Re: [PATCH] Initialise mmu_notifier_range correctly

From: Matthew Wilcox
Date: Thu Jan 03 2019 - 09:43:19 EST


On Thu, Jan 03, 2019 at 09:31:16AM -0500, Jerome Glisse wrote:
> On Wed, Jan 02, 2019 at 04:21:26PM -0800, Matthew Wilcox wrote:
> >
> > One of the paths in follow_pte_pmd() initialised the mmu_notifier_range
> > incorrectly.
> >
> > Signed-off-by: Matthew Wilcox <willy@xxxxxxxxxxxxx>
> > Fixes: ac46d4f3c432 ("mm/mmu_notifier: use structure for invalidate_range_start/end calls v2")
> > Tested-by: Dave Chinner <dchinner@xxxxxxxxxx>
>
> Actually now that i have read the code again this is not ok to
> do so. The caller of follow_pte_pmd() will call range_init and
> follow pmd will only update the range address. So existing code
> is ok.

I think you need to re-read your own patch.

`git show ac46d4f3c43241ffa23d5bf36153a0830c0e02cc`

@@ -4058,10 +4059,10 @@ static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address,
if (!pmdpp)
goto out;

- if (start && end) {
- *start = address & PMD_MASK;
- *end = *start + PMD_SIZE;
- mmu_notifier_invalidate_range_start(mm, *start, *end);
+ if (range) {
+ mmu_notifier_range_init(range, mm, address & PMD_MASK,
+ (address & PMD_MASK) + PMD_SIZE);
+ mmu_notifier_invalidate_range_start(range);

... so it's fine to call range_init() *here*.

@@ -4069,17 +4070,17 @@ static int __follow_pte_pmd(struct mm_struct *mm, unsign
ed long address,
[...]
if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
goto out;

- if (start && end) {
- *start = address & PAGE_MASK;
- *end = *start + PAGE_SIZE;
- mmu_notifier_invalidate_range_start(mm, *start, *end);
+ if (range) {
+ range->start = address & PAGE_MASK;
+ range->end = range->start + PAGE_SIZE;
+ mmu_notifier_invalidate_range_start(range);

... but then *not* here later in the same function? You're not making
any sense.