src/bin/lttng-relayd/main.cpp

   1 /*
   2  * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
   3  * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
   4  * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
   5  * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   6  *
   7  * SPDX-License-Identifier: GPL-2.0-only
   8  *
   9  */
  10
  11 #define _LGPL_SOURCE
  12 #include "backward-compatibility-group-by.hpp"
  13 #include "cmd.hpp"
  14 #include "connection.hpp"
  15 #include "ctf-trace.hpp"
  16 #include "health-relayd.hpp"
  17 #include "index.hpp"
  18 #include "live.hpp"
  19 #include "lttng-relayd.hpp"
  20 #include "session.hpp"
  21 #include "sessiond-trace-chunks.hpp"
  22 #include "stream.hpp"
  23 #include "tcp_keep_alive.hpp"
  24 #include "testpoint.hpp"
  25 #include "tracefile-array.hpp"
  26 #include "utils.hpp"
  27 #include "version.hpp"
  28 #include "viewer-stream.hpp"
  29
  30 #include <common/align.hpp>
  31 #include <common/buffer-view.hpp>
  32 #include <common/common.hpp>
  33 #include <common/compat/endian.hpp>
  34 #include <common/compat/getenv.hpp>
  35 #include <common/compat/poll.hpp>
  36 #include <common/compat/socket.hpp>
  37 #include <common/daemonize.hpp>
  38 #include <common/defaults.hpp>
  39 #include <common/dynamic-buffer.hpp>
  40 #include <common/fd-tracker/fd-tracker.hpp>
  41 #include <common/fd-tracker/utils.hpp>
  42 #include <common/futex.hpp>
  43 #include <common/ini-config/ini-config.hpp>
  44 #include <common/path.hpp>
  45 #include <common/sessiond-comm/inet.hpp>
  46 #include <common/sessiond-comm/relayd.hpp>
  47 #include <common/sessiond-comm/sessiond-comm.hpp>
  48 #include <common/string-utils/format.hpp>
  49 #include <common/uri.hpp>
  50 #include <common/utils.hpp>
  51
  52 #include <lttng/lttng.h>
  53
  54 #include <algorithm>
  55 #include <ctype.h>
  56 #include <fcntl.h>
  57 #include <getopt.h>
  58 #include <grp.h>
  59 #include <inttypes.h>
  60 #include <limits.h>
  61 #include <pthread.h>
  62 #include <signal.h>
  63 #include <stdio.h>
  64 #include <stdlib.h>
  65 #include <string.h>
  66 #include <strings.h>
  67 #include <sys/mman.h>
  68 #include <sys/mount.h>
  69 #include <sys/resource.h>
  70 #include <sys/socket.h>
  71 #include <sys/stat.h>
  72 #include <sys/types.h>
  73 #include <sys/wait.h>
  74 #include <unistd.h>
  75 #include <urcu/futex.h>
  76 #include <urcu/rculist.h>
  77 #include <urcu/uatomic.h>
  78
  79 static const char *help_msg =
  80 #ifdef LTTNG_EMBED_HELP
  81 #include <lttng-relayd.8.h>
  82 #else
  83         nullptr
  84 #endif
  85         ;
  86
  87 enum relay_connection_status {
  88         RELAY_CONNECTION_STATUS_OK,
  89         /* An error occurred while processing an event on the connection. */
  90         RELAY_CONNECTION_STATUS_ERROR,
  91         /* Connection closed/shutdown cleanly. */
  92         RELAY_CONNECTION_STATUS_CLOSED,
  93 };
  94
  95 /* command line options */
  96 char *opt_output_path, *opt_working_directory;
  97 static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
  98 enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
  99
 100 /* Argument variables */
 101 int lttng_opt_quiet; /* not static in error.h */
 102 int lttng_opt_verbose; /* not static in error.h */
 103 int lttng_opt_mi; /* not static in error.h */
 104
 105 /*
 106  * We need to wait for listener and live listener threads, as well as
 107  * health check thread, before being ready to signal readiness.
 108  */
 109 #define NR_LTTNG_RELAY_READY 3
 110 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
 111
 112 /* Size of receive buffer. */
 113 #define RECV_DATA_BUFFER_SIZE 65536
 114
 115 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
 116 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
 117
 118 static struct lttng_uri *control_uri;
 119 static struct lttng_uri *data_uri;
 120 static struct lttng_uri *live_uri;
 121
 122 const char *progname;
 123
 124 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
 125 static int tracing_group_name_override;
 126
 127 const char *const config_section_name = "relayd";
 128
 129 /*
 130  * This pipe is used to inform the worker thread that a command is queued and
 131  * ready to be processed.
 132  */
 133 static int relay_conn_pipe[2] = { -1, -1 };
 134
 135 /* Shared between threads */
 136 static int dispatch_thread_exit;
 137
 138 static pthread_t listener_thread;
 139 static pthread_t dispatcher_thread;
 140 static pthread_t worker_thread;
 141 static pthread_t health_thread;
 142
 143 /*
 144  * last_relay_stream_id_lock protects last_relay_stream_id increment
 145  * atomicity on 32-bit architectures.
 146  */
 147 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
 148 static uint64_t last_relay_stream_id;
 149
 150 /*
 151  * Relay command queue.
 152  *
 153  * The relay_thread_listener and relay_thread_dispatcher communicate with this
 154  * queue.
 155  */
 156 static struct relay_conn_queue relay_conn_queue;
 157
 158 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
 159 static unsigned int lttng_opt_fd_pool_size = -1;
 160
 161 /* Global relay stream hash table. */
 162 struct lttng_ht *relay_streams_ht;
 163
 164 /* Global relay viewer stream hash table. */
 165 struct lttng_ht *viewer_streams_ht;
 166
 167 /* Global relay sessions hash table. */
 168 struct lttng_ht *sessions_ht;
 169
 170 /* Relayd health monitoring */
 171 struct health_app *health_relayd;
 172
 173 struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
 174
 175 /* Global fd tracker. */
 176 struct fd_tracker *the_fd_tracker;
 177
 178 static struct option long_options[] = {
 179         {
 180                 "control-port",
 181                 1,
 182                 nullptr,
 183                 'C',
 184         },
 185         {
 186                 "data-port",
 187                 1,
 188                 nullptr,
 189                 'D',
 190         },
 191         {
 192                 "live-port",
 193                 1,
 194                 nullptr,
 195                 'L',
 196         },
 197         {
 198                 "daemonize",
 199                 0,
 200                 nullptr,
 201                 'd',
 202         },
 203         {
 204                 "background",
 205                 0,
 206                 nullptr,
 207                 'b',
 208         },
 209         {
 210                 "group",
 211                 1,
 212                 nullptr,
 213                 'g',
 214         },
 215         {
 216                 "fd-pool-size",
 217                 1,
 218                 nullptr,
 219                 '\0',
 220         },
 221         {
 222                 "help",
 223                 0,
 224                 nullptr,
 225                 'h',
 226         },
 227         {
 228                 "output",
 229                 1,
 230                 nullptr,
 231                 'o',
 232         },
 233         {
 234                 "verbose",
 235                 0,
 236                 nullptr,
 237                 'v',
 238         },
 239         { "config", 1, nullptr, 'f' },
 240         { "version", 0, nullptr, 'V' },
 241         {
 242                 "working-directory",
 243                 1,
 244                 nullptr,
 245                 'w',
 246         },
 247         {
 248                 "group-output-by-session",
 249                 0,
 250                 nullptr,
 251                 's',
 252         },
 253         {
 254                 "group-output-by-host",
 255                 0,
 256                 nullptr,
 257                 'p',
 258         },
 259         { "disallow-clear", 0, nullptr, 'x' },
 260         {
 261                 nullptr,
 262                 0,
 263                 nullptr,
 264                 0,
 265         },
 266 };
 267
 268 static const char *config_ignore_options[] = { "help", "config", "version" };
 269
 270 static void print_version()
 271 {
 272         fprintf(stdout, "%s\n", VERSION);
 273 }
 274
 275 static void relayd_config_log()
 276 {
 277         DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
 278             GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
 279             EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
 280         if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
 281                 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
 282         }
 283         if (EXTRA_VERSION_PATCHES[0] != '\0') {
 284                 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
 285         }
 286 }
 287
 288 /*
 289  * Take an option from the getopt output and set it in the right variable to be
 290  * used later.
 291  *
 292  * Return 0 on success else a negative value.
 293  */
 294 static int set_option(int opt, const char *arg, const char *optname)
 295 {
 296         int ret;
 297
 298         switch (opt) {
 299         case 0:
 300                 if (!strcmp(optname, "fd-pool-size")) {
 301                         unsigned long v;
 302
 303                         errno = 0;
 304                         v = strtoul(arg, nullptr, 0);
 305                         if (errno != 0 || !isdigit((unsigned char) arg[0])) {
 306                                 ERR("Wrong value in --fd-pool-size parameter: %s", arg);
 307                                 ret = -1;
 308                                 goto end;
 309                         }
 310                         if (v >= UINT_MAX) {
 311                                 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s",
 312                                     arg);
 313                                 ret = -1;
 314                                 goto end;
 315                         }
 316                         lttng_opt_fd_pool_size = (unsigned int) v;
 317                 } else {
 318                         fprintf(stderr, "unknown option %s", optname);
 319                         if (arg) {
 320                                 fprintf(stderr, " with arg %s\n", arg);
 321                         }
 322                 }
 323                 break;
 324         case 'C':
 325                 if (lttng_is_setuid_setgid()) {
 326                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 327                              "-C, --control-port");
 328                 } else {
 329                         ret = uri_parse(arg, &control_uri);
 330                         if (ret < 0) {
 331                                 ERR("Invalid control URI specified");
 332                                 goto end;
 333                         }
 334                         if (control_uri->port == 0) {
 335                                 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
 336                         }
 337                 }
 338                 break;
 339         case 'D':
 340                 if (lttng_is_setuid_setgid()) {
 341                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 342                              "-D, -data-port");
 343                 } else {
 344                         ret = uri_parse(arg, &data_uri);
 345                         if (ret < 0) {
 346                                 ERR("Invalid data URI specified");
 347                                 goto end;
 348                         }
 349                         if (data_uri->port == 0) {
 350                                 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
 351                         }
 352                 }
 353                 break;
 354         case 'L':
 355                 if (lttng_is_setuid_setgid()) {
 356                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 357                              "-L, -live-port");
 358                 } else {
 359                         ret = uri_parse(arg, &live_uri);
 360                         if (ret < 0) {
 361                                 ERR("Invalid live URI specified");
 362                                 goto end;
 363                         }
 364                         if (live_uri->port == 0) {
 365                                 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
 366                         }
 367                 }
 368                 break;
 369         case 'd':
 370                 opt_daemon = 1;
 371                 break;
 372         case 'b':
 373                 opt_background = 1;
 374                 break;
 375         case 'g':
 376                 if (lttng_is_setuid_setgid()) {
 377                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 378                              "-g, --group");
 379                 } else {
 380                         tracing_group_name = strdup(arg);
 381                         if (tracing_group_name == nullptr) {
 382                                 ret = -errno;
 383                                 PERROR("strdup");
 384                                 goto end;
 385                         }
 386                         tracing_group_name_override = 1;
 387                 }
 388                 break;
 389         case 'h':
 390                 ret = utils_show_help(8, "lttng-relayd", help_msg);
 391                 if (ret) {
 392                         ERR("Cannot show --help for `lttng-relayd`");
 393                         perror("exec");
 394                 }
 395                 exit(EXIT_FAILURE);
 396         case 'V':
 397                 opt_print_version = 1;
 398                 break;
 399         case 'o':
 400                 if (lttng_is_setuid_setgid()) {
 401                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 402                              "-o, --output");
 403                 } else {
 404                         ret = asprintf(&opt_output_path, "%s", arg);
 405                         if (ret < 0) {
 406                                 ret = -errno;
 407                                 PERROR("asprintf opt_output_path");
 408                                 goto end;
 409                         }
 410                 }
 411                 break;
 412         case 'w':
 413                 if (lttng_is_setuid_setgid()) {
 414                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 415                              "-w, --working-directory");
 416                 } else {
 417                         ret = asprintf(&opt_working_directory, "%s", arg);
 418                         if (ret < 0) {
 419                                 ret = -errno;
 420                                 PERROR("asprintf opt_working_directory");
 421                                 goto end;
 422                         }
 423                 }
 424                 break;
 425
 426         case 'v':
 427                 /* Verbose level can increase using multiple -v */
 428                 if (arg) {
 429                         lttng_opt_verbose = config_parse_value(arg);
 430                 } else {
 431                         /* Only 3 level of verbosity (-vvv). */
 432                         if (lttng_opt_verbose < 3) {
 433                                 lttng_opt_verbose += 1;
 434                         }
 435                 }
 436                 break;
 437         case 's':
 438                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 439                         ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
 440                         exit(EXIT_FAILURE);
 441                 }
 442                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
 443                 break;
 444         case 'p':
 445                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 446                         ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
 447                         exit(EXIT_FAILURE);
 448                 }
 449                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 450                 break;
 451         case 'x':
 452                 /* Disallow clear */
 453                 opt_allow_clear = 0;
 454                 break;
 455         default:
 456                 /* Unknown option or other error.
 457                  * Error is printed by getopt, just return */
 458                 ret = -1;
 459                 goto end;
 460         }
 461
 462         /* All good. */
 463         ret = 0;
 464
 465 end:
 466         return ret;
 467 }
 468
 469 /*
 470  * config_entry_handler_cb used to handle options read from a config file.
 471  * See config_entry_handler_cb comment in common/config/session-config.h for the
 472  * return value conventions.
 473  */
 474 static int config_entry_handler(const struct config_entry *entry,
 475                                 void *unused __attribute__((unused)))
 476 {
 477         int ret = 0, i;
 478
 479         if (!entry || !entry->name || !entry->value) {
 480                 ret = -EINVAL;
 481                 goto end;
 482         }
 483
 484         /* Check if the option is to be ignored */
 485         for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
 486                 if (!strcmp(entry->name, config_ignore_options[i])) {
 487                         goto end;
 488                 }
 489         }
 490
 491         for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
 492                 /* Ignore if entry name is not fully matched. */
 493                 if (strcmp(entry->name, long_options[i].name) != 0) {
 494                         continue;
 495                 }
 496
 497                 /*
 498                  * If the option takes no argument on the command line,
 499                  * we have to check if the value is "true". We support
 500                  * non-zero numeric values, true, on and yes.
 501                  */
 502                 if (!long_options[i].has_arg) {
 503                         ret = config_parse_value(entry->value);
 504                         if (ret <= 0) {
 505                                 if (ret) {
 506                                         WARN("Invalid configuration value \"%s\" for option %s",
 507                                              entry->value,
 508                                              entry->name);
 509                                 }
 510                                 /* False, skip boolean config option. */
 511                                 goto end;
 512                         }
 513                 }
 514
 515                 ret = set_option(long_options[i].val, entry->value, entry->name);
 516                 goto end;
 517         }
 518
 519         WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
 520
 521 end:
 522         return ret;
 523 }
 524
 525 static int parse_env_options()
 526 {
 527         int ret = 0;
 528         char *value = nullptr;
 529
 530         value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
 531         if (value) {
 532                 opt_working_directory = strdup(value);
 533                 if (!opt_working_directory) {
 534                         ERR("Failed to allocate working directory string (\"%s\")", value);
 535                         ret = -1;
 536                 }
 537         }
 538         return ret;
 539 }
 540
 541 static int set_fd_pool_size()
 542 {
 543         int ret = 0;
 544         struct rlimit rlimit;
 545
 546         ret = getrlimit(RLIMIT_NOFILE, &rlimit);
 547         if (ret) {
 548                 PERROR("Failed to get file descriptor limit");
 549                 ret = -1;
 550                 goto end;
 551         }
 552
 553         DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
 554             (uint64_t) rlimit.rlim_cur,
 555             (uint64_t) rlimit.rlim_max);
 556         if (lttng_opt_fd_pool_size == -1) {
 557                 /* Use default value (soft limit - reserve). */
 558                 if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 559                         ERR("The process' file number limit is too low (%" PRIu64
 560                             "). The process' file number limit must be set to at least %i.",
 561                             (uint64_t) rlimit.rlim_cur,
 562                             DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 563                         ret = -1;
 564                         goto end;
 565                 }
 566                 lttng_opt_fd_pool_size = rlimit.rlim_cur - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 567                 goto end;
 568         }
 569
 570         if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 571                 ERR("File descriptor pool size must be set to at least %d",
 572                     DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 573                 ret = -1;
 574                 goto end;
 575         }
 576
 577         if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
 578                 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
 579                     ").",
 580                     lttng_opt_fd_pool_size,
 581                     (uint64_t) rlimit.rlim_cur);
 582                 ret = -1;
 583                 goto end;
 584         }
 585
 586         DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
 587             lttng_opt_fd_pool_size,
 588             lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
 589         lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 590 end:
 591         return ret;
 592 }
 593
 594 static int set_options(int argc, char **argv)
 595 {
 596         int c, ret = 0, option_index = 0, retval = 0;
 597         int orig_optopt = optopt, orig_optind = optind;
 598         char *default_address, *optstring;
 599         char *config_path = nullptr;
 600
 601         optstring = utils_generate_optstring(long_options,
 602                                              sizeof(long_options) / sizeof(struct option));
 603         if (!optstring) {
 604                 retval = -ENOMEM;
 605                 goto exit;
 606         }
 607
 608         /* Check for the --config option */
 609
 610         while ((c = getopt_long(argc, argv, optstring, long_options, &option_index)) != -1) {
 611                 if (c == '?') {
 612                         retval = -EINVAL;
 613                         goto exit;
 614                 } else if (c != 'f') {
 615                         continue;
 616                 }
 617
 618                 if (lttng_is_setuid_setgid()) {
 619                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 620                              "-f, --config");
 621                 } else {
 622                         free(config_path);
 623                         config_path = utils_expand_path(optarg);
 624                         if (!config_path) {
 625                                 ERR("Failed to resolve path: %s", optarg);
 626                         }
 627                 }
 628         }
 629
 630         ret = config_get_section_entries(
 631                 config_path, config_section_name, config_entry_handler, nullptr);
 632         if (ret) {
 633                 if (ret > 0) {
 634                         ERR("Invalid configuration option at line %i", ret);
 635                 }
 636                 retval = -1;
 637                 goto exit;
 638         }
 639
 640         /* Reset getopt's global state */
 641         optopt = orig_optopt;
 642         optind = orig_optind;
 643         while (true) {
 644                 c = getopt_long(argc, argv, optstring, long_options, &option_index);
 645                 if (c == -1) {
 646                         break;
 647                 }
 648
 649                 ret = set_option(c, optarg, long_options[option_index].name);
 650                 if (ret < 0) {
 651                         retval = -1;
 652                         goto exit;
 653                 }
 654         }
 655
 656         /* assign default values */
 657         if (control_uri == nullptr) {
 658                 ret = asprintf(&default_address,
 659                                "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
 660                                DEFAULT_NETWORK_CONTROL_PORT);
 661                 if (ret < 0) {
 662                         PERROR("asprintf default data address");
 663                         retval = -1;
 664                         goto exit;
 665                 }
 666
 667                 ret = uri_parse(default_address, &control_uri);
 668                 free(default_address);
 669                 if (ret < 0) {
 670                         ERR("Invalid control URI specified");
 671                         retval = -1;
 672                         goto exit;
 673                 }
 674         }
 675         if (data_uri == nullptr) {
 676                 ret = asprintf(&default_address,
 677                                "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
 678                                DEFAULT_NETWORK_DATA_PORT);
 679                 if (ret < 0) {
 680                         PERROR("asprintf default data address");
 681                         retval = -1;
 682                         goto exit;
 683                 }
 684
 685                 ret = uri_parse(default_address, &data_uri);
 686                 free(default_address);
 687                 if (ret < 0) {
 688                         ERR("Invalid data URI specified");
 689                         retval = -1;
 690                         goto exit;
 691                 }
 692         }
 693         if (live_uri == nullptr) {
 694                 ret = asprintf(&default_address,
 695                                "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
 696                                DEFAULT_NETWORK_VIEWER_PORT);
 697                 if (ret < 0) {
 698                         PERROR("asprintf default viewer control address");
 699                         retval = -1;
 700                         goto exit;
 701                 }
 702
 703                 ret = uri_parse(default_address, &live_uri);
 704                 free(default_address);
 705                 if (ret < 0) {
 706                         ERR("Invalid viewer control URI specified");
 707                         retval = -1;
 708                         goto exit;
 709                 }
 710         }
 711         ret = set_fd_pool_size();
 712         if (ret) {
 713                 retval = -1;
 714                 goto exit;
 715         }
 716
 717         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 718                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 719         }
 720         if (opt_allow_clear) {
 721                 /* Check if env variable exists. */
 722                 const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 723                 if (value) {
 724                         ret = config_parse_value(value);
 725                         if (ret < 0) {
 726                                 ERR("Invalid value for %s specified",
 727                                     DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 728                                 retval = -1;
 729                                 goto exit;
 730                         }
 731                         opt_allow_clear = !ret;
 732                 }
 733         }
 734
 735 exit:
 736         free(config_path);
 737         free(optstring);
 738         return retval;
 739 }
 740
 741 static void print_global_objects()
 742 {
 743         print_viewer_streams();
 744         print_relay_streams();
 745         print_sessions();
 746 }
 747
 748 static int noop_close(void *data __attribute__((unused)), int *fds __attribute__((unused)))
 749 {
 750         return 0;
 751 }
 752
 753 static void untrack_stdio()
 754 {
 755         int fds[] = { fileno(stdout), fileno(stderr) };
 756
 757         /*
 758          * noop_close is used since we don't really want to close
 759          * the stdio output fds; we merely want to stop tracking them.
