bitmap.h

   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 3
  10 #define BITMAP_MINOR 39
  11
  12 /*
  13  * in-memory bitmap:
  14  *
  15  * Use 16 bit block counters to track pending writes to each "chunk".
  16  * The 2 high order bits are special-purpose, the first is a flag indicating
  17  * whether a resync is needed.  The second is a flag indicating whether a
  18  * resync is active.
  19  * This means that the counter is actually 14 bits:
  20  *
  21  * +--------+--------+------------------------------------------------+
  22  * | resync | resync |               counter                          |
  23  * | needed | active |                                                |
  24  * |  (0-1) |  (0-1) |              (0-16383)                         |
  25  * +--------+--------+------------------------------------------------+
  26  *
  27  * The "resync needed" bit is set when:
  28  *    a '1' bit is read from storage at startup.
  29  *    a write request fails on some drives
  30  *    a resync is aborted on a chunk with 'resync active' set
  31  * It is cleared (and resync-active set) when a resync starts across all drives
  32  * of the chunk.
  33  *
  34  *
  35  * The "resync active" bit is set when:
  36  *    a resync is started on all drives, and resync_needed is set.
  37  *       resync_needed will be cleared (as long as resync_active wasn't already set).
  38  * It is cleared when a resync completes.
  39  *
  40  * The counter counts pending write requests, plus the on-disk bit.
  41  * When the counter is '1' and the resync bits are clear, the on-disk
  42  * bit can be cleared aswell, thus setting the counter to 0.
  43  * When we set a bit, or in the counter (to start a write), if the fields is
  44  * 0, we first set the disk bit and set the counter to 1.
  45  *
  46  * If the counter is 0, the on-disk bit is clear and the stipe is clean
  47  * Anything that dirties the stipe pushes the counter to 2 (at least)
  48  * and sets the on-disk bit (lazily).
  49  * If a periodic sweep find the counter at 2, it is decremented to 1.
  50  * If the sweep find the counter at 1, the on-disk bit is cleared and the
  51  * counter goes to zero.
  52  *
  53  * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
  54  * counters as a fallback when "page" memory cannot be allocated:
  55  *
  56  * Normal case (page memory allocated):
  57  *
  58  *     page pointer (32-bit)
  59  *
  60  *     [ ] ------+
  61  *               |
  62  *               +-------> [   ][   ]..[   ] (4096 byte page == 2048 counters)
  63  *                          c1   c2    c2048
  64  *
  65  * Hijacked case (page memory allocation failed):
  66  *
  67  *     hijacked page pointer (32-bit)
  68  *
  69  *     [                  ][              ] (no page memory allocated)
  70  *      counter #1 (16-bit) counter #2 (16-bit)
  71  *
  72  */
  73
  74 #ifdef __KERNEL__
  75
  76 #define PAGE_BITS (PAGE_SIZE << 3)
  77 #define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
  78
  79 typedef __u16 bitmap_counter_t;
  80 #define COUNTER_BITS 16
  81 #define COUNTER_BIT_SHIFT 4
  82 #define COUNTER_BYTE_RATIO (COUNTER_BITS / 8)
  83 #define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3)
  84
  85 #define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1)))
  86 #define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2)))
  87 #define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1)
  88 #define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK)
  89 #define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK)
  90 #define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX)
  91
  92 /* how many counters per page? */
  93 #define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS)
  94 /* same, except a shift value for more efficient bitops */
  95 #define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT)
  96 /* same, except a mask value for more efficient bitops */
  97 #define PAGE_COUNTER_MASK  (PAGE_COUNTER_RATIO - 1)
  98
  99 #define BITMAP_BLOCK_SIZE 512
 100 #define BITMAP_BLOCK_SHIFT 9
 101
 102 /* how many blocks per chunk? (this is variable) */
 103 #define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->chunksize >> BITMAP_BLOCK_SHIFT)
 104 #define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT)
 105 #define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1)
 106
 107 /* when hijacked, the counters and bits represent even larger "chunks" */
 108 /* there will be 1024 chunks represented by each counter in the page pointers */
 109 #define PAGEPTR_BLOCK_RATIO(bitmap) \
 110                         (CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1)
 111 #define PAGEPTR_BLOCK_SHIFT(bitmap) \
 112                         (CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1)
 113 #define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1)
 114
 115 /*
 116  * on-disk bitmap:
 117  *
 118  * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
 119  * file a page at a time. There's a superblock at the start of the file.
 120  */
 121
 122 /* map chunks (bits) to file pages - offset by the size of the superblock */
 123 #define CHUNK_BIT_OFFSET(chunk) ((chunk) + (sizeof(bitmap_super_t) << 3))
 124
 125 #endif
 126
 127 /*
 128  * bitmap structures:
 129  */
 130
 131 #define BITMAP_MAGIC 0x6d746962
 132
 133 /* use these for bitmap->flags and bitmap->sb->state bit-fields */
 134 enum bitmap_state {
 135         BITMAP_ACTIVE = 0x001, /* the bitmap is in use */
 136         BITMAP_STALE  = 0x002  /* the bitmap file is out of date or had -EIO */
 137 };
 138
 139 /* the superblock at the front of the bitmap file -- little endian */
 140 typedef struct bitmap_super_s {
 141         __u32 magic;        /*  0  BITMAP_MAGIC */
 142         __u32 version;      /*  4  the bitmap major for now, could change... */
 143         __u8  uuid[16];     /*  8  128 bit uuid - must match md device uuid */
 144         __u64 events;       /* 24  event counter for the bitmap (1)*/
 145         __u64 events_cleared;/*32  event counter when last bit cleared (2) */
 146         __u64 sync_size;    /* 40  the size of the md device's sync range(3) */
 147         __u32 state;        /* 48  bitmap state information */
 148         __u32 chunksize;    /* 52  the bitmap chunk size in bytes */
 149         __u32 daemon_sleep; /* 56  seconds between disk flushes */
 150         __u32 write_behind; /* 60  number of outstanding write-behind writes */
 151
 152         __u8  pad[256 - 64]; /* set to zero */
 153 } bitmap_super_t;
 154
 155 /* notes:
 156  * (1) This event counter is updated before the eventcounter in the md superblock
 157  *    When a bitmap is loaded, it is only accepted if this event counter is equal
 158  *    to, or one greater than, the event counter in the superblock.
