On Tue, Dec 14, 2021, Tadeusz Struk wrote:
> Syzbot reported an use-after-free bug in update_accessed_dirty_bits().
> Fix this by checking if the memremap'ed pointer is still valid.
...
> Fixes: bd53cb35a3e9 ("X86/KVM: Handle PFNs outside of kernel reach when touching GPTEs")
> Link: https://syzkaller.appspot.com/bug?id=6cb6102a0a7b0c52060753dd62d070a1d1e71347
> Reported-by: syzbot+6cde2282daa792c49ab8@xxxxxxxxxxxxxxxxxxxxxxxxx
> Signed-off-by: Tadeusz Struk <tadeusz.struk@xxxxxxxxxx>
> ---
> arch/x86/kvm/mmu/paging_tmpl.h | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/mmu/paging_tmpl.h b/arch/x86/kvm/mmu/paging_tmpl.h
> index 708a5d297fe1..5cf4815d1c45 100644
> --- a/arch/x86/kvm/mmu/paging_tmpl.h
> +++ b/arch/x86/kvm/mmu/paging_tmpl.h
> @@ -174,7 +174,7 @@ static int FNAME(cmpxchg_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
> pfn = ((vaddr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Isn't this code flat out wrong? vm_pgoff is usually the offset relative to the
file and has nothing to do with the pfn. I see that remap_pfn_range_notrack()
stuffs "vma->vm_pgoff = pfn", but that's a weird quirk of that particular usage
of VM_PFNMAP that I'm guessing happened to align with the original usage of this
mess. But unless there's magic I'm missing, vm_pgoff is not guaranteed to have
any relation to the pfn for any ol' VM_PFNMAP vma.
In other words, I suspect pfn and paddr are complete garbage, and adding the
access_ok() check masks that.
> paddr = pfn << PAGE_SHIFT;
> table = memremap(paddr, PAGE_SIZE, MEMREMAP_WB);
> - if (!table) {
> + if (!table || !access_ok(table, PAGE_SIZE)) {
> mmap_read_unlock(current->mm);
> return -EFAULT;
> }
> --
> 2.33.1
>
