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(relay_conn_pipe[1], &new_conn, sizeof(new_conn)); /* NOLINT
1325                                                                                                sizeof
1326                                                                                                used
1327                                                                                                on a
1328                                                                                                pointer.
1329                                                                                              */
1330                         if (ret < 0) {
1331                                 PERROR("write connection pipe");
1332                                 connection_put(new_conn);
1333                                 goto error;
1334                         }
1335                 } while (node != nullptr);
1336
1337                 /* Futex wait on queue. Blocking call on futex() */
1338                 health_poll_entry();
1339                 futex_nto1_wait(&relay_conn_queue.futex);
1340                 health_poll_exit();
1341         }
1342
1343         /* Normal exit, no error */
1344         err = 0;
1345
1346 error:
1347 error_testpoint:
1348         if (err) {
1349                 health_error();
1350                 ERR("Health error occurred in %s", __func__);
1351         }
1352         health_unregister(health_relayd);
1353         DBG("Dispatch thread dying");
1354         lttng_relay_stop_threads();
1355         return nullptr;
1356 }
1357
1358 static bool session_streams_have_index(const struct relay_session *session)
1359 {
1360         return session->minor >= 4 && !session->snapshot;
1361 }
1362
1363 /*
1364  * Handle the RELAYD_CREATE_SESSION command.
1365  *
1366  * On success, send back the session id or else return a negative value.
1367  */
1368 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1369                                 struct relay_connection *conn,
1370                                 const struct lttng_buffer_view *payload)
1371 {
1372         int ret = 0;
1373         ssize_t send_ret;
1374         struct relay_session *session = nullptr;
1375         struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1376         char session_name[LTTNG_NAME_MAX] = {};
1377         char hostname[LTTNG_HOST_NAME_MAX] = {};
1378         uint32_t live_timer = 0;
1379         bool snapshot = false;
1380         bool session_name_contains_creation_timestamp = false;
1381         /* Left nil for peers < 2.11. */
1382         char base_path[LTTNG_PATH_MAX] = {};
1383         lttng_uuid sessiond_uuid = {};
1384         LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1385         LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1386         LTTNG_OPTIONAL(time_t) creation_time = {};
1387         struct lttng_dynamic_buffer reply_payload;
1388
1389         lttng_dynamic_buffer_init(&reply_payload);
1390
1391         if (conn->minor < 4) {
1392                 /* From 2.1 to 2.3 */
1393                 ret = 0;
1394         } else if (conn->minor >= 4 && conn->minor < 11) {
1395                 /* From 2.4 to 2.10 */
1396                 ret = cmd_create_session_2_4(
1397                         payload, session_name, hostname, &live_timer, &snapshot);
1398         } else {
1399                 bool has_current_chunk;
1400                 uint64_t current_chunk_id_value;
1401                 time_t creation_time_value;
1402                 uint64_t id_sessiond_value;
1403
1404                 /* From 2.11 to ... */
1405                 ret = cmd_create_session_2_11(payload,
1406                                               session_name,
1407                                               hostname,
1408                                               base_path,
1409                                               &live_timer,
1410                                               &snapshot,
1411                                               &id_sessiond_value,
1412                                               sessiond_uuid,
1413                                               &has_current_chunk,
1414                                               &current_chunk_id_value,
1415                                               &creation_time_value,
1416                                               &session_name_contains_creation_timestamp);
1417                 if (lttng_uuid_is_nil(sessiond_uuid)) {
1418                         /* The nil UUID is reserved for pre-2.11 clients. */
1419                         ERR("Illegal nil UUID announced by peer in create session command");
1420                         ret = -1;
1421                         goto send_reply;
1422                 }
1423                 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1424                 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1425                 if (has_current_chunk) {
1426                         LTTNG_OPTIONAL_SET(&current_chunk_id, current_chunk_id_value);
1427                 }
1428         }
1429
1430         if (ret < 0) {
1431                 goto send_reply;
1432         }
1433
1434         session = session_create(session_name,
1435                                  hostname,
1436                                  base_path,
1437                                  live_timer,
1438                                  snapshot,
1439                                  sessiond_uuid,
1440                                  id_sessiond.is_set ? &id_sessiond.value : nullptr,
1441                                  current_chunk_id.is_set ? &current_chunk_id.value : nullptr,
1442                                  creation_time.is_set ? &creation_time.value : nullptr,
1443                                  conn->major,
1444                                  conn->minor,
1445                                  session_name_contains_creation_timestamp);
1446         if (!session) {
1447                 ret = -1;
1448                 goto send_reply;
1449         }
1450         LTTNG_ASSERT(!conn->session);
1451         conn->session = session;
1452         DBG("Created session %" PRIu64, session->id);
1453
1454         reply.generic.session_id = htobe64(session->id);
1455
1456 send_reply:
1457         if (ret < 0) {
1458                 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1459         } else {
1460                 reply.generic.ret_code = htobe32(LTTNG_OK);
1461         }
1462
1463         if (conn->minor < 11) {
1464                 /* From 2.1 to 2.10 */
1465                 ret = lttng_dynamic_buffer_append(
1466                         &reply_payload, &reply.generic, sizeof(reply.generic));
1467                 if (ret) {
1468                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1469                         ret = -1;
1470                         goto end;
1471                 }
1472         } else {
1473                 const uint32_t output_path_length = session ? strlen(session->output_path) + 1 : 0;
1474
1475                 reply.output_path_length = htobe32(output_path_length);
1476                 ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
1477                 if (ret) {
1478                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1479                         goto end;
1480                 }
1481
1482                 if (output_path_length) {
1483                         ret = lttng_dynamic_buffer_append(
1484                                 &reply_payload, session->output_path, output_path_length);
1485                         if (ret) {
1486                                 ERR("Failed to append \"create session\" command reply path to payload buffer");
1487                                 goto end;
1488                         }
1489                 }
1490         }
1491
1492         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
1493         if (send_ret < (ssize_t) reply_payload.size) {
1494                 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1495                     reply_payload.size,
1496                     send_ret);
1497                 ret = -1;
1498         }
1499 end:
1500         if (ret < 0 && session) {
1501                 session_put(session);
1502         }
1503         lttng_dynamic_buffer_reset(&reply_payload);
1504         return ret;
1505 }
1506
1507 /*
1508  * When we have received all the streams and the metadata for a channel,
1509  * we make them visible to the viewer threads.
1510  */
1511 static void publish_connection_local_streams(struct relay_connection *conn)
1512 {
1513         struct relay_stream *stream;
1514         struct relay_session *session = conn->session;
1515
1516         /*
1517          * We publish all streams belonging to a session atomically wrt
1518          * session lock.
1519          */
1520         pthread_mutex_lock(&session->lock);
1521         lttng::urcu::read_lock_guard read_lock;
1522         cds_list_for_each_entry_rcu(stream, &session->recv_list, recv_node)
1523         {
1524                 stream_publish(stream);
1525         }
1526
1527         /*
1528          * Inform the viewer that there are new streams in the session.
1529          */
1530         if (session->viewer_attached) {
1531                 uatomic_set(&session->new_streams, 1);
1532         }
1533         pthread_mutex_unlock(&session->lock);
1534 }
1535
1536 static int conform_channel_path(char *channel_path)
1537 {
1538         int ret = 0;
1539
1540         if (strstr("../", channel_path)) {
1541                 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1542                     channel_path);
1543                 ret = -1;
1544                 goto end;
1545         }
1546
1547         if (*channel_path == '/') {
1548                 const size_t len = strlen(channel_path);
1549
1550                 /*
1551                  * Channel paths from peers prior to 2.11 are expressed as an
1552                  * absolute path that is, in reality, relative to the relay
1553                  * daemon's output directory. Remove the leading slash so it
1554                  * is correctly interpreted as a relative path later on.
1555                  *
1556                  * len (and not len - 1) is used to copy the trailing NULL.
1557                  */
1558                 bcopy(channel_path + 1, channel_path, len);
1559         }
1560 end:
1561         return ret;
1562 }
1563
1564 /*
1565  * relay_add_stream: allocate a new stream for a session
1566  */
1567 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1568                             struct relay_connection *conn,
1569                             const struct lttng_buffer_view *payload)
1570 {
1571         int ret;
1572         ssize_t send_ret;
1573         struct relay_session *session = conn->session;
1574         struct relay_stream *stream = nullptr;
1575         struct lttcomm_relayd_status_stream reply;
1576         struct ctf_trace *trace = nullptr;
1577         uint64_t stream_handle = -1ULL;
1578         char *path_name = nullptr, *channel_name = nullptr;
1579         uint64_t tracefile_size = 0, tracefile_count = 0;
1580         LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1581
1582         if (!session || !conn->version_check_done) {
1583                 ERR("Trying to add a stream before version check");
1584                 ret = -1;
1585                 goto end_no_session;
1586         }
1587
1588         if (session->minor == 1) {
1589                 /* For 2.1 */
1590                 ret = cmd_recv_stream_2_1(payload, &path_name, &channel_name);
1591         } else if (session->minor > 1 && session->minor < 11) {
1592                 /* From 2.2 to 2.10 */
1593                 ret = cmd_recv_stream_2_2(
1594                         payload, &path_name, &channel_name, &tracefile_size, &tracefile_count);
1595         } else {
1596                 /* From 2.11 to ... */
1597                 ret = cmd_recv_stream_2_11(payload,
1598                                            &path_name,
1599                                            &channel_name,
1600                                            &tracefile_size,
1601                                            &tracefile_count,
1602                                            &stream_chunk_id.value);
1603                 stream_chunk_id.is_set = true;
1604         }
1605
1606         if (ret < 0) {
1607                 goto send_reply;
1608         }
1609
1610         if (conform_channel_path(path_name)) {
1611                 goto send_reply;
1612         }
1613
1614         /*
1615          * Backward compatibility for --group-output-by-session.
1616          * Prior to lttng 2.11, the complete path is passed by the stream.
1617          * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1618          * >=2.11 the chunk is responsible for the output path. When dealing
1619          * with producer < 2.11 the chunk output_path is the root output path
1620          * and the stream carries the complete path (path_name).
1621          * To support --group-output-by-session with older producer (<2.11), we
1622          * need to craft the path based on the stream path.
1623          */
1624         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1625                 if (conn->minor < 4) {
1626                         /*
1627                          * From 2.1 to 2.3, the session_name is not passed on
1628                          * the RELAYD_CREATE_SESSION command. The session name
1629                          * is necessary to detect the presence of a base_path
1630                          * inside the stream path. Without it we cannot perform
1631                          * a valid group-output-by-session transformation.
1632                          */
1633                         WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1634                              " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1635                              session->id,
1636                              path_name);
1637                 } else if (conn->minor >= 4 && conn->minor < 11) {
1638                         char *group_by_session_path_name;
1639
1640                         LTTNG_ASSERT(session->session_name[0] != '\0');
1641
1642                         group_by_session_path_name = backward_compat_group_by_session(
1643                                 path_name, session->session_name, session->creation_time.value);
1644                         if (!group_by_session_path_name) {
1645                                 ERR("Failed to apply group by session to stream of session %" PRIu64,
1646                                     session->id);
1647                                 goto send_reply;
1648                         }
1649
1650                         DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1651                             path_name,
1652                             group_by_session_path_name);
1653
1654                         free(path_name);
1655                         path_name = group_by_session_path_name;
1656                 }
1657         }
1658
1659         trace = ctf_trace_get_by_path_or_create(session, path_name);
1660         if (!trace) {
1661                 goto send_reply;
1662         }
1663
1664         /* This stream here has one reference on the trace. */
1665         pthread_mutex_lock(&last_relay_stream_id_lock);
1666         stream_handle = ++last_relay_stream_id;
1667         pthread_mutex_unlock(&last_relay_stream_id_lock);
1668
1669         /* We pass ownership of path_name and channel_name. */
1670         stream = stream_create(
1671                 trace, stream_handle, path_name, channel_name, tracefile_size, tracefile_count);
1672         path_name = nullptr;
1673         channel_name = nullptr;
1674
1675         /*
1676          * Streams are the owners of their trace. Reference to trace is
1677          * kept within stream_create().
1678          */
1679         ctf_trace_put(trace);
1680
1681 send_reply:
1682         memset(&reply, 0, sizeof(reply));
1683         reply.handle = htobe64(stream_handle);
1684         if (!stream) {
1685                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1686         } else {
1687                 reply.ret_code = htobe32(LTTNG_OK);
1688         }
1689
1690         send_ret = conn->sock->ops->sendmsg(
1691                 conn->sock, &reply, sizeof(struct lttcomm_relayd_status_stream), 0);
1692         if (send_ret < (ssize_t) sizeof(reply)) {
1693                 ERR("Failed to send \"add stream\" command reply (ret = %zd)", send_ret);
1694                 ret = -1;
1695         }
1696
1697 end_no_session:
1698         free(path_name);
1699         free(channel_name);
1700         return ret;
1701 }
1702
1703 /*
1704  * relay_close_stream: close a specific stream
1705  */
1706 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1707                               struct relay_connection *conn,
1708                               const struct lttng_buffer_view *payload)
1709 {
1710         int ret;
1711         ssize_t send_ret;
1712         struct relay_session *session = conn->session;
1713         struct lttcomm_relayd_close_stream stream_info;
1714         struct lttcomm_relayd_generic_reply reply;
1715         struct relay_stream *stream;
1716
1717         DBG("Close stream received");
1718
1719         if (!session || !conn->version_check_done) {
1720                 ERR("Trying to close a stream before version check");
1721                 ret = -1;
1722                 goto end_no_session;
1723         }
1724
1725         if (payload->size < sizeof(stream_info)) {
1726                 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1727                     sizeof(stream_info),
1728                     payload->size);
1729                 ret = -1;
1730                 goto end_no_session;
1731         }
1732         memcpy(&stream_info, payload->data, sizeof(stream_info));
1733         stream_info.stream_id = be64toh(stream_info.stream_id);
1734         stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1735
1736         stream = stream_get_by_id(stream_info.stream_id);
1737         if (!stream) {
1738                 ret = -1;
1739                 goto end;
1740         }
1741
1742         /*
1743          * Set last_net_seq_num before the close flag. Required by data
1744          * pending check.
