projects/linux·blob master · refs

194 lines · 5.5 KiB

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
52struct r1conf {
53 struct mddev *mddev;
54 struct raid1_info *mirrors; /* twice 'raid_disks' to
55 * allow for replacements.
56 */
57 int raid_disks;
58 int nonrot_disks;
59
60 spinlock_t device_lock;
61
62 /* list of 'struct r1bio' that need to be processed by raid1d,
63 * whether to retry a read, writeout a resync or recovery
64 * block, or anything else.
65 */
66 struct list_head retry_list;
67 /* A separate list of r1bio which just need raid_end_bio_io called.
68 * This mustn't happen for writes which had any errors if the superblock
69 * needs to be written.
70 */
71 struct list_head bio_end_io_list;
72
73 /* queue pending writes to be submitted on unplug */
74 struct bio_list pending_bio_list;
75
76 /* for use when syncing mirrors:
77 * We don't allow both normal IO and resync/recovery IO at
78 * the same time - resync/recovery can only happen when there
79 * is no other IO. So when either is active, the other has to wait.
80 * See more details description in raid1.c near raise_barrier().
81 */
82 wait_queue_head_t wait_barrier;
83 spinlock_t resync_lock;
84 atomic_t nr_sync_pending;
85 atomic_t *nr_pending;
86 atomic_t *nr_waiting;
87 atomic_t *nr_queued;
88 atomic_t *barrier;
89 int array_frozen;
90
91 /* Set to 1 if a full sync is needed, (fresh device added).
92 * Cleared when a sync completes.
93 */
94 int fullsync;
95
96 mempool_t *r1bio_pool;
97 mempool_t r1buf_pool;
98
99 struct bio_set bio_split;
100
101 /* temporary buffer to synchronous IO when attempting to repair
102 * a read error.
103 */
104 struct page *tmppage;
105
106 /* When taking over an array from a different personality, we store
107 * the new thread here until we fully activate the array.
108 */
109 struct md_thread __rcu *thread;
110
111 /* Keep track of cluster resync window to send to other
112 * nodes.
113 */
114 sector_t cluster_sync_low;
115 sector_t cluster_sync_high;
116
117};
118
119/*
120 * this is our 'private' RAID1 bio.
121 *
122 * it contains information about what kind of IO operations were started
123 * for this RAID1 operation, and about their status:
124 */
125
126struct r1bio {
127 atomic_t remaining; /* 'have we finished' count,
128 * used from IRQ handlers
129 */
130 atomic_t behind_remaining; /* number of write-behind ios remaining
131 * in this BehindIO request
132 */
133 sector_t sector;
134 int sectors;
135 unsigned long state;
136 struct mddev *mddev;
137 /*
138 * original bio going to /dev/mdx
139 */
140 struct bio *master_bio;
141 /*
142 * if the IO is in READ direction, then this is where we read
143 */
144 int read_disk;
145
146 struct list_head retry_list;
147
148 /*
149 * When R1BIO_BehindIO is set, we store pages for write behind
150 * in behind_master_bio.
151 */
152 struct bio *behind_master_bio;
153
154 /*
155 * if the IO is in WRITE direction, then multiple bios are used.
156 * We choose the number when they are allocated.
157 */
158 struct bio *bios[];
159 /* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/
160};
161
162/* bits for r1bio.state */
163enum r1bio_state {
164 R1BIO_Uptodate,
165 R1BIO_IsSync,
166 R1BIO_BehindIO,
167/* Set ReadError on bios that experience a readerror so that
168 * raid1d knows what to do with them.
169 */
170 R1BIO_ReadError,
171/* For write-behind requests, we call bi_end_io when
172 * the last non-write-behind device completes, providing
173 * any write was successful. Otherwise we call when
174 * any write-behind write succeeds, otherwise we call
175 * with failure when last write completes (and all failed).
176 *
177 * And for bio_split errors, record that bi_end_io was called
178 * with this flag...
179 */
180 R1BIO_Returned,
181/* If a write for this request means we can clear some
182 * known-bad-block records, we set this flag
183 */
184 R1BIO_MadeGood,
185 R1BIO_WriteError,
186 R1BIO_FailFast,
187};
188
189static inline int sector_to_idx(sector_t sector)
190{
191 return hash_long(sector >> BARRIER_UNIT_SECTOR_BITS,
192 BARRIER_BUCKETS_NR_BITS);
193}
194#endif