On Wed, Nov 24, 2021 at 03:00:00PM -0800, Linus Torvalds wrote:

On Wed, Nov 24, 2021 at 12:04 PM Matthew Wilcox [off-list ref] wrote:

quoted

(where __copy_to_user_nofault() is a new function that does exactly what
copy_to_user_nofault() does, but returns the number of bytes copied)

If we want the "how many bytes" part, then we should just make
copy_to_user_nofault() have the same semantics as a plain
copy_to_user().

IOW, change it to return "number of bytes not copied".

Looking at the current uses, such a change would be trivial. The only
case that wants a 0/-EFAULT error is the bpf_probe_write_user(),
everybody else already just wants "zero for success", so changing
copy_to_user_nofault() would be trivial.

I agree, if we want the number of byte not copied, we should just change
copy_{to,from}_user_nofault() and their callers (I can count three
each).

For this specific btrfs case, if we want go with tuning the offset based
on the fault address, we'd need copy_to_user_nofault() (or a new
function) to be exact. IOW, fall back to byte-at-a-time copy until it
hits the real faulting address.

-- 
Catalin