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 #include <getopt.h>
  21 #include <grp.h>
  22 #include <limits.h>
  23 #include <pthread.h>
  24 #include <signal.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <sys/mman.h>
  29 #include <sys/mount.h>
  30 #include <sys/resource.h>
  31 #include <sys/socket.h>
  32 #include <sys/stat.h>
  33 #include <sys/types.h>
  34 #include <sys/wait.h>
  35 #include <inttypes.h>
  36 #include <urcu/futex.h>
  37 #include <urcu/uatomic.h>
  38 #include <unistd.h>
  39 #include <fcntl.h>
  40 #include <config.h>
  41
  42 #include <lttng/lttng.h>
  43 #include <common/common.h>
  44 #include <common/compat/poll.h>
  45 #include <common/compat/socket.h>
  46 #include <common/compat/endian.h>
  47 #include <common/defaults.h>
  48 #include <common/futex.h>
  49 #include <common/index/index.h>
  50 #include <common/sessiond-comm/sessiond-comm.h>
  51 #include <common/sessiond-comm/inet.h>
  52 #include <common/sessiond-comm/relayd.h>
  53 #include <common/uri.h>
  54 #include <common/utils.h>
  55
  56 #include "cmd.h"
  57 #include "live.h"
  58 #include "lttng-relayd.h"
  59 #include "utils.h"
  60 #include "health-relayd.h"
  61 #include "testpoint.h"
  62 #include "viewer-stream.h"
  63 #include "stream.h"
  64 #include "session.h"
  65 #include "ctf-trace.h"
  66 #include "connection.h"
  67
  68 static struct lttng_uri *live_uri;
  69
  70 /*
  71  * This pipe is used to inform the worker thread that a command is queued and
  72  * ready to be processed.
  73  */
  74 static int live_conn_pipe[2] = { -1, -1 };
  75
  76 /* Shared between threads */
  77 static int live_dispatch_thread_exit;
  78
  79 static pthread_t live_listener_thread;
  80 static pthread_t live_dispatcher_thread;
  81 static pthread_t live_worker_thread;
  82
  83 /*
  84  * Relay command queue.
  85  *
  86  * The live_thread_listener and live_thread_dispatcher communicate with this
  87  * queue.
  88  */
  89 static struct relay_conn_queue viewer_conn_queue;
  90
  91 static uint64_t last_relay_viewer_session_id;
  92
  93 /*
  94  * Cleanup the daemon
  95  */
  96 static
  97 void cleanup(void)
  98 {
  99         DBG("Cleaning up");
 100
 101         free(live_uri);
 102 }
 103
 104 /*
 105  * Receive a request buffer using a given socket, destination allocated buffer
 106  * of length size.
 107  *
 108  * Return the size of the received message or else a negative value on error
 109  * with errno being set by recvmsg() syscall.
 110  */
 111 static
 112 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 113 {
 114         ssize_t ret;
 115
 116         assert(sock);
 117         assert(buf);
 118
 119         ret = sock->ops->recvmsg(sock, buf, size, 0);
 120         if (ret < 0 || ret != size) {
 121                 if (ret == 0) {
 122                         /* Orderly shutdown. Not necessary to print an error. */
 123                         DBG("Socket %d did an orderly shutdown", sock->fd);
 124                 } else {
 125                         ERR("Relay failed to receive request.");
 126                 }
 127                 ret = -1;
 128         }
 129
 130         return ret;
 131 }
 132
 133 /*
 134  * Send a response buffer using a given socket, source allocated buffer of
 135  * length size.
 136  *
 137  * Return the size of the sent message or else a negative value on error with
 138  * errno being set by sendmsg() syscall.
 139  */
 140 static
 141 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 142 {
 143         ssize_t ret;
 144
 145         assert(sock);
 146         assert(buf);
 147
 148         ret = sock->ops->sendmsg(sock, buf, size, 0);
 149         if (ret < 0) {
 150                 ERR("Relayd failed to send response.");
 151         }
 152
 153         return ret;
 154 }
 155
 156 /*
 157  * Atomically check if new streams got added in one of the sessions attached
 158  * and reset the flag to 0.
 159  *
 160  * Returns 1 if new streams got added, 0 if nothing changed, a negative value
 161  * on error.
 162  */
 163 static
 164 int check_new_streams(struct relay_connection *conn)
 165 {
 166         struct relay_session *session;
 167         unsigned long current_val;
 168         int ret = 0;
 169
 170         if (!conn->viewer_session) {
 171                 goto end;
 172         }
 173         cds_list_for_each_entry(session,
 174                         &conn->viewer_session->sessions_head,
 175                         viewer_session_list) {
 176                 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
 177                 ret = current_val;
 178                 if (ret == 1) {
 179                         goto end;
 180                 }
 181         }
 182
 183 end:
 184         return ret;
 185 }
 186
 187 /*
 188  * Send viewer streams to the given socket. The ignore_sent_flag indicates if
 189  * this function should ignore the sent flag or not.
 190  *
 191  * Return 0 on success or else a negative value.
 192  */
 193 static
 194 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
 195                 struct relay_session *session, unsigned int ignore_sent_flag)
 196 {
 197         ssize_t ret;
 198         struct lttng_viewer_stream send_stream;
 199         struct lttng_ht_iter iter;
 200         struct relay_viewer_stream *vstream;
 201
 202         assert(session);
 203
 204         rcu_read_lock();
 205
 206         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
 207                         stream_n.node) {
 208                 struct ctf_trace *ctf_trace;
 209
 210                 health_code_update();
 211
 212                 /* Ignore if not the same session. */
 213                 if (vstream->session_id != session->id ||
 214                                 (!ignore_sent_flag && vstream->sent_flag)) {
 215                         continue;
 216                 }
 217
 218                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 219                                 vstream->path_name);
 220                 assert(ctf_trace);
 221
 222                 send_stream.id = htobe64(vstream->stream_handle);
 223                 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
 224                 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
 225                 strncpy(send_stream.path_name, vstream->path_name,
 226                                 sizeof(send_stream.path_name));
 227                 strncpy(send_stream.channel_name, vstream->channel_name,
 228                                 sizeof(send_stream.channel_name));
 229
 230                 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
 231                 ret = send_response(sock, &send_stream, sizeof(send_stream));
 232                 if (ret < 0) {
 233                         goto end_unlock;
 234                 }
 235                 vstream->sent_flag = 1;
 236         }
 237
 238         ret = 0;
 239
 240 end_unlock:
 241         rcu_read_unlock();
 242         return ret;
 243 }
 244
 245 /*
 246  * Create every viewer stream possible for the given session with the seek
 247  * type. Three counters *can* be return which are in order the total amount of
 248  * viewer stream of the session, the number of unsent stream and the number of
 249  * stream created. Those counters can be NULL and thus will be ignored.
 250  *
 251  * Return 0 on success or else a negative value.
 252  */
 253 static
 254 int make_viewer_streams(struct relay_session *session,
 255                 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
 256                 uint32_t *nb_created)
 257 {
 258         int ret;
 259         struct lttng_ht_iter iter;
 260         struct ctf_trace *ctf_trace;
 261
 262         assert(session);
 263
 264         /*
 265          * This is to make sure we create viewer streams for a full received
 266          * channel. For instance, if we have 8 streams for a channel that are
 267          * concurrently being flagged ready, we can end up creating just a subset
 268          * of the 8 streams (the ones that are flagged). This lock avoids this
 269          * limbo state.
 270          */
 271         pthread_mutex_lock(&session->viewer_ready_lock);
 272
 273         /*
 274          * Create viewer streams for relay streams that are ready to be used for a
 275          * the given session id only.
 276          */
 277         rcu_read_lock();
 278         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 279                         node.node) {
 280                 struct relay_stream *stream;
 281
 282                 health_code_update();
 283
 284                 if (ctf_trace->invalid_flag) {
 285                         continue;
 286                 }
 287
 288                 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
 289                         struct relay_viewer_stream *vstream;
 290
 291                         if (!stream->viewer_ready) {
 292                                 continue;
 293                         }
 294
 295                         vstream = viewer_stream_find_by_id(stream->stream_handle);
 296                         if (!vstream) {
 297                                 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
 298                                 if (!vstream) {
 299                                         ret = -1;
 300                                         goto error_unlock;
 301                                 }
 302                                 /* Acquire reference to ctf_trace. */
 303                                 ctf_trace_get_ref(ctf_trace);
 304
 305                                 if (nb_created) {
 306                                         /* Update number of created stream counter. */
 307                                         (*nb_created)++;
 308                                 }
 309                         } else if (!vstream->sent_flag && nb_unsent) {
 310                                 /* Update number of unsent stream counter. */
 311                                 (*nb_unsent)++;
 312                         }
 313                         /* Update number of total stream counter. */
 314                         if (nb_total) {
 315                                 (*nb_total)++;
 316                         }
 317                 }
 318         }
 319
 320         ret = 0;
 321
 322 error_unlock:
 323         rcu_read_unlock();
 324         pthread_mutex_unlock(&session->viewer_ready_lock);
 325         return ret;
 326 }
 327
 328 /*
 329  * Write to writable pipe used to notify a thread.
 330  */
 331 static
 332 int notify_thread_pipe(int wpipe)
 333 {
 334         ssize_t ret;
 335
 336         ret = lttng_write(wpipe, "!", 1);
 337         if (ret < 1) {
 338                 PERROR("write poll pipe");
 339         }
 340
 341         return (int) ret;
 342 }
 343
 344 /*
 345  * Stop all threads by closing the thread quit pipe.
 346  */
 347 static
 348 void stop_threads(void)
 349 {
 350         int ret;
 351
 352         /* Stopping all threads */
 353         DBG("Terminating all live threads");
 354         ret = notify_thread_pipe(thread_quit_pipe[1]);
 355         if (ret < 0) {
 356                 ERR("write error on thread quit pipe");
 357         }
 358
 359         /* Dispatch thread */
 360         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 361         futex_nto1_wake(&viewer_conn_queue.futex);
 362 }
 363
 364 /*
 365  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 366  */
 367 static
 368 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 369 {
 370         int ret;
 371
 372         if (events == NULL || size == 0) {
 373                 ret = -1;
 374                 goto error;
 375         }
 376
 377         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 378         if (ret < 0) {
 379                 goto error;
 380         }
 381
 382         /* Add quit pipe */
 383         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 384         if (ret < 0) {
 385                 goto error;
 386         }
 387
 388         return 0;
 389
 390 error:
 391         return ret;
 392 }
 393
 394 /*
 395  * Check if the thread quit pipe was triggered.
 396  *
 397  * Return 1 if it was triggered else 0;
 398  */
 399 static
 400 int check_thread_quit_pipe(int fd, uint32_t events)
 401 {
 402         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 403                 return 1;
 404         }
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * Create and init socket from uri.
 411  */
 412 static
 413 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 414 {
 415         int ret;
 416         struct lttcomm_sock *sock = NULL;
 417
 418         sock = lttcomm_alloc_sock_from_uri(uri);
 419         if (sock == NULL) {
 420                 ERR("Allocating socket");
 421                 goto error;
 422         }
 423
 424         ret = lttcomm_create_sock(sock);
 425         if (ret < 0) {
 426                 goto error;
 427         }
 428         DBG("Listening on sock %d for live", sock->fd);
 429
 430         ret = sock->ops->bind(sock);
 431         if (ret < 0) {
 432                 goto error;
 433         }
 434
 435         ret = sock->ops->listen(sock, -1);
 436         if (ret < 0) {
 437                 goto error;
 438
 439         }
 440
 441         return sock;
 442
 443 error:
 444         if (sock) {
 445                 lttcomm_destroy_sock(sock);
 446         }
 447         return NULL;
 448 }
 449
 450 /*
 451  * This thread manages the listening for new connections on the network
 452  */
 453 static
 454 void *thread_listener(void *data)
 455 {
 456         int i, ret, pollfd, err = -1;
 457         uint32_t revents, nb_fd;
 458         struct lttng_poll_event events;
 459         struct lttcomm_sock *live_control_sock;
 460
 461         DBG("[thread] Relay live listener started");
 462
 463         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 464
 465         health_code_update();
 466
 467         live_control_sock = init_socket(live_uri);
 468         if (!live_control_sock) {
 469                 goto error_sock_control;
 470         }
 471
 472         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 473         ret = create_thread_poll_set(&events, 2);
 474         if (ret < 0) {
 475                 goto error_create_poll;
 476         }
 477
 478         /* Add the control socket */
 479         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 480         if (ret < 0) {
 481                 goto error_poll_add;
 482         }
 483
 484         lttng_relay_notify_ready();
 485
 486         if (testpoint(relayd_thread_live_listener)) {
 487                 goto error_testpoint;
 488         }
 489
 490         while (1) {
 491                 health_code_update();
 492
 493                 DBG("Listener accepting live viewers connections");
 494
 495 restart:
 496                 health_poll_entry();
 497                 ret = lttng_poll_wait(&events, -1);
 498                 health_poll_exit();
 499                 if (ret < 0) {
 500                         /*
 501                          * Restart interrupted system call.
 502                          */
 503                         if (errno == EINTR) {
 504                                 goto restart;
 505                         }
 506                         goto error;
 507                 }
 508                 nb_fd = ret;
 509
 510                 DBG("Relay new viewer connection received");
 511                 for (i = 0; i < nb_fd; i++) {
 512                         health_code_update();
 513
 514                         /* Fetch once the poll data */
 515                         revents = LTTNG_POLL_GETEV(&events, i);
 516                         pollfd = LTTNG_POLL_GETFD(&events, i);
 517
 518                         /* Thread quit pipe has been closed. Killing thread. */
 519                         ret = check_thread_quit_pipe(pollfd, revents);
 520                         if (ret) {
 521                                 err = 0;
 522                                 goto exit;
 523                         }
 524
 525                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 526                                 ERR("socket poll error");
 527                                 goto error;
 528                         } else if (revents & LPOLLIN) {
 529                                 /*
 530                                  * Get allocated in this thread, enqueued to a global queue,
 531                                  * dequeued and freed in the worker thread.
 532                                  */
 533                                 int val = 1;
 534                                 struct relay_connection *new_conn;
 535                                 struct lttcomm_sock *newsock;
 536
 537                                 new_conn = connection_create();
 538                                 if (!new_conn) {
 539                                         goto error;
 540                                 }
 541
 542                                 newsock = live_control_sock->ops->accept(live_control_sock);
 543                                 if (!newsock) {
 544                                         PERROR("accepting control sock");
 545                                         connection_free(new_conn);
 546                                         goto error;
 547                                 }
 548                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 549
 550                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 551                                                 sizeof(val));
 552                                 if (ret < 0) {
 553                                         PERROR("setsockopt inet");
 554                                         lttcomm_destroy_sock(newsock);
 555                                         connection_free(new_conn);
 556                                         goto error;
 557                                 }
 558                                 new_conn->sock = newsock;
 559
 560                                 /* Enqueue request for the dispatcher thread. */
 561                                 cds_wfq_enqueue(&viewer_conn_queue.queue, &new_conn->qnode);
 562
 563                                 /*
 564                                  * Wake the dispatch queue futex. Implicit memory barrier with
 565                                  * the exchange in cds_wfq_enqueue.
 566                                  */
 567                                 futex_nto1_wake(&viewer_conn_queue.futex);
 568                         }
 569                 }
 570         }
 571
 572 exit:
 573 error:
 574 error_poll_add:
 575 error_testpoint:
 576         lttng_poll_clean(&events);
 577 error_create_poll:
 578         if (live_control_sock->fd >= 0) {
 579                 ret = live_control_sock->ops->close(live_control_sock);
 580                 if (ret) {
 581                         PERROR("close");
 582                 }
 583         }
 584         lttcomm_destroy_sock(live_control_sock);
 585 error_sock_control:
 586         if (err) {
 587                 health_error();
 588                 DBG("Live viewer listener thread exited with error");
 589         }
 590         health_unregister(health_relayd);
 591         DBG("Live viewer listener thread cleanup complete");
 592         stop_threads();
 593         return NULL;
 594 }
 595
 596 /*
 597  * This thread manages the dispatching of the requests to worker threads
 598  */
 599 static
 600 void *thread_dispatcher(void *data)
 601 {
 602         int err = -1;
 603         ssize_t ret;
 604         struct cds_wfq_node *node;
 605         struct relay_connection *conn = NULL;
 606
 607         DBG("[thread] Live viewer relay dispatcher started");
 608
 609         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 610
 611         if (testpoint(relayd_thread_live_dispatcher)) {
 612                 goto error_testpoint;
 613         }
 614
 615         health_code_update();
 616
 617         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 618                 health_code_update();
 619
 620                 /* Atomically prepare the queue futex */
 621                 futex_nto1_prepare(&viewer_conn_queue.futex);
 622
 623                 do {
 624                         health_code_update();
 625
 626                         /* Dequeue commands */
 627                         node = cds_wfq_dequeue_blocking(&viewer_conn_queue.queue);
 628                         if (node == NULL) {
 629                                 DBG("Woken up but nothing in the live-viewer "
 630                                                 "relay command queue");
 631                                 /* Continue thread execution */
 632                                 break;
 633                         }
 634                         conn = caa_container_of(node, struct relay_connection, qnode);
 635                         DBG("Dispatching viewer request waiting on sock %d",
 636                                         conn->sock->fd);
 637
 638                         /*
 639                          * Inform worker thread of the new request. This call is blocking
 640                          * so we can be assured that the data will be read at some point in
 641                          * time or wait to the end of the world :)
 642                          */
 643                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 644                         if (ret < 0) {
 645                                 PERROR("write conn pipe");
 646                                 connection_destroy(conn);
 647                                 goto error;
 648                         }
 649                 } while (node != NULL);
 650
 651                 /* Futex wait on queue. Blocking call on futex() */
 652                 health_poll_entry();
 653                 futex_nto1_wait(&viewer_conn_queue.futex);
 654                 health_poll_exit();
 655         }
 656
 657         /* Normal exit, no error */
 658         err = 0;
 659
 660 error:
 661 error_testpoint:
 662         if (err) {
 663                 health_error();
 664                 ERR("Health error occurred in %s", __func__);
 665         }
 666         health_unregister(health_relayd);
 667         DBG("Live viewer dispatch thread dying");
 668         stop_threads();
 669         return NULL;
 670 }
 671
 672 /*
 673  * Establish connection with the viewer and check the versions.
 674  *
 675  * Return 0 on success or else negative value.
 676  */
 677 static
 678 int viewer_connect(struct relay_connection *conn)
 679 {
 680         int ret;
 681         struct lttng_viewer_connect reply, msg;
 682
 683         assert(conn);
 684
 685         conn->version_check_done = 1;
 686
 687         health_code_update();
 688
 689         DBG("Viewer is establishing a connection to the relayd.");
 690
 691         ret = recv_request(conn->sock, &msg, sizeof(msg));
 692         if (ret < 0) {
 693                 goto end;
 694         }
 695
 696         health_code_update();
 697
 698         memset(&reply, 0, sizeof(reply));
 699         reply.major = RELAYD_VERSION_COMM_MAJOR;
 700         reply.minor = RELAYD_VERSION_COMM_MINOR;
 701
 702         /* Major versions must be the same */
 703         if (reply.major != be32toh(msg.major)) {
 704                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 705                                 reply.major, be32toh(msg.major));
 706                 ret = -1;
 707                 goto end;
 708         }
 709
 710         conn->major = reply.major;
 711         /* We adapt to the lowest compatible version */
 712         if (reply.minor <= be32toh(msg.minor)) {
 713                 conn->minor = reply.minor;
 714         } else {
 715                 conn->minor = be32toh(msg.minor);
 716         }
 717
 718         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 719                 conn->type = RELAY_VIEWER_COMMAND;
 720         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 721                 conn->type = RELAY_VIEWER_NOTIFICATION;
 722         } else {
 723                 ERR("Unknown connection type : %u", be32toh(msg.type));
 724                 ret = -1;
 725                 goto end;
 726         }
 727
 728         reply.major = htobe32(reply.major);
 729         reply.minor = htobe32(reply.minor);
 730         if (conn->type == RELAY_VIEWER_COMMAND) {
 731                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 732         }
 733
 734         health_code_update();
 735
 736         ret = send_response(conn->sock, &reply, sizeof(reply));
 737         if (ret < 0) {
 738                 goto end;
 739         }
 740
 741         health_code_update();
 742
 743         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 744         ret = 0;
 745
 746 end:
 747         return ret;
 748 }
 749
 750 /*
 751  * Send the viewer the list of current sessions.
 752  *
 753  * Return 0 on success or else a negative value.
 754  */
 755 static
 756 int viewer_list_sessions(struct relay_connection *conn)
 757 {
 758         int ret;
 759         struct lttng_viewer_list_sessions session_list;
 760         unsigned long count;
 761         long approx_before, approx_after;
 762         struct lttng_ht_iter iter;
 763         struct lttng_viewer_session send_session;
 764         struct relay_session *session;
 765
 766         DBG("List sessions received");
 767
 768         rcu_read_lock();
 769         cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
 770                         &approx_after);
 771         session_list.sessions_count = htobe32(count);
 772
 773         health_code_update();
 774
 775         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 776         if (ret < 0) {
 777                 goto end_unlock;
 778         }
 779
 780         health_code_update();
 781
 782         cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
 783                         session_n.node) {
 784                 health_code_update();
 785
 786                 strncpy(send_session.session_name, session->session_name,
 787                                 sizeof(send_session.session_name));
 788                 strncpy(send_session.hostname, session->hostname,
 789                                 sizeof(send_session.hostname));
 790                 send_session.id = htobe64(session->id);
 791                 send_session.live_timer = htobe32(session->live_timer);
 792                 send_session.clients = htobe32(session->viewer_refcount);
 793                 send_session.streams = htobe32(session->stream_count);
 794
 795                 health_code_update();
 796
 797                 ret = send_response(conn->sock, &send_session, sizeof(send_session));
 798                 if (ret < 0) {
 799                         goto end_unlock;
 800                 }
 801         }
 802         health_code_update();
 803
 804         rcu_read_unlock();
 805         ret = 0;
 806         goto end;
 807
 808 end_unlock:
 809         rcu_read_unlock();
 810
 811 end:
 812         return ret;
 813 }
 814
 815 /*
 816  * Check if a connection is attached to a session.
 817  * Return 1 if attached, 0 if not attached, a negative value on error.
 818  */
 819 static
 820 int session_attached(struct relay_connection *conn, uint64_t session_id)
 821 {
 822         struct relay_session *session;
 823         int found = 0;
 824
 825         if (!conn->viewer_session) {
 826                 goto end;
 827         }
 828         cds_list_for_each_entry(session,
 829                         &conn->viewer_session->sessions_head,
 830                         viewer_session_list) {
 831                 if (session->id == session_id) {
 832                         found = 1;
 833                         goto end;
 834                 }
 835         }
 836
 837 end:
 838         return found;
 839 }
 840
 841 /*
 842  * Delete all streams for a specific session ID.
