src/bin/lttng-relayd/live.c

   1 /*
   2  * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
   3  *                      David Goulet <dgoulet@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _GNU_SOURCE
  20 #define _LGPL_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 <unistd.h>
  40 #include <fcntl.h>
  41 #include <config.h>
  42
  43 #include <lttng/lttng.h>
  44 #include <common/common.h>
  45 #include <common/compat/poll.h>
  46 #include <common/compat/socket.h>
  47 #include <common/compat/endian.h>
  48 #include <common/defaults.h>
  49 #include <common/futex.h>
  50 #include <common/index/index.h>
  51 #include <common/sessiond-comm/sessiond-comm.h>
  52 #include <common/sessiond-comm/inet.h>
  53 #include <common/sessiond-comm/relayd.h>
  54 #include <common/uri.h>
  55 #include <common/utils.h>
  56
  57 #include "cmd.h"
  58 #include "live.h"
  59 #include "lttng-relayd.h"
  60 #include "utils.h"
  61 #include "health-relayd.h"
  62 #include "testpoint.h"
  63 #include "viewer-stream.h"
  64 #include "stream.h"
  65 #include "session.h"
  66 #include "ctf-trace.h"
  67 #include "connection.h"
  68
  69 static struct lttng_uri *live_uri;
  70
  71 /*
  72  * This pipe is used to inform the worker thread that a command is queued and
  73  * ready to be processed.
  74  */
  75 static int live_conn_pipe[2] = { -1, -1 };
  76
  77 /* Shared between threads */
  78 static int live_dispatch_thread_exit;
  79
  80 static pthread_t live_listener_thread;
  81 static pthread_t live_dispatcher_thread;
  82 static pthread_t live_worker_thread;
  83
  84 /*
  85  * Relay command queue.
  86  *
  87  * The live_thread_listener and live_thread_dispatcher communicate with this
  88  * queue.
  89  */
  90 static struct relay_conn_queue viewer_conn_queue;
  91
  92 static uint64_t last_relay_viewer_session_id;
  93
  94 /*
  95  * Cleanup the daemon
  96  */
  97 static
  98 void cleanup_relayd_live(void)
  99 {
 100         DBG("Cleaning up");
 101
 102         free(live_uri);
 103 }
 104
 105 /*
 106  * Receive a request buffer using a given socket, destination allocated buffer
 107  * of length size.
 108  *
 109  * Return the size of the received message or else a negative value on error
 110  * with errno being set by recvmsg() syscall.
 111  */
 112 static
 113 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 114 {
 115         ssize_t ret;
 116
 117         assert(sock);
 118         assert(buf);
 119
 120         ret = sock->ops->recvmsg(sock, buf, size, 0);
 121         if (ret < 0 || ret != size) {
 122                 if (ret == 0) {
 123                         /* Orderly shutdown. Not necessary to print an error. */
 124                         DBG("Socket %d did an orderly shutdown", sock->fd);
 125                 } else {
 126                         ERR("Relay failed to receive request.");
 127                 }
 128                 ret = -1;
 129         }
 130
 131         return ret;
 132 }
 133
 134 /*
 135  * Send a response buffer using a given socket, source allocated buffer of
 136  * length size.
 137  *
 138  * Return the size of the sent message or else a negative value on error with
 139  * errno being set by sendmsg() syscall.
 140  */
 141 static
 142 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 143 {
 144         ssize_t ret;
 145
 146         assert(sock);
 147         assert(buf);
 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         cds_list_for_each_entry(session,
 175                         &conn->viewer_session->sessions_head,
 176                         viewer_session_list) {
 177                 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
 178                 ret = current_val;
 179                 if (ret == 1) {
 180                         goto end;
 181                 }
 182         }
 183
 184 end:
 185         return ret;
 186 }
 187
 188 /*
 189  * Send viewer streams to the given socket. The ignore_sent_flag indicates if
 190  * this function should ignore the sent flag or not.
 191  *
 192  * Return 0 on success or else a negative value.
 193  */
 194 static
 195 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
 196                 struct relay_session *session, unsigned int ignore_sent_flag)
 197 {
 198         ssize_t ret;
 199         struct lttng_viewer_stream send_stream;
 200         struct lttng_ht_iter iter;
 201         struct relay_viewer_stream *vstream;
 202
 203         assert(session);
 204
 205         rcu_read_lock();
 206
 207         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
 208                         stream_n.node) {
 209                 struct ctf_trace *ctf_trace;
 210
 211                 health_code_update();
 212
 213                 /* Ignore if not the same session. */
 214                 if (vstream->session_id != session->id ||
 215                                 (!ignore_sent_flag && vstream->sent_flag)) {
 216                         continue;
 217                 }
 218
 219                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 220                                 vstream->path_name);
 221                 assert(ctf_trace);
 222
 223                 send_stream.id = htobe64(vstream->stream_handle);
 224                 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
 225                 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
 226                 strncpy(send_stream.path_name, vstream->path_name,
 227                                 sizeof(send_stream.path_name));
 228                 strncpy(send_stream.channel_name, vstream->channel_name,
 229                                 sizeof(send_stream.channel_name));
 230
 231                 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
 232                 ret = send_response(sock, &send_stream, sizeof(send_stream));
 233                 if (ret < 0) {
 234                         goto end_unlock;
 235                 }
 236                 vstream->sent_flag = 1;
 237         }
 238
 239         ret = 0;
 240
 241 end_unlock:
 242         rcu_read_unlock();
 243         return ret;
 244 }
 245
 246 /*
 247  * Create every viewer stream possible for the given session with the seek
 248  * type. Three counters *can* be return which are in order the total amount of
 249  * viewer stream of the session, the number of unsent stream and the number of
 250  * stream created. Those counters can be NULL and thus will be ignored.
 251  *
 252  * Return 0 on success or else a negative value.
 253  */
 254 static
 255 int make_viewer_streams(struct relay_session *session,
 256                 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
 257                 uint32_t *nb_created)
 258 {
 259         int ret;
 260         struct lttng_ht_iter iter;
 261         struct ctf_trace *ctf_trace;
 262
 263         assert(session);
 264
 265         /*
 266          * This is to make sure we create viewer streams for a full received
 267          * channel. For instance, if we have 8 streams for a channel that are
 268          * concurrently being flagged ready, we can end up creating just a subset
 269          * of the 8 streams (the ones that are flagged). This lock avoids this
 270          * limbo state.
 271          */
 272         pthread_mutex_lock(&session->viewer_ready_lock);
 273
 274         /*
 275          * Create viewer streams for relay streams that are ready to be used for a
 276          * the given session id only.
 277          */
 278         rcu_read_lock();
 279         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 280                         node.node) {
 281                 struct relay_stream *stream;
 282
 283                 health_code_update();
 284
 285                 if (ctf_trace->invalid_flag) {
 286                         continue;
 287                 }
 288
 289                 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
 290                         struct relay_viewer_stream *vstream;
 291
 292                         if (!stream->viewer_ready) {
 293                                 continue;
 294                         }
 295
 296                         vstream = viewer_stream_find_by_id(stream->stream_handle);
 297                         if (!vstream) {
 298                                 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
 299                                 if (!vstream) {
 300                                         ret = -1;
 301                                         goto error_unlock;
 302                                 }
 303                                 /* Acquire reference to ctf_trace. */
 304                                 ctf_trace_get_ref(ctf_trace);
 305
 306                                 if (nb_created) {
 307                                         /* Update number of created stream counter. */
 308                                         (*nb_created)++;
 309                                 }
 310                         } else if (!vstream->sent_flag && nb_unsent) {
 311                                 /* Update number of unsent stream counter. */
 312                                 (*nb_unsent)++;
 313                         }
 314                         /* Update number of total stream counter. */
 315                         if (nb_total) {
 316                                 (*nb_total)++;
 317                         }
 318                 }
 319         }
 320
 321         ret = 0;
 322
 323 error_unlock:
 324         rcu_read_unlock();
 325         pthread_mutex_unlock(&session->viewer_ready_lock);
 326         return ret;
 327 }
 328
 329 int relayd_live_stop(void)
 330 {
 331         /* Stop dispatch thread */
 332         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 333         futex_nto1_wake(&viewer_conn_queue.futex);
 334         return 0;
 335 }
 336
 337 /*
 338  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 339  */
 340 static
 341 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 342 {
 343         int ret;
 344
 345         if (events == NULL || size == 0) {
 346                 ret = -1;
 347                 goto error;
 348         }
 349
 350         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 351         if (ret < 0) {
 352                 goto error;
 353         }
 354
 355         /* Add quit pipe */
 356         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 357         if (ret < 0) {
 358                 goto error;
 359         }
 360
 361         return 0;
 362
 363 error:
 364         return ret;
 365 }
 366
 367 /*
 368  * Check if the thread quit pipe was triggered.
 369  *
 370  * Return 1 if it was triggered else 0;
 371  */
 372 static
 373 int check_thread_quit_pipe(int fd, uint32_t events)
 374 {
 375         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 376                 return 1;
 377         }
 378
 379         return 0;
 380 }
 381
 382 /*
 383  * Create and init socket from uri.
 384  */
 385 static
 386 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 387 {
 388         int ret;
 389         struct lttcomm_sock *sock = NULL;
 390
 391         sock = lttcomm_alloc_sock_from_uri(uri);
 392         if (sock == NULL) {
 393                 ERR("Allocating socket");
 394                 goto error;
 395         }
 396
 397         ret = lttcomm_create_sock(sock);
 398         if (ret < 0) {
 399                 goto error;
 400         }
 401         DBG("Listening on sock %d for live", sock->fd);
 402
 403         ret = sock->ops->bind(sock);
 404         if (ret < 0) {
 405                 goto error;
 406         }
 407
 408         ret = sock->ops->listen(sock, -1);
 409         if (ret < 0) {
 410                 goto error;
 411
 412         }
 413
 414         return sock;
 415
 416 error:
 417         if (sock) {
 418                 lttcomm_destroy_sock(sock);
 419         }
 420         return NULL;
 421 }
 422
 423 /*
 424  * This thread manages the listening for new connections on the network
 425  */
 426 static
 427 void *thread_listener(void *data)
 428 {
 429         int i, ret, pollfd, err = -1;
 430         uint32_t revents, nb_fd;
 431         struct lttng_poll_event events;
 432         struct lttcomm_sock *live_control_sock;
 433
 434         DBG("[thread] Relay live listener started");
 435
 436         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 437
 438         health_code_update();
 439
 440         live_control_sock = init_socket(live_uri);
 441         if (!live_control_sock) {
 442                 goto error_sock_control;
 443         }
 444
 445         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 446         ret = create_thread_poll_set(&events, 2);
 447         if (ret < 0) {
 448                 goto error_create_poll;
 449         }
 450
 451         /* Add the control socket */
 452         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 453         if (ret < 0) {
 454                 goto error_poll_add;
 455         }
 456
 457         lttng_relay_notify_ready();
 458
 459         if (testpoint(relayd_thread_live_listener)) {
 460                 goto error_testpoint;
 461         }
 462
 463         while (1) {
 464                 health_code_update();
 465
 466                 DBG("Listener accepting live viewers connections");
 467
 468 restart:
 469                 health_poll_entry();
 470                 ret = lttng_poll_wait(&events, -1);
 471                 health_poll_exit();
 472                 if (ret < 0) {
 473                         /*
 474                          * Restart interrupted system call.
 475                          */
 476                         if (errno == EINTR) {
 477                                 goto restart;
 478                         }
 479                         goto error;
 480                 }
 481                 nb_fd = ret;
 482
 483                 DBG("Relay new viewer connection received");
 484                 for (i = 0; i < nb_fd; i++) {
 485                         health_code_update();
 486
 487                         /* Fetch once the poll data */
 488                         revents = LTTNG_POLL_GETEV(&events, i);
 489                         pollfd = LTTNG_POLL_GETFD(&events, i);
 490
 491                         /* Thread quit pipe has been closed. Killing thread. */
 492                         ret = check_thread_quit_pipe(pollfd, revents);
 493                         if (ret) {
 494                                 err = 0;
 495                                 goto exit;
 496                         }
 497
 498                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 499                                 ERR("socket poll error");
 500                                 goto error;
 501                         } else if (revents & LPOLLIN) {
 502                                 /*
 503                                  * Get allocated in this thread, enqueued to a global queue,
 504                                  * dequeued and freed in the worker thread.
 505                                  */
 506                                 int val = 1;
 507                                 struct relay_connection *new_conn;
 508                                 struct lttcomm_sock *newsock;
 509
 510                                 new_conn = connection_create();
 511                                 if (!new_conn) {
 512                                         goto error;
 513                                 }
 514
 515                                 newsock = live_control_sock->ops->accept(live_control_sock);
 516                                 if (!newsock) {
 517                                         PERROR("accepting control sock");
 518                                         connection_free(new_conn);
 519                                         goto error;
 520                                 }
 521                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 522
 523                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 524                                                 sizeof(val));
 525                                 if (ret < 0) {
 526                                         PERROR("setsockopt inet");
 527                                         lttcomm_destroy_sock(newsock);
 528                                         connection_free(new_conn);
 529                                         goto error;
 530                                 }
 531                                 new_conn->sock = newsock;
 532
 533                                 /* Enqueue request for the dispatcher thread. */
 534                                 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
 535                                                  &new_conn->qnode);
 536
 537                                 /*
 538                                  * Wake the dispatch queue futex. Implicit memory barrier with
 539                                  * the exchange in cds_wfcq_enqueue.
