blob: 0ed22dff56e0eb4a57ac8019ca2307f4d8eff5e0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _RAID5_H
2#define _RAID5_H
3
Linus Torvalds1da177e2005-04-16 15:20:36 -07004#include <linux/raid/xor.h>
5
6/*
7 *
8 * Each stripe contains one buffer per disc. Each buffer can be in
9 * one of a number of states stored in "flags". Changes between
10 * these states happen *almost* exclusively under a per-stripe
11 * spinlock. Some very specific changes can happen in bi_end_io, and
12 * these are not protected by the spin lock.
13 *
14 * The flag bits that are used to represent these states are:
15 * R5_UPTODATE and R5_LOCKED
16 *
17 * State Empty == !UPTODATE, !LOCK
18 * We have no data, and there is no active request
19 * State Want == !UPTODATE, LOCK
20 * A read request is being submitted for this block
21 * State Dirty == UPTODATE, LOCK
22 * Some new data is in this buffer, and it is being written out
23 * State Clean == UPTODATE, !LOCK
24 * We have valid data which is the same as on disc
25 *
26 * The possible state transitions are:
27 *
28 * Empty -> Want - on read or write to get old data for parity calc
29 * Empty -> Dirty - on compute_parity to satisfy write/sync request.(RECONSTRUCT_WRITE)
30 * Empty -> Clean - on compute_block when computing a block for failed drive
31 * Want -> Empty - on failed read
32 * Want -> Clean - on successful completion of read request
33 * Dirty -> Clean - on successful completion of write request
34 * Dirty -> Clean - on failed write
35 * Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
36 *
37 * The Want->Empty, Want->Clean, Dirty->Clean, transitions
38 * all happen in b_end_io at interrupt time.
39 * Each sets the Uptodate bit before releasing the Lock bit.
40 * This leaves one multi-stage transition:
41 * Want->Dirty->Clean
42 * This is safe because thinking that a Clean buffer is actually dirty
43 * will at worst delay some action, and the stripe will be scheduled
44 * for attention after the transition is complete.
45 *
46 * There is one possibility that is not covered by these states. That
47 * is if one drive has failed and there is a spare being rebuilt. We
48 * can't distinguish between a clean block that has been generated
49 * from parity calculations, and a clean block that has been
50 * successfully written to the spare ( or to parity when resyncing).
51 * To distingush these states we have a stripe bit STRIPE_INSYNC that
52 * is set whenever a write is scheduled to the spare, or to the parity
53 * disc if there is no spare. A sync request clears this bit, and
54 * when we find it set with no buffers locked, we know the sync is
55 * complete.
56 *
57 * Buffers for the md device that arrive via make_request are attached
58 * to the appropriate stripe in one of two lists linked on b_reqnext.
59 * One list (bh_read) for read requests, one (bh_write) for write.
60 * There should never be more than one buffer on the two lists
61 * together, but we are not guaranteed of that so we allow for more.
62 *
63 * If a buffer is on the read list when the associated cache buffer is
64 * Uptodate, the data is copied into the read buffer and it's b_end_io
65 * routine is called. This may happen in the end_request routine only
66 * if the buffer has just successfully been read. end_request should
67 * remove the buffers from the list and then set the Uptodate bit on
68 * the buffer. Other threads may do this only if they first check
69 * that the Uptodate bit is set. Once they have checked that they may
70 * take buffers off the read queue.
71 *
72 * When a buffer on the write list is committed for write it is copied
73 * into the cache buffer, which is then marked dirty, and moved onto a
74 * third list, the written list (bh_written). Once both the parity
75 * block and the cached buffer are successfully written, any buffer on
76 * a written list can be returned with b_end_io.
77 *
78 * The write list and read list both act as fifos. The read list is
79 * protected by the device_lock. The write and written lists are
80 * protected by the stripe lock. The device_lock, which can be
81 * claimed while the stipe lock is held, is only for list
82 * manipulations and will only be held for a very short time. It can
83 * be claimed from interrupts.
