Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 2 | /* |
| 3 | * bitmap.h: Copyright (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 |
| 4 | * |
| 5 | * additions: Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc. |
| 6 | */ |
| 7 | #ifndef BITMAP_H |
| 8 | #define BITMAP_H 1 |
| 9 | |
NeilBrown | bd926c6 | 2005-11-08 21:39:32 -0800 | [diff] [blame] | 10 | #define BITMAP_MAJOR_LO 3 |
| 11 | /* version 4 insists the bitmap is in little-endian order |
| 12 | * with version 3, it is host-endian which is non-portable |
Goldwyn Rodrigues | 3c462c8 | 2015-08-19 07:35:54 +1000 | [diff] [blame] | 13 | * Version 5 is currently set only for clustered devices |
NeilBrown | bd926c6 | 2005-11-08 21:39:32 -0800 | [diff] [blame] | 14 | */ |
| 15 | #define BITMAP_MAJOR_HI 4 |
Goldwyn Rodrigues | 3c462c8 | 2015-08-19 07:35:54 +1000 | [diff] [blame] | 16 | #define BITMAP_MAJOR_CLUSTERED 5 |
NeilBrown | bd926c6 | 2005-11-08 21:39:32 -0800 | [diff] [blame] | 17 | #define BITMAP_MAJOR_HOSTENDIAN 3 |
| 18 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 19 | /* |
| 20 | * in-memory bitmap: |
| 21 | * |
| 22 | * Use 16 bit block counters to track pending writes to each "chunk". |
| 23 | * The 2 high order bits are special-purpose, the first is a flag indicating |
| 24 | * whether a resync is needed. The second is a flag indicating whether a |
| 25 | * resync is active. |
| 26 | * This means that the counter is actually 14 bits: |
| 27 | * |
| 28 | * +--------+--------+------------------------------------------------+ |
| 29 | * | resync | resync | counter | |
| 30 | * | needed | active | | |
| 31 | * | (0-1) | (0-1) | (0-16383) | |
| 32 | * +--------+--------+------------------------------------------------+ |
| 33 | * |
| 34 | * The "resync needed" bit is set when: |
| 35 | * a '1' bit is read from storage at startup. |
| 36 | * a write request fails on some drives |
| 37 | * a resync is aborted on a chunk with 'resync active' set |
| 38 | * It is cleared (and resync-active set) when a resync starts across all drives |
| 39 | * of the chunk. |
| 40 | * |
| 41 | * |
| 42 | * The "resync active" bit is set when: |
| 43 | * a resync is started on all drives, and resync_needed is set. |
| 44 | * resync_needed will be cleared (as long as resync_active wasn't already set). |
| 45 | * It is cleared when a resync completes. |
| 46 | * |
| 47 | * The counter counts pending write requests, plus the on-disk bit. |
| 48 | * 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] | 49 | * bit can be cleared as well, thus setting the counter to 0. |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 50 | * When we set a bit, or in the counter (to start a write), if the fields is |
| 51 | * 0, we first set the disk bit and set the counter to 1. |
| 52 | * |
Guoqing Jiang | d85326c | 2016-03-14 17:01:39 +0800 | [diff] [blame] | 53 | * If the counter is 0, the on-disk bit is clear and the stripe is clean |
| 54 | * Anything that dirties the stripe pushes the counter to 2 (at least) |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 55 | * and sets the on-disk bit (lazily). |
| 56 | * If a periodic sweep find the counter at 2, it is decremented to 1. |
| 57 | * If the sweep find the counter at 1, the on-disk bit is cleared and the |
| 58 | * counter goes to zero. |
| 59 | * |
| 60 | * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block |
| 61 | * counters as a fallback when "page" memory cannot be allocated: |
| 62 | * |
| 63 | * Normal case (page memory allocated): |
| 64 | * |
| 65 | * page pointer (32-bit) |
| 66 | * |
| 67 | * [ ] ------+ |
| 68 | * | |
| 69 | * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters) |
| 70 | * c1 c2 c2048 |
| 71 | * |
| 72 | * Hijacked case (page memory allocation failed): |
| 73 | * |
| 74 | * hijacked page pointer (32-bit) |
| 75 | * |
| 76 | * [ ][ ] (no page memory allocated) |
| 77 | * counter #1 (16-bit) counter #2 (16-bit) |
| 78 | * |
| 79 | */ |
| 80 | |
| 81 | #ifdef __KERNEL__ |
| 82 | |
