Re: [Cluster-devel] [RFC PATCH 3/5] gfs2: Add a dynamic buffer backed by a vector of pages

[Bob Peterson <rpeterso@redhat.com>]

----- Original Message -----
> +static int vp_alloc_pages(struct vp_ctx *vpx, int start, int end)
> +{
> +	int i;
> +
> +	for (i = start; i < end; i++) {
> +		vpx->vp_pages[i] = alloc_page(GFP_KERNEL | GFP_NOFS);
> +		if (vpx->vp_pages[i] == NULL)
> +			goto free;
> +	}
> +	return 0;
> +free:
> +	for (i = start; i < end; i++)
> +		if (vpx->vp_pages[i]) {
> +			__free_page(vpx->vp_pages[i]);
> +			vpx->vp_pages[i] = NULL;
> +		}
> +	return -ENOMEM;
> +}

I'm concerned that if we have a value for vpx->vp_pages[i] that is
nonzero (uninitialized) _after_ an element in which we couldn't
allocate a page, we could try to free the nonzero value. For that
reason, I always prefer to do something like this instead:

free:
     while (i > start) {
          i--;
          __free_page(vpx->vp_pages[i]);
          vpx->vp_pages[i] = NULL;
     }
     return -ENOMEM;

(snip)
> +struct vp_ctx* vp_get_vpx(struct vbuf *vb)
> +{
> +	struct vp_ctx *vpx = NULL;
> +
> +	if (!vb || !vb->v_opaque)
> +		goto out;
> +
> +	vpx = vb->v_opaque;
> +	if (vpx->vp_magic != VP_MAGIC) {
> +		vpx = NULL;
> +		goto out;

This goto is unnecessary, and if you eliminate it, you don't need the brackets.

> +	}
> +out:
> +	return vpx;
> +}

Regards,

Bob Peterson
Red Hat File Systems