blob: f6536399677abcb43e973017ab1f3773bdad90a3 [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*/
Shaohua Lif6bed0e2015-08-13 14:31:59 -0700226
Artur Paszkiewicz3418d032017-03-09 09:59:59 +0100227 union {
228 struct r5l_io_unit *log_io;
229 struct ppl_io_unit *ppl_io;
230 };
231
Shaohua Lif6bed0e2015-08-13 14:31:59 -0700232 struct list_head log_list;
Song Liua39f7af2016-11-17 15:24:40 -0800233 sector_t log_start; /* first meta block on the journal */
234 struct list_head r5c; /* for r5c_cache->stripe_in_journal */
Artur Paszkiewicz3418d032017-03-09 09:59:59 +0100235
236 struct page *ppl_page; /* partial parity of this stripe */
Dan Williams417b8d42009-10-16 16:25:22 +1100237 /**
238 * struct stripe_operations
Dan Williams91c00922007-01-02 13:52:30 -0700239 * @target - STRIPE_OP_COMPUTE_BLK target
Dan Williams417b8d42009-10-16 16:25:22 +1100240 * @target2 - 2nd compute target in the raid6 case
241 * @zero_sum_result - P and Q verification flags
242 * @request - async service request flags for raid_run_ops
Dan Williams91c00922007-01-02 13:52:30 -0700243 */
244 struct stripe_operations {
Dan Williamsac6b53b2009-07-14 13:40:19 -0700245 int target, target2;
Dan Williamsad283ea2009-08-29 19:09:26 -0700246 enum sum_check_flags zero_sum_result;
Dan Williams91c00922007-01-02 13:52:30 -0700247 } ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 struct r5dev {
NeilBrown671488c2011-12-23 10:17:52 +1100249 /* rreq and rvec are used for the replacement device when
250 * writing data to both devices.
251 */
252 struct bio req, rreq;
253 struct bio_vec vec, rvec;
Shaohua Lid592a992014-05-21 17:57:44 +0800254 struct page *page, *orig_page;
Dan Williams91c00922007-01-02 13:52:30 -0700255 struct bio *toread, *read, *towrite, *written;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 sector_t sector; /* sector of this page */
257 unsigned long flags;
Shaohua Lif6bed0e2015-08-13 14:31:59 -0700258 u32 log_checksum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 } dev[1]; /* allocated with extra space depending of RAID geometry */
260};
Dan Williamsa4456852007-07-09 11:56:43 -0700261
262/* stripe_head_state - collects and tracks the dynamic state of a stripe_head
NeilBrownc4c16632011-07-26 11:34:20 +1000263 * for handle_stripe.
Dan Williamsa4456852007-07-09 11:56:43 -0700264 */
265struct stripe_head_state {
NeilBrown9a3e1102011-12-23 10:17:53 +1100266 /* 'syncing' means that we need to read all devices, either
267 * to check/correct parity, or to reconstruct a missing device.
268 * 'replacing' means we are replacing one or more drives and
269 * the source is valid at this point so we don't need to
270 * read all devices, just the replacement targets.
