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_relayd_live(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 int stop_threads(void)
 349 {
 350         int ret, retval = 0;
 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                 retval = -1;
 358         }
 359
 360         /* Dispatch thread */
 361         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 362         futex_nto1_wake(&viewer_conn_queue.futex);
 363         return retval;
 364 }
 365
 366 /*
 367  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 368  */
 369 static
 370 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 371 {
 372         int ret;
 373
 374         if (events == NULL || size == 0) {
 375                 ret = -1;
 376                 goto error;
 377         }
 378
 379         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 380         if (ret < 0) {
 381                 goto error;
 382         }
 383
 384         /* Add quit pipe */
 385         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 386         if (ret < 0) {
 387                 goto error;
 388         }
 389
 390         return 0;
 391
 392 error:
 393         return ret;
 394 }
 395
 396 /*
 397  * Check if the thread quit pipe was triggered.
 398  *
 399  * Return 1 if it was triggered else 0;
 400  */
 401 static
 402 int check_thread_quit_pipe(int fd, uint32_t events)
 403 {
 404         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 405                 return 1;
 406         }
 407
 408         return 0;
 409 }
 410
 411 /*
 412  * Create and init socket from uri.
 413  */
 414 static
 415 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 416 {
 417         int ret;
 418         struct lttcomm_sock *sock = NULL;
 419
 420         sock = lttcomm_alloc_sock_from_uri(uri);
 421         if (sock == NULL) {
 422                 ERR("Allocating socket");
 423                 goto error;
 424         }
 425
 426         ret = lttcomm_create_sock(sock);
 427         if (ret < 0) {
 428                 goto error;
 429         }
 430         DBG("Listening on sock %d for live", sock->fd);
 431
 432         ret = sock->ops->bind(sock);
 433         if (ret < 0) {
 434                 goto error;
 435         }
 436
 437         ret = sock->ops->listen(sock, -1);
 438         if (ret < 0) {
 439                 goto error;
 440
 441         }
 442
 443         return sock;
 444
 445 error:
 446         if (sock) {
 447                 lttcomm_destroy_sock(sock);
 448         }
 449         return NULL;
 450 }
 451
 452 /*
 453  * This thread manages the listening for new connections on the network
 454  */
 455 static
 456 void *thread_listener(void *data)
 457 {
 458         int i, ret, pollfd, err = -1;
 459         uint32_t revents, nb_fd;
 460         struct lttng_poll_event events;
 461         struct lttcomm_sock *live_control_sock;
 462
 463         DBG("[thread] Relay live listener started");
 464
 465         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 466
 467         health_code_update();
 468
 469         live_control_sock = init_socket(live_uri);
 470         if (!live_control_sock) {
 471                 goto error_sock_control;
 472         }
 473
 474         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 475         ret = create_thread_poll_set(&events, 2);
 476         if (ret < 0) {
 477                 goto error_create_poll;
 478         }
 479
 480         /* Add the control socket */
 481         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 482         if (ret < 0) {
 483                 goto error_poll_add;
 484         }
 485
 486         lttng_relay_notify_ready();
 487
 488         if (testpoint(relayd_thread_live_listener)) {
 489                 goto error_testpoint;
 490         }
 491
 492         while (1) {
 493                 health_code_update();
 494
 495                 DBG("Listener accepting live viewers connections");
 496
 497 restart:
 498                 health_poll_entry();
 499                 ret = lttng_poll_wait(&events, -1);
 500                 health_poll_exit();
 501                 if (ret < 0) {
 502                         /*
 503                          * Restart interrupted system call.
 504                          */
 505                         if (errno == EINTR) {
 506                                 goto restart;
 507                         }
 508                         goto error;
 509                 }
 510                 nb_fd = ret;
 511
 512                 DBG("Relay new viewer connection received");
 513                 for (i = 0; i < nb_fd; i++) {
 514                         health_code_update();
 515
 516                         /* Fetch once the poll data */
 517                         revents = LTTNG_POLL_GETEV(&events, i);
 518                         pollfd = LTTNG_POLL_GETFD(&events, i);
 519
 520                         if (!revents) {
 521                                 /* No activity for this FD (poll implementation). */
 522                                 continue;
 523                         }
 524
 525                         /* Thread quit pipe has been closed. Killing thread. */
 526                         ret = check_thread_quit_pipe(pollfd, revents);
 527                         if (ret) {
 528                                 err = 0;
 529                                 goto exit;
 530                         }
 531
 532                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 533                                 ERR("socket poll error");
 534                                 goto error;
 535                         } else if (revents & LPOLLIN) {
 536                                 /*
 537                                  * Get allocated in this thread, enqueued to a global queue,
 538                                  * dequeued and freed in the worker thread.
 539                                  */
 540                                 int val = 1;
 541                                 struct relay_connection *new_conn;
 542                                 struct lttcomm_sock *newsock;
 543
 544                                 new_conn = connection_create();
 545                                 if (!new_conn) {
 546                                         goto error;
 547                                 }
 548
 549                                 newsock = live_control_sock->ops->accept(live_control_sock);
 550                                 if (!newsock) {
 551                                         PERROR("accepting control sock");
 552                                         connection_free(new_conn);
 553                                         goto error;
 554                                 }
 555                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 556
 557                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 558                                                 sizeof(val));
 559                                 if (ret < 0) {
 560                                         PERROR("setsockopt inet");
 561                                         lttcomm_destroy_sock(newsock);
 562                                         connection_free(new_conn);
 563                                         goto error;
 564                                 }
 565                                 new_conn->sock = newsock;
 566
 567                                 /* Enqueue request for the dispatcher thread. */
 568                                 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
 569                                                  &new_conn->qnode);
 570
 571                                 /*
 572                                  * Wake the dispatch queue futex. Implicit memory barrier with
 573                                  * the exchange in cds_wfcq_enqueue.
