src/bin/lttng-relayd/live.c

   1 /*
   2  * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
   3  *                      David Goulet <dgoulet@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _GNU_SOURCE
  20 #define _LGPL_SOURCE
  21 #include <getopt.h>
  22 #include <grp.h>
  23 #include <limits.h>
  24 #include <pthread.h>
  25 #include <signal.h>
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include <sys/mman.h>
  30 #include <sys/mount.h>
  31 #include <sys/resource.h>
  32 #include <sys/socket.h>
  33 #include <sys/stat.h>
  34 #include <sys/types.h>
  35 #include <sys/wait.h>
  36 #include <inttypes.h>
  37 #include <urcu/futex.h>
  38 #include <urcu/uatomic.h>
  39 #include <unistd.h>
  40 #include <fcntl.h>
  41 #include <config.h>
  42
  43 #include <lttng/lttng.h>
  44 #include <common/common.h>
  45 #include <common/compat/poll.h>
  46 #include <common/compat/socket.h>
  47 #include <common/compat/endian.h>
  48 #include <common/defaults.h>
  49 #include <common/futex.h>
  50 #include <common/index/index.h>
  51 #include <common/sessiond-comm/sessiond-comm.h>
  52 #include <common/sessiond-comm/inet.h>
  53 #include <common/sessiond-comm/relayd.h>
  54 #include <common/uri.h>
  55 #include <common/utils.h>
  56
  57 #include "cmd.h"
  58 #include "live.h"
  59 #include "lttng-relayd.h"
  60 #include "utils.h"
  61 #include "health-relayd.h"
  62 #include "testpoint.h"
  63 #include "viewer-stream.h"
  64 #include "stream.h"
  65 #include "session.h"
  66 #include "ctf-trace.h"
  67 #include "connection.h"
  68
  69 static struct lttng_uri *live_uri;
  70
  71 /*
  72  * This pipe is used to inform the worker thread that a command is queued and
  73  * ready to be processed.
  74  */
  75 static int live_conn_pipe[2] = { -1, -1 };
  76
  77 /* Shared between threads */
  78 static int live_dispatch_thread_exit;
  79
  80 static pthread_t live_listener_thread;
  81 static pthread_t live_dispatcher_thread;
  82 static pthread_t live_worker_thread;
  83
  84 /*
  85  * Relay command queue.
  86  *
  87  * The live_thread_listener and live_thread_dispatcher communicate with this
  88  * queue.
  89  */
  90 static struct relay_conn_queue viewer_conn_queue;
  91
  92 static uint64_t last_relay_viewer_session_id;
  93
  94 /*
  95  * Cleanup the daemon
  96  */
  97 static
  98 void cleanup_relayd_live(void)
  99 {
 100         DBG("Cleaning up");
 101
 102         free(live_uri);
 103 }
 104
 105 /*
 106  * Receive a request buffer using a given socket, destination allocated buffer
 107  * of length size.
 108  *
 109  * Return the size of the received message or else a negative value on error
 110  * with errno being set by recvmsg() syscall.
 111  */
 112 static
 113 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 114 {
 115         ssize_t ret;
 116
 117         assert(sock);
 118         assert(buf);
 119
 120         ret = sock->ops->recvmsg(sock, buf, size, 0);
 121         if (ret < 0 || ret != size) {
 122                 if (ret == 0) {
 123                         /* Orderly shutdown. Not necessary to print an error. */
 124                         DBG("Socket %d did an orderly shutdown", sock->fd);
 125                 } else {
 126                         ERR("Relay failed to receive request.");
 127                 }
 128                 ret = -1;
 129         }
 130
 131         return ret;
 132 }
 133
 134 /*
 135  * Send a response buffer using a given socket, source allocated buffer of
 136  * length size.
 137  *
 138  * Return the size of the sent message or else a negative value on error with
 139  * errno being set by sendmsg() syscall.
 140  */
 141 static
 142 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 143 {
 144         ssize_t ret;
 145
 146         assert(sock);
 147         assert(buf);
 148
 149         ret = sock->ops->sendmsg(sock, buf, size, 0);
 150         if (ret < 0) {
 151                 ERR("Relayd failed to send response.");
 152         }
 153
 154         return ret;
 155 }
 156
 157 /*
 158  * Atomically check if new streams got added in one of the sessions attached
 159  * and reset the flag to 0.
 160  *
 161  * Returns 1 if new streams got added, 0 if nothing changed, a negative value
 162  * on error.
 163  */
 164 static
 165 int check_new_streams(struct relay_connection *conn)
 166 {
 167         struct relay_session *session;
 168         unsigned long current_val;
 169         int ret = 0;
 170
 171         if (!conn->viewer_session) {
 172                 goto end;
 173         }
 174         cds_list_for_each_entry(session,
 175                         &conn->viewer_session->sessions_head,
 176                         viewer_session_list) {
 177                 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
 178                 ret = current_val;
 179                 if (ret == 1) {
 180                         goto end;
 181                 }
 182         }
 183
 184 end:
 185         return ret;
 186 }
 187
 188 /*
 189  * Send viewer streams to the given socket. The ignore_sent_flag indicates if
 190  * this function should ignore the sent flag or not.
 191  *
 192  * Return 0 on success or else a negative value.
 193  */
 194 static
 195 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
 196                 struct relay_session *session, unsigned int ignore_sent_flag)
 197 {
 198         ssize_t ret;
 199         struct lttng_viewer_stream send_stream;
 200         struct lttng_ht_iter iter;
 201         struct relay_viewer_stream *vstream;
 202
 203         assert(session);
 204
 205         rcu_read_lock();
 206
 207         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
 208                         stream_n.node) {
 209                 struct ctf_trace *ctf_trace;
 210
 211                 health_code_update();
 212
 213                 /* Ignore if not the same session. */
 214                 if (vstream->session_id != session->id ||
 215                                 (!ignore_sent_flag && vstream->sent_flag)) {
 216                         continue;
 217                 }
 218
 219                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 220                                 vstream->path_name);
 221                 assert(ctf_trace);
 222
 223                 send_stream.id = htobe64(vstream->stream_handle);
 224                 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
 225                 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
 226                 strncpy(send_stream.path_name, vstream->path_name,
 227                                 sizeof(send_stream.path_name));
 228                 strncpy(send_stream.channel_name, vstream->channel_name,
 229                                 sizeof(send_stream.channel_name));
 230
 231                 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
 232                 ret = send_response(sock, &send_stream, sizeof(send_stream));
 233                 if (ret < 0) {
 234                         goto end_unlock;
 235                 }
 236                 vstream->sent_flag = 1;
 237         }
 238
 239         ret = 0;
 240
 241 end_unlock:
 242         rcu_read_unlock();
 243         return ret;
 244 }
 245
 246 /*
 247  * Create every viewer stream possible for the given session with the seek
 248  * type. Three counters *can* be return which are in order the total amount of
 249  * viewer stream of the session, the number of unsent stream and the number of
 250  * stream created. Those counters can be NULL and thus will be ignored.
 251  *
 252  * Return 0 on success or else a negative value.
 253  */
 254 static
 255 int make_viewer_streams(struct relay_session *session,
 256                 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
 257                 uint32_t *nb_created)
 258 {
 259         int ret;
 260         struct lttng_ht_iter iter;
 261         struct ctf_trace *ctf_trace;
 262
 263         assert(session);
 264
 265         /*
 266          * This is to make sure we create viewer streams for a full received
 267          * channel. For instance, if we have 8 streams for a channel that are
 268          * concurrently being flagged ready, we can end up creating just a subset
 269          * of the 8 streams (the ones that are flagged). This lock avoids this
 270          * limbo state.
 271          */
 272         pthread_mutex_lock(&session->viewer_ready_lock);
 273
 274         /*
 275          * Create viewer streams for relay streams that are ready to be used for a
 276          * the given session id only.
