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/defaults.h>
  47 #include <common/futex.h>
  48 #include <common/index/index.h>
  49 #include <common/sessiond-comm/sessiond-comm.h>
  50 #include <common/sessiond-comm/inet.h>
  51 #include <common/sessiond-comm/relayd.h>
  52 #include <common/uri.h>
  53 #include <common/utils.h>
  54
  55 #include "cmd.h"
  56 #include "live.h"
  57 #include "lttng-relayd.h"
  58 #include "utils.h"
  59 #include "health-relayd.h"
  60 #include "testpoint.h"
  61 #include "viewer-stream.h"
  62 #include "stream.h"
  63 #include "session.h"
  64 #include "ctf-trace.h"
  65 #include "connection.h"
  66
  67 static struct lttng_uri *live_uri;
  68
  69 /*
  70  * This pipe is used to inform the worker thread that a command is queued and
  71  * ready to be processed.
  72  */
  73 static int live_conn_pipe[2] = { -1, -1 };
  74
  75 /* Shared between threads */
  76 static int live_dispatch_thread_exit;
  77
  78 static pthread_t live_listener_thread;
  79 static pthread_t live_dispatcher_thread;
  80 static pthread_t live_worker_thread;
  81
  82 /*
  83  * Relay command queue.
  84  *
  85  * The live_thread_listener and live_thread_dispatcher communicate with this
  86  * queue.
  87  */
  88 static struct relay_conn_queue viewer_conn_queue;
  89
  90 static uint64_t last_relay_viewer_session_id;
  91
  92 /*
  93  * Cleanup the daemon
  94  */
  95 static
  96 void cleanup(void)
  97 {
  98         DBG("Cleaning up");
  99
 100         free(live_uri);
 101 }
 102
 103 /*
 104  * Receive a request buffer using a given socket, destination allocated buffer
 105  * of length size.
 106  *
 107  * Return the size of the received message or else a negative value on error
 108  * with errno being set by recvmsg() syscall.
 109  */
 110 static
 111 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 112 {
 113         ssize_t ret;
 114
 115         assert(sock);
 116         assert(buf);
 117
 118         ret = sock->ops->recvmsg(sock, buf, size, 0);
 119         if (ret < 0 || ret != size) {
 120                 if (ret == 0) {
 121                         /* Orderly shutdown. Not necessary to print an error. */
 122                         DBG("Socket %d did an orderly shutdown", sock->fd);
 123                 } else {
 124                         ERR("Relay failed to receive request.");
 125                 }
 126                 ret = -1;
 127         }
 128
 129         return ret;
 130 }
 131
 132 /*
 133  * Send a response buffer using a given socket, source allocated buffer of
 134  * length size.
 135  *
 136  * Return the size of the sent message or else a negative value on error with
 137  * errno being set by sendmsg() syscall.
 138  */
 139 static
 140 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 141 {
 142         ssize_t ret;
 143
 144         assert(sock);
 145         assert(buf);
 146
 147         ret = sock->ops->sendmsg(sock, buf, size, 0);
 148         if (ret < 0) {
 149                 ERR("Relayd failed to send response.");
 150         }
 151
 152         return ret;
 153 }
 154
 155 /*
 156  * Atomically check if new streams got added in one of the sessions attached
 157  * and reset the flag to 0.
 158  *
 159  * Returns 1 if new streams got added, 0 if nothing changed, a negative value
 160  * on error.
 161  */
 162 static
 163 int check_new_streams(struct relay_connection *conn)
 164 {
 165         struct relay_session *session;
 166         unsigned long current_val;
 167         int ret = 0;
 168
 169         if (!conn->viewer_session) {
 170                 goto end;
 171         }
 172         cds_list_for_each_entry(session,
 173                         &conn->viewer_session->sessions_head,
 174                         viewer_session_list) {
 175                 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
 176                 ret = current_val;
 177                 if (ret == 1) {
 178                         goto end;
 179                 }
 180         }
 181
 182 end:
 183         return ret;
 184 }
 185
 186 /*
 187  * Send viewer streams to the given socket. The ignore_sent_flag indicates if
 188  * this function should ignore the sent flag or not.
 189  *
 190  * Return 0 on success or else a negative value.
 191  */
 192 static
 193 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
 194                 struct relay_session *session, unsigned int ignore_sent_flag)
 195 {
 196         ssize_t ret;
 197         struct lttng_viewer_stream send_stream;
 198         struct lttng_ht_iter iter;
 199         struct relay_viewer_stream *vstream;
 200
 201         assert(session);
 202
 203         rcu_read_lock();
 204
 205         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
 206                         stream_n.node) {
 207                 struct ctf_trace *ctf_trace;
 208
 209                 health_code_update();
 210
 211                 /* Ignore if not the same session. */
 212                 if (vstream->session_id != session->id ||
 213                                 (!ignore_sent_flag && vstream->sent_flag)) {
 214                         continue;
 215                 }
 216
 217                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 218                                 vstream->path_name);
 219                 assert(ctf_trace);
 220
 221                 send_stream.id = htobe64(vstream->stream_handle);
 222                 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
 223                 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
 224                 strncpy(send_stream.path_name, vstream->path_name,
 225                                 sizeof(send_stream.path_name));
 226                 strncpy(send_stream.channel_name, vstream->channel_name,
 227                                 sizeof(send_stream.channel_name));
 228
 229                 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
 230                 ret = send_response(sock, &send_stream, sizeof(send_stream));
 231                 if (ret < 0) {
 232                         goto end_unlock;
 233                 }
 234                 vstream->sent_flag = 1;
 235         }
 236
 237         ret = 0;
 238
 239 end_unlock:
 240         rcu_read_unlock();
 241         return ret;
 242 }
 243
 244 /*
 245  * Create every viewer stream possible for the given session with the seek
 246  * type. Three counters *can* be return which are in order the total amount of
 247  * viewer stream of the session, the number of unsent stream and the number of
 248  * stream created. Those counters can be NULL and thus will be ignored.
 249  *
 250  * Return 0 on success or else a negative value.
 251  */
 252 static
 253 int make_viewer_streams(struct relay_session *session,
 254                 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
 255                 uint32_t *nb_created)
 256 {
 257         int ret;
 258         struct lttng_ht_iter iter;
 259         struct ctf_trace *ctf_trace;
 260
 261         assert(session);
 262
 263         /*
 264          * This is to make sure we create viewer streams for a full received
 265          * channel. For instance, if we have 8 streams for a channel that are
 266          * concurrently being flagged ready, we can end up creating just a subset
 267          * of the 8 streams (the ones that are flagged). This lock avoids this
 268          * limbo state.
 269          */
 270         pthread_mutex_lock(&session->viewer_ready_lock);
 271
 272         /*
 273          * Create viewer streams for relay streams that are ready to be used for a
 274          * the given session id only.
 275          */
 276         rcu_read_lock();
 277         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 278                         node.node) {
 279                 struct relay_stream *stream;
 280
 281                 health_code_update();
 282
 283                 if (ctf_trace->invalid_flag) {
 284                         continue;
 285                 }
 286
 287                 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
 288                         struct relay_viewer_stream *vstream;
 289
 290                         if (!stream->viewer_ready) {
 291                                 continue;
 292                         }
 293
 294                         vstream = viewer_stream_find_by_id(stream->stream_handle);
 295                         if (!vstream) {
 296                                 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
 297                                 if (!vstream) {
 298                                         ret = -1;
 299                                         goto error_unlock;
 300                                 }
 301                                 /* Acquire reference to ctf_trace. */
 302                                 ctf_trace_get_ref(ctf_trace);
 303
 304                                 if (nb_created) {
 305                                         /* Update number of created stream counter. */
 306                                         (*nb_created)++;
 307                                 }
 308                         } else if (!vstream->sent_flag && nb_unsent) {
 309                                 /* Update number of unsent stream counter. */
 310                                 (*nb_unsent)++;
 311                         }
 312                         /* Update number of total stream counter. */
 313                         if (nb_total) {
 314                                 (*nb_total)++;
 315                         }
 316                 }
 317         }
 318
 319         ret = 0;
 320
 321 error_unlock:
 322         rcu_read_unlock();
 323         pthread_mutex_unlock(&session->viewer_ready_lock);
 324         return ret;
 325 }
 326
 327 /*
 328  * Write to writable pipe used to notify a thread.
