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/urcu.hpp>
  50 #include <common/uri.hpp>
  51 #include <common/utils.hpp>
  52
  53 #include <lttng/lttng.h>
  54
  55 #include <algorithm>
  56 #include <ctype.h>
  57 #include <fcntl.h>
  58 #include <getopt.h>
  59 #include <grp.h>
  60 #include <inttypes.h>
  61 #include <limits.h>
  62 #include <pthread.h>
  63 #include <signal.h>
  64 #include <stdio.h>
  65 #include <stdlib.h>
  66 #include <string.h>
  67 #include <strings.h>
  68 #include <sys/mman.h>
  69 #include <sys/mount.h>
  70 #include <sys/resource.h>
  71 #include <sys/socket.h>
  72 #include <sys/stat.h>
  73 #include <sys/types.h>
  74 #include <sys/wait.h>
  75 #include <unistd.h>
  76 #include <urcu/futex.h>
  77 #include <urcu/rculist.h>
  78 #include <urcu/uatomic.h>
  79
  80 static const char *help_msg =
  81 #ifdef LTTNG_EMBED_HELP
  82 #include <lttng-relayd.8.h>
  83 #else
  84         nullptr
  85 #endif
  86         ;
  87
  88 enum relay_connection_status {
  89         RELAY_CONNECTION_STATUS_OK,
  90         /* An error occurred while processing an event on the connection. */
  91         RELAY_CONNECTION_STATUS_ERROR,
  92         /* Connection closed/shutdown cleanly. */
  93         RELAY_CONNECTION_STATUS_CLOSED,
  94 };
  95
  96 /* command line options */
  97 char *opt_output_path, *opt_working_directory;
  98 static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
  99 enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
 100
 101 /* Argument variables */
 102 int lttng_opt_quiet; /* not static in error.h */
 103 int lttng_opt_verbose; /* not static in error.h */
 104 int lttng_opt_mi; /* not static in error.h */
 105
 106 /*
 107  * We need to wait for listener and live listener threads, as well as
 108  * health check thread, before being ready to signal readiness.
 109  */
 110 #define NR_LTTNG_RELAY_READY 3
 111 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
 112
 113 /* Size of receive buffer. */
 114 #define RECV_DATA_BUFFER_SIZE 65536
 115
 116 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
 117 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
 118
 119 static struct lttng_uri *control_uri;
 120 static struct lttng_uri *data_uri;
 121 static struct lttng_uri *live_uri;
 122
 123 const char *progname;
 124
 125 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
 126 static int tracing_group_name_override;
 127
 128 const char *const config_section_name = "relayd";
 129
 130 /*
 131  * This pipe is used to inform the worker thread that a command is queued and
 132  * ready to be processed.
 133  */
 134 static int relay_conn_pipe[2] = { -1, -1 };
 135
 136 /* Shared between threads */
 137 static int dispatch_thread_exit;
 138
 139 static pthread_t listener_thread;
 140 static pthread_t dispatcher_thread;
 141 static pthread_t worker_thread;
 142 static pthread_t health_thread;
 143
 144 /*
 145  * last_relay_stream_id_lock protects last_relay_stream_id increment
 146  * atomicity on 32-bit architectures.
 147  */
 148 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
 149 static uint64_t last_relay_stream_id;
 150
 151 /*
 152  * Relay command queue.
 153  *
 154  * The relay_thread_listener and relay_thread_dispatcher communicate with this
 155  * queue.
 156  */
 157 static struct relay_conn_queue relay_conn_queue;
 158
 159 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
 160 static unsigned int lttng_opt_fd_pool_size = -1;
 161
 162 /* Global relay stream hash table. */
 163 struct lttng_ht *relay_streams_ht;
 164
 165 /* Global relay viewer stream hash table. */
 166 struct lttng_ht *viewer_streams_ht;
 167
 168 /* Global relay sessions hash table. */
 169 struct lttng_ht *sessions_ht;
 170
 171 /* Relayd health monitoring */
 172 struct health_app *health_relayd;
 173
 174 struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
 175
 176 /* Global fd tracker. */
 177 struct fd_tracker *the_fd_tracker;
 178
 179 static struct option long_options[] = {
 180         {
 181                 "control-port",
 182                 1,
 183                 nullptr,
 184                 'C',
 185         },
 186         {
 187                 "data-port",
 188                 1,
 189                 nullptr,
 190                 'D',
 191         },
 192         {
 193                 "live-port",
 194                 1,
 195                 nullptr,
 196                 'L',
 197         },
 198         {
 199                 "daemonize",
 200                 0,
 201                 nullptr,
 202                 'd',
 203         },
 204         {
 205                 "background",
 206                 0,
 207                 nullptr,
 208                 'b',
 209         },
 210         {
 211                 "group",
 212                 1,
 213                 nullptr,
 214                 'g',
 215         },
 216         {
 217                 "fd-pool-size",
 218                 1,
 219                 nullptr,
 220                 '\0',
 221         },
 222         {
 223                 "help",
 224                 0,
 225                 nullptr,
 226                 'h',
 227         },
 228         {
 229                 "output",
 230                 1,
 231                 nullptr,
 232                 'o',
 233         },
 234         {
 235                 "verbose",
 236                 0,
 237                 nullptr,
 238                 'v',
 239         },
 240         { "config", 1, nullptr, 'f' },
 241         { "version", 0, nullptr, 'V' },
 242         {
 243                 "working-directory",
 244                 1,
 245                 nullptr,
 246                 'w',
 247         },
 248         {
 249                 "group-output-by-session",
 250                 0,
 251                 nullptr,
 252                 's',
 253         },
 254         {
 255                 "group-output-by-host",
 256                 0,
 257                 nullptr,
 258                 'p',
 259         },
 260         { "disallow-clear", 0, nullptr, 'x' },
 261         {
 262                 nullptr,
 263                 0,
 264                 nullptr,
 265                 0,
 266         },
 267 };
 268
 269 static const char *config_ignore_options[] = { "help", "config", "version" };
 270
 271 static void print_version()
 272 {
 273         fprintf(stdout, "%s\n", VERSION);
 274 }
 275
 276 static void relayd_config_log()
 277 {
 278         DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
 279             GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
 280             EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
 281         if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
 282                 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
 283         }
 284         if (EXTRA_VERSION_PATCHES[0] != '\0') {
 285                 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
 286         }
 287 }
 288
 289 /*
 290  * Take an option from the getopt output and set it in the right variable to be
 291  * used later.
 292  *
 293  * Return 0 on success else a negative value.
 294  */
 295 static int set_option(int opt, const char *arg, const char *optname)
 296 {
 297         int ret;
 298
 299         switch (opt) {
 300         case 0:
 301                 if (!strcmp(optname, "fd-pool-size")) {
 302                         unsigned long v;
 303
 304                         errno = 0;
 305                         v = strtoul(arg, nullptr, 0);
 306                         if (errno != 0 || !isdigit((unsigned char) arg[0])) {
 307                                 ERR("Wrong value in --fd-pool-size parameter: %s", arg);
 308                                 ret = -1;
 309                                 goto end;
 310                         }
 311                         if (v >= UINT_MAX) {
 312                                 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s",
 313                                     arg);
 314                                 ret = -1;
 315                                 goto end;
 316                         }
 317                         lttng_opt_fd_pool_size = (unsigned int) v;
 318                 } else {
 319                         fprintf(stderr, "unknown option %s", optname);
 320                         if (arg) {
 321                                 fprintf(stderr, " with arg %s\n", arg);
 322                         }
 323                 }
 324                 break;
 325         case 'C':
 326                 if (lttng_is_setuid_setgid()) {
 327                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 328                              "-C, --control-port");
 329                 } else {
 330                         ret = uri_parse(arg, &control_uri);
 331                         if (ret < 0) {
 332                                 ERR("Invalid control URI specified");
 333                                 goto end;
 334                         }
 335                         if (control_uri->port == 0) {
 336                                 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
 337                         }
 338                 }
 339                 break;
 340         case 'D':
 341                 if (lttng_is_setuid_setgid()) {
 342                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 343                              "-D, -data-port");
 344                 } else {
 345                         ret = uri_parse(arg, &data_uri);
 346                         if (ret < 0) {
 347                                 ERR("Invalid data URI specified");
 348                                 goto end;
 349                         }
 350                         if (data_uri->port == 0) {
 351                                 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
 352                         }
 353                 }
 354                 break;
 355         case 'L':
 356                 if (lttng_is_setuid_setgid()) {
 357                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 358                              "-L, -live-port");
 359                 } else {
 360                         ret = uri_parse(arg, &live_uri);
 361                         if (ret < 0) {
 362                                 ERR("Invalid live URI specified");
 363                                 goto end;
 364                         }
 365                         if (live_uri->port == 0) {
 366                                 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
 367                         }
 368                 }
 369                 break;
 370         case 'd':
 371                 opt_daemon = 1;
 372                 break;
 373         case 'b':
 374                 opt_background = 1;
 375                 break;
 376         case 'g':
 377                 if (lttng_is_setuid_setgid()) {
 378                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 379                              "-g, --group");
 380                 } else {
 381                         tracing_group_name = strdup(arg);
 382                         if (tracing_group_name == nullptr) {
 383                                 ret = -errno;
 384                                 PERROR("strdup");
 385                                 goto end;
 386                         }
 387                         tracing_group_name_override = 1;
 388                 }
 389                 break;
 390         case 'h':
 391                 ret = utils_show_help(8, "lttng-relayd", help_msg);
 392                 if (ret) {
 393                         ERR("Cannot show --help for `lttng-relayd`");
 394                         perror("exec");
 395                 }
 396                 exit(EXIT_FAILURE);
 397         case 'V':
 398                 opt_print_version = 1;
 399                 break;
 400         case 'o':
 401                 if (lttng_is_setuid_setgid()) {
 402                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 403                              "-o, --output");
 404                 } else {
 405                         ret = asprintf(&opt_output_path, "%s", arg);
 406                         if (ret < 0) {
 407                                 ret = -errno;
 408                                 PERROR("asprintf opt_output_path");
 409                                 goto end;
 410                         }
 411                 }
 412                 break;
 413         case 'w':
 414                 if (lttng_is_setuid_setgid()) {
 415                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 416                              "-w, --working-directory");
 417                 } else {
 418                         ret = asprintf(&opt_working_directory, "%s", arg);
 419                         if (ret < 0) {
 420                                 ret = -errno;
 421                                 PERROR("asprintf opt_working_directory");
 422                                 goto end;
 423                         }
 424                 }
 425                 break;
 426
 427         case 'v':
 428                 /* Verbose level can increase using multiple -v */
 429                 if (arg) {
 430                         lttng_opt_verbose = config_parse_value(arg);
 431                 } else {
 432                         /* Only 3 level of verbosity (-vvv). */
 433                         if (lttng_opt_verbose < 3) {
 434                                 lttng_opt_verbose += 1;
 435                         }
 436                 }
 437                 break;
 438         case 's':
 439                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 440                         ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
 441                         exit(EXIT_FAILURE);
 442                 }
 443                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
 444                 break;
 445         case 'p':
 446                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 447                         ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
 448                         exit(EXIT_FAILURE);
 449                 }
 450                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 451                 break;
 452         case 'x':
 453                 /* Disallow clear */
 454                 opt_allow_clear = 0;
 455                 break;
 456         default:
 457                 /* Unknown option or other error.
 458                  * Error is printed by getopt, just return */
 459                 ret = -1;
 460                 goto end;
 461         }
 462
 463         /* All good. */
 464         ret = 0;
 465
 466 end:
 467         return ret;
 468 }
 469
 470 /*
 471  * config_entry_handler_cb used to handle options read from a config file.
 472  * See config_entry_handler_cb comment in common/config/session-config.h for the
 473  * return value conventions.
 474  */
 475 static int config_entry_handler(const struct config_entry *entry,
 476                                 void *unused __attribute__((unused)))
 477 {
 478         int ret = 0, i;
 479
 480         if (!entry || !entry->name || !entry->value) {
 481                 ret = -EINVAL;
 482                 goto end;
 483         }
 484
 485         /* Check if the option is to be ignored */
 486         for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
 487                 if (!strcmp(entry->name, config_ignore_options[i])) {
 488                         goto end;
 489                 }
 490         }
 491
 492         for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
 493                 /* Ignore if entry name is not fully matched. */
 494                 if (strcmp(entry->name, long_options[i].name) != 0) {
 495                         continue;
 496                 }
 497
 498                 /*
 499                  * If the option takes no argument on the command line,
 500                  * we have to check if the value is "true". We support
 501                  * non-zero numeric values, true, on and yes.
 502                  */
 503                 if (!long_options[i].has_arg) {
 504                         ret = config_parse_value(entry->value);
 505                         if (ret <= 0) {
 506                                 if (ret) {
 507                                         WARN("Invalid configuration value \"%s\" for option %s",
 508                                              entry->value,
 509                                              entry->name);
 510                                 }
 511                                 /* False, skip boolean config option. */
 512                                 goto end;
 513                         }
 514                 }
 515
 516                 ret = set_option(long_options[i].val, entry->value, entry->name);
 517                 goto end;
 518         }
 519
 520         WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
 521
 522 end:
 523         return ret;
 524 }
 525
 526 static int parse_env_options()
 527 {
 528         int ret = 0;
 529         char *value = nullptr;
 530
 531         value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
 532         if (value) {
 533                 opt_working_directory = strdup(value);
 534                 if (!opt_working_directory) {
 535                         ERR("Failed to allocate working directory string (\"%s\")", value);
 536                         ret = -1;
 537                 }
 538         }
 539         return ret;
 540 }
 541
 542 static int set_fd_pool_size()
 543 {
 544         int ret = 0;
 545         struct rlimit rlimit;
 546
 547         ret = getrlimit(RLIMIT_NOFILE, &rlimit);
 548         if (ret) {
 549                 PERROR("Failed to get file descriptor limit");
 550                 ret = -1;
 551                 goto end;
 552         }
 553
 554         DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
 555             (uint64_t) rlimit.rlim_cur,
 556             (uint64_t) rlimit.rlim_max);
 557         if (lttng_opt_fd_pool_size == -1) {
 558                 /* Use default value (soft limit - reserve). */
 559                 if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 560                         ERR("The process' file number limit is too low (%" PRIu64
 561                             "). The process' file number limit must be set to at least %i.",
 562                             (uint64_t) rlimit.rlim_cur,
 563                             DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 564                         ret = -1;
 565                         goto end;
 566                 }
 567                 lttng_opt_fd_pool_size = rlimit.rlim_cur - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 568                 goto end;
 569         }
 570
 571         if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 572                 ERR("File descriptor pool size must be set to at least %d",
 573                     DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 574                 ret = -1;
 575                 goto end;
 576         }
 577
 578         if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
 579                 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
 580                     ").",
 581                     lttng_opt_fd_pool_size,
 582                     (uint64_t) rlimit.rlim_cur);
 583                 ret = -1;
 584                 goto end;
 585         }
 586
 587         DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
 588             lttng_opt_fd_pool_size,
 589             lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
 590         lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 591 end:
 592         return ret;
 593 }
 594
 595 static int set_options(int argc, char **argv)
 596 {
 597         int c, ret = 0, option_index = 0, retval = 0;
 598         int orig_optopt = optopt, orig_optind = optind;
 599         char *default_address, *optstring;
 600         char *config_path = nullptr;
 601
 602         optstring = utils_generate_optstring(long_options,
 603                                              sizeof(long_options) / sizeof(struct option));
 604         if (!optstring) {
 605                 retval = -ENOMEM;
 606                 goto exit;
 607         }
 608
 609         /* Check for the --config option */
 610
 611         while ((c = getopt_long(argc, argv, optstring, long_options, &option_index)) != -1) {
 612                 if (c == '?') {
 613                         retval = -EINVAL;
 614                         goto exit;
 615                 } else if (c != 'f') {
 616                         continue;
 617                 }
 618
 619                 if (lttng_is_setuid_setgid()) {
 620                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 621                              "-f, --config");
 622                 } else {
 623                         free(config_path);
 624                         config_path = utils_expand_path(optarg);
 625                         if (!config_path) {
 626                                 ERR("Failed to resolve path: %s", optarg);
 627                         }
 628                 }
 629         }
 630
 631         ret = config_get_section_entries(
 632                 config_path, config_section_name, config_entry_handler, nullptr);
 633         if (ret) {
 634                 if (ret > 0) {
 635                         ERR("Invalid configuration option at line %i", ret);
 636                 }
 637                 retval = -1;
 638                 goto exit;
 639         }
 640
 641         /* Reset getopt's global state */
 642         optopt = orig_optopt;
 643         optind = orig_optind;
 644         while (true) {
 645                 c = getopt_long(argc, argv, optstring, long_options, &option_index);
 646                 if (c == -1) {
 647                         break;
 648                 }
 649
 650                 ret = set_option(c, optarg, long_options[option_index].name);
 651                 if (ret < 0) {
 652                         retval = -1;
 653                         goto exit;
 654                 }
 655         }
 656
 657         /* assign default values */
 658         if (control_uri == nullptr) {
 659                 ret = asprintf(&default_address,
 660                                "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
 661                                DEFAULT_NETWORK_CONTROL_PORT);
 662                 if (ret < 0) {
 663                         PERROR("asprintf default data address");
 664                         retval = -1;
 665                         goto exit;
 666                 }
 667
 668                 ret = uri_parse(default_address, &control_uri);
 669                 free(default_address);
 670                 if (ret < 0) {
 671                         ERR("Invalid control URI specified");
 672                         retval = -1;
 673                         goto exit;
 674                 }
 675         }
 676         if (data_uri == nullptr) {
 677                 ret = asprintf(&default_address,
 678                                "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
 679                                DEFAULT_NETWORK_DATA_PORT);
 680                 if (ret < 0) {
 681                         PERROR("asprintf default data address");
 682                         retval = -1;
 683                         goto exit;
 684                 }
 685
 686                 ret = uri_parse(default_address, &data_uri);
 687                 free(default_address);
 688                 if (ret < 0) {
 689                         ERR("Invalid data URI specified");
 690                         retval = -1;
 691                         goto exit;
 692                 }
 693         }
 694         if (live_uri == nullptr) {
 695                 ret = asprintf(&default_address,
 696                                "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
 697                                DEFAULT_NETWORK_VIEWER_PORT);
 698                 if (ret < 0) {
 699                         PERROR("asprintf default viewer control address");
 700                         retval = -1;
 701                         goto exit;
 702                 }
 703
 704                 ret = uri_parse(default_address, &live_uri);
 705                 free(default_address);
 706                 if (ret < 0) {
 707                         ERR("Invalid viewer control URI specified");
 708                         retval = -1;
 709                         goto exit;
 710                 }
 711         }
 712         ret = set_fd_pool_size();
 713         if (ret) {
 714                 retval = -1;
 715                 goto exit;
 716         }
 717
 718         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 719                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 720         }
 721         if (opt_allow_clear) {
 722                 /* Check if env variable exists. */
 723                 const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 724                 if (value) {
 725                         ret = config_parse_value(value);
 726                         if (ret < 0) {
 727                                 ERR("Invalid value for %s specified",
 728                                     DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 729                                 retval = -1;
 730                                 goto exit;
 731                         }
 732                         opt_allow_clear = !ret;
 733                 }
 734         }
 735
 736 exit:
 737         free(config_path);
 738         free(optstring);
 739         return retval;
 740 }
 741
 742 static void print_global_objects()
 743 {
 744         print_viewer_streams();
 745         print_relay_streams();
 746         print_sessions();
 747 }
 748
 749 static int noop_close(void *data __attribute__((unused)), int *fds __attribute__((unused)))
 750 {
 751         return 0;
 752 }
 753
 754 static void untrack_stdio()
 755 {
 756         int fds[] = { fileno(stdout), fileno(stderr) };
 757
 758         /*
 759          * noop_close is used since we don't really want to close
 760          * the stdio output fds; we merely want to stop tracking them.
 761          */
 762         (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker, fds, 2, noop_close, nullptr);
 763 }
 764
 765 /*
 766  * Cleanup the daemon
 767  */
 768 static void relayd_cleanup()
 769 {
 770         print_global_objects();
 771
 772         DBG("Cleaning up");
 773
 774         if (viewer_streams_ht)
 775                 lttng_ht_destroy(viewer_streams_ht);
 776         if (relay_streams_ht)
 777                 lttng_ht_destroy(relay_streams_ht);
 778         if (sessions_ht)
 779                 lttng_ht_destroy(sessions_ht);
 780
 781         free(opt_output_path);
 782         free(opt_working_directory);
 783
 784         if (health_relayd) {
 785                 health_app_destroy(health_relayd);
 786         }
 787         /* Close thread quit pipes */
 788         if (health_quit_pipe[0] != -1) {
 789                 (void) fd_tracker_util_pipe_close(the_fd_tracker, health_quit_pipe);
 790         }
 791         relayd_close_thread_quit_pipe();
 792         if (sessiond_trace_chunk_registry) {
 793                 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry);
 794         }
 795         if (the_fd_tracker) {
 796                 untrack_stdio();
 797                 /*
 798                  * fd_tracker_destroy() will log the contents of the fd-tracker
 799                  * if a leak is detected.
 800                  */
 801                 fd_tracker_destroy(the_fd_tracker);
 802         }
 803
 804         uri_free(control_uri);
 805         uri_free(data_uri);
 806         /* Live URI is freed in the live thread. */
 807
 808         if (tracing_group_name_override) {
 809                 free((void *) tracing_group_name);
 810         }
 811 }
 812
 813 static int notify_health_quit_pipe(int *pipe)
 814 {
 815         ssize_t ret;
 816
 817         ret = lttng_write(pipe[1], "4", 1);
 818         if (ret < 1) {
 819                 PERROR("write relay health quit");
 820                 goto end;
 821         }
 822         ret = 0;
 823 end:
 824         return ret;
 825 }
 826
 827 /*
 828  * Stop all relayd and relayd-live threads.
 829  */
 830 int lttng_relay_stop_threads()
 831 {
 832         int retval = 0;
 833
 834         /* Stopping all threads */
 835         DBG("Terminating all threads");
 836         if (relayd_notify_thread_quit_pipe()) {
 837                 ERR("write error on thread quit pipe");
 838                 retval = -1;
 839         }
 840
 841         if (notify_health_quit_pipe(health_quit_pipe)) {
 842                 ERR("write error on health quit pipe");
 843         }
 844
 845         /* Dispatch thread */
 846         CMM_STORE_SHARED(dispatch_thread_exit, 1);
 847         futex_nto1_wake(&relay_conn_queue.futex);
 848
 849         if (relayd_live_stop()) {
 850                 ERR("Error stopping live threads");
 851                 retval = -1;
 852         }
 853         return retval;
 854 }
 855
 856 /*
 857  * Signal handler for the daemon
 858  *
 859  * Simply stop all worker threads, leaving main() return gracefully after
 860  * joining all threads and calling cleanup().
 861  */
 862 static void sighandler(int sig)
 863 {
 864         switch (sig) {
 865         case SIGINT:
 866                 DBG("SIGINT caught");
 867                 if (lttng_relay_stop_threads()) {
 868                         ERR("Error stopping threads");
 869                 }
 870                 break;
 871         case SIGTERM:
 872                 DBG("SIGTERM caught");
 873                 if (lttng_relay_stop_threads()) {
 874                         ERR("Error stopping threads");
 875                 }
 876                 break;
 877         case SIGUSR1:
 878                 CMM_STORE_SHARED(recv_child_signal, 1);
 879                 break;
 880         default:
 881                 break;
 882         }
 883 }
 884
 885 /*
 886  * Setup signal handler for :
 887  *              SIGINT, SIGTERM, SIGPIPE
 888  */
 889 static int set_signal_handler()
 890 {
 891         int ret = 0;
 892         struct sigaction sa;
 893         sigset_t sigset;
 894
 895         if ((ret = sigemptyset(&sigset)) < 0) {
 896                 PERROR("sigemptyset");
 897                 return ret;
 898         }
 899
 900         sa.sa_mask = sigset;
 901         sa.sa_flags = 0;
 902
 903         sa.sa_handler = sighandler;
 904         if ((ret = sigaction(SIGTERM, &sa, nullptr)) < 0) {
 905                 PERROR("sigaction");
 906                 return ret;
 907         }
 908
 909         if ((ret = sigaction(SIGINT, &sa, nullptr)) < 0) {
 910                 PERROR("sigaction");
 911                 return ret;
 912         }
 913
 914         if ((ret = sigaction(SIGUSR1, &sa, nullptr)) < 0) {
 915                 PERROR("sigaction");
 916                 return ret;
 917         }
 918
 919         sa.sa_handler = SIG_IGN;
 920         if ((ret = sigaction(SIGPIPE, &sa, nullptr)) < 0) {
 921                 PERROR("sigaction");
 922                 return ret;
 923         }
 924
 925         DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
 926
 927         return ret;
 928 }
 929
 930 void lttng_relay_notify_ready()
 931 {
 932         /* Notify the parent of the fork() process that we are ready. */
 933         if (opt_daemon || opt_background) {
 934                 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
 935                         kill(child_ppid, SIGUSR1);
 936                 }
 937         }
 938 }
 939
 940 /*
 941  * Init health quit pipe.
 942  *
 943  * Return -1 on error or 0 if all pipes are created.
 944  */
 945 static int init_health_quit_pipe()
 946 {
 947         return fd_tracker_util_pipe_open_cloexec(
 948                 the_fd_tracker, "Health quit pipe", health_quit_pipe);
 949 }
 950
 951 static int create_sock(void *data, int *out_fd)
 952 {
 953         int ret;
 954         struct lttcomm_sock *sock = (lttcomm_sock *) data;
 955
 956         ret = lttcomm_create_sock(sock);
 957         if (ret < 0) {
 958                 goto end;
 959         }
 960
 961         *out_fd = sock->fd;
 962 end:
 963         return ret;
 964 }
 965
 966 static int close_sock(void *data, int *in_fd __attribute__((unused)))
 967 {
 968         struct lttcomm_sock *sock = (lttcomm_sock *) data;
 969
 970         return sock->ops->close(sock);
 971 }
 972
 973 static int accept_sock(void *data, int *out_fd)
 974 {
 975         int ret = 0;
 976         /* Socks is an array of in_sock, out_sock. */
 977         struct lttcomm_sock **socks = (lttcomm_sock **) data;
 978         struct lttcomm_sock *in_sock = socks[0];
 979
 980         socks[1] = in_sock->ops->accept(in_sock);
 981         if (!socks[1]) {
 982                 ret = -1;
 983                 goto end;
 984         }
 985         *out_fd = socks[1]->fd;
 986 end:
 987         return ret;
 988 }
 989
 990 /*
 991  * Create and init socket from uri.
 992  */
 993 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri, const char *name)
 994 {
 995         int ret, sock_fd;
 996         struct lttcomm_sock *sock = nullptr;
 997         char uri_str[PATH_MAX];
 998         char *formated_name = nullptr;
 999
1000         sock = lttcomm_alloc_sock_from_uri(uri);
1001         if (sock == nullptr) {
1002                 ERR("Allocating socket");
1003                 goto error;
1004         }
1005
1006         /*
1007          * Don't fail to create the socket if the name can't be built as it is
1008          * only used for debugging purposes.
1009          */
1010         ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
1011         uri_str[sizeof(uri_str) - 1] = '\0';
1012         if (ret >= 0) {
1013                 ret = asprintf(&formated_name, "%s socket @ %s", name, uri_str);
1014                 if (ret < 0) {
1015                         formated_name = nullptr;
1016                 }
1017         }
1018
1019         ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker,
1020                                                &sock_fd,
1021                                                (const char **) (formated_name ? &formated_name :
1022                                                                                 nullptr),
1023                                                1,
1024                                                create_sock,
1025                                                sock);
1026         if (ret) {
1027                 PERROR("Failed to open \"%s\" relay socket", formated_name ?: "Unknown");
1028                 goto error;
1029         }
1030         DBG("Listening on %s socket %d", name, sock->fd);
1031
1032         ret = sock->ops->bind(sock);
1033         if (ret < 0) {
1034                 PERROR("Failed to bind socket");
1035                 goto error;
1036         }
1037
1038         ret = sock->ops->listen(sock, -1);
1039         if (ret < 0) {
1040                 goto error;
1041         }
1042
1043         free(formated_name);
1044         return sock;
1045
1046 error:
1047         if (sock) {
1048                 lttcomm_destroy_sock(sock);
1049         }
1050         free(formated_name);
1051         return nullptr;
1052 }
1053
1054 static struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1055                                                const char *name)
1056 {
1057         int out_fd, ret;
1058         struct lttcomm_sock *socks[2] = { listening_sock, nullptr };
1059         struct lttcomm_sock *new_sock = nullptr;
1060
1061         ret = fd_tracker_open_unsuspendable_fd(
1062                 the_fd_tracker, &out_fd, (const char **) &name, 1, accept_sock, &socks);
1063         if (ret) {
1064                 goto end;
1065         }
1066         new_sock = socks[1];
1067         DBG("%s accepted, socket %d", name, new_sock->fd);
1068 end:
1069         return new_sock;
1070 }
1071
1072 /*
1073  * This thread manages the listening for new connections on the network
1074  */
1075 static void *relay_thread_listener(void *data __attribute__((unused)))
1076 {
1077         int i, ret, err = -1;
1078         uint32_t nb_fd;
1079         struct lttng_poll_event events;
1080         struct lttcomm_sock *control_sock, *data_sock;
1081
1082         DBG("[thread] Relay listener started");
1083
1084         rcu_register_thread();
1085         health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1086
1087         health_code_update();
1088
1089         control_sock = relay_socket_create(control_uri, "Control listener");
1090         if (!control_sock) {
1091                 goto error_sock_control;
1092         }
1093
1094         data_sock = relay_socket_create(data_uri, "Data listener");
1095         if (!data_sock) {
1096                 goto error_sock_relay;
1097         }
1098
1099         /*
1100          * Pass 3 as size here for the thread quit pipe, control and
1101          * data socket.
1102          */
1103         ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1104         if (ret < 0) {
1105                 goto error_create_poll;
1106         }
1107
1108         /* Add the control socket */
1109         ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1110         if (ret < 0) {
1111                 goto error_poll_add;
1112         }
1113
1114         /* Add the data socket */
1115         ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1116         if (ret < 0) {
1117                 goto error_poll_add;
1118         }
1119
1120         lttng_relay_notify_ready();
1121
1122         if (testpoint(relayd_thread_listener)) {
1123                 goto error_testpoint;
1124         }
1125
1126         while (true) {
1127                 health_code_update();
1128
1129                 DBG("Listener accepting connections");
1130
1131         restart:
1132                 health_poll_entry();
1133                 ret = lttng_poll_wait(&events, -1);
1134                 health_poll_exit();
1135                 if (ret < 0) {
1136                         /*
1137                          * Restart interrupted system call.
1138                          */
1139                         if (errno == EINTR) {
1140                                 goto restart;
1141                         }
1142                         goto error;
1143                 }
1144
1145                 nb_fd = ret;
1146
1147                 DBG("Relay new connection received");
1148                 for (i = 0; i < nb_fd; i++) {
1149                         /* Fetch once the poll data */
1150                         const auto revents = LTTNG_POLL_GETEV(&events, i);
1151                         const auto pollfd = LTTNG_POLL_GETFD(&events, i);
1152
1153                         health_code_update();
1154
1155                         /* Activity on thread quit pipe, exiting. */
1156                         if (relayd_is_thread_quit_pipe(pollfd)) {
1157                                 DBG("Activity on thread quit pipe");
1158                                 err = 0;
1159                                 goto exit;
1160                         }
1161
1162                         if (revents & LPOLLIN) {
1163                                 /*
1164                                  * A new connection is requested, therefore a
1165                                  * sessiond/consumerd connection is allocated in
1166                                  * this thread, enqueued to a global queue and
1167                                  * dequeued (and freed) in the worker thread.
1168                                  */
1169                                 int val = 1;
1170                                 struct relay_connection *new_conn;
1171                                 struct lttcomm_sock *newsock = nullptr;
1172                                 enum connection_type type;
1173
1174                                 if (pollfd == data_sock->fd) {
1175                                         type = RELAY_DATA;
1176                                         newsock = accept_relayd_sock(data_sock,
1177                                                                      "Data socket to relayd");
1178                                 } else {
1179                                         LTTNG_ASSERT(pollfd == control_sock->fd);
1180                                         type = RELAY_CONTROL;
1181                                         newsock = accept_relayd_sock(control_sock,
1182                                                                      "Control socket to relayd");
1183                                 }
1184                                 if (!newsock) {
1185                                         PERROR("accepting sock");
1186                                         goto error;
1187                                 }
1188
1189                                 ret = setsockopt(
1190                                         newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
1191                                 if (ret < 0) {
1192                                         PERROR("setsockopt inet");
1193                                         lttcomm_destroy_sock(newsock);
1194                                         goto error;
1195                                 }
1196
1197                                 ret = socket_apply_keep_alive_config(newsock->fd);
1198                                 if (ret < 0) {
1199                                         ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1200                                             newsock->fd);
1201                                         lttcomm_destroy_sock(newsock);
1202                                         goto error;
1203                                 }
1204
1205                                 new_conn = connection_create(newsock, type);
1206                                 if (!new_conn) {
1207                                         lttcomm_destroy_sock(newsock);
1208                                         goto error;
1209                                 }
1210
1211                                 /* Enqueue request for the dispatcher thread. */
1212                                 cds_wfcq_head_ptr_t head;
1213                                 head.h = &relay_conn_queue.head;
1214                                 cds_wfcq_enqueue(head, &relay_conn_queue.tail, &new_conn->qnode);
1215
1216                                 /*
1217                                  * Wake the dispatch queue futex.
1218                                  * Implicit memory barrier with the
1219                                  * exchange in cds_wfcq_enqueue.
1220                                  */
1221                                 futex_nto1_wake(&relay_conn_queue.futex);
1222                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1223                                 ERR("socket poll error");
1224                                 goto error;
1225                         } else {
1226                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1227                                 goto error;
1228                         }
1229                 }
1230         }
1231
1232 exit:
1233 error:
1234 error_poll_add:
1235 error_testpoint:
1236         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1237 error_create_poll:
1238         if (data_sock->fd >= 0) {
1239                 int data_sock_fd = data_sock->fd;
1240
1241                 ret = fd_tracker_close_unsuspendable_fd(
1242                         the_fd_tracker, &data_sock_fd, 1, close_sock, data_sock);
1243                 if (ret) {
1244                         PERROR("Failed to close the data listener socket file descriptor");
1245                 }
1246                 data_sock->fd = -1;
1247         }
1248         lttcomm_destroy_sock(data_sock);
1249 error_sock_relay:
1250         if (control_sock->fd >= 0) {
1251                 int control_sock_fd = control_sock->fd;
1252
1253                 ret = fd_tracker_close_unsuspendable_fd(
1254                         the_fd_tracker, &control_sock_fd, 1, close_sock, control_sock);
1255                 if (ret) {
1256                         PERROR("Failed to close the control listener socket file descriptor");
1257                 }
1258                 control_sock->fd = -1;
1259         }
1260         lttcomm_destroy_sock(control_sock);
1261 error_sock_control:
1262         if (err) {
1263                 health_error();
1264                 ERR("Health error occurred in %s", __func__);
1265         }
1266         health_unregister(health_relayd);
1267         rcu_unregister_thread();
1268         DBG("Relay listener thread cleanup complete");
1269         lttng_relay_stop_threads();
1270         return nullptr;
1271 }
1272
1273 /*
1274  * This thread manages the dispatching of the requests to worker threads
1275  */
1276 static void *relay_thread_dispatcher(void *data __attribute__((unused)))
1277 {
1278         int err = -1;
1279         ssize_t ret;
1280         struct cds_wfcq_node *node;
1281         struct relay_connection *new_conn = nullptr;
1282
1283         DBG("[thread] Relay dispatcher started");
1284
1285         health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1286
1287         if (testpoint(relayd_thread_dispatcher)) {
1288                 goto error_testpoint;
1289         }
1290
1291         health_code_update();
1292
1293         for (;;) {
1294                 health_code_update();
1295
1296                 /* Atomically prepare the queue futex */
1297                 futex_nto1_prepare(&relay_conn_queue.futex);
1298
1299                 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1300                         break;
1301                 }
1302
1303                 do {
1304                         health_code_update();
1305
1306                         /* Dequeue commands */
1307                         node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1308                                                          &relay_conn_queue.tail);
1309                         if (node == nullptr) {
1310                                 DBG("Woken up but nothing in the relay command queue");
1311                                 /* Continue thread execution */
1312                                 break;
1313                         }
1314                         new_conn = lttng::utils::container_of(node, &relay_connection::qnode);
1315
1316                         DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1317
1318                         /*
1319                          * Inform worker thread of the new request. This
1320                          * call is blocking so we can be assured that
1321                          * the data will be read at some point in time
1322                          * or wait to the end of the world :)
1323                          */
1324                         ret = lttng_write(
1325                                 relay_conn_pipe[1], &new_conn, sizeof(new_conn)); /* NOLINT
1326                                                                                      sizeof
1327                                                                                      used
1328                                                                                      on a
1329                                                                                      pointer.
