History

md: move headers out of include/linux/raid/

Move the headers with the local structures for the disciplines and bitmap.h into drivers/md/ so that they are more easily grepable for hacking and not far away. md.h is left where it is for now as there are some uses from the outside.

Christoph Hellwig committed 17 years ago

Commit ef740c31 parent 2a40a8a16 files · +13 −14Browse files

Showing the changes to include/linux/raid (7 of 16 files). Show all

Unified · Split

7 files
  1. include/linux/raid/bitmap.h +0 −288
  2. include/linux/raid/linear.h +0 −31
  3. include/linux/raid/multipath.h +0 −42
  4. include/linux/raid/raid0.h +0 −30
  5. include/linux/raid/raid1.h +0 −134
  6. include/linux/raid/raid10.h +0 −123
  7. include/linux/raid/raid5.h +0 −402
include/linux/raid/bitmap.h+0−288
@@ -1,288 +0,0 @@
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
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
16#define BITMAP_MINOR 39
17
18/*
19 * in-memory bitmap:
20 *
21 * Use 16 bit block counters to track pending writes to each "chunk".
22 * The 2 high order bits are special-purpose, the first is a flag indicating
23 * whether a resync is needed. The second is a flag indicating whether a
24 * resync is active.
25 * This means that the counter is actually 14 bits:
26 *
27 * +--------+--------+------------------------------------------------+
28 * | resync | resync | counter |
29 * | needed | active | |
30 * | (0-1) | (0-1) | (0-16383) |
31 * +--------+--------+------------------------------------------------+
32 *
33 * The "resync needed" bit is set when:
34 * a '1' bit is read from storage at startup.
35 * a write request fails on some drives
36 * a resync is aborted on a chunk with 'resync active' set
37 * It is cleared (and resync-active set) when a resync starts across all drives
38 * of the chunk.
39 *
40 *
41 * The "resync active" bit is set when:
42 * a resync is started on all drives, and resync_needed is set.
43 * resync_needed will be cleared (as long as resync_active wasn't already set).
44 * It is cleared when a resync completes.
45 *
46 * The counter counts pending write requests, plus the on-disk bit.
47 * When the counter is '1' and the resync bits are clear, the on-disk
48 * bit can be cleared aswell, thus setting the counter to 0.
49 * When we set a bit, or in the counter (to start a write), if the fields is
50 * 0, we first set the disk bit and set the counter to 1.
51 *
52 * If the counter is 0, the on-disk bit is clear and the stipe is clean
53 * Anything that dirties the stipe pushes the counter to 2 (at least)
54 * and sets the on-disk bit (lazily).
55 * If a periodic sweep find the counter at 2, it is decremented to 1.
56 * If the sweep find the counter at 1, the on-disk bit is cleared and the
57 * counter goes to zero.
58 *
59 * Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
60 * counters as a fallback when "page" memory cannot be allocated:
61 *
62 * Normal case (page memory allocated):
63 *
64 * page pointer (32-bit)
65 *
66 * [ ] ------+
67 * |
68 * +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters)
69 * c1 c2 c2048
70 *
71 * Hijacked case (page memory allocation failed):
72 *
73 * hijacked page pointer (32-bit)
74 *
75 * [ ][ ] (no page memory allocated)
76 * counter #1 (16-bit) counter #2 (16-bit)
77 *
78 */
79
80#ifdef __KERNEL__
81
82#define PAGE_BITS (PAGE_SIZE << 3)
83#define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
84
85typedef __u16 bitmap_counter_t;
86#define COUNTER_BITS 16
87#define COUNTER_BIT_SHIFT 4
88#define COUNTER_BYTE_RATIO (COUNTER_BITS / 8)
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_SIZE 512
106#define BITMAP_BLOCK_SHIFT 9
107
108/* how many blocks per chunk? (this is variable) */
109#define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->chunksize >> BITMAP_BLOCK_SHIFT)
110#define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT)
111#define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1)
112
113/* when hijacked, the counters and bits represent even larger "chunks" */
114/* there will be 1024 chunks represented by each counter in the page pointers */
115#define PAGEPTR_BLOCK_RATIO(bitmap) \
116 (CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1)
117#define PAGEPTR_BLOCK_SHIFT(bitmap) \
118 (CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1)
119#define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1)
120
121/*
122 * on-disk bitmap:
123 *
124 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
125 * file a page at a time. There's a superblock at the start of the file.