 329  */
 330 static
 331 int notify_thread_pipe(int wpipe)
 332 {
 333         ssize_t ret;
 334
 335         ret = lttng_write(wpipe, "!", 1);
 336         if (ret < 1) {
 337                 PERROR("write poll pipe");
 338         }
 339
 340         return (int) ret;
 341 }
 342
 343 /*
 344  * Stop all threads by closing the thread quit pipe.
 345  */
 346 static
 347 void stop_threads(void)
 348 {
 349         int ret;
 350
 351         /* Stopping all threads */
 352         DBG("Terminating all live threads");
 353         ret = notify_thread_pipe(thread_quit_pipe[1]);
 354         if (ret < 0) {
 355                 ERR("write error on thread quit pipe");
 356         }
 357
 358         /* Dispatch thread */
 359         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 360         futex_nto1_wake(&viewer_conn_queue.futex);
 361 }
 362
 363 /*
 364  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 365  */
 366 static
 367 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 368 {
 369         int ret;
 370
 371         if (events == NULL || size == 0) {
 372                 ret = -1;
 373                 goto error;
 374         }
 375
 376         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 377         if (ret < 0) {
 378                 goto error;
 379         }
 380
 381         /* Add quit pipe */
 382         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 383         if (ret < 0) {
 384                 goto error;
 385         }
 386
 387         return 0;
 388
 389 error:
 390         return ret;
 391 }
 392
 393 /*
 394  * Check if the thread quit pipe was triggered.
 395  *
 396  * Return 1 if it was triggered else 0;
 397  */
 398 static
 399 int check_thread_quit_pipe(int fd, uint32_t events)
 400 {
 401         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 402                 return 1;
 403         }
 404
 405         return 0;
 406 }
 407
 408 /*
 409  * Create and init socket from uri.
 410  */
 411 static
 412 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 413 {
 414         int ret;
 415         struct lttcomm_sock *sock = NULL;
 416
 417         sock = lttcomm_alloc_sock_from_uri(uri);
 418         if (sock == NULL) {
 419                 ERR("Allocating socket");
 420                 goto error;
 421         }
 422
 423         ret = lttcomm_create_sock(sock);
 424         if (ret < 0) {
 425                 goto error;
 426         }
 427         DBG("Listening on sock %d for live", sock->fd);
 428
 429         ret = sock->ops->bind(sock);
 430         if (ret < 0) {
 431                 goto error;
 432         }
 433
 434         ret = sock->ops->listen(sock, -1);
 435         if (ret < 0) {
 436                 goto error;
 437
 438         }
 439
 440         return sock;
 441
 442 error:
 443         if (sock) {
 444                 lttcomm_destroy_sock(sock);
 445         }
 446         return NULL;
 447 }
 448
 449 /*
 450  * This thread manages the listening for new connections on the network
 451  */
 452 static
 453 void *thread_listener(void *data)
 454 {
 455         int i, ret, pollfd, err = -1;
 456         uint32_t revents, nb_fd;
 457         struct lttng_poll_event events;
 458         struct lttcomm_sock *live_control_sock;
 459
 460         DBG("[thread] Relay live listener started");
 461
 462         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 463
 464         health_code_update();
 465
 466         live_control_sock = init_socket(live_uri);
 467         if (!live_control_sock) {
 468                 goto error_sock_control;
 469         }
 470
 471         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 472         ret = create_thread_poll_set(&events, 2);
 473         if (ret < 0) {
 474                 goto error_create_poll;
 475         }
 476
 477         /* Add the control socket */
 478         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 479         if (ret < 0) {
 480                 goto error_poll_add;
 481         }
 482
 483         lttng_relay_notify_ready();
 484
 485         if (testpoint(relayd_thread_live_listener)) {
 486                 goto error_testpoint;
 487         }
 488
 489         while (1) {
 490                 health_code_update();
 491
 492                 DBG("Listener accepting live viewers connections");
 493
 494 restart:
 495                 health_poll_entry();
 496                 ret = lttng_poll_wait(&events, -1);
 497                 health_poll_exit();
 498                 if (ret < 0) {
 499                         /*
 500                          * Restart interrupted system call.
 501                          */
 502                         if (errno == EINTR) {
 503                                 goto restart;
 504                         }
 505                         goto error;
 506                 }
 507                 nb_fd = ret;
 508
 509                 DBG("Relay new viewer connection received");
 510                 for (i = 0; i < nb_fd; i++) {
 511                         health_code_update();
 512
 513                         /* Fetch once the poll data */
 514                         revents = LTTNG_POLL_GETEV(&events, i);
 515                         pollfd = LTTNG_POLL_GETFD(&events, i);
 516
 517                         /* Thread quit pipe has been closed. Killing thread. */
 518                         ret = check_thread_quit_pipe(pollfd, revents);
 519                         if (ret) {
 520                                 err = 0;
 521                                 goto exit;
 522                         }
 523
 524                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 525                                 ERR("socket poll error");
 526                                 goto error;
 527                         } else if (revents & LPOLLIN) {
 528                                 /*
 529                                  * Get allocated in this thread, enqueued to a global queue,
 530                                  * dequeued and freed in the worker thread.
 531                                  */
 532                                 int val = 1;
 533                                 struct relay_connection *new_conn;
 534                                 struct lttcomm_sock *newsock;
 535
 536                                 new_conn = connection_create();
 537                                 if (!new_conn) {
 538                                         goto error;
 539                                 }
 540
 541                                 newsock = live_control_sock->ops->accept(live_control_sock);
 542                                 if (!newsock) {
 543                                         PERROR("accepting control sock");
 544                                         connection_free(new_conn);
 545                                         goto error;
 546                                 }
 547                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 548
 549                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 550                                                 sizeof(val));
 551                                 if (ret < 0) {
 552                                         PERROR("setsockopt inet");
 553                                         lttcomm_destroy_sock(newsock);
 554                                         connection_free(new_conn);
 555                                         goto error;
 556                                 }
 557                                 new_conn->sock = newsock;
 558
 559                                 /* Enqueue request for the dispatcher thread. */
 560                                 cds_wfq_enqueue(&viewer_conn_queue.queue, &new_conn->qnode);
 561
 562                                 /*
 563                                  * Wake the dispatch queue futex. Implicit memory barrier with
 564                                  * the exchange in cds_wfq_enqueue.
