Re: [PATCH] crypto: testmgr - allocate buffers with __GFP_COMP

From: Matthew Wilcox
Date: Mon Apr 15 2019 - 22:19:05 EST


On Mon, Apr 15, 2019 at 10:46:15AM +0800, Herbert Xu wrote:
> On Sun, Apr 14, 2019 at 07:24:12PM -0700, Matthew Wilcox wrote:
> > On Thu, Apr 11, 2019 at 01:32:32PM -0700, Kees Cook wrote:
> > > > @@ -156,7 +156,8 @@ static int __testmgr_alloc_buf(char *buf[XBUFSIZE], int order)
> > > > int i;
> > > >
> > > > for (i = 0; i < XBUFSIZE; i++) {
> > > > - buf[i] = (char *)__get_free_pages(GFP_KERNEL, order);
> > > > + buf[i] = (char *)__get_free_pages(GFP_KERNEL | __GFP_COMP,
> > > > + order);
> > >
> > > Is there a reason __GFP_COMP isn't automatically included in all page
> > > allocations? (Or rather, it seems like the exception is when things
> > > should NOT be considered part of the same allocation, so something
> > > like __GFP_SINGLE should exist?.)
> >
> > The question is not whether or not things should be considered part of the
> > same allocation. The question is whether the allocation is of a compound
> > page or of N consecutive pages. Now you're asking what the difference is,
> > and it's whether you need to be able to be able to call compound_head(),
> > compound_order(), PageTail() or use a compound_dtor. If you don't, then
> > you can save some time at allocation & free by not specifying __GFP_COMP.
>
> Thanks for clarifying Matthew.
>
> Eric, this means that we should not use __GFP_COMP here just to
> silent what is clearly a broken warning.

I agree; if the crypto code is never going to try to go from the address of
a byte in the allocation back to the head page, then there's no need to
specify GFP_COMP.

But that leaves us in the awkward situation where
HARDENED_USERCOPY_PAGESPAN does need to be able to figure out whether
'ptr + n - 1' lies within the same allocation as ptr. Without using
a compound page, there's no indication in the VM structures that these
two pages were allocated as part of the same allocation.

We could force all multi-page allocations to be compound pages if
HARDENED_USERCOPY_PAGESPAN is enabled, but I worry that could break
something. We could make it catch fewer problems by succeeding if the
page is not compound. I don't know, these all seem like bad choices
to me.