271 */
272 int syncing, expanding, expanded, replacing;
Dan Williamsa4456852007-07-09 11:56:43 -0700273 int locked, uptodate, to_read, to_write, failed, written;
Dan Williamsb5e98d62007-01-02 13:52:31 -0700274 int to_fill, compute, req_compute, non_overwrite;
Song Liu1e6d6902016-11-17 15:24:39 -0800275 int injournal, just_cached;
NeilBrownf2b3b442011-07-26 11:35:19 +1000276 int failed_num[2];
NeilBrownf2b3b442011-07-26 11:35:19 +1000277 int p_failed, q_failed;
NeilBrownc5709ef2011-07-26 11:35:20 +1000278 int dec_preread_active;
279 unsigned long ops_request;
280
NeilBrown3cb03002011-10-11 16:45:26 +1100281 struct md_rdev *blocked_rdev;
NeilBrownbc2607f2011-07-28 11:39:22 +1000282 int handle_bad_blocks;
Shaohua Li6e74a9c2015-10-08 21:54:08 -0700283 int log_failed;
Song Liud7bd3982016-11-23 22:50:39 -0800284 int waiting_extra_page;
Dan Williamsa4456852007-07-09 11:56:43 -0700285};
286
NeilBrown671488c2011-12-23 10:17:52 +1100287/* Flags for struct r5dev.flags */
288enum r5dev_flags {
289 R5_UPTODATE, /* page contains current data */
290 R5_LOCKED, /* IO has been submitted on "req" */
NeilBrown977df362011-12-23 10:17:53 +1100291 R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */
NeilBrown671488c2011-12-23 10:17:52 +1100292 R5_OVERWRITE, /* towrite covers whole page */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293/* and some that are internal to handle_stripe */
NeilBrown671488c2011-12-23 10:17:52 +1100294 R5_Insync, /* rdev && rdev->in_sync at start */
295 R5_Wantread, /* want to schedule a read */
296 R5_Wantwrite,
297 R5_Overlap, /* There is a pending overlapping request
298 * on this block */
majianpeng3f9e7c12012-07-31 10:04:21 +1000299 R5_ReadNoMerge, /* prevent bio from merging in block-layer */
NeilBrown671488c2011-12-23 10:17:52 +1100300 R5_ReadError, /* seen a read error here recently */
301 R5_ReWrite, /* have tried to over-write the readerror */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
NeilBrown671488c2011-12-23 10:17:52 +1100303 R5_Expanded, /* This block now has post-expand data */
304 R5_Wantcompute, /* compute_block in progress treat as
305 * uptodate
306 */
307 R5_Wantfill, /* dev->toread contains a bio that needs
308 * filling
309 */
310 R5_Wantdrain, /* dev->towrite needs to be drained */
311 R5_WantFUA, /* Write should be FUA */
Shaohua Libc0934f2012-05-22 13:55:05 +1000312 R5_SyncIO, /* The IO is sync */
NeilBrown671488c2011-12-23 10:17:52 +1100313 R5_WriteError, /* got a write error - need to record it */
314 R5_MadeGood, /* A bad block has been fixed by writing to it */
315 R5_ReadRepl, /* Will/did read from replacement rather than orig */
316 R5_MadeGoodRepl,/* A bad block on the replacement device has been
317 * fixed by writing to it */
NeilBrown9a3e1102011-12-23 10:17:53 +1100318 R5_NeedReplace, /* This device has a replacement which is not
319 * up-to-date at this stripe. */
320 R5_WantReplace, /* We need to update the replacement, we have read
321 * data in, and now is a good time to write it out.
322 */
Shaohua Li620125f2012-10-11 13:49:05 +1100323 R5_Discard, /* Discard the stripe */
Shaohua Lid592a992014-05-21 17:57:44 +0800324 R5_SkipCopy, /* Don't copy data from bio to stripe cache */
Song Liu2ded3702016-11-17 15:24:38 -0800325 R5_InJournal, /* data being written is in the journal device.
326 * if R5_InJournal is set for parity pd_idx, all the
327 * data and parity being written are in the journal
328 * device
329 */
Song Liu86aa1392017-01-12 17:22:41 -0800330 R5_OrigPageUPTDODATE, /* with write back cache, we read old data into
331 * dev->orig_page for prexor. When this flag is
332 * set, orig_page contains latest data in the
333 * raid disk.
