[PATCH v2 2/2] arm64/mm: fix incorrect file_map_count for invalid pmd/pud

[Will Deacon]

On Thu, Nov 17, 2022 at 03:56:02PM +0800, Liu Shixin wrote:
> The page table check trigger BUG_ON() unexpectedly when split hugepage:
> 
>  ------------[ cut here ]------------
>  kernel BUG at mm/page_table_check.c:119!
>  Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
>  Dumping ftrace buffer:
>     (ftrace buffer empty)
>  Modules linked in:
>  CPU: 7 PID: 210 Comm: transhuge-stres Not tainted 6.1.0-rc3+ #748
>  Hardware name: linux,dummy-virt (DT)
>  pstate: 20000005 (nzCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>  pc : page_table_check_set.isra.0+0x398/0x468
>  lr : page_table_check_set.isra.0+0x1c0/0x468
> [...]
>  Call trace:
>   page_table_check_set.isra.0+0x398/0x468
>   __page_table_check_pte_set+0x160/0x1c0
>   __split_huge_pmd_locked+0x900/0x1648
>   __split_huge_pmd+0x28c/0x3b8
>   unmap_page_range+0x428/0x858
>   unmap_single_vma+0xf4/0x1c8
>   zap_page_range+0x2b0/0x410
>   madvise_vma_behavior+0xc44/0xe78
>   do_madvise+0x280/0x698
>   __arm64_sys_madvise+0x90/0xe8
>   invoke_syscall.constprop.0+0xdc/0x1d8
>   do_el0_svc+0xf4/0x3f8
>   el0_svc+0x58/0x120
>   el0t_64_sync_handler+0xb8/0xc0
>   el0t_64_sync+0x19c/0x1a0
> [...]
> 
> On arm64, pmd_leaf() will return true even if the pmd is invalid due to
> pmd_present_invalid() check. So in pmdp_invalidate() the file_map_count
> will not only decrease once but also increase once. Then in set_pte_at(),
> the file_map_count increase again, and so trigger BUG_ON() unexpectedly.
> 
> Fix this problem by adding pmd_valid() in pmd_user_accessible_page().
> Moreover, add pud_valid() for pud_user_accessible_page() too.
> 
> Fixes: 42b2547137f5 ("arm64/mm: enable ARCH_SUPPORTS_PAGE_TABLE_CHECK")
> Reported-by: Denys Vlasenko <dvlasenk at redhat.com>
> Signed-off-by: Liu Shixin <liushixin2 at huawei.com>
> Acked-by: Pasha Tatashin <pasha.tatashin at soleen.com>
> ---
>  arch/arm64/include/asm/pgtable.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
> index edf6625ce965..3bc64199aa2e 100644
> --- a/arch/arm64/include/asm/pgtable.h
> +++ b/arch/arm64/include/asm/pgtable.h
> @@ -863,12 +863,12 @@ static inline bool pte_user_accessible_page(pte_t pte)
>  
>  static inline bool pmd_user_accessible_page(pmd_t pmd)
>  {
> -	return pmd_leaf(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
> +	return pmd_valid(pmd) && pmd_leaf(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));

Hmm, doesn't this have a funny interaction with PROT_NONE where the pmd is
invalid but present? If you don't care about PROT_NONE, then you could just
do:

  pmd_valid(pmd) && !pmd_table(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd))

but if you do care then you could do:

  pmd_leaf(pmd) && !pmd_present_invalid(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd))

>  static inline bool pud_user_accessible_page(pud_t pud)
>  {
> -	return pud_leaf(pud) && pud_user(pud);
> +	return pud_valid(pud) && pud_leaf(pud) && pud_user(pud);

Not caused by this patch, but why don't we have something like a
pud_user_exec() check here like we do for the pte and pmd levels?

Will