BSD 4_3_Tahoe release

[unix-history] / usr / src / man / man5 / fs.5

.\" Copyright (c) 1983 Regents of the University of California.
.\" All rights reserved.  The Berkeley software License Agreement
.\" specifies the terms and conditions for redistribution.
.\"
.\"	@(#)fs.5	6.3 (Berkeley) 5/2/88
.\"
.TH FS 5 "May 2, 1988"
.UC 5
.SH NAME
fs, inode \- format of file system volume
.SH SYNOPSIS
.B #include <sys/types.h>
.br
.B #include <sys/fs.h>
.br
.B #include <sys/inode.h>
.SH DESCRIPTION
Every file system storage volume (disk, nine-track tape, for instance)
has a common format for certain vital information.
Every such volume is divided into a certain number of blocks.
The block size is a parameter of the file system.
Sectors beginning at BBLOCK and continuing for BBSIZE are used to
contain a label and for some hardware
primary and secondary bootstrapping programs.
.PP
The actual file system begins at sector SBLOCK with the
.I "super block"
that is of size SBSIZE.
The layout of the super block as defined by the include file
.RI < sys/fs.h >
is:
.PP
.nf
.ta \w'\ \ \ \ 'u +\w'daddr_t\ \ 'u +\w'fs_fsmnt[MAXMNTLEN];\ \ 'u
#define	FS_MAGIC 0x011954
struct fs {
	struct	fs *fs_link;	/* linked list of file systems */
	struct	fs *fs_rlink;	/*     used for incore super blocks */
	daddr_t	fs_sblkno;	/* addr of super-block in filesys */
	daddr_t	fs_cblkno;	/* offset of cyl-block in filesys */
	daddr_t	fs_iblkno;	/* offset of inode-blocks in filesys */
	daddr_t	fs_dblkno;	/* offset of first data after cg */
	long	fs_cgoffset;	/* cylinder group offset in cylinder */
	long	fs_cgmask;	/* used to calc mod fs_ntrak */
	time_t 	fs_time;    	/* last time written */
	long	fs_size;	/* number of blocks in fs */
	long	fs_dsize;	/* number of data blocks in fs */
	long	fs_ncg;	/* number of cylinder groups */
	long	fs_bsize;	/* size of basic blocks in fs */
	long	fs_fsize;	/* size of frag blocks in fs */
	long	fs_frag;	/* number of frags in a block in fs */
/* these are configuration parameters */
	long	fs_minfree;	/* minimum percentage of free blocks */
	long	fs_rotdelay;	/* num of ms for optimal next block */
	long	fs_rps;	/* disk revolutions per second */
/* these fields can be computed from the others */
	long	fs_bmask;	/* ``blkoff'' calc of blk offsets */
	long	fs_fmask;	/* ``fragoff'' calc of frag offsets */
	long	fs_bshift;	/* ``lblkno'' calc of logical blkno */
	long	fs_fshift;	/* ``numfrags'' calc number of frags */
/* these are configuration parameters */
	long	fs_maxcontig;	/* max number of contiguous blks */
	long	fs_maxbpg;	/* max number of blks per cyl group */
/* these fields can be computed from the others */
	long	fs_fragshift;	/* block to frag shift */
	long	fs_fsbtodb;	/* fsbtodb and dbtofsb shift constant */
	long	fs_sbsize;	/* actual size of super block */
	long	fs_csmask;	/* csum block offset */
	long	fs_csshift;	/* csum block number */
	long	fs_nindir;	/* value of NINDIR */
	long	fs_inopb;	/* value of INOPB */
	long	fs_nspf;	/* value of NSPF */
/* yet another configuration parameter */
	long	fs_optim;	/* optimization preference, see below */
/* these fields are derived from the hardware */
	long	fs_npsect;	/* # sectors/track including spares */
	long	fs_interleave;	/* hardware sector interleave */
	long	fs_trackskew;	/* sector 0 skew, per track */
	long	fs_headswitch;	/* head switch time, usec */
	long	fs_trkseek;	/* track-to-track seek, usec */
/* sizes determined by number of cylinder groups and their sizes */
	daddr_t fs_csaddr;	/* blk addr of cyl grp summary area */
	long	fs_cssize;	/* size of cyl grp summary area */
	long	fs_cgsize;	/* cylinder group size */
/* these fields are derived from the hardware */
	long	fs_ntrak;	/* tracks per cylinder */
	long	fs_nsect;	/* sectors per track */
	long  	fs_spc;   	/* sectors per cylinder */
/* this comes from the disk driver partitioning */
	long	fs_ncyl;   	/* cylinders in file system */
/* these fields can be computed from the others */
	long	fs_cpg;	/* cylinders per group */
	long	fs_ipg;	/* inodes per group */
	long	fs_fpg;	/* blocks per group * fs_frag */
/* this data must be re-computed after crashes */
	struct	csum fs_cstotal;	/* cylinder summary information */
/* these fields are cleared at mount time */
	char   	fs_fmod;    	/* super block modified flag */
	char   	fs_clean;    	/* file system is clean flag */
	char   	fs_ronly;   	/* mounted read-only flag */
	char   	fs_flags;   	/* currently unused flag */
	char	fs_fsmnt[MAXMNTLEN];	/* name mounted on */
/* these fields retain the current block allocation info */
	long	fs_cgrotor;	/* last cg searched */
	struct	csum *fs_csp[MAXCSBUFS]; /* list of fs_cs info buffers */
	long	fs_cpc;	/* cyl per cycle in postbl */
	short	fs_opostbl[16][8];	/* old rotation block list head */
	long	fs_sparecon[56];	/* reserved for future constants */
	quad	fs_qbmask;	/* ~fs_bmask - for use with quad size */
	quad	fs_qfmask;	/* ~fs_fmask - for use with quad size */
	long	fs_postblformat;	/* format of positional layout tables */
	long	fs_nrpos;	/* number of rotaional positions */
	long	fs_postbloff;	/* (short) rotation block list head */
	long	fs_rotbloff;	/* (u_char) blocks for each rotation */
	long	fs_magic;	/* magic number */
	u_char	fs_space[1];	/* list of blocks for each rotation */
/* actually longer */
};
.fi
.LP
Each disk drive contains some number of file systems.