 574                                  */
 575                                 futex_nto1_wake(&viewer_conn_queue.futex);
 576                         }
 577                 }
 578         }
 579
 580 exit:
 581 error:
 582 error_poll_add:
 583 error_testpoint:
 584         lttng_poll_clean(&events);
 585 error_create_poll:
 586         if (live_control_sock->fd >= 0) {
 587                 ret = live_control_sock->ops->close(live_control_sock);
 588                 if (ret) {
 589                         PERROR("close");
 590                 }
 591         }
 592         lttcomm_destroy_sock(live_control_sock);
 593 error_sock_control:
 594         if (err) {
 595                 health_error();
 596                 DBG("Live viewer listener thread exited with error");
 597         }
 598         health_unregister(health_relayd);
 599         DBG("Live viewer listener thread cleanup complete");
 600         if (stop_threads()) {
 601                 ERR("Error stopping live threads");
 602         }
 603         return NULL;
 604 }
 605
 606 /*
 607  * This thread manages the dispatching of the requests to worker threads
 608  */
 609 static
 610 void *thread_dispatcher(void *data)
 611 {
 612         int err = -1;
 613         ssize_t ret;
 614         struct cds_wfcq_node *node;
 615         struct relay_connection *conn = NULL;
 616
 617         DBG("[thread] Live viewer relay dispatcher started");
 618
 619         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 620
 621         if (testpoint(relayd_thread_live_dispatcher)) {
 622                 goto error_testpoint;
 623         }
 624
 625         health_code_update();
 626
 627         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 628                 health_code_update();
 629
 630                 /* Atomically prepare the queue futex */
 631                 futex_nto1_prepare(&viewer_conn_queue.futex);
 632
 633                 do {
 634                         health_code_update();
 635
 636                         /* Dequeue commands */
 637                         node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
 638                                                          &viewer_conn_queue.tail);
 639                         if (node == NULL) {
 640                                 DBG("Woken up but nothing in the live-viewer "
 641                                                 "relay command queue");
 642                                 /* Continue thread execution */
 643                                 break;
 644                         }
 645                         conn = caa_container_of(node, struct relay_connection, qnode);
 646                         DBG("Dispatching viewer request waiting on sock %d",
 647                                         conn->sock->fd);
 648
 649                         /*
 650                          * Inform worker thread of the new request. This call is blocking
 651                          * so we can be assured that the data will be read at some point in
 652                          * time or wait to the end of the world :)
 653                          */
 654                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 655                         if (ret < 0) {
 656                                 PERROR("write conn pipe");
 657                                 connection_destroy(conn);
 658                                 goto error;
 659                         }
 660                 } while (node != NULL);
 661
 662                 /* Futex wait on queue. Blocking call on futex() */
 663                 health_poll_entry();
 664                 futex_nto1_wait(&viewer_conn_queue.futex);
 665                 health_poll_exit();
 666         }
 667
 668         /* Normal exit, no error */
 669         err = 0;
 670
 671 error:
 672 error_testpoint:
 673         if (err) {
 674                 health_error();
 675                 ERR("Health error occurred in %s", __func__);
 676         }
 677         health_unregister(health_relayd);
 678         DBG("Live viewer dispatch thread dying");
 679         if (stop_threads()) {
 680                 ERR("Error stopping live threads");
 681         }
 682         return NULL;
 683 }
 684
 685 /*
 686  * Establish connection with the viewer and check the versions.
 687  *
 688  * Return 0 on success or else negative value.
 689  */
 690 static
 691 int viewer_connect(struct relay_connection *conn)
 692 {
 693         int ret;
 694         struct lttng_viewer_connect reply, msg;
 695
 696         assert(conn);
 697
 698         conn->version_check_done = 1;
 699
 700         health_code_update();
 701
 702         DBG("Viewer is establishing a connection to the relayd.");
 703
 704         ret = recv_request(conn->sock, &msg, sizeof(msg));
 705         if (ret < 0) {
 706                 goto end;
 707         }
 708
 709         health_code_update();
 710
 711         memset(&reply, 0, sizeof(reply));
 712         reply.major = RELAYD_VERSION_COMM_MAJOR;
 713         reply.minor = RELAYD_VERSION_COMM_MINOR;
 714
 715         /* Major versions must be the same */
 716         if (reply.major != be32toh(msg.major)) {
 717                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 718                                 reply.major, be32toh(msg.major));
 719                 ret = -1;
 720                 goto end;
 721         }
 722
 723         conn->major = reply.major;
 724         /* We adapt to the lowest compatible version */
 725         if (reply.minor <= be32toh(msg.minor)) {
 726                 conn->minor = reply.minor;
 727         } else {
 728                 conn->minor = be32toh(msg.minor);
 729         }
 730
 731         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 732                 conn->type = RELAY_VIEWER_COMMAND;
 733         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 734                 conn->type = RELAY_VIEWER_NOTIFICATION;
 735         } else {
 736                 ERR("Unknown connection type : %u", be32toh(msg.type));
 737                 ret = -1;
 738                 goto end;
 739         }
 740
 741         reply.major = htobe32(reply.major);
 742         reply.minor = htobe32(reply.minor);
 743         if (conn->type == RELAY_VIEWER_COMMAND) {
 744                 /*
 745                  * Increment outside of htobe64 macro, because can be used more than once
 746                  * within the macro, and thus the operation may be undefined.
 747                  */
 748                 last_relay_viewer_session_id++;
 749                 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
 750         }
 751
 752         health_code_update();
 753
 754         ret = send_response(conn->sock, &reply, sizeof(reply));
 755         if (ret < 0) {
 756                 goto end;
 757         }
 758
 759         health_code_update();
 760
 761         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 762         ret = 0;
 763
 764 end:
 765         return ret;
 766 }
 767
 768 /*
 769  * Send the viewer the list of current sessions.
 770  *
 771  * Return 0 on success or else a negative value.
 772  */
 773 static
 774 int viewer_list_sessions(struct relay_connection *conn)
 775 {
 776         int ret;
 777         struct lttng_viewer_list_sessions session_list;
 778         unsigned long count;
 779         long approx_before, approx_after;
 780         struct lttng_ht_iter iter;
 781         struct lttng_viewer_session send_session;
 782         struct relay_session *session;
 783
 784         DBG("List sessions received");
 785
 786         rcu_read_lock();
 787         cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
 788                         &approx_after);
 789         session_list.sessions_count = htobe32(count);
 790
 791         health_code_update();
 792
 793         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 794         if (ret < 0) {
 795                 goto end_unlock;
 796         }
 797
 798         health_code_update();
 799
 800         cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
 801                         session_n.node) {
 802                 health_code_update();
 803
 804                 strncpy(send_session.session_name, session->session_name,
 805                                 sizeof(send_session.session_name));
 806                 strncpy(send_session.hostname, session->hostname,
 807                                 sizeof(send_session.hostname));
 808                 send_session.id = htobe64(session->id);
 809                 send_session.live_timer = htobe32(session->live_timer);
 810                 send_session.clients = htobe32(session->viewer_refcount);
 811                 send_session.streams = htobe32(session->stream_count);
 812
 813                 health_code_update();
 814
 815                 ret = send_response(conn->sock, &send_session, sizeof(send_session));
 816                 if (ret < 0) {
 817                         goto end_unlock;
 818                 }
 819         }
 820         health_code_update();
 821
 822         ret = 0;
 823 end_unlock:
 824         rcu_read_unlock();
 825         return ret;
 826 }
 827
 828 /*
 829  * Check if a connection is attached to a session.
 830  * Return 1 if attached, 0 if not attached, a negative value on error.
 831  */
 832 static
 833 int session_attached(struct relay_connection *conn, uint64_t session_id)
 834 {
 835         struct relay_session *session;
 836         int found = 0;
 837
 838         if (!conn->viewer_session) {
 839                 goto end;
 840         }
 841         cds_list_for_each_entry(session,
 842                         &conn->viewer_session->sessions_head,
 843                         viewer_session_list) {
 844                 if (session->id == session_id) {
 845                         found = 1;
 846                         goto end;
 847                 }
 848         }
 849
 850 end:
 851         return found;
 852 }
 853
 854 /*
 855  * Delete all streams for a specific session ID.
 856  */
 857 static void destroy_viewer_streams_by_session(struct relay_session *session)
 858 {
 859         struct relay_viewer_stream *stream;
 860         struct lttng_ht_iter iter;
 861
 862         assert(session);
 863
 864         rcu_read_lock();
 865         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
 866                         stream_n.node) {
 867                 struct ctf_trace *ctf_trace;
 868
 869                 health_code_update();
 870                 if (stream->session_id != session->id) {
 871                         continue;
 872                 }
 873
 874                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 875                                 stream->path_name);
 876                 assert(ctf_trace);
 877
 878                 viewer_stream_delete(stream);
 879
 880                 if (stream->metadata_flag) {
 881                         ctf_trace->metadata_sent = 0;
 882                         ctf_trace->viewer_metadata_stream = NULL;
 883                 }
 884
 885                 viewer_stream_destroy(ctf_trace, stream);
 886         }
 887         rcu_read_unlock();
 888 }
 889
 890 static void try_destroy_streams(struct relay_session *session)
 891 {
 892         struct ctf_trace *ctf_trace;
 893         struct lttng_ht_iter iter;
 894
 895         assert(session);
 896
 897         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 898                         node.node) {
 899                 /* Attempt to destroy the ctf trace of that session. */
 900                 ctf_trace_try_destroy(session, ctf_trace);
 901         }
 902 }
 903
 904 /*
 905  * Cleanup a session.
 906  */
 907 static void cleanup_session(struct relay_connection *conn,
 908                 struct relay_session *session)
 909 {
 910         /*
 911          * Very important that this is done before destroying the session so we
 912          * can put back every viewer stream reference from the ctf_trace.
 913          */
 914         destroy_viewer_streams_by_session(session);
 915         try_destroy_streams(session);
 916         cds_list_del(&session->viewer_session_list);
 917         session_viewer_try_destroy(conn->sessions_ht, session);
 918 }
 919
 920 /*
 921  * Send the viewer the list of current sessions.
 922  */
 923 static
 924 int viewer_get_new_streams(struct relay_connection *conn)
 925 {
 926         int ret, send_streams = 0;
 927         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
 928         struct lttng_viewer_new_streams_request request;
 929         struct lttng_viewer_new_streams_response response;
 930         struct relay_session *session;
 931         uint64_t session_id;
 932
 933         assert(conn);
 934
 935         DBG("Get new streams received");
 936
 937         health_code_update();
 938
 939         /* Receive the request from the connected client. */
 940         ret = recv_request(conn->sock, &request, sizeof(request));
 941         if (ret < 0) {
 942                 goto error;
 943         }
 944         session_id = be64toh(request.session_id);
 945
 946         health_code_update();
 947
 948         memset(&response, 0, sizeof(response));
 949
 950         rcu_read_lock();
 951         session = session_find_by_id(conn->sessions_ht, session_id);
 952         if (!session) {
 953                 DBG("Relay session %" PRIu64 " not found", session_id);
 954                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 955                 goto send_reply;
 956         }
 957
 958         if (!session_attached(conn, session_id)) {
 959                 send_streams = 0;
 960                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 961                 goto send_reply;
 962         }
 963
 964         send_streams = 1;
 965         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 966
 967         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
 968                         &nb_created);
 969         if (ret < 0) {
 970                 goto end_unlock;
 971         }
 972         /* Only send back the newly created streams with the unsent ones. */
 973         nb_streams = nb_created + nb_unsent;
 974         response.streams_count = htobe32(nb_streams);
 975
 976         /*
 977          * If the session is closed and we have no new streams to send,
 978          * it means that the viewer has already received the whole trace
 979          * for this session and should now close it.
 980          */
 981         if (nb_streams == 0 && session->close_flag) {
 982                 send_streams = 0;
 983                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 984                 /*
 985                  * Remove the session from the attached list of the connection
 986                  * and try to destroy it.
 987                  */
 988                 cds_list_del(&session->viewer_session_list);
 989                 cleanup_session(conn, session);
 990                 goto send_reply;
 991         }
 992
 993 send_reply:
 994         health_code_update();
 995         ret = send_response(conn->sock, &response, sizeof(response));
 996         if (ret < 0) {
 997                 goto end_unlock;
 998         }
 999         health_code_update();
1000
1001         /*
1002          * Unknown or empty session, just return gracefully, the viewer knows what
1003          * is happening.
1004          */
1005         if (!send_streams || !nb_streams) {
1006                 ret = 0;
1007                 goto end_unlock;
1008         }
1009
1010         /*
1011          * Send stream and *DON'T* ignore the sent flag so every viewer streams
1012          * that were not sent from that point will be sent to the viewer.
1013          */
1014         ret = send_viewer_streams(conn->sock, session, 0);
1015         if (ret < 0) {
1016                 goto end_unlock;
1017         }
1018
1019 end_unlock:
1020         rcu_read_unlock();
1021 error:
1022         return ret;
1023 }
1024
1025 /*
1026  * Send the viewer the list of current sessions.
