Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | This is the implementation of the SystemV/Coherent filesystem for Linux. |
| 2 | It grew out of separate filesystem implementations |
| 3 | |
| 4 | Xenix FS Doug Evans <dje@cygnus.com> June 1992 |
| 5 | SystemV FS Paul B. Monday <pmonday@eecs.wsu.edu> March-June 1993 |
| 6 | Coherent FS B. Haible <haible@ma2s2.mathematik.uni-karlsruhe.de> June 1993 |
| 7 | |
| 8 | and was merged together in July 1993. |
| 9 | |
| 10 | These filesystems are rather similar. Here is a comparison with Minix FS: |
| 11 | |
| 12 | * Linux fdisk reports on partitions |
| 13 | - Minix FS 0x81 Linux/Minix |
| 14 | - Xenix FS ?? |
| 15 | - SystemV FS ?? |
| 16 | - Coherent FS 0x08 AIX bootable |
| 17 | |
| 18 | * Size of a block or zone (data allocation unit on disk) |
| 19 | - Minix FS 1024 |
| 20 | - Xenix FS 1024 (also 512 ??) |
| 21 | - SystemV FS 1024 (also 512 and 2048) |
| 22 | - Coherent FS 512 |
| 23 | |
| 24 | * General layout: all have one boot block, one super block and |
| 25 | separate areas for inodes and for directories/data. |
| 26 | On SystemV Release 2 FS (e.g. Microport) the first track is reserved and |
| 27 | all the block numbers (including the super block) are offset by one track. |
| 28 | |
| 29 | * Byte ordering of "short" (16 bit entities) on disk: |
| 30 | - Minix FS little endian 0 1 |
| 31 | - Xenix FS little endian 0 1 |
| 32 | - SystemV FS little endian 0 1 |
| 33 | - Coherent FS little endian 0 1 |
| 34 | Of course, this affects only the file system, not the data of files on it! |
| 35 | |
| 36 | * Byte ordering of "long" (32 bit entities) on disk: |
| 37 | - Minix FS little endian 0 1 2 3 |
| 38 | - Xenix FS little endian 0 1 2 3 |
| 39 | - SystemV FS little endian 0 1 2 3 |
| 40 | - Coherent FS PDP-11 2 3 0 1 |
| 41 | Of course, this affects only the file system, not the data of files on it! |
| 42 | |
| 43 | * Inode on disk: "short", 0 means non-existent, the root dir ino is: |
| 44 | - Minix FS 1 |
| 45 | - Xenix FS, SystemV FS, Coherent FS 2 |
| 46 | |
| 47 | * Maximum number of hard links to a file: |
| 48 | - Minix FS 250 |
| 49 | - Xenix FS ?? |
| 50 | - SystemV FS ?? |
| 51 | - Coherent FS >=10000 |
| 52 | |
| 53 | * Free inode management: |
| 54 | - Minix FS a bitmap |
| 55 | - Xenix FS, SystemV FS, Coherent FS |
| 56 | There is a cache of a certain number of free inodes in the super-block. |
| 57 | When it is exhausted, new free inodes are found using a linear search. |
| 58 | |
| 59 | * Free block management: |
| 60 | - Minix FS a bitmap |
| 61 | - Xenix FS, SystemV FS, Coherent FS |
| 62 | Free blocks are organized in a "free list". Maybe a misleading term, |
| 63 | since it is not true that every free block contains a pointer to |
| 64 | the next free block. Rather, the free blocks are organized in chunks |
| 65 | of limited size, and every now and then a free block contains pointers |
| 66 | to the free blocks pertaining to the next chunk; the first of these |
| 67 | contains pointers and so on. The list terminates with a "block number" |
| 68 | 0 on Xenix FS and SystemV FS, with a block zeroed out on Coherent FS. |
| 69 | |
| 70 | * Super-block location: |
| 71 | - Minix FS block 1 = bytes 1024..2047 |
| 72 | - Xenix FS block 1 = bytes 1024..2047 |
| 73 | - SystemV FS bytes 512..1023 |
| 74 | - Coherent FS block 1 = bytes 512..1023 |
| 75 | |
| 76 | * Super-block layout: |
| 77 | - Minix FS |
| 78 | unsigned short s_ninodes; |
| 79 | unsigned short s_nzones; |
| 80 | unsigned short s_imap_blocks; |
