ext4: cleanup GFP flags inside resize path

ext4: cleanup GFP flags inside resize path

We must use GFP_NOFS instead GFP_KERNEL inside ext4_mb_add_groupinfo
and ext4_calculate_overhead() because they are called from inside a
journal transaction. Call trace:

ioctl
 ->ext4_group_add
   ->journal_start
   ->ext4_setup_new_descs
     ->ext4_mb_add_groupinfo -> GFP_KERNEL
   ->ext4_flex_group_add
     ->ext4_update_super
       ->ext4_calculate_overhead  -> GFP_KERNEL
   ->journal_stop

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>