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 _GNU_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 "lttng-sessiond.h"
  31 #include "session.h"
  32 #include "utils.h"
  33
  34 /*
  35  * Note that there is not port here. It's set after this URI is parsed so we
  36  * can let the user define a custom one. However, localhost is ALWAYS the
  37  * default listening address.
  38  */
  39 static const char *default_reg_uri =
  40         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS;
  41
  42 /*
  43  * Update agent application using the given socket. This is done just after
  44  * registration was successful.
  45  *
  46  * This is a quite heavy call in terms of locking since the session list lock
  47  * AND session lock are acquired.
  48  */
  49 static void update_agent_app(struct agent_app *app)
  50 {
  51         struct ltt_session *session, *stmp;
  52         struct ltt_session_list *list;
  53
  54         list = session_get_list();
  55         assert(list);
  56
  57         session_lock_list();
  58         cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
  59                 session_lock(session);
  60                 if (session->ust_session) {
  61                         struct agent *agt;
  62
  63                         agt = trace_ust_find_agent(session->ust_session, app->domain);
  64                         if (agt) {
  65                                 agent_update(agt, app->sock->fd);
  66                         }
  67                 }
  68                 session_unlock(session);
  69         }
  70         session_unlock_list();
  71 }
  72
  73 /*
  74  * Destroy a agent application by socket.
  75  */
  76 static void destroy_agent_app(int sock)
  77 {
  78         struct agent_app *app;
  79
  80         assert(sock >= 0);
  81
  82         /*
  83          * Not finding an application is a very important error that should NEVER
  84          * happen. The hash table deletion is ONLY done through this call even on
  85          * thread cleanup.
  86          */
  87         rcu_read_lock();
  88         app = agent_find_app_by_sock(sock);
  89         assert(app);
  90         rcu_read_unlock();
  91
  92         /* RCU read side lock is taken in this function call. */
  93         agent_delete_app(app);
  94
  95         /* The application is freed in a RCU call but the socket is closed here. */
  96         agent_destroy_app(app);
  97 }
  98
  99 /*
 100  * Cleanup remaining agent apps in the hash table. This should only be called in
 101  * the exit path of the thread.
 102  */
 103 static void clean_agent_apps_ht(void)
 104 {
 105         struct lttng_ht_node_ulong *node;
 106         struct lttng_ht_iter iter;
 107
 108         DBG3("[agent-thread] Cleaning agent apps ht");
 109
 110         rcu_read_lock();
 111         cds_lfht_for_each_entry(agent_apps_ht_by_sock->ht, &iter.iter, node, node) {
 112                 struct agent_app *app;
 113
 114                 app = caa_container_of(node, struct agent_app, node);
 115                 destroy_agent_app(app->sock->fd);
 116         }
 117         rcu_read_unlock();
 118 }
 119
 120 /*
 121  * Create and init socket from uri.
 122  */
 123 static struct lttcomm_sock *init_tcp_socket(void)
 124 {
 125         int ret;
 126         struct lttng_uri *uri = NULL;
 127         struct lttcomm_sock *sock = NULL;
 128
 129         /*
 130          * This should never fail since the URI is hardcoded and the port is set
 131          * before this thread is launched.
 132          */
 133         ret = uri_parse(default_reg_uri, &uri);
 134         assert(ret);
 135         assert(agent_tcp_port);
 136         uri->port = agent_tcp_port;
 137
 138         sock = lttcomm_alloc_sock_from_uri(uri);
 139         uri_free(uri);
 140         if (sock == NULL) {
 141                 ERR("[agent-thread] agent allocating TCP socket");
 142                 goto error;
 143         }
 144
 145         ret = lttcomm_create_sock(sock);
 146         if (ret < 0) {
 147                 goto error;
 148         }
 149
 150         ret = sock->ops->bind(sock);
 151         if (ret < 0) {
 152                 WARN("Another session daemon is using this agent port. Agent support "
 153                                 "will be deactivated to prevent interfering with the tracing.");
 154                 goto error;
 155         }
 156
 157         ret = sock->ops->listen(sock, -1);
 158         if (ret < 0) {
 159                 goto error;
 160         }
 161
 162         DBG("[agent-thread] Listening on TCP port %u and socket %d",
 163                         agent_tcp_port, sock->fd);
 164
 165         return sock;
 166
 167 error:
 168         if (sock) {
 169                 lttcomm_destroy_sock(sock);
 170         }
 171         return NULL;
 172 }
 173
 174 /*
 175  * Close and destroy the given TCP socket.
 176  */
 177 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 178 {
 179         assert(sock);
 180
 181         DBG3("[agent-thread] Destroy TCP socket on port %u", agent_tcp_port);
 182
 183         /* This will return gracefully if fd is invalid. */
 184         sock->ops->close(sock);
 185         lttcomm_destroy_sock(sock);
 186 }
 187
 188 /*
 189  * Handle a new agent registration using the reg socket. After that, a new
 190  * agent application is added to the global hash table and attach to an UST app
 191  * object. If r_app is not NULL, the created app is set to the pointer.
 192  *
 193  * Return the new FD created upon accept() on success or else a negative errno
 194  * value.
 195  */
 196 static int handle_registration(struct lttcomm_sock *reg_sock,
 197                 struct agent_app **r_app)
 198 {
 199         int ret;
 200         pid_t pid;
 201         ssize_t size;
 202         enum lttng_domain_type domain;
 203         struct agent_app *app;
 204         struct agent_register_msg msg;
 205         struct lttcomm_sock *new_sock;
 206
 207         assert(reg_sock);
 208
 209         new_sock = reg_sock->ops->accept(reg_sock);
 210         if (!new_sock) {
 211                 ret = -ENOTCONN;
 212                 goto error;
 213         }
 214
 215         size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
 216         if (size < sizeof(msg)) {
 217                 ret = -errno;
 218                 goto error_socket;
 219         }
 220         domain = be32toh(msg.domain);
 221         pid = be32toh(msg.pid);
 222
 223         DBG2("[agent-thread] New registration for pid %d domain %d on socket %d",
 224                         pid, domain, new_sock->fd);
 225
 226         app = agent_create_app(pid, domain, new_sock);
 227         if (!app) {
 228                 ret = -ENOMEM;
 229                 goto error_socket;
 230         }
 231
 232         /*
 233          * Add before assigning the socket value to the UST app so it can be found
 234          * concurrently.
 235          */
 236         agent_add_app(app);
 237
 238         /*
 239          * We don't need to attach the agent app to the app. If we ever do so, we
 240          * should consider both registration order of agent before app and app
 241          * before agent.
 242          */
 243
 244         if (r_app) {
 245                 *r_app = app;
 246         }
 247
 248         return new_sock->fd;
 249
 250 error_socket:
 251         new_sock->ops->close(new_sock);
 252         lttcomm_destroy_sock(new_sock);
 253 error:
 254         return ret;
 255 }
 256
 257 /*
 258  * This thread manage application notify communication.
 259  */
 260 void *agent_thread_manage_registration(void *data)
 261 {
 262         int i, ret, pollfd;
 263         uint32_t revents, nb_fd;
 264         struct lttng_poll_event events;
 265         struct lttcomm_sock *reg_sock;
 266
 267         DBG("[agent-thread] Manage agent application registration.");
 268
 269         rcu_register_thread();
 270         rcu_thread_online();
 271
 272         /* Agent initialization call MUST be called before starting the thread. */
 273         assert(agent_apps_ht_by_sock);
 274
 275         /* Create pollset with size 2, quit pipe and socket. */
 276         ret = sessiond_set_thread_pollset(&events, 2);
 277         if (ret < 0) {
 278                 goto error_poll_create;
 279         }
 280
 281         reg_sock = init_tcp_socket();
 282         if (!reg_sock) {
 283                 goto error_tcp_socket;
 284         }
 285
 286         /* Add create valid TCP socket to poll set. */
 287         ret = lttng_poll_add(&events, reg_sock->fd,
 288                         LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 289         if (ret < 0) {
 290                 goto error;
 291         }
 292
 293         while (1) {
 294                 DBG3("[agent-thread] Manage agent polling on %d fds",
 295                                 LTTNG_POLL_GETNB(&events));
 296
 297                 /* Inifinite blocking call, waiting for transmission */
 298 restart:
 299                 ret = lttng_poll_wait(&events, -1);
 300                 if (ret < 0) {
 301                         /*
 302                          * Restart interrupted system call.
