[RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Seewer Philippe]

Basically dracut has a problem with netroot when booting with multiple 
nics. If we have multiple interfaces online at the same time, we don't 
really know which interface to use for default-route or dhcp root-path 
source.

This series of patches adresses this problem by delegating the real 
interface configuration to netroot and introducing a new cmdline 
argument. I'll try to explain to solution with two use cases:


Case 1: Multiple interfaces with ip=dhcp or ip=client-ip... but no <dev>

The problem here is that we can't up just every eth that pops up. 
Default-route or whatever will get confused.

The solution here is that ifup and/or dhclient script just write the 
necessary ip statements into /tmp/net.$dev.* and netroot (which is 
synchronized by a lock) just uses one interface at a time, deconfiguring 
it if mounting fails.


Case 2: Multiple interfaces with multiple ip= lines

This is a case where the user expects all interfaces to be up before 
mounting. Think bonding, iscsi multipathing, etc.

The solution here is that the ip cmdline parsers enforces the presence 
of <dev> for multiple ip= lines and further demands that the new 
argument bootdev contains the name of the primary interface to be used 
for routing etc. Configuration is of course again delegated to netroot 
but in this case netroot takes additional care to "wait" for all 
required interfaces to be up.


In order to ensure correct ip= lines I've added a lot of cmdline 
parseing and some mostly minor rewrites of actual netroot handlers.


Thanks for reviewing (or accepting) the stuff,
Philippe
--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Warren Togami]

When you post this as a git repo, could you please include more 
descriptive changelogs on each commit?

dracut_install $(ls {/usr,}$LIBDIR/libnfsidmap*.so* 2>/dev/null )
dracut_install $(ls {/usr,}$LIBDIR/libnss*.so 2>/dev/null)
-dracut_install grep

instmods nfs sunrpc ipv6

some commits like [3/5] contain unrelated changes that turn out to be 
good, but they are not explained in the changelog.

Warren Togami
wtogami-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org
--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Warren Togami]

On 06/12/2009 04:35 PM, Warren Togami wrote:
> When you post this as a git repo, could you please include more
> descriptive changelogs on each commit?
>
> dracut_install $(ls {/usr,}$LIBDIR/libnfsidmap*.so* 2>/dev/null )
> dracut_install $(ls {/usr,}$LIBDIR/libnss*.so 2>/dev/null)
> -dracut_install grep
>
> instmods nfs sunrpc ipv6
>
> some commits like [3/5] contain unrelated changes that turn out to be
> good, but they are not explained in the changelog.
>

I'd like to implement the bridging support next, but it would really be 
good to do so on top of cleaned up patches from Seewer instead of 
conflicting with these changes.

Warren

--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[David Dillow]

On Fri, 2009-06-12 at 17:11 +0200, Seewer Philippe wrote:

> In order to ensure correct ip= lines I've added a lot of cmdline 
> parseing and some mostly minor rewrites of actual netroot handlers.
>
> Thanks for reviewing (or accepting) the stuff,
> Philippe

The series broke the NFS and NBD test suites, and I've not looked into
why yet. It dies on the first test.

I've started digging in and trying to review some of the changes, but
I'm just not going to have time before I leave; perhaps I can sneak away
for a bit next week.

Perhaps a theory of operation/design document as to what the flow is
expected to be, or what is handled where would help?

--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Seewer Philippe]

David Dillow wrote:
> On Fri, 2009-06-12 at 17:11 +0200, Seewer Philippe wrote:
> 
>> In order to ensure correct ip= lines I've added a lot of cmdline 
>> parseing and some mostly minor rewrites of actual netroot handlers.
>>
>> Thanks for reviewing (or accepting) the stuff,
>> Philippe
> 
> The series broke the NFS and NBD test suites, and I've not looked into
> why yet. It dies on the first test.

That's ... weird? I wasn't able to run the qemu/kvm test locally (don't 
ask, please). But I replicated the test arguments and ran them through 
the scripts. Should work, and nfsroot booted flawlessly Friday afternoon.

What's the die() error message?

> I've started digging in and trying to review some of the changes, but
> I'm just not going to have time before I leave; perhaps I can sneak away
> for a bit next week.

Thanks for looking into it!

> Perhaps a theory of operation/design document as to what the flow is
> expected to be, or what is handled where would help?

Is it really that complex? I'll write down my thoughts and add in a few 
examples.

Regards,
Philippe
--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[David Dillow]

On Sat, 2009-06-13 at 10:17 +0200, Seewer Philippe wrote:
> 
> David Dillow wrote:
> > The series broke the NFS and NBD test suites, and I've not looked into
> > why yet. It dies on the first test.
> 
> That's ... weird? I wasn't able to run the qemu/kvm test locally (don't 
> ask, please). But I replicated the test arguments and ran them through 
> the scripts. Should work, and nfsroot booted flawlessly Friday afternoon.
> 
> What's the die() error message?

The only thing printed for either test was
Warning: No ip= argument(s) provided, defaulting to DHCP

It never made it past the first subtest.

It's possible that I screwed up applying them, but git apply-mbox
applied all 6 (5+missing) in order without issue.

