2.4.2 on Solaris - Crashes in mailbox_unlock_index

[Bron Gondwana]

On Sat, Oct 30, 2010 at 10:45:16PM +0000, Andy Fiddaman wrote:
> 
> Hi,
> 
> I've just upgraded to Cyrus 2.4 from 2.3 on Solaris 10 and I'm seeing lots
> of crashes with various utilities in mailbox_unlock_index().
> 
> (gdb) where
> #0  0xfffffd7ffef41e4a in _lwp_kill () from /lib/64/libc.so.1
> #1  0xfffffd7ffef3a753 in thr_kill () from /lib/64/libc.so.1
> #2  0xfffffd7ffeee6f89 in raise () from /lib/64/libc.so.1
> #3  0xfffffd7ffeec5f00 in abort () from /lib/64/libc.so.1
> #4  0x000000000041a1fb in mailbox_unlock_index (mailbox=0x58cae8,
> sdata=0x0)
>     at mailbox.c:1544
> #5  0x0000000000418c04 in mailbox_close (mailboxptr=0xfffffd7fffdfdd70)
>     at mailbox.c:1003

I don't suppose the stacktrace went any further up than that?  I'm
more interested in the call-site of mailbox_close, because that's
where a dirty mailbox will be being closed.
 
> Just after the upgrade, I was seeing this during attempted index upgrades
> (from lmtpd) and reconstruct was failing with the same problem. I removed
> the cyrus.* files and then reconstruct completed correctly, and subsequent
> reconstructs were fine too.
> 
> Everything now seems to be working correctly from the user's perspective,
> but I'm still getting these crashes with ipurge and cyr_expire.
> 
> Is anyone else seeing this?

Almost certainly a failure to commit before closing.  Which 2.4 are you
using?  2.4.2, or 2.4.0?  I'm pretty sure we fixed one of these in 2.4.2.
There's a fix for chk_cyrus that's not released yet, but I don't think
it's that.

Bron.