raid5.h 24 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638
  1. #ifndef _RAID5_H
  2. #define _RAID5_H
  3. #include <linux/raid/xor.h>
  4. #include <linux/dmaengine.h>
  5. /*
  6. *
  7. * Each stripe contains one buffer per device. Each buffer can be in
  8. * one of a number of states stored in "flags". Changes between
  9. * these states happen *almost* exclusively under the protection of the
  10. * STRIPE_ACTIVE flag. Some very specific changes can happen in bi_end_io, and
  11. * these are not protected by STRIPE_ACTIVE.
  12. *
  13. * The flag bits that are used to represent these states are:
  14. * R5_UPTODATE and R5_LOCKED
  15. *
  16. * State Empty == !UPTODATE, !LOCK
  17. * We have no data, and there is no active request
  18. * State Want == !UPTODATE, LOCK
  19. * A read request is being submitted for this block
  20. * State Dirty == UPTODATE, LOCK
  21. * Some new data is in this buffer, and it is being written out
  22. * State Clean == UPTODATE, !LOCK
  23. * We have valid data which is the same as on disc
  24. *
  25. * The possible state transitions are:
  26. *
  27. * Empty -> Want - on read or write to get old data for parity calc
  28. * Empty -> Dirty - on compute_parity to satisfy write/sync request.
  29. * Empty -> Clean - on compute_block when computing a block for failed drive
  30. * Want -> Empty - on failed read
  31. * Want -> Clean - on successful completion of read request
  32. * Dirty -> Clean - on successful completion of write request
  33. * Dirty -> Clean - on failed write
  34. * Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
  35. *
  36. * The Want->Empty, Want->Clean, Dirty->Clean, transitions
  37. * all happen in b_end_io at interrupt time.
  38. * Each sets the Uptodate bit before releasing the Lock bit.
  39. * This leaves one multi-stage transition:
  40. * Want->Dirty->Clean
  41. * This is safe because thinking that a Clean buffer is actually dirty
  42. * will at worst delay some action, and the stripe will be scheduled
  43. * for attention after the transition is complete.
  44. *
  45. * There is one possibility that is not covered by these states. That
  46. * is if one drive has failed and there is a spare being rebuilt. We
  47. * can't distinguish between a clean block that has been generated
  48. * from parity calculations, and a clean block that has been
  49. * successfully written to the spare ( or to parity when resyncing).
  50. * To distinguish these states we have a stripe bit STRIPE_INSYNC that
  51. * is set whenever a write is scheduled to the spare, or to the parity
  52. * disc if there is no spare. A sync request clears this bit, and
  53. * when we find it set with no buffers locked, we know the sync is
  54. * complete.
  55. *
  56. * Buffers for the md device that arrive via make_request are attached
  57. * to the appropriate stripe in one of two lists linked on b_reqnext.
  58. * One list (bh_read) for read requests, one (bh_write) for write.
  59. * There should never be more than one buffer on the two lists
  60. * together, but we are not guaranteed of that so we allow for more.
  61. *
  62. * If a buffer is on the read list when the associated cache buffer is
  63. * Uptodate, the data is copied into the read buffer and it's b_end_io
  64. * routine is called. This may happen in the end_request routine only
  65. * if the buffer has just successfully been read. end_request should
  66. * remove the buffers from the list and then set the Uptodate bit on
  67. * the buffer. Other threads may do this only if they first check
  68. * that the Uptodate bit is set. Once they have checked that they may
  69. * take buffers off the read queue.
  70. *
  71. * When a buffer on the write list is committed for write it is copied
  72. * into the cache buffer, which is then marked dirty, and moved onto a
  73. * third list, the written list (bh_written). Once both the parity
  74. * block and the cached buffer are successfully written, any buffer on
  75. * a written list can be returned with b_end_io.
  76. *
  77. * The write list and read list both act as fifos. The read list,
  78. * write list and written list are protected by the device_lock.
  79. * The device_lock is only for list manipulations and will only be
  80. * held for a very short time. It can be claimed from interrupts.