334 */
NeilBrown671488c2011-12-23 10:17:52 +1100335};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
337/*
338 * Stripe state
339 */
NeilBrown83206d62011-07-26 11:19:49 +1000340enum {
NeilBrownc4c16632011-07-26 11:34:20 +1000341 STRIPE_ACTIVE,
NeilBrown83206d62011-07-26 11:19:49 +1000342 STRIPE_HANDLE,
343 STRIPE_SYNC_REQUESTED,
344 STRIPE_SYNCING,
345 STRIPE_INSYNC,
NeilBrownf94c0b62013-07-22 12:57:21 +1000346 STRIPE_REPLACED,
NeilBrown83206d62011-07-26 11:19:49 +1000347 STRIPE_PREREAD_ACTIVE,
348 STRIPE_DELAYED,
349 STRIPE_DEGRADED,
350 STRIPE_BIT_DELAY,
351 STRIPE_EXPANDING,
352 STRIPE_EXPAND_SOURCE,
353 STRIPE_EXPAND_READY,
354 STRIPE_IO_STARTED, /* do not count towards 'bypass_count' */
355 STRIPE_FULL_WRITE, /* all blocks are set to be overwritten */
356 STRIPE_BIOFILL_RUN,
357 STRIPE_COMPUTE_RUN,
358 STRIPE_OPS_REQ_PENDING,
Shaohua Li8811b592012-08-02 08:33:00 +1000359 STRIPE_ON_UNPLUG_LIST,
NeilBrownf8dfcff2013-03-12 12:18:06 +1100360 STRIPE_DISCARD,
Shaohua Li773ca822013-08-27 17:50:39 +0800361 STRIPE_ON_RELEASE_LIST,
shli@kernel.orgda41ba62014-12-15 12:57:03 +1100362 STRIPE_BATCH_READY,
shli@kernel.org72ac7332014-12-15 12:57:03 +1100363 STRIPE_BATCH_ERR,
NeilBrownd0852df52015-05-27 08:43:45 +1000364 STRIPE_BITMAP_PENDING, /* Being added to bitmap, don't add
365 * to batch yet.
366 */
Song Liu2ded3702016-11-17 15:24:38 -0800367 STRIPE_LOG_TRAPPED, /* trapped into log (see raid5-cache.c)
368 * this bit is used in two scenarios:
369 *
370 * 1. write-out phase
371 * set in first entry of r5l_write_stripe
372 * clear in second entry of r5l_write_stripe
373 * used to bypass logic in handle_stripe
374 *
375 * 2. caching phase
376 * set in r5c_try_caching_write()
377 * clear when journal write is done
378 * used to initiate r5c_cache_data()
379 * also used to bypass logic in handle_stripe
380 */
381 STRIPE_R5C_CACHING, /* the stripe is in caching phase
382 * see more detail in the raid5-cache.c
383 */
Song Liu1e6d6902016-11-17 15:24:39 -0800384 STRIPE_R5C_PARTIAL_STRIPE, /* in r5c cache (to-be/being handled or
385 * in conf->r5c_partial_stripe_list)
386 */
387 STRIPE_R5C_FULL_STRIPE, /* in r5c cache (to-be/being handled or
388 * in conf->r5c_full_stripe_list)
389 */
Song Liu3bddb7f2016-11-18 16:46:50 -0800390 STRIPE_R5C_PREFLUSH, /* need to flush journal device */
NeilBrown83206d62011-07-26 11:19:49 +1000391};
Dan Williams417b8d42009-10-16 16:25:22 +1100392
NeilBrown1b956f72015-05-21 12:40:26 +1000393#define STRIPE_EXPAND_SYNC_FLAGS \
shli@kernel.orgdabc4ec2014-12-15 12:57:04 +1100394 ((1 << STRIPE_EXPAND_SOURCE) |\
395 (1 << STRIPE_EXPAND_READY) |\
396 (1 << STRIPE_EXPANDING) |\
397 (1 << STRIPE_SYNC_REQUESTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398/*
Dan Williamsecc65c92008-06-28 08:31:57 +1000399 * Operation request flags
Dan Williams91c00922007-01-02 13:52:30 -0700400 */
NeilBrownede7ee82011-12-23 10:17:52 +1100401enum {
402 STRIPE_OP_BIOFILL,
403 STRIPE_OP_COMPUTE_BLK,
404 STRIPE_OP_PREXOR,
405 STRIPE_OP_BIODRAIN,
406 STRIPE_OP_RECONSTRUCT,
407 STRIPE_OP_CHECK,
Artur Paszkiewicz3418d032017-03-09 09:59:59 +0100408 STRIPE_OP_PARTIAL_PARITY,