126 */
127
128/* map chunks (bits) to file pages - offset by the size of the superblock */
129#define CHUNK_BIT_OFFSET(chunk) ((chunk) + (sizeof(bitmap_super_t) << 3))
130
131#endif
132
133/*
134 * bitmap structures:
135 */
136
137#define BITMAP_MAGIC 0x6d746962
138
139/* use these for bitmap->flags and bitmap->sb->state bit-fields */
140enum bitmap_state {
141 BITMAP_STALE = 0x002, /* the bitmap file is out of date or had -EIO */
142 BITMAP_WRITE_ERROR = 0x004, /* A write error has occurred */
143 BITMAP_HOSTENDIAN = 0x8000,
144};
145
146/* the superblock at the front of the bitmap file -- little endian */
147typedef struct bitmap_super_s {
148 __le32 magic; /* 0 BITMAP_MAGIC */
149 __le32 version; /* 4 the bitmap major for now, could change... */
150 __u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */
151 __le64 events; /* 24 event counter for the bitmap (1)*/
152 __le64 events_cleared;/*32 event counter when last bit cleared (2) */
153 __le64 sync_size; /* 40 the size of the md device's sync range(3) */
154 __le32 state; /* 48 bitmap state information */
155 __le32 chunksize; /* 52 the bitmap chunk size in bytes */
156 __le32 daemon_sleep; /* 56 seconds between disk flushes */
157 __le32 write_behind; /* 60 number of outstanding write-behind writes */
158
159 __u8 pad[256 - 64]; /* set to zero */
160} bitmap_super_t;
161
162/* notes:
163 * (1) This event counter is updated before the eventcounter in the md superblock
164 * When a bitmap is loaded, it is only accepted if this event counter is equal
165 * to, or one greater than, the event counter in the superblock.
166 * (2) This event counter is updated when the other one is *if*and*only*if* the
167 * array is not degraded. As bits are not cleared when the array is degraded,
168 * this represents the last time that any bits were cleared.
169 * If a device is being added that has an event count with this value or
170 * higher, it is accepted as conforming to the bitmap.
171 * (3)This is the number of sectors represented by the bitmap, and is the range that
172 * resync happens across. For raid1 and raid5/6 it is the size of individual
173 * devices. For raid10 it is the size of the array.
174 */
175
176#ifdef __KERNEL__
177
178/* the in-memory bitmap is represented by bitmap_pages */
179struct bitmap_page {
180 /*
181 * map points to the actual memory page
182 */
183 char *map;
184 /*
185 * in emergencies (when map cannot be alloced), hijack the map
186 * pointer and use it as two counters itself
187 */
188 unsigned int hijacked:1;
189 /*
190 * count of dirty bits on the page
191 */
192 unsigned int count:31;
193};
194
195/* keep track of bitmap file pages that have pending writes on them */
196struct page_list {
197 struct list_head list;
198 struct page *page;
199};
200
201/* the main bitmap structure - one per mddev */
202struct bitmap {
203 struct bitmap_page *bp;
204 unsigned long pages; /* total number of pages in the bitmap */
205 unsigned long missing_pages; /* number of pages not yet allocated */
206
207 mddev_t *mddev; /* the md device that the bitmap is for */
208
209 int counter_bits; /* how many bits per block counter */
210
211 /* bitmap chunksize -- how much data does each bit represent? */
212 unsigned long chunksize;
213 unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */
214 unsigned long chunks; /* total number of data chunks for the array */
215
216 /* We hold a count on the chunk currently being synced, and drop
217 * it when the last block is started. If the resync is aborted
218 * midway, we need to be able to drop that count, so we remember
219 * the counted chunk..
220 */
221 unsigned long syncchunk;
222
223 __u64 events_cleared;
224 int need_sync;
225
226 /* bitmap spinlock */
227 spinlock_t lock;
228
229 long offset; /* offset from superblock if file is NULL */
230 struct file *file; /* backing disk file */
231 struct page *sb_page; /* cached copy of the bitmap file superblock */
232 struct page **filemap; /* list of cache pages for the file */
233 unsigned long *filemap_attr; /* attributes associated w/ filemap pages */
234 unsigned long file_pages; /* number of pages in the file */
235 int last_page_size; /* bytes in the last page */
236
237 unsigned long flags;
238
239 int allclean;
240
241 unsigned long max_write_behind; /* write-behind mode */
242 atomic_t behind_writes;
243
244 /*
245 * the bitmap daemon - periodically wakes up and sweeps the bitmap
246 * file, cleaning up bits and flushing out pages to disk as necessary
247 */
248 unsigned long daemon_lastrun; /* jiffies of last run */