  81. *
  82. *
  83. * Stripes in the stripe cache can be on one of two lists (or on
  84. * neither). The "inactive_list" contains stripes which are not
  85. * currently being used for any request. They can freely be reused
  86. * for another stripe. The "handle_list" contains stripes that need
  87. * to be handled in some way. Both of these are fifo queues. Each
  88. * stripe is also (potentially) linked to a hash bucket in the hash
  89. * table so that it can be found by sector number. Stripes that are
  90. * not hashed must be on the inactive_list, and will normally be at
  91. * the front. All stripes start life this way.
  92. *
  93. * The inactive_list, handle_list and hash bucket lists are all protected by the
  94. * device_lock.
  95. * - stripes have a reference counter. If count==0, they are on a list.
  96. * - If a stripe might need handling, STRIPE_HANDLE is set.
  97. * - When refcount reaches zero, then if STRIPE_HANDLE it is put on
  98. * handle_list else inactive_list
  99. *
  100. * This, combined with the fact that STRIPE_HANDLE is only ever
  101. * cleared while a stripe has a non-zero count means that if the
  102. * refcount is 0 and STRIPE_HANDLE is set, then it is on the
  103. * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
  104. * the stripe is on inactive_list.
  105. *
  106. * The possible transitions are:
  107. * activate an unhashed/inactive stripe (get_active_stripe())
  108. * lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
  109. * activate a hashed, possibly active stripe (get_active_stripe())
  110. * lockdev check-hash if(!cnt++)unlink-stripe unlockdev
  111. * attach a request to an active stripe (add_stripe_bh())
  112. * lockdev attach-buffer unlockdev
  113. * handle a stripe (handle_stripe())
  114. * setSTRIPE_ACTIVE, clrSTRIPE_HANDLE ...
  115. * (lockdev check-buffers unlockdev) ..
  116. * change-state ..
  117. * record io/ops needed clearSTRIPE_ACTIVE schedule io/ops
  118. * release an active stripe (release_stripe())
  119. * lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
  120. *
  121. * The refcount counts each thread that have activated the stripe,
  122. * plus raid5d if it is handling it, plus one for each active request
  123. * on a cached buffer, and plus one if the stripe is undergoing stripe
  124. * operations.
  125. *
  126. * The stripe operations are:
  127. * -copying data between the stripe cache and user application buffers
  128. * -computing blocks to save a disk access, or to recover a missing block
  129. * -updating the parity on a write operation (reconstruct write and
  130. * read-modify-write)
  131. * -checking parity correctness
  132. * -running i/o to disk
  133. * These operations are carried out by raid5_run_ops which uses the async_tx
  134. * api to (optionally) offload operations to dedicated hardware engines.
  135. * When requesting an operation handle_stripe sets the pending bit for the
  136. * operation and increments the count. raid5_run_ops is then run whenever
  137. * the count is non-zero.
  138. * There are some critical dependencies between the operations that prevent some
  139. * from being requested while another is in flight.
  140. * 1/ Parity check operations destroy the in cache version of the parity block,
  141. * so we prevent parity dependent operations like writes and compute_blocks
  142. * from starting while a check is in progress. Some dma engines can perform
  143. * the check without damaging the parity block, in these cases the parity
  144. * block is re-marked up to date (assuming the check was successful) and is
  145. * not re-read from disk.
  146. * 2/ When a write operation is requested we immediately lock the affected
  147. * blocks, and mark them as not up to date. This causes new read requests
  148. * to be held off, as well as parity checks and compute block operations.
  149. * 3/ Once a compute block operation has been requested handle_stripe treats
  150. * that block as if it is up to date. raid5_run_ops guaruntees that any
  151. * operation that is dependent on the compute block result is initiated after
  152. * the compute block completes.
  153. */
  154. /*
  155. * Operations state - intermediate states that are visible outside of
  156. * STRIPE_ACTIVE.
