Re: segfault with just running "rpm -qa"

[Jeff Johnson <n3npq nc rr com>]

Robert P. J. Day wrote:

> On Wed, 24 Sep 2003, Jeff Johnson wrote:
>
>  
>
> > LD_ASSUME_KERNEL is OK, but rebuilding rpm-4.2.1 (or rpm-4.2) without 
> > the 2 occurences
> > of --enable-posixmutexes in autogen;sh and rpm.spec will solve the 
> > problem too.
> >
> > You will lose unified process/thread locks and won't care a bit, as 
> > process locks are
> > all that's needed atm for concurrent access from applications that wish 
> > to access rpmdb.
> >    
>
> well, i *was* going to reboot under bk3 just to test, closed down most of
> my xterms, and did one more "rpm -qa" just to verify the problem.
> and it worked fine.  go figure.
>
> i didn't do *any* rebuilding, didn't use LD_ASSUME_KERNEL or anything
> out of the ordinary.  mind you, i still have the ubiquitous 
> /var/lib/rpm/__db* files, but a simple "rpm -qa" is back to working.
>  

"ubiquitous" -- yes, the __db* files are persistent, created by 1st root 
execution
of rpm. Note that __db* are not lock files, but rather files that 
contain locks.

Yes, __db* files can be blown away whenever, the only risk is that there 
will be a small
lock window race introduced. The lock race hardly ever matters, but is 
*THE* reason
why rpm cannot "clean up" __db* files and why the files *MUST* be 
persistent.

"working" -- Make sure you test "rpm -qa" as root, rpm as non-root 
honors exclusive
locks, but does not create shared locks (yet).

73 de Jeff