src/bin/lttng-sessiond/agent-thread.cpp

   1 /*
   2  * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
   3  *
   4  * SPDX-License-Identifier: GPL-2.0-only
   5  *
   6  */
   7
   8 #define _LGPL_SOURCE
   9
  10 #include <common/common.hpp>
  11 #include <common/sessiond-comm/sessiond-comm.hpp>
  12 #include <common/uri.hpp>
  13 #include <common/utils.hpp>
  14
  15 #include <common/compat/endian.hpp>
  16
  17 #include "fd-limit.hpp"
  18 #include "agent-thread.hpp"
  19 #include "agent.hpp"
  20 #include "lttng-sessiond.hpp"
  21 #include "session.hpp"
  22 #include "utils.hpp"
  23 #include "thread.hpp"
  24
  25 struct thread_notifiers {
  26         struct lttng_pipe *quit_pipe;
  27         sem_t ready;
  28 };
  29
  30 struct agent_app_id {
  31         pid_t pid;
  32         enum lttng_domain_type domain;
  33 };
  34
  35 struct agent_protocol_version {
  36         unsigned int major, minor;
  37 };
  38
  39 static int agent_tracing_enabled = -1;
  40
  41 /*
  42  * Note that there is not port here. It's set after this URI is parsed so we
  43  * can let the user define a custom one. However, localhost is ALWAYS the
  44  * default listening address.
  45  */
  46 static const char *default_reg_uri =
  47         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS;
  48
  49 /*
  50  * Update agent application using the given socket. This is done just after
  51  * registration was successful.
  52  *
  53  * This will acquire the various sessions' lock; none must be held by the
  54  * caller.
  55  * The caller must hold the session list lock.
  56  */
  57 static void update_agent_app(const struct agent_app *app)
  58 {
  59         struct ltt_session *session, *stmp;
  60         struct ltt_session_list *list;
  61         struct agent *trigger_agent;
  62         struct lttng_ht_iter iter;
  63
  64         list = session_get_list();
  65         LTTNG_ASSERT(list);
  66
  67         cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
  68                 if (!session_get(session)) {
  69                         continue;
  70                 }
  71
  72                 session_lock(session);
  73                 if (session->ust_session) {
  74                         const struct agent *agt;
  75
  76                         rcu_read_lock();
  77                         agt = trace_ust_find_agent(session->ust_session, app->domain);
  78                         if (agt) {
  79                                 agent_update(agt, app);
  80                         }
  81                         rcu_read_unlock();
  82                 }
  83                 session_unlock(session);
  84                 session_put(session);
  85         }
  86
  87         rcu_read_lock();
  88         /*
  89          * We are protected against the addition of new events by the session
  90          * list lock being held.
  91          */
  92         cds_lfht_for_each_entry(the_trigger_agents_ht_by_domain->ht,
  93                         &iter.iter, trigger_agent, node.node) {
  94                 agent_update(trigger_agent, app);
  95         }
  96         rcu_read_unlock();
  97 }
  98
  99 /*
 100  * Create and init socket from uri.
 101  */
 102 static struct lttcomm_sock *init_tcp_socket(void)
 103 {
 104         int ret;
 105         struct lttng_uri *uri = NULL;
 106         struct lttcomm_sock *sock = NULL;
 107         unsigned int port;
 108         bool bind_succeeded = false;
 109
 110         /*
 111          * This should never fail since the URI is hardcoded and the port is set
 112          * before this thread is launched.
