[prev in list] [next in list] [prev in thread] [next in thread] 

List:       snap-users
Subject:    (KAME-snap 8726) Re: MBufs strange problem in IPSec, race
From:       Shoichi Sakane <sakane () kame ! net>
Date:       2004-08-02 17:10:57
Message-ID: 20040803021057O.sakane () kame ! net
[Download RAW message or body]

hi, Yvan.

i dont have much time and a environment to check it.
could we have your script or your racoon configuration
to reproduce your problem ?  then somebody can match with the problem.

> Ok.
> 
> I made another test, here are some more informations:
> 
> I have an IPSec link between 2 FreeBSD 4.10 gates (one with the
> standard FreeBSD/racoon code, the other with some modifications, but
> nothing around the code concerned, AFAIK), with 15 phases2.
> 
> On the initiator side, I have a script to generate some traffic for
> each phase2, to force tunnels negociation.
> 
> If I run this configuration, I have the problems mentionned in my
> previous mail.
> 
> If I add a sleep(1) in my traffic generator between each packet
> generated, I don't have the problem anymore !
> 
> I tested it about 100 times in each situation, I have the problem
> quite always without sleep, and I never had it when I add the
> sleep(1).
> 
> 
> In the source code, I tried to understant where the used structures
> are created, and if I'm right, it leads to soalloc() in
> kern/uipc_socket.c, where I found the following comment:
> /* XXX race condition for reentrant kernel */
> 
> I tried a splnet() lock and I also tried to report the zlock() to
> soalloc(), but it didn't change my problem.
> 
> 
> If someone have a suggestion, an idea, a patch, or whatever else could
> help me, thanks !!
> 
> 
> Yvan.
> 
[prev in list] [next in list] [prev in thread] [next in thread]