On Sun, Aug 30, 2009 at 05:35:51PM +0100, Jamie Lokier wrote:

I like that.  It looks really clear and self-documenting, if
vfs_fsync_range does what it sounds like, which is a nice change.

If I've guessed right what that code does, proper O_RSYNC will be easy:

int generic_sync_before_read(struct file *file, loff_t pos, loff_t count)
{
	int is_sync = ((file->f_flags & O_SYNC)
		       || IS_SYNC(file->f_mapping->host));
	int is_dsync = is_sync || (file->f_flags & O_DSYNC);

	if (!is_dsync || !(file->f_flags & O_RSYNC))
		return 0;
	return vfs_fsync_range(file, file->f_ath.denty, pos,
			       pos + count - 1, is_sync);
}

Yes. something like this.

(By the way, did I mention Irix has range-fsync and range-fdatasync
system calls too :-) (actually fcntls))

Linux has sync_file_range which currently is a perfect way to lose your
synced' data, but with two more flags and calls to ->fsync we could
turn it into range-fsync/fdatasync.  I'm not sure if that's a good
idea or if we should just add a sys_fdatasync_rage systems call.

I don't quite see the point of a range-fsync, but it could be easily
implemented as a flag.