2 * bitmap.h: Copyright (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
4 * additions: Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
10 #define BITMAP_MINOR 38
15 * Use 16 bit block counters to track pending writes to each "chunk".
16 * The 2 high order bits are special-purpose, the first is a flag indicating
17 * whether a resync is needed. The second is a flag indicating whether a
19 * This means that the counter is actually 14 bits:
21 * +--------+--------+------------------------------------------------+
22 * | resync | resync | counter |
23 * | needed | active | |
24 * | (0-1) | (0-1) | (0-16383) |
25 * +--------+--------+------------------------------------------------+
27 * The "resync needed" bit is set when:
28 * a '1' bit is read from storage at startup.
29 * a write request fails on some drives
30 * a resync is aborted on a chunk with 'resync active' set
31 * It is cleared (and resync-active set) when a resync starts across all drives
35 * The "resync active" bit is set when:
36 * a resync is started on all drives, and resync_needed is set.
37 * resync_needed will be cleared (as long as resync_active wasn't already set).
38 * It is cleared when a resync completes.
40 * The counter counts pending write requests, plus the on-disk bit.
41 * When the counter is '1' and the resync bits are clear, the on-disk
42 * bit can be cleared aswell, thus setting the counter to 0.
43 * When we set a bit, or in the counter (to start a write), if the fields is
44 * 0, we first set the disk bit and set the counter to 1.
46 * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
47 * counters as a fallback when "page" memory cannot be allocated:
49 * Normal case (page memory allocated):
51 * page pointer (32-bit)
55 * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters)
58 * Hijacked case (page memory allocation failed):
60 * hijacked page pointer (32-bit)
62 * [ ][ ] (no page memory allocated)
63 * counter #1 (16-bit) counter #2 (16-bit)
69 #define PAGE_BITS (PAGE_SIZE << 3)
70 #define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
72 typedef __u16 bitmap_counter_t
;
73 #define COUNTER_BITS 16
74 #define COUNTER_BIT_SHIFT 4
75 #define COUNTER_BYTE_RATIO (COUNTER_BITS / 8)
76 #define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3)
78 #define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1)))
79 #define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2)))
80 #define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1)
81 #define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK)
82 #define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK)
83 #define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX)
85 /* how many counters per page? */
86 #define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS)
87 /* same, except a shift value for more efficient bitops */
88 #define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT)
89 /* same, except a mask value for more efficient bitops */
90 #define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1)
92 #define BITMAP_BLOCK_SIZE 512
93 #define BITMAP_BLOCK_SHIFT 9
95 /* how many blocks per chunk? (this is variable) */
96 #define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->chunksize >> BITMAP_BLOCK_SHIFT)
97 #define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT)
98 #define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1)
100 /* when hijacked, the counters and bits represent even larger "chunks" */
101 /* there will be 1024 chunks represented by each counter in the page pointers */
102 #define PAGEPTR_BLOCK_RATIO(bitmap) \
103 (CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1)
104 #define PAGEPTR_BLOCK_SHIFT(bitmap) \
105 (CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1)
106 #define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1)
111 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
112 * file a page at a time. There's a superblock at the start of the file.
115 /* map chunks (bits) to file pages - offset by the size of the superblock */
116 #define CHUNK_BIT_OFFSET(chunk) ((chunk) + (sizeof(bitmap_super_t) << 3))
124 #define BITMAP_MAGIC 0x6d746962
126 /* use these for bitmap->flags and bitmap->sb->state bit-fields */
128 BITMAP_ACTIVE
= 0x001, /* the bitmap is in use */
129 BITMAP_STALE
= 0x002 /* the bitmap file is out of date or had -EIO */
132 /* the superblock at the front of the bitmap file -- little endian */
133 typedef struct bitmap_super_s
{
134 __u32 magic
; /* 0 BITMAP_MAGIC */
135 __u32 version
; /* 4 the bitmap major for now, could change... */
136 __u8 uuid
[16]; /* 8 128 bit uuid - must match md device uuid */
137 __u64 events
; /* 24 event counter for the bitmap (1)*/
138 __u64 events_cleared
;/*32 event counter when last bit cleared (2) */
139 __u64 sync_size
; /* 40 the size of the md device's sync range(3) */
140 __u32 state
; /* 48 bitmap state information */
141 __u32 chunksize
; /* 52 the bitmap chunk size in bytes */
142 __u32 daemon_sleep
; /* 56 seconds between disk flushes */
144 __u8 pad
[256 - 60]; /* set to zero */
148 * (1) This event counter is updated before the eventcounter in the md superblock
149 * When a bitmap is loaded, it is only accepted if this event counter is equal
150 * to, or one greater than, the event counter in the superblock.