 113          */
 114         ret = uri_parse(default_reg_uri, &uri);
 115         LTTNG_ASSERT(ret);
 116         LTTNG_ASSERT(the_config.agent_tcp_port.begin > 0);
 117         uri->port = the_config.agent_tcp_port.begin;
 118
 119         sock = lttcomm_alloc_sock_from_uri(uri);
 120         uri_free(uri);
 121         if (sock == NULL) {
 122                 ERR("agent allocating TCP socket");
 123                 goto error;
 124         }
 125
 126         ret = lttcomm_create_sock(sock);
 127         if (ret < 0) {
 128                 goto error;
 129         }
 130
 131         for (port = the_config.agent_tcp_port.begin;
 132                         port <= the_config.agent_tcp_port.end; port++) {
 133                 ret = lttcomm_sock_set_port(sock, (uint16_t) port);
 134                 if (ret) {
 135                         ERR("Failed to set port %u on socket",
 136                                         port);
 137                         goto error;
 138                 }
 139                 DBG3("Trying to bind on port %u", port);
 140                 ret = sock->ops->bind(sock);
 141                 if (!ret) {
 142                         bind_succeeded = true;
 143                         break;
 144                 }
 145
 146                 if (errno == EADDRINUSE) {
 147                         DBG("Failed to bind to port %u since it is already in use",
 148                                         port);
 149                 } else {
 150                         PERROR("Failed to bind to port %u", port);
 151                         goto error;
 152                 }
 153         }
 154
 155         if (!bind_succeeded) {
 156                 if (the_config.agent_tcp_port.begin ==
 157                                 the_config.agent_tcp_port.end) {
 158                         WARN("Another process is already using the agent port %i. "
 159                              "Agent support will be deactivated.",
 160                                         the_config.agent_tcp_port.begin);
 161                         goto error;
 162                 } else {
 163                         WARN("All ports in the range [%i, %i] are already in use. "
 164                              "Agent support will be deactivated.",
 165                                         the_config.agent_tcp_port.begin,
 166                                         the_config.agent_tcp_port.end);
 167                         goto error;
 168                 }
 169         }
 170
 171         ret = sock->ops->listen(sock, -1);
 172         if (ret < 0) {
 173                 goto error;
 174         }
 175
 176         DBG("Listening on TCP port %u and socket %d",
 177                         port, sock->fd);
 178
 179         return sock;
 180
 181 error:
 182         if (sock) {
 183                 lttcomm_destroy_sock(sock);
 184         }
 185         return NULL;
 186 }
 187
 188 /*
 189  * Close and destroy the given TCP socket.
 190  */
 191 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 192 {
 193         int ret;
 194         uint16_t port;
 195
 196         LTTNG_ASSERT(sock);
 197
 198         ret = lttcomm_sock_get_port(sock, &port);
 199         if (ret) {
 200                 ERR("Failed to get port of agent TCP socket");
 201                 port = 0;
 202         }
 203
 204         DBG3("Destroy TCP socket on port %" PRIu16,
 205                         port);
 206
 207         /* This will return gracefully if fd is invalid. */
 208         sock->ops->close(sock);
 209         lttcomm_destroy_sock(sock);
 210 }
 211
 212 static const char *domain_type_str(enum lttng_domain_type domain_type)
 213 {
 214         switch (domain_type) {
 215         case LTTNG_DOMAIN_NONE:
 216                 return "none";
 217         case LTTNG_DOMAIN_KERNEL:
 218                 return "kernel";
 219         case LTTNG_DOMAIN_UST:
 220                 return "ust";
 221         case LTTNG_DOMAIN_JUL:
 222                 return "jul";
 223         case LTTNG_DOMAIN_LOG4J:
 224                 return "log4j";
 225         case LTTNG_DOMAIN_PYTHON:
 226                 return "python";
 227         default:
 228                 return "unknown";
 229         }
 230 }
 231
 232 static bool is_agent_protocol_version_supported(
 233                 const struct agent_protocol_version *version)
 234 {
 235         const bool is_supported = version->major == AGENT_MAJOR_VERSION &&
 236                         version->minor == AGENT_MINOR_VERSION;
 237
 238         if (!is_supported) {
 239                 WARN("Refusing agent connection: unsupported protocol version %ui.%ui, expected %i.%i",
 240                                 version->major, version->minor,
 241                                 AGENT_MAJOR_VERSION, AGENT_MINOR_VERSION);
 242         }
 243
 244         return is_supported;
 245 }
 246
 247 /*
 248  * Handle a new agent connection on the registration socket.
 249  *
 250  * Returns 0 on success, or else a negative errno value.
 251  * On success, the resulting socket is returned through `agent_app_socket`
 252  * and the application's reported id is updated through `agent_app_id`.
 253  */
 254 static int accept_agent_connection(
 255                 struct lttcomm_sock *reg_sock,
 256                 struct agent_app_id *agent_app_id,
 257                 struct lttcomm_sock **agent_app_socket)
 258 {
 259         int ret;
 260         struct agent_protocol_version agent_version;
 261         ssize_t size;
 262         struct agent_register_msg msg;
 263         struct lttcomm_sock *new_sock;
 264
 265         LTTNG_ASSERT(reg_sock);
 266
 267         new_sock = reg_sock->ops->accept(reg_sock);
 268         if (!new_sock) {
 269                 ret = -ENOTCONN;
 270                 goto end;
 271         }
 272
 273         size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
 274         if (size < sizeof(msg)) {
 275                 if (size < 0) {
 276                         PERROR("Failed to register new agent application");
 277                 } else if (size != 0) {
 278                         ERR("Failed to register new agent application: invalid registration message length: expected length = %zu, message length = %zd",
 279                                         sizeof(msg), size);
 280                 } else {
 281                         DBG("Failed to register new agent application: connection closed");
 282                 }
 283                 ret = -EINVAL;
 284                 goto error_close_socket;
 285         }
 286
 287         agent_version = (struct agent_protocol_version) {
 288                 be32toh(msg.major_version),
 289                 be32toh(msg.minor_version),
 290         };
 291
 292         /* Test communication protocol version of the registering agent. */
 293         if (!is_agent_protocol_version_supported(&agent_version)) {
 294                 ret = -EINVAL;
 295                 goto error_close_socket;
 296         }
 297
 298         *agent_app_id = (struct agent_app_id) {
 299                 .pid = (pid_t) be32toh(msg.pid),
 300                 .domain = (lttng_domain_type) be32toh(msg.domain),
 301         };
 302
 303         DBG2("New registration for agent application: pid = %ld, domain = %s, socket fd = %d",
 304                         (long) agent_app_id->pid,
 305                         domain_type_str(agent_app_id->domain), new_sock->fd);
 306
 307         *agent_app_socket = new_sock;
 308         new_sock = NULL;
 309         ret = 0;
 310         goto end;
 311
 312 error_close_socket:
 313         new_sock->ops->close(new_sock);
 314         lttcomm_destroy_sock(new_sock);
 315 end:
 316         return ret;
 317 }
 318
 319 bool agent_tracing_is_enabled(void)
 320 {
 321         int enabled;
 322
 323         enabled = uatomic_read(&agent_tracing_enabled);
 324         LTTNG_ASSERT(enabled != -1);
 325         return enabled == 1;
 326 }
 327
 328 /*
 329  * Write agent TCP port using the rundir.
 330  */
 331 static int write_agent_port(uint16_t port)
 332 {
 333         return utils_create_pid_file(
 334                         (pid_t) port, the_config.agent_port_file_path.value);
 335 }
 336
 337 static
 338 void mark_thread_as_ready(struct thread_notifiers *notifiers)
 339 {
 340         DBG("Marking agent management thread as ready");
 341         sem_post(&notifiers->ready);
 342 }
 343
 344 static
 345 void wait_until_thread_is_ready(struct thread_notifiers *notifiers)
 346 {
 347         DBG("Waiting for agent management thread to be ready");
 348         sem_wait(&notifiers->ready);
 349         DBG("Agent management thread is ready");
 350 }
 351
 352 /*
 353  * This thread manage application notify communication.
 354  */
 355 static void *thread_agent_management(void *data)
 356 {
 357         int i, ret, pollfd;
 358         uint32_t revents, nb_fd;
 359         struct lttng_poll_event events;
 360         struct lttcomm_sock *reg_sock;
 361         struct thread_notifiers *notifiers = (thread_notifiers *) data;
 362         const int quit_pipe_read_fd = lttng_pipe_get_readfd(
 363                         notifiers->quit_pipe);
 364
 365         DBG("Manage agent application registration.");
 366
 367         rcu_register_thread();
 368         rcu_thread_online();
 369
 370         /* Agent initialization call MUST be called before starting the thread. */
 371         LTTNG_ASSERT(the_agent_apps_ht_by_sock);
 372
 373         /* Create pollset with size 2, quit pipe and registration socket. */
 374         ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
 375         if (ret < 0) {
 376                 goto error_poll_create;
 377         }
 378
 379         ret = lttng_poll_add(&events, quit_pipe_read_fd,
 380                         LPOLLIN | LPOLLERR);
 381         if (ret < 0) {
 382                 goto error_tcp_socket;
 383         }
 384
 385         reg_sock = init_tcp_socket();
 386         if (reg_sock) {
 387                 uint16_t port;
 388
 389                 ret = lttcomm_sock_get_port(reg_sock, &port);
 390                 LTTNG_ASSERT(ret == 0);
 391
 392                 ret = write_agent_port(port);
 393                 if (ret) {
 394                         ERR("Failed to create agent port file: agent tracing will be unavailable");
 395                         /* Don't prevent the launch of the sessiond on error. */
 396                         mark_thread_as_ready(notifiers);
 397                         goto error;
 398                 }
 399         } else {
 400                 /* Don't prevent the launch of the sessiond on error. */
 401                 mark_thread_as_ready(notifiers);
 402                 goto error_tcp_socket;
 403         }
 404
 405         /*
 406          * Signal that the agent thread is ready. The command thread
 407          * may start to query whether or not agent tracing is enabled.
 408          */
 409         uatomic_set(&agent_tracing_enabled, 1);
 410         mark_thread_as_ready(notifiers);
 411
 412         /* Add TCP socket to the poll set. */
 413         ret = lttng_poll_add(&events, reg_sock->fd,
 414                         LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 415         if (ret < 0) {
 416                 goto error;
 417         }
 418
 419         while (1) {
 420                 DBG3("Manage agent polling");
 421
 422                 /* Inifinite blocking call, waiting for transmission */
 423 restart:
 424                 ret = lttng_poll_wait(&events, -1);
 425                 DBG3("Manage agent return from poll on %d fds",
 426                                 LTTNG_POLL_GETNB(&events));
 427                 if (ret < 0) {
 428                         /*
 429                          * Restart interrupted system call.
 430                          */
 431                         if (errno == EINTR) {
 432                                 goto restart;
 433                         }
 434                         goto error;
 435                 }
 436                 nb_fd = ret;
 437                 DBG3("%d fd ready", nb_fd);
 438
 439                 for (i = 0; i < nb_fd; i++) {
 440                         /* Fetch once the poll data */
 441                         revents = LTTNG_POLL_GETEV(&events, i);
 442                         pollfd = LTTNG_POLL_GETFD(&events, i);
 443
 444                         /* Thread quit pipe has been closed. Killing thread. */
 445                         if (pollfd == quit_pipe_read_fd) {
 446                                 goto exit;
 447                         }
 448
 449                         /* Activity on the registration socket. */
 450                         if (revents & LPOLLIN) {
 451                                 struct agent_app_id new_app_id;
 452                                 struct agent_app *new_app = NULL;
 453                                 struct lttcomm_sock *new_app_socket;
 454                                 int new_app_socket_fd;
 455
 456                                 LTTNG_ASSERT(pollfd == reg_sock->fd);
 457
 458                                 ret = accept_agent_connection(
 459                                         reg_sock, &new_app_id, &new_app_socket);
 460                                 if (ret < 0) {
 461                                         /* Errors are already logged. */
 462                                         continue;
 463                                 }
 464
 465                                 /*
 466                                  * new_app_socket's ownership has been
 467                                  * transferred to the new agent app.
