Re: [v4 PATCH] fs/proc: task_mmu.c: don't read mapcount for migration entry

[Yang Shi <shy828301@gmail.com>]

On Thu, Mar 23, 2023 at 3:11 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>
> On 3/23/23 11:08, David Hildenbrand wrote:
> > On 23.03.23 10:52, Vlastimil Babka wrote:
> >> On 2/3/22 19:26, Yang Shi wrote:
> >>> --- a/fs/proc/task_mmu.c
> >>> +++ b/fs/proc/task_mmu.c
> >>> @@ -440,7 +440,8 @@ static void smaps_page_accumulate(struct mem_size_stats *mss,
> >>>   }
> >>>
> >>>   static void smaps_account(struct mem_size_stats *mss, struct page *page,
> >>> -           bool compound, bool young, bool dirty, bool locked)
> >>> +           bool compound, bool young, bool dirty, bool locked,
> >>> +           bool migration)
> >>>   {
> >>>     int i, nr = compound ? compound_nr(page) : 1;
> >>>     unsigned long size = nr * PAGE_SIZE;
> >>> @@ -467,8 +468,15 @@ static void smaps_account(struct mem_size_stats *mss, struct page *page,
> >>>      * page_count(page) == 1 guarantees the page is mapped exactly once.
> >>>      * If any subpage of the compound page mapped with PTE it would elevate
> >>>      * page_count().
> >>> +    *
> >>> +    * The page_mapcount() is called to get a snapshot of the mapcount.
> >>> +    * Without holding the page lock this snapshot can be slightly wrong as
> >>> +    * we cannot always read the mapcount atomically.  It is not safe to
> >>> +    * call page_mapcount() even with PTL held if the page is not mapped,
> >>> +    * especially for migration entries.  Treat regular migration entries
> >>> +    * as mapcount == 1.
> >>>      */
> >>> -   if (page_count(page) == 1) {
> >>> +   if ((page_count(page) == 1) || migration) {
> >>
> >> Since this is now apparently a CVE-2023-1582 for whatever RHeasons...
> >>
> >> wonder if the patch actually works as intended when
> >> (page_count() || migration) is in this particular order and not the other one?
> >
> > Only the page_mapcount() call to a page that should be problematic, not
> > the page_count() call. There might be the rare chance of the page
>
> Oh right, page_mapcount() vs page_count(), I need more coffee.
>
> > getting remove due to memory offlining... but we're still holding the
> > page table lock with the migration entry, so we should be protected
> > against that.
> >
> > Regarding the CVE, IIUC the main reason for the CVE should be
> > RHEL-specific -- which behaves differently than other code bases; for
> > other code bases, it's just a way to trigger a BUG_ON as described here.

Out of curiosity, is there any public link for this CVE? Google search
can't find it.

>
> That's good to know so at least my bogus mail was useful for that, thanks!