| 1 | /* |
|---|
| 2 | md_p.h : physical layout of Linux RAID devices |
|---|
| 3 | Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman |
|---|
| 4 | |
|---|
| 5 | This program is free software; you can redistribute it and/or modify |
|---|
| 6 | it under the terms of the GNU General Public License as published by |
|---|
| 7 | the Free Software Foundation; either version 2, or (at your option) |
|---|
| 8 | any later version. |
|---|
| 9 | |
|---|
| 10 | You should have received a copy of the GNU General Public License |
|---|
| 11 | (for example /usr/src/linux/COPYING); if not, write to the Free |
|---|
| 12 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
|---|
| 13 | */ |
|---|
| 14 | |
|---|
| 15 | #ifndef _MD_P_H |
|---|
| 16 | #define _MD_P_H |
|---|
| 17 | |
|---|
| 18 | #include <asm/types.h> |
|---|
| 19 | |
|---|
| 20 | /* |
|---|
| 21 | * RAID superblock. |
|---|
| 22 | * |
|---|
| 23 | * The RAID superblock maintains some statistics on each RAID configuration. |
|---|
| 24 | * Each real device in the RAID set contains it near the end of the device. |
|---|
| 25 | * Some of the ideas are copied from the ext2fs implementation. |
|---|
| 26 | * |
|---|
| 27 | * We currently use 4096 bytes as follows: |
|---|
| 28 | * |
|---|
| 29 | * word offset function |
|---|
| 30 | * |
|---|
| 31 | * 0 - 31 Constant generic RAID device information. |
|---|
| 32 | * 32 - 63 Generic state information. |
|---|
| 33 | * 64 - 127 Personality specific information. |
|---|
| 34 | * 128 - 511 12 32-words descriptors of the disks in the raid set. |
|---|
| 35 | * 512 - 911 Reserved. |
|---|
| 36 | * 912 - 1023 Disk specific descriptor. |
|---|
| 37 | */ |
|---|
| 38 | |
|---|
| 39 | /* |
|---|
| 40 | * If x is the real device size in bytes, we return an apparent size of: |
|---|
| 41 | * |
|---|
| 42 | * y = (x & ~(MD_RESERVED_BYTES - 1)) - MD_RESERVED_BYTES |
|---|
| 43 | * |
|---|
| 44 | * and place the 4kB superblock at offset y. |
|---|
| 45 | */ |
|---|
| 46 | #define MD_RESERVED_BYTES (64 * 1024) |
|---|
| 47 | #define MD_RESERVED_SECTORS (MD_RESERVED_BYTES / 512) |
|---|
| 48 | #define MD_RESERVED_BLOCKS (MD_RESERVED_BYTES / BLOCK_SIZE) |
|---|
| 49 | |
|---|
| 50 | #define MD_NEW_SIZE_SECTORS(x) ((x & ~(MD_RESERVED_SECTORS - 1)) - MD_RESERVED_SECTORS) |
|---|
| 51 | #define MD_NEW_SIZE_BLOCKS(x) ((x & ~(MD_RESERVED_BLOCKS - 1)) - MD_RESERVED_BLOCKS) |
|---|
| 52 | |
|---|
| 53 | #define MD_SB_BYTES 4096 |
|---|
| 54 | #define MD_SB_WORDS (MD_SB_BYTES / 4) |
|---|
| 55 | #define MD_SB_BLOCKS (MD_SB_BYTES / BLOCK_SIZE) |
|---|
| 56 | #define MD_SB_SECTORS (MD_SB_BYTES / 512) |
|---|
| 57 | |
|---|
| 58 | /* |
|---|
| 59 | * The following are counted in 32-bit words |
|---|
| 60 | */ |
|---|
| 61 | #define MD_SB_GENERIC_OFFSET 0 |
|---|
| 62 | #define MD_SB_PERSONALITY_OFFSET 64 |
|---|
| 63 | #define MD_SB_DISKS_OFFSET 128 |
|---|
