Re: [RFC PATCH 01/14] userfaultfd: set dirty and young on writeprotect

From: David Hildenbrand <david@redhat.com>
To: Peter Xu <peterx@redhat.com>
Cc: Nadav Amit <nadav.amit@gmail.com>,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	Andrew Morton <akpm@linux-foundation.org>,
	Mike Rapoport <rppt@linux.ibm.com>,
	Axel Rasmussen <axelrasmussen@google.com>,
	Nadav Amit <namit@vmware.com>,
	Andrea Arcangeli <aarcange@redhat.com>,
	Andrew Cooper <andrew.cooper3@citrix.com>,
	Andy Lutomirski <luto@kernel.org>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Will Deacon <will@kernel.org>, Yu Zhao <yuzhao@google.com>,
	Nick Piggin <npiggin@gmail.com>
Subject: Re: [RFC PATCH 01/14] userfaultfd: set dirty and young on writeprotect
Date: Wed, 20 Jul 2022 21:55:35 +0200	[thread overview]
Message-ID: <4ad140b5-1d5b-2486-0893-7886a9cdfd76@redhat.com> (raw)
In-Reply-To: <YthcC78q1hdd7mNT@xz-m1.local>

On 20.07.22 21:48, Peter Xu wrote:
> On Wed, Jul 20, 2022 at 09:33:35PM +0200, David Hildenbrand wrote:
>> On 20.07.22 21:15, Peter Xu wrote:
>>> On Wed, Jul 20, 2022 at 05:10:37PM +0200, David Hildenbrand wrote:
>>>> For pagecache pages it may as well be *plain wrong* to bypass the write
>>>> fault handler and simply mark pages dirty+map them writable.
>>>
>>> Could you elaborate?
>>
>> Write-fault handling for some filesystems (that even require this
>> "slow path") is a bit special.
>>
>> For example, do_shared_fault() might have to call page_mkwrite().
>>
>> AFAIK file systems use that for lazy allocation of disk blocks.
>> If you simply go ahead and map a !dirty pagecache page writable
>> and mark it dirty, it will not trigger page_mkwrite() and you might
>> end up corrupting data.
>>
>> That's why we the old change_pte_range() code never touched
>> anything if the pte wasn't already dirty.
> 
> I don't think that pte_dirty() check was for the pagecache code. For any fs
> that has page_mkwrite() defined, it'll already have vma_wants_writenotify()
> return 1, so we'll never try to add write bit, hence we'll never even try
> to check pte_dirty().
> 

I might be too tired, but the whole reason we had this magic before my
commit in place was only for the pagecache.

With vma_wants_writenotify()=0 you can directly map the pages writable
and don't have to do these advanced checks here. In a writable
MAP_SHARED VMA you'll already have pte_write().

We only get !pte_write() in case we have vma_wants_writenotify()=1 ...

  try_change_writable = vma_wants_writenotify(vma, vma->vm_page_prot);

and that's the code that checked the dirty bit after all to decide --
amongst other things -- if we can simply map it writable without going
via the write fault handler and triggering do_shared_fault() .

See crazy/ugly FOLL_FORCE code in GUP that similarly checks the dirty bit.

But yeah, it's all confusing so I might just be wrong regarding
pagecache pages.

-- 
Thanks,

David / dhildenb