719         }
1720         return ret;
1721 }
1722
1723 /*
1724  * Create a viewer session.
1725  *
1726  * Return 0 on success or else a negative value.
1727  */
1728 static
1729 int viewer_create_session(struct relay_connection *conn)
1730 {
1731         int ret;
1732         struct lttng_viewer_create_session_response resp;
1733
1734         DBG("Viewer create session received");
1735
1736         memset(&resp, 0, sizeof(resp));
1737         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1738         conn->viewer_session = viewer_session_create();
1739         if (!conn->viewer_session) {
1740                 ERR("Allocation viewer session");
1741                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1742                 goto send_reply;
1743         }
1744
1745 send_reply:
1746         health_code_update();
1747         ret = send_response(conn->sock, &resp, sizeof(resp));
1748         if (ret < 0) {
1749                 goto end;
1750         }
1751         health_code_update();
1752         ret = 0;
1753
1754 end:
1755         return ret;
1756 }
1757
1758
1759 /*
1760  * live_relay_unknown_command: send -1 if received unknown command
1761  */
1762 static
1763 void live_relay_unknown_command(struct relay_connection *conn)
1764 {
1765         struct lttcomm_relayd_generic_reply reply;
1766
1767         memset(&reply, 0, sizeof(reply));
1768         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1769         (void) send_response(conn->sock, &reply, sizeof(reply));
1770 }
1771
1772 /*
1773  * Process the commands received on the control socket
1774  */
1775 static
1776 int process_control(struct lttng_viewer_cmd *recv_hdr,
1777                 struct relay_connection *conn)
1778 {
1779         int ret = 0;
1780         uint32_t msg_value;
1781
1782         msg_value = be32toh(recv_hdr->cmd);
1783
1784         /*
1785          * Make sure we've done the version check before any command other then a
1786          * new client connection.
1787          */
1788         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1789                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1790                 ret = -1;
1791                 goto end;
1792         }
1793
1794         switch (msg_value) {
1795         case LTTNG_VIEWER_CONNECT:
1796                 ret = viewer_connect(conn);
1797                 break;
1798         case LTTNG_VIEWER_LIST_SESSIONS:
1799                 ret = viewer_list_sessions(conn);
1800                 break;
1801         case LTTNG_VIEWER_ATTACH_SESSION:
1802                 ret = viewer_attach_session(conn);
1803                 break;
1804         case LTTNG_VIEWER_GET_NEXT_INDEX:
1805                 ret = viewer_get_next_index(conn);
1806                 break;
1807         case LTTNG_VIEWER_GET_PACKET:
1808                 ret = viewer_get_packet(conn);
1809                 break;
1810         case LTTNG_VIEWER_GET_METADATA:
1811                 ret = viewer_get_metadata(conn);
1812                 break;
1813         case LTTNG_VIEWER_GET_NEW_STREAMS:
1814                 ret = viewer_get_new_streams(conn);
1815                 break;
1816         case LTTNG_VIEWER_CREATE_SESSION:
1817                 ret = viewer_create_session(conn);
1818                 break;
1819         default:
1820                 ERR("Received unknown viewer command (%u)",
1821                                 be32toh(recv_hdr->cmd));
1822                 live_relay_unknown_command(conn);
1823                 ret = -1;
1824                 goto end;
1825         }
1826
1827 end:
1828         return ret;
1829 }
1830
1831 static
1832 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1833 {
1834         int ret;
1835
1836         (void) lttng_poll_del(events, pollfd);
1837
1838         ret = close(pollfd);
1839         if (ret < 0) {
1840                 ERR("Closing pollfd %d", pollfd);
1841         }
1842 }
1843
1844 /*
1845  * This thread does the actual work
1846  */
1847 static
1848 void *thread_worker(void *data)
1849 {
1850         int ret, err = -1;
1851         uint32_t nb_fd;
1852         struct lttng_poll_event events;
1853         struct lttng_ht *viewer_connections_ht;
1854         struct lttng_ht_iter iter;
1855         struct lttng_viewer_cmd recv_hdr;
1856         struct relay_connection *destroy_conn;
1857
1858         DBG("[thread] Live viewer relay worker started");
1859
1860         rcu_register_thread();
1861
1862         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1863
1864         if (testpoint(relayd_thread_live_worker)) {
1865                 goto error_testpoint;
1866         }
1867
1868         /* table of connections indexed on socket */
1869         viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1870         if (!viewer_connections_ht) {
1871                 goto viewer_connections_ht_error;
1872         }
1873
1874         ret = create_thread_poll_set(&events, 2);
1875         if (ret < 0) {
1876                 goto error_poll_create;
1877         }
1878
1879         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1880         if (ret < 0) {
1881                 goto error;
1882         }
1883
1884 restart:
1885         while (1) {
1886                 int i;
1887
1888                 health_code_update();
1889
1890                 /* Infinite blocking call, waiting for transmission */
1891                 DBG3("Relayd live viewer worker thread polling...");
1892                 health_poll_entry();
1893                 ret = lttng_poll_wait(&events, -1);
1894                 health_poll_exit();
1895                 if (ret < 0) {
1896                         /*
1897                          * Restart interrupted system call.
1898                          */
1899                         if (errno == EINTR) {
1900                                 goto restart;
1901                         }
1902                         goto error;
1903                 }
1904
1905                 nb_fd = ret;
1906
1907                 /*
1908                  * Process control. The control connection is prioritised so we don't
1909                  * starve it with high throughput tracing data on the data
1910                  * connection.
1911                  */
1912                 for (i = 0; i < nb_fd; i++) {
1913                         /* Fetch once the poll data */
1914                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1915                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1916
1917                         health_code_update();
1918
1919                         if (!revents) {
1920                                 /* No activity for this FD (poll implementation). */
1921                                 continue;
1922                         }
1923
1924                         /* Thread quit pipe has been closed. Killing thread. */
1925                         ret = check_thread_quit_pipe(pollfd, revents);
1926                         if (ret) {
1927                                 err = 0;
1928                                 goto exit;
1929                         }
1930
1931                         /* Inspect the relay conn pipe for new connection. */
1932                         if (pollfd == live_conn_pipe[0]) {
1933                                 if (revents & LPOLLIN) {
1934                                         struct relay_connection *conn;
1935
1936                                         ret = lttng_read(live_conn_pipe[0],
1937                                                         &conn, sizeof(conn));
1938                                         if (ret < 0) {
1939                                                 goto error;
1940                                         }
1941                                         lttng_poll_add(&events, conn->sock->fd,
1942                                                         LPOLLIN | LPOLLRDHUP);
1943                                         connection_ht_add(viewer_connections_ht, conn);
1944                                         DBG("Connection socket %d added to poll", conn->sock->fd);
1945                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1946                                         ERR("Relay live pipe error");
1947                                         goto error;
1948                                 } else {
1949                                         ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1950                                         goto error;
1951                                 }
1952                         } else {
1953                                 /* Connection activity. */
1954                                 struct relay_connection *conn;
1955
1956                                 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
1957                                 if (!conn) {
1958                                         continue;
1959                                 }
1960
1961                                 if (revents & LPOLLIN) {
1962                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1963                                                         sizeof(recv_hdr), 0);
1964                                         if (ret <= 0) {
1965                                                 /* Connection closed. */
1966                                                 cleanup_connection_pollfd(&events, pollfd);
1967                                                 /* Put "create" ownership reference. */
1968                                                 connection_put(conn);
1969                                                 DBG("Viewer control conn closed with %d", pollfd);
1970                                         } else {
1971                                                 ret = process_control(&recv_hdr, conn);
1972                                                 if (ret < 0) {
1973                                                         /* Clear the session on error. */
1974                                                         cleanup_connection_pollfd(&events, pollfd);
1975                                                         /* Put "create" ownership reference. */
1976                                                         connection_put(conn);
1977                                                         DBG("Viewer connection closed with %d", pollfd);
1978                                                 }
1979                                         }
1980                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1981                                         cleanup_connection_pollfd(&events, pollfd);
1982                                         /* Put "create" ownership reference. */
1983                                         connection_put(conn);
1984                                 } else {
1985                                         ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1986                                         connection_put(conn);
1987                                         goto error;
1988                                 }
1989                                 /* Put local "get_by_sock" reference. */
1990                                 connection_put(conn);
1991                         }
1992                 }
1993         }
1994
1995 exit:
1996 error:
1997         lttng_poll_clean(&events);
1998
1999         /* Cleanup reamaining connection object. */
2000         rcu_read_lock();
2001         cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2002                         destroy_conn,
2003                         sock_n.node) {
2004                 health_code_update();
2005                 connection_put(destroy_conn);
2006         }
2007         rcu_read_unlock();
2008 error_poll_create:
2009         lttng_ht_destroy(viewer_connections_ht);
2010 viewer_connections_ht_error:
2011         /* Close relay conn pipes */
2012         utils_close_pipe(live_conn_pipe);
2013         if (err) {
2014                 DBG("Viewer worker thread exited with error");
2015         }
2016         DBG("Viewer worker thread cleanup complete");
2017 error_testpoint:
2018         if (err) {
2019                 health_error();
2020                 ERR("Health error occurred in %s", __func__);
2021         }
2022         health_unregister(health_relayd);
2023         if (lttng_relay_stop_threads()) {
2024                 ERR("Error stopping threads");
2025         }
2026         rcu_unregister_thread();
2027         return NULL;
2028 }
2029
2030 /*
2031  * Create the relay command pipe to wake thread_manage_apps.
