blob: cfd7395de8fd36c779092617c23ad6d36fad7bea [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
NeilBrown32a76272005-06-21 17:17:14 -07002/*
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
NeilBrownbd926c62005-11-08 21:39:32 -080010#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
Goldwyn Rodrigues3c462c82015-08-19 07:35:54 +100013 * Version 5 is currently set only for clustered devices
NeilBrownbd926c62005-11-08 21:39:32 -080014 */
15#define BITMAP_MAJOR_HI 4
Goldwyn Rodrigues3c462c82015-08-19 07:35:54 +100016#define BITMAP_MAJOR_CLUSTERED 5
NeilBrownbd926c62005-11-08 21:39:32 -080017#define BITMAP_MAJOR_HOSTENDIAN 3
18
NeilBrown32a76272005-06-21 17:17:14 -070019/*
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
Lucas De Marchi25985ed2011-03-30 22:57:33 -030049 * bit can be cleared as well, thus setting the counter to 0.
NeilBrown32a76272005-06-21 17:17:14 -070050 * 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 *
Guoqing Jiangd85326c2016-03-14 17:01:39 +080053 * 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)
NeilBrown32a76272005-06-21 17:17:14 -070055 * 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
86typedef __u16 bitmap_counter_t;
87#define COUNTER_BITS 16
88#define COUNTER_BIT_SHIFT 4
NeilBrown32a76272005-06-21 17:17:14 -070089#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
NeilBrown32a76272005-06-21 17:17:14 -0700105#define BITMAP_BLOCK_SHIFT 9
106
NeilBrown32a76272005-06-21 17:17:14 -0700107#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 */
116enum bitmap_state {
NeilBrownb405fe92012-05-22 13:55:15 +1000117 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,
NeilBrown32a76272005-06-21 17:17:14 -0700120};
121
122/* the superblock at the front of the bitmap file -- little endian */
123typedef struct bitmap_super_s {
NeilBrown4f2e6392006-10-21 10:24:09 -0700124 __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 */
NeilBrown1dff2b82012-05-22 13:55:34 +1000134 __le32 sectors_reserved; /* 64 number of 512-byte sectors that are
135 * reserved for the bitmap. */
Goldwyn Rodrigues183bdf52014-03-07 10:30:43 -0600136 __le32 nodes; /* 68 the maximum number of nodes in cluster. */
Goldwyn Rodriguescf921cc2014-03-30 00:42:49 -0500137 __u8 cluster_name[64]; /* 72 cluster name to which this md belongs */
138 __u8 pad[256 - 136]; /* set to zero */
NeilBrown32a76272005-06-21 17:17:14 -0700139} 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 */
158struct 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 /*
NeilBrownbf07bb72012-05-22 13:55:06 +1000169 * 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 /*
NeilBrown32a76272005-06-21 17:17:14 -0700174 * count of dirty bits on the page
175 */
NeilBrownbf07bb72012-05-22 13:55:06 +1000176 unsigned int count:30;
NeilBrown32a76272005-06-21 17:17:14 -0700177};
178
NeilBrown32a76272005-06-21 17:17:14 -0700179/* the main bitmap structure - one per mddev */
180struct bitmap {
NeilBrown40cffcc2012-05-22 13:55:24 +1000181
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;
NeilBrown32a76272005-06-21 17:17:14 -0700194
NeilBrownfd01b882011-10-11 16:47:53 +1100195 struct mddev *mddev; /* the md device that the bitmap is for */
NeilBrown32a76272005-06-21 17:17:14 -0700196
NeilBrown32a76272005-06-21 17:17:14 -0700197 __u64 events_cleared;
Neil Browna0da84f2008-06-28 08:31:22 +1000198 int need_sync;
NeilBrown32a76272005-06-21 17:17:14 -0700199
NeilBrown1ec885c2012-05-22 13:55:10 +1000200 struct bitmap_storage {
201 struct file *file; /* backing disk file */
202 struct page *sb_page; /* cached copy of the bitmap
203 * file superblock */
204 struct page **filemap; /* list of cache pages for
205 * the file */
206 unsigned long *filemap_attr; /* attributes associated
207 * w/ filemap pages */
208 unsigned long file_pages; /* number of pages in the file*/
NeilBrown9b1215c2012-05-22 13:55:11 +1000209 unsigned long bytes; /* total bytes in the bitmap */
NeilBrown1ec885c2012-05-22 13:55:10 +1000210 } storage;
NeilBrown32a76272005-06-21 17:17:14 -0700211
