mm/pagewalk.c: report holes in hugetlb ranges

commit 373c455 upstream.

This matters at least for the mincore syscall, which will otherwise copy
uninitialized memory from the page allocator to userspace.  It is
probably also a correctness error for /proc/$pid/pagemap, but I haven't
tested that.

Removing the `walk->hugetlb_entry` condition in walk_hugetlb_range() has
no effect because the caller already checks for that.

This only reports holes in hugetlb ranges to callers who have specified
a hugetlb_entry callback.

This issue was found using an AFL-based fuzzer.

v2:
 - don't crash on ->pte_hole==NULL (Andrew Morton)
 - add Cc stable (Andrew Morton)

Changed for 4.4/4.9 stable backport:
 - fix up conflict in the huge_pte_offset() call

Fixes: 1e25a27 ("mincore: apply page table walker on do_mincore()")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

mm/pagewalk.c

@@ -142,8 +142,12 @@ static int walk_hugetlb_range(unsigned long addr, unsigned long end,
 	do {
 		next = hugetlb_entry_end(h, addr, end);
 		pte = huge_pte_offset(walk->mm, addr & hmask);
-		if (pte && walk->hugetlb_entry)
+
+		if (pte)
 			err = walk->hugetlb_entry(pte, hmask, addr, next, walk);
+		else if (walk->pte_hole)
+			err = walk->pte_hole(addr, next, walk);
+
 		if (err)
 			break;
 	} while (addr = next, addr != end);