  157. * In general _idle indicates nothing is running, _run indicates a data
  158. * processing operation is active, and _result means the data processing result
  159. * is stable and can be acted upon. For simple operations like biofill and
  160. * compute that only have an _idle and _run state they are indicated with
  161. * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
  162. */
  163. /**
  164. * enum check_states - handles syncing / repairing a stripe
  165. * @check_state_idle - check operations are quiesced
  166. * @check_state_run - check operation is running
  167. * @check_state_result - set outside lock when check result is valid
  168. * @check_state_compute_run - check failed and we are repairing
  169. * @check_state_compute_result - set outside lock when compute result is valid
  170. */
  171. enum check_states {
  172. check_state_idle = 0,
  173. check_state_run, /* xor parity check */
  174. check_state_run_q, /* q-parity check */
  175. check_state_run_pq, /* pq dual parity check */
  176. check_state_check_result,
  177. check_state_compute_run, /* parity repair */
  178. check_state_compute_result,
  179. };
  180. /**
  181. * enum reconstruct_states - handles writing or expanding a stripe
  182. */
  183. enum reconstruct_states {
  184. reconstruct_state_idle = 0,
  185. reconstruct_state_prexor_drain_run, /* prexor-write */
  186. reconstruct_state_drain_run, /* write */
  187. reconstruct_state_run, /* expand */
  188. reconstruct_state_prexor_drain_result,
  189. reconstruct_state_drain_result,
  190. reconstruct_state_result,
  191. };
  192. struct stripe_head {
  193. struct hlist_node hash;
  194. struct list_head lru; /* inactive_list or handle_list */
  195. struct llist_node release_list;
  196. struct r5conf *raid_conf;
  197. short generation; /* increments with every
  198. * reshape */
  199. sector_t sector; /* sector of this row */
  200. short pd_idx; /* parity disk index */
  201. short qd_idx; /* 'Q' disk index for raid6 */
  202. short ddf_layout;/* use DDF ordering to calculate Q */
  203. short hash_lock_index;
  204. unsigned long state; /* state flags */
  205. atomic_t count; /* nr of active thread/requests */
  206. int bm_seq; /* sequence number for bitmap flushes */
  207. int disks; /* disks in stripe */
  208. int overwrite_disks; /* total overwrite disks in stripe,
  209. * this is only checked when stripe
  210. * has STRIPE_BATCH_READY
  211. */
  212. enum check_states check_state;
  213. enum reconstruct_states reconstruct_state;
  214. spinlock_t stripe_lock;
  215. int cpu;
  216. struct r5worker_group *group;
  217. struct stripe_head *batch_head; /* protected by stripe lock */
  218. spinlock_t batch_lock; /* only header's lock is useful */
  219. struct list_head batch_list; /* protected by head's batch lock*/
  220. struct r5l_io_unit *log_io;
  221. struct list_head log_list;
  222. /**
  223. * struct stripe_operations
  224. * @target - STRIPE_OP_COMPUTE_BLK target
  225. * @target2 - 2nd compute target in the raid6 case
  226. * @zero_sum_result - P and Q verification flags
  227. * @request - async service request flags for raid_run_ops
  228. */
  229. struct stripe_operations {
  230. int target, target2;
  231. enum sum_check_flags zero_sum_result;
  232. } ops;
  233. struct r5dev {
  234. /* rreq and rvec are used for the replacement device when
  235. * writing data to both devices.
  236. */
  237. struct bio req, rreq;
  238. struct bio_vec vec, rvec;
  239. struct page *page, *orig_page;
  240. struct bio *toread, *read, *towrite, *written;
  241. sector_t sector; /* sector of this page */
  242. unsigned long flags;
  243. u32 log_checksum;
  244. } dev[1]; /* allocated with extra space depending of RAID geometry */
  245. };
  246. /* stripe_head_state - collects and tracks the dynamic state of a stripe_head
  247. * for handle_stripe.
  248. */
  249. struct stripe_head_state {
  250. /* 'syncing' means that we need to read all devices, either
  251. * to check/correct parity, or to reconstruct a missing device.
  252. * 'replacing' means we are replacing one or more drives and
  253. * the source is valid at this point so we don't need to
  254. * read all devices, just the replacement targets.
