From: Christian Grothoff <christian@grothoff.org>
Date: Thu, 21 May 2015 10:07:58 +0000 (+0000)
Subject: with respect to #3615: we might (theoretically) get a migration stop before we finish... 
X-Git-Tag: initial-import-from-subversion-38251~1881
X-Git-Url: https://git.librecmc.org/?a=commitdiff_plain;h=7fb5550c4dd74db2bc71ec188a645697416ebb11;p=oweals%2Fgnunet.git

with respect to #3615: we might (theoretically) get a migration stop before we finished the peerstore lookup; in that case, we would do a GSF_push_start() twice (once via revive migration, and once from peerstore lookup succeeding), and then on disconnect only stop once, keeping the migration going for a disconnected peer.  That still does not explain the bug report, as the '*cp' in the report doesn't show BADFOOD, and we'd expect a free'd cp in this case to be used for the transmission. Still, this is a possible issue which can be fixed by ignoring migration stop messages until after peerstore lookup succeeded
---

diff --git a/src/fs/gnunet-service-fs_cp.c b/src/fs/gnunet-service-fs_cp.c
index 708308f7f..5dcd7689e 100644
--- a/src/fs/gnunet-service-fs_cp.c
+++ b/src/fs/gnunet-service-fs_cp.c
@@ -741,7 +741,8 @@ GSF_handle_p2p_migration_stop_ (void *cls,
               GNUNET_i2s (other),
 	      GNUNET_STRINGS_relative_time_to_string (bt, GNUNET_YES));
   cp->ppd.migration_blocked_until = GNUNET_TIME_relative_to_absolute (bt);
-  if (NULL == cp->mig_revive_task)
+  if ( (NULL == cp->mig_revive_task) &&
+       (NULL == cp->respect_iterate_req) )
   {
     GSF_push_stop_ (cp);
     cp->mig_revive_task =
diff --git a/src/fs/gnunet-service-fs_push.c b/src/fs/gnunet-service-fs_push.c
index d1bdad1d5..73813c68e 100644
--- a/src/fs/gnunet-service-fs_push.c
+++ b/src/fs/gnunet-service-fs_push.c
@@ -611,6 +611,7 @@ GSF_push_start_ (struct GSF_ConnectedPeer *peer)
       break;
   if (NULL != mrp)
   {
+    /* same peer added twice, must not happen */
     GNUNET_break (0);
     return;
   }