src/bin/lttng-consumerd/lttng-consumerd.c

   1 /*
   2  * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
   3  * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * SPDX-License-Identifier: GPL-2.0-only
   6  *
   7  */
   8
   9 #define _LGPL_SOURCE
  10 #include <fcntl.h>
  11 #include <getopt.h>
  12 #include <grp.h>
  13 #include <limits.h>
  14 #include <pthread.h>
  15 #include <signal.h>
  16 #include <stdio.h>
  17 #include <stdlib.h>
  18 #include <string.h>
  19 #include <sys/ipc.h>
  20 #include <sys/resource.h>
  21 #include <sys/shm.h>
  22 #include <sys/socket.h>
  23 #include <sys/stat.h>
  24 #include <sys/types.h>
  25 #include <urcu/list.h>
  26 #include <poll.h>
  27 #include <unistd.h>
  28 #include <sys/mman.h>
  29 #include <assert.h>
  30 #include <urcu/compiler.h>
  31 #include <ulimit.h>
  32
  33 #include <common/defaults.h>
  34 #include <common/common.h>
  35 #include <common/consumer/consumer.h>
  36 #include <common/consumer/consumer-timer.h>
  37 #include <common/compat/poll.h>
  38 #include <common/compat/getenv.h>
  39 #include <common/sessiond-comm/sessiond-comm.h>
  40 #include <common/utils.h>
  41
  42 #include "lttng-consumerd.h"
  43 #include "health-consumerd.h"
  44
  45 /* threads (channel handling, poll, metadata, sessiond) */
  46
  47 static pthread_t channel_thread, data_thread, metadata_thread,
  48                 sessiond_thread, metadata_timer_thread, health_thread;
  49 static bool metadata_timer_thread_online;
  50
  51 /* to count the number of times the user pressed ctrl+c */
  52 static int sigintcount = 0;
  53
  54 /* Argument variables */
  55 int lttng_opt_quiet;    /* not static in error.h */
  56 int lttng_opt_verbose;  /* not static in error.h */
  57 int lttng_opt_mi;       /* not static in error.h */
  58
  59 static int opt_daemon;
  60 static const char *progname;
  61 static char command_sock_path[PATH_MAX]; /* Global command socket path */
  62 static char error_sock_path[PATH_MAX]; /* Global error path */
  63 static enum lttng_consumer_type opt_type = LTTNG_CONSUMER_KERNEL;
  64
  65 /* the liblttngconsumerd context */
  66 static struct lttng_consumer_local_data *ctx;
  67
  68 /* Consumerd health monitoring */
  69 struct health_app *health_consumerd;
  70
  71 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
  72
  73 int lttng_consumer_ready = NR_LTTNG_CONSUMER_READY;
  74
  75 enum lttng_consumer_type lttng_consumer_get_type(void)
  76 {
  77         if (!ctx) {
  78                 return LTTNG_CONSUMER_UNKNOWN;
  79         }
  80         return ctx->type;
  81 }
  82
  83 /*
  84  * Signal handler for the daemon
  85  */
  86 static void sighandler(int sig)
  87 {
  88         if (sig == SIGINT && sigintcount++ == 0) {
  89                 DBG("ignoring first SIGINT");
  90                 return;
  91         }
  92
  93         if (ctx) {
  94                 lttng_consumer_should_exit(ctx);
  95         }
  96 }
  97
  98 /*
  99  * Setup signal handler for :
 100  *      SIGINT, SIGTERM, SIGPIPE
 101  */
 102 static int set_signal_handler(void)
 103 {
 104         int ret = 0;
 105         struct sigaction sa;
 106         sigset_t sigset;
 107
 108         if ((ret = sigemptyset(&sigset)) < 0) {
 109                 PERROR("sigemptyset");
 110                 return ret;
 111         }
 112
 113         sa.sa_mask = sigset;
 114         sa.sa_flags = 0;
 115
 116         sa.sa_handler = sighandler;
 117         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
 118                 PERROR("sigaction");
 119                 return ret;
 120         }
 121
 122         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
 123                 PERROR("sigaction");
 124                 return ret;
 125         }
 126
 127         sa.sa_handler = SIG_IGN;
 128         if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
 129                 PERROR("sigaction");
 130                 return ret;
 131         }
 132
 133         return ret;
 134 }
 135
 136 /*
 137  * Usage function on stream file.
