xref: /linux/drivers/md/raid1.h (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
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 struct raid1_info {
30 	struct md_rdev	*rdev;
31 	sector_t	head_position;
32 
33 	/* When choose the best device for a read (read_balance())
34 	 * we try to keep sequential reads one the same device
35 	 */
36 	sector_t	next_seq_sect;
37 	sector_t	seq_start;
38 };
39 
40 /*
41  * memory pools need a pointer to the mddev, so they can force an unplug
42  * when memory is tight, and a count of the number of drives that the
43  * pool was allocated for, so they know how much to allocate and free.
44  * mddev->raid_disks cannot be used, as it can change while a pool is active
45  * These two datums are stored in a kmalloced struct.
46  * The 'raid_disks' here is twice the raid_disks in r1conf.
47  * This allows space for each 'real' device can have a replacement in the
48  * second half of the array.
49  */
50 
51 struct pool_info {
52 	struct mddev *mddev;
53 	int	raid_disks;
54 };
55 
56 struct r1conf {
57 	struct mddev		*mddev;
58 	struct raid1_info	*mirrors;	/* twice 'raid_disks' to
59 						 * allow for replacements.
60 						 */
61 	int			raid_disks;
62 
63 	spinlock_t		device_lock;
64 
65 	/* list of 'struct r1bio' that need to be processed by raid1d,
66 	 * whether to retry a read, writeout a resync or recovery
67 	 * block, or anything else.
68 	 */
69 	struct list_head	retry_list;
70 	/* A separate list of r1bio which just need raid_end_bio_io called.
71 	 * This mustn't happen for writes which had any errors if the superblock
72 	 * needs to be written.
73 	 */
74 	struct list_head	bio_end_io_list;
75 
76 	/* queue pending writes to be submitted on unplug */
77 	struct bio_list		pending_bio_list;
78 	int			pending_count;
79 
80 	/* for use when syncing mirrors:
81 	 * We don't allow both normal IO and resync/recovery IO at
82 	 * the same time - resync/recovery can only happen when there
83 	 * is no other IO.  So when either is active, the other has to wait.
84 	 * See more details description in raid1.c near raise_barrier().
85 	 */
86 	wait_queue_head_t	wait_barrier;
87 	spinlock_t		resync_lock;
88 	atomic_t		nr_sync_pending;
89 	atomic_t		*nr_pending;
90 	atomic_t		*nr_waiting;
91 	atomic_t		*nr_queued;
92 	atomic_t		*barrier;
93 	int			array_frozen;
94 
95 	/* Set to 1 if a full sync is needed, (fresh device added).
96 	 * Cleared when a sync completes.
97 	 */
98 	int			fullsync;
99 
100 	/* When the same as mddev->recovery_disabled we don't allow
101 	 * recovery to be attempted as we expect a read error.
102 	 */
103 	int			recovery_disabled;
104 
105 	/* poolinfo contains information about the content of the
106 	 * mempools - it changes when the array grows or shrinks
107 	 */
108 	struct pool_info	*poolinfo;
109 	mempool_t		*r1bio_pool;
110 	mempool_t		*r1buf_pool;
111 
112 	struct bio_set		*bio_split;
113 
114 	/* temporary buffer to synchronous IO when attempting to repair
115 	 * a read error.
116 	 */
117 	struct page		*tmppage;
118 
119 	/* When taking over an array from a different personality, we store
120 	 * the new thread here until we fully activate the array.
121 	 */
122 	struct md_thread	*thread;
123 
124 	/* Keep track of cluster resync window to send to other
125 	 * nodes.
126 	 */
127 	sector_t		cluster_sync_low;
128 	sector_t		cluster_sync_high;
129 
130 };
131 
132 /*
133  * this is our 'private' RAID1 bio.
134  *
135  * it contains information about what kind of IO operations were started
136  * for this RAID1 operation, and about their status:
137  */
138 
139 struct r1bio {
140 	atomic_t		remaining; /* 'have we finished' count,
141 					    * used from IRQ handlers
142 					    */
143 	atomic_t		behind_remaining; /* number of write-behind ios remaining
144 						 * in this BehindIO request
145 						 */
146 	sector_t		sector;
147 	int			sectors;
148 	unsigned long		state;
149 	struct mddev		*mddev;
150 	/*
151 	 * original bio going to /dev/mdx
152 	 */
153 	struct bio		*master_bio;
154 	/*
155 	 * if the IO is in READ direction, then this is where we read
156 	 */
157 	int			read_disk;
158 
159 	struct list_head	retry_list;
160 
161 	/*
162 	 * When R1BIO_BehindIO is set, we store pages for write behind
163 	 * in behind_master_bio.
164 	 */
165 	struct bio		*behind_master_bio;
166 
167 	/*
168 	 * if the IO is in WRITE direction, then multiple bios are used.
169 	 * We choose the number when they are allocated.
170 	 */
171 	struct bio		*bios[0];
172 	/* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/
173 };
174 
175 /* bits for r1bio.state */
176 enum r1bio_state {
177 	R1BIO_Uptodate,
178 	R1BIO_IsSync,
179 	R1BIO_Degraded,
180 	R1BIO_BehindIO,
181 /* Set ReadError on bios that experience a readerror so that
182  * raid1d knows what to do with them.
183  */
184 	R1BIO_ReadError,
185 /* For write-behind requests, we call bi_end_io when
186  * the last non-write-behind device completes, providing
187  * any write was successful.  Otherwise we call when
188  * any write-behind write succeeds, otherwise we call
189  * with failure when last write completes (and all failed).
190  * Record that bi_end_io was called with this flag...
191  */
192 	R1BIO_Returned,
193 /* If a write for this request means we can clear some
194  * known-bad-block records, we set this flag
195  */
196 	R1BIO_MadeGood,
197 	R1BIO_WriteError,
198 	R1BIO_FailFast,
199 };
200 
201 static inline int sector_to_idx(sector_t sector)
202 {
203 	return hash_long(sector >> BARRIER_UNIT_SECTOR_BITS,
204 			 BARRIER_BUCKETS_NR_BITS);
205 }
206 #endif
207