Kerberos throwing SIGABORT in exit processing

[Ken Raeburn]

On May 12, 2008, at 08:43, John Hascall wrote:
> Has anyone else seen anything like this?
> The application completes successfully,
> then dies somewhere deep in exit processing
> (app's log):
>
> update_server: [info] Closing connection.
> Error detected by libpthread: Invalid mutex.
> Detected by file "/usr/src/lib/libpthread/pthread_mutex.c", line  
> 334, function "pthread_mutex_unlock".
> See pthread(3) for information.
> exited on Abort trap signal

Interesting... does this happen reproducibly?

I've seen reports of crashes in the thread support before, but aside  
from a race condition Ezra reported (is this program multithreaded?),  
mostly they seem to do with programs that load multiple versions of  
the com_err library and I guess finalization code for one gets run  
without it having been properly initialized, or something.  Can you  
give me some more details on this case?  For example, what libraries  
are getting loaded by this program?

I don't recall if gdb on i386-netbsd supports hardware watchpoints,  
but if it does, could you trace changes to "key_lock" in util/support/ 
threads.c and see if it's maybe getting destroyed more than once?

Ken