249 unsigned long daemon_sleep; /* how many seconds between updates? */
250 unsigned long last_end_sync; /* when we lasted called end_sync to
251 * update bitmap with resync progress */
252
253 atomic_t pending_writes; /* pending writes to the bitmap file */
254 wait_queue_head_t write_wait;
255 wait_queue_head_t overflow_wait;
256
257};
258
259/* the bitmap API */
260
261/* these are used only by md/bitmap */
262int bitmap_create(mddev_t *mddev);
263void bitmap_flush(mddev_t *mddev);
264void bitmap_destroy(mddev_t *mddev);
265
266void bitmap_print_sb(struct bitmap *bitmap);
267void bitmap_update_sb(struct bitmap *bitmap);
268
269int bitmap_setallbits(struct bitmap *bitmap);
270void bitmap_write_all(struct bitmap *bitmap);
271
272void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
273
274/* these are exported */
275int bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
276 unsigned long sectors, int behind);
277void bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
278 unsigned long sectors, int success, int behind);
279int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int degraded);
280void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted);
281void bitmap_close_sync(struct bitmap *bitmap);
282void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector);
283
284void bitmap_unplug(struct bitmap *bitmap);
285void bitmap_daemon_work(struct bitmap *bitmap);
286#endif
287
288#endif
include/linux/raid/linear.h+0−31
@@ -1,31 +0,0 @@
1#ifndef _LINEAR_H
2#define _LINEAR_H
3
4#include <linux/raid/md.h>
5
6struct dev_info {
7 mdk_rdev_t *rdev;
8 sector_t num_sectors;
9 sector_t start_sector;
10};
11
12typedef struct dev_info dev_info_t;
13
14struct linear_private_data
15{
16 struct linear_private_data *prev; /* earlier version */
17 dev_info_t **hash_table;
18 sector_t spacing;
19 sector_t array_sectors;
20 int sector_shift; /* shift before dividing
21 * by spacing
22 */
23 dev_info_t disks[0];
24};
25
26
27typedef struct linear_private_data linear_conf_t;
28
29#define mddev_to_conf(mddev) ((linear_conf_t *) mddev->private)
30
31#endif
include/linux/raid/multipath.h+0−42
@@ -1,42 +0,0 @@
1#ifndef _MULTIPATH_H
2#define _MULTIPATH_H
3
4#include <linux/raid/md.h>
5
6struct multipath_info {
7 mdk_rdev_t *rdev;
8};
9
10struct multipath_private_data {
11 mddev_t *mddev;
12 struct multipath_info *multipaths;
13 int raid_disks;
14 int working_disks;
15 spinlock_t device_lock;
16 struct list_head retry_list;
17
18 mempool_t *pool;
19};
20
21typedef struct multipath_private_data multipath_conf_t;
22
23/*
24 * this is the only point in the RAID code where we violate
25 * C type safety. mddev->private is an 'opaque' pointer.
26 */
27#define mddev_to_conf(mddev) ((multipath_conf_t *) mddev->private)
28
29/*
30 * this is our 'private' 'collective' MULTIPATH buffer head.
31 * it contains information about what kind of IO operations were started
32 * for this MULTIPATH operation, and about their status:
33 */
34
35struct multipath_bh {
36 mddev_t *mddev;
37 struct bio *master_bio;
38 struct bio bio;
39 int path;
40 struct list_head retry_list;
41};
42#endif
include/linux/raid/raid0.h+0−30
@@ -1,30 +0,0 @@
1#ifndef _RAID0_H
2#define _RAID0_H
3
4#include <linux/raid/md.h>
5
6struct strip_zone
7{
8 sector_t zone_start; /* Zone offset in md_dev (in sectors) */
9 sector_t dev_start; /* Zone offset in real dev (in sectors) */
10 sector_t sectors; /* Zone size in sectors */
11 int nb_dev; /* # of devices attached to the zone */
12 mdk_rdev_t **dev; /* Devices attached to the zone */
13};
14
15struct raid0_private_data
16{
17 struct strip_zone **hash_table; /* Table of indexes into strip_zone */
18 struct strip_zone *strip_zone;
19 mdk_rdev_t **devlist; /* lists of rdevs, pointed to by strip_zone->dev */
20 int nr_strip_zones;
21
22 sector_t spacing;
23 int sector_shift; /* shift this before divide by spacing */
24};
25
26typedef struct raid0_private_data raid0_conf_t;
27
28#define mddev_to_conf(mddev) ((raid0_conf_t *) mddev->private)
29
30#endif
include/linux/raid/raid1.h+0−134
@@ -1,134 +0,0 @@
1#ifndef _RAID1_H
2#define _RAID1_H
3
4#include <linux/raid/md.h>
5
6typedef struct mirror_info mirror_info_t;
7
8struct mirror_info {
9 mdk_rdev_t *rdev;
10 sector_t head_position;
11};
12
13/*
14 * memory pools need a pointer to the mddev, so they can force an unplug
15 * when memory is tight, and a count of the number of drives that the
16 * pool was allocated for, so they know how much to allocate and free.
17 * mddev->raid_disks cannot be used, as it can change while a pool is active
18 * These two datums are stored in a kmalloced struct.
19 */
20
21struct pool_info {
22 mddev_t *mddev;
23 int raid_disks;
24};
25
26
27typedef struct r1bio_s r1bio_t;
28
29struct r1_private_data_s {
30 mddev_t *mddev;
31 mirror_info_t *mirrors;
32 int raid_disks;
33 int last_used;
34 sector_t next_seq_sect;
35 spinlock_t device_lock;
36
37 struct list_head retry_list;
38 /* queue pending writes and submit them on unplug */
39 struct bio_list pending_bio_list;
40 /* queue of writes that have been unplugged */
41 struct bio_list flushing_bio_list;
42
43 /* for use when syncing mirrors: */
44
45 spinlock_t resync_lock;
46 int nr_pending;
47 int nr_waiting;
48 int nr_queued;
49 int barrier;
50 sector_t next_resync;
51 int fullsync; /* set to 1 if a full sync is needed,
52 * (fresh device added).
53 * Cleared when a sync completes.
54 */
55
56 wait_queue_head_t wait_barrier;
57
58 struct pool_info *poolinfo;
59
60 struct page *tmppage;
61
62 mempool_t *r1bio_pool;
63 mempool_t *r1buf_pool;
64};
65
66typedef struct r1_private_data_s conf_t;
67
68/*
69 * this is the only point in the RAID code where we violate
70 * C type safety. mddev->private is an 'opaque' pointer.
71 */
72#define mddev_to_conf(mddev) ((conf_t *) mddev->private)
73
74/*
75 * this is our 'private' RAID1 bio.
76 *
77 * it contains information about what kind of IO operations were started
78 * for this RAID1 operation, and about their status:
79 */
80
81struct r1bio_s {
82 atomic_t remaining; /* 'have we finished' count,
83 * used from IRQ handlers
84 */
85 atomic_t behind_remaining; /* number of write-behind ios remaining
86 * in this BehindIO request
87 */
88 sector_t sector;
89 int sectors;
90 unsigned long state;
91 mddev_t *mddev;
92 /*
93 * original bio going to /dev/mdx
94 */
95 struct bio *master_bio;
96 /*
97 * if the IO is in READ direction, then this is where we read
98 */
99 int read_disk;
100
101 struct list_head retry_list;
102 struct bitmap_update *bitmap_update;
103 /*
104 * if the IO is in WRITE direction, then multiple bios are used.
105 * We choose the number when they are allocated.
106 */
107 struct bio *bios[0];
108 /* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/
109};
110
111/* when we get a read error on a read-only array, we redirect to another
112 * device without failing the first device, or trying to over-write to
113 * correct the read error. To keep track of bad blocks on a per-bio
114 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
115 */
116#define IO_BLOCKED ((struct bio*)1)
117
118/* bits for r1bio.state */
119#define R1BIO_Uptodate 0
120#define R1BIO_IsSync 1
121#define R1BIO_Degraded 2
122#define R1BIO_BehindIO 3
123#define R1BIO_Barrier 4
124#define R1BIO_BarrierRetry 5
125/* For write-behind requests, we call bi_end_io when
126 * the last non-write-behind device completes, providing
127 * any write was successful. Otherwise we call when
128 * any write-behind write succeeds, otherwise we call
129 * with failure when last write completes (and all failed).
130 * Record that bi_end_io was called with this flag...
131 */
132#define R1BIO_Returned 6
133
134#endif
include/linux/raid/raid10.h+0−123
@@ -1,123 +0,0 @@
1#ifndef _RAID10_H
2#define _RAID10_H
3
4#include <linux/raid/md.h>
5
6typedef struct mirror_info mirror_info_t;
7
8struct mirror_info {
9 mdk_rdev_t *rdev;
10 sector_t head_position;
11};
12
13typedef struct r10bio_s r10bio_t;
14
15struct r10_private_data_s {
16 mddev_t *mddev;
17 mirror_info_t *mirrors;
18 int raid_disks;
19 spinlock_t device_lock;
20
21 /* geometry */
22 int near_copies; /* number of copies layed out raid0 style */
23 int far_copies; /* number of copies layed out
24 * at large strides across drives
25 */
26 int far_offset; /* far_copies are offset by 1 stripe
27 * instead of many
28 */
29 int copies; /* near_copies * far_copies.
30 * must be <= raid_disks
31 */
32 sector_t stride; /* distance between far copies.
33 * This is size / far_copies unless
34 * far_offset, in which case it is
35 * 1 stripe.
36 */
37
38 int chunk_shift; /* shift from chunks to sectors */
39 sector_t chunk_mask;
40
41 struct list_head retry_list;
42 /* queue pending writes and submit them on unplug */
43 struct bio_list pending_bio_list;
44
45
46 spinlock_t resync_lock;
47 int nr_pending;
48 int nr_waiting;
49 int nr_queued;
50 int barrier;
51 sector_t next_resync;
52 int fullsync; /* set to 1 if a full sync is needed,
53 * (fresh device added).
54 * Cleared when a sync completes.
55 */
56
57 wait_queue_head_t wait_barrier;
58
59 mempool_t *r10bio_pool;
60 mempool_t *r10buf_pool;
61 struct page *tmppage;
62};
63
64typedef struct r10_private_data_s conf_t;
65
66/*
67 * this is the only point in the RAID code where we violate
68 * C type safety. mddev->private is an 'opaque' pointer.
69 */
70#define mddev_to_conf(mddev) ((conf_t *) mddev->private)
71
72/*
73 * this is our 'private' RAID10 bio.
74 *
75 * it contains information about what kind of IO operations were started
76 * for this RAID10 operation, and about their status:
77 */
78
79struct r10bio_s {
80 atomic_t remaining; /* 'have we finished' count,
81 * used from IRQ handlers
82 */
83 sector_t sector; /* virtual sector number */
84 int sectors;
85 unsigned long state;
86 mddev_t *mddev;
87 /*
88 * original bio going to /dev/mdx
89 */
90 struct bio *master_bio;
91 /*
92 * if the IO is in READ direction, then this is where we read
93 */
94 int read_slot;
95
96 struct list_head retry_list;
97 /*
98 * if the IO is in WRITE direction, then multiple bios are used,
99 * one for each copy.
100 * When resyncing we also use one for each copy.
101 * When reconstructing, we use 2 bios, one for read, one for write.
102 * We choose the number when they are allocated.
103 */
104 struct {
105 struct bio *bio;
106 sector_t addr;
107 int devnum;
108 } devs[0];
109};
110
111/* when we get a read error on a read-only array, we redirect to another
112 * device without failing the first device, or trying to over-write to
113 * correct the read error. To keep track of bad blocks on a per-bio
114 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
115 */
116#define IO_BLOCKED ((struct bio*)1)
117
118/* bits for r10bio.state */
119#define R10BIO_Uptodate 0
120#define R10BIO_IsSync 1
121#define R10BIO_IsRecover 2
122#define R10BIO_Degraded 3
123#endif
include/linux/raid/raid5.h+0−402
@@ -1,402 +0,0 @@
1#ifndef _RAID5_H
2#define _RAID5_H
3
4#include <linux/raid/md.h>
5#include <linux/raid/xor.h>
6
7/*
8 *
9 * Each stripe contains one buffer per disc. Each buffer can be in
10 * one of a number of states stored in "flags". Changes between
11 * these states happen *almost* exclusively under a per-stripe
12 * spinlock. Some very specific changes can happen in bi_end_io, and
13 * these are not protected by the spin lock.
14 *
15 * The flag bits that are used to represent these states are:
16 * R5_UPTODATE and R5_LOCKED
17 *
18 * State Empty == !UPTODATE, !LOCK
19 * We have no data, and there is no active request
20 * State Want == !UPTODATE, LOCK
21 * A read request is being submitted for this block
22 * State Dirty == UPTODATE, LOCK
23 * Some new data is in this buffer, and it is being written out
24 * State Clean == UPTODATE, !LOCK
25 * We have valid data which is the same as on disc
26 *
27 * The possible state transitions are:
28 *
29 * Empty -> Want - on read or write to get old data for parity calc
30 * Empty -> Dirty - on compute_parity to satisfy write/sync request.(RECONSTRUCT_WRITE)
31 * Empty -> Clean - on compute_block when computing a block for failed drive
32 * Want -> Empty - on failed read
33 * Want -> Clean - on successful completion of read request
34 * Dirty -> Clean - on successful completion of write request
35 * Dirty -> Clean - on failed write
36 * Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
37 *
38 * The Want->Empty, Want->Clean, Dirty->Clean, transitions
39 * all happen in b_end_io at interrupt time.
40 * Each sets the Uptodate bit before releasing the Lock bit.
41 * This leaves one multi-stage transition:
42 * Want->Dirty->Clean
43 * This is safe because thinking that a Clean buffer is actually dirty
44 * will at worst delay some action, and the stripe will be scheduled
45 * for attention after the transition is complete.
46 *
47 * There is one possibility that is not covered by these states. That
48 * is if one drive has failed and there is a spare being rebuilt. We
49 * can't distinguish between a clean block that has been generated
50 * from parity calculations, and a clean block that has been
51 * successfully written to the spare ( or to parity when resyncing).
52 * To distingush these states we have a stripe bit STRIPE_INSYNC that
53 * is set whenever a write is scheduled to the spare, or to the parity
54 * disc if there is no spare. A sync request clears this bit, and
55 * when we find it set with no buffers locked, we know the sync is
56 * complete.
