Re: [PATCH, RFC] mke2fs: get device topology values from blkid

[Andreas Dilger <adilger@sun.com>]

On Sep 18, 2009  15:40 -0400, Martin K. Petersen wrote:
> >>>>> "Eric" == Eric Sandeen <sandeen@redhat.com> writes:
> >> Also, are you guys affected by the
> >> previously-acked-sectors-are-now-gone problems with 512-byte
> >> logical/4KB physical drives?
> 
> Eric> previously-acked-sectors-are-now-gone?  I guess I haven't been
> Eric> keeping up.  What do you mean by this?
> 
> We already discussed this on irc.  But in case anybody else are
> wondering...
> 
> On a disk with 4KB physical blocks emulating 512-byte logical blocks the
> drive firmware must resort to read-modify-write cycles and that opens up
> a new error scenario:
> 
> 4KB physical block: |                       0                       |
> 512b logical block: |  0  |  1  |  2  |  3  |  4  |  5  |  6  |  7  |
>                                                                  ^ ERROR
> 
> In this case we have successfully written LBA 0 - 6.  However, when the
> drive attempts to write LBA 7 it gets an I/O error and the previous 7
> logical blocks (that have previously been acknowledged as written) are
> lost.  IOW, the drive write atomicity is at the physical block level and
> not the logical ditto.

In some sense, this is no worse than "real" sectors 0-6 going bad right
after they were written, or in fact even having the writes silently fail.

Yes, there is more chance that writing sector 7 (due to 4k sector r-m-w)
will cause collateral damage, but the truth even today is that disks
are not going to fail a single 512-byte sector at one time, but more
likely 64kB (or whatever the remapping unit size is), so this isn't
really introducing a new failure mode.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.