1027  */
1028 static
1029 int viewer_attach_session(struct relay_connection *conn)
1030 {
1031         int send_streams = 0;
1032         ssize_t ret;
1033         uint32_t nb_streams = 0;
1034         enum lttng_viewer_seek seek_type;
1035         struct lttng_viewer_attach_session_request request;
1036         struct lttng_viewer_attach_session_response response;
1037         struct relay_session *session;
1038
1039         assert(conn);
1040
1041         health_code_update();
1042
1043         /* Receive the request from the connected client. */
1044         ret = recv_request(conn->sock, &request, sizeof(request));
1045         if (ret < 0) {
1046                 goto error;
1047         }
1048
1049         health_code_update();
1050
1051         memset(&response, 0, sizeof(response));
1052
1053         if (!conn->viewer_session) {
1054                 DBG("Client trying to attach before creating a live viewer session");
1055                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1056                 goto send_reply;
1057         }
1058
1059         rcu_read_lock();
1060         session = session_find_by_id(conn->sessions_ht,
1061                         be64toh(request.session_id));
1062         if (!session) {
1063                 DBG("Relay session %" PRIu64 " not found",
1064                                 be64toh(request.session_id));
1065                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1066                 goto send_reply;
1067         }
1068         session_viewer_attach(session);
1069         DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1070
1071         if (uatomic_read(&session->viewer_refcount) > 1) {
1072                 DBG("Already a viewer attached");
1073                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1074                 session_viewer_detach(session);
1075                 goto send_reply;
1076         } else if (session->live_timer == 0) {
1077                 DBG("Not live session");
1078                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1079                 goto send_reply;
1080         } else {
1081                 send_streams = 1;
1082                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1083                 cds_list_add(&session->viewer_session_list,
1084                                 &conn->viewer_session->sessions_head);
1085         }
1086
1087         switch (be32toh(request.seek)) {
1088         case LTTNG_VIEWER_SEEK_BEGINNING:
1089         case LTTNG_VIEWER_SEEK_LAST:
1090                 seek_type = be32toh(request.seek);
1091                 break;
1092         default:
1093                 ERR("Wrong seek parameter");
1094                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1095                 send_streams = 0;
1096                 goto send_reply;
1097         }
1098
1099         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1100         if (ret < 0) {
1101                 goto end_unlock;
1102         }
1103         response.streams_count = htobe32(nb_streams);
1104
1105 send_reply:
1106         health_code_update();
1107         ret = send_response(conn->sock, &response, sizeof(response));
1108         if (ret < 0) {
1109                 goto end_unlock;
1110         }
1111         health_code_update();
1112
1113         /*
1114          * Unknown or empty session, just return gracefully, the viewer knows what
1115          * is happening.
1116          */
1117         if (!send_streams || !nb_streams) {
1118                 ret = 0;
1119                 goto end_unlock;
1120         }
1121
1122         /* Send stream and ignore the sent flag. */
1123         ret = send_viewer_streams(conn->sock, session, 1);
1124         if (ret < 0) {
1125                 goto end_unlock;
1126         }
1127
1128 end_unlock:
1129         rcu_read_unlock();
1130 error:
1131         return ret;
1132 }
1133
1134 /*
1135  * Open the index file if needed for the given vstream.
1136  *
1137  * If an index file is successfully opened, the index_read_fd of the stream is
1138  * set with it.
1139  *
1140  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1141  */
1142 static int try_open_index(struct relay_viewer_stream *vstream,
1143                 struct relay_stream *rstream)
1144 {
1145         int ret = 0;
1146
1147         assert(vstream);
1148         assert(rstream);
1149
1150         if (vstream->index_read_fd >= 0) {
1151                 goto end;
1152         }
1153
1154         /*
1155          * First time, we open the index file and at least one index is ready.  The
1156          * race between the read and write of the total_index_received is
1157          * acceptable here since the client will be notified to simply come back
1158          * and get the next index.
1159          */
1160         if (rstream->total_index_received <= 0) {
1161                 ret = -ENOENT;
1162                 goto end;
1163         }
1164         ret = index_open(vstream->path_name, vstream->channel_name,
1165                         vstream->tracefile_count, vstream->tracefile_count_current);
1166         if (ret >= 0) {
1167                 vstream->index_read_fd = ret;
1168                 ret = 0;
1169                 goto end;
1170         }
1171
1172 end:
1173         return ret;
1174 }
1175
1176 /*
1177  * Check the status of the index for the given stream. This function updates
1178  * the index structure if needed and can destroy the vstream also for the HUP
1179  * situation.
1180  *
1181  * Return 0 means that we can proceed with the index. A value of 1 means that
1182  * the index has been updated and is ready to be send to the client. A negative
1183  * value indicates an error that can't be handled.
1184  */
1185 static int check_index_status(struct relay_viewer_stream *vstream,
1186                 struct relay_stream *rstream, struct ctf_trace *trace,
1187                 struct lttng_viewer_index *index)
1188 {
1189         int ret;
1190
1191         assert(vstream);
1192         assert(rstream);
1193         assert(index);
1194         assert(trace);
1195
1196         if (!rstream->close_flag) {
1197                 /* Rotate on abort (overwrite). */
1198                 if (vstream->abort_flag) {
1199                         DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1200                                         vstream->stream_handle);
1201                         ret = viewer_stream_rotate(vstream, rstream);
1202                         if (ret < 0) {
1203                                 goto error;
1204                         } else if (ret == 1) {
1205                                 /* EOF */
1206                                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1207                                 goto hup;
1208                         }
1209                         /* ret == 0 means successful so we continue. */
1210                 }
1211
1212                 /* Check if we are in the same trace file at this point. */
1213                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1214                         if (rstream->beacon_ts_end != -1ULL &&
1215                                         vstream->last_sent_index == rstream->total_index_received) {
1216                                 /*
1217                                  * We've received a synchronization beacon and the last index
1218                                  * available has been sent, the index for now is inactive.
1219                                  */
1220                                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1221                                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1222                                 index->stream_id = htobe64(rstream->ctf_stream_id);
1223                                 goto index_ready;
1224                         } else if (rstream->total_index_received <= vstream->last_sent_index
1225                                         && !vstream->close_write_flag) {
1226                                 /*
1227                                  * Reader and writer are working in the same tracefile, so we care
1228                                  * about the number of index received and sent. Otherwise, we read
1229                                  * up to EOF.
1230                                  */
1231                                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1232                                 goto index_ready;
1233                         }
1234                 }
1235                 /* Nothing to do with the index, continue with it. */
1236                 ret = 0;
1237         } else if (rstream->close_flag && vstream->close_write_flag &&
1238                         vstream->total_index_received == vstream->last_sent_index) {
1239                 /* Last index sent and current tracefile closed in write */
1240                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1241                 goto hup;
1242         } else {
1243                 vstream->close_write_flag = 1;
1244                 ret = 0;
1245         }
1246
1247 error:
1248         return ret;
1249
1250 hup:
1251         viewer_stream_delete(vstream);
1252         viewer_stream_destroy(trace, vstream);
1253 index_ready:
1254         return 1;
1255 }
1256
1257 /*
1258  * Send the next index for a stream.
1259  *
1260  * Return 0 on success or else a negative value.
