tmpfs: fix highmem swapoff crash regression

Commit 778dd89 ("tmpfs: fix race between umount and swapoff")
forgot the new rules for strict atomic kmap nesting, causing

  WARNING: at arch/x86/mm/highmem_32.c:81

from __kunmap_atomic(), then

  BUG: unable to handle kernel paging request at fffb9000

from shmem_swp_set() when shmem_unuse_inode() is handling swapoff with
highmem in use.  My disgrace again.

See
  https://bugzilla.kernel.org/show_bug.cgi?id=35352

Reported-by: Witold Baryluk <baryluk@smp.if.uj.edu.pl>
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/shmem.c

@@ -916,11 +916,12 @@ static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, s
 			if (size > ENTRIES_PER_PAGE)
 				size = ENTRIES_PER_PAGE;
 			offset = shmem_find_swp(entry, ptr, ptr+size);
+			shmem_swp_unmap(ptr);
 			if (offset >= 0) {
 				shmem_dir_unmap(dir);
+				ptr = shmem_swp_map(subdir);
 				goto found;
 			}
-			shmem_swp_unmap(ptr);
 		}
 	}
 lost1: