On Tue, Jul 07, 2020 at 04:52:37PM +0100, Matthew Wilcox wrote:

But userspace has to _do_ something with that information anyway.  So
it must already have somewhere to put that information.

I do think that interpretation of that field should be a separate flag
from WRITE_APPEND so apps which genuinely don't care about where the I/O
ended up don't have to allocate some temporary storage.  eg a logging
application which just needs to know that it managed to append to the
end of the log and doesn't need to do anything if it's successful.

I agree with the concept of a flag for just returning the write
location.  Because if you don't need that O_APPEND is all you need
anyway.