 565                                  */
 566                                 futex_nto1_wake(&viewer_conn_queue.futex);
 567                         }
 568                 }
 569         }
 570
 571 exit:
 572 error:
 573 error_poll_add:
 574 error_testpoint:
 575         lttng_poll_clean(&events);
 576 error_create_poll:
 577         if (live_control_sock->fd >= 0) {
 578                 ret = live_control_sock->ops->close(live_control_sock);
 579                 if (ret) {
 580                         PERROR("close");
 581                 }
 582         }
 583         lttcomm_destroy_sock(live_control_sock);
 584 error_sock_control:
 585         if (err) {
 586                 health_error();
 587                 DBG("Live viewer listener thread exited with error");
 588         }
 589         health_unregister(health_relayd);
 590         DBG("Live viewer listener thread cleanup complete");
 591         stop_threads();
 592         return NULL;
 593 }
 594
 595 /*
 596  * This thread manages the dispatching of the requests to worker threads
 597  */
 598 static
 599 void *thread_dispatcher(void *data)
 600 {
 601         int err = -1;
 602         ssize_t ret;
 603         struct cds_wfq_node *node;
 604         struct relay_connection *conn = NULL;
 605
 606         DBG("[thread] Live viewer relay dispatcher started");
 607
 608         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 609
 610         if (testpoint(relayd_thread_live_dispatcher)) {
 611                 goto error_testpoint;
 612         }
 613
 614         health_code_update();
 615
 616         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 617                 health_code_update();
 618
 619                 /* Atomically prepare the queue futex */
 620                 futex_nto1_prepare(&viewer_conn_queue.futex);
 621
 622                 do {
 623                         health_code_update();
 624
 625                         /* Dequeue commands */
 626                         node = cds_wfq_dequeue_blocking(&viewer_conn_queue.queue);
 627                         if (node == NULL) {
 628                                 DBG("Woken up but nothing in the live-viewer "
 629                                                 "relay command queue");
 630                                 /* Continue thread execution */
 631                                 break;
 632                         }
 633                         conn = caa_container_of(node, struct relay_connection, qnode);
 634                         DBG("Dispatching viewer request waiting on sock %d",
 635                                         conn->sock->fd);
 636
 637                         /*
 638                          * Inform worker thread of the new request. This call is blocking
 639                          * so we can be assured that the data will be read at some point in
 640                          * time or wait to the end of the world :)
 641                          */
 642                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 643                         if (ret < 0) {
 644                                 PERROR("write conn pipe");
 645                                 connection_destroy(conn);
 646                                 goto error;
 647                         }
 648                 } while (node != NULL);
 649
 650                 /* Futex wait on queue. Blocking call on futex() */
 651                 health_poll_entry();
 652                 futex_nto1_wait(&viewer_conn_queue.futex);
 653                 health_poll_exit();
 654         }
 655
 656         /* Normal exit, no error */
 657         err = 0;
 658
 659 error:
 660 error_testpoint:
 661         if (err) {
 662                 health_error();
 663                 ERR("Health error occurred in %s", __func__);
 664         }
 665         health_unregister(health_relayd);
 666         DBG("Live viewer dispatch thread dying");
 667         stop_threads();
 668         return NULL;
 669 }
 670
 671 /*
 672  * Establish connection with the viewer and check the versions.
 673  *
 674  * Return 0 on success or else negative value.
 675  */
 676 static
 677 int viewer_connect(struct relay_connection *conn)
 678 {
 679         int ret;
 680         struct lttng_viewer_connect reply, msg;
 681
 682         assert(conn);
 683
 684         conn->version_check_done = 1;
 685
 686         health_code_update();
 687
 688         DBG("Viewer is establishing a connection to the relayd.");
 689
 690         ret = recv_request(conn->sock, &msg, sizeof(msg));
 691         if (ret < 0) {
 692                 goto end;
 693         }
 694
 695         health_code_update();
 696
 697         memset(&reply, 0, sizeof(reply));
 698         reply.major = RELAYD_VERSION_COMM_MAJOR;
 699         reply.minor = RELAYD_VERSION_COMM_MINOR;
 700
 701         /* Major versions must be the same */
 702         if (reply.major != be32toh(msg.major)) {
 703                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 704                                 reply.major, be32toh(msg.major));
 705                 ret = -1;
 706                 goto end;
 707         }
 708
 709         conn->major = reply.major;
 710         /* We adapt to the lowest compatible version */
 711         if (reply.minor <= be32toh(msg.minor)) {
 712                 conn->minor = reply.minor;
 713         } else {
 714                 conn->minor = be32toh(msg.minor);
 715         }
 716
 717         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 718                 conn->type = RELAY_VIEWER_COMMAND;
 719         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 720                 conn->type = RELAY_VIEWER_NOTIFICATION;
 721         } else {
 722                 ERR("Unknown connection type : %u", be32toh(msg.type));
 723                 ret = -1;
 724                 goto end;
 725         }
 726
 727         reply.major = htobe32(reply.major);
 728         reply.minor = htobe32(reply.minor);
 729         if (conn->type == RELAY_VIEWER_COMMAND) {
 730                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 731         }
 732
 733         health_code_update();
 734
 735         ret = send_response(conn->sock, &reply, sizeof(reply));
 736         if (ret < 0) {
 737                 goto end;
 738         }
 739
 740         health_code_update();
 741
 742         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 743         ret = 0;
 744
 745 end:
 746         return ret;
 747 }
 748
 749 /*
 750  * Send the viewer the list of current sessions.
 751  *
 752  * Return 0 on success or else a negative value.
 753  */
 754 static
 755 int viewer_list_sessions(struct relay_connection *conn)
 756 {
 757         int ret;
 758         struct lttng_viewer_list_sessions session_list;
 759         unsigned long count;
 760         long approx_before, approx_after;
 761         struct lttng_ht_iter iter;
 762         struct lttng_viewer_session send_session;
 763         struct relay_session *session;
 764
 765         DBG("List sessions received");
 766
 767         rcu_read_lock();
 768         cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
 769                         &approx_after);
 770         session_list.sessions_count = htobe32(count);
 771
 772         health_code_update();
 773
 774         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 775         if (ret < 0) {
 776                 goto end_unlock;
 777         }
 778
 779         health_code_update();
 780
 781         cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
 782                         session_n.node) {
 783                 health_code_update();
 784
 785                 strncpy(send_session.session_name, session->session_name,
 786                                 sizeof(send_session.session_name));
 787                 strncpy(send_session.hostname, session->hostname,
 788                                 sizeof(send_session.hostname));
 789                 send_session.id = htobe64(session->id);
 790                 send_session.live_timer = htobe32(session->live_timer);
 791                 send_session.clients = htobe32(session->viewer_refcount);
 792                 send_session.streams = htobe32(session->stream_count);
 793
 794                 health_code_update();
 795
 796                 ret = send_response(conn->sock, &send_session, sizeof(send_session));
 797                 if (ret < 0) {
 798                         goto end_unlock;
 799                 }
 800         }
 801         health_code_update();
 802
 803         rcu_read_unlock();
 804         ret = 0;
 805         goto end;
 806
 807 end_unlock:
 808         rcu_read_unlock();
 809
 810 end:
 811         return ret;
 812 }
 813
 814 /*
 815  * Check if a connection is attached to a session.
 816  * Return 1 if attached, 0 if not attached, a negative value on error.
 817  */
 818 static
 819 int session_attached(struct relay_connection *conn, uint64_t session_id)
 820 {
 821         struct relay_session *session;
 822         int found = 0;
 823
 824         if (!conn->viewer_session) {
 825                 goto end;
 826         }
 827         cds_list_for_each_entry(session,
 828                         &conn->viewer_session->sessions_head,
 829                         viewer_session_list) {
 830                 if (session->id == session_id) {
 831                         found = 1;
 832                         goto end;
 833                 }
 834         }
 835
 836 end:
 837         return found;
 838 }
 839
 840 /*
 841  * Delete all streams for a specific session ID.
