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