1745          */
1746         pthread_mutex_lock(&stream->lock);
1747         stream->last_net_seq_num = stream_info.last_net_seq_num;
1748         pthread_mutex_unlock(&stream->lock);
1749
1750         /*
1751          * This is one of the conditions which may trigger a stream close
1752          * with the others being:
1753          *     1) A close command is received for a stream
1754          *     2) The control connection owning the stream is closed
1755          *     3) We have received all of the stream's data _after_ a close
1756          *        request.
1757          */
1758         try_stream_close(stream);
1759         stream_put(stream);
1760         ret = 0;
1761
1762 end:
1763         memset(&reply, 0, sizeof(reply));
1764         if (ret < 0) {
1765                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1766         } else {
1767                 reply.ret_code = htobe32(LTTNG_OK);
1768         }
1769         send_ret = conn->sock->ops->sendmsg(
1770                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1771         if (send_ret < (ssize_t) sizeof(reply)) {
1772                 ERR("Failed to send \"close stream\" command reply (ret = %zd)", send_ret);
1773                 ret = -1;
1774         }
1775
1776 end_no_session:
1777         return ret;
1778 }
1779
1780 /*
1781  * relay_reset_metadata: reset a metadata stream
1782  */
1783 static int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1784                                 struct relay_connection *conn,
1785                                 const struct lttng_buffer_view *payload)
1786 {
1787         int ret;
1788         ssize_t send_ret;
1789         struct relay_session *session = conn->session;
1790         struct lttcomm_relayd_reset_metadata stream_info;
1791         struct lttcomm_relayd_generic_reply reply;
1792         struct relay_stream *stream;
1793
1794         DBG("Reset metadata received");
1795
1796         if (!session || !conn->version_check_done) {
1797                 ERR("Trying to reset a metadata stream before version check");
1798                 ret = -1;
1799                 goto end_no_session;
1800         }
1801
1802         if (payload->size < sizeof(stream_info)) {
1803                 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1804                     sizeof(stream_info),
1805                     payload->size);
1806                 ret = -1;
1807                 goto end_no_session;
1808         }
1809         memcpy(&stream_info, payload->data, sizeof(stream_info));
1810         stream_info.stream_id = be64toh(stream_info.stream_id);
1811         stream_info.version = be64toh(stream_info.version);
1812
1813         DBG("Update metadata to version %" PRIu64, stream_info.version);
1814
1815         /* Unsupported for live sessions for now. */
1816         if (session->live_timer != 0) {
1817                 ret = -1;
1818                 goto end;
1819         }
1820
1821         stream = stream_get_by_id(stream_info.stream_id);
1822         if (!stream) {
1823                 ret = -1;
1824                 goto end;
1825         }
1826         pthread_mutex_lock(&stream->lock);
1827         if (!stream->is_metadata) {
1828                 ret = -1;
1829                 goto end_unlock;
1830         }
1831
1832         ret = stream_reset_file(stream);
1833         if (ret < 0) {
1834                 ERR("Failed to reset metadata stream %" PRIu64 ": stream_path = %s, channel = %s",
1835                     stream->stream_handle,
1836                     stream->path_name,
1837                     stream->channel_name);
1838                 goto end_unlock;
1839         }
1840 end_unlock:
1841         pthread_mutex_unlock(&stream->lock);
1842         stream_put(stream);
1843
1844 end:
1845         memset(&reply, 0, sizeof(reply));
1846         if (ret < 0) {
1847                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1848         } else {
1849                 reply.ret_code = htobe32(LTTNG_OK);
1850         }
1851         send_ret = conn->sock->ops->sendmsg(
1852                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1853         if (send_ret < (ssize_t) sizeof(reply)) {
1854                 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)", send_ret);
1855                 ret = -1;
1856         }
1857
1858 end_no_session:
1859         return ret;
1860 }
1861
1862 /*
1863  * relay_unknown_command: send -1 if received unknown command
1864  */
1865 static void relay_unknown_command(struct relay_connection *conn)
1866 {
1867         struct lttcomm_relayd_generic_reply reply;
1868         ssize_t send_ret;
1869
1870         memset(&reply, 0, sizeof(reply));
1871         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1872         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1873         if (send_ret < sizeof(reply)) {
1874                 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1875         }
1876 }
1877
1878 /*
1879  * relay_start: send an acknowledgment to the client to tell if we are
1880  * ready to receive data. We are ready if a session is established.
1881  */
1882 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1883                        struct relay_connection *conn,
1884                        const struct lttng_buffer_view *payload __attribute__((unused)))
1885 {
1886         int ret = 0;
1887         ssize_t send_ret;
1888         struct lttcomm_relayd_generic_reply reply;
1889         struct relay_session *session = conn->session;
1890
1891         if (!session) {
1892                 DBG("Trying to start the streaming without a session established");
1893                 ret = htobe32(LTTNG_ERR_UNK);
1894         }
1895
1896         memset(&reply, 0, sizeof(reply));
1897         reply.ret_code = htobe32(LTTNG_OK);
1898         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1899         if (send_ret < (ssize_t) sizeof(reply)) {
1900                 ERR("Failed to send \"relay_start\" command reply (ret = %zd)", send_ret);
1901                 ret = -1;
1902         }
1903
1904         return ret;
1905 }
1906
1907 /*
1908  * relay_recv_metadata: receive the metadata for the session.
1909  */
1910 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1911                                struct relay_connection *conn,
1912                                const struct lttng_buffer_view *payload)
1913 {
1914         int ret = 0;
1915         struct relay_session *session = conn->session;
1916         struct lttcomm_relayd_metadata_payload metadata_payload_header;
1917         struct relay_stream *metadata_stream;
1918         uint64_t metadata_payload_size;
1919         struct lttng_buffer_view packet_view;
1920
1921         if (!session) {
1922                 ERR("Metadata sent before version check");
1923                 ret = -1;
1924                 goto end;
1925         }
1926
1927         if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1928                 ERR("Incorrect data size");
1929                 ret = -1;
1930                 goto end;
1931         }
1932         metadata_payload_size =
1933                 recv_hdr->data_size - sizeof(struct lttcomm_relayd_metadata_payload);
1934
1935         memcpy(&metadata_payload_header, payload->data, sizeof(metadata_payload_header));
1936         metadata_payload_header.stream_id = be64toh(metadata_payload_header.stream_id);
1937         metadata_payload_header.padding_size = be32toh(metadata_payload_header.padding_size);
1938
1939         metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1940         if (!metadata_stream) {
1941                 ret = -1;
1942                 goto end;
1943         }
1944
1945         packet_view = lttng_buffer_view_from_view(
1946                 payload, sizeof(metadata_payload_header), metadata_payload_size);
1947         if (!lttng_buffer_view_is_valid(&packet_view)) {
1948                 ERR("Invalid metadata packet length announced by header");
1949                 ret = -1;
1950                 goto end_put;
1951         }
1952
1953         pthread_mutex_lock(&metadata_stream->lock);
1954         ret = stream_write(metadata_stream, &packet_view, metadata_payload_header.padding_size);
1955         pthread_mutex_unlock(&metadata_stream->lock);
1956         if (ret) {
1957                 ret = -1;
1958                 goto end_put;
1959         }
1960 end_put:
1961         stream_put(metadata_stream);
1962 end:
1963         return ret;
1964 }
1965
1966 /*
1967  * relay_send_version: send relayd version number
1968  */
1969 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1970                               struct relay_connection *conn,
1971                               const struct lttng_buffer_view *payload)
1972 {
1973         int ret;
1974         ssize_t send_ret;
1975         struct lttcomm_relayd_version reply, msg;
1976         bool compatible = true;
1977
1978         conn->version_check_done = true;
1979
1980         /* Get version from the other side. */
1981         if (payload->size < sizeof(msg)) {
1982                 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1983                     sizeof(msg),
1984                     payload->size);
1985                 ret = -1;
1986                 goto end;
1987         }
1988
1989         memcpy(&msg, payload->data, sizeof(msg));
1990         msg.major = be32toh(msg.major);
1991         msg.minor = be32toh(msg.minor);
1992
1993         memset(&reply, 0, sizeof(reply));
1994         reply.major = RELAYD_VERSION_COMM_MAJOR;
1995         reply.minor = RELAYD_VERSION_COMM_MINOR;
1996
1997         /* Major versions must be the same */
1998         if (reply.major != msg.major) {
1999                 DBG("Incompatible major versions (%u vs %u), deleting session",
2000                     reply.major,
2001                     msg.major);
2002                 compatible = false;
2003         }
2004
2005         conn->major = reply.major;
2006         /* We adapt to the lowest compatible version */
2007         if (reply.minor <= msg.minor) {
2008                 conn->minor = reply.minor;
2009         } else {
2010                 conn->minor = msg.minor;
2011         }
2012
2013         reply.major = htobe32(reply.major);
2014         reply.minor = htobe32(reply.minor);
2015         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2016         if (send_ret < (ssize_t) sizeof(reply)) {
2017                 ERR("Failed to send \"send version\" command reply (ret = %zd)", send_ret);
2018                 ret = -1;
2019                 goto end;
2020         } else {
2021                 ret = 0;
2022         }
2023
2024         if (!compatible) {
2025                 ret = -1;
2026                 goto end;
2027         }
2028
2029         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
2030
2031 end:
2032         return ret;
2033 }
2034
2035 /*
2036  * Check for data pending for a given stream id from the session daemon.
2037  */
2038 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2039                               struct relay_connection *conn,
2040                               const struct lttng_buffer_view *payload)
2041 {
2042         struct relay_session *session = conn->session;
2043         struct lttcomm_relayd_data_pending msg;
2044         struct lttcomm_relayd_generic_reply reply;
2045         struct relay_stream *stream;
2046         ssize_t send_ret;
2047         int ret;
2048         uint64_t stream_seq;
2049
2050         DBG("Data pending command received");
2051
2052         if (!session || !conn->version_check_done) {
2053                 ERR("Trying to check for data before version check");
2054                 ret = -1;
2055                 goto end_no_session;
2056         }
2057
2058         if (payload->size < sizeof(msg)) {
2059                 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2060                     sizeof(msg),
2061                     payload->size);
2062                 ret = -1;
2063                 goto end_no_session;
2064         }
2065         memcpy(&msg, payload->data, sizeof(msg));
2066         msg.stream_id = be64toh(msg.stream_id);
2067         msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2068
2069         stream = stream_get_by_id(msg.stream_id);
2070         if (stream == nullptr) {
2071                 ret = -1;
2072                 goto end;
2073         }
2074
2075         pthread_mutex_lock(&stream->lock);
2076
2077         if (session_streams_have_index(session)) {
2078                 /*
2079                  * Ensure that both the index and stream data have been
2080                  * flushed up to the requested point.
2081                  */
2082                 stream_seq = std::min(stream->prev_data_seq, stream->prev_index_seq);
2083         } else {
2084                 stream_seq = stream->prev_data_seq;
2085         }
2086         DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2087             ", prev_index_seq %" PRIu64 ", and last_seq %" PRIu64,
2088             msg.stream_id,
2089             stream->prev_data_seq,
2090             stream->prev_index_seq,
2091             msg.last_net_seq_num);
2092
2093         /* Avoid wrapping issue */
2094         if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2095                 /* Data has in fact been written and is NOT pending */
2096                 ret = 0;
2097         } else {
2098                 /* Data still being streamed thus pending */
2099                 ret = 1;
2100         }
2101
2102         stream->data_pending_check_done = true;
2103         pthread_mutex_unlock(&stream->lock);
2104
2105         stream_put(stream);
2106 end:
2107
2108         memset(&reply, 0, sizeof(reply));
2109         reply.ret_code = htobe32(ret);
2110         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2111         if (send_ret < (ssize_t) sizeof(reply)) {
2112                 ERR("Failed to send \"data pending\" command reply (ret = %zd)", send_ret);
2113                 ret = -1;
2114         }
2115
2116 end_no_session:
2117         return ret;
2118 }
2119
2120 /*
2121  * Wait for the control socket to reach a quiescent state.
2122  *
2123  * Note that for now, when receiving this command from the session
2124  * daemon, this means that every subsequent commands or data received on
2125  * the control socket has been handled. So, this is why we simply return
2126  * OK here.
