Re: [PATCH 4/6] fs: Establish locking order for unrelated directories

[Jan Kara <jack@suse.cz>]

On Fri 26-05-23 11:45:15, Christian Brauner wrote:
> On Thu, May 25, 2023 at 12:16:10PM +0200, Jan Kara wrote:
> > Currently the locking order of inode locks for directories that are not
> > in ancestor relationship is not defined because all operations that
> > needed to lock two directories like this were serialized by
> > sb->s_vfs_rename_mutex. However some filesystems need to lock two
> > subdirectories for RENAME_EXCHANGE operations and for this we need the
> > locking order established even for two tree-unrelated directories.
> > Provide a helper function lock_two_inodes() that establishes lock
> > ordering for any two inodes and use it in lock_two_directories().
> > 
> > CC: stable@vger.kernel.org
> > Signed-off-by: Jan Kara <jack@suse.cz>
> > ---
> >  fs/inode.c    | 34 ++++++++++++++++++++++++++++++++++
> >  fs/internal.h |  2 ++
> >  fs/namei.c    |  4 ++--
> >  3 files changed, 38 insertions(+), 2 deletions(-)
> > 
> > diff --git a/fs/inode.c b/fs/inode.c
> > index 577799b7855f..2015fa50d34a 100644
> > --- a/fs/inode.c
> > +++ b/fs/inode.c
> > @@ -1103,6 +1103,40 @@ void discard_new_inode(struct inode *inode)
> >  }
> >  EXPORT_SYMBOL(discard_new_inode);
> >  
> > +/**
> > + * lock_two_inodes - lock two inodes (may be regular files but also dirs)
> > + *
> > + * Lock any non-NULL argument. The caller must make sure that if he is passing
> > + * in two directories, one is not ancestor of the other.  Zero, one or two
> > + * objects may be locked by this function.
> > + *
> > + * @inode1: first inode to lock
> > + * @inode2: second inode to lock
> > + * @subclass1: inode lock subclass for the first lock obtained
> > + * @subclass2: inode lock subclass for the second lock obtained
> > + */
> > +void lock_two_inodes(struct inode *inode1, struct inode *inode2,
> > +		     unsigned subclass1, unsigned subclass2)
> > +{
> > +	if (!inode1 || !inode2)
> > +		goto lock;
> 
> Before this change in
> 
> lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
> 
> the swap() would cause the non-NULL inode to always be locked with
> I_MUTEX_NONDIR2. Now it can be either I_MUTEX_NORMAL or I_MUTEX_NONDIR2.
> Is that change intentional?

Kind of. I don't think we really care so I didn't bother to complicate the
code for this. If you think keeping the lockdep class consistent is worth
it, I can modify the patch...

								Honza

-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR