Re: [PATCH v5] mtd: Do not corrupt backing device of device node inode

[David Woodhouse <dwmw2@infradead.org>]

On Mon, 2010-05-17 at 16:55 +0300, Kirill A. Shutemov wrote:
> +       if (!mtd->inode) {
> +               struct inode *mtd_inode;
> +               mtd_inode = new_inode(mtd_inode_mnt->mnt_sb);
> +               if (!mtd_inode) {
> +                       put_mtd_device(mtd);
> +                       ret = -ENOMEM;
> +                       goto out;
> +               }
> +               mtd_inode->i_mode = S_IFCHR;
> +               mtd_inode->i_rdev = inode->i_rdev;
> +               if (mtd->backing_dev_info) {
> +                       mtd_inode->i_data.backing_dev_info =
> +                               mtd->backing_dev_info;
> +               }
> +
> +               if (unlikely(cmpxchg(&mtd->inode, NULL, mtd_inode))) {
> +                       /* Somebody has already initialized mtd->inode
> */
> +                       iput(mtd_inode);
> +               }
> +       }
> +
> +       igrab(mtd->inode); 

Now you're just reimplementing iget().

Can we just use iget_locked() to get the appropriate inode (using the
mtd device number as the inode number)? Then we don't need to bother
storing it in mtd->inode at all, and we can iput() the last refcount on
it when the chardevice is closed -- we don't need it to stick around
until the MTD device is destroyed.

-- 
dwmw2