151 * (2) This event counter is updated when the other one is *if*and*only*if* the
152 * array is not degraded. As bits are not cleared when the array is degraded,
153 * this represents the last time that any bits were cleared.
154 * If a device is being added that has an event count with this value or
155 * higher, it is accepted as conforming to the bitmap.
156 * (3)This is the number of sectors represented by the bitmap, and is the range that
157 * resync happens across. For raid1 and raid5/6 it is the size of individual
158 * devices. For raid10 it is the size of the array.
163 /* the in-memory bitmap is represented by bitmap_pages */
166 * map points to the actual memory page
170 * in emergencies (when map cannot be alloced), hijack the map
171 * pointer and use it as two counters itself
173 unsigned int hijacked
;
175 * count of dirty bits on the page
180 /* keep track of bitmap file pages that have pending writes on them */
182 struct list_head list
;
186 /* the main bitmap structure - one per mddev */
188 struct bitmap_page
*bp
;
189 unsigned long pages
; /* total number of pages in the bitmap */
190 unsigned long missing_pages
; /* number of pages not yet allocated */
192 mddev_t
*mddev
; /* the md device that the bitmap is for */
194 int counter_bits
; /* how many bits per block counter */
196 /* bitmap chunksize -- how much data does each bit represent? */
197 unsigned long chunksize
;
198 unsigned long chunkshift
; /* chunksize = 2^chunkshift (for bitops) */
199 unsigned long chunks
; /* total number of data chunks for the array */
201 /* We hold a count on the chunk currently being synced, and drop
202 * it when the last block is started. If the resync is aborted
203 * midway, we need to be able to drop that count, so we remember
204 * the counted chunk..
206 unsigned long syncchunk
;
208 __u64 events_cleared
;
210 /* bitmap spinlock */
213 struct file
*file
; /* backing disk file */
214 struct page
*sb_page
; /* cached copy of the bitmap file superblock */
215 struct page
**filemap
; /* list of cache pages for the file */
216 unsigned long *filemap_attr
; /* attributes associated w/ filemap pages */
217 unsigned long file_pages
; /* number of pages in the file */
222 * the bitmap daemon - periodically wakes up and sweeps the bitmap
223 * file, cleaning up bits and flushing out pages to disk as necessary
225 mdk_thread_t
*daemon
;
226 unsigned long daemon_sleep
; /* how many seconds between updates? */
229 * bitmap write daemon - this daemon performs writes to the bitmap file
230 * this thread is only needed because of a limitation in ext3 (jbd)
231 * that does not allow a task to have two journal transactions ongoing
232 * simultaneously (even if the transactions are for two different
233 * filesystems) -- in the case of bitmap, that would be the filesystem
234 * that the bitmap file resides on and the filesystem that is mounted
235 * on the md device -- see current->journal_info in jbd/transaction.c
237 mdk_thread_t
*write_daemon
;
238 mdk_thread_t
*writeback_daemon
;
239 spinlock_t write_lock
;
240 struct semaphore write_ready
;
241 struct semaphore write_done
;
242 unsigned long writes_pending
;
243 wait_queue_head_t write_wait
;
244 struct list_head write_pages
;
245 struct list_head complete_pages
;
246 mempool_t
*write_pool
;
251 /* these are used only by md/bitmap */
252 int bitmap_create(mddev_t
*mddev
);
253 void bitmap_destroy(mddev_t
*mddev
);
254 int bitmap_active(struct bitmap
*bitmap
);
256 char *file_path(struct file
*file
, char *buf
, int count
);
257 void bitmap_print_sb(struct bitmap
*bitmap
);
258 int bitmap_update_sb(struct bitmap
*bitmap
);
260 int bitmap_setallbits(struct bitmap
*bitmap
);
262 /* these are exported */
263 void bitmap_startwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
);
264 void bitmap_endwrite(struct bitmap
*bitmap
, sector_t offset
, unsigned long sectors
,
266 int bitmap_start_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
);
267 void bitmap_end_sync(struct bitmap
*bitmap
, sector_t offset
, int *blocks
, int aborted
);
268 void bitmap_close_sync(struct bitmap
*bitmap
);
270 int bitmap_unplug(struct bitmap
*bitmap
);