initramfs support

Bryan Kadzban bryan at kadzban.is-a-geek.net
Sat Jul 7 11:05:56 PDT 2007


-----BEGIN PGP SIGNED MESSAGE-----
Hash: RIPEMD160

Dan Nicholson wrote:
> Simon Geard has been using your initramfs setup.

Uh oh, a user!  :-P

(Nah, just kidding.  It should be working fine now, especially for a
non-LVM and non-dmraid setup.)

> If you have anything specific you want tested, maybe you can ask him.

I'm doing some testing inside qemu right now, at Alexander's suggestion.
(It's certainly easier than rebooting 12 times, like I did when I first
set up the basic package and started running it on my machine...)

I'm a bit more interested in what types of setups the initramfs needs to
support, though.  From the initial bug, it sounds like root-on-LVM2 and
root-on-dmraid (and combinations of the two) require it.  It looks like
hibernation also requires it (to fire up the hibernated image).  I don't
know if any other setup does.

But Simon, if you do find something that doesn't work right, let me
know.  :-)

> http://linuxfromscratch.org/pipermail/blfs-support/2007-July/063195.html

- From that thread:

>> So what, fstab contains /dev/disk/by-label/BootDisk for the root 
>> entry?
> 
> Normal fstab entries, but it gets parsed and any LABEL= or UUID= 
> entries get temporary udev rules that use vol_id to test the LABEL
> and creates the /dev/root symlink. I didn't write that part. I don't
> know how Bryan's handling this offhand.

I'm not parsing fstab; I'm requiring a root= option.  Since I have a
full udev setup (including persistent symlinks for disks), you can pass
root=/dev/disk/by-uuid/xxxxx or root=/dev/disk/by-label/xxxx, and it
should work.  (And with the expanded (and dynamic) space available for
the command line in 2.6.21, long root device names are possible.)

However, I don't know whether root=LABEL=xxx or root=UUID=xxx work; I
believe mount parses up /proc/partitions and opens each device for that.
If it does, then it'll get confused on a dmraid mirror LABEL= or UUID=
(since all component disks will have the same partition structure and FS
metadata) -- and it'll probably mount just one disk's partition -- but
it might work on LVM2.  It'll definitely work on normal IDE or SCSI.

(Although root=LABEL= or root=UUID= on dmraid or LVM don't make any
sense anyway: the /dev/mapper/* and /dev/<VG>/<LV> names are already
persistent.)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFGj9YES5vET1Wea5wRA5TGAJ0ZLKSB5gaCkXmdfdzAyNf+EyKnJQCgvbyY
U7M/ytfEvOMhhlcoXhWYavI=
=edQd
-----END PGP SIGNATURE-----



More information about the lfs-dev mailing list