NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 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 | |
NeilBrown | bd926c6 | 2005-11-08 21:39:32 -0800 | [diff] [blame] | 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 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 16 | /* |
| 17 | * in-memory bitmap: |
| 18 | * |
| 19 | * Use 16 bit block counters to track pending writes to each "chunk". |
| 20 | * The 2 high order bits are special-purpose, the first is a flag indicating |
| 21 | * whether a resync is needed. The second is a flag indicating whether a |
| 22 | * resync is active. |
| 23 | * This means that the counter is actually 14 bits: |
| 24 | * |
| 25 | * +--------+--------+------------------------------------------------+ |
| 26 | * | resync | resync | counter | |
| 27 | * | needed | active | | |
| 28 | * | (0-1) | (0-1) | (0-16383) | |
| 29 | * +--------+--------+------------------------------------------------+ |
| 30 | * |
| 31 | * The "resync needed" bit is set when: |
| 32 | * a '1' bit is read from storage at startup. |
| 33 | * a write request fails on some drives |
| 34 | * a resync is aborted on a chunk with 'resync active' set |
| 35 | * It is cleared (and resync-active set) when a resync starts across all drives |
| 36 | * of the chunk. |
| 37 | * |
| 38 | * |
| 39 | * The "resync active" bit is set when: |
| 40 | * a resync is started on all drives, and resync_needed is set. |
| 41 | * resync_needed will be cleared (as long as resync_active wasn't already set). |
| 42 | * It is cleared when a resync completes. |
| 43 | * |
| 44 | * The counter counts pending write requests, plus the on-disk bit. |
| 45 | * When the counter is '1' and the resync bits are clear, the on-disk |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 46 | * bit can be cleared as well, thus setting the counter to 0. |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 47 | * When we set a bit, or in the counter (to start a write), if the fields is |
| 48 | * 0, we first set the disk bit and set the counter to 1. |
| 49 | * |
| 50 | * If the counter is 0, the on-disk bit is clear and the stipe is clean |
| 51 | * Anything that dirties the stipe pushes the counter to 2 (at least) |
| 52 | * and sets the on-disk bit (lazily). |
| 53 | * If a periodic sweep find the counter at 2, it is decremented to 1. |
| 54 | * If the sweep find the counter at 1, the on-disk bit is cleared and the |
| 55 | * counter goes to zero. |
| 56 | * |
| 57 | * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block |
| 58 | * counters as a fallback when "page" memory cannot be allocated: |
| 59 | * |
| 60 | * Normal case (page memory allocated): |
| 61 | * |
| 62 | * page pointer (32-bit) |
| 63 | * |
| 64 | * [ ] ------+ |
| 65 | * | |
| 66 | * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters) |
| 67 | * c1 c2 c2048 |
| 68 | * |
| 69 | * Hijacked case (page memory allocation failed): |
| 70 | * |
| 71 | * hijacked page pointer (32-bit) |
| 72 | * |
| 73 | * [ ][ ] (no page memory allocated) |
| 74 | * counter #1 (16-bit) counter #2 (16-bit) |
| 75 | * |
| 76 | */ |
| 77 | |
| 78 | #ifdef __KERNEL__ |
| 79 | |
| 80 | #define PAGE_BITS (PAGE_SIZE << 3) |
| 81 | #define PAGE_BIT_SHIFT (PAGE_SHIFT + 3) |
| 82 | |
| 83 | typedef __u16 bitmap_counter_t; |
| 84 | #define COUNTER_BITS 16 |
| 85 | #define COUNTER_BIT_SHIFT 4 |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 86 | #define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3) |
| 87 | |
| 88 | #define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1))) |
| 89 | #define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2))) |
| 90 | #define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1) |
| 91 | #define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK) |
| 92 | #define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK) |
| 93 | #define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX) |
| 94 | |
| 95 | /* how many counters per page? */ |
| 96 | #define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS) |
| 97 | /* same, except a shift value for more efficient bitops */ |
| 98 | #define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT) |
| 99 | /* same, except a mask value for more efficient bitops */ |
| 100 | #define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1) |
| 101 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 102 | #define BITMAP_BLOCK_SHIFT 9 |
| 103 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 104 | #endif |
| 105 | |
| 106 | /* |
| 107 | * bitmap structures: |
| 108 | */ |
| 109 | |
| 110 | #define BITMAP_MAGIC 0x6d746962 |
| 111 | |
| 112 | /* use these for bitmap->flags and bitmap->sb->state bit-fields */ |
| 113 | enum bitmap_state { |
NeilBrown | b405fe9 | 2012-05-22 13:55:15 +1000 | [diff] [blame] | 114 | BITMAP_STALE = 1, /* the bitmap file is out of date or had -EIO */ |
| 115 | BITMAP_WRITE_ERROR = 2, /* A write error has occurred */ |