1261  */
1262 static
1263 int viewer_get_next_index(struct relay_connection *conn)
1264 {
1265         int ret;
1266         ssize_t read_ret;
1267         struct lttng_viewer_get_next_index request_index;
1268         struct lttng_viewer_index viewer_index;
1269         struct ctf_packet_index packet_index;
1270         struct relay_viewer_stream *vstream;
1271         struct relay_stream *rstream;
1272         struct ctf_trace *ctf_trace;
1273         struct relay_session *session;
1274
1275         assert(conn);
1276
1277         DBG("Viewer get next index");
1278
1279         health_code_update();
1280
1281         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1282         if (ret < 0) {
1283                 goto end;
1284         }
1285         health_code_update();
1286
1287         rcu_read_lock();
1288         vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1289         if (!vstream) {
1290                 ret = -1;
1291                 goto end_unlock;
1292         }
1293
1294         session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1295         if (!session) {
1296                 ret = -1;
1297                 goto end_unlock;
1298         }
1299
1300         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1301         assert(ctf_trace);
1302
1303         memset(&viewer_index, 0, sizeof(viewer_index));
1304
1305         /*
1306          * The viewer should not ask for index on metadata stream.
1307          */
1308         if (vstream->metadata_flag) {
1309                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1310                 goto send_reply;
1311         }
1312
1313         rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1314         assert(rstream);
1315
1316         /* Try to open an index if one is needed for that stream. */
1317         ret = try_open_index(vstream, rstream);
1318         if (ret < 0) {
1319                 if (ret == -ENOENT) {
1320                         /*
1321                          * The index is created only when the first data packet arrives, it
1322                          * might not be ready at the beginning of the session
1323                          */
1324                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1325                 } else {
1326                         /* Unhandled error. */
1327                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1328                 }
1329                 goto send_reply;
1330         }
1331
1332         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1333         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1334         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1335         if (ret < 0) {
1336                 goto end_unlock;
1337         } else if (ret == 1) {
1338                 /*
1339                  * This means the viewer index data structure has been populated by the
1340                  * check call thus we now send back the reply to the client.
1341                  */
1342                 goto send_reply;
1343         }
1344         /* At this point, ret MUST be 0 thus we continue with the get. */
1345         assert(!ret);
1346
1347         if (!ctf_trace->metadata_received ||
1348                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1349                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1350         }
1351
1352         ret = check_new_streams(conn);
1353         if (ret < 0) {
1354                 goto end_unlock;
1355         } else if (ret == 1) {
1356                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1357         }
1358
1359         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1360         pthread_mutex_lock(&vstream->overwrite_lock);
1361         if (vstream->abort_flag) {
1362                 /* The file is being overwritten by the writer, we cannot use it. */
1363                 pthread_mutex_unlock(&vstream->overwrite_lock);
1364                 ret = viewer_stream_rotate(vstream, rstream);
1365                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1366                 if (ret < 0) {
1367                         goto end_unlock;
1368                 } else if (ret == 1) {
1369                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1370                         viewer_stream_delete(vstream);
1371                         viewer_stream_destroy(ctf_trace, vstream);
1372                 } else {
1373                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1374                 }
1375                 goto send_reply;
1376         }
1377
1378         read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1379                         sizeof(packet_index));
1380         pthread_mutex_unlock(&vstream->overwrite_lock);
1381         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1382         if (read_ret < 0) {
1383                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1384                 viewer_stream_delete(vstream);
1385                 viewer_stream_destroy(ctf_trace, vstream);
1386                 goto send_reply;
1387         } else if (read_ret < sizeof(packet_index)) {
1388                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1389                 if (vstream->close_write_flag) {
1390                         ret = viewer_stream_rotate(vstream, rstream);
1391                         if (ret < 0) {
1392                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1393                                 goto end_unlock;
1394                         } else if (ret == 1) {
1395                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1396                                 viewer_stream_delete(vstream);
1397                                 viewer_stream_destroy(ctf_trace, vstream);
1398                         } else {
1399                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1400                         }
1401                 } else {
1402                         ERR("Relay reading index file %d", vstream->index_read_fd);
1403                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1404                 }
1405                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1406                 goto send_reply;
1407         } else {
1408                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1409                 vstream->last_sent_index++;
1410         }
1411
1412         /*
1413          * Indexes are stored in big endian, no need to switch before sending.
1414          */
1415         viewer_index.offset = packet_index.offset;
1416         viewer_index.packet_size = packet_index.packet_size;
1417         viewer_index.content_size = packet_index.content_size;
1418         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1419         viewer_index.timestamp_end = packet_index.timestamp_end;
1420         viewer_index.events_discarded = packet_index.events_discarded;
1421         viewer_index.stream_id = packet_index.stream_id;
1422
1423 send_reply:
1424         viewer_index.flags = htobe32(viewer_index.flags);
1425         health_code_update();
1426
1427         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1428         if (ret < 0) {
1429                 goto end_unlock;
1430         }
1431         health_code_update();
1432
1433         DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1434                         vstream->last_sent_index, vstream->stream_handle);
1435
1436 end_unlock:
1437         rcu_read_unlock();
1438
1439 end:
1440         return ret;
1441 }
1442
1443 /*
1444  * Send the next index for a stream
1445  *
1446  * Return 0 on success or else a negative value.
1447  */
1448 static
1449 int viewer_get_packet(struct relay_connection *conn)
1450 {
1451         int ret, send_data = 0;
1452         char *data = NULL;
1453         uint32_t len = 0;
1454         ssize_t read_len;
1455         struct lttng_viewer_get_packet get_packet_info;
1456         struct lttng_viewer_trace_packet reply;
1457         struct relay_viewer_stream *stream;
1458         struct relay_session *session;
1459         struct ctf_trace *ctf_trace;
1460
1461         assert(conn);
1462
1463         DBG2("Relay get data packet");
1464
1465         health_code_update();
1466
1467         ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1468         if (ret < 0) {
1469                 goto end;
1470         }
1471         health_code_update();
1472
1473         /* From this point on, the error label can be reached. */
1474         memset(&reply, 0, sizeof(reply));
1475
1476         rcu_read_lock();
1477         stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1478         if (!stream) {
1479                 goto error;
1480         }
1481
1482         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1483         if (!session) {
1484                 ret = -1;
1485                 goto error;
1486         }
1487
1488         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1489                         stream->path_name);
1490         assert(ctf_trace);
1491
1492         /*
1493          * First time we read this stream, we need open the tracefile, we should
1494          * only arrive here if an index has already been sent to the viewer, so the
1495          * tracefile must exist, if it does not it is a fatal error.