84 *
85 *
86 * Stripes in the stripe cache can be on one of two lists (or on
87 * neither). The "inactive_list" contains stripes which are not
88 * currently being used for any request. They can freely be reused
89 * for another stripe. The "handle_list" contains stripes that need
90 * to be handled in some way. Both of these are fifo queues. Each
91 * stripe is also (potentially) linked to a hash bucket in the hash
92 * table so that it can be found by sector number. Stripes that are
93 * not hashed must be on the inactive_list, and will normally be at
94 * the front. All stripes start life this way.
95 *
96 * The inactive_list, handle_list and hash bucket lists are all protected by the
97 * device_lock.
98 * - stripes on the inactive_list never have their stripe_lock held.
99 * - stripes have a reference counter. If count==0, they are on a list.
100 * - If a stripe might need handling, STRIPE_HANDLE is set.
101 * - When refcount reaches zero, then if STRIPE_HANDLE it is put on
102 * handle_list else inactive_list
103 *
104 * This, combined with the fact that STRIPE_HANDLE is only ever
105 * cleared while a stripe has a non-zero count means that if the
106 * refcount is 0 and STRIPE_HANDLE is set, then it is on the
107 * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
108 * the stripe is on inactive_list.
109 *
110 * The possible transitions are:
111 * activate an unhashed/inactive stripe (get_active_stripe())
112 * lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
113 * activate a hashed, possibly active stripe (get_active_stripe())
114 * lockdev check-hash if(!cnt++)unlink-stripe unlockdev
115 * attach a request to an active stripe (add_stripe_bh())
116 * lockdev attach-buffer unlockdev
117 * handle a stripe (handle_stripe())
Dan Williams91c00922007-01-02 13:52:30 -0700118 * lockstripe clrSTRIPE_HANDLE ...
119 * (lockdev check-buffers unlockdev) ..
120 * change-state ..
121 * record io/ops needed unlockstripe schedule io/ops
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 * release an active stripe (release_stripe())
123 * lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
124 *
125 * The refcount counts each thread that have activated the stripe,
126 * plus raid5d if it is handling it, plus one for each active request
Dan Williams91c00922007-01-02 13:52:30 -0700127 * on a cached buffer, and plus one if the stripe is undergoing stripe
128 * operations.
129 *
130 * Stripe operations are performed outside the stripe lock,
131 * the stripe operations are:
132 * -copying data between the stripe cache and user application buffers
133 * -computing blocks to save a disk access, or to recover a missing block
134 * -updating the parity on a write operation (reconstruct write and
135 * read-modify-write)
136 * -checking parity correctness
137 * -running i/o to disk
138 * These operations are carried out by raid5_run_ops which uses the async_tx
139 * api to (optionally) offload operations to dedicated hardware engines.
140 * When requesting an operation handle_stripe sets the pending bit for the
141 * operation and increments the count. raid5_run_ops is then run whenever
142 * the count is non-zero.
143 * There are some critical dependencies between the operations that prevent some
144 * from being requested while another is in flight.
145 * 1/ Parity check operations destroy the in cache version of the parity block,
146 * so we prevent parity dependent operations like writes and compute_blocks
147 * from starting while a check is in progress. Some dma engines can perform
148 * the check without damaging the parity block, in these cases the parity
149 * block is re-marked up to date (assuming the check was successful) and is
150 * not re-read from disk.
151 * 2/ When a write operation is requested we immediately lock the affected
152 * blocks, and mark them as not up to date. This causes new read requests
153 * to be held off, as well as parity checks and compute block operations.