 760          */
 761         (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker, fds, 2, noop_close, nullptr);
 762 }
 763
 764 /*
 765  * Cleanup the daemon
 766  */
 767 static void relayd_cleanup()
 768 {
 769         print_global_objects();
 770
 771         DBG("Cleaning up");
 772
 773         if (viewer_streams_ht)
 774                 lttng_ht_destroy(viewer_streams_ht);
 775         if (relay_streams_ht)
 776                 lttng_ht_destroy(relay_streams_ht);
 777         if (sessions_ht)
 778                 lttng_ht_destroy(sessions_ht);
 779
 780         free(opt_output_path);
 781         free(opt_working_directory);
 782
 783         if (health_relayd) {
 784                 health_app_destroy(health_relayd);
 785         }
 786         /* Close thread quit pipes */
 787         if (health_quit_pipe[0] != -1) {
 788                 (void) fd_tracker_util_pipe_close(the_fd_tracker, health_quit_pipe);
 789         }
 790         relayd_close_thread_quit_pipe();
 791         if (sessiond_trace_chunk_registry) {
 792                 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry);
 793         }
 794         if (the_fd_tracker) {
 795                 untrack_stdio();
 796                 /*
 797                  * fd_tracker_destroy() will log the contents of the fd-tracker
 798                  * if a leak is detected.
 799                  */
 800                 fd_tracker_destroy(the_fd_tracker);
 801         }
 802
 803         uri_free(control_uri);
 804         uri_free(data_uri);
 805         /* Live URI is freed in the live thread. */
 806
 807         if (tracing_group_name_override) {
 808                 free((void *) tracing_group_name);
 809         }
 810 }
 811
 812 static int notify_health_quit_pipe(int *pipe)
 813 {
 814         ssize_t ret;
 815
 816         ret = lttng_write(pipe[1], "4", 1);
 817         if (ret < 1) {
 818                 PERROR("write relay health quit");
 819                 goto end;
 820         }
 821         ret = 0;
 822 end:
 823         return ret;
 824 }
 825
 826 /*
 827  * Stop all relayd and relayd-live threads.
 828  */
 829 int lttng_relay_stop_threads()
 830 {
 831         int retval = 0;
 832
 833         /* Stopping all threads */
 834         DBG("Terminating all threads");
 835         if (relayd_notify_thread_quit_pipe()) {
 836                 ERR("write error on thread quit pipe");
 837                 retval = -1;
 838         }
 839
 840         if (notify_health_quit_pipe(health_quit_pipe)) {
 841                 ERR("write error on health quit pipe");
 842         }
 843
 844         /* Dispatch thread */
 845         CMM_STORE_SHARED(dispatch_thread_exit, 1);
 846         futex_nto1_wake(&relay_conn_queue.futex);
 847
 848         if (relayd_live_stop()) {
 849                 ERR("Error stopping live threads");
 850                 retval = -1;
 851         }
 852         return retval;
 853 }
 854
 855 /*
 856  * Signal handler for the daemon
 857  *
 858  * Simply stop all worker threads, leaving main() return gracefully after
 859  * joining all threads and calling cleanup().
 860  */
 861 static void sighandler(int sig)
 862 {
 863         switch (sig) {
 864         case SIGINT:
 865                 DBG("SIGINT caught");
 866                 if (lttng_relay_stop_threads()) {
 867                         ERR("Error stopping threads");
 868                 }
 869                 break;
 870         case SIGTERM:
 871                 DBG("SIGTERM caught");
 872                 if (lttng_relay_stop_threads()) {
 873                         ERR("Error stopping threads");
 874                 }
 875                 break;
 876         case SIGUSR1:
 877                 CMM_STORE_SHARED(recv_child_signal, 1);
 878                 break;
 879         default:
 880                 break;
 881         }
 882 }
 883
 884 /*
 885  * Setup signal handler for :
 886  *              SIGINT, SIGTERM, SIGPIPE
 887  */
 888 static int set_signal_handler()
 889 {
 890         int ret = 0;
 891         struct sigaction sa;
 892         sigset_t sigset;
 893
 894         if ((ret = sigemptyset(&sigset)) < 0) {
 895                 PERROR("sigemptyset");
 896                 return ret;
 897         }
 898
 899         sa.sa_mask = sigset;
 900         sa.sa_flags = 0;
 901
 902         sa.sa_handler = sighandler;
 903         if ((ret = sigaction(SIGTERM, &sa, nullptr)) < 0) {
 904                 PERROR("sigaction");
 905                 return ret;
 906         }
 907
 908         if ((ret = sigaction(SIGINT, &sa, nullptr)) < 0) {
 909                 PERROR("sigaction");
 910                 return ret;
 911         }
 912
 913         if ((ret = sigaction(SIGUSR1, &sa, nullptr)) < 0) {
 914                 PERROR("sigaction");
 915                 return ret;
 916         }
 917
 918         sa.sa_handler = SIG_IGN;
 919         if ((ret = sigaction(SIGPIPE, &sa, nullptr)) < 0) {
 920                 PERROR("sigaction");
 921                 return ret;
 922         }
 923
 924         DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
 925
 926         return ret;
 927 }
 928
 929 void lttng_relay_notify_ready()
 930 {
 931         /* Notify the parent of the fork() process that we are ready. */
 932         if (opt_daemon || opt_background) {
 933                 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
 934                         kill(child_ppid, SIGUSR1);
 935                 }
 936         }
 937 }
 938
 939 /*
 940  * Init health quit pipe.
 941  *
 942  * Return -1 on error or 0 if all pipes are created.
 943  */
 944 static int init_health_quit_pipe()
 945 {
 946         return fd_tracker_util_pipe_open_cloexec(
 947                 the_fd_tracker, "Health quit pipe", health_quit_pipe);
 948 }
 949
 950 static int create_sock(void *data, int *out_fd)
 951 {
 952         int ret;
 953         struct lttcomm_sock *sock = (lttcomm_sock *) data;
 954
 955         ret = lttcomm_create_sock(sock);
 956         if (ret < 0) {
 957                 goto end;
 958         }
 959
 960         *out_fd = sock->fd;
 961 end:
 962         return ret;
 963 }
 964
 965 static int close_sock(void *data, int *in_fd __attribute__((unused)))
 966 {
 967         struct lttcomm_sock *sock = (lttcomm_sock *) data;
 968
 969         return sock->ops->close(sock);
 970 }
 971
 972 static int accept_sock(void *data, int *out_fd)
 973 {
 974         int ret = 0;
 975         /* Socks is an array of in_sock, out_sock. */
 976         struct lttcomm_sock **socks = (lttcomm_sock **) data;
 977         struct lttcomm_sock *in_sock = socks[0];
 978
 979         socks[1] = in_sock->ops->accept(in_sock);
 980         if (!socks[1]) {
 981                 ret = -1;
 982                 goto end;
 983         }
 984         *out_fd = socks[1]->fd;
 985 end:
 986         return ret;
 987 }
 988
 989 /*
 990  * Create and init socket from uri.
 991  */
 992 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri, const char *name)
 993 {
 994         int ret, sock_fd;
 995         struct lttcomm_sock *sock = nullptr;
 996         char uri_str[PATH_MAX];
 997         char *formated_name = nullptr;
 998
 999         sock = lttcomm_alloc_sock_from_uri(uri);
1000         if (sock == nullptr) {
1001                 ERR("Allocating socket");
1002                 goto error;
1003         }
1004
1005         /*
1006          * Don't fail to create the socket if the name can't be built as it is
1007          * only used for debugging purposes.
1008          */
1009         ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
1010         uri_str[sizeof(uri_str) - 1] = '\0';
1011         if (ret >= 0) {
1012                 ret = asprintf(&formated_name, "%s socket @ %s", name, uri_str);
1013                 if (ret < 0) {
1014                         formated_name = nullptr;
1015                 }
1016         }
1017
1018         ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker,
1019                                                &sock_fd,
1020                                                (const char **) (formated_name ? &formated_name :
1021                                                                                 nullptr),
1022                                                1,
1023                                                create_sock,
1024                                                sock);
1025         if (ret) {
1026                 PERROR("Failed to open \"%s\" relay socket", formated_name ?: "Unknown");
1027                 goto error;
1028         }
1029         DBG("Listening on %s socket %d", name, sock->fd);
1030
1031         ret = sock->ops->bind(sock);
1032         if (ret < 0) {
1033                 PERROR("Failed to bind socket");
1034                 goto error;
1035         }
1036
1037         ret = sock->ops->listen(sock, -1);
1038         if (ret < 0) {
1039                 goto error;
1040         }
1041
1042         free(formated_name);
1043         return sock;
1044
1045 error:
1046         if (sock) {
1047                 lttcomm_destroy_sock(sock);
1048         }
1049         free(formated_name);
1050         return nullptr;
1051 }
1052
1053 static struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1054                                                const char *name)
1055 {
1056         int out_fd, ret;
1057         struct lttcomm_sock *socks[2] = { listening_sock, nullptr };
1058         struct lttcomm_sock *new_sock = nullptr;
1059
1060         ret = fd_tracker_open_unsuspendable_fd(
1061                 the_fd_tracker, &out_fd, (const char **) &name, 1, accept_sock, &socks);
1062         if (ret) {
1063                 goto end;
1064         }
1065         new_sock = socks[1];
1066         DBG("%s accepted, socket %d", name, new_sock->fd);
1067 end:
1068         return new_sock;
1069 }
1070
1071 /*
1072  * This thread manages the listening for new connections on the network
1073  */
1074 static void *relay_thread_listener(void *data __attribute__((unused)))
1075 {
1076         int i, ret, err = -1;
1077         uint32_t nb_fd;
1078         struct lttng_poll_event events;
1079         struct lttcomm_sock *control_sock, *data_sock;
1080
1081         DBG("[thread] Relay listener started");
1082
1083         rcu_register_thread();
1084         health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1085
1086         health_code_update();
1087
1088         control_sock = relay_socket_create(control_uri, "Control listener");
1089         if (!control_sock) {
1090                 goto error_sock_control;
1091         }
1092
1093         data_sock = relay_socket_create(data_uri, "Data listener");
1094         if (!data_sock) {
1095                 goto error_sock_relay;
1096         }
1097
1098         /*
1099          * Pass 3 as size here for the thread quit pipe, control and
1100          * data socket.
1101          */
1102         ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1103         if (ret < 0) {
1104                 goto error_create_poll;
1105         }
1106
1107         /* Add the control socket */
1108         ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1109         if (ret < 0) {
1110                 goto error_poll_add;
1111         }
1112
1113         /* Add the data socket */
1114         ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1115         if (ret < 0) {
1116                 goto error_poll_add;
1117         }
1118
1119         lttng_relay_notify_ready();
1120
1121         if (testpoint(relayd_thread_listener)) {
1122                 goto error_testpoint;
1123         }
1124
1125         while (true) {
1126                 health_code_update();
1127
1128                 DBG("Listener accepting connections");
1129
1130         restart:
1131                 health_poll_entry();
1132                 ret = lttng_poll_wait(&events, -1);
1133                 health_poll_exit();
1134                 if (ret < 0) {
1135                         /*
1136                          * Restart interrupted system call.
1137                          */
1138                         if (errno == EINTR) {
1139                                 goto restart;
1140                         }
1141                         goto error;
1142                 }
1143
1144                 nb_fd = ret;
1145
1146                 DBG("Relay new connection received");
1147                 for (i = 0; i < nb_fd; i++) {
1148                         /* Fetch once the poll data */
1149                         const auto revents = LTTNG_POLL_GETEV(&events, i);
1150                         const auto pollfd = LTTNG_POLL_GETFD(&events, i);
1151
1152                         health_code_update();
1153
1154                         /* Activity on thread quit pipe, exiting. */
1155                         if (relayd_is_thread_quit_pipe(pollfd)) {
1156                                 DBG("Activity on thread quit pipe");
1157                                 err = 0;
1158                                 goto exit;
1159                         }
1160
1161                         if (revents & LPOLLIN) {
1162                                 /*
1163                                  * A new connection is requested, therefore a
1164                                  * sessiond/consumerd connection is allocated in
1165                                  * this thread, enqueued to a global queue and
1166                                  * dequeued (and freed) in the worker thread.
1167                                  */
1168                                 int val = 1;
1169                                 struct relay_connection *new_conn;
1170                                 struct lttcomm_sock *newsock = nullptr;
1171                                 enum connection_type type;
1172
1173                                 if (pollfd == data_sock->fd) {
1174                                         type = RELAY_DATA;
1175                                         newsock = accept_relayd_sock(data_sock,
1176                                                                      "Data socket to relayd");
1177                                 } else {
1178                                         LTTNG_ASSERT(pollfd == control_sock->fd);
1179                                         type = RELAY_CONTROL;
1180                                         newsock = accept_relayd_sock(control_sock,
1181                                                                      "Control socket to relayd");
1182                                 }
1183                                 if (!newsock) {
1184                                         PERROR("accepting sock");
1185                                         goto error;
1186                                 }
1187
1188                                 ret = setsockopt(
1189                                         newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
1190                                 if (ret < 0) {
1191                                         PERROR("setsockopt inet");
1192                                         lttcomm_destroy_sock(newsock);
1193                                         goto error;
1194                                 }
1195
1196                                 ret = socket_apply_keep_alive_config(newsock->fd);
1197                                 if (ret < 0) {
1198                                         ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1199                                             newsock->fd);
1200                                         lttcomm_destroy_sock(newsock);
1201                                         goto error;
1202                                 }
1203
1204                                 new_conn = connection_create(newsock, type);
1205                                 if (!new_conn) {
1206                                         lttcomm_destroy_sock(newsock);
1207                                         goto error;
1208                                 }
1209
1210                                 /* Enqueue request for the dispatcher thread. */
1211                                 cds_wfcq_head_ptr_t head;
1212                                 head.h = &relay_conn_queue.head;
1213                                 cds_wfcq_enqueue(head, &relay_conn_queue.tail, &new_conn->qnode);
1214
1215                                 /*
1216                                  * Wake the dispatch queue futex.
1217                                  * Implicit memory barrier with the
1218                                  * exchange in cds_wfcq_enqueue.
