bitmap.h 9.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270
  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. #define BITMAP_MAJOR_LO 3
  9. /* version 4 insists the bitmap is in little-endian order
  10. * with version 3, it is host-endian which is non-portable
  11. */
  12. #define BITMAP_MAJOR_HI 4
  13. #define BITMAP_MAJOR_HOSTENDIAN 3
  14. /*
  15. * in-memory bitmap:
  16. *
  17. * Use 16 bit block counters to track pending writes to each "chunk".
  18. * The 2 high order bits are special-purpose, the first is a flag indicating
  19. * whether a resync is needed. The second is a flag indicating whether a
  20. * resync is active.
  21. * This means that the counter is actually 14 bits:
  22. *
  23. * +--------+--------+------------------------------------------------+
  24. * | resync | resync | counter |
  25. * | needed | active | |
  26. * | (0-1) | (0-1) | (0-16383) |
  27. * +--------+--------+------------------------------------------------+
  28. *
  29. * The "resync needed" bit is set when:
  30. * a '1' bit is read from storage at startup.
  31. * a write request fails on some drives
  32. * a resync is aborted on a chunk with 'resync active' set
  33. * It is cleared (and resync-active set) when a resync starts across all drives
  34. * of the chunk.
  35. *
  36. *
  37. * The "resync active" bit is set when:
  38. * a resync is started on all drives, and resync_needed is set.
  39. * resync_needed will be cleared (as long as resync_active wasn't already set).
  40. * It is cleared when a resync completes.
  41. *
  42. * The counter counts pending write requests, plus the on-disk bit.
  43. * When the counter is '1' and the resync bits are clear, the on-disk
  44. * bit can be cleared as well, thus setting the counter to 0.
  45. * When we set a bit, or in the counter (to start a write), if the fields is
  46. * 0, we first set the disk bit and set the counter to 1.
  47. *
  48. * If the counter is 0, the on-disk bit is clear and the stipe is clean
  49. * Anything that dirties the stipe pushes the counter to 2 (at least)
  50. * and sets the on-disk bit (lazily).
  51. * If a periodic sweep find the counter at 2, it is decremented to 1.
  52. * If the sweep find the counter at 1, the on-disk bit is cleared and the
  53. * counter goes to zero.
  54. *
  55. * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
  56. * counters as a fallback when "page" memory cannot be allocated:
  57. *
  58. * Normal case (page memory allocated):
  59. *
  60. * page pointer (32-bit)
  61. *
  62. * [ ] ------+
  63. * |
  64. * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters)
  65. * c1 c2 c2048
  66. *
  67. * Hijacked case (page memory allocation failed):
  68. *
  69. * hijacked page pointer (32-bit)
  70. *
  71. * [ ][ ] (no page memory allocated)
  72. * counter #1 (16-bit) counter #2 (16-bit)
  73. *
  74. */
  75. #ifdef __KERNEL__
  76. #define PAGE_BITS (PAGE_SIZE << 3)
  77. #define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
  78. typedef __u16 bitmap_counter_t;
  79. #define COUNTER_BITS 16
  80. #define COUNTER_BIT_SHIFT 4
  81. #define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3)
  82. #define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1)))
  83. #define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2)))
  84. #define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1)
  85. #define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK)
  86. #define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK)
  87. #define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX)
  88. /* how many counters per page? */
  89. #define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS)
  90. /* same, except a shift value for more efficient bitops */
  91. #define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT)
  92. /* same, except a mask value for more efficient bitops */
  93. #define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1)
  94. #define BITMAP_BLOCK_SHIFT 9
  95. #endif
  96. /*
  97. * bitmap structures:
  98. */
  99. #define BITMAP_MAGIC 0x6d746962
  100. /* use these for bitmap->flags and bitmap->sb->state bit-fields */
  101. enum bitmap_state {
  102. BITMAP_STALE = 1, /* the bitmap file is out of date or had -EIO */
  103. BITMAP_WRITE_ERROR = 2, /* A write error has occurred */
  104. BITMAP_HOSTENDIAN =15,
  105. };
  106. /* the superblock at the front of the bitmap file -- little endian */
  107. typedef struct bitmap_super_s {
  108. __le32 magic; /* 0 BITMAP_MAGIC */
  109. __le32 version; /* 4 the bitmap major for now, could change... */
  110. __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */
  111. __le64 events; /* 24 event counter for the bitmap (1)*/
  112. __le64 events_cleared;/*32 event counter when last bit cleared (2) */
  113. __le64 sync_size; /* 40 the size of the md device's sync range(3) */
  114. __le32 state; /* 48 bitmap state information */
  115. __le32 chunksize; /* 52 the bitmap chunk size in bytes */
  116. __le32 daemon_sleep; /* 56 seconds between disk flushes */
  117. __le32 write_behind; /* 60 number of outstanding write-behind writes */
  118. __le32 sectors_reserved; /* 64 number of 512-byte sectors that are
  119. * reserved for the bitmap. */
  120. __le32 nodes; /* 68 the maximum number of nodes in cluster. */
  121. __u8 cluster_name[64]; /* 72 cluster name to which this md belongs */
  122. __u8 pad[256 - 136]; /* set to zero */
  123. } bitmap_super_t;
  124. /* notes:
  125. * (1) This event counter is updated before the eventcounter in the md superblock
  126. * When a bitmap is loaded, it is only accepted if this event counter is equal
  127. * to, or one greater than, the event counter in the superblock.