 277          */
 278         rcu_read_lock();
 279         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 280                         node.node) {
 281                 struct relay_stream *stream;
 282
 283                 health_code_update();
 284
 285                 if (ctf_trace->invalid_flag) {
 286                         continue;
 287                 }
 288
 289                 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
 290                         struct relay_viewer_stream *vstream;
 291
 292                         if (!stream->viewer_ready) {
 293                                 continue;
 294                         }
 295
 296                         vstream = viewer_stream_find_by_id(stream->stream_handle);
 297                         if (!vstream) {
 298                                 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
 299                                 if (!vstream) {
 300                                         ret = -1;
 301                                         goto error_unlock;
 302                                 }
 303                                 /* Acquire reference to ctf_trace. */
 304                                 ctf_trace_get_ref(ctf_trace);
 305
 306                                 if (nb_created) {
 307                                         /* Update number of created stream counter. */
 308                                         (*nb_created)++;
 309                                 }
 310                         } else if (!vstream->sent_flag && nb_unsent) {
 311                                 /* Update number of unsent stream counter. */
 312                                 (*nb_unsent)++;
 313                         }
 314                         /* Update number of total stream counter. */
 315                         if (nb_total) {
 316                                 (*nb_total)++;
 317                         }
 318                 }
 319         }
 320
 321         ret = 0;
 322
 323 error_unlock:
 324         rcu_read_unlock();
 325         pthread_mutex_unlock(&session->viewer_ready_lock);
 326         return ret;
 327 }
 328
 329 int relayd_live_stop(void)
 330 {
 331         /* Stop dispatch thread */
 332         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 333         futex_nto1_wake(&viewer_conn_queue.futex);
 334         return 0;
 335 }
 336
 337 /*
 338  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 339  */
 340 static
 341 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 342 {
 343         int ret;
 344
 345         if (events == NULL || size == 0) {
 346                 ret = -1;
 347                 goto error;
 348         }
 349
 350         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 351         if (ret < 0) {
 352                 goto error;
 353         }
 354
 355         /* Add quit pipe */
 356         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 357         if (ret < 0) {
 358                 goto error;
 359         }
 360
 361         return 0;
 362
 363 error:
 364         return ret;
 365 }
 366
 367 /*
 368  * Check if the thread quit pipe was triggered.
 369  *
 370  * Return 1 if it was triggered else 0;
 371  */
 372 static
 373 int check_thread_quit_pipe(int fd, uint32_t events)
 374 {
 375         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 376                 return 1;
 377         }
 378
 379         return 0;
 380 }
 381
 382 /*
 383  * Create and init socket from uri.
 384  */
 385 static
 386 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 387 {
 388         int ret;
 389         struct lttcomm_sock *sock = NULL;
 390
 391         sock = lttcomm_alloc_sock_from_uri(uri);
 392         if (sock == NULL) {
 393                 ERR("Allocating socket");
 394                 goto error;
 395         }
 396
 397         ret = lttcomm_create_sock(sock);
 398         if (ret < 0) {
 399                 goto error;
 400         }
 401         DBG("Listening on sock %d for live", sock->fd);
 402
 403         ret = sock->ops->bind(sock);
 404         if (ret < 0) {
 405                 goto error;
 406         }
 407
 408         ret = sock->ops->listen(sock, -1);
 409         if (ret < 0) {
 410                 goto error;
 411
 412         }
 413
 414         return sock;
 415
 416 error:
 417         if (sock) {
 418                 lttcomm_destroy_sock(sock);
 419         }
 420         return NULL;
 421 }
 422
 423 /*
 424  * This thread manages the listening for new connections on the network
 425  */
 426 static
 427 void *thread_listener(void *data)
 428 {
 429         int i, ret, pollfd, err = -1;
 430         uint32_t revents, nb_fd;
 431         struct lttng_poll_event events;
 432         struct lttcomm_sock *live_control_sock;
 433
 434         DBG("[thread] Relay live listener started");
 435
 436         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 437
 438         health_code_update();
 439
 440         live_control_sock = init_socket(live_uri);
 441         if (!live_control_sock) {
 442                 goto error_sock_control;
 443         }
 444
 445         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 446         ret = create_thread_poll_set(&events, 2);
 447         if (ret < 0) {
 448                 goto error_create_poll;
 449         }
 450
 451         /* Add the control socket */
 452         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 453         if (ret < 0) {
 454                 goto error_poll_add;
 455         }
 456
 457         lttng_relay_notify_ready();
 458
 459         if (testpoint(relayd_thread_live_listener)) {
 460                 goto error_testpoint;
 461         }
 462
 463         while (1) {
 464                 health_code_update();
 465
 466                 DBG("Listener accepting live viewers connections");
 467
 468 restart:
 469                 health_poll_entry();
 470                 ret = lttng_poll_wait(&events, -1);
 471                 health_poll_exit();
 472                 if (ret < 0) {
 473                         /*
 474                          * Restart interrupted system call.
 475                          */
 476                         if (errno == EINTR) {
 477                                 goto restart;
 478                         }
 479                         goto error;
 480                 }
 481                 nb_fd = ret;
 482
 483                 DBG("Relay new viewer connection received");
 484                 for (i = 0; i < nb_fd; i++) {
 485                         health_code_update();
 486
 487                         /* Fetch once the poll data */
 488                         revents = LTTNG_POLL_GETEV(&events, i);
 489                         pollfd = LTTNG_POLL_GETFD(&events, i);
 490
 491                         if (!revents) {
 492                                 /* No activity for this FD (poll implementation). */
 493                                 continue;
 494                         }
 495
 496                         /* Thread quit pipe has been closed. Killing thread. */
 497                         ret = check_thread_quit_pipe(pollfd, revents);
 498                         if (ret) {
 499                                 err = 0;
 500                                 goto exit;
 501                         }
 502
 503                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 504                                 ERR("socket poll error");
 505                                 goto error;
 506                         } else if (revents & LPOLLIN) {
 507                                 /*
 508                                  * Get allocated in this thread, enqueued to a global queue,
 509                                  * dequeued and freed in the worker thread.
 510                                  */
 511                                 int val = 1;
 512                                 struct relay_connection *new_conn;
 513                                 struct lttcomm_sock *newsock;
 514
 515                                 new_conn = connection_create();
 516                                 if (!new_conn) {
 517                                         goto error;
 518                                 }
 519
 520                                 newsock = live_control_sock->ops->accept(live_control_sock);
 521                                 if (!newsock) {
 522                                         PERROR("accepting control sock");
 523                                         connection_free(new_conn);
 524                                         goto error;
 525                                 }
 526                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 527
 528                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 529                                                 sizeof(val));
 530                                 if (ret < 0) {
 531                                         PERROR("setsockopt inet");
 532                                         lttcomm_destroy_sock(newsock);
 533                                         connection_free(new_conn);
 534                                         goto error;
 535                                 }
 536                                 new_conn->sock = newsock;
 537
 538                                 /* Enqueue request for the dispatcher thread. */
 539                                 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
 540                                                  &new_conn->qnode);
 541
 542                                 /*
 543                                  * Wake the dispatch queue futex. Implicit memory barrier with
 544                                  * the exchange in cds_wfcq_enqueue.