 540                                  */
 541                                 futex_nto1_wake(&viewer_conn_queue.futex);
 542                         }
 543                 }
 544         }
 545
 546 exit:
 547 error:
 548 error_poll_add:
 549 error_testpoint:
 550         lttng_poll_clean(&events);
 551 error_create_poll:
 552         if (live_control_sock->fd >= 0) {
 553                 ret = live_control_sock->ops->close(live_control_sock);
 554                 if (ret) {
 555                         PERROR("close");
 556                 }
 557         }
 558         lttcomm_destroy_sock(live_control_sock);
 559 error_sock_control:
 560         if (err) {
 561                 health_error();
 562                 DBG("Live viewer listener thread exited with error");
 563         }
 564         health_unregister(health_relayd);
 565         DBG("Live viewer listener thread cleanup complete");
 566         if (lttng_relay_stop_threads()) {
 567                 ERR("Error stopping threads");
 568         }
 569         return NULL;
 570 }
 571
 572 /*
 573  * This thread manages the dispatching of the requests to worker threads
 574  */
 575 static
 576 void *thread_dispatcher(void *data)
 577 {
 578         int err = -1;
 579         ssize_t ret;
 580         struct cds_wfcq_node *node;
 581         struct relay_connection *conn = NULL;
 582
 583         DBG("[thread] Live viewer relay dispatcher started");
 584
 585         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 586
 587         if (testpoint(relayd_thread_live_dispatcher)) {
 588                 goto error_testpoint;
 589         }
 590
 591         health_code_update();
 592
 593         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 594                 health_code_update();
 595
 596                 /* Atomically prepare the queue futex */
 597                 futex_nto1_prepare(&viewer_conn_queue.futex);
 598
 599                 do {
 600                         health_code_update();
 601
 602                         /* Dequeue commands */
 603                         node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
 604                                                          &viewer_conn_queue.tail);
 605                         if (node == NULL) {
 606                                 DBG("Woken up but nothing in the live-viewer "
 607                                                 "relay command queue");
 608                                 /* Continue thread execution */
 609                                 break;
 610                         }
 611                         conn = caa_container_of(node, struct relay_connection, qnode);
 612                         DBG("Dispatching viewer request waiting on sock %d",
 613                                         conn->sock->fd);
 614
 615                         /*
 616                          * Inform worker thread of the new request. This call is blocking
 617                          * so we can be assured that the data will be read at some point in
 618                          * time or wait to the end of the world :)
 619                          */
 620                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 621                         if (ret < 0) {
 622                                 PERROR("write conn pipe");
 623                                 connection_destroy(conn);
 624                                 goto error;
 625                         }
 626                 } while (node != NULL);
 627
 628                 /* Futex wait on queue. Blocking call on futex() */
 629                 health_poll_entry();
 630                 futex_nto1_wait(&viewer_conn_queue.futex);
 631                 health_poll_exit();
 632         }
 633
 634         /* Normal exit, no error */
 635         err = 0;
 636
 637 error:
 638 error_testpoint:
 639         if (err) {
 640                 health_error();
 641                 ERR("Health error occurred in %s", __func__);
 642         }
 643         health_unregister(health_relayd);
 644         DBG("Live viewer dispatch thread dying");
 645         if (lttng_relay_stop_threads()) {
 646                 ERR("Error stopping threads");
 647         }
 648         return NULL;
 649 }
 650
 651 /*
 652  * Establish connection with the viewer and check the versions.
 653  *
 654  * Return 0 on success or else negative value.
 655  */
 656 static
 657 int viewer_connect(struct relay_connection *conn)
 658 {
 659         int ret;
 660         struct lttng_viewer_connect reply, msg;
 661
 662         assert(conn);
 663
 664         conn->version_check_done = 1;
 665
 666         health_code_update();
 667
 668         DBG("Viewer is establishing a connection to the relayd.");
 669
 670         ret = recv_request(conn->sock, &msg, sizeof(msg));
 671         if (ret < 0) {
 672                 goto end;
 673         }
 674
 675         health_code_update();
 676
 677         memset(&reply, 0, sizeof(reply));
 678         reply.major = RELAYD_VERSION_COMM_MAJOR;
 679         reply.minor = RELAYD_VERSION_COMM_MINOR;
 680
 681         /* Major versions must be the same */
 682         if (reply.major != be32toh(msg.major)) {
 683                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 684                                 reply.major, be32toh(msg.major));
 685                 ret = -1;
 686                 goto end;
 687         }
 688
 689         conn->major = reply.major;
 690         /* We adapt to the lowest compatible version */
 691         if (reply.minor <= be32toh(msg.minor)) {
 692                 conn->minor = reply.minor;
 693         } else {
 694                 conn->minor = be32toh(msg.minor);
 695         }
 696
 697         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 698                 conn->type = RELAY_VIEWER_COMMAND;
 699         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 700                 conn->type = RELAY_VIEWER_NOTIFICATION;
 701         } else {
 702                 ERR("Unknown connection type : %u", be32toh(msg.type));
 703                 ret = -1;
 704                 goto end;
 705         }
 706
 707         reply.major = htobe32(reply.major);
 708         reply.minor = htobe32(reply.minor);
 709         if (conn->type == RELAY_VIEWER_COMMAND) {
 710                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 711         }
 712
 713         health_code_update();
 714
 715         ret = send_response(conn->sock, &reply, sizeof(reply));
 716         if (ret < 0) {
 717                 goto end;
 718         }
 719
 720         health_code_update();
 721
 722         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 723         ret = 0;
 724
 725 end:
 726         return ret;
 727 }
 728
 729 /*
 730  * Send the viewer the list of current sessions.
 731  *
 732  * Return 0 on success or else a negative value.
 733  */
 734 static
 735 int viewer_list_sessions(struct relay_connection *conn)
 736 {
 737         int ret;
 738         struct lttng_viewer_list_sessions session_list;
 739         unsigned long count;
 740         long approx_before, approx_after;
 741         struct lttng_ht_iter iter;
 742         struct lttng_viewer_session send_session;
 743         struct relay_session *session;
 744
 745         DBG("List sessions received");
 746
 747         rcu_read_lock();
 748         cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
 749                         &approx_after);
 750         session_list.sessions_count = htobe32(count);
 751
 752         health_code_update();
 753
 754         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 755         if (ret < 0) {
 756                 goto end_unlock;
 757         }
 758
 759         health_code_update();
 760
 761         cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
 762                         session_n.node) {
 763                 health_code_update();
 764
 765                 strncpy(send_session.session_name, session->session_name,
 766                                 sizeof(send_session.session_name));
 767                 strncpy(send_session.hostname, session->hostname,
 768                                 sizeof(send_session.hostname));
 769                 send_session.id = htobe64(session->id);
 770                 send_session.live_timer = htobe32(session->live_timer);
 771                 send_session.clients = htobe32(session->viewer_refcount);
 772                 send_session.streams = htobe32(session->stream_count);
 773
 774                 health_code_update();
 775
 776                 ret = send_response(conn->sock, &send_session, sizeof(send_session));
 777                 if (ret < 0) {
 778                         goto end_unlock;
 779                 }
 780         }
 781         health_code_update();
 782
 783         ret = 0;
 784 end_unlock:
 785         rcu_read_unlock();
 786         return ret;
 787 }
 788
 789 /*
 790  * Check if a connection is attached to a session.
 791  * Return 1 if attached, 0 if not attached, a negative value on error.
 792  */
 793 static
 794 int session_attached(struct relay_connection *conn, uint64_t session_id)
 795 {
 796         struct relay_session *session;
 797         int found = 0;
 798
 799         if (!conn->viewer_session) {
 800                 goto end;
 801         }
 802         cds_list_for_each_entry(session,
 803                         &conn->viewer_session->sessions_head,
 804                         viewer_session_list) {
 805                 if (session->id == session_id) {
 806                         found = 1;
 807                         goto end;
 808                 }
 809         }
 810
 811 end:
 812         return found;
 813 }
 814
 815 /*
 816  * Delete all streams for a specific session ID.
 817  */
 818 static void destroy_viewer_streams_by_session(struct relay_session *session)
 819 {
 820         struct relay_viewer_stream *stream;
 821         struct lttng_ht_iter iter;
 822
 823         assert(session);
 824
 825         rcu_read_lock();
 826         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
 827                         stream_n.node) {
 828                 struct ctf_trace *ctf_trace;
 829
 830                 health_code_update();
 831                 if (stream->session_id != session->id) {
 832                         continue;
 833                 }
 834
 835                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 836                                 stream->path_name);
 837                 assert(ctf_trace);
 838
 839                 viewer_stream_delete(stream);
 840
 841                 if (stream->metadata_flag) {
 842                         ctf_trace->metadata_sent = 0;
 843                         ctf_trace->viewer_metadata_stream = NULL;
 844                 }
 845
 846                 viewer_stream_destroy(ctf_trace, stream);
 847         }
 848         rcu_read_unlock();
 849 }
 850
 851 static void try_destroy_streams(struct relay_session *session)
 852 {
 853         struct ctf_trace *ctf_trace;
 854         struct lttng_ht_iter iter;
 855
 856         assert(session);
 857
 858         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 859                         node.node) {
 860                 /* Attempt to destroy the ctf trace of that session. */
 861                 ctf_trace_try_destroy(session, ctf_trace);
 862         }
 863 }
 864
 865 /*
 866  * Cleanup a session.
