1. UNIX File System (UFS)
Chapter Five

2. Topics Background: Old System V File System On Disk Format
inodes
fragments
block allocation
Consistency Issues

3. Features Traditional BSD Local File System
Compatible with Berkeley 4.3 Tahoe release
Digital Enhancements
increased file and path sizes
file block clustering
File on File Layer for STREAMS
Unified Buffer Cache Integration
Support for Enhanced File Attributes
Notes:
V4.0 Release Notes Limits
Supported Theoretical
file size 128 GB 1 TB

4. History: The old System V File System Layout
Boot
Block
Super
Block
Inode
Table
Data
Notes:
File names are limited to 14 characters.
Each file has an on-disk inode structure including:
Mode and type of file
Link count - number of directory entries that point to this inode
Owner - user ID and group ID
Size (bytes)
Disk map - an array of direct pointers to the data blocks for the file
The System V file system uses a simple but inefficient layout:
Bootblock - used on some systems to contain a bootstrap program
Superblock - describes the file system: total size, size of inode list, head of free block list, list of free inodes, flags, number of free blocks and inodes
Inode table - area for allocating inodes for files
Data region - remainder of the file system is for data blocks and indirect pointer blocks

5. UNIX File System Layout
Alternate
Super
Block
Boot
Cylinder
Group
Inode
Table
Data
Cylinder Group 0
Cylinder Group N
Partition
Notes:
The figure above illustrates UNIX file system layout. The figure is not drawn to scale. The data area is the largest by far.
File system space is divided into cylinder groups, providing:
Less head movement between inode table and data blocks, for faster performance
Duplication of file system structure for increased reliability
Cylinder groups require more overhead for storing extra structure, but provide greater performance.
The superblock is replicated in a different position in each cylinder group to prevent loss of critical data in case an entire cylinder or surface is damaged.

6. UFS Block Types bootblocks superblocks inodes data blocks
cylinder group information
Notes:
The file system subdivides a disk partition into one or more cylinder groups. A cylinder group is comprised of one or more consecutive cylinders on the disk. Associated with each cylinder group is some book keeping that includes a redundant copy of the superblock, space for inodes, a bit map describing available blocks in cylinder group. For each cylinder group a static number of inodes is allocated at file creation time.
Cylinder group information begins a floating offset from the beginning from the cylinder group. This prevents a single platter, track or cylinder failure from destroying all superblocks.

7. Inodes State of an individual file disk inode incore inode
stored on disk
sized to fit evenly in a disk block
incore inode
in-memory
disk inode information plus information related to being a currently open file
for UFS, vnode v_data private is an incore inode
Notes:
Disk Inode: Information that must be saved permanently;
Modes (Permissions)
Link Count
Owner and Group IDs
Access Times
File Size
Generation Number
Pointers to data blocks or sym link info (fastlink)
Incore Inode: Additional information needed when the file is referenced;
Pointers to inode lists
Device
Inode number
Pointer to mount structure
Offset of last entry found if a directory
Disk inode

8. Disk Inode 1 2 3 Mode Link Count UID GID Size Access Date Modify
Create 15
Pointers
Actual
Block
Gener-
ation
Number 1 2 3
Data Block Pointers
Level of
Indirection
Data
(not implemented)
SUID
SGID
Sticky
File type
User
Group
Other
127
Common to all File Systems
Specific to this File System
Notes:
A disk inode (struct dinode) is defined in kernel/ufs/dinode.h and is 128 bytes in size. The disk address pointers it contains are each 4 bytes.