A file system consists of a number of cylinder groups.
Each cylinder group has inodes and data.
.LP
A file system is described by its super-block, which in turn
describes the cylinder groups.  The super-block is critical
data and is replicated in each cylinder group to protect against
catastrophic loss.  This is done at file system creation
time and the critical
super-block data does not change, so the copies need not be
referenced further unless disaster strikes.
.LP
Addresses stored in inodes are capable of addressing fragments
of `blocks'. File system blocks of at most size MAXBSIZE can 
be optionally broken into 2, 4, or 8 pieces, each of which is
addressable; these pieces may be DEV_BSIZE, or some multiple of
a DEV_BSIZE unit.
.LP
Large files consist of exclusively large data blocks.  To avoid
undue wasted disk space, the last data block of a small file is
allocated as only as many fragments of a large block as are
necessary.  The file system format retains only a single pointer
to such a fragment, which is a piece of a single large block that
has been divided.  The size of such a fragment is determinable from
information in the inode, using the ``blksize(fs, ip, lbn)'' macro.
.LP
The file system records space availability at the fragment level;
to determine block availability, aligned fragments are examined.
.LP
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).
.LP
.I fs_minfree
gives the minimum acceptable percentage of file system
blocks that may be free. If the freelist drops below this level
only the super-user may continue to allocate blocks.
.I Fs_minfree
may be set to 0 if no reserve of free blocks is deemed necessary,
however severe performance degradations will be observed if the
file system is run at greater than 90% full; thus the default
value of
.I fs_minfree
is 10%.
.LP
Empirically the best trade-off between block fragmentation and
overall disk utilization at a loading of 90% comes with a
fragmentation of 8, thus the default fragment size is an eighth
of the block size.
.LP
.I fs_optim
specifies whether the file system should try to minimize the time spent
allocating blocks, or if it should attempt to minimize the space
fragmentation on the disk.
If the value of fs_minfree (see above) is less than 10%,
then the file system defaults to optimizing for space to avoid
running out of full sized blocks.
If the value of minfree is greater than or equal to 10%,
fragmentation is unlikely to be problematical, and
the file system defaults to optimizing for time.
.LP
.I Cylinder group related
.IR limits :
Each cylinder keeps track of the availability of blocks at different
rotational positions, so that sequential blocks can be laid out
with minimum rotational latency. With the default of 8 distinguished
rotational positions, the resolution of the
summary information is 2ms for a typical 3600 rpm drive.
.LP
.I fs_rotdelay
gives the minimum number of milliseconds to initiate
another disk transfer on the same cylinder.  It is used in
determining the rotationally optimal layout for disk blocks
within a file; the default value for
.I fs_rotdelay
is 2ms.
.LP
Each file system has a statically allocated number of inodes.
An inode is allocated for each NBPI bytes of disk space.
The inode allocation strategy is extremely conservative.
.LP
MINBSIZE is the smallest allowable block size.
With a MINBSIZE of 4096
it is possible to create files of size
2^32 with only two levels of indirection.
MINBSIZE must be big enough to hold a cylinder group block,
thus changes to (struct cg) must keep its size within MINBSIZE.
Note that super blocks are never more than size SBSIZE.
.LP
The path name on which the file system is mounted is maintained in
.IR fs_fsmnt .
MAXMNTLEN defines the amount of space allocated in 
the super block for this name.
The limit on the amount of summary information per file system
is defined by MAXCSBUFS.
For a 4096 byte block size, it is currently parameterized for a
maximum of two million cylinders.
.LP
Per cylinder group information is summarized in blocks allocated
from the first cylinder group's data blocks. 
These blocks are read in from
.I fs_csaddr
(size
.IR fs_cssize )
in addition to the super block.
.LP
.B N.B.:
sizeof (struct csum) must be a power of two in order for
the ``fs_cs'' macro to work.
.LP
.I Super block for a file
.IR system :
The size of the rotational layout tables
is limited by the fact that the super block is of size SBSIZE.
The size of these tables is
.B inversely
proportional to the block
size of the file system. The size of the tables is
increased when sector sizes are not powers of two,
as this increases the number of cylinders
included before the rotational pattern repeats (
.IR fs_cpc ).
The size of the rotational layout
tables is derived from the number of bytes remaining in (struct fs).
.LP
The number of blocks of data per cylinder group
is limited because cylinder groups are at most one block.
The inode and free block tables
must fit into a single block after deducting space for
the cylinder group structure (struct cg).
.LP
.IR 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.
For further information, see the include file
.RI < sys/inode.h >.