1330                                                                                    */
1331                         if (ret < 0) {
1332                                 PERROR("write connection pipe");
1333                                 connection_put(new_conn);
1334                                 goto error;
1335                         }
1336                 } while (node != nullptr);
1337
1338                 /* Futex wait on queue. Blocking call on futex() */
1339                 health_poll_entry();
1340                 futex_nto1_wait(&relay_conn_queue.futex);
1341                 health_poll_exit();
1342         }
1343
1344         /* Normal exit, no error */
1345         err = 0;
1346
1347 error:
1348 error_testpoint:
1349         if (err) {
1350                 health_error();
1351                 ERR("Health error occurred in %s", __func__);
1352         }
1353         health_unregister(health_relayd);
1354         DBG("Dispatch thread dying");
1355         lttng_relay_stop_threads();
1356         return nullptr;
1357 }
1358
1359 static bool session_streams_have_index(const struct relay_session *session)
1360 {
1361         return session->minor >= 4 && !session->snapshot;
1362 }
1363
1364 /*
1365  * Handle the RELAYD_CREATE_SESSION command.
1366  *
1367  * On success, send back the session id or else return a negative value.
1368  */
1369 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1370                                 struct relay_connection *conn,
1371                                 const struct lttng_buffer_view *payload)
1372 {
1373         int ret = 0;
1374         ssize_t send_ret;
1375         struct relay_session *session = nullptr;
1376         struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1377         char session_name[LTTNG_NAME_MAX] = {};
1378         char hostname[LTTNG_HOST_NAME_MAX] = {};
1379         uint32_t live_timer = 0;
1380         bool snapshot = false;
1381         bool session_name_contains_creation_timestamp = false;
1382         /* Left nil for peers < 2.11. */
1383         char base_path[LTTNG_PATH_MAX] = {};
1384         lttng_uuid sessiond_uuid = {};
1385         LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1386         LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1387         LTTNG_OPTIONAL(time_t) creation_time = {};
1388         struct lttng_dynamic_buffer reply_payload;
1389
1390         lttng_dynamic_buffer_init(&reply_payload);
1391
1392         if (conn->minor < 4) {
1393                 /* From 2.1 to 2.3 */
1394                 ret = 0;
1395         } else if (conn->minor >= 4 && conn->minor < 11) {
1396                 /* From 2.4 to 2.10 */
1397                 ret = cmd_create_session_2_4(
1398                         payload, session_name, hostname, &live_timer, &snapshot);
1399         } else {
1400                 bool has_current_chunk;
1401                 uint64_t current_chunk_id_value;
1402                 time_t creation_time_value;
1403                 uint64_t id_sessiond_value;
1404
1405                 /* From 2.11 to ... */
1406                 ret = cmd_create_session_2_11(payload,
1407                                               session_name,
1408                                               hostname,
1409                                               base_path,
1410                                               &live_timer,
1411                                               &snapshot,
1412                                               &id_sessiond_value,
1413                                               sessiond_uuid,
1414                                               &has_current_chunk,
1415                                               &current_chunk_id_value,
1416                                               &creation_time_value,
1417                                               &session_name_contains_creation_timestamp);
1418                 if (lttng_uuid_is_nil(sessiond_uuid)) {
1419                         /* The nil UUID is reserved for pre-2.11 clients. */
1420                         ERR("Illegal nil UUID announced by peer in create session command");
1421                         ret = -1;
1422                         goto send_reply;
1423                 }
1424                 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1425                 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1426                 if (has_current_chunk) {
1427                         LTTNG_OPTIONAL_SET(&current_chunk_id, current_chunk_id_value);
1428                 }
1429         }
1430
1431         if (ret < 0) {
1432                 goto send_reply;
1433         }
1434
1435         session = session_create(session_name,
1436                                  hostname,
1437                                  base_path,
1438                                  live_timer,
1439                                  snapshot,
1440                                  sessiond_uuid,
1441                                  id_sessiond.is_set ? &id_sessiond.value : nullptr,
1442                                  current_chunk_id.is_set ? &current_chunk_id.value : nullptr,
1443                                  creation_time.is_set ? &creation_time.value : nullptr,
1444                                  conn->major,
1445                                  conn->minor,
1446                                  session_name_contains_creation_timestamp);
1447         if (!session) {
1448                 ret = -1;
1449                 goto send_reply;
1450         }
1451         LTTNG_ASSERT(!conn->session);
1452         conn->session = session;
1453         DBG("Created session %" PRIu64, session->id);
1454
1455         reply.generic.session_id = htobe64(session->id);
1456
1457 send_reply:
1458         if (ret < 0) {
1459                 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1460         } else {
1461                 reply.generic.ret_code = htobe32(LTTNG_OK);
1462         }
1463
1464         if (conn->minor < 11) {
1465                 /* From 2.1 to 2.10 */
1466                 ret = lttng_dynamic_buffer_append(
1467                         &reply_payload, &reply.generic, sizeof(reply.generic));
1468                 if (ret) {
1469                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1470                         ret = -1;
1471                         goto end;
1472                 }
1473         } else {
1474                 const uint32_t output_path_length = session ? strlen(session->output_path) + 1 : 0;
1475
1476                 reply.output_path_length = htobe32(output_path_length);
1477                 ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
1478                 if (ret) {
1479                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1480                         goto end;
1481                 }
1482
1483                 if (output_path_length) {
1484                         ret = lttng_dynamic_buffer_append(
1485                                 &reply_payload, session->output_path, output_path_length);
1486                         if (ret) {
1487                                 ERR("Failed to append \"create session\" command reply path to payload buffer");
1488                                 goto end;
1489                         }
1490                 }
1491         }
1492
1493         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
1494         if (send_ret < (ssize_t) reply_payload.size) {
1495                 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1496                     reply_payload.size,
1497                     send_ret);
1498                 ret = -1;
1499         }
1500 end:
1501         if (ret < 0 && session) {
1502                 session_put(session);
1503         }
1504         lttng_dynamic_buffer_reset(&reply_payload);
1505         return ret;
1506 }
1507
1508 /*
1509  * When we have received all the streams and the metadata for a channel,
1510  * we make them visible to the viewer threads.
1511  */
1512 static void publish_connection_local_streams(struct relay_connection *conn)
1513 {
1514         struct relay_stream *stream;
1515         struct relay_session *session = conn->session;
1516
1517         /*
1518          * We publish all streams belonging to a session atomically wrt
1519          * session lock.
1520          */
1521         pthread_mutex_lock(&session->lock);
1522         lttng::urcu::read_lock_guard read_lock;
1523         cds_list_for_each_entry_rcu(stream, &session->recv_list, recv_node)
1524         {
1525                 stream_publish(stream);
1526         }
1527
1528         /*
1529          * Inform the viewer that there are new streams in the session.
1530          */
1531         if (session->viewer_attached) {
1532                 uatomic_set(&session->new_streams, 1);
1533         }
1534         pthread_mutex_unlock(&session->lock);
1535 }
1536
1537 static int conform_channel_path(char *channel_path)
1538 {
1539         int ret = 0;
1540
1541         if (strstr("../", channel_path)) {
1542                 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1543                     channel_path);
1544                 ret = -1;
1545                 goto end;
1546         }
1547
1548         if (*channel_path == '/') {
1549                 const size_t len = strlen(channel_path);
1550
1551                 /*
1552                  * Channel paths from peers prior to 2.11 are expressed as an
1553                  * absolute path that is, in reality, relative to the relay
1554                  * daemon's output directory. Remove the leading slash so it
1555                  * is correctly interpreted as a relative path later on.
1556                  *
1557                  * len (and not len - 1) is used to copy the trailing NULL.
1558                  */
1559                 bcopy(channel_path + 1, channel_path, len);
1560         }
1561 end:
1562         return ret;
1563 }
1564
1565 /*
1566  * relay_add_stream: allocate a new stream for a session
1567  */
1568 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1569                             struct relay_connection *conn,
1570                             const struct lttng_buffer_view *payload)
1571 {
1572         int ret;
1573         ssize_t send_ret;
1574         struct relay_session *session = conn->session;
1575         struct relay_stream *stream = nullptr;
1576         struct lttcomm_relayd_status_stream reply;
1577         struct ctf_trace *trace = nullptr;
1578         uint64_t stream_handle = -1ULL;
1579         char *path_name = nullptr, *channel_name = nullptr;
1580         uint64_t tracefile_size = 0, tracefile_count = 0;
1581         LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1582
1583         if (!session || !conn->version_check_done) {
1584                 ERR("Trying to add a stream before version check");
1585                 ret = -1;
1586                 goto end_no_session;
1587         }
1588
1589         if (session->minor == 1) {
1590                 /* For 2.1 */
1591                 ret = cmd_recv_stream_2_1(payload, &path_name, &channel_name);
1592         } else if (session->minor > 1 && session->minor < 11) {
1593                 /* From 2.2 to 2.10 */
1594                 ret = cmd_recv_stream_2_2(
1595                         payload, &path_name, &channel_name, &tracefile_size, &tracefile_count);
1596         } else {
1597                 /* From 2.11 to ... */
1598                 ret = cmd_recv_stream_2_11(payload,
1599                                            &path_name,
1600                                            &channel_name,
1601                                            &tracefile_size,
1602                                            &tracefile_count,
1603                                            &stream_chunk_id.value);
1604                 stream_chunk_id.is_set = true;
1605         }
1606
1607         if (ret < 0) {
1608                 goto send_reply;
1609         }
1610
1611         if (conform_channel_path(path_name)) {
1612                 goto send_reply;
1613         }
1614
1615         /*
1616          * Backward compatibility for --group-output-by-session.
1617          * Prior to lttng 2.11, the complete path is passed by the stream.
1618          * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1619          * >=2.11 the chunk is responsible for the output path. When dealing
1620          * with producer < 2.11 the chunk output_path is the root output path
1621          * and the stream carries the complete path (path_name).
1622          * To support --group-output-by-session with older producer (<2.11), we
1623          * need to craft the path based on the stream path.
1624          */
1625         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1626                 if (conn->minor < 4) {
1627                         /*
1628                          * From 2.1 to 2.3, the session_name is not passed on
1629                          * the RELAYD_CREATE_SESSION command. The session name
1630                          * is necessary to detect the presence of a base_path
1631                          * inside the stream path. Without it we cannot perform
1632                          * a valid group-output-by-session transformation.
1633                          */
1634                         WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1635                              " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1636                              session->id,
1637                              path_name);
1638                 } else if (conn->minor >= 4 && conn->minor < 11) {
1639                         char *group_by_session_path_name;
1640
1641                         LTTNG_ASSERT(session->session_name[0] != '\0');
1642
1643                         group_by_session_path_name = backward_compat_group_by_session(
1644                                 path_name, session->session_name, session->creation_time.value);
1645                         if (!group_by_session_path_name) {
1646                                 ERR("Failed to apply group by session to stream of session %" PRIu64,
1647                                     session->id);
1648                                 goto send_reply;
1649                         }
1650
1651                         DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1652                             path_name,
1653                             group_by_session_path_name);
1654
1655                         free(path_name);
1656                         path_name = group_by_session_path_name;
1657                 }
1658         }
1659
1660         trace = ctf_trace_get_by_path_or_create(session, path_name);
1661         if (!trace) {
1662                 goto send_reply;
1663         }
1664
1665         /* This stream here has one reference on the trace. */
1666         pthread_mutex_lock(&last_relay_stream_id_lock);
1667         stream_handle = ++last_relay_stream_id;
1668         pthread_mutex_unlock(&last_relay_stream_id_lock);
1669
1670         /* We pass ownership of path_name and channel_name. */
1671         stream = stream_create(
1672                 trace, stream_handle, path_name, channel_name, tracefile_size, tracefile_count);
1673         path_name = nullptr;
1674         channel_name = nullptr;
1675
1676         /*
1677          * Streams are the owners of their trace. Reference to trace is
1678          * kept within stream_create().
1679          */
1680         ctf_trace_put(trace);
1681
1682 send_reply:
1683         memset(&reply, 0, sizeof(reply));
1684         reply.handle = htobe64(stream_handle);
1685         if (!stream) {
1686                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1687         } else {
1688                 reply.ret_code = htobe32(LTTNG_OK);
1689         }
1690
1691         send_ret = conn->sock->ops->sendmsg(
1692                 conn->sock, &reply, sizeof(struct lttcomm_relayd_status_stream), 0);
1693         if (send_ret < (ssize_t) sizeof(reply)) {
1694                 ERR("Failed to send \"add stream\" command reply (ret = %zd)", send_ret);
1695                 ret = -1;
1696         }
1697
1698 end_no_session:
1699         free(path_name);
1700         free(channel_name);
1701         return ret;
1702 }
1703
1704 /*
1705  * relay_close_stream: close a specific stream
1706  */
1707 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1708                               struct relay_connection *conn,
1709                               const struct lttng_buffer_view *payload)
1710 {
1711         int ret;
1712         ssize_t send_ret;
1713         struct relay_session *session = conn->session;
1714         struct lttcomm_relayd_close_stream stream_info;
1715         struct lttcomm_relayd_generic_reply reply;
1716         struct relay_stream *stream;
1717
1718         DBG("Close stream received");
1719
1720         if (!session || !conn->version_check_done) {
1721                 ERR("Trying to close a stream before version check");
1722                 ret = -1;
1723                 goto end_no_session;
1724         }
1725
1726         if (payload->size < sizeof(stream_info)) {
1727                 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1728                     sizeof(stream_info),
1729                     payload->size);
1730                 ret = -1;
1731                 goto end_no_session;
1732         }
1733         memcpy(&stream_info, payload->data, sizeof(stream_info));
1734         stream_info.stream_id = be64toh(stream_info.stream_id);
1735         stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1736
1737         stream = stream_get_by_id(stream_info.stream_id);
1738         if (!stream) {
1739                 ret = -1;
1740                 goto end;
1741         }
1742
1743         /*
1744          * Set last_net_seq_num before the close flag. Required by data
1745          * pending check.
1746          */
1747         pthread_mutex_lock(&stream->lock);
1748         stream->last_net_seq_num = stream_info.last_net_seq_num;
1749         pthread_mutex_unlock(&stream->lock);
1750
1751         /*
1752          * This is one of the conditions which may trigger a stream close
1753          * with the others being:
1754          *     1) A close command is received for a stream
1755          *     2) The control connection owning the stream is closed
1756          *     3) We have received all of the stream's data _after_ a close
1757          *        request.
1758          */
1759         try_stream_close(stream);
1760         stream_put(stream);
1761         ret = 0;
1762
1763 end:
1764         memset(&reply, 0, sizeof(reply));
1765         if (ret < 0) {
1766                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1767         } else {
1768                 reply.ret_code = htobe32(LTTNG_OK);
1769         }
1770         send_ret = conn->sock->ops->sendmsg(
1771                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1772         if (send_ret < (ssize_t) sizeof(reply)) {
1773                 ERR("Failed to send \"close stream\" command reply (ret = %zd)", send_ret);
1774                 ret = -1;
1775         }
1776
1777 end_no_session:
1778         return ret;
1779 }
1780
1781 /*
1782  * relay_reset_metadata: reset a metadata stream
1783  */
1784 static int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1785                                 struct relay_connection *conn,
1786                                 const struct lttng_buffer_view *payload)
1787 {
1788         int ret;
1789         ssize_t send_ret;
1790         struct relay_session *session = conn->session;
1791         struct lttcomm_relayd_reset_metadata stream_info;
1792         struct lttcomm_relayd_generic_reply reply;
1793         struct relay_stream *stream;
1794
1795         DBG("Reset metadata received");
1796
1797         if (!session || !conn->version_check_done) {
1798                 ERR("Trying to reset a metadata stream before version check");
1799                 ret = -1;
1800                 goto end_no_session;
1801         }
1802
1803         if (payload->size < sizeof(stream_info)) {
1804                 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1805                     sizeof(stream_info),
1806                     payload->size);
1807                 ret = -1;
1808                 goto end_no_session;
1809         }
1810         memcpy(&stream_info, payload->data, sizeof(stream_info));
1811         stream_info.stream_id = be64toh(stream_info.stream_id);
1812         stream_info.version = be64toh(stream_info.version);
1813
1814         DBG("Update metadata to version %" PRIu64, stream_info.version);
1815
1816         /* Unsupported for live sessions for now. */
1817         if (session->live_timer != 0) {
1818                 ret = -1;
1819                 goto end;
1820         }
1821
1822         stream = stream_get_by_id(stream_info.stream_id);
1823         if (!stream) {
1824                 ret = -1;
1825                 goto end;
1826         }
1827         pthread_mutex_lock(&stream->lock);
1828         if (!stream->is_metadata) {
1829                 ret = -1;
1830                 goto end_unlock;
1831         }
1832
1833         ret = stream_reset_file(stream);
1834         if (ret < 0) {
1835                 ERR("Failed to reset metadata stream %" PRIu64 ": stream_path = %s, channel = %s",
1836                     stream->stream_handle,
1837                     stream->path_name,
1838                     stream->channel_name);
1839                 goto end_unlock;
1840         }
1841 end_unlock:
1842         pthread_mutex_unlock(&stream->lock);
1843         stream_put(stream);
1844
1845 end:
1846         memset(&reply, 0, sizeof(reply));
1847         if (ret < 0) {
1848                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1849         } else {
1850                 reply.ret_code = htobe32(LTTNG_OK);
1851         }
1852         send_ret = conn->sock->ops->sendmsg(
1853                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1854         if (send_ret < (ssize_t) sizeof(reply)) {
1855                 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)", send_ret);
1856                 ret = -1;
1857         }
1858
1859 end_no_session:
1860         return ret;
1861 }
1862
1863 /*
1864  * relay_unknown_command: send -1 if received unknown command
1865  */
1866 static void relay_unknown_command(struct relay_connection *conn)
1867 {
1868         struct lttcomm_relayd_generic_reply reply;
1869         ssize_t send_ret;
1870
1871         memset(&reply, 0, sizeof(reply));
1872         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1873         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1874         if (send_ret < sizeof(reply)) {
1875                 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1876         }
1877 }
1878
1879 /*
1880  * relay_start: send an acknowledgment to the client to tell if we are
1881  * ready to receive data. We are ready if a session is established.
