[Dovecot] Connection refused with auth-master afterupgradingto Dovecot 1.0 rc 28

[Jason Warner]

> On 10.4.2007, at 21.33, Jason Warner wrote:
> 
> > I still didn't get a core dump in /var/run/dovecot. Should I be
> > looking
> > somewhere else? That is the directory that the base_dir variable is
> > set to
> > in my dovecot.conf file.
> 
> Well, another way would be to attach gdb directly into dovecot-auth
> while it's still running:
> 
> gdb attach `pidof dovecot-auth`
> cont
> (wait for crash)
> bt full
> fr 4
> p *request
> 
> Anyway I should be able to figure out the assert by looking at the
> code too. Just a bit more difficult, since a couple of minutes wasn't
> enough. :)

Here's the best I could do:

(gdb) bt full
#0  0x004b9402 in __kernel_vsyscall ()
No symbol table info available.
#1  0x00686d40 in raise () from /lib/libc.so.6
No symbol table info available.
#2  0x00688591 in abort () from /lib/libc.so.6
No symbol table info available.
#3  0x0806bb9a in i_error ()
No symbol table info available.
#4  0x0806baac in i_panic ()
No symbol table info available.
#5  0x080552ce in auth_request_lookup_credentials_callback ()
No symbol table info available.
#6  0x0805f2bd in passdb_cache_init ()
No symbol table info available.
#7  0x0805f684 in passdb_cache_init ()
No symbol table info available.
#8  0x0805906b in db_ldap_connect ()
No symbol table info available.
#9  0x0806ef30 in io_loop_handler_run ()
No symbol table info available.
#10 0x0806e27c in io_loop_run ()
No symbol table info available.
#11 0x0805a51e in main ()
No symbol table info available.

It looks like a stack trace, but the p *request just gives "No symbol table
info available" for all 11 integers.

> 
> > I tried this with both the allow_all_users set and without it set.
> > That was
> > the fix though. By adding allow_all_users to my userdb static args,
> > everything is running just fine.
> 
> Only problem with that is that it doesn't know if the user exists or
> not. So depending on how you've set up permissions, it's possible
> that it delivers mails to unknown users as well, creating the
> directories if needed..

That is a problem. I'll go back to Postfix's virtual delivery method until I
don't have to use the allow_all_users flag.