FreeBSD MBRs

[Christopher Key]

Ivan Voras wrote:
> Christopher Key wrote:
>
>> I've a machine with 3 SATA drives.  The first (ad8) with a standard
>> FreeBSD install in a single slice with /boot/boot0 MBR.  The remaining
>> two drives (ad10, ad12) are in a RAID1 mirror with 3 slices, and used
>> for storing data.  They have the /boot/mbr MBR.
>
> Ok. Let's call them the "first (ad8), second (ad10) and third (ad12) 
> drive"
>
>> After booting off various USB flash drives to try and update the BIOS on
>> my machine, it got into a state where during startup, it would display
>> 'Missing operating system' and hang.  What seems to have been happening
>> is that it was trying to boot from one of my data store drives, despite
>> the boot order of the disks set in the BIOS.
>
> Flashing the BIOS often resets all settings, including boot order. 
> Depending on your BIOS, controller and motherboard, you might need to 
> first select the active controller, then the drive on the controller. 
> In your case, you need to finally tell BIOS to boot of the first drive 
> (ad8).
>
> Another thing that might be happening (though judging from what you 
> said elsewhere it's not probable) is that the changed BIOS code 
> enumerates devices differently than it used to, so ad8 is no longer 
> the drive with boot0. This case will not hurt your RAID if you used 
> gmirror, it might for other systems.
>
>> The only solution that I found was to start booting from a USB flash
>> drive with a boot0 MBR, and to hit F5 to change to booting from my first
>> drive  After this, the machine then reboots quite happily until I hit F5
>> again, in which case I get the same 'Missing operating behavior'.  This
>> persists even while power cycling the machine.
>
> Ok, so now you're booting from the first drive, and with F5 you're 
> telling your boot loader to skip it and move on to the second drive, 
> which has the standard mbr (this mbr is what's displaying the "Missing 
> operating system" message).
>
>> I had imagined the boot process to be entirely stateless, certainly
>> across power cycles.  The BIOS executes the MBR on the first drive in
>> its boot boot.  The boot0 MBR then allowed you to either execute the
>> boot sector from any of the slices on the current drive, or to execute
>> the MBR from the next drive in the list.
>
> boot0 will by default remember what's the last option you booted from 
> and use it as the default for the next time. It may not be the best 
> thing for you, but I found it to be extremely useful behaviour in 
> several cases, since it allowed me to "script" a boot order when the 
> BIOS and the disk controller didn't get along for booting.
>
>> However, this clearly isn't what's happening.  Is it boot0 remembering
>> my F5 key stroke, or is it more likely that the BIOS is remembering
>> something?  Does anyone have any recommendations to avoid this in the
>> future?  Is putting boot0 on all three drives a good idea perhaps?
>
> It will not hurt in any case to put boot0 on any drives (as long as 
> you do it with the appropriate utility, else you may destroy the 
> partition table). In your case, it will only be useless.
>
> Here's something to try (I didn't try it): Do you have active 
> partitions on the second and the third drive? If so, you might want to 
> unmark them, and hopefully the boot loader won't display the drives in 
> the boot menu. Maybe.
>
> Or, if you're seriously worried about hitting F5 during boot, you 
> might try an alternative boot loader such as sysutils/extipl or 
> sysutils/grub.
Hello All,

Many thanks Ivan, Lowell, Jerry and Alex for your feedback.  I've done 
some experimentation, and had a read of the boot0 and mbr source code, 
and I think I know what's going on now:

When executed, boot0 retains a copy of the partition (slice) table in 
memory, and in it, marks each slice as being inactive.  When you hit 
F1-F4 to boot one of those partitions, it marks the selected slice as 
active, rewrites the partition table to the disk, and executes the 
appropriate boot sector for that slice.  If on the other hand you hit 
F5, it writes the partition table back out *without an active slice*, 
and executes the MBR from the next disk.

In my case, when I hit F5, boot/mbr was executed from my second disk, 
which displays 'Missing operating system' if there is an active slice 
but no magic number in the boot sector for that slice.  Now, as far as I 
can tell, when I next attempted to boot my system having hit F5 
previously, the BIOS didn't attempt to boot from my first drive because 
there was no active slice.  It instead tried to boot from my second 
drive, hence giving me the message 'Missing operating system'.

Now, when I inserted the USB drive and tried to boot, the BIOS saw the 
there was an active slice on it, and executed boot0 from it.  When I 
then hit F5 to switch to my first harddrive, boot0 was loaded fro it, 
and showed its default option as F5, having remembered that from last 
time.  If I then hit F1 to boot from the first slice on that disk, boot0 
would mark the first slice as active in the partition table and write 
that back to the disk before booting normally.  The next time I booted 
the machine, the BIOS would then see that there was an active slice on 
my first disk, load boot0 from it, and the machine would boot normally.

I think that this explains the behaviour I was seeing, and I certainly 
have a slightly better grasp of what's going on now.  I think that the 
confusion was because I was assuming that the BIOS was executing boot0 
from my first disk, which was *immediately* executing boot/mbr from the 
second disk without giving me an option to intervene.  In fact, boot0 
was never being executed, and the BIOS itself was going straight to the 
second disk, being the first one with an active slice.

I now intend to configure my mbrs as follows:

I'll keep boot0 on my first disk, with updates *disabled*, and the 
default option as F1
I'll put boot0 on my USB drive, again with updates disabled, but the the 
default option as F5.  Hence, if I leave the USB drive in by mistake, 
the system will still boot normally, although I can override that and 
boot from the disk instead.
I'll put an empty mbr on my second and third disks, so that boot0 will 
never allow me to try to boot from them.  (dd if=/dev/zero 
of=/tmp/zerombr count=1;  fdisk -B -b/tmp/zerombr /dev/mirror/gm0)

Once again, many thanks for the advice,

Regards,

Chris