mm: pagewalk: Fix race between unmap and page walker

The mmap lock protects the page walker from changes to the page tables
during the walk.  However a read lock is insufficient to protect those
areas which don't have a VMA as munmap() detaches the VMAs before
downgrading to a read lock and actually tearing down PTEs/page tables.

For users of walk_page_range() the solution is to simply call pte_hole()
immediately without checking the actual page tables when a VMA is not
present. We now never call __walk_page_range() without a valid vma.

For walk_page_range_novma() the locking requirements are tightened to
require the mmap write lock to be taken, and then walking the pgd
directly with 'no_vma' set.

This in turn means that all page walkers either have a valid vma, or
it's that special 'novma' case for page table debugging.  As a result,
all the odd '(!walk->vma && !walk->no_vma)' tests can be removed.

Fixes: dd2283f ("mm: mmap: zap pages with read mmap_sem in munmap")
Reported-by: Jann Horn <[email protected]>
Signed-off-by: Steven Price <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Thomas Hellström <[email protected]>
Cc: Konstantin Khlebnikov <[email protected]>
Cc: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Author: Steven Price

arch/riscv/mm/pageattr.c

@@ -118,10 +118,10 @@ static int __set_memory(unsigned long addr, int numpages, pgprot_t set_mask,
 	if (!numpages)
 		return 0;
 
-	mmap_read_lock(&init_mm);
+	mmap_write_lock(&init_mm);
 	ret =  walk_page_range_novma(&init_mm, start, end, &pageattr_ops, NULL,
 				     &masks);
-	mmap_read_unlock(&init_mm);
+	mmap_write_unlock(&init_mm);
 
 	flush_tlb_kernel_range(start, end);
 

mm/pagewalk.c

@@ -110,7 +110,7 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
 	do {
 again:
 		next = pmd_addr_end(addr, end);
-		if (pmd_none(*pmd) || (!walk->vma && !walk->no_vma)) {
+		if (pmd_none(*pmd)) {
 			if (ops->pte_hole)
 				err = ops->pte_hole(addr, next, depth, walk);
 			if (err)
@@ -171,7 +171,7 @@ static int walk_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end,
 	do {
  again:
 		next = pud_addr_end(addr, end);
-		if (pud_none(*pud) || (!walk->vma && !walk->no_vma)) {
+		if (pud_none(*pud)) {
 			if (ops->pte_hole)
 				err = ops->pte_hole(addr, next, depth, walk);
 			if (err)
@@ -366,19 +366,19 @@ static int __walk_page_range(unsigned long start, unsigned long end,
 	struct vm_area_struct *vma = walk->vma;
 	const struct mm_walk_ops *ops = walk->ops;
 
-	if (vma && ops->pre_vma) {
+	if (ops->pre_vma) {
 		err = ops->pre_vma(start, end, walk);
 		if (err)
 			return err;
 	}
 
-	if (vma && is_vm_hugetlb_page(vma)) {
+	if (is_vm_hugetlb_page(vma)) {
 		if (ops->hugetlb_entry)
 			err = walk_hugetlb_range(start, end, walk);
 	} else
 		err = walk_pgd_range(start, end, walk);
 
-	if (vma && ops->post_vma)
+	if (ops->post_vma)
 		ops->post_vma(walk);
 
 	return err;
@@ -450,9 +450,13 @@ int walk_page_range(struct mm_struct *mm, unsigned long start,
 		if (!vma) { /* after the last vma */
 			walk.vma = NULL;
 			next = end;
+			if (ops->pte_hole)
+				err = ops->pte_hole(start, next, -1, &walk);
 		} else if (start < vma->vm_start) { /* outside vma */
 			walk.vma = NULL;
 			next = min(end, vma->vm_start);
+			if (ops->pte_hole)
+				err = ops->pte_hole(start, next, -1, &walk);
 		} else { /* inside vma */
 			walk.vma = vma;
 			next = min(end, vma->vm_end);
@@ -470,9 +474,8 @@ int walk_page_range(struct mm_struct *mm, unsigned long start,
 			}
 			if (err < 0)
 				break;
-		}
-		if (walk.vma || walk.ops->pte_hole)
 			err = __walk_page_range(start, next, &walk);
+		}
 		if (err)
 			break;
 	} while (start = next, start < end);
@@ -501,9 +504,9 @@ int walk_page_range_novma(struct mm_struct *mm, unsigned long start,
 	if (start >= end || !walk.mm)
 		return -EINVAL;
 
-	mmap_assert_locked(walk.mm);
+	mmap_assert_write_locked(walk.mm);
 
-	return __walk_page_range(start, end, &walk);
+	return walk_pgd_range(start, end, &walk);
 }
 
 int walk_page_vma(struct vm_area_struct *vma, const struct mm_walk_ops *ops,

mm/ptdump.c

@@ -152,13 +152,13 @@ void ptdump_walk_pgd(struct ptdump_state *st, struct mm_struct *mm, pgd_t *pgd)
 {
 	const struct ptdump_range *range = st->range;
 
-	mmap_read_lock(mm);
+	mmap_write_lock(mm);
 	while (range->start != range->end) {
 		walk_page_range_novma(mm, range->start, range->end,
 				      &ptdump_ops, pgd, st);
 		range++;
 	}
-	mmap_read_unlock(mm);
+	mmap_write_unlock(mm);
 
 	/* Flush out the last page */
 	st->note_page(st, 0, -1, 0);