hi,

> Its problem is that ffs with UBC does read-modify-write
> even when full block write.

yes.

> I wrote a quick hack patch and measured its through put with dd.
> My patch also eliminates unnessesary ubc page faults.

but your patch seems wrong, at least for write(2), if i read it correctly.

because ufs_balloc_range_with_pages keeps pages busy,
it can deadlock if some of them are mapped to the userspace and
a page fault occurs on one of them in uiomove.

i guess that Chuck Silvers disabled overwrite-optimization
for this deadlock problem at the first place.

YAMAMOTO Takashi