1496          */
1497         if (stream->read_fd < 0) {
1498                 char fullpath[PATH_MAX];
1499
1500                 if (stream->tracefile_count > 0) {
1501                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1502                                         stream->channel_name,
1503                                         stream->tracefile_count_current);
1504                 } else {
1505                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1506                                         stream->channel_name);
1507                 }
1508                 if (ret < 0) {
1509                         goto error;
1510                 }
1511                 ret = open(fullpath, O_RDONLY);
1512                 if (ret < 0) {
1513                         PERROR("Relay opening trace file");
1514                         goto error;
1515                 }
1516                 stream->read_fd = ret;
1517         }
1518
1519         if (!ctf_trace->metadata_received ||
1520                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1521                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1522                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1523                 goto send_reply;
1524         }
1525
1526         ret = check_new_streams(conn);
1527         if (ret < 0) {
1528                 goto end_unlock;
1529         } else if (ret == 1) {
1530                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1531                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1532                 goto send_reply;
1533         }
1534
1535         len = be32toh(get_packet_info.len);
1536         data = zmalloc(len);
1537         if (!data) {
1538                 PERROR("relay data zmalloc");
1539                 goto error;
1540         }
1541
1542         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1543         if (ret < 0) {
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("lseek");
1550                         goto error;
1551                 }
1552                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1553                 goto send_reply;
1554         }
1555         read_len = lttng_read(stream->read_fd, data, len);
1556         if (read_len < len) {
1557                 /*
1558                  * If the read fd was closed by the streaming side, the
1559                  * abort_flag will be set to 1, otherwise it is an error.
1560                  */
1561                 if (stream->abort_flag == 0) {
1562                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1563                                         stream->read_fd,
1564                                         be64toh(get_packet_info.offset));
1565                         goto error;
1566                 } else {
1567                         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1568                         goto send_reply;
1569                 }
1570         }
1571         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1572         reply.len = htobe32(len);
1573         send_data = 1;
1574         goto send_reply;
1575
1576 error:
1577         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1578
1579 send_reply:
1580         reply.flags = htobe32(reply.flags);
1581
1582         health_code_update();
1583
1584         ret = send_response(conn->sock, &reply, sizeof(reply));
1585         if (ret < 0) {
1586                 goto end_unlock;
1587         }
1588         health_code_update();
1589
1590         if (send_data) {
1591                 health_code_update();
1592                 ret = send_response(conn->sock, data, len);
1593                 if (ret < 0) {
1594                         goto end_unlock;
1595                 }
1596                 health_code_update();
1597         }
1598
1599         DBG("Sent %u bytes for stream %" PRIu64, len,
1600                         be64toh(get_packet_info.stream_id));
1601
1602 end_unlock:
1603         free(data);
1604         rcu_read_unlock();
1605
1606 end:
1607         return ret;
1608 }
1609
1610 /*
1611  * Send the session's metadata
1612  *
1613  * Return 0 on success else a negative value.
1614  */
1615 static
1616 int viewer_get_metadata(struct relay_connection *conn)
1617 {
1618         int ret = 0;
1619         ssize_t read_len;
1620         uint64_t len = 0;
1621         char *data = NULL;
1622         struct lttng_viewer_get_metadata request;
1623         struct lttng_viewer_metadata_packet reply;
1624         struct relay_viewer_stream *stream;
1625         struct ctf_trace *ctf_trace;
1626         struct relay_session *session;
1627
1628         assert(conn);
1629
1630         DBG("Relay get metadata");
1631
1632         health_code_update();
1633
1634         ret = recv_request(conn->sock, &request, sizeof(request));
1635         if (ret < 0) {
1636                 goto end;
1637         }
1638         health_code_update();
1639
1640         memset(&reply, 0, sizeof(reply));
1641
1642         rcu_read_lock();
1643         stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1644         if (!stream || !stream->metadata_flag) {
1645                 ERR("Invalid metadata stream");
1646                 goto error;
1647         }
1648
1649         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1650         if (!session) {
1651                 ret = -1;
1652                 goto error;
1653         }
1654
1655         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1656                         stream->path_name);
1657         assert(ctf_trace);
1658         assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1659
1660         len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1661         if (len == 0) {
1662                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1663                 goto send_reply;
1664         }
1665
1666         /* first time, we open the metadata file */
1667         if (stream->read_fd < 0) {
1668                 char fullpath[PATH_MAX];
1669
1670                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1671                                 stream->channel_name);
1672                 if (ret < 0) {
1673                         goto error;
1674                 }
1675                 ret = open(fullpath, O_RDONLY);
1676                 if (ret < 0) {
1677                         PERROR("Relay opening metadata file");
1678                         goto error;
1679                 }
1680                 stream->read_fd = ret;
1681         }
1682
1683         reply.len = htobe64(len);
1684         data = zmalloc(len);
1685         if (!data) {
1686                 PERROR("viewer metadata zmalloc");
1687                 goto error;
1688         }
1689
1690         read_len = lttng_read(stream->read_fd, data, len);
1691         if (read_len < len) {
1692                 PERROR("Relay reading metadata file");
1693                 goto error;
1694         }
1695         ctf_trace->metadata_sent += read_len;
1696         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1697         goto send_reply;
1698
1699 error:
1700         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1701
1702 send_reply:
1703         health_code_update();
1704         ret = send_response(conn->sock, &reply, sizeof(reply));
1705         if (ret < 0) {
1706                 goto end_unlock;
1707         }
1708         health_code_update();
1709
1710         if (len > 0) {
1711                 ret = send_response(conn->sock, data, len);
1712                 if (ret < 0) {
1713                         goto end_unlock;
1714                 }
1715         }
1716
1717         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1718                         be64toh(request.stream_id));
1719
1720         DBG("Metadata sent");
1721
1722 end_unlock:
1723         free(data);
1724         rcu_read_unlock();
1725 end:
1726         return ret;
1727 }
1728
1729 /*
1730  * Create a viewer session.
1731  *
1732  * Return 0 on success or else a negative value.
