[PATCH] arm64: fix address fault during mapping fdt region

[Mark Rutland]

On Mon, Aug 01, 2016 at 11:50:39AM +0200, Ard Biesheuvel wrote:
> On 1 August 2016 at 11:42, zijun_hu <zijun_hu at zoho.com> wrote:
> > From 07b9216ec3494515e7a6c41e0333eb8782427db3 Mon Sep 17 00:00:00 2001
> > From: zijun_hu <zijun_hu at htc.com>
> > Date: Mon, 1 Aug 2016 17:04:59 +0800
> > Subject: [PATCH] arm64: fix address fault during mapping fdt region
> >
> > fdt_check_header() accesses other fileds of fdt header but
> > the first 8 bytes such as version; so accessing unmapped
> > address fault happens if fdt region locates below align
> > boundary nearly during mapping fdt region, or expressed as
> > (offset + sizeof(struct fdt_header)) > SWAPPER_BLOCK_SIZE
> >
> > fdt header size at least is mapped in order to avoid the issue
> >
> > Signed-off-by: zijun_hu <zijun_hu at htc.com>
> > ---
> >  arch/arm64/mm/mmu.c | 14 ++++++++++++--
> >  1 file changed, 12 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> > index 0f85a46..0d72b71 100644
> > --- a/arch/arm64/mm/mmu.c
> > +++ b/arch/arm64/mm/mmu.c
> > @@ -744,6 +744,7 @@ void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
> >         const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
> >         int offset;
> >         void *dt_virt;
> > +       int dt_header_map_size;
> >
> >         /*
> >          * Check whether the physical FDT address is set and meets the minimum
> > @@ -774,9 +775,18 @@ void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
> >         offset = dt_phys % SWAPPER_BLOCK_SIZE;
> >         dt_virt = (void *)dt_virt_base + offset;
> >
> > +       /*
> > +        * fdt_check_header() maybe access any field of fdt header not
> > +        * the first 8 bytes only, so map fdt header size at least for
> > +        * checking fdt header without address fault more portably
> > +        */
> > +       BUILD_BUG_ON(sizeof(struct fdt_header) > SWAPPER_BLOCK_SIZE);
> > +       dt_header_map_size = round_up(offset + sizeof(struct fdt_header),
> > +                       SWAPPER_BLOCK_SIZE);
> > +
> >         /* map the first chunk so we can read the size from the header */
> >         create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
> > -                       dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
> > +                       dt_virt_base, dt_header_map_size, prot);
> >
> >         if (fdt_check_header(dt_virt) != 0)
> >                 return NULL;
> > @@ -785,7 +795,7 @@ void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
> >         if (*size > MAX_FDT_SIZE)
> >                 return NULL;
> >
> > -       if (offset + *size > SWAPPER_BLOCK_SIZE)
> > +       if (offset + *size > dt_header_map_size)
> >                 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
> >                                round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
> >
> 
> Couldn't we simply do this instead?
> 
> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index 0f85a46c3e18..e8d3b04a2b57 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
> @@ -778,7 +778,7 @@ void *__init __fixmap_remap_fdt(phys_addr_t
> dt_phys, int *size, pgprot_t prot)
>         create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
>                         dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
> 
> -       if (fdt_check_header(dt_virt) != 0)
> +       if (fdt_magic(dt_virt) != FDT_MAGIC)
>                 return NULL;
> 
>         *size = fdt_totalsize(dt_virt);
> 
> We are simply looking for a size field. The OF code will call
> fdt_check_header() again, so anything that is checked there in
> addition to the magic field will still be checked.

That looks much nicer to me, and deferring the header check also looks
fine.

If you add a comment above the fdt_magic() check, pointing out why we
can only safely access the magic and totalsize fields (and thus can't
call fdt_check_header() yet):

Acked-by: Mark Rutland <mark.rutland at arm.com>

Thanks,
Mark.