I've recently updated the nfs-utils in rawhide with the
latest patches from the SourceForge CVS tree and the
latest CITI patches (1.0.7-4).

In testing these patches, I notice that when the server was started
and a SIGHUP was sent to rpc.idmapd to open the nfs4.nametoid/channel
and nfs4.idtoname/channel files, the second open (the nfs4.idtoname one)
failed because the path (i.e. ic->ic_path) was NULL.

Now the reason the ic_path was NULL was because it was never set
during the call to nfsdopen(). nfsdopen() looks like:
nfsdopen(char *path)
{
     return ((nfsdopenone(&nfsd_ic[IC_NAMEID], IC_NAMEID, path) == 0 &&
             nfsdopenone(&nfsd_ic[IC_IDNAME], IC_IDNAME, path) == 0) ? 0 
: -1);
}

Note: the call to nfsdopenone() is how the path is set in each nfsd_ic[]
entry and nfsdopen() is only called once.

So when rpc.idmap comes up and the first call to nfsdopenone() fails
(because the server is not running) the path in nfsd_ic[IC_IDNAME] is
never filled in because the second nfsdopenone() never happen...

Now there was a CITI patche (idmapd_revert_fix_reopen_on_sighup.dif)
that tried to address this problem but did seem to fix it.. The
attached patch fix the problem by initializing both nfsd_ic[IC_IDNAME]
and nfsd_ic[IC_NAMEID] structures with the needed info...
I figured since there is no way of changing these paths or filenames
by command line args, why not just set them during compile time...
so that's what this patch does.

This patch also changes how nfsdreopen_one() handles the
case where the event has already been set. Unlike the CITI
patch (idmapd_revert_fix_reopen_on_sighup.dif) which just
just does not register the second event, my patch deletes
the old event and the registers the new one. It just seems like
the right thing to do since a SIGHUP means a new server just
started so we probably should create a new event as well...

steved.