FS(5) File Formats Manual FS(5)
NAME
fs, inode - format of file system volume (2BSD)
SYNOPSIS
#include <sys/types.h>
#include <sys/fs.h>
#include <sys/inode.h>
DESCRIPTION
Every file system storage volume (e.g. disk) has a common format for certain vital information. Every such volume is divided into a cer-
tain number of blocks. The block size is DEV_BSIZE bytes; specified in <sys/param.h> - currently 1024.
Each disk drive contains some number of file systems each laid out on a contiguous partition of the disk. A file system consists of a boot
block, followed by a super block, followed by an inode area, followed by a data block area which takes up the remainder of the disk parti-
tion. The layout of the super block as defined in <sys/fs.h> is:
#define MAXMNTLEN 12
/*
* Structure of the super-block
*/
struct fs
{
u_short fs_isize; /* first block after i-list */
daddr_t fs_fsize; /* size in blocks of entire volume */
short fs_nfree; /* number of addresses in fs_free */
daddr_t fs_free[NICFREE]; /* free block list */
short fs_ninode; /* number of inodes in fs_inode */
ino_t fs_inode[NICINOD]; /* free inode list */
char fs_flock; /* lock during free list manipulation */
char fs_ilock; /* lock during i-list manipulation */
char fs_fmod; /* super block modified flag */
char fs_ronly; /* mounted read-only flag */
time_t fs_time; /* last super block update */
daddr_t fs_tfree; /* total free blocks */
ino_t fs_tinode; /* total free inodes */
short fs_step; /* optimal step in free list pattern */
short fs_cyl; /* number of blocks per pattern */
char fs_fsmnt[MAXMNTLEN]; /* ordinary file mounted on */
ino_t fs_lasti; /* start place for circular search */
ino_t fs_nbehind; /* est # free inodes before s_lasti */
u_short fs_flags; /* mount time flags */
};
File system: A file system is described by its super-block. Block 0 of each file system partition is unused and is available to contain a
bootstrap program, pack label, or other information. Block 1 (SUPERB) is the super block. The inode area starts immediately after the
super-block, in block 2. Fs_isize is the address of the first block after the inode area. Thus the inode area is fs_isize-2 blocks long.
Fs_fsize is the address of the first block not potentially available for allocation to a file. Thus the data block area is fs_fsize -
fs_isize blocks long.
Super block: The path name on which the file system is mounted is maintained in fs_fsmnt. Fs_flock, fs_ilock, fs_fmod, fs_ronly and
fs_flags are flags maintained in the in core copy of the super block while its file system is mounted and their values on disk are immate-
rial. Fs_fmod is used as a flag to indicate that the super-block has changed and should be copied to the disk during the next periodic
update of file system information. Fs_ronly is a write-protection indicator. It is a copy of the mount flags fs_flags anded with
MNT_RDONLY(see/sys/h/mount.h).
Fs_time is the last time the super-block of the file system was changed. During a reboot, the fs_time of the super-block for the root file
system is used to set the system's idea of the time.
Inode: The inode is the focus of all file activity in the UNIX file system. There is a unique inode allocated for each active file, each
current directory, each mounted-on file, text file, and the root. An inode is `named' by its device/i-number pair.
Inodes are 64 bytes long, so 16 of them fit into a block if DEV_BSIZE is 1024. The root inode is the root of the file system. Inode 0
can't be used for normal purposes and historically bad blocks were linked to inode 1, thus the root inode is 2 (inode 1 is no longer used
for this purpose, however numerous dump tapes make this assumption, so we are stuck with it). No other i-number has a built-in meaning.
The format of an inode as given in <sys/inode.h> is:
/*
* Inode structure as it appears on
* a disk block.
*/
struct dinode {
u_short di_mode; /* mode and type of file */
short di_nlink; /* number of links to file */
uid_t di_uid; /* owner's user id */
gid_t di_gid; /* owner's group id */
off_t di_size; /* number of bytes in file */
daddr_t di_addr[7]; /* 7 block addresses 4 bytes each */
u_short di_reserved[5];/* pad of 10 to make total size 64 */
u_short di_flags;
time_t di_atime; /* time last accessed */
time_t di_mtime; /* time last modified */
time_t di_ctime; /* time created */
};
/*
* 28 of the di_addr address bytes are used; 7 addresses of 4
* bytes each: 4 direct (4Kb directly accessible) and 3 indirect.
*/
#define NADDR 7
/* modes */
#define IFMT 0170000 /* type of file */
#define IFCHR 0020000 /* character special */
#define IFDIR 0040000 /* directory */
#define IFBLK 0060000 /* block special */
#define IFREG 0100000 /* regular */
#define IFLNK 0120000 /* symbolic link */
#define IFSOCK 0140000 /* socket */
#define ISUID 04000 /* set user id on execution */
#define ISGID 02000 /* set group id on execution */
#define ISVTX 01000 /* save swapped text even after use */
#define IREAD 0400 /* read, write, execute permissions */
#define IWRITE 0200
#define IEXEC 0100
Di_mode identifies the type of file the inode represents; it is encoded identically to the st_mode field of stat(2). Di_nlink is the num-
ber of directory entries (links) that refer to this inode. Di_uid and di_gid are the owner's user and group IDs. Di_size is the number of
bytes in the file. Di_atime and di_mtime are the times of last access and modification of the file contents (read, write or create);
Di_ctime records the time of last modification to the inode or to the file, and is used to determine whether it should be dumped by
dump(8).
Special files are recognized by their modes. A block-type special file is one which can potentially be mounted as a file system; a charac-
ter-type special file cannot, though it is not necessarily character-oriented. For special files, the first two bytes of the di_addr field
are occupied by the device code (see types(5)). The device codes of block and character special files overlap.