212 unsigned long flags;
213
NeilBrown8311c292008-03-04 14:29:30 -0800214 int allclean;
215
NeilBrown4b6d2872005-09-09 16:23:47 -0700216 atomic_t behind_writes;
Paul Clements696fcd52010-03-08 16:02:37 +1100217 unsigned long behind_writes_used; /* highest actual value at runtime */
NeilBrown4b6d2872005-09-09 16:23:47 -0700218
NeilBrown32a76272005-06-21 17:17:14 -0700219 /*
220 * the bitmap daemon - periodically wakes up and sweeps the bitmap
221 * file, cleaning up bits and flushing out pages to disk as necessary
222 */
223 unsigned long daemon_lastrun; /* jiffies of last run */
NeilBrownb47490c2008-02-06 01:39:50 -0800224 unsigned long last_end_sync; /* when we lasted called end_sync to
225 * update bitmap with resync progress */
NeilBrown32a76272005-06-21 17:17:14 -0700226
NeilBrownd785a062006-06-26 00:27:48 -0700227 atomic_t pending_writes; /* pending writes to the bitmap file */
228 wait_queue_head_t write_wait;
Neil Brownda6e1a32007-02-08 14:20:37 -0800229 wait_queue_head_t overflow_wait;
NeilBrowne5551902010-03-31 11:21:44 +1100230 wait_queue_head_t behind_wait;
NeilBrownd785a062006-06-26 00:27:48 -0700231
Tejun Heo324a56e2013-12-11 14:11:53 -0500232 struct kernfs_node *sysfs_can_clear;
Goldwyn Rodriguesb97e92572014-06-06 11:50:56 -0500233 int cluster_slot; /* Slot offset for clustered env */
NeilBrown32a76272005-06-21 17:17:14 -0700234};
235
236/* the bitmap API */
237
238/* these are used only by md/bitmap */
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700239struct bitmap *md_bitmap_create(struct mddev *mddev, int slot);
240int md_bitmap_load(struct mddev *mddev);
241void md_bitmap_flush(struct mddev *mddev);
242void md_bitmap_destroy(struct mddev *mddev);
NeilBrown32a76272005-06-21 17:17:14 -0700243
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700244void md_bitmap_print_sb(struct bitmap *bitmap);
245void md_bitmap_update_sb(struct bitmap *bitmap);
246void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700247
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700248int md_bitmap_setallbits(struct bitmap *bitmap);
249void md_bitmap_write_all(struct bitmap *bitmap);
NeilBrown32a76272005-06-21 17:17:14 -0700250
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700251void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
Paul Clements9b1d1da2006-10-03 01:15:49 -0700252
NeilBrown32a76272005-06-21 17:17:14 -0700253/* these are exported */
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700254int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
255 unsigned long sectors, int behind);
256void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
NeilBrown4b6d2872005-09-09 16:23:47 -0700257 unsigned long sectors, int success, int behind);
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700258int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int degraded);
259void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted);
260void md_bitmap_close_sync(struct bitmap *bitmap);
261void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force);
262void md_bitmap_sync_with_cluster(struct mddev *mddev,
263 sector_t old_lo, sector_t old_hi,
264 sector_t new_lo, sector_t new_hi);
NeilBrown32a76272005-06-21 17:17:14 -0700265
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700266void md_bitmap_unplug(struct bitmap *bitmap);
267void md_bitmap_daemon_work(struct mddev *mddev);
NeilBrownd60b4792012-05-22 13:55:25 +1000268
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700269int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks,
270 int chunksize, int init);
Guoqing Jiangb98938d2017-03-01 16:42:39 +0800271struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot);
Andy Shevchenkoe64e40182018-08-01 15:20:50 -0700272int md_bitmap_copy_from_slot(struct mddev *mddev, int slot,
273 sector_t *lo, sector_t *hi, bool clear_bits);
274void md_bitmap_free(struct bitmap *bitmap);
275void md_bitmap_wait_behind_writes(struct mddev *mddev);
NeilBrown32a76272005-06-21 17:17:14 -0700276#endif
277
278#endif