tmpfs: fix shmem_getpage_gfp() VM_BUG_ON

commit 215c02b upstream.

Fuzzing with trinity hit the "impossible" VM_BUG_ON(error) (which Fedora
has converted to WARNING) in shmem_getpage_gfp():

  WARNING: at mm/shmem.c:1151 shmem_getpage_gfp+0xa5c/0xa70()
  Pid: 29795, comm: trinity-child4 Not tainted 3.7.0-rc2+ #49
  Call Trace:
    warn_slowpath_common+0x7f/0xc0
    warn_slowpath_null+0x1a/0x20
    shmem_getpage_gfp+0xa5c/0xa70
    shmem_fault+0x4f/0xa0
    __do_fault+0x71/0x5c0
    handle_pte_fault+0x97/0xae0
    handle_mm_fault+0x289/0x350
    __do_page_fault+0x18e/0x530
    do_page_fault+0x2b/0x50
    page_fault+0x28/0x30
    tracesys+0xe1/0xe6

Thanks to Johannes for pointing to truncation: free_swap_and_cache()
only does a trylock on the page, so the page lock we've held since
before confirming swap is not enough to protect against truncation.

What cleanup is needed in this case? Just delete_from_swap_cache(),
which takes care of the memcg uncharge.

Signed-off-by: Hugh Dickins <[email protected]>
Reported-by: Dave Jones <[email protected]>
Cc: Johannes Weiner <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>

Author: Hugh Dickins

mm/shmem.c

@@ -1156,8 +1156,20 @@ static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 		if (!error) {
 			error = shmem_add_to_page_cache(page, mapping, index,
 						gfp, swp_to_radix_entry(swap));
-			/* We already confirmed swap, and make no allocation */
-			VM_BUG_ON(error);
+			/*
+			 * We already confirmed swap under page lock, and make
+			 * no memory allocation here, so usually no possibility
+			 * of error; but free_swap_and_cache() only trylocks a
+			 * page, so it is just possible that the entry has been
+			 * truncated or holepunched since swap was confirmed.
+			 * shmem_undo_range() will have done some of the
+			 * unaccounting, now delete_from_swap_cache() will do
+			 * the rest (including mem_cgroup_uncharge_swapcache).
+			 * Reset swap.val? No, leave it so "failed" goes back to
+			 * "repeat": reading a hole and writing should succeed.
+			 */
+			if (error)
+				delete_from_swap_cache(page);
 		}
 		if (error)
 			goto failed;