1 /* 2 * bitmap.h: Copyright (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 3 * 4 * additions: Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. 5 */ 6 #ifndef BITMAP_H 7 #define BITMAP_H 1 8 9 #define BITMAP_MAJOR_LO 3 10 /* version 4 insists the bitmap is in little-endian order 11 * with version 3, it is host-endian which is non-portable 12 */ 13 #define BITMAP_MAJOR_HI 4 14 #define BITMAP_MAJOR_HOSTENDIAN 3 15 16 #define BITMAP_MINOR 39 17 18 /* 19 * in-memory bitmap: 20 * 21 * Use 16 bit block counters to track pending writes to each "chunk". 22 * The 2 high order bits are special-purpose, the first is a flag indicating 23 * whether a resync is needed. The second is a flag indicating whether a 24 * resync is active. 25 * This means that the counter is actually 14 bits: 26 * 27 * +--------+--------+------------------------------------------------+ 28 * | resync | resync | counter | 29 * | needed | active | | 30 * | (0-1) | (0-1) | (0-16383) | 31 * +--------+--------+------------------------------------------------+ 32 * 33 * The "resync needed" bit is set when: 34 * a '1' bit is read from storage at startup. 35 * a write request fails on some drives 36 * a resync is aborted on a chunk with 'resync active' set 37 * It is cleared (and resync-active set) when a resync starts across all drives 38 * of the chunk. 39 * 40 * 41 * The "resync active" bit is set when: 42 * a resync is started on all drives, and resync_needed is set. 43 * resync_needed will be cleared (as long as resync_active wasn't already set). 44 * It is cleared when a resync completes. 45 * 46 * The counter counts pending write requests, plus the on-disk bit. 47 * When the counter is '1' and the resync bits are clear, the on-disk 48 * bit can be cleared as well, thus setting the counter to 0. 49 * When we set a bit, or in the counter (to start a write), if the fields is 50 * 0, we first set the disk bit and set the counter to 1. 51 * 52 * If the counter is 0, the on-disk bit is clear and the stipe is clean 53 * Anything that dirties the stipe pushes the counter to 2 (at least) 54 * and sets the on-disk bit (lazily). 55 * If a periodic sweep find the counter at 2, it is decremented to 1. 56 * If the sweep find the counter at 1, the on-disk bit is cleared and the 57 * counter goes to zero. 58 * 59 * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block 60 * counters as a fallback when "page" memory cannot be allocated: 61 * 62 * Normal case (page memory allocated): 63 * 64 * page pointer (32-bit) 65 * 66 * [ ] ------+ 67 * | 68 * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters) 69 * c1 c2 c2048 70 * 71 * Hijacked case (page memory allocation failed): 72 * 73 * hijacked page pointer (32-bit) 74 * 75 * [ ][ ] (no page memory allocated) 76 * counter #1 (16-bit) counter #2 (16-bit) 77 * 78 */ 79 80 #ifdef __KERNEL__ 81 82 #define PAGE_BITS (PAGE_SIZE << 3) 83 #define PAGE_BIT_SHIFT (PAGE_SHIFT + 3) 84 85 typedef __u16 bitmap_counter_t; 86 #define COUNTER_BITS 16 87 #define COUNTER_BIT_SHIFT 4 88 #define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3) 89 90 #define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1))) 91 #define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2))) 92 #define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1) 93 #define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK) 94 #define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK) 95 #define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX) 96 97 /* how many counters per page? */ 98 #define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS) 99 /* same, except a shift value for more efficient bitops */ 100 #define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT) 101 /* same, except a mask value for more efficient bitops */ 102 #define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1) 103 104 #define BITMAP_BLOCK_SIZE 512 105 #define BITMAP_BLOCK_SHIFT 9 106 107 /* how many blocks per chunk? (this is variable) */ 108 #define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->mddev->bitmap_info.chunksize >> BITMAP_BLOCK_SHIFT) 109 #define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT) 110 #define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1) 111 112 /* when hijacked, the counters and bits represent even larger "chunks" */ 113 /* there will be 1024 chunks represented by each counter in the page pointers */ 114 #define PAGEPTR_BLOCK_RATIO(bitmap) \ 115 (CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1) 116 #define PAGEPTR_BLOCK_SHIFT(bitmap) \ 117 (CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1) 118 #define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1) 119 120 #endif 121 122 /* 123 * bitmap structures: 124 */ 125 126 #define BITMAP_MAGIC 0x6d746962 127 128 /* use these for bitmap->flags and bitmap->sb->state bit-fields */ 129 enum bitmap_state { 130 BITMAP_STALE = 0x002, /* the bitmap file is out of date or had -EIO */ 131 BITMAP_WRITE_ERROR = 0x004, /* A write error has occurred */ 132 BITMAP_HOSTENDIAN = 0x8000, 133 }; 134 135 /* the superblock at