 867  */
 868 static void cleanup_session(struct relay_connection *conn,
 869                 struct relay_session *session)
 870 {
 871         /*
 872          * Very important that this is done before destroying the session so we
 873          * can put back every viewer stream reference from the ctf_trace.
 874          */
 875         destroy_viewer_streams_by_session(session);
 876         try_destroy_streams(session);
 877         cds_list_del(&session->viewer_session_list);
 878         session_viewer_try_destroy(conn->sessions_ht, session);
 879 }
 880
 881 /*
 882  * Send the viewer the list of current sessions.
 883  */
 884 static
 885 int viewer_get_new_streams(struct relay_connection *conn)
 886 {
 887         int ret, send_streams = 0;
 888         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
 889         struct lttng_viewer_new_streams_request request;
 890         struct lttng_viewer_new_streams_response response;
 891         struct relay_session *session;
 892         uint64_t session_id;
 893
 894         assert(conn);
 895
 896         DBG("Get new streams received");
 897
 898         health_code_update();
 899
 900         /* Receive the request from the connected client. */
 901         ret = recv_request(conn->sock, &request, sizeof(request));
 902         if (ret < 0) {
 903                 goto error;
 904         }
 905         session_id = be64toh(request.session_id);
 906
 907         health_code_update();
 908
 909         memset(&response, 0, sizeof(response));
 910
 911         rcu_read_lock();
 912         session = session_find_by_id(conn->sessions_ht, session_id);
 913         if (!session) {
 914                 DBG("Relay session %" PRIu64 " not found", session_id);
 915                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 916                 goto send_reply;
 917         }
 918
 919         if (!session_attached(conn, session_id)) {
 920                 send_streams = 0;
 921                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 922                 goto send_reply;
 923         }
 924
 925         send_streams = 1;
 926         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 927
 928         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
 929                         &nb_created);
 930         if (ret < 0) {
 931                 goto end_unlock;
 932         }
 933         /* Only send back the newly created streams with the unsent ones. */
 934         nb_streams = nb_created + nb_unsent;
 935         response.streams_count = htobe32(nb_streams);
 936
 937         /*
 938          * If the session is closed and we have no new streams to send,
 939          * it means that the viewer has already received the whole trace
 940          * for this session and should now close it.
 941          */
 942         if (nb_streams == 0 && session->close_flag) {
 943                 send_streams = 0;
 944                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 945                 /*
 946                  * Remove the session from the attached list of the connection
 947                  * and try to destroy it.
 948                  */
 949                 cds_list_del(&session->viewer_session_list);
 950                 cleanup_session(conn, session);
 951                 goto send_reply;
 952         }
 953
 954 send_reply:
 955         health_code_update();
 956         ret = send_response(conn->sock, &response, sizeof(response));
 957         if (ret < 0) {
 958                 goto end_unlock;
 959         }
 960         health_code_update();
 961
 962         /*
 963          * Unknown or empty session, just return gracefully, the viewer knows what
 964          * is happening.
 965          */
 966         if (!send_streams || !nb_streams) {
 967                 ret = 0;
 968                 goto end_unlock;
 969         }
 970
 971         /*
 972          * Send stream and *DON'T* ignore the sent flag so every viewer streams
 973          * that were not sent from that point will be sent to the viewer.
 974          */
 975         ret = send_viewer_streams(conn->sock, session, 0);
 976         if (ret < 0) {
 977                 goto end_unlock;
 978         }
 979
 980 end_unlock:
 981         rcu_read_unlock();
 982 error:
 983         return ret;
 984 }
 985
 986 /*
 987  * Send the viewer the list of current sessions.
 988  */
 989 static
 990 int viewer_attach_session(struct relay_connection *conn)
 991 {
 992         int send_streams = 0;
 993         ssize_t ret;
 994         uint32_t nb_streams = 0;
 995         enum lttng_viewer_seek seek_type;
 996         struct lttng_viewer_attach_session_request request;
 997         struct lttng_viewer_attach_session_response response;
 998         struct relay_session *session;
 999
1000         assert(conn);
1001
1002         health_code_update();
1003
1004         /* Receive the request from the connected client. */
1005         ret = recv_request(conn->sock, &request, sizeof(request));
1006         if (ret < 0) {
1007                 goto error;
1008         }
1009
1010         health_code_update();
1011
1012         memset(&response, 0, sizeof(response));
1013
1014         if (!conn->viewer_session) {
1015                 DBG("Client trying to attach before creating a live viewer session");
1016                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1017                 goto send_reply;
1018         }
1019
1020         rcu_read_lock();
1021         session = session_find_by_id(conn->sessions_ht,
1022                         be64toh(request.session_id));
1023         if (!session) {
1024                 DBG("Relay session %" PRIu64 " not found",
1025                                 be64toh(request.session_id));
1026                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1027                 goto send_reply;
1028         }
1029         session_viewer_attach(session);
1030         DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1031
1032         if (uatomic_read(&session->viewer_refcount) > 1) {
1033                 DBG("Already a viewer attached");
1034                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1035                 session_viewer_detach(session);
1036                 goto send_reply;
1037         } else if (session->live_timer == 0) {
1038                 DBG("Not live session");
1039                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1040                 goto send_reply;
1041         } else {
1042                 send_streams = 1;
1043                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1044                 cds_list_add(&session->viewer_session_list,
1045                                 &conn->viewer_session->sessions_head);
1046         }
1047
1048         switch (be32toh(request.seek)) {
1049         case LTTNG_VIEWER_SEEK_BEGINNING:
1050         case LTTNG_VIEWER_SEEK_LAST:
1051                 seek_type = be32toh(request.seek);
1052                 break;
1053         default:
1054                 ERR("Wrong seek parameter");
1055                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1056                 send_streams = 0;
1057                 goto send_reply;
1058         }
1059
1060         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1061         if (ret < 0) {
1062                 goto end_unlock;
1063         }
1064         response.streams_count = htobe32(nb_streams);
1065
1066 send_reply:
1067         health_code_update();
1068         ret = send_response(conn->sock, &response, sizeof(response));
1069         if (ret < 0) {
1070                 goto end_unlock;
1071         }
1072         health_code_update();
1073
1074         /*
1075          * Unknown or empty session, just return gracefully, the viewer knows what
1076          * is happening.
1077          */
1078         if (!send_streams || !nb_streams) {
1079                 ret = 0;
1080                 goto end_unlock;
1081         }
1082
1083         /* Send stream and ignore the sent flag. */
1084         ret = send_viewer_streams(conn->sock, session, 1);
1085         if (ret < 0) {
1086                 goto end_unlock;
1087         }
1088
1089 end_unlock:
1090         rcu_read_unlock();
1091 error:
1092         return ret;
1093 }
1094
1095 /*
1096  * Open the index file if needed for the given vstream.
1097  *
1098  * If an index file is successfully opened, the index_read_fd of the stream is
1099  * set with it.