2127  */
2128 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr
2129                                    __attribute__((unused)),
2130                                    struct relay_connection *conn,
2131                                    const struct lttng_buffer_view *payload)
2132 {
2133         int ret;
2134         ssize_t send_ret;
2135         struct relay_stream *stream;
2136         struct lttcomm_relayd_quiescent_control msg;
2137         struct lttcomm_relayd_generic_reply reply;
2138
2139         DBG("Checking quiescent state on control socket");
2140
2141         if (!conn->session || !conn->version_check_done) {
2142                 ERR("Trying to check for data before version check");
2143                 ret = -1;
2144                 goto end_no_session;
2145         }
2146
2147         if (payload->size < sizeof(msg)) {
2148                 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2149                     sizeof(msg),
2150                     payload->size);
2151                 ret = -1;
2152                 goto end_no_session;
2153         }
2154         memcpy(&msg, payload->data, sizeof(msg));
2155         msg.stream_id = be64toh(msg.stream_id);
2156
2157         stream = stream_get_by_id(msg.stream_id);
2158         if (!stream) {
2159                 goto reply;
2160         }
2161         pthread_mutex_lock(&stream->lock);
2162         stream->data_pending_check_done = true;
2163         pthread_mutex_unlock(&stream->lock);
2164
2165         DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2166         stream_put(stream);
2167 reply:
2168         memset(&reply, 0, sizeof(reply));
2169         reply.ret_code = htobe32(LTTNG_OK);
2170         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2171         if (send_ret < (ssize_t) sizeof(reply)) {
2172                 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)", send_ret);
2173                 ret = -1;
2174         } else {
2175                 ret = 0;
2176         }
2177
2178 end_no_session:
2179         return ret;
2180 }
2181
2182 /*
2183  * Initialize a data pending command. This means that a consumer is about
2184  * to ask for data pending for each stream it holds. Simply iterate over
2185  * all streams of a session and set the data_pending_check_done flag.
2186  *
2187  * This command returns to the client a LTTNG_OK code.
2188  */
2189 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2190                                     struct relay_connection *conn,
2191                                     const struct lttng_buffer_view *payload)
2192 {
2193         int ret;
2194         ssize_t send_ret;
2195         struct lttng_ht_iter iter;
2196         struct lttcomm_relayd_begin_data_pending msg;
2197         struct lttcomm_relayd_generic_reply reply;
2198         struct relay_stream *stream;
2199
2200         LTTNG_ASSERT(recv_hdr);
2201         LTTNG_ASSERT(conn);
2202
2203         DBG("Init streams for data pending");
2204
2205         if (!conn->session || !conn->version_check_done) {
2206                 ERR("Trying to check for data before version check");
2207                 ret = -1;
2208                 goto end_no_session;
2209         }
2210
2211         if (payload->size < sizeof(msg)) {
2212                 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2213                     sizeof(msg),
2214                     payload->size);
2215                 ret = -1;
2216                 goto end_no_session;
2217         }
2218         memcpy(&msg, payload->data, sizeof(msg));
2219         msg.session_id = be64toh(msg.session_id);
2220
2221         /*
2222          * Iterate over all streams to set the begin data pending flag.
2223          * For now, the streams are indexed by stream handle so we have
2224          * to iterate over all streams to find the one associated with
2225          * the right session_id.
2226          */
2227         {
2228                 lttng::urcu::read_lock_guard read_lock;
2229
2230                 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2231                         if (!stream_get(stream)) {
2232                                 continue;
2233                         }
2234
2235                         if (stream->trace->session->id == msg.session_id) {
2236                                 pthread_mutex_lock(&stream->lock);
2237                                 stream->data_pending_check_done = false;
2238                                 pthread_mutex_unlock(&stream->lock);
2239                                 DBG("Set begin data pending flag to stream %" PRIu64,
2240                                     stream->stream_handle);
2241                         }
2242
2243                         stream_put(stream);
2244                 }
2245         }
2246
2247         memset(&reply, 0, sizeof(reply));
2248         /* All good, send back reply. */
2249         reply.ret_code = htobe32(LTTNG_OK);
2250
2251         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2252         if (send_ret < (ssize_t) sizeof(reply)) {
2253                 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)", send_ret);
2254                 ret = -1;
2255         } else {
2256                 ret = 0;
2257         }
2258
2259 end_no_session:
2260         return ret;
2261 }
2262
2263 /*
2264  * End data pending command. This will check, for a given session id, if
2265  * each stream associated with it has its data_pending_check_done flag
2266  * set. If not, this means that the client lost track of the stream but
2267  * the data is still being streamed on our side. In this case, we inform
2268  * the client that data is in flight.
2269  *
2270  * Return to the client if there is data in flight or not with a ret_code.
2271  */
2272 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2273                                   struct relay_connection *conn,
2274                                   const struct lttng_buffer_view *payload)
2275 {
2276         int ret;
2277         ssize_t send_ret;
2278         struct lttng_ht_iter iter;
2279         struct lttcomm_relayd_end_data_pending msg;
2280         struct lttcomm_relayd_generic_reply reply;
2281         struct relay_stream *stream;
2282         uint32_t is_data_inflight = 0;
2283
2284         DBG("End data pending command");
2285
2286         if (!conn->session || !conn->version_check_done) {
2287                 ERR("Trying to check for data before version check");
2288                 ret = -1;
2289                 goto end_no_session;
2290         }
2291
2292         if (payload->size < sizeof(msg)) {
2293                 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2294                     sizeof(msg),
2295                     payload->size);
2296                 ret = -1;
2297                 goto end_no_session;
2298         }
2299         memcpy(&msg, payload->data, sizeof(msg));
2300         msg.session_id = be64toh(msg.session_id);
2301
2302         /*
2303          * Iterate over all streams to see if the begin data pending
2304          * flag is set.
2305          */
2306         {
2307                 lttng::urcu::read_lock_guard read_lock;
2308
2309                 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2310                         if (!stream_get(stream)) {
2311                                 continue;
2312                         }
2313
2314                         if (stream->trace->session->id != msg.session_id) {
2315                                 stream_put(stream);
2316                                 continue;
2317                         }
2318
2319                         pthread_mutex_lock(&stream->lock);
2320                         if (!stream->data_pending_check_done) {
2321                                 uint64_t stream_seq;
2322
2323                                 if (session_streams_have_index(conn->session)) {
2324                                         /*
2325                                          * Ensure that both the index and stream data have been
2326                                          * flushed up to the requested point.
2327                                          */
2328                                         stream_seq = std::min(stream->prev_data_seq,
2329                                                               stream->prev_index_seq);
2330                                 } else {
2331                                         stream_seq = stream->prev_data_seq;
2332                                 }
2333
2334                                 if (!stream->closed ||
2335                                     !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2336                                         is_data_inflight = 1;
2337                                         DBG("Data is still in flight for stream %" PRIu64,
2338                                             stream->stream_handle);
2339                                         pthread_mutex_unlock(&stream->lock);
2340                                         stream_put(stream);
2341                                         break;
2342                                 }
2343                         }
2344
2345                         pthread_mutex_unlock(&stream->lock);
2346                         stream_put(stream);
2347                 }
2348         }
2349
2350         memset(&reply, 0, sizeof(reply));
2351         /* All good, send back reply. */
2352         reply.ret_code = htobe32(is_data_inflight);
2353
2354         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2355         if (send_ret < (ssize_t) sizeof(reply)) {
2356                 ERR("Failed to send \"end data pending\" command reply (ret = %zd)", send_ret);
2357                 ret = -1;
2358         } else {
2359                 ret = 0;
2360         }
2361
2362 end_no_session:
2363         return ret;
2364 }
2365
2366 /*
2367  * Receive an index for a specific stream.
2368  *
2369  * Return 0 on success else a negative value.
2370  */
2371 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2372                             struct relay_connection *conn,
2373                             const struct lttng_buffer_view *payload)
2374 {
2375         int ret;
2376         ssize_t send_ret;
2377         struct relay_session *session = conn->session;
2378         struct lttcomm_relayd_index index_info;
2379         struct lttcomm_relayd_generic_reply reply;
2380         struct relay_stream *stream;
2381         size_t msg_len;
2382
2383         LTTNG_ASSERT(conn);
2384
2385         DBG("Relay receiving index");
2386
2387         if (!session || !conn->version_check_done) {
2388                 ERR("Trying to close a stream before version check");
2389                 ret = -1;
2390                 goto end_no_session;
2391         }
2392
2393         msg_len = lttcomm_relayd_index_len(lttng_to_index_major(conn->major, conn->minor),
2394                                            lttng_to_index_minor(conn->major, conn->minor));
2395         if (payload->size < msg_len) {
2396                 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2397                     msg_len,
2398                     payload->size);
2399                 ret = -1;
2400                 goto end_no_session;
2401         }
2402         memcpy(&index_info, payload->data, msg_len);
2403         index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2404         index_info.net_seq_num = be64toh(index_info.net_seq_num);
2405         index_info.packet_size = be64toh(index_info.packet_size);
2406         index_info.content_size = be64toh(index_info.content_size);
2407         index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2408         index_info.timestamp_end = be64toh(index_info.timestamp_end);
2409         index_info.events_discarded = be64toh(index_info.events_discarded);
2410         index_info.stream_id = be64toh(index_info.stream_id);
2411
2412         if (conn->minor >= 8) {
2413                 index_info.stream_instance_id = be64toh(index_info.stream_instance_id);
2414                 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2415         } else {
2416                 index_info.stream_instance_id = -1ULL;
2417                 index_info.packet_seq_num = -1ULL;
2418         }
2419
2420         stream = stream_get_by_id(index_info.relay_stream_id);
2421         if (!stream) {
2422                 ERR("stream_get_by_id not found");
2423                 ret = -1;
2424                 goto end;
2425         }
2426
2427         pthread_mutex_lock(&stream->lock);
2428         ret = stream_add_index(stream, &index_info);
2429         pthread_mutex_unlock(&stream->lock);
2430         if (ret) {
2431                 goto end_stream_put;
2432         }
2433
2434 end_stream_put:
2435         stream_put(stream);
2436 end:
2437         memset(&reply, 0, sizeof(reply));
2438         if (ret < 0) {
2439                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2440         } else {
2441                 reply.ret_code = htobe32(LTTNG_OK);
2442         }
2443         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2444         if (send_ret < (ssize_t) sizeof(reply)) {
2445                 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2446                 ret = -1;
2447         }
2448
2449 end_no_session:
2450         return ret;
2451 }
2452
2453 /*
2454  * Receive the streams_sent message.
2455  *
2456  * Return 0 on success else a negative value.
2457  */
2458 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2459                               struct relay_connection *conn,
2460                               const struct lttng_buffer_view *payload __attribute__((unused)))
2461 {
2462         int ret;
2463         ssize_t send_ret;
2464         struct lttcomm_relayd_generic_reply reply;
2465
2466         LTTNG_ASSERT(conn);
2467
2468         DBG("Relay receiving streams_sent");
2469
2470         if (!conn->session || !conn->version_check_done) {
2471                 ERR("Trying to close a stream before version check");
2472                 ret = -1;
2473                 goto end_no_session;
2474         }
2475
2476         /*
2477          * Publish every pending stream in the connection recv list which are
2478          * now ready to be used by the viewer.
2479          */
2480         publish_connection_local_streams(conn);
2481
2482         memset(&reply, 0, sizeof(reply));
2483         reply.ret_code = htobe32(LTTNG_OK);
2484         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2485         if (send_ret < (ssize_t) sizeof(reply)) {
2486                 ERR("Failed to send \"streams sent\" command reply (ret = %zd)", send_ret);
2487                 ret = -1;
2488         } else {
2489                 /* Success. */
2490                 ret = 0;
2491         }
2492
2493 end_no_session:
2494         return ret;
2495 }
2496
2497 static ssize_t
2498 relay_unpack_rotate_streams_header(const struct lttng_buffer_view *payload,
2499                                    struct lttcomm_relayd_rotate_streams *_rotate_streams)
2500 {
2501         struct lttcomm_relayd_rotate_streams rotate_streams;
2502         /*
2503          * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2504          * This is the smallest version of this structure, but it can be larger;
2505          * this variable is updated once the proper size of the structure is known.
2506          *
2507          * See comment at the declaration of this structure for more information.
2508          */
2509         ssize_t header_len = sizeof(struct lttcomm_relayd_rotate_streams_packed);
2510         size_t expected_payload_size_no_padding, expected_payload_size_3_bytes_padding,
2511                 expected_payload_size_7_bytes_padding;
2512
2513         if (payload->size < header_len) {
2514                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2515                     header_len,
2516                     payload->size);
2517                 goto error;
2518         }
2519
2520         /*
2521          * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2522          * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2523          *
2524          * We start by "unpacking" `stream_count` to figure out the padding length
2525          * emited by our peer.
2526          */
2527         {
2528                 decltype(rotate_streams.stream_count) stream_count;
2529
2530                 memcpy(&stream_count, payload->data, sizeof(stream_count));
2531                 rotate_streams.stream_count = be32toh(stream_count);
2532         }
2533
2534         rotate_streams.new_chunk_id = LTTNG_OPTIONAL_INIT_UNSET;
2535
2536         /*
2537          * Payload size expected given the possible padding lengths in
2538          * `struct lttcomm_relayd_rotate_streams`.
2539          */
2540         expected_payload_size_no_padding =
2541                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2542                 sizeof(lttcomm_relayd_rotate_streams_packed);
2543         expected_payload_size_3_bytes_padding =
2544                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2545                 sizeof(lttcomm_relayd_rotate_streams_3_bytes_padding);
2546         expected_payload_size_7_bytes_padding =
2547                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2548                 sizeof(lttcomm_relayd_rotate_streams_7_bytes_padding);
2549
2550         if (payload->size == expected_payload_size_no_padding) {
2551                 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams;
2552
2553                 /*
2554                  * This handles cases where someone might build with
2555                  * -fpack-struct or any other toolchain that wouldn't produce
2556                  * padding to align `value`.
2557                  */
2558                 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2559
2560                 header_len = sizeof(packed_rotate_streams);
2561                 memcpy(&packed_rotate_streams, payload->data, header_len);
2562
2563                 /* Unpack the packed structure to the natively-packed version. */
2564                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2565                         .is_set = !!packed_rotate_streams.new_chunk_id.is_set,
2566                         .value = be64toh(packed_rotate_streams.new_chunk_id.value),
2567                 };
2568                 _rotate_streams->stream_count = be32toh(packed_rotate_streams.stream_count);
2569         } else if (payload->size == expected_payload_size_3_bytes_padding) {
2570                 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams;
2571
2572                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2573
2574                 header_len = sizeof(padded_rotate_streams);
2575                 memcpy(&padded_rotate_streams, payload->data, header_len);
2576
2577                 /* Unpack the 3-byte padded structure to the natively-packed version. */
2578                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2579                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2580                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2581                 };
2582                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2583         } else if (payload->size == expected_payload_size_7_bytes_padding) {
2584                 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams;
2585
2586                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2587
2588                 header_len = sizeof(padded_rotate_streams);
2589                 memcpy(&padded_rotate_streams, payload->data, header_len);
2590
2591                 /* Unpack the 7-byte padded structure to the natively-packed version. */
2592                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2593                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2594                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2595                 };
2596                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2597
2598                 header_len = sizeof(padded_rotate_streams);
2599         } else {
2600                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2601                     expected_payload_size_no_padding,
2602                     expected_payload_size_3_bytes_padding,
2603                     expected_payload_size_7_bytes_padding,
2604                     payload->size);
2605                 goto error;
2606         }
2607
2608         return header_len;
2609 error:
2610         return -1;
2611 }
2612
2613 /*
2614  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2615  * session rotation feature (not the tracefile rotation feature).
2616  */
2617 static int relay_rotate_session_streams(const struct lttcomm_relayd_hdr *recv_hdr
2618                                         __attribute__((unused)),
2619                                         struct relay_connection *conn,
2620                                         const struct lttng_buffer_view *payload)
2621 {
2622         int ret = 0;
2623         uint32_t i;
2624         ssize_t send_ret;
2625         enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2626         struct relay_session *session = conn->session;
2627         struct lttcomm_relayd_rotate_streams rotate_streams;
2628         struct lttcomm_relayd_generic_reply reply = {};
2629         struct relay_stream *stream = nullptr;
2630         struct lttng_trace_chunk *next_trace_chunk = nullptr;
2631         struct lttng_buffer_view stream_positions;
2632         char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2633         const char *chunk_id_str = "none";
2634         ssize_t header_len;
2635
2636         if (!session || !conn->version_check_done) {
2637                 ERR("Trying to rotate a stream before version check");
2638                 ret = -1;
2639                 goto end_no_reply;
2640         }
2641
2642         if (session->major == 2 && session->minor < 11) {
2643                 ERR("Unsupported feature before 2.11");
2644                 ret = -1;
2645                 goto end_no_reply;
2646         }
2647
2648         header_len = relay_unpack_rotate_streams_header(payload, &rotate_streams);
2649         if (header_len < 0) {
2650                 ret = -1;
2651                 goto end_no_reply;
2652         }
2653
2654         if (rotate_streams.new_chunk_id.is_set) {
2655                 /*
2656                  * Retrieve the trace chunk the stream must transition to. As
2657                  * per the protocol, this chunk should have been created
2658                  * before this command is received.
2659                  */
2660                 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2661                         sessiond_trace_chunk_registry,
2662                         session->sessiond_uuid,
2663                         conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2664                                                             conn->session->id,
2665                         rotate_streams.new_chunk_id.value);
2666                 if (!next_trace_chunk) {
2667                         char uuid_str[LTTNG_UUID_STR_LEN];
2668
2669                         lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2670                         ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2671                             ", trace_chunk_id = %" PRIu64,
2672                             uuid_str,
2673                             session->id,
2674                             rotate_streams.new_chunk_id.value);
2675                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2676                         ret = -1;
2677                         goto end;
2678                 }
2679
2680                 ret = snprintf(chunk_id_buf,
2681                                sizeof(chunk_id_buf),
2682                                "%" PRIu64,
2683                                rotate_streams.new_chunk_id.value);
2684                 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2685                         chunk_id_str = "formatting error";
2686                 } else {
2687                         chunk_id_str = chunk_id_buf;
2688                 }
2689         }
2690
2691         DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2692             rotate_streams.stream_count,
2693             session->session_name,
2694             chunk_id_str);
2695
2696         stream_positions = lttng_buffer_view_from_view(payload, header_len, -1);
2697         if (!stream_positions.data ||
2698             stream_positions.size < (rotate_streams.stream_count *
2699                                      sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2700                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2701                 ret = -1;
2702                 goto end;
2703         }
2704
2705         for (i = 0; i < rotate_streams.stream_count; i++) {
2706                 struct lttcomm_relayd_stream_rotation_position *position_comm =
2707                         &((typeof(position_comm)) stream_positions.data)[i];
2708                 const struct lttcomm_relayd_stream_rotation_position pos = {
2709                         .stream_id = be64toh(position_comm->stream_id),
2710                         .rotate_at_seq_num = be64toh(position_comm->rotate_at_seq_num),
2711                 };
2712
2713                 stream = stream_get_by_id(pos.stream_id);
2714                 if (!stream) {
2715                         reply_code = LTTNG_ERR_INVALID;
2716                         ret = -1;
2717                         goto end;
2718                 }
2719
2720                 pthread_mutex_lock(&stream->lock);
2721                 ret = stream_set_pending_rotation(stream, next_trace_chunk, pos.rotate_at_seq_num);
2722                 pthread_mutex_unlock(&stream->lock);
2723                 if (ret) {
2724                         reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2725                         goto end;
2726                 }
2727
2728                 stream_put(stream);
2729                 stream = nullptr;
2730         }
2731
2732         reply_code = LTTNG_OK;
2733         ret = 0;
2734 end:
2735         if (stream) {
2736                 stream_put(stream);
2737         }
2738
2739         reply.ret_code = htobe32((uint32_t) reply_code);
2740         send_ret = conn->sock->ops->sendmsg(
2741                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2742         if (send_ret < (ssize_t) sizeof(reply)) {
2743                 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)", send_ret);
2744                 ret = -1;
2745         }
2746 end_no_reply:
2747         lttng_trace_chunk_put(next_trace_chunk);
2748         return ret;
2749 }
2750
2751 /*
2752  * relay_create_trace_chunk: create a new trace chunk
2753  */
2754 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2755                                     __attribute__((unused)),
2756                                     struct relay_connection *conn,
2757                                     const struct lttng_buffer_view *payload)
2758 {
2759         int ret = 0;
2760         ssize_t send_ret;
2761         struct relay_session *session = conn->session;
2762         struct lttcomm_relayd_create_trace_chunk *msg;
2763         struct lttcomm_relayd_generic_reply reply = {};
2764         struct lttng_buffer_view header_view;
2765         struct lttng_trace_chunk *chunk = nullptr, *published_chunk = nullptr;
2766         enum lttng_error_code reply_code = LTTNG_OK;
2767         enum lttng_trace_chunk_status chunk_status;
2768         const char *new_path;
2769
2770         if (!session || !conn->version_check_done) {
2771                 ERR("Trying to create a trace chunk before version check");
2772                 ret = -1;
2773                 goto end_no_reply;
2774         }
2775
2776         if (session->major == 2 && session->minor < 11) {
2777                 ERR("Chunk creation command is unsupported before 2.11");
2778                 ret = -1;
2779                 goto end_no_reply;
2780         }
2781
2782         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2783         if (!lttng_buffer_view_is_valid(&header_view)) {
2784                 ERR("Failed to receive payload of chunk creation command");
2785                 ret = -1;
2786                 goto end_no_reply;
2787         }
2788
2789         /* Convert to host endianness. */
2790         msg = (typeof(msg)) header_view.data;
2791         msg->chunk_id = be64toh(msg->chunk_id);
2792         msg->creation_timestamp = be64toh(msg->creation_timestamp);
2793         msg->override_name_length = be32toh(msg->override_name_length);
2794
2795         pthread_mutex_lock(&conn->session->lock);
2796         session->ongoing_rotation = true;
2797         if (session->current_trace_chunk &&
2798             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2799                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2800                                                              DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2801                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2802                         ERR("Failed to rename old chunk");
2803                         ret = -1;
2804                         reply_code = LTTNG_ERR_UNK;
2805                         goto end;
2806                 }
2807         }
2808         if (!session->current_trace_chunk) {
2809                 if (!session->has_rotated) {
2810                         new_path = "";
2811                 } else {
2812                         new_path = nullptr;
2813                 }
2814         } else {
2815                 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2816         }
2817         chunk = lttng_trace_chunk_create(msg->chunk_id, msg->creation_timestamp, new_path);
2818         if (!chunk) {
2819                 ERR("Failed to create trace chunk in trace chunk creation command");
2820                 ret = -1;
2821                 reply_code = LTTNG_ERR_NOMEM;
2822                 goto end;
2823         }
2824         lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2825
2826         if (msg->override_name_length) {
2827                 const char *name;
2828                 const struct lttng_buffer_view chunk_name_view = lttng_buffer_view_from_view(
2829                         payload, sizeof(*msg), msg->override_name_length);
2830
2831                 if (!lttng_buffer_view_is_valid(&chunk_name_view)) {
2832                         ERR("Invalid payload of chunk creation command (protocol error): buffer too short for expected name length");
2833                         ret = -1;
2834                         reply_code = LTTNG_ERR_INVALID;
2835                         goto end;
2836                 }
2837
2838                 name = chunk_name_view.data;
2839                 if (name[msg->override_name_length - 1]) {
2840                         ERR("Invalid payload of chunk creation command (protocol error): name is not null-terminated");
2841                         ret = -1;
2842                         reply_code = LTTNG_ERR_INVALID;
2843                         goto end;
2844                 }
2845
2846                 chunk_status = lttng_trace_chunk_override_name(chunk, chunk_name_view.data);
2847                 switch (chunk_status) {
2848                 case LTTNG_TRACE_CHUNK_STATUS_OK:
2849                         break;
2850                 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2851                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2852                         reply_code = LTTNG_ERR_INVALID;
2853                         ret = -1;
2854                         goto end;
2855                 default:
2856                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2857                         reply_code = LTTNG_ERR_UNK;
2858                         ret = -1;
2859                         goto end;
2860                 }
2861         }
2862
2863         chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2864         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2865                 reply_code = LTTNG_ERR_UNK;
2866                 ret = -1;
2867                 goto end;
2868         }
2869
2870         LTTNG_ASSERT(conn->session->output_directory);
2871         chunk_status = lttng_trace_chunk_set_as_owner(chunk, conn->session->output_directory);
2872         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2873                 reply_code = LTTNG_ERR_UNK;
2874                 ret = -1;
2875                 goto end;
2876         }
2877
2878         published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2879                 sessiond_trace_chunk_registry,
2880                 conn->session->sessiond_uuid,
2881                 conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2882                                                     conn->session->id,
2883                 chunk);
2884         if (!published_chunk) {
2885                 char uuid_str[LTTNG_UUID_STR_LEN];
2886
2887                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2888                 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
2889                     ", chunk_id = %" PRIu64,
2890                     uuid_str,
2891                     conn->session->id,
2892                     msg->chunk_id);
2893                 ret = -1;
2894                 reply_code = LTTNG_ERR_NOMEM;
2895                 goto end;
2896         }
2897
2898         if (conn->session->pending_closure_trace_chunk) {
2899                 /*
2900                  * Invalid; this means a second create_trace_chunk command was
2901                  * received before a close_trace_chunk.
