Re: [PATCH] Handle vma regions with no backing page

[Andrea Arcangeli <andrea@qumranet.com>]

On Tue, Apr 29, 2008 at 09:32:09AM -0500, Anthony Liguori wrote:
> +		vma = find_vma(current->mm, addr);
> +		if (vma == NULL) {
> +			get_page(bad_page);
> +			return page_to_pfn(bad_page);
> +		}

Here you must check vm_start address, find_vma only checks addr <
vm_end but there's no guarantee addr >= vm_start yet.

> +
> +		BUG_ON(!(vma->vm_flags & VM_IO));

For consistency we should return bad_page and not bug on, VM_IO and
VM_PFNMAP can theoretically not be set at the same time, otherwise
get_user_pages would be buggy checking against VM_PFNMAP|VM_IO. I
doubt anybody isn't setting VM_IO before calling remap_pfn_range but
anyway...

Secondly the really correct check is against VM_PFNMAP. This is
because PFNMAP is set at the same time of vm_pgoff = pfn. VM_IO is not
even if in theory if a driver uses ->fault instead of remap_pfn_range,
shouldn't set VM_IO and it should only set VM_RESERVED. VM_IO is about
keeping gdb/coredump out as they could mess with the hardware if they
read, PFNMAP is about remap_pfn_range having been called and pgoff
pointing to the first pfn mapped at vm_start address.

Patch is in the right direction, way to go!

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone