Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _RAID1_H
3#define _RAID1_H
4
5/*
6 * each barrier unit size is 64MB fow now
7 * note: it must be larger than RESYNC_DEPTH
8 */
9#define BARRIER_UNIT_SECTOR_BITS 17
10#define BARRIER_UNIT_SECTOR_SIZE (1<<17)
11/*
12 * In struct r1conf, the following members are related to I/O barrier
13 * buckets,
14 * atomic_t *nr_pending;
15 * atomic_t *nr_waiting;
16 * atomic_t *nr_queued;
17 * atomic_t *barrier;
18 * Each of them points to array of atomic_t variables, each array is
19 * designed to have BARRIER_BUCKETS_NR elements and occupy a single
20 * memory page. The data width of atomic_t variables is 4 bytes, equal
21 * to 1<<(ilog2(sizeof(atomic_t))), BARRIER_BUCKETS_NR_BITS is defined
22 * as (PAGE_SHIFT - ilog2(sizeof(int))) to make sure an array of
23 * atomic_t variables with BARRIER_BUCKETS_NR elements just exactly
24 * occupies a single memory page.
25 */
26#define BARRIER_BUCKETS_NR_BITS (PAGE_SHIFT - ilog2(sizeof(atomic_t)))
27#define BARRIER_BUCKETS_NR (1<<BARRIER_BUCKETS_NR_BITS)
28
29/* Note: raid1_info.rdev can be set to NULL asynchronously by raid1_remove_disk.
30 * There are three safe ways to access raid1_info.rdev.
31 * 1/ when holding mddev->reconfig_mutex
32 * 2/ when resync/recovery is known to be happening - i.e. in code that is
33 * called as part of performing resync/recovery.
34 * 3/ while holding rcu_read_lock(), use rcu_dereference to get the pointer
35 * and if it is non-NULL, increment rdev->nr_pending before dropping the
36 * RCU lock.
37 * When .rdev is set to NULL, the nr_pending count checked again and if it has
38 * been incremented, the pointer is put back in .rdev.
39 */
40
41struct raid1_info {
42 struct md_rdev *rdev;
43 sector_t head_position;
44
45 /* When choose the best device for a read (read_balance())
46 * we try to keep sequential reads one the same device
47 */
48 sector_t next_seq_sect;
49 sector_t seq_start;
50};
51
52/*
53 * memory pools need a pointer to the mddev, so they can force an unplug
54 * when memory is tight, and a count of the number of drives that the
55 * pool was allocated for, so they know how much to allocate and free.
56 * mddev->raid_disks cannot be used, as it can change while a pool is active
57 * These two datums are stored in a kmalloced struct.
58 * The 'raid_disks' here is twice the raid_disks in r1conf.
59 * This allows space for each 'real' device can have a replacement in the
60 * second half of the array.
61 */
62
63struct pool_info {
64 struct mddev *mddev;
65 int raid_disks;
66};
67
68struct r1conf {
69 struct mddev *mddev;
70 struct raid1_info *mirrors; /* twice 'raid_disks' to
71 * allow for replacements.
72 */
73 int raid_disks;
74
75 spinlock_t device_lock;
76
77 /* list of 'struct r1bio' that need to be processed by raid1d,
78 * whether to retry a read, writeout a resync or recovery
79 * block, or anything else.
80 */
81 struct list_head retry_list;
82 /* A separate list of r1bio which just need raid_end_bio_io called.
83 * This mustn't happen for writes which had any errors if the superblock
84 * needs to be written.
85 */
86 struct list_head bio_end_io_list;
87
88 /* queue pending writes to be submitted on unplug */
89 struct bio_list pending_bio_list;
90
91 /* for use when syncing mirrors:
92 * We don't allow both normal IO and resync/recovery IO at
93 * the same time - resync/recovery can only happen when there
94 * is no other IO. So when either is active, the other has to wait.
95 * See more details description in raid1.c near raise_barrier().
96 */
97 wait_queue_head_t wait_barrier;
98 spinlock_t resync_lock;
99 atomic_t nr_sync_pending;
100 atomic_t *nr_pending;
101 atomic_t *nr_waiting;
102 atomic_t *nr_queued;
103 atomic_t *barrier;
104 int array_frozen;
105
106 /* Set to 1 if a full sync is needed, (fresh device added).
107 * Cleared when a sync completes.
108 */
109 int fullsync;
110
111 /* When the same as mddev->recovery_disabled we don't allow
112 * recovery to be attempted as we expect a read error.
113 */
114 int recovery_disabled;
115
116 /* poolinfo contains information about the content of the
117 * mempools - it changes when the array grows or shrinks
118 */
119 struct pool_info *poolinfo;
120 mempool_t r1bio_pool;
121 mempool_t r1buf_pool;
122
123 struct bio_set bio_split;
124
125 /* temporary buffer to synchronous IO when attempting to repair
126 * a read error.
127 */
128 struct page *tmppage;
129
130 /* When taking over an array from a different personality, we store
131 * the new thread here until we fully activate the array.
132 */
133 struct md_thread __rcu *thread;
134
135 /* Keep track of cluster resync window to send to other
136 * nodes.
137 */
138 sector_t cluster_sync_low;
139 sector_t cluster_sync_high;
140
141};
142
143/*
144 * this is our 'private' RAID1 bio.
145 *
146 * it contains information about what kind of IO operations were started
147 * for this RAID1 operation, and about their status:
148 */
149
150struct r1bio {
151 atomic_t remaining; /* 'have we finished' count,
152 * used from IRQ handlers
153 */
154 atomic_t behind_remaining; /* number of write-behind ios remaining
155 * in this BehindIO request
156 */
157 sector_t sector;
158 int sectors;
159 unsigned long state;
160 struct mddev *mddev;
161 /*
162 * original bio going to /dev/mdx
163 */
164 struct bio *master_bio;
165 /*
166 * if the IO is in READ direction, then this is where we read
167 */
168 int read_disk;
169
170 struct list_head retry_list;
171
172 /*
173 * When R1BIO_BehindIO is set, we store pages for write behind
174 * in behind_master_bio.
175 */
176 struct bio *behind_master_bio;
177
178 /*
179 * if the IO is in WRITE direction, then multiple bios are used.
180 * We choose the number when they are allocated.
181 */
182 struct bio *bios[];
183 /* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/
184};
185
186/* bits for r1bio.state */
187enum r1bio_state {
188 R1BIO_Uptodate,
189 R1BIO_IsSync,
190 R1BIO_Degraded,
191 R1BIO_BehindIO,
192/* Set ReadError on bios that experience a readerror so that
193 * raid1d knows what to do with them.
194 */
195 R1BIO_ReadError,
196/* For write-behind requests, we call bi_end_io when
197 * the last non-write-behind device completes, providing
198 * any write was successful. Otherwise we call when
199 * any write-behind write succeeds, otherwise we call
200 * with failure when last write completes (and all failed).
201 * Record that bi_end_io was called with this flag...
202 */
203 R1BIO_Returned,
204/* If a write for this request means we can clear some
205 * known-bad-block records, we set this flag
206 */
207 R1BIO_MadeGood,
208 R1BIO_WriteError,
209 R1BIO_FailFast,
210};
211
212static inline int sector_to_idx(sector_t sector)
213{
214 return hash_long(sector >> BARRIER_UNIT_SECTOR_BITS,
215 BARRIER_BUCKETS_NR_BITS);
216}
217#endif
1#ifndef _RAID1_H
2#define _RAID1_H
3
4struct raid1_info {
5 struct md_rdev *rdev;
6 sector_t head_position;
7
8 /* When choose the best device for a read (read_balance())
9 * we try to keep sequential reads one the same device
10 */
11 sector_t next_seq_sect;
12 sector_t seq_start;
13};
14
15/*
16 * memory pools need a pointer to the mddev, so they can force an unplug
17 * when memory is tight, and a count of the number of drives that the
18 * pool was allocated for, so they know how much to allocate and free.
19 * mddev->raid_disks cannot be used, as it can change while a pool is active
20 * These two datums are stored in a kmalloced struct.
21 * The 'raid_disks' here is twice the raid_disks in r1conf.
22 * This allows space for each 'real' device can have a replacement in the
23 * second half of the array.
24 */
25
26struct pool_info {
27 struct mddev *mddev;
28 int raid_disks;
29};
30
31struct r1conf {
32 struct mddev *mddev;
33 struct raid1_info *mirrors; /* twice 'raid_disks' to
34 * allow for replacements.
35 */
36 int raid_disks;
37
38 /* During resync, read_balancing is only allowed on the part
39 * of the array that has been resynced. 'next_resync' tells us
40 * where that is.