1219                                  */
1220                                 futex_nto1_wake(&relay_conn_queue.futex);
1221                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1222                                 ERR("socket poll error");
1223                                 goto error;
1224                         } else {
1225                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1226                                 goto error;
1227                         }
1228                 }
1229         }
1230
1231 exit:
1232 error:
1233 error_poll_add:
1234 error_testpoint:
1235         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1236 error_create_poll:
1237         if (data_sock->fd >= 0) {
1238                 int data_sock_fd = data_sock->fd;
1239
1240                 ret = fd_tracker_close_unsuspendable_fd(
1241                         the_fd_tracker, &data_sock_fd, 1, close_sock, data_sock);
1242                 if (ret) {
1243                         PERROR("Failed to close the data listener socket file descriptor");
1244                 }
1245                 data_sock->fd = -1;
1246         }
1247         lttcomm_destroy_sock(data_sock);
1248 error_sock_relay:
1249         if (control_sock->fd >= 0) {
1250                 int control_sock_fd = control_sock->fd;
1251
1252                 ret = fd_tracker_close_unsuspendable_fd(
1253                         the_fd_tracker, &control_sock_fd, 1, close_sock, control_sock);
1254                 if (ret) {
1255                         PERROR("Failed to close the control listener socket file descriptor");
1256                 }
1257                 control_sock->fd = -1;
1258         }
1259         lttcomm_destroy_sock(control_sock);
1260 error_sock_control:
1261         if (err) {
1262                 health_error();
1263                 ERR("Health error occurred in %s", __func__);
1264         }
1265         health_unregister(health_relayd);
1266         rcu_unregister_thread();
1267         DBG("Relay listener thread cleanup complete");
1268         lttng_relay_stop_threads();
1269         return nullptr;
1270 }
1271
1272 /*
1273  * This thread manages the dispatching of the requests to worker threads
1274  */
1275 static void *relay_thread_dispatcher(void *data __attribute__((unused)))
1276 {
1277         int err = -1;
1278         ssize_t ret;
1279         struct cds_wfcq_node *node;
1280         struct relay_connection *new_conn = nullptr;
1281
1282         DBG("[thread] Relay dispatcher started");
1283
1284         health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1285
1286         if (testpoint(relayd_thread_dispatcher)) {
1287                 goto error_testpoint;
1288         }
1289
1290         health_code_update();
1291
1292         for (;;) {
1293                 health_code_update();
1294
1295                 /* Atomically prepare the queue futex */
1296                 futex_nto1_prepare(&relay_conn_queue.futex);
1297
1298                 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1299                         break;
1300                 }
1301
1302                 do {
1303                         health_code_update();
1304
1305                         /* Dequeue commands */
1306                         node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1307                                                          &relay_conn_queue.tail);
1308                         if (node == nullptr) {
1309                                 DBG("Woken up but nothing in the relay command queue");
1310                                 /* Continue thread execution */
1311                                 break;
1312                         }
1313                         new_conn = lttng::utils::container_of(node, &relay_connection::qnode);
1314
1315                         DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1316
1317                         /*
1318                          * Inform worker thread of the new request. This
1319                          * call is blocking so we can be assured that
1320                          * the data will be read at some point in time
1321                          * or wait to the end of the world :)
1322                          */
1323                         ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn)); /* NOLINT
1324                                                                                                sizeof
1325                                                                                                used
1326                                                                                                on a
1327                                                                                                pointer.
1328                                                                                              */
1329                         if (ret < 0) {
1330                                 PERROR("write connection pipe");
1331                                 connection_put(new_conn);
1332                                 goto error;
1333                         }
1334                 } while (node != nullptr);
1335
1336                 /* Futex wait on queue. Blocking call on futex() */
1337                 health_poll_entry();
1338                 futex_nto1_wait(&relay_conn_queue.futex);
1339                 health_poll_exit();
1340         }
1341
1342         /* Normal exit, no error */
1343         err = 0;
1344
1345 error:
1346 error_testpoint:
1347         if (err) {
1348                 health_error();
1349                 ERR("Health error occurred in %s", __func__);
1350         }
1351         health_unregister(health_relayd);
1352         DBG("Dispatch thread dying");
1353         lttng_relay_stop_threads();
1354         return nullptr;
1355 }
1356
1357 static bool session_streams_have_index(const struct relay_session *session)
1358 {
1359         return session->minor >= 4 && !session->snapshot;
1360 }
1361
1362 /*
1363  * Handle the RELAYD_CREATE_SESSION command.
1364  *
1365  * On success, send back the session id or else return a negative value.
1366  */
1367 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1368                                 struct relay_connection *conn,
1369                                 const struct lttng_buffer_view *payload)
1370 {
1371         int ret = 0;
1372         ssize_t send_ret;
1373         struct relay_session *session = nullptr;
1374         struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1375         char session_name[LTTNG_NAME_MAX] = {};
1376         char hostname[LTTNG_HOST_NAME_MAX] = {};
1377         uint32_t live_timer = 0;
1378         bool snapshot = false;
1379         bool session_name_contains_creation_timestamp = false;
1380         /* Left nil for peers < 2.11. */
1381         char base_path[LTTNG_PATH_MAX] = {};
1382         lttng_uuid sessiond_uuid = {};
1383         LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1384         LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1385         LTTNG_OPTIONAL(time_t) creation_time = {};
1386         struct lttng_dynamic_buffer reply_payload;
1387
1388         lttng_dynamic_buffer_init(&reply_payload);
1389
1390         if (conn->minor < 4) {
1391                 /* From 2.1 to 2.3 */
1392                 ret = 0;
1393         } else if (conn->minor >= 4 && conn->minor < 11) {
1394                 /* From 2.4 to 2.10 */
1395                 ret = cmd_create_session_2_4(
1396                         payload, session_name, hostname, &live_timer, &snapshot);
1397         } else {
1398                 bool has_current_chunk;
1399                 uint64_t current_chunk_id_value;
1400                 time_t creation_time_value;
1401                 uint64_t id_sessiond_value;
1402
1403                 /* From 2.11 to ... */
1404                 ret = cmd_create_session_2_11(payload,
1405                                               session_name,
1406                                               hostname,
1407                                               base_path,
1408                                               &live_timer,
1409                                               &snapshot,
1410                                               &id_sessiond_value,
1411                                               sessiond_uuid,
1412                                               &has_current_chunk,
1413                                               &current_chunk_id_value,
1414                                               &creation_time_value,
1415                                               &session_name_contains_creation_timestamp);
1416                 if (lttng_uuid_is_nil(sessiond_uuid)) {
1417                         /* The nil UUID is reserved for pre-2.11 clients. */
1418                         ERR("Illegal nil UUID announced by peer in create session command");
1419                         ret = -1;
1420                         goto send_reply;
1421                 }
1422                 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1423                 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1424                 if (has_current_chunk) {
1425                         LTTNG_OPTIONAL_SET(&current_chunk_id, current_chunk_id_value);
1426                 }
1427         }
1428
1429         if (ret < 0) {
1430                 goto send_reply;
1431         }
1432
1433         session = session_create(session_name,
1434                                  hostname,
1435                                  base_path,
1436                                  live_timer,
1437                                  snapshot,
1438                                  sessiond_uuid,
1439                                  id_sessiond.is_set ? &id_sessiond.value : nullptr,
1440                                  current_chunk_id.is_set ? &current_chunk_id.value : nullptr,
1441                                  creation_time.is_set ? &creation_time.value : nullptr,
1442                                  conn->major,
1443                                  conn->minor,
1444                                  session_name_contains_creation_timestamp);
1445         if (!session) {
1446                 ret = -1;
1447                 goto send_reply;
1448         }
1449         LTTNG_ASSERT(!conn->session);
1450         conn->session = session;
1451         DBG("Created session %" PRIu64, session->id);
1452
1453         reply.generic.session_id = htobe64(session->id);
1454
1455 send_reply:
1456         if (ret < 0) {
1457                 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1458         } else {
1459                 reply.generic.ret_code = htobe32(LTTNG_OK);
1460         }
1461
1462         if (conn->minor < 11) {
1463                 /* From 2.1 to 2.10 */
1464                 ret = lttng_dynamic_buffer_append(
1465                         &reply_payload, &reply.generic, sizeof(reply.generic));
1466                 if (ret) {
1467                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1468                         ret = -1;
1469                         goto end;
1470                 }
1471         } else {
1472                 const uint32_t output_path_length = session ? strlen(session->output_path) + 1 : 0;
1473
1474                 reply.output_path_length = htobe32(output_path_length);
1475                 ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
1476                 if (ret) {
1477                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1478                         goto end;
1479                 }
1480
1481                 if (output_path_length) {
1482                         ret = lttng_dynamic_buffer_append(
1483                                 &reply_payload, session->output_path, output_path_length);
1484                         if (ret) {
1485                                 ERR("Failed to append \"create session\" command reply path to payload buffer");
1486                                 goto end;
1487                         }
1488                 }
1489         }
1490
1491         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
1492         if (send_ret < (ssize_t) reply_payload.size) {
1493                 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1494                     reply_payload.size,
1495                     send_ret);
1496                 ret = -1;
1497         }
1498 end:
1499         if (ret < 0 && session) {
1500                 session_put(session);
1501         }
1502         lttng_dynamic_buffer_reset(&reply_payload);
1503         return ret;
1504 }
1505
1506 /*
1507  * When we have received all the streams and the metadata for a channel,
1508  * we make them visible to the viewer threads.
1509  */
1510 static void publish_connection_local_streams(struct relay_connection *conn)
1511 {
1512         struct relay_stream *stream;
1513         struct relay_session *session = conn->session;
1514
1515         /*
1516          * We publish all streams belonging to a session atomically wrt
1517          * session lock.
1518          */
1519         pthread_mutex_lock(&session->lock);
1520         rcu_read_lock();
1521         cds_list_for_each_entry_rcu(stream, &session->recv_list, recv_node)
1522         {
1523                 stream_publish(stream);
1524         }
1525         rcu_read_unlock();
1526
1527         /*
1528          * Inform the viewer that there are new streams in the session.
1529          */
1530         if (session->viewer_attached) {
1531                 uatomic_set(&session->new_streams, 1);
1532         }
1533         pthread_mutex_unlock(&session->lock);
1534 }
1535
1536 static int conform_channel_path(char *channel_path)
1537 {
1538         int ret = 0;
1539
1540         if (strstr("../", channel_path)) {
1541                 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1542                     channel_path);
1543                 ret = -1;
1544                 goto end;
1545         }
1546
1547         if (*channel_path == '/') {
1548                 const size_t len = strlen(channel_path);
1549
1550                 /*
1551                  * Channel paths from peers prior to 2.11 are expressed as an
1552                  * absolute path that is, in reality, relative to the relay
1553                  * daemon's output directory. Remove the leading slash so it
1554                  * is correctly interpreted as a relative path later on.
1555                  *
1556                  * len (and not len - 1) is used to copy the trailing NULL.
1557                  */
1558                 bcopy(channel_path + 1, channel_path, len);
1559         }
1560 end:
1561         return ret;
1562 }
1563
1564 /*
1565  * relay_add_stream: allocate a new stream for a session
1566  */
1567 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1568                             struct relay_connection *conn,
1569                             const struct lttng_buffer_view *payload)
1570 {
1571         int ret;
1572         ssize_t send_ret;
1573         struct relay_session *session = conn->session;
1574         struct relay_stream *stream = nullptr;
1575         struct lttcomm_relayd_status_stream reply;
1576         struct ctf_trace *trace = nullptr;
1577         uint64_t stream_handle = -1ULL;
1578         char *path_name = nullptr, *channel_name = nullptr;
1579         uint64_t tracefile_size = 0, tracefile_count = 0;
1580         LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1581
1582         if (!session || !conn->version_check_done) {
1583                 ERR("Trying to add a stream before version check");
1584                 ret = -1;
1585                 goto end_no_session;
1586         }
1587
1588         if (session->minor == 1) {
1589                 /* For 2.1 */
1590                 ret = cmd_recv_stream_2_1(payload, &path_name, &channel_name);
1591         } else if (session->minor > 1 && session->minor < 11) {
1592                 /* From 2.2 to 2.10 */
1593                 ret = cmd_recv_stream_2_2(
1594                         payload, &path_name, &channel_name, &tracefile_size, &tracefile_count);
1595         } else {
1596                 /* From 2.11 to ... */
1597                 ret = cmd_recv_stream_2_11(payload,
1598                                            &path_name,
1599                                            &channel_name,
1600                                            &tracefile_size,
1601                                            &tracefile_count,
1602                                            &stream_chunk_id.value);
1603                 stream_chunk_id.is_set = true;
1604         }
1605
1606         if (ret < 0) {
1607                 goto send_reply;
1608         }
1609
1610         if (conform_channel_path(path_name)) {
1611                 goto send_reply;
1612         }
1613
1614         /*
1615          * Backward compatibility for --group-output-by-session.
1616          * Prior to lttng 2.11, the complete path is passed by the stream.
1617          * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1618          * >=2.11 the chunk is responsible for the output path. When dealing
1619          * with producer < 2.11 the chunk output_path is the root output path
1620          * and the stream carries the complete path (path_name).
1621          * To support --group-output-by-session with older producer (<2.11), we
1622          * need to craft the path based on the stream path.
1623          */
1624         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1625                 if (conn->minor < 4) {
1626                         /*
1627                          * From 2.1 to 2.3, the session_name is not passed on
1628                          * the RELAYD_CREATE_SESSION command. The session name
1629                          * is necessary to detect the presence of a base_path
1630                          * inside the stream path. Without it we cannot perform
1631                          * a valid group-output-by-session transformation.
1632                          */
1633                         WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1634                              " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1635                              session->id,
1636                              path_name);
1637                 } else if (conn->minor >= 4 && conn->minor < 11) {
1638                         char *group_by_session_path_name;
1639
1640                         LTTNG_ASSERT(session->session_name[0] != '\0');
1641
1642                         group_by_session_path_name = backward_compat_group_by_session(
1643                                 path_name, session->session_name, session->creation_time.value);
1644                         if (!group_by_session_path_name) {
1645                                 ERR("Failed to apply group by session to stream of session %" PRIu64,
1646                                     session->id);
1647                                 goto send_reply;
1648                         }
1649
1650                         DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1651                             path_name,
1652                             group_by_session_path_name);
1653
1654                         free(path_name);
1655                         path_name = group_by_session_path_name;
1656                 }
1657         }
1658
1659         trace = ctf_trace_get_by_path_or_create(session, path_name);
1660         if (!trace) {
1661                 goto send_reply;
1662         }
1663
1664         /* This stream here has one reference on the trace. */
1665         pthread_mutex_lock(&last_relay_stream_id_lock);
1666         stream_handle = ++last_relay_stream_id;
1667         pthread_mutex_unlock(&last_relay_stream_id_lock);
1668
1669         /* We pass ownership of path_name and channel_name. */
1670         stream = stream_create(
1671                 trace, stream_handle, path_name, channel_name, tracefile_size, tracefile_count);
1672         path_name = nullptr;
1673         channel_name = nullptr;
1674
1675         /*
1676          * Streams are the owners of their trace. Reference to trace is
1677          * kept within stream_create().
1678          */
1679         ctf_trace_put(trace);
1680
1681 send_reply:
1682         memset(&reply, 0, sizeof(reply));
1683         reply.handle = htobe64(stream_handle);
1684         if (!stream) {
1685                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1686         } else {
1687                 reply.ret_code = htobe32(LTTNG_OK);
1688         }
1689
1690         send_ret = conn->sock->ops->sendmsg(
1691                 conn->sock, &reply, sizeof(struct lttcomm_relayd_status_stream), 0);
1692         if (send_ret < (ssize_t) sizeof(reply)) {
1693                 ERR("Failed to send \"add stream\" command reply (ret = %zd)", send_ret);
1694                 ret = -1;
1695         }
1696
1697 end_no_session:
1698         free(path_name);
1699         free(channel_name);
1700         return ret;
1701 }
1702
1703 /*
1704  * relay_close_stream: close a specific stream
1705  */
1706 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1707                               struct relay_connection *conn,
1708                               const struct lttng_buffer_view *payload)
1709 {
1710         int ret;
1711         ssize_t send_ret;
1712         struct relay_session *session = conn->session;
1713         struct lttcomm_relayd_close_stream stream_info;
1714         struct lttcomm_relayd_generic_reply reply;
1715         struct relay_stream *stream;
1716
1717         DBG("Close stream received");
1718
1719         if (!session || !conn->version_check_done) {
1720                 ERR("Trying to close a stream before version check");
1721                 ret = -1;
1722                 goto end_no_session;
1723         }
1724
1725         if (payload->size < sizeof(stream_info)) {
1726                 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1727                     sizeof(stream_info),
1728                     payload->size);
1729                 ret = -1;
1730                 goto end_no_session;
1731         }
1732         memcpy(&stream_info, payload->data, sizeof(stream_info));
1733         stream_info.stream_id = be64toh(stream_info.stream_id);
1734         stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1735
1736         stream = stream_get_by_id(stream_info.stream_id);
1737         if (!stream) {
1738                 ret = -1;
1739                 goto end;
1740         }
1741
1742         /*
1743          * Set last_net_seq_num before the close flag. Required by data
1744          * pending check.
1745          */
1746         pthread_mutex_lock(&stream->lock);
1747         stream->last_net_seq_num = stream_info.last_net_seq_num;
1748         pthread_mutex_unlock(&stream->lock);
1749
1750         /*
1751          * This is one of the conditions which may trigger a stream close
1752          * with the others being:
1753          *     1) A close command is received for a stream
1754          *     2) The control connection owning the stream is closed
1755          *     3) We have received all of the stream's data _after_ a close
1756          *        request.