154 * 3/ Once a compute block operation has been requested handle_stripe treats
155 * that block as if it is up to date. raid5_run_ops guaruntees that any
156 * operation that is dependent on the compute block result is initiated after
157 * the compute block completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 */
159
Dan Williamsecc65c92008-06-28 08:31:57 +1000160/*
161 * Operations state - intermediate states that are visible outside of sh->lock
162 * In general _idle indicates nothing is running, _run indicates a data
163 * processing operation is active, and _result means the data processing result
164 * is stable and can be acted upon. For simple operations like biofill and
165 * compute that only have an _idle and _run state they are indicated with
166 * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
167 */
168/**
169 * enum check_states - handles syncing / repairing a stripe
170 * @check_state_idle - check operations are quiesced
171 * @check_state_run - check operation is running
172 * @check_state_result - set outside lock when check result is valid
173 * @check_state_compute_run - check failed and we are repairing
174 * @check_state_compute_result - set outside lock when compute result is valid
175 */
176enum check_states {
177 check_state_idle = 0,
178 check_state_run, /* parity check */
179 check_state_check_result,
180 check_state_compute_run, /* parity repair */
181 check_state_compute_result,
182};
183
184/**
185 * enum reconstruct_states - handles writing or expanding a stripe
186 */
187enum reconstruct_states {
188 reconstruct_state_idle = 0,
Dan Williamsd8ee0722008-06-28 08:32:06 +1000189 reconstruct_state_prexor_drain_run, /* prexor-write */
Dan Williamsecc65c92008-06-28 08:31:57 +1000190 reconstruct_state_drain_run, /* write */
191 reconstruct_state_run, /* expand */
Dan Williamsd8ee0722008-06-28 08:32:06 +1000192 reconstruct_state_prexor_drain_result,
Dan Williamsecc65c92008-06-28 08:31:57 +1000193 reconstruct_state_drain_result,
194 reconstruct_state_result,
195};
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197struct stripe_head {
NeilBrownfccddba2006-01-06 00:20:33 -0800198 struct hlist_node hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 struct list_head lru; /* inactive_list or handle_list */
200 struct raid5_private_data *raid_conf;
201 sector_t sector; /* sector of this row */
202 int pd_idx; /* parity disk index */
203 unsigned long state; /* state flags */
204 atomic_t count; /* nr of active thread/requests */
205 spinlock_t lock;
NeilBrown72626682005-09-09 16:23:54 -0700206 int bm_seq; /* sequence number for bitmap flushes */
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800207 int disks; /* disks in stripe */
Dan Williamsecc65c92008-06-28 08:31:57 +1000208 enum check_states check_state;
Dan Williams600aa102008-06-28 08:32:05 +1000209 enum reconstruct_states reconstruct_state;
Dan Williams91c00922007-01-02 13:52:30 -0700210 /* stripe_operations
Dan Williams91c00922007-01-02 13:52:30 -0700211 * @target - STRIPE_OP_COMPUTE_BLK target
Dan Williams91c00922007-01-02 13:52:30 -0700212 */
213 struct stripe_operations {
Dan Williams91c00922007-01-02 13:52:30 -0700214 int target;
Dan Williams91c00922007-01-02 13:52:30 -0700215 u32 zero_sum_result;
216 } ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 struct r5dev {
218 struct bio req;
219 struct bio_vec vec;
220 struct page *page;
Dan Williams91c00922007-01-02 13:52:30 -0700221 struct bio *toread, *read, *towrite, *written;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 sector_t sector; /* sector of this page */
223 unsigned long flags;
224 } dev[1]; /* allocated with extra space depending of RAID geometry */
225};
Dan Williamsa4456852007-07-09 11:56:43 -0700226
227/* stripe_head_state - collects and tracks the dynamic state of a stripe_head
228 * for handle_stripe. It is only valid under spin_lock(sh->lock);
229 */
230struct stripe_head_state {
231 int syncing, expanding, expanded;
