On Dec 2, 2012, at 7:57 PM, John Reiser <jrei...@bitwagon.com> wrote:

> Fourteen years ago in 1998 I published a boot loader for i386 that loaded 
> linux
> kernel and initrd using filesystem lookup by name from ext2, occupying 446 
> bytes
> of MBR plus 2*510 bytes of bootblock from ext2.  The space also allowed some
> tens of bytes of kernel command line.  So "must use block lists" was not true 
> then.

Yeah obviously I asked the wrong question, again. I will digress instead of 
asking a new (bad) question:

1.  Another possible work around.
In addition to the "do nothing" and "install to MBR + MBR gap" options, is the 
"make a core.img and grub.cfg, but do not install anything to MBR, the MBR gap, 
or use block lists" option. This core.img could be called by another boot 
manager, presumably already installed and in use by the user, the one they 
don't want stepped on in the first place. If they load the 
/boot/grub2/core.img, it will in turn load the grub.cfg and they get a GRUB 
menu as expected. To do this anaconda would use grub2-mkimage instead of 
grub2-install; it would use grub2-mkconfig as it currently does to generate the 
grub.cfg.

I think that work around satisfies most concerns, albeit without automatically 
linking the existing boot manager to the core.img file. The user wouldh have to 
do that themselves (rather than relying merely on a change in the MBR active 
flag).

2.
Incidentally, Btrfs has a 64KB offset before the start of the file system, 
which I confirmed on the btrfs list is intended for a boot mananager/loader. 
Because of how btrfs works, block lists are not merely "not recommended" but 
proscribed, the --force option fails, but as it turns out you don't need to use 
block lists with Btrfs because core.img and then some will happily fit in the 
rather enormous boot loader region completely contiguously intact. And I've 
confirmed that it does work, both installation of GRUB2 into this region, as 
well as it being loaded and functioning (ergo 100% Btrfs boot, even with boot 
and root in subvols).

So the solution to the problem maybe for F19/F20 is actually Btrfs by default, 
and then grub2-install to a partition can be employed once again and simply 
just work without extra effort on the part of the user as is the case with work 
around 1.


Chris Murphy
-- 
devel mailing list
devel@lists.fedoraproject.org
https://admin.fedoraproject.org/mailman/listinfo/devel

Reply via email to