the front of the bitmap file -- little endian */ 136 typedef struct bitmap_super_s { 137 __le32 magic; /* 0 BITMAP_MAGIC */ 138 __le32 version; /* 4 the bitmap major for now, could change... */ 139 __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */ 140 __le64 events; /* 24 event counter for the bitmap (1)*/ 141 __le64 events_cleared;/*32 event counter when last bit cleared (2) */ 142 __le64 sync_size; /* 40 the size of the md device's sync range(3) */ 143 __le32 state; /* 48 bitmap state information */ 144 __le32 chunksize; /* 52 the bitmap chunk size in bytes */ 145 __le32 daemon_sleep; /* 56 seconds between disk flushes */ 146 __le32 write_behind; /* 60 number of outstanding write-behind writes */ 147 148 __u8 pad[256 - 64]; /* set to zero */ 149 } bitmap_super_t; 150 151 /* notes: 152 * (1) This event counter is updated before the eventcounter in the md superblock 153 * When a bitmap is loaded, it is only accepted if this event counter is equal 154 * to, or one greater than, the event counter in the superblock. 155 * (2) This event counter is updated when the other one is *if*and*only*if* the 156 * array is not degraded. As bits are not cleared when the array is degraded, 157 * this represents the last time that any bits were cleared. 158 * If a device is being added that has an event count with this value or 159 * higher, it is accepted as conforming to the bitmap. 160 * (3)This is the number of sectors represented by the bitmap, and is the range that 161 * resync happens across. For raid1 and raid5/6 it is the size of individual 162 * devices. For raid10 it is the size of the array. 163 */ 164 165 #ifdef __KERNEL__ 166 167 /* the in-memory bitmap is represented by bitmap_pages */ 168 struct bitmap_page { 169 /* 170 * map points to the actual memory page 171 */ 172 char *map; 173 /* 174 * in emergencies (when map cannot be alloced), hijack the map 175 * pointer and use it as two counters itself 176 */ 177 unsigned int hijacked:1; 178 /* 179 * count of dirty bits on the page 180 */ 181 unsigned int count:31; 182 }; 183 184 /* keep track of bitmap file pages that have pending writes on them */ 185 struct page_list { 186 struct list_head list; 187 struct page *page; 188 }; 189 190 /* the main bitmap structure - one per mddev */ 191 struct bitmap { 192 struct bitmap_page *bp; 193 unsigned long pages; /* total number of pages in the bitmap */ 194 unsigned long missing_pages; /* number of pages not yet allocated */ 195 196 struct mddev *mddev; /* the md device that the bitmap is for */ 197 198 /* bitmap chunksize -- how much data does each bit represent? */ 199 unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */ 200 unsigned long chunks; /* total number of data chunks for the array */ 201 202 __u64 events_cleared; 203 int need_sync; 204 205 /* bitmap spinlock */ 206 spinlock_t lock; 207 208 struct file *file; /* backing disk file */ 209 struct page *sb_page; /* cached copy of the bitmap file superblock */ 210 struct page **filemap; /* list of cache pages for the file */ 211 unsigned long *filemap_attr; /* attributes associated w/ filemap pages */ 212 unsigned long file_pages; /* number of pages in the file */ 213 int last_page_size; /* bytes in the last page */ 214 215 unsigned long flags; 216 217 int allclean; 218 219 atomic_t behind_writes; 220 unsigned long behind_writes_used; /* highest actual value at runtime */ 221 222 /* 223 * the bitmap daemon - periodically wakes up and sweeps the bitmap 224 * file, cleaning up bits and flushing out pages to disk as necessary 225 */ 226 unsigned long daemon_lastrun; /* jiffies of last run */ 227 unsigned long last_end_sync; /* when we lasted called end_sync to 228 * update bitmap with resync progress */ 229 230 atomic_t pending_writes; /* pending writes to the bitmap file */ 231 wait_queue_head_t write_wait; 232 wait_queue_head_t overflow_wait; 233 wait_queue_head_t behind_wait; 234 235 struct sysfs_dirent *sysfs_can_clear; 236 }; 237 238 /* the bitmap API */ 239 240 /* these are used only by md/bitmap */ 241 int bitmap_create(struct mddev *mddev); 242 int bitmap_load(struct mddev *mddev); 243 void bitmap_flush(struct mddev *mddev); 244 void bitmap_destroy(struct mddev *mddev); 245 246 void bitmap_print_sb(struct bitmap *bitmap); 247 void bitmap_update_sb(struct bitmap *bitmap); 248 249 int bitmap_setallbits(struct bitmap *bitmap); 250 void bitmap_write_all(struct bitmap *bitmap); 251 252 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e); 253 254 /* these are exported */ 255 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, 256 unsigned long sectors, int behind); 257 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, 258 unsigned long sectors, int success, int behind); 259 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded); 260 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted); 261 void bitmap_close_sync(struct bitmap *bitmap); 262 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector); 263 264 void bitmap_unplug(struct bitmap *bitmap); 265 void bitmap_daemon_work(struct mddev *mddev); 266 #endif 267 268 #endif 269