> Looking at old code I think I have an idea what needs to be done.
> The quick hack for vnd remains.
> The adjustment for end of patition is done to b_bcount instead of
> b_resid.
> The block count to byte comparison used to be correct, so back out
> to the original or similar code.
> This would then work for all other drivers as well.

can you explain "The block count to byte comparison"?

except that "bp->b_blkno + p->p_offset" seems bogus,
most of variables there seem to be in DEV_BSIZE.

YAMAMOTO Takashi