57 *
58 * Buffers for the md device that arrive via make_request are attached
59 * to the appropriate stripe in one of two lists linked on b_reqnext.
60 * One list (bh_read) for read requests, one (bh_write) for write.
61 * There should never be more than one buffer on the two lists
62 * together, but we are not guaranteed of that so we allow for more.
63 *
64 * If a buffer is on the read list when the associated cache buffer is
65 * Uptodate, the data is copied into the read buffer and it's b_end_io
66 * routine is called. This may happen in the end_request routine only
67 * if the buffer has just successfully been read. end_request should
68 * remove the buffers from the list and then set the Uptodate bit on
69 * the buffer. Other threads may do this only if they first check
70 * that the Uptodate bit is set. Once they have checked that they may
71 * take buffers off the read queue.
72 *
73 * When a buffer on the write list is committed for write it is copied
74 * into the cache buffer, which is then marked dirty, and moved onto a
75 * third list, the written list (bh_written). Once both the parity
76 * block and the cached buffer are successfully written, any buffer on
77 * a written list can be returned with b_end_io.
78 *
79 * The write list and read list both act as fifos. The read list is
80 * protected by the device_lock. The write and written lists are
81 * protected by the stripe lock. The device_lock, which can be
82 * claimed while the stipe lock is held, is only for list
83 * manipulations and will only be held for a very short time. It can
84 * be claimed from interrupts.
85 *
86 *
87 * Stripes in the stripe cache can be on one of two lists (or on
88 * neither). The "inactive_list" contains stripes which are not
89 * currently being used for any request. They can freely be reused
90 * for another stripe. The "handle_list" contains stripes that need
91 * to be handled in some way. Both of these are fifo queues. Each
92 * stripe is also (potentially) linked to a hash bucket in the hash
93 * table so that it can be found by sector number. Stripes that are
94 * not hashed must be on the inactive_list, and will normally be at
95 * the front. All stripes start life this way.
96 *
97 * The inactive_list, handle_list and hash bucket lists are all protected by the
98 * device_lock.
99 * - stripes on the inactive_list never have their stripe_lock held.
100 * - stripes have a reference counter. If count==0, they are on a list.
101 * - If a stripe might need handling, STRIPE_HANDLE is set.
102 * - When refcount reaches zero, then if STRIPE_HANDLE it is put on
103 * handle_list else inactive_list
104 *
105 * This, combined with the fact that STRIPE_HANDLE is only ever
106 * cleared while a stripe has a non-zero count means that if the
107 * refcount is 0 and STRIPE_HANDLE is set, then it is on the
108 * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
109 * the stripe is on inactive_list.
110 *
111 * The possible transitions are:
112 * activate an unhashed/inactive stripe (get_active_stripe())
113 * lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
114 * activate a hashed, possibly active stripe (get_active_stripe())
115 * lockdev check-hash if(!cnt++)unlink-stripe unlockdev
116 * attach a request to an active stripe (add_stripe_bh())
117 * lockdev attach-buffer unlockdev
118 * handle a stripe (handle_stripe())
119 * lockstripe clrSTRIPE_HANDLE ...
120 * (lockdev check-buffers unlockdev) ..
121 * change-state ..
122 * record io/ops needed unlockstripe schedule io/ops
123 * release an active stripe (release_stripe())
124 * lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
125 *
126 * The refcount counts each thread that have activated the stripe,
127 * plus raid5d if it is handling it, plus one for each active request
128 * on a cached buffer, and plus one if the stripe is undergoing stripe
129 * operations.
130 *
131 * Stripe operations are performed outside the stripe lock,
132 * the stripe operations are:
133 * -copying data between the stripe cache and user application buffers
134 * -computing blocks to save a disk access, or to recover a missing block
135 * -updating the parity on a write operation (reconstruct write and
136 * read-modify-write)
137 * -checking parity correctness
138 * -running i/o to disk
139 * These operations are carried out by raid5_run_ops which uses the async_tx
140 * api to (optionally) offload operations to dedicated hardware engines.
141 * When requesting an operation handle_stripe sets the pending bit for the
142 * operation and increments the count. raid5_run_ops is then run whenever
143 * the count is non-zero.
144 * There are some critical dependencies between the operations that prevent some
145 * from being requested while another is in flight.
146 * 1/ Parity check operations destroy the in cache version of the parity block,
147 * so we prevent parity dependent operations like writes and compute_blocks
148 * from starting while a check is in progress. Some dma engines can perform
149 * the check without damaging the parity block, in these cases the parity
150 * block is re-marked up to date (assuming the check was successful) and is
151 * not re-read from disk.
152 * 2/ When a write operation is requested we immediately lock the affected
153 * blocks, and mark them as not up to date. This causes new read requests
154 * to be held off, as well as parity checks and compute block operations.