 303                          */
 304                         if (errno == EINTR) {
 305                                 goto restart;
 306                         }
 307                         goto error;
 308                 }
 309                 nb_fd = ret;
 310                 DBG3("[agent-thread] %d fd ready", nb_fd);
 311
 312                 for (i = 0; i < nb_fd; i++) {
 313                         /* Fetch once the poll data */
 314                         revents = LTTNG_POLL_GETEV(&events, i);
 315                         pollfd = LTTNG_POLL_GETFD(&events, i);
 316
 317                         /* Thread quit pipe has been closed. Killing thread. */
 318                         ret = sessiond_check_thread_quit_pipe(pollfd, revents);
 319                         if (ret) {
 320                                 goto exit;
 321                         }
 322
 323                         /*
 324                          * Check first if this is a POLLERR since POLLIN is also included
 325                          * in an error value thus checking first.
 326                          */
 327                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 328                                 /* Removing from the poll set */
 329                                 ret = lttng_poll_del(&events, pollfd);
 330                                 if (ret < 0) {
 331                                         goto error;
 332                                 }
 333
 334                                 destroy_agent_app(pollfd);
 335                         } else if (revents & (LPOLLIN)) {
 336                                 int new_fd;
 337                                 struct agent_app *app = NULL;
 338
 339                                 /* Pollin event of agent app socket should NEVER happen. */
 340                                 assert(pollfd == reg_sock->fd);
 341
 342                                 new_fd = handle_registration(reg_sock, &app);
 343                                 if (new_fd < 0) {
 344                                         WARN("[agent-thread] agent registration failed. Ignoring.");
 345                                         /* Somehow the communication failed. Just continue. */
 346                                         continue;
 347                                 }
 348                                 /* Should not have a NULL app on success. */
 349                                 assert(app);
 350
 351                                 /* Only add poll error event to only detect shutdown. */
 352                                 ret = lttng_poll_add(&events, new_fd,
 353                                                 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 354                                 if (ret < 0) {
 355                                         destroy_agent_app(new_fd);
 356                                         continue;
 357                                 }
 358
 359                                 /* Update newly registered app. */
 360                                 update_agent_app(app);
 361
 362                                 /* On failure, the poll will detect it and clean it up. */
 363                                 (void) agent_send_registration_done(app);
 364                         } else {
 365                                 ERR("Unknown poll events %u for sock %d", revents, pollfd);
 366                                 continue;
 367                         }
 368                 }
 369         }
 370
 371 exit:
 372         /* Whatever happens, try to delete it and exit. */
 373         (void) lttng_poll_del(&events, reg_sock->fd);
 374 error:
 375         destroy_tcp_socket(reg_sock);
 376 error_tcp_socket:
 377         lttng_poll_clean(&events);
 378 error_poll_create:
 379         DBG("[agent-thread] is cleaning up and stopping.");
 380
 381         if (agent_apps_ht_by_sock) {
 382                 clean_agent_apps_ht();
 383                 lttng_ht_destroy(agent_apps_ht_by_sock);
 384         }
 385
 386         rcu_thread_offline();
 387         rcu_unregister_thread();
 388         return NULL;
 389 }