1757          */
1758         try_stream_close(stream);
1759         stream_put(stream);
1760         ret = 0;
1761
1762 end:
1763         memset(&reply, 0, sizeof(reply));
1764         if (ret < 0) {
1765                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1766         } else {
1767                 reply.ret_code = htobe32(LTTNG_OK);
1768         }
1769         send_ret = conn->sock->ops->sendmsg(
1770                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1771         if (send_ret < (ssize_t) sizeof(reply)) {
1772                 ERR("Failed to send \"close stream\" command reply (ret = %zd)", send_ret);
1773                 ret = -1;
1774         }
1775
1776 end_no_session:
1777         return ret;
1778 }
1779
1780 /*
1781  * relay_reset_metadata: reset a metadata stream
1782  */
1783 static int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1784                                 struct relay_connection *conn,
1785                                 const struct lttng_buffer_view *payload)
1786 {
1787         int ret;
1788         ssize_t send_ret;
1789         struct relay_session *session = conn->session;
1790         struct lttcomm_relayd_reset_metadata stream_info;
1791         struct lttcomm_relayd_generic_reply reply;
1792         struct relay_stream *stream;
1793
1794         DBG("Reset metadata received");
1795
1796         if (!session || !conn->version_check_done) {
1797                 ERR("Trying to reset a metadata stream before version check");
1798                 ret = -1;
1799                 goto end_no_session;
1800         }
1801
1802         if (payload->size < sizeof(stream_info)) {
1803                 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1804                     sizeof(stream_info),
1805                     payload->size);
1806                 ret = -1;
1807                 goto end_no_session;
1808         }
1809         memcpy(&stream_info, payload->data, sizeof(stream_info));
1810         stream_info.stream_id = be64toh(stream_info.stream_id);
1811         stream_info.version = be64toh(stream_info.version);
1812
1813         DBG("Update metadata to version %" PRIu64, stream_info.version);
1814
1815         /* Unsupported for live sessions for now. */
1816         if (session->live_timer != 0) {
1817                 ret = -1;
1818                 goto end;
1819         }
1820
1821         stream = stream_get_by_id(stream_info.stream_id);
1822         if (!stream) {
1823                 ret = -1;
1824                 goto end;
1825         }
1826         pthread_mutex_lock(&stream->lock);
1827         if (!stream->is_metadata) {
1828                 ret = -1;
1829                 goto end_unlock;
1830         }
1831
1832         ret = stream_reset_file(stream);
1833         if (ret < 0) {
1834                 ERR("Failed to reset metadata stream %" PRIu64 ": stream_path = %s, channel = %s",
1835                     stream->stream_handle,
1836                     stream->path_name,
1837                     stream->channel_name);
1838                 goto end_unlock;
1839         }
1840 end_unlock:
1841         pthread_mutex_unlock(&stream->lock);
1842         stream_put(stream);
1843
1844 end:
1845         memset(&reply, 0, sizeof(reply));
1846         if (ret < 0) {
1847                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1848         } else {
1849                 reply.ret_code = htobe32(LTTNG_OK);
1850         }
1851         send_ret = conn->sock->ops->sendmsg(
1852                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1853         if (send_ret < (ssize_t) sizeof(reply)) {
1854                 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)", send_ret);
1855                 ret = -1;
1856         }
1857
1858 end_no_session:
1859         return ret;
1860 }
1861
1862 /*
1863  * relay_unknown_command: send -1 if received unknown command
1864  */
1865 static void relay_unknown_command(struct relay_connection *conn)
1866 {
1867         struct lttcomm_relayd_generic_reply reply;
1868         ssize_t send_ret;
1869
1870         memset(&reply, 0, sizeof(reply));
1871         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1872         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1873         if (send_ret < sizeof(reply)) {
1874                 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1875         }
1876 }
1877
1878 /*
1879  * relay_start: send an acknowledgment to the client to tell if we are
1880  * ready to receive data. We are ready if a session is established.
1881  */
1882 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1883                        struct relay_connection *conn,
1884                        const struct lttng_buffer_view *payload __attribute__((unused)))
1885 {
1886         int ret = 0;
1887         ssize_t send_ret;
1888         struct lttcomm_relayd_generic_reply reply;
1889         struct relay_session *session = conn->session;
1890
1891         if (!session) {
1892                 DBG("Trying to start the streaming without a session established");
1893                 ret = htobe32(LTTNG_ERR_UNK);
1894         }
1895
1896         memset(&reply, 0, sizeof(reply));
1897         reply.ret_code = htobe32(LTTNG_OK);
1898         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1899         if (send_ret < (ssize_t) sizeof(reply)) {
1900                 ERR("Failed to send \"relay_start\" command reply (ret = %zd)", send_ret);
1901                 ret = -1;
1902         }
1903
1904         return ret;
1905 }
1906
1907 /*
1908  * relay_recv_metadata: receive the metadata for the session.
1909  */
1910 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1911                                struct relay_connection *conn,
1912                                const struct lttng_buffer_view *payload)
1913 {
1914         int ret = 0;
1915         struct relay_session *session = conn->session;
1916         struct lttcomm_relayd_metadata_payload metadata_payload_header;
1917         struct relay_stream *metadata_stream;
1918         uint64_t metadata_payload_size;
1919         struct lttng_buffer_view packet_view;
1920
1921         if (!session) {
1922                 ERR("Metadata sent before version check");
1923                 ret = -1;
1924                 goto end;
1925         }
1926
1927         if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1928                 ERR("Incorrect data size");
1929                 ret = -1;
1930                 goto end;
1931         }
1932         metadata_payload_size =
1933                 recv_hdr->data_size - sizeof(struct lttcomm_relayd_metadata_payload);
1934
1935         memcpy(&metadata_payload_header, payload->data, sizeof(metadata_payload_header));
1936         metadata_payload_header.stream_id = be64toh(metadata_payload_header.stream_id);
1937         metadata_payload_header.padding_size = be32toh(metadata_payload_header.padding_size);
1938
1939         metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1940         if (!metadata_stream) {
1941                 ret = -1;
1942                 goto end;
1943         }
1944
1945         packet_view = lttng_buffer_view_from_view(
1946                 payload, sizeof(metadata_payload_header), metadata_payload_size);
1947         if (!lttng_buffer_view_is_valid(&packet_view)) {
1948                 ERR("Invalid metadata packet length announced by header");
1949                 ret = -1;
1950                 goto end_put;
1951         }
1952
1953         pthread_mutex_lock(&metadata_stream->lock);
1954         ret = stream_write(metadata_stream, &packet_view, metadata_payload_header.padding_size);
1955         pthread_mutex_unlock(&metadata_stream->lock);
1956         if (ret) {
1957                 ret = -1;
1958                 goto end_put;
1959         }
1960 end_put:
1961         stream_put(metadata_stream);
1962 end:
1963         return ret;
1964 }
1965
1966 /*
1967  * relay_send_version: send relayd version number
1968  */
1969 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1970                               struct relay_connection *conn,
1971                               const struct lttng_buffer_view *payload)
1972 {
1973         int ret;
1974         ssize_t send_ret;
1975         struct lttcomm_relayd_version reply, msg;
1976         bool compatible = true;
1977
1978         conn->version_check_done = true;
1979
1980         /* Get version from the other side. */
1981         if (payload->size < sizeof(msg)) {
1982                 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1983                     sizeof(msg),
1984                     payload->size);
1985                 ret = -1;
1986                 goto end;
1987         }
1988
1989         memcpy(&msg, payload->data, sizeof(msg));
1990         msg.major = be32toh(msg.major);
1991         msg.minor = be32toh(msg.minor);
1992
1993         memset(&reply, 0, sizeof(reply));
1994         reply.major = RELAYD_VERSION_COMM_MAJOR;
1995         reply.minor = RELAYD_VERSION_COMM_MINOR;
1996
1997         /* Major versions must be the same */
1998         if (reply.major != msg.major) {
1999                 DBG("Incompatible major versions (%u vs %u), deleting session",
2000                     reply.major,
2001                     msg.major);
2002                 compatible = false;
2003         }
2004
2005         conn->major = reply.major;
2006         /* We adapt to the lowest compatible version */
2007         if (reply.minor <= msg.minor) {
2008                 conn->minor = reply.minor;
2009         } else {
2010                 conn->minor = msg.minor;
2011         }
2012
2013         reply.major = htobe32(reply.major);
2014         reply.minor = htobe32(reply.minor);
2015         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2016         if (send_ret < (ssize_t) sizeof(reply)) {
2017                 ERR("Failed to send \"send version\" command reply (ret = %zd)", send_ret);
2018                 ret = -1;
2019                 goto end;
2020         } else {
2021                 ret = 0;
2022         }
2023
2024         if (!compatible) {
2025                 ret = -1;
2026                 goto end;
2027         }
2028
2029         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
2030
2031 end:
2032         return ret;
2033 }
2034
2035 /*
2036  * Check for data pending for a given stream id from the session daemon.
2037  */
2038 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2039                               struct relay_connection *conn,
2040                               const struct lttng_buffer_view *payload)
2041 {
2042         struct relay_session *session = conn->session;
2043         struct lttcomm_relayd_data_pending msg;
2044         struct lttcomm_relayd_generic_reply reply;
2045         struct relay_stream *stream;
2046         ssize_t send_ret;
2047         int ret;
2048         uint64_t stream_seq;
2049
2050         DBG("Data pending command received");
2051
2052         if (!session || !conn->version_check_done) {
2053                 ERR("Trying to check for data before version check");
2054                 ret = -1;
2055                 goto end_no_session;
2056         }
2057
2058         if (payload->size < sizeof(msg)) {
2059                 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2060                     sizeof(msg),
2061                     payload->size);
2062                 ret = -1;
2063                 goto end_no_session;
2064         }
2065         memcpy(&msg, payload->data, sizeof(msg));
2066         msg.stream_id = be64toh(msg.stream_id);
2067         msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2068
2069         stream = stream_get_by_id(msg.stream_id);
2070         if (stream == nullptr) {
2071                 ret = -1;
2072                 goto end;
2073         }
2074
2075         pthread_mutex_lock(&stream->lock);
2076
2077         if (session_streams_have_index(session)) {
2078                 /*
2079                  * Ensure that both the index and stream data have been
2080                  * flushed up to the requested point.
2081                  */
2082                 stream_seq = std::min(stream->prev_data_seq, stream->prev_index_seq);
2083         } else {
2084                 stream_seq = stream->prev_data_seq;
2085         }
2086         DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2087             ", prev_index_seq %" PRIu64 ", and last_seq %" PRIu64,
2088             msg.stream_id,
2089             stream->prev_data_seq,
2090             stream->prev_index_seq,
2091             msg.last_net_seq_num);
2092
2093         /* Avoid wrapping issue */
2094         if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2095                 /* Data has in fact been written and is NOT pending */
2096                 ret = 0;
2097         } else {
2098                 /* Data still being streamed thus pending */
2099                 ret = 1;
2100         }
2101
2102         stream->data_pending_check_done = true;
2103         pthread_mutex_unlock(&stream->lock);
2104
2105         stream_put(stream);
2106 end:
2107
2108         memset(&reply, 0, sizeof(reply));
2109         reply.ret_code = htobe32(ret);
2110         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2111         if (send_ret < (ssize_t) sizeof(reply)) {
2112                 ERR("Failed to send \"data pending\" command reply (ret = %zd)", send_ret);
2113                 ret = -1;
2114         }
2115
2116 end_no_session:
2117         return ret;
2118 }
2119
2120 /*
2121  * Wait for the control socket to reach a quiescent state.
2122  *
2123  * Note that for now, when receiving this command from the session
2124  * daemon, this means that every subsequent commands or data received on
2125  * the control socket has been handled. So, this is why we simply return
2126  * OK here.
2127  */
2128 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr
2129                                    __attribute__((unused)),
2130                                    struct relay_connection *conn,
2131                                    const struct lttng_buffer_view *payload)
2132 {
2133         int ret;
2134         ssize_t send_ret;
2135         struct relay_stream *stream;
2136         struct lttcomm_relayd_quiescent_control msg;
2137         struct lttcomm_relayd_generic_reply reply;
2138
2139         DBG("Checking quiescent state on control socket");
2140
2141         if (!conn->session || !conn->version_check_done) {
2142                 ERR("Trying to check for data before version check");
2143                 ret = -1;
2144                 goto end_no_session;
2145         }
2146
2147         if (payload->size < sizeof(msg)) {
2148                 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2149                     sizeof(msg),
2150                     payload->size);
2151                 ret = -1;
2152                 goto end_no_session;
2153         }
2154         memcpy(&msg, payload->data, sizeof(msg));
2155         msg.stream_id = be64toh(msg.stream_id);
2156
2157         stream = stream_get_by_id(msg.stream_id);
2158         if (!stream) {
2159                 goto reply;
2160         }
2161         pthread_mutex_lock(&stream->lock);
2162         stream->data_pending_check_done = true;
2163         pthread_mutex_unlock(&stream->lock);
2164
2165         DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2166         stream_put(stream);
2167 reply:
2168         memset(&reply, 0, sizeof(reply));
2169         reply.ret_code = htobe32(LTTNG_OK);
2170         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2171         if (send_ret < (ssize_t) sizeof(reply)) {
2172                 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)", send_ret);
2173                 ret = -1;
2174         } else {
2175                 ret = 0;
2176         }
2177
2178 end_no_session:
2179         return ret;
2180 }
2181
2182 /*
2183  * Initialize a data pending command. This means that a consumer is about
2184  * to ask for data pending for each stream it holds. Simply iterate over
2185  * all streams of a session and set the data_pending_check_done flag.
2186  *
2187  * This command returns to the client a LTTNG_OK code.
2188  */
2189 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2190                                     struct relay_connection *conn,
2191                                     const struct lttng_buffer_view *payload)
2192 {
2193         int ret;
2194         ssize_t send_ret;
2195         struct lttng_ht_iter iter;
2196         struct lttcomm_relayd_begin_data_pending msg;
2197         struct lttcomm_relayd_generic_reply reply;
2198         struct relay_stream *stream;
2199
2200         LTTNG_ASSERT(recv_hdr);
2201         LTTNG_ASSERT(conn);
2202
2203         DBG("Init streams for data pending");
2204
2205         if (!conn->session || !conn->version_check_done) {
2206                 ERR("Trying to check for data before version check");
2207                 ret = -1;
2208                 goto end_no_session;
2209         }
2210
2211         if (payload->size < sizeof(msg)) {
2212                 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2213                     sizeof(msg),
2214                     payload->size);
2215                 ret = -1;
2216                 goto end_no_session;
2217         }
2218         memcpy(&msg, payload->data, sizeof(msg));
2219         msg.session_id = be64toh(msg.session_id);
2220
2221         /*
2222          * Iterate over all streams to set the begin data pending flag.
2223          * For now, the streams are indexed by stream handle so we have
2224          * to iterate over all streams to find the one associated with
2225          * the right session_id.
2226          */
2227         rcu_read_lock();
2228         cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2229                 if (!stream_get(stream)) {
2230                         continue;
2231                 }
2232                 if (stream->trace->session->id == msg.session_id) {
2233                         pthread_mutex_lock(&stream->lock);
2234                         stream->data_pending_check_done = false;
2235                         pthread_mutex_unlock(&stream->lock);
2236                         DBG("Set begin data pending flag to stream %" PRIu64,
2237                             stream->stream_handle);
2238                 }
2239                 stream_put(stream);
2240         }
2241         rcu_read_unlock();
2242
2243         memset(&reply, 0, sizeof(reply));
2244         /* All good, send back reply. */
2245         reply.ret_code = htobe32(LTTNG_OK);
2246
2247         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2248         if (send_ret < (ssize_t) sizeof(reply)) {
2249                 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)", send_ret);
2250                 ret = -1;
2251         } else {
2252                 ret = 0;
2253         }
2254
2255 end_no_session:
2256         return ret;
2257 }
2258
2259 /*
2260  * End data pending command. This will check, for a given session id, if
2261  * each stream associated with it has its data_pending_check_done flag
2262  * set. If not, this means that the client lost track of the stream but
2263  * the data is still being streamed on our side. In this case, we inform
2264  * the client that data is in flight.
2265  *
2266  * Return to the client if there is data in flight or not with a ret_code.
2267  */
2268 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2269                                   struct relay_connection *conn,
2270                                   const struct lttng_buffer_view *payload)
2271 {
2272         int ret;
2273         ssize_t send_ret;
2274         struct lttng_ht_iter iter;
2275         struct lttcomm_relayd_end_data_pending msg;
2276         struct lttcomm_relayd_generic_reply reply;
2277         struct relay_stream *stream;
2278         uint32_t is_data_inflight = 0;
2279
2280         DBG("End data pending command");
2281
2282         if (!conn->session || !conn->version_check_done) {
2283                 ERR("Trying to check for data before version check");
2284                 ret = -1;
2285                 goto end_no_session;
2286         }
2287
2288         if (payload->size < sizeof(msg)) {
2289                 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2290                     sizeof(msg),
2291                     payload->size);
2292                 ret = -1;
2293                 goto end_no_session;
2294         }
2295         memcpy(&msg, payload->data, sizeof(msg));
2296         msg.session_id = be64toh(msg.session_id);
2297
2298         /*
2299          * Iterate over all streams to see if the begin data pending
2300          * flag is set.
2301          */
2302         rcu_read_lock();
2303         cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2304                 if (!stream_get(stream)) {
2305                         continue;
2306                 }
2307                 if (stream->trace->session->id != msg.session_id) {
2308                         stream_put(stream);
2309                         continue;
2310                 }
2311                 pthread_mutex_lock(&stream->lock);
2312                 if (!stream->data_pending_check_done) {
2313                         uint64_t stream_seq;
2314
2315                         if (session_streams_have_index(conn->session)) {
2316                                 /*
2317                                  * Ensure that both the index and stream data have been
2318                                  * flushed up to the requested point.
2319                                  */
2320                                 stream_seq =
2321                                         std::min(stream->prev_data_seq, stream->prev_index_seq);
2322                         } else {
2323                                 stream_seq = stream->prev_data_seq;
2324                         }
2325                         if (!stream->closed ||
2326                             !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2327                                 is_data_inflight = 1;
2328                                 DBG("Data is still in flight for stream %" PRIu64,
2329                                     stream->stream_handle);
2330                                 pthread_mutex_unlock(&stream->lock);
2331                                 stream_put(stream);
2332                                 break;
2333                         }
2334                 }
2335                 pthread_mutex_unlock(&stream->lock);
2336                 stream_put(stream);
2337         }
2338         rcu_read_unlock();
2339
2340         memset(&reply, 0, sizeof(reply));
2341         /* All good, send back reply. */
2342         reply.ret_code = htobe32(is_data_inflight);
2343
2344         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2345         if (send_ret < (ssize_t) sizeof(reply)) {
2346                 ERR("Failed to send \"end data pending\" command reply (ret = %zd)", send_ret);
2347                 ret = -1;
2348         } else {
2349                 ret = 0;
2350         }
2351
2352 end_no_session:
2353         return ret;
2354 }
2355
2356 /*
2357  * Receive an index for a specific stream.
