src/msg/msg_mailbox.c

   1 #include "mailbox.h"
   2 #include "msg/private.h"
   3
   4 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_mailbox, msg,
   5                                 "Logging specific to MSG (mailbox)");
   6
   7 static xbt_dict_t msg_mailboxes = NULL;
   8
   9 void MSG_mailbox_mod_init(void)
  10 {
  11   msg_mailboxes = xbt_dict_new();
  12 }
  13
  14 void MSG_mailbox_mod_exit(void)
  15 {
  16   xbt_dict_free(&msg_mailboxes);
  17 }
  18
  19 msg_mailbox_t MSG_mailbox_create(const char *alias)
  20 {
  21   msg_mailbox_t mailbox = xbt_new0(s_msg_mailbox_t, 1);
  22
  23   mailbox->tasks = xbt_fifo_new();
  24   mailbox->cond = NULL;
  25   mailbox->alias = alias ? xbt_strdup(alias) : NULL;
  26   mailbox->hostname = NULL;
  27
  28   return mailbox;
  29 }
  30
  31 msg_mailbox_t MSG_mailbox_new(const char *alias)
  32 {
  33   msg_mailbox_t mailbox = MSG_mailbox_create(alias);
  34
  35   /* add the mbox in the dictionary */
  36   xbt_dict_set(msg_mailboxes, alias, mailbox, MSG_mailbox_free);
  37
  38   return mailbox;
  39 }
  40
  41 void MSG_mailbox_free(void *mailbox)
  42 {
  43   msg_mailbox_t _mailbox = (msg_mailbox_t) mailbox;
  44
  45   if (_mailbox->hostname)
  46     free(_mailbox->hostname);
  47
  48   xbt_fifo_free(_mailbox->tasks);
  49   free(_mailbox->alias);
  50
  51   free(_mailbox);
  52 }
  53
  54 smx_cond_t MSG_mailbox_get_cond(msg_mailbox_t mailbox)
  55 {
  56   return mailbox->cond;
  57 }
  58
  59 void MSG_mailbox_remove(msg_mailbox_t mailbox, m_task_t task)
  60 {
  61   xbt_fifo_remove(mailbox->tasks, task);
  62 }
  63
  64 int MSG_mailbox_is_empty(msg_mailbox_t mailbox)
  65 {
  66   return (NULL == xbt_fifo_get_first_item(mailbox->tasks));
  67 }
  68
  69 m_task_t MSG_mailbox_pop_head(msg_mailbox_t mailbox)
  70 {
  71   return (m_task_t) xbt_fifo_shift(mailbox->tasks);
  72 }
  73
  74 m_task_t MSG_mailbox_get_head(msg_mailbox_t mailbox)
  75 {
  76   xbt_fifo_item_t item;
  77
  78   if (!(item = xbt_fifo_get_first_item(mailbox->tasks)))
  79     return NULL;
  80
  81   return (m_task_t) xbt_fifo_get_item_content(item);
  82 }
  83
  84
  85 m_task_t MSG_mailbox_get_first_host_task(msg_mailbox_t mailbox, m_host_t host)
  86 {
  87   m_task_t task = NULL;
  88   xbt_fifo_item_t item = NULL;
  89
  90   xbt_fifo_foreach(mailbox->tasks, item, task, m_task_t)
  91     if (task->simdata->source == host) {
  92     xbt_fifo_remove_item(mailbox->tasks, item);
  93     return task;
  94   }
  95
  96   return NULL;
  97 }
  98
  99 int
 100 MSG_mailbox_get_count_host_waiting_tasks(msg_mailbox_t mailbox, m_host_t host)
 101 {
 102   m_task_t task = NULL;
 103   xbt_fifo_item_t item = NULL;
 104   int count = 0;
 105
 106   xbt_fifo_foreach(mailbox->tasks, item, task, m_task_t) {
 107     if (task->simdata->source == host)
 108       count++;
 109   }
 110
 111   return count;
 112 }
 113
 114 void MSG_mailbox_set_cond(msg_mailbox_t mailbox, smx_cond_t cond)
 115 {
 116   mailbox->cond = cond;
 117 }
 118
 119 const char *MSG_mailbox_get_alias(msg_mailbox_t mailbox)
 120 {
 121   return mailbox->alias;
 122 }
 123
 124 const char *MSG_mailbox_get_hostname(msg_mailbox_t mailbox)
 125 {
 126   return mailbox->hostname;
 127 }
 128
 129 void MSG_mailbox_set_hostname(msg_mailbox_t mailbox, const char *hostname)
 130 {
 131   mailbox->hostname = xbt_strdup(hostname);
 132 }
 133
 134 msg_mailbox_t MSG_mailbox_get_by_alias(const char *alias)
 135 {
 136
 137   msg_mailbox_t mailbox = xbt_dict_get_or_null(msg_mailboxes, alias);
 138
 139   if (!mailbox) {
 140     mailbox = MSG_mailbox_new(alias);
 141     MSG_mailbox_set_hostname(mailbox, MSG_host_self()->name);
 142   }
 143
 144   return mailbox;
 145 }
