Commit | Line | Data |
---|---|---|
32a7627c N |
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 | ||
bd926c63 N |
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 | ||
4b6d287f | 16 | #define BITMAP_MINOR 39 |
32a7627c N |
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 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 | ||
85 | typedef __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) */ | |
42a04b50 | 109 | #define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->mddev->bitmap_info.chunksize >> BITMAP_BLOCK_SHIFT) |
32a7627c N |
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 */ | |
140 | enum bitmap_state { | |
bd926c63 | 141 | BITMAP_STALE = 0x002, /* the bitmap file is out of date or had -EIO */ |
d785a06a | 142 | BITMAP_WRITE_ERROR = 0x004, /* A write error has occurred */ |
bd926c63 | 143 | BITMAP_HOSTENDIAN = 0x8000, |
32a7627c N |
144 | }; |
145 | ||
146 | /* the superblock at the front of the bitmap file -- little endian */ | |
147 | typedef struct bitmap_super_s { | |
4f2e639a N |
148 | __le32 magic; /* 0 BITMAP_MAGIC */ |
149 | __le32 version; /* 4 the bitmap major for now, could change... */ | |
150 | __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */ | |
151 | __le64 events; /* 24 event counter for the bitmap (1)*/ | |
152 | __le64 events_cleared;/*32 event counter when last bit cleared (2) */ | |
153 | __le64 sync_size; /* 40 the size of the md device's sync range(3) */ | |
154 | __le32 state; /* 48 bitmap state information */ | |
155 | __le32 chunksize; /* 52 the bitmap chunk size in bytes */ | |
156 | __le32 daemon_sleep; /* 56 seconds between disk flushes */ | |
157 | __le32 write_behind; /* 60 number of outstanding write-behind writes */ | |
32a7627c | 158 | |
4b6d287f | 159 | __u8 pad[256 - 64]; /* set to zero */ |
32a7627c N |
160 | } bitmap_super_t; |
161 | ||
162 | /* notes: | |
163 | * (1) This event counter is updated before the eventcounter in the md superblock | |
164 | * When a bitmap is loaded, it is only accepted if this event counter is equal | |
165 | * to, or one greater than, the event counter in the superblock. | |
166 | * (2) This event counter is updated when the other one is *if*and*only*if* the | |
167 | * array is not degraded. As bits are not cleared when the array is degraded, | |
168 | * this represents the last time that any bits were cleared. | |
169 | * If a device is being added that has an event count with this value or | |
170 | * higher, it is accepted as conforming to the bitmap. | |
171 | * (3)This is the number of sectors represented by the bitmap, and is the range that | |
172 | * resync happens across. For raid1 and raid5/6 it is the size of individual | |
173 | * devices. For raid10 it is the size of the array. | |
174 | */ | |
175 | ||
176 | #ifdef __KERNEL__ | |
177 | ||
178 | /* the in-memory bitmap is represented by bitmap_pages */ | |
179 | struct bitmap_page { | |
180 | /* | |
181 | * map points to the actual memory page | |
182 | */ | |
183 | char *map; | |
184 | /* | |
185 | * in emergencies (when map cannot be alloced), hijack the map | |
186 | * pointer and use it as two counters itself | |
187 | */ | |
188 | unsigned int hijacked:1; | |
189 | /* | |
190 | * count of dirty bits on the page | |
191 | */ | |
192 | unsigned int count:31; | |
193 | }; | |
194 | ||
195 | /* keep track of bitmap file pages that have pending writes on them */ | |
196 | struct page_list { | |
197 | struct list_head list; | |
198 | struct page *page; | |
199 | }; | |
200 | ||
201 | /* the main bitmap structure - one per mddev */ | |
202 | struct bitmap { | |
203 | struct bitmap_page *bp; | |
204 | unsigned long pages; /* total number of pages in the bitmap */ | |
205 | unsigned long missing_pages; /* number of pages not yet allocated */ | |
206 | ||
207 | mddev_t *mddev; /* the md device that the bitmap is for */ | |
208 | ||
209 | int counter_bits; /* how many bits per block counter */ | |
210 | ||
211 | /* bitmap chunksize -- how much data does each bit represent? */ | |
32a7627c N |
212 | unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */ |
213 | unsigned long chunks; /* total number of data chunks for the array */ | |
214 | ||
215 | /* We hold a count on the chunk currently being synced, and drop | |
216 | * it when the last block is started. If the resync is aborted | |
217 | * midway, we need to be able to drop that count, so we remember | |
218 | * the counted chunk.. | |
219 | */ | |
220 | unsigned long syncchunk; | |
221 | ||
222 | __u64 events_cleared; | |
a0da84f3 | 223 | int need_sync; |
32a7627c N |
224 | |
225 | /* bitmap spinlock */ | |
226 | spinlock_t lock; | |
227 | ||
228 | struct file *file; /* backing disk file */ | |
229 | struct page *sb_page; /* cached copy of the bitmap file superblock */ | |
230 | struct page **filemap; /* list of cache pages for the file */ | |
231 | unsigned long *filemap_attr; /* attributes associated w/ filemap pages */ | |
232 | unsigned long file_pages; /* number of pages in the file */ | |
ab6085c7 | 233 | int last_page_size; /* bytes in the last page */ |
32a7627c N |
234 | |
235 | unsigned long flags; | |
236 | ||
8311c29d N |
237 | int allclean; |
238 | ||
4b6d287f N |
239 | atomic_t behind_writes; |
240 | ||
32a7627c N |
241 | /* |
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 */ | |
b47490c9 N |
246 | unsigned long last_end_sync; /* when we lasted called end_sync to |
247 | * update bitmap with resync progress */ | |
32a7627c | 248 | |
d785a06a N |
249 | atomic_t pending_writes; /* pending writes to the bitmap file */ |
250 | wait_queue_head_t write_wait; | |
da6e1a32 | 251 | wait_queue_head_t overflow_wait; |
d785a06a | 252 | |
32a7627c N |
253 | }; |
254 | ||
255 | /* the bitmap API */ | |
256 | ||
257 | /* these are used only by md/bitmap */ | |
258 | int bitmap_create(mddev_t *mddev); | |
6b8b3e8a | 259 | void bitmap_flush(mddev_t *mddev); |
32a7627c | 260 | void bitmap_destroy(mddev_t *mddev); |
32a7627c | 261 | |
32a7627c | 262 | void bitmap_print_sb(struct bitmap *bitmap); |
4ad13663 | 263 | void bitmap_update_sb(struct bitmap *bitmap); |
32a7627c N |
264 | |
265 | int bitmap_setallbits(struct bitmap *bitmap); | |
a654b9d8 | 266 | void bitmap_write_all(struct bitmap *bitmap); |
32a7627c | 267 | |
9b1d1dac PC |
268 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e); |
269 | ||
32a7627c | 270 | /* these are exported */ |
4b6d287f N |
271 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, |
272 | unsigned long sectors, int behind); | |
273 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, | |
274 | unsigned long sectors, int success, int behind); | |
6a806c51 | 275 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int degraded); |
32a7627c N |
276 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted); |
277 | void bitmap_close_sync(struct bitmap *bitmap); | |
b47490c9 | 278 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector); |
32a7627c | 279 | |
4ad13663 | 280 | void bitmap_unplug(struct bitmap *bitmap); |
aa5cbd10 | 281 | void bitmap_daemon_work(mddev_t *mddev); |
32a7627c N |
282 | #endif |
283 | ||
284 | #endif |