On Tue, Jan 04, 2011 at 07:03:23PM +1100, Nick Piggin wrote: > On Tue, Jan 04, 2011 at 01:57:37AM -0500, Christoph Hellwig wrote: > > > Also giving filesystems the flexibility to do the data writeout can > > > be more optimal by doing all writeout at once or ordering to suit their > > > writeback patterns. Having range data could give some performance > > > advantages writing back mappings or commit operations over network. I > > > don't see it as a big complexity. It gives a chance to do range fsyncs > > > and sync_file_range and such properly too. > > > > We currently do all that just fine before calling into ->fsync. > > What do you mean we do all that just fine? A filesystem can't schedule > data submission and waiting optimally versus metadata, it can't do > metadata operations or remote commits corresponding to data ranges, and > it doesn't support nfs sync operations. And actually I think it is much better to have sync_inode, which means we'll be able to get rid of commit_metadata (which should be an inode operation anyway, not an export operation which really should deal with exporting filesystems to a non-vfs namespace, not nfsd hacks). commit_metadata would just be sync_inode with a null range or no data sync flag set. -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
