Re: [PATCH 1/5] brd: fix oops if write concurrent with discard

[Christoph Hellwig <hch@infradead.org>]

On Fri, Apr 18, 2025 at 05:38:22PM +0800, Yu Kuai wrote:
>  	copy = min_t(size_t, n, PAGE_SIZE - offset);
>  	page = brd_lookup_page(brd, sector);
> -	BUG_ON(!page);
> -
> -	dst = kmap_atomic(page);
> -	memcpy(dst + offset, src, copy);
> -	kunmap_atomic(dst);
> +	if (page) {
> +		dst = kmap_local_page(page);
> +		memcpy(dst + offset, src, copy);
> +		kunmap_local(dst);
> +	}

I don't see how this can fix any race, it just narrows down the
race window.  To fix the race for real, copy_to_brd_setup needs
to return a page and keep a reference to it for the caller.  The
caller then only needs to operate on a single page.

We'll also need to do something similar for brd_lookup_page to
ensure the page reference doesn't go away after the xarray lookup
but before using the page.

> Also fix following checkpatch warnings:
> WARNING: Deprecated use of 'kmap_atomic', prefer 'kmap_local_page' instead
> WARNING: Deprecated use of 'kunmap_atomic', prefer 'kunmap_local' instead

This really should be using bvec_kmap_local.  I actually have an
entire series to fix that and clean up some of the surroundings that I
need to send out.  Let me dust that off because it might help with the
above mentioned fixes as well.