On Tue, Nov 10, 2020 at 11:20:28PM +0000, Al Viro wrote: > On Tue, Nov 10, 2020 at 09:35:11PM +0000, Al Viro wrote: > > On Tue, Nov 10, 2020 at 09:32:53PM +0000, Al Viro wrote: > > > > > AFAICS, not all callers want that semantics, but I think it's worth > > > a new primitive. I'm not saying it should be a prereq for your > > > series, but either that or an explicit iov_iter_revert() is needed. > > > > Seeing that it already went into mainline, it needs a followup fix. > > And since it's not -stable fodder (AFAICS), I'd rather go with > > adding a new primitive... > > Any objections to the following? > > Fix seq_read_iter() behaviour on full pipe > > generic_file_splice_read() will purge what we'd left in pipe in case > of error; it will *not* do so in case of short write, so we must make > sure that reported amount of data stored by ->read_iter() matches the > reality. > > It's not a rare situation (and we already have it open-coded in at least > one place), so let's introduce a new primitive - copy_to_iter_full(). > Similar to copy_from_iter_full(), it returns true if we had been able > to copy everything we'd been asked to and false otherwise. Iterator > is advanced only on success. > > Signed-off-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx> Looks ok to me.
