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