 468                                  */
 469                                 new_app = agent_create_app(new_app_id.pid,
 470                                                 new_app_id.domain,
 471                                                 new_app_socket);
 472                                 if (!new_app) {
 473                                         new_app_socket->ops->close(
 474                                                         new_app_socket);
 475                                         continue;
 476                                 }
 477                                 new_app_socket_fd = new_app_socket->fd;
 478                                 new_app_socket = NULL;
 479
 480                                 /*
 481                                  * Since this is a command socket (write then
 482                                  * read), only add poll error event to only
 483                                  * detect shutdown.
 484                                  */
 485                                 ret = lttng_poll_add(&events, new_app_socket_fd,
 486                                                 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 487                                 if (ret < 0) {
 488                                         agent_destroy_app(new_app);
 489                                         continue;
 490                                 }
 491
 492                                 /*
 493                                  * Prevent sessions from being modified while
 494                                  * the agent application's configuration is
 495                                  * updated.
 496                                  */
 497                                 session_lock_list();
 498
 499                                 /*
 500                                  * Update the newly registered applications's
 501                                  * configuration.
 502                                  */
 503                                 update_agent_app(new_app);
 504
 505                                 ret = agent_send_registration_done(new_app);
 506                                 if (ret < 0) {
 507                                         agent_destroy_app(new_app);
 508                                         /* Removing from the poll set. */
 509                                         ret = lttng_poll_del(&events,
 510                                                         new_app_socket_fd);
 511                                         if (ret < 0) {
 512                                                 session_unlock_list();
 513                                                 goto error;
 514                                         }
 515                                         continue;
 516                                 }
 517
 518                                 /* Publish the new agent app. */
 519                                 agent_add_app(new_app);
 520
 521                                 session_unlock_list();
 522                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 523                                 /* Removing from the poll set */
 524                                 ret = lttng_poll_del(&events, pollfd);
 525                                 if (ret < 0) {
 526                                         goto error;
 527                                 }
 528                                 agent_destroy_app_by_sock(pollfd);
 529                         } else {
 530                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
 531                                 goto error;
 532                         }
 533                 }
 534         }
 535
 536 exit:
 537         /* Whatever happens, try to delete it and exit. */
 538         (void) lttng_poll_del(&events, reg_sock->fd);
 539 error:
 540         destroy_tcp_socket(reg_sock);
 541 error_tcp_socket:
 542         lttng_poll_clean(&events);
 543 error_poll_create:
 544         uatomic_set(&agent_tracing_enabled, 0);
 545         DBG("Cleaning up and stopping.");
 546         rcu_thread_offline();
 547         rcu_unregister_thread();
 548         return NULL;
 549 }
 550
 551 static bool shutdown_agent_management_thread(void *data)
 552 {
 553         struct thread_notifiers *notifiers = (thread_notifiers *) data;
 554         const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe);
 555
 556         return notify_thread_pipe(write_fd) == 1;
 557 }
 558
 559 static void cleanup_agent_management_thread(void *data)
 560 {
 561         struct thread_notifiers *notifiers = (thread_notifiers *) data;
 562
 563         lttng_pipe_destroy(notifiers->quit_pipe);
 564         sem_destroy(&notifiers->ready);
 565         free(notifiers);
 566 }
 567
 568 bool launch_agent_management_thread(void)
 569 {
 570         struct thread_notifiers *notifiers;
 571         struct lttng_thread *thread;
 572
 573         notifiers = (thread_notifiers *) zmalloc(sizeof(*notifiers));
 574         if (!notifiers) {
 575                 goto error_alloc;
 576         }
 577
 578         sem_init(&notifiers->ready, 0, 0);
 579         notifiers->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
 580         if (!notifiers->quit_pipe) {
 581                 goto error;
 582         }
 583         thread = lttng_thread_create("Agent management",
 584                         thread_agent_management,
 585                         shutdown_agent_management_thread,
 586                         cleanup_agent_management_thread,
 587                         notifiers);
 588         if (!thread) {
 589                 goto error;
 590         }
 591         wait_until_thread_is_ready(notifiers);
 592         lttng_thread_put(thread);
 593         return true;
 594 error:
 595         cleanup_agent_management_thread(notifiers);
 596 error_alloc:
 597         return false;
 598 }