blob: 7686fcb62157117c67f7d320ec7b6081569d0398 [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>
Dan Williamsad283ea2009-08-29 19:09:26 -07005#include <linux/dmaengine.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07006
7/*
8 *
NeilBrownc4c16632011-07-26 11:34:20 +10009 * Each stripe contains one buffer per device. Each buffer can be in
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * one of a number of states stored in "flags". Changes between
NeilBrownc4c16632011-07-26 11:34:20 +100011 * these states happen *almost* exclusively under the protection of the
12 * STRIPE_ACTIVE flag. Some very specific changes can happen in bi_end_io, and
13 * these are not protected by STRIPE_ACTIVE.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 *
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
NeilBrownede7ee82011-12-23 10:17:52 +110030 * Empty -> Dirty - on compute_parity to satisfy write/sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 * 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).
Michael Opdenackeraa5e5dc2013-09-18 06:00:43 +020052 * To distinguish these states we have a stripe bit STRIPE_INSYNC that
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * 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 *
NeilBrownc4c16632011-07-26 11:34:20 +100079 * The write list and read list both act as fifos. The read list,
80 * write list and written list are protected by the device_lock.
81 * The device_lock is only for list manipulations and will only be
82 * held for a very short time. It can be claimed from interrupts.
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 *
84 *
85 * Stripes in the stripe cache can be on one of two lists (or on
86 * neither). The "inactive_list" contains stripes which are not
87 * currently being used for any request. They can freely be reused
88 * for another stripe. The "handle_list" contains stripes that need
89 * to be handled in some way. Both of these are fifo queues. Each
90 * stripe is also (potentially) linked to a hash bucket in the hash
91 * table so that it can be found by sector number. Stripes that are
92 * not hashed must be on the inactive_list, and will normally be at
93 * the front. All stripes start life this way.
94 *
95 * The inactive_list, handle_list and hash bucket lists are all protected by the
96 * device_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * - stripes have a reference counter. If count==0, they are on a list.
98 * - If a stripe might need handling, STRIPE_HANDLE is set.
99 * - When refcount reaches zero, then if STRIPE_HANDLE it is put on
100 * handle_list else inactive_list
101 *
102 * This, combined with the fact that STRIPE_HANDLE is only ever
103 * cleared while a stripe has a non-zero count means that if the
104 * refcount is 0 and STRIPE_HANDLE is set, then it is on the
105 * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
106 * the stripe is on inactive_list.
107 *
108 * The possible transitions are:
109 * activate an unhashed/inactive stripe (get_active_stripe())
110 * lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
111 * activate a hashed, possibly active stripe (get_active_stripe())
112 * lockdev check-hash if(!cnt++)unlink-stripe unlockdev
113 * attach a request to an active stripe (add_stripe_bh())
114 * lockdev attach-buffer unlockdev
115 * handle a stripe (handle_stripe())
NeilBrownc4c16632011-07-26 11:34:20 +1000116 * setSTRIPE_ACTIVE, clrSTRIPE_HANDLE ...
Dan Williams91c00922007-01-02 13:52:30 -0700117 * (lockdev check-buffers unlockdev) ..
118 * change-state ..
NeilBrownc4c16632011-07-26 11:34:20 +1000119 * record io/ops needed clearSTRIPE_ACTIVE schedule io/ops
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 * 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
Dan Williams91c00922007-01-02 13:52:30 -0700125 * on a cached buffer, and plus one if the stripe is undergoing stripe
126 * operations.
127 *
NeilBrownc4c16632011-07-26 11:34:20 +1000128 * The stripe operations are:
Dan Williams91c00922007-01-02 13:52:30 -0700129 * -copying data between the stripe cache and user application buffers
130 * -computing blocks to save a disk access, or to recover a missing block
131 * -updating the parity on a write operation (reconstruct write and
132 * read-modify-write)
133 * -checking parity correctness
134 * -running i/o to disk
135 * These operations are carried out by raid5_run_ops which uses the async_tx
136 * api to (optionally) offload operations to dedicated hardware engines.
