[PATCH] arm64: mte: Allow PTRACE_PEEKMTETAGS access to the zero page

[Luis Machado]

On 2/10/21 3:03 PM, Catalin Marinas wrote:
> The ptrace(PTRACE_PEEKMTETAGS) implementation checks whether the user
> page has valid tags (mapped with PROT_MTE) by testing the PG_mte_tagged
> page flag. If this bit is cleared, ptrace(PTRACE_PEEKMTETAGS) returns
> -EIO.
> 
> A newly created (PROT_MTE) mapping points to the zero page which had its
> tags zeroed during cpu_enable_mte(). If there were no prior writes to
> this mapping, ptrace(PTRACE_PEEKMTETAGS) fails with -EIO since the zero
> page does not have the PG_mte_tagged flag set.
> 
> Set PG_mte_tagged on the zero page when its tags are cleared during
> boot. In addition, to avoid ptrace(PTRACE_PEEKMTETAGS) succeeding on
> !PROT_MTE mappings pointing to the zero page, change the
> __access_remote_tags() check to (vm_flags & VM_MTE) instead of
> PG_mte_tagged.
> 
> Signed-off-by: Catalin Marinas <catalin.marinas at arm.com>
> Fixes: 34bfeea4a9e9 ("arm64: mte: Clear the tags when a page is mapped in user-space with PROT_MTE")
> Cc: <stable at vger.kernel.org> # 5.10.x
> Cc: Will Deacon <will at kernel.org>
> Reported-by: Luis Machado <luis.machado at linaro.org>
> ---
> 
> The fix is actually checking VM_MTE instead of PG_mte_tagged in
> __access_remote_tags() but I added the WARN_ON(!PG_mte_tagged) and
> setting the flag on the zero page in case we break this assumption in
> the future.
> 
>   arch/arm64/kernel/cpufeature.c | 6 +-----
>   arch/arm64/kernel/mte.c        | 3 ++-
>   2 files changed, 3 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
> index e99eddec0a46..3e6331b64932 100644
> --- a/arch/arm64/kernel/cpufeature.c
> +++ b/arch/arm64/kernel/cpufeature.c
> @@ -1701,16 +1701,12 @@ static void bti_enable(const struct arm64_cpu_capabilities *__unused)
>   #ifdef CONFIG_ARM64_MTE
>   static void cpu_enable_mte(struct arm64_cpu_capabilities const *cap)
>   {
> -	static bool cleared_zero_page = false;
> -
>   	/*
>   	 * Clear the tags in the zero page. This needs to be done via the
>   	 * linear map which has the Tagged attribute.
>   	 */
> -	if (!cleared_zero_page) {
> -		cleared_zero_page = true;
> +	if (!test_and_set_bit(PG_mte_tagged, &ZERO_PAGE(0)->flags))
>   		mte_clear_page_tags(lm_alias(empty_zero_page));
> -	}
>   
>   	kasan_init_hw_tags_cpu();
>   }
> diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c
> index dc9ada64feed..80b62fe49dcf 100644
> --- a/arch/arm64/kernel/mte.c
> +++ b/arch/arm64/kernel/mte.c
> @@ -329,11 +329,12 @@ static int __access_remote_tags(struct mm_struct *mm, unsigned long addr,
>   		 * would cause the existing tags to be cleared if the page
>   		 * was never mapped with PROT_MTE.
>   		 */
> -		if (!test_bit(PG_mte_tagged, &page->flags)) {
> +		if (!(vma->vm_flags & VM_MTE)) {
>   			ret = -EOPNOTSUPP;
>   			put_page(page);
>   			break;
>   		}
> +		WARN_ON_ONCE(!test_bit(PG_mte_tagged, &page->flags));
>   
>   		/* limit access to the end of the page */
>   		offset = offset_in_page(addr);
> 

Thanks. I gave this a try and it works as expected. So memory that is 
PROT_MTE but has not been accessed yet can be inspected with PEEKMTETAGS 
without getting an EIO back.