Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
1da177e4 LT |
2 | #ifndef _RAID5_H |
3 | #define _RAID5_H | |
4 | ||
1da177e4 | 5 | #include <linux/raid/xor.h> |
ad283ea4 | 6 | #include <linux/dmaengine.h> |
770b1d21 | 7 | #include <linux/local_lock.h> |
1da177e4 LT |
8 | |
9 | /* | |
10 | * | |
c4c1663b | 11 | * Each stripe contains one buffer per device. Each buffer can be in |
1da177e4 | 12 | * one of a number of states stored in "flags". Changes between |
c4c1663b N |
13 | * these states happen *almost* exclusively under the protection of the |
14 | * STRIPE_ACTIVE flag. Some very specific changes can happen in bi_end_io, and | |
15 | * these are not protected by STRIPE_ACTIVE. | |
1da177e4 LT |
16 | * |
17 | * The flag bits that are used to represent these states are: | |
18 | * R5_UPTODATE and R5_LOCKED | |
19 | * | |
20 | * State Empty == !UPTODATE, !LOCK | |
21 | * We have no data, and there is no active request | |
22 | * State Want == !UPTODATE, LOCK | |
23 | * A read request is being submitted for this block | |
24 | * State Dirty == UPTODATE, LOCK | |
25 | * Some new data is in this buffer, and it is being written out | |
26 | * State Clean == UPTODATE, !LOCK | |
27 | * We have valid data which is the same as on disc | |
28 | * | |
29 | * The possible state transitions are: | |
30 | * | |
31 | * Empty -> Want - on read or write to get old data for parity calc | |
ede7ee8b | 32 | * Empty -> Dirty - on compute_parity to satisfy write/sync request. |
1da177e4 LT |
33 | * Empty -> Clean - on compute_block when computing a block for failed drive |
34 | * Want -> Empty - on failed read | |
35 | * Want -> Clean - on successful completion of read request | |
36 | * Dirty -> Clean - on successful completion of write request | |
37 | * Dirty -> Clean - on failed write | |
38 | * Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW) | |
39 | * | |
40 | * The Want->Empty, Want->Clean, Dirty->Clean, transitions | |
41 | * all happen in b_end_io at interrupt time. | |
42 | * Each sets the Uptodate bit before releasing the Lock bit. | |
43 | * This leaves one multi-stage transition: | |
44 | * Want->Dirty->Clean | |
45 | * This is safe because thinking that a Clean buffer is actually dirty | |
46 | * will at worst delay some action, and the stripe will be scheduled | |
47 | * for attention after the transition is complete. | |
48 | * | |
49 | * There is one possibility that is not covered by these states. That | |
50 | * is if one drive has failed and there is a spare being rebuilt. We | |
51 | * can't distinguish between a clean block that has been generated | |
52 | * from parity calculations, and a clean block that has been | |
53 | * successfully written to the spare ( or to parity when resyncing). | |
aa5e5dc2 | 54 | * To distinguish these states we have a stripe bit STRIPE_INSYNC that |
1da177e4 LT |
55 | * is set whenever a write is scheduled to the spare, or to the parity |
56 | * disc if there is no spare. A sync request clears this bit, and | |
57 | * when we find it set with no buffers locked, we know the sync is | |
58 | * complete. | |
59 | * | |
60 | * Buffers for the md device that arrive via make_request are attached | |
61 | * to the appropriate stripe in one of two lists linked on b_reqnext. | |
62 | * One list (bh_read) for read requests, one (bh_write) for write. | |
63 | * There should never be more than one buffer on the two lists | |
64 | * together, but we are not guaranteed of that so we allow for more. | |
65 | * | |
66 | * If a buffer is on the read list when the associated cache buffer is | |
67 | * Uptodate, the data is copied into the read buffer and it's b_end_io | |
68 | * routine is called. This may happen in the end_request routine only | |
69 | * if the buffer has just successfully been read. end_request should | |
70 | * remove the buffers from the list and then set the Uptodate bit on | |
71 | * the buffer. Other threads may do this only if they first check | |
72 | * that the Uptodate bit is set. Once they have checked that they may | |
73 | * take buffers off the read queue. | |
74 | * | |
75 | * When a buffer on the write list is committed for write it is copied | |
76 | * into the cache buffer, which is then marked dirty, and moved onto a | |
77 | * third list, the written list (bh_written). Once both the parity | |
78 | * block and the cached buffer are successfully written, any buffer on | |
79 | * a written list can be returned with b_end_io. | |
80 | * | |
c4c1663b N |
81 | * The write list and read list both act as fifos. The read list, |
