md/raid10: ensure device failure recorded before write request returns.
[pandora-kernel.git] / drivers / md / raid10.h
1 #ifndef _RAID10_H
2 #define _RAID10_H
3
4 struct raid10_info {
5         struct md_rdev  *rdev, *replacement;
6         sector_t        head_position;
7         int             recovery_disabled;      /* matches
8                                                  * mddev->recovery_disabled
9                                                  * when we shouldn't try
10                                                  * recovering this device.
11                                                  */
12 };
13
14 struct r10conf {
15         struct mddev            *mddev;
16         struct raid10_info      *mirrors;
17         struct raid10_info      *mirrors_new, *mirrors_old;
18         spinlock_t              device_lock;
19
20         /* geometry */
21         struct geom {
22                 int             raid_disks;
23                 int             near_copies;  /* number of copies laid out
24                                                * raid0 style */
25                 int             far_copies;   /* number of copies laid out
26                                                * at large strides across drives
27                                                */
28                 int             far_offset;   /* far_copies are offset by 1
29                                                * stripe instead of many
30                                                */
31                 sector_t        stride;       /* distance between far copies.
32                                                * This is size / far_copies unless
33                                                * far_offset, in which case it is
34                                                * 1 stripe.
35                                                */
36                 int             far_set_size; /* The number of devices in a set,
37                                                * where a 'set' are devices that
38                                                * contain far/offset copies of
39                                                * each other.
40                                                */
41                 int             chunk_shift; /* shift from chunks to sectors */
42                 sector_t        chunk_mask;
43         } prev, geo;
44         int                     copies;       /* near_copies * far_copies.
45                                                * must be <= raid_disks
46                                                */
47
48         sector_t                dev_sectors;  /* temp copy of
49                                                * mddev->dev_sectors */
50         sector_t                reshape_progress;
51         sector_t                reshape_safe;
52         unsigned long           reshape_checkpoint;
53         sector_t                offset_diff;
54
55         struct list_head        retry_list;
56         /* A separate list of r1bio which just need raid_end_bio_io called.
57          * This mustn't happen for writes which had any errors if the superblock
58          * needs to be written.
59          */
60         struct list_head        bio_end_io_list;
61
62         /* queue pending writes and submit them on unplug */
63         struct bio_list         pending_bio_list;
64         int                     pending_count;
65
66         spinlock_t              resync_lock;
67         int                     nr_pending;
68         int                     nr_waiting;
69         int                     nr_queued;
70         int                     barrier;
71         sector_t                next_resync;
72         int                     fullsync;  /* set to 1 if a full sync is needed,
73                                             * (fresh device added).
74                                             * Cleared when a sync completes.
75                                             */
76         int                     have_replacement; /* There is at least one
77                                                    * replacement device.
78                                                    */
79         wait_queue_head_t       wait_barrier;
80
81         mempool_t               *r10bio_pool;
82         mempool_t               *r10buf_pool;
83         struct page             *tmppage;
84
85         /* When taking over an array from a different personality, we store
86          * the new thread here until we fully activate the array.
87          */
88         struct md_thread        *thread;
89 };
90
91 /*
92  * this is our 'private' RAID10 bio.
93  *
94  * it contains information about what kind of IO operations were started
95  * for this RAID10 operation, and about their status:
96  */
97
98 struct r10bio {
99         atomic_t                remaining; /* 'have we finished' count,
100                                             * used from IRQ handlers
101                                             */
102         sector_t                sector; /* virtual sector number */
103         int                     sectors;
104         unsigned long           state;
105         struct mddev            *mddev;
106         /*
107          * original bio going to /dev/mdx
108          */
109         struct bio              *master_bio;
110         /*
111          * if the IO is in READ direction, then this is where we read
112          */
113         int                     read_slot;
114
115         struct list_head        retry_list;
116         /*
117          * if the IO is in WRITE direction, then multiple bios are used,
118          * one for each copy.
119          * When resyncing we also use one for each copy.
120          * When reconstructing, we use 2 bios, one for read, one for write.
121          * We choose the number when they are allocated.
122          * We sometimes need an extra bio to write to the replacement.
123          */
124         struct r10dev {
125                 struct bio      *bio;
126                 union {
127                         struct bio      *repl_bio; /* used for resync and
128                                                     * writes */
129                         struct md_rdev  *rdev;     /* used for reads
130                                                     * (read_slot >= 0) */
131                 };
132                 sector_t        addr;
133                 int             devnum;
134         } devs[0];
135 };
136
137 /* bits for r10bio.state */
138 enum r10bio_state {
139         R10BIO_Uptodate,
140         R10BIO_IsSync,
141         R10BIO_IsRecover,
142         R10BIO_IsReshape,
143         R10BIO_Degraded,
144 /* Set ReadError on bios that experience a read error
145  * so that raid10d knows what to do with them.
146  */
147         R10BIO_ReadError,
148 /* If a write for this request means we can clear some
149  * known-bad-block records, we set this flag.
150  */
151         R10BIO_MadeGood,
152         R10BIO_WriteError,
153 /* During a reshape we might be performing IO on the
154  * 'previous' part of the array, in which case this
155  * flag is set
156  */
157         R10BIO_Previous,
158 };
159 #endif