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
  86 MSG_mailbox_get_first_host_task(msg_mailbox_t mailbox, m_host_t host)
  87 {
  88   m_task_t task = NULL;
  89   xbt_fifo_item_t item = NULL;
  90
  91   xbt_fifo_foreach(mailbox->tasks, item, task, m_task_t)
  92       if (task->simdata->source == host) {
  93     xbt_fifo_remove_item(mailbox->tasks, item);
  94     return task;
  95   }
  96
  97   return NULL;
  98 }
  99
 100 int
 101 MSG_mailbox_get_count_host_waiting_tasks(msg_mailbox_t mailbox,
 102                                          m_host_t host)
 103 {
 104   m_task_t task = NULL;
 105   xbt_fifo_item_t item = NULL;
 106   int count = 0;
 107
 108   xbt_fifo_foreach(mailbox->tasks, item, task, m_task_t) {
 109     if (task->simdata->source == host)
 110       count++;
 111   }
 112
 113   return count;
 114 }
 115
 116 void MSG_mailbox_set_cond(msg_mailbox_t mailbox, smx_cond_t cond)
 117 {
 118   mailbox->cond = cond;
 119 }
 120
 121 const char *MSG_mailbox_get_alias(msg_mailbox_t mailbox)
 122 {
 123   return mailbox->alias;
 124 }
 125
 126 const char *MSG_mailbox_get_hostname(msg_mailbox_t mailbox)
 127 {
 128   return mailbox->hostname;
 129 }
 130
 131 void MSG_mailbox_set_hostname(msg_mailbox_t mailbox, const char *hostname)
 132 {
 133   mailbox->hostname = xbt_strdup(hostname);
 134 }
 135
 136 msg_mailbox_t MSG_mailbox_get_by_alias(const char *alias)
 137 {
 138
 139   msg_mailbox_t mailbox = xbt_dict_get_or_null(msg_mailboxes, alias);
 140
 141   if (!mailbox) {
 142     mailbox = MSG_mailbox_new(alias);
 143     MSG_mailbox_set_hostname(mailbox, MSG_host_self()->name);
 144   }
 145
 146   return mailbox;
 147 }
 148
 149 msg_mailbox_t
 150 MSG_mailbox_get_by_channel(m_host_t host, m_channel_t channel)
 151 {
 152   xbt_assert0((host != NULL), "Invalid host");
 153   xbt_assert1((channel >= 0)
 154               && (channel < msg_global->max_channel), "Invalid channel %d",
 155               channel);
 156
 157   return host->simdata->mailboxes[(size_t) channel];
 158 }
 159
 160 MSG_error_t
 161 MSG_mailbox_get_task_ext(msg_mailbox_t mailbox, m_task_t * task,
 162                          m_host_t host, double timeout)
 163 {
 164   m_process_t process = MSG_process_self();
 165   m_task_t t = NULL;
 166   m_host_t h = NULL;
 167   simdata_task_t t_simdata = NULL;
 168   simdata_host_t h_simdata = NULL;
 169   double start_time = SIMIX_get_clock();
 170
 171   smx_cond_t cond = NULL;       //conditional wait if the task isn't on the channel yet
 172
 173   CHECK_HOST();
 174
 175   /* Sanity check */
 176   xbt_assert0(task, "Null pointer for the task storage");
 177
 178   if (*task)
 179     CRITICAL0
 180         ("MSG_task_get() was asked to write in a non empty task struct.");
 181
 182   /* Get the task */
 183   h = MSG_host_self();
 184   h_simdata = h->simdata;
 185
 186   SIMIX_mutex_lock(h->simdata->mutex);
 187
 188   if (MSG_mailbox_get_cond(mailbox)) {
 189     CRITICAL1("A process is already blocked on the channel %s",
 190               MSG_mailbox_get_alias(mailbox));
 191     SIMIX_cond_display_info(MSG_mailbox_get_cond(mailbox));
 192     xbt_die("Go fix your code!");
 193   }
 194
 195   while (1) {
 196     /* if the mailbox is empty (has no task */
 197     if (!MSG_mailbox_is_empty(mailbox)) {
 198       if (!host) {
 199         /* pop the head of the mailbox */
 200         t = MSG_mailbox_pop_head(mailbox);
 201         break;
 202       } else {
 203         /* get the first task of the host */
 204         if ((t = MSG_mailbox_get_first_host_task(mailbox, host)))
 205           break;
 206       }
 207     }
 208
 209     if ((timeout > 0) && (SIMIX_get_clock()-start_time>=timeout)) {
 210       SIMIX_mutex_unlock(h->simdata->mutex);
 211       MSG_mailbox_set_cond(mailbox, NULL);
 212       SIMIX_cond_destroy(cond);
 213       MSG_RETURN(MSG_TRANSFER_FAILURE);
 214     }
 215
 216     if(!cond) {
 217       cond = SIMIX_cond_init();
 218       MSG_mailbox_set_cond(mailbox, cond);
 219     }
 220
 221     if (timeout > 0)
 222       SIMIX_cond_wait_timeout(cond, h->simdata->mutex, timeout-start_time);
 223     else
 224       SIMIX_cond_wait(MSG_mailbox_get_cond(mailbox), h->simdata->mutex);
 225
 226     if (SIMIX_host_get_state(h_simdata->smx_host) == 0) {