 842  */
 843 static void destroy_viewer_streams_by_session(struct relay_session *session)
 844 {
 845         struct relay_viewer_stream *stream;
 846         struct lttng_ht_iter iter;
 847
 848         assert(session);
 849
 850         rcu_read_lock();
 851         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
 852                         stream_n.node) {
 853                 struct ctf_trace *ctf_trace;
 854
 855                 health_code_update();
 856                 if (stream->session_id != session->id) {
 857                         continue;
 858                 }
 859
 860                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 861                                 stream->path_name);
 862                 assert(ctf_trace);
 863
 864                 viewer_stream_delete(stream);
 865
 866                 if (stream->metadata_flag) {
 867                         ctf_trace->metadata_sent = 0;
 868                         ctf_trace->viewer_metadata_stream = NULL;
 869                 }
 870
 871                 viewer_stream_destroy(ctf_trace, stream);
 872         }
 873         rcu_read_unlock();
 874 }
 875
 876 static void try_destroy_streams(struct relay_session *session)
 877 {
 878         struct ctf_trace *ctf_trace;
 879         struct lttng_ht_iter iter;
 880
 881         assert(session);
 882
 883         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 884                         node.node) {
 885                 /* Attempt to destroy the ctf trace of that session. */
 886                 ctf_trace_try_destroy(session, ctf_trace);
 887         }
 888 }
 889
 890 /*
 891  * Cleanup a session.
 892  */
 893 static void cleanup_session(struct relay_connection *conn,
 894                 struct relay_session *session)
 895 {
 896         /*
 897          * Very important that this is done before destroying the session so we
 898          * can put back every viewer stream reference from the ctf_trace.
 899          */
 900         destroy_viewer_streams_by_session(session);
 901         try_destroy_streams(session);
 902         cds_list_del(&session->viewer_session_list);
 903         session_viewer_try_destroy(conn->sessions_ht, session);
 904 }
 905
 906 /*
 907  * Send the viewer the list of current sessions.
 908  */
 909 static
 910 int viewer_get_new_streams(struct relay_connection *conn)
 911 {
 912         int ret, send_streams = 0;
 913         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
 914         struct lttng_viewer_new_streams_request request;
 915         struct lttng_viewer_new_streams_response response;
 916         struct relay_session *session;
 917         uint64_t session_id;
 918
 919         assert(conn);
 920
 921         DBG("Get new streams received");
 922
 923         health_code_update();
 924
 925         /* Receive the request from the connected client. */
 926         ret = recv_request(conn->sock, &request, sizeof(request));
 927         if (ret < 0) {
 928                 goto error;
 929         }
 930         session_id = be64toh(request.session_id);
 931
 932         health_code_update();
 933
 934         memset(&response, 0, sizeof(response));
 935
 936         rcu_read_lock();
 937         session = session_find_by_id(conn->sessions_ht, session_id);
 938         if (!session) {
 939                 DBG("Relay session %" PRIu64 " not found", session_id);
 940                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 941                 goto send_reply;
 942         }
 943
 944         if (!session_attached(conn, session_id)) {
 945                 send_streams = 0;
 946                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 947                 goto send_reply;
 948         }
 949
 950         send_streams = 1;
 951         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 952
 953         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
 954                         &nb_created);
 955         if (ret < 0) {
 956                 goto end_unlock;
 957         }
 958         /* Only send back the newly created streams with the unsent ones. */
 959         nb_streams = nb_created + nb_unsent;
 960         response.streams_count = htobe32(nb_streams);
 961
 962         /*
 963          * If the session is closed and we have no new streams to send,
 964          * it means that the viewer has already received the whole trace
 965          * for this session and should now close it.
 966          */
 967         if (nb_streams == 0 && session->close_flag) {
 968                 send_streams = 0;
 969                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 970                 /*
 971                  * Remove the session from the attached list of the connection
 972                  * and try to destroy it.
 973                  */
 974                 cds_list_del(&session->viewer_session_list);
 975                 cleanup_session(conn, session);
 976                 goto send_reply;
 977         }
 978
 979 send_reply:
 980         health_code_update();
 981         ret = send_response(conn->sock, &response, sizeof(response));
 982         if (ret < 0) {
 983                 goto end_unlock;
 984         }
 985         health_code_update();
 986
 987         /*
 988          * Unknown or empty session, just return gracefully, the viewer knows what
 989          * is happening.
 990          */
 991         if (!send_streams || !nb_streams) {
 992                 ret = 0;
 993                 goto end_unlock;
 994         }
 995
 996         /*
 997          * Send stream and *DON'T* ignore the sent flag so every viewer streams
 998          * that were not sent from that point will be sent to the viewer.
 999          */
1000         ret = send_viewer_streams(conn->sock, session, 0);
1001         if (ret < 0) {
1002                 goto end_unlock;
1003         }
1004
1005 end_unlock:
1006         rcu_read_unlock();
1007 error:
1008         return ret;
1009 }
1010
1011 /*
1012  * Send the viewer the list of current sessions.
1013  */
1014 static
1015 int viewer_attach_session(struct relay_connection *conn)
1016 {
1017         int send_streams = 0;
1018         ssize_t ret;
1019         uint32_t nb_streams = 0;
1020         enum lttng_viewer_seek seek_type;
1021         struct lttng_viewer_attach_session_request request;
1022         struct lttng_viewer_attach_session_response response;
1023         struct relay_session *session;
1024
1025         assert(conn);
1026
1027         health_code_update();
1028
1029         /* Receive the request from the connected client. */
1030         ret = recv_request(conn->sock, &request, sizeof(request));
1031         if (ret < 0) {
1032                 goto error;
1033         }
1034
1035         health_code_update();
1036
1037         memset(&response, 0, sizeof(response));
1038
1039         if (!conn->viewer_session) {
1040                 DBG("Client trying to attach before creating a live viewer session");
1041                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1042                 goto send_reply;
1043         }
1044
1045         rcu_read_lock();
1046         session = session_find_by_id(conn->sessions_ht,
1047                         be64toh(request.session_id));
1048         if (!session) {
1049                 DBG("Relay session %" PRIu64 " not found",
1050                                 be64toh(request.session_id));
1051                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1052                 goto send_reply;
1053         }
1054         session_viewer_attach(session);
1055         DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1056
1057         if (uatomic_read(&session->viewer_refcount) > 1) {
1058                 DBG("Already a viewer attached");
1059                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1060                 session_viewer_detach(session);
1061                 goto send_reply;
1062         } else if (session->live_timer == 0) {
1063                 DBG("Not live session");
1064                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1065                 goto send_reply;
1066         } else {
1067                 send_streams = 1;
1068                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1069                 cds_list_add(&session->viewer_session_list,
1070                                 &conn->viewer_session->sessions_head);
1071         }
1072
1073         switch (be32toh(request.seek)) {
1074         case LTTNG_VIEWER_SEEK_BEGINNING:
1075         case LTTNG_VIEWER_SEEK_LAST:
1076                 seek_type = be32toh(request.seek);
1077                 break;
1078         default:
1079                 ERR("Wrong seek parameter");
1080                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1081                 send_streams = 0;
1082                 goto send_reply;
1083         }
1084
1085         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1086         if (ret < 0) {
1087                 goto end_unlock;
1088         }
1089         response.streams_count = htobe32(nb_streams);
1090
1091 send_reply:
1092         health_code_update();
1093         ret = send_response(conn->sock, &response, sizeof(response));
1094         if (ret < 0) {
1095                 goto end_unlock;
1096         }
1097         health_code_update();
1098
1099         /*
1100          * Unknown or empty session, just return gracefully, the viewer knows what
1101          * is happening.