82 | * write list and written list are protected by the device_lock. | |
83 | * The device_lock is only for list manipulations and will only be | |
84 | * held for a very short time. It can be claimed from interrupts. | |
1da177e4 LT |
85 | * |
86 | * | |
87 | * Stripes in the stripe cache can be on one of two lists (or on | |
88 | * neither). The "inactive_list" contains stripes which are not | |
89 | * currently being used for any request. They can freely be reused | |
90 | * for another stripe. The "handle_list" contains stripes that need | |
91 | * to be handled in some way. Both of these are fifo queues. Each | |
92 | * stripe is also (potentially) linked to a hash bucket in the hash | |
93 | * table so that it can be found by sector number. Stripes that are | |
94 | * not hashed must be on the inactive_list, and will normally be at | |
95 | * the front. All stripes start life this way. | |
96 | * | |
97 | * The inactive_list, handle_list and hash bucket lists are all protected by the | |
98 | * device_lock. | |
1da177e4 LT |
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()) | |
c4c1663b | 118 | * setSTRIPE_ACTIVE, clrSTRIPE_HANDLE ... |
91c00924 DW |
119 | * (lockdev check-buffers unlockdev) .. |
120 | * change-state .. | |
c4c1663b | 121 | * record io/ops needed clearSTRIPE_ACTIVE schedule io/ops |
1da177e4 LT |
122 | * 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 | |
91c00924 DW |
127 | * on a cached buffer, and plus one if the stripe is undergoing stripe |
128 | * operations. | |
129 | * | |
c4c1663b | 130 | * The stripe operations are: |
91c00924 DW |
131 | * -copying data between the stripe cache and user application buffers |
132 | * -computing blocks to save a disk access, or to recover a missing block | |
133 | * -updating the parity on a write operation (reconstruct write and | |
134 | * read-modify-write) | |
135 | * -checking parity correctness | |
136 | * -running i/o to disk | |
137 | * These operations are carried out by raid5_run_ops which uses the async_tx | |
138 | * api to (optionally) offload operations to dedicated hardware engines. | |
139 | * When requesting an operation handle_stripe sets the pending bit for the | |
140 | * operation and increments the count. raid5_run_ops is then run whenever | |
141 | * the count is non-zero. | |
142 | * There are some critical dependencies between the operations that prevent some | |
143 | * from being requested while another is in flight. | |
144 | * 1/ Parity check operations destroy the in cache version of the parity block, | |
145 | * so we prevent parity dependent operations like writes and compute_blocks | |
146 | * from starting while a check is in progress. Some dma engines can perform | |
147 | * the check without damaging the parity block, in these cases the parity | |
148 | * block is re-marked up to date (assuming the check was successful) and is | |
149 | * not re-read from disk. | |
150 | * 2/ When a write operation is requested we immediately lock the affected | |
151 | * blocks, and mark them as not up to date. This causes new read requests | |
152 | * to be held off, as well as parity checks and compute block operations. | |
153 | * 3/ Once a compute block operation has been requested handle_stripe treats | |
154 | * that block as if it is up to date. raid5_run_ops guaruntees that any | |
155 | * operation that is dependent on the compute block result is initiated after | |
156 | * the compute block completes. | |
1da177e4 LT |
157 | */ |
158 | ||
ecc65c9b | 159 | /* |
f72ffdd6 | 160 | * Operations state - intermediate states that are visible outside of |
c4c1663b | 161 | * STRIPE_ACTIVE. |
ecc65c9b DW |
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 | */ | |
176 | enum check_states { | |
177 | check_state_idle = 0, | |
ac6b53b6 DW |
178 | check_state_run, /* xor parity check */ |
179 | check_state_run_q, /* q-parity check */ | |
180 | check_state_run_pq, /* pq dual parity check */ | |
ecc65c9b DW |
181 | check_state_check_result, |
182 | check_state_compute_run, /* parity repair */ | |
183 | check_state_compute_result, | |
184 | }; | |
185 | ||
186 | /** | |
187 | * enum reconstruct_states - handles writing or expanding a stripe | |
188 | */ | |
189 | enum reconstruct_states { | |
190 | reconstruct_state_idle = 0, | |
d8ee0728 | 191 | reconstruct_state_prexor_drain_run, /* prexor-write */ |
ecc65c9b DW |
192 | reconstruct_state_drain_run, /* write */ |
193 | reconstruct_state_run, /* expand */ | |
d8ee0728 | 194 | reconstruct_state_prexor_drain_result, |
ecc65c9b DW |
195 | reconstruct_state_drain_result, |
196 | reconstruct_state_result, | |
197 | }; | |
198 | ||
046169f0 | 199 | #define DEFAULT_STRIPE_SIZE 4096 |
1da177e4 | 200 | struct stripe_head { |
fccddba0 | 201 | struct hlist_node hash; |
d0dabf7e | 202 | struct list_head lru; /* inactive_list or handle_list */ |
773ca82f | 203 | struct llist_node release_list; |
d1688a6d | 204 | struct r5conf *raid_conf; |
86b42c71 N |
205 | short generation; /* increments with every |
206 | * reshape */ | |
d0dabf7e N |
207 | sector_t sector; /* sector of this row */ |
208 | short pd_idx; /* parity disk index */ | |
209 | short qd_idx; /* 'Q' disk index for raid6 */ | |
67cc2b81 | 210 | short ddf_layout;/* use DDF ordering to calculate Q */ |
566c09c5 | 211 | short hash_lock_index; |
d0dabf7e N |
212 | unsigned long state; /* state flags */ |
213 | atomic_t count; /* nr of active thread/requests */ | |
72626685 | 214 | int bm_seq; /* sequence number for bitmap flushes */ |
d0dabf7e | 215 | int disks; /* disks in stripe */ |
7a87f434 | 216 | int overwrite_disks; /* total overwrite disks in stripe, |
217 | * this is only checked when stripe | |
218 | * has STRIPE_BATCH_READY | |
219 | */ | |
ecc65c9b | 220 | enum check_states check_state; |
600aa109 | 221 | enum reconstruct_states reconstruct_state; |
b17459c0 | 222 | spinlock_t stripe_lock; |
851c30c9 | 223 | int cpu; |
bfc90cb0 | 224 | struct r5worker_group *group; |
59fc630b | 225 | |
226 | struct stripe_head *batch_head; /* protected by stripe lock */ | |
227 | spinlock_t batch_lock; /* only header's lock is useful */ | |
228 | struct list_head batch_list; /* protected by head's batch lock*/ | |
f6bed0ef | 229 | |
3418d036 AP |
230 | union { |
231 | struct r5l_io_unit *log_io; | |
232 | struct ppl_io_unit *ppl_io; | |
233 | }; | |
234 | ||
f6bed0ef | 235 | struct list_head log_list; |
a39f7afd SL |
236 | sector_t log_start; /* first meta block on the journal */ |
237 | struct list_head r5c; /* for r5c_cache->stripe_in_journal */ | |
3418d036 AP |
238 | |
239 | struct page *ppl_page; /* partial parity of this stripe */ | |
417b8d4a DW |
240 | /** |
241 | * struct stripe_operations | |
91c00924 | 242 | * @target - STRIPE_OP_COMPUTE_BLK target |
417b8d4a DW |
243 | * @target2 - 2nd compute target in the raid6 case |
244 | * @zero_sum_result - P and Q verification flags | |
245 | * @request - async service request flags for raid_run_ops | |
91c00924 DW |
246 | */ |
247 | struct stripe_operations { | |
ac6b53b6 | 248 | int target, target2; |
ad283ea4 | 249 | enum sum_check_flags zero_sum_result; |
91c00924 | 250 | } ops; |
046169f0 YY |
251 | |
252 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE | |
253 | /* These pages will be used by bios in dev[i] */ | |
254 | struct page **pages; | |
255 | int nr_pages; /* page array size */ | |
256 | int stripes_per_page; | |
257 | #endif | |
1da177e4 | 258 | struct r5dev { |
671488cc N |
259 | /* rreq and rvec are used for the replacement device when |
260 | * writing data to both devices. | |
261 | */ | |
262 | struct bio req, rreq; | |
263 | struct bio_vec vec, rvec; | |
d592a996 | 264 | struct page *page, *orig_page; |
7aba13b7 | 265 | unsigned int offset; /* offset of the page */ |
91c00924 | 266 | struct bio *toread, *read, *towrite, *written; |
1da177e4 LT |
267 | sector_t sector; /* sector of this page */ |
268 | unsigned long flags; | |
f6bed0ef | 269 | u32 log_checksum; |
2cd259a7 | 270 | unsigned short write_hint; |
1da177e4 LT |
271 | } dev[1]; /* allocated with extra space depending of RAID geometry */ |
272 | }; | |
a4456856 DW |
273 | |
274 | /* stripe_head_state - collects and tracks the dynamic state of a stripe_head | |
c4c1663b | 275 | * for handle_stripe. |
a4456856 DW |
276 | */ |
277 | struct stripe_head_state { | |
9a3e1101 N |
278 | /* 'syncing' means that we need to read all devices, either |
279 | * to check/correct parity, or to reconstruct a missing device. | |
280 | * 'replacing' means we are replacing one or more drives and | |
281 | * the source is valid at this point so we don't need to | |
282 | * read all devices, just the replacement targets. | |
283 | */ | |
284 | int syncing, expanding, expanded, replacing; | |
a4456856 | 285 | int locked, uptodate, to_read, to_write, failed, written; |
b5e98d65 | 286 | int to_fill, compute, req_compute, non_overwrite; |
1e6d690b | 287 | int injournal, just_cached; |
f2b3b44d | 288 | int failed_num[2]; |
f2b3b44d | 289 | int p_failed, q_failed; |
c5709ef6 N |
290 | int dec_preread_active; |
291 | unsigned long ops_request; | |
292 | ||
3cb03002 | 293 | struct md_rdev *blocked_rdev; |
bc2607f3 | 294 | int handle_bad_blocks; |
6e74a9cf | 295 | int log_failed; |
