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

   1 /*
   2  * Copyright (C) 2013 - David Goulet <dgoulet@efficios.com>
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License, version 2 only, as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 51
  15  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _LGPL_SOURCE
  19 #include <assert.h>
  20
  21 #include <common/common.h>
  22 #include <common/sessiond-comm/sessiond-comm.h>
  23 #include <common/uri.h>
  24 #include <common/utils.h>
  25
  26 #include <common/compat/endian.h>
  27
  28 #include "fd-limit.h"
  29 #include "agent-thread.h"
  30 #include "agent.h"
  31 #include "lttng-sessiond.h"
  32 #include "session.h"
  33 #include "utils.h"
  34
  35 static int agent_tracing_enabled = -1;
  36
  37 /*
  38  * Note that there is not port here. It's set after this URI is parsed so we
  39  * can let the user define a custom one. However, localhost is ALWAYS the
  40  * default listening address.
  41  */
  42 static const char *default_reg_uri =
  43         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS;
  44
  45 /*
  46  * Update agent application using the given socket. This is done just after
  47  * registration was successful.
  48  *
  49  * This is a quite heavy call in terms of locking since the session list lock
  50  * AND session lock are acquired.
  51  */
  52 static void update_agent_app(struct agent_app *app)
  53 {
  54         struct ltt_session *session, *stmp;
  55         struct ltt_session_list *list;
  56
  57         list = session_get_list();
  58         assert(list);
  59
  60         session_lock_list();
  61         cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
  62                 if (!session_get(session)) {
  63                         continue;
  64                 }
  65
  66                 session_lock(session);
  67                 if (session->ust_session) {
  68                         struct agent *agt;
  69
  70                         rcu_read_lock();
  71                         agt = trace_ust_find_agent(session->ust_session, app->domain);
  72                         if (agt) {
  73                                 agent_update(agt, app->sock->fd);
  74                         }
  75                         rcu_read_unlock();
  76                 }
  77                 session_unlock(session);
  78                 session_put(session);
  79         }
  80         session_unlock_list();
  81 }
  82
  83 /*
  84  * Create and init socket from uri.
  85  */
  86 static struct lttcomm_sock *init_tcp_socket(void)
  87 {
  88         int ret;
  89         struct lttng_uri *uri = NULL;
  90         struct lttcomm_sock *sock = NULL;
  91         unsigned int port;
  92         bool bind_succeeded = false;
  93
  94         /*
  95          * This should never fail since the URI is hardcoded and the port is set
  96          * before this thread is launched.
  97          */
  98         ret = uri_parse(default_reg_uri, &uri);
  99         assert(ret);
 100         assert(config.agent_tcp_port.begin > 0);
 101         uri->port = config.agent_tcp_port.begin;
 102
 103         sock = lttcomm_alloc_sock_from_uri(uri);
 104         uri_free(uri);
 105         if (sock == NULL) {
 106                 ERR("[agent-thread] agent allocating TCP socket");
 107                 goto error;
 108         }
 109
 110         ret = lttcomm_create_sock(sock);
 111         if (ret < 0) {
 112                 goto error;
 113         }
 114
 115         for (port = config.agent_tcp_port.begin;
 116                         port <= config.agent_tcp_port.end; port++) {
 117                 ret = lttcomm_sock_set_port(sock, (uint16_t) port);
 118                 if (ret) {
 119                         ERR("[agent-thread] Failed to set port %u on socket",
 120                                         port);
 121                         goto error;
 122                 }
 123                 DBG3("[agent-thread] Trying to bind on port %u", port);
 124                 ret = sock->ops->bind(sock);
 125                 if (!ret) {
 126                         bind_succeeded = true;
 127                         break;
 128                 }
 129
 130                 if (errno == EADDRINUSE) {
 131                         DBG("Failed to bind to port %u since it is already in use",
 132                                         port);
 133                 } else {
 134                         PERROR("Failed to bind to port %u", port);
 135                         goto error;
 136                 }
 137         }
 138
 139         if (!bind_succeeded) {
 140                 if (config.agent_tcp_port.begin == config.agent_tcp_port.end) {
 141                         WARN("Another process is already using the agent port %i. "
 142                                         "Agent support will be deactivated.",
 143                                         config.agent_tcp_port.begin);
 144                         goto error;
 145                 } else {
 146                         WARN("All ports in the range [%i, %i] are already in use. "
 147                                         "Agent support will be deactivated.",
 148                                         config.agent_tcp_port.begin,
 149                                         config.agent_tcp_port.end);
 150                         goto error;
 151                 }
 152         }
 153
 154         ret = sock->ops->listen(sock, -1);
 155         if (ret < 0) {
 156                 goto error;
 157         }
 158
 159         DBG("[agent-thread] Listening on TCP port %u and socket %d",
 160                         port, sock->fd);
 161
 162         return sock;
 163
 164 error:
 165         if (sock) {
 166                 lttcomm_destroy_sock(sock);
 167         }
 168         return NULL;
 169 }
 170
 171 /*
 172  * Close and destroy the given TCP socket.