232 int locked, uptodate, to_read, to_write, failed, written;
Dan Williamsb5e98d62007-01-02 13:52:31 -0700233 int to_fill, compute, req_compute, non_overwrite;
Dan Williamsa4456852007-07-09 11:56:43 -0700234 int failed_num;
Dan Williamsecc65c92008-06-28 08:31:57 +1000235 unsigned long ops_request;
Dan Williamsa4456852007-07-09 11:56:43 -0700236};
237
238/* r6_state - extra state data only relevant to r6 */
239struct r6_state {
240 int p_failed, q_failed, qd_idx, failed_num[2];
241};
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243/* Flags */
244#define R5_UPTODATE 0 /* page contains current data */
245#define R5_LOCKED 1 /* IO has been submitted on "req" */
246#define R5_OVERWRITE 2 /* towrite covers whole page */
247/* and some that are internal to handle_stripe */
248#define R5_Insync 3 /* rdev && rdev->in_sync at start */
249#define R5_Wantread 4 /* want to schedule a read */
250#define R5_Wantwrite 5
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251#define R5_Overlap 7 /* There is a pending overlapping request on this block */
NeilBrown4e5314b2005-11-08 21:39:22 -0800252#define R5_ReadError 8 /* seen a read error here recently */
253#define R5_ReWrite 9 /* have tried to over-write the readerror */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
NeilBrownccfcc3c2006-03-27 01:18:09 -0800255#define R5_Expanded 10 /* This block now has post-expand data */
Dan Williams91c00922007-01-02 13:52:30 -0700256#define R5_Wantcompute 11 /* compute_block in progress treat as
257 * uptodate
258 */
259#define R5_Wantfill 12 /* dev->toread contains a bio that needs
260 * filling
261 */
Dan Williamsd8ee0722008-06-28 08:32:06 +1000262#define R5_Wantdrain 13 /* dev->towrite needs to be drained */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263/*
264 * Write method
265 */
266#define RECONSTRUCT_WRITE 1
267#define READ_MODIFY_WRITE 2
268/* not a write method, but a compute_parity mode */
269#define CHECK_PARITY 3
270
271/*
272 * Stripe state
273 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274#define STRIPE_HANDLE 2
275#define STRIPE_SYNCING 3
276#define STRIPE_INSYNC 4
277#define STRIPE_PREREAD_ACTIVE 5
278#define STRIPE_DELAYED 6
NeilBrown72626682005-09-09 16:23:54 -0700279#define STRIPE_DEGRADED 7
280#define STRIPE_BIT_DELAY 8
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800281#define STRIPE_EXPANDING 9
NeilBrownccfcc3c2006-03-27 01:18:09 -0800282#define STRIPE_EXPAND_SOURCE 10
283#define STRIPE_EXPAND_READY 11
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700284#define STRIPE_IO_STARTED 12 /* do not count towards 'bypass_count' */
285#define STRIPE_FULL_WRITE 13 /* all blocks are set to be overwritten */
Dan Williamsecc65c92008-06-28 08:31:57 +1000286#define STRIPE_BIOFILL_RUN 14
287#define STRIPE_COMPUTE_RUN 15
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288/*
Dan Williamsecc65c92008-06-28 08:31:57 +1000289 * Operation request flags
Dan Williams91c00922007-01-02 13:52:30 -0700290 */
291#define STRIPE_OP_BIOFILL 0
292#define STRIPE_OP_COMPUTE_BLK 1
293#define STRIPE_OP_PREXOR 2
294#define STRIPE_OP_BIODRAIN 3
295#define STRIPE_OP_POSTXOR 4
296#define STRIPE_OP_CHECK 5
Dan Williams91c00922007-01-02 13:52:30 -0700297
Dan Williams91c00922007-01-02 13:52:30 -0700298/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 * Plugging:
300 *
301 * To improve write throughput, we need to delay the handling of some
302 * stripes until there has been a chance that several write requests
303 * for the one stripe have all been collected.
304 * In particular, any write request that would require pre-reading
305 * is put on a "delayed" queue until there are no stripes currently
306 * in a pre-read phase. Further, if the "delayed" queue is empty when
307 * a stripe is put on it then we "plug" the queue and do not process it
308 * until an unplug call is made. (the unplug_io_fn() is called).
309 *
310 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
311 * it to the count of prereading stripes.