NeilBrownede7ee82011-12-23 10:17:52 +1100409};
Markus Stockhausen584acdd2014-12-15 12:57:05 +1100410
411/*
412 * RAID parity calculation preferences
413 */
414enum {
415 PARITY_DISABLE_RMW = 0,
416 PARITY_ENABLE_RMW,
Markus Stockhausend06f1912014-12-15 12:57:05 +1100417 PARITY_PREFER_RMW,
Markus Stockhausen584acdd2014-12-15 12:57:05 +1100418};
419
420/*
421 * Pages requested from set_syndrome_sources()
422 */
423enum {
424 SYNDROME_SRC_ALL,
425 SYNDROME_SRC_WANT_DRAIN,
426 SYNDROME_SRC_WRITTEN,
427};
Dan Williams91c00922007-01-02 13:52:30 -0700428/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 * Plugging:
430 *
431 * To improve write throughput, we need to delay the handling of some
432 * stripes until there has been a chance that several write requests
433 * for the one stripe have all been collected.
434 * In particular, any write request that would require pre-reading
435 * is put on a "delayed" queue until there are no stripes currently
436 * in a pre-read phase. Further, if the "delayed" queue is empty when
437 * a stripe is put on it then we "plug" the queue and do not process it
438 * until an unplug call is made. (the unplug_io_fn() is called).
439 *
440 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
441 * it to the count of prereading stripes.
442 * When write is initiated, or the stripe refcnt == 0 (just in case) we
443 * clear the PREREAD_ACTIVE flag and decrement the count
NeilBrownb5c124a2006-10-03 01:15:45 -0700444 * Whenever the 'handle' queue is empty and the device is not plugged, we
445 * move any strips from delayed to handle and clear the DELAYED flag and set
446 * PREREAD_ACTIVE.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 * In stripe_handle, if we find pre-reading is necessary, we do it if
448 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
NeilBrownc4c16632011-07-26 11:34:20 +1000449 * HANDLE gets cleared if stripe_handle leaves nothing locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 */
Christoph Hellwigef740c32009-03-31 14:27:03 +1100451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452struct disk_info {
NeilBrown671488c2011-12-23 10:17:52 +1100453 struct md_rdev *rdev, *replacement;
Song Liud7bd3982016-11-23 22:50:39 -0800454 struct page *extra_page; /* extra page to use in prexor */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455};
456
Song Liu937621c2016-11-17 15:24:37 -0800457/*
458 * Stripe cache
459 */
460
461#define NR_STRIPES 256
462#define STRIPE_SIZE PAGE_SIZE
463#define STRIPE_SHIFT (PAGE_SHIFT - 9)
464#define STRIPE_SECTORS (STRIPE_SIZE>>9)
465#define IO_THRESHOLD 1
466#define BYPASS_THRESHOLD 1
467#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
468#define HASH_MASK (NR_HASH - 1)
469#define MAX_STRIPE_BATCH 8
470
471/* bio's attached to a stripe+device for I/O are linked together in bi_sector
472 * order without overlap. There may be several bio's per stripe+device, and
473 * a bio could span several devices.
474 * When walking this list for a particular stripe+device, we must never proceed
475 * beyond a bio that extends past this device, as the next bio might no longer
476 * be valid.
