[OE-core] [PATCH] image_types: Ensure /init exists for cpio rootfs archives

[Andrea Adami]

On Fri, Jan 6, 2012 at 11:58 PM, Andrea Adami <andrea.adami at gmail.com> wrote:
> On Fri, Jan 6, 2012 at 11:57 AM, Richard Purdie
> <richard.purdie at linuxfoundation.org> wrote:
>> On Fri, 2012-01-06 at 02:00 +0100, Andrea Adami wrote:
>>> On Fri, Jan 6, 2012 at 12:36 AM, Darren Hart <dvhart at linux.intel.com> wrote:
>>> >
>>> >
>>> > On 01/05/2012 04:30 AM, Richard Purdie wrote:
>>> >> On Thu, 2012-01-05 at 01:05 +0100, Andrea Adami wrote:
>>> >>> I'd say we have to test for the in-existence of the file instead of
>>> >>> blindly touching it.
>>> >>> Smthg like
>>> >>>
>>> >>> if [! -e ${IMAGE_ROOTFS}/init]; then
>>> >>> touch ${IMAGE_ROOTFS}/init
>>> >>> fi
>>> >
>>> > Why? Touch creates it if it doesn't exist and updates the modtime if it
>>> > does. It doesn't modify the content of the file if it exists. What is
>>> > the motivation for testing for it's existence?
>>> >
>>> touch fails with Permission denied because one package in my rootfs
>>> already created the symlink
>>>
>>> >>>
>>> >>>
>>> >>> Now, repeating that block for all 4 cpio images is a bit ugly... better ideas?
>>> >>
>>> >> Make it a shell function we call? I'd also check for
>>> >> ${base_sbindir}/init first too?
>>> >
>>> > Why check for that? So we can link to it instead of creating an empty file?
>>>
>>> Nonsense, kernel looks for /init
>>> ( http://www.kernel.org/doc/Documentation/filesystems/ramfs-rootfs-initramfs.txt
>>> )
>>
>> Yes, please ignore my comment about /sbin/init, I'm misremembering
>> something. We should deal with the symlink case though.
>>
>> Cheers,
>>
>> Richard
>>
>>
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core at lists.openembedded.org
>> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core
>
> ok, I think the issue is the behavior of touch when trying to change
> the timestamp of a symlink.
>
> http://fossplanet.com/f13/bug-7122-touch-gnu-coreutils-7-4-no-way-touch-symlink-57418/
>
> Using 'touch -h' seems solving the issue and the image is built.
> However, the timestamp of the /init symlink now differs from the
> timestamp of the binary.
>
> Regards
>
> Andrea


Ok, now I think I understand the issue:

the /init symlink in rootfs is absolute, being its path is pointing to
/usr/bin/kexecboot in our case.
Now, until this becomes the real rootfs, the symlink appears broken
and that's why adding the -h (do not dereference) fixes the case of
existent symlink.

Unfortunately it appears touch -h fails if /init is a file and not a
symlink (tested now).

Afterall, I think it would be easier to add a guard around touch as
initially proposed.

Just my 2 cents

Andrea