 545                                  */
 546                                 futex_nto1_wake(&viewer_conn_queue.futex);
 547                         }
 548                 }
 549         }
 550
 551 exit:
 552 error:
 553 error_poll_add:
 554 error_testpoint:
 555         lttng_poll_clean(&events);
 556 error_create_poll:
 557         if (live_control_sock->fd >= 0) {
 558                 ret = live_control_sock->ops->close(live_control_sock);
 559                 if (ret) {
 560                         PERROR("close");
 561                 }
 562         }
 563         lttcomm_destroy_sock(live_control_sock);
 564 error_sock_control:
 565         if (err) {
 566                 health_error();
 567                 DBG("Live viewer listener thread exited with error");
 568         }
 569         health_unregister(health_relayd);
 570         DBG("Live viewer listener thread cleanup complete");
 571         if (lttng_relay_stop_threads()) {
 572                 ERR("Error stopping threads");
 573         }
 574         return NULL;
 575 }
 576
 577 /*
 578  * This thread manages the dispatching of the requests to worker threads
 579  */
 580 static
 581 void *thread_dispatcher(void *data)
 582 {
 583         int err = -1;
 584         ssize_t ret;
 585         struct cds_wfcq_node *node;
 586         struct relay_connection *conn = NULL;
 587
 588         DBG("[thread] Live viewer relay dispatcher started");
 589
 590         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 591
 592         if (testpoint(relayd_thread_live_dispatcher)) {
 593                 goto error_testpoint;
 594         }
 595
 596         health_code_update();
 597
 598         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 599                 health_code_update();
 600
 601                 /* Atomically prepare the queue futex */
 602                 futex_nto1_prepare(&viewer_conn_queue.futex);
 603
 604                 do {
 605                         health_code_update();
 606
 607                         /* Dequeue commands */
 608                         node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
 609                                                          &viewer_conn_queue.tail);
 610                         if (node == NULL) {
 611                                 DBG("Woken up but nothing in the live-viewer "
 612                                                 "relay command queue");
 613                                 /* Continue thread execution */
 614                                 break;
 615                         }
 616                         conn = caa_container_of(node, struct relay_connection, qnode);
 617                         DBG("Dispatching viewer request waiting on sock %d",
 618                                         conn->sock->fd);
 619
 620                         /*
 621                          * Inform worker thread of the new request. This call is blocking
 622                          * so we can be assured that the data will be read at some point in
 623                          * time or wait to the end of the world :)
 624                          */
 625                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 626                         if (ret < 0) {
 627                                 PERROR("write conn pipe");
 628                                 connection_destroy(conn);
 629                                 goto error;
 630                         }
 631                 } while (node != NULL);
 632
 633                 /* Futex wait on queue. Blocking call on futex() */
 634                 health_poll_entry();
 635                 futex_nto1_wait(&viewer_conn_queue.futex);
 636                 health_poll_exit();
 637         }
 638
 639         /* Normal exit, no error */
 640         err = 0;
 641
 642 error:
 643 error_testpoint:
 644         if (err) {
 645                 health_error();
 646                 ERR("Health error occurred in %s", __func__);
 647         }
 648         health_unregister(health_relayd);
 649         DBG("Live viewer dispatch thread dying");
 650         if (lttng_relay_stop_threads()) {
 651                 ERR("Error stopping threads");
 652         }
 653         return NULL;
 654 }
 655
 656 /*
 657  * Establish connection with the viewer and check the versions.
 658  *
 659  * Return 0 on success or else negative value.
 660  */
 661 static
 662 int viewer_connect(struct relay_connection *conn)
 663 {
 664         int ret;
 665         struct lttng_viewer_connect reply, msg;
 666
 667         assert(conn);
 668
 669         conn->version_check_done = 1;
 670
 671         health_code_update();
 672
 673         DBG("Viewer is establishing a connection to the relayd.");
 674
 675         ret = recv_request(conn->sock, &msg, sizeof(msg));
 676         if (ret < 0) {
 677                 goto end;
 678         }
 679
 680         health_code_update();
 681
 682         memset(&reply, 0, sizeof(reply));
 683         reply.major = RELAYD_VERSION_COMM_MAJOR;
 684         reply.minor = RELAYD_VERSION_COMM_MINOR;
 685
 686         /* Major versions must be the same */
 687         if (reply.major != be32toh(msg.major)) {
 688                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 689                                 reply.major, be32toh(msg.major));
 690                 ret = -1;
 691                 goto end;
 692         }
 693
 694         conn->major = reply.major;
 695         /* We adapt to the lowest compatible version */
 696         if (reply.minor <= be32toh(msg.minor)) {
 697                 conn->minor = reply.minor;
 698         } else {
 699                 conn->minor = be32toh(msg.minor);
 700         }
 701
 702         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 703                 conn->type = RELAY_VIEWER_COMMAND;
 704         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 705                 conn->type = RELAY_VIEWER_NOTIFICATION;
 706         } else {
 707                 ERR("Unknown connection type : %u", be32toh(msg.type));
 708                 ret = -1;
 709                 goto end;
 710         }
 711
 712         reply.major = htobe32(reply.major);
 713         reply.minor = htobe32(reply.minor);
 714         if (conn->type == RELAY_VIEWER_COMMAND) {
 715                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 716         }
 717
 718         health_code_update();
 719
 720         ret = send_response(conn->sock, &reply, sizeof(reply));
 721         if (ret < 0) {
 722                 goto end;
 723         }
 724
 725         health_code_update();
 726
 727         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 728         ret = 0;
 729
 730 end:
 731         return ret;
 732 }
 733
 734 /*
 735  * Send the viewer the list of current sessions.
 736  *
 737  * Return 0 on success or else a negative value.
 738  */
 739 static
 740 int viewer_list_sessions(struct relay_connection *conn)
 741 {
 742         int ret;
 743         struct lttng_viewer_list_sessions session_list;
 744         unsigned long count;
 745         long approx_before, approx_after;
 746         struct lttng_ht_iter iter;
 747         struct lttng_viewer_session send_session;
 748         struct relay_session *session;
 749
 750         DBG("List sessions received");
 751
 752         rcu_read_lock();
 753         cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
 754                         &approx_after);
 755         session_list.sessions_count = htobe32(count);
 756
 757         health_code_update();
 758
 759         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 760         if (ret < 0) {
 761                 goto end_unlock;
 762         }
 763
 764         health_code_update();
 765
 766         cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
 767                         session_n.node) {
 768                 health_code_update();
 769
 770                 strncpy(send_session.session_name, session->session_name,
 771                                 sizeof(send_session.session_name));
 772                 strncpy(send_session.hostname, session->hostname,
 773                                 sizeof(send_session.hostname));
 774                 send_session.id = htobe64(session->id);
 775                 send_session.live_timer = htobe32(session->live_timer);
 776                 send_session.clients = htobe32(session->viewer_refcount);
 777                 send_session.streams = htobe32(session->stream_count);
 778
 779                 health_code_update();
 780
 781                 ret = send_response(conn->sock, &send_session, sizeof(send_session));
 782                 if (ret < 0) {
 783                         goto end_unlock;
 784                 }
 785         }
 786         health_code_update();
 787
 788         ret = 0;
 789 end_unlock:
 790         rcu_read_unlock();
 791         return ret;
 792 }
 793
 794 /*
 795  * Check if a connection is attached to a session.
 796  * Return 1 if attached, 0 if not attached, a negative value on error.
 797  */
 798 static
 799 int session_attached(struct relay_connection *conn, uint64_t session_id)
 800 {
 801         struct relay_session *session;
 802         int found = 0;
 803
 804         if (!conn->viewer_session) {
 805                 goto end;
 806         }
 807         cds_list_for_each_entry(session,
 808                         &conn->viewer_session->sessions_head,
 809                         viewer_session_list) {
 810                 if (session->id == session_id) {
 811                         found = 1;
 812                         goto end;
 813                 }
 814         }
 815
 816 end:
 817         return found;
 818 }
 819
 820 /*
 821  * Delete all streams for a specific session ID.
 822  */
 823 static void destroy_viewer_streams_by_session(struct relay_session *session)
 824 {
 825         struct relay_viewer_stream *stream;
 826         struct lttng_ht_iter iter;
 827
 828         assert(session);
 829
 830         rcu_read_lock();
 831         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
 832                         stream_n.node) {
 833                 struct ctf_trace *ctf_trace;
 834
 835                 health_code_update();
 836                 if (stream->session_id != session->id) {
 837                         continue;
 838                 }
 839
 840                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 841                                 stream->path_name);
 842                 assert(ctf_trace);
 843
 844                 viewer_stream_delete(stream);
 845
 846                 if (stream->metadata_flag) {
 847                         ctf_trace->metadata_sent = 0;
 848                         ctf_trace->viewer_metadata_stream = NULL;
 849                 }
 850
 851                 viewer_stream_destroy(ctf_trace, stream);
 852         }
 853         rcu_read_unlock();
 854 }
 855
 856 static void try_destroy_streams(struct relay_session *session)
 857 {
 858         struct ctf_trace *ctf_trace;
 859         struct lttng_ht_iter iter;
 860
 861         assert(session);
 862
 863         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 864                         node.node) {
 865                 /* Attempt to destroy the ctf trace of that session. */
 866                 ctf_trace_try_destroy(session, ctf_trace);
 867         }
 868 }
 869
 870 /*
 871  * Cleanup a session.
 872  */
 873 static void cleanup_session(struct relay_connection *conn,
 874                 struct relay_session *session)
 875 {
 876         /*
 877          * Very important that this is done before destroying the session so we
 878          * can put back every viewer stream reference from the ctf_trace.
 879          */
 880         destroy_viewer_streams_by_session(session);
 881         try_destroy_streams(session);
 882         cds_list_del(&session->viewer_session_list);
 883         session_viewer_try_destroy(conn->sessions_ht, session);
 884 }
 885
 886 /*
 887  * Send the viewer the list of current sessions.
 888  */
 889 static
 890 int viewer_get_new_streams(struct relay_connection *conn)
 891 {
 892         int ret, send_streams = 0;
 893         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
 894         struct lttng_viewer_new_streams_request request;
 895         struct lttng_viewer_new_streams_response response;
 896         struct relay_session *session;
 897         uint64_t session_id;
 898
 899         assert(conn);
 900
 901         DBG("Get new streams received");
 902
 903         health_code_update();
 904
 905         /* Receive the request from the connected client. */
 906         ret = recv_request(conn->sock, &request, sizeof(request));
 907         if (ret < 0) {
 908                 goto error;
 909         }
 910         session_id = be64toh(request.session_id);
 911
 912         health_code_update();
 913
 914         memset(&response, 0, sizeof(response));
 915
 916         rcu_read_lock();
 917         session = session_find_by_id(conn->sessions_ht, session_id);
 918         if (!session) {
 919                 DBG("Relay session %" PRIu64 " not found", session_id);
 920                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 921                 goto send_reply;
 922         }
 923
 924         if (!session_attached(conn, session_id)) {
 925                 send_streams = 0;
 926                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 927                 goto send_reply;
 928         }
 929
 930         send_streams = 1;
 931         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 932
 933         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
 934                         &nb_created);
 935         if (ret < 0) {
 936                 goto end_unlock;
 937         }
 938         /* Only send back the newly created streams with the unsent ones. */
 939         nb_streams = nb_created + nb_unsent;
 940         response.streams_count = htobe32(nb_streams);
 941
 942         /*
 943          * If the session is closed and we have no new streams to send,
 944          * it means that the viewer has already received the whole trace
 945          * for this session and should now close it.
 946          */
 947         if (nb_streams == 0 && session->close_flag) {
 948                 send_streams = 0;
 949                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 950                 /*
 951                  * Remove the session from the attached list of the connection
 952                  * and try to destroy it.
 953                  */
 954                 cds_list_del(&session->viewer_session_list);
 955                 cleanup_session(conn, session);
 956                 goto send_reply;
 957         }
 958
 959 send_reply:
 960         health_code_update();
 961         ret = send_response(conn->sock, &response, sizeof(response));
 962         if (ret < 0) {
 963                 goto end_unlock;
 964         }
 965         health_code_update();
 966
 967         /*
 968          * Unknown or empty session, just return gracefully, the viewer knows what
 969          * is happening.
 970          */
 971         if (!send_streams || !nb_streams) {
 972                 ret = 0;
 973                 goto end_unlock;
 974         }
 975
 976         /*
 977          * Send stream and *DON'T* ignore the sent flag so every viewer streams
 978          * that were not sent from that point will be sent to the viewer.
 979          */
 980         ret = send_viewer_streams(conn->sock, session, 0);
 981         if (ret < 0) {
 982                 goto end_unlock;
 983         }
 984
 985 end_unlock:
 986         rcu_read_unlock();
 987 error:
 988         return ret;
 989 }
 990
 991 /*
 992  * Send the viewer the list of current sessions.
 993  */
 994 static
 995 int viewer_attach_session(struct relay_connection *conn)
 996 {
 997         int send_streams = 0;
 998         ssize_t ret;
 999         uint32_t nb_streams = 0;
1000         enum lttng_viewer_seek seek_type;
1001         struct lttng_viewer_attach_session_request request;
1002         struct lttng_viewer_attach_session_response response;
1003         struct relay_session *session;
1004
1005         assert(conn);
1006
1007         health_code_update();
1008
1009         /* Receive the request from the connected client. */
1010         ret = recv_request(conn->sock, &request, sizeof(request));
1011         if (ret < 0) {
1012                 goto error;
1013         }
1014
1015         health_code_update();
1016
1017         memset(&response, 0, sizeof(response));
1018
1019         if (!conn->viewer_session) {
1020                 DBG("Client trying to attach before creating a live viewer session");
1021                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1022                 goto send_reply;
1023         }
1024
1025         rcu_read_lock();
1026         session = session_find_by_id(conn->sessions_ht,
1027                         be64toh(request.session_id));
1028         if (!session) {
1029                 DBG("Relay session %" PRIu64 " not found",
1030                                 be64toh(request.session_id));
1031                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1032                 goto send_reply;
1033         }
1034         session_viewer_attach(session);
1035         DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1036
1037         if (uatomic_read(&session->viewer_refcount) > 1) {
1038                 DBG("Already a viewer attached");
1039                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1040                 session_viewer_detach(session);
1041                 goto send_reply;
1042         } else if (session->live_timer == 0) {
1043                 DBG("Not live session");
1044                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1045                 goto send_reply;
1046         } else {
1047                 send_streams = 1;
1048                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1049                 cds_list_add(&session->viewer_session_list,
1050                                 &conn->viewer_session->sessions_head);
1051         }
1052
1053         switch (be32toh(request.seek)) {
1054         case LTTNG_VIEWER_SEEK_BEGINNING:
1055         case LTTNG_VIEWER_SEEK_LAST:
1056                 seek_type = be32toh(request.seek);
1057                 break;
1058         default:
1059                 ERR("Wrong seek parameter");
1060                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1061                 send_streams = 0;
1062                 goto send_reply;
1063         }
1064
1065         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1066         if (ret < 0) {
1067                 goto end_unlock;
1068         }
1069         response.streams_count = htobe32(nb_streams);
1070
1071 send_reply:
1072         health_code_update();
1073         ret = send_response(conn->sock, &response, sizeof(response));
1074         if (ret < 0) {
1075                 goto end_unlock;
1076         }
1077         health_code_update();
1078
1079         /*
1080          * Unknown or empty session, just return gracefully, the viewer knows what
1081          * is happening.
1082          */
1083         if (!send_streams || !nb_streams) {
1084                 ret = 0;
1085                 goto end_unlock;
1086         }
1087
1088         /* Send stream and ignore the sent flag. */
1089         ret = send_viewer_streams(conn->sock, session, 1);
1090         if (ret < 0) {
1091                 goto end_unlock;
1092         }
1093
1094 end_unlock:
1095         rcu_read_unlock();
1096 error:
1097         return ret;
1098 }
1099
1100 /*
1101  * Open the index file if needed for the given vstream.
1102  *
1103  * If an index file is successfully opened, the index_read_fd of the stream is
1104  * set with it.
1105  *
1106  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1107  */
1108 static int try_open_index(struct relay_viewer_stream *vstream,
1109                 struct relay_stream *rstream)
1110 {
1111         int ret = 0;
1112
1113         assert(vstream);
1114         assert(rstream);
1115
1116         if (vstream->index_read_fd >= 0) {
1117                 goto end;
1118         }
1119
1120         /*
1121          * First time, we open the index file and at least one index is ready.  The
1122          * race between the read and write of the total_index_received is
1123          * acceptable here since the client will be notified to simply come back
1124          * and get the next index.
1125          */
1126         if (rstream->total_index_received <= 0) {
1127                 ret = -ENOENT;
1128                 goto end;
1129         }
1130         ret = index_open(vstream->path_name, vstream->channel_name,
1131                         vstream->tracefile_count, vstream->tracefile_count_current);
1132         if (ret >= 0) {
1133                 vstream->index_read_fd = ret;
1134                 ret = 0;
1135                 goto end;
1136         }
1137
1138 end:
1139         return ret;
1140 }
1141
1142 /*
1143  * Check the status of the index for the given stream. This function updates
1144  * the index structure if needed and can destroy the vstream also for the HUP
1145  * situation.
1146  *
1147  * Return 0 means that we can proceed with the index. A value of 1 means that
1148  * the index has been updated and is ready to be send to the client. A negative
1149  * value indicates an error that can't be handled.
1150  */
1151 static int check_index_status(struct relay_viewer_stream *vstream,
1152                 struct relay_stream *rstream, struct ctf_trace *trace,
1153                 struct lttng_viewer_index *index)
1154 {
1155         int ret;
1156
1157         assert(vstream);
1158         assert(rstream);
1159         assert(index);
1160         assert(trace);
1161
1162         if (!rstream->close_flag) {
1163                 /* Rotate on abort (overwrite). */
1164                 if (vstream->abort_flag) {
1165                         DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1166                                         vstream->stream_handle);
1167                         ret = viewer_stream_rotate(vstream, rstream);
1168                         if (ret < 0) {
1169                                 goto error;
1170                         } else if (ret == 1) {
1171                                 /* EOF */
1172                                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1173                                 goto hup;
1174                         }
1175                         /* ret == 0 means successful so we continue. */
1176                 }
1177
1178                 /* Check if we are in the same trace file at this point. */
1179                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1180                         if (rstream->beacon_ts_end != -1ULL &&
1181                                         vstream->last_sent_index == rstream->total_index_received) {
1182                                 /*
1183                                  * We've received a synchronization beacon and the last index
1184                                  * available has been sent, the index for now is inactive.
1185                                  */
1186                                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1187                                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1188                                 index->stream_id = htobe64(rstream->ctf_stream_id);
1189                                 goto index_ready;
1190                         } else if (rstream->total_index_received <= vstream->last_sent_index
1191                                         && !vstream->close_write_flag) {
1192                                 /*
1193                                  * Reader and writer are working in the same tracefile, so we care
1194                                  * about the number of index received and sent. Otherwise, we read
1195                                  * up to EOF.
1196                                  */
1197                                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1198                                 goto index_ready;
1199                         }
1200                 }
1201                 /* Nothing to do with the index, continue with it. */
1202                 ret = 0;
1203         } else if (rstream->close_flag && vstream->close_write_flag &&
1204                         vstream->total_index_received == vstream->last_sent_index) {
1205                 /* Last index sent and current tracefile closed in write */
1206                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1207                 goto hup;
1208         } else {
1209                 vstream->close_write_flag = 1;
1210                 ret = 0;
1211         }
1212
1213 error:
1214         return ret;
1215
1216 hup:
1217         viewer_stream_delete(vstream);
1218         viewer_stream_destroy(trace, vstream);
1219 index_ready:
1220         return 1;
1221 }
1222
1223 /*
1224  * Send the next index for a stream.
1225  *
1226  * Return 0 on success or else a negative value.
1227  */
1228 static
1229 int viewer_get_next_index(struct relay_connection *conn)
1230 {
1231         int ret;
1232         ssize_t read_ret;
1233         struct lttng_viewer_get_next_index request_index;
1234         struct lttng_viewer_index viewer_index;
1235         struct ctf_packet_index packet_index;
1236         struct relay_viewer_stream *vstream;
1237         struct relay_stream *rstream;
1238         struct ctf_trace *ctf_trace;
1239         struct relay_session *session;
1240
1241         assert(conn);
1242
1243         DBG("Viewer get next index");
1244
1245         health_code_update();
1246
1247         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1248         if (ret < 0) {
1249                 goto end;
1250         }
1251         health_code_update();
1252
1253         rcu_read_lock();
1254         vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1255         if (!vstream) {
1256                 ret = -1;
1257                 goto end_unlock;
1258         }
1259
1260         session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1261         if (!session) {
1262                 ret = -1;
1263                 goto end_unlock;
1264         }
1265
1266         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1267         assert(ctf_trace);
1268
1269         memset(&viewer_index, 0, sizeof(viewer_index));
1270
1271         /*
1272          * The viewer should not ask for index on metadata stream.
1273          */
1274         if (vstream->metadata_flag) {
1275                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1276                 goto send_reply;
1277         }
1278
1279         rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1280         assert(rstream);
1281
1282         /* Try to open an index if one is needed for that stream. */
1283         ret = try_open_index(vstream, rstream);
1284         if (ret < 0) {
1285                 if (ret == -ENOENT) {
1286                         /*
1287                          * The index is created only when the first data packet arrives, it
1288                          * might not be ready at the beginning of the session
1289                          */
1290                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1291                 } else {
1292                         /* Unhandled error. */
1293                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1294                 }
1295                 goto send_reply;
1296         }
1297
1298         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1299         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1300         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1301         if (ret < 0) {
1302                 goto end_unlock;
1303         } else if (ret == 1) {
1304                 /*
1305                  * This means the viewer index data structure has been populated by the
1306                  * check call thus we now send back the reply to the client.
1307                  */
1308                 goto send_reply;
1309         }
1310         /* At this point, ret MUST be 0 thus we continue with the get. */
1311         assert(!ret);
1312
1313         if (!ctf_trace->metadata_received ||
1314                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1315                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1316         }
1317
1318         ret = check_new_streams(conn);
1319         if (ret < 0) {
1320                 goto end_unlock;
1321         } else if (ret == 1) {
1322                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1323         }
1324
1325         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1326         pthread_mutex_lock(&vstream->overwrite_lock);
1327         if (vstream->abort_flag) {
1328                 /* The file is being overwritten by the writer, we cannot use it. */
1329                 pthread_mutex_unlock(&vstream->overwrite_lock);
1330                 ret = viewer_stream_rotate(vstream, rstream);
1331                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1332                 if (ret < 0) {
1333                         goto end_unlock;
1334                 } else if (ret == 1) {
1335                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1336                         viewer_stream_delete(vstream);
1337                         viewer_stream_destroy(ctf_trace, vstream);
1338                 } else {
1339                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1340                 }
1341                 goto send_reply;
1342         }
1343
1344         read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1345                         sizeof(packet_index));
1346         pthread_mutex_unlock(&vstream->overwrite_lock);
1347         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1348         if (read_ret < 0) {
1349                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1350                 viewer_stream_delete(vstream);
1351                 viewer_stream_destroy(ctf_trace, vstream);
1352                 goto send_reply;
1353         } else if (read_ret < sizeof(packet_index)) {
1354                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1355                 if (vstream->close_write_flag) {
1356                         ret = viewer_stream_rotate(vstream, rstream);
1357                         if (ret < 0) {
1358                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1359                                 goto end_unlock;
1360                         } else if (ret == 1) {
1361                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1362                                 viewer_stream_delete(vstream);
1363                                 viewer_stream_destroy(ctf_trace, vstream);
1364                         } else {
1365                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1366                         }
1367                 } else {
1368                         ERR("Relay reading index file %d", vstream->index_read_fd);
1369                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1370                 }
1371                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1372                 goto send_reply;
1373         } else {
1374                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1375                 vstream->last_sent_index++;
1376         }
1377
1378         /*
1379          * Indexes are stored in big endian, no need to switch before sending.