477 * This function is used to determine the 'next' bio in the list, given the
478 * sector of the current stripe+device
479 */
480static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
481{
482 int sectors = bio_sectors(bio);
483
484 if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS)
485 return bio->bi_next;
486 else
487 return NULL;
488}
489
Shaohua Li566c09c2013-11-14 15:16:17 +1100490/* NOTE NR_STRIPE_HASH_LOCKS must remain below 64.
491 * This is because we sometimes take all the spinlocks
492 * and creating that much locking depth can cause
493 * problems.
494 */
495#define NR_STRIPE_HASH_LOCKS 8
496#define STRIPE_HASH_LOCKS_MASK (NR_STRIPE_HASH_LOCKS - 1)
497
Shaohua Li851c30c2013-08-28 14:30:16 +0800498struct r5worker {
499 struct work_struct work;
500 struct r5worker_group *group;
Shaohua Li566c09c2013-11-14 15:16:17 +1100501 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Libfc90cb2013-08-29 15:40:32 +0800502 bool working;
Shaohua Li851c30c2013-08-28 14:30:16 +0800503};
504
505struct r5worker_group {
506 struct list_head handle_list;
Shaohua Li535ae4e2017-02-15 19:37:32 -0800507 struct list_head loprio_list;
Shaohua Li851c30c2013-08-28 14:30:16 +0800508 struct r5conf *conf;
509 struct r5worker *workers;
Shaohua Libfc90cb2013-08-29 15:40:32 +0800510 int stripes_cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +0800511};
512
Heinz Mauelshagen78e470c2017-03-22 17:44:37 +0100513/*
514 * r5c journal modes of the array: write-back or write-through.
515 * write-through mode has identical behavior as existing log only
516 * implementation.
517 */
518enum r5c_journal_mode {
519 R5C_JOURNAL_MODE_WRITE_THROUGH = 0,
520 R5C_JOURNAL_MODE_WRITE_BACK = 1,
521};
522
Song Liua39f7af2016-11-17 15:24:40 -0800523enum r5_cache_state {
524 R5_INACTIVE_BLOCKED, /* release of inactive stripes blocked,
525 * waiting for 25% to be free
526 */
527 R5_ALLOC_MORE, /* It might help to allocate another
528 * stripe.
529 */
530 R5_DID_ALLOC, /* A stripe was allocated, don't allocate
531 * more until at least one has been
532 * released. This avoids flooding
533 * the cache.
534 */
535 R5C_LOG_TIGHT, /* log device space tight, need to
536 * prioritize stripes at last_checkpoint
537 */
538 R5C_LOG_CRITICAL, /* log device is running out of space,
539 * only process stripes that are already
540 * occupying the log
541 */
Song Liud7bd3982016-11-23 22:50:39 -0800542 R5C_EXTRA_PAGE_IN_USE, /* a stripe is using disk_info.extra_page
543 * for prexor
544 */
Song Liua39f7af2016-11-17 15:24:40 -0800545};
546
Shaohua Liaaf9f122017-03-03 22:06:12 -0800547#define PENDING_IO_MAX 512
548#define PENDING_IO_ONE_FLUSH 128
549struct r5pending_data {
550 struct list_head sibling;
551 sector_t sector; /* stripe sector */
552 struct bio_list bios;
553};
554
NeilBrownd1688a62011-10-11 16:49:52 +1100555struct r5conf {
NeilBrownfccddba2006-01-06 00:20:33 -0800556 struct hlist_head *stripe_hashtbl;
Shaohua Li566c09c2013-11-14 15:16:17 +1100557 /* only protect corresponding hash list and inactive_list */
558 spinlock_t hash_locks[NR_STRIPE_HASH_LOCKS];
NeilBrownfd01b882011-10-11 16:47:53 +1100559 struct mddev *mddev;
Andre Noll09c9e5f2009-06-18 08:45:55 +1000560 int chunk_sectors;
Markus Stockhausen584acdd2014-12-15 12:57:05 +1100561 int level, algorithm, rmw_level;
NeilBrown16a53ec2006-06-26 00:27:38 -0700562 int max_degraded;
NeilBrown02c2de82006-10-03 01:15:47 -0700563 int raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 int max_nr_stripes;
NeilBrownedbe83a2015-02-26 12:47:56 +1100565 int min_nr_stripes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
NeilBrownfef9c612009-03-31 15:16:46 +1100567 /* reshape_progress is the leading edge of a 'reshape'
568 * It has value MaxSector when no reshape is happening
569 * If delta_disks < 0, it is the last sector we started work on,
570 * else is it the next sector to work on.