137 * When requesting an operation handle_stripe sets the pending bit for the
138 * operation and increments the count. raid5_run_ops is then run whenever
139 * the count is non-zero.
140 * There are some critical dependencies between the operations that prevent some
141 * from being requested while another is in flight.
142 * 1/ Parity check operations destroy the in cache version of the parity block,
143 * so we prevent parity dependent operations like writes and compute_blocks
144 * from starting while a check is in progress. Some dma engines can perform
145 * the check without damaging the parity block, in these cases the parity
146 * block is re-marked up to date (assuming the check was successful) and is
147 * not re-read from disk.
148 * 2/ When a write operation is requested we immediately lock the affected
149 * blocks, and mark them as not up to date. This causes new read requests
150 * to be held off, as well as parity checks and compute block operations.
151 * 3/ Once a compute block operation has been requested handle_stripe treats
152 * that block as if it is up to date. raid5_run_ops guaruntees that any
153 * operation that is dependent on the compute block result is initiated after
154 * the compute block completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 */
156
Dan Williamsecc65c92008-06-28 08:31:57 +1000157/*
NeilBrownf72ffdd2014-09-30 14:23:59 +1000158 * Operations state - intermediate states that are visible outside of
NeilBrownc4c16632011-07-26 11:34:20 +1000159 * STRIPE_ACTIVE.
Dan Williamsecc65c92008-06-28 08:31:57 +1000160 * In general _idle indicates nothing is running, _run indicates a data
161 * processing operation is active, and _result means the data processing result
162 * is stable and can be acted upon. For simple operations like biofill and
163 * compute that only have an _idle and _run state they are indicated with
164 * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
165 */
166/**
167 * enum check_states - handles syncing / repairing a stripe
168 * @check_state_idle - check operations are quiesced
169 * @check_state_run - check operation is running
170 * @check_state_result - set outside lock when check result is valid
171 * @check_state_compute_run - check failed and we are repairing
172 * @check_state_compute_result - set outside lock when compute result is valid
173 */
174enum check_states {
175 check_state_idle = 0,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700176 check_state_run, /* xor parity check */
177 check_state_run_q, /* q-parity check */
178 check_state_run_pq, /* pq dual parity check */
Dan Williamsecc65c92008-06-28 08:31:57 +1000179 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;
NeilBrownd0dabf72009-03-31 14:39:38 +1100199 struct list_head lru; /* inactive_list or handle_list */
Shaohua Li773ca822013-08-27 17:50:39 +0800200 struct llist_node release_list;
NeilBrownd1688a62011-10-11 16:49:52 +1100201 struct r5conf *raid_conf;
NeilBrown86b42c72009-03-31 15:19:03 +1100202 short generation; /* increments with every
203 * reshape */
NeilBrownd0dabf72009-03-31 14:39:38 +1100204 sector_t sector; /* sector of this row */
205 short pd_idx; /* parity disk index */
206 short qd_idx; /* 'Q' disk index for raid6 */