| 83 | #define PAGE_BITS (PAGE_SIZE << 3) |
| 84 | #define PAGE_BIT_SHIFT (PAGE_SHIFT + 3) |
| 85 | |
| 86 | typedef __u16 bitmap_counter_t; |
| 87 | #define COUNTER_BITS 16 |
| 88 | #define COUNTER_BIT_SHIFT 4 |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 89 | #define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3) |
| 90 | |
| 91 | #define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1))) |
| 92 | #define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2))) |
| 93 | #define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1) |
| 94 | #define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK) |
| 95 | #define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK) |
| 96 | #define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX) |
| 97 | |
| 98 | /* how many counters per page? */ |
| 99 | #define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS) |
| 100 | /* same, except a shift value for more efficient bitops */ |
| 101 | #define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT) |
| 102 | /* same, except a mask value for more efficient bitops */ |
| 103 | #define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1) |
| 104 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 105 | #define BITMAP_BLOCK_SHIFT 9 |
| 106 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 107 | #endif |
| 108 | |
| 109 | /* |
| 110 | * bitmap structures: |
| 111 | */ |
| 112 | |
| 113 | #define BITMAP_MAGIC 0x6d746962 |
| 114 | |
| 115 | /* use these for bitmap->flags and bitmap->sb->state bit-fields */ |
| 116 | enum bitmap_state { |
NeilBrown | b405fe9 | 2012-05-22 13:55:15 +1000 | [diff] [blame] | 117 | BITMAP_STALE = 1, /* the bitmap file is out of date or had -EIO */ |
| 118 | BITMAP_WRITE_ERROR = 2, /* A write error has occurred */ |
| 119 | BITMAP_HOSTENDIAN =15, |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 120 | }; |
| 121 | |
| 122 | /* the superblock at the front of the bitmap file -- little endian */ |
| 123 | typedef struct bitmap_super_s { |
NeilBrown | 4f2e639 | 2006-10-21 10:24:09 -0700 | [diff] [blame] | 124 | __le32 magic; /* 0 BITMAP_MAGIC */ |
| 125 | __le32 version; /* 4 the bitmap major for now, could change... */ |
| 126 | __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */ |
| 127 | __le64 events; /* 24 event counter for the bitmap (1)*/ |
| 128 | __le64 events_cleared;/*32 event counter when last bit cleared (2) */ |
| 129 | __le64 sync_size; /* 40 the size of the md device's sync range(3) */ |
| 130 | __le32 state; /* 48 bitmap state information */ |
| 131 | __le32 chunksize; /* 52 the bitmap chunk size in bytes */ |
| 132 | __le32 daemon_sleep; /* 56 seconds between disk flushes */ |
| 133 | __le32 write_behind; /* 60 number of outstanding write-behind writes */ |
NeilBrown | 1dff2b8 | 2012-05-22 13:55:34 +1000 | [diff] [blame] | 134 | __le32 sectors_reserved; /* 64 number of 512-byte sectors that are |
| 135 | * reserved for the bitmap. */ |
Goldwyn Rodrigues | 183bdf5 | 2014-03-07 10:30:43 -0600 | [diff] [blame] | 136 | __le32 nodes; /* 68 the maximum number of nodes in cluster. */ |
Goldwyn Rodrigues | cf921cc | 2014-03-30 00:42:49 -0500 | [diff] [blame] | 137 | __u8 cluster_name[64]; /* 72 cluster name to which this md belongs */ |
| 138 | __u8 pad[256 - 136]; /* set to zero */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 139 | } bitmap_super_t; |
| 140 | |
| 141 | /* notes: |
| 142 | * (1) This event counter is updated before the eventcounter in the md superblock |
| 143 | * When a bitmap is loaded, it is only accepted if this event counter is equal |
| 144 | * to, or one greater than, the event counter in the superblock. |
| 145 | * (2) This event counter is updated when the other one is *if*and*only*if* the |
| 146 | * array is not degraded. As bits are not cleared when the array is degraded, |
| 147 | * this represents the last time that any bits were cleared. |
| 148 | * If a device is being added that has an event count with this value or |
