drivers/md/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_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  * Version 5 is currently set only for clustered devices
  13  */
  14 #define BITMAP_MAJOR_HI 4
  15 #define BITMAP_MAJOR_CLUSTERED 5
  16 #define BITMAP_MAJOR_HOSTENDIAN 3
  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 stripe is clean
  53  * Anything that dirties the stripe 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_SHIFT 9
 105
 106 #endif
 107
 108 /*
 109  * bitmap structures:
 110  */
 111
 112 #define BITMAP_MAGIC 0x6d746962
 113
 114 /* use these for bitmap->flags and bitmap->sb->state bit-fields */
 115 enum bitmap_state {
 116         BITMAP_STALE       = 1,  /* the bitmap file is out of date or had -EIO */
 117         BITMAP_WRITE_ERROR = 2, /* A write error has occurred */
 118         BITMAP_HOSTENDIAN  =15,
 119 };
 120
 121 /* the superblock at the front of the bitmap file -- little endian */
 122 typedef struct bitmap_super_s {
 123         __le32 magic;        /*  0  BITMAP_MAGIC */
 124         __le32 version;      /*  4  the bitmap major for now, could change... */
 125         __u8  uuid[16];      /*  8  128 bit uuid - must match md device uuid */
 126         __le64 events;       /* 24  event counter for the bitmap (1)*/
 127         __le64 events_cleared;/*32  event counter when last bit cleared (2) */
 128         __le64 sync_size;    /* 40  the size of the md device's sync range(3) */
 129         __le32 state;        /* 48  bitmap state information */
 130         __le32 chunksize;    /* 52  the bitmap chunk size in bytes */
 131         __le32 daemon_sleep; /* 56  seconds between disk flushes */
 132         __le32 write_behind; /* 60  number of outstanding write-behind writes */
 133         __le32 sectors_reserved; /* 64 number of 512-byte sectors that are
 134                                   * reserved for the bitmap. */
 135         __le32 nodes;        /* 68 the maximum number of nodes in cluster. */
 136         __u8 cluster_name[64]; /* 72 cluster name to which this md belongs */
 137         __u8  pad[256 - 136]; /* set to zero */
 138 } bitmap_super_t;
 139
 140 /* notes:
 141  * (1) This event counter is updated before the eventcounter in the md superblock
 142  *    When a bitmap is loaded, it is only accepted if this event counter is equal
 143  *    to, or one greater than, the event counter in the superblock.
 144  * (2) This event counter is updated when the other one is *if*and*only*if* the
 145  *    array is not degraded.  As bits are not cleared when the array is degraded,
 146  *    this represents the last time that any bits were cleared.
 147  *    If a device is being added that has an event count with this value or
 148  *    higher, it is accepted as conforming to the bitmap.
 149  * (3)This is the number of sectors represented by the bitmap, and is the range that
 150  *    resync happens across.  For raid1 and raid5/6 it is the size of individual
 151  *    devices.  For raid10 it is the size of the array.
 152  */
 153
 154 #ifdef __KERNEL__
 155
 156 /* the in-memory bitmap is represented by bitmap_pages */
 157 struct bitmap_page {
 158         /*
 159          * map points to the actual memory page
 160          */
 161         char *map;
 162         /*
 163          * in emergencies (when map cannot be alloced), hijack the map
 164          * pointer and use it as two counters itself
 165          */
 166         unsigned int hijacked:1;
 167         /*
 168          * If any counter in this page is '1' or '2' - and so could be
 169          * cleared then that page is marked as 'pending'
 170          */
 171         unsigned int pending:1;
 172         /*
 173          * count of dirty bits on the page
 174          */
 175         unsigned int  count:30;
 176 };
 177
 178 /* the main bitmap structure - one per mddev */
 179 struct bitmap {
 180
 181         struct bitmap_counts {
 182                 spinlock_t lock;
 183                 struct bitmap_page *bp;
 184                 unsigned long pages;            /* total number of pages
 185                                                  * in the bitmap */
 186                 unsigned long missing_pages;    /* number of pages
 187                                                  * not yet allocated */
 188                 unsigned long chunkshift;       /* chunksize = 2^chunkshift
 189                                                  * (for bitops) */
 190                 unsigned long chunks;           /* Total number of data
 191                                                  * chunks for the array */
 192         } counts;
 193
 194         struct mddev *mddev; /* the md device that the bitmap is for */
 195
 196         __u64   events_cleared;
 197         int need_sync;
 198
 199         struct bitmap_storage {
 200                 struct file *file;              /* backing disk file */
 201                 struct page *sb_page;           /* cached copy of the bitmap
 202                                                  * file superblock */
 203                 struct page **filemap;          /* list of cache pages for
 204                                                  * the file */
 205                 unsigned long *filemap_attr;    /* attributes associated
 206                                                  * w/ filemap pages */
 207                 unsigned long file_pages;       /* number of pages in the file*/
 208                 unsigned long bytes;            /* total bytes in the bitmap */
 209         } storage;
 210
 211         unsigned long flags;
 212
 213         int allclean;
 214
 215         atomic_t behind_writes;
 216         unsigned long behind_writes_used; /* highest actual value at runtime */
 217
 218         /*
 219          * the bitmap daemon - periodically wakes up and sweeps the bitmap
 220          * file, cleaning up bits and flushing out pages to disk as necessary
 221          */
 222         unsigned long daemon_lastrun; /* jiffies of last run */
 223         unsigned long last_end_sync; /* when we lasted called end_sync to
 224                                       * update bitmap with resync progress */
 225
 226         atomic_t pending_writes; /* pending writes to the bitmap file */
 227         wait_queue_head_t write_wait;
 228         wait_queue_head_t overflow_wait;
 229         wait_queue_head_t behind_wait;
 230
 231         struct kernfs_node *sysfs_can_clear;
 232         int cluster_slot;               /* Slot offset for clustered env */
 233 };
 234
 235 /* the bitmap API */
 236
 237 /* these are used only by md/bitmap */
 238 struct bitmap *bitmap_create(struct mddev *mddev, int slot);
 239 int bitmap_load(struct mddev *mddev);
 240 void bitmap_flush(struct mddev *mddev);
 241 void bitmap_destroy(struct mddev *mddev);
 242
 243 void bitmap_print_sb(struct bitmap *bitmap);
 244 void bitmap_update_sb(struct bitmap *bitmap);
 245 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap);
 246
 247 int  bitmap_setallbits(struct bitmap *bitmap);
 248 void bitmap_write_all(struct bitmap *bitmap);
 249
 250 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
 251
 252 /* these are exported */
 253 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
 254                         unsigned long sectors, int behind);
 255 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
 256                         unsigned long sectors, int success, int behind);
 257 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded);
 258 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted);
 259 void bitmap_close_sync(struct bitmap *bitmap);
 260 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force);
 261
 262 void bitmap_unplug(struct bitmap *bitmap);
 263 void bitmap_daemon_work(struct mddev *mddev);
 264
 265 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
 266                   int chunksize, int init);
 267 int bitmap_copy_from_slot(struct mddev *mddev, int slot,
 268                                 sector_t *lo, sector_t *hi, bool clear_bits);
 269 #endif
 270
 271 #endif