[Pkg-sysvinit-devel] Bug#620788: Automount problem with backports nfs-common and squeeze initscripts

[Daniel Kahn Gillmor]

(this is in reference to a debian-backports discussion about NFS 
mounting failures using rpcbind on squeeze, archived at: 
https://lists.debian.org/debian-backports/2012/03/msg00038.html )

On 03/12/2012 02:38 PM, Alessandro Grassi wrote:
> The NFS sources don't get mounted directly by init (in which case that line
> would have worked), instead they are mounted by the post-up hook in
> ifupdown (see /etc/network/if-up.d/mountnfs), as i mentioned in my original
> mail.

ah, i understand now, thanks for the explanation.

> That script belongs to "initscripts" (which, if my understanding is
> correct, would be very dangerous to "backport" once patched) and has a
> hardcoded check for /etc/init.d/portmap, so it has no knowledge of either
> insserv.conf nor rpcbind.

right.  the difference seems to be:

--- 2.88dsf-13.1/etc/network/if-up.d/mountnfs	2012-03-12
+++ 2.88dsf-22/etc/network/if-up.d/mountnfs	2011-12-28
@@ -98,9 +100,10 @@
  	# and/or rpc.idmapd, and loads the right kernel modules if
  	# applicable) if we use Kerberos and/or NFSv4 mounts.
  	#
-	if [ "$start_nfs" = yes ] && [ -x /etc/init.d/portmap ] && [ -x 
/etc/init.d/nfs-common ]
+	if [ "$start_nfs" = yes ] && [ -x /etc/init.d/nfs-common ]
  	then
-		/etc/init.d/portmap start
+		[ -x /etc/init.d/portmap ] && /etc/init.d/portmap start
+		[ -x /etc/init.d/rpcbind ] && /etc/init.d/rpcbind start
  		/etc/init.d/nfs-common start
  	fi


This appears to come from #620788 (cc'ed here), which i've just 
unarchived in the BTS.

> I'm not really an expert on these matters, but since rpcbind can
> effectively replace portmap, it could be a solution to declare it as a
> portmap replacement in debian/control and make it provide the rpcbind ->
> portmap symlink on /etc/init.d to make things work.
> It's not very clean but would allow not to touch the distribution's files
> too much.

ugh, this sounds pretty painful to me, and probably not really 
appropriate for squeeze-backports or debian stable.

> Another solution could be to package the patched file and distribute a
> patched version of initscripts in the backports, but i don't know how hard
> to maintain that would be.

This approach (using the patch above) actually seems plausible to me as 
something that we might want to fix in a squeeze point release, since i 
think rpcbind should be a viable alternative to portmap in stable anyway.

sysvinit maintainers, release team -- what do y'all think of this 
proposal?  I'd be happy to prepare a package with this targeted fix for 
squeeze-proposed-updates if you think that would be acceptable.

Please let me know!

Regards,

	--dkg