NeilBrown67cc2b82009-03-31 14:39:38 +1100207 short ddf_layout;/* use DDF ordering to calculate Q */
Shaohua Li566c09c2013-11-14 15:16:17 +1100208 short hash_lock_index;
NeilBrownd0dabf72009-03-31 14:39:38 +1100209 unsigned long state; /* state flags */
210 atomic_t count; /* nr of active thread/requests */
NeilBrown72626682005-09-09 16:23:54 -0700211 int bm_seq; /* sequence number for bitmap flushes */
NeilBrownd0dabf72009-03-31 14:39:38 +1100212 int disks; /* disks in stripe */
shli@kernel.org7a87f432014-12-15 12:57:03 +1100213 int overwrite_disks; /* total overwrite disks in stripe,
214 * this is only checked when stripe
215 * has STRIPE_BATCH_READY
216 */
Dan Williamsecc65c92008-06-28 08:31:57 +1000217 enum check_states check_state;
Dan Williams600aa102008-06-28 08:32:05 +1000218 enum reconstruct_states reconstruct_state;
Shaohua Lib17459c2012-07-19 16:01:31 +1000219 spinlock_t stripe_lock;
Shaohua Li851c30c2013-08-28 14:30:16 +0800220 int cpu;
Shaohua Libfc90cb2013-08-29 15:40:32 +0800221 struct r5worker_group *group;
shli@kernel.org59fc6302014-12-15 12:57:03 +1100222
223 struct stripe_head *batch_head; /* protected by stripe lock */
224 spinlock_t batch_lock; /* only header's lock is useful */
225 struct list_head batch_list; /* protected by head's batch lock*/
Dan Williams417b8d42009-10-16 16:25:22 +1100226 /**
227 * struct stripe_operations
Dan Williams91c00922007-01-02 13:52:30 -0700228 * @target - STRIPE_OP_COMPUTE_BLK target
Dan Williams417b8d42009-10-16 16:25:22 +1100229 * @target2 - 2nd compute target in the raid6 case
230 * @zero_sum_result - P and Q verification flags
231 * @request - async service request flags for raid_run_ops
Dan Williams91c00922007-01-02 13:52:30 -0700232 */
233 struct stripe_operations {
Dan Williamsac6b53b2009-07-14 13:40:19 -0700234 int target, target2;
Dan Williamsad283ea2009-08-29 19:09:26 -0700235 enum sum_check_flags zero_sum_result;
Dan Williams91c00922007-01-02 13:52:30 -0700236 } ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 struct r5dev {
NeilBrown671488c2011-12-23 10:17:52 +1100238 /* rreq and rvec are used for the replacement device when
239 * writing data to both devices.
240 */
241 struct bio req, rreq;
242 struct bio_vec vec, rvec;
Shaohua Lid592a992014-05-21 17:57:44 +0800243 struct page *page, *orig_page;
Dan Williams91c00922007-01-02 13:52:30 -0700244 struct bio *toread, *read, *towrite, *written;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 sector_t sector; /* sector of this page */
246 unsigned long flags;
247 } dev[1]; /* allocated with extra space depending of RAID geometry */
248};
Dan Williamsa4456852007-07-09 11:56:43 -0700249
250/* stripe_head_state - collects and tracks the dynamic state of a stripe_head
NeilBrownc4c16632011-07-26 11:34:20 +1000251 * for handle_stripe.
Dan Williamsa4456852007-07-09 11:56:43 -0700252 */
253struct stripe_head_state {
NeilBrown9a3e1102011-12-23 10:17:53 +1100254 /* 'syncing' means that we need to read all devices, either
255 * to check/correct parity, or to reconstruct a missing device.
256 * 'replacing' means we are replacing one or more drives and
257 * the source is valid at this point so we don't need to
258 * read all devices, just the replacement targets.
