md/bitmap: copy correct data for bitmap super
authorShaohua Li <shli@fb.com>
Thu, 17 Aug 2017 17:35:11 +0000 (10:35 -0700)
committerShaohua Li <shli@fb.com>
Thu, 24 Aug 2017 17:04:54 +0000 (10:04 -0700)
raid5 cache could write bitmap superblock before bitmap superblock is
initialized. The bitmap superblock is less than 512B. The current code will
only copy the superblock to a new page and write the whole 512B, which will
zero the the data after the superblock. Unfortunately the data could include
bitmap, which we should preserve. The patch will make superblock read do 4k
chunk and we always copy the 4k data to new page, so the superblock write will
old data to disk and we don't change the bitmap.

Reported-by: Song Liu <songliubraving@fb.com>
Reviewed-by: Song Liu <songliubraving@fb.com>
Cc: stable@vger.kernel.org (4.10+)
Signed-off-by: Shaohua Li <shli@fb.com>
drivers/md/bitmap.c

index 40f3cd7eab0fc69ef0ea3e79efa079a33fc78aaf..67e992185a24cf73914ebc77b8075e25f6514aee 100644 (file)
@@ -625,7 +625,7 @@ re_read:
                err = read_sb_page(bitmap->mddev,
                                   offset,
                                   sb_page,
-                                  0, sizeof(bitmap_super_t));
+                                  0, PAGE_SIZE);
        }
        if (err)
                return err;
@@ -2118,7 +2118,7 @@ int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
        if (store.sb_page && bitmap->storage.sb_page)
                memcpy(page_address(store.sb_page),
                       page_address(bitmap->storage.sb_page),
-                      sizeof(bitmap_super_t));
+                      PAGE_SIZE);
        bitmap_file_unmap(&bitmap->storage);
        bitmap->storage = store;