Ident trouble?

[Russ Allbery]

Jeffrey M Vinocur <jeff at litech.org> writes:

> I have no idea why since rc.c hasn't changed in ages, but since the
> 2.4.2 prerelease or so I've had innd hang several times in the following
> fashion.  I suppose it could be coincident with some other system
> changes around here, but even if there's something wrong with our
> identd, innd shouldn't die as a result...

> #0  0x40127a98 in read () from /lib/libpthread.so.0
> #1  0xbffff0a0 in ?? ()
> #2  0x080653f8 in GoodIdent (fd=96, identd=0x80d4b10 "news") at rc.c:190
> #3  0x080661f9 in RCreader (cp=0xbffff0a0) at rc.c:642
> #4  0x0805b173 in CHANreadloop () at chan.c:1015
> #5  0x0805d630 in main (ac=2, av=0x8243880) at innd.c:666

The ident code currently implements no timeouts whatsoever.  Hm.  That's a
problem.

I'm afraid that the correct solution here is going to be annoying.  We
really need to put the connection into a different mode, open a new file
descriptor for the ident callback, create a new type of channel for
accepting the ident callback, and put that channel into the main select
loop.  That's... incredibly annoying, but I don't see any good
alternatives if we want to support ident.

I assume that you do have ident checking turned on (via setting identd in
incoming.conf)?

This is more work than I'm going to be able to do right now.  I'll make a
note to look at this later, and in the meantime will add a caution to the
man page.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.