1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _RAID5_H
3 #define _RAID5_H
4 
5 #include <linux/raid/xor.h>
6 #include <linux/dmaengine.h>
7 
8 /*
9  *
10  * Each stripe contains one buffer per device.  Each buffer can be in
11  * one of a number of states stored in "flags".  Changes between
12  * these states happen *almost* exclusively under the protection of the
13  * STRIPE_ACTIVE flag.  Some very specific changes can happen in bi_end_io, and
14  * these are not protected by STRIPE_ACTIVE.
15  *
16  * The flag bits that are used to represent these states are:
17  *   R5_UPTODATE and R5_LOCKED
18  *
19  * State Empty == !UPTODATE, !LOCK
20  *        We have no data, and there is no active request
21  * State Want == !UPTODATE, LOCK
22  *        A read request is being submitted for this block
23  * State Dirty == UPTODATE, LOCK
24  *        Some new data is in this buffer, and it is being written out
25  * State Clean == UPTODATE, !LOCK
26  *        We have valid data which is the same as on disc
27  *
28  * The possible state transitions are:
29  *
30  *  Empty -> Want   - on read or write to get old data for  parity calc
31  *  Empty -> Dirty  - on compute_parity to satisfy write/sync request.
32  *  Empty -> Clean  - on compute_block when computing a block for failed drive
33  *  Want  -> Empty  - on failed read
34  *  Want  -> Clean  - on successful completion of read request
35  *  Dirty -> Clean  - on successful completion of write request
36  *  Dirty -> Clean  - on failed write
37  *  Clean -> Dirty  - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
38  *
39  * The Want->Empty, Want->Clean, Dirty->Clean, transitions
40  * all happen in b_end_io at interrupt time.
41  * Each sets the Uptodate bit before releasing the Lock bit.
42  * This leaves one multi-stage transition:
43  *    Want->Dirty->Clean
44  * This is safe because thinking that a Clean buffer is actually dirty
45  * will at worst delay some action, and the stripe will be scheduled
46  * for attention after the transition is complete.
47  *
48  * There is one possibility that is not covered by these states.  That
49  * is if one drive has failed and there is a spare being rebuilt.  We
50  * can't distinguish between a clean block that has been generated
51  * from parity calculations, and a clean block that has been
52  * successfully written to the spare ( or to parity when resyncing).
53  * To distinguish these states we have a stripe bit STRIPE_INSYNC that
54  * is set whenever a write is scheduled to the spare, or to the parity
55  * disc if there is no spare.  A sync request clears this bit, and
56  * when we find it set with no buffers locked, we know the sync is
57  * complete.
58  *
59  * Buffers for the md device that arrive via make_request are attached
60  * to the appropriate stripe in one of two lists linked on b_reqnext.
61  * One list (bh_read) for read requests, one (bh_write) for write.
62  * There should never be more than one buffer on the two lists
63  * together, but we are not guaranteed of that so we allow for more.
64  *
65  * If a buffer is on the read list when the associated cache buffer is
66  * Uptodate, the data is copied into the read buffer and it's b_end_io
67  * routine is called.  This may happen in the end_request routine only
68  * if the buffer has just successfully been read.  end_request should
69  * remove the buffers from the list and then set the Uptodate bit on
70  * the buffer.  Other threads may do this only if they first check
71  * that the Uptodate bit is set.  Once they have checked that they may
72  * take buffers off the read queue.
73  *
74  * When a buffer on the write list is committed for write it is copied
75  * into the cache buffer, which is then marked dirty, and moved onto a
76  * third list, the written list (bh_written).  Once both the parity
77  * block and the cached buffer are successfully written, any buffer on
78  * a written list can be returned with b_end_io.
79  *
80  * The write list and read list both act as fifos.  The read list,
81  * write list and written list are protected by the device_lock.
82  * The device_lock is only for list manipulations and will only be
83  * held for a very short time.  It can be claimed from interrupts.
