Bugzilla Bug 197730

modprobe during the initramfs fails. It does not show any modules loading
automatically, then if I use shell and run modprobe manually, I get this: 
# modprobe dm-mod
modprobe: exec of insmod: No such file or directory
modprobe: failed to load module dm-mod

Manually running insmod for each of the needed modules (with their deps first)
works, so it's just modprobe that is broken somehow.

My system:
Linux grubbs-int 2.6.24-rc1-pmp-gc1698780 #3 SMP Wed Oct 31 19:53:34 PST 2007
x86_64 Intel(R) Core(TM)2 CPU 6400 @ 2.13GHz GenuineIntel GNU/Linux

I'm digging into it more now to try and find out why.

Ok, something is broken in the $PATH handling of that busybox.

If I do:
# modprobe FOO
I get the failure
If I do:
# cd /sbin
# modprobe FOO
It works fine.

Upgrading to blocker/P1.

Perhaps this particular issue is why genkernel was building a static insmod and
friends instead of using the busybox-provided ones. We've got a few options
here:

* cd to /sbin before running insmod in /linuxrc
* go forward with the upgrade to busybox-1.4.2 and see if it fixes the issue
* restore the code that builds the static insmod and friends

All of these are pretty easy, although, I'm not sure any of them are actually
desirable.

Robin, what is the PATH in the initramfs shell?

Hmm, I can't reproduce this with an install cd kernel/initramfs produced by
genkernel-3.4.9_pre3. As far as I know, nothing has changed between _pre3 and
_pre5 that might account for this. I'll bump to _pre5 and rebuild and see if I
can reproduce.

I just rebuild my installcd with genkernel-3.4.9_pre5, and there it is. It's
the same version of busybox in both. The PATH looks sane:

/ # echo $PATH
/usr/local/bin:/usr/bin:/sbin:/bin
/ # which insmod
/sbin/insmod

The only change that I can see that might have affected busybox is the mdadm
patch  from bug #172128

agaffney: I dug a bit more and found some more on it.

In _pre3, there was the generic/modprobe script that was installed as the
initramfs modprobe, while in _pre5, the busybox modprobe is used.

So then I theorize that the _pre3 in-busybox modprobe was probably broken
already, and we just didn't run into it because of the custom modprobe script.

OK.  I think I have a fix in SVN.  I simply reverted the removal of the script
until we can investigate if it is needed with newer busybox versions.

OK.  This should be fixed in 3.4.9_pre6.