Fix possible crashes and leaks with dsends during processes cleanup

[simgrid.git] / src / simix / smx_process.c

diff --git a/src/simix/smx_process.c b/src/simix/smx_process.c
index 3296198..f2f550c 100644 (file)
--- a/src/simix/smx_process.c
+++ b/src/simix/smx_process.c
@@ -35,7 +35,51 @@ XBT_INLINE smx_process_t SIMIX_process_self(void)
  */
 void SIMIX_process_cleanup(smx_process_t process)
 {
-  XBT_DEBUG("Cleanup process %s", process->name);
+  XBT_DEBUG("Cleanup process %s (%p), waiting action %p",
+      process->name, process, process->waiting_action);
+
+  /* cancel non-blocking communications */
+  smx_action_t action;
+  while ((action = xbt_fifo_pop(process->comms))) {
+
+    /* make sure no one will finish the comm after this process is destroyed */
+    SIMIX_comm_cancel(action);
+
+    if (action->comm.src_proc == process) {
+      XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
+          action, action->comm.detached, action->state, action->comm.src_proc, action->comm.dst_proc);
+      action->comm.src_proc = NULL;
+
+      if (action->comm.detached) {
+         if (action->comm.refcount == 0) {
+           /* I'm not supposed to destroy a detached comm from the sender side,
+            * unless there is no receiver matching the rdv */
+           action->comm.refcount++;
+           SIMIX_comm_destroy(action);
+         }
+      }
+      else {
+        SIMIX_comm_destroy(action);
+      }
+    }
+    else if (action->comm.dst_proc == process){
+      XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
+          action, action->state, action->comm.src_proc, action->comm.dst_proc);
+      action->comm.dst_proc = NULL;
+
+      if (action->comm.detached && action->comm.refcount == 1
+          && action->comm.src_proc != NULL) {
+        /* the comm will be freed right now, remove it from the sender */
+        xbt_fifo_remove(action->comm.src_proc->comms, action);
+      }
+      SIMIX_comm_destroy(action);
+    }
+    else {
+      xbt_die("Communication action %p is in my list but I'm not the sender "
+          "or the receiver", action);
+    }
+  }
+
   /*xbt_swag_remove(process, simix_global->process_to_run);*/
   xbt_swag_remove(process, simix_global->process_list);
   xbt_swag_remove(process, process->smx_host->process_list);
@@ -61,6 +105,8 @@ void SIMIX_process_empty_trash(void)
     if (process->properties)
       xbt_dict_free(&process->properties);
 
+    xbt_fifo_free(process->comms);
+
     free(process->name);
     process->name = NULL;
     free(process);
@@ -143,6 +189,7 @@ void SIMIX_process_create(smx_process_t *process,
     (*process)->name = xbt_strdup(name);
     (*process)->smx_host = host;
     (*process)->data = data;
+    (*process)->comms = xbt_fifo_new();
 
     XBT_VERB("Create context %s", (*process)->name);
     (*process)->context = SIMIX_context_new(code, argc, argv,
@@ -166,6 +213,25 @@ void SIMIX_process_create(smx_process_t *process,
   }
 }
 
+/**
+ * \brief Executes the processes from simix_global->process_to_run.
+ *
+ * The processes of simix_global->process_to_run are run (in parallel if
+ * possible).  On exit, simix_global->process_to_run is empty, and
+ * simix_global->process_that_ran contains the list of processes that just ran.
+ * The two lists are swapped so, be careful when using them before and after a
+ * call to this function.
+ */
+void SIMIX_process_runall(void)
+{
+  SIMIX_context_runall();
+
+  xbt_dynar_t tmp = simix_global->process_that_ran;
+  simix_global->process_that_ran = simix_global->process_to_run;
+  simix_global->process_to_run = tmp;
+  xbt_dynar_reset(simix_global->process_to_run);
+}
+
 /**
  * \brief Internal function to kill a SIMIX process.
  *
@@ -183,6 +249,7 @@ void SIMIX_process_kill(smx_process_t process) {
   process->suspended = 0;
   /* FIXME: set doexception to 0 also? */
 
+  /* destroy the blocking action if any */
   if (process->waiting_action) {
 
     switch (process->waiting_action->type) {
@@ -260,10 +327,17 @@ void SIMIX_pre_process_suspend(smx_req_t req)
 
 void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 {
+  xbt_assert((process != NULL), "Invalid parameters");
+
+  if (process->suspended) {
+    XBT_DEBUG("Process '%s' is already suspended", process->name);
+    return;
+  }
+
   process->suspended = 1;
 
   /* If we are suspending another process, and it is waiting on an action,
-     suspend it's action. */
+     suspend its action. */
   if (process != issuer) {
 
     if (process->waiting_action) {
@@ -284,7 +358,8 @@ void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
           break;
 
         default:
-          THROW_IMPOSSIBLE;
+          xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
+              process->waiting_action->type);
       }
     }
   }
@@ -294,6 +369,11 @@ void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 {
   xbt_assert((process != NULL), "Invalid parameters");
 
+  if (!process->suspended) {
+    XBT_DEBUG("Process '%s' is not suspended", process->name);
+    return;
+  }
+
   process->suspended = 0;
 
   /* If we are resuming another process, resume the action it was waiting for
@@ -318,7 +398,8 @@ void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
           break;
 
         default:
-          THROW_IMPOSSIBLE;
+          xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
+              process->waiting_action->type);
       }
     }
     else {
@@ -477,9 +558,6 @@ void SIMIX_process_sleep_destroy(smx_action_t action)
   XBT_DEBUG("Destroy action %p", action);
   if (action->sleep.surf_sleep)
     action->sleep.surf_sleep->model_type->action_unref(action->sleep.surf_sleep);
-#ifdef HAVE_TRACING
-  TRACE_smx_action_destroy(action);
-#endif
   xbt_mallocator_release(simix_global->action_mallocator, action);
 }
 
@@ -507,7 +585,7 @@ void SIMIX_process_yield(void)
   SIMIX_context_suspend(self->context);
 
   /* Ok, maestro returned control to us */
-  XBT_DEBUG("Maestro returned control to me: '%s'", self->name);
+  XBT_DEBUG("Control returned to me: '%s'", self->name);
 
   if (self->context->iwannadie){
     XBT_DEBUG("I wanna die!");