d7bd398e | 296 | int waiting_extra_page; |
a4456856 DW |
297 | }; |
298 | ||
671488cc N |
299 | /* Flags for struct r5dev.flags */ |
300 | enum r5dev_flags { | |
301 | R5_UPTODATE, /* page contains current data */ | |
302 | R5_LOCKED, /* IO has been submitted on "req" */ | |
977df362 | 303 | R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */ |
671488cc | 304 | R5_OVERWRITE, /* towrite covers whole page */ |
1da177e4 | 305 | /* and some that are internal to handle_stripe */ |
671488cc N |
306 | R5_Insync, /* rdev && rdev->in_sync at start */ |
307 | R5_Wantread, /* want to schedule a read */ | |
308 | R5_Wantwrite, | |
309 | R5_Overlap, /* There is a pending overlapping request | |
310 | * on this block */ | |
3f9e7c14 | 311 | R5_ReadNoMerge, /* prevent bio from merging in block-layer */ |
671488cc N |
312 | R5_ReadError, /* seen a read error here recently */ |
313 | R5_ReWrite, /* have tried to over-write the readerror */ | |
1da177e4 | 314 | |
671488cc N |
315 | R5_Expanded, /* This block now has post-expand data */ |
316 | R5_Wantcompute, /* compute_block in progress treat as | |
317 | * uptodate | |
318 | */ | |
319 | R5_Wantfill, /* dev->toread contains a bio that needs | |
320 | * filling | |
321 | */ | |
322 | R5_Wantdrain, /* dev->towrite needs to be drained */ | |
323 | R5_WantFUA, /* Write should be FUA */ | |
bc0934f0 | 324 | R5_SyncIO, /* The IO is sync */ |
671488cc N |
325 | R5_WriteError, /* got a write error - need to record it */ |
326 | R5_MadeGood, /* A bad block has been fixed by writing to it */ | |
327 | R5_ReadRepl, /* Will/did read from replacement rather than orig */ | |
328 | R5_MadeGoodRepl,/* A bad block on the replacement device has been | |
329 | * fixed by writing to it */ | |
9a3e1101 N |
330 | R5_NeedReplace, /* This device has a replacement which is not |
331 | * up-to-date at this stripe. */ | |
332 | R5_WantReplace, /* We need to update the replacement, we have read | |
333 | * data in, and now is a good time to write it out. | |
334 | */ | |
620125f2 | 335 | R5_Discard, /* Discard the stripe */ |
d592a996 | 336 | R5_SkipCopy, /* Don't copy data from bio to stripe cache */ |
2ded3703 SL |
337 | R5_InJournal, /* data being written is in the journal device. |
338 | * if R5_InJournal is set for parity pd_idx, all the | |
339 | * data and parity being written are in the journal | |
340 | * device | |
341 | */ | |
86aa1397 SL |
342 | R5_OrigPageUPTDODATE, /* with write back cache, we read old data into |
343 | * dev->orig_page for prexor. When this flag is | |
344 | * set, orig_page contains latest data in the | |
345 | * raid disk. | |
346 | */ | |
671488cc | 347 | }; |
1da177e4 LT |
348 | |
349 | /* | |
350 | * Stripe state | |
351 | */ | |
83206d66 | 352 | enum { |
c4c1663b | 353 | STRIPE_ACTIVE, |
83206d66 N |
354 | STRIPE_HANDLE, |
355 | STRIPE_SYNC_REQUESTED, | |
356 | STRIPE_SYNCING, | |
357 | STRIPE_INSYNC, | |
f94c0b66 | 358 | STRIPE_REPLACED, |
83206d66 N |
359 | STRIPE_PREREAD_ACTIVE, |
360 | STRIPE_DELAYED, | |
361 | STRIPE_DEGRADED, | |
362 | STRIPE_BIT_DELAY, | |
363 | STRIPE_EXPANDING, | |
364 | STRIPE_EXPAND_SOURCE, | |
365 | STRIPE_EXPAND_READY, | |
366 | STRIPE_IO_STARTED, /* do not count towards 'bypass_count' */ | |
367 | STRIPE_FULL_WRITE, /* all blocks are set to be overwritten */ | |
368 | STRIPE_BIOFILL_RUN, | |
369 | STRIPE_COMPUTE_RUN, | |
8811b596 | 370 | STRIPE_ON_UNPLUG_LIST, |
f8dfcffd | 371 | STRIPE_DISCARD, |
773ca82f | 372 | STRIPE_ON_RELEASE_LIST, |
da41ba65 | 373 | STRIPE_BATCH_READY, |
72ac7330 | 374 | STRIPE_BATCH_ERR, |
d0852df5 N |
375 | STRIPE_BITMAP_PENDING, /* Being added to bitmap, don't add |
376 | * to batch yet. | |
377 | */ | |
2ded3703 SL |
378 | STRIPE_LOG_TRAPPED, /* trapped into log (see raid5-cache.c) |
379 | * this bit is used in two scenarios: | |
380 | * | |
381 | * 1. write-out phase | |
382 | * set in first entry of r5l_write_stripe | |
383 | * clear in second entry of r5l_write_stripe | |
384 | * used to bypass logic in handle_stripe | |
385 | * | |
386 | * 2. caching phase | |
387 | * set in r5c_try_caching_write() | |
388 | * clear when journal write is done | |
389 | * used to initiate r5c_cache_data() | |
390 | * also used to bypass logic in handle_stripe | |
391 | */ | |
392 | STRIPE_R5C_CACHING, /* the stripe is in caching phase | |
393 | * see more detail in the raid5-cache.c | |
394 | */ | |
1e6d690b SL |
395 | STRIPE_R5C_PARTIAL_STRIPE, /* in r5c cache (to-be/being handled or |
396 | * in conf->r5c_partial_stripe_list) | |
397 | */ | |
398 | STRIPE_R5C_FULL_STRIPE, /* in r5c cache (to-be/being handled or | |