| 149 | * higher, it is accepted as conforming to the bitmap. |
| 150 | * (3)This is the number of sectors represented by the bitmap, and is the range that |
| 151 | * resync happens across. For raid1 and raid5/6 it is the size of individual |
| 152 | * devices. For raid10 it is the size of the array. |
| 153 | */ |
| 154 | |
| 155 | #ifdef __KERNEL__ |
| 156 | |
| 157 | /* the in-memory bitmap is represented by bitmap_pages */ |
| 158 | struct bitmap_page { |
| 159 | /* |
| 160 | * map points to the actual memory page |
| 161 | */ |
| 162 | char *map; |
| 163 | /* |
| 164 | * in emergencies (when map cannot be alloced), hijack the map |
| 165 | * pointer and use it as two counters itself |
| 166 | */ |
| 167 | unsigned int hijacked:1; |
| 168 | /* |
NeilBrown | bf07bb7 | 2012-05-22 13:55:06 +1000 | [diff] [blame] | 169 | * If any counter in this page is '1' or '2' - and so could be |
| 170 | * cleared then that page is marked as 'pending' |
| 171 | */ |
| 172 | unsigned int pending:1; |
| 173 | /* |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 174 | * count of dirty bits on the page |
| 175 | */ |
NeilBrown | bf07bb7 | 2012-05-22 13:55:06 +1000 | [diff] [blame] | 176 | unsigned int count:30; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 177 | }; |
| 178 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 179 | /* the main bitmap structure - one per mddev */ |
| 180 | struct bitmap { |
NeilBrown | 40cffcc | 2012-05-22 13:55:24 +1000 | [diff] [blame] | 181 | |
| 182 | struct bitmap_counts { |
| 183 | spinlock_t lock; |
| 184 | struct bitmap_page *bp; |
| 185 | unsigned long pages; /* total number of pages |
| 186 | * in the bitmap */ |
| 187 | unsigned long missing_pages; /* number of pages |
| 188 | * not yet allocated */ |
| 189 | unsigned long chunkshift; /* chunksize = 2^chunkshift |
| 190 | * (for bitops) */ |
| 191 | unsigned long chunks; /* Total number of data |
| 192 | * chunks for the array */ |
| 193 | } counts; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 194 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 195 | struct mddev *mddev; /* the md device that the bitmap is for */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 196 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 197 | __u64 events_cleared; |
Neil Brown | a0da84f | 2008-06-28 08:31:22 +1000 | [diff] [blame] | 198 | int need_sync; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 199 | |
NeilBrown | 1ec885c | 2012-05-22 13:55:10 +1000 | [diff] [blame] | 200 | struct bitmap_storage { |
| 201 | struct file *file; /* backing disk file */ |
| 202 | struct page *sb_page; /* cached copy of the bitmap |
| 203 | * file superblock */ |
| 204 | struct page **filemap; /* list of cache pages for |
| 205 | * the file */ |
| 206 | unsigned long *filemap_attr; /* attributes associated |
| 207 | * w/ filemap pages */ |
| 208 | unsigned long file_pages; /* number of pages in the file*/ |
NeilBrown | 9b1215c | 2012-05-22 13:55:11 +1000 | [diff] [blame] | 209 | unsigned long bytes; /* total bytes in the bitmap */ |
NeilBrown | 1ec885c | 2012-05-22 13:55:10 +1000 | [diff] [blame] | 210 | } storage; |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 211 | |
| 212 | unsigned long flags; |
| 213 | |
NeilBrown | 8311c29 | 2008-03-04 14:29:30 -0800 | [diff] [blame] | 214 | int allclean; |
| 215 | |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 216 | atomic_t behind_writes; |
Paul Clements | 696fcd5 | 2010-03-08 16:02:37 +1100 | [diff] [blame] | 217 | unsigned long behind_writes_used; /* highest actual value at runtime */ |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 218 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 219 | /* |
| 220 | * the bitmap daemon - periodically wakes up and sweeps the bitmap |
| 221 | * file, cleaning up bits and flushing out pages to disk as necessary |
| 222 | */ |
| 223 | unsigned long daemon_lastrun; /* jiffies of last run */ |
NeilBrown | b47490c | 2008-02-06 01:39:50 -0800 | [diff] [blame] | 224 | unsigned long last_end_sync; /* when we lasted called end_sync to |
