blob: e2d999a62ba970adac48e74a793a4d9a7cf65694 [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
NeilBrown32a76272005-06-21 17:17:14 -070016/*
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
Lucas De Marchi25985ed2011-03-30 22:57:33 -030046 * bit can be cleared as well, thus setting the counter to 0.
NeilBrown32a76272005-06-21 17:17:14 -070047 * 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
83typedef __u16 bitmap_counter_t;
84#define COUNTER_BITS 16
85#define COUNTER_BIT_SHIFT 4
NeilBrown32a76272005-06-21 17:17:14 -070086#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
NeilBrown32a76272005-06-21 17:17:14 -0700102#define BITMAP_BLOCK_SHIFT 9
103
NeilBrown32a76272005-06-21 17:17:14 -0700104#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 */
113enum bitmap_state {
NeilBrownb405fe92012-05-22 13:55:15 +1000114 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,
NeilBrown32a76272005-06-21 17:17:14 -0700117};
118
119/* the superblock at the front of the bitmap file -- little endian */
120typedef struct bitmap_super_s {
NeilBrown4f2e6392006-10-21 10:24:09 -0700121 __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 */
NeilBrown32a76272005-06-21 17:17:14 -0700131
NeilBrown4b6d2872005-09-09 16:23:47 -0700132 __u8 pad[256 - 64]; /* set to zero */
NeilBrown32a76272005-06-21 17:17:14 -0700133} bitmap_super_t;
134
135/* notes:
136 * (1) This event counter is updated before the eventcounter in the md superblock
137 * When a bitmap is loaded, it is only accepted if this event counter is equal
138 * to, or one greater than, the event counter in the superblock.
139 * (2) This event counter is updated when the other one is *if*and*only*if* the
140 * array is not degraded. As bits are not cleared when the array is degraded,
141 * this represents the last time that any bits were cleared.
142 * If a device is being added that has an event count with this value or
143 * higher, it is accepted as conforming to the bitmap.
144 * (3)This is the number of sectors represented by the bitmap, and is the range that
145 * resync happens across. For raid1 and raid5/6 it is the size of individual
146 * devices. For raid10 it is the size of the array.
147 */
148
149#ifdef __KERNEL__
150
151/* the in-memory bitmap is represented by bitmap_pages */
152struct bitmap_page {
153 /*
154 * map points to the actual memory page
155 */
156 char *map;
157 /*
158 * in emergencies (when map cannot be alloced), hijack the map
159 * pointer and use it as two counters itself
160 */
161 unsigned int hijacked:1;
162 /*
NeilBrownbf07bb72012-05-22 13:55:06 +1000163 * If any counter in this page is '1' or '2' - and so could be
164 * cleared then that page is marked as 'pending'
165 */
166 unsigned int pending:1;
167 /*
NeilBrown32a76272005-06-21 17:17:14 -0700168 * count of dirty bits on the page
169 */
NeilBrownbf07bb72012-05-22 13:55:06 +1000170 unsigned int count:30;
NeilBrown32a76272005-06-21 17:17:14 -0700171};
172
NeilBrown32a76272005-06-21 17:17:14 -0700173/* the main bitmap structure - one per mddev */
174struct bitmap {
175 struct bitmap_page *bp;
176 unsigned long pages; /* total number of pages in the bitmap */
177 unsigned long missing_pages; /* number of pages not yet allocated */
178
NeilBrownfd01b882011-10-11 16:47:53 +1100179 struct mddev *mddev; /* the md device that the bitmap is for */
NeilBrown32a76272005-06-21 17:17:14 -0700180
NeilBrown32a76272005-06-21 17:17:14 -0700181 /* bitmap chunksize -- how much data does each bit represent? */
NeilBrown61a0d802012-03-19 12:46:41 +1100182 unsigned long chunkshift; /* chunksize = 2^(chunkshift+9) (for bitops) */
NeilBrown32a76272005-06-21 17:17:14 -0700183 unsigned long chunks; /* total number of data chunks for the array */
184
NeilBrown32a76272005-06-21 17:17:14 -0700185 __u64 events_cleared;
Neil Browna0da84f2008-06-28 08:31:22 +1000186 int need_sync;