2902                  */
2903                 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2904                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2905                 ret = -1;
2906                 goto end;
2907         }
2908         conn->session->pending_closure_trace_chunk = conn->session->current_trace_chunk;
2909         conn->session->current_trace_chunk = published_chunk;
2910         published_chunk = nullptr;
2911         if (!conn->session->pending_closure_trace_chunk) {
2912                 session->ongoing_rotation = false;
2913         }
2914 end:
2915         pthread_mutex_unlock(&conn->session->lock);
2916         reply.ret_code = htobe32((uint32_t) reply_code);
2917         send_ret = conn->sock->ops->sendmsg(
2918                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2919         if (send_ret < (ssize_t) sizeof(reply)) {
2920                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
2921                 ret = -1;
2922         }
2923 end_no_reply:
2924         lttng_trace_chunk_put(chunk);
2925         lttng_trace_chunk_put(published_chunk);
2926         return ret;
2927 }
2928
2929 /*
2930  * relay_close_trace_chunk: close a trace chunk
2931  */
2932 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2933                                    __attribute__((unused)),
2934                                    struct relay_connection *conn,
2935                                    const struct lttng_buffer_view *payload)
2936 {
2937         int ret = 0, buf_ret;
2938         ssize_t send_ret;
2939         struct relay_session *session = conn->session;
2940         struct lttcomm_relayd_close_trace_chunk *msg;
2941         struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2942         struct lttng_buffer_view header_view;
2943         struct lttng_trace_chunk *chunk = nullptr;
2944         enum lttng_error_code reply_code = LTTNG_OK;
2945         enum lttng_trace_chunk_status chunk_status;
2946         uint64_t chunk_id;
2947         LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2948         time_t close_timestamp;
2949         char closed_trace_chunk_path[LTTNG_PATH_MAX];
2950         size_t path_length = 0;
2951         const char *chunk_name = nullptr;
2952         struct lttng_dynamic_buffer reply_payload;
2953         const char *new_path;
2954
2955         lttng_dynamic_buffer_init(&reply_payload);
2956
2957         if (!session || !conn->version_check_done) {
2958                 ERR("Trying to close a trace chunk before version check");
2959                 ret = -1;
2960                 goto end_no_reply;
2961         }
2962
2963         if (session->major == 2 && session->minor < 11) {
2964                 ERR("Chunk close command is unsupported before 2.11");
2965                 ret = -1;
2966                 goto end_no_reply;
2967         }
2968
2969         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2970         if (!lttng_buffer_view_is_valid(&header_view)) {
2971                 ERR("Failed to receive payload of chunk close command");
2972                 ret = -1;
2973                 goto end_no_reply;
2974         }
2975
2976         /* Convert to host endianness. */
2977         msg = (typeof(msg)) header_view.data;
2978         chunk_id = be64toh(msg->chunk_id);
2979         close_timestamp = (time_t) be64toh(msg->close_timestamp);
2980         close_command.value = (lttng_trace_chunk_command_type) be32toh(msg->close_command.value);
2981         close_command.is_set = msg->close_command.is_set;
2982
2983         chunk = sessiond_trace_chunk_registry_get_chunk(sessiond_trace_chunk_registry,
2984                                                         conn->session->sessiond_uuid,
2985                                                         conn->session->id_sessiond.is_set ?
2986                                                                 conn->session->id_sessiond.value :
2987                                                                 conn->session->id,
2988                                                         chunk_id);
2989         if (!chunk) {
2990                 char uuid_str[LTTNG_UUID_STR_LEN];
2991
2992                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2993                 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
2994                     ", chunk_id = %" PRIu64,
2995                     uuid_str,
2996                     conn->session->id,
2997                     msg->chunk_id);
2998                 ret = -1;
2999                 reply_code = LTTNG_ERR_NOMEM;
3000                 goto end;
3001         }
3002
3003         pthread_mutex_lock(&session->lock);
3004         if (close_command.is_set && close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
3005                 /*
3006                  * Clear command. It is a protocol error to ask for a
3007                  * clear on a relay which does not allow it. Querying
3008                  * the configuration allows figuring out whether
3009                  * clearing is allowed before doing the clear.
3010                  */
3011                 if (!opt_allow_clear) {
3012                         ret = -1;
3013                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3014                         goto end_unlock_session;
3015                 }
3016         }
3017         if (session->pending_closure_trace_chunk && session->pending_closure_trace_chunk != chunk) {
3018                 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3019                     session->session_name);
3020                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3021                 ret = -1;
3022                 goto end_unlock_session;
3023         }
3024
3025         if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
3026             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
3027                 if (close_command.is_set &&
3028                     close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
3029                     !session->has_rotated) {
3030                         /* New chunk stays in session output directory. */
3031                         new_path = "";
3032                 } else {
3033                         /* Use chunk name for new chunk. */
3034                         new_path = nullptr;
3035                 }
3036                 /* Rename new chunk path. */
3037                 chunk_status =
3038                         lttng_trace_chunk_rename_path(session->current_trace_chunk, new_path);
3039                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3040                         ret = -1;
3041                         goto end_unlock_session;
3042                 }
3043                 session->ongoing_rotation = false;
3044         }
3045         if ((!close_command.is_set ||
3046              close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
3047             !lttng_trace_chunk_get_name_overridden(chunk)) {
3048                 const char *old_path;
3049
3050                 if (!session->has_rotated) {
3051                         old_path = "";
3052                 } else {
3053                         old_path = nullptr;
3054                 }
3055                 /* We need to move back the .tmp_old_chunk to its rightful place. */
3056                 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
3057                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3058                         ret = -1;
3059                         goto end_unlock_session;
3060                 }
3061         }
3062         chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, close_timestamp);
3063         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3064                 ERR("Failed to set trace chunk close timestamp");
3065                 ret = -1;
3066                 reply_code = LTTNG_ERR_UNK;
3067                 goto end_unlock_session;
3068         }
3069
3070         if (close_command.is_set) {
3071                 chunk_status = lttng_trace_chunk_set_close_command(chunk, close_command.value);
3072                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3073                         ret = -1;
3074                         reply_code = LTTNG_ERR_INVALID;
3075                         goto end_unlock_session;
3076                 }
3077         }
3078         chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, nullptr);
3079         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3080                 ERR("Failed to get chunk name");
3081                 ret = -1;
3082                 reply_code = LTTNG_ERR_UNK;
3083                 goto end_unlock_session;
3084         }
3085         if (!session->has_rotated && !session->snapshot) {
3086                 ret = lttng_strncpy(closed_trace_chunk_path,
3087                                     session->output_path,
3088                                     sizeof(closed_trace_chunk_path));
3089                 if (ret) {
3090                         ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3091                             strlen(session->output_path),
3092                             sizeof(closed_trace_chunk_path));
3093                         reply_code = LTTNG_ERR_NOMEM;
3094                         ret = -1;
3095                         goto end_unlock_session;
3096                 }
3097         } else {
3098                 if (session->snapshot) {
3099                         ret = snprintf(closed_trace_chunk_path,
3100                                        sizeof(closed_trace_chunk_path),
3101                                        "%s/%s",
3102                                        session->output_path,
3103                                        chunk_name);
3104                 } else {
3105                         ret = snprintf(closed_trace_chunk_path,
3106                                        sizeof(closed_trace_chunk_path),
3107                                        "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY "/%s",
3108                                        session->output_path,
3109                                        chunk_name);
3110                 }
3111                 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3112                         ERR("Failed to format closed trace chunk resulting path");
3113                         reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3114                         ret = -1;
3115                         goto end_unlock_session;
3116                 }
3117         }
3118         if (close_command.is_set &&
3119             close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3120                 session->has_rotated = true;
3121         }
3122         DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3123         path_length = strlen(closed_trace_chunk_path) + 1;
3124         if (path_length > UINT32_MAX) {
3125                 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3126                 ret = -1;
3127                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3128                 goto end_unlock_session;
3129         }
3130
3131         if (session->current_trace_chunk == chunk) {
3132                 /*
3133                  * After a trace chunk close command, no new streams
3134                  * referencing the chunk may be created. Hence, on the
3135                  * event that no new trace chunk have been created for
3136                  * the session, the reference to the current trace chunk
3137                  * is released in order to allow it to be reclaimed when
3138                  * the last stream releases its reference to it.
3139                  */
3140                 lttng_trace_chunk_put(session->current_trace_chunk);
3141                 session->current_trace_chunk = nullptr;
3142         }
3143         lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3144         session->pending_closure_trace_chunk = nullptr;
3145 end_unlock_session:
3146         pthread_mutex_unlock(&session->lock);
3147
3148 end:
3149         reply.generic.ret_code = htobe32((uint32_t) reply_code);
3150         reply.path_length = htobe32((uint32_t) path_length);
3151         buf_ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
3152         if (buf_ret) {
3153                 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3154                 goto end_no_reply;
3155         }
3156
3157         if (reply_code == LTTNG_OK) {
3158                 buf_ret = lttng_dynamic_buffer_append(
3159                         &reply_payload, closed_trace_chunk_path, path_length);
3160                 if (buf_ret) {
3161                         ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3162                         goto end_no_reply;
3163                 }
3164         }
3165
3166         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
3167         if (send_ret < reply_payload.size) {
3168                 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3169                     reply_payload.size,
3170                     send_ret);
3171                 ret = -1;
3172                 goto end_no_reply;
3173         }
3174 end_no_reply:
3175         lttng_trace_chunk_put(chunk);
3176         lttng_dynamic_buffer_reset(&reply_payload);
3177         return ret;
3178 }
3179
3180 /*
3181  * relay_trace_chunk_exists: check if a trace chunk exists
3182  */
3183 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr
3184                                     __attribute__((unused)),
3185                                     struct relay_connection *conn,
3186                                     const struct lttng_buffer_view *payload)
3187 {
3188         int ret = 0;
3189         ssize_t send_ret;
3190         struct relay_session *session = conn->session;
3191         struct lttcomm_relayd_trace_chunk_exists *msg;
3192         struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3193         struct lttng_buffer_view header_view;
3194         uint64_t chunk_id;
3195         bool chunk_exists;
3196
3197         if (!session || !conn->version_check_done) {
3198                 ERR("Trying to check for the presence of a trace chunk before version check");
3199                 ret = -1;
3200                 goto end_no_reply;
3201         }
3202
3203         if (session->major == 2 && session->minor < 11) {
3204                 ERR("Chunk exists command is unsupported before 2.11");
3205                 ret = -1;
3206                 goto end_no_reply;
3207         }
3208
3209         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3210         if (!lttng_buffer_view_is_valid(&header_view)) {
3211                 ERR("Failed to receive payload of chunk exists command");
3212                 ret = -1;
3213                 goto end_no_reply;
3214         }
3215
3216         /* Convert to host endianness. */
3217         msg = (typeof(msg)) header_view.data;
3218         chunk_id = be64toh(msg->chunk_id);
3219
3220         ret = sessiond_trace_chunk_registry_chunk_exists(sessiond_trace_chunk_registry,
3221                                                          conn->session->sessiond_uuid,
3222                                                          conn->session->id,
3223                                                          chunk_id,
3224                                                          &chunk_exists);
3225         /*
3226          * If ret is not 0, send the reply and report the error to the caller.
3227          * It is a protocol (or internal) error and the session/connection
3228          * should be torn down.