1102          */
1103         if (!send_streams || !nb_streams) {
1104                 ret = 0;
1105                 goto end_unlock;
1106         }
1107
1108         /* Send stream and ignore the sent flag. */
1109         ret = send_viewer_streams(conn->sock, session, 1);
1110         if (ret < 0) {
1111                 goto end_unlock;
1112         }
1113
1114 end_unlock:
1115         rcu_read_unlock();
1116 error:
1117         return ret;
1118 }
1119
1120 /*
1121  * Open the index file if needed for the given vstream.
1122  *
1123  * If an index file is successfully opened, the index_read_fd of the stream is
1124  * set with it.
1125  *
1126  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1127  */
1128 static int try_open_index(struct relay_viewer_stream *vstream,
1129                 struct relay_stream *rstream)
1130 {
1131         int ret = 0;
1132
1133         assert(vstream);
1134         assert(rstream);
1135
1136         if (vstream->index_read_fd >= 0) {
1137                 goto end;
1138         }
1139
1140         /*
1141          * First time, we open the index file and at least one index is ready.  The
1142          * race between the read and write of the total_index_received is
1143          * acceptable here since the client will be notified to simply come back
1144          * and get the next index.
1145          */
1146         if (rstream->total_index_received <= 0) {
1147                 ret = -ENOENT;
1148                 goto end;
1149         }
1150         ret = index_open(vstream->path_name, vstream->channel_name,
1151                         vstream->tracefile_count, vstream->tracefile_count_current);
1152         if (ret >= 0) {
1153                 vstream->index_read_fd = ret;
1154                 ret = 0;
1155                 goto end;
1156         }
1157
1158 end:
1159         return ret;
1160 }
1161
1162 /*
1163  * Check the status of the index for the given stream. This function updates
1164  * the index structure if needed and can destroy the vstream also for the HUP
1165  * situation.
1166  *
1167  * Return 0 means that we can proceed with the index. A value of 1 means that
1168  * the index has been updated and is ready to be send to the client. A negative
1169  * value indicates an error that can't be handled.
1170  */
1171 static int check_index_status(struct relay_viewer_stream *vstream,
1172                 struct relay_stream *rstream, struct ctf_trace *trace,
1173                 struct lttng_viewer_index *index)
1174 {
1175         int ret;
1176
1177         assert(vstream);
1178         assert(rstream);
1179         assert(index);
1180         assert(trace);
1181
1182         if (!rstream->close_flag) {
1183                 /* Rotate on abort (overwrite). */
1184                 if (vstream->abort_flag) {
1185                         DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1186                                         vstream->stream_handle);
1187                         ret = viewer_stream_rotate(vstream, rstream);
1188                         if (ret < 0) {
1189                                 goto error;
1190                         } else if (ret == 1) {
1191                                 /* EOF */
1192                                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1193                                 goto hup;
1194                         }
1195                         /* ret == 0 means successful so we continue. */
1196                 }
1197
1198                 /* Check if we are in the same trace file at this point. */
1199                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1200                         if (rstream->beacon_ts_end != -1ULL &&
1201                                         vstream->last_sent_index == rstream->total_index_received) {
1202                                 /*
1203                                  * We've received a synchronization beacon and the last index
1204                                  * available has been sent, the index for now is inactive.
1205                                  */
1206                                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1207                                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1208                                 goto index_ready;
1209                         } else if (rstream->total_index_received <= vstream->last_sent_index
1210                                         && !vstream->close_write_flag) {
1211                                 /*
1212                                  * Reader and writer are working in the same tracefile, so we care
1213                                  * about the number of index received and sent. Otherwise, we read
1214                                  * up to EOF.
1215                                  */
1216                                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1217                                 goto index_ready;
1218                         }
1219                 }
1220                 /* Nothing to do with the index, continue with it. */
1221                 ret = 0;
1222         } else if (rstream->close_flag && vstream->close_write_flag &&
1223                         vstream->total_index_received == vstream->last_sent_index) {
1224                 /* Last index sent and current tracefile closed in write */
1225                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1226                 goto hup;
1227         } else {
1228                 vstream->close_write_flag = 1;
1229                 ret = 0;
1230         }
1231
1232 error:
1233         return ret;
1234
1235 hup:
1236         viewer_stream_delete(vstream);
1237         viewer_stream_destroy(trace, vstream);
1238 index_ready:
1239         return 1;
1240 }
1241
1242 /*
1243  * Send the next index for a stream.
1244  *
1245  * Return 0 on success or else a negative value.
1246  */
1247 static
1248 int viewer_get_next_index(struct relay_connection *conn)
1249 {
1250         int ret;
1251         ssize_t read_ret;
1252         struct lttng_viewer_get_next_index request_index;
1253         struct lttng_viewer_index viewer_index;
1254         struct ctf_packet_index packet_index;
1255         struct relay_viewer_stream *vstream;
1256         struct relay_stream *rstream;
1257         struct ctf_trace *ctf_trace;
1258         struct relay_session *session;
1259
1260         assert(conn);
1261
1262         DBG("Viewer get next index");
1263
1264         health_code_update();
1265
1266         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1267         if (ret < 0) {
1268                 goto end;
1269         }
1270         health_code_update();
1271
1272         rcu_read_lock();
1273         vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1274         if (!vstream) {
1275                 ret = -1;
1276                 goto end_unlock;
1277         }
1278
1279         session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1280         if (!session) {
1281                 ret = -1;
1282                 goto end_unlock;
1283         }
1284
1285         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1286         assert(ctf_trace);
1287
1288         memset(&viewer_index, 0, sizeof(viewer_index));
1289
1290         /*
1291          * The viewer should not ask for index on metadata stream.
1292          */
1293         if (vstream->metadata_flag) {
1294                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1295                 goto send_reply;
1296         }
1297
1298         rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1299         assert(rstream);
1300
1301         /* Try to open an index if one is needed for that stream. */
1302         ret = try_open_index(vstream, rstream);
1303         if (ret < 0) {
1304                 if (ret == -ENOENT) {
1305                         /*
1306                          * The index is created only when the first data packet arrives, it
1307                          * might not be ready at the beginning of the session
1308                          */
1309                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1310                 } else {
1311                         /* Unhandled error. */
1312                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1313                 }
1314                 goto send_reply;
1315         }
1316
1317         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1318         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1319         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1320         if (ret < 0) {
1321                 goto end;
1322         } else if (ret == 1) {
1323                 /*
1324                  * This means the viewer index data structure has been populated by the
1325                  * check call thus we now send back the reply to the client.
1326                  */
1327                 goto send_reply;
1328         }
1329         /* At this point, ret MUST be 0 thus we continue with the get. */
1330         assert(!ret);
1331
1332         if (!ctf_trace->metadata_received ||
1333                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1334                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1335         }
1336
1337         ret = check_new_streams(conn);
1338         if (ret < 0) {
1339                 goto end_unlock;
1340         } else if (ret == 1) {
1341                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1342         }
1343
1344         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1345         pthread_mutex_lock(&vstream->overwrite_lock);
1346         if (vstream->abort_flag) {
1347                 /* The file is being overwritten by the writer, we cannot use it. */
1348                 pthread_mutex_unlock(&vstream->overwrite_lock);
1349                 ret = viewer_stream_rotate(vstream, rstream);
1350                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1351                 if (ret < 0) {
1352                         goto end_unlock;
1353                 } else if (ret == 1) {
1354                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1355                         viewer_stream_delete(vstream);
1356                         viewer_stream_destroy(ctf_trace, vstream);
1357                 } else {
1358                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1359                 }
1360                 goto send_reply;
1361         }
1362
1363         read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1364                         sizeof(packet_index));
1365         pthread_mutex_unlock(&vstream->overwrite_lock);
1366         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1367         if (read_ret < 0) {
1368                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1369                 viewer_stream_delete(vstream);
1370                 viewer_stream_destroy(ctf_trace, vstream);
1371                 goto send_reply;
1372         } else if (read_ret < sizeof(packet_index)) {
1373                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1374                 if (vstream->close_write_flag) {
1375                         ret = viewer_stream_rotate(vstream, rstream);
1376                         if (ret < 0) {
1377                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1378                                 goto end_unlock;
1379                         } else if (ret == 1) {
1380                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1381                                 viewer_stream_delete(vstream);
1382                                 viewer_stream_destroy(ctf_trace, vstream);
1383                         } else {
1384                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1385                         }
1386                 } else {
1387                         ERR("Relay reading index file %d", vstream->index_read_fd);
1388                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1389                 }
1390                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1391                 goto send_reply;
1392         } else {
1393                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1394                 vstream->last_sent_index++;
1395         }
1396
1397         /*
1398          * Indexes are stored in big endian, no need to switch before sending.
