1 /* SPDX-License-Identifier: GPL-2.0 */
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
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
13 * Version 5 is currently set only for clustered devices
14 */
15 #define BITMAP_MAJOR_HI 4
16 #define BITMAP_MAJOR_CLUSTERED 5
17 #define BITMAP_MAJOR_HOSTENDIAN 3
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 as well, 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 *
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)
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
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_SHIFT 9
106
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 {
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,
120 };
121
122 /* the superblock at the front of the bitmap file -- little endian */
123 typedef struct bitmap_super_s {
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 */
134 __le32 sectors_reserved; /* 64 number of 512-byte sectors that are
135 * reserved for the bitmap. */
136 __le32 nodes; /* 68 the maximum number of nodes in cluster. */
137 __u8 cluster_name[64]; /* 72 cluster name to which this md belongs */
138 __u8 pad[256 - 136]; /* set to zero */
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 /*
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 /*
174 * count of dirty bits on the page
175 */
176 unsigned int count:30;
177 };
178
179 /* the main bitmap structure - one per mddev */
180 struct bitmap {
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;
194
195 struct mddev *mddev; /* the md device that the bitmap is for */
196
197 __u64 events_cleared;
198 int need_sync;
199
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 unsigned long sb_index;
205 struct page **filemap; /* list of cache pages for
206 * the file */
207 unsigned long *filemap_attr; /* attributes associated
208 * w/ filemap pages */
209 unsigned long file_pages; /* number of pages in the file*/
210 unsigned long bytes; /* total bytes in the bitmap */
211 } storage;
212
213 unsigned long flags;
214
215 int allclean;
216
217 atomic_t behind_writes;
218 unsigned long behind_writes_used; /* highest actual value at runtime */
219
220 /*
221 * the bitmap daemon - periodically wakes up and sweeps the bitmap
222 * file, cleaning up bits and flushing out pages to disk as necessary
223 */
224 unsigned long daemon_lastrun; /* jiffies of last run */
225 unsigned long last_end_sync; /* when we lasted called end_sync to
226 * update bitmap with resync progress */
227
228 atomic_t pending_writes; /* pending writes to the bitmap file */
229 wait_queue_head_t write_wait;
230 wait_queue_head_t overflow_wait;
231 wait_queue_head_t behind_wait;
232
233 struct kernfs_node *sysfs_can_clear;
234 int cluster_slot; /* Slot offset for clustered env */
235 };
236
237 /* the bitmap API */
238
239 /* these are used only by md/bitmap */
240 struct bitmap *md_bitmap_create(struct mddev *mddev, int slot);
241 int md_bitmap_load(struct mddev *mddev);
242 void md_bitmap_flush(struct mddev *mddev);
243 void md_bitmap_destroy(struct mddev *mddev);
244
245 void md_bitmap_print_sb(struct bitmap *bitmap);
246 void md_bitmap_update_sb(struct bitmap *bitmap);
247 void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap);
248
249 int md_bitmap_setallbits(struct bitmap *bitmap);
250 void md_bitmap_write_all(struct bitmap *bitmap);
251
252 void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
253
254 /* these are exported */
255 int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
256 unsigned long sectors, int behind);
257 void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
258 unsigned long sectors, int success, int behind);
259 int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded);
260 void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted);
261 void md_bitmap_close_sync(struct bitmap *bitmap);
262 void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force);
263 void md_bitmap_sync_with_cluster(struct mddev *mddev,
264 sector_t old_lo, sector_t old_hi,
265 sector_t new_lo, sector_t new_hi);
266
267 void md_bitmap_unplug(struct bitmap *bitmap);
268 void md_bitmap_unplug_async(struct bitmap *bitmap);
269 void md_bitmap_daemon_work(struct mddev *mddev);
270
271 int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
272 int chunksize, int init);
273 struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot);
274 int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
275 sector_t *lo, sector_t *hi, bool clear_bits);
276 void md_bitmap_free(struct bitmap *bitmap);
277 void md_bitmap_wait_behind_writes(struct mddev *mddev);
278
md_bitmap_enabled(struct bitmap * bitmap)279 static inline bool md_bitmap_enabled(struct bitmap *bitmap)
280 {
281 return bitmap && bitmap->storage.filemap &&
282 !test_bit(BITMAP_STALE, &bitmap->flags);
283 }
284
285 #endif
286
287 #endif
288