| 116 | BITMAP_HOSTENDIAN =15, |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 117 | }; |
| 118 | |
| 119 | /* the superblock at the front of the bitmap file -- little endian */ |
| 120 | typedef struct bitmap_super_s { |
NeilBrown | 4f2e639 | 2006-10-21 10:24:09 -0700 | [diff] [blame] | 121 | __le32 magic; /* 0 BITMAP_MAGIC */ |
| 122 | __le32 version; /* 4 the bitmap major for now, could change... */ |
| 123 | __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */ |
| 124 | __le64 events; /* 24 event counter for the bitmap (1)*/ |
| 125 | __le64 events_cleared;/*32 event counter when last bit cleared (2) */ |
| 126 | __le64 sync_size; /* 40 the size of the md device's sync range(3) */ |
| 127 | __le32 state; /* 48 bitmap state information */ |
| 128 | __le32 chunksize; /* 52 the bitmap chunk size in bytes */ |
| 129 | __le32 daemon_sleep; /* 56 seconds between disk flushes */ |
| 130 | __le32 write_behind; /* 60 number of outstanding write-behind writes */ |
NeilBrown | 1dff2b8 | 2012-05-22 13:55:34 +1000 | [diff] [blame] | 131 | __le32 sectors_reserved; /* 64 number of 512-byte sectors that are |
| 132 | * reserved for the bitmap. */ |
Goldwyn Rodrigues | 183bdf5 | 2014-03-07 10:30:43 -0600 | [diff] [blame] | 133 | __le32 nodes; /* 68 the maximum number of nodes in cluster. */ |
Goldwyn Rodrigues | cf921cc | 2014-03-30 00:42:49 -0500 | [diff] [blame] | 134 | __u8 cluster_name[64]; /* 72 cluster name to which this md belongs */ |
| 135 | __u8 pad[256 - 136]; /* set to zero */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 136 | } bitmap_super_t; |
| 137 | |
| 138 | /* notes: |
| 139 | * (1) This event counter is updated before the eventcounter in the md superblock |
| 140 | * When a bitmap is loaded, it is only accepted if this event counter is equal |
| 141 | * to, or one greater than, the event counter in the superblock. |
| 142 | * (2) This event counter is updated when the other one is *if*and*only*if* the |
| 143 | * array is not degraded. As bits are not cleared when the array is degraded, |
| 144 | * this represents the last time that any bits were cleared. |
| 145 | * If a device is being added that has an event count with this value or |
| 146 | * higher, it is accepted as conforming to the bitmap. |
| 147 | * (3)This is the number of sectors represented by the bitmap, and is the range that |
| 148 | * resync happens across. For raid1 and raid5/6 it is the size of individual |
| 149 | * devices. For raid10 it is the size of the array. |
| 150 | */ |
| 151 | |
| 152 | #ifdef __KERNEL__ |
| 153 | |
| 154 | /* the in-memory bitmap is represented by bitmap_pages */ |
| 155 | struct bitmap_page { |
| 156 | /* |
| 157 | * map points to the actual memory page |
| 158 | */ |
| 159 | char *map; |
| 160 | /* |
| 161 | * in emergencies (when map cannot be alloced), hijack the map |
| 162 | * pointer and use it as two counters itself |
| 163 | */ |
| 164 | unsigned int hijacked:1; |
| 165 | /* |
NeilBrown | bf07bb7 | 2012-05-22 13:55:06 +1000 | [diff] [blame] | 166 | * If any counter in this page is '1' or '2' - and so could be |
| 167 | * cleared then that page is marked as 'pending' |
| 168 | */ |
| 169 | unsigned int pending:1; |
| 170 | /* |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 171 | * count of dirty bits on the page |
| 172 | */ |
NeilBrown | bf07bb7 | 2012-05-22 13:55:06 +1000 | [diff] [blame] | 173 | unsigned int count:30; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 174 | }; |
| 175 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 176 | /* the main bitmap structure - one per mddev */ |
| 177 | struct bitmap { |
NeilBrown | 40cffcc | 2012-05-22 13:55:24 +1000 | [diff] [blame] | 178 | |
| 179 | struct bitmap_counts { |
| 180 | spinlock_t lock; |
| 181 | struct bitmap_page *bp; |
| 182 | unsigned long pages; /* total number of pages |
| 183 | * in the bitmap */ |
| 184 | unsigned long missing_pages; /* number of pages |
| 185 | * not yet allocated */ |
| 186 | unsigned long chunkshift; /* chunksize = 2^chunkshift |
| 187 | * (for bitops) */ |
| 188 | unsigned long chunks; /* Total number of data |
| 189 | * chunks for the array */ |
| 190 | } counts; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 191 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 192 | struct mddev *mddev; /* the md device that the bitmap is for */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 193 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 194 | __u64 events_cleared; |
Neil Brown | a0da84f | 2008-06-28 08:31:22 +1000 | [diff] [blame] | 195 | int need_sync; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 196 | |
NeilBrown | 1ec885c | 2012-05-22 13:55:10 +1000 | [diff] [blame] | 197 | struct bitmap_storage { |
| 198 | struct file *file; /* backing disk file */ |
| 199 | struct page *sb_page; /* cached copy of the bitmap |
| 200 | * file superblock */ |
| 201 | struct page **filemap; /* list of cache pages for |