 146
 147 msg_mailbox_t MSG_mailbox_get_by_channel(m_host_t host, m_channel_t channel)
 148 {
 149   xbt_assert0((host != NULL), "Invalid host");
 150   xbt_assert1((channel >= 0)
 151               && (channel < msg_global->max_channel), "Invalid channel %d",
 152               channel);
 153
 154   return host->simdata->mailboxes[(size_t) channel];
 155 }
 156
 157 MSG_error_t
 158 MSG_mailbox_get_task_ext(msg_mailbox_t mailbox, m_task_t * task,
 159                          m_host_t host, double timeout)
 160 {
 161   m_process_t process = MSG_process_self();
 162   m_task_t t = NULL;
 163   m_host_t h = NULL;
 164   simdata_task_t t_simdata = NULL;
 165   simdata_host_t h_simdata = NULL;
 166   double start_time = SIMIX_get_clock();
 167
 168   smx_cond_t cond = NULL;       //conditional wait if the task isn't on the channel yet
 169
 170   CHECK_HOST();
 171
 172   /* Sanity check */
 173   xbt_assert0(task, "Null pointer for the task storage");
 174
 175   if (*task)
 176     CRITICAL0
 177       ("MSG_task_get() was asked to write in a non empty task struct.");
 178
 179   /* Get the task */
 180   h = MSG_host_self();
 181   h_simdata = h->simdata;
 182
 183   SIMIX_mutex_lock(h->simdata->mutex);
 184
 185   if (MSG_mailbox_get_cond(mailbox)) {
 186     CRITICAL1("A process is already blocked on the channel %s",
 187               MSG_mailbox_get_alias(mailbox));
 188     SIMIX_cond_display_info(MSG_mailbox_get_cond(mailbox));
 189     xbt_die("Go fix your code!");
 190   }
 191
 192   while (1) {
 193     /* if the mailbox is empty (has no task */
 194     if (!MSG_mailbox_is_empty(mailbox)) {
 195       if (!host) {
 196         /* pop the head of the mailbox */
 197         t = MSG_mailbox_pop_head(mailbox);
 198         break;
 199       } else {
 200         /* get the first task of the host */
 201         if ((t = MSG_mailbox_get_first_host_task(mailbox, host)))
 202           break;
 203       }
 204     }
 205
 206     if ((timeout > 0) && (SIMIX_get_clock() - start_time >= timeout)) {
 207       SIMIX_mutex_unlock(h->simdata->mutex);
 208       MSG_mailbox_set_cond(mailbox, NULL);
 209       SIMIX_cond_destroy(cond);
 210       MSG_RETURN(MSG_TRANSFER_FAILURE);
 211     }
 212
 213     if (!cond) {
 214       cond = SIMIX_cond_init();
 215       MSG_mailbox_set_cond(mailbox, cond);
 216     }
 217
 218     if (timeout > 0)
 219       SIMIX_cond_wait_timeout(cond, h->simdata->mutex, timeout - start_time);
 220     else
 221       SIMIX_cond_wait(MSG_mailbox_get_cond(mailbox), h->simdata->mutex);
 222
 223     if (SIMIX_host_get_state(h_simdata->smx_host) == 0) {
 224       SIMIX_mutex_unlock(h->simdata->mutex);
 225       MSG_mailbox_set_cond(mailbox, NULL);
 226       SIMIX_cond_destroy(cond);
 227       MSG_RETURN(MSG_HOST_FAILURE);
 228     }
 229   }
 230
 231
 232   DEBUG1("OK, got a task (%s)", t->name);
 233   /* clean conditional */
 234   if (cond) {
 235     MSG_mailbox_set_cond(mailbox, NULL);
 236     SIMIX_cond_destroy(cond);
 237   }
 238
 239   SIMIX_mutex_unlock(h->simdata->mutex);
 240
 241   t_simdata = t->simdata;
 242   t_simdata->receiver = process;
 243   *task = t;
 244
 245   SIMIX_mutex_lock(t_simdata->mutex);
 246
 247   /* Transfer */
 248   /* create SIMIX action to the communication */
 249   t_simdata->comm =
 250     SIMIX_action_communicate(t_simdata->sender->simdata->m_host->simdata->
 251                              smx_host,
 252                              process->simdata->m_host->simdata->smx_host,
 253                              t->name, t_simdata->message_size,
 254                              t_simdata->rate);
 255
 256   /* This is a hack. We know that both the receiver and the sender will
