Re: [PATCH v4 07/17] fs/dcache.c: add shrink_dcache_inode()

[Eric Biggers <ebiggers@kernel.org>]

On Wed, Apr 03, 2019 at 07:34:12PM +0100, Al Viro wrote:
> On Tue, Apr 02, 2019 at 08:45:50AM -0700, Eric Biggers wrote:
> > From: Eric Biggers <ebiggers@google.com>
> > 
> > When a filesystem encryption key is removed, we need all files which had
> > been "unlocked" (had ->i_crypt_info set up) with it to appear "locked"
> > again.  This is most easily done by evicting the inodes.  This can
> > currently be done using 'echo 2 > /proc/sys/vm/drop_caches'; however,
> > that is overkill and not usable by non-root users.
> > 
> > To evict just the needed inodes we also need the ability to evict those
> > inodes' dentries, since an inode is pinned by its dentries.  Therefore,
> > add a function shrink_dcache_inode() which iterates through an inode's
> > dentries and evicts any unused ones as well as any unused descendants
> > (since there may be negative dentries pinning the inode's dentries).
> 
> Huh?
> 
> > + * Evict all unused aliases of the specified inode from the dcache.  This is
> > + * intended to be used when trying to evict a specific inode, since inodes are
> > + * pinned by their dentries.  We also have to descend to ->d_subdirs for each
> > + * alias, since aliases may be pinned by negative child dentries.
> > + */
> > +void shrink_dcache_inode(struct inode *inode)
> > +{
> > +	for (;;) {
> > +		struct select_data data;
> > +		struct dentry *dentry;
> > +
> > +		INIT_LIST_HEAD(&data.dispose);
> > +		data.start = NULL;
> > +		data.found = 0;
> > +
> > +		spin_lock(&inode->i_lock);
> > +		hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias)
> > +			d_walk(dentry, &data, select_collect);
> > +		spin_unlock(&inode->i_lock);
> > +
> > +		if (!data.found)
> > +			break;
> > +
> > +		shrink_dentry_list(&data.dispose);
> > +		cond_resched();
> 
> This is... odd.  What's wrong with
> 	if (S_ISDIR(inode->i_mode)) {
> 		dentry = d_find_any_alias(inode);
> 		if (dentry) {
> 			shrink_dcache_parent(dentry);
> 			dput(dentry);
> 		}
> 	}
> 	d_prune_aliases(inode);
> instead of that thing?

That works, as far as I can tell, so I'll do that instead.

I don't think I noticed that d_prune_aliases() existed when I wrote this.

Thanks for the suggestion!

- Eric