1380          */
1381         viewer_index.offset = packet_index.offset;
1382         viewer_index.packet_size = packet_index.packet_size;
1383         viewer_index.content_size = packet_index.content_size;
1384         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1385         viewer_index.timestamp_end = packet_index.timestamp_end;
1386         viewer_index.events_discarded = packet_index.events_discarded;
1387         viewer_index.stream_id = packet_index.stream_id;
1388
1389 send_reply:
1390         viewer_index.flags = htobe32(viewer_index.flags);
1391         health_code_update();
1392
1393         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1394         if (ret < 0) {
1395                 goto end_unlock;
1396         }
1397         health_code_update();
1398
1399         DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1400                         vstream->last_sent_index, vstream->stream_handle);
1401
1402 end_unlock:
1403         rcu_read_unlock();
1404
1405 end:
1406         return ret;
1407 }
1408
1409 /*
1410  * Send the next index for a stream
1411  *
1412  * Return 0 on success or else a negative value.
1413  */
1414 static
1415 int viewer_get_packet(struct relay_connection *conn)
1416 {
1417         int ret, send_data = 0;
1418         char *data = NULL;
1419         uint32_t len = 0;
1420         ssize_t read_len;
1421         struct lttng_viewer_get_packet get_packet_info;
1422         struct lttng_viewer_trace_packet reply;
1423         struct relay_viewer_stream *stream;
1424         struct relay_session *session;
1425         struct ctf_trace *ctf_trace;
1426
1427         assert(conn);
1428
1429         DBG2("Relay get data packet");
1430
1431         health_code_update();
1432
1433         ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1434         if (ret < 0) {
1435                 goto end;
1436         }
1437         health_code_update();
1438
1439         /* From this point on, the error label can be reached. */
1440         memset(&reply, 0, sizeof(reply));
1441
1442         rcu_read_lock();
1443         stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1444         if (!stream) {
1445                 goto error;
1446         }
1447
1448         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1449         if (!session) {
1450                 ret = -1;
1451                 goto error;
1452         }
1453
1454         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1455                         stream->path_name);
1456         assert(ctf_trace);
1457
1458         /*
1459          * First time we read this stream, we need open the tracefile, we should
1460          * only arrive here if an index has already been sent to the viewer, so the
1461          * tracefile must exist, if it does not it is a fatal error.
1462          */
1463         if (stream->read_fd < 0) {
1464                 char fullpath[PATH_MAX];
1465
1466                 if (stream->tracefile_count > 0) {
1467                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1468                                         stream->channel_name,
1469                                         stream->tracefile_count_current);
1470                 } else {
1471                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1472                                         stream->channel_name);
1473                 }
1474                 if (ret < 0) {
1475                         goto error;
1476                 }
1477                 ret = open(fullpath, O_RDONLY);
1478                 if (ret < 0) {
1479                         PERROR("Relay opening trace file");
1480                         goto error;
1481                 }
1482                 stream->read_fd = ret;
1483         }
1484
1485         if (!ctf_trace->metadata_received ||
1486                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1487                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1488                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1489                 goto send_reply;
1490         }
1491
1492         ret = check_new_streams(conn);
1493         if (ret < 0) {
1494                 goto end_unlock;
1495         } else if (ret == 1) {
1496                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1497                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1498                 goto send_reply;
1499         }
1500
1501         len = be32toh(get_packet_info.len);
1502         data = zmalloc(len);
1503         if (!data) {
1504                 PERROR("relay data zmalloc");
1505                 goto error;
1506         }
1507
1508         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1509         if (ret < 0) {
1510                 /*
1511                  * If the read fd was closed by the streaming side, the
1512                  * abort_flag will be set to 1, otherwise it is an error.
1513                  */
1514                 if (stream->abort_flag == 0) {
1515                         PERROR("lseek");
1516                         goto error;
1517                 }
1518                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1519                 goto send_reply;
1520         }
1521         read_len = lttng_read(stream->read_fd, data, len);
1522         if (read_len < len) {
1523                 /*
1524                  * If the read fd was closed by the streaming side, the
1525                  * abort_flag will be set to 1, otherwise it is an error.
1526                  */
1527                 if (stream->abort_flag == 0) {
1528                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1529                                         stream->read_fd,
1530                                         be64toh(get_packet_info.offset));
1531                         goto error;
1532                 } else {
1533                         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1534                         goto send_reply;
1535                 }
1536         }
1537         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1538         reply.len = htobe32(len);
1539         send_data = 1;
1540         goto send_reply;
1541
1542 error:
1543         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1544
1545 send_reply:
1546         reply.flags = htobe32(reply.flags);
1547
1548         health_code_update();
1549
1550         ret = send_response(conn->sock, &reply, sizeof(reply));
1551         if (ret < 0) {
1552                 goto end_unlock;
1553         }
1554         health_code_update();
1555
1556         if (send_data) {
1557                 health_code_update();
1558                 ret = send_response(conn->sock, data, len);
1559                 if (ret < 0) {
1560                         goto end_unlock;
1561                 }
1562                 health_code_update();
1563         }
1564
1565         DBG("Sent %u bytes for stream %" PRIu64, len,
1566                         be64toh(get_packet_info.stream_id));
1567
1568 end_unlock:
1569         free(data);
1570         rcu_read_unlock();
1571
1572 end:
1573         return ret;
1574 }
1575
1576 /*
1577  * Send the session's metadata
1578  *
1579  * Return 0 on success else a negative value.
1580  */
1581 static
1582 int viewer_get_metadata(struct relay_connection *conn)
1583 {
1584         int ret = 0;
1585         ssize_t read_len;
1586         uint64_t len = 0;
1587         char *data = NULL;
1588         struct lttng_viewer_get_metadata request;
1589         struct lttng_viewer_metadata_packet reply;
1590         struct relay_viewer_stream *stream;
1591         struct ctf_trace *ctf_trace;
1592         struct relay_session *session;
1593
1594         assert(conn);
1595
1596         DBG("Relay get metadata");
1597
1598         health_code_update();
1599
1600         ret = recv_request(conn->sock, &request, sizeof(request));
1601         if (ret < 0) {
1602                 goto end;
1603         }
1604         health_code_update();
1605
1606         memset(&reply, 0, sizeof(reply));
1607
1608         rcu_read_lock();
1609         stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1610         if (!stream || !stream->metadata_flag) {
1611                 ERR("Invalid metadata stream");
1612                 goto error;
1613         }
1614
1615         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1616         if (!session) {
1617                 ret = -1;
1618                 goto error;
1619         }
1620
1621         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1622                         stream->path_name);
1623         assert(ctf_trace);
1624         assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1625
1626         len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1627         if (len == 0) {
1628                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1629                 goto send_reply;
1630         }
1631
1632         /* first time, we open the metadata file */
1633         if (stream->read_fd < 0) {
1634                 char fullpath[PATH_MAX];
1635
1636                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1637                                 stream->channel_name);
1638                 if (ret < 0) {
1639                         goto error;
1640                 }
1641                 ret = open(fullpath, O_RDONLY);
1642                 if (ret < 0) {
1643                         PERROR("Relay opening metadata file");
1644                         goto error;
1645                 }
1646                 stream->read_fd = ret;
1647         }
1648
1649         reply.len = htobe64(len);
1650         data = zmalloc(len);
1651         if (!data) {
1652                 PERROR("viewer metadata zmalloc");
1653                 goto error;
1654         }
1655
1656         read_len = lttng_read(stream->read_fd, data, len);
1657         if (read_len < len) {
1658                 PERROR("Relay reading metadata file");
1659                 goto error;
1660         }
1661         ctf_trace->metadata_sent += read_len;
1662         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1663         goto send_reply;
1664
1665 error:
1666         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1667
1668 send_reply:
1669         health_code_update();
1670         ret = send_response(conn->sock, &reply, sizeof(reply));
1671         if (ret < 0) {
1672                 goto end_unlock;
1673         }
1674         health_code_update();
1675
1676         if (len > 0) {
1677                 ret = send_response(conn->sock, data, len);
1678                 if (ret < 0) {
1679                         goto end_unlock;
1680                 }
1681         }
1682
1683         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1684                         be64toh(request.stream_id));
1685
1686         DBG("Metadata sent");
1687
1688 end_unlock:
1689         free(data);
1690         rcu_read_unlock();
1691 end:
1692         return ret;
1693 }
1694
1695 /*
1696  * Create a viewer session.
1697  *
1698  * Return 0 on success or else a negative value.
1699  */
1700 static
1701 int viewer_create_session(struct relay_connection *conn)
1702 {
1703         int ret;
1704         struct lttng_viewer_create_session_response resp;
1705
1706         DBG("Viewer create session received");
1707
1708         memset(&resp, 0, sizeof(resp));
1709         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1710         conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1711         if (!conn->viewer_session) {
1712                 ERR("Allocation viewer session");
1713                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1714                 goto send_reply;
1715         }
1716         CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1717
1718 send_reply:
1719         health_code_update();
1720         ret = send_response(conn->sock, &resp, sizeof(resp));
1721         if (ret < 0) {
1722                 goto end;
1723         }
1724         health_code_update();
1725         ret = 0;
1726
1727 end:
1728         return ret;
1729 }
1730
1731
1732 /*
1733  * live_relay_unknown_command: send -1 if received unknown command
1734  */
1735 static
1736 void live_relay_unknown_command(struct relay_connection *conn)
1737 {
1738         struct lttcomm_relayd_generic_reply reply;
1739
1740         memset(&reply, 0, sizeof(reply));
1741         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1742         (void) send_response(conn->sock, &reply, sizeof(reply));
1743 }
1744
1745 /*
1746  * Process the commands received on the control socket
1747  */
1748 static
1749 int process_control(struct lttng_viewer_cmd *recv_hdr,
1750                 struct relay_connection *conn)
1751 {
1752         int ret = 0;
1753         uint32_t msg_value;
1754
1755         assert(recv_hdr);
1756         assert(conn);
1757
1758         msg_value = be32toh(recv_hdr->cmd);
1759
1760         /*
1761          * Make sure we've done the version check before any command other then a
1762          * new client connection.
1763          */
1764         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1765                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1766                 ret = -1;
1767                 goto end;
1768         }
1769
1770         switch (msg_value) {
1771         case LTTNG_VIEWER_CONNECT:
1772                 ret = viewer_connect(conn);
1773                 break;
1774         case LTTNG_VIEWER_LIST_SESSIONS:
1775                 ret = viewer_list_sessions(conn);
1776                 break;
1777         case LTTNG_VIEWER_ATTACH_SESSION:
1778                 ret = viewer_attach_session(conn);
1779                 break;
1780         case LTTNG_VIEWER_GET_NEXT_INDEX:
1781                 ret = viewer_get_next_index(conn);
1782                 break;
1783         case LTTNG_VIEWER_GET_PACKET:
1784                 ret = viewer_get_packet(conn);
1785                 break;
1786         case LTTNG_VIEWER_GET_METADATA:
1787                 ret = viewer_get_metadata(conn);
1788                 break;
1789         case LTTNG_VIEWER_GET_NEW_STREAMS:
1790                 ret = viewer_get_new_streams(conn);
1791                 break;
1792         case LTTNG_VIEWER_CREATE_SESSION:
1793                 ret = viewer_create_session(conn);
1794                 break;
1795         default:
1796                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1797                 live_relay_unknown_command(conn);
1798                 ret = -1;
1799                 goto end;
1800         }
1801
1802 end:
1803         return ret;
1804 }
1805
1806 static
1807 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1808 {
1809         int ret;
1810
1811         assert(events);
1812
1813         (void) lttng_poll_del(events, pollfd);
1814
1815         ret = close(pollfd);
1816         if (ret < 0) {
1817                 ERR("Closing pollfd %d", pollfd);
1818         }
1819 }
1820
1821 /*
1822  * Delete and destroy a connection.
1823  *
1824  * RCU read side lock MUST be acquired.
1825  */
1826 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1827                 struct relay_connection *conn)
1828 {
1829         struct relay_session *session, *tmp_session;
1830
1831         assert(relay_connections_ht);
1832         assert(conn);
1833
1834         connection_delete(relay_connections_ht, conn);
1835
1836         if (!conn->viewer_session) {
1837                 goto end;
1838         }
1839
1840         rcu_read_lock();
1841         cds_list_for_each_entry_safe(session, tmp_session,
1842                         &conn->viewer_session->sessions_head,
1843                         viewer_session_list) {
1844                 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1845                 cleanup_session(conn, session);
1846         }
1847         rcu_read_unlock();
1848
1849 end:
1850         connection_destroy(conn);
1851 }
1852
1853 /*
1854  * This thread does the actual work
1855  */
1856 static
1857 void *thread_worker(void *data)
1858 {
1859         int ret, err = -1;
1860         uint32_t nb_fd;
1861         struct relay_connection *conn;
1862         struct lttng_poll_event events;
1863         struct lttng_ht *relay_connections_ht;
1864         struct lttng_ht_iter iter;
1865         struct lttng_viewer_cmd recv_hdr;
1866         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1867         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1868
1869         DBG("[thread] Live viewer relay worker started");
1870
1871         rcu_register_thread();
1872
1873         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1874
1875         if (testpoint(relayd_thread_live_worker)) {
1876                 goto error_testpoint;
1877         }
1878
1879         /* table of connections indexed on socket */
1880         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1881         if (!relay_connections_ht) {
1882                 goto relay_connections_ht_error;
1883         }
1884
1885         ret = create_thread_poll_set(&events, 2);
1886         if (ret < 0) {
1887                 goto error_poll_create;
1888         }
1889
1890         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1891         if (ret < 0) {
1892                 goto error;
1893         }
1894
1895 restart:
1896         while (1) {
1897                 int i;
1898
1899                 health_code_update();
1900
1901                 /* Infinite blocking call, waiting for transmission */
1902                 DBG3("Relayd live viewer worker thread polling...");
1903                 health_poll_entry();
1904                 ret = lttng_poll_wait(&events, -1);
1905                 health_poll_exit();
1906                 if (ret < 0) {
1907                         /*
1908                          * Restart interrupted system call.
1909                          */
1910                         if (errno == EINTR) {
1911                                 goto restart;
1912                         }
1913                         goto error;
1914                 }
1915
1916                 nb_fd = ret;
1917
1918                 /*
1919                  * Process control. The control connection is prioritised so we don't
1920                  * starve it with high throughput tracing data on the data
1921                  * connection.
1922                  */
1923                 for (i = 0; i < nb_fd; i++) {
1924                         /* Fetch once the poll data */
1925                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1926                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1927
1928                         health_code_update();
1929
1930                         if (!revents) {
1931                                 /* No activity for this FD (poll implementation). */
1932                                 continue;
1933                         }
1934
1935                         /* Thread quit pipe has been closed. Killing thread. */
1936                         ret = check_thread_quit_pipe(pollfd, revents);
1937                         if (ret) {
1938                                 err = 0;
1939                                 goto exit;
1940                         }
1941
1942                         /* Inspect the relay conn pipe for new connection */
1943                         if (pollfd == live_conn_pipe[0]) {
1944                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1945                                         ERR("Relay live pipe error");
1946                                         goto error;
1947                                 } else if (revents & LPOLLIN) {
1948                                         ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1949                                         if (ret < 0) {
1950                                                 goto error;
1951                                         }
1952                                         conn->sessions_ht = sessions_ht;
1953                                         connection_init(conn);
1954                                         lttng_poll_add(&events, conn->sock->fd,
1955                                                         LPOLLIN | LPOLLRDHUP);
1956                                         rcu_read_lock();
1957                                         lttng_ht_add_unique_ulong(relay_connections_ht,
1958                                                         &conn->sock_n);
1959                                         rcu_read_unlock();
1960                                         DBG("Connection socket %d added", conn->sock->fd);
1961                                 }
1962                         } else {
1963                                 rcu_read_lock();
1964                                 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1965                                 /* If not found, there is a synchronization issue. */
1966                                 assert(conn);
1967
1968                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1969                                         cleanup_connection_pollfd(&events, pollfd);
1970                                         destroy_connection(relay_connections_ht, conn);
1971                                 } else if (revents & LPOLLIN) {
1972                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1973                                                         sizeof(recv_hdr), 0);
1974                                         if (ret <= 0) {
1975                                                 /* Connection closed */
1976                                                 cleanup_connection_pollfd(&events, pollfd);
1977                                                 destroy_connection(relay_connections_ht, conn);
1978                                                 DBG("Viewer control conn closed with %d", pollfd);
1979                                         } else {
1980                                                 ret = process_control(&recv_hdr, conn);
1981                                                 if (ret < 0) {
1982                                                         /* Clear the session on error. */
1983                                                         cleanup_connection_pollfd(&events, pollfd);
1984                                                         destroy_connection(relay_connections_ht, conn);
1985                                                         DBG("Viewer connection closed with %d", pollfd);
1986                                                 }
1987                                         }
1988                                 }
1989                                 rcu_read_unlock();
1990                         }
1991                 }
1992         }
1993
1994 exit:
1995 error:
1996         lttng_poll_clean(&events);
1997
1998         /* Cleanup reamaining connection object. */
1999         rcu_read_lock();
2000         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2001                         sock_n.node) {
2002                 health_code_update();
2003                 destroy_connection(relay_connections_ht, conn);
2004         }
2005         rcu_read_unlock();
2006 error_poll_create:
2007         lttng_ht_destroy(relay_connections_ht);
2008 relay_connections_ht_error:
2009         /* Close relay conn pipes */
2010         utils_close_pipe(live_conn_pipe);
2011         if (err) {
2012                 DBG("Viewer worker thread exited with error");
2013         }
2014         DBG("Viewer worker thread cleanup complete");
2015 error_testpoint:
2016         if (err) {
2017                 health_error();
2018                 ERR("Health error occurred in %s", __func__);
2019         }
2020         health_unregister(health_relayd);
2021         if (lttng_relay_stop_threads()) {
2022                 ERR("Error stopping threads");
2023         }
2024         rcu_unregister_thread();
2025         return NULL;
2026 }
2027
2028 /*
2029  * Create the relay command pipe to wake thread_manage_apps.
2030  * Closed in cleanup().
2031  */
2032 static int create_conn_pipe(void)
2033 {
2034         return utils_create_pipe_cloexec(live_conn_pipe);
2035 }
2036
2037 int relayd_live_join(void)
2038 {
2039         int ret, retval = 0;
2040         void *status;
2041
2042         ret = pthread_join(live_listener_thread, &status);
2043         if (ret) {
2044                 errno = ret;
2045                 PERROR("pthread_join live listener");
2046                 retval = -1;
2047         }
2048
2049         ret = pthread_join(live_worker_thread, &status);
2050         if (ret) {
2051                 errno = ret;
2052                 PERROR("pthread_join live worker");
2053                 retval = -1;
2054         }
2055
2056         ret = pthread_join(live_dispatcher_thread, &status);
2057         if (ret) {
2058                 errno = ret;
2059                 PERROR("pthread_join live dispatcher");
2060                 retval = -1;
2061         }
2062
2063         cleanup_relayd_live();
2064
2065         return retval;
2066 }
2067
2068 /*
2069  * main
2070  */
2071 int relayd_live_create(struct lttng_uri *uri,
2072                 struct relay_local_data *relay_ctx)
2073 {
2074         int ret = 0, retval = 0;
2075         void *status;
2076         int is_root;
2077
2078         if (!uri) {
2079                 retval = -1;
2080                 goto exit_init_data;
2081         }
2082         live_uri = uri;
2083
2084         /* Check if daemon is UID = 0 */
2085         is_root = !getuid();
2086
2087         if (!is_root) {
2088                 if (live_uri->port < 1024) {
2089                         ERR("Need to be root to use ports < 1024");
2090                         retval = -1;
2091                         goto exit_init_data;
2092                 }
2093         }
2094
2095         /* Setup the thread apps communication pipe. */
2096         if (create_conn_pipe()) {
2097                 retval = -1;
2098                 goto exit_init_data;
2099         }
2100
2101         /* Init relay command queue. */
2102         cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2103
2104         /* Set up max poll set size */
2105         if (lttng_poll_set_max_size()) {
2106                 retval = -1;
2107                 goto exit_init_data;
2108         }
2109
2110         /* Setup the dispatcher thread */
2111         ret = pthread_create(&live_dispatcher_thread, NULL,
2112                         thread_dispatcher, (void *) NULL);
2113         if (ret) {
2114                 errno = ret;
2115                 PERROR("pthread_create viewer dispatcher");
2116                 retval = -1;
2117                 goto exit_dispatcher_thread;
2118         }
2119
2120         /* Setup the worker thread */
2121         ret = pthread_create(&live_worker_thread, NULL,
2122                         thread_worker, relay_ctx);
2123         if (ret) {
2124                 errno = ret;
2125                 PERROR("pthread_create viewer worker");
2126                 retval = -1;
2127                 goto exit_worker_thread;
2128         }
2129
2130         /* Setup the listener thread */
2131         ret = pthread_create(&live_listener_thread, NULL,
2132                         thread_listener, (void *) NULL);
2133         if (ret) {
2134                 errno = ret;
2135                 PERROR("pthread_create viewer listener");
2136                 retval = -1;
2137                 goto exit_listener_thread;
2138         }
2139
2140         /*
2141          * All OK, started all threads.
2142          */
2143         return retval;
2144
2145
2146         ret = pthread_join(live_listener_thread, &status);
2147         if (ret) {
2148                 errno = ret;
2149                 PERROR("pthread_join live listener");
2150                 retval = -1;
2151         }
2152 exit_listener_thread:
2153
2154         ret = pthread_join(live_worker_thread, &status);
2155         if (ret) {
2156                 errno = ret;
2157                 PERROR("pthread_join live worker");
2158                 retval = -1;
2159         }
2160 exit_worker_thread:
2161
2162         ret = pthread_join(live_dispatcher_thread, &status);
2163         if (ret) {
2164                 errno = ret;
2165                 PERROR("pthread_join live dispatcher");
2166                 retval = -1;
2167         }
2168 exit_dispatcher_thread:
2169
2170 exit_init_data:
2171         cleanup_relayd_live();
2172
2173         return retval;
2174 }