blob: 03636d7918fef80b565b0e1141247be35c79bd88 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _RAID5_H
2#define _RAID5_H
3
4#include <linux/raid/md.h>
5#include <linux/raid/xor.h>
6
7/*
8 *
9 * Each stripe contains one buffer per disc. Each buffer can be in
10 * one of a number of states stored in "flags". Changes between
11 * these states happen *almost* exclusively under a per-stripe
12 * spinlock. Some very specific changes can happen in bi_end_io, and
13 * these are not protected by the spin lock.
14 *
15 * The flag bits that are used to represent these states are:
16 * R5_UPTODATE and R5_LOCKED
17 *
18 * State Empty == !UPTODATE, !LOCK
19 * We have no data, and there is no active request
20 * State Want == !UPTODATE, LOCK
21 * A read request is being submitted for this block
22 * State Dirty == UPTODATE, LOCK
23 * Some new data is in this buffer, and it is being written out
24 * State Clean == UPTODATE, !LOCK
25 * We have valid data which is the same as on disc
26 *
27 * The possible state transitions are:
28 *
29 * Empty -> Want - on read or write to get old data for parity calc
30 * Empty -> Dirty - on compute_parity to satisfy write/sync request.(RECONSTRUCT_WRITE)
31 * Empty -> Clean - on compute_block when computing a block for failed drive
32 * Want -> Empty - on failed read
33 * Want -> Clean - on successful completion of read request
34 * Dirty -> Clean - on successful completion of write request
35 * Dirty -> Clean - on failed write
36 * Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
37 *
38 * The Want->Empty, Want->Clean, Dirty->Clean, transitions
39 * all happen in b_end_io at interrupt time.
40 * Each sets the Uptodate bit before releasing the Lock bit.
41 * This leaves one multi-stage transition:
42 * Want->Dirty->Clean
43 * This is safe because thinking that a Clean buffer is actually dirty
44 * will at worst delay some action, and the stripe will be scheduled
45 * for attention after the transition is complete.
46 *
47 * There is one possibility that is not covered by these states. That
48 * is if one drive has failed and there is a spare being rebuilt. We
49 * can't distinguish between a clean block that has been generated
50 * from parity calculations, and a clean block that has been
51 * successfully written to the spare ( or to parity when resyncing).
52 * To distingush these states we have a stripe bit STRIPE_INSYNC that
53 * is set whenever a write is scheduled to the spare, or to the parity
54 * disc if there is no spare. A sync request clears this bit, and
55 * when we find it set with no buffers locked, we know the sync is
56 * complete.
57 *
58 * Buffers for the md device that arrive via make_request are attached
59 * to the appropriate stripe in one of two lists linked on b_reqnext.
60 * One list (bh_read) for read requests, one (bh_write) for write.
61 * There should never be more than one buffer on the two lists
62 * together, but we are not guaranteed of that so we allow for more.
63 *
64 * If a buffer is on the read list when the associated cache buffer is
65 * Uptodate, the data is copied into the read buffer and it's b_end_io
66 * routine is called. This may happen in the end_request routine only
67 * if the buffer has just successfully been read. end_request should
68 * remove the buffers from the list and then set the Uptodate bit on
69 * the buffer. Other threads may do this only if they first check
70 * that the Uptodate bit is set. Once they have checked that they may
71 * take buffers off the read queue.
72 *
73 * When a buffer on the write list is committed for write it is copied
74 * into the cache buffer, which is then marked dirty, and moved onto a
75 * third list, the written list (bh_written). Once both the parity
76 * block and the cached buffer are successfully written, any buffer on
77 * a written list can be returned with b_end_io.
78 *
79 * The write list and read list both act as fifos. The read list is
80 * protected by the device_lock. The write and written lists are
81 * protected by the stripe lock. The device_lock, which can be
82 * claimed while the stipe lock is held, is only for list
83 * manipulations and will only be held for a very short time. It can
84 * be claimed from interrupts.
85 *
86 *
87 * Stripes in the stripe cache can be on one of two lists (or on
88 * neither). The "inactive_list" contains stripes which are not
89 * currently being used for any request. They can freely be reused
90 * for another stripe. The "handle_list" contains stripes that need
91 * to be handled in some way. Both of these are fifo queues. Each
92 * stripe is also (potentially) linked to a hash bucket in the hash
93 * table so that it can be found by sector number. Stripes that are
94 * not hashed must be on the inactive_list, and will normally be at
95 * the front. All stripes start life this way.