  255. */
  256. int syncing, expanding, expanded, replacing;
  257. int locked, uptodate, to_read, to_write, failed, written;
  258. int to_fill, compute, req_compute, non_overwrite;
  259. int failed_num[2];
  260. int p_failed, q_failed;
  261. int dec_preread_active;
  262. unsigned long ops_request;
  263. struct bio_list return_bi;
  264. struct md_rdev *blocked_rdev;
  265. int handle_bad_blocks;
  266. int log_failed;
  267. };
  268. /* Flags for struct r5dev.flags */
  269. enum r5dev_flags {
  270. R5_UPTODATE, /* page contains current data */
  271. R5_LOCKED, /* IO has been submitted on "req" */
  272. R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */
  273. R5_OVERWRITE, /* towrite covers whole page */
  274. /* and some that are internal to handle_stripe */
  275. R5_Insync, /* rdev && rdev->in_sync at start */
  276. R5_Wantread, /* want to schedule a read */
  277. R5_Wantwrite,
  278. R5_Overlap, /* There is a pending overlapping request
  279. * on this block */
  280. R5_ReadNoMerge, /* prevent bio from merging in block-layer */
  281. R5_ReadError, /* seen a read error here recently */
  282. R5_ReWrite, /* have tried to over-write the readerror */
  283. R5_Expanded, /* This block now has post-expand data */
  284. R5_Wantcompute, /* compute_block in progress treat as
  285. * uptodate
  286. */
  287. R5_Wantfill, /* dev->toread contains a bio that needs
  288. * filling
  289. */
  290. R5_Wantdrain, /* dev->towrite needs to be drained */
  291. R5_WantFUA, /* Write should be FUA */
  292. R5_SyncIO, /* The IO is sync */
  293. R5_WriteError, /* got a write error - need to record it */
  294. R5_MadeGood, /* A bad block has been fixed by writing to it */
  295. R5_ReadRepl, /* Will/did read from replacement rather than orig */
  296. R5_MadeGoodRepl,/* A bad block on the replacement device has been
  297. * fixed by writing to it */
  298. R5_NeedReplace, /* This device has a replacement which is not
  299. * up-to-date at this stripe. */
  300. R5_WantReplace, /* We need to update the replacement, we have read
  301. * data in, and now is a good time to write it out.
  302. */
  303. R5_Discard, /* Discard the stripe */
  304. R5_SkipCopy, /* Don't copy data from bio to stripe cache */
  305. };
  306. /*
  307. * Stripe state
  308. */
  309. enum {
  310. STRIPE_ACTIVE,
  311. STRIPE_HANDLE,
  312. STRIPE_SYNC_REQUESTED,
  313. STRIPE_SYNCING,
  314. STRIPE_INSYNC,
  315. STRIPE_REPLACED,
  316. STRIPE_PREREAD_ACTIVE,
  317. STRIPE_DELAYED,
  318. STRIPE_DEGRADED,
  319. STRIPE_BIT_DELAY,
  320. STRIPE_EXPANDING,
  321. STRIPE_EXPAND_SOURCE,
  322. STRIPE_EXPAND_READY,
  323. STRIPE_IO_STARTED, /* do not count towards 'bypass_count' */
  324. STRIPE_FULL_WRITE, /* all blocks are set to be overwritten */
  325. STRIPE_BIOFILL_RUN,
  326. STRIPE_COMPUTE_RUN,
  327. STRIPE_OPS_REQ_PENDING,
  328. STRIPE_ON_UNPLUG_LIST,
  329. STRIPE_DISCARD,
  330. STRIPE_ON_RELEASE_LIST,
  331. STRIPE_BATCH_READY,
  332. STRIPE_BATCH_ERR,
  333. STRIPE_BITMAP_PENDING, /* Being added to bitmap, don't add
  334. * to batch yet.