1100  *
1101  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1102  */
1103 static int try_open_index(struct relay_viewer_stream *vstream,
1104                 struct relay_stream *rstream)
1105 {
1106         int ret = 0;
1107
1108         assert(vstream);
1109         assert(rstream);
1110
1111         if (vstream->index_read_fd >= 0) {
1112                 goto end;
1113         }
1114
1115         /*
1116          * First time, we open the index file and at least one index is ready.  The
1117          * race between the read and write of the total_index_received is
1118          * acceptable here since the client will be notified to simply come back
1119          * and get the next index.
1120          */
1121         if (rstream->total_index_received <= 0) {
1122                 ret = -ENOENT;
1123                 goto end;
1124         }
1125         ret = index_open(vstream->path_name, vstream->channel_name,
1126                         vstream->tracefile_count, vstream->tracefile_count_current);
1127         if (ret >= 0) {
1128                 vstream->index_read_fd = ret;
1129                 ret = 0;
1130                 goto end;
1131         }
1132
1133 end:
1134         return ret;
1135 }
1136
1137 /*
1138  * Check the status of the index for the given stream. This function updates
1139  * the index structure if needed and can destroy the vstream also for the HUP
1140  * situation.
1141  *
1142  * Return 0 means that we can proceed with the index. A value of 1 means that
1143  * the index has been updated and is ready to be send to the client. A negative
1144  * value indicates an error that can't be handled.
1145  */
1146 static int check_index_status(struct relay_viewer_stream *vstream,
1147                 struct relay_stream *rstream, struct ctf_trace *trace,
1148                 struct lttng_viewer_index *index)
1149 {
1150         int ret;
1151
1152         assert(vstream);
1153         assert(rstream);
1154         assert(index);
1155         assert(trace);
1156
1157         if (!rstream->close_flag) {
1158                 /* Rotate on abort (overwrite). */
1159                 if (vstream->abort_flag) {
1160                         DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1161                                         vstream->stream_handle);
1162                         ret = viewer_stream_rotate(vstream, rstream);
1163                         if (ret < 0) {
1164                                 goto error;
1165                         } else if (ret == 1) {
1166                                 /* EOF */
1167                                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1168                                 goto hup;
1169                         }
1170                         /* ret == 0 means successful so we continue. */
1171                 }
1172
1173                 /* Check if we are in the same trace file at this point. */
1174                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1175                         if (rstream->beacon_ts_end != -1ULL &&
1176                                         vstream->last_sent_index == rstream->total_index_received) {
1177                                 /*
1178                                  * We've received a synchronization beacon and the last index
1179                                  * available has been sent, the index for now is inactive.
1180                                  */
1181                                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1182                                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1183                                 index->stream_id = htobe64(rstream->ctf_stream_id);
1184                                 goto index_ready;
1185                         } else if (rstream->total_index_received <= vstream->last_sent_index
1186                                         && !vstream->close_write_flag) {
1187                                 /*
1188                                  * Reader and writer are working in the same tracefile, so we care
1189                                  * about the number of index received and sent. Otherwise, we read
1190                                  * up to EOF.
1191                                  */
1192                                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1193                                 goto index_ready;
1194                         }
1195                 }
1196                 /* Nothing to do with the index, continue with it. */
1197                 ret = 0;
1198         } else if (rstream->close_flag && vstream->close_write_flag &&
1199                         vstream->total_index_received == vstream->last_sent_index) {
1200                 /* Last index sent and current tracefile closed in write */
1201                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1202                 goto hup;
1203         } else {
1204                 vstream->close_write_flag = 1;
1205                 ret = 0;
1206         }
1207
1208 error:
1209         return ret;
1210
1211 hup:
1212         viewer_stream_delete(vstream);
1213         viewer_stream_destroy(trace, vstream);
1214 index_ready:
1215         return 1;
1216 }
1217
1218 /*
1219  * Send the next index for a stream.
1220  *
1221  * Return 0 on success or else a negative value.
1222  */
1223 static
1224 int viewer_get_next_index(struct relay_connection *conn)
1225 {
1226         int ret;
1227         ssize_t read_ret;
1228         struct lttng_viewer_get_next_index request_index;
1229         struct lttng_viewer_index viewer_index;
1230         struct ctf_packet_index packet_index;
1231         struct relay_viewer_stream *vstream;
1232         struct relay_stream *rstream;
1233         struct ctf_trace *ctf_trace;
1234         struct relay_session *session;
1235
1236         assert(conn);
1237
1238         DBG("Viewer get next index");
1239
1240         health_code_update();
1241
1242         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1243         if (ret < 0) {
1244                 goto end;
1245         }
1246         health_code_update();
1247
1248         rcu_read_lock();
1249         vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1250         if (!vstream) {
1251                 ret = -1;
1252                 goto end_unlock;
1253         }
1254
1255         session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1256         if (!session) {
1257                 ret = -1;
1258                 goto end_unlock;
1259         }
1260
1261         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1262         assert(ctf_trace);
1263
1264         memset(&viewer_index, 0, sizeof(viewer_index));
1265
1266         /*
1267          * The viewer should not ask for index on metadata stream.
1268          */
1269         if (vstream->metadata_flag) {
1270                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1271                 goto send_reply;
1272         }
1273
1274         rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1275         assert(rstream);
1276
1277         /* Try to open an index if one is needed for that stream. */
1278         ret = try_open_index(vstream, rstream);
1279         if (ret < 0) {
1280                 if (ret == -ENOENT) {
1281                         /*
1282                          * The index is created only when the first data packet arrives, it
1283                          * might not be ready at the beginning of the session
1284                          */
1285                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1286                 } else {
1287                         /* Unhandled error. */
1288                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1289                 }
1290                 goto send_reply;
1291         }
1292
1293         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1294         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1295         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1296         if (ret < 0) {
1297                 goto end_unlock;
1298         } else if (ret == 1) {
1299                 /*
1300                  * This means the viewer index data structure has been populated by the
1301                  * check call thus we now send back the reply to the client.
1302                  */
1303                 goto send_reply;
1304         }
1305         /* At this point, ret MUST be 0 thus we continue with the get. */
1306         assert(!ret);
1307
1308         if (!ctf_trace->metadata_received ||
1309                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1310                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1311         }
1312
1313         ret = check_new_streams(conn);
1314         if (ret < 0) {
1315                 goto end_unlock;
1316         } else if (ret == 1) {
1317                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1318         }
1319
1320         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1321         pthread_mutex_lock(&vstream->overwrite_lock);
1322         if (vstream->abort_flag) {
1323                 /* The file is being overwritten by the writer, we cannot use it. */
1324                 pthread_mutex_unlock(&vstream->overwrite_lock);
1325                 ret = viewer_stream_rotate(vstream, rstream);
1326                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1327                 if (ret < 0) {
1328                         goto end_unlock;
1329                 } else if (ret == 1) {
1330                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1331                         viewer_stream_delete(vstream);
1332                         viewer_stream_destroy(ctf_trace, vstream);
1333                 } else {
1334                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1335                 }
1336                 goto send_reply;
1337         }
1338
1339         read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1340                         sizeof(packet_index));
1341         pthread_mutex_unlock(&vstream->overwrite_lock);
1342         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1343         if (read_ret < 0) {
1344                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1345                 viewer_stream_delete(vstream);
1346                 viewer_stream_destroy(ctf_trace, vstream);
1347                 goto send_reply;
1348         } else if (read_ret < sizeof(packet_index)) {
1349                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1350                 if (vstream->close_write_flag) {
1351                         ret = viewer_stream_rotate(vstream, rstream);
1352                         if (ret < 0) {
1353                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1354                                 goto end_unlock;
1355                         } else if (ret == 1) {
1356                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1357                                 viewer_stream_delete(vstream);
1358                                 viewer_stream_destroy(ctf_trace, vstream);
1359                         } else {
1360                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1361                         }
1362                 } else {
1363                         ERR("Relay reading index file %d", vstream->index_read_fd);
1364                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1365                 }
1366                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1367                 goto send_reply;
1368         } else {
1369                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1370                 vstream->last_sent_index++;
1371         }
1372
1373         /*
1374          * Indexes are stored in big endian, no need to switch before sending.