259 */
260 int syncing, expanding, expanded, replacing;
Dan Williamsa4456852007-07-09 11:56:43 -0700261 int locked, uptodate, to_read, to_write, failed, written;
Dan Williamsb5e98d62007-01-02 13:52:31 -0700262 int to_fill, compute, req_compute, non_overwrite;
NeilBrownf2b3b442011-07-26 11:35:19 +1000263 int failed_num[2];
NeilBrownf2b3b442011-07-26 11:35:19 +1000264 int p_failed, q_failed;
NeilBrownc5709ef2011-07-26 11:35:20 +1000265 int dec_preread_active;
266 unsigned long ops_request;
267
NeilBrown34a6f802015-08-14 12:07:57 +1000268 struct bio_list return_bi;
NeilBrown3cb03002011-10-11 16:45:26 +1100269 struct md_rdev *blocked_rdev;
NeilBrownbc2607f2011-07-28 11:39:22 +1000270 int handle_bad_blocks;
Dan Williamsa4456852007-07-09 11:56:43 -0700271};
272
NeilBrown671488c2011-12-23 10:17:52 +1100273/* Flags for struct r5dev.flags */
274enum r5dev_flags {
275 R5_UPTODATE, /* page contains current data */
276 R5_LOCKED, /* IO has been submitted on "req" */
NeilBrown977df362011-12-23 10:17:53 +1100277 R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */
NeilBrown671488c2011-12-23 10:17:52 +1100278 R5_OVERWRITE, /* towrite covers whole page */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279/* and some that are internal to handle_stripe */
NeilBrown671488c2011-12-23 10:17:52 +1100280 R5_Insync, /* rdev && rdev->in_sync at start */
281 R5_Wantread, /* want to schedule a read */
282 R5_Wantwrite,
283 R5_Overlap, /* There is a pending overlapping request
284 * on this block */
majianpeng3f9e7c12012-07-31 10:04:21 +1000285 R5_ReadNoMerge, /* prevent bio from merging in block-layer */
NeilBrown671488c2011-12-23 10:17:52 +1100286 R5_ReadError, /* seen a read error here recently */
287 R5_ReWrite, /* have tried to over-write the readerror */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
NeilBrown671488c2011-12-23 10:17:52 +1100289 R5_Expanded, /* This block now has post-expand data */
290 R5_Wantcompute, /* compute_block in progress treat as
291 * uptodate
292 */
293 R5_Wantfill, /* dev->toread contains a bio that needs
294 * filling
295 */
296 R5_Wantdrain, /* dev->towrite needs to be drained */
297 R5_WantFUA, /* Write should be FUA */
Shaohua Libc0934f2012-05-22 13:55:05 +1000298 R5_SyncIO, /* The IO is sync */
NeilBrown671488c2011-12-23 10:17:52 +1100299 R5_WriteError, /* got a write error - need to record it */
300 R5_MadeGood, /* A bad block has been fixed by writing to it */
301 R5_ReadRepl, /* Will/did read from replacement rather than orig */
302 R5_MadeGoodRepl,/* A bad block on the replacement device has been
303 * fixed by writing to it */
NeilBrown9a3e1102011-12-23 10:17:53 +1100304 R5_NeedReplace, /* This device has a replacement which is not
305 * up-to-date at this stripe. */
306 R5_WantReplace, /* We need to update the replacement, we have read
307 * data in, and now is a good time to write it out.
308 */
Shaohua Li620125f2012-10-11 13:49:05 +1100309 R5_Discard, /* Discard the stripe */
Shaohua Lid592a992014-05-21 17:57:44 +0800310 R5_SkipCopy, /* Don't copy data from bio to stripe cache */
NeilBrown671488c2011-12-23 10:17:52 +1100311};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313/*
314 * Stripe state
315 */
NeilBrown83206d62011-07-26 11:19:49 +1000316enum {
NeilBrownc4c16632011-07-26 11:34:20 +1000317 STRIPE_ACTIVE,
NeilBrown83206d62011-07-26 11:19:49 +1000318 STRIPE_HANDLE,
319 STRIPE_SYNC_REQUESTED,
320 STRIPE_SYNCING,
321 STRIPE_INSYNC,
NeilBrownf94c0b62013-07-22 12:57:21 +1000322 STRIPE_REPLACED,
NeilBrown83206d62011-07-26 11:19:49 +1000323 STRIPE_PREREAD_ACTIVE,
324 STRIPE_DELAYED,
325 STRIPE_DEGRADED,
326 STRIPE_BIT_DELAY,
327 STRIPE_EXPANDING,
328 STRIPE_EXPAND_SOURCE,
329 STRIPE_EXPAND_READY,
330 STRIPE_IO_STARTED, /* do not count towards 'bypass_count' */
331 STRIPE_FULL_WRITE, /* all blocks are set to be overwritten */
332 STRIPE_BIOFILL_RUN,
333 STRIPE_COMPUTE_RUN,
334 STRIPE_OPS_REQ_PENDING,
Shaohua Li8811b592012-08-02 08:33:00 +1000335 STRIPE_ON_UNPLUG_LIST,
NeilBrownf8dfcff2013-03-12 12:18:06 +1100336 STRIPE_DISCARD,
Shaohua Li773ca822013-08-27 17:50:39 +0800337 STRIPE_ON_RELEASE_LIST,
shli@kernel.orgda41ba62014-12-15 12:57:03 +1100338 STRIPE_BATCH_READY,
shli@kernel.org72ac7332014-12-15 12:57:03 +1100339 STRIPE_BATCH_ERR,
NeilBrownd0852df52015-05-27 08:43:45 +1000340 STRIPE_BITMAP_PENDING, /* Being added to bitmap, don't add
341 * to batch yet.