NeilBrown32a76272005-06-21 17:17:14 -0700187
188 /* bitmap spinlock */
189 spinlock_t lock;
190
NeilBrown1ec885c2012-05-22 13:55:10 +1000191 struct bitmap_storage {
192 struct file *file; /* backing disk file */
193 struct page *sb_page; /* cached copy of the bitmap
194 * file superblock */
195 struct page **filemap; /* list of cache pages for
196 * the file */
197 unsigned long *filemap_attr; /* attributes associated
198 * w/ filemap pages */
199 unsigned long file_pages; /* number of pages in the file*/
NeilBrown9b1215c2012-05-22 13:55:11 +1000200 unsigned long bytes; /* total bytes in the bitmap */
NeilBrown1ec885c2012-05-22 13:55:10 +1000201 } storage;
NeilBrown32a76272005-06-21 17:17:14 -0700202
203 unsigned long flags;
204
NeilBrown8311c292008-03-04 14:29:30 -0800205 int allclean;
206
NeilBrown4b6d2872005-09-09 16:23:47 -0700207 atomic_t behind_writes;
Paul Clements696fcd52010-03-08 16:02:37 +1100208 unsigned long behind_writes_used; /* highest actual value at runtime */
NeilBrown4b6d2872005-09-09 16:23:47 -0700209
NeilBrown32a76272005-06-21 17:17:14 -0700210 /*
211 * the bitmap daemon - periodically wakes up and sweeps the bitmap
212 * file, cleaning up bits and flushing out pages to disk as necessary
213 */
214 unsigned long daemon_lastrun; /* jiffies of last run */
NeilBrownb47490c2008-02-06 01:39:50 -0800215 unsigned long last_end_sync; /* when we lasted called end_sync to
216 * update bitmap with resync progress */
NeilBrown32a76272005-06-21 17:17:14 -0700217
NeilBrownd785a062006-06-26 00:27:48 -0700218 atomic_t pending_writes; /* pending writes to the bitmap file */
219 wait_queue_head_t write_wait;
Neil Brownda6e1a32007-02-08 14:20:37 -0800220 wait_queue_head_t overflow_wait;
NeilBrowne5551902010-03-31 11:21:44 +1100221 wait_queue_head_t behind_wait;
NeilBrownd785a062006-06-26 00:27:48 -0700222
NeilBrownece5cff2009-12-14 12:49:56 +1100223 struct sysfs_dirent *sysfs_can_clear;
NeilBrown32a76272005-06-21 17:17:14 -0700224};
225
226/* the bitmap API */
227
228/* these are used only by md/bitmap */
NeilBrownfd01b882011-10-11 16:47:53 +1100229int bitmap_create(struct mddev *mddev);
230int bitmap_load(struct mddev *mddev);
231void bitmap_flush(struct mddev *mddev);
232void bitmap_destroy(struct mddev *mddev);
NeilBrown32a76272005-06-21 17:17:14 -0700233
NeilBrown32a76272005-06-21 17:17:14 -0700234void bitmap_print_sb(struct bitmap *bitmap);
NeilBrown4ad13662007-07-17 04:06:13 -0700235void bitmap_update_sb(struct bitmap *bitmap);
NeilBrown57148962012-03-19 12:46:40 +1100236void bitmap_status(struct seq_file *seq, struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700237
238int bitmap_setallbits(struct bitmap *bitmap);
NeilBrowna654b9d82005-06-21 17:17:27 -0700239void bitmap_write_all(struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700240
Paul Clements9b1d1da2006-10-03 01:15:49 -0700241void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
242
NeilBrown32a76272005-06-21 17:17:14 -0700243/* these are exported */
NeilBrown4b6d2872005-09-09 16:23:47 -0700244int bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
245 unsigned long sectors, int behind);
246void bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
247 unsigned long sectors, int success, int behind);
NeilBrown57dab0b2010-10-19 10:03:39 +1100248int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded);
249void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted);
NeilBrown32a76272005-06-21 17:17:14 -0700250void bitmap_close_sync(struct bitmap *bitmap);
NeilBrownb47490c2008-02-06 01:39:50 -0800251void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector);
NeilBrown32a76272005-06-21 17:17:14 -0700252
NeilBrown4ad13662007-07-17 04:06:13 -0700253void bitmap_unplug(struct bitmap *bitmap);
NeilBrownfd01b882011-10-11 16:47:53 +1100254void bitmap_daemon_work(struct mddev *mddev);
NeilBrown32a76272005-06-21 17:17:14 -0700255#endif
256
257#endif