On Tue, Dec 14, 2021 at 6:11 PM Catalin Marinas <catalin.marinas@xxxxxxx> wrote:
>
> On Mon, Dec 13, 2021 at 10:54:21PM +0100, andrey.konovalov@xxxxxxxxx wrote:
> > diff --git a/arch/arm64/include/asm/vmalloc.h b/arch/arm64/include/asm/vmalloc.h
> > index b9185503feae..3d35adf365bf 100644
> > --- a/arch/arm64/include/asm/vmalloc.h
> > +++ b/arch/arm64/include/asm/vmalloc.h
> > @@ -25,4 +25,14 @@ static inline bool arch_vmap_pmd_supported(pgprot_t prot)
> >
> > #endif
> >
> > +#define arch_vmalloc_pgprot_modify arch_vmalloc_pgprot_modify
> > +static inline pgprot_t arch_vmalloc_pgprot_modify(pgprot_t prot)
> > +{
> > + if (IS_ENABLED(CONFIG_KASAN_HW_TAGS) &&
> > + (pgprot_val(prot) == pgprot_val(PAGE_KERNEL)))
> > + prot = pgprot_tagged(prot);
> > +
> > + return prot;
> > +}
> > +
> > #endif /* _ASM_ARM64_VMALLOC_H */
> > diff --git a/include/linux/vmalloc.h b/include/linux/vmalloc.h
> > index 28becb10d013..760caeedd749 100644
> > --- a/include/linux/vmalloc.h
> > +++ b/include/linux/vmalloc.h
> > @@ -115,6 +115,13 @@ static inline int arch_vmap_pte_supported_shift(unsigned long size)
> > }
> > #endif
> >
> > +#ifndef arch_vmalloc_pgprot_modify
> > +static inline pgprot_t arch_vmalloc_pgprot_modify(pgprot_t prot)
> > +{
> > + return prot;
> > +}
> > +#endif
> > +
> > /*
> > * Highlevel APIs for driver use
> > */
> > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > index 837ed355bfc6..58bd2f7f86d7 100644
> > --- a/mm/vmalloc.c
> > +++ b/mm/vmalloc.c
> > @@ -3060,6 +3060,8 @@ void *__vmalloc_node_range(unsigned long size, unsigned long align,
> > return NULL;
> > }
> >
> > + prot = arch_vmalloc_pgprot_modify(prot);
> > +
> > if (vmap_allow_huge && !(vm_flags & VM_NO_HUGE_VMAP)) {
> > unsigned long size_per_node;
>
> I wonder whether we could fix the prot bits in the caller instead and we
> won't need to worry about the exec or the module_alloc() case. Something
> like:
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index d2a00ad4e1dd..4e8c61255b92 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -3112,7 +3112,7 @@ void *__vmalloc_node(unsigned long size, unsigned long align,
> gfp_t gfp_mask, int node, const void *caller)
> {
> return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
> - gfp_mask, PAGE_KERNEL, 0, node, caller);
> + gfp_mask, pgprot_hwasan(PAGE_KERNEL), 0, node, caller);
> }
> /*
> * This is only for performance analysis of vmalloc and stress purpose.
> @@ -3161,7 +3161,7 @@ EXPORT_SYMBOL(vmalloc);
> void *vmalloc_no_huge(unsigned long size)
> {
> return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
> - GFP_KERNEL, PAGE_KERNEL, VM_NO_HUGE_VMAP,
> + GFP_KERNEL, pgprot_hwasan(PAGE_KERNEL), VM_NO_HUGE_VMAP,
> NUMA_NO_NODE, __builtin_return_address(0));
> }
> EXPORT_SYMBOL(vmalloc_no_huge);
>
> with pgprot_hwasan() defined to pgprot_tagged() only if KASAN_HW_TAGS is
> enabled.
And also change kasan_unpoison_vmalloc() to tag only if
pgprot_tagged() has been applied, I assume.
Hm. Then __vmalloc_node_range() callers will never get tagged memory
unless requested. I suppose that's OK, most of them untag the pointer
anyway.
But this won't work for SW_TAGS mode, which is also affected by the
exec issue and needs those kasan_reset_tag()s in module_alloc()/BPF.
We could invent some virtual protection bit for it and reuse
pgprot_hwasan(). Not sure if this would be acceptable.