342 */
NeilBrown83206d62011-07-26 11:19:49 +1000343};
Dan Williams417b8d42009-10-16 16:25:22 +1100344
NeilBrown1b956f72015-05-21 12:40:26 +1000345#define STRIPE_EXPAND_SYNC_FLAGS \
shli@kernel.orgdabc4ec2014-12-15 12:57:04 +1100346 ((1 << STRIPE_EXPAND_SOURCE) |\
347 (1 << STRIPE_EXPAND_READY) |\
348 (1 << STRIPE_EXPANDING) |\
349 (1 << STRIPE_SYNC_REQUESTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350/*
Dan Williamsecc65c92008-06-28 08:31:57 +1000351 * Operation request flags
Dan Williams91c00922007-01-02 13:52:30 -0700352 */
NeilBrownede7ee82011-12-23 10:17:52 +1100353enum {
354 STRIPE_OP_BIOFILL,
355 STRIPE_OP_COMPUTE_BLK,
356 STRIPE_OP_PREXOR,
357 STRIPE_OP_BIODRAIN,
358 STRIPE_OP_RECONSTRUCT,
359 STRIPE_OP_CHECK,
360};
Markus Stockhausen584acdd2014-12-15 12:57:05 +1100361
362/*
363 * RAID parity calculation preferences
364 */
365enum {
366 PARITY_DISABLE_RMW = 0,
367 PARITY_ENABLE_RMW,
Markus Stockhausend06f1912014-12-15 12:57:05 +1100368 PARITY_PREFER_RMW,
Markus Stockhausen584acdd2014-12-15 12:57:05 +1100369};
370
371/*
372 * Pages requested from set_syndrome_sources()
373 */
374enum {
375 SYNDROME_SRC_ALL,
376 SYNDROME_SRC_WANT_DRAIN,
377 SYNDROME_SRC_WRITTEN,
378};
Dan Williams91c00922007-01-02 13:52:30 -0700379/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 * Plugging:
381 *
382 * To improve write throughput, we need to delay the handling of some
383 * stripes until there has been a chance that several write requests
384 * for the one stripe have all been collected.
385 * In particular, any write request that would require pre-reading
386 * is put on a "delayed" queue until there are no stripes currently
387 * in a pre-read phase. Further, if the "delayed" queue is empty when
388 * a stripe is put on it then we "plug" the queue and do not process it
389 * until an unplug call is made. (the unplug_io_fn() is called).
390 *
391 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
392 * it to the count of prereading stripes.
393 * When write is initiated, or the stripe refcnt == 0 (just in case) we
394 * clear the PREREAD_ACTIVE flag and decrement the count
NeilBrownb5c124a2006-10-03 01:15:45 -0700395 * Whenever the 'handle' queue is empty and the device is not plugged, we
396 * move any strips from delayed to handle and clear the DELAYED flag and set
397 * PREREAD_ACTIVE.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * In stripe_handle, if we find pre-reading is necessary, we do it if
399 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
NeilBrownc4c16632011-07-26 11:34:20 +1000400 * HANDLE gets cleared if stripe_handle leaves nothing locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 */
Christoph Hellwigef740c32009-03-31 14:27:03 +1100402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403struct disk_info {
NeilBrown671488c2011-12-23 10:17:52 +1100404 struct md_rdev *rdev, *replacement;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405};
406
Shaohua Li566c09c2013-11-14 15:16:17 +1100407/* NOTE NR_STRIPE_HASH_LOCKS must remain below 64.