399 | * in conf->r5c_full_stripe_list) | |
400 | */ | |
3bddb7f8 | 401 | STRIPE_R5C_PREFLUSH, /* need to flush journal device */ |
83206d66 | 402 | }; |
417b8d4a | 403 | |
1b956f7a | 404 | #define STRIPE_EXPAND_SYNC_FLAGS \ |
dabc4ec6 | 405 | ((1 << STRIPE_EXPAND_SOURCE) |\ |
406 | (1 << STRIPE_EXPAND_READY) |\ | |
407 | (1 << STRIPE_EXPANDING) |\ | |
408 | (1 << STRIPE_SYNC_REQUESTED)) | |
91c00924 | 409 | /* |
ecc65c9b | 410 | * Operation request flags |
91c00924 | 411 | */ |
ede7ee8b N |
412 | enum { |
413 | STRIPE_OP_BIOFILL, | |
414 | STRIPE_OP_COMPUTE_BLK, | |
415 | STRIPE_OP_PREXOR, | |
416 | STRIPE_OP_BIODRAIN, | |
417 | STRIPE_OP_RECONSTRUCT, | |
418 | STRIPE_OP_CHECK, | |
3418d036 | 419 | STRIPE_OP_PARTIAL_PARITY, |
ede7ee8b | 420 | }; |
584acdd4 MS |
421 | |
422 | /* | |
423 | * RAID parity calculation preferences | |
424 | */ | |
425 | enum { | |
426 | PARITY_DISABLE_RMW = 0, | |
427 | PARITY_ENABLE_RMW, | |
d06f191f | 428 | PARITY_PREFER_RMW, |
584acdd4 MS |
429 | }; |
430 | ||
431 | /* | |
432 | * Pages requested from set_syndrome_sources() | |
433 | */ | |
434 | enum { | |
435 | SYNDROME_SRC_ALL, | |
436 | SYNDROME_SRC_WANT_DRAIN, | |
437 | SYNDROME_SRC_WRITTEN, | |
438 | }; | |
1da177e4 LT |
439 | /* |
440 | * Plugging: | |
441 | * | |
442 | * To improve write throughput, we need to delay the handling of some | |
443 | * stripes until there has been a chance that several write requests | |
444 | * for the one stripe have all been collected. | |
445 | * In particular, any write request that would require pre-reading | |
446 | * is put on a "delayed" queue until there are no stripes currently | |
447 | * in a pre-read phase. Further, if the "delayed" queue is empty when | |
448 | * a stripe is put on it then we "plug" the queue and do not process it | |
449 | * until an unplug call is made. (the unplug_io_fn() is called). | |
450 | * | |
451 | * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add | |
452 | * it to the count of prereading stripes. | |
453 | * When write is initiated, or the stripe refcnt == 0 (just in case) we | |
454 | * clear the PREREAD_ACTIVE flag and decrement the count | |
b5c124af N |
455 | * Whenever the 'handle' queue is empty and the device is not plugged, we |
456 | * move any strips from delayed to handle and clear the DELAYED flag and set | |
457 | * PREREAD_ACTIVE. | |
1da177e4 LT |
458 | * In stripe_handle, if we find pre-reading is necessary, we do it if |
459 | * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue. | |
c4c1663b | 460 | * HANDLE gets cleared if stripe_handle leaves nothing locked. |
1da177e4 | 461 | */ |
ef740c37 | 462 | |
f2785b52 N |
463 | /* Note: disk_info.rdev can be set to NULL asynchronously by raid5_remove_disk. |
464 | * There are three safe ways to access disk_info.rdev. | |
465 | * 1/ when holding mddev->reconfig_mutex | |
466 | * 2/ when resync/recovery/reshape is known to be happening - i.e. in code that | |
467 | * is called as part of performing resync/recovery/reshape. | |
468 | * 3/ while holding rcu_read_lock(), use rcu_dereference to get the pointer | |
469 | * and if it is non-NULL, increment rdev->nr_pending before dropping the RCU | |
470 | * lock. | |
471 | * When .rdev is set to NULL, the nr_pending count checked again and if | |
472 | * it has been incremented, the pointer is put back in .rdev. | |
473 | */ | |
474 | ||
1da177e4 | 475 | struct disk_info { |
b0920ede LG |
476 | struct md_rdev __rcu *rdev; |
477 | struct md_rdev __rcu *replacement; | |
d7bd398e | 478 | struct page *extra_page; /* extra page to use in prexor */ |
1da177e4 LT |
479 | }; |
480 | ||
937621c3 SL |
481 | /* |
482 | * Stripe cache | |
483 | */ | |
484 | ||
485 | #define NR_STRIPES 256 | |
e2368582 YY |
486 | |
487 | #if PAGE_SIZE == DEFAULT_STRIPE_SIZE | |
937621c3 SL |
488 | #define STRIPE_SIZE PAGE_SIZE |
489 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
490 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
e2368582 YY |
491 | #endif |
492 | ||
937621c3 SL |
493 | #define IO_THRESHOLD 1 |
494 | #define BYPASS_THRESHOLD 1 | |
495 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) | |
496 | #define HASH_MASK (NR_HASH - 1) | |
497 | #define MAX_STRIPE_BATCH 8 | |
498 | ||
566c09c5 SL |
499 | /* NOTE NR_STRIPE_HASH_LOCKS must remain below 64. |
500 | * This is because we sometimes take all the spinlocks | |
501 | * and creating that much locking depth can cause | |
502 | * problems. | |
503 | */ | |
504 | #define NR_STRIPE_HASH_LOCKS 8 | |
505 | #define STRIPE_HASH_LOCKS_MASK (NR_STRIPE_HASH_LOCKS - 1) | |
506 | ||
851c30c9 SL |