 138  */
 139 static void usage(FILE *fp)
 140 {
 141         fprintf(fp, "Usage: %s OPTIONS\n\nOptions:\n", progname);
 142         fprintf(fp, "  -h, --help                         "
 143                         "Display this usage.\n");
 144         fprintf(fp, "  -c, --consumerd-cmd-sock PATH      "
 145                         "Specify path for the command socket\n");
 146         fprintf(fp, "  -e, --consumerd-err-sock PATH      "
 147                         "Specify path for the error socket\n");
 148         fprintf(fp, "  -d, --daemonize                    "
 149                         "Start as a daemon.\n");
 150         fprintf(fp, "  -q, --quiet                        "
 151                         "No output at all.\n");
 152         fprintf(fp, "  -v, --verbose                      "
 153                         "Verbose mode. Activate DBG() macro.\n");
 154         fprintf(fp, "  -V, --version                      "
 155                         "Show version number.\n");
 156         fprintf(fp, "  -g, --group NAME                   "
 157                         "Specify the tracing group name. (default: tracing)\n");
 158         fprintf(fp, "  -k, --kernel                       "
 159                         "Consumer kernel buffers (default).\n");
 160         fprintf(fp, "  -u, --ust                          "
 161                         "Consumer UST buffers.%s\n",
 162 #ifdef HAVE_LIBLTTNG_UST_CTL
 163                         ""
 164 #else
 165                         " (support not compiled in)"
 166 #endif
 167                         );
 168 }
 169
 170 /*
 171  * daemon argument parsing
 172  */
 173 static int parse_args(int argc, char **argv)
 174 {
 175         int c, ret = 0;
 176
 177         static struct option long_options[] = {
 178                 { "consumerd-cmd-sock", 1, 0, 'c' },
 179                 { "consumerd-err-sock", 1, 0, 'e' },
 180                 { "daemonize", 0, 0, 'd' },
 181                 { "group", 1, 0, 'g' },
 182                 { "help", 0, 0, 'h' },
 183                 { "quiet", 0, 0, 'q' },
 184                 { "verbose", 0, 0, 'v' },
 185                 { "version", 0, 0, 'V' },
 186                 { "kernel", 0, 0, 'k' },
 187 #ifdef HAVE_LIBLTTNG_UST_CTL
 188                 { "ust", 0, 0, 'u' },
 189 #endif
 190                 { NULL, 0, 0, 0 }
 191         };
 192
 193         while (1) {
 194                 int option_index = 0;
 195                 c = getopt_long(argc, argv, "dhqvVku" "c:e:g:",
 196                                 long_options, &option_index);
 197                 if (c == -1) {
 198                         break;
 199                 }
 200
 201                 switch (c) {
 202                 case 0:
 203                         fprintf(stderr, "option %s",
 204                                 long_options[option_index].name);
 205                         if (optarg) {
 206                                 fprintf(stderr, " with arg %s\n", optarg);
 207                                 ret = -1;
 208                                 goto end;
 209                         }
 210                         break;
 211                 case 'c':
 212                         if (lttng_is_setuid_setgid()) {
 213                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 214                                         "-c, --consumerd-cmd-sock");
 215                         } else {
 216                                 snprintf(command_sock_path, PATH_MAX, "%s", optarg);
 217                         }
 218                         break;
 219                 case 'e':
 220                         if (lttng_is_setuid_setgid()) {
 221                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 222                                         "-e, --consumerd-err-sock");
 223                         } else {
 224                                 snprintf(error_sock_path, PATH_MAX, "%s", optarg);
 225                         }
 226                         break;
 227                 case 'd':
 228                         opt_daemon = 1;
 229                         break;
 230                 case 'g':
 231                         if (lttng_is_setuid_setgid()) {
 232                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 233                                         "-g, --group");
 234                         } else {
 235                                 tracing_group_name = optarg;
 236                         }
 237                         break;
 238                 case 'h':
 239                         usage(stdout);
 240                         exit(EXIT_SUCCESS);
 241                 case 'q':
 242                         lttng_opt_quiet = 1;
 243                         break;
 244                 case 'v':
 245                         lttng_opt_verbose = 3;
 246                         break;
 247                 case 'V':
 248                         fprintf(stdout, "%s\n", VERSION);
 249                         exit(EXIT_SUCCESS);
 250                 case 'k':
 251                         opt_type = LTTNG_CONSUMER_KERNEL;
 252                         break;
 253 #ifdef HAVE_LIBLTTNG_UST_CTL
 254                 case 'u':
 255 # if (CAA_BITS_PER_LONG == 64)
 256                         opt_type = LTTNG_CONSUMER64_UST;
 257 # elif (CAA_BITS_PER_LONG == 32)
 258                         opt_type = LTTNG_CONSUMER32_UST;
 259 # else
 260 #  error "Unknown bitness"
 261 # endif
 262                         break;
 263 #endif
 264                 default:
 265                         usage(stderr);
 266                         ret = -1;
 267                         goto end;
 268                 }
 269         }
 270 end:
 271         return ret;
 272 }
 273
 274 /*
 275  * Set open files limit to unlimited. This daemon can open a large number of
 276  * file descriptors in order to consumer multiple kernel traces.
