blob: 6db9a4c153556a7236caf4dbf376b27900af3f15 [file] [log] [blame]
NeilBrown32a76272005-06-21 17:17:14 -07001/*
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
NeilBrownbd926c62005-11-08 21:39:32 -08009#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
NeilBrown4b6d2872005-09-09 16:23:47 -070016#define BITMAP_MINOR 39
NeilBrown32a76272005-06-21 17:17:14 -070017
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 aswell, 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 stipe is clean
53 * Anything that dirties the stipe 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
85typedef __u16 bitmap_counter_t;
86#define COUNTER_BITS 16
87#define COUNTER_BIT_SHIFT 4
88#define COUNTER_BYTE_RATIO (COUNTER_BITS / 8)
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
105#define BITMAP_BLOCK_SIZE 512
106#define BITMAP_BLOCK_SHIFT 9
107
108/* how many blocks per chunk? (this is variable) */
109#define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->chunksize >> BITMAP_BLOCK_SHIFT)
110#define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT)
111#define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1)
112
113/* when hijacked, the counters and bits represent even larger "chunks" */
114/* there will be 1024 chunks represented by each counter in the page pointers */
115#define PAGEPTR_BLOCK_RATIO(bitmap) \
116 (CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1)
117#define PAGEPTR_BLOCK_SHIFT(bitmap) \
118 (CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1)
119#define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1)
120
121/*
122 * on-disk bitmap:
123 *
124 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
125 * file a page at a time. There's a superblock at the start of the file.
126 */
127
128/* map chunks (bits) to file pages - offset by the size of the superblock */
129#define CHUNK_BIT_OFFSET(chunk) ((chunk) + (sizeof(bitmap_super_t) << 3))
130
131#endif
132
133/*
134 * bitmap structures:
135 */
136
137#define BITMAP_MAGIC 0x6d746962
138
139/* use these for bitmap->flags and bitmap->sb->state bit-fields */
140enum bitmap_state {
141 BITMAP_ACTIVE = 0x001, /* the bitmap is in use */
NeilBrownbd926c62005-11-08 21:39:32 -0800142 BITMAP_STALE = 0x002, /* the bitmap file is out of date or had -EIO */
NeilBrownd785a062006-06-26 00:27:48 -0700143 BITMAP_WRITE_ERROR = 0x004, /* A write error has occurred */
NeilBrownbd926c62005-11-08 21:39:32 -0800144 BITMAP_HOSTENDIAN = 0x8000,
NeilBrown32a76272005-06-21 17:17:14 -0700145};
146
147/* the superblock at the front of the bitmap file -- little endian */
148typedef struct bitmap_super_s {
NeilBrown4f2e6392006-10-21 10:24:09 -0700149 __le32 magic; /* 0 BITMAP_MAGIC */
150 __le32 version; /* 4 the bitmap major for now, could change... */
151 __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */
152 __le64 events; /* 24 event counter for the bitmap (1)*/
153 __le64 events_cleared;/*32 event counter when last bit cleared (2) */
154 __le64 sync_size; /* 40 the size of the md device's sync range(3) */
155 __le32 state; /* 48 bitmap state information */
156 __le32 chunksize; /* 52 the bitmap chunk size in bytes */
157 __le32 daemon_sleep; /* 56 seconds between disk flushes */
158 __le32 write_behind; /* 60 number of outstanding write-behind writes */
NeilBrown32a76272005-06-21 17:17:14 -0700159
NeilBrown4b6d2872005-09-09 16:23:47 -0700160 __u8 pad[256 - 64]; /* set to zero */
NeilBrown32a76272005-06-21 17:17:14 -0700161} bitmap_super_t;
162
163/* notes:
164 * (1) This event counter is updated before the eventcounter in the md superblock
165 * When a bitmap is loaded, it is only accepted if this event counter is equal
166 * to, or one greater than, the event counter in the superblock.
167 * (2) This event counter is updated when the other one is *if*and*only*if* the
168 * array is not degraded. As bits are not cleared when the array is degraded,
169 * this represents the last time that any bits were cleared.
170 * If a device is being added that has an event count with this value or
171 * higher, it is accepted as conforming to the bitmap.
172 * (3)This is the number of sectors represented by the bitmap, and is the range that
173 * resync happens across. For raid1 and raid5/6 it is the size of individual
174 * devices. For raid10 it is the size of the array.
