]>
Commit | Line | Data |
---|---|---|
c82f047c NB |
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 | ||
dcec9ee5 NB |
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 | |
6d9c7c25 | 15 | #define BITMAP_MAJOR_CLUSTERED 5 |
dcec9ee5 | 16 | |
dfd4d8ee | 17 | #define BITMAP_MINOR 39 |
c82f047c NB |
18 | |
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 | |
49 | * bit can be cleared aswell, thus setting the counter to 0. | |
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 | * | |
dfd4d8ee NB |
53 | * If the counter is 0, the on-disk bit is clear and the stipe is clean |
54 | * Anything that dirties the stipe pushes the counter to 2 (at least) | |
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 | * | |
c82f047c NB |
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 | |
89 | #define COUNTER_BYTE_RATIO (COUNTER_BITS / 8) | |
90 | #define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3) | |
91 | ||
92 | #define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1))) | |
93 | #define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2))) | |
94 | #define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1) | |
95 | #define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK) | |
96 | #define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK) | |
97 | #define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX) | |
98 | ||
99 | /* how many counters per page? */ | |
100 | #define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS) | |
101 | /* same, except a shift value for more efficient bitops */ | |
102 | #define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT) | |
103 | /* same, except a mask value for more efficient bitops */ | |
104 | #define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1) | |
105 | ||
106 | #define BITMAP_BLOCK_SIZE 512 | |
107 | #define BITMAP_BLOCK_SHIFT 9 | |
108 | ||
109 | /* how many blocks per chunk? (this is variable) */ | |
110 | #define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->chunksize >> BITMAP_BLOCK_SHIFT) | |
111 | #define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT) | |
112 | #define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1) | |
113 | ||
114 | /* when hijacked, the counters and bits represent even larger "chunks" */ | |
115 | /* there will be 1024 chunks represented by each counter in the page pointers */ | |
116 | #define PAGEPTR_BLOCK_RATIO(bitmap) \ | |
117 | (CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1) | |
118 | #define PAGEPTR_BLOCK_SHIFT(bitmap) \ | |
119 | (CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1) | |
120 | #define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1) | |
121 | ||
122 | /* | |
123 | * on-disk bitmap: | |
124 | * | |
125 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
126 | * file a page at a time. There's a superblock at the start of the file. | |
127 | */ | |
128 | ||
129 | /* map chunks (bits) to file pages - offset by the size of the superblock */ | |
130 | #define CHUNK_BIT_OFFSET(chunk) ((chunk) + (sizeof(bitmap_super_t) << 3)) | |
131 | ||
132 | #endif | |
133 | ||
134 | /* | |
135 | * bitmap structures: | |
136 | */ | |
137 | ||
138 | #define BITMAP_MAGIC 0x6d746962 | |
139 | ||
140 | /* use these for bitmap->flags and bitmap->sb->state bit-fields */ | |
141 | enum bitmap_state { | |
142 | BITMAP_ACTIVE = 0x001, /* the bitmap is in use */ | |
143 | BITMAP_STALE = 0x002 /* the bitmap file is out of date or had -EIO */ | |
144 | }; | |
145 | ||
146 | /* the superblock at the front of the bitmap file -- little endian */ | |
147 | typedef struct bitmap_super_s { | |
148 | __u32 magic; /* 0 BITMAP_MAGIC */ | |
149 | __u32 version; /* 4 the bitmap major for now, could change... */ | |
150 | __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */ | |
151 | __u64 events; /* 24 event counter for the bitmap (1)*/ | |
152 | __u64 events_cleared;/*32 event counter when last bit cleared (2) */ | |
153 | __u64 sync_size; /* 40 the size of the md device's sync range(3) */ | |
154 | __u32 state; /* 48 bitmap state information */ | |
155 | __u32 chunksize; /* 52 the bitmap chunk size in bytes */ | |
156 | __u32 daemon_sleep; /* 56 seconds between disk flushes */ | |
dfd4d8ee | 157 | __u32 write_behind; /* 60 number of outstanding write-behind writes */ |
95a05b37 GJ |
158 | __u32 sectors_reserved; /* 64 number of 512-byte sectors that are |
159 | * reserved for the bitmap. */ | |
160 | __u32 nodes; /* 68 the maximum number of nodes in cluster. */ | |
161 | __u8 cluster_name[64]; /* 72 cluster name to which this md belongs */ | |
162 | __u8 pad[256 - 136]; /* set to zero */ | |
c82f047c NB |
163 | } bitmap_super_t; |
164 | ||
165 | /* notes: | |
166 | * (1) This event counter is updated before the eventcounter in the md superblock | |
167 | * When a bitmap is loaded, it is only accepted if this event counter is equal | |
168 | * to, or one greater than, the event counter in the superblock. | |
aba69144 | 169 | * (2) This event counter is updated when the other one is *if*and*only*if* the |
c82f047c NB |
