[PATCH v4 12/25] mm/memory: Enhance insert_page_into_pte_locked() to create writable mappings

[David Hildenbrand]

On 06.01.25 03:07, Alistair Popple wrote:
> On Fri, Dec 20, 2024 at 08:06:48PM +0100, David Hildenbrand wrote:
>> On 20.12.24 20:01, David Hildenbrand wrote:
>>> On 17.12.24 06:12, Alistair Popple wrote:
>>>> In preparation for using insert_page() for DAX, enhance
>>>> insert_page_into_pte_locked() to handle establishing writable
>>>> mappings.  Recall that DAX returns VM_FAULT_NOPAGE after installing a
>>>> PTE which bypasses the typical set_pte_range() in finish_fault.
>>>>
>>>> Signed-off-by: Alistair Popple <apopple at nvidia.com>
>>>> Suggested-by: Dan Williams <dan.j.williams at intel.com>
>>>>
>>>> ---
>>>>
>>>> Changes since v2:
>>>>
>>>>     - New patch split out from "mm/memory: Add dax_insert_pfn"
>>>> ---
>>>>     mm/memory.c | 45 +++++++++++++++++++++++++++++++++++++--------
>>>>     1 file changed, 37 insertions(+), 8 deletions(-)
>>>>
>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>> index 06bb29e..cd82952 100644
>>>> --- a/mm/memory.c
>>>> +++ b/mm/memory.c
>>>> @@ -2126,19 +2126,47 @@ static int validate_page_before_insert(struct vm_area_struct *vma,
>>>>     }
>>>>     static int insert_page_into_pte_locked(struct vm_area_struct *vma, pte_t *pte,
>>>> -			unsigned long addr, struct page *page, pgprot_t prot)
>>>> +				unsigned long addr, struct page *page,
>>>> +				pgprot_t prot, bool mkwrite)
>>>>     {
>>>>     	struct folio *folio = page_folio(page);
>>>> +	pte_t entry = ptep_get(pte);
>>>>     	pte_t pteval;
>>>> -	if (!pte_none(ptep_get(pte)))
>>>> -		return -EBUSY;
>>>> +	if (!pte_none(entry)) {
>>>> +		if (!mkwrite)
>>>> +			return -EBUSY;
>>>> +
>>>> +		/*
>>>> +		 * For read faults on private mappings the PFN passed in may not
>>>> +		 * match the PFN we have mapped if the mapped PFN is a writeable
>>>> +		 * COW page.  In the mkwrite case we are creating a writable PTE
>>>> +		 * for a shared mapping and we expect the PFNs to match. If they
>>>> +		 * don't match, we are likely racing with block allocation and
>>>> +		 * mapping invalidation so just skip the update.
>>>> +		 */
>>>
>>> Would it make sense to instead have here
>>>
>>> /* See insert_pfn(). */
>>>
>>> But ...
>>>
>>>> +		if (pte_pfn(entry) != page_to_pfn(page)) {
>>>> +			WARN_ON_ONCE(!is_zero_pfn(pte_pfn(entry)));
>>>> +			return -EFAULT;
>>>> +		}
>>>> +		entry = maybe_mkwrite(entry, vma);
>>>> +		entry = pte_mkyoung(entry);
>>>> +		if (ptep_set_access_flags(vma, addr, pte, entry, 1))
>>>> +			update_mmu_cache(vma, addr, pte);
>>>
>>> ... I am not sure if we want the above at all. Someone inserted a page,
>>> which is refcounted + mapcounted already.
>>>
>>> Now you ignore that and do like the second insertion "worked" ?
>>>
>>> No, that feels wrong, I suspect you will run into refcount+mapcount issues.
>>>
>>> If there is already something, inserting must fail IMHO. If you want to
>>> change something to upgrade write permissions, then a different
>>> interface should be used.
>>
>> Ah, now I realize that the early exit saves you because we won't adjust the
>> refcount +mapcount.
> 
> Right.
>   
>> I still wonder if that really belongs in here, I would prefer to not play
>> such tricks to upgrade write permissions if possible.
> 
> As you have pointed out this was all inspired (ie. mostly copied)
> from the existing insert_pfn() implementation which is used from
> vmf_insert_mixed{_mkwrite}().
> 
> I agree a different interface to upgrade permissions would be nice. However
> it's tricky because in general callers of these functions (eg. FS DAX) aren't
> aware if the page is already mapped by a PTE/PMD. They only know a fault has
> occured and the faulting permissions.
> 
> This wouldn't be impossible to fix - the mm does provide vm_ops->page_mkwrite()
> for permission upgrades. The difficulty is that most filesystems that support
> FS DAX (ie. ext4, XFS) don't treat a vm_ops->page_mkwrite() call any differently
> from a vm_ops->fault() call due to write fault. Therefore the FS DAX code is
> unaware of whether or not this is a permission upgrade or initial writeable
> mapping of the page in the VMA.
> 
> A further issue in there is currently no vm_ops->huge_mkwrite() callback.
> 
> Obviously this could all be plumbed through the MM/FS layers, but that would
> require a separate patch series. Given the current implementation has no issues
> beyond the cosmetic I'd rather not delay this series any longer, especially as
> the cosmetic defect is largely pre-existing (vmf_insert_mixed{_mkwrite}() could
> have equally had a separate upgrade interface).

Fine with me, just stumbled over it an thought "that looks odd".

-- 
Cheers,

David / dhildenb