 227       SIMIX_mutex_unlock(h->simdata->mutex);
 228       MSG_mailbox_set_cond(mailbox, NULL);
 229       SIMIX_cond_destroy(cond);
 230       MSG_RETURN(MSG_HOST_FAILURE);
 231     }
 232   }
 233
 234
 235   DEBUG1("OK, got a task (%s)", t->name);
 236   /* clean conditional */
 237   if (cond) {
 238     MSG_mailbox_set_cond(mailbox, NULL);
 239     SIMIX_cond_destroy(cond);
 240   }
 241
 242   SIMIX_mutex_unlock(h->simdata->mutex);
 243
 244   t_simdata = t->simdata;
 245   t_simdata->receiver = process;
 246   *task = t;
 247
 248   SIMIX_mutex_lock(t_simdata->mutex);
 249
 250   /* Transfer */
 251   /* create SIMIX action to the communication */
 252   t_simdata->comm =
 253       SIMIX_action_communicate(t_simdata->sender->simdata->m_host->
 254                                simdata->smx_host,
 255                                process->simdata->m_host->simdata->smx_host,
 256                                t->name, t_simdata->message_size,
 257                                t_simdata->rate);
 258
 259   /* This is a hack. We know that both the receiver and the sender will
 260      need to look at the content of t_simdata->comm. And it needs to be
 261      destroyed. However, we don't known whether the receiver or the sender
 262      will get to it first. So by setting with refcount to 2 we can enforce
 263      that things happen correctly. An alternative would be to only do ++ and
 264      -- on this refcount and to sprinkle them judiciously throughout the code,
 265      which appears perhaps worse? Or perhaps the refcount field of
 266      task->simdata can be used for this? At any rate, this will do for now */
 267   t_simdata->comm->refcount = 2;
 268
 269   /* if the process is suspend, create the action but stop its execution, it will be restart when the sender process resume */
 270   if (MSG_process_is_suspended(t_simdata->sender)) {
 271     DEBUG1("Process sender (%s) suspended", t_simdata->sender->name);
 272     SIMIX_action_set_priority(t_simdata->comm, 0);
 273   }
 274
 275   process->simdata->waiting_task = t;
 276   SIMIX_register_action_to_condition(t_simdata->comm, t_simdata->cond);
 277
 278   while (1) {
 279     SIMIX_cond_wait(t_simdata->cond, t_simdata->mutex);
 280
 281     if (SIMIX_action_get_state(t_simdata->comm) != SURF_ACTION_RUNNING)
 282       break;
 283   }
 284
 285   SIMIX_unregister_action_to_condition(t_simdata->comm, t_simdata->cond);
 286   process->simdata->waiting_task = NULL;
 287
 288   /* If sender still around (it didn't free the comm yet), note that it's not waiting anymore */
 289   if (t_simdata->comm->refcount == 2) {
 290     t->simdata->sender->simdata->waiting_task = NULL;
 291   }
 292
 293   /* for this process, don't need to change in get function */
 294   SIMIX_mutex_unlock(t_simdata->mutex);
 295
 296
 297   if (SIMIX_action_get_state(t_simdata->comm) == SURF_ACTION_DONE) {
 298     if (t_simdata->comm->refcount == 1) {
 299       SIMIX_action_destroy(t_simdata->comm);
 300       t_simdata->comm = NULL;
 301     } else {
 302       t_simdata->comm->refcount --;
 303     }
 304     t_simdata->refcount --;
 305     MSG_RETURN(MSG_OK);
 306   } else if (SIMIX_host_get_state(h_simdata->smx_host) == 0) {
 307     if (t_simdata->comm->refcount == 1) {
 308       SIMIX_action_destroy(t_simdata->comm);
 309       t_simdata->comm = NULL;
 310     } else {
 311       t_simdata->comm->refcount --;
 312     }
 313     t_simdata->refcount --;
 314     MSG_RETURN(MSG_HOST_FAILURE);
 315   } else {
 316     if (t_simdata->comm->refcount ==1 ) {
 317       SIMIX_action_destroy(t_simdata->comm);
 318       t_simdata->comm = NULL;
 319     } else {
 320       t_simdata->comm->refcount --;
 321     }
 322     t_simdata->refcount --;
 323     MSG_RETURN(MSG_TRANSFER_FAILURE);
 324   }
 325 }
 326
 327 MSG_error_t
 328 MSG_mailbox_put_with_timeout(msg_mailbox_t mailbox, m_task_t task,
 329                              double timeout)
 330 {
 331   m_process_t process = MSG_process_self();
 332   const char *hostname;
 333   simdata_task_t t_simdata = NULL;
 334   m_host_t local_host = NULL;
 335   m_host_t remote_host = NULL;
 336   smx_cond_t cond = NULL;
 337
 338   CHECK_HOST();
 339
 340   t_simdata = task->simdata;