84  *
85  *
86  * Stripes in the stripe cache can be on one of two lists (or on
87  * neither).  The "inactive_list" contains stripes which are not
88  * currently being used for any request.  They can freely be reused
89  * for another stripe.  The "handle_list" contains stripes that need
90  * to be handled in some way.  Both of these are fifo queues.  Each
91  * stripe is also (potentially) linked to a hash bucket in the hash
92  * table so that it can be found by sector number.  Stripes that are
93  * not hashed must be on the inactive_list, and will normally be at
94  * the front.  All stripes start life this way.
95  *
96  * The inactive_list, handle_list and hash bucket lists are all protected by the
97  * device_lock.
98  *  - stripes have a reference counter. If count==0, they are on a list.
99  *  - If a stripe might need handling, STRIPE_HANDLE is set.
100  *  - When refcount reaches zero, then if STRIPE_HANDLE it is put on
101  *    handle_list else inactive_list
102  *
103  * This, combined with the fact that STRIPE_HANDLE is only ever
104  * cleared while a stripe has a non-zero count means that if the
105  * refcount is 0 and STRIPE_HANDLE is set, then it is on the
106  * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
107  * the stripe is on inactive_list.
108  *
109  * The possible transitions are:
110  *  activate an unhashed/inactive stripe (get_active_stripe())
111  *     lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
112  *  activate a hashed, possibly active stripe (get_active_stripe())
113  *     lockdev check-hash if(!cnt++)unlink-stripe unlockdev
114  *  attach a request to an active stripe (add_stripe_bh())
115  *     lockdev attach-buffer unlockdev
116  *  handle a stripe (handle_stripe())
117  *     setSTRIPE_ACTIVE,  clrSTRIPE_HANDLE ...
118  *		(lockdev check-buffers unlockdev) ..
119  *		change-state ..
120  *		record io/ops needed clearSTRIPE_ACTIVE schedule io/ops
121  *  release an active stripe (release_stripe())
122  *     lockdev if (!--cnt) { if  STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
123  *
124  * The refcount counts each thread that have activated the stripe,
125  * plus raid5d if it is handling it, plus one for each active request
126  * on a cached buffer, and plus one if the stripe is undergoing stripe
127  * operations.
128  *
129  * The stripe operations are:
130  * -copying data between the stripe cache and user application buffers
131  * -computing blocks to save a disk access, or to recover a missing block
132  * -updating the parity on a write operation (reconstruct write and
133  *  read-modify-write)
134  * -checking parity correctness
135  * -running i/o to disk
136  * These operations are carried out by raid5_run_ops which uses the async_tx
137  * api to (optionally) offload operations to dedicated hardware engines.
138  * When requesting an operation handle_stripe sets the pending bit for the
139  * operation and increments the count.  raid5_run_ops is then run whenever
140  * the count is non-zero.
141  * There are some critical dependencies between the operations that prevent some
142  * from being requested while another is in flight.
143  * 1/ Parity check operations destroy the in cache version of the parity block,
144  *    so we prevent parity dependent operations like writes and compute_blocks
145  *    from starting while a check is in progress.  Some dma engines can perform
146  *    the check without damaging the parity block, in these cases the parity
147  *    block is re-marked up to date (assuming the check was successful) and is
148  *    not re-read from disk.
149  * 2/ When a write operation is requested we immediately lock the affected
150  *    blocks, and mark them as not up to date.  This causes new read requests
151  *    to be held off, as well as parity checks and compute block operations.
152  * 3/ Once a compute block operation has been requested handle_stripe treats
153  *    that block as if it is up to date.  raid5_run_ops guaruntees that any
154  *    operation that is dependent on the compute block result is initiated after
155  *    the compute block completes.
156  */
157 
158 /*
159  * Operations state - intermediate states that are visible outside of
160  *   STRIPE_ACTIVE.
161  * In general _idle indicates nothing is running, _run indicates a data
162  * processing operation is active, and _result means the data processing result
163  * is stable and can be acted upon.  For simple operations like biofill and
164  * compute that only have an _idle and _run state they are indicated with
165  * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
166  */
167 /**
168  * enum check_states - handles syncing / repairing a stripe
169  * @check_state_idle - check operations are quiesced
170  * @check_state_run - check operation is running
171  * @check_state_result - set outside lock when check result is valid
172  * @check_state_compute_run - check failed and we are repairing
173  * @check_state_compute_result - set outside lock when compute result is valid
174  */
175 enum check_states {
176 	check_state_idle = 0,
177 	check_state_run, /* xor parity check */
178 	check_state_run_q, /* q-parity check */
179 	check_state_run_pq, /* pq dual parity check */
180 	check_state_check_result,
181 	check_state_compute_run, /* parity repair */
182 	check_state_compute_result,
183 };
184 
185 /**
186  * enum reconstruct_states - handles writing or expanding a stripe
187  */
188 enum reconstruct_states {
189 	reconstruct_state_idle = 0,
190 	reconstruct_state_prexor_drain_run,	/* prexor-write */
191 	reconstruct_state_drain_run,		/* write */
192 	reconstruct_state_run,			/* expand */
193 	reconstruct_state_prexor_drain_result,
194 	reconstruct_state_drain_result,
195 	reconstruct_state_result,
196 };
197 
198 struct stripe_head {
199 	struct hlist_node	hash;
200 	struct list_head	lru;	      /* inactive_list or handle_list */
201 	struct llist_node	release_list;
202 	struct r5conf		*raid_conf;
203 	short			generation;	/* increments with every
204 						 * reshape */
205 	sector_t		sector;		/* sector of this row */
206 	short			pd_idx;		/* parity disk index */
207 	short			qd_idx;		/* 'Q' disk index for raid6 */
208 	short			ddf_layout;/* use DDF ordering to calculate Q */
209 	short			hash_lock_index;
210 	unsigned long		state;		/* state flags */
211 	atomic_t		count;	      /* nr of active thread/requests */
212 	int			bm_seq;	/* sequence number for bitmap flushes */
213 	int			disks;		/* disks in stripe */
214 	int			overwrite_disks; /* total overwrite disks in stripe,
215 						  * this is only checked when stripe
216 						  * has STRIPE_BATCH_READY
217 						  */
218 	enum check_states	check_state;
219 	enum reconstruct_states reconstruct_state;
220 	spinlock_t		stripe_lock;
221 	int			cpu;
222 	struct r5worker_group	*group;
223 
224 	struct stripe_head	*batch_head; /* protected by stripe lock */
225 	spinlock_t		batch_lock; /* only header's lock is useful */
226 	struct list_head	batch_list; /* protected by head's batch lock*/
227 
228 	union {
229 		struct r5l_io_unit	*log_io;
230 		struct ppl_io_unit	*ppl_io;
231 	};
232 
233 	struct list_head	log_list;
234 	sector_t		log_start; /* first meta block on the journal */
235 	struct list_head	r5c; /* for r5c_cache->stripe_in_journal */
236 
237 	struct page		*ppl_page; /* partial parity of this stripe */
238 	/**
239 	 * struct stripe_operations
240 	 * @target - STRIPE_OP_COMPUTE_BLK target
241 	 * @target2 - 2nd compute target in the raid6 case
242 	 * @zero_sum_result - P and Q verification flags
243 	 * @request - async service request flags for raid_run_ops
244 	 */
245 	struct stripe_operations {
246 		int 		     target, target2;
247 		enum sum_check_flags zero_sum_result;
248 	} ops;
249 	struct r5dev {
250 		/* rreq and rvec are used for the replacement device when
251 		 * writing data to both devices.
252 		 */
253 		struct bio	req, rreq;
254 		struct bio_vec	vec, rvec;
255 		struct page	*page, *orig_page;
256 		struct bio	*toread, *read, *towrite, *written;
257 		sector_t	sector;			/* sector of this page */
258 		unsigned long	flags;
259 		u32		log_checksum;
260 		unsigned short	write_hint;
261 	} dev[1]; /* allocated with extra space depending of RAID geometry */
262 };
263 
264 /* stripe_head_state - collects and tracks the dynamic state of a stripe_head
265  *     for handle_stripe.
266  */
267 struct stripe_head_state {
268 	/* 'syncing' means that we need to read all devices, either
269 	 * to check/correct parity, or to reconstruct a missing device.