312 * When write is initiated, or the stripe refcnt == 0 (just in case) we
313 * clear the PREREAD_ACTIVE flag and decrement the count
NeilBrownb5c124a2006-10-03 01:15:45 -0700314 * Whenever the 'handle' queue is empty and the device is not plugged, we
315 * move any strips from delayed to handle and clear the DELAYED flag and set
316 * PREREAD_ACTIVE.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 * In stripe_handle, if we find pre-reading is necessary, we do it if
318 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
319 * HANDLE gets cleared if stripe_handle leave nothing locked.
320 */
Christoph Hellwigef740c32009-03-31 14:27:03 +1100321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323struct disk_info {
324 mdk_rdev_t *rdev;
325};
326
327struct raid5_private_data {
NeilBrownfccddba2006-01-06 00:20:33 -0800328 struct hlist_head *stripe_hashtbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 mddev_t *mddev;
330 struct disk_info *spare;
331 int chunk_size, level, algorithm;
NeilBrown16a53ec2006-06-26 00:27:38 -0700332 int max_degraded;
NeilBrown02c2de82006-10-03 01:15:47 -0700333 int raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 int max_nr_stripes;
335
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800336 /* used during an expand */
337 sector_t expand_progress; /* MaxSector when no expand happening */
NeilBrownb578d552006-03-27 01:18:12 -0800338 sector_t expand_lo; /* from here up to expand_progress it out-of-bounds
339 * as we haven't flushed the metadata yet
340 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800341 int previous_raid_disks;
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 struct list_head handle_list; /* stripes needing handling */
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700344 struct list_head hold_list; /* preread ready stripes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 struct list_head delayed_list; /* stripes that have plugged requests */
NeilBrown72626682005-09-09 16:23:54 -0700346 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800347 struct bio *retry_read_aligned; /* currently retrying aligned bios */
348 struct bio *retry_read_aligned_list; /* aligned bios retry list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 atomic_t preread_active_stripes; /* stripes with scheduled io */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800350 atomic_t active_aligned_reads;
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700351 atomic_t pending_full_writes; /* full write backlog */
352 int bypass_count; /* bypassed prereads */
353 int bypass_threshold; /* preread nice */
354 struct list_head *last_hold; /* detect hold_list promotions */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
NeilBrownf6705572006-03-27 01:18:11 -0800356 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
NeilBrownad01c9e2006-03-27 01:18:07 -0800357 /* unfortunately we need two cache names as we temporarily have
358 * two caches.
359 */
360 int active_name;
361 char cache_name[2][20];
Christoph Lametere18b8902006-12-06 20:33:20 -0800362 struct kmem_cache *slab_cache; /* for allocating stripes */
NeilBrown72626682005-09-09 16:23:54 -0700363
364 int seq_flush, seq_write;
365 int quiesce;
366
367 int fullsync; /* set to 1 if a full sync is needed,
368 * (fresh device added).
369 * Cleared when a sync completes.
370 */
371
NeilBrownca65b732006-01-06 00:20:17 -0800372 struct page *spare_page; /* Used when checking P/Q in raid6 */
373
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 /*
375 * Free stripes pool
376 */
377 atomic_t active_stripes;
378 struct list_head inactive_list;
379 wait_queue_head_t wait_for_stripe;
380 wait_queue_head_t wait_for_overlap;
381 int inactive_blocked; /* release of inactive stripes blocked,
382 * waiting for 25% to be free
NeilBrownad01c9e2006-03-27 01:18:07 -0800383 */
384 int pool_size; /* number of disks in stripeheads in pool */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 spinlock_t device_lock;
NeilBrownb55e6bf2006-03-27 01:18:06 -0800386 struct disk_info *disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387};
388
389typedef struct raid5_private_data raid5_conf_t;
390
391#define mddev_to_conf(mddev) ((raid5_conf_t *) mddev->private)
392
393/*
394 * Our supported algorithms
395 */
396#define ALGORITHM_LEFT_ASYMMETRIC 0
397#define ALGORITHM_RIGHT_ASYMMETRIC 1
398#define ALGORITHM_LEFT_SYMMETRIC 2
399#define ALGORITHM_RIGHT_SYMMETRIC 3
400
401#endif