 257      need to look at the content of t_simdata->comm. And it needs to be
 258      destroyed. However, we don't known whether the receiver or the sender
 259      will get to it first. So by setting with refcount to 2 we can enforce
 260      that things happen correctly. An alternative would be to only do ++ and
 261      -- on this refcount and to sprinkle them judiciously throughout the code,
 262      which appears perhaps worse? Or perhaps the refcount field of
 263      task->simdata can be used for this? At any rate, this will do for now */
 264   t_simdata->comm->refcount = 2;
 265
 266   /* if the process is suspend, create the action but stop its execution, it will be restart when the sender process resume */
 267   if (MSG_process_is_suspended(t_simdata->sender)) {
 268     DEBUG1("Process sender (%s) suspended", t_simdata->sender->name);
 269     SIMIX_action_set_priority(t_simdata->comm, 0);
 270   }
 271
 272   process->simdata->waiting_task = t;
 273   SIMIX_register_action_to_condition(t_simdata->comm, t_simdata->cond);
 274
 275   while (1) {
 276     SIMIX_cond_wait(t_simdata->cond, t_simdata->mutex);
 277
 278     if (SIMIX_action_get_state(t_simdata->comm) != SURF_ACTION_RUNNING)
 279       break;
 280   }
 281
 282   SIMIX_unregister_action_to_condition(t_simdata->comm, t_simdata->cond);
 283   process->simdata->waiting_task = NULL;
 284
 285   /* If sender still around (it didn't free the comm yet), note that it's not waiting anymore */
 286   if (t_simdata->comm->refcount == 2) {
 287     t->simdata->sender->simdata->waiting_task = NULL;
 288   }
 289
 290   /* for this process, don't need to change in get function */
 291   SIMIX_mutex_unlock(t_simdata->mutex);
 292
 293
 294   if (SIMIX_action_get_state(t_simdata->comm) == SURF_ACTION_DONE) {
 295     if (t_simdata->comm->refcount == 1) {
 296       SIMIX_action_destroy(t_simdata->comm);
 297       t_simdata->comm = NULL;
 298     } else {
 299       t_simdata->comm->refcount--;
 300     }
 301     t_simdata->refcount--;
 302     MSG_RETURN(MSG_OK);
 303   } else if (SIMIX_host_get_state(h_simdata->smx_host) == 0) {
 304     if (t_simdata->comm->refcount == 1) {
 305       SIMIX_action_destroy(t_simdata->comm);
 306       t_simdata->comm = NULL;
 307     } else {
 308       t_simdata->comm->refcount--;
 309     }
 310     t_simdata->refcount--;
 311     MSG_RETURN(MSG_HOST_FAILURE);
 312   } else {
 313     if (t_simdata->comm->refcount == 1) {
 314       SIMIX_action_destroy(t_simdata->comm);
 315       t_simdata->comm = NULL;
 316     } else {
 317       t_simdata->comm->refcount--;
 318     }
 319     t_simdata->refcount--;
 320     MSG_RETURN(MSG_TRANSFER_FAILURE);
 321   }
 322 }
 323
 324 MSG_error_t
 325 MSG_mailbox_put_with_timeout(msg_mailbox_t mailbox, m_task_t task,
 326                              double timeout)
 327 {
 328   m_process_t process = MSG_process_self();
 329   const char *hostname;
 330   simdata_task_t t_simdata = NULL;
 331   m_host_t local_host = NULL;
 332   m_host_t remote_host = NULL;
 333   smx_cond_t cond = NULL;
 334
 335   CHECK_HOST();
 336
 337   t_simdata = task->simdata;
 338   t_simdata->sender = process;
 339   t_simdata->source = MSG_process_get_host(process);
 340
 341   xbt_assert0(t_simdata->refcount == 1,
 342               "This task is still being used somewhere else. You cannot send it now. Go fix your code!");
 343
 344   t_simdata->comm = NULL;
 345
 346   t_simdata->refcount++;
 347   local_host = ((simdata_process_t) process->simdata)->m_host;
 348   msg_global->sent_msg++;
 349
 350   /* get the host name containing the mailbox */
 351   hostname = MSG_mailbox_get_hostname(mailbox);
 352
 353   remote_host = MSG_get_host_by_name(hostname);
 354