1882  */
1883 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1884                        struct relay_connection *conn,
1885                        const struct lttng_buffer_view *payload __attribute__((unused)))
1886 {
1887         int ret = 0;
1888         ssize_t send_ret;
1889         struct lttcomm_relayd_generic_reply reply;
1890         struct relay_session *session = conn->session;
1891
1892         if (!session) {
1893                 DBG("Trying to start the streaming without a session established");
1894                 ret = htobe32(LTTNG_ERR_UNK);
1895         }
1896
1897         memset(&reply, 0, sizeof(reply));
1898         reply.ret_code = htobe32(LTTNG_OK);
1899         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1900         if (send_ret < (ssize_t) sizeof(reply)) {
1901                 ERR("Failed to send \"relay_start\" command reply (ret = %zd)", send_ret);
1902                 ret = -1;
1903         }
1904
1905         return ret;
1906 }
1907
1908 /*
1909  * relay_recv_metadata: receive the metadata for the session.
1910  */
1911 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1912                                struct relay_connection *conn,
1913                                const struct lttng_buffer_view *payload)
1914 {
1915         int ret = 0;
1916         struct relay_session *session = conn->session;
1917         struct lttcomm_relayd_metadata_payload metadata_payload_header;
1918         struct relay_stream *metadata_stream;
1919         uint64_t metadata_payload_size;
1920         struct lttng_buffer_view packet_view;
1921
1922         if (!session) {
1923                 ERR("Metadata sent before version check");
1924                 ret = -1;
1925                 goto end;
1926         }
1927
1928         if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1929                 ERR("Incorrect data size");
1930                 ret = -1;
1931                 goto end;
1932         }
1933         metadata_payload_size =
1934                 recv_hdr->data_size - sizeof(struct lttcomm_relayd_metadata_payload);
1935
1936         memcpy(&metadata_payload_header, payload->data, sizeof(metadata_payload_header));
1937         metadata_payload_header.stream_id = be64toh(metadata_payload_header.stream_id);
1938         metadata_payload_header.padding_size = be32toh(metadata_payload_header.padding_size);
1939
1940         metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1941         if (!metadata_stream) {
1942                 ret = -1;
1943                 goto end;
1944         }
1945
1946         packet_view = lttng_buffer_view_from_view(
1947                 payload, sizeof(metadata_payload_header), metadata_payload_size);
1948         if (!lttng_buffer_view_is_valid(&packet_view)) {
1949                 ERR("Invalid metadata packet length announced by header");
1950                 ret = -1;
1951                 goto end_put;
1952         }
1953
1954         pthread_mutex_lock(&metadata_stream->lock);
1955         ret = stream_write(metadata_stream, &packet_view, metadata_payload_header.padding_size);
1956         pthread_mutex_unlock(&metadata_stream->lock);
1957         if (ret) {
1958                 ret = -1;
1959                 goto end_put;
1960         }
1961 end_put:
1962         stream_put(metadata_stream);
1963 end:
1964         return ret;
1965 }
1966
1967 /*
1968  * relay_send_version: send relayd version number
1969  */
1970 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1971                               struct relay_connection *conn,
1972                               const struct lttng_buffer_view *payload)
1973 {
1974         int ret;
1975         ssize_t send_ret;
1976         struct lttcomm_relayd_version reply, msg;
1977         bool compatible = true;
1978
1979         conn->version_check_done = true;
1980
1981         /* Get version from the other side. */
1982         if (payload->size < sizeof(msg)) {
1983                 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1984                     sizeof(msg),
1985                     payload->size);
1986                 ret = -1;
1987                 goto end;
1988         }
1989
1990         memcpy(&msg, payload->data, sizeof(msg));
1991         msg.major = be32toh(msg.major);
1992         msg.minor = be32toh(msg.minor);
1993
1994         memset(&reply, 0, sizeof(reply));
1995         reply.major = RELAYD_VERSION_COMM_MAJOR;
1996         reply.minor = RELAYD_VERSION_COMM_MINOR;
1997
1998         /* Major versions must be the same */
1999         if (reply.major != msg.major) {
2000                 DBG("Incompatible major versions (%u vs %u), deleting session",
2001                     reply.major,
2002                     msg.major);
2003                 compatible = false;
2004         }
2005
2006         conn->major = reply.major;
2007         /* We adapt to the lowest compatible version */
2008         if (reply.minor <= msg.minor) {
2009                 conn->minor = reply.minor;
2010         } else {
2011                 conn->minor = msg.minor;
2012         }
2013
2014         reply.major = htobe32(reply.major);
2015         reply.minor = htobe32(reply.minor);
2016         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2017         if (send_ret < (ssize_t) sizeof(reply)) {
2018                 ERR("Failed to send \"send version\" command reply (ret = %zd)", send_ret);
2019                 ret = -1;
2020                 goto end;
2021         } else {
2022                 ret = 0;
2023         }
2024
2025         if (!compatible) {
2026                 ret = -1;
2027                 goto end;
2028         }
2029
2030         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
2031
2032 end:
2033         return ret;
2034 }
2035
2036 /*
2037  * Check for data pending for a given stream id from the session daemon.
2038  */
2039 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2040                               struct relay_connection *conn,
2041                               const struct lttng_buffer_view *payload)
2042 {
2043         struct relay_session *session = conn->session;
2044         struct lttcomm_relayd_data_pending msg;
2045         struct lttcomm_relayd_generic_reply reply;
2046         struct relay_stream *stream;
2047         ssize_t send_ret;
2048         int ret;
2049         uint64_t stream_seq;
2050
2051         DBG("Data pending command received");
2052
2053         if (!session || !conn->version_check_done) {
2054                 ERR("Trying to check for data before version check");
2055                 ret = -1;
2056                 goto end_no_session;
2057         }
2058
2059         if (payload->size < sizeof(msg)) {
2060                 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2061                     sizeof(msg),
2062                     payload->size);
2063                 ret = -1;
2064                 goto end_no_session;
2065         }
2066         memcpy(&msg, payload->data, sizeof(msg));
2067         msg.stream_id = be64toh(msg.stream_id);
2068         msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2069
2070         stream = stream_get_by_id(msg.stream_id);
2071         if (stream == nullptr) {
2072                 ret = -1;
2073                 goto end;
2074         }
2075
2076         pthread_mutex_lock(&stream->lock);
2077
2078         if (session_streams_have_index(session)) {
2079                 /*
2080                  * Ensure that both the index and stream data have been
2081                  * flushed up to the requested point.
2082                  */
2083                 stream_seq = std::min(stream->prev_data_seq, stream->prev_index_seq);
2084         } else {
2085                 stream_seq = stream->prev_data_seq;
2086         }
2087         DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2088             ", prev_index_seq %" PRIu64 ", and last_seq %" PRIu64,
2089             msg.stream_id,
2090             stream->prev_data_seq,
2091             stream->prev_index_seq,
2092             msg.last_net_seq_num);
2093
2094         /* Avoid wrapping issue */
2095         if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2096                 /* Data has in fact been written and is NOT pending */
2097                 ret = 0;
2098         } else {
2099                 /* Data still being streamed thus pending */
2100                 ret = 1;
2101         }
2102
2103         stream->data_pending_check_done = true;
2104         pthread_mutex_unlock(&stream->lock);
2105
2106         stream_put(stream);
2107 end:
2108
2109         memset(&reply, 0, sizeof(reply));
2110         reply.ret_code = htobe32(ret);
2111         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2112         if (send_ret < (ssize_t) sizeof(reply)) {
2113                 ERR("Failed to send \"data pending\" command reply (ret = %zd)", send_ret);
2114                 ret = -1;
2115         }
2116
2117 end_no_session:
2118         return ret;
2119 }
2120
2121 /*
2122  * Wait for the control socket to reach a quiescent state.
2123  *
2124  * Note that for now, when receiving this command from the session
2125  * daemon, this means that every subsequent commands or data received on
2126  * the control socket has been handled. So, this is why we simply return
2127  * OK here.
2128  */
2129 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr
2130                                    __attribute__((unused)),
2131                                    struct relay_connection *conn,
2132                                    const struct lttng_buffer_view *payload)
2133 {
2134         int ret;
2135         ssize_t send_ret;
2136         struct relay_stream *stream;
2137         struct lttcomm_relayd_quiescent_control msg;
2138         struct lttcomm_relayd_generic_reply reply;
2139
2140         DBG("Checking quiescent state on control socket");
2141
2142         if (!conn->session || !conn->version_check_done) {
2143                 ERR("Trying to check for data before version check");
2144                 ret = -1;
2145                 goto end_no_session;
2146         }
2147
2148         if (payload->size < sizeof(msg)) {
2149                 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2150                     sizeof(msg),
2151                     payload->size);
2152                 ret = -1;
2153                 goto end_no_session;
2154         }
2155         memcpy(&msg, payload->data, sizeof(msg));
2156         msg.stream_id = be64toh(msg.stream_id);
2157
2158         stream = stream_get_by_id(msg.stream_id);
2159         if (!stream) {
2160                 goto reply;
2161         }
2162         pthread_mutex_lock(&stream->lock);
2163         stream->data_pending_check_done = true;
2164         pthread_mutex_unlock(&stream->lock);
2165
2166         DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2167         stream_put(stream);
2168 reply:
2169         memset(&reply, 0, sizeof(reply));
2170         reply.ret_code = htobe32(LTTNG_OK);
2171         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2172         if (send_ret < (ssize_t) sizeof(reply)) {
2173                 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)", send_ret);
2174                 ret = -1;
2175         } else {
2176                 ret = 0;
2177         }
2178
2179 end_no_session:
2180         return ret;
2181 }
2182
2183 /*
2184  * Initialize a data pending command. This means that a consumer is about
2185  * to ask for data pending for each stream it holds. Simply iterate over
2186  * all streams of a session and set the data_pending_check_done flag.
2187  *
2188  * This command returns to the client a LTTNG_OK code.
2189  */
2190 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2191                                     struct relay_connection *conn,
2192                                     const struct lttng_buffer_view *payload)
2193 {
2194         int ret;
2195         ssize_t send_ret;
2196         struct lttng_ht_iter iter;
2197         struct lttcomm_relayd_begin_data_pending msg;
2198         struct lttcomm_relayd_generic_reply reply;
2199         struct relay_stream *stream;
2200
2201         LTTNG_ASSERT(recv_hdr);
2202         LTTNG_ASSERT(conn);
2203
2204         DBG("Init streams for data pending");
2205
2206         if (!conn->session || !conn->version_check_done) {
2207                 ERR("Trying to check for data before version check");
2208                 ret = -1;
2209                 goto end_no_session;
2210         }
2211
2212         if (payload->size < sizeof(msg)) {
2213                 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2214                     sizeof(msg),
2215                     payload->size);
2216                 ret = -1;
2217                 goto end_no_session;
2218         }
2219         memcpy(&msg, payload->data, sizeof(msg));
2220         msg.session_id = be64toh(msg.session_id);
2221
2222         /*
2223          * Iterate over all streams to set the begin data pending flag.
2224          * For now, the streams are indexed by stream handle so we have
2225          * to iterate over all streams to find the one associated with
2226          * the right session_id.
2227          */
2228         {
2229                 lttng::urcu::read_lock_guard read_lock;
2230
2231                 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2232                         if (!stream_get(stream)) {
2233                                 continue;
2234                         }
2235
2236                         if (stream->trace->session->id == msg.session_id) {
2237                                 pthread_mutex_lock(&stream->lock);
2238                                 stream->data_pending_check_done = false;
2239                                 pthread_mutex_unlock(&stream->lock);
2240                                 DBG("Set begin data pending flag to stream %" PRIu64,
2241                                     stream->stream_handle);
2242                         }
2243
2244                         stream_put(stream);
2245                 }
2246         }
2247
2248         memset(&reply, 0, sizeof(reply));
2249         /* All good, send back reply. */
2250         reply.ret_code = htobe32(LTTNG_OK);
2251
2252         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2253         if (send_ret < (ssize_t) sizeof(reply)) {
2254                 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)", send_ret);
2255                 ret = -1;
2256         } else {
2257                 ret = 0;
2258         }
2259
2260 end_no_session:
2261         return ret;
2262 }
2263
2264 /*
2265  * End data pending command. This will check, for a given session id, if
2266  * each stream associated with it has its data_pending_check_done flag
2267  * set. If not, this means that the client lost track of the stream but
2268  * the data is still being streamed on our side. In this case, we inform
2269  * the client that data is in flight.
2270  *
2271  * Return to the client if there is data in flight or not with a ret_code.
2272  */
2273 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2274                                   struct relay_connection *conn,
2275                                   const struct lttng_buffer_view *payload)
2276 {
2277         int ret;
2278         ssize_t send_ret;
2279         struct lttng_ht_iter iter;
2280         struct lttcomm_relayd_end_data_pending msg;
2281         struct lttcomm_relayd_generic_reply reply;
2282         struct relay_stream *stream;
2283         uint32_t is_data_inflight = 0;
2284
2285         DBG("End data pending command");
2286
2287         if (!conn->session || !conn->version_check_done) {
2288                 ERR("Trying to check for data before version check");
2289                 ret = -1;
2290                 goto end_no_session;
2291         }
2292
2293         if (payload->size < sizeof(msg)) {
2294                 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2295                     sizeof(msg),
2296                     payload->size);
2297                 ret = -1;
2298                 goto end_no_session;
2299         }
2300         memcpy(&msg, payload->data, sizeof(msg));
2301         msg.session_id = be64toh(msg.session_id);
2302
2303         /*
2304          * Iterate over all streams to see if the begin data pending
2305          * flag is set.
2306          */
2307         {
2308                 lttng::urcu::read_lock_guard read_lock;
2309
2310                 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2311                         if (!stream_get(stream)) {
2312                                 continue;
2313                         }
2314
2315                         if (stream->trace->session->id != msg.session_id) {
2316                                 stream_put(stream);
2317                                 continue;
2318                         }
2319
2320                         pthread_mutex_lock(&stream->lock);
2321                         if (!stream->data_pending_check_done) {
2322                                 uint64_t stream_seq;
2323
2324                                 if (session_streams_have_index(conn->session)) {
2325                                         /*
2326                                          * Ensure that both the index and stream data have been
2327                                          * flushed up to the requested point.
2328                                          */
2329                                         stream_seq = std::min(stream->prev_data_seq,
2330                                                               stream->prev_index_seq);
2331                                 } else {
2332                                         stream_seq = stream->prev_data_seq;
2333                                 }
2334
2335                                 if (!stream->closed ||
2336                                     !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2337                                         is_data_inflight = 1;
2338                                         DBG("Data is still in flight for stream %" PRIu64,
2339                                             stream->stream_handle);
2340                                         pthread_mutex_unlock(&stream->lock);
2341                                         stream_put(stream);
2342                                         break;
2343                                 }
2344                         }
2345
2346                         pthread_mutex_unlock(&stream->lock);
2347                         stream_put(stream);
2348                 }
2349         }
2350
2351         memset(&reply, 0, sizeof(reply));
2352         /* All good, send back reply. */
2353         reply.ret_code = htobe32(is_data_inflight);
2354
2355         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2356         if (send_ret < (ssize_t) sizeof(reply)) {
2357                 ERR("Failed to send \"end data pending\" command reply (ret = %zd)", send_ret);
2358                 ret = -1;
2359         } else {
2360                 ret = 0;
2361         }
2362
2363 end_no_session:
2364         return ret;
2365 }
2366
2367 /*
2368  * Receive an index for a specific stream.
2369  *
2370  * Return 0 on success else a negative value.
