unloading the ipv6 module

unloading the ipv6 module

[Daniel Lezcano]

Hi,

I saw in net/ipv6/Kconfig a comment saying
"IPv6 as module will cause a CRASH if you try to unload it".

If I refer to this comment, that means unloading of ipv6 is not 
supported (which makes sense),
will it be one day supported?

 -- Daniel

Re: unloading the ipv6 module

[David Miller]

From: Daniel Lezcano <daniel.lezcano@free.fr>
Date: Thu, 26 Feb 2009 14:22:12 +0100

> I saw in net/ipv6/Kconfig a comment saying
> "IPv6 as module will cause a CRASH if you try to unload it".
> 
> If I refer to this comment, that means unloading of ipv6 is not
> supported (which makes sense), will it be one day supported?

It's very unlikely to be supported any time soon.

Maybe "one day" as in 5 years from now at the earliest.

It's just such an enormous set of problems to solve to make it work,
and there are so many more urgent issues to work on.

Re: unloading the ipv6 module

[Daniel Lezcano]

David Miller wrote:
> From: Daniel Lezcano <daniel.lezcano@free.fr>
> Date: Thu, 26 Feb 2009 14:22:12 +0100
>
>   
>> I saw in net/ipv6/Kconfig a comment saying
>> "IPv6 as module will cause a CRASH if you try to unload it".
>>
>> If I refer to this comment, that means unloading of ipv6 is not
>> supported (which makes sense), will it be one day supported?
>>     
>
> It's very unlikely to be supported any time soon.
>
> Maybe "one day" as in 5 years from now at the earliest.
>
> It's just such an enormous set of problems to solve to make it work,
> and there are so many more urgent issues to work on.
>   
I was wondering if it would have make sense to remove the unloading 
module code from ipv6 which is big at all and not testable.

Re: unloading the ipv6 module

[David Miller]

From: Daniel Lezcano <daniel.lezcano@free.fr>
Date: Thu, 26 Feb 2009 14:43:02 +0100

> I was wondering if it would have make sense to remove the unloading
> module code from ipv6 which is big at all and not testable.

The majority of that code is needed anyways for the module load
failure paths.

Re: unloading the ipv6 module

[Daniel Lezcano]

David Miller wrote:
> From: Daniel Lezcano <daniel.lezcano@free.fr>
> Date: Thu, 26 Feb 2009 14:43:02 +0100
>
>   
>> I was wondering if it would have make sense to remove the unloading
>> module code from ipv6 which is big at all and not testable.
>>     
>
> The majority of that code is needed anyways for the module load
> failure paths.
Right, I missed this point. I have my answer :)
 Thanks
  -- Daniel

Re: unloading the ipv6 module

[Matti Aarnio]

On Thu, Feb 26, 2009 at 05:24:11AM -0800, David Miller wrote:
> From: Daniel Lezcano <daniel.lezcano@free.fr>
> Date: Thu, 26 Feb 2009 14:22:12 +0100
> 
> > I saw in net/ipv6/Kconfig a comment saying
> > "IPv6 as module will cause a CRASH if you try to unload it".
> > 
> > If I refer to this comment, that means unloading of ipv6 is not
> > supported (which makes sense), will it be one day supported?
> 
> It's very unlikely to be supported any time soon.
> 
> Maybe "one day" as in 5 years from now at the earliest.

My guess is that somebody reasonably proficient coder with
a week of free time can get it to unload properly.

I did it once in 2.4 days, I recall..  Not completely, but mostly.

> It's just such an enormous set of problems to solve to make it work,
> and there are so many more urgent issues to work on.

Like refactoring the system to have IPv4 specific codes,
IPv6 specific codes, and common codes...   But DaveM has
probably other items in mind.

  /Matti Aarnio

Re: unloading the ipv6 module

[Stephen Hemminger]

On Thu, 26 Feb 2009 20:09:20 +0200
Matti Aarnio <matti.aarnio@zmailer.org> wrote:

> On Thu, Feb 26, 2009 at 05:24:11AM -0800, David Miller wrote:
> > From: Daniel Lezcano <daniel.lezcano@free.fr>
> > Date: Thu, 26 Feb 2009 14:22:12 +0100
> > 
> > > I saw in net/ipv6/Kconfig a comment saying
> > > "IPv6 as module will cause a CRASH if you try to unload it".
> > > 
> > > If I refer to this comment, that means unloading of ipv6 is not
> > > supported (which makes sense), will it be one day supported?
> > 
> > It's very unlikely to be supported any time soon.
> > 
> > Maybe "one day" as in 5 years from now at the earliest.
> 
> My guess is that somebody reasonably proficient coder with
> a week of free time can get it to unload properly.
> 
> I did it once in 2.4 days, I recall..  Not completely, but mostly.
> 
> > It's just such an enormous set of problems to solve to make it work,
> > and there are so many more urgent issues to work on.
> 
> Like refactoring the system to have IPv4 specific codes,
> IPv6 specific codes, and common codes...   But DaveM has
> probably other items in mind.
> 

The issue is the reference counting. IPV6 has some internal sockets
that stay open and hold therefore hold references to itself.
These reference counts must be accounted for and managed properly.

Re: unloading the ipv6 module

[David Miller]

From: Matti Aarnio <matti.aarnio@zmailer.org>
Date: Thu, 26 Feb 2009 20:09:20 +0200

> On Thu, Feb 26, 2009 at 05:24:11AM -0800, David Miller wrote:
> > It's just such an enormous set of problems to solve to make it work,
> > and there are so many more urgent issues to work on.
> 
> Like refactoring the system to have IPv4 specific codes,
> IPv6 specific codes, and common codes...   But DaveM has
> probably other items in mind.

The biggest problem is the reference counting of objects.

And if you make ipv6 sockets grab references to the ipv6 module,
the ipv6 module itself creates a few for ICMP6 processing, so
you somehow have to get rid of those in order for the module
unload to be invokable at all.

There are hundreds of cases like that you have to find solutions for
before you can cleanly unload ipv6.