 355   if (!remote_host)
 356     THROW1(not_found_error, 0, "Host %s not fount", hostname);
 357
 358
 359   DEBUG4("Trying to send a task (%g kB) from %s to %s on the channel %s",
 360          t_simdata->message_size / 1000, local_host->name,
 361          remote_host->name, MSG_mailbox_get_alias(mailbox));
 362
 363   SIMIX_mutex_lock(remote_host->simdata->mutex);
 364
 365   /* put the task in the mailbox */
 366   xbt_fifo_push(mailbox->tasks, task);
 367
 368   if ((cond = MSG_mailbox_get_cond(mailbox))) {
 369     DEBUG0("Somebody is listening. Let's wake him up!");
 370     SIMIX_cond_signal(cond);
 371   }
 372
 373   SIMIX_mutex_unlock(remote_host->simdata->mutex);
 374
 375   SIMIX_mutex_lock(t_simdata->mutex);
 376
 377   process->simdata->waiting_task = task;
 378
 379   if (timeout > 0) {
 380     xbt_ex_t e;
 381     double time;
 382     double time_elapsed;
 383     time = SIMIX_get_clock();
 384
 385     TRY {
 386       /*verify if the action that ends is the correct. Call the wait_timeout with the new time. If the timeout occurs, an exception is raised */
 387       while (1) {
 388         time_elapsed = SIMIX_get_clock() - time;
 389         SIMIX_cond_wait_timeout(t_simdata->cond, t_simdata->mutex,
 390                                 timeout - time_elapsed);
 391
 392         if ((t_simdata->comm != NULL)
 393             && (SIMIX_action_get_state(t_simdata->comm) !=
 394                 SURF_ACTION_RUNNING))
 395           break;
 396       }
 397     }
 398     CATCH(e) {
 399       if (e.category == timeout_error) {
 400         xbt_ex_free(e);
 401         /* verify if the timeout happened and the communication didn't started yet */
 402         if (t_simdata->comm == NULL) {
 403           process->simdata->waiting_task = NULL;
 404
 405           /* remove the task from the mailbox */
 406           MSG_mailbox_remove(mailbox, task);
 407
 408           if (t_simdata->receiver && t_simdata->receiver->simdata) {    /* receiver still around */
 409             t_simdata->receiver->simdata->waiting_task = NULL;
 410           }
 411
 412           SIMIX_mutex_unlock(t_simdata->mutex);
 413           MSG_RETURN(MSG_TRANSFER_FAILURE);
 414         }
 415       } else {
 416         RETHROW;
 417       }
 418     }
 419   } else {
 420     while (1) {
 421       SIMIX_cond_wait(t_simdata->cond, t_simdata->mutex);
 422
 423       if (SIMIX_action_get_state(t_simdata->comm) != SURF_ACTION_RUNNING)
 424         break;
 425     }
 426   }
 427
 428   DEBUG1("Action terminated %s", task->name);
 429   process->simdata->waiting_task = NULL;
 430
 431   if (t_simdata->comm->refcount == 2) { //receiver didn't free it yet: he's still around
 432     t_simdata->receiver->simdata->waiting_task = NULL;
 433   }
 434
 435   SIMIX_mutex_unlock(task->simdata->mutex);
 436
 437   if (SIMIX_action_get_state(t_simdata->comm) == SURF_ACTION_DONE) {
 438     if (t_simdata->comm->refcount == 1) {
 439       SIMIX_action_destroy(t_simdata->comm);
 440       t_simdata->comm = NULL;
 441     } else {
 442       t_simdata->comm->refcount--;
 443     }
 444     MSG_RETURN(MSG_OK);
 445   } else if (SIMIX_host_get_state(local_host->simdata->smx_host) == 0) {
 446     if (t_simdata->comm->refcount == 1) {
 447       SIMIX_action_destroy(t_simdata->comm);
 448       t_simdata->comm = NULL;
 449     } else {
 450       t_simdata->comm->refcount--;
 451     }
 452     MSG_RETURN(MSG_HOST_FAILURE);
 453   } else {
 454     if (t_simdata->comm->refcount == 1) {
 455       SIMIX_action_destroy(t_simdata->comm);
 456       t_simdata->comm = NULL;
 457     } else {
 458       t_simdata->comm->refcount--;
 459     }
 460     MSG_RETURN(MSG_TRANSFER_FAILURE);
 461   }
 462 }