408 * This is because we sometimes take all the spinlocks
409 * and creating that much locking depth can cause
410 * problems.
411 */
412#define NR_STRIPE_HASH_LOCKS 8
413#define STRIPE_HASH_LOCKS_MASK (NR_STRIPE_HASH_LOCKS - 1)
414
Shaohua Li851c30c2013-08-28 14:30:16 +0800415struct r5worker {
416 struct work_struct work;
417 struct r5worker_group *group;
Shaohua Li566c09c2013-11-14 15:16:17 +1100418 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Libfc90cb2013-08-29 15:40:32 +0800419 bool working;
Shaohua Li851c30c2013-08-28 14:30:16 +0800420};
421
422struct r5worker_group {
423 struct list_head handle_list;
424 struct r5conf *conf;
425 struct r5worker *workers;
Shaohua Libfc90cb2013-08-29 15:40:32 +0800426 int stripes_cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +0800427};
428
NeilBrownd1688a62011-10-11 16:49:52 +1100429struct r5conf {
NeilBrownfccddba2006-01-06 00:20:33 -0800430 struct hlist_head *stripe_hashtbl;
Shaohua Li566c09c2013-11-14 15:16:17 +1100431 /* only protect corresponding hash list and inactive_list */
432 spinlock_t hash_locks[NR_STRIPE_HASH_LOCKS];
NeilBrownfd01b882011-10-11 16:47:53 +1100433 struct mddev *mddev;
Andre Noll09c9e5f2009-06-18 08:45:55 +1000434 int chunk_sectors;
Markus Stockhausen584acdd2014-12-15 12:57:05 +1100435 int level, algorithm, rmw_level;
NeilBrown16a53ec2006-06-26 00:27:38 -0700436 int max_degraded;
NeilBrown02c2de82006-10-03 01:15:47 -0700437 int raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 int max_nr_stripes;
NeilBrownedbe83a2015-02-26 12:47:56 +1100439 int min_nr_stripes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
NeilBrownfef9c612009-03-31 15:16:46 +1100441 /* reshape_progress is the leading edge of a 'reshape'
442 * It has value MaxSector when no reshape is happening
443 * If delta_disks < 0, it is the last sector we started work on,
444 * else is it the next sector to work on.
445 */
446 sector_t reshape_progress;
447 /* reshape_safe is the trailing edge of a reshape. We know that
448 * before (or after) this address, all reshape has completed.
449 */
450 sector_t reshape_safe;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800451 int previous_raid_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +1000452 int prev_chunk_sectors;
453 int prev_algo;
NeilBrown86b42c72009-03-31 15:19:03 +1100454 short generation; /* increments with every reshape */
NeilBrownc46501b2013-08-27 15:52:13 +1000455 seqcount_t gen_lock; /* lock against generation changes */
NeilBrownc8f517c2009-03-31 15:28:40 +1100456 unsigned long reshape_checkpoint; /* Time we last updated
457 * metadata */
NeilBrownb5254dd2012-05-21 09:27:01 +1000458 long long min_offset_diff; /* minimum difference between
459 * data_offset and
460 * new_data_offset across all
461 * devices. May be negative,
462 * but is closest to zero.
463 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 struct list_head handle_list; /* stripes needing handling */
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700466 struct list_head hold_list; /* preread ready stripes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 struct list_head delayed_list; /* stripes that have plugged requests */
NeilBrown72626682005-09-09 16:23:54 -0700468 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800469 struct bio *retry_read_aligned; /* currently retrying aligned bios */
470 struct bio *retry_read_aligned_list; /* aligned bios retry list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 atomic_t preread_active_stripes; /* stripes with scheduled io */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800472 atomic_t active_aligned_reads;
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700473 atomic_t pending_full_writes; /* full write backlog */
474 int bypass_count; /* bypassed prereads */
475 int bypass_threshold; /* preread nice */
Shaohua Lid592a992014-05-21 17:57:44 +0800476 int skip_copy; /* Don't copy data from bio to stripe cache */
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700477 struct list_head *last_hold; /* detect hold_list promotions */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
NeilBrownc3cce6c2015-08-14 12:47:33 +1000479 /* bios to have bi_end_io called after metadata is synced */
480 struct bio_list return_bi;
481
NeilBrownf6705572006-03-27 01:18:11 -0800482 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
NeilBrownad01c9e2006-03-27 01:18:07 -0800483 /* unfortunately we need two cache names as we temporarily have
484 * two caches.
485 */
486 int active_name;
NeilBrownf4be6b42010-06-01 19:37:25 +1000487 char cache_name[2][32];
NeilBrown2d5b5692015-07-06 12:49:23 +1000488 struct kmem_cache *slab_cache; /* for allocating stripes */
489 struct mutex cache_size_mutex; /* Protect changes to cache size */
NeilBrown72626682005-09-09 16:23:54 -0700490
491 int seq_flush, seq_write;
492 int quiesce;
493
494 int fullsync; /* set to 1 if a full sync is needed,
495 * (fresh device added).
496 * Cleared when a sync completes.
497 */
NeilBrown7f0da592011-07-28 11:39:22 +1000498 int recovery_disabled;
Dan Williams36d1c642009-07-14 11:48:22 -0700499 /* per cpu variables */
500 struct raid5_percpu {
501 struct page *spare_page; /* Used when checking P/Q in raid6 */
shli@kernel.org46d5b782014-12-15 12:57:02 +1100502 struct flex_array *scribble; /* space for constructing buffer
Dan Williamsd6f38f32009-07-14 11:50:52 -0700503 * lists and performing address
504 * conversions
505 */
Tejun Heoa29d8b82010-02-02 14:39:15 +0900506 } __percpu *percpu;
Dan Williams36d1c642009-07-14 11:48:22 -0700507#ifdef CONFIG_HOTPLUG_CPU
508 struct notifier_block cpu_notify;
509#endif
NeilBrownca65b732006-01-06 00:20:17 -0800510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 /*
512 * Free stripes pool
513 */
514 atomic_t active_stripes;
Shaohua Li566c09c2013-11-14 15:16:17 +1100515 struct list_head inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Li4bda5562013-11-14 15:16:17 +1100516 atomic_t empty_inactive_list_nr;
Shaohua Li773ca822013-08-27 17:50:39 +0800517 struct llist_head released_stripes;
Yuanhan Liub1b46482015-05-08 18:19:06 +1000518 wait_queue_head_t wait_for_quiescent;
Yuanhan Liue9e4c372015-05-08 18:19:07 +1000519 wait_queue_head_t wait_for_stripe[NR_STRIPE_HASH_LOCKS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 wait_queue_head_t wait_for_overlap;
NeilBrown54233992015-02-26 12:21:04 +1100521 unsigned long cache_state;
522#define R5_INACTIVE_BLOCKED 1 /* release of inactive stripes blocked,
523 * waiting for 25% to be free
524 */
NeilBrownedbe83a2015-02-26 12:47:56 +1100525#define R5_ALLOC_MORE 2 /* It might help to allocate another
526 * stripe.
527 */
528#define R5_DID_ALLOC 4 /* A stripe was allocated, don't allocate
529 * more until at least one has been
530 * released. This avoids flooding
531 * the cache.
532 */
533 struct shrinker shrinker;
NeilBrownad01c9e2006-03-27 01:18:07 -0800534 int pool_size; /* number of disks in stripeheads in pool */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 spinlock_t device_lock;
NeilBrownb55e6bf2006-03-27 01:18:06 -0800536 struct disk_info *disks;
NeilBrown91adb562009-03-31 14:39:39 +1100537
538 /* When taking over an array from a different personality, we store
539 * the new thread here until we fully activate the array.