2371  */
2372 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2373                             struct relay_connection *conn,
2374                             const struct lttng_buffer_view *payload)
2375 {
2376         int ret;
2377         ssize_t send_ret;
2378         struct relay_session *session = conn->session;
2379         struct lttcomm_relayd_index index_info;
2380         struct lttcomm_relayd_generic_reply reply;
2381         struct relay_stream *stream;
2382         size_t msg_len;
2383
2384         LTTNG_ASSERT(conn);
2385
2386         DBG("Relay receiving index");
2387
2388         if (!session || !conn->version_check_done) {
2389                 ERR("Trying to close a stream before version check");
2390                 ret = -1;
2391                 goto end_no_session;
2392         }
2393
2394         msg_len = lttcomm_relayd_index_len(lttng_to_index_major(conn->major, conn->minor),
2395                                            lttng_to_index_minor(conn->major, conn->minor));
2396         if (payload->size < msg_len) {
2397                 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2398                     msg_len,
2399                     payload->size);
2400                 ret = -1;
2401                 goto end_no_session;
2402         }
2403         memcpy(&index_info, payload->data, msg_len);
2404         index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2405         index_info.net_seq_num = be64toh(index_info.net_seq_num);
2406         index_info.packet_size = be64toh(index_info.packet_size);
2407         index_info.content_size = be64toh(index_info.content_size);
2408         index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2409         index_info.timestamp_end = be64toh(index_info.timestamp_end);
2410         index_info.events_discarded = be64toh(index_info.events_discarded);
2411         index_info.stream_id = be64toh(index_info.stream_id);
2412
2413         if (conn->minor >= 8) {
2414                 index_info.stream_instance_id = be64toh(index_info.stream_instance_id);
2415                 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2416         } else {
2417                 index_info.stream_instance_id = -1ULL;
2418                 index_info.packet_seq_num = -1ULL;
2419         }
2420
2421         stream = stream_get_by_id(index_info.relay_stream_id);
2422         if (!stream) {
2423                 ERR("stream_get_by_id not found");
2424                 ret = -1;
2425                 goto end;
2426         }
2427
2428         pthread_mutex_lock(&stream->lock);
2429         ret = stream_add_index(stream, &index_info);
2430         pthread_mutex_unlock(&stream->lock);
2431         if (ret) {
2432                 goto end_stream_put;
2433         }
2434
2435 end_stream_put:
2436         stream_put(stream);
2437 end:
2438         memset(&reply, 0, sizeof(reply));
2439         if (ret < 0) {
2440                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2441         } else {
2442                 reply.ret_code = htobe32(LTTNG_OK);
2443         }
2444         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2445         if (send_ret < (ssize_t) sizeof(reply)) {
2446                 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2447                 ret = -1;
2448         }
2449
2450 end_no_session:
2451         return ret;
2452 }
2453
2454 /*
2455  * Receive the streams_sent message.
2456  *
2457  * Return 0 on success else a negative value.
2458  */
2459 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2460                               struct relay_connection *conn,
2461                               const struct lttng_buffer_view *payload __attribute__((unused)))
2462 {
2463         int ret;
2464         ssize_t send_ret;
2465         struct lttcomm_relayd_generic_reply reply;
2466
2467         LTTNG_ASSERT(conn);
2468
2469         DBG("Relay receiving streams_sent");
2470
2471         if (!conn->session || !conn->version_check_done) {
2472                 ERR("Trying to close a stream before version check");
2473                 ret = -1;
2474                 goto end_no_session;
2475         }
2476
2477         /*
2478          * Publish every pending stream in the connection recv list which are
2479          * now ready to be used by the viewer.
2480          */
2481         publish_connection_local_streams(conn);
2482
2483         memset(&reply, 0, sizeof(reply));
2484         reply.ret_code = htobe32(LTTNG_OK);
2485         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2486         if (send_ret < (ssize_t) sizeof(reply)) {
2487                 ERR("Failed to send \"streams sent\" command reply (ret = %zd)", send_ret);
2488                 ret = -1;
2489         } else {
2490                 /* Success. */
2491                 ret = 0;
2492         }
2493
2494 end_no_session:
2495         return ret;
2496 }
2497
2498 static ssize_t
2499 relay_unpack_rotate_streams_header(const struct lttng_buffer_view *payload,
2500                                    struct lttcomm_relayd_rotate_streams *_rotate_streams)
2501 {
2502         struct lttcomm_relayd_rotate_streams rotate_streams;
2503         /*
2504          * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2505          * This is the smallest version of this structure, but it can be larger;
2506          * this variable is updated once the proper size of the structure is known.
2507          *
2508          * See comment at the declaration of this structure for more information.
2509          */
2510         ssize_t header_len = sizeof(struct lttcomm_relayd_rotate_streams_packed);
2511         size_t expected_payload_size_no_padding, expected_payload_size_3_bytes_padding,
2512                 expected_payload_size_7_bytes_padding;
2513
2514         if (payload->size < header_len) {
2515                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2516                     header_len,
2517                     payload->size);
2518                 goto error;
2519         }
2520
2521         /*
2522          * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2523          * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2524          *
2525          * We start by "unpacking" `stream_count` to figure out the padding length
2526          * emited by our peer.
2527          */
2528         {
2529                 decltype(rotate_streams.stream_count) stream_count;
2530
2531                 memcpy(&stream_count, payload->data, sizeof(stream_count));
2532                 rotate_streams.stream_count = be32toh(stream_count);
2533         }
2534
2535         rotate_streams.new_chunk_id = LTTNG_OPTIONAL_INIT_UNSET;
2536
2537         /*
2538          * Payload size expected given the possible padding lengths in
2539          * `struct lttcomm_relayd_rotate_streams`.
2540          */
2541         expected_payload_size_no_padding =
2542                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2543                 sizeof(lttcomm_relayd_rotate_streams_packed);
2544         expected_payload_size_3_bytes_padding =
2545                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2546                 sizeof(lttcomm_relayd_rotate_streams_3_bytes_padding);
2547         expected_payload_size_7_bytes_padding =
2548                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2549                 sizeof(lttcomm_relayd_rotate_streams_7_bytes_padding);
2550
2551         if (payload->size == expected_payload_size_no_padding) {
2552                 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams;
2553
2554                 /*
2555                  * This handles cases where someone might build with
2556                  * -fpack-struct or any other toolchain that wouldn't produce
2557                  * padding to align `value`.
2558                  */
2559                 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2560
2561                 header_len = sizeof(packed_rotate_streams);
2562                 memcpy(&packed_rotate_streams, payload->data, header_len);
2563
2564                 /* Unpack the packed structure to the natively-packed version. */
2565                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2566                         .is_set = !!packed_rotate_streams.new_chunk_id.is_set,
2567                         .value = be64toh(packed_rotate_streams.new_chunk_id.value),
2568                 };
2569                 _rotate_streams->stream_count = be32toh(packed_rotate_streams.stream_count);
2570         } else if (payload->size == expected_payload_size_3_bytes_padding) {
2571                 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams;
2572
2573                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2574
2575                 header_len = sizeof(padded_rotate_streams);
2576                 memcpy(&padded_rotate_streams, payload->data, header_len);
2577
2578                 /* Unpack the 3-byte padded structure to the natively-packed version. */
2579                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2580                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2581                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2582                 };
2583                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2584         } else if (payload->size == expected_payload_size_7_bytes_padding) {
2585                 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams;
2586
2587                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2588
2589                 header_len = sizeof(padded_rotate_streams);
2590                 memcpy(&padded_rotate_streams, payload->data, header_len);
2591
2592                 /* Unpack the 7-byte padded structure to the natively-packed version. */
2593                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2594                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2595                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2596                 };
2597                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2598
2599                 header_len = sizeof(padded_rotate_streams);
2600         } else {
2601                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2602                     expected_payload_size_no_padding,
2603                     expected_payload_size_3_bytes_padding,
2604                     expected_payload_size_7_bytes_padding,
2605                     payload->size);
2606                 goto error;
2607         }
2608
2609         return header_len;
2610 error:
2611         return -1;
2612 }
2613
2614 /*
2615  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2616  * session rotation feature (not the tracefile rotation feature).
2617  */
2618 static int relay_rotate_session_streams(const struct lttcomm_relayd_hdr *recv_hdr
2619                                         __attribute__((unused)),
2620                                         struct relay_connection *conn,
2621                                         const struct lttng_buffer_view *payload)
2622 {
2623         int ret = 0;
2624         uint32_t i;
2625         ssize_t send_ret;
2626         enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2627         struct relay_session *session = conn->session;
2628         struct lttcomm_relayd_rotate_streams rotate_streams;
2629         struct lttcomm_relayd_generic_reply reply = {};
2630         struct relay_stream *stream = nullptr;
2631         struct lttng_trace_chunk *next_trace_chunk = nullptr;
2632         struct lttng_buffer_view stream_positions;
2633         char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2634         const char *chunk_id_str = "none";
2635         ssize_t header_len;
2636
2637         if (!session || !conn->version_check_done) {
2638                 ERR("Trying to rotate a stream before version check");
2639                 ret = -1;
2640                 goto end_no_reply;
2641         }
2642
2643         if (session->major == 2 && session->minor < 11) {
2644                 ERR("Unsupported feature before 2.11");
2645                 ret = -1;
2646                 goto end_no_reply;
2647         }
2648
2649         header_len = relay_unpack_rotate_streams_header(payload, &rotate_streams);
2650         if (header_len < 0) {
2651                 ret = -1;
2652                 goto end_no_reply;
2653         }
2654
2655         if (rotate_streams.new_chunk_id.is_set) {
2656                 /*
2657                  * Retrieve the trace chunk the stream must transition to. As
2658                  * per the protocol, this chunk should have been created
2659                  * before this command is received.
2660                  */
2661                 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2662                         sessiond_trace_chunk_registry,
2663                         session->sessiond_uuid,
2664                         conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2665                                                             conn->session->id,
2666                         rotate_streams.new_chunk_id.value);
2667                 if (!next_trace_chunk) {
2668                         char uuid_str[LTTNG_UUID_STR_LEN];
2669
2670                         lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2671                         ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2672                             ", trace_chunk_id = %" PRIu64,
2673                             uuid_str,
2674                             session->id,
2675                             rotate_streams.new_chunk_id.value);
2676                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2677                         ret = -1;
2678                         goto end;
2679                 }
2680
2681                 ret = snprintf(chunk_id_buf,
2682                                sizeof(chunk_id_buf),
2683                                "%" PRIu64,
2684                                rotate_streams.new_chunk_id.value);
2685                 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2686                         chunk_id_str = "formatting error";
2687                 } else {
2688                         chunk_id_str = chunk_id_buf;
2689                 }
2690         }
2691
2692         DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2693             rotate_streams.stream_count,
2694             session->session_name,
2695             chunk_id_str);
2696
2697         stream_positions = lttng_buffer_view_from_view(payload, header_len, -1);
2698         if (!stream_positions.data ||
2699             stream_positions.size < (rotate_streams.stream_count *
2700                                      sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2701                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2702                 ret = -1;
2703                 goto end;
2704         }
2705
2706         for (i = 0; i < rotate_streams.stream_count; i++) {
2707                 struct lttcomm_relayd_stream_rotation_position *position_comm =
2708                         &((typeof(position_comm)) stream_positions.data)[i];
2709                 const struct lttcomm_relayd_stream_rotation_position pos = {
2710                         .stream_id = be64toh(position_comm->stream_id),
2711                         .rotate_at_seq_num = be64toh(position_comm->rotate_at_seq_num),
2712                 };
2713
2714                 stream = stream_get_by_id(pos.stream_id);
2715                 if (!stream) {
2716                         reply_code = LTTNG_ERR_INVALID;
2717                         ret = -1;
2718                         goto end;
2719                 }
2720
2721                 pthread_mutex_lock(&stream->lock);
2722                 ret = stream_set_pending_rotation(stream, next_trace_chunk, pos.rotate_at_seq_num);
2723                 pthread_mutex_unlock(&stream->lock);
2724                 if (ret) {
2725                         reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2726                         goto end;
2727                 }
2728
2729                 stream_put(stream);
2730                 stream = nullptr;
2731         }
2732
2733         reply_code = LTTNG_OK;
2734         ret = 0;
2735 end:
2736         if (stream) {
2737                 stream_put(stream);
2738         }
2739
2740         reply.ret_code = htobe32((uint32_t) reply_code);
2741         send_ret = conn->sock->ops->sendmsg(
2742                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2743         if (send_ret < (ssize_t) sizeof(reply)) {
2744                 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)", send_ret);
2745                 ret = -1;
2746         }
2747 end_no_reply:
2748         lttng_trace_chunk_put(next_trace_chunk);
2749         return ret;
2750 }
2751
2752 /*
2753  * relay_create_trace_chunk: create a new trace chunk
2754  */
2755 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2756                                     __attribute__((unused)),
2757                                     struct relay_connection *conn,
2758                                     const struct lttng_buffer_view *payload)
2759 {
2760         int ret = 0;
2761         ssize_t send_ret;
2762         struct relay_session *session = conn->session;
2763         struct lttcomm_relayd_create_trace_chunk *msg;
2764         struct lttcomm_relayd_generic_reply reply = {};
2765         struct lttng_buffer_view header_view;
2766         struct lttng_trace_chunk *chunk = nullptr, *published_chunk = nullptr;
2767         enum lttng_error_code reply_code = LTTNG_OK;
2768         enum lttng_trace_chunk_status chunk_status;
2769         const char *new_path;
2770
2771         if (!session || !conn->version_check_done) {
2772                 ERR("Trying to create a trace chunk before version check");
2773                 ret = -1;
2774                 goto end_no_reply;
2775         }
2776
2777         if (session->major == 2 && session->minor < 11) {
2778                 ERR("Chunk creation command is unsupported before 2.11");
2779                 ret = -1;
2780                 goto end_no_reply;
2781         }
2782
2783         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2784         if (!lttng_buffer_view_is_valid(&header_view)) {
2785                 ERR("Failed to receive payload of chunk creation command");
2786                 ret = -1;
2787                 goto end_no_reply;
2788         }
2789
2790         /* Convert to host endianness. */
2791         msg = (typeof(msg)) header_view.data;
2792         msg->chunk_id = be64toh(msg->chunk_id);
2793         msg->creation_timestamp = be64toh(msg->creation_timestamp);
2794         msg->override_name_length = be32toh(msg->override_name_length);
2795
2796         pthread_mutex_lock(&conn->session->lock);
2797         session->ongoing_rotation = true;
2798         if (session->current_trace_chunk &&
2799             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2800                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2801                                                              DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2802                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2803                         ERR("Failed to rename old chunk");
2804                         ret = -1;
2805                         reply_code = LTTNG_ERR_UNK;
2806                         goto end;
2807                 }
2808         }
2809         if (!session->current_trace_chunk) {
2810                 if (!session->has_rotated) {
2811                         new_path = "";
2812                 } else {
2813                         new_path = nullptr;
2814                 }
2815         } else {
2816                 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2817         }
2818         chunk = lttng_trace_chunk_create(msg->chunk_id, msg->creation_timestamp, new_path);
2819         if (!chunk) {
2820                 ERR("Failed to create trace chunk in trace chunk creation command");
2821                 ret = -1;
2822                 reply_code = LTTNG_ERR_NOMEM;
2823                 goto end;
2824         }
2825         lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2826
2827         if (msg->override_name_length) {
2828                 const char *name;
2829                 const struct lttng_buffer_view chunk_name_view = lttng_buffer_view_from_view(
2830                         payload, sizeof(*msg), msg->override_name_length);
2831
2832                 if (!lttng_buffer_view_is_valid(&chunk_name_view)) {
2833                         ERR("Invalid payload of chunk creation command (protocol error): buffer too short for expected name length");
2834                         ret = -1;
2835                         reply_code = LTTNG_ERR_INVALID;
2836                         goto end;
2837                 }
2838
2839                 name = chunk_name_view.data;
2840                 if (name[msg->override_name_length - 1]) {
2841                         ERR("Invalid payload of chunk creation command (protocol error): name is not null-terminated");
2842                         ret = -1;
2843                         reply_code = LTTNG_ERR_INVALID;
2844                         goto end;
2845                 }
2846
2847                 chunk_status = lttng_trace_chunk_override_name(chunk, chunk_name_view.data);
2848                 switch (chunk_status) {
2849                 case LTTNG_TRACE_CHUNK_STATUS_OK:
2850                         break;
2851                 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2852                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2853                         reply_code = LTTNG_ERR_INVALID;
2854                         ret = -1;
2855                         goto end;
2856                 default:
2857                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2858                         reply_code = LTTNG_ERR_UNK;
2859                         ret = -1;
2860                         goto end;
2861                 }
2862         }
2863
2864         chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2865         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2866                 reply_code = LTTNG_ERR_UNK;
2867                 ret = -1;
2868                 goto end;
2869         }
2870
2871         LTTNG_ASSERT(conn->session->output_directory);
2872         chunk_status = lttng_trace_chunk_set_as_owner(chunk, conn->session->output_directory);
2873         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2874                 reply_code = LTTNG_ERR_UNK;
2875                 ret = -1;
2876                 goto end;
2877         }
2878
2879         published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2880                 sessiond_trace_chunk_registry,
2881                 conn->session->sessiond_uuid,
2882                 conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2883                                                     conn->session->id,
2884                 chunk);
2885         if (!published_chunk) {
2886                 char uuid_str[LTTNG_UUID_STR_LEN];
2887
2888                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2889                 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
2890                     ", chunk_id = %" PRIu64,
2891                     uuid_str,
2892                     conn->session->id,
2893                     msg->chunk_id);
2894                 ret = -1;
2895                 reply_code = LTTNG_ERR_NOMEM;
2896                 goto end;
2897         }
2898
2899         if (conn->session->pending_closure_trace_chunk) {
2900                 /*
2901                  * Invalid; this means a second create_trace_chunk command was
2902                  * received before a close_trace_chunk.
