On 2017-04-11 09:15, Vlad Khorsun wrote: > 11.04.2017 2:33, Adriano dos Santos Fernandes wrote: >> Here is a better description of the problem and a possible fix that >> Vlad >> can use and discard, as he surely better understand this part. >> >> Sorry for some imprecision in my previous description. >> >> Jaybird synchronized the op_queue_events send/response. >> >> Remote queue_events synchronizes with the *sync* port lock. >> >> When event callback is called by the engine, before delivery the >> event, >> remote marks it as dead (event->rvnt_iface = NULL) but using the >> *async* >> port lock. >> >> Even the different locks (sync / async) is not the underling problem >> by >> the way events are marked and verified as dead. >> >> But after event is marked as dead, que_events can concurrently reuse >> it, >> and reuse its rvnt_id. >> >> The problem is that the callback reads rvnt_id after mark it as dead. >> >> So here is my proposed patch. > > Your description is correct and i've considered this kind of fix > too. > Another one was to include search for unused event at > rem_port::que_events > under scope of async port lock. Finally i committed a bit more > efficient > (i think) fix which allows to not extend async port lock scope. > > Thanks, > Vlad > > PS Mark, snapshot build of v4 contains a fix and could be tested
Thanks, I will try to test it tonight if I have time, otherwise tomorrow evening. Is this something that can easily be backported to 3.0 and 2.5? Mark ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot Firebird-Devel mailing list, web interface at https://lists.sourceforge.net/lists/listinfo/firebird-devel