 277  */
 278 static void set_ulimit(void)
 279 {
 280         int ret;
 281         struct rlimit lim;
 282
 283         /* The kernel does not allowed an infinite limit for open files */
 284         lim.rlim_cur = 65535;
 285         lim.rlim_max = 65535;
 286
 287         ret = setrlimit(RLIMIT_NOFILE, &lim);
 288         if (ret < 0) {
 289                 PERROR("failed to set open files limit");
 290         }
 291 }
 292
 293 /*
 294  * main
 295  */
 296 int main(int argc, char **argv)
 297 {
 298         int ret = 0, retval = 0;
 299         void *status;
 300         struct lttng_consumer_local_data *tmp_ctx;
 301
 302         rcu_register_thread();
 303
 304         if (run_as_create_worker(argv[0], NULL, NULL) < 0) {
 305                 goto exit_set_signal_handler;
 306         }
 307
 308         if (set_signal_handler()) {
 309                 retval = -1;
 310                 goto exit_set_signal_handler;
 311         }
 312
 313         /* Parse arguments */
 314         progname = argv[0];
 315         if (parse_args(argc, argv)) {
 316                 retval = -1;
 317                 goto exit_options;
 318         }
 319
 320         /* Daemonize */
 321         if (opt_daemon) {
 322                 int i;
 323
 324                 /*
 325                  * fork
 326                  * child: setsid, close FD 0, 1, 2, chdir /
 327                  * parent: exit (if fork is successful)
 328                  */
 329                 ret = daemon(0, 0);
 330                 if (ret < 0) {
 331                         PERROR("daemon");
 332                         retval = -1;
 333                         goto exit_options;
 334                 }
 335                 /*
 336                  * We are in the child. Make sure all other file
 337                  * descriptors are closed, in case we are called with
 338                  * more opened file descriptors than the standard ones.
 339                  */
 340                 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
 341                         (void) close(i);
 342                 }
 343         }
 344
 345         /*
 346          * Starting from here, we can create threads. This needs to be after
 347          * lttng_daemonize due to RCU.
 348          */
 349
 350         health_consumerd = health_app_create(NR_HEALTH_CONSUMERD_TYPES);
 351         if (!health_consumerd) {
 352                 retval = -1;
 353                 goto exit_health_consumerd_cleanup;
 354         }
 355
 356         if (*command_sock_path == '\0') {
 357                 switch (opt_type) {
 358                 case LTTNG_CONSUMER_KERNEL:
 359                         ret = snprintf(command_sock_path, PATH_MAX,
 360                                         DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
 361                                         DEFAULT_LTTNG_RUNDIR);
 362                         if (ret < 0) {
 363                                 retval = -1;
 364                                 goto exit_init_data;
 365                         }
 366                         break;
 367                 case LTTNG_CONSUMER64_UST:
 368                         ret = snprintf(command_sock_path, PATH_MAX,
 369                                         DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
 370                                         DEFAULT_LTTNG_RUNDIR);
 371                         if (ret < 0) {
 372                                 retval = -1;
 373                                 goto exit_init_data;
 374                         }
 375                         break;
 376                 case LTTNG_CONSUMER32_UST:
 377                         ret = snprintf(command_sock_path, PATH_MAX,
 378                                         DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
 379                                         DEFAULT_LTTNG_RUNDIR);
 380                         if (ret < 0) {
 381                                 retval = -1;
 382                                 goto exit_init_data;
 383                         }
 384                         break;
 385                 default:
 386                         ERR("Unknown consumerd type");
 387                         retval = -1;
 388                         goto exit_init_data;
 389                 }
 390         }
 391
 392         /* Init */
 393         if (lttng_consumer_init()) {
 394                 retval = -1;
 395                 goto exit_init_data;
 396         }
 397
 398         /* Initialize communication library */
 399         lttcomm_init();
 400         /* Initialize TCP timeout values */
 401         lttcomm_inet_init();
 402
 403         if (!getuid()) {
 404                 /* Set limit for open files */
 405                 set_ulimit();
 406         }
 407
 408         /* create the consumer instance with and assign the callbacks */
 409         ctx = lttng_consumer_create(opt_type, lttng_consumer_read_subbuffer,
 410                 NULL, lttng_consumer_on_recv_stream, NULL);
 411         if (!ctx) {
 412                 retval = -1;
 413                 goto exit_init_data;
 414         }
 415
 416         lttng_consumer_set_command_sock_path(ctx, command_sock_path);
 417         if (*error_sock_path == '\0') {
 418                 switch (opt_type) {
 419                 case LTTNG_CONSUMER_KERNEL:
 420                         ret = snprintf(error_sock_path, PATH_MAX,
 421                                         DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
 422                                         DEFAULT_LTTNG_RUNDIR);
 423                         if (ret < 0) {
 424                                 retval = -1;