155 * 3/ Once a compute block operation has been requested handle_stripe treats
156 * that block as if it is up to date. raid5_run_ops guaruntees that any
157 * operation that is dependent on the compute block result is initiated after
158 * the compute block completes.
159 */
160
161/*
162 * Operations state - intermediate states that are visible outside of sh->lock
163 * In general _idle indicates nothing is running, _run indicates a data
164 * processing operation is active, and _result means the data processing result
165 * is stable and can be acted upon. For simple operations like biofill and
166 * compute that only have an _idle and _run state they are indicated with
167 * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
168 */
169/**
170 * enum check_states - handles syncing / repairing a stripe
171 * @check_state_idle - check operations are quiesced
172 * @check_state_run - check operation is running
173 * @check_state_result - set outside lock when check result is valid
174 * @check_state_compute_run - check failed and we are repairing
175 * @check_state_compute_result - set outside lock when compute result is valid
176 */
177enum check_states {
178 check_state_idle = 0,
179 check_state_run, /* parity check */
180 check_state_check_result,
181 check_state_compute_run, /* parity repair */
182 check_state_compute_result,
183};
184
185/**
186 * enum reconstruct_states - handles writing or expanding a stripe
187 */
188enum reconstruct_states {
189 reconstruct_state_idle = 0,
190 reconstruct_state_prexor_drain_run, /* prexor-write */
191 reconstruct_state_drain_run, /* write */
192 reconstruct_state_run, /* expand */
193 reconstruct_state_prexor_drain_result,
194 reconstruct_state_drain_result,
195 reconstruct_state_result,
196};
197
198struct stripe_head {
199 struct hlist_node hash;
200 struct list_head lru; /* inactive_list or handle_list */
201 struct raid5_private_data *raid_conf;
202 sector_t sector; /* sector of this row */
203 int pd_idx; /* parity disk index */
204 unsigned long state; /* state flags */
205 atomic_t count; /* nr of active thread/requests */
206 spinlock_t lock;
207 int bm_seq; /* sequence number for bitmap flushes */
208 int disks; /* disks in stripe */
209 enum check_states check_state;
210 enum reconstruct_states reconstruct_state;
211 /* stripe_operations
212 * @target - STRIPE_OP_COMPUTE_BLK target
213 */
214 struct stripe_operations {
215 int target;
216 u32 zero_sum_result;
217 } ops;
218 struct r5dev {
219 struct bio req;
220 struct bio_vec vec;
221 struct page *page;
222 struct bio *toread, *read, *towrite, *written;
223 sector_t sector; /* sector of this page */
224 unsigned long flags;
225 } dev[1]; /* allocated with extra space depending of RAID geometry */
226};
227
228/* stripe_head_state - collects and tracks the dynamic state of a stripe_head
229 * for handle_stripe. It is only valid under spin_lock(sh->lock);
230 */
231struct stripe_head_state {
232 int syncing, expanding, expanded;
233 int locked, uptodate, to_read, to_write, failed, written;
234 int to_fill, compute, req_compute, non_overwrite;
235 int failed_num;
236 unsigned long ops_request;
237};
238
239/* r6_state - extra state data only relevant to r6 */
240struct r6_state {
241 int p_failed, q_failed, qd_idx, failed_num[2];
242};
243
244/* Flags */
245#define R5_UPTODATE 0 /* page contains current data */
246#define R5_LOCKED 1 /* IO has been submitted on "req" */
247#define R5_OVERWRITE 2 /* towrite covers whole page */
248/* and some that are internal to handle_stripe */
249#define R5_Insync 3 /* rdev && rdev->in_sync at start */
250#define R5_Wantread 4 /* want to schedule a read */
251#define R5_Wantwrite 5
252#define R5_Overlap 7 /* There is a pending overlapping request on this block */
253#define R5_ReadError 8 /* seen a read error here recently */
254#define R5_ReWrite 9 /* have tried to over-write the readerror */
255
256#define R5_Expanded 10 /* This block now has post-expand data */
257#define R5_Wantcompute 11 /* compute_block in progress treat as
258 * uptodate
259 */
260#define R5_Wantfill 12 /* dev->toread contains a bio that needs
261 * filling
262 */
263#define R5_Wantdrain 13 /* dev->towrite needs to be drained */
264/*
265 * Write method
266 */
267#define RECONSTRUCT_WRITE 1
268#define READ_MODIFY_WRITE 2
269/* not a write method, but a compute_parity mode */
270#define CHECK_PARITY 3
271
272/*
273 * Stripe state
274 */
275#define STRIPE_HANDLE 2
276#define STRIPE_SYNCING 3
277#define STRIPE_INSYNC 4
278#define STRIPE_PREREAD_ACTIVE 5
279#define STRIPE_DELAYED 6
280#define STRIPE_DEGRADED 7
281#define STRIPE_BIT_DELAY 8
282#define STRIPE_EXPANDING 9
283#define STRIPE_EXPAND_SOURCE 10
284#define STRIPE_EXPAND_READY 11
285#define STRIPE_IO_STARTED 12 /* do not count towards 'bypass_count' */
286#define STRIPE_FULL_WRITE 13 /* all blocks are set to be overwritten */
287#define STRIPE_BIOFILL_RUN 14
288#define STRIPE_COMPUTE_RUN 15
289/*
290 * Operation request flags
291 */
292#define STRIPE_OP_BIOFILL 0
293#define STRIPE_OP_COMPUTE_BLK 1
294#define STRIPE_OP_PREXOR 2
295#define STRIPE_OP_BIODRAIN 3
296#define STRIPE_OP_POSTXOR 4
297#define STRIPE_OP_CHECK 5
298
299/*
300 * Plugging:
301 *
302 * To improve write throughput, we need to delay the handling of some
303 * stripes until there has been a chance that several write requests
304 * for the one stripe have all been collected.