> > I've started digging in and trying to review some of the changes, but
> > I'm just not going to have time before I leave; perhaps I can sneak away
> > for a bit next week.
> 
> Thanks for looking into it!
> 
> > Perhaps a theory of operation/design document as to what the flow is
> > expected to be, or what is handled where would help?
> 
> Is it really that complex? I'll write down my thoughts and add in a few 
> examples.

I think part of the problem is that your series does too much in each
step. You have style changes, network device handling changes,
dhclient.conf changes, and command line parsing changes in the series.
Each one of those should be its own patch.

I would suggest getting the dhclient.conf and network device handling
changes in first; the dhclient.conf change should be uncontroversial and
we need to get the discussion on the handling going -- that is holding
back other progress. The command line handling is important, but lower
priority at the moment in my opinion.

Also, please try again to get the test suite running locally, and make
sure any support you add/drop is reflected in the test suite in the same
patch, if at all possible. This will help demonstrate the effect of the
changes you are making.

--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Seewer Philippe]

David Dillow wrote:
[snip]
>>> Perhaps a theory of operation/design document as to what the flow is
>>> expected to be, or what is handled where would help?
>> Is it really that complex? I'll write down my thoughts and add in a few 
>> examples.
> 
> I think part of the problem is that your series does too much in each
> step. You have style changes, network device handling changes,
> dhclient.conf changes, and command line parsing changes in the series.
> Each one of those should be its own patch.
> 
> I would suggest getting the dhclient.conf and network device handling
> changes in first; the dhclient.conf change should be uncontroversial and
> we need to get the discussion on the handling going -- that is holding
> back other progress. The command line handling is important, but lower
> priority at the moment in my opinion.

Hmmm... yes and no. To me, the command line handling is important as 
well since at least the ip= parser needs to work for the network handling.

> Also, please try again to get the test suite running locally, and make
> sure any support you add/drop is reflected in the test suite in the same
> patch, if at all possible. This will help demonstrate the effect of the
> changes you are making.

ok, thanks for the great feedback!

Working on getting the test-suite running right now, will try to split 
my patches later the way you suggested. I guess apologies are in order 
for the "mess".

Regards,
Philippe
--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Warren Togami]

On 06/13/2009 08:49 AM, Seewer Philippe wrote:
>
> Working on getting the test-suite running right now, will try to split
> my patches later the way you suggested. I guess apologies are in order
> for the "mess".

Hi,

We are still waiting for a git repo to pull from so we can review this 
patch set.  Please split unrelated changes into their own commits and 
write detailed changelog entries on each.

Warren
--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Seewer Philippe]

Warren Togami wrote:
> On 06/13/2009 08:49 AM, Seewer Philippe wrote:
>>
>> Working on getting the test-suite running right now, will try to split
>> my patches later the way you suggested. I guess apologies are in order
>> for the "mess".
> 
> Hi,
> 
> We are still waiting for a git repo to pull from so we can review this 
> patch set.  Please split unrelated changes into their own commits and 
> write detailed changelog entries on each.

There you go

http://github.com/seewer/dracut-seewer/commits/master

Patches are split as atomically as made sense (to me). Except for the 
ISCSI test-suite which I've ignored, all tests now pass. Accordingly the 
tree contains some fixes that weren't in the patches last week, this is 
mainly to fix the test-suites to run with Debian based distros and some 
enhancments that were discovered while running the test cases.

As for further examples how the multinic stuff should work, I've tried 
document an example in the two last multinic patches:

http://github.com/seewer/dracut-seewer/commit/db8158439dac10e7462203d3de44fd794bd35e2b
http://github.com/seewer/dracut-seewer/commit/8ecd9d010a46999263b629b16954156e34083244

Apologies it took longer than expected, the only thing I can say there 
is that Murphy is always right.

Regards,
Philippe
--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC PATCH 0/5] Multiple interfaces support and lots of cmdline parsing

[Harald Hoyer]

On 06/17/2009 08:49 AM, Seewer Philippe wrote:
> Warren Togami wrote:
>> On 06/13/2009 08:49 AM, Seewer Philippe wrote:
>>>
>>> Working on getting the test-suite running right now, will try to split
>>> my patches later the way you suggested. I guess apologies are in order
>>> for the "mess".
>>
>> Hi,
>>
>> We are still waiting for a git repo to pull from so we can review this
>> patch set. Please split unrelated changes into their own commits and
>> write detailed changelog entries on each.
>
> There you go
>
> http://github.com/seewer/dracut-seewer/commits/master
>
> Patches are split as atomically as made sense (to me). Except for the
> ISCSI test-suite which I've ignored, all tests now pass. Accordingly the
> tree contains some fixes that weren't in the patches last week, this is
> mainly to fix the test-suites to run with Debian based distros and some
> enhancments that were discovered while running the test cases.
>
> As for further examples how the multinic stuff should work, I've tried
> document an example in the two last multinic patches:
>
> http://github.com/seewer/dracut-seewer/commit/db8158439dac10e7462203d3de44fd794bd35e2b
>
> http://github.com/seewer/dracut-seewer/commit/8ecd9d010a46999263b629b16954156e34083244
>
>
> Apologies it took longer than expected, the only thing I can say there
> is that Murphy is always right.
>
> Regards,
> Philippe
> --


Pushed with some additional changes.
--
To unsubscribe from this list: send the line "unsubscribe initramfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html