The online resizing mechanism, despite its seemingly complex task, is actually rather simple in its implementation. In order to avoid a large amount of complexity it is only possible to increase the size of a filesystem while it is mounted. This addresses the primary requirement that a filesystem that is (nearly) full can have space added to it without interrupting the use of that system. The online resizing code depends on the underlying block device to handle all aspects of its own resizing prior to the start of filesystem resizing, and does nothing itself to manipulate the partition tables of LVM/MD block devices.
The ext2/3 filesystem is divided into one or more block allocation groups of a fixed size, with possibly a partial block group at the end of the filesystem [4]. The layout of each block group (where the inode and block allocation bitmaps and the inode table are stored) is kept in the group descriptor table. This table is stored at the start of at the first block group, and consists of one or more filesystem blocks, depending on the size of the filesystem. Backup copies of the group descriptor table are kept in more groups if the filesystem is large enough.
There are three primary phases by which a filesystem is grown. The first, and simplest, is to expand the last partial block group (if any) to be a full block group. The second phase is to add a new block group to an existing block in the group descriptor table. The third phase is to add a new block to the group descriptor table and add a new group to that block. All filesystem resizes are done incrementally, going through one or more of the phases to add free space to the end of the filesystem until the desired size is reached.