On Wed, 8 Jan 2003, Costin Manolache wrote:

> Date: Wed, 08 Jan 2003 19:37:28 -0800
> From: Costin Manolache <[EMAIL PROTECTED]>
> Reply-To: Tomcat Developers List <[EMAIL PROTECTED]>
> To: [EMAIL PROTECTED]
> Subject: Re: Duplicate session IDs are *common*
>
> Schnitzer, Jeff wrote:
>
> > For whatever reason, be it the seed algorithm or the hashing algorithm
> > or something else that degenerates the randomness - the duplicate
> > session ID problem is very, very common.
> >
> > I discovered this problem because a few of our users suddenly found
> > themselves with the sessions from administrative accounts.  Luckily they
> > alerted us instead of causing mayhem.  There were at least three
> > separate occasions of this in the last week - that we heard about.
> >
> > We have also seen this a number of times with other game components -
> > users suddenly finding themselves logged in as other people.
> >
> > It probably explains the recent post to tomcat-user included below.
> >
> > Here at my company this problem caused about as much panic as a wildfire
> > breaking out in the machine room (read: LOTS).  I humbly suggest raising
> > the level of concern a bit; post a security bulletin, etc.
>
> What version of tomcat ? Are you on Linux ? What randomClass are you using ?
> How many sessions are usually generated ?
>
> The default is java.security.SecureRandom - and should give enough
> randomness. There is a change on head ( that would work with 5.0 - but
> it can be backported ) that allow you to use /dev/urandom ( or another
> source - it can be a pipe or something like that ).
>

Even though duplicate session ids are *possible* (because Tomcat doesn't
check for it), the 99.9% cause of this kind of a problem is application
programming errors like using an instance (or static) variable in a
servlet to store per-request state information.  That only works if the
servlet never processes more than one request at a time.  Based on my
experience, I don't believe this is a *common* problem, but it is a
problem that needs to be addressed.

>
> Costin
>

Craig


--
To unsubscribe, e-mail:   <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>

Reply via email to