Kernel changes for initramfs generation

[Colin Watson]

On Wed, Nov 30, 2005 at 02:43:37PM -0500, Jeff Bailey wrote:
> On mar, 2005-11-29 at 10:48 +0000, Colin Watson wrote:
> > That said, what will happen if initramfs-tools and its other
> > dependencies are unpacked but not configured? This is quite likely on
> > initial installation or on upgrades from earlier versions of the
> > distribution. Will update-initramfs bomb out (this would be OK) or would
> > it generate an incorrect initramfs (this would be rather less good)?
> 
> update-initramfs just copies pieces around, so it has no reasonably way
> of knowing this.  What happens now?  I suspect most of the pieces
> already just cope in some useful way, since mkinitrd was already being
> called at some random point - Whenever the kernel was unpacked.

If initramfs-tools and its dependencies all operate like Essential
packages and ensure that they work even when unconfigured, then this
would be a non-problem.

> > Also, what is the correct error handling behaviour for postinst scripts
> > calling update-initramfs? Should errors from update-initramfs cause the
> > postinst to abort, or should they be ignored? The above question is
> > relevant here too, since packages that are dependencies of
> > initramfs-tools may not be able to generate a correct initramfs yet.
> 
> Hmm.  My initial reaction is that failures from u-i should be ignored,
> since they most likely just indicate that the user has a custom
> initramfs and u-i is being a coward and refusing the break the system.
> 
> If each package that has something to do with the initramfs regenerates
> it, then by the time everything is unpacked and configured, the
> initramfs will be correct.
> 
> There could be an issue processing a config file or a hook script which
> should result in a more serious failure.  Unless the code has changed
> much since I wrote it, that case wasn't trapped.

OK, thanks. Can all this be noted in the spec?

Perhaps a command-line option is called for that causes update-initramfs
to exit zero in the case of a custom initramfs.

Cheers,

-- 
Colin Watson                                       [cjwatson at ubuntu.com]