Bruce Martin wrote:
> 
> Okay, one more thing I've discovered: when the crash occurs under fsck, it always occurs with exactly the same disk address. For instance, while it is cleaning /dev/wd0a I get
>   wd0e: Error reading fsbn 48 of 48-51 (wd0 bn 204848 cn 216 tn 11 sn 35)
> 
> Interesting that it is an error reading wd0e when it is trying to fsck wd0a.

I think these are red-herrings, although since they're reproducible they
may be useful for debugging purposes. fsck(8) uses the raw device, as you
were with dd(1), so it will crash for the same reason.

Since the filesystems are usable when mounted (via the block device) this
all seems to point to a software problem which is being tickled by the
raw disk device.

Cheers, Steve