41 */
42 sector_t next_resync;
43
44 /* When raid1 starts resync, we divide array into four partitions
45 * |---------|--------------|---------------------|-------------|
46 * next_resync start_next_window end_window
47 * start_next_window = next_resync + NEXT_NORMALIO_DISTANCE
48 * end_window = start_next_window + NEXT_NORMALIO_DISTANCE
49 * current_window_requests means the count of normalIO between
50 * start_next_window and end_window.
51 * next_window_requests means the count of normalIO after end_window.
52 * */
53 sector_t start_next_window;
54 int current_window_requests;
55 int next_window_requests;
56
57 spinlock_t device_lock;
58
59 /* list of 'struct r1bio' that need to be processed by raid1d,
60 * whether to retry a read, writeout a resync or recovery
61 * block, or anything else.
62 */
63 struct list_head retry_list;
64
65 /* queue pending writes to be submitted on unplug */
66 struct bio_list pending_bio_list;
67 int pending_count;
68
69 /* for use when syncing mirrors:
70 * We don't allow both normal IO and resync/recovery IO at
71 * the same time - resync/recovery can only happen when there
72 * is no other IO. So when either is active, the other has to wait.
73 * See more details description in raid1.c near raise_barrier().
74 */
75 wait_queue_head_t wait_barrier;
76 spinlock_t resync_lock;
77 int nr_pending;
78 int nr_waiting;
79 int nr_queued;
80 int barrier;
81 int array_frozen;
82
83 /* Set to 1 if a full sync is needed, (fresh device added).
84 * Cleared when a sync completes.
85 */
86 int fullsync;
87
88 /* When the same as mddev->recovery_disabled we don't allow
89 * recovery to be attempted as we expect a read error.
90 */
91 int recovery_disabled;
92
93
94 /* poolinfo contains information about the content of the
95 * mempools - it changes when the array grows or shrinks
96 */
97 struct pool_info *poolinfo;
98 mempool_t *r1bio_pool;
99 mempool_t *r1buf_pool;
100
101 /* temporary buffer to synchronous IO when attempting to repair
102 * a read error.
103 */
104 struct page *tmppage;
105
106
107 /* When taking over an array from a different personality, we store
108 * the new thread here until we fully activate the array.
109 */
110 struct md_thread *thread;
111};
112
113/*
114 * this is our 'private' RAID1 bio.
115 *
116 * it contains information about what kind of IO operations were started
117 * for this RAID1 operation, and about their status:
118 */
119
120struct r1bio {
121 atomic_t remaining; /* 'have we finished' count,
122 * used from IRQ handlers
123 */
124 atomic_t behind_remaining; /* number of write-behind ios remaining
125 * in this BehindIO request
126 */
127 sector_t sector;
128 sector_t start_next_window;
129 int sectors;
130 unsigned long state;
131 struct mddev *mddev;
132 /*
133 * original bio going to /dev/mdx
134 */
135 struct bio *master_bio;
136 /*
137 * if the IO is in READ direction, then this is where we read
138 */
139 int read_disk;
140
141 struct list_head retry_list;
142 /* Next two are only valid when R1BIO_BehindIO is set */
143 struct bio_vec *behind_bvecs;
144 int behind_page_count;
145 /*
146 * if the IO is in WRITE direction, then multiple bios are used.
147 * We choose the number when they are allocated.
148 */
149 struct bio *bios[0];
150 /* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/
151};
152
153/* bits for r1bio.state */
154#define R1BIO_Uptodate 0
155#define R1BIO_IsSync 1
156#define R1BIO_Degraded 2
157#define R1BIO_BehindIO 3
158/* Set ReadError on bios that experience a readerror so that
159 * raid1d knows what to do with them.
160 */
161#define R1BIO_ReadError 4
162/* For write-behind requests, we call bi_end_io when
163 * the last non-write-behind device completes, providing
164 * any write was successful. Otherwise we call when
165 * any write-behind write succeeds, otherwise we call
166 * with failure when last write completes (and all failed).
167 * Record that bi_end_io was called with this flag...
168 */
169#define R1BIO_Returned 6
170/* If a write for this request means we can clear some
171 * known-bad-block records, we set this flag
172 */
173#define R1BIO_MadeGood 7
174#define R1BIO_WriteError 8
175
176extern int md_raid1_congested(struct mddev *mddev, int bits);
177
178#endif