Re: [PATCH 2/9] vfs: export do_splice_direct() to modules

[Al Viro <viro@ZenIV.linux.org.uk>]

On Tue, Mar 19, 2013 at 09:38:31PM +0000, Al Viro wrote:
> On Tue, Mar 19, 2013 at 09:25:43PM +0100, Jan Kara wrote:
> > > BTW, having sb_start_write() buried in individual ->splice_write() is
> > > asking for trouble; could you describe the rules for that?  E.g. where
> > > does it nest wrt filesystem-private locks?  XFS iolock, for example...
> >   Generally, the freeze protection should be the outermost lock taken (so
> > that we mitigate possibility of blocking readers when waiting for fs to
> > unfreeze). So it ranks above i_mutex, or XFS' ilock and iolock.
> 
> Welcome to deadlock, then:
> xfs_file_splice_write()
> ...
>         xfs_ilock(ip, XFS_IOLOCK_EXCL);
> ...
>         ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
> 
> > It seems that I screwed this up for ->splice_write() :-| If we are going to
> > move out sb_start_write() out of filesystems' hands into do_splice_from()
> > then we should likely do the same with ->aio_write(). Hmm?

... and then there's ext4_file_dio_write(), which cheerfully ignores freeze.
... and udf expanding from inline files to separately allocated before it
gets to sb_start_write()
... and a bunch of guys doing generic_write_sync() after generic_aio_file_write()
... and a lot of other fun stuff.  Ouch...

OK, it's going to be an interesting series - aforementioned tentative patch
was badly incomplete ;-/