1733  */
1734 static
1735 int viewer_create_session(struct relay_connection *conn)
1736 {
1737         int ret;
1738         struct lttng_viewer_create_session_response resp;
1739
1740         DBG("Viewer create session received");
1741
1742         memset(&resp, 0, sizeof(resp));
1743         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1744         conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1745         if (!conn->viewer_session) {
1746                 ERR("Allocation viewer session");
1747                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1748                 goto send_reply;
1749         }
1750         CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1751
1752 send_reply:
1753         health_code_update();
1754         ret = send_response(conn->sock, &resp, sizeof(resp));
1755         if (ret < 0) {
1756                 goto end;
1757         }
1758         health_code_update();
1759         ret = 0;
1760
1761 end:
1762         return ret;
1763 }
1764
1765
1766 /*
1767  * live_relay_unknown_command: send -1 if received unknown command
1768  */
1769 static
1770 void live_relay_unknown_command(struct relay_connection *conn)
1771 {
1772         struct lttcomm_relayd_generic_reply reply;
1773
1774         memset(&reply, 0, sizeof(reply));
1775         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1776         (void) send_response(conn->sock, &reply, sizeof(reply));
1777 }
1778
1779 /*
1780  * Process the commands received on the control socket
1781  */
1782 static
1783 int process_control(struct lttng_viewer_cmd *recv_hdr,
1784                 struct relay_connection *conn)
1785 {
1786         int ret = 0;
1787         uint32_t msg_value;
1788
1789         assert(recv_hdr);
1790         assert(conn);
1791
1792         msg_value = be32toh(recv_hdr->cmd);
1793
1794         /*
1795          * Make sure we've done the version check before any command other then a
1796          * new client connection.
1797          */
1798         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1799                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1800                 ret = -1;
1801                 goto end;
1802         }
1803
1804         switch (msg_value) {
1805         case LTTNG_VIEWER_CONNECT:
1806                 ret = viewer_connect(conn);
1807                 break;
1808         case LTTNG_VIEWER_LIST_SESSIONS:
1809                 ret = viewer_list_sessions(conn);
1810                 break;
1811         case LTTNG_VIEWER_ATTACH_SESSION:
1812                 ret = viewer_attach_session(conn);
1813                 break;
1814         case LTTNG_VIEWER_GET_NEXT_INDEX:
1815                 ret = viewer_get_next_index(conn);
1816                 break;
1817         case LTTNG_VIEWER_GET_PACKET:
1818                 ret = viewer_get_packet(conn);
1819                 break;
1820         case LTTNG_VIEWER_GET_METADATA:
1821                 ret = viewer_get_metadata(conn);
1822                 break;
1823         case LTTNG_VIEWER_GET_NEW_STREAMS:
1824                 ret = viewer_get_new_streams(conn);
1825                 break;
1826         case LTTNG_VIEWER_CREATE_SESSION:
1827                 ret = viewer_create_session(conn);
1828                 break;
1829         default:
1830                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1831                 live_relay_unknown_command(conn);
1832                 ret = -1;
1833                 goto end;
1834         }
1835
1836 end:
1837         return ret;
1838 }
1839
1840 static
1841 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1842 {
1843         int ret;
1844
1845         assert(events);
1846
1847         (void) lttng_poll_del(events, pollfd);
1848
1849         ret = close(pollfd);
1850         if (ret < 0) {
1851                 ERR("Closing pollfd %d", pollfd);
1852         }
1853 }
1854
1855 /*
1856  * Delete and destroy a connection.
1857  *
1858  * RCU read side lock MUST be acquired.
1859  */
1860 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1861                 struct relay_connection *conn)
1862 {
1863         struct relay_session *session, *tmp_session;
1864
1865         assert(relay_connections_ht);
1866         assert(conn);
1867
1868         connection_delete(relay_connections_ht, conn);
1869
1870         if (!conn->viewer_session) {
1871                 goto end;
1872         }
1873
1874         rcu_read_lock();
1875         cds_list_for_each_entry_safe(session, tmp_session,
1876                         &conn->viewer_session->sessions_head,
1877                         viewer_session_list) {
1878                 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1879                 cleanup_session(conn, session);
1880         }
1881         rcu_read_unlock();
1882
1883 end:
1884         connection_destroy(conn);
1885 }
1886
1887 /*
1888  * This thread does the actual work
1889  */
1890 static
1891 void *thread_worker(void *data)
1892 {
1893         int ret, err = -1;
1894         uint32_t nb_fd;
1895         struct relay_connection *conn;
1896         struct lttng_poll_event events;
1897         struct lttng_ht *relay_connections_ht;
1898         struct lttng_ht_iter iter;
1899         struct lttng_viewer_cmd recv_hdr;
1900         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1901         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1902
1903         DBG("[thread] Live viewer relay worker started");
1904
1905         rcu_register_thread();
1906
1907         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1908
1909         if (testpoint(relayd_thread_live_worker)) {
1910                 goto error_testpoint;
1911         }
1912
1913         /* table of connections indexed on socket */
1914         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1915         if (!relay_connections_ht) {
1916                 goto relay_connections_ht_error;
1917         }
1918
1919         ret = create_thread_poll_set(&events, 2);
1920         if (ret < 0) {
1921                 goto error_poll_create;
1922         }
1923
1924         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1925         if (ret < 0) {
1926                 goto error;
1927         }
1928
1929 restart:
1930         while (1) {
1931                 int i;
1932
1933                 health_code_update();
1934
1935                 /* Infinite blocking call, waiting for transmission */
1936                 DBG3("Relayd live viewer worker thread polling...");
1937                 health_poll_entry();
1938                 ret = lttng_poll_wait(&events, -1);
1939                 health_poll_exit();
1940                 if (ret < 0) {
1941                         /*
1942                          * Restart interrupted system call.
1943                          */
1944                         if (errno == EINTR) {
1945                                 goto restart;
1946                         }
1947                         goto error;
1948                 }
1949
1950                 nb_fd = ret;
1951
1952                 /*
1953                  * Process control. The control connection is prioritised so we don't
1954                  * starve it with high throughput tracing data on the data
1955                  * connection.
