Re: current best practice for booting >2TiB volumes?

[eric%cirr.com@localhost (Eric Schnoebelen)]

David Laight writes:
- On Sun, Feb 20, 2011 at 10:57:27PM -0600, Eric Schnoebelen wrote:
- > 
- > David Laight writes:
- > - On Fri, Feb 18, 2011 at 06:28:34PM -0500, Matthew Mondor wrote:
- > - > On Fri, 18 Feb 2011 12:09:39 -0600
- > - > eric%cirr.com@localhost (Eric Schnoebelen) wrote:
- > - > 
- > - > > When I attempt to boot (after relocating wd2/3 to wd0/1), the gpt
- > - > > mbr loads bootxx_ffsv1, but bootxx_ffsv1 complains it can't find
- > - > > /boot in the root filesystem.
- > - 
- > - The bootxx code probably requires that to root fs be at the start of
- > - the raid volume.
- > 
- > Hmm, that might be the problem.  I was treating it like a wdXa,
- > where the NetBSD partion starts at 64 (offset 63).  I'll try
- > recreating the FS with root starting at offset 0.
- 
- The RF 'disk' is usually formatted like a bios partition.
- So sector 1 is a disklabel (etc).
- (but the boot code won't look at it).

yes, that was the problem.  Moving the a partion to the begining
of the raid array worked.

- > - It might also get confused by any disklabel in the pbr (presumably
- > - you don't have one).
- > 
- > installboot(8) does complain about there being no pbr..
- 
- Installboot adds the code to the pbr! It avoids writing to the sector
- that might contain a label.

It did complain about the PBR being invalid, at least the first
time it was executed..  but that seems to be of no consequence
in this case.

- > - Certainly 'boot from RF' relied on the bootxx code adding in 64 sectors
- > - and trying again after an initial file open failed.
- > -
- > - But I can't remember if this was in any way related to detecting
- > - an RF entry in the pbr's disklabel.
- > - 
- > - Your disks may have larger physical sectors. So you really want
- > - to align the fs fragments/blocks onto physical sectors. The offset
- > - 63 sector offset (one nominal track) isn't a good value!
- > 
- > I was reading that as offset of 63, with sectors 0-63 (aka 64)
- > being unused, and sector 64 being the first sector of my
- > partion.
- > 
- > Am I mis-interpreting?
- 
- Yes - the offset of 63 means that that sectore 0-62 are unused.
- I (think it was me) changed the default offsets for large volumes
- to be 1M (and 1M alignments) instead of track/cylinder alignment.

You'd think that after 15 years with NetBSD, I'd remember more
of this.. Of course, the fact I have to muck with it so seldom
says something about the reliability of NetBSD as an OS..


--
Eric Schnoebelen                eric%cirr.com@localhost         
http://www.cirr.com
        %SYSTEM-F-ANARCHISM, The operating system has been overthrown