Re: [patch] updated inotify for 2.6.12-rc3.

[Al Viro <viro@parcelfarce.linux.theplanet.co.uk>]

> +static int inotify_ignore(struct inotify_device *dev, s32 wd)
> +{
> +	struct inotify_watch *watch;
> +	struct inode *inode;
> +
> +	down(&dev->sem);
> +	watch = idr_find(&dev->idr, wd);
> +	if (unlikely(!watch)) {
> +		up(&dev->sem);
> +		return -EINVAL;
> +	}
> +	get_inotify_watch(watch);
> +	up(&dev->sem);
> +
> +	inode = watch->inode;
> +	down(&inode->inotify_sem);
> +	down(&dev->sem);
> +	remove_watch(watch, dev);
> +	up(&dev->sem);
> +	up(&inode->inotify_sem);
> +	put_inotify_watch(watch);
> +
> +	return 0;
> +}

So what happens if
	* something is holding inotify_sem right now
	* ten threads call that on the same watch
	* all of them get to down(&inode->inotify_sem); and block there,
having acquired ten references to the watch
	* after whatever had been holding ->inotify_sem in the first place
releases it, they will one by one go through the rest of function.  And
drop _20_ references to the watch.  9 of those - after we kfree() the
watch...