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 | ||
32a7627c N |
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 | |
25985edc | 46 | * bit can be cleared as well, thus setting the counter to 0. |
32a7627c N |
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 | |
32a7627c N |
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 | ||
32a7627c N |
102 | #define BITMAP_BLOCK_SHIFT 9 |
103 | ||
32a7627c N |
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 { | |
b405fe91 N |
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, | |
32a7627c N |
117 | }; |
118 | ||
119 | /* the superblock at the front of the bitmap file -- little endian */ | |
120 | typedef struct bitmap_super_s { | |
4f2e639a N |
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 */ | |
1dff2b87 N |
131 | __le32 sectors_reserved; /* 64 number of 512-byte sectors that are |
132 | * reserved for the bitmap. */ | |
32a7627c | 133 | |
1dff2b87 | 134 | __u8 pad[256 - 68]; /* set to zero */ |
32a7627c N |
135 | } bitmap_super_t; |
136 | ||
137 | /* notes: | |
138 | * (1) This event counter is updated before the eventcounter in the md superblock | |
139 | * When a bitmap is loaded, it is only accepted if this event counter is equal | |
140 | * to, or one greater than, the event counter in the superblock. | |
141 | * (2) This event counter is updated when the other one is *if*and*only*if* the | |
142 | * array is not degraded. As bits are not cleared when the array is degraded, | |
143 | * this represents the last time that any bits were cleared. | |
144 | * If a device is being added that has an event count with this value or | |
145 | * higher, it is accepted as conforming to the bitmap. | |
146 | * (3)This is the number of sectors represented by the bitmap, and is the range that | |
147 | * resync happens across. For raid1 and raid5/6 it is the size of individual | |
148 | * devices. For raid10 it is the size of the array. | |
149 | */ | |
150 | ||
151 | #ifdef __KERNEL__ | |
152 | ||
153 | /* the in-memory bitmap is represented by bitmap_pages */ | |
154 | struct bitmap_page { | |
155 | /* | |
156 | * map points to the actual memory page | |
157 | */ | |
158 | char *map; | |
159 | /* | |
160 | * in emergencies (when map cannot be alloced), hijack the map | |
161 | * pointer and use it as two counters itself | |
162 | */ | |
163 | unsigned int hijacked:1; | |
bf07bb7d N |
164 | /* |
165 | * If any counter in this page is '1' or '2' - and so could be | |
166 | * cleared then that page is marked as 'pending' | |
167 | */ | |
168 | unsigned int pending:1; | |
32a7627c N |
169 | /* |
170 | * count of dirty bits on the page | |
171 | */ | |
bf07bb7d | 172 | unsigned int count:30; |
32a7627c N |
173 | }; |
174 | ||
32a7627c N |
175 | /* the main bitmap structure - one per mddev */ |
176 | struct bitmap { | |
32a7627c | 177 | |
40cffcc0 N |
178 | struct bitmap_counts { |
179 | spinlock_t lock; | |
180 | struct bitmap_page *bp; | |
181 | unsigned long pages; /* total number of pages | |
182 | * in the bitmap */ | |
183 | unsigned long missing_pages; /* number of pages | |
184 | * not yet allocated */ | |
185 | unsigned long chunkshift; /* chunksize = 2^chunkshift | |
186 | * (for bitops) */ | |
187 | unsigned long chunks; /* Total number of data | |
188 | * chunks for the array */ | |
189 | } counts; | |
32a7627c | 190 | |
40cffcc0 | 191 | struct mddev *mddev; /* the md device that the bitmap is for */ |
32a7627c | 192 | |
32a7627c | 193 | __u64 events_cleared; |
a0da84f3 | 194 | int need_sync; |
32a7627c | 195 | |
1ec885cd N |
196 | struct bitmap_storage { |
197 | struct file *file; /* backing disk file */ | |
198 | struct page *sb_page; /* cached copy of the bitmap | |
199 | * file superblock */ | |
200 | struct page **filemap; /* list of cache pages for | |
201 | * the file */ | |
202 | unsigned long *filemap_attr; /* attributes associated | |
203 | * w/ filemap pages */ | |
204 | unsigned long file_pages; /* number of pages in the file*/ | |
9b1215c1 | 205 | unsigned long bytes; /* total bytes in the bitmap */ |
1ec885cd | 206 | } storage; |
32a7627c N |
207 | |
208 | unsigned long flags; | |
209 | ||
8311c29d N |
210 | int allclean; |
211 | ||
4b6d287f | 212 | atomic_t behind_writes; |
696fcd53 | 213 | unsigned long behind_writes_used; /* highest actual value at runtime */ |
4b6d287f | 214 | |
32a7627c N |
215 | /* |
216 | * the bitmap daemon - periodically wakes up and sweeps the bitmap | |
217 | * file, cleaning up bits and flushing out pages to disk as necessary | |
218 | */ | |
219 | unsigned long daemon_lastrun; /* jiffies of last run */ | |
b47490c9 N |
220 | unsigned long last_end_sync; /* when we lasted called end_sync to |
221 | * update bitmap with resync progress */ | |
32a7627c | 222 | |
d785a06a N |
223 | atomic_t pending_writes; /* pending writes to the bitmap file */ |
224 | wait_queue_head_t write_wait; | |
da6e1a32 | 225 | wait_queue_head_t overflow_wait; |
e555190d | 226 | wait_queue_head_t behind_wait; |
d785a06a | 227 | |
324a56e1 | 228 | struct kernfs_node *sysfs_can_clear; |
32a7627c N |
229 | }; |
230 | ||
231 | /* the bitmap API */ | |
232 | ||
233 | /* these are used only by md/bitmap */ | |
fd01b88c N |
234 | int bitmap_create(struct mddev *mddev); |
235 | int bitmap_load(struct mddev *mddev); | |
236 | void bitmap_flush(struct mddev *mddev); | |
237 | void bitmap_destroy(struct mddev *mddev); | |
32a7627c | 238 | |
32a7627c | 239 | void bitmap_print_sb(struct bitmap *bitmap); |
4ad13663 | 240 | void bitmap_update_sb(struct bitmap *bitmap); |
57148964 | 241 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap); |
32a7627c N |
242 | |
243 | int bitmap_setallbits(struct bitmap *bitmap); | |
a654b9d8 | 244 | void bitmap_write_all(struct bitmap *bitmap); |
32a7627c | 245 | |
9b1d1dac PC |
246 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e); |
247 | ||
32a7627c | 248 | /* these are exported */ |
4b6d287f N |
249 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, |
250 | unsigned long sectors, int behind); | |
251 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, | |
252 | unsigned long sectors, int success, int behind); | |
57dab0bd N |
253 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded); |
254 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted); | |
32a7627c | 255 | void bitmap_close_sync(struct bitmap *bitmap); |
b47490c9 | 256 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector); |
32a7627c | 257 | |
4ad13663 | 258 | void bitmap_unplug(struct bitmap *bitmap); |
fd01b88c | 259 | void bitmap_daemon_work(struct mddev *mddev); |
d60b479d N |
260 | |
261 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, | |
262 | int chunksize, int init); | |
32a7627c N |
263 | #endif |
264 | ||
265 | #endif |