2903                  */
2904                 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2905                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2906                 ret = -1;
2907                 goto end;
2908         }
2909         conn->session->pending_closure_trace_chunk = conn->session->current_trace_chunk;
2910         conn->session->current_trace_chunk = published_chunk;
2911         published_chunk = nullptr;
2912         if (!conn->session->pending_closure_trace_chunk) {
2913                 session->ongoing_rotation = false;
2914         }
2915 end:
2916         pthread_mutex_unlock(&conn->session->lock);
2917         reply.ret_code = htobe32((uint32_t) reply_code);
2918         send_ret = conn->sock->ops->sendmsg(
2919                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2920         if (send_ret < (ssize_t) sizeof(reply)) {
2921                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
2922                 ret = -1;
2923         }
2924 end_no_reply:
2925         lttng_trace_chunk_put(chunk);
2926         lttng_trace_chunk_put(published_chunk);
2927         return ret;
2928 }
2929
2930 /*
2931  * relay_close_trace_chunk: close a trace chunk
2932  */
2933 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2934                                    __attribute__((unused)),
2935                                    struct relay_connection *conn,
2936                                    const struct lttng_buffer_view *payload)
2937 {
2938         int ret = 0, buf_ret;
2939         ssize_t send_ret;
2940         struct relay_session *session = conn->session;
2941         struct lttcomm_relayd_close_trace_chunk *msg;
2942         struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2943         struct lttng_buffer_view header_view;
2944         struct lttng_trace_chunk *chunk = nullptr;
2945         enum lttng_error_code reply_code = LTTNG_OK;
2946         enum lttng_trace_chunk_status chunk_status;
2947         uint64_t chunk_id;
2948         LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2949         time_t close_timestamp;
2950         char closed_trace_chunk_path[LTTNG_PATH_MAX];
2951         size_t path_length = 0;
2952         const char *chunk_name = nullptr;
2953         struct lttng_dynamic_buffer reply_payload;
2954         const char *new_path;
2955
2956         lttng_dynamic_buffer_init(&reply_payload);
2957
2958         if (!session || !conn->version_check_done) {
2959                 ERR("Trying to close a trace chunk before version check");
2960                 ret = -1;
2961                 goto end_no_reply;
2962         }
2963
2964         if (session->major == 2 && session->minor < 11) {
2965                 ERR("Chunk close command is unsupported before 2.11");
2966                 ret = -1;
2967                 goto end_no_reply;
2968         }
2969
2970         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2971         if (!lttng_buffer_view_is_valid(&header_view)) {
2972                 ERR("Failed to receive payload of chunk close command");
2973                 ret = -1;
2974                 goto end_no_reply;
2975         }
2976
2977         /* Convert to host endianness. */
2978         msg = (typeof(msg)) header_view.data;
2979         chunk_id = be64toh(msg->chunk_id);
2980         close_timestamp = (time_t) be64toh(msg->close_timestamp);
2981         close_command.value = (lttng_trace_chunk_command_type) be32toh(msg->close_command.value);
2982         close_command.is_set = msg->close_command.is_set;
2983
2984         chunk = sessiond_trace_chunk_registry_get_chunk(sessiond_trace_chunk_registry,
2985                                                         conn->session->sessiond_uuid,
2986                                                         conn->session->id_sessiond.is_set ?
2987                                                                 conn->session->id_sessiond.value :
2988                                                                 conn->session->id,
2989                                                         chunk_id);
2990         if (!chunk) {
2991                 char uuid_str[LTTNG_UUID_STR_LEN];
2992
2993                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2994                 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
2995                     ", chunk_id = %" PRIu64,
2996                     uuid_str,
2997                     conn->session->id,
2998                     msg->chunk_id);
2999                 ret = -1;
3000                 reply_code = LTTNG_ERR_NOMEM;
3001                 goto end;
3002         }
3003
3004         pthread_mutex_lock(&session->lock);
3005         if (close_command.is_set && close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
3006                 /*
3007                  * Clear command. It is a protocol error to ask for a
3008                  * clear on a relay which does not allow it. Querying
3009                  * the configuration allows figuring out whether
3010                  * clearing is allowed before doing the clear.
3011                  */
3012                 if (!opt_allow_clear) {
3013                         ret = -1;
3014                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3015                         goto end_unlock_session;
3016                 }
3017         }
3018         if (session->pending_closure_trace_chunk && session->pending_closure_trace_chunk != chunk) {
3019                 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3020                     session->session_name);
3021                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3022                 ret = -1;
3023                 goto end_unlock_session;
3024         }
3025
3026         if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
3027             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
3028                 if (close_command.is_set &&
3029                     close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
3030                     !session->has_rotated) {
3031                         /* New chunk stays in session output directory. */
3032                         new_path = "";
3033                 } else {
3034                         /* Use chunk name for new chunk. */
3035                         new_path = nullptr;
3036                 }
3037                 /* Rename new chunk path. */
3038                 chunk_status =
3039                         lttng_trace_chunk_rename_path(session->current_trace_chunk, new_path);
3040                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3041                         ret = -1;
3042                         goto end_unlock_session;
3043                 }
3044                 session->ongoing_rotation = false;
3045         }
3046         if ((!close_command.is_set ||
3047              close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
3048             !lttng_trace_chunk_get_name_overridden(chunk)) {
3049                 const char *old_path;
3050
3051                 if (!session->has_rotated) {
3052                         old_path = "";
3053                 } else {
3054                         old_path = nullptr;
3055                 }
3056                 /* We need to move back the .tmp_old_chunk to its rightful place. */
3057                 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
3058                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3059                         ret = -1;
3060                         goto end_unlock_session;
3061                 }
3062         }
3063         chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, close_timestamp);
3064         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3065                 ERR("Failed to set trace chunk close timestamp");
3066                 ret = -1;
3067                 reply_code = LTTNG_ERR_UNK;
3068                 goto end_unlock_session;
3069         }
3070
3071         if (close_command.is_set) {
3072                 chunk_status = lttng_trace_chunk_set_close_command(chunk, close_command.value);
3073                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3074                         ret = -1;
3075                         reply_code = LTTNG_ERR_INVALID;
3076                         goto end_unlock_session;
3077                 }
3078         }
3079         chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, nullptr);
3080         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3081                 ERR("Failed to get chunk name");
3082                 ret = -1;
3083                 reply_code = LTTNG_ERR_UNK;
3084                 goto end_unlock_session;
3085         }
3086         if (!session->has_rotated && !session->snapshot) {
3087                 ret = lttng_strncpy(closed_trace_chunk_path,
3088                                     session->output_path,
3089                                     sizeof(closed_trace_chunk_path));
3090                 if (ret) {
3091                         ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3092                             strlen(session->output_path),
3093                             sizeof(closed_trace_chunk_path));
3094                         reply_code = LTTNG_ERR_NOMEM;
3095                         ret = -1;
3096                         goto end_unlock_session;
3097                 }
3098         } else {
3099                 if (session->snapshot) {
3100                         ret = snprintf(closed_trace_chunk_path,
3101                                        sizeof(closed_trace_chunk_path),
3102                                        "%s/%s",
3103                                        session->output_path,
3104                                        chunk_name);
3105                 } else {
3106                         ret = snprintf(closed_trace_chunk_path,
3107                                        sizeof(closed_trace_chunk_path),
3108                                        "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY "/%s",
3109                                        session->output_path,
3110                                        chunk_name);
3111                 }
3112                 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3113                         ERR("Failed to format closed trace chunk resulting path");
3114                         reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3115                         ret = -1;
3116                         goto end_unlock_session;
3117                 }
3118         }
3119         if (close_command.is_set &&
3120             close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3121                 session->has_rotated = true;
3122         }
3123         DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3124         path_length = strlen(closed_trace_chunk_path) + 1;
3125         if (path_length > UINT32_MAX) {
3126                 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3127                 ret = -1;
3128                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3129                 goto end_unlock_session;
3130         }
3131
3132         if (session->current_trace_chunk == chunk) {
3133                 /*
3134                  * After a trace chunk close command, no new streams
3135                  * referencing the chunk may be created. Hence, on the
3136                  * event that no new trace chunk have been created for
3137                  * the session, the reference to the current trace chunk
3138                  * is released in order to allow it to be reclaimed when
3139                  * the last stream releases its reference to it.
3140                  */
3141                 lttng_trace_chunk_put(session->current_trace_chunk);
3142                 session->current_trace_chunk = nullptr;
3143         }
3144         lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3145         session->pending_closure_trace_chunk = nullptr;
3146 end_unlock_session:
3147         pthread_mutex_unlock(&session->lock);
3148
3149 end:
3150         reply.generic.ret_code = htobe32((uint32_t) reply_code);
3151         reply.path_length = htobe32((uint32_t) path_length);
3152         buf_ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
3153         if (buf_ret) {
3154                 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3155                 goto end_no_reply;
3156         }
3157
3158         if (reply_code == LTTNG_OK) {
3159                 buf_ret = lttng_dynamic_buffer_append(
3160                         &reply_payload, closed_trace_chunk_path, path_length);
3161                 if (buf_ret) {
3162                         ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3163                         goto end_no_reply;
3164                 }
3165         }
3166
3167         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
3168         if (send_ret < reply_payload.size) {
3169                 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3170                     reply_payload.size,
3171                     send_ret);
3172                 ret = -1;
3173                 goto end_no_reply;
3174         }
3175 end_no_reply:
3176         lttng_trace_chunk_put(chunk);
3177         lttng_dynamic_buffer_reset(&reply_payload);
3178         return ret;
3179 }
3180
3181 /*
3182  * relay_trace_chunk_exists: check if a trace chunk exists
3183  */
3184 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr
3185                                     __attribute__((unused)),
3186                                     struct relay_connection *conn,
3187                                     const struct lttng_buffer_view *payload)
3188 {
3189         int ret = 0;
3190         ssize_t send_ret;
3191         struct relay_session *session = conn->session;
3192         struct lttcomm_relayd_trace_chunk_exists *msg;
3193         struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3194         struct lttng_buffer_view header_view;
3195         uint64_t chunk_id;
3196         bool chunk_exists;
3197
3198         if (!session || !conn->version_check_done) {
3199                 ERR("Trying to check for the presence of a trace chunk before version check");
3200                 ret = -1;
3201                 goto end_no_reply;
3202         }
3203
3204         if (session->major == 2 && session->minor < 11) {
3205                 ERR("Chunk exists command is unsupported before 2.11");
3206                 ret = -1;
3207                 goto end_no_reply;
3208         }
3209
3210         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3211         if (!lttng_buffer_view_is_valid(&header_view)) {
3212                 ERR("Failed to receive payload of chunk exists command");
3213                 ret = -1;
3214                 goto end_no_reply;
3215         }
3216
3217         /* Convert to host endianness. */
3218         msg = (typeof(msg)) header_view.data;
3219         chunk_id = be64toh(msg->chunk_id);
3220
3221         ret = sessiond_trace_chunk_registry_chunk_exists(sessiond_trace_chunk_registry,
3222                                                          conn->session->sessiond_uuid,
3223                                                          conn->session->id,
3224                                                          chunk_id,
3225                                                          &chunk_exists);
3226         /*
3227          * If ret is not 0, send the reply and report the error to the caller.
3228          * It is a protocol (or internal) error and the session/connection
3229          * should be torn down.
3230          */
3231         reply.generic.ret_code =
3232                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3233         reply.trace_chunk_exists = ret == 0 ? chunk_exists : 0;
3234
3235         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3236         if (send_ret < (ssize_t) sizeof(reply)) {
3237                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
3238                 ret = -1;
3239         }
3240 end_no_reply:
3241         return ret;
3242 }
3243
3244 /*
3245  * relay_get_configuration: query whether feature is available
3246  */
3247 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr
3248                                    __attribute__((unused)),
3249                                    struct relay_connection *conn,
3250                                    const struct lttng_buffer_view *payload)
3251 {
3252         int ret = 0;
3253         ssize_t send_ret;
3254         struct lttcomm_relayd_get_configuration *msg;
3255         struct lttcomm_relayd_get_configuration_reply reply = {};
3256         struct lttng_buffer_view header_view;
3257         uint64_t query_flags = 0;
3258         uint64_t result_flags = 0;
3259
3260         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3261         if (!lttng_buffer_view_is_valid(&header_view)) {
3262                 ERR("Failed to receive payload of chunk close command");
3263                 ret = -1;
3264                 goto end_no_reply;
3265         }
3266
3267         /* Convert to host endianness. */
3268         msg = (typeof(msg)) header_view.data;
3269         query_flags = be64toh(msg->query_flags);
3270
3271         if (query_flags) {
3272                 ret = LTTNG_ERR_INVALID_PROTOCOL;
3273                 goto reply;
3274         }
3275         if (opt_allow_clear) {
3276                 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3277         }
3278         ret = 0;
3279 reply:
3280         reply.generic.ret_code =
3281                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3282         reply.relayd_configuration_flags = htobe64(result_flags);
3283
3284         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3285         if (send_ret < (ssize_t) sizeof(reply)) {
3286                 ERR("Failed to send \"get configuration\" command reply (ret = %zd)", send_ret);
3287                 ret = -1;
3288         }
3289 end_no_reply:
3290         return ret;
3291 }
3292
3293 static int relay_process_control_command(struct relay_connection *conn,
3294                                          const struct lttcomm_relayd_hdr *header,
3295                                          const struct lttng_buffer_view *payload)
3296 {
3297         int ret = 0;
3298
3299         DBG3("Processing \"%s\" command for socket %i",
3300              lttcomm_relayd_command_str((lttcomm_relayd_command) header->cmd),
3301              conn->sock->fd);
3302         switch (header->cmd) {
3303         case RELAYD_CREATE_SESSION:
3304                 ret = relay_create_session(header, conn, payload);
3305                 break;
3306         case RELAYD_ADD_STREAM:
3307                 ret = relay_add_stream(header, conn, payload);
3308                 break;
3309         case RELAYD_START_DATA:
3310                 ret = relay_start(header, conn, payload);
3311                 break;
3312         case RELAYD_SEND_METADATA:
3313                 ret = relay_recv_metadata(header, conn, payload);
3314                 break;
3315         case RELAYD_VERSION:
3316                 ret = relay_send_version(header, conn, payload);
3317                 break;
3318         case RELAYD_CLOSE_STREAM:
3319                 ret = relay_close_stream(header, conn, payload);
3320                 break;
3321         case RELAYD_DATA_PENDING:
3322                 ret = relay_data_pending(header, conn, payload);
3323                 break;
3324         case RELAYD_QUIESCENT_CONTROL:
3325                 ret = relay_quiescent_control(header, conn, payload);
3326                 break;
3327         case RELAYD_BEGIN_DATA_PENDING:
3328                 ret = relay_begin_data_pending(header, conn, payload);
3329                 break;
3330         case RELAYD_END_DATA_PENDING:
3331                 ret = relay_end_data_pending(header, conn, payload);
3332                 break;
3333         case RELAYD_SEND_INDEX:
3334                 ret = relay_recv_index(header, conn, payload);
3335                 break;
3336         case RELAYD_STREAMS_SENT:
3337                 ret = relay_streams_sent(header, conn, payload);
3338                 break;
3339         case RELAYD_RESET_METADATA:
3340                 ret = relay_reset_metadata(header, conn, payload);
3341                 break;
3342         case RELAYD_ROTATE_STREAMS:
3343                 ret = relay_rotate_session_streams(header, conn, payload);
3344                 break;
3345         case RELAYD_CREATE_TRACE_CHUNK:
3346                 ret = relay_create_trace_chunk(header, conn, payload);
3347                 break;
3348         case RELAYD_CLOSE_TRACE_CHUNK:
3349                 ret = relay_close_trace_chunk(header, conn, payload);
3350                 break;
3351         case RELAYD_TRACE_CHUNK_EXISTS:
3352                 ret = relay_trace_chunk_exists(header, conn, payload);
3353                 break;
3354         case RELAYD_GET_CONFIGURATION:
3355                 ret = relay_get_configuration(header, conn, payload);
3356                 break;
3357         case RELAYD_UPDATE_SYNC_INFO:
3358         default:
3359                 ERR("Received unknown command (%u)", header->cmd);
3360                 relay_unknown_command(conn);
3361                 ret = -1;
3362                 goto end;
3363         }
3364
3365 end:
3366         return ret;
3367 }
3368
3369 static enum relay_connection_status
3370 relay_process_control_receive_payload(struct relay_connection *conn)
3371 {
3372         int ret = 0;
3373         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3374         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3375         struct ctrl_connection_state_receive_payload *state =
3376                 &conn->protocol.ctrl.state.receive_payload;
3377         struct lttng_buffer_view payload_view;
3378
3379         if (state->left_to_receive == 0) {
3380                 /* Short-circuit for payload-less commands. */
3381                 goto reception_complete;
3382         }
3383
3384         ret = conn->sock->ops->recvmsg(conn->sock,
3385                                        reception_buffer->data + state->received,
3386                                        state->left_to_receive,
3387                                        MSG_DONTWAIT);
3388         if (ret < 0) {
3389                 DIAGNOSTIC_PUSH
3390                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3391                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3392                         DIAGNOSTIC_POP
3393                         PERROR("Unable to receive command payload on sock %d", conn->sock->fd);
3394                         status = RELAY_CONNECTION_STATUS_ERROR;
3395                 }
3396                 goto end;
3397         } else if (ret == 0) {
3398                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3399                 status = RELAY_CONNECTION_STATUS_CLOSED;
3400                 goto end;
3401         }
3402
3403         LTTNG_ASSERT(ret > 0);
3404         LTTNG_ASSERT(ret <= state->left_to_receive);
3405
3406         state->left_to_receive -= ret;
3407         state->received += ret;
3408
3409         if (state->left_to_receive > 0) {
3410                 /*
3411                  * Can't transition to the protocol's next state, wait to
3412                  * receive the rest of the header.
3413                  */
3414                 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
3415                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3416                      state->received,
3417                      state->left_to_receive,
3418                      conn->sock->fd);
3419                 goto end;
3420         }
3421
3422 reception_complete:
3423         DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3424             conn->sock->fd,
3425             state->received);
3426         /*
3427          * The payload required to process the command has been received.
3428          * A view to the reception buffer is forwarded to the various
3429          * commands and the state of the control is reset on success.
3430          *
3431          * Commands are responsible for sending their reply to the peer.