96 *
97 * The inactive_list, handle_list and hash bucket lists are all protected by the
98 * device_lock.
99 * - stripes on the inactive_list never have their stripe_lock held.
100 * - stripes have a reference counter. If count==0, they are on a list.
101 * - If a stripe might need handling, STRIPE_HANDLE is set.
102 * - When refcount reaches zero, then if STRIPE_HANDLE it is put on
103 * handle_list else inactive_list
104 *
105 * This, combined with the fact that STRIPE_HANDLE is only ever
106 * cleared while a stripe has a non-zero count means that if the
107 * refcount is 0 and STRIPE_HANDLE is set, then it is on the
108 * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
109 * the stripe is on inactive_list.
110 *
111 * The possible transitions are:
112 * activate an unhashed/inactive stripe (get_active_stripe())
113 * lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
114 * activate a hashed, possibly active stripe (get_active_stripe())
115 * lockdev check-hash if(!cnt++)unlink-stripe unlockdev
116 * attach a request to an active stripe (add_stripe_bh())
117 * lockdev attach-buffer unlockdev
118 * handle a stripe (handle_stripe())
119 * lockstripe clrSTRIPE_HANDLE ... (lockdev check-buffers unlockdev) .. change-state .. record io needed unlockstripe schedule io
120 * release an active stripe (release_stripe())
121 * lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
122 *
123 * The refcount counts each thread that have activated the stripe,
124 * plus raid5d if it is handling it, plus one for each active request
125 * on a cached buffer.
126 */
127
128struct stripe_head {
NeilBrownfccddba2006-01-06 00:20:33 -0800129 struct hlist_node hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 struct list_head lru; /* inactive_list or handle_list */
131 struct raid5_private_data *raid_conf;
132 sector_t sector; /* sector of this row */
133 int pd_idx; /* parity disk index */
134 unsigned long state; /* state flags */
135 atomic_t count; /* nr of active thread/requests */
136 spinlock_t lock;
NeilBrown72626682005-09-09 16:23:54 -0700137 int bm_seq; /* sequence number for bitmap flushes */
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800138 int disks; /* disks in stripe */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 struct r5dev {
140 struct bio req;
141 struct bio_vec vec;
142 struct page *page;
143 struct bio *toread, *towrite, *written;
144 sector_t sector; /* sector of this page */
145 unsigned long flags;
146 } dev[1]; /* allocated with extra space depending of RAID geometry */
147};
148/* Flags */
149#define R5_UPTODATE 0 /* page contains current data */
150#define R5_LOCKED 1 /* IO has been submitted on "req" */
151#define R5_OVERWRITE 2 /* towrite covers whole page */
152/* and some that are internal to handle_stripe */
153#define R5_Insync 3 /* rdev && rdev->in_sync at start */
154#define R5_Wantread 4 /* want to schedule a read */
155#define R5_Wantwrite 5
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156#define R5_Overlap 7 /* There is a pending overlapping request on this block */
NeilBrown4e5314b2005-11-08 21:39:22 -0800157#define R5_ReadError 8 /* seen a read error here recently */
158#define R5_ReWrite 9 /* have tried to over-write the readerror */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
NeilBrownccfcc3c2006-03-27 01:18:09 -0800160#define R5_Expanded 10 /* This block now has post-expand data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161/*
162 * Write method
163 */
164#define RECONSTRUCT_WRITE 1
165#define READ_MODIFY_WRITE 2
166/* not a write method, but a compute_parity mode */
167#define CHECK_PARITY 3
168
169/*
170 * Stripe state
171 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172#define STRIPE_HANDLE 2
173#define STRIPE_SYNCING 3
174#define STRIPE_INSYNC 4
175#define STRIPE_PREREAD_ACTIVE 5
176#define STRIPE_DELAYED 6
NeilBrown72626682005-09-09 16:23:54 -0700177#define STRIPE_DEGRADED 7
178#define STRIPE_BIT_DELAY 8
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800179#define STRIPE_EXPANDING 9
NeilBrownccfcc3c2006-03-27 01:18:09 -0800180#define STRIPE_EXPAND_SOURCE 10
181#define STRIPE_EXPAND_READY 11
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182/*
183 * Plugging:
184 *
185 * To improve write throughput, we need to delay the handling of some
186 * stripes until there has been a chance that several write requests
187 * for the one stripe have all been collected.