  335. */
  336. STRIPE_LOG_TRAPPED, /* trapped into log */
  337. };
  338. #define STRIPE_EXPAND_SYNC_FLAGS \
  339. ((1 << STRIPE_EXPAND_SOURCE) |\
  340. (1 << STRIPE_EXPAND_READY) |\
  341. (1 << STRIPE_EXPANDING) |\
  342. (1 << STRIPE_SYNC_REQUESTED))
  343. /*
  344. * Operation request flags
  345. */
  346. enum {
  347. STRIPE_OP_BIOFILL,
  348. STRIPE_OP_COMPUTE_BLK,
  349. STRIPE_OP_PREXOR,
  350. STRIPE_OP_BIODRAIN,
  351. STRIPE_OP_RECONSTRUCT,
  352. STRIPE_OP_CHECK,
  353. };
  354. /*
  355. * RAID parity calculation preferences
  356. */
  357. enum {
  358. PARITY_DISABLE_RMW = 0,
  359. PARITY_ENABLE_RMW,
  360. PARITY_PREFER_RMW,
  361. };
  362. /*
  363. * Pages requested from set_syndrome_sources()
  364. */
  365. enum {
  366. SYNDROME_SRC_ALL,
  367. SYNDROME_SRC_WANT_DRAIN,
  368. SYNDROME_SRC_WRITTEN,
  369. };
  370. /*
  371. * Plugging:
  372. *
  373. * To improve write throughput, we need to delay the handling of some
  374. * stripes until there has been a chance that several write requests
  375. * for the one stripe have all been collected.
  376. * In particular, any write request that would require pre-reading
  377. * is put on a "delayed" queue until there are no stripes currently
  378. * in a pre-read phase. Further, if the "delayed" queue is empty when
  379. * a stripe is put on it then we "plug" the queue and do not process it
  380. * until an unplug call is made. (the unplug_io_fn() is called).
  381. *
  382. * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
  383. * it to the count of prereading stripes.
  384. * When write is initiated, or the stripe refcnt == 0 (just in case) we
  385. * clear the PREREAD_ACTIVE flag and decrement the count
  386. * Whenever the 'handle' queue is empty and the device is not plugged, we
  387. * move any strips from delayed to handle and clear the DELAYED flag and set
  388. * PREREAD_ACTIVE.
  389. * In stripe_handle, if we find pre-reading is necessary, we do it if
  390. * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
  391. * HANDLE gets cleared if stripe_handle leaves nothing locked.
  392. */
  393. struct disk_info {
  394. struct md_rdev *rdev, *replacement;
  395. };
  396. /* NOTE NR_STRIPE_HASH_LOCKS must remain below 64.
  397. * This is because we sometimes take all the spinlocks
  398. * and creating that much locking depth can cause
  399. * problems.
  400. */
  401. #define NR_STRIPE_HASH_LOCKS 8
  402. #define STRIPE_HASH_LOCKS_MASK (NR_STRIPE_HASH_LOCKS - 1)
  403. struct r5worker {
  404. struct work_struct work;
  405. struct r5worker_group *group;
  406. struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
  407. bool working;
  408. };
  409. struct r5worker_group {
  410. struct list_head handle_list;
  411. struct r5conf *conf;
  412. struct r5worker *workers;
  413. int stripes_cnt;
  414. };
  415. struct r5conf {
  416. struct hlist_head *stripe_hashtbl;
  417. /* only protect corresponding hash list and inactive_list */
  418. spinlock_t hash_locks[NR_STRIPE_HASH_LOCKS];
  419. struct mddev *mddev;
  420. int chunk_sectors;
  421. int level, algorithm, rmw_level;
  422. int max_degraded;
  423. int raid_disks;
  424. int max_nr_stripes;
  425. int min_nr_stripes;
  426. /* reshape_progress is the leading edge of a 'reshape'
  427. * It has value MaxSector when no reshape is happening
  428. * If delta_disks < 0, it is the last sector we started work on,
  429. * else is it the next sector to work on.
  430. */
  431. sector_t reshape_progress;
  432. /* reshape_safe is the trailing edge of a reshape. We know that
  433. * before (or after) this address, all reshape has completed.