270 	 * 'replacing' means we are replacing one or more drives and
271 	 * the source is valid at this point so we don't need to
272 	 * read all devices, just the replacement targets.
273 	 */
274 	int syncing, expanding, expanded, replacing;
275 	int locked, uptodate, to_read, to_write, failed, written;
276 	int to_fill, compute, req_compute, non_overwrite;
277 	int injournal, just_cached;
278 	int failed_num[2];
279 	int p_failed, q_failed;
280 	int dec_preread_active;
281 	unsigned long ops_request;
282 
283 	struct md_rdev *blocked_rdev;
284 	int handle_bad_blocks;
285 	int log_failed;
286 	int waiting_extra_page;
287 };
288 
289 /* Flags for struct r5dev.flags */
290 enum r5dev_flags {
291 	R5_UPTODATE,	/* page contains current data */
292 	R5_LOCKED,	/* IO has been submitted on "req" */
293 	R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */
294 	R5_OVERWRITE,	/* towrite covers whole page */
295 /* and some that are internal to handle_stripe */
296 	R5_Insync,	/* rdev && rdev->in_sync at start */
297 	R5_Wantread,	/* want to schedule a read */
298 	R5_Wantwrite,
299 	R5_Overlap,	/* There is a pending overlapping request
300 			 * on this block */
301 	R5_ReadNoMerge, /* prevent bio from merging in block-layer */
302 	R5_ReadError,	/* seen a read error here recently */
303 	R5_ReWrite,	/* have tried to over-write the readerror */
304 
305 	R5_Expanded,	/* This block now has post-expand data */
306 	R5_Wantcompute,	/* compute_block in progress treat as
307 			 * uptodate
308 			 */
309 	R5_Wantfill,	/* dev->toread contains a bio that needs
310 			 * filling
311 			 */
312 	R5_Wantdrain,	/* dev->towrite needs to be drained */
313 	R5_WantFUA,	/* Write should be FUA */
314 	R5_SyncIO,	/* The IO is sync */
315 	R5_WriteError,	/* got a write error - need to record it */
316 	R5_MadeGood,	/* A bad block has been fixed by writing to it */
317 	R5_ReadRepl,	/* Will/did read from replacement rather than orig */
318 	R5_MadeGoodRepl,/* A bad block on the replacement device has been
319 			 * fixed by writing to it */
320 	R5_NeedReplace,	/* This device has a replacement which is not
321 			 * up-to-date at this stripe. */
322 	R5_WantReplace, /* We need to update the replacement, we have read
323 			 * data in, and now is a good time to write it out.
324 			 */
325 	R5_Discard,	/* Discard the stripe */
326 	R5_SkipCopy,	/* Don't copy data from bio to stripe cache */
327 	R5_InJournal,	/* data being written is in the journal device.
328 			 * if R5_InJournal is set for parity pd_idx, all the
329 			 * data and parity being written are in the journal
330 			 * device
331 			 */
332 	R5_OrigPageUPTDODATE,	/* with write back cache, we read old data into
333 				 * dev->orig_page for prexor. When this flag is
334 				 * set, orig_page contains latest data in the
335 				 * raid disk.
336 				 */
337 };
338 
339 /*
340  * Stripe state
341  */
342 enum {
343 	STRIPE_ACTIVE,
344 	STRIPE_HANDLE,
345 	STRIPE_SYNC_REQUESTED,
346 	STRIPE_SYNCING,
347 	STRIPE_INSYNC,
348 	STRIPE_REPLACED,
349 	STRIPE_PREREAD_ACTIVE,
350 	STRIPE_DELAYED,
351 	STRIPE_DEGRADED,
352 	STRIPE_BIT_DELAY,
353 	STRIPE_EXPANDING,
354 	STRIPE_EXPAND_SOURCE,
355 	STRIPE_EXPAND_READY,
356 	STRIPE_IO_STARTED,	/* do not count towards 'bypass_count' */
357 	STRIPE_FULL_WRITE,	/* all blocks are set to be overwritten */
358 	STRIPE_BIOFILL_RUN,
359 	STRIPE_COMPUTE_RUN,
360 	STRIPE_OPS_REQ_PENDING,
361 	STRIPE_ON_UNPLUG_LIST,
362 	STRIPE_DISCARD,
363 	STRIPE_ON_RELEASE_LIST,
364 	STRIPE_BATCH_READY,
365 	STRIPE_BATCH_ERR,
366 	STRIPE_BITMAP_PENDING,	/* Being added to bitmap, don't add
367 				 * to batch yet.
