Ffs

The Berkeley FFS resides on a disk partition in blocks greater or equal to 4096 bytes (no less). It has a super block that contains filesystem specific data (eg. when the last fsck took place) which is at block 0. It also has backup superblocks in case the first superblock is corrupted (fsck -b to repair a broken superblock and check the filesystem with an alternate superblock), the next superblock is at block 32 and the others are made known at newfs time.

# newfs /dev/rsvnd0a
Warning: cylinder groups must have a multiple of 2 cylinders
Warning: 64 sector(s) in last cylinder unallocated
/dev/rsvnd0a:   65536 sectors in 656 cylinders of 1 tracks, 100 sectors
        32.0MB in 3 cyl groups (250 c/g, 12.21MB/g, 3136 i/g)
super-block backups (for fsck -b #) at:
 32, 25032, 50032,

Here in the example the backups are 32, 25032 and 50032. To find out the alternate superblocks after a filesystem has already been created use newfs -N.

Next (not counting the super-blocks) the filesystem is split into 2 parts. One for a long sequence of inodes and the other for data. The inodes are the reference points for objects and data in the filesystem and they hold information such as permissions, file link count, file size, atime, mtime, ctime, disk blocks used by file, chflags, file owner and file group. The size of a UFS1 inode is 128 bytes. By default an inode exists for every 8192 bytes of filesystem space, this is tweakable with newfs.

For more information on FFS see A fast filesystem for UNIX.

A breakdown of a disk

I created a virtual disk, it is 32MB big:

# ls -lh disk
-rw-r--r--  1 root  wheel  32.0M Oct 28 08:55 disk

WTF? It's a file! Ok so lets make it a disk device:

# vnconfig -cv svnd0 disk
svnd0: 33554432 bytes on disk

let's make a new partition on it called a:

# disklabel -E /dev/rsvnd0c
disklabel: Can't get bios geometry: Device not configured
Initial label editor (enter '?' for help at any prompt)
> a
partition: [a] 
offset: [0] 
size: [65536] 
FS type: [4.2BSD] 
> p a
device: /dev/rsvnd0c
type: SCSI
disk: vnd device
label: fictitious
bytes/sector: 512
sectors/track: 100
tracks/cylinder: 1
sectors/cylinder: 100
cylinders: 655
total sectors: 65536
free sectors: 0
rpm: 3600
16 partitions:
#             size        offset  fstype [fsize bsize  cpg]
  a:         65536             0  4.2BSD   2048 16384   16 # Cyl     0 -   655*
  c:         65536             0  unused      0     0      # Cyl     0 -   655*
> q
Write new label?: [y] y
# 

Now write the filesystem on it with newfs:

# newfs /dev/rsvnd0a
Warning: cylinder groups must have a multiple of 8 cylinders
Warning: 64 sector(s) in last cylinder unallocated
/dev/rsvnd0a:   65536 sectors in 656 cylinders of 1 tracks, 100 sectors
       32.0MB in 1 cyl groups (656 c/g, 32.03MB/g, 4096 i/g)
super-block backups (for fsck -b #) at:
 32,

mount the new filesystem to /mnt:

# mount /dev/svnd0a /mnt
# df /mnt
Filesystem  512-blocks      Used     Avail Capacity  Mounted on
/dev/svnd0a      64412         4     61188     0%    /mnt
# df -i /mnt
Filesystem  512-blocks      Used     Avail Capacity iused   ifree  %iused  Mounted on
/dev/svnd0a      64412         4     61188     0%       1    4093     0%   /mnt

OK let's count it up, we have 65536 blocks in total for the disk, 4094 inodes exist at 128 bytes size which is 1023 blocks, and we have 2 superblocks (0 and 32) which are 16384 bytes big each or 32 blocks, then the filesystem space claims 64412 blocks, this gives us a total of 65499 blocks which isn't quite the complete disksize of 65536 blocks, however I left the disklabel unaccounted for which also takes up some space.

Now let's fill the drive:

$ dd if=/dev/zero of=fill bs=1
/mnt: write failed, file system is full
dd: fill: No space left on device
31326209+0 records in
31326208+0 records out
31326208 bytes transferred in 40.458 secs (774289 bytes/sec)
$ df /mnt
Filesystem  512-blocks      Used     Avail Capacity  Mounted on
/dev/svnd0a      64412     61220       -28   100%    /mnt

100% capacity, but the used blocks aren't all the blocks that the filesystem claims, so let's repeat that last step but this time as root:

# dd if=/dev/zero of=fill bs=1
/mnt: write failed, file system is full
dd: fill: No space left on device
32931841+0 records in
32931840+0 records out
32931840 bytes transferred in 44.276 secs (743779 bytes/sec)
# df /mnt
Filesystem  512-blocks      Used     Avail Capacity  Mounted on
/dev/svnd0a      64412     64356     -3164   105%    /mnt

Lovely almost all used up and it can't go further. Note that we're using up 105% capacity this is built into ffs and can be tweaked at newfs time. The superuser gets 5% above 100% of capacity noone else.

UFS2

FreeBSD and NetBSD have a second version of FFS that allows filesystem sizes greater than 1 TB. This is UFS2. The size of an UFS2 inode is 256 bytes and thus UFS1 and UFS2 are not compatible.

Soft Updates

You will want to check out soft-updates if you have ffs, and most likely install it on all partitions except / (example for FreeBSD):

$ less -XF /usr/src/sys/contrib/softupdates/README

To see if you have soft-updates enabled: on FreeBSD:

$ mount 
/dev/ad0s1a on / (ufs, local)
devfs on /dev (devfs, local)
/dev/ad0s1g on /backup (ufs, local, soft-updates)
/dev/ad0s1e on /tmp (ufs, local, soft-updates)
/dev/ad0s1f on /usr (ufs, local, soft-updates)
/dev/ad0s1d on /var (ufs, local, soft-updates)

on OpenBSD

$  mount
/dev/wd0a on / type ffs (local, softdep)
/dev/wd0e on /usr type ffs (NFS exported, local, nodev, softdep)
/dev/wd0d on /var type ffs (local, nodev, nosuid, softdep)
mfs:22707 on /tmp type mfs (asynchronous, local, nodev, nosuid, size=261120 512-blocks)