2358  *
2359  * Return 0 on success else a negative value.
2360  */
2361 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2362                             struct relay_connection *conn,
2363                             const struct lttng_buffer_view *payload)
2364 {
2365         int ret;
2366         ssize_t send_ret;
2367         struct relay_session *session = conn->session;
2368         struct lttcomm_relayd_index index_info;
2369         struct lttcomm_relayd_generic_reply reply;
2370         struct relay_stream *stream;
2371         size_t msg_len;
2372
2373         LTTNG_ASSERT(conn);
2374
2375         DBG("Relay receiving index");
2376
2377         if (!session || !conn->version_check_done) {
2378                 ERR("Trying to close a stream before version check");
2379                 ret = -1;
2380                 goto end_no_session;
2381         }
2382
2383         msg_len = lttcomm_relayd_index_len(lttng_to_index_major(conn->major, conn->minor),
2384                                            lttng_to_index_minor(conn->major, conn->minor));
2385         if (payload->size < msg_len) {
2386                 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2387                     msg_len,
2388                     payload->size);
2389                 ret = -1;
2390                 goto end_no_session;
2391         }
2392         memcpy(&index_info, payload->data, msg_len);
2393         index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2394         index_info.net_seq_num = be64toh(index_info.net_seq_num);
2395         index_info.packet_size = be64toh(index_info.packet_size);
2396         index_info.content_size = be64toh(index_info.content_size);
2397         index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2398         index_info.timestamp_end = be64toh(index_info.timestamp_end);
2399         index_info.events_discarded = be64toh(index_info.events_discarded);
2400         index_info.stream_id = be64toh(index_info.stream_id);
2401
2402         if (conn->minor >= 8) {
2403                 index_info.stream_instance_id = be64toh(index_info.stream_instance_id);
2404                 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2405         } else {
2406                 index_info.stream_instance_id = -1ULL;
2407                 index_info.packet_seq_num = -1ULL;
2408         }
2409
2410         stream = stream_get_by_id(index_info.relay_stream_id);
2411         if (!stream) {
2412                 ERR("stream_get_by_id not found");
2413                 ret = -1;
2414                 goto end;
2415         }
2416
2417         pthread_mutex_lock(&stream->lock);
2418         ret = stream_add_index(stream, &index_info);
2419         pthread_mutex_unlock(&stream->lock);
2420         if (ret) {
2421                 goto end_stream_put;
2422         }
2423
2424 end_stream_put:
2425         stream_put(stream);
2426 end:
2427         memset(&reply, 0, sizeof(reply));
2428         if (ret < 0) {
2429                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2430         } else {
2431                 reply.ret_code = htobe32(LTTNG_OK);
2432         }
2433         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2434         if (send_ret < (ssize_t) sizeof(reply)) {
2435                 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2436                 ret = -1;
2437         }
2438
2439 end_no_session:
2440         return ret;
2441 }
2442
2443 /*
2444  * Receive the streams_sent message.
2445  *
2446  * Return 0 on success else a negative value.
2447  */
2448 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2449                               struct relay_connection *conn,
2450                               const struct lttng_buffer_view *payload __attribute__((unused)))
2451 {
2452         int ret;
2453         ssize_t send_ret;
2454         struct lttcomm_relayd_generic_reply reply;
2455
2456         LTTNG_ASSERT(conn);
2457
2458         DBG("Relay receiving streams_sent");
2459
2460         if (!conn->session || !conn->version_check_done) {
2461                 ERR("Trying to close a stream before version check");
2462                 ret = -1;
2463                 goto end_no_session;
2464         }
2465
2466         /*
2467          * Publish every pending stream in the connection recv list which are
2468          * now ready to be used by the viewer.
2469          */
2470         publish_connection_local_streams(conn);
2471
2472         memset(&reply, 0, sizeof(reply));
2473         reply.ret_code = htobe32(LTTNG_OK);
2474         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2475         if (send_ret < (ssize_t) sizeof(reply)) {
2476                 ERR("Failed to send \"streams sent\" command reply (ret = %zd)", send_ret);
2477                 ret = -1;
2478         } else {
2479                 /* Success. */
2480                 ret = 0;
2481         }
2482
2483 end_no_session:
2484         return ret;
2485 }
2486
2487 static ssize_t
2488 relay_unpack_rotate_streams_header(const struct lttng_buffer_view *payload,
2489                                    struct lttcomm_relayd_rotate_streams *_rotate_streams)
2490 {
2491         struct lttcomm_relayd_rotate_streams rotate_streams;
2492         /*
2493          * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2494          * This is the smallest version of this structure, but it can be larger;
2495          * this variable is updated once the proper size of the structure is known.
2496          *
2497          * See comment at the declaration of this structure for more information.
2498          */
2499         ssize_t header_len = sizeof(struct lttcomm_relayd_rotate_streams_packed);
2500         size_t expected_payload_size_no_padding, expected_payload_size_3_bytes_padding,
2501                 expected_payload_size_7_bytes_padding;
2502
2503         if (payload->size < header_len) {
2504                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2505                     header_len,
2506                     payload->size);
2507                 goto error;
2508         }
2509
2510         /*
2511          * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2512          * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2513          *
2514          * We start by "unpacking" `stream_count` to figure out the padding length
2515          * emited by our peer.
2516          */
2517         {
2518                 decltype(rotate_streams.stream_count) stream_count;
2519
2520                 memcpy(&stream_count, payload->data, sizeof(stream_count));
2521                 rotate_streams.stream_count = be32toh(stream_count);
2522         }
2523
2524         rotate_streams.new_chunk_id = LTTNG_OPTIONAL_INIT_UNSET;
2525
2526         /*
2527          * Payload size expected given the possible padding lengths in
2528          * `struct lttcomm_relayd_rotate_streams`.
2529          */
2530         expected_payload_size_no_padding =
2531                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2532                 sizeof(lttcomm_relayd_rotate_streams_packed);
2533         expected_payload_size_3_bytes_padding =
2534                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2535                 sizeof(lttcomm_relayd_rotate_streams_3_bytes_padding);
2536         expected_payload_size_7_bytes_padding =
2537                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2538                 sizeof(lttcomm_relayd_rotate_streams_7_bytes_padding);
2539
2540         if (payload->size == expected_payload_size_no_padding) {
2541                 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams;
2542
2543                 /*
2544                  * This handles cases where someone might build with
2545                  * -fpack-struct or any other toolchain that wouldn't produce
2546                  * padding to align `value`.
2547                  */
2548                 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2549
2550                 header_len = sizeof(packed_rotate_streams);
2551                 memcpy(&packed_rotate_streams, payload->data, header_len);
2552
2553                 /* Unpack the packed structure to the natively-packed version. */
2554                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2555                         .is_set = !!packed_rotate_streams.new_chunk_id.is_set,
2556                         .value = be64toh(packed_rotate_streams.new_chunk_id.value),
2557                 };
2558                 _rotate_streams->stream_count = be32toh(packed_rotate_streams.stream_count);
2559         } else if (payload->size == expected_payload_size_3_bytes_padding) {
2560                 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams;
2561
2562                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2563
2564                 header_len = sizeof(padded_rotate_streams);
2565                 memcpy(&padded_rotate_streams, payload->data, header_len);
2566
2567                 /* Unpack the 3-byte padded structure to the natively-packed version. */
2568                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2569                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2570                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2571                 };
2572                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2573         } else if (payload->size == expected_payload_size_7_bytes_padding) {
2574                 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams;
2575
2576                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2577
2578                 header_len = sizeof(padded_rotate_streams);
2579                 memcpy(&padded_rotate_streams, payload->data, header_len);
2580
2581                 /* Unpack the 7-byte padded structure to the natively-packed version. */
2582                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2583                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2584                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2585                 };
2586                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2587
2588                 header_len = sizeof(padded_rotate_streams);
2589         } else {
2590                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2591                     expected_payload_size_no_padding,
2592                     expected_payload_size_3_bytes_padding,
2593                     expected_payload_size_7_bytes_padding,
2594                     payload->size);
2595                 goto error;
2596         }
2597
2598         return header_len;
2599 error:
2600         return -1;
2601 }
2602
2603 /*
2604  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2605  * session rotation feature (not the tracefile rotation feature).
2606  */
2607 static int relay_rotate_session_streams(const struct lttcomm_relayd_hdr *recv_hdr
2608                                         __attribute__((unused)),
2609                                         struct relay_connection *conn,
2610                                         const struct lttng_buffer_view *payload)
2611 {
2612         int ret = 0;
2613         uint32_t i;
2614         ssize_t send_ret;
2615         enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2616         struct relay_session *session = conn->session;
2617         struct lttcomm_relayd_rotate_streams rotate_streams;
2618         struct lttcomm_relayd_generic_reply reply = {};
2619         struct relay_stream *stream = nullptr;
2620         struct lttng_trace_chunk *next_trace_chunk = nullptr;
2621         struct lttng_buffer_view stream_positions;
2622         char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2623         const char *chunk_id_str = "none";
2624         ssize_t header_len;
2625
2626         if (!session || !conn->version_check_done) {
2627                 ERR("Trying to rotate a stream before version check");
2628                 ret = -1;
2629                 goto end_no_reply;
2630         }
2631
2632         if (session->major == 2 && session->minor < 11) {
2633                 ERR("Unsupported feature before 2.11");
2634                 ret = -1;
2635                 goto end_no_reply;
2636         }
2637
2638         header_len = relay_unpack_rotate_streams_header(payload, &rotate_streams);
2639         if (header_len < 0) {
2640                 ret = -1;
2641                 goto end_no_reply;
2642         }
2643
2644         if (rotate_streams.new_chunk_id.is_set) {
2645                 /*
2646                  * Retrieve the trace chunk the stream must transition to. As
2647                  * per the protocol, this chunk should have been created
2648                  * before this command is received.
2649                  */
2650                 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2651                         sessiond_trace_chunk_registry,
2652                         session->sessiond_uuid,
2653                         conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2654                                                             conn->session->id,
2655                         rotate_streams.new_chunk_id.value);
2656                 if (!next_trace_chunk) {
2657                         char uuid_str[LTTNG_UUID_STR_LEN];
2658
2659                         lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2660                         ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2661                             ", trace_chunk_id = %" PRIu64,
2662                             uuid_str,
2663                             session->id,
2664                             rotate_streams.new_chunk_id.value);
2665                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2666                         ret = -1;
2667                         goto end;
2668                 }
2669
2670                 ret = snprintf(chunk_id_buf,
2671                                sizeof(chunk_id_buf),
2672                                "%" PRIu64,
2673                                rotate_streams.new_chunk_id.value);
2674                 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2675                         chunk_id_str = "formatting error";
2676                 } else {
2677                         chunk_id_str = chunk_id_buf;
2678                 }
2679         }
2680
2681         DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2682             rotate_streams.stream_count,
2683             session->session_name,
2684             chunk_id_str);
2685
2686         stream_positions = lttng_buffer_view_from_view(payload, header_len, -1);
2687         if (!stream_positions.data ||
2688             stream_positions.size < (rotate_streams.stream_count *
2689                                      sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2690                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2691                 ret = -1;
2692                 goto end;
2693         }
2694
2695         for (i = 0; i < rotate_streams.stream_count; i++) {
2696                 struct lttcomm_relayd_stream_rotation_position *position_comm =
2697                         &((typeof(position_comm)) stream_positions.data)[i];
2698                 const struct lttcomm_relayd_stream_rotation_position pos = {
2699                         .stream_id = be64toh(position_comm->stream_id),
2700                         .rotate_at_seq_num = be64toh(position_comm->rotate_at_seq_num),
2701                 };
2702
2703                 stream = stream_get_by_id(pos.stream_id);
2704                 if (!stream) {
2705                         reply_code = LTTNG_ERR_INVALID;
2706                         ret = -1;
2707                         goto end;
2708                 }
2709
2710                 pthread_mutex_lock(&stream->lock);
2711                 ret = stream_set_pending_rotation(stream, next_trace_chunk, pos.rotate_at_seq_num);
2712                 pthread_mutex_unlock(&stream->lock);
2713                 if (ret) {
2714                         reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2715                         goto end;
2716                 }
2717
2718                 stream_put(stream);
2719                 stream = nullptr;
2720         }
2721
2722         reply_code = LTTNG_OK;
2723         ret = 0;
2724 end:
2725         if (stream) {
2726                 stream_put(stream);
2727         }
2728
2729         reply.ret_code = htobe32((uint32_t) reply_code);
2730         send_ret = conn->sock->ops->sendmsg(
2731                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2732         if (send_ret < (ssize_t) sizeof(reply)) {
2733                 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)", send_ret);
2734                 ret = -1;
2735         }
2736 end_no_reply:
2737         lttng_trace_chunk_put(next_trace_chunk);
2738         return ret;
2739 }
2740
2741 /*
2742  * relay_create_trace_chunk: create a new trace chunk
2743  */
2744 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2745                                     __attribute__((unused)),
2746                                     struct relay_connection *conn,
2747                                     const struct lttng_buffer_view *payload)
2748 {
2749         int ret = 0;
2750         ssize_t send_ret;
2751         struct relay_session *session = conn->session;
2752         struct lttcomm_relayd_create_trace_chunk *msg;
2753         struct lttcomm_relayd_generic_reply reply = {};
2754         struct lttng_buffer_view header_view;
2755         struct lttng_trace_chunk *chunk = nullptr, *published_chunk = nullptr;
2756         enum lttng_error_code reply_code = LTTNG_OK;
2757         enum lttng_trace_chunk_status chunk_status;
2758         const char *new_path;
2759
2760         if (!session || !conn->version_check_done) {
2761                 ERR("Trying to create a trace chunk before version check");
2762                 ret = -1;
2763                 goto end_no_reply;
2764         }
2765
2766         if (session->major == 2 && session->minor < 11) {
2767                 ERR("Chunk creation command is unsupported before 2.11");
2768                 ret = -1;
2769                 goto end_no_reply;
2770         }
2771
2772         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2773         if (!lttng_buffer_view_is_valid(&header_view)) {
2774                 ERR("Failed to receive payload of chunk creation command");
2775                 ret = -1;
2776                 goto end_no_reply;
2777         }
2778
2779         /* Convert to host endianness. */
2780         msg = (typeof(msg)) header_view.data;
2781         msg->chunk_id = be64toh(msg->chunk_id);
2782         msg->creation_timestamp = be64toh(msg->creation_timestamp);
2783         msg->override_name_length = be32toh(msg->override_name_length);
2784
2785         pthread_mutex_lock(&conn->session->lock);
2786         session->ongoing_rotation = true;
2787         if (session->current_trace_chunk &&
2788             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2789                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2790                                                              DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2791                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2792                         ERR("Failed to rename old chunk");
2793                         ret = -1;
2794                         reply_code = LTTNG_ERR_UNK;
2795                         goto end;
2796                 }
2797         }
2798         if (!session->current_trace_chunk) {
2799                 if (!session->has_rotated) {
2800                         new_path = "";
2801                 } else {
2802                         new_path = nullptr;
2803                 }
2804         } else {
2805                 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2806         }
2807         chunk = lttng_trace_chunk_create(msg->chunk_id, msg->creation_timestamp, new_path);
2808         if (!chunk) {
2809                 ERR("Failed to create trace chunk in trace chunk creation command");
2810                 ret = -1;
2811                 reply_code = LTTNG_ERR_NOMEM;
2812                 goto end;
2813         }
2814         lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2815
2816         if (msg->override_name_length) {
2817                 const char *name;
2818                 const struct lttng_buffer_view chunk_name_view = lttng_buffer_view_from_view(
2819                         payload, sizeof(*msg), msg->override_name_length);
2820
2821                 if (!lttng_buffer_view_is_valid(&chunk_name_view)) {
2822                         ERR("Invalid payload of chunk creation command (protocol error): buffer too short for expected name length");
2823                         ret = -1;
2824                         reply_code = LTTNG_ERR_INVALID;
2825                         goto end;
2826                 }
2827
2828                 name = chunk_name_view.data;
2829                 if (name[msg->override_name_length - 1]) {
2830                         ERR("Invalid payload of chunk creation command (protocol error): name is not null-terminated");
2831                         ret = -1;
2832                         reply_code = LTTNG_ERR_INVALID;
2833                         goto end;
2834                 }
2835
2836                 chunk_status = lttng_trace_chunk_override_name(chunk, chunk_name_view.data);
2837                 switch (chunk_status) {
2838                 case LTTNG_TRACE_CHUNK_STATUS_OK:
2839                         break;
2840                 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2841                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2842                         reply_code = LTTNG_ERR_INVALID;
2843                         ret = -1;
2844                         goto end;
2845                 default:
2846                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2847                         reply_code = LTTNG_ERR_UNK;
2848                         ret = -1;
2849                         goto end;
2850                 }
2851         }
2852
2853         chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2854         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2855                 reply_code = LTTNG_ERR_UNK;
2856                 ret = -1;
2857                 goto end;
2858         }
2859
2860         LTTNG_ASSERT(conn->session->output_directory);
2861         chunk_status = lttng_trace_chunk_set_as_owner(chunk, conn->session->output_directory);
2862         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2863                 reply_code = LTTNG_ERR_UNK;
2864                 ret = -1;
2865                 goto end;
2866         }
2867
2868         published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2869                 sessiond_trace_chunk_registry,
2870                 conn->session->sessiond_uuid,
2871                 conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2872                                                     conn->session->id,
2873                 chunk);
2874         if (!published_chunk) {
2875                 char uuid_str[LTTNG_UUID_STR_LEN];
2876
2877                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2878                 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
2879                     ", chunk_id = %" PRIu64,
2880                     uuid_str,
2881                     conn->session->id,
2882                     msg->chunk_id);
2883                 ret = -1;
2884                 reply_code = LTTNG_ERR_NOMEM;
2885                 goto end;
2886         }
2887
2888         if (conn->session->pending_closure_trace_chunk) {
2889                 /*
2890                  * Invalid; this means a second create_trace_chunk command was
2891                  * received before a close_trace_chunk.
