On Tue, Nov 19, 2019 at 08:34:54AM -0800, Darrick J. Wong wrote: > > The slight concern I have with this is that that would change e.g. the > > behavior of IOMAP_REPORT. We could specialcase IOMAP_REPORT but then it > > gets kind of ugly. And it seems kind of fuzzy when do we truncate the > > extent with i_size and when not... Generally i_size is kind of a side-band > > thing for block mapping operations so if we could leave it out of > > ->iomap_begin I'd find that nicer. > > <nod> Yes. I'd prefer if the caller deals with any i_size limiting and not the iomap methods themselves. For now I'm tempted to just go with the iov_iter_revert scheme. Note that I particularly like it, but it matches the most common direct I/O implementation at least.