305 * In particular, any write request that would require pre-reading
306 * is put on a "delayed" queue until there are no stripes currently
307 * in a pre-read phase. Further, if the "delayed" queue is empty when
308 * a stripe is put on it then we "plug" the queue and do not process it
309 * until an unplug call is made. (the unplug_io_fn() is called).
310 *
311 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
312 * it to the count of prereading stripes.
313 * When write is initiated, or the stripe refcnt == 0 (just in case) we
314 * clear the PREREAD_ACTIVE flag and decrement the count
315 * Whenever the 'handle' queue is empty and the device is not plugged, we
316 * move any strips from delayed to handle and clear the DELAYED flag and set
317 * PREREAD_ACTIVE.
318 * In stripe_handle, if we find pre-reading is necessary, we do it if
319 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
320 * HANDLE gets cleared if stripe_handle leave nothing locked.
321 */
322
323
324struct disk_info {
325 mdk_rdev_t *rdev;
326};
327
328struct raid5_private_data {
329 struct hlist_head *stripe_hashtbl;
330 mddev_t *mddev;
331 struct disk_info *spare;
332 int chunk_size, level, algorithm;
333 int max_degraded;
334 int raid_disks;
335 int max_nr_stripes;
336
337 /* used during an expand */
338 sector_t expand_progress; /* MaxSector when no expand happening */
339 sector_t expand_lo; /* from here up to expand_progress it out-of-bounds
340 * as we haven't flushed the metadata yet
341 */
342 int previous_raid_disks;
343
344 struct list_head handle_list; /* stripes needing handling */
345 struct list_head hold_list; /* preread ready stripes */
346 struct list_head delayed_list; /* stripes that have plugged requests */
347 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
348 struct bio *retry_read_aligned; /* currently retrying aligned bios */
349 struct bio *retry_read_aligned_list; /* aligned bios retry list */
350 atomic_t preread_active_stripes; /* stripes with scheduled io */
351 atomic_t active_aligned_reads;
352 atomic_t pending_full_writes; /* full write backlog */
353 int bypass_count; /* bypassed prereads */
354 int bypass_threshold; /* preread nice */
355 struct list_head *last_hold; /* detect hold_list promotions */
356
357 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
358 /* unfortunately we need two cache names as we temporarily have
359 * two caches.
360 */
361 int active_name;
362 char cache_name[2][20];
363 struct kmem_cache *slab_cache; /* for allocating stripes */
364
365 int seq_flush, seq_write;
366 int quiesce;
367
368 int fullsync; /* set to 1 if a full sync is needed,
369 * (fresh device added).
370 * Cleared when a sync completes.
371 */
372
373 struct page *spare_page; /* Used when checking P/Q in raid6 */
374
375 /*
376 * Free stripes pool
377 */
378 atomic_t active_stripes;
379 struct list_head inactive_list;
380 wait_queue_head_t wait_for_stripe;
381 wait_queue_head_t wait_for_overlap;
382 int inactive_blocked; /* release of inactive stripes blocked,
383 * waiting for 25% to be free
384 */
385 int pool_size; /* number of disks in stripeheads in pool */
386 spinlock_t device_lock;
387 struct disk_info *disks;
388};
389
390typedef struct raid5_private_data raid5_conf_t;
391
392#define mddev_to_conf(mddev) ((raid5_conf_t *) mddev->private)
393
394/*
395 * Our supported algorithms
396 */
397#define ALGORITHM_LEFT_ASYMMETRIC 0
398#define ALGORITHM_RIGHT_ASYMMETRIC 1
399#define ALGORITHM_LEFT_SYMMETRIC 2
400#define ALGORITHM_RIGHT_SYMMETRIC 3
401
402#endif