[PATCH 03/10] mm/hugetlb: Protect follow_huge_(pud|pgd) functions from race

[Anshuman Khandual <khandual@linux.vnet.ibm.com>]

follow_huge_(pmd|pud|pgd) functions are used to walk the page table and
fetch the page struct during 'follow_page_mask' call. There are possible
race conditions faced by these functions which arise out of simultaneous
calls of move_pages() and freeing of huge pages. This was fixed partly
by the previous commit e66f17ff7177 ("mm/hugetlb: take page table lock
in follow_huge_pmd()") for only PMD based huge pages.

After implementing similar logic, functions like follow_huge_(pud|pgd)
are now safe from above mentioned race conditions and also can support
FOLL_GET. Generic version of the function 'follow_huge_addr' has been
left as it is and its upto the architecture to decide on it.

Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
---
 include/linux/mm.h | 33 +++++++++++++++++++++++++++
 mm/hugetlb.c       | 67 ++++++++++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 91 insertions(+), 9 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index ffcff53..734182a 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1751,6 +1751,19 @@ static inline void pgtable_page_dtor(struct page *page)
 		NULL: pte_offset_kernel(pmd, address))
 
 #if USE_SPLIT_PMD_PTLOCKS
+static struct page *pgd_to_page(pgd_t *pgd)
+{
+	unsigned long mask = ~(PTRS_PER_PGD * sizeof(pgd_t) - 1);
+
+	return virt_to_page((void *)((unsigned long) pgd & mask));
+}
+
+static struct page *pud_to_page(pud_t *pud)
+{
+	unsigned long mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1);
+
+	return virt_to_page((void *)((unsigned long) pud & mask));
+}
 
 static struct page *pmd_to_page(pmd_t *pmd)
 {
@@ -1758,6 +1771,16 @@ static struct page *pmd_to_page(pmd_t *pmd)
 	return virt_to_page((void *)((unsigned long) pmd & mask));
 }
 
+static inline spinlock_t *pgd_lockptr(struct mm_struct *mm, pgd_t *pgd)
+{
+	return ptlock_ptr(pgd_to_page(pgd));
+}
+
+static inline spinlock_t *pud_lockptr(struct mm_struct *mm, pud_t *pud)
+{
+	return ptlock_ptr(pud_to_page(pud));
+}
+
 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
 {
 	return ptlock_ptr(pmd_to_page(pmd));
@@ -1783,6 +1806,16 @@ static inline void pgtable_pmd_page_dtor(struct page *page)
 
 #else
 
+static inline spinlock_t *pgd_lockptr(struct mm_struct *mm, pgd_t *pgd)
+{
+	return &mm->page_table_lock;
+}
+
+static inline spinlock_t *pud_lockptr(struct mm_struct *mm, pud_t *pud)
+{
+	return &mm->page_table_lock;
+}
+
 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
 {
 	return &mm->page_table_lock;
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 5ea3158..e84e479 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -4346,21 +4346,70 @@ struct page * __weak
 follow_huge_pud(struct mm_struct *mm, unsigned long address,
 		pud_t *pud, int flags)
 {
-	if (flags & FOLL_GET)
-		return NULL;
-
-	return pte_page(*(pte_t *)pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT);
+	struct page *page = NULL;
+	spinlock_t *ptl;
+retry:
+	ptl = pud_lockptr(mm, pud);
+	spin_lock(ptl);
+	/*
+	 * make sure that the address range covered by this pud is not
+	 * unmapped from other threads.
+	 */
+	if (!pud_huge(*pud))
+		goto out;
+	if (pud_present(*pud)) {
+		page = pud_page(*pud) + ((address & ~PUD_MASK) >> PAGE_SHIFT);
+		if (flags & FOLL_GET)
+			get_page(page);
+	} else {
+		if (is_hugetlb_entry_migration(huge_ptep_get((pte_t *)pud))) {
+			spin_unlock(ptl);
+			__migration_entry_wait(mm, (pte_t *)pud, ptl);
+			goto retry;
+		}
+		/*
+		 * hwpoisoned entry is treated as no_page_table in
+		 * follow_page_mask().
+		 */
+	}
+out:
+	spin_unlock(ptl);
+	return page;
 }
 
 struct page * __weak
 follow_huge_pgd(struct mm_struct *mm, unsigned long address,
 		pgd_t *pgd, int flags)
 {
-	if (flags & FOLL_GET)
-		return NULL;
-
-	return pte_page(*(pte_t *)pgd) +
-				((address & ~PGDIR_MASK) >> PAGE_SHIFT);
+	struct page *page = NULL;
+	spinlock_t *ptl;
+retry:
+	ptl = pgd_lockptr(mm, pgd);
+	spin_lock(ptl);
+	/*
+	 * make sure that the address range covered by this pgd is not
+	 * unmapped from other threads.
+	 */
+	if (!pgd_huge(*pgd))
+		goto out;
+	if (pgd_present(*pgd)) {
+		page = pgd_page(*pgd) + ((address & ~PGDIR_MASK) >> PAGE_SHIFT);
+		if (flags & FOLL_GET)
+			get_page(page);
+	} else {
+		if (is_hugetlb_entry_migration(huge_ptep_get((pte_t *)pgd))) {
+			spin_unlock(ptl);
+			__migration_entry_wait(mm, (pte_t *)pgd, ptl);
+			goto retry;
+		}
+		/*
+		 * hwpoisoned entry is treated as no_page_table in
+		 * follow_page_mask().
+		 */
+	}
+out:
+	spin_unlock(ptl);
+	return page;
 }
 
 #ifdef CONFIG_MEMORY_FAILURE
-- 
2.1.0