2892                  */
2893                 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2894                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2895                 ret = -1;
2896                 goto end;
2897         }
2898         conn->session->pending_closure_trace_chunk = conn->session->current_trace_chunk;
2899         conn->session->current_trace_chunk = published_chunk;
2900         published_chunk = nullptr;
2901         if (!conn->session->pending_closure_trace_chunk) {
2902                 session->ongoing_rotation = false;
2903         }
2904 end:
2905         pthread_mutex_unlock(&conn->session->lock);
2906         reply.ret_code = htobe32((uint32_t) reply_code);
2907         send_ret = conn->sock->ops->sendmsg(
2908                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2909         if (send_ret < (ssize_t) sizeof(reply)) {
2910                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
2911                 ret = -1;
2912         }
2913 end_no_reply:
2914         lttng_trace_chunk_put(chunk);
2915         lttng_trace_chunk_put(published_chunk);
2916         return ret;
2917 }
2918
2919 /*
2920  * relay_close_trace_chunk: close a trace chunk
2921  */
2922 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2923                                    __attribute__((unused)),
2924                                    struct relay_connection *conn,
2925                                    const struct lttng_buffer_view *payload)
2926 {
2927         int ret = 0, buf_ret;
2928         ssize_t send_ret;
2929         struct relay_session *session = conn->session;
2930         struct lttcomm_relayd_close_trace_chunk *msg;
2931         struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2932         struct lttng_buffer_view header_view;
2933         struct lttng_trace_chunk *chunk = nullptr;
2934         enum lttng_error_code reply_code = LTTNG_OK;
2935         enum lttng_trace_chunk_status chunk_status;
2936         uint64_t chunk_id;
2937         LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2938         time_t close_timestamp;
2939         char closed_trace_chunk_path[LTTNG_PATH_MAX];
2940         size_t path_length = 0;
2941         const char *chunk_name = nullptr;
2942         struct lttng_dynamic_buffer reply_payload;
2943         const char *new_path;
2944
2945         lttng_dynamic_buffer_init(&reply_payload);
2946
2947         if (!session || !conn->version_check_done) {
2948                 ERR("Trying to close a trace chunk before version check");
2949                 ret = -1;
2950                 goto end_no_reply;
2951         }
2952
2953         if (session->major == 2 && session->minor < 11) {
2954                 ERR("Chunk close command is unsupported before 2.11");
2955                 ret = -1;
2956                 goto end_no_reply;
2957         }
2958
2959         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2960         if (!lttng_buffer_view_is_valid(&header_view)) {
2961                 ERR("Failed to receive payload of chunk close command");
2962                 ret = -1;
2963                 goto end_no_reply;
2964         }
2965
2966         /* Convert to host endianness. */
2967         msg = (typeof(msg)) header_view.data;
2968         chunk_id = be64toh(msg->chunk_id);
2969         close_timestamp = (time_t) be64toh(msg->close_timestamp);
2970         close_command.value = (lttng_trace_chunk_command_type) be32toh(msg->close_command.value);
2971         close_command.is_set = msg->close_command.is_set;
2972
2973         chunk = sessiond_trace_chunk_registry_get_chunk(sessiond_trace_chunk_registry,
2974                                                         conn->session->sessiond_uuid,
2975                                                         conn->session->id_sessiond.is_set ?
2976                                                                 conn->session->id_sessiond.value :
2977                                                                 conn->session->id,
2978                                                         chunk_id);
2979         if (!chunk) {
2980                 char uuid_str[LTTNG_UUID_STR_LEN];
2981
2982                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2983                 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
2984                     ", chunk_id = %" PRIu64,
2985                     uuid_str,
2986                     conn->session->id,
2987                     msg->chunk_id);
2988                 ret = -1;
2989                 reply_code = LTTNG_ERR_NOMEM;
2990                 goto end;
2991         }
2992
2993         pthread_mutex_lock(&session->lock);
2994         if (close_command.is_set && close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
2995                 /*
2996                  * Clear command. It is a protocol error to ask for a
2997                  * clear on a relay which does not allow it. Querying
2998                  * the configuration allows figuring out whether
2999                  * clearing is allowed before doing the clear.
3000                  */
3001                 if (!opt_allow_clear) {
3002                         ret = -1;
3003                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3004                         goto end_unlock_session;
3005                 }
3006         }
3007         if (session->pending_closure_trace_chunk && session->pending_closure_trace_chunk != chunk) {
3008                 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3009                     session->session_name);
3010                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3011                 ret = -1;
3012                 goto end_unlock_session;
3013         }
3014
3015         if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
3016             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
3017                 if (close_command.is_set &&
3018                     close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
3019                     !session->has_rotated) {
3020                         /* New chunk stays in session output directory. */
3021                         new_path = "";
3022                 } else {
3023                         /* Use chunk name for new chunk. */
3024                         new_path = nullptr;
3025                 }
3026                 /* Rename new chunk path. */
3027                 chunk_status =
3028                         lttng_trace_chunk_rename_path(session->current_trace_chunk, new_path);
3029                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3030                         ret = -1;
3031                         goto end_unlock_session;
3032                 }
3033                 session->ongoing_rotation = false;
3034         }
3035         if ((!close_command.is_set ||
3036              close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
3037             !lttng_trace_chunk_get_name_overridden(chunk)) {
3038                 const char *old_path;
3039
3040                 if (!session->has_rotated) {
3041                         old_path = "";
3042                 } else {
3043                         old_path = nullptr;
3044                 }
3045                 /* We need to move back the .tmp_old_chunk to its rightful place. */
3046                 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
3047                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3048                         ret = -1;
3049                         goto end_unlock_session;
3050                 }
3051         }
3052         chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, close_timestamp);
3053         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3054                 ERR("Failed to set trace chunk close timestamp");
3055                 ret = -1;
3056                 reply_code = LTTNG_ERR_UNK;
3057                 goto end_unlock_session;
3058         }
3059
3060         if (close_command.is_set) {
3061                 chunk_status = lttng_trace_chunk_set_close_command(chunk, close_command.value);
3062                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3063                         ret = -1;
3064                         reply_code = LTTNG_ERR_INVALID;
3065                         goto end_unlock_session;
3066                 }
3067         }
3068         chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, nullptr);
3069         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3070                 ERR("Failed to get chunk name");
3071                 ret = -1;
3072                 reply_code = LTTNG_ERR_UNK;
3073                 goto end_unlock_session;
3074         }
3075         if (!session->has_rotated && !session->snapshot) {
3076                 ret = lttng_strncpy(closed_trace_chunk_path,
3077                                     session->output_path,
3078                                     sizeof(closed_trace_chunk_path));
3079                 if (ret) {
3080                         ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3081                             strlen(session->output_path),
3082                             sizeof(closed_trace_chunk_path));
3083                         reply_code = LTTNG_ERR_NOMEM;
3084                         ret = -1;
3085                         goto end_unlock_session;
3086                 }
3087         } else {
3088                 if (session->snapshot) {
3089                         ret = snprintf(closed_trace_chunk_path,
3090                                        sizeof(closed_trace_chunk_path),
3091                                        "%s/%s",
3092                                        session->output_path,
3093                                        chunk_name);
3094                 } else {
3095                         ret = snprintf(closed_trace_chunk_path,
3096                                        sizeof(closed_trace_chunk_path),
3097                                        "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY "/%s",
3098                                        session->output_path,
3099                                        chunk_name);
3100                 }
3101                 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3102                         ERR("Failed to format closed trace chunk resulting path");
3103                         reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3104                         ret = -1;
3105                         goto end_unlock_session;
3106                 }
3107         }
3108         if (close_command.is_set &&
3109             close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3110                 session->has_rotated = true;
3111         }
3112         DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3113         path_length = strlen(closed_trace_chunk_path) + 1;
3114         if (path_length > UINT32_MAX) {
3115                 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3116                 ret = -1;
3117                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3118                 goto end_unlock_session;
3119         }
3120
3121         if (session->current_trace_chunk == chunk) {
3122                 /*
3123                  * After a trace chunk close command, no new streams
3124                  * referencing the chunk may be created. Hence, on the
3125                  * event that no new trace chunk have been created for
3126                  * the session, the reference to the current trace chunk
3127                  * is released in order to allow it to be reclaimed when
3128                  * the last stream releases its reference to it.
3129                  */
3130                 lttng_trace_chunk_put(session->current_trace_chunk);
3131                 session->current_trace_chunk = nullptr;
3132         }
3133         lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3134         session->pending_closure_trace_chunk = nullptr;
3135 end_unlock_session:
3136         pthread_mutex_unlock(&session->lock);
3137
3138 end:
3139         reply.generic.ret_code = htobe32((uint32_t) reply_code);
3140         reply.path_length = htobe32((uint32_t) path_length);
3141         buf_ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
3142         if (buf_ret) {
3143                 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3144                 goto end_no_reply;
3145         }
3146
3147         if (reply_code == LTTNG_OK) {
3148                 buf_ret = lttng_dynamic_buffer_append(
3149                         &reply_payload, closed_trace_chunk_path, path_length);
3150                 if (buf_ret) {
3151                         ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3152                         goto end_no_reply;
3153                 }
3154         }
3155
3156         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
3157         if (send_ret < reply_payload.size) {
3158                 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3159                     reply_payload.size,
3160                     send_ret);
3161                 ret = -1;
3162                 goto end_no_reply;
3163         }
3164 end_no_reply:
3165         lttng_trace_chunk_put(chunk);
3166         lttng_dynamic_buffer_reset(&reply_payload);
3167         return ret;
3168 }
3169
3170 /*
3171  * relay_trace_chunk_exists: check if a trace chunk exists
3172  */
3173 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr
3174                                     __attribute__((unused)),
3175                                     struct relay_connection *conn,
3176                                     const struct lttng_buffer_view *payload)
3177 {
3178         int ret = 0;
3179         ssize_t send_ret;
3180         struct relay_session *session = conn->session;
3181         struct lttcomm_relayd_trace_chunk_exists *msg;
3182         struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3183         struct lttng_buffer_view header_view;
3184         uint64_t chunk_id;
3185         bool chunk_exists;
3186
3187         if (!session || !conn->version_check_done) {
3188                 ERR("Trying to check for the presence of a trace chunk before version check");
3189                 ret = -1;
3190                 goto end_no_reply;
3191         }
3192
3193         if (session->major == 2 && session->minor < 11) {
3194                 ERR("Chunk exists command is unsupported before 2.11");
3195                 ret = -1;
3196                 goto end_no_reply;
3197         }
3198
3199         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3200         if (!lttng_buffer_view_is_valid(&header_view)) {
3201                 ERR("Failed to receive payload of chunk exists command");
3202                 ret = -1;
3203                 goto end_no_reply;
3204         }
3205
3206         /* Convert to host endianness. */
3207         msg = (typeof(msg)) header_view.data;
3208         chunk_id = be64toh(msg->chunk_id);
3209
3210         ret = sessiond_trace_chunk_registry_chunk_exists(sessiond_trace_chunk_registry,
3211                                                          conn->session->sessiond_uuid,
3212                                                          conn->session->id,
3213                                                          chunk_id,
3214                                                          &chunk_exists);
3215         /*
3216          * If ret is not 0, send the reply and report the error to the caller.
3217          * It is a protocol (or internal) error and the session/connection
3218          * should be torn down.
3219          */
3220         reply.generic.ret_code =
3221                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3222         reply.trace_chunk_exists = ret == 0 ? chunk_exists : 0;
3223
3224         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3225         if (send_ret < (ssize_t) sizeof(reply)) {
3226                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
3227                 ret = -1;
3228         }
3229 end_no_reply:
3230         return ret;
3231 }
3232
3233 /*
3234  * relay_get_configuration: query whether feature is available
3235  */
3236 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr
3237                                    __attribute__((unused)),
3238                                    struct relay_connection *conn,
3239                                    const struct lttng_buffer_view *payload)
3240 {
3241         int ret = 0;
3242         ssize_t send_ret;
3243         struct lttcomm_relayd_get_configuration *msg;
3244         struct lttcomm_relayd_get_configuration_reply reply = {};
3245         struct lttng_buffer_view header_view;
3246         uint64_t query_flags = 0;
3247         uint64_t result_flags = 0;
3248
3249         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3250         if (!lttng_buffer_view_is_valid(&header_view)) {
3251                 ERR("Failed to receive payload of chunk close command");
3252                 ret = -1;
3253                 goto end_no_reply;
3254         }
3255
3256         /* Convert to host endianness. */
3257         msg = (typeof(msg)) header_view.data;
3258         query_flags = be64toh(msg->query_flags);
3259
3260         if (query_flags) {
3261                 ret = LTTNG_ERR_INVALID_PROTOCOL;
3262                 goto reply;
3263         }
3264         if (opt_allow_clear) {
3265                 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3266         }
3267         ret = 0;
3268 reply:
3269         reply.generic.ret_code =
3270                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3271         reply.relayd_configuration_flags = htobe64(result_flags);
3272
3273         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3274         if (send_ret < (ssize_t) sizeof(reply)) {
3275                 ERR("Failed to send \"get configuration\" command reply (ret = %zd)", send_ret);
3276                 ret = -1;
3277         }
3278 end_no_reply:
3279         return ret;
3280 }
3281
3282 static int relay_process_control_command(struct relay_connection *conn,
3283                                          const struct lttcomm_relayd_hdr *header,
3284                                          const struct lttng_buffer_view *payload)
3285 {
3286         int ret = 0;
3287
3288         DBG3("Processing \"%s\" command for socket %i",
3289              lttcomm_relayd_command_str((lttcomm_relayd_command) header->cmd),
3290              conn->sock->fd);
3291         switch (header->cmd) {
3292         case RELAYD_CREATE_SESSION:
3293                 ret = relay_create_session(header, conn, payload);
3294                 break;
3295         case RELAYD_ADD_STREAM:
3296                 ret = relay_add_stream(header, conn, payload);
3297                 break;
3298         case RELAYD_START_DATA:
3299                 ret = relay_start(header, conn, payload);
3300                 break;
3301         case RELAYD_SEND_METADATA:
3302                 ret = relay_recv_metadata(header, conn, payload);
3303                 break;
3304         case RELAYD_VERSION:
3305                 ret = relay_send_version(header, conn, payload);
3306                 break;
3307         case RELAYD_CLOSE_STREAM:
3308                 ret = relay_close_stream(header, conn, payload);
3309                 break;
3310         case RELAYD_DATA_PENDING:
3311                 ret = relay_data_pending(header, conn, payload);
3312                 break;
3313         case RELAYD_QUIESCENT_CONTROL:
3314                 ret = relay_quiescent_control(header, conn, payload);
3315                 break;
3316         case RELAYD_BEGIN_DATA_PENDING:
3317                 ret = relay_begin_data_pending(header, conn, payload);
3318                 break;
3319         case RELAYD_END_DATA_PENDING:
3320                 ret = relay_end_data_pending(header, conn, payload);
3321                 break;
3322         case RELAYD_SEND_INDEX:
3323                 ret = relay_recv_index(header, conn, payload);
3324                 break;
3325         case RELAYD_STREAMS_SENT:
3326                 ret = relay_streams_sent(header, conn, payload);
3327                 break;
3328         case RELAYD_RESET_METADATA:
3329                 ret = relay_reset_metadata(header, conn, payload);
3330                 break;
3331         case RELAYD_ROTATE_STREAMS:
3332                 ret = relay_rotate_session_streams(header, conn, payload);
3333                 break;
3334         case RELAYD_CREATE_TRACE_CHUNK:
3335                 ret = relay_create_trace_chunk(header, conn, payload);
3336                 break;
3337         case RELAYD_CLOSE_TRACE_CHUNK:
3338                 ret = relay_close_trace_chunk(header, conn, payload);
3339                 break;
3340         case RELAYD_TRACE_CHUNK_EXISTS:
3341                 ret = relay_trace_chunk_exists(header, conn, payload);
3342                 break;
3343         case RELAYD_GET_CONFIGURATION:
3344                 ret = relay_get_configuration(header, conn, payload);
3345                 break;
3346         case RELAYD_UPDATE_SYNC_INFO:
3347         default:
3348                 ERR("Received unknown command (%u)", header->cmd);
3349                 relay_unknown_command(conn);
3350                 ret = -1;
3351                 goto end;
3352         }
3353
3354 end:
3355         return ret;
3356 }
3357
3358 static enum relay_connection_status
3359 relay_process_control_receive_payload(struct relay_connection *conn)
3360 {
3361         int ret = 0;
3362         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3363         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3364         struct ctrl_connection_state_receive_payload *state =
3365                 &conn->protocol.ctrl.state.receive_payload;
3366         struct lttng_buffer_view payload_view;
3367
3368         if (state->left_to_receive == 0) {
3369                 /* Short-circuit for payload-less commands. */
3370                 goto reception_complete;
3371         }
3372
3373         ret = conn->sock->ops->recvmsg(conn->sock,
3374                                        reception_buffer->data + state->received,
3375                                        state->left_to_receive,
3376                                        MSG_DONTWAIT);
3377         if (ret < 0) {
3378                 DIAGNOSTIC_PUSH
3379                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3380                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3381                         DIAGNOSTIC_POP
3382                         PERROR("Unable to receive command payload on sock %d", conn->sock->fd);
3383                         status = RELAY_CONNECTION_STATUS_ERROR;
3384                 }
3385                 goto end;
3386         } else if (ret == 0) {
3387                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3388                 status = RELAY_CONNECTION_STATUS_CLOSED;
3389                 goto end;
3390         }
3391
3392         LTTNG_ASSERT(ret > 0);
3393         LTTNG_ASSERT(ret <= state->left_to_receive);
3394
3395         state->left_to_receive -= ret;
3396         state->received += ret;
3397
3398         if (state->left_to_receive > 0) {
3399                 /*
3400                  * Can't transition to the protocol's next state, wait to
3401                  * receive the rest of the header.
