[PATCH 1/3] mm/gup: fix races when looking up a CONT-PTE size hugetlb page

[Baolin Wang <baolin.wang@linux.alibaba.com>]

On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb:
2M and 1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size
specified.

When looking up a CONT-PTE size hugetlb page by follow_page(), it will
use pte_offset_map_lock() to get the pte lock for the CONT-PTE size
hugetlb in follow_page_pte(). However this pte lock is incorrect for
the CONT-PTE size hugetlb, since we should use mm->page_table_lock
by huge_pte_lockptr().

That means the pte entry of the CONT-PTE size hugetlb under current
pte lock is unstable in follow_page_pte(), we can continue to migrate
or poison the pte entry of the CONT-PTE size hugetlb, which can cause
some potential race issues, since the pte entry is unstable, and
following pte_xxx() validation is also incorrect in follow_page_pte(),
even though they are under the 'pte lock'.

To fix this issue, we should validate if it is a CONT-PTE size VMA
at first, and use huge_pte_lockptr() to get the correct pte lock
and get the pte value by huge_ptep_get() to make the pte entry stable
under the correct pte lock.

Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 mm/gup.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)

diff --git a/mm/gup.c b/mm/gup.c
index 5aa7531..3b2fa86 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -534,8 +534,26 @@ static struct page *follow_page_pte(struct vm_area_struct *vma,
 	if (unlikely(pmd_bad(*pmd)))
 		return no_page_table(vma, flags);
 
-	ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
-	pte = *ptep;
+	/*
+	 * Considering PTE level hugetlb, like continuous-PTE hugetlb on
+	 * ARM64 architecture.
+	 */
+	if (is_vm_hugetlb_page(vma)) {
+		struct hstate *hstate = hstate_vma(vma);
+		unsigned long size = huge_page_size(hstate);
+
+		ptep = huge_pte_offset(mm, address, size);
+		if (!ptep)
+			return no_page_table(vma, flags);
+
+		ptl = huge_pte_lockptr(hstate, mm, ptep);
+		spin_lock(ptl);
+		pte = huge_ptep_get(ptep);
+	} else {
+		ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
+		pte = *ptep;
+	}
+
 	if (!pte_present(pte)) {
 		swp_entry_t entry;
 		/*
-- 
1.8.3.1