 425                                 goto exit_init_data;
 426                         }
 427                         break;
 428                 case LTTNG_CONSUMER64_UST:
 429                         ret = snprintf(error_sock_path, PATH_MAX,
 430                                         DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
 431                                         DEFAULT_LTTNG_RUNDIR);
 432                         if (ret < 0) {
 433                                 retval = -1;
 434                                 goto exit_init_data;
 435                         }
 436                         break;
 437                 case LTTNG_CONSUMER32_UST:
 438                         ret = snprintf(error_sock_path, PATH_MAX,
 439                                         DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
 440                                         DEFAULT_LTTNG_RUNDIR);
 441                         if (ret < 0) {
 442                                 retval = -1;
 443                                 goto exit_init_data;
 444                         }
 445                         break;
 446                 default:
 447                         ERR("Unknown consumerd type");
 448                         retval = -1;
 449                         goto exit_init_data;
 450                 }
 451         }
 452
 453         /* Connect to the socket created by lttng-sessiond to report errors */
 454         DBG("Connecting to error socket %s", error_sock_path);
 455         ret = lttcomm_connect_unix_sock(error_sock_path);
 456         /*
 457          * Not a fatal error, but all communication with lttng-sessiond will
 458          * fail.
 459          */
 460         if (ret < 0) {
 461                 WARN("Cannot connect to error socket (is lttng-sessiond started?)");
 462         }
 463         lttng_consumer_set_error_sock(ctx, ret);
 464
 465         /*
 466          * Block RT signals used for UST periodical metadata flush and the live
 467          * timer in main, and create a dedicated thread to handle these signals.
 468          */
 469         if (consumer_signal_init()) {
 470                 retval = -1;
 471                 goto exit_init_data;
 472         }
 473
 474         ctx->type = opt_type;
 475
 476         if (utils_create_pipe(health_quit_pipe)) {
 477                 retval = -1;
 478                 goto exit_health_pipe;
 479         }
 480
 481         /* Create thread to manage the client socket */
 482         ret = pthread_create(&health_thread, default_pthread_attr(),
 483                         thread_manage_health, (void *) NULL);
 484         if (ret) {
 485                 errno = ret;
 486                 PERROR("pthread_create health");
 487                 retval = -1;
 488                 goto exit_health_thread;
 489         }
 490
 491         /*
 492          * Wait for health thread to be initialized before letting the
 493          * sessiond thread reply to the sessiond that we are ready.
 494          */
 495         while (uatomic_read(&lttng_consumer_ready)) {
 496                 usleep(100000);
 497         }
 498         cmm_smp_mb();   /* Read ready before following operations */
 499
 500         /*
 501          * Create the thread to manage the UST metadata periodic timer and
 502          * live timer.
 503          */
 504         ret = pthread_create(&metadata_timer_thread, NULL,
 505                         consumer_timer_thread, (void *) ctx);
 506         if (ret) {
 507                 errno = ret;
 508                 PERROR("pthread_create");
 509                 retval = -1;
 510                 goto exit_metadata_timer_thread;
 511         }
 512         metadata_timer_thread_online = true;
 513
 514         /* Create thread to manage channels */
 515         ret = pthread_create(&channel_thread, default_pthread_attr(),
 516                         consumer_thread_channel_poll,
 517                         (void *) ctx);
 518         if (ret) {
 519                 errno = ret;
 520                 PERROR("pthread_create");
 521                 retval = -1;
 522                 goto exit_channel_thread;
 523         }
 524
 525         /* Create thread to manage the polling/writing of trace metadata */
 526         ret = pthread_create(&metadata_thread, default_pthread_attr(),
 527                         consumer_thread_metadata_poll,
 528                         (void *) ctx);
 529         if (ret) {
 530                 errno = ret;
 531                 PERROR("pthread_create");
 532                 retval = -1;
 533                 goto exit_metadata_thread;
 534         }
 535
 536         /* Create thread to manage the polling/writing of trace data */
 537         ret = pthread_create(&data_thread, default_pthread_attr(),
 538                         consumer_thread_data_poll, (void *) ctx);
 539         if (ret) {
 540                 errno = ret;
 541                 PERROR("pthread_create");
 542                 retval = -1;
 543                 goto exit_data_thread;
 544         }
 545
 546         /* Create the thread to manage the reception of fds */
 547         ret = pthread_create(&sessiond_thread, default_pthread_attr(),
 548                         consumer_thread_sessiond_poll,
 549                         (void *) ctx);
 550         if (ret) {
 551                 errno = ret;
 552                 PERROR("pthread_create");
 553                 retval = -1;
 554                 goto exit_sessiond_thread;
 555         }
 556
 557
 558         /*
 559          * This is where we start awaiting program completion (e.g. through
 560          * signal that asks threads to teardown.