 341   t_simdata->sender = process;
 342   t_simdata->source = MSG_process_get_host(process);
 343
 344   xbt_assert0(t_simdata->refcount  == 1,
 345               "This task is still being used somewhere else. You cannot send it now. Go fix your code!");
 346
 347   t_simdata->comm = NULL;
 348
 349   t_simdata->refcount ++;
 350   local_host = ((simdata_process_t) process->simdata)->m_host;
 351   msg_global->sent_msg++;
 352
 353   /* get the host name containing the mailbox */
 354   hostname = MSG_mailbox_get_hostname(mailbox);
 355
 356   remote_host = MSG_get_host_by_name(hostname);
 357
 358   if (!remote_host)
 359     THROW1(not_found_error, 0, "Host %s not fount", hostname);
 360
 361
 362   DEBUG4("Trying to send a task (%g kB) from %s to %s on the channel %s",
 363                   t_simdata->message_size / 1000, local_host->name,
 364                   remote_host->name, MSG_mailbox_get_alias(mailbox));
 365
 366   SIMIX_mutex_lock(remote_host->simdata->mutex);
 367
 368   /* put the task in the mailbox */
 369   xbt_fifo_push(mailbox->tasks, task);
 370
 371   if ((cond = MSG_mailbox_get_cond(mailbox))) {
 372     DEBUG0("Somebody is listening. Let's wake him up!");
 373     SIMIX_cond_signal(cond);
 374   }
 375
 376   SIMIX_mutex_unlock(remote_host->simdata->mutex);
 377
 378   SIMIX_mutex_lock(t_simdata->mutex);
 379
 380   process->simdata->waiting_task = task;
 381
 382   if (timeout > 0) {
 383     xbt_ex_t e;
 384     double time;
 385     double time_elapsed;
 386     time = SIMIX_get_clock();
 387
 388     TRY {
 389       /*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 */
 390       while (1) {
 391         time_elapsed = SIMIX_get_clock() - time;
 392         SIMIX_cond_wait_timeout(t_simdata->cond, t_simdata->mutex,
 393                                 timeout - time_elapsed);
 394
 395         if ((t_simdata->comm != NULL)
 396             && (SIMIX_action_get_state(t_simdata->comm) !=
 397                 SURF_ACTION_RUNNING))
 398           break;
 399       }
 400     }
 401     CATCH(e) {
 402       if (e.category == timeout_error) {
 403         xbt_ex_free(e);
 404         /* verify if the timeout happened and the communication didn't started yet */
 405         if (t_simdata->comm == NULL) {
 406           process->simdata->waiting_task = NULL;
 407
 408           /* remove the task from the mailbox */
 409           MSG_mailbox_remove(mailbox, task);
 410
 411           if (t_simdata->receiver && t_simdata->receiver->simdata) { /* receiver still around */
 412             t_simdata->receiver->simdata->waiting_task = NULL;
 413           }
 414
 415           SIMIX_mutex_unlock(t_simdata->mutex);
 416           MSG_RETURN(MSG_TRANSFER_FAILURE);
 417         }
 418       } else {
 419         RETHROW;
 420       }
 421     }
 422   } else {
 423     while (1) {
 424       SIMIX_cond_wait(t_simdata->cond, t_simdata->mutex);
 425
 426       if (SIMIX_action_get_state(t_simdata->comm) !=
 427           SURF_ACTION_RUNNING)
 428         break;
 429     }
 430   }
 431
 432   DEBUG1("Action terminated %s", task->name);
 433   process->simdata->waiting_task = NULL;
 434
 435   if (t_simdata->comm->refcount == 2) { //receiver didn't free it yet: he's still around
 436     t_simdata->receiver->simdata->waiting_task = NULL;
 437   }
 438
 439   SIMIX_mutex_unlock(task->simdata->mutex);
 440
 441   if (SIMIX_action_get_state(t_simdata->comm) == SURF_ACTION_DONE) {
 442     if (t_simdata->comm->refcount == 1) {
 443       SIMIX_action_destroy(t_simdata->comm);
 444       t_simdata->comm = NULL;
 445     } else {
 446       t_simdata->comm->refcount --;
 447     }
 448     MSG_RETURN(MSG_OK);
 449   } else if (SIMIX_host_get_state(local_host->simdata->smx_host) == 0) {
 450     if (t_simdata->comm->refcount == 1) {
 451       SIMIX_action_destroy(t_simdata->comm);
 452       t_simdata->comm = NULL;
 453     } else {
 454       t_simdata->comm->refcount --;
 455     }
 456     MSG_RETURN(MSG_HOST_FAILURE);
 457   } else {
 458     if (t_simdata->comm->refcount == 1) {
 459       SIMIX_action_destroy(t_simdata->comm);
 460       t_simdata->comm = NULL;
 461     } else {
 462       t_simdata->comm->refcount --;
 463     }
 464     MSG_RETURN(MSG_TRANSFER_FAILURE);
 465   }
 466 }