| 225 | * update bitmap with resync progress */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 226 | |
NeilBrown | d785a06 | 2006-06-26 00:27:48 -0700 | [diff] [blame] | 227 | atomic_t pending_writes; /* pending writes to the bitmap file */ |
| 228 | wait_queue_head_t write_wait; |
Neil Brown | da6e1a3 | 2007-02-08 14:20:37 -0800 | [diff] [blame] | 229 | wait_queue_head_t overflow_wait; |
NeilBrown | e555190 | 2010-03-31 11:21:44 +1100 | [diff] [blame] | 230 | wait_queue_head_t behind_wait; |
NeilBrown | d785a06 | 2006-06-26 00:27:48 -0700 | [diff] [blame] | 231 | |
Tejun Heo | 324a56e | 2013-12-11 14:11:53 -0500 | [diff] [blame] | 232 | struct kernfs_node *sysfs_can_clear; |
Goldwyn Rodrigues | b97e9257 | 2014-06-06 11:50:56 -0500 | [diff] [blame] | 233 | int cluster_slot; /* Slot offset for clustered env */ |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 234 | }; |
| 235 | |
| 236 | /* the bitmap API */ |
| 237 | |
| 238 | /* these are used only by md/bitmap */ |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 239 | struct bitmap *md_bitmap_create(struct mddev *mddev, int slot); |
| 240 | int md_bitmap_load(struct mddev *mddev); |
| 241 | void md_bitmap_flush(struct mddev *mddev); |
| 242 | void md_bitmap_destroy(struct mddev *mddev); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 243 | |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 244 | void md_bitmap_print_sb(struct bitmap *bitmap); |
| 245 | void md_bitmap_update_sb(struct bitmap *bitmap); |
| 246 | void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 247 | |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 248 | int md_bitmap_setallbits(struct bitmap *bitmap); |
| 249 | void md_bitmap_write_all(struct bitmap *bitmap); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 250 | |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 251 | void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e); |
Paul Clements | 9b1d1da | 2006-10-03 01:15:49 -0700 | [diff] [blame] | 252 | |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 253 | /* these are exported */ |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 254 | int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, |
| 255 | unsigned long sectors, int behind); |
| 256 | void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset, |
NeilBrown | 4b6d287 | 2005-09-09 16:23:47 -0700 | [diff] [blame] | 257 | unsigned long sectors, int success, int behind); |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 258 | int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded); |
| 259 | void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted); |
| 260 | void md_bitmap_close_sync(struct bitmap *bitmap); |
| 261 | void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force); |
| 262 | void md_bitmap_sync_with_cluster(struct mddev *mddev, |
| 263 | sector_t old_lo, sector_t old_hi, |
| 264 | sector_t new_lo, sector_t new_hi); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 265 | |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 266 | void md_bitmap_unplug(struct bitmap *bitmap); |
| 267 | void md_bitmap_daemon_work(struct mddev *mddev); |
NeilBrown | d60b479 | 2012-05-22 13:55:25 +1000 | [diff] [blame] | 268 | |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 269 | int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
| 270 | int chunksize, int init); |
Guoqing Jiang | b98938d | 2017-03-01 16:42:39 +0800 | [diff] [blame] | 271 | struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot); |
Andy Shevchenko | e64e4018 | 2018-08-01 15:20:50 -0700 | [diff] [blame] | 272 | int md_bitmap_copy_from_slot(struct mddev *mddev, int slot, |
| 273 | sector_t *lo, sector_t *hi, bool clear_bits); |
| 274 | void md_bitmap_free(struct bitmap *bitmap); |
| 275 | void md_bitmap_wait_behind_writes(struct mddev *mddev); |
NeilBrown | 32a7627 | 2005-06-21 17:17:14 -0700 | [diff] [blame] | 276 | #endif |
| 277 | |
| 278 | #endif |