3229          */
3230         reply.generic.ret_code =
3231                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3232         reply.trace_chunk_exists = ret == 0 ? chunk_exists : 0;
3233
3234         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3235         if (send_ret < (ssize_t) sizeof(reply)) {
3236                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
3237                 ret = -1;
3238         }
3239 end_no_reply:
3240         return ret;
3241 }
3242
3243 /*
3244  * relay_get_configuration: query whether feature is available
3245  */
3246 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr
3247                                    __attribute__((unused)),
3248                                    struct relay_connection *conn,
3249                                    const struct lttng_buffer_view *payload)
3250 {
3251         int ret = 0;
3252         ssize_t send_ret;
3253         struct lttcomm_relayd_get_configuration *msg;
3254         struct lttcomm_relayd_get_configuration_reply reply = {};
3255         struct lttng_buffer_view header_view;
3256         uint64_t query_flags = 0;
3257         uint64_t result_flags = 0;
3258
3259         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3260         if (!lttng_buffer_view_is_valid(&header_view)) {
3261                 ERR("Failed to receive payload of chunk close command");
3262                 ret = -1;
3263                 goto end_no_reply;
3264         }
3265
3266         /* Convert to host endianness. */
3267         msg = (typeof(msg)) header_view.data;
3268         query_flags = be64toh(msg->query_flags);
3269
3270         if (query_flags) {
3271                 ret = LTTNG_ERR_INVALID_PROTOCOL;
3272                 goto reply;
3273         }
3274         if (opt_allow_clear) {
3275                 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3276         }
3277         ret = 0;
3278 reply:
3279         reply.generic.ret_code =
3280                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3281         reply.relayd_configuration_flags = htobe64(result_flags);
3282
3283         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3284         if (send_ret < (ssize_t) sizeof(reply)) {
3285                 ERR("Failed to send \"get configuration\" command reply (ret = %zd)", send_ret);
3286                 ret = -1;
3287         }
3288 end_no_reply:
3289         return ret;
3290 }
3291
3292 static int relay_process_control_command(struct relay_connection *conn,
3293                                          const struct lttcomm_relayd_hdr *header,
3294                                          const struct lttng_buffer_view *payload)
3295 {
3296         int ret = 0;
3297
3298         DBG3("Processing \"%s\" command for socket %i",
3299              lttcomm_relayd_command_str((lttcomm_relayd_command) header->cmd),
3300              conn->sock->fd);
3301         switch (header->cmd) {
3302         case RELAYD_CREATE_SESSION:
3303                 ret = relay_create_session(header, conn, payload);
3304                 break;
3305         case RELAYD_ADD_STREAM:
3306                 ret = relay_add_stream(header, conn, payload);
3307                 break;
3308         case RELAYD_START_DATA:
3309                 ret = relay_start(header, conn, payload);
3310                 break;
3311         case RELAYD_SEND_METADATA:
3312                 ret = relay_recv_metadata(header, conn, payload);
3313                 break;
3314         case RELAYD_VERSION:
3315                 ret = relay_send_version(header, conn, payload);
3316                 break;
3317         case RELAYD_CLOSE_STREAM:
3318                 ret = relay_close_stream(header, conn, payload);
3319                 break;
3320         case RELAYD_DATA_PENDING:
3321                 ret = relay_data_pending(header, conn, payload);
3322                 break;
3323         case RELAYD_QUIESCENT_CONTROL:
3324                 ret = relay_quiescent_control(header, conn, payload);
3325                 break;
3326         case RELAYD_BEGIN_DATA_PENDING:
3327                 ret = relay_begin_data_pending(header, conn, payload);
3328                 break;
3329         case RELAYD_END_DATA_PENDING:
3330                 ret = relay_end_data_pending(header, conn, payload);
3331                 break;
3332         case RELAYD_SEND_INDEX:
3333                 ret = relay_recv_index(header, conn, payload);
3334                 break;
3335         case RELAYD_STREAMS_SENT:
3336                 ret = relay_streams_sent(header, conn, payload);
3337                 break;
3338         case RELAYD_RESET_METADATA:
3339                 ret = relay_reset_metadata(header, conn, payload);
3340                 break;
3341         case RELAYD_ROTATE_STREAMS:
3342                 ret = relay_rotate_session_streams(header, conn, payload);
3343                 break;
3344         case RELAYD_CREATE_TRACE_CHUNK:
3345                 ret = relay_create_trace_chunk(header, conn, payload);
3346                 break;
3347         case RELAYD_CLOSE_TRACE_CHUNK:
3348                 ret = relay_close_trace_chunk(header, conn, payload);
3349                 break;
3350         case RELAYD_TRACE_CHUNK_EXISTS:
3351                 ret = relay_trace_chunk_exists(header, conn, payload);
3352                 break;
3353         case RELAYD_GET_CONFIGURATION:
3354                 ret = relay_get_configuration(header, conn, payload);
3355                 break;
3356         case RELAYD_UPDATE_SYNC_INFO:
3357         default:
3358                 ERR("Received unknown command (%u)", header->cmd);
3359                 relay_unknown_command(conn);
3360                 ret = -1;
3361                 goto end;
3362         }
3363
3364 end:
3365         return ret;
3366 }
3367
3368 static enum relay_connection_status
3369 relay_process_control_receive_payload(struct relay_connection *conn)
3370 {
3371         int ret = 0;
3372         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3373         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3374         struct ctrl_connection_state_receive_payload *state =
3375                 &conn->protocol.ctrl.state.receive_payload;
3376         struct lttng_buffer_view payload_view;
3377
3378         if (state->left_to_receive == 0) {
3379                 /* Short-circuit for payload-less commands. */
3380                 goto reception_complete;
3381         }
3382
3383         ret = conn->sock->ops->recvmsg(conn->sock,
3384                                        reception_buffer->data + state->received,
3385                                        state->left_to_receive,
3386                                        MSG_DONTWAIT);
3387         if (ret < 0) {
3388                 DIAGNOSTIC_PUSH
3389                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3390                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3391                         DIAGNOSTIC_POP
3392                         PERROR("Unable to receive command payload on sock %d", conn->sock->fd);
3393                         status = RELAY_CONNECTION_STATUS_ERROR;
3394                 }
3395                 goto end;
3396         } else if (ret == 0) {
3397                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3398                 status = RELAY_CONNECTION_STATUS_CLOSED;
3399                 goto end;
3400         }
3401
3402         LTTNG_ASSERT(ret > 0);
3403         LTTNG_ASSERT(ret <= state->left_to_receive);
3404
3405         state->left_to_receive -= ret;
3406         state->received += ret;
3407
3408         if (state->left_to_receive > 0) {
3409                 /*
3410                  * Can't transition to the protocol's next state, wait to
3411                  * receive the rest of the header.
3412                  */
3413                 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
3414                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3415                      state->received,
3416                      state->left_to_receive,
3417                      conn->sock->fd);
3418                 goto end;
3419         }
3420
3421 reception_complete:
3422         DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3423             conn->sock->fd,
3424             state->received);
3425         /*
3426          * The payload required to process the command has been received.
3427          * A view to the reception buffer is forwarded to the various
3428          * commands and the state of the control is reset on success.
3429          *
3430          * Commands are responsible for sending their reply to the peer.
3431          */
3432         payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer, 0, -1);
3433         ret = relay_process_control_command(conn, &state->header, &payload_view);
3434         if (ret < 0) {
3435                 status = RELAY_CONNECTION_STATUS_ERROR;
3436                 goto end;
3437         }
3438
3439         ret = connection_reset_protocol_state(conn);
3440         if (ret) {
3441                 status = RELAY_CONNECTION_STATUS_ERROR;
3442         }
3443 end:
3444         return status;
3445 }
3446
3447 static enum relay_connection_status
3448 relay_process_control_receive_header(struct relay_connection *conn)
3449 {
3450         int ret = 0;
3451         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3452         struct lttcomm_relayd_hdr header;
3453         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3454         struct ctrl_connection_state_receive_header *state =
3455                 &conn->protocol.ctrl.state.receive_header;
3456
3457         LTTNG_ASSERT(state->left_to_receive != 0);
3458
3459         ret = conn->sock->ops->recvmsg(conn->sock,
3460                                        reception_buffer->data + state->received,
3461                                        state->left_to_receive,
3462                                        MSG_DONTWAIT);
3463         if (ret < 0) {
3464                 DIAGNOSTIC_PUSH
3465                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3466                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3467                         DIAGNOSTIC_POP
3468                         PERROR("Unable to receive control command header on sock %d",
3469                                conn->sock->fd);
3470                         status = RELAY_CONNECTION_STATUS_ERROR;
3471                 }
3472                 goto end;
3473         } else if (ret == 0) {
3474                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3475                 status = RELAY_CONNECTION_STATUS_CLOSED;
3476                 goto end;
3477         }
3478
3479         LTTNG_ASSERT(ret > 0);
3480         LTTNG_ASSERT(ret <= state->left_to_receive);
3481
3482         state->left_to_receive -= ret;
3483         state->received += ret;
3484
3485         if (state->left_to_receive > 0) {
3486                 /*
3487                  * Can't transition to the protocol's next state, wait to
3488                  * receive the rest of the header.
3489                  */
3490                 DBG3("Partial reception of control connection protocol header (received %" PRIu64
3491                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3492                      state->received,
3493                      state->left_to_receive,
3494                      conn->sock->fd);
3495                 goto end;
3496         }
3497
3498         /* Transition to next state: receiving the command's payload. */
3499         conn->protocol.ctrl.state_id = CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3500         memcpy(&header, reception_buffer->data, sizeof(header));
3501         header.circuit_id = be64toh(header.circuit_id);
3502         header.data_size = be64toh(header.data_size);
3503         header.cmd = be32toh(header.cmd);
3504         header.cmd_version = be32toh(header.cmd_version);
3505         memcpy(&conn->protocol.ctrl.state.receive_payload.header, &header, sizeof(header));
3506
3507         DBG("Done receiving control command header: fd = %i, cmd = %s, cmd_version = %" PRIu32
3508             ", payload size = %" PRIu64 " bytes",
3509             conn->sock->fd,
3510             lttcomm_relayd_command_str((enum lttcomm_relayd_command) header.cmd),
3511             header.cmd_version,
3512             header.data_size);
3513
3514         if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3515                 ERR("Command header indicates a payload (%" PRIu64
3516                     " bytes) that exceeds the maximal payload size allowed on a control connection.",
3517                     header.data_size);
3518                 status = RELAY_CONNECTION_STATUS_ERROR;
3519                 goto end;
3520         }
3521
3522         conn->protocol.ctrl.state.receive_payload.left_to_receive = header.data_size;
3523         conn->protocol.ctrl.state.receive_payload.received = 0;
3524         ret = lttng_dynamic_buffer_set_size(reception_buffer, header.data_size);
3525         if (ret) {
3526                 status = RELAY_CONNECTION_STATUS_ERROR;
3527                 goto end;
3528         }
3529
3530         if (header.data_size == 0) {
3531                 /*
3532                  * Manually invoke the next state as the poll loop
3533                  * will not wake-up to allow us to proceed further.
3534                  */
3535                 status = relay_process_control_receive_payload(conn);
3536         }
3537 end:
3538         return status;
3539 }
3540
3541 /*
3542  * Process the commands received on the control socket
3543  */
3544 static enum relay_connection_status relay_process_control(struct relay_connection *conn)
3545 {
3546         enum relay_connection_status status;
3547
3548         switch (conn->protocol.ctrl.state_id) {
3549         case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3550                 status = relay_process_control_receive_header(conn);
3551                 break;
3552         case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3553                 status = relay_process_control_receive_payload(conn);
3554                 break;
3555         default:
3556                 ERR("Unknown control connection protocol state encountered.");
3557                 abort();
3558         }
3559
3560         return status;
3561 }
3562
3563 static enum relay_connection_status relay_process_data_receive_header(struct relay_connection *conn)
3564 {
3565         int ret;
3566         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3567         struct data_connection_state_receive_header *state =
3568                 &conn->protocol.data.state.receive_header;
3569         struct lttcomm_relayd_data_hdr header;
3570         struct relay_stream *stream;
3571
3572         LTTNG_ASSERT(state->left_to_receive != 0);
3573
3574         ret = conn->sock->ops->recvmsg(conn->sock,
3575                                        state->header_reception_buffer + state->received,
3576                                        state->left_to_receive,
3577                                        MSG_DONTWAIT);
3578         if (ret < 0) {
3579                 DIAGNOSTIC_PUSH
3580                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3581                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3582                         DIAGNOSTIC_POP
3583                         PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3584                         status = RELAY_CONNECTION_STATUS_ERROR;
3585                 }
3586                 goto end;
3587         } else if (ret == 0) {
3588                 /* Orderly shutdown. Not necessary to print an error. */
3589                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3590                 status = RELAY_CONNECTION_STATUS_CLOSED;
3591                 goto end;
3592         }
3593
3594         LTTNG_ASSERT(ret > 0);
3595         LTTNG_ASSERT(ret <= state->left_to_receive);
3596
3597         state->left_to_receive -= ret;
3598         state->received += ret;
3599
3600         if (state->left_to_receive > 0) {
3601                 /*
3602                  * Can't transition to the protocol's next state, wait to
3603                  * receive the rest of the header.
3604                  */
3605                 DBG3("Partial reception of data connection header (received %" PRIu64
3606                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3607                      state->received,
3608                      state->left_to_receive,
3609                      conn->sock->fd);
3610                 goto end;
3611         }
3612
3613         /* Transition to next state: receiving the payload. */
3614         conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3615
3616         memcpy(&header, state->header_reception_buffer, sizeof(header));
3617         header.circuit_id = be64toh(header.circuit_id);
3618         header.stream_id = be64toh(header.stream_id);
3619         header.data_size = be32toh(header.data_size);
3620         header.net_seq_num = be64toh(header.net_seq_num);
3621         header.padding_size = be32toh(header.padding_size);
3622         memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3623
3624         conn->protocol.data.state.receive_payload.left_to_receive = header.data_size;
3625         conn->protocol.data.state.receive_payload.received = 0;
3626         conn->protocol.data.state.receive_payload.rotate_index = false;
3627
3628         DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
3629             ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64
3630             ", padding_size = %" PRIu32,
3631             conn->sock->fd,
3632             header.circuit_id,
3633             header.stream_id,
3634             header.data_size,
3635             header.net_seq_num,
3636             header.padding_size);
3637
3638         stream = stream_get_by_id(header.stream_id);
3639         if (!stream) {
3640                 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3641                     header.stream_id);
3642                 /* Protocol error. */
3643                 status = RELAY_CONNECTION_STATUS_ERROR;
3644                 goto end;
3645         }
3646
3647         pthread_mutex_lock(&stream->lock);
3648         /* Prepare stream for the reception of a new packet. */
3649         ret = stream_init_packet(
3650                 stream, header.data_size, &conn->protocol.data.state.receive_payload.rotate_index);
3651         pthread_mutex_unlock(&stream->lock);
3652         if (ret) {
3653                 ERR("Failed to rotate stream output file");
3654                 status = RELAY_CONNECTION_STATUS_ERROR;
3655                 goto end_stream_unlock;
3656         }
3657
3658 end_stream_unlock:
3659         stream_put(stream);
3660 end:
3661         return status;
3662 }
3663
3664 static enum relay_connection_status
3665 relay_process_data_receive_payload(struct relay_connection *conn)
3666 {
3667         int ret;
3668         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3669         struct relay_stream *stream;
3670         struct data_connection_state_receive_payload *state =
3671                 &conn->protocol.data.state.receive_payload;
3672         const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3673         char data_buffer[chunk_size];
3674         bool partial_recv = false;
3675         bool new_stream = false, close_requested = false, index_flushed = false;
3676         uint64_t left_to_receive = state->left_to_receive;
3677         struct relay_session *session;
3678
3679         DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64
3680              " bytes received, %" PRIu64 " bytes left to receive",
3681              state->header.stream_id,
3682              state->header.net_seq_num,
3683              state->received,
3684              left_to_receive);
3685
3686         stream = stream_get_by_id(state->header.stream_id);
3687         if (!stream) {
3688                 /* Protocol error. */
3689                 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3690                     state->header.stream_id);
3691                 status = RELAY_CONNECTION_STATUS_ERROR;
3692                 goto end;
3693         }
3694
3695         pthread_mutex_lock(&stream->lock);
3696         session = stream->trace->session;
3697         if (!conn->session) {
3698                 ret = connection_set_session(conn, session);
3699                 if (ret) {
3700                         status = RELAY_CONNECTION_STATUS_ERROR;
3701                         goto end_stream_unlock;
3702                 }
3703         }
3704
3705         /*
3706          * The size of the "chunk" received on any iteration is bounded by:
3707          *   - the data left to receive,
3708          *   - the data immediately available on the socket,
3709          *   - the on-stack data buffer
3710          */
3711         while (left_to_receive > 0 && !partial_recv) {
3712                 size_t recv_size = std::min<uint64_t>(left_to_receive, chunk_size);
3713                 struct lttng_buffer_view packet_chunk;
3714
3715                 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, MSG_DONTWAIT);
3716                 if (ret < 0) {
3717                         DIAGNOSTIC_PUSH
3718                         DIAGNOSTIC_IGNORE_LOGICAL_OP
3719                         if (errno != EAGAIN && errno != EWOULDBLOCK) {
3720                                 DIAGNOSTIC_POP
3721                                 PERROR("Socket %d error", conn->sock->fd);
3722                                 status = RELAY_CONNECTION_STATUS_ERROR;
3723                         }
3724                         goto end_stream_unlock;
3725                 } else if (ret == 0) {
3726                         /* No more data ready to be consumed on socket. */
3727                         DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3728                              state->header.stream_id);
3729                         status = RELAY_CONNECTION_STATUS_CLOSED;
3730                         break;
3731                 } else if (ret < (int) recv_size) {
3732                         /*
3733                          * All the data available on the socket has been
3734                          * consumed.
3735                          */
3736                         partial_recv = true;
3737                         recv_size = ret;
3738                 }
3739
3740                 packet_chunk = lttng_buffer_view_init(data_buffer, 0, recv_size);
3741                 LTTNG_ASSERT(packet_chunk.data);
3742
3743                 ret = stream_write(stream, &packet_chunk, 0);
3744                 if (ret) {
3745                         ERR("Relay error writing data to file");
3746                         status = RELAY_CONNECTION_STATUS_ERROR;
3747                         goto end_stream_unlock;
3748                 }
3749
3750                 left_to_receive -= recv_size;
3751                 state->received += recv_size;
3752                 state->left_to_receive = left_to_receive;
3753         }
3754
3755         if (state->left_to_receive > 0) {
3756                 /*
3757                  * Did not receive all the data expected, wait for more data to
3758                  * become available on the socket.
3759                  */
3760                 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64
3761                      " bytes received, %" PRIu64 " bytes left to receive",
3762                      state->header.stream_id,
3763                      state->received,
3764                      state->left_to_receive);
3765                 goto end_stream_unlock;
3766         }
3767
3768         ret = stream_write(stream, nullptr, state->header.padding_size);
3769         if (ret) {
3770                 status = RELAY_CONNECTION_STATUS_ERROR;
3771                 goto end_stream_unlock;
3772         }
3773
3774         if (session_streams_have_index(session)) {
3775                 ret = stream_update_index(stream,
3776                                           state->header.net_seq_num,
3777                                           state->rotate_index,
3778                                           &index_flushed,
3779                                           state->header.data_size + state->header.padding_size);
3780                 if (ret < 0) {
3781                         ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64
3782                             " ret %d",
3783                             stream->stream_handle,
3784                             state->header.net_seq_num,
3785                             ret);
3786                         status = RELAY_CONNECTION_STATUS_ERROR;
3787                         goto end_stream_unlock;
3788                 }
3789         }
3790
3791         if (stream->prev_data_seq == -1ULL) {
3792                 new_stream = true;
3793         }
3794
3795         ret = stream_complete_packet(stream,
3796                                      state->header.data_size + state->header.padding_size,
3797                                      state->header.net_seq_num,
3798                                      index_flushed);
3799         if (ret) {
3800                 status = RELAY_CONNECTION_STATUS_ERROR;
3801                 goto end_stream_unlock;
3802         }
3803
3804         /*
3805          * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3806          * contents of *state which are aliased (union) to the same location as
3807          * the new state. Don't use it beyond this point.
3808          */
3809         connection_reset_protocol_state(conn);
3810         state = nullptr;
3811
3812 end_stream_unlock:
3813         close_requested = stream->close_requested;
3814         pthread_mutex_unlock(&stream->lock);
3815         if (close_requested && left_to_receive == 0) {
3816                 try_stream_close(stream);
3817         }
3818
3819         if (new_stream) {
3820                 pthread_mutex_lock(&session->lock);
3821                 uatomic_set(&session->new_streams, 1);
3822                 pthread_mutex_unlock(&session->lock);
3823         }
3824
3825         stream_put(stream);
3826 end:
3827         return status;
3828 }
3829
3830 /*
3831  * relay_process_data: Process the data received on the data socket
3832  */
3833 static enum relay_connection_status relay_process_data(struct relay_connection *conn)
3834 {
3835         enum relay_connection_status status;
3836
3837         switch (conn->protocol.data.state_id) {
3838         case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3839                 status = relay_process_data_receive_header(conn);
3840                 break;
3841         case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3842                 status = relay_process_data_receive_payload(conn);
3843                 break;
3844         default:
3845                 ERR("Unexpected data connection communication state.");
3846                 abort();
3847         }
3848
3849         return status;
3850 }
3851
3852 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3853 {
3854         int ret;
3855
3856         (void) lttng_poll_del(events, pollfd);
3857
3858         ret = fd_tracker_close_unsuspendable_fd(
3859                 the_fd_tracker, &pollfd, 1, fd_tracker_util_close_fd, nullptr);
3860         if (ret < 0) {
3861                 ERR("Closing pollfd %d", pollfd);
3862         }
3863 }
3864
3865 static void relay_thread_close_connection(struct lttng_poll_event *events,
3866                                           int pollfd,
3867                                           struct relay_connection *conn)
3868 {
3869         const char *type_str;
3870
3871         switch (conn->type) {
3872         case RELAY_DATA:
3873                 type_str = "Data";
3874                 break;
3875         case RELAY_CONTROL:
3876                 type_str = "Control";
3877                 break;
3878         case RELAY_VIEWER_COMMAND:
3879                 type_str = "Viewer Command";
3880                 break;
3881         case RELAY_VIEWER_NOTIFICATION:
3882                 type_str = "Viewer Notification";
3883                 break;
3884         default:
3885                 type_str = "Unknown";
3886         }
3887         cleanup_connection_pollfd(events, pollfd);
3888         connection_put(conn);
3889         DBG("%s connection closed with %d", type_str, pollfd);
3890 }
3891
3892 /*
3893  * This thread does the actual work
3894  */
3895 static void *relay_thread_worker(void *data __attribute__((unused)))
3896 {
3897         int ret, err = -1, last_seen_data_fd = -1;
3898         uint32_t nb_fd;
3899         struct lttng_poll_event events;
3900         struct lttng_ht *relay_connections_ht;
3901         struct lttng_ht_iter iter;
3902         struct relay_connection *destroy_conn = nullptr;
3903
3904         DBG("[thread] Relay worker started");
3905
3906         rcu_register_thread();
3907
3908         health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3909
3910         if (testpoint(relayd_thread_worker)) {
3911                 goto error_testpoint;
3912         }
3913
3914         health_code_update();
3915
3916         /* table of connections indexed on socket */
3917         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3918         if (!relay_connections_ht) {
3919                 goto relay_connections_ht_error;
3920         }
3921
3922         ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3923         if (ret < 0) {
3924                 goto error_poll_create;
3925         }
3926
3927         ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3928         if (ret < 0) {
3929                 goto error;
3930         }
3931
3932 restart:
3933         while (true) {
3934                 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3935
3936                 health_code_update();
3937
3938                 /* Infinite blocking call, waiting for transmission */
3939                 DBG3("Relayd worker thread polling...");
3940                 health_poll_entry();
3941                 ret = lttng_poll_wait(&events, -1);
3942                 health_poll_exit();
3943                 if (ret < 0) {
3944                         /*
3945                          * Restart interrupted system call.
3946                          */
3947                         if (errno == EINTR) {
3948                                 goto restart;
3949                         }
3950                         goto error;
3951                 }
3952
3953                 nb_fd = ret;
3954
3955                 /*
3956                  * Process control. The control connection is
3957                  * prioritized so we don't starve it with high
3958                  * throughput tracing data on the data connection.
3959                  */
3960                 for (i = 0; i < nb_fd; i++) {
3961                         /* Fetch once the poll data */
3962                         const auto revents = LTTNG_POLL_GETEV(&events, i);
3963                         const auto pollfd = LTTNG_POLL_GETFD(&events, i);
3964
3965                         health_code_update();
3966
3967                         /* Activity on thread quit pipe, exiting. */
3968                         if (relayd_is_thread_quit_pipe(pollfd)) {
3969                                 DBG("Activity on thread quit pipe");
3970                                 err = 0;
3971                                 goto exit;
3972                         }
3973
3974                         /* Inspect the relay conn pipe for new connection */
3975                         if (pollfd == relay_conn_pipe[0]) {
3976                                 if (revents & LPOLLIN) {
3977                                         struct relay_connection *conn;
3978
3979                                         ret = lttng_read(relay_conn_pipe[0],
3980                                                          &conn,
3981                                                          sizeof(conn)); /* NOLINT sizeof used on a
3982                                                                            pointer. */
3983                                         if (ret < 0) {
3984                                                 goto error;
3985                                         }
3986                                         ret = lttng_poll_add(
3987                                                 &events, conn->sock->fd, LPOLLIN | LPOLLRDHUP);
3988                                         if (ret) {
3989                                                 ERR("Failed to add new connection file descriptor to poll set");
3990                                                 goto error;
3991                                         }
3992                                         connection_ht_add(relay_connections_ht, conn);
3993                                         DBG("Connection socket %d added", conn->sock->fd);
3994                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3995                                         ERR("Relay connection pipe error");
3996                                         goto error;
3997                                 } else {
3998                                         ERR("Unexpected poll events %u for sock %d",
3999                                             revents,
4000                                             pollfd);
4001                                         goto error;
4002                                 }
4003                         } else {
4004                                 struct relay_connection *ctrl_conn;
4005
4006                                 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4007                                 /* If not found, there is a synchronization issue. */
4008                                 LTTNG_ASSERT(ctrl_conn);
4009
4010                                 if (ctrl_conn->type == RELAY_DATA) {
4011                                         if (revents & LPOLLIN) {
4012                                                 /*
4013                                                  * Flag the last seen data fd not deleted. It will
4014                                                  * be used as the last seen fd if any fd gets
4015                                                  * deleted in this first loop.
4016                                                  */
4017                                                 last_notdel_data_fd = pollfd;
4018                                         }
4019                                         goto put_ctrl_connection;
4020                                 }
4021                                 LTTNG_ASSERT(ctrl_conn->type == RELAY_CONTROL);
4022
4023                                 if (revents & LPOLLIN) {
4024                                         enum relay_connection_status status;
4025
4026                                         status = relay_process_control(ctrl_conn);
4027                                         if (status != RELAY_CONNECTION_STATUS_OK) {
4028                                                 /*
4029                                                  * On socket error flag the session as aborted to
4030                                                  * force the cleanup of its stream otherwise it can
4031                                                  * leak during the lifetime of the relayd.
4032                                                  *
4033                                                  * This prevents situations in which streams can be
4034                                                  * left opened because an index was received, the
4035                                                  * control connection is closed, and the data
4036                                                  * connection is closed (uncleanly) before the
4037                                                  * packet's data provided.
4038                                                  *
4039                                                  * Since the control connection encountered an
4040                                                  * error, it is okay to be conservative and close
4041                                                  * the session right now as we can't rely on the
4042                                                  * protocol being respected anymore.
4043                                                  */
4044                                                 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4045                                                         session_abort(ctrl_conn->session);
4046                                                 }
4047
4048                                                 /* Clear the connection on error or close. */
4049                                                 relay_thread_close_connection(
4050                                                         &events, pollfd, ctrl_conn);
4051                                         }
4052                                         seen_control = 1;
4053                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4054                                         relay_thread_close_connection(&events, pollfd, ctrl_conn);
4055                                         if (last_seen_data_fd == pollfd) {
4056                                                 last_seen_data_fd = last_notdel_data_fd;
4057                                         }
4058                                 } else {
4059                                         ERR("Unexpected poll events %u for control sock %d",
4060                                             revents,
4061                                             pollfd);
4062                                         connection_put(ctrl_conn);
4063                                         goto error;
4064                                 }
4065                         put_ctrl_connection:
4066                                 connection_put(ctrl_conn);
4067                         }
4068                 }
4069
4070                 /*
4071                  * The last loop handled a control request, go back to poll to make
4072                  * sure we prioritise the control socket.