571 */
572 sector_t reshape_progress;
573 /* reshape_safe is the trailing edge of a reshape. We know that
574 * before (or after) this address, all reshape has completed.
575 */
576 sector_t reshape_safe;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800577 int previous_raid_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +1000578 int prev_chunk_sectors;
579 int prev_algo;
NeilBrown86b42c72009-03-31 15:19:03 +1100580 short generation; /* increments with every reshape */
NeilBrownc46501b2013-08-27 15:52:13 +1000581 seqcount_t gen_lock; /* lock against generation changes */
NeilBrownc8f517c2009-03-31 15:28:40 +1100582 unsigned long reshape_checkpoint; /* Time we last updated
583 * metadata */
NeilBrownb5254dd2012-05-21 09:27:01 +1000584 long long min_offset_diff; /* minimum difference between
585 * data_offset and
586 * new_data_offset across all
587 * devices. May be negative,
588 * but is closest to zero.
589 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 struct list_head handle_list; /* stripes needing handling */
Shaohua Li535ae4e2017-02-15 19:37:32 -0800592 struct list_head loprio_list; /* low priority stripes */
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700593 struct list_head hold_list; /* preread ready stripes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 struct list_head delayed_list; /* stripes that have plugged requests */
NeilBrown72626682005-09-09 16:23:54 -0700595 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800596 struct bio *retry_read_aligned; /* currently retrying aligned bios */
NeilBrown0472a422017-03-15 14:05:13 +1100597 unsigned int retry_read_offset; /* sector offset into retry_read_aligned */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800598 struct bio *retry_read_aligned_list; /* aligned bios retry list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 atomic_t preread_active_stripes; /* stripes with scheduled io */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800600 atomic_t active_aligned_reads;
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700601 atomic_t pending_full_writes; /* full write backlog */
602 int bypass_count; /* bypassed prereads */
603 int bypass_threshold; /* preread nice */
Shaohua Lid592a992014-05-21 17:57:44 +0800604 int skip_copy; /* Don't copy data from bio to stripe cache */
Dan Williams8b3e6cd2008-04-28 02:15:53 -0700605 struct list_head *last_hold; /* detect hold_list promotions */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
NeilBrownf6705572006-03-27 01:18:11 -0800607 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
NeilBrownad01c9e2006-03-27 01:18:07 -0800608 /* unfortunately we need two cache names as we temporarily have
609 * two caches.
610 */
611 int active_name;
NeilBrownf4be6b42010-06-01 19:37:25 +1000612 char cache_name[2][32];
NeilBrown2d5b5692015-07-06 12:49:23 +1000613 struct kmem_cache *slab_cache; /* for allocating stripes */
614 struct mutex cache_size_mutex; /* Protect changes to cache size */
NeilBrown72626682005-09-09 16:23:54 -0700615
616 int seq_flush, seq_write;
617 int quiesce;
618
619 int fullsync; /* set to 1 if a full sync is needed,
620 * (fresh device added).
621 * Cleared when a sync completes.
