Re: [PATCH v2 2/7] mm/gup: check if both GUP_GET and GUP_PIN are set in __get_user_pages() earlier

[David Hildenbrand]

On 01.04.25 16:34, Baoquan He wrote:
> On 04/01/25 at 10:02am, David Hildenbrand wrote:
> > On 31.03.25 10:13, Baoquan He wrote:
> > > In __get_user_pages(), it will traverse page table and take a reference
> > > to the page the given user address corresponds to if GUP_GET or GUP_PIN
> > > is et. However, it's not supported both GUP_GET and GUP_PIN are set.
> > > This check should be done earlier, but not doing it till entering into
> > > follow_page_pte() and failed.
> > >
> > > Here move the checking to the beginning of __get_user_pages().
> > >
> > > Signed-off-by: Baoquan He <bhe@xxxxxxxxxx>
> > > ---
> > > v1->v2:
> > > - Fix code bug caused by copy-and-paste error, this is reported by
> > >     lkp test robot.
> > >
> > >    mm/gup.c | 10 +++++-----
> > >    1 file changed, 5 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/mm/gup.c b/mm/gup.c
> > > index 73777b1de679..f9bce14ed3cd 100644
> > > --- a/mm/gup.c
> > > +++ b/mm/gup.c
> > > @@ -847,11 +847,6 @@ static struct page *follow_page_pte(struct vm_area_struct *vma,
> > >    	pte_t *ptep, pte;
> > >    	int ret;
> > > -	/* FOLL_GET and FOLL_PIN are mutually exclusive. */
> > > -	if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) ==
> > > -			 (FOLL_PIN | FOLL_GET)))
> > > -		return ERR_PTR(-EINVAL);
> > > -
> > >    	ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
> > >    	if (!ptep)
> > >    		return no_page_table(vma, flags, address);
> > > @@ -1434,6 +1429,11 @@ static long __get_user_pages(struct mm_struct *mm,
> > >    	VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN)));
> > > +	/* FOLL_GET and FOLL_PIN are mutually exclusive. */
> > > +	if (WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) ==
> > > +			 (FOLL_PIN | FOLL_GET)))
> > > +		return -EINVAL;
> > > +
> >
> > We already have that check in is_valid_gup_args(), that catches all external
> > users that could possibly mess this up.
>
> Right.
>
> > So we can just convert that into a VM_WARN_ON_ONCE(), and while doing that,
> > we should convert the VM_BUG_ON() above to a VM_WARN_ON_ONCE() as well.
>
> Sounds great to me, will put below change into this patch of v3 as suggested.
> Thanks.
>
> diff --git a/mm/gup.c b/mm/gup.c
> index 9e4ed09c578b..d551da9549b1 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -1427,10 +1427,10 @@ static long __get_user_pages(struct mm_struct *mm,
>   
>   	start = untagged_addr_remote(mm, start);
>   
> -	VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN)));
> +	VM_WARN_ON_ONCE(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN)));
>   
>   	/* FOLL_GET and FOLL_PIN are mutually exclusive. */
> -	if (WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) ==
> +	if (VM_WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) ==
>   			 (FOLL_PIN | FOLL_GET)))

That won't work (or shouldn't work :) ). Just drop the handling, it's a 
sanity check we don't expect to ever trigger, just like the old 
VM_BUG_ON above.

--
Cheers,

David / dhildenb