 173  */
 174 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 175 {
 176         int ret;
 177         uint16_t port;
 178
 179         assert(sock);
 180
 181         ret = lttcomm_sock_get_port(sock, &port);
 182         if (ret) {
 183                 ERR("[agent-thread] Failed to get port of agent TCP socket");
 184                 port = 0;
 185         }
 186
 187         DBG3("[agent-thread] Destroy TCP socket on port %" PRIu16,
 188                         port);
 189
 190         /* This will return gracefully if fd is invalid. */
 191         sock->ops->close(sock);
 192         lttcomm_destroy_sock(sock);
 193 }
 194
 195 /*
 196  * Handle a new agent registration using the reg socket. After that, a new
 197  * agent application is added to the global hash table and attach to an UST app
 198  * object. If r_app is not NULL, the created app is set to the pointer.
 199  *
 200  * Return the new FD created upon accept() on success or else a negative errno
 201  * value.
 202  */
 203 static int handle_registration(struct lttcomm_sock *reg_sock,
 204                 struct agent_app **r_app)
 205 {
 206         int ret;
 207         pid_t pid;
 208         uint32_t major_version, minor_version;
 209         ssize_t size;
 210         enum lttng_domain_type domain;
 211         struct agent_app *app;
 212         struct agent_register_msg msg;
 213         struct lttcomm_sock *new_sock;
 214
 215         assert(reg_sock);
 216
 217         new_sock = reg_sock->ops->accept(reg_sock);
 218         if (!new_sock) {
 219                 ret = -ENOTCONN;
 220                 goto error;
 221         }
 222
 223         size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
 224         if (size < sizeof(msg)) {
 225                 ret = -EINVAL;
 226                 goto error_socket;
 227         }
 228         domain = be32toh(msg.domain);
 229         pid = be32toh(msg.pid);
 230         major_version = be32toh(msg.major_version);
 231         minor_version = be32toh(msg.minor_version);
 232
 233         /* Test communication protocol version of the registring agent. */
 234         if (major_version != AGENT_MAJOR_VERSION) {
 235                 ret = -EINVAL;
 236                 goto error_socket;
 237         }
 238         if (minor_version != AGENT_MINOR_VERSION) {
 239                 ret = -EINVAL;
 240                 goto error_socket;
 241         }
 242
 243         DBG2("[agent-thread] New registration for pid %d domain %d on socket %d",
 244                         pid, domain, new_sock->fd);
 245
 246         app = agent_create_app(pid, domain, new_sock);
 247         if (!app) {
 248                 ret = -ENOMEM;
 249                 goto error_socket;
 250         }
 251
 252         /*
 253          * Add before assigning the socket value to the UST app so it can be found
 254          * concurrently.
 255          */
 256         agent_add_app(app);
 257
 258         /*
 259          * We don't need to attach the agent app to the app. If we ever do so, we
 260          * should consider both registration order of agent before app and app
 261          * before agent.
 262          */
 263
 264         if (r_app) {
 265                 *r_app = app;
 266         }
 267
 268         return new_sock->fd;
 269
 270 error_socket:
 271         new_sock->ops->close(new_sock);
 272         lttcomm_destroy_sock(new_sock);
 273 error:
 274         return ret;
 275 }
 276
 277 bool agent_tracing_is_enabled(void)
 278 {
 279         int enabled;
 280
 281         enabled = uatomic_read(&agent_tracing_enabled);
 282         assert(enabled != -1);
 283         return enabled == 1;
 284 }
 285
 286 /*
 287  * Write agent TCP port using the rundir.
 288  */
 289 static int write_agent_port(uint16_t port)
 290 {
 291         return utils_create_pid_file((pid_t) port,
 292                         config.agent_port_file_path.value);
 293 }
 294
 295 /*
 296  * This thread manage application notify communication.
 297  */
 298 void *agent_thread_manage_registration(void *data)
 299 {
 300         int i, ret, pollfd;
 301         uint32_t revents, nb_fd;
 302         struct lttng_poll_event events;
 303         struct lttcomm_sock *reg_sock;
 304
 305         DBG("[agent-thread] Manage agent application registration.");
 306
 307         rcu_register_thread();
 308         rcu_thread_online();
 309
 310         /* Agent initialization call MUST be called before starting the thread. */
 311         assert(agent_apps_ht_by_sock);
 312
 313         /* Create pollset with size 2, quit pipe and socket. */
 314         ret = sessiond_set_thread_pollset(&events, 2);
 315         if (ret < 0) {
 316                 sessiond_notify_ready();
 317                 goto error_poll_create;
 318         }
 319
 320         reg_sock = init_tcp_socket();
 321         if (reg_sock) {
 322                 uint16_t port;
 323
 324                 assert(lttcomm_sock_get_port(reg_sock, &port) == 0);
 325
 326                 ret = write_agent_port(port);
 327                 if (ret) {
 328                         ERR("[agent-thread] Failed to create agent port file: agent tracing will be unavailable");
 329                         /* Don't prevent the launch of the sessiond on error. */
 330                         sessiond_notify_ready();
 331                         goto error;
 332                 }
 333         } else {
 334                 /* Don't prevent the launch of the sessiond on error. */
 335                 sessiond_notify_ready();
 336                 goto error_tcp_socket;
 337         }
 338
 339         /*
 340          * Signal that the agent thread is ready. The command thread
 341          * may start to query whether or not agent tracing is enabled.
 342          */
 343         uatomic_set(&agent_tracing_enabled, 1);
 344         sessiond_notify_ready();
 345
 346         /* Add TCP socket to poll set. */
 347         ret = lttng_poll_add(&events, reg_sock->fd,
 348                         LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 349         if (ret < 0) {
 350                 goto error;
 351         }
 352
 353         while (1) {
 354                 DBG3("[agent-thread] Manage agent polling");
 355
 356                 /* Inifinite blocking call, waiting for transmission */
 357 restart:
 358                 ret = lttng_poll_wait(&events, -1);
 359                 DBG3("[agent-thread] Manage agent return from poll on %d fds",
 360                                 LTTNG_POLL_GETNB(&events));
 361                 if (ret < 0) {
 362                         /*
 363                          * Restart interrupted system call.
 364                          */
 365                         if (errno == EINTR) {
 366                                 goto restart;
 367                         }
 368                         goto error;
 369                 }
 370                 nb_fd = ret;
 371                 DBG3("[agent-thread] %d fd ready", nb_fd);
 372
 373                 for (i = 0; i < nb_fd; i++) {
 374                         /* Fetch once the poll data */
 375                         revents = LTTNG_POLL_GETEV(&events, i);
 376                         pollfd = LTTNG_POLL_GETFD(&events, i);
 377
 378                         if (!revents) {
 379                                 /* No activity for this FD (poll implementation). */
 380                                 continue;
 381                         }
 382
 383                         /* Thread quit pipe has been closed. Killing thread. */
 384                         ret = sessiond_check_thread_quit_pipe(pollfd, revents);
 385                         if (ret) {
 386                                 goto exit;
 387                         }
 388
 389                         if (revents & LPOLLIN) {
 390                                 int new_fd;
 391                                 struct agent_app *app = NULL;
 392
 393                                 assert(pollfd == reg_sock->fd);
 394                                 new_fd = handle_registration(reg_sock, &app);
 395                                 if (new_fd < 0) {
 396                                         continue;
 397                                 }
 398                                 /* Should not have a NULL app on success. */
 399                                 assert(app);
 400
 401                                 /*
 402                                  * Since this is a command socket (write then read),
 403                                  * only add poll error event to only detect shutdown.
 404                                  */
 405                                 ret = lttng_poll_add(&events, new_fd,
 406                                                 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 407                                 if (ret < 0) {
 408                                         agent_destroy_app_by_sock(new_fd);
 409                                         continue;
 410                                 }
 411
 412                                 /* Update newly registered app. */
 413                                 update_agent_app(app);
 414
 415                                 /* On failure, the poll will detect it and clean it up. */
 416                                 ret = agent_send_registration_done(app);
 417                                 if (ret < 0) {
 418                                         /* Removing from the poll set */
 419                                         ret = lttng_poll_del(&events, new_fd);
 420                                         if (ret < 0) {
 421                                                 goto error;
 422                                         }
 423                                         agent_destroy_app_by_sock(new_fd);
 424                                         continue;
 425                                 }
 426                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 427                                 /* Removing from the poll set */
 428                                 ret = lttng_poll_del(&events, pollfd);
 429                                 if (ret < 0) {
 430                                         goto error;
 431                                 }
 432                                 agent_destroy_app_by_sock(pollfd);
 433                         } else {
 434                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
 435                                 goto error;
 436                         }
 437                 }
 438         }
 439
 440 exit:
 441         /* Whatever happens, try to delete it and exit. */
 442         (void) lttng_poll_del(&events, reg_sock->fd);
 443 error:
 444         destroy_tcp_socket(reg_sock);
 445 error_tcp_socket:
 446         lttng_poll_clean(&events);
 447 error_poll_create:
 448         uatomic_set(&agent_tracing_enabled, 0);
 449         DBG("[agent-thread] is cleaning up and stopping.");
 450
 451         rcu_thread_offline();
 452         rcu_unregister_thread();
 453         return NULL;
 454 }