klibc - Jan 2006 - symlinks and initramfs (was klibc-1.2.1: kinit works...)

On Mon Jan 30 08:56:06 PST 2006, H. Peter Anvin wrote:
> That being said, there is no reason that you can't run code before it;
> to make that simpler, in fact, I'm trying to set it up so that the
> default initramfs looks like:
>
> /dev/console
> /dev/zero
> /dev/null
> /kinit
> /init -> /kinit
I would request we not put in symlinks.  The problem is that they are 
difficult to override.  Loading a later cpio has the effect of 
(attempting to) write over the current target of the symlink.  There is 
no "remove existing file or directory" entry in cpio.  This and the 
fact that initramfs only extends files are the two biggest limitations 
I've run into during the past 3 plus years.

We would be better off with a #! script instead, although that means 
the target has to accept at least the file name.

milton

On 1/31/06, Milton Miller <miltonm@bga.com> wrote:
> I would request we not put in symlinks.  The problem is that they are
> difficult to override.  Loading a later cpio has the effect of
> (attempting to) write over the current target of the symlink.  There is
> no "remove existing file or directory" entry in cpio.  This and
the
> fact that initramfs only extends files are the two biggest limitations
> I've run into during the past 3 plus years.
I kind of thought that was the point he was making.  /kinit exists as
a hard file, symlinked to /init so that a user built initramfs can
replace /init easier, while still keeping /kinit in place.  Sure, you
could load an initramfs which overwrites /kinit while still expecting
/init to work as normal, but at that point you're just trying to
protect fools from themselves.

Aaron

Chris Wedgwood wrote:
> On Wed, Feb 01, 2006 at 06:26:35PM -0800, H. Peter Anvin wrote:
> 
>>Since the issue in question was overwriting a symbolic link with
>>either another symbolic link or a file, it should be a non-issue.
> 
> On reflection though I think it should just unlink in both cases.
Which two cases?

	-hpa

On Thu, Feb 02, 2006 at 10:10:29AM -0800, H. Peter Anvin wrote:
> Unless linkable.
linkable? what do you mean?

Chris Wedgwood wrote:
> On Thu, Feb 02, 2006 at 10:10:29AM -0800, H. Peter Anvin wrote:
> 
>>Unless linkable.
> 
> linkable? what do you mean?
Match the conditions for making a hard link.

	-hpa