Re: [PATCH] arm64/mm: fold check for KFENCE into can_set_direct_map()

[Mike Rapoport]

On Thu, Sep 22, 2022 at 08:21:38AM +0530, Anshuman Khandual wrote:
> 
> On 9/21/22 20:49, Mike Rapoport wrote:
> > Hi Anshuman,
> > 
> > On Wed, Sep 21, 2022 at 05:09:19PM +0530, Anshuman Khandual wrote:
> >>
> >>
> >> On 9/21/22 13:18, Mike Rapoport wrote:
> >>> From: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> >>>
> >>> KFENCE requires linear map to be mapped at page granularity, so that it
> >>> is possible to protect/unprotect single pages, just like with
> >>> rodata_full and DEBUG_PAGEALLOC.
> >>>
> >>> Instead of repating
> >>>
> >>> 	can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE)
> >>>
> >>> make can_set_direct_map() handle the KFENCE case.
> >>>
> >>> This also prevents potential false positives in kernel_page_present()
> >>> that may return true for non-present page if CONFIG_KFENCE is enabled.
> >>>
> >>> Signed-off-by: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> >>> ---
> >>>  arch/arm64/mm/mmu.c      | 8 ++------
> >>>  arch/arm64/mm/pageattr.c | 8 +++++++-
> >>>  2 files changed, 9 insertions(+), 7 deletions(-)
> >>>
> >>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> >>> index e7ad44585f40..c5065abec55a 100644
> >>> --- a/arch/arm64/mm/mmu.c
> >>> +++ b/arch/arm64/mm/mmu.c
> >>> @@ -535,7 +535,7 @@ static void __init map_mem(pgd_t *pgdp)
> >>>  	 */
> >>>  	BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
> >>>  
> >>> -	if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
> >>> +	if (can_set_direct_map())
> >>>  		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
> >>>  
> >>>  	/*
> >>> @@ -1547,11 +1547,7 @@ int arch_add_memory(int nid, u64 start, u64 size,
> >>>  
> >>>  	VM_BUG_ON(!mhp_range_allowed(start, size, true));
> >>>  
> >>> -	/*
> >>> -	 * KFENCE requires linear map to be mapped at page granularity, so that
> >>> -	 * it is possible to protect/unprotect single pages in the KFENCE pool.
> >>> -	 */
> >>> -	if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
> >>> +	if (can_set_direct_map())
> >>>  		flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
> >>>  
> >>>  	__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
> >>> diff --git a/arch/arm64/mm/pageattr.c b/arch/arm64/mm/pageattr.c
> >>> index 64e985eaa52d..d107c3d434e2 100644
> >>> --- a/arch/arm64/mm/pageattr.c
> >>> +++ b/arch/arm64/mm/pageattr.c
> >>> @@ -21,7 +21,13 @@ bool rodata_full __ro_after_init = IS_ENABLED(CONFIG_RODATA_FULL_DEFAULT_ENABLED
> >>>  
> >>>  bool can_set_direct_map(void)
> >>>  {
> >>> -	return rodata_full || debug_pagealloc_enabled();
> >>> +	/*
> >>> +	 * rodata_full, DEBUG_PAGEALLOC and KFENCE require linear map to be
> >>> +	 * mapped at page granularity, so that it is possible to
> >>> +	 * protect/unprotect single pages.
> >>> +	 */
> >>> +	return rodata_full || debug_pagealloc_enabled() ||
> >>> +		IS_ENABLED(CONFIG_KFENCE);
> >>>  }
> >>
> >> Changing can_set_direct_map() also changes behaviour for other functions such as 
> >>
> >> set_direct_map_default_noflush()
> >> set_direct_map_invalid_noflush()
> >> __kernel_map_pages()
> >>
> >> Is that okay ?
> >  
> > Yes. Since KFENCE disables block mappings, these will actually change the
> > page tables.
> > Actually, before this change the test for can_set_direct_map() in these
> > functions was false negative when CONFIG_KFENCE=y
> 
> Okay but then should not this have a "Fixes:" tag as well ?
 
I feel that this is more of a theoretical bug and it's not worth
backporting to stable.
   
> >>>  static int change_page_range(pte_t *ptep, unsigned long addr, void *data)
> > 

-- 
Sincerely yours,
Mike.