  434. */
  435. sector_t reshape_safe;
  436. int previous_raid_disks;
  437. int prev_chunk_sectors;
  438. int prev_algo;
  439. short generation; /* increments with every reshape */
  440. seqcount_t gen_lock; /* lock against generation changes */
  441. unsigned long reshape_checkpoint; /* Time we last updated
  442. * metadata */
  443. long long min_offset_diff; /* minimum difference between
  444. * data_offset and
  445. * new_data_offset across all
  446. * devices. May be negative,
  447. * but is closest to zero.
  448. */
  449. struct list_head handle_list; /* stripes needing handling */
  450. struct list_head hold_list; /* preread ready stripes */
  451. struct list_head delayed_list; /* stripes that have plugged requests */
  452. struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
  453. struct bio *retry_read_aligned; /* currently retrying aligned bios */
  454. struct bio *retry_read_aligned_list; /* aligned bios retry list */
  455. atomic_t preread_active_stripes; /* stripes with scheduled io */
  456. atomic_t active_aligned_reads;
  457. atomic_t pending_full_writes; /* full write backlog */
  458. int bypass_count; /* bypassed prereads */
  459. int bypass_threshold; /* preread nice */
  460. int skip_copy; /* Don't copy data from bio to stripe cache */
  461. struct list_head *last_hold; /* detect hold_list promotions */
  462. /* bios to have bi_end_io called after metadata is synced */
  463. struct bio_list return_bi;
  464. atomic_t reshape_stripes; /* stripes with pending writes for reshape */
  465. /* unfortunately we need two cache names as we temporarily have
  466. * two caches.
  467. */
  468. int active_name;
  469. char cache_name[2][32];
  470. struct kmem_cache *slab_cache; /* for allocating stripes */
  471. struct mutex cache_size_mutex; /* Protect changes to cache size */
  472. int seq_flush, seq_write;
  473. int quiesce;
  474. int fullsync; /* set to 1 if a full sync is needed,
  475. * (fresh device added).
  476. * Cleared when a sync completes.
  477. */
  478. int recovery_disabled;
  479. /* per cpu variables */
  480. struct raid5_percpu {
  481. struct page *spare_page; /* Used when checking P/Q in raid6 */
  482. struct flex_array *scribble; /* space for constructing buffer
  483. * lists and performing address
  484. * conversions
  485. */
  486. } __percpu *percpu;
  487. int scribble_disks;
  488. int scribble_sectors;
  489. #ifdef CONFIG_HOTPLUG_CPU
  490. struct notifier_block cpu_notify;
  491. #endif
  492. /*
  493. * Free stripes pool
  494. */
  495. atomic_t active_stripes;
  496. struct list_head inactive_list[NR_STRIPE_HASH_LOCKS];
  497. atomic_t empty_inactive_list_nr;
  498. struct llist_head released_stripes;
  499. wait_queue_head_t wait_for_quiescent;
  500. wait_queue_head_t wait_for_stripe;
  501. wait_queue_head_t wait_for_overlap;
  502. unsigned long cache_state;
  503. #define R5_INACTIVE_BLOCKED 1 /* release of inactive stripes blocked,
  504. * waiting for 25% to be free
  505. */
  506. #define R5_ALLOC_MORE 2 /* It might help to allocate another
  507. * stripe.
  508. */
  509. #define R5_DID_ALLOC 4 /* A stripe was allocated, don't allocate
  510. * more until at least one has been
  511. * released. This avoids flooding
  512. * the cache.
  513. */
  514. struct shrinker shrinker;
  515. int pool_size; /* number of disks in stripeheads in pool */
  516. spinlock_t device_lock;
  517. struct disk_info *disks;
  518. /* When taking over an array from a different personality, we store
  519. * the new thread here until we fully activate the array.
