Re: [RFC][PATCHSET] sorting out RCU-delayed stuff in ->destroy_inode()

[Andreas Dilger <adilger@dilger.ca>]

[-- Attachment #1: Type: text/plain, Size: 2519 bytes --]


> On Apr 29, 2019, at 10:26 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> 
> On Mon, Apr 29, 2019 at 10:18:04PM -0600, Andreas Dilger wrote:
>>> 
>>> 	void			*i_private; /* fs or device private pointer */
>>> +	void (*free_inode)(struct inode *);
>> 
>> It seems like a waste to increase the size of every struct inode just to access
>> a static pointer.  Is this the only place that ->free_inode() is called?  Why
>> not move the ->free_inode() pointer into inode->i_fop->free_inode() so that it
>> is still directly accessible at this point.
> 
> i_op, surely?

Yes, i_op is what I was thinking.

> In any case, increasing sizeof(struct inode) is not a problem -

> if anything, I'd turn ->i_fop into an anon union with that.  As in,
> 
> diff --git a/fs/inode.c b/fs/inode.c
> index fb45590d284e..627e1766503a 100644
> --- a/fs/inode.c
> +++ b/fs/inode.c
> @@ -211,8 +211,8 @@ EXPORT_SYMBOL(free_inode_nonrcu);
> static void i_callback(struct rcu_head *head)
> {
> 	struct inode *inode = container_of(head, struct inode, i_rcu);
> -	if (inode->i_sb->s_op->free_inode)
> -		inode->i_sb->s_op->free_inode(inode);
> +	if (inode->free_inode)
> +		inode->free_inode(inode);
> 	else
> 		free_inode_nonrcu(inode);
> }
> @@ -236,6 +236,7 @@ static struct inode *alloc_inode(struct super_block *sb)
> 			if (!ops->free_inode)
> 				return NULL;
> 		}
> +		inode->free_inode = ops->free_inode;
> 		i_callback(&inode->i_rcu);
> 		return NULL;
> 	}

> @@ -276,6 +277,7 @@ static void destroy_inode(struct inode *inode)
> 		if (!ops->free_inode)
> 			return;
> 	}
> +	inode->free_inode = ops->free_inode;
> 	call_rcu(&inode->i_rcu, i_callback);
> }

This seems like kind of a hack.  I guess your goal is to have ->free_inode
accessible regardless of whether the filesystem has installed its own ->i_op
methods or not, and i_fop is no longer used by this point.

That said, this seems better than increasing the size of struct inode.

> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 2e9b9f87caca..92732286b748 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -694,7 +694,10 @@ struct inode {
> #ifdef CONFIG_IMA
> 	atomic_t		i_readcount; /* struct files open RO */
> #endif
> -	const struct file_operations	*i_fop;	/* former ->i_op->default_file_ops */
> +	union {
> +		const struct file_operations	*i_fop;	/* former ->i_op->default_file_ops */
> +		void (*free_inode)(struct inode *);
> +	};


Cheers, Andreas






[-- Attachment #2: Message signed with OpenPGP --]
[-- Type: application/pgp-signature, Size: 873 bytes --]