Re: [PATCH hotfix 6.12 v2 4/8] mm: resolve faulty mmap_region() error path behaviour

[Vlastimil Babka <vbabka@suse.cz>]

On 10/28/24 21:22, Linus Torvalds wrote:
> On Mon, 28 Oct 2024 at 10:18, Lorenzo Stoakes
> <lorenzo.stoakes@oracle.com> wrote:
>>
>> I'm genuinely not opposed to a horrible, awful:
>>
>> #ifdef CONFIG_ARM64
>>         if (file && file->f_ops == shmem_file_operations)
>>                 vm_flags |= VM_MTE_ALLOWED;
>> #endif
>>
>> Early in the operation prior to the arch_validate_flags() check.
> 
> I would just put it inside the arm64 code itself.
> 
> IOW, get rid of the VM_MTE_ALLOWED flag entirely, and just make the
> arm64 arch_validate_flags() code do something like
> 
>         if (flags & VM_MTE) {
>                 if (file->f_ops != shmem_file_operations)
>                         return false;
>         }
> 
> and be done with it.

VM_MTE_ALLOWED is also set by arm64's arch_calc_vm_flag_bits():

        if (system_supports_mte() && (flags & MAP_ANONYMOUS))
                return VM_MTE_ALLOWED;

And there's also this in arch/arm64/include/asm/page.h

#define VM_DATA_DEFAULT_FLAGS     (VM_DATA_FLAGS_TSK_EXEC | VM_MTE_ALLOWED)

So it would need to all be considered in the validation if we wanted to
replace VM_MTE_ALLOWED completely?

> Considering that we only have that horrendous arch_validate_flags()
> for two architectures, and that they both just have magical special
> cases for MTE-like behavior, I do think that just making it be a hack
> inside those functions is the way to go.
> 
>               Linus