On Tue 27-06-23 21:51:05, Christoph Hellwig wrote: > Can you please resend this outside this thread? I really cant't see > what's new or old here if you have a reply-to in the old thread. > > On Tue, Jun 27, 2023 at 06:55:22PM +0200, Ahelenia Ziemiańska wrote: > > In 1/3 I've applied if/else if/else tree like you said, > > and expounded a bit in the message. > > > > This is less pretty now, however, since it turns out that > > iter_file_splice_write() already marks the out fd as written because it > > writes to it via vfs_iter_write(), and that sent a double notification. > > It seems like vfs_iter_write is the wrong level to implement > ->splice_write given that the the ->splice_write caller has already > checked f_mode, done the equivalent of rw_verify_area and > should do the fsnotify_modify. I'd suggest to just open code the > relevant parts of vfs_iocb_iter_write in iter_file_splice_write. Yeah, looking into the code I agree (with a small remark that unlike vfs_iocb_iter_write() this particular variant also needs to work with files providing only ->write and not ->write_iter). But we can live with duplicate events for now and this seems like a rather separate cleanup to do. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