170 | * array is not degraded. As bits are not cleared when the array is degraded, |
171 | * this represents the last time that any bits were cleared. | |
172 | * If a device is being added that has an event count with this value or | |
173 | * higher, it is accepted as conforming to the bitmap. | |
174 | * (3)This is the number of sectors represented by the bitmap, and is the range that | |
175 | * resync happens across. For raid1 and raid5/6 it is the size of individual | |
176 | * devices. For raid10 it is the size of the array. | |
177 | */ | |
178 | ||
179 | #ifdef __KERNEL__ | |
180 | ||
181 | /* the in-memory bitmap is represented by bitmap_pages */ | |
182 | struct bitmap_page { | |
183 | /* | |
184 | * map points to the actual memory page | |
185 | */ | |
186 | char *map; | |
187 | /* | |
188 | * in emergencies (when map cannot be alloced), hijack the map | |
189 | * pointer and use it as two counters itself | |
190 | */ | |
191 | unsigned int hijacked; | |
192 | /* | |
193 | * count of dirty bits on the page | |
aba69144 | 194 | */ |
c82f047c NB |
195 | int count; |
196 | }; | |
197 | ||
198 | /* keep track of bitmap file pages that have pending writes on them */ | |
199 | struct page_list { | |
200 | struct list_head list; | |
201 | struct page *page; | |
202 | }; | |
203 | ||
204 | /* the main bitmap structure - one per mddev */ | |
205 | struct bitmap { | |
206 | struct bitmap_page *bp; | |
207 | unsigned long pages; /* total number of pages in the bitmap */ | |
208 | unsigned long missing_pages; /* number of pages not yet allocated */ | |
209 | ||
210 | mddev_t *mddev; /* the md device that the bitmap is for */ | |
211 | ||
212 | int counter_bits; /* how many bits per block counter */ | |
213 | ||
214 | /* bitmap chunksize -- how much data does each bit represent? */ | |
215 | unsigned long chunksize; | |
216 | unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */ | |
217 | unsigned long chunks; /* total number of data chunks for the array */ | |
218 | ||
219 | /* We hold a count on the chunk currently being synced, and drop | |
220 | * it when the last block is started. If the resync is aborted | |
221 | * midway, we need to be able to drop that count, so we remember | |
222 | * the counted chunk.. | |
223 | */ | |
224 | unsigned long syncchunk; | |
225 | ||
226 | __u64 events_cleared; | |
227 | ||
228 | /* bitmap spinlock */ | |
229 | spinlock_t lock; | |
230 | ||
231 | struct file *file; /* backing disk file */ | |
232 | struct page *sb_page; /* cached copy of the bitmap file superblock */ | |
233 | struct page **filemap; /* list of cache pages for the file */ | |
234 | unsigned long *filemap_attr; /* attributes associated w/ filemap pages */ | |
235 | unsigned long file_pages; /* number of pages in the file */ | |
236 | ||
237 | unsigned long flags; | |
238 | ||
239 | /* | |
240 | * the bitmap daemon - periodically wakes up and sweeps the bitmap | |
241 | * file, cleaning up bits and flushing out pages to disk as necessary | |
242 | */ | |
243 | mdk_thread_t *daemon; | |
244 | unsigned long daemon_sleep; /* how many seconds between updates? */ | |
245 | ||
246 | /* | |
247 | * bitmap write daemon - this daemon performs writes to the bitmap file | |
248 | * this thread is only needed because of a limitation in ext3 (jbd) | |
249 | * that does not allow a task to have two journal transactions ongoing | |
250 | * simultaneously (even if the transactions are for two different | |
251 | * filesystems) -- in the case of bitmap, that would be the filesystem | |
252 | * that the bitmap file resides on and the filesystem that is mounted | |
253 | * on the md device -- see current->journal_info in jbd/transaction.c | |
254 | */ | |
255 | mdk_thread_t *write_daemon; | |
256 | mdk_thread_t *writeback_daemon; | |
257 | spinlock_t write_lock; | |
258 | struct semaphore write_ready; | |
259 | struct semaphore write_done; | |
260 | unsigned long writes_pending; | |
261 | wait_queue_head_t write_wait; | |
262 | struct list_head write_pages; | |
263 | struct list_head complete_pages; | |
264 | mempool_t *write_pool; | |
265 | }; | |
266 | ||
267 | /* the bitmap API */ | |
268 | ||
269 | /* these are used only by md/bitmap */ | |
270 | int bitmap_create(mddev_t *mddev); | |
271 | void bitmap_destroy(mddev_t *mddev); | |
272 | int bitmap_active(struct bitmap *bitmap); | |
273 | ||
274 | char *file_path(struct file *file, char *buf, int count); | |
275 | void bitmap_print_sb(struct bitmap *bitmap); | |
276 | int bitmap_update_sb(struct bitmap *bitmap); | |
277 | ||
278 | int bitmap_setallbits(struct bitmap *bitmap); | |
279 | ||
280 | /* these are exported */ | |
281 | void bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors); | |
282 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
283 | int success); | |
284 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks); | |
285 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted); | |
286 | void bitmap_close_sync(struct bitmap *bitmap); | |
287 | ||
288 | int bitmap_unplug(struct bitmap *bitmap); | |
289 | #endif | |
290 | ||
291 | #endif |