507 | struct r5worker { |
508 | struct work_struct work; | |
509 | struct r5worker_group *group; | |
566c09c5 | 510 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
bfc90cb0 | 511 | bool working; |
851c30c9 SL |
512 | }; |
513 | ||
514 | struct r5worker_group { | |
515 | struct list_head handle_list; | |
535ae4eb | 516 | struct list_head loprio_list; |
851c30c9 SL |
517 | struct r5conf *conf; |
518 | struct r5worker *workers; | |
bfc90cb0 | 519 | int stripes_cnt; |
851c30c9 SL |
520 | }; |
521 | ||
78e470c2 HM |
522 | /* |
523 | * r5c journal modes of the array: write-back or write-through. | |
524 | * write-through mode has identical behavior as existing log only | |
525 | * implementation. | |
526 | */ | |
527 | enum r5c_journal_mode { | |
528 | R5C_JOURNAL_MODE_WRITE_THROUGH = 0, | |
529 | R5C_JOURNAL_MODE_WRITE_BACK = 1, | |
530 | }; | |
531 | ||
a39f7afd SL |
532 | enum r5_cache_state { |
533 | R5_INACTIVE_BLOCKED, /* release of inactive stripes blocked, | |
534 | * waiting for 25% to be free | |
535 | */ | |
536 | R5_ALLOC_MORE, /* It might help to allocate another | |
537 | * stripe. | |
538 | */ | |
539 | R5_DID_ALLOC, /* A stripe was allocated, don't allocate | |
540 | * more until at least one has been | |
541 | * released. This avoids flooding | |
542 | * the cache. | |
543 | */ | |
544 | R5C_LOG_TIGHT, /* log device space tight, need to | |
545 | * prioritize stripes at last_checkpoint | |
546 | */ | |
547 | R5C_LOG_CRITICAL, /* log device is running out of space, | |
548 | * only process stripes that are already | |
549 | * occupying the log | |
550 | */ | |
d7bd398e SL |
551 | R5C_EXTRA_PAGE_IN_USE, /* a stripe is using disk_info.extra_page |
552 | * for prexor | |
553 | */ | |
a39f7afd SL |
554 | }; |
555 | ||
aaf9f12e SL |
556 | #define PENDING_IO_MAX 512 |
557 | #define PENDING_IO_ONE_FLUSH 128 | |
558 | struct r5pending_data { | |
559 | struct list_head sibling; | |
560 | sector_t sector; /* stripe sector */ | |
561 | struct bio_list bios; | |
562 | }; | |
563 | ||
3d9a644c LG |
564 | struct raid5_percpu { |
565 | struct page *spare_page; /* Used when checking P/Q in raid6 */ | |
566 | void *scribble; /* space for constructing buffer | |
567 | * lists and performing address | |
568 | * conversions | |
569 | */ | |
570 | int scribble_obj_size; | |
571 | local_lock_t lock; | |
572 | }; | |
573 | ||
d1688a6d | 574 | struct r5conf { |
fccddba0 | 575 | struct hlist_head *stripe_hashtbl; |
566c09c5 SL |
576 | /* only protect corresponding hash list and inactive_list */ |
577 | spinlock_t hash_locks[NR_STRIPE_HASH_LOCKS]; | |
fd01b88c | 578 | struct mddev *mddev; |
09c9e5fa | 579 | int chunk_sectors; |
584acdd4 | 580 | int level, algorithm, rmw_level; |
16a53ecc | 581 | int max_degraded; |
02c2de8c | 582 | int raid_disks; |
1da177e4 | 583 | int max_nr_stripes; |
edbe83ab | 584 | int min_nr_stripes; |
e2368582 YY |
585 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
586 | unsigned long stripe_size; | |
587 | unsigned int stripe_shift; | |
588 | unsigned long stripe_sectors; | |
589 | #endif | |
1da177e4 | 590 | |
fef9c61f N |
591 | /* reshape_progress is the leading edge of a 'reshape' |
592 | * It has value MaxSector when no reshape is happening | |
593 | * If delta_disks < 0, it is the last sector we started work on, | |
594 | * else is it the next sector to work on. | |
595 | */ | |
596 | sector_t reshape_progress; | |
597 | /* reshape_safe is the trailing edge of a reshape. We know that | |
598 | * before (or after) this address, all reshape has completed. | |
599 | */ | |
600 | sector_t reshape_safe; | |
7ecaa1e6 | 601 | int previous_raid_disks; |
09c9e5fa AN |
602 | int prev_chunk_sectors; |
603 | int prev_algo; | |
86b42c71 | 604 | short generation; /* increments with every reshape */ |
0a87b25f | 605 | seqcount_spinlock_t gen_lock; /* lock against generation changes */ |
c8f517c4 N |
606 | unsigned long reshape_checkpoint; /* Time we last updated |
607 | * metadata */ | |
b5254dd5 N |
608 | long long min_offset_diff; /* minimum difference between |
609 | * data_offset and | |
610 | * new_data_offset across all | |
611 | * devices. May be negative, | |
612 | * but is closest to zero. | |
613 | */ | |
7ecaa1e6 | 614 | |
1da177e4 | 615 | struct list_head handle_list; /* stripes needing handling */ |
535ae4eb | 616 | struct list_head loprio_list; /* low priority stripes */ |
8b3e6cdc | 617 | struct list_head hold_list; /* preread ready stripes */ |
1da177e4 | 618 | struct list_head delayed_list; /* stripes that have plugged requests */ |
72626685 | 619 | struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */ |
46031f9a | 620 | struct bio *retry_read_aligned; /* currently retrying aligned bios */ |
0472a42b | 621 | unsigned int retry_read_offset; /* sector offset into retry_read_aligned */ |
46031f9a | 622 | struct bio *retry_read_aligned_list; /* aligned bios retry list */ |
1da177e4 | 623 | atomic_t preread_active_stripes; /* stripes with scheduled io */ |
46031f9a | 624 | atomic_t active_aligned_reads; |
8b3e6cdc DW |
625 | atomic_t pending_full_writes; /* full write backlog */ |
626 | int bypass_count; /* bypassed prereads */ | |
627 | int bypass_threshold; /* preread nice */ | |
d592a996 | 628 | int skip_copy; /* Don't copy data from bio to stripe cache */ |
8b3e6cdc | 629 | struct list_head *last_hold; /* detect hold_list promotions */ |
1da177e4 | 630 | |
f6705578 | 631 | atomic_t reshape_stripes; /* stripes with pending writes for reshape */ |
ad01c9e3 N |
632 | /* unfortunately we need two cache names as we temporarily have |
633 | * two caches. | |
634 | */ | |
635 | int active_name; | |
f4be6b43 | 636 | char cache_name[2][32]; |
2d5b569b N |
637 | struct kmem_cache *slab_cache; /* for allocating stripes */ |
638 | struct mutex cache_size_mutex; /* Protect changes to cache size */ | |
72626685 N |
639 | |
640 | int seq_flush, seq_write; | |
641 | int quiesce; | |
642 | ||
643 | int fullsync; /* set to 1 if a full sync is needed, | |
644 | * (fresh device added). | |
645 | * Cleared when a sync completes. | |
646 | */ | |
7f0da59b | 647 | int recovery_disabled; |
36d1c647 | 648 | /* per cpu variables */ |
3d9a644c | 649 | struct raid5_percpu __percpu *percpu; |
27a353c0 SL |
650 | int scribble_disks; |
651 | int scribble_sectors; | |
29c6d1bb | 652 | struct hlist_node node; |
ca65b73b | 653 | |
1da177e4 LT |
654 | /* |
655 | * Free stripes pool | |
656 | */ | |
657 | atomic_t active_stripes; | |
566c09c5 | 658 | struct list_head inactive_list[NR_STRIPE_HASH_LOCKS]; |
1e6d690b SL |
659 | |
660 | atomic_t r5c_cached_full_stripes; | |
661 | struct list_head r5c_full_stripe_list; | |
662 | atomic_t r5c_cached_partial_stripes; | |
663 | struct list_head r5c_partial_stripe_list; | |
e33fbb9c SL |
664 | atomic_t r5c_flushing_full_stripes; |
665 | atomic_t r5c_flushing_partial_stripes; | |
1e6d690b | 666 | |
4bda556a | 667 | atomic_t empty_inactive_list_nr; |
773ca82f | 668 | struct llist_head released_stripes; |
b1b46486 | 669 | wait_queue_head_t wait_for_quiescent; |
6ab2a4b8 | 670 | wait_queue_head_t wait_for_stripe; |
1da177e4 | 671 | wait_queue_head_t wait_for_overlap; |
5423399a | 672 | unsigned long cache_state; |
edbe83ab | 673 | struct shrinker shrinker; |
ad01c9e3 | 674 | int pool_size; /* number of disks in stripeheads in pool */ |
1da177e4 | 675 | spinlock_t device_lock; |
b55e6bfc | 676 | struct disk_info *disks; |
afeee514 | 677 | struct bio_set bio_split; |
91adb564 N |
678 | |
679 | /* When taking over an array from a different personality, we store | |
680 | * the new thread here until we fully activate the array. | |
681 | */ | |
2b8bf345 | 682 | struct md_thread *thread; |
566c09c5 | 683 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
851c30c9 SL |
684 | struct r5worker_group *worker_groups; |
685 | int group_cnt; | |
686 | int worker_cnt_per_group; | |
f6bed0ef | 687 | struct r5l_log *log; |
3418d036 | 688 | void *log_private; |
765d704d | 689 | |
765d704d SL |
690 | spinlock_t pending_bios_lock; |
691 | bool batch_bio_dispatch; | |
aaf9f12e SL |
692 | struct r5pending_data *pending_data; |
693 | struct list_head free_list; | |
694 | struct list_head pending_list; | |
695 | int pending_data_cnt; | |
696 | struct r5pending_data *next_pending_data; | |
1da177e4 LT |
697 | }; |
698 | ||
e2368582 YY |
699 | #if PAGE_SIZE == DEFAULT_STRIPE_SIZE |
700 | #define RAID5_STRIPE_SIZE(conf) STRIPE_SIZE | |
701 | #define RAID5_STRIPE_SHIFT(conf) STRIPE_SHIFT | |
702 | #define RAID5_STRIPE_SECTORS(conf) STRIPE_SECTORS | |
703 | #else | |
704 | #define RAID5_STRIPE_SIZE(conf) ((conf)->stripe_size) | |
705 | #define RAID5_STRIPE_SHIFT(conf) ((conf)->stripe_shift) | |
706 | #define RAID5_STRIPE_SECTORS(conf) ((conf)->stripe_sectors) | |
707 | #endif | |
c911c46c YY |
708 | |
709 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
710 | * order without overlap. There may be several bio's per stripe+device, and | |
711 | * a bio could span several devices. | |
712 | * When walking this list for a particular stripe+device, we must never proceed | |
713 | * beyond a bio that extends past this device, as the next bio might no longer | |
714 | * be valid. | |
715 | * This function is used to determine the 'next' bio in the list, given the | |
716 | * sector of the current stripe+device | |
717 | */ | |
718 | static inline struct bio *r5_next_bio(struct r5conf *conf, struct bio *bio, sector_t sector) | |
719 | { | |
720 | if (bio_end_sector(bio) < sector + RAID5_STRIPE_SECTORS(conf)) | |
721 | return bio->bi_next; | |
722 | else | |
723 | return NULL; | |
724 | } | |
5423399a | 725 | |
1da177e4 LT |
726 | /* |
727 | * Our supported algorithms | |
728 | */ | |
99c0fb5f N |
729 | #define ALGORITHM_LEFT_ASYMMETRIC 0 /* Rotating Parity N with Data Restart */ |
730 | #define ALGORITHM_RIGHT_ASYMMETRIC 1 /* Rotating Parity 0 with Data Restart */ | |
731 | #define ALGORITHM_LEFT_SYMMETRIC 2 /* Rotating Parity N with Data Continuation */ | |
732 | #define ALGORITHM_RIGHT_SYMMETRIC 3 /* Rotating Parity 0 with Data Continuation */ | |
1da177e4 | 733 | |
99c0fb5f N |
734 | /* Define non-rotating (raid4) algorithms. These allow |
735 | * conversion of raid4 to raid5. | |
736 | */ | |
737 | #define ALGORITHM_PARITY_0 4 /* P or P,Q are initial devices */ | |
738 | #define ALGORITHM_PARITY_N 5 /* P or P,Q are final devices. */ | |
739 | ||
740 | /* DDF RAID6 layouts differ from md/raid6 layouts in two ways. | |
741 | * Firstly, the exact positioning of the parity block is slightly | |
742 | * different between the 'LEFT_*' modes of md and the "_N_*" modes | |
743 | * of DDF. | |
744 | * Secondly, or order of datablocks over which the Q syndrome is computed | |
745 | * is different. | |
746 | * Consequently we have different layouts for DDF/raid6 than md/raid6. | |
747 | * These layouts are from the DDFv1.2 spec. | |
748 | * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but | |
749 | * leaves RLQ=3 as 'Vendor Specific' | |
750 | */ | |
751 | ||
752 | #define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */ | |
753 | #define ALGORITHM_ROTATING_N_RESTART 9 /* DDF PRL=6 RLQ=2 */ | |
754 | #define ALGORITHM_ROTATING_N_CONTINUE 10 /*DDF PRL=6 RLQ=3 */ | |
755 | ||
99c0fb5f N |
756 | /* For every RAID5 algorithm we define a RAID6 algorithm |
757 | * with exactly the same layout for data and parity, and | |
758 | * with the Q block always on the last device (N-1). | |
759 | * This allows trivial conversion from RAID5 to RAID6 | |
760 | */ | |
761 | #define ALGORITHM_LEFT_ASYMMETRIC_6 16 | |
762 | #define ALGORITHM_RIGHT_ASYMMETRIC_6 17 | |
763 | #define ALGORITHM_LEFT_SYMMETRIC_6 18 | |
764 | #define ALGORITHM_RIGHT_SYMMETRIC_6 19 | |
765 | #define ALGORITHM_PARITY_0_6 20 | |
766 | #define ALGORITHM_PARITY_N_6 ALGORITHM_PARITY_N | |
767 | ||
768 | static inline int algorithm_valid_raid5(int layout) | |
769 | { | |
770 | return (layout >= 0) && | |
771 | (layout <= 5); | |
772 | } | |
773 | static inline int algorithm_valid_raid6(int layout) | |
774 | { | |
775 | return (layout >= 0 && layout <= 5) | |
776 | || | |
e4424fee | 777 | (layout >= 8 && layout <= 10) |
99c0fb5f N |
778 | || |
779 | (layout >= 16 && layout <= 20); | |
780 | } | |
781 | ||
782 | static inline int algorithm_is_DDF(int layout) | |
783 | { | |
784 | return layout >= 8 && layout <= 10; | |
785 | } | |
11d8a6e3 | 786 | |
046169f0 YY |
787 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
788 | /* | |
789 | * Return offset of the corresponding page for r5dev. | |
790 | */ | |
791 | static inline int raid5_get_page_offset(struct stripe_head *sh, int disk_idx) | |
792 | { | |
793 | return (disk_idx % sh->stripes_per_page) * RAID5_STRIPE_SIZE(sh->raid_conf); | |
794 | } | |
795 | ||
796 | /* | |
797 | * Return corresponding page address for r5dev. | |
798 | */ | |
799 | static inline struct page * | |
800 | raid5_get_dev_page(struct stripe_head *sh, int disk_idx) | |
801 | { | |
802 | return sh->pages[disk_idx / sh->stripes_per_page]; | |
803 | } | |
804 | #endif | |
805 | ||
d1688a6d | 806 | extern void md_raid5_kick_device(struct r5conf *conf); |
fd01b88c | 807 | extern int raid5_set_cache_size(struct mddev *mddev, int size); |
6d036f7d SL |
808 | extern sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous); |
809 | extern void raid5_release_stripe(struct stripe_head *sh); | |
810 | extern sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, | |
811 | int previous, int *dd_idx, | |
812 | struct stripe_head *sh); | |
813 | extern struct stripe_head * | |
814 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
815 | int previous, int noblock, int noquiesce); | |
2e38a37f | 816 | extern int raid5_calc_degraded(struct r5conf *conf); |
78e470c2 | 817 | extern int r5c_journal_mode_set(struct mddev *mddev, int journal_mode); |
1da177e4 | 818 | #endif |