Re: [PATCH 10/11] VFS: take a shared lock for create/remove directory operations.

[NeilBrown]

On Mon, 23 Dec 2024, Al Viro wrote:
> On Fri, Dec 20, 2024 at 01:54:28PM +1100, NeilBrown wrote:
> 
> > Once the exclusive "update" lock is obtained on the dentry we must make
> > sure it wasn't unlinked or renamed while we slept.  If it was we repeat
> > the lookup.
> 
> > +		if (
> > +			/* Was unlinked while we waited ?*/
> > +			d_unhashed(dentry) ||
> > +			/* Or was dentry renamed ?? */
> > +			dentry->d_parent != base ||
> > +			dentry->d_name.hash != last->hash ||
> > +			!d_same_name(dentry, base, last)
> > +		) {
> > +			rcu_read_unlock();
> > +			spin_unlock(&dentry->d_lock);
> > +			lock_map_release(&dentry->d_update_map);
> > +			dput(dentry);
> > +			goto retry;
> > +		}
> > +		rcu_read_unlock();
> > +	}
> > +	dentry->d_flags |= DCACHE_PAR_UPDATE;
> > +	spin_unlock(&dentry->d_lock);
> 
> ... and now __d_unalias() moves it to new place, making all the checks
> you've just done completely useless.
> 

... Yes, thanks.

So I need __d_unalias() to effectively do a "try_lock" of
DCACHE_PAR_UPDATE and hold the lock across __d_move().
I think that would address the concerned you raised.

Did you see anything else that might be problematic?

Thanks,
NeilBrown