| 64 | #define MD_SB_DESCRIPTOR_OFFSET 992 |
|---|
| 65 | |
|---|
| 66 | #define MD_SB_GENERIC_CONSTANT_WORDS 32 |
|---|
| 67 | #define MD_SB_GENERIC_STATE_WORDS 32 |
|---|
| 68 | #define MD_SB_GENERIC_WORDS (MD_SB_GENERIC_CONSTANT_WORDS + MD_SB_GENERIC_STATE_WORDS) |
|---|
| 69 | #define MD_SB_PERSONALITY_WORDS 64 |
|---|
| 70 | #define MD_SB_DESCRIPTOR_WORDS 32 |
|---|
| 71 | #define MD_SB_DISKS 27 |
|---|
| 72 | #define MD_SB_DISKS_WORDS (MD_SB_DISKS*MD_SB_DESCRIPTOR_WORDS) |
|---|
| 73 | #define MD_SB_RESERVED_WORDS (1024 - MD_SB_GENERIC_WORDS - MD_SB_PERSONALITY_WORDS - MD_SB_DISKS_WORDS - MD_SB_DESCRIPTOR_WORDS) |
|---|
| 74 | #define MD_SB_EQUAL_WORDS (MD_SB_GENERIC_WORDS + MD_SB_PERSONALITY_WORDS + MD_SB_DISKS_WORDS) |
|---|
| 75 | |
|---|
| 76 | /* |
|---|
| 77 | * Device "operational" state bits |
|---|
| 78 | */ |
|---|
| 79 | #define MD_DISK_FAULTY 0 /* disk is faulty / operational */ |
|---|
| 80 | #define MD_DISK_ACTIVE 1 /* disk is running or spare disk */ |
|---|
| 81 | #define MD_DISK_SYNC 2 /* disk is in sync with the raid set */ |
|---|
| 82 | #define MD_DISK_REMOVED 3 /* disk is in sync with the raid set */ |
|---|
| 83 | |
|---|
| 84 | typedef struct mdp_device_descriptor_s { |
|---|
| 85 | __u32 number; /* 0 Device number in the entire set */ |
|---|
| 86 | __u32 major; /* 1 Device major number */ |
|---|
| 87 | __u32 minor; /* 2 Device minor number */ |
|---|
| 88 | __u32 raid_disk; /* 3 The role of the device in the raid set */ |
|---|
| 89 | __u32 state; /* 4 Operational state */ |
|---|
| 90 | __u32 reserved[MD_SB_DESCRIPTOR_WORDS - 5]; |
|---|
| 91 | } mdp_disk_t; |
|---|
| 92 | |
|---|
| 93 | #define MD_SB_MAGIC 0xa92b4efc |
|---|
| 94 | |
|---|
| 95 | /* |
|---|
| 96 | * Superblock state bits |
|---|
| 97 | */ |
|---|
| 98 | #define MD_SB_CLEAN 0 |
|---|
| 99 | #define MD_SB_ERRORS 1 |
|---|
| 100 | |
|---|
| 101 | typedef struct mdp_superblock_s { |
|---|
| 102 | /* |
|---|
| 103 | * Constant generic information |
|---|
| 104 | */ |
|---|
| 105 | __u32 md_magic; /* 0 MD identifier */ |
|---|
| 106 | __u32 major_version; /* 1 major version to which the set conforms */ |
|---|
| 107 | __u32 minor_version; /* 2 minor version ... */ |
|---|
| 108 | __u32 patch_version; /* 3 patchlevel version ... */ |
|---|
| 109 | __u32 gvalid_words; /* 4 Number of used words in this section */ |
|---|
| 110 | __u32 set_uuid0; /* 5 Raid set identifier */ |
|---|
| 111 | __u32 ctime; /* 6 Creation time */ |
|---|
| 112 | __u32 level; /* 7 Raid personality */ |
|---|
| 113 | __u32 size; /* 8 Apparent size of each individual disk */ |
|---|
| 114 | __u32 nr_disks; /* 9 total disks in the raid set */ |
|---|
| 115 | __u32 raid_disks; /* 10 disks in a fully functional raid set */ |
|---|
| 116 | __u32 md_minor; /* 11 preferred MD minor device number */ |
|---|
| 117 | __u32 not_persistent; /* 12 does it have a persistent superblock */ |
|---|