1375          */
1376         viewer_index.offset = packet_index.offset;
1377         viewer_index.packet_size = packet_index.packet_size;
1378         viewer_index.content_size = packet_index.content_size;
1379         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1380         viewer_index.timestamp_end = packet_index.timestamp_end;
1381         viewer_index.events_discarded = packet_index.events_discarded;
1382         viewer_index.stream_id = packet_index.stream_id;
1383
1384 send_reply:
1385         viewer_index.flags = htobe32(viewer_index.flags);
1386         health_code_update();
1387
1388         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1389         if (ret < 0) {
1390                 goto end_unlock;
1391         }
1392         health_code_update();
1393
1394         DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1395                         vstream->last_sent_index, vstream->stream_handle);
1396
1397 end_unlock:
1398         rcu_read_unlock();
1399
1400 end:
1401         return ret;
1402 }
1403
1404 /*
1405  * Send the next index for a stream
1406  *
1407  * Return 0 on success or else a negative value.
1408  */
1409 static
1410 int viewer_get_packet(struct relay_connection *conn)
1411 {
1412         int ret, send_data = 0;
1413         char *data = NULL;
1414         uint32_t len = 0;
1415         ssize_t read_len;
1416         struct lttng_viewer_get_packet get_packet_info;
1417         struct lttng_viewer_trace_packet reply;
1418         struct relay_viewer_stream *stream;
1419         struct relay_session *session;
1420         struct ctf_trace *ctf_trace;
1421
1422         assert(conn);
1423
1424         DBG2("Relay get data packet");
1425
1426         health_code_update();
1427
1428         ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1429         if (ret < 0) {
1430                 goto end;
1431         }
1432         health_code_update();
1433
1434         /* From this point on, the error label can be reached. */
1435         memset(&reply, 0, sizeof(reply));
1436
1437         rcu_read_lock();
1438         stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1439         if (!stream) {
1440                 goto error;
1441         }
1442
1443         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1444         if (!session) {
1445                 ret = -1;
1446                 goto error;
1447         }
1448
1449         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1450                         stream->path_name);
1451         assert(ctf_trace);
1452
1453         /*
1454          * First time we read this stream, we need open the tracefile, we should
1455          * only arrive here if an index has already been sent to the viewer, so the
1456          * tracefile must exist, if it does not it is a fatal error.
1457          */
1458         if (stream->read_fd < 0) {
1459                 char fullpath[PATH_MAX];
1460
1461                 if (stream->tracefile_count > 0) {
1462                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1463                                         stream->channel_name,
1464                                         stream->tracefile_count_current);
1465                 } else {
1466                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1467                                         stream->channel_name);
1468                 }
1469                 if (ret < 0) {
1470                         goto error;
1471                 }
1472                 ret = open(fullpath, O_RDONLY);
1473                 if (ret < 0) {
1474                         PERROR("Relay opening trace file");
1475                         goto error;
1476                 }
1477                 stream->read_fd = ret;
1478         }
1479
1480         if (!ctf_trace->metadata_received ||
1481                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1482                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1483                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1484                 goto send_reply;
1485         }
1486
1487         ret = check_new_streams(conn);
1488         if (ret < 0) {
1489                 goto end_unlock;
1490         } else if (ret == 1) {
1491                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1492                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1493                 goto send_reply;
1494         }
1495
1496         len = be32toh(get_packet_info.len);
1497         data = zmalloc(len);
1498         if (!data) {
1499                 PERROR("relay data zmalloc");
1500                 goto error;
1501         }
1502
1503         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1504         if (ret < 0) {
1505                 /*
1506                  * If the read fd was closed by the streaming side, the
1507                  * abort_flag will be set to 1, otherwise it is an error.
1508                  */
1509                 if (stream->abort_flag == 0) {
1510                         PERROR("lseek");
1511                         goto error;
1512                 }
1513                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1514                 goto send_reply;
1515         }
1516         read_len = lttng_read(stream->read_fd, data, len);
1517         if (read_len < len) {
1518                 /*
1519                  * If the read fd was closed by the streaming side, the
1520                  * abort_flag will be set to 1, otherwise it is an error.
1521                  */
1522                 if (stream->abort_flag == 0) {
1523                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1524                                         stream->read_fd,
1525                                         be64toh(get_packet_info.offset));
1526                         goto error;
1527                 } else {
1528                         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1529                         goto send_reply;
1530                 }
1531         }
1532         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1533         reply.len = htobe32(len);
1534         send_data = 1;
1535         goto send_reply;
1536
1537 error:
1538         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1539
1540 send_reply:
1541         reply.flags = htobe32(reply.flags);
1542
1543         health_code_update();
1544
1545         ret = send_response(conn->sock, &reply, sizeof(reply));
1546         if (ret < 0) {
1547                 goto end_unlock;
1548         }
1549         health_code_update();
1550
1551         if (send_data) {
1552                 health_code_update();
1553                 ret = send_response(conn->sock, data, len);
1554                 if (ret < 0) {
1555                         goto end_unlock;
1556                 }
1557                 health_code_update();
1558         }
1559
1560         DBG("Sent %u bytes for stream %" PRIu64, len,
1561                         be64toh(get_packet_info.stream_id));
1562
1563 end_unlock:
1564         free(data);
1565         rcu_read_unlock();
1566
1567 end:
1568         return ret;
1569 }
1570
1571 /*
1572  * Send the session's metadata
1573  *
1574  * Return 0 on success else a negative value.
1575  */
1576 static
1577 int viewer_get_metadata(struct relay_connection *conn)
1578 {
1579         int ret = 0;
1580         ssize_t read_len;
1581         uint64_t len = 0;
1582         char *data = NULL;
1583         struct lttng_viewer_get_metadata request;
1584         struct lttng_viewer_metadata_packet reply;
1585         struct relay_viewer_stream *stream;
1586         struct ctf_trace *ctf_trace;
1587         struct relay_session *session;
1588
1589         assert(conn);
1590
1591         DBG("Relay get metadata");
1592
1593         health_code_update();
1594
1595         ret = recv_request(conn->sock, &request, sizeof(request));
1596         if (ret < 0) {
1597                 goto end;
1598         }
1599         health_code_update();
1600
1601         memset(&reply, 0, sizeof(reply));
1602
1603         rcu_read_lock();
1604         stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1605         if (!stream || !stream->metadata_flag) {
1606                 ERR("Invalid metadata stream");
1607                 goto error;
1608         }
1609
1610         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1611         if (!session) {
1612                 ret = -1;
1613                 goto error;
1614         }
1615
1616         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1617                         stream->path_name);
1618         assert(ctf_trace);
1619         assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1620
1621         len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1622         if (len == 0) {
1623                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1624                 goto send_reply;
1625         }
1626
1627         /* first time, we open the metadata file */
1628         if (stream->read_fd < 0) {
1629                 char fullpath[PATH_MAX];
1630
1631                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1632                                 stream->channel_name);
1633                 if (ret < 0) {
1634                         goto error;
1635                 }
1636                 ret = open(fullpath, O_RDONLY);
1637                 if (ret < 0) {
1638                         PERROR("Relay opening metadata file");
1639                         goto error;
1640                 }
1641                 stream->read_fd = ret;
1642         }
1643
1644         reply.len = htobe64(len);
1645         data = zmalloc(len);
1646         if (!data) {
1647                 PERROR("viewer metadata zmalloc");
1648                 goto error;
1649         }
1650
1651         read_len = lttng_read(stream->read_fd, data, len);
1652         if (read_len < len) {
1653                 PERROR("Relay reading metadata file");
1654                 goto error;
1655         }
1656         ctf_trace->metadata_sent += read_len;
1657         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1658         goto send_reply;
1659
1660 error:
1661         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1662
1663 send_reply:
1664         health_code_update();
1665         ret = send_response(conn->sock, &reply, sizeof(reply));
1666         if (ret < 0) {
1667                 goto end_unlock;
1668         }
1669         health_code_update();
1670
1671         if (len > 0) {
1672                 ret = send_response(conn->sock, data, len);
1673                 if (ret < 0) {
1674                         goto end_unlock;
1675                 }
1676         }
1677
1678         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1679                         be64toh(request.stream_id));
1680
1681         DBG("Metadata sent");
1682
1683 end_unlock:
1684         free(data);
1685         rcu_read_unlock();
1686 end:
1687         return ret;
1688 }
1689
1690 /*
1691  * Create a viewer session.
