Re: [PATCH v2 1/1] fs/splice: add missing callback for inaccessible pages

[Dave Hansen]

On 5/1/20 12:18 AM, Christian Borntraeger wrote:
>> 					unlock_page();
>> 	get_page();
>> 	// ^ OK because I have a ref
>> 	// do DMA on inaccessible page
>>
>> Because the make_secure_pte() code isn't looking for a *specific*
>> 'expected' value, it has no way of noticing that the extra ref snuck in
>> there.
> I think the expected calcution is actually doing that,giving back the minimum
> value when no one else has any references that are valid for I/O.
> 
> But I might not have understood what you are trying to tell me?

I was wrong.  I was looking at migrate_page_move_mapping():

>         int expected_count = expected_page_refs(mapping, page) + extra_count;
...
>         xas_lock_irq(&xas);
>         if (page_count(page) != expected_count || xas_load(&xas) != page) {
>                 xas_unlock_irq(&xas);
>                 return -EAGAIN;
>         }
> 
>         if (!page_ref_freeze(page, expected_count)) {
>                 xas_unlock_irq(&xas);
>                 return -EAGAIN;
>         }

I saw the check for page_count(page) *and* the page_ref_freeze() call.
My assumption was that both were needed.  My assumption was wrong.  (I
think the migrate_page_move_mapping() code may actually be doing a
superfluous check.)

The larger point, though, is that the s390 code ensures no extra
references exist upon entering make_secure_pte(), but it still has no
mechanism to prevent future, new references to page cache pages from
being created.

The one existing user of expected_page_refs() freezes the refs then
*removes* the page from the page cache (that's what the xas_lock_irq()
is for).  That stops *new* refs from being acquired.

The s390 code is missing an equivalent mechanism.

One example:

	page_freeze_refs();
	// page->_count==0 now
					find_get_page();
					// ^ sees a "freed" page
	page_unfreeze_refs();

find_get_page() will either fail to *find* the page because it will see
page->_refcount==0 think it is freed (not great), or it will
VM_BUG_ON_PAGE() in __page_cache_add_speculative().

My bigger point is that this patches doesn't systematically stop finding
page cache pages that are arch-inaccessible.  This patch hits *one* of
those sites.