9. Inode Numbers Inodes numbered from 1
Position #1,
originally an inode for bad data blocks, no longer used for that purpose.
Position #2
inode for the root directory of the file system
UFS file identity is <device, inumber>
Notes:

10. Fragments (1) Small Blocks Require More Overhead Large Blocks WasteX
FILE 1
FILE 2
Large Blocks Waste
Space X
FILE 1
FILE 2
X X X X X X X X
Notes:
The UNIX file system:
Uses a large block size (such as 8192 bytes) to maximize data transfer
Allows blocks to be split into fragments to reduce wasted space
The above figures illustrate the differences using different block sizes.
The Xs represent data and the boxes represent block size.
In the first two figures, we see file systems with only one block size, no fragment size. A small block size requires more overhead and more time to handle large files. A large block size wastes space when files are small.
UFS does not bother allocating fragments from file > 96 KB assuming they will grow larger.
Large Blocks and
Small Fragments
FILE 1
FILE 2
X X X X X X X X X

11. Fragments (2) UFS Blocks are 8192 Bytes Fragment
good for modern disk drives
bad for small files
space wasted with internal fragmentation
Fragment
Subdivision of a Block
1024 Bytes
Consecutive fragments are use for the last partial block of small files
Notes:
When the size of a file is increased;
If the file is empty, block size units are allocated and any remainder placed in a fragment(s)
If the existing fragment or block can contain the new data it is placed their
If the file already has fragments allocated, a new block is allocated and the existing fragment is copied into it an then the new data added at the end.

12. Fragment (3)
Cylinder Group Block has one bit set for each fragment in the cylinder group
117
118
119
Block #
Notes:
free bits

13. Disk Block Allocation Whenever Possible
First 96KB Direct Block Allocation
in cylinder containing inode
Each Additional 2 Meg
cylinder groups with "lots" of free space
Allocate consecutive disk blocks
Works well until disk reaches 90% full
Notes:
Data blocks of a file are placed in the same cylinder group when possible. A limit is placed on the percentage of a cylinder group that an individual file is allowed to consume before its additional blocks are allocated from a different cylinder group.

14. Inode Allocation Regular Files Directories
in cylinder group of parent directory
Directories
in different cylinder group than parent directory
Notes:
Regular files of a directory are placed in the same cylinder group as the directory when possible to minimize rotational delays for applications that try to modify all the files of a directory at the same time.
Directories are allocated in the cylinder group with the fewest number of directories already and greater than average free inode count.

15. Directories in UFS Collection of inode:file_name pairs
four byte inode number
two byte length of entry
two byte length of name
null terminated name
padding to next four byte boundary
Deleted entries become padding for previous
Entries must fit in a 512-byte sector so disk write is atomic
Notes:

16. Symbolic Links Hard Links: ln Soft Links: ln -s
new directory entry added referencing an existing inode
link count in existing inode incremented
Soft Links: ln -s
new directory entry added referencing a new inode
new inode is of type sym link
if link path short enough
stored in data block map area of inode
referred to as a fast link
else
link path stored in a data block
Notes:
A symbolic link is created in BSD systems by writing a pathname into the data blocks of a file and marking the file (vnode) type as VLNK.
The pathname translation mechanism checks for this type and reads the data as part of the pathname. This requires extra disk operations.
Digital UNIX UFS contains an optimization for symbolic links to reduce I/O called fast symbolic links.

17. UFS Consistency If Machine Crashes Disaster if the lost block
writes to data blocks may be lost
Disaster if the lost block
is an indirect block containing the disk addresses of newly written data blocks already on disk.
is a directory block containing new file references newly created files with written inodes.
contains free list information about newly allocated blocks.
Notes:

18. UFS Solution Some Writes are Synchronous
Any changed block that refers to other blocks
30 second update process sync's all mounted file systems
Directory entries are not written across sector boundaries
fsck utility after reboot
must know usual order of UFS synchronous writes
Notes:

19. Source References kernel/ufs/fs.h kernel/ufs/dinode.h
definitions of super block (struct fs) and cylinder groups (struct csum)
kernel/ufs/dinode.h
definition on disk inode (struct dinode)
kernel/ufs/inode.h
definition of in-memory inode (struct inode)
kernel/ufs/ufs_vnops.c
implementation of vnode ops for ufs
kernel/ufs/ufsmount.h
definition of the ufs mount structure
kernel/ufs/ufs_vfsops.c
implementation of the ufs vfs ops.