Re: PATCH (updated): 'lilo -R' functionality

[Yoshinori K. Okuji]

[I added bug-grub into Cc:, because we don't talk about secrets.]

At Tue, 10 Dec 2002 16:58:39 +0000,
Andrew Stribblehill wrote:
> How about this...

Sorry, I don't like this, either. Perhaps I should have made it clear
why I don't like it.

The reason is that it always modifies the boot image directly. This is
dangerous, especially on a running OS, because the OS may relocate the
data of the image in a disk physically. Also, this is not desirable,
if using a flash memory, because the same sector is rewritten
frequently, so it is deteriorated very soon.

You might want to insist that savedefault does the same thing. That's
right, but I really regret that savedefault was implemented this
way. So, if you want to add one more feature into savedefault, please
consider redesigning it.

Possibly NVRAM is a good solution, if available. But NVRAM is not
standardized in PC, unfortunately, so using NVRAM is very
machine-specific.

Another way is to use a file on a disk as a substitute for
NVRAM. Although GRUB doesn't support writing files, GRUB can write
data into disks. If my understanding is correct, it is safe for GRUB
to modify the contents of a file, as long as the size of the file
doesn't change.

So I can think of this kind of scenario:

1) Create a NVRAM-like file, say, "/boot/grub/nvram.dat" on an OS. The
   file must be bigger than the size of required data. 512 bytes might
   be sufficient.

2) When setting a boot-once flag, you write something like this to the
   file:

bootonce 1 # The entry number.
end # Prevent GRUB from reading garbage.

3) When GRUB starts up, read "/boot/grub/nvram.dat" if exists. If it
   finds the line "bootonce blahblah", GRUB rewrites the file so that
   it won't boot the entry in next boot time, like this:

end

   And, GRUB boots the entry.

4) In next boot, GRUB doesn't find bootonce, so it boots up the
   default entry.

I don't know if this idea is really nice, because this can be
error-prone (consider how many error checks would be required...).

Thanks,
Okuji