Disk addresses of plain files and directories are kept in the array di_addr. For a DEV_BSIZE of 1K bytes, 7 addresses are kept in di_addr
using 28 of the 40 bytes. The first 4 addresses specify device blocks directly. The last 3 addresses are singly, doubly and triply indi-
rect and point to blocks containing 256 further block pointers. There are 3 block addresses reserved as a pad to bring the total size of
an inode to 64 bytes. All block addresses are of type daddr_t (see types(5)).
For block b in a file to exist, it is not necessary that all blocks less than b exist. A zero block number indicates that the correspond-
ing block has never been allocated. Such a missing block reads as if it contained all zero bytes.
Free block list: The free data block list for each volume is maintained as follows. Fs_free[1], ... , fs_free[fs_nfree-1], contain up to
NICFREE free block numbers (NICFREE is a configuration constant defined in <sys/param.h>). Fs_free[0] is the block address of the head of
a chain of blocks constituting the free list. The layout of each block of the free chain as defined in <sys/fs.h> is:
struct fblk
{
short df_nfree; /* number of addresses in df_free */
daddr_t df_free[NICFREE]; /* free block list */
};
The fields df_nfree and df_free in a free block are used exactly like fs_nfree and fs_free in the super block.
The algorithm used to allocate a block is: decrement fs_nfree, and the new block number is fs_free[fs_nfree]. If the new block address is
0, there are no blocks left, so give an error. If fs_nfree became 0, read the new block into fs_nfree and fs_free.
To free a block: check if fs_nfree is NICFREE; if so, copy fs_nfree and the fs_free array into the newly freed block, write it out, and set
fs_nfree to 0. In any event set fs_free[fs_nfree] to the freed block's address and increment fs_nfree.
Fs_isize and fs_fsize are used by the system to check for bad block addresses; if an `impossible' block address is allocated from or
returned to the free list, a diagnostic is written on the console. Moreover, the free array is cleared, to prevent further allocation from
a presumably corrupted free list.
Fs_step and fs_cyl determine the block interleaving of files for fastest access; traditionally these were referred to as s_m and s_n
respectively. Fs_step is the distance between successive blocks and fs_cyl is the number of blocks before the pattern repeats. A file
system's interleaving factors are determined when it is first created by mkfs(8). Mkfs lays out the initial free list with these parame-
ters and fsck(8) can be used to rebuild the free list optimally (and assign new interleaving factors if necessary).
Free inode list: Fs_ninode is the number of free inode numbers in the fs_inode array.
To allocate an inode: if fs_ninode is greater than 0, decrement it and return fs_inode[fs_ninode]. If it was 0, read through the inode
area and place the numbers of all free inodes (up to NICINOD) into the fs_inode array, then try again. If a search for free inodes is nec-
essary, the search will start at the beginning of the inode area if fs_nbehind >= 4 x NICINOD, otherwise starting at fs_lasti and continu-
ing at the beginning of the inode area if NICINOD free inodes aren't found when the end of the inode area is reached. When a search com-
pletes the i-number of the first inode of the last block scanned in the search is left in fs_lasti.
To free an inode, provided fs_ninode is less than NICINODE, place its number into fs_inode[fs_ninode] and increment fs_ninode. If fs_nin-
ode is already NICINODE, don't bother to enter the freed inode into any table (fs_inode is only to speed up the allocation process; the
information as to whether the inode is really free or not is maintained in the inode itself). If the i-number of the freed inode is less
than fs_lasti increment fs_nbehind.
SEE ALSO
stat(2), dir(5), types(5), dcheck(8), fsck(8), icheck(8), mkfs(8), mount(8)
BUGS
It isn't the 4BSD fast file system. The 2BSD file system is a direct descendent of the V7 file system and exists little changed from that
ancestor. There are many performance holes in the file system.
Some changes from the original V7 file system have resulted in better performance: The larger block size (1Kb as opposed to the 512 byte
block size of V7) cuts the average number of system calls necessary to access a file by a factor of two; the smaller (in core) inodes
allowed by the smaller number of direct links kept in inodes saves valuable kernel data space allowing the kernel buffer cache to be made
larger while sacrificing only 1Kb of direct file accessing; and starting free inode searches at the position the last search ended cuts the
time to gather free inodes significantly.
However, the separation of inodes and data blocks into completely different areas of the disk, the handling of the free list, the lack of
any file allocation layout policy encouraging locality such as that found in the 4BSD file system and the still too small block size often
leads to extremely poor performance.
The separation of inodes and data blocks in the file system means that to access a file a seek will have to be made to the beginning of the
disk partition containing the file system followed another to the the actual data blocks of the file (often quite distant from the inode
area).
The free list which is laid out at file system creation for optimal file block allocation, becomes scrambled over time on an active file
system. This process is slowed down by the kernel which always frees blocks from unlink'ed or truncated files in reverse order thereby
maintaining strings of optimally laid out free blocks in the free list. Eventually, however, since both freed and allocated blocks use the
head of the free list, it's possible (and quite probable) to have most of the free list laid out optimally with the first portion totally
scrambled. As a trade off, a file system's free list may be rebuilt fairly frequently via icheck -s or fsck -s and most blocks allocated
will be localized as close to the the inode area as possible. Because of this problem, files are sometimes scattered across a file system
generating an unpleasant amount of disk arm movement. A nasty oscillation also occurs in the free block list when fs_nfree hovers around
NICFREE and 0 causing the free array to be constantly written out and read back in as blocks are freed and allocated.
For a more in depth analysis of the 2BSD file system, its shortcomings, and a description of the changes made for the 4BSD file system see
"A Fast File System for UNIX" by M. McKusick; W. Joy; S. Leffler; and R. Fabry.
3rd Berkeley Distribution January 27, 1996 FS(5)