1399          */
1400         viewer_index.offset = packet_index.offset;
1401         viewer_index.packet_size = packet_index.packet_size;
1402         viewer_index.content_size = packet_index.content_size;
1403         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1404         viewer_index.timestamp_end = packet_index.timestamp_end;
1405         viewer_index.events_discarded = packet_index.events_discarded;
1406         viewer_index.stream_id = packet_index.stream_id;
1407
1408 send_reply:
1409         viewer_index.flags = htobe32(viewer_index.flags);
1410         health_code_update();
1411
1412         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1413         if (ret < 0) {
1414                 goto end_unlock;
1415         }
1416         health_code_update();
1417
1418         DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1419                         vstream->last_sent_index, vstream->stream_handle);
1420
1421 end_unlock:
1422         rcu_read_unlock();
1423
1424 end:
1425         return ret;
1426 }
1427
1428 /*
1429  * Send the next index for a stream
1430  *
1431  * Return 0 on success or else a negative value.
1432  */
1433 static
1434 int viewer_get_packet(struct relay_connection *conn)
1435 {
1436         int ret, send_data = 0;
1437         char *data = NULL;
1438         uint32_t len = 0;
1439         ssize_t read_len;
1440         struct lttng_viewer_get_packet get_packet_info;
1441         struct lttng_viewer_trace_packet reply;
1442         struct relay_viewer_stream *stream;
1443         struct relay_session *session;
1444         struct ctf_trace *ctf_trace;
1445
1446         assert(conn);
1447
1448         DBG2("Relay get data packet");
1449
1450         health_code_update();
1451
1452         ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1453         if (ret < 0) {
1454                 goto end;
1455         }
1456         health_code_update();
1457
1458         /* From this point on, the error label can be reached. */
1459         memset(&reply, 0, sizeof(reply));
1460
1461         rcu_read_lock();
1462         stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1463         if (!stream) {
1464                 goto error;
1465         }
1466
1467         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1468         if (!session) {
1469                 ret = -1;
1470                 goto error;
1471         }
1472
1473         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1474                         stream->path_name);
1475         assert(ctf_trace);
1476
1477         /*
1478          * First time we read this stream, we need open the tracefile, we should
1479          * only arrive here if an index has already been sent to the viewer, so the
1480          * tracefile must exist, if it does not it is a fatal error.
1481          */
1482         if (stream->read_fd < 0) {
1483                 char fullpath[PATH_MAX];
1484
1485                 if (stream->tracefile_count > 0) {
1486                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1487                                         stream->channel_name,
1488                                         stream->tracefile_count_current);
1489                 } else {
1490                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1491                                         stream->channel_name);
1492                 }
1493                 if (ret < 0) {
1494                         goto error;
1495                 }
1496                 ret = open(fullpath, O_RDONLY);
1497                 if (ret < 0) {
1498                         PERROR("Relay opening trace file");
1499                         goto error;
1500                 }
1501                 stream->read_fd = ret;
1502         }
1503
1504         if (!ctf_trace->metadata_received ||
1505                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1506                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1507                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1508                 goto send_reply;
1509         }
1510
1511         ret = check_new_streams(conn);
1512         if (ret < 0) {
1513                 goto end_unlock;
1514         } else if (ret == 1) {
1515                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1516                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1517                 goto send_reply;
1518         }
1519
1520         len = be32toh(get_packet_info.len);
1521         data = zmalloc(len);
1522         if (!data) {
1523                 PERROR("relay data zmalloc");
1524                 goto error;
1525         }
1526
1527         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1528         if (ret < 0) {
1529                 /*
1530                  * If the read fd was closed by the streaming side, the
1531                  * abort_flag will be set to 1, otherwise it is an error.
1532                  */
1533                 if (stream->abort_flag == 0) {
1534                         PERROR("lseek");
1535                         goto error;
1536                 }
1537                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1538                 goto send_reply;
1539         }
1540         read_len = lttng_read(stream->read_fd, data, len);
1541         if (read_len < len) {
1542                 /*
1543                  * If the read fd was closed by the streaming side, the
1544                  * abort_flag will be set to 1, otherwise it is an error.
1545                  */
1546                 if (stream->abort_flag == 0) {
1547                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1548                                         stream->read_fd,
1549                                         be64toh(get_packet_info.offset));
1550                         goto error;
1551                 } else {
1552                         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1553                         goto send_reply;
1554                 }
1555         }
1556         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1557         reply.len = htobe32(len);
1558         send_data = 1;
1559         goto send_reply;
1560
1561 error:
1562         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1563
1564 send_reply:
1565         reply.flags = htobe32(reply.flags);
1566
1567         health_code_update();
1568
1569         ret = send_response(conn->sock, &reply, sizeof(reply));
1570         if (ret < 0) {
1571                 goto end_unlock;
1572         }
1573         health_code_update();
1574
1575         if (send_data) {
1576                 health_code_update();
1577                 ret = send_response(conn->sock, data, len);
1578                 if (ret < 0) {
1579                         goto end_unlock;
1580                 }
1581                 health_code_update();
1582         }
1583
1584         DBG("Sent %u bytes for stream %" PRIu64, len,
1585                         be64toh(get_packet_info.stream_id));
1586
1587 end_unlock:
1588         free(data);
1589         rcu_read_unlock();
1590
1591 end:
1592         return ret;
1593 }
1594
1595 /*
1596  * Send the session's metadata
1597  *
1598  * Return 0 on success else a negative value.