2032  * Closed in cleanup().
2033  */
2034 static int create_conn_pipe(void)
2035 {
2036         return utils_create_pipe_cloexec(live_conn_pipe);
2037 }
2038
2039 int relayd_live_join(void)
2040 {
2041         int ret, retval = 0;
2042         void *status;
2043
2044         ret = pthread_join(live_listener_thread, &status);
2045         if (ret) {
2046                 errno = ret;
2047                 PERROR("pthread_join live listener");
2048                 retval = -1;
2049         }
2050
2051         ret = pthread_join(live_worker_thread, &status);
2052         if (ret) {
2053                 errno = ret;
2054                 PERROR("pthread_join live worker");
2055                 retval = -1;
2056         }
2057
2058         ret = pthread_join(live_dispatcher_thread, &status);
2059         if (ret) {
2060                 errno = ret;
2061                 PERROR("pthread_join live dispatcher");
2062                 retval = -1;
2063         }
2064
2065         cleanup_relayd_live();
2066
2067         return retval;
2068 }
2069
2070 /*
2071  * main
2072  */
2073 int relayd_live_create(struct lttng_uri *uri)
2074 {
2075         int ret = 0, retval = 0;
2076         void *status;
2077         int is_root;
2078
2079         if (!uri) {
2080                 retval = -1;
2081                 goto exit_init_data;
2082         }
2083         live_uri = uri;
2084
2085         /* Check if daemon is UID = 0 */
2086         is_root = !getuid();
2087
2088         if (!is_root) {
2089                 if (live_uri->port < 1024) {
2090                         ERR("Need to be root to use ports < 1024");
2091                         retval = -1;
2092                         goto exit_init_data;
2093                 }
2094         }
2095
2096         /* Setup the thread apps communication pipe. */
2097         if (create_conn_pipe()) {
2098                 retval = -1;
2099                 goto exit_init_data;
2100         }
2101
2102         /* Init relay command queue. */
2103         cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2104
2105         /* Set up max poll set size */
2106         lttng_poll_set_max_size();
2107
2108         /* Setup the dispatcher thread */
2109         ret = pthread_create(&live_dispatcher_thread, NULL,
2110                         thread_dispatcher, (void *) NULL);
2111         if (ret) {
2112                 errno = ret;
2113                 PERROR("pthread_create viewer dispatcher");
2114                 retval = -1;
2115                 goto exit_dispatcher_thread;
2116         }
2117
2118         /* Setup the worker thread */
2119         ret = pthread_create(&live_worker_thread, NULL,
2120                         thread_worker, NULL);
2121         if (ret) {
2122                 errno = ret;
2123                 PERROR("pthread_create viewer worker");
2124                 retval = -1;
2125                 goto exit_worker_thread;
2126         }
2127
2128         /* Setup the listener thread */
2129         ret = pthread_create(&live_listener_thread, NULL,
2130                         thread_listener, (void *) NULL);
2131         if (ret) {
2132                 errno = ret;
2133                 PERROR("pthread_create viewer listener");
2134                 retval = -1;
2135                 goto exit_listener_thread;
2136         }
2137
2138         /*
2139          * All OK, started all threads.
2140          */
2141         return retval;
2142
2143         /*
2144          * Join on the live_listener_thread should anything be added after
2145          * the live_listener thread's creation.
2146          */
2147
2148 exit_listener_thread:
2149
2150         ret = pthread_join(live_worker_thread, &status);
2151         if (ret) {
2152                 errno = ret;
2153                 PERROR("pthread_join live worker");
2154                 retval = -1;
2155         }
2156 exit_worker_thread:
2157
2158         ret = pthread_join(live_dispatcher_thread, &status);
2159         if (ret) {
2160                 errno = ret;
2161                 PERROR("pthread_join live dispatcher");
2162                 retval = -1;
2163         }
2164 exit_dispatcher_thread:
2165
2166 exit_init_data:
2167         cleanup_relayd_live();
2168
2169         return retval;
2170 }