Problems with kprop and incremental propagation - works but bugs found

Matej Zagiba matej.zagiba at fmph.uniba.sk
Tue Nov 16 04:23:00 EST 2010


Hello,

  thank You for help, I was reading the source code, but I got lost :-)

Anyway, I noticed kpropd has an -F switch to specify path to database, but it's used only in
full replication when passing arguments to kbd_util. It would be nice to use this argument
to override config files in incremental propagation as well.

I suppose not too many people uses incremental propagation, it relatively new feature,
but I must say after getting it work, it's perfect solution.

  thanx for all the help

   Matej Zagiba


On 11/16/2010 01:14 AM, Greg Hudson wrote:
> I can confirm two bugs that you have encountered and worked around:
>
> 1. kprop uses krb5_sname_to_principal() to determine its client
> principal, and does not understand the referral realm.  So it does not
> work without a -r parameter unless the profile's domain_realm section
> can map the local hostname.  You worked around this by correcting your
> existing domain_realm section in your profile.
>
> A reasonable, if not perfect, fix here is to do what kpropd does in a
> similiar piece of code: substitute the default realm for the referral
> realm when using the result of krb5_sname_to_principal() as a client
> principal.
>
> 2. kpropd, when processing incremental updates, modifies the KDB using
> ulog_replay(), but does not initialize its context to use the KDC
> profile, so it uses only settings from krb5.conf to find the KDB.  You
> worked around this with symlinks.  An alternative workaround would be to
> put the KDB configuration into krb5.conf instead of kdc.conf.  (In the
> past, it used to be required to put KDB configuration into krb5.conf.
> That odd requirement was relaxed somewhere around krb5 1.5 for most
> programs which run on the KDC, but a few have escaped the net, including
> kpropd.)
>
> I will open issues for both bugs and try to get them fixed for 1.9.
> Thanks for your investigative work.
>
>



More information about the Kerberos mailing list