1599  */
1600 static
1601 int viewer_get_metadata(struct relay_connection *conn)
1602 {
1603         int ret = 0;
1604         ssize_t read_len;
1605         uint64_t len = 0;
1606         char *data = NULL;
1607         struct lttng_viewer_get_metadata request;
1608         struct lttng_viewer_metadata_packet reply;
1609         struct relay_viewer_stream *stream;
1610         struct ctf_trace *ctf_trace;
1611         struct relay_session *session;
1612
1613         assert(conn);
1614
1615         DBG("Relay get metadata");
1616
1617         health_code_update();
1618
1619         ret = recv_request(conn->sock, &request, sizeof(request));
1620         if (ret < 0) {
1621                 goto end;
1622         }
1623         health_code_update();
1624
1625         memset(&reply, 0, sizeof(reply));
1626
1627         rcu_read_lock();
1628         stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1629         if (!stream || !stream->metadata_flag) {
1630                 ERR("Invalid metadata stream");
1631                 goto error;
1632         }
1633
1634         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1635         if (!session) {
1636                 ret = -1;
1637                 goto error;
1638         }
1639
1640         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1641                         stream->path_name);
1642         assert(ctf_trace);
1643         assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1644
1645         len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1646         if (len == 0) {
1647                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1648                 goto send_reply;
1649         }
1650
1651         /* first time, we open the metadata file */
1652         if (stream->read_fd < 0) {
1653                 char fullpath[PATH_MAX];
1654
1655                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1656                                 stream->channel_name);
1657                 if (ret < 0) {
1658                         goto error;
1659                 }
1660                 ret = open(fullpath, O_RDONLY);
1661                 if (ret < 0) {
1662                         PERROR("Relay opening metadata file");
1663                         goto error;
1664                 }
1665                 stream->read_fd = ret;
1666         }
1667
1668         reply.len = htobe64(len);
1669         data = zmalloc(len);
1670         if (!data) {
1671                 PERROR("viewer metadata zmalloc");
1672                 goto error;
1673         }
1674
1675         read_len = lttng_read(stream->read_fd, data, len);
1676         if (read_len < len) {
1677                 PERROR("Relay reading metadata file");
1678                 goto error;
1679         }
1680         ctf_trace->metadata_sent += read_len;
1681         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1682         goto send_reply;
1683
1684 error:
1685         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1686
1687 send_reply:
1688         health_code_update();
1689         ret = send_response(conn->sock, &reply, sizeof(reply));
1690         if (ret < 0) {
1691                 goto end_unlock;
1692         }
1693         health_code_update();
1694
1695         if (len > 0) {
1696                 ret = send_response(conn->sock, data, len);
1697                 if (ret < 0) {
1698                         goto end_unlock;
1699                 }
1700         }
1701
1702         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1703                         be64toh(request.stream_id));
1704
1705         DBG("Metadata sent");
1706
1707 end_unlock:
1708         free(data);
1709         rcu_read_unlock();
1710 end:
1711         return ret;
1712 }
1713
1714 /*
1715  * Create a viewer session.
1716  *
1717  * Return 0 on success or else a negative value.
1718  */
1719 static
1720 int viewer_create_session(struct relay_connection *conn)
1721 {
1722         int ret;
1723         struct lttng_viewer_create_session_response resp;
1724
1725         DBG("Viewer create session received");
1726
1727         memset(&resp, 0, sizeof(resp));
1728         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1729         conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1730         if (!conn->viewer_session) {
1731                 ERR("Allocation viewer session");
1732                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1733                 goto send_reply;
1734         }
1735         CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1736
1737 send_reply:
1738         health_code_update();
1739         ret = send_response(conn->sock, &resp, sizeof(resp));
1740         if (ret < 0) {
1741                 goto end;
1742         }
1743         health_code_update();
1744         ret = 0;
1745
1746 end:
1747         return ret;
1748 }
1749
1750
1751 /*
1752  * live_relay_unknown_command: send -1 if received unknown command
1753  */
1754 static
1755 void live_relay_unknown_command(struct relay_connection *conn)
1756 {
1757         struct lttcomm_relayd_generic_reply reply;
1758
1759         memset(&reply, 0, sizeof(reply));
1760         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1761         (void) send_response(conn->sock, &reply, sizeof(reply));
1762 }
1763
1764 /*
1765  * Process the commands received on the control socket
1766  */
1767 static
1768 int process_control(struct lttng_viewer_cmd *recv_hdr,
1769                 struct relay_connection *conn)
1770 {
1771         int ret = 0;
1772         uint32_t msg_value;
1773
1774         assert(recv_hdr);
1775         assert(conn);
1776
1777         msg_value = be32toh(recv_hdr->cmd);
1778
1779         /*
1780          * Make sure we've done the version check before any command other then a
1781          * new client connection.
1782          */
1783         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1784                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1785                 ret = -1;
1786                 goto end;
1787         }
1788
1789         switch (msg_value) {
1790         case LTTNG_VIEWER_CONNECT:
1791                 ret = viewer_connect(conn);
1792                 break;
1793         case LTTNG_VIEWER_LIST_SESSIONS:
1794                 ret = viewer_list_sessions(conn);
1795                 break;
1796         case LTTNG_VIEWER_ATTACH_SESSION:
1797                 ret = viewer_attach_session(conn);
1798                 break;
1799         case LTTNG_VIEWER_GET_NEXT_INDEX:
1800                 ret = viewer_get_next_index(conn);
1801                 break;
1802         case LTTNG_VIEWER_GET_PACKET:
1803                 ret = viewer_get_packet(conn);
1804                 break;
1805         case LTTNG_VIEWER_GET_METADATA:
1806                 ret = viewer_get_metadata(conn);
1807                 break;
1808         case LTTNG_VIEWER_GET_NEW_STREAMS:
1809                 ret = viewer_get_new_streams(conn);
1810                 break;
1811         case LTTNG_VIEWER_CREATE_SESSION:
1812                 ret = viewer_create_session(conn);
1813                 break;
1814         default:
1815                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1816                 live_relay_unknown_command(conn);
1817                 ret = -1;
1818                 goto end;
1819         }
1820
1821 end:
1822         return ret;
1823 }
1824
1825 static
1826 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1827 {
1828         int ret;
1829
1830         assert(events);
1831
1832         (void) lttng_poll_del(events, pollfd);
1833
1834         ret = close(pollfd);
1835         if (ret < 0) {
1836                 ERR("Closing pollfd %d", pollfd);
1837         }
1838 }
1839
1840 /*
1841  * Delete and destroy a connection.
1842  *
1843  * RCU read side lock MUST be acquired.
1844  */
1845 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1846                 struct relay_connection *conn)
1847 {
1848         struct relay_session *session, *tmp_session;
1849
1850         assert(relay_connections_ht);
1851         assert(conn);
1852
1853         connection_delete(relay_connections_ht, conn);
1854
1855         if (!conn->viewer_session) {
1856                 goto end;
1857         }
1858
1859         rcu_read_lock();
1860         cds_list_for_each_entry_safe(session, tmp_session,
1861                         &conn->viewer_session->sessions_head,
1862                         viewer_session_list) {
1863                 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1864                 cleanup_session(conn, session);
1865         }
1866         rcu_read_unlock();
1867
1868 end:
1869         connection_destroy(conn);
1870 }
1871
1872 /*
1873  * This thread does the actual work
1874  */
1875 static
1876 void *thread_worker(void *data)
1877 {
1878         int ret, err = -1;
1879         uint32_t nb_fd;
1880         struct relay_connection *conn;
1881         struct lttng_poll_event events;
1882         struct lttng_ht *relay_connections_ht;
1883         struct lttng_ht_iter iter;
1884         struct lttng_viewer_cmd recv_hdr;
1885         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1886         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1887
1888         DBG("[thread] Live viewer relay worker started");
1889
1890         rcu_register_thread();
1891
1892         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1893
1894         if (testpoint(relayd_thread_live_worker)) {
1895                 goto error_testpoint;
1896         }
1897
1898         /* table of connections indexed on socket */
1899         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1900         if (!relay_connections_ht) {
1901                 goto relay_connections_ht_error;
1902         }
1903
1904         ret = create_thread_poll_set(&events, 2);
1905         if (ret < 0) {
1906                 goto error_poll_create;
1907         }
1908
1909         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1910         if (ret < 0) {
1911                 goto error;
1912         }
1913
1914 restart:
1915         while (1) {
1916                 int i;
1917
1918                 health_code_update();
1919
1920                 /* Infinite blocking call, waiting for transmission */
1921                 DBG3("Relayd live viewer worker thread polling...");
1922                 health_poll_entry();
1923                 ret = lttng_poll_wait(&events, -1);
1924                 health_poll_exit();
1925                 if (ret < 0) {
1926                         /*
1927                          * Restart interrupted system call.
1928                          */
1929                         if (errno == EINTR) {
1930                                 goto restart;
1931                         }
1932                         goto error;
1933                 }
1934
1935                 nb_fd = ret;
1936
1937                 /*
1938                  * Process control. The control connection is prioritised so we don't
1939                  * starve it with high throughput tracing data on the data
1940                  * connection.