4073                  */
4074                 if (seen_control) {
4075                         continue;
4076                 }
4077
4078                 if (last_seen_data_fd >= 0) {
4079                         for (i = 0; i < nb_fd; i++) {
4080                                 int pollfd = LTTNG_POLL_GETFD(&events, i);
4081
4082                                 health_code_update();
4083
4084                                 if (last_seen_data_fd == pollfd) {
4085                                         idx = i;
4086                                         break;
4087                                 }
4088                         }
4089                 }
4090
4091                 /* Process data connection. */
4092                 for (i = idx + 1; i < nb_fd; i++) {
4093                         /* Fetch the poll data. */
4094                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
4095                         int pollfd = LTTNG_POLL_GETFD(&events, i);
4096                         struct relay_connection *data_conn;
4097
4098                         health_code_update();
4099
4100                         if (!revents) {
4101                                 /* No activity for this FD (poll implementation). */
4102                                 continue;
4103                         }
4104
4105                         /* Skip the command pipe. It's handled in the first loop. */
4106                         if (pollfd == relay_conn_pipe[0]) {
4107                                 continue;
4108                         }
4109
4110                         data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4111                         if (!data_conn) {
4112                                 /* Skip it. Might be removed before. */
4113                                 continue;
4114                         }
4115                         if (data_conn->type == RELAY_CONTROL) {
4116                                 goto put_data_connection;
4117                         }
4118                         LTTNG_ASSERT(data_conn->type == RELAY_DATA);
4119
4120                         if (revents & LPOLLIN) {
4121                                 enum relay_connection_status status;
4122
4123                                 status = relay_process_data(data_conn);
4124                                 /* Connection closed or error. */
4125                                 if (status != RELAY_CONNECTION_STATUS_OK) {
4126                                         /*
4127                                          * On socket error flag the session as aborted to force
4128                                          * the cleanup of its stream otherwise it can leak
4129                                          * during the lifetime of the relayd.
4130                                          *
4131                                          * This prevents situations in which streams can be
4132                                          * left opened because an index was received, the
4133                                          * control connection is closed, and the data
4134                                          * connection is closed (uncleanly) before the packet's
4135                                          * data provided.
4136                                          *
4137                                          * Since the data connection encountered an error,
4138                                          * it is okay to be conservative and close the
4139                                          * session right now as we can't rely on the protocol
4140                                          * being respected anymore.
4141                                          */
4142                                         if (status == RELAY_CONNECTION_STATUS_ERROR) {
4143                                                 session_abort(data_conn->session);
4144                                         }
4145                                         relay_thread_close_connection(&events, pollfd, data_conn);
4146                                         /*
4147                                          * Every goto restart call sets the last seen fd where
4148                                          * here we don't really care since we gracefully
4149                                          * continue the loop after the connection is deleted.
4150                                          */
4151                                 } else {
4152                                         /* Keep last seen port. */
4153                                         last_seen_data_fd = pollfd;
4154                                         connection_put(data_conn);
4155                                         goto restart;
4156                                 }
4157                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4158                                 relay_thread_close_connection(&events, pollfd, data_conn);
4159                         } else {
4160                                 ERR("Unknown poll events %u for data sock %d", revents, pollfd);
4161                         }
4162                 put_data_connection:
4163                         connection_put(data_conn);
4164                 }
4165                 last_seen_data_fd = -1;
4166         }
4167
4168         /* Normal exit, no error */
4169         ret = 0;
4170
4171 exit:
4172 error:
4173         /* Cleanup remaining connection object. */
4174         {
4175                 lttng::urcu::read_lock_guard read_lock;
4176
4177                 cds_lfht_for_each_entry (
4178                         relay_connections_ht->ht, &iter.iter, destroy_conn, sock_n.node) {
4179                         health_code_update();
4180
4181                         session_abort(destroy_conn->session);
4182
4183                         /*
4184                          * No need to grab another ref, because we own
4185                          * destroy_conn.
4186                          */
4187                         relay_thread_close_connection(
4188                                 &events, destroy_conn->sock->fd, destroy_conn);
4189                 }
4190         }
4191
4192         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4193 error_poll_create:
4194         lttng_ht_destroy(relay_connections_ht);
4195 relay_connections_ht_error:
4196         /* Close relay conn pipes */
4197         (void) fd_tracker_util_pipe_close(the_fd_tracker, relay_conn_pipe);
4198         if (err) {
4199                 DBG("Thread exited with error");
4200         }
4201         DBG("Worker thread cleanup complete");
4202 error_testpoint:
4203         if (err) {
4204                 health_error();
4205                 ERR("Health error occurred in %s", __func__);
4206         }
4207         health_unregister(health_relayd);
4208         rcu_unregister_thread();
4209         lttng_relay_stop_threads();
4210         return nullptr;
4211 }
4212
4213 /*
4214  * Create the relay command pipe to wake thread_manage_apps.
4215  * Closed in cleanup().
4216  */
4217 static int create_relay_conn_pipe()
4218 {
4219         return fd_tracker_util_pipe_open_cloexec(
4220                 the_fd_tracker, "Relayd connection pipe", relay_conn_pipe);
4221 }
4222
4223 static int stdio_open(void *data __attribute__((unused)), int *fds)
4224 {
4225         fds[0] = fileno(stdout);
4226         fds[1] = fileno(stderr);
4227         return 0;
4228 }
4229
4230 static int track_stdio()
4231 {
4232         int fds[2];
4233         const char *names[] = { "stdout", "stderr" };
4234
4235         return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds, names, 2, stdio_open, nullptr);
4236 }
4237
4238 /*
4239  * main
4240  */
4241 int main(int argc, char **argv)
4242 {
4243         bool thread_is_rcu_registered = false;
4244         int ret = 0, retval = 0;
4245         void *status;
4246         char *unlinked_file_directory_path = nullptr, *output_path = nullptr;
4247
4248         /* Parse environment variables */
4249         ret = parse_env_options();
4250         if (ret) {
4251                 retval = -1;
4252                 goto exit_options;
4253         }
4254
4255         /*
4256          * Parse arguments.
4257          * Command line arguments overwrite environment.
4258          */
4259         progname = argv[0];
4260         if (set_options(argc, argv)) {
4261                 retval = -1;
4262                 goto exit_options;
4263         }
4264
4265         if (set_signal_handler()) {
4266                 retval = -1;
4267                 goto exit_options;
4268         }
4269
4270         relayd_config_log();
4271
4272         if (opt_print_version) {
4273                 print_version();
4274                 retval = 0;
4275                 goto exit_options;
4276         }
4277
4278         ret = fclose(stdin);
4279         if (ret) {
4280                 PERROR("Failed to close stdin");
4281                 goto exit_options;
4282         }
4283
4284         DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4285
4286         /* Try to create directory if -o, --output is specified. */
4287         if (opt_output_path) {
4288                 if (*opt_output_path != '/') {
4289                         ERR("Please specify an absolute path for -o, --output PATH");
4290                         retval = -1;
4291                         goto exit_options;
4292                 }
4293
4294                 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG, -1, -1);
4295                 if (ret < 0) {
4296                         ERR("Unable to create %s", opt_output_path);
4297                         retval = -1;
4298                         goto exit_options;
4299                 }
4300         }
4301
4302         /* Daemonize */
4303         if (opt_daemon || opt_background) {
4304                 ret = lttng_daemonize(&child_ppid, &recv_child_signal, !opt_background);
4305                 if (ret < 0) {
4306                         retval = -1;
4307                         goto exit_options;
4308                 }
4309         }
4310
4311         if (opt_working_directory) {
4312                 ret = utils_change_working_directory(opt_working_directory);
4313                 if (ret) {
4314                         /* All errors are already logged. */
4315                         goto exit_options;
4316                 }
4317         }
4318
4319         sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4320         if (!sessiond_trace_chunk_registry) {
4321                 ERR("Failed to initialize session daemon trace chunk registry");
4322                 retval = -1;
4323                 goto exit_options;
4324         }
4325
4326         /*
4327          * The RCU thread registration (and use, through the fd-tracker's
4328          * creation) is done after the daemonization to allow us to not
4329          * deal with liburcu's fork() management as the call RCU needs to
4330          * be restored.
4331          */
4332         rcu_register_thread();
4333         thread_is_rcu_registered = true;
4334
4335         output_path = create_output_path("");
4336         if (!output_path) {
4337                 ERR("Failed to get output path");
4338                 retval = -1;
4339                 goto exit_options;
4340         }
4341         ret = asprintf(&unlinked_file_directory_path,
4342                        "%s/%s",
4343                        output_path,
4344                        DEFAULT_UNLINKED_FILES_DIRECTORY);
4345         free(output_path);
4346         if (ret < 0) {
4347                 ERR("Failed to format unlinked file directory path");
4348                 retval = -1;
4349                 goto exit_options;
4350         }
4351         the_fd_tracker = fd_tracker_create(unlinked_file_directory_path, lttng_opt_fd_pool_size);
4352         free(unlinked_file_directory_path);
4353         if (!the_fd_tracker) {
4354                 retval = -1;
4355                 goto exit_options;
4356         }
4357
4358         ret = track_stdio();
4359         if (ret) {
4360                 retval = -1;
4361                 goto exit_options;
4362         }
4363
4364         /* Initialize thread health monitoring */
4365         health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4366         if (!health_relayd) {
4367                 PERROR("health_app_create error");
4368                 retval = -1;
4369                 goto exit_options;
4370         }
4371
4372         /* Create thread quit pipe */
4373         if (relayd_init_thread_quit_pipe()) {
4374                 retval = -1;
4375                 goto exit_options;
4376         }
4377
4378         /* Setup the thread apps communication pipe. */
4379         if (create_relay_conn_pipe()) {
4380                 retval = -1;
4381                 goto exit_options;
4382         }
4383
4384         /* Init relay command queue. */
4385         cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4386
4387         /* Initialize communication library */
4388         lttcomm_init();
4389         lttcomm_inet_init();
4390
4391         /* tables of sessions indexed by session ID */
4392         sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4393         if (!sessions_ht) {
4394                 retval = -1;
4395                 goto exit_options;
4396         }
4397
4398         /* tables of streams indexed by stream ID */
4399         relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4400         if (!relay_streams_ht) {
4401                 retval = -1;
4402                 goto exit_options;
4403         }
4404
4405         /* tables of streams indexed by stream ID */
4406         viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4407         if (!viewer_streams_ht) {
4408                 retval = -1;
4409                 goto exit_options;
4410         }
4411
4412         ret = init_health_quit_pipe();
4413         if (ret) {
4414                 retval = -1;
4415                 goto exit_options;
4416         }
4417
4418         /* Create thread to manage the client socket */
4419         ret = pthread_create(&health_thread,
4420                              default_pthread_attr(),
4421                              thread_manage_health_relayd,
4422                              (void *) nullptr);
4423         if (ret) {
4424                 errno = ret;
4425                 PERROR("pthread_create health");
4426                 retval = -1;
4427                 goto exit_options;
4428         }
4429
4430         /* Setup the dispatcher thread */
4431         ret = pthread_create(&dispatcher_thread,
4432                              default_pthread_attr(),
4433                              relay_thread_dispatcher,
4434                              (void *) nullptr);
4435         if (ret) {
4436                 errno = ret;
4437                 PERROR("pthread_create dispatcher");
4438                 retval = -1;
4439                 goto exit_dispatcher_thread;
4440         }
4441
4442         /* Setup the worker thread */
4443         ret = pthread_create(&worker_thread, default_pthread_attr(), relay_thread_worker, nullptr);
4444         if (ret) {
4445                 errno = ret;
4446                 PERROR("pthread_create worker");
4447                 retval = -1;
4448                 goto exit_worker_thread;
4449         }
4450
4451         /* Setup the listener thread */
4452         ret = pthread_create(
4453                 &listener_thread, default_pthread_attr(), relay_thread_listener, (void *) nullptr);
4454         if (ret) {
4455                 errno = ret;
4456                 PERROR("pthread_create listener");
4457                 retval = -1;
4458                 goto exit_listener_thread;
4459         }
4460
4461         ret = relayd_live_create(live_uri);
4462         if (ret) {
4463                 ERR("Starting live viewer threads");
4464                 retval = -1;
4465                 goto exit_live;
4466         }
4467
4468         /*
4469          * This is where we start awaiting program completion (e.g. through
4470          * signal that asks threads to teardown).
4471          */
4472
4473         ret = relayd_live_join();
4474         if (ret) {
4475                 retval = -1;
4476         }
4477 exit_live:
4478
4479         ret = pthread_join(listener_thread, &status);
4480         if (ret) {
4481                 errno = ret;
4482                 PERROR("pthread_join listener_thread");
4483                 retval = -1;
4484         }
4485
4486 exit_listener_thread:
4487         ret = pthread_join(worker_thread, &status);
4488         if (ret) {
4489                 errno = ret;
4490                 PERROR("pthread_join worker_thread");
4491                 retval = -1;
4492         }
4493
4494 exit_worker_thread:
4495         ret = pthread_join(dispatcher_thread, &status);
4496         if (ret) {
4497                 errno = ret;
4498                 PERROR("pthread_join dispatcher_thread");
4499                 retval = -1;
4500         }
4501 exit_dispatcher_thread:
4502
4503         ret = pthread_join(health_thread, &status);
4504         if (ret) {
4505                 errno = ret;
4506                 PERROR("pthread_join health_thread");
4507                 retval = -1;
4508         }
4509 exit_options:
4510         /*
4511          * Wait for all pending call_rcu work to complete before tearing
4512          * down data structures. call_rcu worker may be trying to
4513          * perform lookups in those structures.
4514          */
4515         rcu_barrier();
4516         relayd_cleanup();
4517
4518         /* Ensure all prior call_rcu are done. */
4519         rcu_barrier();
4520
4521         if (thread_is_rcu_registered) {
4522                 rcu_unregister_thread();
4523         }
4524
4525         if (!retval) {
4526                 exit(EXIT_SUCCESS);
4527         } else {
4528                 exit(EXIT_FAILURE);
4529         }
4530 }