  520. */
  521. struct md_thread *thread;
  522. struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
  523. struct r5worker_group *worker_groups;
  524. int group_cnt;
  525. int worker_cnt_per_group;
  526. struct r5l_log *log;
  527. };
  528. /*
  529. * Our supported algorithms
  530. */
  531. #define ALGORITHM_LEFT_ASYMMETRIC 0 /* Rotating Parity N with Data Restart */
  532. #define ALGORITHM_RIGHT_ASYMMETRIC 1 /* Rotating Parity 0 with Data Restart */
  533. #define ALGORITHM_LEFT_SYMMETRIC 2 /* Rotating Parity N with Data Continuation */
  534. #define ALGORITHM_RIGHT_SYMMETRIC 3 /* Rotating Parity 0 with Data Continuation */
  535. /* Define non-rotating (raid4) algorithms. These allow
  536. * conversion of raid4 to raid5.
  537. */
  538. #define ALGORITHM_PARITY_0 4 /* P or P,Q are initial devices */
  539. #define ALGORITHM_PARITY_N 5 /* P or P,Q are final devices. */
  540. /* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
  541. * Firstly, the exact positioning of the parity block is slightly
  542. * different between the 'LEFT_*' modes of md and the "_N_*" modes
  543. * of DDF.
  544. * Secondly, or order of datablocks over which the Q syndrome is computed
  545. * is different.
  546. * Consequently we have different layouts for DDF/raid6 than md/raid6.
  547. * These layouts are from the DDFv1.2 spec.
  548. * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
  549. * leaves RLQ=3 as 'Vendor Specific'
  550. */
  551. #define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
  552. #define ALGORITHM_ROTATING_N_RESTART 9 /* DDF PRL=6 RLQ=2 */
  553. #define ALGORITHM_ROTATING_N_CONTINUE 10 /*DDF PRL=6 RLQ=3 */
  554. /* For every RAID5 algorithm we define a RAID6 algorithm
  555. * with exactly the same layout for data and parity, and
  556. * with the Q block always on the last device (N-1).
  557. * This allows trivial conversion from RAID5 to RAID6
  558. */
  559. #define ALGORITHM_LEFT_ASYMMETRIC_6 16
  560. #define ALGORITHM_RIGHT_ASYMMETRIC_6 17
  561. #define ALGORITHM_LEFT_SYMMETRIC_6 18
  562. #define ALGORITHM_RIGHT_SYMMETRIC_6 19
  563. #define ALGORITHM_PARITY_0_6 20
  564. #define ALGORITHM_PARITY_N_6 ALGORITHM_PARITY_N
  565. static inline int algorithm_valid_raid5(int layout)
  566. {
  567. return (layout >= 0) &&
  568. (layout <= 5);
  569. }
  570. static inline int algorithm_valid_raid6(int layout)
  571. {
  572. return (layout >= 0 && layout <= 5)
  573. ||
  574. (layout >= 8 && layout <= 10)
  575. ||
  576. (layout >= 16 && layout <= 20);
  577. }
  578. static inline int algorithm_is_DDF(int layout)
  579. {
  580. return layout >= 8 && layout <= 10;
  581. }
  582. extern void md_raid5_kick_device(struct r5conf *conf);
  583. extern int raid5_set_cache_size(struct mddev *mddev, int size);
  584. extern sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous);
  585. extern void raid5_release_stripe(struct stripe_head *sh);
  586. extern sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
  587. int previous, int *dd_idx,
  588. struct stripe_head *sh);
  589. extern struct stripe_head *
  590. raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
  591. int previous, int noblock, int noquiesce);
  592. extern int r5l_init_log(struct r5conf *conf, struct md_rdev *rdev);
  593. extern void r5l_exit_log(struct r5l_log *log);
  594. extern int r5l_write_stripe(struct r5l_log *log, struct stripe_head *head_sh);
  595. extern void r5l_write_stripe_run(struct r5l_log *log);
  596. extern void r5l_flush_stripe_to_raid(struct r5l_log *log);
  597. extern void r5l_stripe_write_finished(struct stripe_head *sh);
  598. extern int r5l_handle_flush_request(struct r5l_log *log, struct bio *bio);
  599. extern void r5l_quiesce(struct r5l_log *log, int state);
  600. extern bool r5l_log_disk_error(struct r5conf *conf);
  601. #endif