  128. * (2) This event counter is updated when the other one is *if*and*only*if* the
  129. * array is not degraded. As bits are not cleared when the array is degraded,
  130. * this represents the last time that any bits were cleared.
  131. * If a device is being added that has an event count with this value or
  132. * higher, it is accepted as conforming to the bitmap.
  133. * (3)This is the number of sectors represented by the bitmap, and is the range that
  134. * resync happens across. For raid1 and raid5/6 it is the size of individual
  135. * devices. For raid10 it is the size of the array.
  136. */
  137. #ifdef __KERNEL__
  138. /* the in-memory bitmap is represented by bitmap_pages */
  139. struct bitmap_page {
  140. /*
  141. * map points to the actual memory page
  142. */
  143. char *map;
  144. /*
  145. * in emergencies (when map cannot be alloced), hijack the map
  146. * pointer and use it as two counters itself
  147. */
  148. unsigned int hijacked:1;
  149. /*
  150. * If any counter in this page is '1' or '2' - and so could be
  151. * cleared then that page is marked as 'pending'
  152. */
  153. unsigned int pending:1;
  154. /*
  155. * count of dirty bits on the page
  156. */
  157. unsigned int count:30;
  158. };
  159. /* the main bitmap structure - one per mddev */
  160. struct bitmap {
  161. struct bitmap_counts {
  162. spinlock_t lock;
  163. struct bitmap_page *bp;
  164. unsigned long pages; /* total number of pages
  165. * in the bitmap */
  166. unsigned long missing_pages; /* number of pages
  167. * not yet allocated */
  168. unsigned long chunkshift; /* chunksize = 2^chunkshift
  169. * (for bitops) */
  170. unsigned long chunks; /* Total number of data
  171. * chunks for the array */
  172. } counts;
  173. struct mddev *mddev; /* the md device that the bitmap is for */
  174. __u64 events_cleared;
  175. int need_sync;
  176. struct bitmap_storage {
  177. struct file *file; /* backing disk file */
  178. struct page *sb_page; /* cached copy of the bitmap
  179. * file superblock */
  180. struct page **filemap; /* list of cache pages for
  181. * the file */
  182. unsigned long *filemap_attr; /* attributes associated
  183. * w/ filemap pages */
  184. unsigned long file_pages; /* number of pages in the file*/
  185. unsigned long bytes; /* total bytes in the bitmap */
  186. } storage;
  187. unsigned long flags;
  188. int allclean;
  189. atomic_t behind_writes;
  190. unsigned long behind_writes_used; /* highest actual value at runtime */
  191. /*
  192. * the bitmap daemon - periodically wakes up and sweeps the bitmap
  193. * file, cleaning up bits and flushing out pages to disk as necessary
  194. */
  195. unsigned long daemon_lastrun; /* jiffies of last run */
  196. unsigned long last_end_sync; /* when we lasted called end_sync to
  197. * update bitmap with resync progress */
  198. atomic_t pending_writes; /* pending writes to the bitmap file */
  199. wait_queue_head_t write_wait;
  200. wait_queue_head_t overflow_wait;
  201. wait_queue_head_t behind_wait;
  202. struct kernfs_node *sysfs_can_clear;
  203. int cluster_slot; /* Slot offset for clustered env */
  204. };
  205. /* the bitmap API */
  206. /* these are used only by md/bitmap */
  207. struct bitmap *bitmap_create(struct mddev *mddev, int slot);
  208. int bitmap_load(struct mddev *mddev);
  209. void bitmap_flush(struct mddev *mddev);
  210. void bitmap_destroy(struct mddev *mddev);
  211. void bitmap_print_sb(struct bitmap *bitmap);
  212. void bitmap_update_sb(struct bitmap *bitmap);
  213. void bitmap_status(struct seq_file *seq, struct bitmap *bitmap);
  214. int bitmap_setallbits(struct bitmap *bitmap);
  215. void bitmap_write_all(struct bitmap *bitmap);
  216. void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
  217. /* these are exported */
  218. int bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
  219. unsigned long sectors, int behind);
  220. void bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
  221. unsigned long sectors, int success, int behind);
  222. int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded);
  223. void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted);
  224. void bitmap_close_sync(struct bitmap *bitmap);
  225. void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector);
  226. void bitmap_unplug(struct bitmap *bitmap);
  227. void bitmap_daemon_work(struct mddev *mddev);
  228. int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
  229. int chunksize, int init);
  230. int bitmap_copy_from_slot(struct mddev *mddev, int slot,
  231. sector_t *lo, sector_t *hi, bool clear_bits);
  232. #endif
  233. #endif