 843  */
 844 static void destroy_viewer_streams_by_session(struct relay_session *session)
 845 {
 846         struct relay_viewer_stream *stream;
 847         struct lttng_ht_iter iter;
 848
 849         assert(session);
 850
 851         rcu_read_lock();
 852         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
 853                         stream_n.node) {
 854                 struct ctf_trace *ctf_trace;
 855
 856                 health_code_update();
 857                 if (stream->session_id != session->id) {
 858                         continue;
 859                 }
 860
 861                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 862                                 stream->path_name);
 863                 assert(ctf_trace);
 864
 865                 viewer_stream_delete(stream);
 866
 867                 if (stream->metadata_flag) {
 868                         ctf_trace->metadata_sent = 0;
 869                         ctf_trace->viewer_metadata_stream = NULL;
 870                 }
 871
 872                 viewer_stream_destroy(ctf_trace, stream);
 873         }
 874         rcu_read_unlock();
 875 }
 876
 877 static void try_destroy_streams(struct relay_session *session)
 878 {
 879         struct ctf_trace *ctf_trace;
 880         struct lttng_ht_iter iter;
 881
 882         assert(session);
 883
 884         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 885                         node.node) {
 886                 /* Attempt to destroy the ctf trace of that session. */
 887                 ctf_trace_try_destroy(session, ctf_trace);
 888         }
 889 }
 890
 891 /*
 892  * Cleanup a session.
 893  */
 894 static void cleanup_session(struct relay_connection *conn,
 895                 struct relay_session *session)
 896 {
 897         /*
 898          * Very important that this is done before destroying the session so we
 899          * can put back every viewer stream reference from the ctf_trace.
 900          */
 901         destroy_viewer_streams_by_session(session);
 902         try_destroy_streams(session);
 903         cds_list_del(&session->viewer_session_list);
 904         session_viewer_try_destroy(conn->sessions_ht, session);
 905 }
 906
 907 /*
 908  * Send the viewer the list of current sessions.
 909  */
 910 static
 911 int viewer_get_new_streams(struct relay_connection *conn)
 912 {
 913         int ret, send_streams = 0;
 914         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
 915         struct lttng_viewer_new_streams_request request;
 916         struct lttng_viewer_new_streams_response response;
 917         struct relay_session *session;
 918         uint64_t session_id;
 919
 920         assert(conn);
 921
 922         DBG("Get new streams received");
 923
 924         health_code_update();
 925
 926         /* Receive the request from the connected client. */
 927         ret = recv_request(conn->sock, &request, sizeof(request));
 928         if (ret < 0) {
 929                 goto error;
 930         }
 931         session_id = be64toh(request.session_id);
 932
 933         health_code_update();
 934
 935         memset(&response, 0, sizeof(response));
 936
 937         rcu_read_lock();
 938         session = session_find_by_id(conn->sessions_ht, session_id);
 939         if (!session) {
 940                 DBG("Relay session %" PRIu64 " not found", session_id);
 941                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 942                 goto send_reply;
 943         }
 944
 945         if (!session_attached(conn, session_id)) {
 946                 send_streams = 0;
 947                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 948                 goto send_reply;
 949         }
 950
 951         send_streams = 1;
 952         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 953
 954         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
 955                         &nb_created);
 956         if (ret < 0) {
 957                 goto end_unlock;
 958         }
 959         /* Only send back the newly created streams with the unsent ones. */
 960         nb_streams = nb_created + nb_unsent;
 961         response.streams_count = htobe32(nb_streams);
 962
 963         /*
 964          * If the session is closed and we have no new streams to send,
 965          * it means that the viewer has already received the whole trace
 966          * for this session and should now close it.
 967          */
 968         if (nb_streams == 0 && session->close_flag) {
 969                 send_streams = 0;
 970                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 971                 /*
 972                  * Remove the session from the attached list of the connection
 973                  * and try to destroy it.
 974                  */
 975                 cds_list_del(&session->viewer_session_list);
 976                 cleanup_session(conn, session);
 977                 goto send_reply;
 978         }
 979
 980 send_reply:
 981         health_code_update();
 982         ret = send_response(conn->sock, &response, sizeof(response));
 983         if (ret < 0) {
 984                 goto end_unlock;
 985         }
 986         health_code_update();
 987
 988         /*
 989          * Unknown or empty session, just return gracefully, the viewer knows what
 990          * is happening.
 991          */
 992         if (!send_streams || !nb_streams) {
 993                 ret = 0;
 994                 goto end_unlock;
 995         }
 996
 997         /*
 998          * Send stream and *DON'T* ignore the sent flag so every viewer streams
 999          * that were not sent from that point will be sent to the viewer.
1000          */
1001         ret = send_viewer_streams(conn->sock, session, 0);
1002         if (ret < 0) {
1003                 goto end_unlock;
1004         }
1005
1006 end_unlock:
1007         rcu_read_unlock();
1008 error:
1009         return ret;
1010 }
1011
1012 /*
1013  * Send the viewer the list of current sessions.
1014  */
1015 static
1016 int viewer_attach_session(struct relay_connection *conn)
1017 {
1018         int send_streams = 0;
1019         ssize_t ret;
1020         uint32_t nb_streams = 0;
1021         enum lttng_viewer_seek seek_type;
1022         struct lttng_viewer_attach_session_request request;
1023         struct lttng_viewer_attach_session_response response;
1024         struct relay_session *session;
1025
1026         assert(conn);
1027
1028         health_code_update();
1029
1030         /* Receive the request from the connected client. */
1031         ret = recv_request(conn->sock, &request, sizeof(request));
1032         if (ret < 0) {
1033                 goto error;
1034         }
1035
1036         health_code_update();
1037
1038         memset(&response, 0, sizeof(response));
1039
1040         if (!conn->viewer_session) {
1041                 DBG("Client trying to attach before creating a live viewer session");
1042                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1043                 goto send_reply;
1044         }
1045
1046         rcu_read_lock();
1047         session = session_find_by_id(conn->sessions_ht,
1048                         be64toh(request.session_id));
1049         if (!session) {
1050                 DBG("Relay session %" PRIu64 " not found",
1051                                 be64toh(request.session_id));
1052                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1053                 goto send_reply;
1054         }
1055         session_viewer_attach(session);
1056         DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1057
1058         if (uatomic_read(&session->viewer_refcount) > 1) {
1059                 DBG("Already a viewer attached");
1060                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1061                 session_viewer_detach(session);
1062                 goto send_reply;
1063         } else if (session->live_timer == 0) {
1064                 DBG("Not live session");
1065                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1066                 goto send_reply;
1067         } else {
1068                 send_streams = 1;
1069                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1070                 cds_list_add(&session->viewer_session_list,
1071                                 &conn->viewer_session->sessions_head);
1072         }
1073
1074         switch (be32toh(request.seek)) {
1075         case LTTNG_VIEWER_SEEK_BEGINNING:
1076         case LTTNG_VIEWER_SEEK_LAST:
1077                 seek_type = be32toh(request.seek);
1078                 break;
1079         default:
1080                 ERR("Wrong seek parameter");
1081                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1082                 send_streams = 0;
1083                 goto send_reply;
1084         }
1085
1086         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1087         if (ret < 0) {
1088                 goto end_unlock;
1089         }
1090         response.streams_count = htobe32(nb_streams);
1091
1092 send_reply:
1093         health_code_update();
1094         ret = send_response(conn->sock, &response, sizeof(response));
1095         if (ret < 0) {
1096                 goto end_unlock;
1097         }
1098         health_code_update();
1099
1100         /*
1101          * Unknown or empty session, just return gracefully, the viewer knows what
1102          * is happening.
1103          */
1104         if (!send_streams || !nb_streams) {
1105                 ret = 0;
1106                 goto end_unlock;
1107         }
1108
1109         /* Send stream and ignore the sent flag. */
1110         ret = send_viewer_streams(conn->sock, session, 1);
1111         if (ret < 0) {
1112                 goto end_unlock;
1113         }
1114
1115 end_unlock:
1116         rcu_read_unlock();
1117 error:
1118         return ret;
1119 }
1120
1121 /*
1122  * Open the index file if needed for the given vstream.
1123  *
1124  * If an index file is successfully opened, the index_read_fd of the stream is
1125  * set with it.
1126  *
1127  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1128  */
1129 static int try_open_index(struct relay_viewer_stream *vstream,
1130                 struct relay_stream *rstream)
1131 {
1132         int ret = 0;
1133
1134         assert(vstream);
1135         assert(rstream);
1136
1137         if (vstream->index_read_fd >= 0) {
1138                 goto end;
1139         }
1140
1141         /*
1142          * First time, we open the index file and at least one index is ready.  The
1143          * race between the read and write of the total_index_received is
1144          * acceptable here since the client will be notified to simply come back
1145          * and get the next index.
1146          */
1147         if (rstream->total_index_received <= 0) {
1148                 ret = -ENOENT;
1149                 goto end;
1150         }
1151         ret = index_open(vstream->path_name, vstream->channel_name,
1152                         vstream->tracefile_count, vstream->tracefile_count_current);
1153         if (ret >= 0) {
1154                 vstream->index_read_fd = ret;
1155                 ret = 0;
1156                 goto end;
1157         }
1158
1159 end:
1160         return ret;
1161 }
1162
1163 /*
1164  * Check the status of the index for the given stream. This function updates
1165  * the index structure if needed and can destroy the vstream also for the HUP
1166  * situation.
1167  *
1168  * Return 0 means that we can proceed with the index. A value of 1 means that
1169  * the index has been updated and is ready to be send to the client. A negative
1170  * value indicates an error that can't be handled.
1171  */
1172 static int check_index_status(struct relay_viewer_stream *vstream,
1173                 struct relay_stream *rstream, struct ctf_trace *trace,
1174                 struct lttng_viewer_index *index)
1175 {
1176         int ret;
1177
1178         assert(vstream);
1179         assert(rstream);
1180         assert(index);
1181         assert(trace);
1182
1183         if (!rstream->close_flag) {
1184                 /* Rotate on abort (overwrite). */
1185                 if (vstream->abort_flag) {
1186                         DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1187                                         vstream->stream_handle);
1188                         ret = viewer_stream_rotate(vstream, rstream);
1189                         if (ret < 0) {
1190                                 goto error;
1191                         } else if (ret == 1) {
1192                                 /* EOF */
1193                                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1194                                 goto hup;
1195                         }
1196                         /* ret == 0 means successful so we continue. */
1197                 }
1198
1199                 /* Check if we are in the same trace file at this point. */
1200                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1201                         if (rstream->beacon_ts_end != -1ULL &&
1202                                         vstream->last_sent_index == rstream->total_index_received) {
1203                                 /*
1204                                  * We've received a synchronization beacon and the last index
1205                                  * available has been sent, the index for now is inactive.
1206                                  */
1207                                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1208                                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1209                                 index->stream_id = htobe64(rstream->ctf_stream_id);
1210                                 goto index_ready;
1211                         } else if (rstream->total_index_received <= vstream->last_sent_index
1212                                         && !vstream->close_write_flag) {
1213                                 /*
1214                                  * Reader and writer are working in the same tracefile, so we care
1215                                  * about the number of index received and sent. Otherwise, we read
1216                                  * up to EOF.
1217                                  */
1218                                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1219                                 goto index_ready;
1220                         }
1221                 }
1222                 /* Nothing to do with the index, continue with it. */
1223                 ret = 0;
1224         } else if (rstream->close_flag && vstream->close_write_flag &&
1225                         vstream->total_index_received == vstream->last_sent_index) {
1226                 /* Last index sent and current tracefile closed in write */
1227                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1228                 goto hup;
1229         } else {
1230                 vstream->close_write_flag = 1;
1231                 ret = 0;
1232         }
1233
1234 error:
1235         return ret;
1236
1237 hup:
1238         viewer_stream_delete(vstream);
1239         viewer_stream_destroy(trace, vstream);
1240 index_ready:
1241         return 1;
1242 }
1243
1244 /*
1245  * Send the next index for a stream.
1246  *
1247  * Return 0 on success or else a negative value.
1248  */
1249 static
1250 int viewer_get_next_index(struct relay_connection *conn)
1251 {
1252         int ret;
1253         ssize_t read_ret;
1254         struct lttng_viewer_get_next_index request_index;
1255         struct lttng_viewer_index viewer_index;
1256         struct ctf_packet_index packet_index;
1257         struct relay_viewer_stream *vstream;
1258         struct relay_stream *rstream;
1259         struct ctf_trace *ctf_trace;
1260         struct relay_session *session;
1261
1262         assert(conn);
1263
1264         DBG("Viewer get next index");
1265
1266         health_code_update();
1267
1268         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1269         if (ret < 0) {
1270                 goto end;
1271         }
1272         health_code_update();
1273
1274         rcu_read_lock();
1275         vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1276         if (!vstream) {
1277                 ret = -1;
1278                 goto end_unlock;
1279         }
1280
1281         session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1282         if (!session) {
1283                 ret = -1;
1284                 goto end_unlock;
1285         }
1286
1287         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1288         assert(ctf_trace);
1289
1290         memset(&viewer_index, 0, sizeof(viewer_index));
1291
1292         /*
1293          * The viewer should not ask for index on metadata stream.
1294          */
1295         if (vstream->metadata_flag) {
1296                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1297                 goto send_reply;
1298         }
1299
1300         rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1301         assert(rstream);
1302
1303         /* Try to open an index if one is needed for that stream. */
1304         ret = try_open_index(vstream, rstream);
1305         if (ret < 0) {
1306                 if (ret == -ENOENT) {
1307                         /*
1308                          * The index is created only when the first data packet arrives, it
1309                          * might not be ready at the beginning of the session
1310                          */
1311                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1312                 } else {
1313                         /* Unhandled error. */
1314                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1315                 }
1316                 goto send_reply;
1317         }
1318
1319         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1320         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1321         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1322         if (ret < 0) {
1323                 goto end;
1324         } else if (ret == 1) {
1325                 /*
1326                  * This means the viewer index data structure has been populated by the
1327                  * check call thus we now send back the reply to the client.