188 * In particular, any write request that would require pre-reading
189 * is put on a "delayed" queue until there are no stripes currently
190 * in a pre-read phase. Further, if the "delayed" queue is empty when
191 * a stripe is put on it then we "plug" the queue and do not process it
192 * until an unplug call is made. (the unplug_io_fn() is called).
193 *
194 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
195 * it to the count of prereading stripes.
196 * When write is initiated, or the stripe refcnt == 0 (just in case) we
197 * clear the PREREAD_ACTIVE flag and decrement the count
NeilBrownb5c124a2006-10-03 01:15:45 -0700198 * Whenever the 'handle' queue is empty and the device is not plugged, we
199 * move any strips from delayed to handle and clear the DELAYED flag and set
200 * PREREAD_ACTIVE.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 * In stripe_handle, if we find pre-reading is necessary, we do it if
202 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
203 * HANDLE gets cleared if stripe_handle leave nothing locked.
204 */
205
206
207struct disk_info {
208 mdk_rdev_t *rdev;
209};
210
211struct raid5_private_data {
NeilBrownfccddba2006-01-06 00:20:33 -0800212 struct hlist_head *stripe_hashtbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 mddev_t *mddev;
214 struct disk_info *spare;
215 int chunk_size, level, algorithm;
NeilBrown16a53ec2006-06-26 00:27:38 -0700216 int max_degraded;
NeilBrown02c2de82006-10-03 01:15:47 -0700217 int raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 int max_nr_stripes;
219
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800220 /* used during an expand */
221 sector_t expand_progress; /* MaxSector when no expand happening */
NeilBrownb578d552006-03-27 01:18:12 -0800222 sector_t expand_lo; /* from here up to expand_progress it out-of-bounds
223 * as we haven't flushed the metadata yet
224 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800225 int previous_raid_disks;
226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 struct list_head handle_list; /* stripes needing handling */
228 struct list_head delayed_list; /* stripes that have plugged requests */
NeilBrown72626682005-09-09 16:23:54 -0700229 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 atomic_t preread_active_stripes; /* stripes with scheduled io */
231
NeilBrownf6705572006-03-27 01:18:11 -0800232 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
NeilBrownad01c9e2006-03-27 01:18:07 -0800233 /* unfortunately we need two cache names as we temporarily have
234 * two caches.
235 */
236 int active_name;
237 char cache_name[2][20];
Christoph Lametere18b8902006-12-06 20:33:20 -0800238 struct kmem_cache *slab_cache; /* for allocating stripes */
NeilBrown72626682005-09-09 16:23:54 -0700239
240 int seq_flush, seq_write;
241 int quiesce;
242
243 int fullsync; /* set to 1 if a full sync is needed,
244 * (fresh device added).
245 * Cleared when a sync completes.
246 */
247
NeilBrownca65b732006-01-06 00:20:17 -0800248 struct page *spare_page; /* Used when checking P/Q in raid6 */
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 /*
251 * Free stripes pool
252 */
253 atomic_t active_stripes;
254 struct list_head inactive_list;
255 wait_queue_head_t wait_for_stripe;
256 wait_queue_head_t wait_for_overlap;
257 int inactive_blocked; /* release of inactive stripes blocked,
258 * waiting for 25% to be free
NeilBrownad01c9e2006-03-27 01:18:07 -0800259 */
260 int pool_size; /* number of disks in stripeheads in pool */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 spinlock_t device_lock;
NeilBrownb55e6bf2006-03-27 01:18:06 -0800262 struct disk_info *disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263};
264
265typedef struct raid5_private_data raid5_conf_t;
266
267#define mddev_to_conf(mddev) ((raid5_conf_t *) mddev->private)
268
269/*
270 * Our supported algorithms
271 */
272#define ALGORITHM_LEFT_ASYMMETRIC 0
273#define ALGORITHM_RIGHT_ASYMMETRIC 1
274#define ALGORITHM_LEFT_SYMMETRIC 2
275#define ALGORITHM_RIGHT_SYMMETRIC 3
276
277#endif