Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
1da177e4 LT |
2 | #ifndef _RAID1_H |
3 | #define _RAID1_H | |
4 | ||
fd76863e | 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, | |
824e47da | 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. | |
fd76863e | 25 | */ |
824e47da | 26 | #define BARRIER_BUCKETS_NR_BITS (PAGE_SHIFT - ilog2(sizeof(atomic_t))) |
fd76863e | 27 | #define BARRIER_BUCKETS_NR (1<<BARRIER_BUCKETS_NR_BITS) |
28 | ||
f2785b52 N |
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 | ||
0eaf822c | 41 | struct raid1_info { |
3cb03002 | 42 | struct md_rdev *rdev; |
1da177e4 | 43 | sector_t head_position; |
be4d3280 SL |
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; | |
12cee5a8 | 49 | sector_t seq_start; |
1da177e4 LT |
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. | |
8f19ccb2 N |
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. | |
1da177e4 LT |
61 | */ |
62 | ||
63 | struct pool_info { | |
fd01b88c | 64 | struct mddev *mddev; |
1da177e4 LT |
65 | int raid_disks; |
66 | }; | |
67 | ||
e8096360 | 68 | struct r1conf { |
fd01b88c | 69 | struct mddev *mddev; |
0eaf822c | 70 | struct raid1_info *mirrors; /* twice 'raid_disks' to |
8f19ccb2 N |
71 | * allow for replacements. |
72 | */ | |
1da177e4 | 73 | int raid_disks; |
ce550c20 | 74 | |
1da177e4 LT |
75 | spinlock_t device_lock; |
76 | ||
9f2c9d12 N |
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. | |
ce550c20 | 80 | */ |
1da177e4 | 81 | struct list_head retry_list; |
55ce74d4 N |
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; | |
191ea9b2 | 87 | |
ce550c20 N |
88 | /* queue pending writes to be submitted on unplug */ |
89 | struct bio_list pending_bio_list; | |
1da177e4 | 90 | |
ce550c20 N |
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; | |
1da177e4 | 98 | spinlock_t resync_lock; |
43ac9b84 | 99 | atomic_t nr_sync_pending; |
824e47da | 100 | atomic_t *nr_pending; |
101 | atomic_t *nr_waiting; | |
102 | atomic_t *nr_queued; | |
103 | atomic_t *barrier; | |
b364e3d0 | 104 | int array_frozen; |
1da177e4 | 105 | |
ce550c20 N |
106 | /* Set to 1 if a full sync is needed, (fresh device added). |
107 | * Cleared when a sync completes. | |
108 | */ | |
109 | int fullsync; | |
1da177e4 | 110 | |
ce550c20 N |
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; | |
1da177e4 | 115 | |
ce550c20 N |
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; | |
afeee514 KO |
120 | mempool_t r1bio_pool; |
121 | mempool_t r1buf_pool; | |
709ae487 | 122 | |
afeee514 | 123 | struct bio_set bio_split; |
c230e7e5 | 124 | |
ce550c20 N |
125 | /* temporary buffer to synchronous IO when attempting to repair |
126 | * a read error. | |
127 | */ | |
128 | struct page *tmppage; | |
129 | ||
709ae487 N |
130 | /* When taking over an array from a different personality, we store |
131 | * the new thread here until we fully activate the array. | |
132 | */ | |
44693154 | 133 | struct md_thread __rcu *thread; |
c40f341f GR |
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 | ||
1da177e4 LT |
141 | }; |
142 | ||
1da177e4 LT |
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 | ||
9f2c9d12 | 150 | struct r1bio { |
1da177e4 LT |
151 | atomic_t remaining; /* 'have we finished' count, |
152 | * used from IRQ handlers | |
153 | */ | |
4b6d287f N |
154 | atomic_t behind_remaining; /* number of write-behind ios remaining |
155 | * in this BehindIO request | |
156 | */ | |
1da177e4 LT |
157 | sector_t sector; |
158 | int sectors; | |
159 | unsigned long state; | |
fd01b88c | 160 | struct mddev *mddev; |
1da177e4 LT |
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; | |
841c1316 ML |
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 | ||
1da177e4 LT |
178 | /* |
179 | * if the IO is in WRITE direction, then multiple bios are used. | |
180 | * We choose the number when they are allocated. | |
181 | */ | |
358369f0 | 182 | struct bio *bios[]; |
191ea9b2 | 183 | /* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/ |
1da177e4 LT |
184 | }; |
185 | ||
186 | /* bits for r1bio.state */ | |
be306c29 N |
187 | enum r1bio_state { |
188 | R1BIO_Uptodate, | |
189 | R1BIO_IsSync, | |
190 | R1BIO_Degraded, | |
191 | R1BIO_BehindIO, | |
d2eb35ac N |
192 | /* Set ReadError on bios that experience a readerror so that |
193 | * raid1d knows what to do with them. | |
194 | */ | |
be306c29 | 195 | R1BIO_ReadError, |
4b6d287f N |
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 | */ | |
be306c29 | 203 | R1BIO_Returned, |
4367af55 N |
204 | /* If a write for this request means we can clear some |
205 | * known-bad-block records, we set this flag | |
206 | */ | |
be306c29 N |
207 | R1BIO_MadeGood, |
208 | R1BIO_WriteError, | |
2e52d449 | 209 | R1BIO_FailFast, |
be306c29 | 210 | }; |
fd76863e | 211 | |
212 | static inline int sector_to_idx(sector_t sector) | |
213 | { | |
214 | return hash_long(sector >> BARRIER_UNIT_SECTOR_BITS, | |
215 | BARRIER_BUCKETS_NR_BITS); | |
216 | } | |
1da177e4 | 217 | #endif |