1328                  */
1329                 goto send_reply;
1330         }
1331         /* At this point, ret MUST be 0 thus we continue with the get. */
1332         assert(!ret);
1333
1334         if (!ctf_trace->metadata_received ||
1335                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1336                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1337         }
1338
1339         ret = check_new_streams(conn);
1340         if (ret < 0) {
1341                 goto end_unlock;
1342         } else if (ret == 1) {
1343                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1344         }
1345
1346         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1347         pthread_mutex_lock(&vstream->overwrite_lock);
1348         if (vstream->abort_flag) {
1349                 /* The file is being overwritten by the writer, we cannot use it. */
1350                 pthread_mutex_unlock(&vstream->overwrite_lock);
1351                 ret = viewer_stream_rotate(vstream, rstream);
1352                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1353                 if (ret < 0) {
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                 goto send_reply;
1363         }
1364
1365         read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1366                         sizeof(packet_index));
1367         pthread_mutex_unlock(&vstream->overwrite_lock);
1368         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1369         if (read_ret < 0) {
1370                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1371                 viewer_stream_delete(vstream);
1372                 viewer_stream_destroy(ctf_trace, vstream);
1373                 goto send_reply;
1374         } else if (read_ret < sizeof(packet_index)) {
1375                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1376                 if (vstream->close_write_flag) {
1377                         ret = viewer_stream_rotate(vstream, rstream);
1378                         if (ret < 0) {
1379                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1380                                 goto end_unlock;
1381                         } else if (ret == 1) {
1382                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1383                                 viewer_stream_delete(vstream);
1384                                 viewer_stream_destroy(ctf_trace, vstream);
1385                         } else {
1386                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1387                         }
1388                 } else {
1389                         ERR("Relay reading index file %d", vstream->index_read_fd);
1390                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1391                 }
1392                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1393                 goto send_reply;
1394         } else {
1395                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1396                 vstream->last_sent_index++;
1397         }
1398
1399         /*
1400          * Indexes are stored in big endian, no need to switch before sending.
1401          */
1402         viewer_index.offset = packet_index.offset;
1403         viewer_index.packet_size = packet_index.packet_size;
1404         viewer_index.content_size = packet_index.content_size;
1405         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1406         viewer_index.timestamp_end = packet_index.timestamp_end;
1407         viewer_index.events_discarded = packet_index.events_discarded;
1408         viewer_index.stream_id = packet_index.stream_id;
1409
1410 send_reply:
1411         viewer_index.flags = htobe32(viewer_index.flags);
1412         health_code_update();
1413
1414         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1415         if (ret < 0) {
1416                 goto end_unlock;
1417         }
1418         health_code_update();
1419
1420         DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1421                         vstream->last_sent_index, vstream->stream_handle);
1422
1423 end_unlock:
1424         rcu_read_unlock();
1425
1426 end:
1427         return ret;
1428 }
1429
1430 /*
1431  * Send the next index for a stream
1432  *
1433  * Return 0 on success or else a negative value.
1434  */
1435 static
1436 int viewer_get_packet(struct relay_connection *conn)
1437 {
1438         int ret, send_data = 0;
1439         char *data = NULL;
1440         uint32_t len = 0;
1441         ssize_t read_len;
1442         struct lttng_viewer_get_packet get_packet_info;
1443         struct lttng_viewer_trace_packet reply;
1444         struct relay_viewer_stream *stream;
1445         struct relay_session *session;
1446         struct ctf_trace *ctf_trace;
1447
1448         assert(conn);
1449
1450         DBG2("Relay get data packet");
1451
1452         health_code_update();
1453
1454         ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1455         if (ret < 0) {
1456                 goto end;
1457         }
1458         health_code_update();
1459
1460         /* From this point on, the error label can be reached. */
1461         memset(&reply, 0, sizeof(reply));
1462
1463         rcu_read_lock();
1464         stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1465         if (!stream) {
1466                 goto error;
1467         }
1468
1469         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1470         if (!session) {
1471                 ret = -1;
1472                 goto error;
1473         }
1474
1475         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1476                         stream->path_name);
1477         assert(ctf_trace);
1478
1479         /*
1480          * First time we read this stream, we need open the tracefile, we should
1481          * only arrive here if an index has already been sent to the viewer, so the
1482          * tracefile must exist, if it does not it is a fatal error.
1483          */
1484         if (stream->read_fd < 0) {
1485                 char fullpath[PATH_MAX];
1486
1487                 if (stream->tracefile_count > 0) {
1488                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1489                                         stream->channel_name,
1490                                         stream->tracefile_count_current);
1491                 } else {
1492                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1493                                         stream->channel_name);
1494                 }
1495                 if (ret < 0) {
1496                         goto error;
1497                 }
1498                 ret = open(fullpath, O_RDONLY);
1499                 if (ret < 0) {
1500                         PERROR("Relay opening trace file");
1501                         goto error;
1502                 }
1503                 stream->read_fd = ret;
1504         }
1505
1506         if (!ctf_trace->metadata_received ||
1507                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1508                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1509                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1510                 goto send_reply;
1511         }
1512
1513         ret = check_new_streams(conn);
1514         if (ret < 0) {
1515                 goto end_unlock;
1516         } else if (ret == 1) {
1517                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1518                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1519                 goto send_reply;
1520         }
1521
1522         len = be32toh(get_packet_info.len);
1523         data = zmalloc(len);
1524         if (!data) {
1525                 PERROR("relay data zmalloc");
1526                 goto error;
1527         }
1528
1529         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1530         if (ret < 0) {
1531                 /*
1532                  * If the read fd was closed by the streaming side, the
1533                  * abort_flag will be set to 1, otherwise it is an error.
1534                  */
1535                 if (stream->abort_flag == 0) {
1536                         PERROR("lseek");
1537                         goto error;
1538                 }
1539                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1540                 goto send_reply;
1541         }
1542         read_len = lttng_read(stream->read_fd, data, len);
1543         if (read_len < len) {
1544                 /*
1545                  * If the read fd was closed by the streaming side, the
1546                  * abort_flag will be set to 1, otherwise it is an error.
1547                  */
1548                 if (stream->abort_flag == 0) {
1549                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1550                                         stream->read_fd,
1551                                         be64toh(get_packet_info.offset));
1552                         goto error;
1553                 } else {
1554                         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1555                         goto send_reply;
1556                 }
1557         }
1558         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1559         reply.len = htobe32(len);
1560         send_data = 1;
1561         goto send_reply;
1562
1563 error:
1564         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1565
1566 send_reply:
1567         reply.flags = htobe32(reply.flags);
1568
1569         health_code_update();
1570
1571         ret = send_response(conn->sock, &reply, sizeof(reply));
1572         if (ret < 0) {
1573                 goto end_unlock;
1574         }
1575         health_code_update();
1576
1577         if (send_data) {
1578                 health_code_update();
1579                 ret = send_response(conn->sock, data, len);
1580                 if (ret < 0) {
1581                         goto end_unlock;
1582                 }
1583                 health_code_update();
1584         }
1585
1586         DBG("Sent %u bytes for stream %" PRIu64, len,
1587                         be64toh(get_packet_info.stream_id));
1588
1589 end_unlock:
1590         free(data);
1591         rcu_read_unlock();
1592
1593 end:
1594         return ret;
1595 }
1596
1597 /*
1598  * Send the session's metadata
1599  *
1600  * Return 0 on success else a negative value.
