Bug 292894 – net.lo dependency on localmount is too late. net.lo should dep on root, and other net.* should be on localmount

• Status: RESOLVED INVALID
• Product: Gentoo Hosted Projects

[Description Robin Johnson 2009-11-12 09:23:30 UTC]

Dealing with bug 232347  for putting critical filesystems (/var) on iSCSI, and I was wondering why net.lo needs localmount.

For non-loopback devices, sure I can see a need for /usr, but for loopback itself we should be able to start right after root.

Patch attached.

[Comment 1 Robin Johnson 2009-11-12 09:23:47 UTC]

Created attachment 209992 [details, diff]
net.lo-earlier.patch

[Comment 2 Roy Marples 2009-11-12 14:09:19 UTC]

This will break people who configure a domain/nameserver against the loopback interface and have resolvconf installed as resolvconf needs /var/run

[Comment 3 Roy Marples 2009-11-12 14:49:14 UTC]

http://roy.marples.name/projects/openresolv/changeset/85fbb935f36c0f1920b250005254b0ba34e96c27

This allows the user to configure the resolvconf state dir, to say /lib[exec]/rc/init.d/resolvconf or /dev/shm/resolvconf

It may be an idea to add a new use flags to the openresolv ebuild.
shm use flag (linux only systems)
and/or
openrc use flag (/lib/rc/init.d/resovlconf)
so that the default var dir is changed and this change works by default.

[Comment 4 Robin Johnson 2010-08-04 22:08:48 UTC]

I propose we take that patch, anybody using resolvconf in that manner needs to add localmount as a dep of net.lo or use the -I option to not use /var.

[Comment 5 William Hubbs 2010-12-04 18:51:25 UTC]

Robin,

I think this bug got lost in the shuffel somewhere.

(In reply to comment #4)
> I propose we take that patch, anybody using resolvconf in that manner needs to
> add localmount as a dep of net.lo or use the -I option to not use /var.

Which patch are you referring to? Comment #3 seems to give a good solution, but you also have a patch in comment #1. Can you clarify?

Thanks,

William

[Comment 6 Robin Johnson 2010-12-04 19:09:58 UTC]

williamh: the resolvconf patch should already be in the tree, so it's just the other one.

The other idea I had is checking /etc/resolv.conf for 127.0.0.1 and doing the same as current if that's detected.

[Comment 7 William Hubbs 2011-04-12 13:36:47 UTC]

Commit 8e92536 contains the patch from comment #1 as well as
documentation on how to adjust dependencies if necessary.

In further testing, I found that net.lo was not starting any earlier, so
commits 85827d4 and 03cd55a were added to fix this. Now the loopback
interface starts right after the root filesystem is mounted read/write.

[Comment 8 William Hubbs 2011-04-15 21:56:30 UTC]

I am reopening this bug because the fixes for it caused bug #363693.

I plan to revert these fixes, so my next question is, should this bug
block stabilization? I'm thinking not, but I want comments from others
on the team.

[Comment 9 William Hubbs 2011-04-17 23:55:46 UTC]

I am removing this from the tracker for stabilization.

It does need to be looked at again, but I think it can wait until after openrc is stable.

[Comment 10 Robin Johnson 2012-02-09 10:03:56 UTC]

WilliamH: I'm thinking this one should go either go away because there are more and more tools that have to live in /usr anyway with their libraries.

[Comment 11 William Hubbs 2012-02-09 21:53:32 UTC]

Agreed, there isn't much we can do about this.

[Comment 12 William Hubbs 2012-02-09 21:55:43 UTC]

There isn't anything we can do about this because more and more tools
are moving to /usr anyway.