622 */
NeilBrown7f0da592011-07-28 11:39:22 +1000623 int recovery_disabled;
Dan Williams36d1c642009-07-14 11:48:22 -0700624 /* per cpu variables */
625 struct raid5_percpu {
626 struct page *spare_page; /* Used when checking P/Q in raid6 */
shli@kernel.org46d5b782014-12-15 12:57:02 +1100627 struct flex_array *scribble; /* space for constructing buffer
Dan Williamsd6f38f32009-07-14 11:50:52 -0700628 * lists and performing address
629 * conversions
630 */
Tejun Heoa29d8b82010-02-02 14:39:15 +0900631 } __percpu *percpu;
Shaohua Li27a353c2016-02-24 17:38:28 -0800632 int scribble_disks;
633 int scribble_sectors;
Sebastian Andrzej Siewior29c6d1b2016-08-18 14:57:24 +0200634 struct hlist_node node;
NeilBrownca65b732006-01-06 00:20:17 -0800635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 /*
637 * Free stripes pool
638 */
639 atomic_t active_stripes;
Shaohua Li566c09c2013-11-14 15:16:17 +1100640 struct list_head inactive_list[NR_STRIPE_HASH_LOCKS];
Song Liu1e6d6902016-11-17 15:24:39 -0800641
642 atomic_t r5c_cached_full_stripes;
643 struct list_head r5c_full_stripe_list;
644 atomic_t r5c_cached_partial_stripes;
645 struct list_head r5c_partial_stripe_list;
Shaohua Lie33fbb92017-02-10 16:18:09 -0800646 atomic_t r5c_flushing_full_stripes;
647 atomic_t r5c_flushing_partial_stripes;
Song Liu1e6d6902016-11-17 15:24:39 -0800648
Shaohua Li4bda5562013-11-14 15:16:17 +1100649 atomic_t empty_inactive_list_nr;
Shaohua Li773ca822013-08-27 17:50:39 +0800650 struct llist_head released_stripes;
Yuanhan Liub1b46482015-05-08 18:19:06 +1000651 wait_queue_head_t wait_for_quiescent;
Shaohua Li6ab2a4b2016-02-25 16:24:42 -0800652 wait_queue_head_t wait_for_stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 wait_queue_head_t wait_for_overlap;
NeilBrown54233992015-02-26 12:21:04 +1100654 unsigned long cache_state;
NeilBrownedbe83a2015-02-26 12:47:56 +1100655 struct shrinker shrinker;
NeilBrownad01c9e2006-03-27 01:18:07 -0800656 int pool_size; /* number of disks in stripeheads in pool */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 spinlock_t device_lock;
NeilBrownb55e6bf2006-03-27 01:18:06 -0800658 struct disk_info *disks;
NeilBrowndd7a8f52017-04-05 14:05:51 +1000659 struct bio_set *bio_split;
NeilBrown91adb562009-03-31 14:39:39 +1100660
661 /* When taking over an array from a different personality, we store
662 * the new thread here until we fully activate the array.
663 */
NeilBrown2b8bf342011-10-11 16:48:23 +1100664 struct md_thread *thread;
Shaohua Li566c09c2013-11-14 15:16:17 +1100665 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Li851c30c2013-08-28 14:30:16 +0800666 struct r5worker_group *worker_groups;
667 int group_cnt;
668 int worker_cnt_per_group;
Shaohua Lif6bed0e2015-08-13 14:31:59 -0700669 struct r5l_log *log;
Artur Paszkiewicz3418d032017-03-09 09:59:59 +0100670 void *log_private;
Shaohua Li765d7042017-01-04 09:33:23 -0800671
Shaohua Li765d7042017-01-04 09:33:23 -0800672 spinlock_t pending_bios_lock;
673 bool batch_bio_dispatch;
Shaohua Liaaf9f122017-03-03 22:06:12 -0800674 struct r5pending_data *pending_data;
675 struct list_head free_list;
676 struct list_head pending_list;
677 int pending_data_cnt;
678 struct r5pending_data *next_pending_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679};
680
NeilBrown54233992015-02-26 12:21:04 +1100681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682/*
683 * Our supported algorithms
684 */
NeilBrown99c0fb52009-03-31 14:39:38 +1100685#define ALGORITHM_LEFT_ASYMMETRIC 0 /* Rotating Parity N with Data Restart */
686#define ALGORITHM_RIGHT_ASYMMETRIC 1 /* Rotating Parity 0 with Data Restart */
687#define ALGORITHM_LEFT_SYMMETRIC 2 /* Rotating Parity N with Data Continuation */
688#define ALGORITHM_RIGHT_SYMMETRIC 3 /* Rotating Parity 0 with Data Continuation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
NeilBrown99c0fb52009-03-31 14:39:38 +1100690/* Define non-rotating (raid4) algorithms. These allow
691 * conversion of raid4 to raid5.
