Re: hung cfservd

[Wipf, Stefan]

I tried to prevent the clients from hanging by introducing
an ExpireAfter but without success.

However, I have found the bug in cfservd.c:

at line 2040:  AuthenticationDialogue returns without releasing
the mutex &MUTEX_SYSCALL if RSA_private_decrypt fails (as is
the case when the client has a bad public key).  This will cause
the next pthread_mutex_lock(&MUTEX_SYSCALL) to block forever.


diff cfservd.c.fix  cfservd.c.original 
2040,2046d2039
< /* release the lock on mutex SYSCALL */
< /* prevents cfservd from freezing up when client connects with bad
public key */
< #if defined HAVE_PTHREAD_H && (defined HAVE_LIBPTHREAD || defined
BUILDTIN_GCC_THREAD)
< if (pthread_mutex_unlock(&MUTEX_SYSCALL) != 0) {
<    CfLog(cferror,"pthread_mutex_unlock failed","lock");
< }
< #endif


Mark.Burgess@iu.hio.no wrote:
>
> On 29 Jan, Wipf, Stefan wrote:
> > - by 'bouncing' cfservd I mean killing it and restarting it.
> > - it's a Ultra-60 with 1 cpu
> 
> Ok - so this is something that I have never seen before.
> Those of you that are having the problem need to figure out
> what is going on. Try running both ends with  -d2 and see
> if you get any clue.
> 
> Please check carefully that all versions are in synch, i.e.
> client and server are the same version and the right version
> of openssl is getting loaded if it's shared libs.
> 
> M
> 
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Work: +47 22453272            Email:  Mark.Burgess@iu.hio.no
> Fax : +47 22453205            WWW  :  http://www.iu.hio.no/~mark
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

-- 
Stefan Wipf
swipf@htc.com