1692  *
1693  * Return 0 on success or else a negative value.
1694  */
1695 static
1696 int viewer_create_session(struct relay_connection *conn)
1697 {
1698         int ret;
1699         struct lttng_viewer_create_session_response resp;
1700
1701         DBG("Viewer create session received");
1702
1703         memset(&resp, 0, sizeof(resp));
1704         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1705         conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1706         if (!conn->viewer_session) {
1707                 ERR("Allocation viewer session");
1708                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1709                 goto send_reply;
1710         }
1711         CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1712
1713 send_reply:
1714         health_code_update();
1715         ret = send_response(conn->sock, &resp, sizeof(resp));
1716         if (ret < 0) {
1717                 goto end;
1718         }
1719         health_code_update();
1720         ret = 0;
1721
1722 end:
1723         return ret;
1724 }
1725
1726
1727 /*
1728  * live_relay_unknown_command: send -1 if received unknown command
1729  */
1730 static
1731 void live_relay_unknown_command(struct relay_connection *conn)
1732 {
1733         struct lttcomm_relayd_generic_reply reply;
1734
1735         memset(&reply, 0, sizeof(reply));
1736         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1737         (void) send_response(conn->sock, &reply, sizeof(reply));
1738 }
1739
1740 /*
1741  * Process the commands received on the control socket
1742  */
1743 static
1744 int process_control(struct lttng_viewer_cmd *recv_hdr,
1745                 struct relay_connection *conn)
1746 {
1747         int ret = 0;
1748         uint32_t msg_value;
1749
1750         assert(recv_hdr);
1751         assert(conn);
1752
1753         msg_value = be32toh(recv_hdr->cmd);
1754
1755         /*
1756          * Make sure we've done the version check before any command other then a
1757          * new client connection.
1758          */
1759         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1760                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1761                 ret = -1;
1762                 goto end;
1763         }
1764
1765         switch (msg_value) {
1766         case LTTNG_VIEWER_CONNECT:
1767                 ret = viewer_connect(conn);
1768                 break;
1769         case LTTNG_VIEWER_LIST_SESSIONS:
1770                 ret = viewer_list_sessions(conn);
1771                 break;
1772         case LTTNG_VIEWER_ATTACH_SESSION:
1773                 ret = viewer_attach_session(conn);
1774                 break;
1775         case LTTNG_VIEWER_GET_NEXT_INDEX:
1776                 ret = viewer_get_next_index(conn);
1777                 break;
1778         case LTTNG_VIEWER_GET_PACKET:
1779                 ret = viewer_get_packet(conn);
1780                 break;
1781         case LTTNG_VIEWER_GET_METADATA:
1782                 ret = viewer_get_metadata(conn);
1783                 break;
1784         case LTTNG_VIEWER_GET_NEW_STREAMS:
1785                 ret = viewer_get_new_streams(conn);
1786                 break;
1787         case LTTNG_VIEWER_CREATE_SESSION:
1788                 ret = viewer_create_session(conn);
1789                 break;
1790         default:
1791                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1792                 live_relay_unknown_command(conn);
1793                 ret = -1;
1794                 goto end;
1795         }
1796
1797 end:
1798         return ret;
1799 }
1800
1801 static
1802 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1803 {
1804         int ret;
1805
1806         assert(events);
1807
1808         (void) lttng_poll_del(events, pollfd);
1809
1810         ret = close(pollfd);
1811         if (ret < 0) {
1812                 ERR("Closing pollfd %d", pollfd);
1813         }
1814 }
1815
1816 /*
1817  * Delete and destroy a connection.
1818  *
1819  * RCU read side lock MUST be acquired.
1820  */
1821 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1822                 struct relay_connection *conn)
1823 {
1824         struct relay_session *session, *tmp_session;
1825
1826         assert(relay_connections_ht);
1827         assert(conn);
1828
1829         connection_delete(relay_connections_ht, conn);
1830
1831         if (!conn->viewer_session) {
1832                 goto end;
1833         }
1834
1835         rcu_read_lock();
1836         cds_list_for_each_entry_safe(session, tmp_session,
1837                         &conn->viewer_session->sessions_head,
1838                         viewer_session_list) {
1839                 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1840                 cleanup_session(conn, session);
1841         }
1842         rcu_read_unlock();
1843
1844 end:
1845         connection_destroy(conn);
1846 }
1847
1848 /*
1849  * This thread does the actual work
1850  */
1851 static
1852 void *thread_worker(void *data)
1853 {
1854         int ret, err = -1;
1855         uint32_t nb_fd;
1856         struct relay_connection *conn;
1857         struct lttng_poll_event events;
1858         struct lttng_ht *relay_connections_ht;
1859         struct lttng_ht_iter iter;
1860         struct lttng_viewer_cmd recv_hdr;
1861         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1862         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1863
1864         DBG("[thread] Live viewer relay worker started");
1865
1866         rcu_register_thread();
1867
1868         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1869
1870         if (testpoint(relayd_thread_live_worker)) {
1871                 goto error_testpoint;
1872         }
1873
1874         /* table of connections indexed on socket */
1875         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1876         if (!relay_connections_ht) {
1877                 goto relay_connections_ht_error;
1878         }
1879
1880         ret = create_thread_poll_set(&events, 2);
1881         if (ret < 0) {
1882                 goto error_poll_create;
1883         }
1884
1885         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1886         if (ret < 0) {
1887                 goto error;
1888         }
1889
1890 restart:
1891         while (1) {
1892                 int i;
1893
1894                 health_code_update();
1895
1896                 /* Infinite blocking call, waiting for transmission */
1897                 DBG3("Relayd live viewer worker thread polling...");
1898                 health_poll_entry();
1899                 ret = lttng_poll_wait(&events, -1);
1900                 health_poll_exit();
1901                 if (ret < 0) {
1902                         /*
1903                          * Restart interrupted system call.
1904                          */
1905                         if (errno == EINTR) {
1906                                 goto restart;
1907                         }
1908                         goto error;
1909                 }
1910
1911                 nb_fd = ret;
1912
1913                 /*
1914                  * Process control. The control connection is prioritised so we don't
1915                  * starve it with high throughput tracing data on the data
1916                  * connection.