540 */
NeilBrown2b8bf342011-10-11 16:48:23 +1100541 struct md_thread *thread;
Shaohua Li566c09c2013-11-14 15:16:17 +1100542 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Li851c30c2013-08-28 14:30:16 +0800543 struct r5worker_group *worker_groups;
544 int group_cnt;
545 int worker_cnt_per_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546};
547
NeilBrown54233992015-02-26 12:21:04 +1100548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549/*
550 * Our supported algorithms
551 */
NeilBrown99c0fb52009-03-31 14:39:38 +1100552#define ALGORITHM_LEFT_ASYMMETRIC 0 /* Rotating Parity N with Data Restart */
553#define ALGORITHM_RIGHT_ASYMMETRIC 1 /* Rotating Parity 0 with Data Restart */
554#define ALGORITHM_LEFT_SYMMETRIC 2 /* Rotating Parity N with Data Continuation */
555#define ALGORITHM_RIGHT_SYMMETRIC 3 /* Rotating Parity 0 with Data Continuation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
NeilBrown99c0fb52009-03-31 14:39:38 +1100557/* Define non-rotating (raid4) algorithms. These allow
558 * conversion of raid4 to raid5.
559 */
560#define ALGORITHM_PARITY_0 4 /* P or P,Q are initial devices */
561#define ALGORITHM_PARITY_N 5 /* P or P,Q are final devices. */
562
563/* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
564 * Firstly, the exact positioning of the parity block is slightly
565 * different between the 'LEFT_*' modes of md and the "_N_*" modes
566 * of DDF.
567 * Secondly, or order of datablocks over which the Q syndrome is computed
568 * is different.
569 * Consequently we have different layouts for DDF/raid6 than md/raid6.
570 * These layouts are from the DDFv1.2 spec.
571 * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
572 * leaves RLQ=3 as 'Vendor Specific'
573 */
574
575#define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
576#define ALGORITHM_ROTATING_N_RESTART 9 /* DDF PRL=6 RLQ=2 */
577#define ALGORITHM_ROTATING_N_CONTINUE 10 /*DDF PRL=6 RLQ=3 */
578
NeilBrown99c0fb52009-03-31 14:39:38 +1100579/* For every RAID5 algorithm we define a RAID6 algorithm
580 * with exactly the same layout for data and parity, and
581 * with the Q block always on the last device (N-1).
582 * This allows trivial conversion from RAID5 to RAID6
583 */
584#define ALGORITHM_LEFT_ASYMMETRIC_6 16
585#define ALGORITHM_RIGHT_ASYMMETRIC_6 17
586#define ALGORITHM_LEFT_SYMMETRIC_6 18
587#define ALGORITHM_RIGHT_SYMMETRIC_6 19
588#define ALGORITHM_PARITY_0_6 20
589#define ALGORITHM_PARITY_N_6 ALGORITHM_PARITY_N
590
591static inline int algorithm_valid_raid5(int layout)
592{
593 return (layout >= 0) &&
594 (layout <= 5);
595}
596static inline int algorithm_valid_raid6(int layout)
597{
598 return (layout >= 0 && layout <= 5)
599 ||
NeilBrowne4424fe2009-10-16 16:27:34 +1100600 (layout >= 8 && layout <= 10)
NeilBrown99c0fb52009-03-31 14:39:38 +1100601 ||
602 (layout >= 16 && layout <= 20);
603}
604
605static inline int algorithm_is_DDF(int layout)
606{
607 return layout >= 8 && layout <= 10;
608}
NeilBrown11d8a6e2010-07-26 11:57:07 +1000609
NeilBrownd1688a62011-10-11 16:49:52 +1100610extern void md_raid5_kick_device(struct r5conf *conf);
NeilBrownfd01b882011-10-11 16:47:53 +1100611extern int raid5_set_cache_size(struct mddev *mddev, int size);
Shaohua Li6d036f72015-08-13 14:31:57 -0700612extern sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous);
613extern void raid5_release_stripe(struct stripe_head *sh);
614extern sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
615 int previous, int *dd_idx,
616 struct stripe_head *sh);
617extern struct stripe_head *
618raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
619 int previous, int noblock, int noquiesce);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620#endif