| 202 | * the file */ |
| 203 | unsigned long *filemap_attr; /* attributes associated |
| 204 | * w/ filemap pages */ |
| 205 | unsigned long file_pages; /* number of pages in the file*/ |
NeilBrown | 9b1215c | 2012-05-22 13:55:11 +1000 | [diff] [blame] | 206 | unsigned long bytes; /* total bytes in the bitmap */ |
NeilBrown | 1ec885c | 2012-05-22 13:55:10 +1000 | [diff] [blame] | 207 | } storage; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 208 | |
| 209 | unsigned long flags; |
| 210 | |
NeilBrown | 8311c29 | 2008-03-04 14:29:30 -0800 | [diff] [blame] | 211 | int allclean; |
| 212 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 213 | atomic_t behind_writes; |
Paul Clements | 696fcd5 | 2010-03-08 16:02:37 +1100 | [diff] [blame] | 214 | unsigned long behind_writes_used; /* highest actual value at runtime */ |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 215 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 216 | /* |
| 217 | * the bitmap daemon - periodically wakes up and sweeps the bitmap |
| 218 | * file, cleaning up bits and flushing out pages to disk as necessary |
| 219 | */ |
| 220 | unsigned long daemon_lastrun; /* jiffies of last run */ |
NeilBrown | b47490c | 2008-02-06 01:39:50 -0800 | [diff] [blame] | 221 | unsigned long last_end_sync; /* when we lasted called end_sync to |
| 222 | * update bitmap with resync progress */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 223 | |
NeilBrown | d785a06 | 2006-06-26 00:27:48 -0700 | [diff] [blame] | 224 | atomic_t pending_writes; /* pending writes to the bitmap file */ |
| 225 | wait_queue_head_t write_wait; |
Neil Brown | da6e1a3 | 2007-02-08 14:20:37 -0800 | [diff] [blame] | 226 | wait_queue_head_t overflow_wait; |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 227 | wait_queue_head_t behind_wait; |
NeilBrown | d785a06 | 2006-06-26 00:27:48 -0700 | [diff] [blame] | 228 | |
Tejun Heo | 324a56e | 2013-12-11 14:11:53 -0500 | [diff] [blame] | 229 | struct kernfs_node *sysfs_can_clear; |
Goldwyn Rodrigues | b97e9257 | 2014-06-06 11:50:56 -0500 | [diff] [blame] | 230 | int cluster_slot; /* Slot offset for clustered env */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 231 | }; |
| 232 | |
| 233 | /* the bitmap API */ |
| 234 | |
| 235 | /* these are used only by md/bitmap */ |
Goldwyn Rodrigues | f9209a3 | 2014-06-06 12:43:49 -0500 | [diff] [blame] | 236 | struct bitmap *bitmap_create(struct mddev *mddev, int slot); |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 237 | int bitmap_load(struct mddev *mddev); |
| 238 | void bitmap_flush(struct mddev *mddev); |
| 239 | void bitmap_destroy(struct mddev *mddev); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 240 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 241 | void bitmap_print_sb(struct bitmap *bitmap); |
NeilBrown | 4ad1366 | 2007-07-17 04:06:13 -0700 | [diff] [blame] | 242 | void bitmap_update_sb(struct bitmap *bitmap); |
NeilBrown | 5714896 | 2012-03-19 12:46:40 +1100 | [diff] [blame] | 243 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 244 | |
| 245 | int bitmap_setallbits(struct bitmap *bitmap); |
NeilBrown | a654b9d8 | 2005-06-21 17:17:27 -0700 | [diff] [blame] | 246 | void bitmap_write_all(struct bitmap *bitmap); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 247 | |
Paul Clements | 9b1d1da | 2006-10-03 01:15:49 -0700 | [diff] [blame] | 248 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e); |
| 249 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 250 | /* these are exported */ |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 251 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, |
| 252 | unsigned long sectors, int behind); |
| 253 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, |
| 254 | unsigned long sectors, int success, int behind); |
NeilBrown | 57dab0b | 2010-10-19 10:03:39 +1100 | [diff] [blame] | 255 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded); |
| 256 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 257 | void bitmap_close_sync(struct bitmap *bitmap); |
NeilBrown | b47490c | 2008-02-06 01:39:50 -0800 | [diff] [blame] | 258 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 259 | |
NeilBrown | 4ad1366 | 2007-07-17 04:06:13 -0700 | [diff] [blame] | 260 | void bitmap_unplug(struct bitmap *bitmap); |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 261 | void bitmap_daemon_work(struct mddev *mddev); |
NeilBrown | d60b479 | 2012-05-22 13:55:25 +1000 | [diff] [blame] | 262 | |
| 263 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
| 264 | int chunksize, int init); |
Goldwyn Rodrigues | 11dd35d | 2014-06-07 00:36:26 -0500 | [diff] [blame] | 265 | int bitmap_copy_from_slot(struct mddev *mddev, int slot, |
Goldwyn Rodrigues | 97f6cd3 | 2015-04-14 10:45:42 -0500 | [diff] [blame] | 266 | sector_t *lo, sector_t *hi, bool clear_bits); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 267 | #endif |
| 268 | |
| 269 | #endif |