| 81 | unsigned short s_zmap_blocks; |
| 82 | unsigned short s_firstdatazone; |
| 83 | unsigned short s_log_zone_size; |
| 84 | unsigned long s_max_size; |
| 85 | unsigned short s_magic; |
| 86 | - Xenix FS, SystemV FS, Coherent FS |
| 87 | unsigned short s_firstdatazone; |
| 88 | unsigned long s_nzones; |
| 89 | unsigned short s_fzone_count; |
| 90 | unsigned long s_fzones[NICFREE]; |
| 91 | unsigned short s_finode_count; |
| 92 | unsigned short s_finodes[NICINOD]; |
| 93 | char s_flock; |
| 94 | char s_ilock; |
| 95 | char s_modified; |
| 96 | char s_rdonly; |
| 97 | unsigned long s_time; |
| 98 | short s_dinfo[4]; -- SystemV FS only |
| 99 | unsigned long s_free_zones; |
| 100 | unsigned short s_free_inodes; |
| 101 | short s_dinfo[4]; -- Xenix FS only |
| 102 | unsigned short s_interleave_m,s_interleave_n; -- Coherent FS only |
| 103 | char s_fname[6]; |
| 104 | char s_fpack[6]; |
| 105 | then they differ considerably: |
| 106 | Xenix FS |
| 107 | char s_clean; |
| 108 | char s_fill[371]; |
| 109 | long s_magic; |
| 110 | long s_type; |
| 111 | SystemV FS |
| 112 | long s_fill[12 or 14]; |
| 113 | long s_state; |
| 114 | long s_magic; |
| 115 | long s_type; |
| 116 | Coherent FS |
| 117 | unsigned long s_unique; |
| 118 | Note that Coherent FS has no magic. |
| 119 | |
| 120 | * Inode layout: |
| 121 | - Minix FS |
| 122 | unsigned short i_mode; |
| 123 | unsigned short i_uid; |
| 124 | unsigned long i_size; |
| 125 | unsigned long i_time; |
| 126 | unsigned char i_gid; |
| 127 | unsigned char i_nlinks; |
| 128 | unsigned short i_zone[7+1+1]; |
| 129 | - Xenix FS, SystemV FS, Coherent FS |
| 130 | unsigned short i_mode; |
| 131 | unsigned short i_nlink; |
| 132 | unsigned short i_uid; |
| 133 | unsigned short i_gid; |
| 134 | unsigned long i_size; |
| 135 | unsigned char i_zone[3*(10+1+1+1)]; |
| 136 | unsigned long i_atime; |
| 137 | unsigned long i_mtime; |
| 138 | unsigned long i_ctime; |
| 139 | |
| 140 | * Regular file data blocks are organized as |
| 141 | - Minix FS |
| 142 | 7 direct blocks |
| 143 | 1 indirect block (pointers to blocks) |
| 144 | 1 double-indirect block (pointer to pointers to blocks) |
| 145 | - Xenix FS, SystemV FS, Coherent FS |
| 146 | 10 direct blocks |
| 147 | 1 indirect block (pointers to blocks) |
| 148 | 1 double-indirect block (pointer to pointers to blocks) |
| 149 | 1 triple-indirect block (pointer to pointers to pointers to blocks) |
| 150 | |
| 151 | * Inode size, inodes per block |
| 152 | - Minix FS 32 32 |
| 153 | - Xenix FS 64 16 |
| 154 | - SystemV FS 64 16 |
| 155 | - Coherent FS 64 8 |
| 156 | |
| 157 | * Directory entry on disk |
| 158 | - Minix FS |
| 159 | unsigned short inode; |
| 160 | char name[14/30]; |
| 161 | - Xenix FS, SystemV FS, Coherent FS |
| 162 | unsigned short inode; |
| 163 | char name[14]; |
| 164 | |
| 165 | * Dir entry size, dir entries per block |
| 166 | - Minix FS 16/32 64/32 |
| 167 | - Xenix FS 16 64 |
| 168 | - SystemV FS 16 64 |
| 169 | - Coherent FS 16 32 |
| 170 | |
| 171 | * How to implement symbolic links such that the host fsck doesn't scream: |
| 172 | - Minix FS normal |
| 173 | - Xenix FS kludge: as regular files with chmod 1000 |
| 174 | - SystemV FS ?? |
| 175 | - Coherent FS kludge: as regular files with chmod 1000 |
| 176 | |
| 177 | |
| 178 | Notation: We often speak of a "block" but mean a zone (the allocation unit) |
| 179 | and not the disk driver's notion of "block". |
| 180 | |
| 181 | |
| 182 | Bruno Haible <haible@ma2s2.mathematik.uni-karlsruhe.de> |