3402                  */
3403                 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
3404                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3405                      state->received,
3406                      state->left_to_receive,
3407                      conn->sock->fd);
3408                 goto end;
3409         }
3410
3411 reception_complete:
3412         DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3413             conn->sock->fd,
3414             state->received);
3415         /*
3416          * The payload required to process the command has been received.
3417          * A view to the reception buffer is forwarded to the various
3418          * commands and the state of the control is reset on success.
3419          *
3420          * Commands are responsible for sending their reply to the peer.
3421          */
3422         payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer, 0, -1);
3423         ret = relay_process_control_command(conn, &state->header, &payload_view);
3424         if (ret < 0) {
3425                 status = RELAY_CONNECTION_STATUS_ERROR;
3426                 goto end;
3427         }
3428
3429         ret = connection_reset_protocol_state(conn);
3430         if (ret) {
3431                 status = RELAY_CONNECTION_STATUS_ERROR;
3432         }
3433 end:
3434         return status;
3435 }
3436
3437 static enum relay_connection_status
3438 relay_process_control_receive_header(struct relay_connection *conn)
3439 {
3440         int ret = 0;
3441         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3442         struct lttcomm_relayd_hdr header;
3443         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3444         struct ctrl_connection_state_receive_header *state =
3445                 &conn->protocol.ctrl.state.receive_header;
3446
3447         LTTNG_ASSERT(state->left_to_receive != 0);
3448
3449         ret = conn->sock->ops->recvmsg(conn->sock,
3450                                        reception_buffer->data + state->received,
3451                                        state->left_to_receive,
3452                                        MSG_DONTWAIT);
3453         if (ret < 0) {
3454                 DIAGNOSTIC_PUSH
3455                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3456                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3457                         DIAGNOSTIC_POP
3458                         PERROR("Unable to receive control command header on sock %d",
3459                                conn->sock->fd);
3460                         status = RELAY_CONNECTION_STATUS_ERROR;
3461                 }
3462                 goto end;
3463         } else if (ret == 0) {
3464                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3465                 status = RELAY_CONNECTION_STATUS_CLOSED;
3466                 goto end;
3467         }
3468
3469         LTTNG_ASSERT(ret > 0);
3470         LTTNG_ASSERT(ret <= state->left_to_receive);
3471
3472         state->left_to_receive -= ret;
3473         state->received += ret;
3474
3475         if (state->left_to_receive > 0) {
3476                 /*
3477                  * Can't transition to the protocol's next state, wait to
3478                  * receive the rest of the header.
3479                  */
3480                 DBG3("Partial reception of control connection protocol header (received %" PRIu64
3481                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3482                      state->received,
3483                      state->left_to_receive,
3484                      conn->sock->fd);
3485                 goto end;
3486         }
3487
3488         /* Transition to next state: receiving the command's payload. */
3489         conn->protocol.ctrl.state_id = CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3490         memcpy(&header, reception_buffer->data, sizeof(header));
3491         header.circuit_id = be64toh(header.circuit_id);
3492         header.data_size = be64toh(header.data_size);
3493         header.cmd = be32toh(header.cmd);
3494         header.cmd_version = be32toh(header.cmd_version);
3495         memcpy(&conn->protocol.ctrl.state.receive_payload.header, &header, sizeof(header));
3496
3497         DBG("Done receiving control command header: fd = %i, cmd = %s, cmd_version = %" PRIu32
3498             ", payload size = %" PRIu64 " bytes",
3499             conn->sock->fd,
3500             lttcomm_relayd_command_str((enum lttcomm_relayd_command) header.cmd),
3501             header.cmd_version,
3502             header.data_size);
3503
3504         if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3505                 ERR("Command header indicates a payload (%" PRIu64
3506                     " bytes) that exceeds the maximal payload size allowed on a control connection.",
3507                     header.data_size);
3508                 status = RELAY_CONNECTION_STATUS_ERROR;
3509                 goto end;
3510         }
3511
3512         conn->protocol.ctrl.state.receive_payload.left_to_receive = header.data_size;
3513         conn->protocol.ctrl.state.receive_payload.received = 0;
3514         ret = lttng_dynamic_buffer_set_size(reception_buffer, header.data_size);
3515         if (ret) {
3516                 status = RELAY_CONNECTION_STATUS_ERROR;
3517                 goto end;
3518         }
3519
3520         if (header.data_size == 0) {
3521                 /*
3522                  * Manually invoke the next state as the poll loop
3523                  * will not wake-up to allow us to proceed further.
3524                  */
3525                 status = relay_process_control_receive_payload(conn);
3526         }
3527 end:
3528         return status;
3529 }
3530
3531 /*
3532  * Process the commands received on the control socket
3533  */
3534 static enum relay_connection_status relay_process_control(struct relay_connection *conn)
3535 {
3536         enum relay_connection_status status;
3537
3538         switch (conn->protocol.ctrl.state_id) {
3539         case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3540                 status = relay_process_control_receive_header(conn);
3541                 break;
3542         case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3543                 status = relay_process_control_receive_payload(conn);
3544                 break;
3545         default:
3546                 ERR("Unknown control connection protocol state encountered.");
3547                 abort();
3548         }
3549
3550         return status;
3551 }
3552
3553 static enum relay_connection_status relay_process_data_receive_header(struct relay_connection *conn)
3554 {
3555         int ret;
3556         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3557         struct data_connection_state_receive_header *state =
3558                 &conn->protocol.data.state.receive_header;
3559         struct lttcomm_relayd_data_hdr header;
3560         struct relay_stream *stream;
3561
3562         LTTNG_ASSERT(state->left_to_receive != 0);
3563
3564         ret = conn->sock->ops->recvmsg(conn->sock,
3565                                        state->header_reception_buffer + state->received,
3566                                        state->left_to_receive,
3567                                        MSG_DONTWAIT);
3568         if (ret < 0) {
3569                 DIAGNOSTIC_PUSH
3570                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3571                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3572                         DIAGNOSTIC_POP
3573                         PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3574                         status = RELAY_CONNECTION_STATUS_ERROR;
3575                 }
3576                 goto end;
3577         } else if (ret == 0) {
3578                 /* Orderly shutdown. Not necessary to print an error. */
3579                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3580                 status = RELAY_CONNECTION_STATUS_CLOSED;
3581                 goto end;
3582         }
3583
3584         LTTNG_ASSERT(ret > 0);
3585         LTTNG_ASSERT(ret <= state->left_to_receive);
3586
3587         state->left_to_receive -= ret;
3588         state->received += ret;
3589
3590         if (state->left_to_receive > 0) {
3591                 /*
3592                  * Can't transition to the protocol's next state, wait to
3593                  * receive the rest of the header.
3594                  */
3595                 DBG3("Partial reception of data connection header (received %" PRIu64
3596                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3597                      state->received,
3598                      state->left_to_receive,
3599                      conn->sock->fd);
3600                 goto end;
3601         }
3602
3603         /* Transition to next state: receiving the payload. */
3604         conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3605
3606         memcpy(&header, state->header_reception_buffer, sizeof(header));
3607         header.circuit_id = be64toh(header.circuit_id);
3608         header.stream_id = be64toh(header.stream_id);
3609         header.data_size = be32toh(header.data_size);
3610         header.net_seq_num = be64toh(header.net_seq_num);
3611         header.padding_size = be32toh(header.padding_size);
3612         memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3613
3614         conn->protocol.data.state.receive_payload.left_to_receive = header.data_size;
3615         conn->protocol.data.state.receive_payload.received = 0;
3616         conn->protocol.data.state.receive_payload.rotate_index = false;
3617
3618         DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
3619             ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64
3620             ", padding_size = %" PRIu32,
3621             conn->sock->fd,
3622             header.circuit_id,
3623             header.stream_id,
3624             header.data_size,
3625             header.net_seq_num,
3626             header.padding_size);
3627
3628         stream = stream_get_by_id(header.stream_id);
3629         if (!stream) {
3630                 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3631                     header.stream_id);
3632                 /* Protocol error. */
3633                 status = RELAY_CONNECTION_STATUS_ERROR;
3634                 goto end;
3635         }
3636
3637         pthread_mutex_lock(&stream->lock);
3638         /* Prepare stream for the reception of a new packet. */
3639         ret = stream_init_packet(
3640                 stream, header.data_size, &conn->protocol.data.state.receive_payload.rotate_index);
3641         pthread_mutex_unlock(&stream->lock);
3642         if (ret) {
3643                 ERR("Failed to rotate stream output file");
3644                 status = RELAY_CONNECTION_STATUS_ERROR;
3645                 goto end_stream_unlock;
3646         }
3647
3648 end_stream_unlock:
3649         stream_put(stream);
3650 end:
3651         return status;
3652 }
3653
3654 static enum relay_connection_status
3655 relay_process_data_receive_payload(struct relay_connection *conn)
3656 {
3657         int ret;
3658         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3659         struct relay_stream *stream;
3660         struct data_connection_state_receive_payload *state =
3661                 &conn->protocol.data.state.receive_payload;
3662         const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3663         char data_buffer[chunk_size];
3664         bool partial_recv = false;
3665         bool new_stream = false, close_requested = false, index_flushed = false;
3666         uint64_t left_to_receive = state->left_to_receive;
3667         struct relay_session *session;
3668
3669         DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64
3670              " bytes received, %" PRIu64 " bytes left to receive",
3671              state->header.stream_id,
3672              state->header.net_seq_num,
3673              state->received,
3674              left_to_receive);
3675
3676         stream = stream_get_by_id(state->header.stream_id);
3677         if (!stream) {
3678                 /* Protocol error. */
3679                 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3680                     state->header.stream_id);
3681                 status = RELAY_CONNECTION_STATUS_ERROR;
3682                 goto end;
3683         }
3684
3685         pthread_mutex_lock(&stream->lock);
3686         session = stream->trace->session;
3687         if (!conn->session) {
3688                 ret = connection_set_session(conn, session);
3689                 if (ret) {
3690                         status = RELAY_CONNECTION_STATUS_ERROR;
3691                         goto end_stream_unlock;
3692                 }
3693         }
3694
3695         /*
3696          * The size of the "chunk" received on any iteration is bounded by:
3697          *   - the data left to receive,
3698          *   - the data immediately available on the socket,
3699          *   - the on-stack data buffer
3700          */
3701         while (left_to_receive > 0 && !partial_recv) {
3702                 size_t recv_size = std::min<uint64_t>(left_to_receive, chunk_size);
3703                 struct lttng_buffer_view packet_chunk;
3704
3705                 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, MSG_DONTWAIT);
3706                 if (ret < 0) {
3707                         DIAGNOSTIC_PUSH
3708                         DIAGNOSTIC_IGNORE_LOGICAL_OP
3709                         if (errno != EAGAIN && errno != EWOULDBLOCK) {
3710                                 DIAGNOSTIC_POP
3711                                 PERROR("Socket %d error", conn->sock->fd);
3712                                 status = RELAY_CONNECTION_STATUS_ERROR;
3713                         }
3714                         goto end_stream_unlock;
3715                 } else if (ret == 0) {
3716                         /* No more data ready to be consumed on socket. */
3717                         DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3718                              state->header.stream_id);
3719                         status = RELAY_CONNECTION_STATUS_CLOSED;
3720                         break;
3721                 } else if (ret < (int) recv_size) {
3722                         /*
3723                          * All the data available on the socket has been
3724                          * consumed.
3725                          */
3726                         partial_recv = true;
3727                         recv_size = ret;
3728                 }
3729
3730                 packet_chunk = lttng_buffer_view_init(data_buffer, 0, recv_size);
3731                 LTTNG_ASSERT(packet_chunk.data);
3732
3733                 ret = stream_write(stream, &packet_chunk, 0);
3734                 if (ret) {
3735                         ERR("Relay error writing data to file");
3736                         status = RELAY_CONNECTION_STATUS_ERROR;
3737                         goto end_stream_unlock;
3738                 }
3739
3740                 left_to_receive -= recv_size;
3741                 state->received += recv_size;
3742                 state->left_to_receive = left_to_receive;
3743         }
3744
3745         if (state->left_to_receive > 0) {
3746                 /*
3747                  * Did not receive all the data expected, wait for more data to
3748                  * become available on the socket.
3749                  */
3750                 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64
3751                      " bytes received, %" PRIu64 " bytes left to receive",
3752                      state->header.stream_id,
3753                      state->received,
3754                      state->left_to_receive);
3755                 goto end_stream_unlock;
3756         }
3757
3758         ret = stream_write(stream, nullptr, state->header.padding_size);
3759         if (ret) {
3760                 status = RELAY_CONNECTION_STATUS_ERROR;
3761                 goto end_stream_unlock;
3762         }
3763
3764         if (session_streams_have_index(session)) {
3765                 ret = stream_update_index(stream,
3766                                           state->header.net_seq_num,
3767                                           state->rotate_index,
3768                                           &index_flushed,
3769                                           state->header.data_size + state->header.padding_size);
3770                 if (ret < 0) {
3771                         ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64
3772                             " ret %d",
3773                             stream->stream_handle,
3774                             state->header.net_seq_num,
3775                             ret);
3776                         status = RELAY_CONNECTION_STATUS_ERROR;
3777                         goto end_stream_unlock;
3778                 }
3779         }
3780
3781         if (stream->prev_data_seq == -1ULL) {
3782                 new_stream = true;
3783         }
3784
3785         ret = stream_complete_packet(stream,
3786                                      state->header.data_size + state->header.padding_size,
3787                                      state->header.net_seq_num,
3788                                      index_flushed);
3789         if (ret) {
3790                 status = RELAY_CONNECTION_STATUS_ERROR;
3791                 goto end_stream_unlock;
3792         }
3793
3794         /*
3795          * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3796          * contents of *state which are aliased (union) to the same location as
3797          * the new state. Don't use it beyond this point.
3798          */
3799         connection_reset_protocol_state(conn);
3800         state = nullptr;
3801
3802 end_stream_unlock:
3803         close_requested = stream->close_requested;
3804         pthread_mutex_unlock(&stream->lock);
3805         if (close_requested && left_to_receive == 0) {
3806                 try_stream_close(stream);
3807         }
3808
3809         if (new_stream) {
3810                 pthread_mutex_lock(&session->lock);
3811                 uatomic_set(&session->new_streams, 1);
3812                 pthread_mutex_unlock(&session->lock);
3813         }
3814
3815         stream_put(stream);
3816 end:
3817         return status;
3818 }
3819
3820 /*
3821  * relay_process_data: Process the data received on the data socket
3822  */
3823 static enum relay_connection_status relay_process_data(struct relay_connection *conn)
3824 {
3825         enum relay_connection_status status;
3826
3827         switch (conn->protocol.data.state_id) {
3828         case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3829                 status = relay_process_data_receive_header(conn);
3830                 break;
3831         case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3832                 status = relay_process_data_receive_payload(conn);
3833                 break;
3834         default:
3835                 ERR("Unexpected data connection communication state.");
3836                 abort();
3837         }
3838
3839         return status;
3840 }
3841
3842 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3843 {
3844         int ret;
3845
3846         (void) lttng_poll_del(events, pollfd);
3847
3848         ret = fd_tracker_close_unsuspendable_fd(
3849                 the_fd_tracker, &pollfd, 1, fd_tracker_util_close_fd, nullptr);
3850         if (ret < 0) {
3851                 ERR("Closing pollfd %d", pollfd);
3852         }
3853 }
3854
3855 static void relay_thread_close_connection(struct lttng_poll_event *events,
3856                                           int pollfd,
3857                                           struct relay_connection *conn)
3858 {
3859         const char *type_str;
3860
3861         switch (conn->type) {
3862         case RELAY_DATA:
3863                 type_str = "Data";
3864                 break;
3865         case RELAY_CONTROL:
3866                 type_str = "Control";
3867                 break;
3868         case RELAY_VIEWER_COMMAND:
3869                 type_str = "Viewer Command";
3870                 break;
3871         case RELAY_VIEWER_NOTIFICATION:
3872                 type_str = "Viewer Notification";
3873                 break;
3874         default:
3875                 type_str = "Unknown";
3876         }
3877         cleanup_connection_pollfd(events, pollfd);
3878         connection_put(conn);
3879         DBG("%s connection closed with %d", type_str, pollfd);
3880 }
3881
3882 /*
3883  * This thread does the actual work
3884  */
3885 static void *relay_thread_worker(void *data __attribute__((unused)))
3886 {
3887         int ret, err = -1, last_seen_data_fd = -1;
3888         uint32_t nb_fd;
3889         struct lttng_poll_event events;
3890         struct lttng_ht *relay_connections_ht;
3891         struct lttng_ht_iter iter;
3892         struct relay_connection *destroy_conn = nullptr;
3893
3894         DBG("[thread] Relay worker started");
3895
3896         rcu_register_thread();
3897
3898         health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3899
3900         if (testpoint(relayd_thread_worker)) {
3901                 goto error_testpoint;
3902         }
3903
3904         health_code_update();
3905
3906         /* table of connections indexed on socket */
3907         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3908         if (!relay_connections_ht) {
3909                 goto relay_connections_ht_error;
3910         }
3911
3912         ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3913         if (ret < 0) {
3914                 goto error_poll_create;
3915         }
3916
3917         ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3918         if (ret < 0) {
3919                 goto error;
3920         }
3921
3922 restart:
3923         while (true) {
3924                 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3925
3926                 health_code_update();
3927
3928                 /* Infinite blocking call, waiting for transmission */
3929                 DBG3("Relayd worker thread polling...");
3930                 health_poll_entry();
3931                 ret = lttng_poll_wait(&events, -1);
3932                 health_poll_exit();
3933                 if (ret < 0) {
3934                         /*
3935                          * Restart interrupted system call.
3936                          */
3937                         if (errno == EINTR) {
3938                                 goto restart;
3939                         }
3940                         goto error;
3941                 }
3942
3943                 nb_fd = ret;
3944
3945                 /*
3946                  * Process control. The control connection is
3947                  * prioritized so we don't starve it with high
3948                  * throughput tracing data on the data connection.