3432          */
3433         payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer, 0, -1);
3434         ret = relay_process_control_command(conn, &state->header, &payload_view);
3435         if (ret < 0) {
3436                 status = RELAY_CONNECTION_STATUS_ERROR;
3437                 goto end;
3438         }
3439
3440         ret = connection_reset_protocol_state(conn);
3441         if (ret) {
3442                 status = RELAY_CONNECTION_STATUS_ERROR;
3443         }
3444 end:
3445         return status;
3446 }
3447
3448 static enum relay_connection_status
3449 relay_process_control_receive_header(struct relay_connection *conn)
3450 {
3451         int ret = 0;
3452         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3453         struct lttcomm_relayd_hdr header;
3454         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3455         struct ctrl_connection_state_receive_header *state =
3456                 &conn->protocol.ctrl.state.receive_header;
3457
3458         LTTNG_ASSERT(state->left_to_receive != 0);
3459
3460         ret = conn->sock->ops->recvmsg(conn->sock,
3461                                        reception_buffer->data + state->received,
3462                                        state->left_to_receive,
3463                                        MSG_DONTWAIT);
3464         if (ret < 0) {
3465                 DIAGNOSTIC_PUSH
3466                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3467                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3468                         DIAGNOSTIC_POP
3469                         PERROR("Unable to receive control command header on sock %d",
3470                                conn->sock->fd);
3471                         status = RELAY_CONNECTION_STATUS_ERROR;
3472                 }
3473                 goto end;
3474         } else if (ret == 0) {
3475                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3476                 status = RELAY_CONNECTION_STATUS_CLOSED;
3477                 goto end;
3478         }
3479
3480         LTTNG_ASSERT(ret > 0);
3481         LTTNG_ASSERT(ret <= state->left_to_receive);
3482
3483         state->left_to_receive -= ret;
3484         state->received += ret;
3485
3486         if (state->left_to_receive > 0) {
3487                 /*
3488                  * Can't transition to the protocol's next state, wait to
3489                  * receive the rest of the header.
3490                  */
3491                 DBG3("Partial reception of control connection protocol header (received %" PRIu64
3492                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3493                      state->received,
3494                      state->left_to_receive,
3495                      conn->sock->fd);
3496                 goto end;
3497         }
3498
3499         /* Transition to next state: receiving the command's payload. */
3500         conn->protocol.ctrl.state_id = CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3501         memcpy(&header, reception_buffer->data, sizeof(header));
3502         header.circuit_id = be64toh(header.circuit_id);
3503         header.data_size = be64toh(header.data_size);
3504         header.cmd = be32toh(header.cmd);
3505         header.cmd_version = be32toh(header.cmd_version);
3506         memcpy(&conn->protocol.ctrl.state.receive_payload.header, &header, sizeof(header));
3507
3508         DBG("Done receiving control command header: fd = %i, cmd = %s, cmd_version = %" PRIu32
3509             ", payload size = %" PRIu64 " bytes",
3510             conn->sock->fd,
3511             lttcomm_relayd_command_str((enum lttcomm_relayd_command) header.cmd),
3512             header.cmd_version,
3513             header.data_size);
3514
3515         if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3516                 ERR("Command header indicates a payload (%" PRIu64
3517                     " bytes) that exceeds the maximal payload size allowed on a control connection.",
3518                     header.data_size);
3519                 status = RELAY_CONNECTION_STATUS_ERROR;
3520                 goto end;
3521         }
3522
3523         conn->protocol.ctrl.state.receive_payload.left_to_receive = header.data_size;
3524         conn->protocol.ctrl.state.receive_payload.received = 0;
3525         ret = lttng_dynamic_buffer_set_size(reception_buffer, header.data_size);
3526         if (ret) {
3527                 status = RELAY_CONNECTION_STATUS_ERROR;
3528                 goto end;
3529         }
3530
3531         if (header.data_size == 0) {
3532                 /*
3533                  * Manually invoke the next state as the poll loop
3534                  * will not wake-up to allow us to proceed further.
3535                  */
3536                 status = relay_process_control_receive_payload(conn);
3537         }
3538 end:
3539         return status;
3540 }
3541
3542 /*
3543  * Process the commands received on the control socket
3544  */
3545 static enum relay_connection_status relay_process_control(struct relay_connection *conn)
3546 {
3547         enum relay_connection_status status;
3548
3549         switch (conn->protocol.ctrl.state_id) {
3550         case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3551                 status = relay_process_control_receive_header(conn);
3552                 break;
3553         case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3554                 status = relay_process_control_receive_payload(conn);
3555                 break;
3556         default:
3557                 ERR("Unknown control connection protocol state encountered.");
3558                 abort();
3559         }
3560
3561         return status;
3562 }
3563
3564 static enum relay_connection_status relay_process_data_receive_header(struct relay_connection *conn)
3565 {
3566         int ret;
3567         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3568         struct data_connection_state_receive_header *state =
3569                 &conn->protocol.data.state.receive_header;
3570         struct lttcomm_relayd_data_hdr header;
3571         struct relay_stream *stream;
3572
3573         LTTNG_ASSERT(state->left_to_receive != 0);
3574
3575         ret = conn->sock->ops->recvmsg(conn->sock,
3576                                        state->header_reception_buffer + state->received,
3577                                        state->left_to_receive,
3578                                        MSG_DONTWAIT);
3579         if (ret < 0) {
3580                 DIAGNOSTIC_PUSH
3581                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3582                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3583                         DIAGNOSTIC_POP
3584                         PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3585                         status = RELAY_CONNECTION_STATUS_ERROR;
3586                 }
3587                 goto end;
3588         } else if (ret == 0) {
3589                 /* Orderly shutdown. Not necessary to print an error. */
3590                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3591                 status = RELAY_CONNECTION_STATUS_CLOSED;
3592                 goto end;
3593         }
3594
3595         LTTNG_ASSERT(ret > 0);
3596         LTTNG_ASSERT(ret <= state->left_to_receive);
3597
3598         state->left_to_receive -= ret;
3599         state->received += ret;
3600
3601         if (state->left_to_receive > 0) {
3602                 /*
3603                  * Can't transition to the protocol's next state, wait to
3604                  * receive the rest of the header.
3605                  */
3606                 DBG3("Partial reception of data connection header (received %" PRIu64
3607                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3608                      state->received,
3609                      state->left_to_receive,
3610                      conn->sock->fd);
3611                 goto end;
3612         }
3613
3614         /* Transition to next state: receiving the payload. */
3615         conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3616
3617         memcpy(&header, state->header_reception_buffer, sizeof(header));
3618         header.circuit_id = be64toh(header.circuit_id);
3619         header.stream_id = be64toh(header.stream_id);
3620         header.data_size = be32toh(header.data_size);
3621         header.net_seq_num = be64toh(header.net_seq_num);
3622         header.padding_size = be32toh(header.padding_size);
3623         memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3624
3625         conn->protocol.data.state.receive_payload.left_to_receive = header.data_size;
3626         conn->protocol.data.state.receive_payload.received = 0;
3627         conn->protocol.data.state.receive_payload.rotate_index = false;
3628
3629         DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
3630             ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64
3631             ", padding_size = %" PRIu32,
3632             conn->sock->fd,
3633             header.circuit_id,
3634             header.stream_id,
3635             header.data_size,
3636             header.net_seq_num,
3637             header.padding_size);
3638
3639         stream = stream_get_by_id(header.stream_id);
3640         if (!stream) {
3641                 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3642                     header.stream_id);
3643                 /* Protocol error. */
3644                 status = RELAY_CONNECTION_STATUS_ERROR;
3645                 goto end;
3646         }
3647
3648         pthread_mutex_lock(&stream->lock);
3649         /* Prepare stream for the reception of a new packet. */
3650         ret = stream_init_packet(
3651                 stream, header.data_size, &conn->protocol.data.state.receive_payload.rotate_index);
3652         pthread_mutex_unlock(&stream->lock);
3653         if (ret) {
3654                 ERR("Failed to rotate stream output file");
3655                 status = RELAY_CONNECTION_STATUS_ERROR;
3656                 goto end_stream_unlock;
3657         }
3658
3659 end_stream_unlock:
3660         stream_put(stream);
3661 end:
3662         return status;
3663 }
3664
3665 static enum relay_connection_status
3666 relay_process_data_receive_payload(struct relay_connection *conn)
3667 {
3668         int ret;
3669         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3670         struct relay_stream *stream;
3671         struct data_connection_state_receive_payload *state =
3672                 &conn->protocol.data.state.receive_payload;
3673         const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3674         char data_buffer[chunk_size];
3675         bool partial_recv = false;
3676         bool new_stream = false, close_requested = false, index_flushed = false;
3677         uint64_t left_to_receive = state->left_to_receive;
3678         struct relay_session *session;
3679
3680         DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64
3681              " bytes received, %" PRIu64 " bytes left to receive",
3682              state->header.stream_id,
3683              state->header.net_seq_num,
3684              state->received,
3685              left_to_receive);
3686
3687         stream = stream_get_by_id(state->header.stream_id);
3688         if (!stream) {
3689                 /* Protocol error. */
3690                 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3691                     state->header.stream_id);
3692                 status = RELAY_CONNECTION_STATUS_ERROR;
3693                 goto end;
3694         }
3695
3696         pthread_mutex_lock(&stream->lock);
3697         session = stream->trace->session;
3698         if (!conn->session) {
3699                 ret = connection_set_session(conn, session);
3700                 if (ret) {
3701                         status = RELAY_CONNECTION_STATUS_ERROR;
3702                         goto end_stream_unlock;
3703                 }
3704         }
3705
3706         /*
3707          * The size of the "chunk" received on any iteration is bounded by:
3708          *   - the data left to receive,
3709          *   - the data immediately available on the socket,
3710          *   - the on-stack data buffer
3711          */
3712         while (left_to_receive > 0 && !partial_recv) {
3713                 size_t recv_size = std::min<uint64_t>(left_to_receive, chunk_size);
3714                 struct lttng_buffer_view packet_chunk;
3715
3716                 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, MSG_DONTWAIT);
3717                 if (ret < 0) {
3718                         DIAGNOSTIC_PUSH
3719                         DIAGNOSTIC_IGNORE_LOGICAL_OP
3720                         if (errno != EAGAIN && errno != EWOULDBLOCK) {
3721                                 DIAGNOSTIC_POP
3722                                 PERROR("Socket %d error", conn->sock->fd);
3723                                 status = RELAY_CONNECTION_STATUS_ERROR;
3724                         }
3725                         goto end_stream_unlock;
3726                 } else if (ret == 0) {
3727                         /* No more data ready to be consumed on socket. */
3728                         DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3729                              state->header.stream_id);
3730                         status = RELAY_CONNECTION_STATUS_CLOSED;
3731                         break;
3732                 } else if (ret < (int) recv_size) {
3733                         /*
3734                          * All the data available on the socket has been
3735                          * consumed.
3736                          */
3737                         partial_recv = true;
3738                         recv_size = ret;
3739                 }
3740
3741                 packet_chunk = lttng_buffer_view_init(data_buffer, 0, recv_size);
3742                 LTTNG_ASSERT(packet_chunk.data);
3743
3744                 ret = stream_write(stream, &packet_chunk, 0);
3745                 if (ret) {
3746                         ERR("Relay error writing data to file");
3747                         status = RELAY_CONNECTION_STATUS_ERROR;
3748                         goto end_stream_unlock;
3749                 }
3750
3751                 left_to_receive -= recv_size;
3752                 state->received += recv_size;
3753                 state->left_to_receive = left_to_receive;
3754         }
3755
3756         if (state->left_to_receive > 0) {
3757                 /*
3758                  * Did not receive all the data expected, wait for more data to
3759                  * become available on the socket.
3760                  */
3761                 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64
3762                      " bytes received, %" PRIu64 " bytes left to receive",
3763                      state->header.stream_id,
3764                      state->received,
3765                      state->left_to_receive);
3766                 goto end_stream_unlock;
3767         }
3768
3769         ret = stream_write(stream, nullptr, state->header.padding_size);
3770         if (ret) {
3771                 status = RELAY_CONNECTION_STATUS_ERROR;
3772                 goto end_stream_unlock;
3773         }
3774
3775         if (session_streams_have_index(session)) {
3776                 ret = stream_update_index(stream,
3777                                           state->header.net_seq_num,
3778                                           state->rotate_index,
3779                                           &index_flushed,
3780                                           state->header.data_size + state->header.padding_size);
3781                 if (ret < 0) {
3782                         ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64
3783                             " ret %d",
3784                             stream->stream_handle,
3785                             state->header.net_seq_num,
3786                             ret);
3787                         status = RELAY_CONNECTION_STATUS_ERROR;
3788                         goto end_stream_unlock;
3789                 }
3790         }
3791
3792         if (stream->prev_data_seq == -1ULL) {
3793                 new_stream = true;
3794         }
3795
3796         ret = stream_complete_packet(stream,
3797                                      state->header.data_size + state->header.padding_size,
3798                                      state->header.net_seq_num,
3799                                      index_flushed);
3800         if (ret) {
3801                 status = RELAY_CONNECTION_STATUS_ERROR;
3802                 goto end_stream_unlock;
3803         }
3804
3805         /*
3806          * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3807          * contents of *state which are aliased (union) to the same location as
3808          * the new state. Don't use it beyond this point.
3809          */
3810         connection_reset_protocol_state(conn);
3811         state = nullptr;
3812
3813 end_stream_unlock:
3814         close_requested = stream->close_requested;
3815         pthread_mutex_unlock(&stream->lock);
3816         if (close_requested && left_to_receive == 0) {
3817                 try_stream_close(stream);
3818         }
3819
3820         if (new_stream) {
3821                 pthread_mutex_lock(&session->lock);
3822                 uatomic_set(&session->new_streams, 1);
3823                 pthread_mutex_unlock(&session->lock);
3824         }
3825
3826         stream_put(stream);
3827 end:
3828         return status;
3829 }
3830
3831 /*
3832  * relay_process_data: Process the data received on the data socket
3833  */
3834 static enum relay_connection_status relay_process_data(struct relay_connection *conn)
3835 {
3836         enum relay_connection_status status;
3837
3838         switch (conn->protocol.data.state_id) {
3839         case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3840                 status = relay_process_data_receive_header(conn);
3841                 break;
3842         case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3843                 status = relay_process_data_receive_payload(conn);
3844                 break;
3845         default:
3846                 ERR("Unexpected data connection communication state.");
3847                 abort();
3848         }
3849
3850         return status;
3851 }
3852
3853 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3854 {
3855         int ret;
3856
3857         (void) lttng_poll_del(events, pollfd);
3858
3859         ret = fd_tracker_close_unsuspendable_fd(
3860                 the_fd_tracker, &pollfd, 1, fd_tracker_util_close_fd, nullptr);
3861         if (ret < 0) {
3862                 ERR("Closing pollfd %d", pollfd);
3863         }
3864 }
3865
3866 static void relay_thread_close_connection(struct lttng_poll_event *events,
3867                                           int pollfd,
3868                                           struct relay_connection *conn)
3869 {
3870         const char *type_str;
3871
3872         switch (conn->type) {
3873         case RELAY_DATA:
3874                 type_str = "Data";
3875                 break;
3876         case RELAY_CONTROL:
3877                 type_str = "Control";
3878                 break;
3879         case RELAY_VIEWER_COMMAND:
3880                 type_str = "Viewer Command";
3881                 break;
3882         case RELAY_VIEWER_NOTIFICATION:
3883                 type_str = "Viewer Notification";
3884                 break;
3885         default:
3886                 type_str = "Unknown";
3887         }
3888         cleanup_connection_pollfd(events, pollfd);
3889         connection_put(conn);
3890         DBG("%s connection closed with %d", type_str, pollfd);
3891 }
3892
3893 /*
3894  * This thread does the actual work
3895  */
3896 static void *relay_thread_worker(void *data __attribute__((unused)))
3897 {
3898         int ret, err = -1, last_seen_data_fd = -1;
3899         uint32_t nb_fd;
3900         struct lttng_poll_event events;
3901         struct lttng_ht *relay_connections_ht;
3902         struct lttng_ht_iter iter;
3903         struct relay_connection *destroy_conn = nullptr;
3904
3905         DBG("[thread] Relay worker started");
3906
3907         rcu_register_thread();
3908
3909         health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3910
3911         if (testpoint(relayd_thread_worker)) {
3912                 goto error_testpoint;
3913         }
3914
3915         health_code_update();
3916
3917         /* table of connections indexed on socket */
3918         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3919         if (!relay_connections_ht) {
3920                 goto relay_connections_ht_error;
3921         }
3922
3923         ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3924         if (ret < 0) {
3925                 goto error_poll_create;
3926         }
3927
3928         ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3929         if (ret < 0) {
3930                 goto error;
3931         }
3932
3933 restart:
3934         while (true) {
3935                 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3936
3937                 health_code_update();
3938
3939                 /* Infinite blocking call, waiting for transmission */
3940                 DBG3("Relayd worker thread polling...");
3941                 health_poll_entry();
3942                 ret = lttng_poll_wait(&events, -1);
3943                 health_poll_exit();
3944                 if (ret < 0) {
3945                         /*
3946                          * Restart interrupted system call.
3947                          */
3948                         if (errno == EINTR) {
3949                                 goto restart;
3950                         }
3951                         goto error;
3952                 }
3953
3954                 nb_fd = ret;
3955
3956                 /*
3957                  * Process control. The control connection is
3958                  * prioritized so we don't starve it with high
3959                  * throughput tracing data on the data connection.