1941                  */
1942                 for (i = 0; i < nb_fd; i++) {
1943                         /* Fetch once the poll data */
1944                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1945                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1946
1947                         health_code_update();
1948
1949                         /* Thread quit pipe has been closed. Killing thread. */
1950                         ret = check_thread_quit_pipe(pollfd, revents);
1951                         if (ret) {
1952                                 err = 0;
1953                                 goto exit;
1954                         }
1955
1956                         /* Inspect the relay conn pipe for new connection */
1957                         if (pollfd == live_conn_pipe[0]) {
1958                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1959                                         ERR("Relay live pipe error");
1960                                         goto error;
1961                                 } else if (revents & LPOLLIN) {
1962                                         ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1963                                         if (ret < 0) {
1964                                                 goto error;
1965                                         }
1966                                         conn->sessions_ht = sessions_ht;
1967                                         connection_init(conn);
1968                                         lttng_poll_add(&events, conn->sock->fd,
1969                                                         LPOLLIN | LPOLLRDHUP);
1970                                         rcu_read_lock();
1971                                         lttng_ht_add_unique_ulong(relay_connections_ht,
1972                                                         &conn->sock_n);
1973                                         rcu_read_unlock();
1974                                         DBG("Connection socket %d added", conn->sock->fd);
1975                                 }
1976                         } else {
1977                                 rcu_read_lock();
1978                                 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1979                                 /* If not found, there is a synchronization issue. */
1980                                 assert(conn);
1981
1982                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1983                                         cleanup_connection_pollfd(&events, pollfd);
1984                                         destroy_connection(relay_connections_ht, conn);
1985                                 } else if (revents & LPOLLIN) {
1986                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1987                                                         sizeof(recv_hdr), 0);
1988                                         if (ret <= 0) {
1989                                                 /* Connection closed */
1990                                                 cleanup_connection_pollfd(&events, pollfd);
1991                                                 destroy_connection(relay_connections_ht, conn);
1992                                                 DBG("Viewer control conn closed with %d", pollfd);
1993                                         } else {
1994                                                 ret = process_control(&recv_hdr, conn);
1995                                                 if (ret < 0) {
1996                                                         /* Clear the session on error. */
1997                                                         cleanup_connection_pollfd(&events, pollfd);
1998                                                         destroy_connection(relay_connections_ht, conn);
1999                                                         DBG("Viewer connection closed with %d", pollfd);
2000                                                 }
2001                                         }
2002                                 }
2003                                 rcu_read_unlock();
2004                         }
2005                 }
2006         }
2007
2008 exit:
2009 error:
2010         lttng_poll_clean(&events);
2011
2012         /* Cleanup reamaining connection object. */
2013         rcu_read_lock();
2014         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2015                         sock_n.node) {
2016                 health_code_update();
2017                 destroy_connection(relay_connections_ht, conn);
2018         }
2019         rcu_read_unlock();
2020 error_poll_create:
2021         lttng_ht_destroy(relay_connections_ht);
2022 relay_connections_ht_error:
2023         /* Close relay conn pipes */
2024         utils_close_pipe(live_conn_pipe);
2025         if (err) {
2026                 DBG("Viewer worker thread exited with error");
2027         }
2028         DBG("Viewer worker thread cleanup complete");
2029 error_testpoint:
2030         if (err) {
2031                 health_error();
2032                 ERR("Health error occurred in %s", __func__);
2033         }
2034         health_unregister(health_relayd);
2035         stop_threads();
2036         rcu_unregister_thread();
2037         return NULL;
2038 }
2039
2040 /*
2041  * Create the relay command pipe to wake thread_manage_apps.
2042  * Closed in cleanup().
2043  */
2044 static int create_conn_pipe(void)
2045 {
2046         int ret;
2047
2048         ret = utils_create_pipe_cloexec(live_conn_pipe);
2049
2050         return ret;
2051 }
2052
2053 void live_stop_threads(void)
2054 {
2055         int ret;
2056         void *status;
2057
2058         stop_threads();
2059
2060         ret = pthread_join(live_listener_thread, &status);
2061         if (ret != 0) {
2062                 PERROR("pthread_join live listener");
2063                 goto error;     /* join error, exit without cleanup */
2064         }
2065
2066         ret = pthread_join(live_worker_thread, &status);
2067         if (ret != 0) {
2068                 PERROR("pthread_join live worker");
2069                 goto error;     /* join error, exit without cleanup */
2070         }
2071
2072         ret = pthread_join(live_dispatcher_thread, &status);
2073         if (ret != 0) {
2074                 PERROR("pthread_join live dispatcher");
2075                 goto error;     /* join error, exit without cleanup */
2076         }
2077
2078         cleanup();
2079
2080 error:
2081         return;
2082 }
2083
2084 /*
2085  * main
2086  */
2087 int live_start_threads(struct lttng_uri *uri,
2088                 struct relay_local_data *relay_ctx)
2089 {
2090         int ret = 0;
2091         void *status;
2092         int is_root;
2093
2094         assert(uri);
2095         live_uri = uri;
2096
2097         /* Check if daemon is UID = 0 */
2098         is_root = !getuid();
2099
2100         if (!is_root) {
2101                 if (live_uri->port < 1024) {
2102                         ERR("Need to be root to use ports < 1024");
2103                         ret = -1;
2104                         goto exit;
2105                 }
2106         }
2107
2108         /* Setup the thread apps communication pipe. */
2109         if ((ret = create_conn_pipe()) < 0) {
2110                 goto exit;
2111         }
2112
2113         /* Init relay command queue. */
2114         cds_wfq_init(&viewer_conn_queue.queue);
2115
2116         /* Set up max poll set size */
2117         lttng_poll_set_max_size();
2118
2119         /* Setup the dispatcher thread */
2120         ret = pthread_create(&live_dispatcher_thread, NULL,
2121                         thread_dispatcher, (void *) NULL);
2122         if (ret != 0) {
2123                 PERROR("pthread_create viewer dispatcher");
2124                 goto exit_dispatcher;
2125         }
2126
2127         /* Setup the worker thread */
2128         ret = pthread_create(&live_worker_thread, NULL,
2129                         thread_worker, relay_ctx);
2130         if (ret != 0) {
2131                 PERROR("pthread_create viewer worker");
2132                 goto exit_worker;
2133         }
2134
2135         /* Setup the listener thread */
2136         ret = pthread_create(&live_listener_thread, NULL,
2137                         thread_listener, (void *) NULL);
2138         if (ret != 0) {
2139                 PERROR("pthread_create viewer listener");
2140                 goto exit_listener;
2141         }
2142
2143         ret = 0;
2144         goto end;
2145
2146 exit_listener:
2147         ret = pthread_join(live_listener_thread, &status);
2148         if (ret != 0) {
2149                 PERROR("pthread_join live listener");
2150                 goto error;     /* join error, exit without cleanup */
2151         }
2152
2153 exit_worker:
2154         ret = pthread_join(live_worker_thread, &status);
2155         if (ret != 0) {
2156                 PERROR("pthread_join live worker");
2157                 goto error;     /* join error, exit without cleanup */
2158         }
2159
2160 exit_dispatcher:
2161         ret = pthread_join(live_dispatcher_thread, &status);
2162         if (ret != 0) {
2163                 PERROR("pthread_join live dispatcher");
2164                 goto error;     /* join error, exit without cleanup */
2165         }
2166
2167 exit:
2168         cleanup();
2169
2170 end:
2171 error:
2172         return ret;
2173 }