1917                  */
1918                 for (i = 0; i < nb_fd; i++) {
1919                         /* Fetch once the poll data */
1920                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1921                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1922
1923                         health_code_update();
1924
1925                         /* Thread quit pipe has been closed. Killing thread. */
1926                         ret = check_thread_quit_pipe(pollfd, revents);
1927                         if (ret) {
1928                                 err = 0;
1929                                 goto exit;
1930                         }
1931
1932                         /* Inspect the relay conn pipe for new connection */
1933                         if (pollfd == live_conn_pipe[0]) {
1934                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1935                                         ERR("Relay live pipe error");
1936                                         goto error;
1937                                 } else if (revents & LPOLLIN) {
1938                                         ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1939                                         if (ret < 0) {
1940                                                 goto error;
1941                                         }
1942                                         conn->sessions_ht = sessions_ht;
1943                                         connection_init(conn);
1944                                         lttng_poll_add(&events, conn->sock->fd,
1945                                                         LPOLLIN | LPOLLRDHUP);
1946                                         rcu_read_lock();
1947                                         lttng_ht_add_unique_ulong(relay_connections_ht,
1948                                                         &conn->sock_n);
1949                                         rcu_read_unlock();
1950                                         DBG("Connection socket %d added", conn->sock->fd);
1951                                 }
1952                         } else {
1953                                 rcu_read_lock();
1954                                 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1955                                 /* If not found, there is a synchronization issue. */
1956                                 assert(conn);
1957
1958                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1959                                         cleanup_connection_pollfd(&events, pollfd);
1960                                         destroy_connection(relay_connections_ht, conn);
1961                                 } else 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                                                 destroy_connection(relay_connections_ht, conn);
1968                                                 DBG("Viewer control conn closed with %d", pollfd);
1969                                         } else {
1970                                                 ret = process_control(&recv_hdr, conn);
1971                                                 if (ret < 0) {
1972                                                         /* Clear the session on error. */
1973                                                         cleanup_connection_pollfd(&events, pollfd);
1974                                                         destroy_connection(relay_connections_ht, conn);
1975                                                         DBG("Viewer connection closed with %d", pollfd);
1976                                                 }
1977                                         }
1978                                 }
1979                                 rcu_read_unlock();
1980                         }
1981                 }
1982         }
1983
1984 exit:
1985 error:
1986         lttng_poll_clean(&events);
1987
1988         /* Cleanup reamaining connection object. */
1989         rcu_read_lock();
1990         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
1991                         sock_n.node) {
1992                 health_code_update();
1993                 destroy_connection(relay_connections_ht, conn);
1994         }
1995         rcu_read_unlock();
1996 error_poll_create:
1997         lttng_ht_destroy(relay_connections_ht);
1998 relay_connections_ht_error:
1999         /* Close relay conn pipes */
2000         utils_close_pipe(live_conn_pipe);
2001         if (err) {
2002                 DBG("Viewer worker thread exited with error");
2003         }
2004         DBG("Viewer worker thread cleanup complete");
2005 error_testpoint:
2006         if (err) {
2007                 health_error();
2008                 ERR("Health error occurred in %s", __func__);
2009         }
2010         health_unregister(health_relayd);
2011         if (lttng_relay_stop_threads()) {
2012                 ERR("Error stopping threads");
2013         }
2014         rcu_unregister_thread();
2015         return NULL;
2016 }
2017
2018 /*
2019  * Create the relay command pipe to wake thread_manage_apps.
2020  * Closed in cleanup().
2021  */
2022 static int create_conn_pipe(void)
2023 {
2024         return utils_create_pipe_cloexec(live_conn_pipe);
2025 }
2026
2027 int relayd_live_join(void)
2028 {
2029         int ret, retval = 0;
2030         void *status;
2031
2032         ret = pthread_join(live_listener_thread, &status);
2033         if (ret) {
2034                 errno = ret;
2035                 PERROR("pthread_join live listener");
2036                 retval = -1;
2037         }
2038
2039         ret = pthread_join(live_worker_thread, &status);
2040         if (ret) {
2041                 errno = ret;
2042                 PERROR("pthread_join live worker");
2043                 retval = -1;
2044         }
2045
2046         ret = pthread_join(live_dispatcher_thread, &status);
2047         if (ret) {
2048                 errno = ret;
2049                 PERROR("pthread_join live dispatcher");
2050                 retval = -1;
2051         }
2052
2053         cleanup_relayd_live();
2054
2055         return retval;
2056 }
2057
2058 /*
2059  * main
2060  */
2061 int relayd_live_create(struct lttng_uri *uri,
2062                 struct relay_local_data *relay_ctx)
2063 {
2064         int ret = 0, retval = 0;
2065         void *status;
2066         int is_root;
2067
2068         if (!uri) {
2069                 retval = -1;
2070                 goto exit_init_data;
2071         }
2072         live_uri = uri;
2073
2074         /* Check if daemon is UID = 0 */
2075         is_root = !getuid();
2076
2077         if (!is_root) {
2078                 if (live_uri->port < 1024) {
2079                         ERR("Need to be root to use ports < 1024");
2080                         retval = -1;
2081                         goto exit_init_data;
2082                 }
2083         }
2084
2085         /* Setup the thread apps communication pipe. */
2086         if (create_conn_pipe()) {
2087                 retval = -1;
2088                 goto exit_init_data;
2089         }
2090
2091         /* Init relay command queue. */
2092         cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2093
2094         /* Set up max poll set size */
2095         lttng_poll_set_max_size();
2096
2097         /* Setup the dispatcher thread */
2098         ret = pthread_create(&live_dispatcher_thread, NULL,
2099                         thread_dispatcher, (void *) NULL);
2100         if (ret) {
2101                 errno = ret;
2102                 PERROR("pthread_create viewer dispatcher");
2103                 retval = -1;
2104                 goto exit_dispatcher_thread;
2105         }
2106
2107         /* Setup the worker thread */
2108         ret = pthread_create(&live_worker_thread, NULL,
2109                         thread_worker, relay_ctx);
2110         if (ret) {
2111                 errno = ret;
2112                 PERROR("pthread_create viewer worker");
2113                 retval = -1;
2114                 goto exit_worker_thread;
2115         }
2116
2117         /* Setup the listener thread */
2118         ret = pthread_create(&live_listener_thread, NULL,
2119                         thread_listener, (void *) NULL);
2120         if (ret) {
2121                 errno = ret;
2122                 PERROR("pthread_create viewer listener");
2123                 retval = -1;
2124                 goto exit_listener_thread;
2125         }
2126
2127         /*
2128          * All OK, started all threads.
2129          */
2130         return retval;
2131
2132
2133         ret = pthread_join(live_listener_thread, &status);
2134         if (ret) {
2135                 errno = ret;
2136                 PERROR("pthread_join live listener");
2137                 retval = -1;
2138         }
2139 exit_listener_thread:
2140
2141         ret = pthread_join(live_worker_thread, &status);
2142         if (ret) {
2143                 errno = ret;
2144                 PERROR("pthread_join live worker");
2145                 retval = -1;
2146         }
2147 exit_worker_thread:
2148
2149         ret = pthread_join(live_dispatcher_thread, &status);
2150         if (ret) {
2151                 errno = ret;
2152                 PERROR("pthread_join live dispatcher");
2153                 retval = -1;
2154         }
2155 exit_dispatcher_thread:
2156
2157 exit_init_data:
2158         cleanup_relayd_live();
2159
2160         return retval;
2161 }