 561          */
 562
 563         ret = pthread_join(sessiond_thread, &status);
 564         if (ret) {
 565                 errno = ret;
 566                 PERROR("pthread_join sessiond_thread");
 567                 retval = -1;
 568         }
 569 exit_sessiond_thread:
 570
 571         ret = pthread_join(data_thread, &status);
 572         if (ret) {
 573                 errno = ret;
 574                 PERROR("pthread_join data_thread");
 575                 retval = -1;
 576         }
 577 exit_data_thread:
 578
 579         ret = pthread_join(metadata_thread, &status);
 580         if (ret) {
 581                 errno = ret;
 582                 PERROR("pthread_join metadata_thread");
 583                 retval = -1;
 584         }
 585 exit_metadata_thread:
 586
 587         ret = pthread_join(channel_thread, &status);
 588         if (ret) {
 589                 errno = ret;
 590                 PERROR("pthread_join channel_thread");
 591                 retval = -1;
 592         }
 593 exit_channel_thread:
 594
 595 exit_metadata_timer_thread:
 596
 597         ret = pthread_join(health_thread, &status);
 598         if (ret) {
 599                 errno = ret;
 600                 PERROR("pthread_join health_thread");
 601                 retval = -1;
 602         }
 603 exit_health_thread:
 604
 605         utils_close_pipe(health_quit_pipe);
 606 exit_health_pipe:
 607
 608 exit_init_data:
 609         /*
 610          * Wait for all pending call_rcu work to complete before tearing
 611          * down data structures. call_rcu worker may be trying to
 612          * perform lookups in those structures.
 613          */
 614         rcu_barrier();
 615         lttng_consumer_cleanup();
 616         /*
 617          * Tearing down the metadata timer thread in a
 618          * non-fully-symmetric fashion compared to its creation in case
 619          * lttng_consumer_cleanup() ends up tearing down timers (which
 620          * requires the timer thread to be alive).
 621          */
 622         if (metadata_timer_thread_online) {
 623                 /*
 624                  * Ensure the metadata timer thread exits only after all other
 625                  * threads are gone, because it is required to perform timer
 626                  * teardown synchronization.
 627                  */
 628                 kill(getpid(), LTTNG_CONSUMER_SIG_EXIT);
 629                 ret = pthread_join(metadata_timer_thread, &status);
 630                 if (ret) {
 631                         errno = ret;
 632                         PERROR("pthread_join metadata_timer_thread");
 633                         retval = -1;
 634                 }
 635                 ret = consumer_timer_thread_get_channel_monitor_pipe();
 636                 if (ret >= 0) {
 637                         ret = close(ret);
 638                         if (ret) {
 639                                 PERROR("close channel monitor pipe");
 640                         }
 641                 }
 642                 metadata_timer_thread_online = false;
 643         }
 644         tmp_ctx = ctx;
 645         ctx = NULL;
 646         cmm_barrier();  /* Clear ctx for signal handler. */
 647         lttng_consumer_destroy(tmp_ctx);
 648
 649         if (health_consumerd) {
 650                 health_app_destroy(health_consumerd);
 651         }
 652         /* Ensure all prior call_rcu are done. */
 653         rcu_barrier();
 654
 655         run_as_destroy_worker();
 656
 657 exit_health_consumerd_cleanup:
 658 exit_options:
 659 exit_set_signal_handler:
 660
 661         rcu_unregister_thread();
 662
 663         if (!retval) {
 664                 exit(EXIT_SUCCESS);
 665         } else {
 666                 exit(EXIT_FAILURE);
 667         }
 668 }