xref: /openbmc/linux/drivers/md/raid1.h (revision 1c2dd16a)
1 #ifndef _RAID1_H
2 #define _RAID1_H
3 
4 /*
5  * each barrier unit size is 64MB fow now
6  * note: it must be larger than RESYNC_DEPTH
7  */
8 #define BARRIER_UNIT_SECTOR_BITS	17
9 #define BARRIER_UNIT_SECTOR_SIZE	(1<<17)
10 /*
11  * In struct r1conf, the following members are related to I/O barrier
12  * buckets,
13  *	atomic_t	*nr_pending;
14  *	atomic_t	*nr_waiting;
15  *	atomic_t	*nr_queued;
16  *	atomic_t	*barrier;
17  * Each of them points to array of atomic_t variables, each array is
18  * designed to have BARRIER_BUCKETS_NR elements and occupy a single
19  * memory page. The data width of atomic_t variables is 4 bytes, equal
20  * to 1<<(ilog2(sizeof(atomic_t))), BARRIER_BUCKETS_NR_BITS is defined
21  * as (PAGE_SHIFT - ilog2(sizeof(int))) to make sure an array of
22  * atomic_t variables with BARRIER_BUCKETS_NR elements just exactly
23  * occupies a single memory page.
24  */
25 #define BARRIER_BUCKETS_NR_BITS		(PAGE_SHIFT - ilog2(sizeof(atomic_t)))
26 #define BARRIER_BUCKETS_NR		(1<<BARRIER_BUCKETS_NR_BITS)
27 
28 struct raid1_info {
29 	struct md_rdev	*rdev;
30 	sector_t	head_position;
31 
32 	/* When choose the best device for a read (read_balance())
33 	 * we try to keep sequential reads one the same device
34 	 */
35 	sector_t	next_seq_sect;
36 	sector_t	seq_start;
37 };
38 
39 /*
40  * memory pools need a pointer to the mddev, so they can force an unplug
41  * when memory is tight, and a count of the number of drives that the
42  * pool was allocated for, so they know how much to allocate and free.
43  * mddev->raid_disks cannot be used, as it can change while a pool is active
44  * These two datums are stored in a kmalloced struct.
45  * The 'raid_disks' here is twice the raid_disks in r1conf.
46  * This allows space for each 'real' device can have a replacement in the
47  * second half of the array.
48  */
49 
50 struct pool_info {
51 	struct mddev *mddev;
52 	int	raid_disks;
53 };
54 
55 struct r1conf {
56 	struct mddev		*mddev;
57 	struct raid1_info	*mirrors;	/* twice 'raid_disks' to
58 						 * allow for replacements.
59 						 */
60 	int			raid_disks;
61 
62 	spinlock_t		device_lock;
63 
64 	/* list of 'struct r1bio' that need to be processed by raid1d,
65 	 * whether to retry a read, writeout a resync or recovery
66 	 * block, or anything else.
67 	 */
68 	struct list_head	retry_list;
69 	/* A separate list of r1bio which just need raid_end_bio_io called.
70 	 * This mustn't happen for writes which had any errors if the superblock
71 	 * needs to be written.
72 	 */
73 	struct list_head	bio_end_io_list;
74 
75 	/* queue pending writes to be submitted on unplug */
76 	struct bio_list		pending_bio_list;
77 	int			pending_count;
78 
79 	/* for use when syncing mirrors:
80 	 * We don't allow both normal IO and resync/recovery IO at
81 	 * the same time - resync/recovery can only happen when there
82 	 * is no other IO.  So when either is active, the other has to wait.
83 	 * See more details description in raid1.c near raise_barrier().
84 	 */
85 	wait_queue_head_t	wait_barrier;
86 	spinlock_t		resync_lock;
87 	atomic_t		*nr_pending;
88 	atomic_t		*nr_waiting;
89 	atomic_t		*nr_queued;
90 	atomic_t		*barrier;
91 	int			array_frozen;
92 
93 	/* Set to 1 if a full sync is needed, (fresh device added).
94 	 * Cleared when a sync completes.
95 	 */
96 	int			fullsync;
97 
98 	/* When the same as mddev->recovery_disabled we don't allow
99 	 * recovery to be attempted as we expect a read error.
100 	 */
101 	int			recovery_disabled;
102 
103 	/* poolinfo contains information about the content of the
104 	 * mempools - it changes when the array grows or shrinks
105 	 */
106 	struct pool_info	*poolinfo;
107 	mempool_t		*r1bio_pool;
108 	mempool_t		*r1buf_pool;
109 
110 	/* temporary buffer to synchronous IO when attempting to repair
111 	 * a read error.
112 	 */
113 	struct page		*tmppage;
114 
115 	/* When taking over an array from a different personality, we store
116 	 * the new thread here until we fully activate the array.
117 	 */
118 	struct md_thread	*thread;
119 
120 	/* Keep track of cluster resync window to send to other
121 	 * nodes.
122 	 */
123 	sector_t		cluster_sync_low;
124 	sector_t		cluster_sync_high;
125 
126 };
127 
128 /*
129  * this is our 'private' RAID1 bio.
130  *
131  * it contains information about what kind of IO operations were started
132  * for this RAID1 operation, and about their status:
133  */
134 
135 struct r1bio {
136 	atomic_t		remaining; /* 'have we finished' count,
137 					    * used from IRQ handlers
138 					    */
139 	atomic_t		behind_remaining; /* number of write-behind ios remaining
140 						 * in this BehindIO request
141 						 */
142 	sector_t		sector;
143 	int			sectors;
144 	unsigned long		state;
145 	struct mddev		*mddev;
146 	/*
147 	 * original bio going to /dev/mdx
148 	 */
149 	struct bio		*master_bio;
150 	/*
151 	 * if the IO is in READ direction, then this is where we read
152 	 */
153 	int			read_disk;
154 
155 	struct list_head	retry_list;
156 	/* Next two are only valid when R1BIO_BehindIO is set */
157 	struct bio_vec		*behind_bvecs;
158 	int			behind_page_count;
159 	/*
160 	 * if the IO is in WRITE direction, then multiple bios are used.
161 	 * We choose the number when they are allocated.
162 	 */
163 	struct bio		*bios[0];
164 	/* DO NOT PUT ANY NEW FIELDS HERE - bios array is contiguously alloced*/
165 };
166 
167 /* bits for r1bio.state */
168 enum r1bio_state {
169 	R1BIO_Uptodate,
170 	R1BIO_IsSync,
171 	R1BIO_Degraded,
172 	R1BIO_BehindIO,
173 /* Set ReadError on bios that experience a readerror so that
174  * raid1d knows what to do with them.
175  */
176 	R1BIO_ReadError,
177 /* For write-behind requests, we call bi_end_io when
178  * the last non-write-behind device completes, providing
179  * any write was successful.  Otherwise we call when
180  * any write-behind write succeeds, otherwise we call
181  * with failure when last write completes (and all failed).
182  * Record that bi_end_io was called with this flag...
183  */
184 	R1BIO_Returned,
185 /* If a write for this request means we can clear some
186  * known-bad-block records, we set this flag
187  */
188 	R1BIO_MadeGood,
189 	R1BIO_WriteError,
190 	R1BIO_FailFast,
191 };
192 
193 static inline int sector_to_idx(sector_t sector)
194 {
195 	return hash_long(sector >> BARRIER_UNIT_SECTOR_BITS,
196 			 BARRIER_BUCKETS_NR_BITS);
197 }
198 #endif
199