| 118 | __u32 set_uuid1; /* 13 Raid set identifier #2 */ |
|---|
| 119 | __u32 set_uuid2; /* 14 Raid set identifier #3 */ |
|---|
| 120 | __u32 set_uuid3; /* 15 Raid set identifier #4 */ |
|---|
| 121 | __u32 gstate_creserved[MD_SB_GENERIC_CONSTANT_WORDS - 16]; |
|---|
| 122 | |
|---|
| 123 | /* |
|---|
| 124 | * Generic state information |
|---|
| 125 | */ |
|---|
| 126 | __u32 utime; /* 0 Superblock update time */ |
|---|
| 127 | __u32 state; /* 1 State bits (clean, ...) */ |
|---|
| 128 | __u32 active_disks; /* 2 Number of currently active disks */ |
|---|
| 129 | __u32 working_disks; /* 3 Number of working disks */ |
|---|
| 130 | __u32 failed_disks; /* 4 Number of failed disks */ |
|---|
| 131 | __u32 spare_disks; /* 5 Number of spare disks */ |
|---|
| 132 | __u32 sb_csum; /* 6 checksum of the whole superblock */ |
|---|
| 133 | #ifdef __BIG_ENDIAN |
|---|
| 134 | __u32 events_hi; /* 7 high-order of superblock update count */ |
|---|
| 135 | __u32 events_lo; /* 8 low-order of superblock update count */ |
|---|
| 136 | __u32 cp_events_hi; /* 9 high-order of checkpoint update count */ |
|---|
| 137 | __u32 cp_events_lo; /* 10 low-order of checkpoint update count */ |
|---|
| 138 | #else |
|---|
| 139 | __u32 events_lo; /* 7 low-order of superblock update count */ |
|---|
| 140 | __u32 events_hi; /* 8 high-order of superblock update count */ |
|---|
| 141 | __u32 cp_events_lo; /* 9 low-order of checkpoint update count */ |
|---|
| 142 | __u32 cp_events_hi; /* 10 high-order of checkpoint update count */ |
|---|
| 143 | #endif |
|---|
| 144 | __u32 recovery_cp; /* 11 recovery checkpoint sector count */ |
|---|
| 145 | __u32 gstate_sreserved[MD_SB_GENERIC_STATE_WORDS - 12]; |
|---|
| 146 | |
|---|
| 147 | /* |
|---|
| 148 | * Personality information |
|---|
| 149 | */ |
|---|
| 150 | __u32 layout; /* 0 the array's physical layout */ |
|---|
| 151 | __u32 chunk_size; /* 1 chunk size in bytes */ |
|---|
| 152 | __u32 root_pv; /* 2 LV root PV */ |
|---|
| 153 | __u32 root_block; /* 3 LV root block */ |
|---|
| 154 | __u32 pstate_reserved[MD_SB_PERSONALITY_WORDS - 4]; |
|---|
| 155 | |
|---|
| 156 | /* |
|---|
| 157 | * Disks information |
|---|
| 158 | */ |
|---|
| 159 | mdp_disk_t disks[MD_SB_DISKS]; |
|---|
| 160 | |
|---|
| 161 | /* |
|---|
| 162 | * Reserved |
|---|
| 163 | */ |
|---|
| 164 | __u32 reserved[MD_SB_RESERVED_WORDS]; |
|---|
| 165 | |
|---|
| 166 | /* |
|---|
| 167 | * Active descriptor |
|---|
| 168 | */ |
|---|
| 169 | mdp_disk_t this_disk; |
|---|
| 170 | |
|---|
| 171 | } mdp_super_t; |
|---|
| 172 | |
|---|
| 173 | static inline __u64 md_event(mdp_super_t *sb) { |
|---|
| 174 | __u64 ev = sb->events_hi; |
|---|
| 175 | return (ev<<32)| sb->events_lo; |
|---|
| 176 | } |
|---|
| 177 | |
|---|
| 178 | /* |
|---|
| 179 | * The version-1 superblock : |
|---|
| 180 | * All numeric fields are little-endian. |
|---|
| 181 | * |
|---|