 159  * (2) This event counter is updated when the other one is *if*and*only*if* the
 160  *    array is not degraded.  As bits are not cleared when the array is degraded,
 161  *    this represents the last time that any bits were cleared.
 162  *    If a device is being added that has an event count with this value or
 163  *    higher, it is accepted as conforming to the bitmap.
 164  * (3)This is the number of sectors represented by the bitmap, and is the range that
 165  *    resync happens across.  For raid1 and raid5/6 it is the size of individual
 166  *    devices.  For raid10 it is the size of the array.
 167  */
 168
 169 #ifdef __KERNEL__
 170
 171 /* the in-memory bitmap is represented by bitmap_pages */
 172 struct bitmap_page {
 173         /*
 174          * map points to the actual memory page
 175          */
 176         char *map;
 177         /*
 178          * in emergencies (when map cannot be alloced), hijack the map
 179          * pointer and use it as two counters itself
 180          */
 181         unsigned int hijacked;
 182         /*
 183          * count of dirty bits on the page
 184          */
 185         int count;
 186 };
 187
 188 /* keep track of bitmap file pages that have pending writes on them */
 189 struct page_list {
 190         struct list_head list;
 191         struct page *page;
 192 };
 193
 194 /* the main bitmap structure - one per mddev */
 195 struct bitmap {
 196         struct bitmap_page *bp;
 197         unsigned long pages; /* total number of pages in the bitmap */
 198         unsigned long missing_pages; /* number of pages not yet allocated */
 199
 200         mddev_t *mddev; /* the md device that the bitmap is for */
 201
 202         int counter_bits; /* how many bits per block counter */
 203
 204         /* bitmap chunksize -- how much data does each bit represent? */
 205         unsigned long chunksize;
 206         unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */
 207         unsigned long chunks; /* total number of data chunks for the array */
 208
 209         /* We hold a count on the chunk currently being synced, and drop
 210          * it when the last block is started.  If the resync is aborted
 211          * midway, we need to be able to drop that count, so we remember
 212          * the counted chunk..
 213          */
 214         unsigned long syncchunk;
 215
 216         __u64   events_cleared;
 217
 218         /* bitmap spinlock */
 219         spinlock_t lock;
 220
 221         struct file *file; /* backing disk file */
 222         struct page *sb_page; /* cached copy of the bitmap file superblock */
 223         struct page **filemap; /* list of cache pages for the file */
 224         unsigned long *filemap_attr; /* attributes associated w/ filemap pages */
 225         unsigned long file_pages; /* number of pages in the file */
 226
 227         unsigned long flags;
 228
 229         /*
 230          * the bitmap daemon - periodically wakes up and sweeps the bitmap
 231          * file, cleaning up bits and flushing out pages to disk as necessary
 232          */
 233         mdk_thread_t *daemon;
 234         unsigned long daemon_sleep; /* how many seconds between updates? */
 235
 236         /*
 237          * bitmap write daemon - this daemon performs writes to the bitmap file
 238          * this thread is only needed because of a limitation in ext3 (jbd)
 239          * that does not allow a task to have two journal transactions ongoing
 240          * simultaneously (even if the transactions are for two different
 241          * filesystems) -- in the case of bitmap, that would be the filesystem
 242          * that the bitmap file resides on and the filesystem that is mounted
 243          * on the md device -- see current->journal_info in jbd/transaction.c
 244          */
 245         mdk_thread_t *write_daemon;
 246         mdk_thread_t *writeback_daemon;
 247         spinlock_t write_lock;
 248         struct semaphore write_ready;
 249         struct semaphore write_done;
 250         unsigned long writes_pending;
 251         wait_queue_head_t write_wait;
 252         struct list_head write_pages;
 253         struct list_head complete_pages;
 254         mempool_t *write_pool;
 255 };
 256
 257 /* the bitmap API */
 258
 259 /* these are used only by md/bitmap */
 260 int  bitmap_create(mddev_t *mddev);
 261 void bitmap_destroy(mddev_t *mddev);
 262 int  bitmap_active(struct bitmap *bitmap);
 263
 264 char *file_path(struct file *file, char *buf, int count);
 265 void bitmap_print_sb(struct bitmap *bitmap);
 266 int bitmap_update_sb(struct bitmap *bitmap);
 267
 268 int  bitmap_setallbits(struct bitmap *bitmap);
 269
 270 /* these are exported */
 271 void bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors);
 272 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
 273                      int success);
 274 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks);
 275 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted);
 276 void bitmap_close_sync(struct bitmap *bitmap);
 277
 278 int bitmap_unplug(struct bitmap *bitmap);
 279 #endif
 280
 281 #endif