1956                  */
1957                 for (i = 0; i < nb_fd; i++) {
1958                         /* Fetch once the poll data */
1959                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1960                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1961
1962                         health_code_update();
1963
1964                         if (!revents) {
1965                                 /* No activity for this FD (poll implementation). */
1966                                 continue;
1967                         }
1968
1969                         /* Thread quit pipe has been closed. Killing thread. */
1970                         ret = check_thread_quit_pipe(pollfd, revents);
1971                         if (ret) {
1972                                 err = 0;
1973                                 goto exit;
1974                         }
1975
1976                         /* Inspect the relay conn pipe for new connection */
1977                         if (pollfd == live_conn_pipe[0]) {
1978                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1979                                         ERR("Relay live pipe error");
1980                                         goto error;
1981                                 } else if (revents & LPOLLIN) {
1982                                         ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1983                                         if (ret < 0) {
1984                                                 goto error;
1985                                         }
1986                                         conn->sessions_ht = sessions_ht;
1987                                         connection_init(conn);
1988                                         lttng_poll_add(&events, conn->sock->fd,
1989                                                         LPOLLIN | LPOLLRDHUP);
1990                                         rcu_read_lock();
1991                                         lttng_ht_add_unique_ulong(relay_connections_ht,
1992                                                         &conn->sock_n);
1993                                         rcu_read_unlock();
1994                                         DBG("Connection socket %d added", conn->sock->fd);
1995                                 }
1996                         } else {
1997                                 rcu_read_lock();
1998                                 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1999                                 /* If not found, there is a synchronization issue. */
2000                                 assert(conn);
2001
2002                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2003                                         cleanup_connection_pollfd(&events, pollfd);
2004                                         destroy_connection(relay_connections_ht, conn);
2005                                 } else if (revents & LPOLLIN) {
2006                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2007                                                         sizeof(recv_hdr), 0);
2008                                         if (ret <= 0) {
2009                                                 /* Connection closed */
2010                                                 cleanup_connection_pollfd(&events, pollfd);
2011                                                 destroy_connection(relay_connections_ht, conn);
2012                                                 DBG("Viewer control conn closed with %d", pollfd);
2013                                         } else {
2014                                                 ret = process_control(&recv_hdr, conn);
2015                                                 if (ret < 0) {
2016                                                         /* Clear the session on error. */
2017                                                         cleanup_connection_pollfd(&events, pollfd);
2018                                                         destroy_connection(relay_connections_ht, conn);
2019                                                         DBG("Viewer connection closed with %d", pollfd);
2020                                                 }
2021                                         }
2022                                 }
2023                                 rcu_read_unlock();
2024                         }
2025                 }
2026         }
2027
2028 exit:
2029 error:
2030         lttng_poll_clean(&events);
2031
2032         /* Cleanup reamaining connection object. */
2033         rcu_read_lock();
2034         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2035                         sock_n.node) {
2036                 health_code_update();
2037                 destroy_connection(relay_connections_ht, conn);
2038         }
2039         rcu_read_unlock();
2040 error_poll_create:
2041         lttng_ht_destroy(relay_connections_ht);
2042 relay_connections_ht_error:
2043         /* Close relay conn pipes */
2044         utils_close_pipe(live_conn_pipe);
2045         if (err) {
2046                 DBG("Viewer worker thread exited with error");
2047         }
2048         DBG("Viewer worker thread cleanup complete");
2049 error_testpoint:
2050         if (err) {
2051                 health_error();
2052                 ERR("Health error occurred in %s", __func__);
2053         }
2054         health_unregister(health_relayd);
2055         if (stop_threads()) {
2056                 ERR("Error stopping live threads");
2057         }
2058         rcu_unregister_thread();
2059         return NULL;
2060 }
2061
2062 /*
2063  * Create the relay command pipe to wake thread_manage_apps.
2064  * Closed in cleanup().
2065  */
2066 static int create_conn_pipe(void)
2067 {
2068         return utils_create_pipe_cloexec(live_conn_pipe);
2069 }
2070
2071 int relayd_live_stop(void)
2072 {
2073         return stop_threads();
2074 }
2075
2076 int relayd_live_join(void)
2077 {
2078         int ret, retval = 0;
2079         void *status;
2080
2081         ret = pthread_join(live_listener_thread, &status);
2082         if (ret) {
2083                 errno = ret;
2084                 PERROR("pthread_join live listener");
2085                 retval = -1;
2086         }
2087
2088         ret = pthread_join(live_worker_thread, &status);
2089         if (ret) {
2090                 errno = ret;
2091                 PERROR("pthread_join live worker");
2092                 retval = -1;
2093         }
2094
2095         ret = pthread_join(live_dispatcher_thread, &status);
2096         if (ret) {
2097                 errno = ret;
2098                 PERROR("pthread_join live dispatcher");
2099                 retval = -1;
2100         }
2101
2102         cleanup_relayd_live();
2103
2104         return retval;
2105 }
2106
2107 /*
2108  * main
2109  */
2110 int relayd_live_create(struct lttng_uri *uri,
2111                 struct relay_local_data *relay_ctx)
2112 {
2113         int ret = 0, retval = 0;
2114         void *status;
2115         int is_root;
2116
2117         if (!uri) {
2118                 retval = -1;
2119                 goto exit_init_data;
2120         }
2121         live_uri = uri;
2122
2123         /* Check if daemon is UID = 0 */
2124         is_root = !getuid();
2125
2126         if (!is_root) {
2127                 if (live_uri->port < 1024) {
2128                         ERR("Need to be root to use ports < 1024");
2129                         retval = -1;
2130                         goto exit_init_data;
2131                 }
2132         }
2133
2134         /* Setup the thread apps communication pipe. */
2135         if (create_conn_pipe()) {
2136                 retval = -1;
2137                 goto exit_init_data;
2138         }
2139
2140         /* Init relay command queue. */
2141         cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2142
2143         /* Set up max poll set size */
2144         lttng_poll_set_max_size();
2145
2146         /* Setup the dispatcher thread */
2147         ret = pthread_create(&live_dispatcher_thread, NULL,
2148                         thread_dispatcher, (void *) NULL);
2149         if (ret) {
2150                 errno = ret;
2151                 PERROR("pthread_create viewer dispatcher");
2152                 retval = -1;
2153                 goto exit_dispatcher_thread;
2154         }
2155
2156         /* Setup the worker thread */
2157         ret = pthread_create(&live_worker_thread, NULL,
2158                         thread_worker, relay_ctx);
2159         if (ret) {
2160                 errno = ret;
2161                 PERROR("pthread_create viewer worker");
2162                 retval = -1;
2163                 goto exit_worker_thread;
2164         }
2165
2166         /* Setup the listener thread */
2167         ret = pthread_create(&live_listener_thread, NULL,
2168                         thread_listener, (void *) NULL);
2169         if (ret) {
2170                 errno = ret;
2171                 PERROR("pthread_create viewer listener");
2172                 retval = -1;
2173                 goto exit_listener_thread;
2174         }
2175
2176         /*
2177          * All OK, started all threads.
2178          */
2179         return retval;
2180
2181
2182         ret = pthread_join(live_listener_thread, &status);
2183         if (ret) {
2184                 errno = ret;
2185                 PERROR("pthread_join live listener");
2186                 retval = -1;
2187         }
2188 exit_listener_thread:
2189
2190         ret = pthread_join(live_worker_thread, &status);
2191         if (ret) {
2192                 errno = ret;
2193                 PERROR("pthread_join live worker");
2194                 retval = -1;
2195         }
2196 exit_worker_thread:
2197
2198         ret = pthread_join(live_dispatcher_thread, &status);
2199         if (ret) {
2200                 errno = ret;
2201                 PERROR("pthread_join live dispatcher");
2202                 retval = -1;
2203         }
2204 exit_dispatcher_thread:
2205
2206 exit_init_data:
2207         cleanup_relayd_live();
2208
2209         return retval;
2210 }