692 */
693#define ALGORITHM_PARITY_0 4 /* P or P,Q are initial devices */
694#define ALGORITHM_PARITY_N 5 /* P or P,Q are final devices. */
695
696/* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
697 * Firstly, the exact positioning of the parity block is slightly
698 * different between the 'LEFT_*' modes of md and the "_N_*" modes
699 * of DDF.
700 * Secondly, or order of datablocks over which the Q syndrome is computed
701 * is different.
702 * Consequently we have different layouts for DDF/raid6 than md/raid6.
703 * These layouts are from the DDFv1.2 spec.
704 * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
705 * leaves RLQ=3 as 'Vendor Specific'
706 */
707
708#define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
709#define ALGORITHM_ROTATING_N_RESTART 9 /* DDF PRL=6 RLQ=2 */
710#define ALGORITHM_ROTATING_N_CONTINUE 10 /*DDF PRL=6 RLQ=3 */
711
NeilBrown99c0fb52009-03-31 14:39:38 +1100712/* For every RAID5 algorithm we define a RAID6 algorithm
713 * with exactly the same layout for data and parity, and
714 * with the Q block always on the last device (N-1).
715 * This allows trivial conversion from RAID5 to RAID6
716 */
717#define ALGORITHM_LEFT_ASYMMETRIC_6 16
718#define ALGORITHM_RIGHT_ASYMMETRIC_6 17
719#define ALGORITHM_LEFT_SYMMETRIC_6 18
720#define ALGORITHM_RIGHT_SYMMETRIC_6 19
721#define ALGORITHM_PARITY_0_6 20
722#define ALGORITHM_PARITY_N_6 ALGORITHM_PARITY_N
723
724static inline int algorithm_valid_raid5(int layout)
725{
726 return (layout >= 0) &&
727 (layout <= 5);
728}
729static inline int algorithm_valid_raid6(int layout)
730{
731 return (layout >= 0 && layout <= 5)
732 ||
NeilBrowne4424fe2009-10-16 16:27:34 +1100733 (layout >= 8 && layout <= 10)
NeilBrown99c0fb52009-03-31 14:39:38 +1100734 ||
735 (layout >= 16 && layout <= 20);
736}
737
738static inline int algorithm_is_DDF(int layout)
739{
740 return layout >= 8 && layout <= 10;
741}
NeilBrown11d8a6e2010-07-26 11:57:07 +1000742
NeilBrownd1688a62011-10-11 16:49:52 +1100743extern void md_raid5_kick_device(struct r5conf *conf);
NeilBrownfd01b882011-10-11 16:47:53 +1100744extern int raid5_set_cache_size(struct mddev *mddev, int size);
Shaohua Li6d036f72015-08-13 14:31:57 -0700745extern sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous);
746extern void raid5_release_stripe(struct stripe_head *sh);
747extern sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
748 int previous, int *dd_idx,
749 struct stripe_head *sh);
750extern struct stripe_head *
751raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
752 int previous, int noblock, int noquiesce);
Song Liu2e38a372017-01-24 10:45:30 -0800753extern int raid5_calc_degraded(struct r5conf *conf);
Heinz Mauelshagen78e470c2017-03-22 17:44:37 +0100754extern int r5c_journal_mode_set(struct mddev *mddev, int journal_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755#endif