3949                  */
3950                 for (i = 0; i < nb_fd; i++) {
3951                         /* Fetch once the poll data */
3952                         const auto revents = LTTNG_POLL_GETEV(&events, i);
3953                         const auto pollfd = LTTNG_POLL_GETFD(&events, i);
3954
3955                         health_code_update();
3956
3957                         /* Activity on thread quit pipe, exiting. */
3958                         if (relayd_is_thread_quit_pipe(pollfd)) {
3959                                 DBG("Activity on thread quit pipe");
3960                                 err = 0;
3961                                 goto exit;
3962                         }
3963
3964                         /* Inspect the relay conn pipe for new connection */
3965                         if (pollfd == relay_conn_pipe[0]) {
3966                                 if (revents & LPOLLIN) {
3967                                         struct relay_connection *conn;
3968
3969                                         ret = lttng_read(relay_conn_pipe[0],
3970                                                          &conn,
3971                                                          sizeof(conn)); /* NOLINT sizeof used on a
3972                                                                            pointer. */
3973                                         if (ret < 0) {
3974                                                 goto error;
3975                                         }
3976                                         ret = lttng_poll_add(
3977                                                 &events, conn->sock->fd, LPOLLIN | LPOLLRDHUP);
3978                                         if (ret) {
3979                                                 ERR("Failed to add new connection file descriptor to poll set");
3980                                                 goto error;
3981                                         }
3982                                         connection_ht_add(relay_connections_ht, conn);
3983                                         DBG("Connection socket %d added", conn->sock->fd);
3984                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3985                                         ERR("Relay connection pipe error");
3986                                         goto error;
3987                                 } else {
3988                                         ERR("Unexpected poll events %u for sock %d",
3989                                             revents,
3990                                             pollfd);
3991                                         goto error;
3992                                 }
3993                         } else {
3994                                 struct relay_connection *ctrl_conn;
3995
3996                                 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
3997                                 /* If not found, there is a synchronization issue. */
3998                                 LTTNG_ASSERT(ctrl_conn);
3999
4000                                 if (ctrl_conn->type == RELAY_DATA) {
4001                                         if (revents & LPOLLIN) {
4002                                                 /*
4003                                                  * Flag the last seen data fd not deleted. It will
4004                                                  * be used as the last seen fd if any fd gets
4005                                                  * deleted in this first loop.
4006                                                  */
4007                                                 last_notdel_data_fd = pollfd;
4008                                         }
4009                                         goto put_ctrl_connection;
4010                                 }
4011                                 LTTNG_ASSERT(ctrl_conn->type == RELAY_CONTROL);
4012
4013                                 if (revents & LPOLLIN) {
4014                                         enum relay_connection_status status;
4015
4016                                         status = relay_process_control(ctrl_conn);
4017                                         if (status != RELAY_CONNECTION_STATUS_OK) {
4018                                                 /*
4019                                                  * On socket error flag the session as aborted to
4020                                                  * force the cleanup of its stream otherwise it can
4021                                                  * leak during the lifetime of the relayd.
4022                                                  *
4023                                                  * This prevents situations in which streams can be
4024                                                  * left opened because an index was received, the
4025                                                  * control connection is closed, and the data
4026                                                  * connection is closed (uncleanly) before the
4027                                                  * packet's data provided.
4028                                                  *
4029                                                  * Since the control connection encountered an
4030                                                  * error, it is okay to be conservative and close
4031                                                  * the session right now as we can't rely on the
4032                                                  * protocol being respected anymore.
4033                                                  */
4034                                                 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4035                                                         session_abort(ctrl_conn->session);
4036                                                 }
4037
4038                                                 /* Clear the connection on error or close. */
4039                                                 relay_thread_close_connection(
4040                                                         &events, pollfd, ctrl_conn);
4041                                         }
4042                                         seen_control = 1;
4043                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4044                                         relay_thread_close_connection(&events, pollfd, ctrl_conn);
4045                                         if (last_seen_data_fd == pollfd) {
4046                                                 last_seen_data_fd = last_notdel_data_fd;
4047                                         }
4048                                 } else {
4049                                         ERR("Unexpected poll events %u for control sock %d",
4050                                             revents,
4051                                             pollfd);
4052                                         connection_put(ctrl_conn);
4053                                         goto error;
4054                                 }
4055                         put_ctrl_connection:
4056                                 connection_put(ctrl_conn);
4057                         }
4058                 }
4059
4060                 /*
4061                  * The last loop handled a control request, go back to poll to make
4062                  * sure we prioritise the control socket.
4063                  */
4064                 if (seen_control) {
4065                         continue;
4066                 }
4067
4068                 if (last_seen_data_fd >= 0) {
4069                         for (i = 0; i < nb_fd; i++) {
4070                                 int pollfd = LTTNG_POLL_GETFD(&events, i);
4071
4072                                 health_code_update();
4073
4074                                 if (last_seen_data_fd == pollfd) {
4075                                         idx = i;
4076                                         break;
4077                                 }
4078                         }
4079                 }
4080
4081                 /* Process data connection. */
4082                 for (i = idx + 1; i < nb_fd; i++) {
4083                         /* Fetch the poll data. */
4084                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
4085                         int pollfd = LTTNG_POLL_GETFD(&events, i);
4086                         struct relay_connection *data_conn;
4087
4088                         health_code_update();
4089
4090                         if (!revents) {
4091                                 /* No activity for this FD (poll implementation). */
4092                                 continue;
4093                         }
4094
4095                         /* Skip the command pipe. It's handled in the first loop. */
4096                         if (pollfd == relay_conn_pipe[0]) {
4097                                 continue;
4098                         }
4099
4100                         data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4101                         if (!data_conn) {
4102                                 /* Skip it. Might be removed before. */
4103                                 continue;
4104                         }
4105                         if (data_conn->type == RELAY_CONTROL) {
4106                                 goto put_data_connection;
4107                         }
4108                         LTTNG_ASSERT(data_conn->type == RELAY_DATA);
4109
4110                         if (revents & LPOLLIN) {
4111                                 enum relay_connection_status status;
4112
4113                                 status = relay_process_data(data_conn);
4114                                 /* Connection closed or error. */
4115                                 if (status != RELAY_CONNECTION_STATUS_OK) {
4116                                         /*
4117                                          * On socket error flag the session as aborted to force
4118                                          * the cleanup of its stream otherwise it can leak
4119                                          * during the lifetime of the relayd.
4120                                          *
4121                                          * This prevents situations in which streams can be
4122                                          * left opened because an index was received, the
4123                                          * control connection is closed, and the data
4124                                          * connection is closed (uncleanly) before the packet's
4125                                          * data provided.
4126                                          *
4127                                          * Since the data connection encountered an error,
4128                                          * it is okay to be conservative and close the
4129                                          * session right now as we can't rely on the protocol
4130                                          * being respected anymore.
4131                                          */
4132                                         if (status == RELAY_CONNECTION_STATUS_ERROR) {
4133                                                 session_abort(data_conn->session);
4134                                         }
4135                                         relay_thread_close_connection(&events, pollfd, data_conn);
4136                                         /*
4137                                          * Every goto restart call sets the last seen fd where
4138                                          * here we don't really care since we gracefully
4139                                          * continue the loop after the connection is deleted.
4140                                          */
4141                                 } else {
4142                                         /* Keep last seen port. */
4143                                         last_seen_data_fd = pollfd;
4144                                         connection_put(data_conn);
4145                                         goto restart;
4146                                 }
4147                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4148                                 relay_thread_close_connection(&events, pollfd, data_conn);
4149                         } else {
4150                                 ERR("Unknown poll events %u for data sock %d", revents, pollfd);
4151                         }
4152                 put_data_connection:
4153                         connection_put(data_conn);
4154                 }
4155                 last_seen_data_fd = -1;
4156         }
4157
4158         /* Normal exit, no error */
4159         ret = 0;
4160
4161 exit:
4162 error:
4163         /* Cleanup remaining connection object. */
4164         rcu_read_lock();
4165         cds_lfht_for_each_entry (relay_connections_ht->ht, &iter.iter, destroy_conn, sock_n.node) {
4166                 health_code_update();
4167
4168                 session_abort(destroy_conn->session);
4169
4170                 /*
4171                  * No need to grab another ref, because we own
4172                  * destroy_conn.
4173                  */
4174                 relay_thread_close_connection(&events, destroy_conn->sock->fd, destroy_conn);
4175         }
4176         rcu_read_unlock();
4177
4178         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4179 error_poll_create:
4180         lttng_ht_destroy(relay_connections_ht);
4181 relay_connections_ht_error:
4182         /* Close relay conn pipes */
4183         (void) fd_tracker_util_pipe_close(the_fd_tracker, relay_conn_pipe);
4184         if (err) {
4185                 DBG("Thread exited with error");
4186         }
4187         DBG("Worker thread cleanup complete");
4188 error_testpoint:
4189         if (err) {
4190                 health_error();
4191                 ERR("Health error occurred in %s", __func__);
4192         }
4193         health_unregister(health_relayd);
4194         rcu_unregister_thread();
4195         lttng_relay_stop_threads();
4196         return nullptr;
4197 }
4198
4199 /*
4200  * Create the relay command pipe to wake thread_manage_apps.
4201  * Closed in cleanup().
4202  */
4203 static int create_relay_conn_pipe()
4204 {
4205         return fd_tracker_util_pipe_open_cloexec(
4206                 the_fd_tracker, "Relayd connection pipe", relay_conn_pipe);
4207 }
4208
4209 static int stdio_open(void *data __attribute__((unused)), int *fds)
4210 {
4211         fds[0] = fileno(stdout);
4212         fds[1] = fileno(stderr);
4213         return 0;
4214 }
4215
4216 static int track_stdio()
4217 {
4218         int fds[2];
4219         const char *names[] = { "stdout", "stderr" };
4220
4221         return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds, names, 2, stdio_open, nullptr);
4222 }
4223
4224 /*
4225  * main
4226  */
4227 int main(int argc, char **argv)
4228 {
4229         bool thread_is_rcu_registered = false;
4230         int ret = 0, retval = 0;
4231         void *status;
4232         char *unlinked_file_directory_path = nullptr, *output_path = nullptr;
4233
4234         /* Parse environment variables */
4235         ret = parse_env_options();
4236         if (ret) {
4237                 retval = -1;
4238                 goto exit_options;
4239         }
4240
4241         /*
4242          * Parse arguments.
4243          * Command line arguments overwrite environment.
4244          */
4245         progname = argv[0];
4246         if (set_options(argc, argv)) {
4247                 retval = -1;
4248                 goto exit_options;
4249         }
4250
4251         if (set_signal_handler()) {
4252                 retval = -1;
4253                 goto exit_options;
4254         }
4255
4256         relayd_config_log();
4257
4258         if (opt_print_version) {
4259                 print_version();
4260                 retval = 0;
4261                 goto exit_options;
4262         }
4263
4264         ret = fclose(stdin);
4265         if (ret) {
4266                 PERROR("Failed to close stdin");
4267                 goto exit_options;
4268         }
4269
4270         DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4271
4272         /* Try to create directory if -o, --output is specified. */
4273         if (opt_output_path) {
4274                 if (*opt_output_path != '/') {
4275                         ERR("Please specify an absolute path for -o, --output PATH");
4276                         retval = -1;
4277                         goto exit_options;
4278                 }
4279
4280                 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG, -1, -1);
4281                 if (ret < 0) {
4282                         ERR("Unable to create %s", opt_output_path);
4283                         retval = -1;
4284                         goto exit_options;
4285                 }
4286         }
4287
4288         /* Daemonize */
4289         if (opt_daemon || opt_background) {
4290                 ret = lttng_daemonize(&child_ppid, &recv_child_signal, !opt_background);
4291                 if (ret < 0) {
4292                         retval = -1;
4293                         goto exit_options;
4294                 }
4295         }
4296
4297         if (opt_working_directory) {
4298                 ret = utils_change_working_directory(opt_working_directory);
4299                 if (ret) {
4300                         /* All errors are already logged. */
4301                         goto exit_options;
4302                 }
4303         }
4304
4305         sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4306         if (!sessiond_trace_chunk_registry) {
4307                 ERR("Failed to initialize session daemon trace chunk registry");
4308                 retval = -1;
4309                 goto exit_options;
4310         }
4311
4312         /*
4313          * The RCU thread registration (and use, through the fd-tracker's
4314          * creation) is done after the daemonization to allow us to not
4315          * deal with liburcu's fork() management as the call RCU needs to
4316          * be restored.
4317          */
4318         rcu_register_thread();
4319         thread_is_rcu_registered = true;
4320
4321         output_path = create_output_path("");
4322         if (!output_path) {
4323                 ERR("Failed to get output path");
4324                 retval = -1;
4325                 goto exit_options;
4326         }
4327         ret = asprintf(&unlinked_file_directory_path,
4328                        "%s/%s",
4329                        output_path,
4330                        DEFAULT_UNLINKED_FILES_DIRECTORY);
4331         free(output_path);
4332         if (ret < 0) {
4333                 ERR("Failed to format unlinked file directory path");
4334                 retval = -1;
4335                 goto exit_options;
4336         }
4337         the_fd_tracker = fd_tracker_create(unlinked_file_directory_path, lttng_opt_fd_pool_size);
4338         free(unlinked_file_directory_path);
4339         if (!the_fd_tracker) {
4340                 retval = -1;
4341                 goto exit_options;
4342         }
4343
4344         ret = track_stdio();
4345         if (ret) {
4346                 retval = -1;
4347                 goto exit_options;
4348         }
4349
4350         /* Initialize thread health monitoring */
4351         health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4352         if (!health_relayd) {
4353                 PERROR("health_app_create error");
4354                 retval = -1;
4355                 goto exit_options;
4356         }
4357
4358         /* Create thread quit pipe */
4359         if (relayd_init_thread_quit_pipe()) {
4360                 retval = -1;
4361                 goto exit_options;
4362         }
4363
4364         /* Setup the thread apps communication pipe. */
4365         if (create_relay_conn_pipe()) {
4366                 retval = -1;
4367                 goto exit_options;
4368         }
4369
4370         /* Init relay command queue. */
4371         cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4372
4373         /* Initialize communication library */
4374         lttcomm_init();
4375         lttcomm_inet_init();
4376
4377         /* tables of sessions indexed by session ID */
4378         sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4379         if (!sessions_ht) {
4380                 retval = -1;
4381                 goto exit_options;
4382         }
4383
4384         /* tables of streams indexed by stream ID */
4385         relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4386         if (!relay_streams_ht) {
4387                 retval = -1;
4388                 goto exit_options;
4389         }
4390
4391         /* tables of streams indexed by stream ID */
4392         viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4393         if (!viewer_streams_ht) {
4394                 retval = -1;
4395                 goto exit_options;
4396         }
4397
4398         ret = init_health_quit_pipe();
4399         if (ret) {
4400                 retval = -1;
4401                 goto exit_options;
4402         }
4403
4404         /* Create thread to manage the client socket */
4405         ret = pthread_create(&health_thread,
4406                              default_pthread_attr(),
4407                              thread_manage_health_relayd,
4408                              (void *) nullptr);
4409         if (ret) {
4410                 errno = ret;
4411                 PERROR("pthread_create health");
4412                 retval = -1;
4413                 goto exit_options;
4414         }
4415
4416         /* Setup the dispatcher thread */
4417         ret = pthread_create(&dispatcher_thread,
4418                              default_pthread_attr(),
4419                              relay_thread_dispatcher,
4420                              (void *) nullptr);
4421         if (ret) {
4422                 errno = ret;
4423                 PERROR("pthread_create dispatcher");
4424                 retval = -1;
4425                 goto exit_dispatcher_thread;
4426         }
4427
4428         /* Setup the worker thread */
4429         ret = pthread_create(&worker_thread, default_pthread_attr(), relay_thread_worker, nullptr);
4430         if (ret) {
4431                 errno = ret;
4432                 PERROR("pthread_create worker");
4433                 retval = -1;
4434                 goto exit_worker_thread;
4435         }
4436
4437         /* Setup the listener thread */
4438         ret = pthread_create(
4439                 &listener_thread, default_pthread_attr(), relay_thread_listener, (void *) nullptr);
4440         if (ret) {
4441                 errno = ret;
4442                 PERROR("pthread_create listener");
4443                 retval = -1;
4444                 goto exit_listener_thread;
4445         }
4446
4447         ret = relayd_live_create(live_uri);
4448         if (ret) {
4449                 ERR("Starting live viewer threads");
4450                 retval = -1;
4451                 goto exit_live;
4452         }
4453
4454         /*
4455          * This is where we start awaiting program completion (e.g. through
4456          * signal that asks threads to teardown).
4457          */
4458
4459         ret = relayd_live_join();
4460         if (ret) {
4461                 retval = -1;
4462         }
4463 exit_live:
4464
4465         ret = pthread_join(listener_thread, &status);
4466         if (ret) {
4467                 errno = ret;
4468                 PERROR("pthread_join listener_thread");
4469                 retval = -1;
4470         }
4471
4472 exit_listener_thread:
4473         ret = pthread_join(worker_thread, &status);
4474         if (ret) {
4475                 errno = ret;
4476                 PERROR("pthread_join worker_thread");
4477                 retval = -1;
4478         }
4479
4480 exit_worker_thread:
4481         ret = pthread_join(dispatcher_thread, &status);
4482         if (ret) {
4483                 errno = ret;
4484                 PERROR("pthread_join dispatcher_thread");
4485                 retval = -1;
4486         }
4487 exit_dispatcher_thread:
4488
4489         ret = pthread_join(health_thread, &status);
4490         if (ret) {
4491                 errno = ret;
4492                 PERROR("pthread_join health_thread");
4493                 retval = -1;
4494         }
4495 exit_options:
4496         /*
4497          * Wait for all pending call_rcu work to complete before tearing
4498          * down data structures. call_rcu worker may be trying to
4499          * perform lookups in those structures.
4500          */
4501         rcu_barrier();
4502         relayd_cleanup();
4503
4504         /* Ensure all prior call_rcu are done. */
4505         rcu_barrier();
4506
4507         if (thread_is_rcu_registered) {
4508                 rcu_unregister_thread();
4509         }
4510
4511         if (!retval) {
4512                 exit(EXIT_SUCCESS);
4513         } else {
4514                 exit(EXIT_FAILURE);
4515         }
4516 }