3960                  */
3961                 for (i = 0; i < nb_fd; i++) {
3962                         /* Fetch once the poll data */
3963                         const auto revents = LTTNG_POLL_GETEV(&events, i);
3964                         const auto pollfd = LTTNG_POLL_GETFD(&events, i);
3965
3966                         health_code_update();
3967
3968                         /* Activity on thread quit pipe, exiting. */
3969                         if (relayd_is_thread_quit_pipe(pollfd)) {
3970                                 DBG("Activity on thread quit pipe");
3971                                 err = 0;
3972                                 goto exit;
3973                         }
3974
3975                         /* Inspect the relay conn pipe for new connection */
3976                         if (pollfd == relay_conn_pipe[0]) {
3977                                 if (revents & LPOLLIN) {
3978                                         struct relay_connection *conn;
3979
3980                                         ret = lttng_read(relay_conn_pipe[0],
3981                                                          &conn,
3982                                                          sizeof(conn)); /* NOLINT sizeof used on a
3983                                                                            pointer. */
3984                                         if (ret < 0) {
3985                                                 goto error;
3986                                         }
3987                                         ret = lttng_poll_add(
3988                                                 &events, conn->sock->fd, LPOLLIN | LPOLLRDHUP);
3989                                         if (ret) {
3990                                                 ERR("Failed to add new connection file descriptor to poll set");
3991                                                 goto error;
3992                                         }
3993                                         connection_ht_add(relay_connections_ht, conn);
3994                                         DBG("Connection socket %d added", conn->sock->fd);
3995                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3996                                         ERR("Relay connection pipe error");
3997                                         goto error;
3998                                 } else {
3999                                         ERR("Unexpected poll events %u for sock %d",
4000                                             revents,
4001                                             pollfd);
4002                                         goto error;
4003                                 }
4004                         } else {
4005                                 struct relay_connection *ctrl_conn;
4006
4007                                 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4008                                 /* If not found, there is a synchronization issue. */
4009                                 LTTNG_ASSERT(ctrl_conn);
4010
4011                                 if (ctrl_conn->type == RELAY_DATA) {
4012                                         if (revents & LPOLLIN) {
4013                                                 /*
4014                                                  * Flag the last seen data fd not deleted. It will
4015                                                  * be used as the last seen fd if any fd gets
4016                                                  * deleted in this first loop.
4017                                                  */
4018                                                 last_notdel_data_fd = pollfd;
4019                                         }
4020                                         goto put_ctrl_connection;
4021                                 }
4022                                 LTTNG_ASSERT(ctrl_conn->type == RELAY_CONTROL);
4023
4024                                 if (revents & LPOLLIN) {
4025                                         enum relay_connection_status status;
4026
4027                                         status = relay_process_control(ctrl_conn);
4028                                         if (status != RELAY_CONNECTION_STATUS_OK) {
4029                                                 /*
4030                                                  * On socket error flag the session as aborted to
4031                                                  * force the cleanup of its stream otherwise it can
4032                                                  * leak during the lifetime of the relayd.
4033                                                  *
4034                                                  * This prevents situations in which streams can be
4035                                                  * left opened because an index was received, the
4036                                                  * control connection is closed, and the data
4037                                                  * connection is closed (uncleanly) before the
4038                                                  * packet's data provided.
4039                                                  *
4040                                                  * Since the control connection encountered an
4041                                                  * error, it is okay to be conservative and close
4042                                                  * the session right now as we can't rely on the
4043                                                  * protocol being respected anymore.
4044                                                  */
4045                                                 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4046                                                         session_abort(ctrl_conn->session);
4047                                                 }
4048
4049                                                 /* Clear the connection on error or close. */
4050                                                 relay_thread_close_connection(
4051                                                         &events, pollfd, ctrl_conn);
4052                                         }
4053                                         seen_control = 1;
4054                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4055                                         relay_thread_close_connection(&events, pollfd, ctrl_conn);
4056                                         if (last_seen_data_fd == pollfd) {
4057                                                 last_seen_data_fd = last_notdel_data_fd;
4058                                         }
4059                                 } else {
4060                                         ERR("Unexpected poll events %u for control sock %d",
4061                                             revents,
4062                                             pollfd);
4063                                         connection_put(ctrl_conn);
4064                                         goto error;
4065                                 }
4066                         put_ctrl_connection:
4067                                 connection_put(ctrl_conn);
4068                         }
4069                 }
4070
4071                 /*
4072                  * The last loop handled a control request, go back to poll to make
4073                  * sure we prioritise the control socket.
4074                  */
4075                 if (seen_control) {
4076                         continue;
4077                 }
4078
4079                 if (last_seen_data_fd >= 0) {
4080                         for (i = 0; i < nb_fd; i++) {
4081                                 int pollfd = LTTNG_POLL_GETFD(&events, i);
4082
4083                                 health_code_update();
4084
4085                                 if (last_seen_data_fd == pollfd) {
4086                                         idx = i;
4087                                         break;
4088                                 }
4089                         }
4090                 }
4091
4092                 /* Process data connection. */
4093                 for (i = idx + 1; i < nb_fd; i++) {
4094                         /* Fetch the poll data. */
4095                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
4096                         int pollfd = LTTNG_POLL_GETFD(&events, i);
4097                         struct relay_connection *data_conn;
4098
4099                         health_code_update();
4100
4101                         if (!revents) {
4102                                 /* No activity for this FD (poll implementation). */
4103                                 continue;
4104                         }
4105
4106                         /* Skip the command pipe. It's handled in the first loop. */
4107                         if (pollfd == relay_conn_pipe[0]) {
4108                                 continue;
4109                         }
4110
4111                         data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4112                         if (!data_conn) {
4113                                 /* Skip it. Might be removed before. */
4114                                 continue;
4115                         }
4116                         if (data_conn->type == RELAY_CONTROL) {
4117                                 goto put_data_connection;
4118                         }
4119                         LTTNG_ASSERT(data_conn->type == RELAY_DATA);
4120
4121                         if (revents & LPOLLIN) {
4122                                 enum relay_connection_status status;
4123
4124                                 status = relay_process_data(data_conn);
4125                                 /* Connection closed or error. */
4126                                 if (status != RELAY_CONNECTION_STATUS_OK) {
4127                                         /*
4128                                          * On socket error flag the session as aborted to force
4129                                          * the cleanup of its stream otherwise it can leak
4130                                          * during the lifetime of the relayd.
4131                                          *
4132                                          * This prevents situations in which streams can be
4133                                          * left opened because an index was received, the
4134                                          * control connection is closed, and the data
4135                                          * connection is closed (uncleanly) before the packet's
4136                                          * data provided.
4137                                          *
4138                                          * Since the data connection encountered an error,
4139                                          * it is okay to be conservative and close the
4140                                          * session right now as we can't rely on the protocol
4141                                          * being respected anymore.
4142                                          */
4143                                         if (status == RELAY_CONNECTION_STATUS_ERROR) {
4144                                                 session_abort(data_conn->session);
4145                                         }
4146                                         relay_thread_close_connection(&events, pollfd, data_conn);
4147                                         /*
4148                                          * Every goto restart call sets the last seen fd where
4149                                          * here we don't really care since we gracefully
4150                                          * continue the loop after the connection is deleted.
4151                                          */
4152                                 } else {
4153                                         /* Keep last seen port. */
4154                                         last_seen_data_fd = pollfd;
4155                                         connection_put(data_conn);
4156                                         goto restart;
4157                                 }
4158                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4159                                 relay_thread_close_connection(&events, pollfd, data_conn);
4160                         } else {
4161                                 ERR("Unknown poll events %u for data sock %d", revents, pollfd);
4162                         }
4163                 put_data_connection:
4164                         connection_put(data_conn);
4165                 }
4166                 last_seen_data_fd = -1;
4167         }
4168
4169         /* Normal exit, no error */
4170         ret = 0;
4171
4172 exit:
4173 error:
4174         /* Cleanup remaining connection object. */
4175         {
4176                 lttng::urcu::read_lock_guard read_lock;
4177
4178                 cds_lfht_for_each_entry (
4179                         relay_connections_ht->ht, &iter.iter, destroy_conn, sock_n.node) {
4180                         health_code_update();
4181
4182                         session_abort(destroy_conn->session);
4183
4184                         /*
4185                          * No need to grab another ref, because we own
4186                          * destroy_conn.
4187                          */
4188                         relay_thread_close_connection(
4189                                 &events, destroy_conn->sock->fd, destroy_conn);
4190                 }
4191         }
4192
4193         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4194 error_poll_create:
4195         lttng_ht_destroy(relay_connections_ht);
4196 relay_connections_ht_error:
4197         /* Close relay conn pipes */
4198         (void) fd_tracker_util_pipe_close(the_fd_tracker, relay_conn_pipe);
4199         if (err) {
4200                 DBG("Thread exited with error");
4201         }
4202         DBG("Worker thread cleanup complete");
4203 error_testpoint:
4204         if (err) {
4205                 health_error();
4206                 ERR("Health error occurred in %s", __func__);
4207         }
4208         health_unregister(health_relayd);
4209         rcu_unregister_thread();
4210         lttng_relay_stop_threads();
4211         return nullptr;
4212 }
4213
4214 /*
4215  * Create the relay command pipe to wake thread_manage_apps.
4216  * Closed in cleanup().
4217  */
4218 static int create_relay_conn_pipe()
4219 {
4220         return fd_tracker_util_pipe_open_cloexec(
4221                 the_fd_tracker, "Relayd connection pipe", relay_conn_pipe);
4222 }
4223
4224 static int stdio_open(void *data __attribute__((unused)), int *fds)
4225 {
4226         fds[0] = fileno(stdout);
4227         fds[1] = fileno(stderr);
4228         return 0;
4229 }
4230
4231 static int track_stdio()
4232 {
4233         int fds[2];
4234         const char *names[] = { "stdout", "stderr" };
4235
4236         return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds, names, 2, stdio_open, nullptr);
4237 }
4238
4239 /*
4240  * main
4241  */
4242 int main(int argc, char **argv)
4243 {
4244         bool thread_is_rcu_registered = false;
4245         int ret = 0, retval = 0;
4246         void *status;
4247         char *unlinked_file_directory_path = nullptr, *output_path = nullptr;
4248
4249         /* Parse environment variables */
4250         ret = parse_env_options();
4251         if (ret) {
4252                 retval = -1;
4253                 goto exit_options;
4254         }
4255
4256         /*
4257          * Parse arguments.
4258          * Command line arguments overwrite environment.
4259          */
4260         progname = argv[0];
4261         if (set_options(argc, argv)) {
4262                 retval = -1;
4263                 goto exit_options;
4264         }
4265
4266         if (set_signal_handler()) {
4267                 retval = -1;
4268                 goto exit_options;
4269         }
4270
4271         relayd_config_log();
4272
4273         if (opt_print_version) {
4274                 print_version();
4275                 retval = 0;
4276                 goto exit_options;
4277         }
4278
4279         ret = fclose(stdin);
4280         if (ret) {
4281                 PERROR("Failed to close stdin");
4282                 goto exit_options;
4283         }
4284
4285         DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4286
4287         /* Try to create directory if -o, --output is specified. */
4288         if (opt_output_path) {
4289                 if (*opt_output_path != '/') {
4290                         ERR("Please specify an absolute path for -o, --output PATH");
4291                         retval = -1;
4292                         goto exit_options;
4293                 }
4294
4295                 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG, -1, -1);
4296                 if (ret < 0) {
4297                         ERR("Unable to create %s", opt_output_path);
4298                         retval = -1;
4299                         goto exit_options;
4300                 }
4301         }
4302
4303         /* Daemonize */
4304         if (opt_daemon || opt_background) {
4305                 ret = lttng_daemonize(&child_ppid, &recv_child_signal, !opt_background);
4306                 if (ret < 0) {
4307                         retval = -1;
4308                         goto exit_options;
4309                 }
4310         }
4311
4312         if (opt_working_directory) {
4313                 ret = utils_change_working_directory(opt_working_directory);
4314                 if (ret) {
4315                         /* All errors are already logged. */
4316                         goto exit_options;
4317                 }
4318         }
4319
4320         sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4321         if (!sessiond_trace_chunk_registry) {
4322                 ERR("Failed to initialize session daemon trace chunk registry");
4323                 retval = -1;
4324                 goto exit_options;
4325         }
4326
4327         /*
4328          * The RCU thread registration (and use, through the fd-tracker's
4329          * creation) is done after the daemonization to allow us to not
4330          * deal with liburcu's fork() management as the call RCU needs to
4331          * be restored.
4332          */
4333         rcu_register_thread();
4334         thread_is_rcu_registered = true;
4335
4336         output_path = create_output_path("");
4337         if (!output_path) {
4338                 ERR("Failed to get output path");
4339                 retval = -1;
4340                 goto exit_options;
4341         }
4342         ret = asprintf(&unlinked_file_directory_path,
4343                        "%s/%s",
4344                        output_path,
4345                        DEFAULT_UNLINKED_FILES_DIRECTORY);
4346         free(output_path);
4347         if (ret < 0) {
4348                 ERR("Failed to format unlinked file directory path");
4349                 retval = -1;
4350                 goto exit_options;
4351         }
4352         the_fd_tracker = fd_tracker_create(unlinked_file_directory_path, lttng_opt_fd_pool_size);
4353         free(unlinked_file_directory_path);
4354         if (!the_fd_tracker) {
4355                 retval = -1;
4356                 goto exit_options;
4357         }
4358
4359         ret = track_stdio();
4360         if (ret) {
4361                 retval = -1;
4362                 goto exit_options;
4363         }
4364
4365         /* Initialize thread health monitoring */
4366         health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4367         if (!health_relayd) {
4368                 PERROR("health_app_create error");
4369                 retval = -1;
4370                 goto exit_options;
4371         }
4372
4373         /* Create thread quit pipe */
4374         if (relayd_init_thread_quit_pipe()) {
4375                 retval = -1;
4376                 goto exit_options;
4377         }
4378
4379         /* Setup the thread apps communication pipe. */
4380         if (create_relay_conn_pipe()) {
4381                 retval = -1;
4382                 goto exit_options;
4383         }
4384
4385         /* Init relay command queue. */
4386         cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4387
4388         /* Initialize communication library */
4389         lttcomm_init();
4390         lttcomm_inet_init();
4391
4392         /* tables of sessions indexed by session ID */
4393         sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4394         if (!sessions_ht) {
4395                 retval = -1;
4396                 goto exit_options;
4397         }
4398
4399         /* tables of streams indexed by stream ID */
4400         relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4401         if (!relay_streams_ht) {
4402                 retval = -1;
4403                 goto exit_options;
4404         }
4405
4406         /* tables of streams indexed by stream ID */
4407         viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4408         if (!viewer_streams_ht) {
4409                 retval = -1;
4410                 goto exit_options;
4411         }
4412
4413         ret = init_health_quit_pipe();
4414         if (ret) {
4415                 retval = -1;
4416                 goto exit_options;
4417         }
4418
4419         /* Create thread to manage the client socket */
4420         ret = pthread_create(&health_thread,
4421                              default_pthread_attr(),
4422                              thread_manage_health_relayd,
4423                              (void *) nullptr);
4424         if (ret) {
4425                 errno = ret;
4426                 PERROR("pthread_create health");
4427                 retval = -1;
4428                 goto exit_options;
4429         }
4430
4431         /* Setup the dispatcher thread */
4432         ret = pthread_create(&dispatcher_thread,
4433                              default_pthread_attr(),
4434                              relay_thread_dispatcher,
4435                              (void *) nullptr);
4436         if (ret) {
4437                 errno = ret;
4438                 PERROR("pthread_create dispatcher");
4439                 retval = -1;
4440                 goto exit_dispatcher_thread;
4441         }
4442
4443         /* Setup the worker thread */
4444         ret = pthread_create(&worker_thread, default_pthread_attr(), relay_thread_worker, nullptr);
4445         if (ret) {
4446                 errno = ret;
4447                 PERROR("pthread_create worker");
4448                 retval = -1;
4449                 goto exit_worker_thread;
4450         }
4451
4452         /* Setup the listener thread */
4453         ret = pthread_create(
4454                 &listener_thread, default_pthread_attr(), relay_thread_listener, (void *) nullptr);
4455         if (ret) {
4456                 errno = ret;
4457                 PERROR("pthread_create listener");
4458                 retval = -1;
4459                 goto exit_listener_thread;
4460         }
4461
4462         ret = relayd_live_create(live_uri);
4463         if (ret) {
4464                 ERR("Starting live viewer threads");
4465                 retval = -1;
4466                 goto exit_live;
4467         }
4468
4469         /*
4470          * This is where we start awaiting program completion (e.g. through
4471          * signal that asks threads to teardown).
4472          */
4473
4474         ret = relayd_live_join();
4475         if (ret) {
4476                 retval = -1;
4477         }
4478 exit_live:
4479
4480         ret = pthread_join(listener_thread, &status);
4481         if (ret) {
4482                 errno = ret;
4483                 PERROR("pthread_join listener_thread");
4484                 retval = -1;
4485         }
4486
4487 exit_listener_thread:
4488         ret = pthread_join(worker_thread, &status);
4489         if (ret) {
4490                 errno = ret;
4491                 PERROR("pthread_join worker_thread");
4492                 retval = -1;
4493         }
4494
4495 exit_worker_thread:
4496         ret = pthread_join(dispatcher_thread, &status);
4497         if (ret) {
4498                 errno = ret;
4499                 PERROR("pthread_join dispatcher_thread");
4500                 retval = -1;
4501         }
4502 exit_dispatcher_thread:
4503
4504         ret = pthread_join(health_thread, &status);
4505         if (ret) {
4506                 errno = ret;
4507                 PERROR("pthread_join health_thread");
4508                 retval = -1;
4509         }
4510 exit_options:
4511         /*
4512          * Wait for all pending call_rcu work to complete before tearing
4513          * down data structures. call_rcu worker may be trying to
4514          * perform lookups in those structures.
4515          */
4516         rcu_barrier();
4517         relayd_cleanup();
4518
4519         /* Ensure all prior call_rcu are done. */
4520         rcu_barrier();
4521
4522         if (thread_is_rcu_registered) {
4523                 rcu_unregister_thread();
4524         }
4525
4526         if (!retval) {
4527                 exit(EXIT_SUCCESS);
4528         } else {
4529                 exit(EXIT_FAILURE);
4530         }
4531 }