175 */
176
177#ifdef __KERNEL__
178
179/* the in-memory bitmap is represented by bitmap_pages */
180struct bitmap_page {
181 /*
182 * map points to the actual memory page
183 */
184 char *map;
185 /*
186 * in emergencies (when map cannot be alloced), hijack the map
187 * pointer and use it as two counters itself
188 */
189 unsigned int hijacked:1;
190 /*
191 * count of dirty bits on the page
192 */
193 unsigned int count:31;
194};
195
196/* keep track of bitmap file pages that have pending writes on them */
197struct page_list {
198 struct list_head list;
199 struct page *page;
200};
201
202/* the main bitmap structure - one per mddev */
203struct bitmap {
204 struct bitmap_page *bp;
205 unsigned long pages; /* total number of pages in the bitmap */
206 unsigned long missing_pages; /* number of pages not yet allocated */
207
208 mddev_t *mddev; /* the md device that the bitmap is for */
209
210 int counter_bits; /* how many bits per block counter */
211
212 /* bitmap chunksize -- how much data does each bit represent? */
213 unsigned long chunksize;
214 unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */
215 unsigned long chunks; /* total number of data chunks for the array */
216
217 /* We hold a count on the chunk currently being synced, and drop
218 * it when the last block is started. If the resync is aborted
219 * midway, we need to be able to drop that count, so we remember
220 * the counted chunk..
221 */
222 unsigned long syncchunk;
223
224 __u64 events_cleared;
225
226 /* bitmap spinlock */
227 spinlock_t lock;
228
NeilBrowna654b9d82005-06-21 17:17:27 -0700229 long offset; /* offset from superblock if file is NULL */
NeilBrown32a76272005-06-21 17:17:14 -0700230 struct file *file; /* backing disk file */
231 struct page *sb_page; /* cached copy of the bitmap file superblock */
232 struct page **filemap; /* list of cache pages for the file */
233 unsigned long *filemap_attr; /* attributes associated w/ filemap pages */
234 unsigned long file_pages; /* number of pages in the file */
235
236 unsigned long flags;
237
NeilBrown4b6d2872005-09-09 16:23:47 -0700238 unsigned long max_write_behind; /* write-behind mode */
239 atomic_t behind_writes;
240
NeilBrown32a76272005-06-21 17:17:14 -0700241 /*
242 * the bitmap daemon - periodically wakes up and sweeps the bitmap
243 * file, cleaning up bits and flushing out pages to disk as necessary
244 */
245 unsigned long daemon_lastrun; /* jiffies of last run */
246 unsigned long daemon_sleep; /* how many seconds between updates? */
247
NeilBrownd785a062006-06-26 00:27:48 -0700248 atomic_t pending_writes; /* pending writes to the bitmap file */
249 wait_queue_head_t write_wait;
Neil Brownda6e1a32007-02-08 14:20:37 -0800250 wait_queue_head_t overflow_wait;
NeilBrownd785a062006-06-26 00:27:48 -0700251
NeilBrown32a76272005-06-21 17:17:14 -0700252};
253
254/* the bitmap API */
255
256/* these are used only by md/bitmap */
257int bitmap_create(mddev_t *mddev);
NeilBrown6b8b3e82005-08-04 12:53:35 -0700258void bitmap_flush(mddev_t *mddev);
NeilBrown32a76272005-06-21 17:17:14 -0700259void bitmap_destroy(mddev_t *mddev);
260int bitmap_active(struct bitmap *bitmap);
261
262char *file_path(struct file *file, char *buf, int count);
263void bitmap_print_sb(struct bitmap *bitmap);
264int bitmap_update_sb(struct bitmap *bitmap);
265
266int bitmap_setallbits(struct bitmap *bitmap);
NeilBrowna654b9d82005-06-21 17:17:27 -0700267void bitmap_write_all(struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700268
Paul Clements9b1d1da2006-10-03 01:15:49 -0700269void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
270
NeilBrown32a76272005-06-21 17:17:14 -0700271/* these are exported */
NeilBrown4b6d2872005-09-09 16:23:47 -0700272int bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
273 unsigned long sectors, int behind);
274void bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
275 unsigned long sectors, int success, int behind);
NeilBrown6a806c52005-07-15 03:56:35 -0700276int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int degraded);
NeilBrown32a76272005-06-21 17:17:14 -0700277void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted);
278void bitmap_close_sync(struct bitmap *bitmap);
279
280int bitmap_unplug(struct bitmap *bitmap);
281int bitmap_daemon_work(struct bitmap *bitmap);
282#endif
283
284#endif