| 182 | * total size: 256 bytes plus 2 per device. |
|---|
| 183 | * 1K allows 384 devices. |
|---|
| 184 | */ |
|---|
| 185 | struct mdp_superblock_1 { |
|---|
| 186 | /* constant array information - 128 bytes */ |
|---|
| 187 | __u32 magic; /* MD_SB_MAGIC: 0xa92b4efc - little endian */ |
|---|
| 188 | __u32 major_version; /* 1 */ |
|---|
| 189 | __u32 feature_map; /* 0 for now */ |
|---|
| 190 | __u32 pad0; /* always set to 0 when writing */ |
|---|
| 191 | |
|---|
| 192 | __u8 set_uuid[16]; /* user-space generated. */ |
|---|
| 193 | char set_name[32]; /* set and interpreted by user-space */ |
|---|
| 194 | |
|---|
| 195 | __u64 ctime; /* lo 40 bits are seconds, top 24 are microseconds or 0*/ |
|---|
| 196 | __u32 level; /* -4 (multipath), -1 (linear), 0,1,4,5 */ |
|---|
| 197 | __u32 layout; /* only for raid5 currently */ |
|---|
| 198 | __u64 size; /* used size of component devices, in 512byte sectors */ |
|---|
| 199 | |
|---|
| 200 | __u32 chunksize; /* in 512byte sectors */ |
|---|
| 201 | __u32 raid_disks; |
|---|
| 202 | __u8 pad1[128-96]; /* set to 0 when written */ |
|---|
| 203 | |
|---|
| 204 | /* constant this-device information - 64 bytes */ |
|---|
| 205 | __u64 data_offset; /* sector start of data, often 0 */ |
|---|
| 206 | __u64 data_size; /* sectors in this device that can be used for data */ |
|---|
| 207 | __u64 super_offset; /* sector start of this superblock */ |
|---|
| 208 | __u64 recovery_offset;/* sectors before this offset (from data_offset) have been recovered */ |
|---|
| 209 | __u32 dev_number; /* permanent identifier of this device - not role in raid */ |
|---|
| 210 | __u32 cnt_corrected_read; /* number of read errors that were corrected by re-writing */ |
|---|
| 211 | __u8 device_uuid[16]; /* user-space setable, ignored by kernel */ |
|---|
| 212 | __u8 pad2[64-56]; /* set to 0 when writing */ |
|---|
| 213 | |
|---|
| 214 | /* array state information - 64 bytes */ |
|---|
| 215 | __u64 utime; /* 40 bits second, 24 btes microseconds */ |
|---|
| 216 | __u64 events; /* incremented when superblock updated */ |
|---|
| 217 | __u64 resync_offset; /* data before this offset (from data_offset) known to be in sync */ |
|---|
| 218 | __u32 sb_csum; /* checksum upto devs[max_dev] */ |
|---|
| 219 | __u32 max_dev; /* size of devs[] array to consider */ |
|---|
| 220 | __u8 pad3[64-32]; /* set to 0 when writing */ |
|---|
| 221 | |
|---|
| 222 | /* device state information. Indexed by dev_number. |
|---|
| 223 | * 2 bytes per device |
|---|
| 224 | * Note there are no per-device state flags. State information is rolled |
|---|
| 225 | * into the 'roles' value. If a device is spare or faulty, then it doesn't |
|---|
| 226 | * have a meaningful role. |
|---|
| 227 | */ |
|---|
| 228 | __u16 dev_roles[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */ |
|---|
| 229 | }; |
|---|
| 230 | |
|---|
| 231 | #endif |
|---|
| 232 | |
|---|
