Re: [patch 3/7] vfs: mountinfo: add mount ID

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, Mar 26, 2008 at 10:11:34PM +0100, Miklos Szeredi wrote:
> +static int mnt_alloc_id(struct vfsmount *mnt)
> +{
> +	int res;
> +
> + retry:
> +	spin_lock(&vfsmount_lock);
> +	res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
> +	spin_unlock(&vfsmount_lock);
> +	if (res == -EAGAIN) {
> +		if (ida_pre_get(&mnt_id_ida, GFP_KERNEL))
> +			goto retry;
> +		res = -ENOMEM;
> +	}
> +	return res;

*Ugh*

Why bother with vfsmount_lock here?  All allocations are done under
namespace_sem.  Moreover, I'd rather replace that 'goto retry' with
a single call of ida_get_new(), since we are serialized anyway.

> @@ -353,6 +386,7 @@ EXPORT_SYMBOL(simple_set_mnt);
>  void free_vfsmnt(struct vfsmount *mnt)
>  {
>  	kfree(mnt->mnt_devname);
> +	mnt_free_id(mnt);
>  	kmem_cache_free(mnt_cache, mnt);
>  }

... and I'd rather do that earlier, e.g. in umount_tree().  At that point
we (a) have namespace_sem and (b) irrevocably kick the sucker out of
any namespace.

I'd rather minimize banging vfsmount_lock like that...