Re: software RAID (fwd)

[Adam Miller]

Here is a response from Bruce Allen.

> > If you have a bad sector, it doesn't go away by writing to it again.  On
> > modern drives, if you see bad sectors the disk is just about dead, and
> > will probably be seen as such by the raid system which will then stop
> > using the disk entirely and expect you to replace it ASAP.

This is false.

Modern ATA and SCSI disk drives have several thousand spare sectors.
When a sector is unreadable (UNC) which means that the ECC codes are
inconsistent, the drive will REALLOCATE the sector, assigning a spare
sector the LBA of the failed sector.  However it will only do this when
you WRITE to the LBA of the failed sector.

> The one exception here is if you have a miswritten sector (usually
> caused by unexpected power-down), which won't read back correctly -
> but running badblocks with one of the 'write-verify' options will
> resurrect it.

Sectors can have inconsistent ECC codes for a number of reasons:
  -- failed write during sudden power-down
  -- damage to magnetic media at this LBA
  -- other reasons

> If you have a drive that has a bad block in it even *after* badblocks has
> re-written it, it's time to replace the drive *now*....

Not true.  Disks which have reallocated large numbers of blocks are
usually failing. But most good disks have some reallocated blocks.

> For the original poster: Breaking the mirror and then re-mirroring
> from the "good" drive *might* recover the bad block when it re-writes
> it.  But don't bet on it...

It won't 'recover' the bad block.  It will write the data (obtained from
the good drive) to a newly allocated spare sector on the bad drive.

Cheers,
	Bruce
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/