btrfs: handle missing chunk mapping more gracefully

[BUG]
During my scrub rework, I did a stupid thing like this:

        bio->bi_iter.bi_sector = stripe->logical;
        btrfs_submit_bio(fs_info, bio, stripe->mirror_num);

Above bi_sector assignment is using logical address directly, which
lacks ">> SECTOR_SHIFT".

This results a read on a range which has no chunk mapping.

This results the following crash:

  BTRFS critical (device dm-1): unable to find logical 11274289152 length 65536
  assertion failed: !IS_ERR(em), in fs/btrfs/volumes.c:6387

Sure this is all my fault, but this shows a possible problem in real
world, that some bit flip in file extents/tree block can point to
unmapped ranges, and trigger above ASSERT(), or if CONFIG_BTRFS_ASSERT
is not configured, cause invalid pointer access.

[PROBLEMS]
In the above call chain, we just don't handle the possible error from
btrfs_get_chunk_map() inside __btrfs_map_block().

[FIX]
The fix is straightforward, replace the ASSERT() with proper error
handling (callers handle errors already).
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/volumes.c

@@ -6363,7 +6363,8 @@ int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
 	ASSERT(op != BTRFS_MAP_DISCARD);
 
 	em = btrfs_get_chunk_map(fs_info, logical, *length);
-	ASSERT(!IS_ERR(em));
+	if (IS_ERR(em))
+		return PTR_ERR(em);
 
 	map = em->map_lookup;
 	data_stripes = nr_data_stripes(map);