Outline

talk about changes:

Explored in more detail: EXT234

EXT2 Released in 1993 with Linux 0.99

EXT3 Released in 2001 with Linux 2.4.15

talk about old hardware and simplicity

Journal

can be applied to multiple partition

helps with:

not losing data in crash

faster recovery in case of crash

solve the orphaned file problem (phantom inode)

Relegated to its own vid

in 2001 there was not a lot of multicore CPU and was non existent from the consumer market

but the writing was on the wall that we needed to optimize parallel load

there is 3 main way they did it:

they stopped updating the superblock s_inodes_count and S_blocks_count, only updating the block group descriptor
Replaced Big Kernel Lock (BKL) by smaller lock in Journal code.
removed sleep_on() by wait_event(). sleep_on() was unsafe, most implementation looked like this:

while (we_have_to_wait)

sleep_on(some_wait_queue);

10x perf boost

Preallocation (Old) used the block bitmap by preallocating block and correcting with full check in case of crash

Reservation (New) is a process 100% done in memory instead of in disk.

You have a reservation window which is 8 block long (unless specified in IOCTL command)

The availability of a Logical Volume Manager (LVM), motivated the desire for on-line resizing.

Here are the 3 part of growing:

Expand the last partial block group (if any) to be a full block group.
Add new entry in Group descriptor table and fill group. up to 16GB increment for 4k block size
Add a new block to the group descriptor table and add a new group to that block (Need to patch the FS before mounting)

superblock:

has some entry to know which feature is compatible or not

special inode:

has a special inode #8 for journal

directory:

that was there since 2nd version of ext2, we have the file type