1601  */
1602 static
1603 int viewer_get_metadata(struct relay_connection *conn)
1604 {
1605         int ret = 0;
1606         ssize_t read_len;
1607         uint64_t len = 0;
1608         char *data = NULL;
1609         struct lttng_viewer_get_metadata request;
1610         struct lttng_viewer_metadata_packet reply;
1611         struct relay_viewer_stream *stream;
1612         struct ctf_trace *ctf_trace;
1613         struct relay_session *session;
1614
1615         assert(conn);
1616
1617         DBG("Relay get metadata");
1618
1619         health_code_update();
1620
1621         ret = recv_request(conn->sock, &request, sizeof(request));
1622         if (ret < 0) {
1623                 goto end;
1624         }
1625         health_code_update();
1626
1627         memset(&reply, 0, sizeof(reply));
1628
1629         rcu_read_lock();
1630         stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1631         if (!stream || !stream->metadata_flag) {
1632                 ERR("Invalid metadata stream");
1633                 goto error;
1634         }
1635
1636         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1637         if (!session) {
1638                 ret = -1;
1639                 goto error;
1640         }
1641
1642         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1643                         stream->path_name);
1644         assert(ctf_trace);
1645         assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1646
1647         len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1648         if (len == 0) {
1649                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1650                 goto send_reply;
1651         }
1652
1653         /* first time, we open the metadata file */
1654         if (stream->read_fd < 0) {
1655                 char fullpath[PATH_MAX];
1656
1657                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1658                                 stream->channel_name);
1659                 if (ret < 0) {
1660                         goto error;
1661                 }
1662                 ret = open(fullpath, O_RDONLY);
1663                 if (ret < 0) {
1664                         PERROR("Relay opening metadata file");
1665                         goto error;
1666                 }
1667                 stream->read_fd = ret;
1668         }
1669
1670         reply.len = htobe64(len);
1671         data = zmalloc(len);
1672         if (!data) {
1673                 PERROR("viewer metadata zmalloc");
1674                 goto error;
1675         }
1676
1677         read_len = lttng_read(stream->read_fd, data, len);
1678         if (read_len < len) {
1679                 PERROR("Relay reading metadata file");
1680                 goto error;
1681         }
1682         ctf_trace->metadata_sent += read_len;
1683         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1684         goto send_reply;
1685
1686 error:
1687         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1688
1689 send_reply:
1690         health_code_update();
1691         ret = send_response(conn->sock, &reply, sizeof(reply));
1692         if (ret < 0) {
1693                 goto end_unlock;
1694         }
1695         health_code_update();
1696
1697         if (len > 0) {
1698                 ret = send_response(conn->sock, data, len);
1699                 if (ret < 0) {
1700                         goto end_unlock;
1701                 }
1702         }
1703
1704         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1705                         be64toh(request.stream_id));
1706
1707         DBG("Metadata sent");
1708
1709 end_unlock:
1710         free(data);
1711         rcu_read_unlock();
1712 end:
1713         return ret;
1714 }
1715
1716 /*
1717  * Create a viewer session.
1718  *
1719  * Return 0 on success or else a negative value.
1720  */
1721 static
1722 int viewer_create_session(struct relay_connection *conn)
1723 {
1724         int ret;
1725         struct lttng_viewer_create_session_response resp;
1726
1727         DBG("Viewer create session received");
1728
1729         memset(&resp, 0, sizeof(resp));
1730         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1731         conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1732         if (!conn->viewer_session) {
1733                 ERR("Allocation viewer session");
1734                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1735                 goto send_reply;
1736         }
1737         CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1738
1739 send_reply:
1740         health_code_update();
1741         ret = send_response(conn->sock, &resp, sizeof(resp));
1742         if (ret < 0) {
1743                 goto end;
1744         }
1745         health_code_update();
1746         ret = 0;
1747
1748 end:
1749         return ret;
1750 }
1751
1752
1753 /*
1754  * live_relay_unknown_command: send -1 if received unknown command
1755  */
1756 static
1757 void live_relay_unknown_command(struct relay_connection *conn)
1758 {
1759         struct lttcomm_relayd_generic_reply reply;
1760
1761         memset(&reply, 0, sizeof(reply));
1762         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1763         (void) send_response(conn->sock, &reply, sizeof(reply));
1764 }
1765
1766 /*
1767  * Process the commands received on the control socket
1768  */
1769 static
1770 int process_control(struct lttng_viewer_cmd *recv_hdr,
1771                 struct relay_connection *conn)
1772 {
1773         int ret = 0;
1774         uint32_t msg_value;
1775
1776         assert(recv_hdr);
1777         assert(conn);
1778
1779         msg_value = be32toh(recv_hdr->cmd);
1780
1781         /*
1782          * Make sure we've done the version check before any command other then a
1783          * new client connection.
1784          */
1785         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1786                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1787                 ret = -1;
1788                 goto end;
1789         }
1790
1791         switch (msg_value) {
1792         case LTTNG_VIEWER_CONNECT:
1793                 ret = viewer_connect(conn);
1794                 break;
1795         case LTTNG_VIEWER_LIST_SESSIONS:
1796                 ret = viewer_list_sessions(conn);
1797                 break;
1798         case LTTNG_VIEWER_ATTACH_SESSION:
1799                 ret = viewer_attach_session(conn);
1800                 break;
1801         case LTTNG_VIEWER_GET_NEXT_INDEX:
1802                 ret = viewer_get_next_index(conn);
1803                 break;
1804         case LTTNG_VIEWER_GET_PACKET:
1805                 ret = viewer_get_packet(conn);
1806                 break;
1807         case LTTNG_VIEWER_GET_METADATA:
1808                 ret = viewer_get_metadata(conn);
1809                 break;
1810         case LTTNG_VIEWER_GET_NEW_STREAMS:
1811                 ret = viewer_get_new_streams(conn);
1812                 break;
1813         case LTTNG_VIEWER_CREATE_SESSION:
1814                 ret = viewer_create_session(conn);
1815                 break;
1816         default:
1817                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1818                 live_relay_unknown_command(conn);
1819                 ret = -1;
1820                 goto end;
1821         }
1822
1823 end:
1824         return ret;
1825 }
1826
1827 static
1828 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1829 {
1830         int ret;
1831
1832         assert(events);
1833
1834         (void) lttng_poll_del(events, pollfd);
1835
1836         ret = close(pollfd);
1837         if (ret < 0) {
1838                 ERR("Closing pollfd %d", pollfd);
1839         }
1840 }
1841
1842 /*
1843  * Delete and destroy a connection.
1844  *
1845  * RCU read side lock MUST be acquired.
1846  */
1847 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1848                 struct relay_connection *conn)
1849 {
1850         struct relay_session *session, *tmp_session;
1851
1852         assert(relay_connections_ht);
1853         assert(conn);
1854
1855         connection_delete(relay_connections_ht, conn);
1856
1857         if (!conn->viewer_session) {
1858                 goto end;
1859         }
1860
1861         rcu_read_lock();
1862         cds_list_for_each_entry_safe(session, tmp_session,
1863                         &conn->viewer_session->sessions_head,
1864                         viewer_session_list) {
1865                 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1866                 cleanup_session(conn, session);
1867         }
1868         rcu_read_unlock();
1869
1870 end:
1871         connection_destroy(conn);
1872 }
1873
1874 /*
1875  * This thread does the actual work
1876  */
1877 static
1878 void *thread_worker(void *data)
1879 {
1880         int ret, err = -1;
1881         uint32_t nb_fd;
1882         struct relay_connection *conn;
1883         struct lttng_poll_event events;
1884         struct lttng_ht *relay_connections_ht;
1885         struct lttng_ht_iter iter;
1886         struct lttng_viewer_cmd recv_hdr;
1887         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1888         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1889
1890         DBG("[thread] Live viewer relay worker started");
1891
1892         rcu_register_thread();
1893
1894         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1895
1896         if (testpoint(relayd_thread_live_worker)) {
1897                 goto error_testpoint;
1898         }
1899
1900         /* table of connections indexed on socket */
1901         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1902         if (!relay_connections_ht) {
1903                 goto relay_connections_ht_error;
1904         }
1905
1906         ret = create_thread_poll_set(&events, 2);
1907         if (ret < 0) {
1908                 goto error_poll_create;
1909         }
1910
1911         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1912         if (ret < 0) {
1913                 goto error;
1914         }
1915
1916 restart:
1917         while (1) {
1918                 int i;
1919
1920                 health_code_update();
1921
1922                 /* Infinite blocking call, waiting for transmission */
1923                 DBG3("Relayd live viewer worker thread polling...");
1924                 health_poll_entry();
1925                 ret = lttng_poll_wait(&events, -1);
1926                 health_poll_exit();
1927                 if (ret < 0) {
1928                         /*
1929                          * Restart interrupted system call.
1930                          */
1931                         if (errno == EINTR) {
1932                                 goto restart;
1933                         }
1934                         goto error;
1935                 }
1936
1937                 nb_fd = ret;
1938
1939                 /*
1940                  * Process control. The control connection is prioritised so we don't
1941                  * starve it with high throughput tracing data on the data
1942                  * connection.
1943                  */
1944                 for (i = 0; i < nb_fd; i++) {
1945                         /* Fetch once the poll data */
1946                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1947                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1948
1949                         health_code_update();
1950
1951                         /* Thread quit pipe has been closed. Killing thread. */
1952                         ret = check_thread_quit_pipe(pollfd, revents);
1953                         if (ret) {
1954                                 err = 0;
1955                                 goto exit;
1956                         }
1957
1958                         /* Inspect the relay conn pipe for new connection */
1959                         if (pollfd == live_conn_pipe[0]) {
1960                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1961                                         ERR("Relay live pipe error");
1962                                         goto error;
1963                                 } else if (revents & LPOLLIN) {
1964                                         ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1965                                         if (ret < 0) {
1966                                                 goto error;
1967                                         }
1968                                         conn->sessions_ht = sessions_ht;
1969                                         connection_init(conn);
1970                                         lttng_poll_add(&events, conn->sock->fd,
1971                                                         LPOLLIN | LPOLLRDHUP);
1972                                         rcu_read_lock();
1973                                         lttng_ht_add_unique_ulong(relay_connections_ht,
1974                                                         &conn->sock_n);
1975                                         rcu_read_unlock();
1976                                         DBG("Connection socket %d added", conn->sock->fd);
1977                                 }
1978                         } else {
1979                                 rcu_read_lock();
1980                                 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1981                                 /* If not found, there is a synchronization issue. */
1982                                 assert(conn);
1983
1984                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1985                                         cleanup_connection_pollfd(&events, pollfd);
1986                                         destroy_connection(relay_connections_ht, conn);
1987                                 } else if (revents & LPOLLIN) {
1988                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1989                                                         sizeof(recv_hdr), 0);
1990                                         if (ret <= 0) {
1991                                                 /* Connection closed */
1992                                                 cleanup_connection_pollfd(&events, pollfd);
1993                                                 destroy_connection(relay_connections_ht, conn);
1994                                                 DBG("Viewer control conn closed with %d", pollfd);
1995                                         } else {
1996                                                 ret = process_control(&recv_hdr, conn);
1997                                                 if (ret < 0) {
1998                                                         /* Clear the session on error. */
1999                                                         cleanup_connection_pollfd(&events, pollfd);
2000                                                         destroy_connection(relay_connections_ht, conn);
2001                                                         DBG("Viewer connection closed with %d", pollfd);
2002                                                 }
2003                                         }
2004                                 }
2005                                 rcu_read_unlock();
2006                         }
2007                 }
2008         }
2009
2010 exit:
2011 error:
2012         lttng_poll_clean(&events);
2013
2014         /* Cleanup reamaining connection object. */
2015         rcu_read_lock();
2016         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2017                         sock_n.node) {
2018                 health_code_update();
2019                 destroy_connection(relay_connections_ht, conn);
2020         }
2021         rcu_read_unlock();
2022 error_poll_create:
2023         lttng_ht_destroy(relay_connections_ht);
2024 relay_connections_ht_error:
2025         /* Close relay conn pipes */
2026         utils_close_pipe(live_conn_pipe);
2027         if (err) {
2028                 DBG("Viewer worker thread exited with error");
2029         }
2030         DBG("Viewer worker thread cleanup complete");
2031 error_testpoint:
2032         if (err) {
2033                 health_error();
2034                 ERR("Health error occurred in %s", __func__);
2035         }
2036         health_unregister(health_relayd);
2037         stop_threads();
2038         rcu_unregister_thread();
2039         return NULL;
2040 }
2041
2042 /*
2043  * Create the relay command pipe to wake thread_manage_apps.
2044  * Closed in cleanup().
2045  */
2046 static int create_conn_pipe(void)
2047 {
2048         int ret;
2049
2050         ret = utils_create_pipe_cloexec(live_conn_pipe);
2051
2052         return ret;
2053 }
2054
2055 void live_stop_threads(void)
2056 {
2057         int ret;
2058         void *status;
2059
2060         stop_threads();
2061
2062         ret = pthread_join(live_listener_thread, &status);
2063         if (ret != 0) {
2064                 PERROR("pthread_join live listener");
2065                 goto error;     /* join error, exit without cleanup */
2066         }
2067
2068         ret = pthread_join(live_worker_thread, &status);
2069         if (ret != 0) {
2070                 PERROR("pthread_join live worker");
2071                 goto error;     /* join error, exit without cleanup */
2072         }
2073
2074         ret = pthread_join(live_dispatcher_thread, &status);
2075         if (ret != 0) {
2076                 PERROR("pthread_join live dispatcher");
2077                 goto error;     /* join error, exit without cleanup */
2078         }
2079
2080         cleanup();
2081
2082 error:
2083         return;
2084 }
2085
2086 /*
2087  * main
2088  */
2089 int live_start_threads(struct lttng_uri *uri,
2090                 struct relay_local_data *relay_ctx)
2091 {
2092         int ret = 0;
2093         void *status;
2094         int is_root;
2095
2096         assert(uri);
2097         live_uri = uri;
2098
2099         /* Check if daemon is UID = 0 */
2100         is_root = !getuid();
2101
2102         if (!is_root) {
2103                 if (live_uri->port < 1024) {
2104                         ERR("Need to be root to use ports < 1024");
2105                         ret = -1;
2106                         goto exit;
2107                 }
2108         }
2109
2110         /* Setup the thread apps communication pipe. */
2111         if ((ret = create_conn_pipe()) < 0) {
2112                 goto exit;
2113         }
2114
2115         /* Init relay command queue. */
2116         cds_wfq_init(&viewer_conn_queue.queue);
2117
2118         /* Set up max poll set size */
2119         lttng_poll_set_max_size();
2120
2121         /* Setup the dispatcher thread */
2122         ret = pthread_create(&live_dispatcher_thread, NULL,
2123                         thread_dispatcher, (void *) NULL);
2124         if (ret != 0) {
2125                 PERROR("pthread_create viewer dispatcher");
2126                 goto exit_dispatcher;
2127         }
2128
2129         /* Setup the worker thread */
2130         ret = pthread_create(&live_worker_thread, NULL,
2131                         thread_worker, relay_ctx);
2132         if (ret != 0) {
2133                 PERROR("pthread_create viewer worker");
2134                 goto exit_worker;
2135         }
2136
2137         /* Setup the listener thread */
2138         ret = pthread_create(&live_listener_thread, NULL,
2139                         thread_listener, (void *) NULL);
2140         if (ret != 0) {
2141                 PERROR("pthread_create viewer listener");
2142                 goto exit_listener;
2143         }
2144
2145         ret = 0;
2146         goto end;
2147
2148 exit_listener:
2149         ret = pthread_join(live_listener_thread, &status);
2150         if (ret != 0) {
2151                 PERROR("pthread_join live listener");
2152                 goto error;     /* join error, exit without cleanup */
2153         }
2154
2155 exit_worker:
2156         ret = pthread_join(live_worker_thread, &status);
2157         if (ret != 0) {
2158                 PERROR("pthread_join live worker");
2159                 goto error;     /* join error, exit without cleanup */
2160         }
2161
2162 exit_dispatcher:
2163         ret = pthread_join(live_dispatcher_thread, &status);
2164         if (ret != 0) {
2165                 PERROR("pthread_join live dispatcher");
2166                 goto error;     /* join error, exit without cleanup */
2167         }
2168
2169 exit:
2170         cleanup();
2171
2172 end:
2173 error:
2174         return ret;
2175 }