368 				 */
369 	STRIPE_LOG_TRAPPED,	/* trapped into log (see raid5-cache.c)
370 				 * this bit is used in two scenarios:
371 				 *
372 				 * 1. write-out phase
373 				 *  set in first entry of r5l_write_stripe
374 				 *  clear in second entry of r5l_write_stripe
375 				 *  used to bypass logic in handle_stripe
376 				 *
377 				 * 2. caching phase
378 				 *  set in r5c_try_caching_write()
379 				 *  clear when journal write is done
380 				 *  used to initiate r5c_cache_data()
381 				 *  also used to bypass logic in handle_stripe
382 				 */
383 	STRIPE_R5C_CACHING,	/* the stripe is in caching phase
384 				 * see more detail in the raid5-cache.c
385 				 */
386 	STRIPE_R5C_PARTIAL_STRIPE,	/* in r5c cache (to-be/being handled or
387 					 * in conf->r5c_partial_stripe_list)
388 					 */
389 	STRIPE_R5C_FULL_STRIPE,	/* in r5c cache (to-be/being handled or
390 				 * in conf->r5c_full_stripe_list)
391 				 */
392 	STRIPE_R5C_PREFLUSH,	/* need to flush journal device */
393 };
394 
395 #define STRIPE_EXPAND_SYNC_FLAGS \
396 	((1 << STRIPE_EXPAND_SOURCE) |\
397 	(1 << STRIPE_EXPAND_READY) |\
398 	(1 << STRIPE_EXPANDING) |\
399 	(1 << STRIPE_SYNC_REQUESTED))
400 /*
401  * Operation request flags
402  */
403 enum {
404 	STRIPE_OP_BIOFILL,
405 	STRIPE_OP_COMPUTE_BLK,
406 	STRIPE_OP_PREXOR,
407 	STRIPE_OP_BIODRAIN,
408 	STRIPE_OP_RECONSTRUCT,
409 	STRIPE_OP_CHECK,
410 	STRIPE_OP_PARTIAL_PARITY,
411 };
412 
413 /*
414  * RAID parity calculation preferences
415  */
416 enum {
417 	PARITY_DISABLE_RMW = 0,
418 	PARITY_ENABLE_RMW,
419 	PARITY_PREFER_RMW,
420 };
421 
422 /*
423  * Pages requested from set_syndrome_sources()
424  */
425 enum {
426 	SYNDROME_SRC_ALL,
427 	SYNDROME_SRC_WANT_DRAIN,
428 	SYNDROME_SRC_WRITTEN,
429 };
430 /*
431  * Plugging:
432  *
433  * To improve write throughput, we need to delay the handling of some
434  * stripes until there has been a chance that several write requests
435  * for the one stripe have all been collected.
436  * In particular, any write request that would require pre-reading
437  * is put on a "delayed" queue until there are no stripes currently
438  * in a pre-read phase.  Further, if the "delayed" queue is empty when
439  * a stripe is put on it then we "plug" the queue and do not process it
440  * until an unplug call is made. (the unplug_io_fn() is called).
441  *
442  * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
443  * it to the count of prereading stripes.
444  * When write is initiated, or the stripe refcnt == 0 (just in case) we
445  * clear the PREREAD_ACTIVE flag and decrement the count
446  * Whenever the 'handle' queue is empty and the device is not plugged, we
447  * move any strips from delayed to handle and clear the DELAYED flag and set
448  * PREREAD_ACTIVE.
449  * In stripe_handle, if we find pre-reading is necessary, we do it if
450  * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
451  * HANDLE gets cleared if stripe_handle leaves nothing locked.
452  */
453 
454 /* Note: disk_info.rdev can be set to NULL asynchronously by raid5_remove_disk.
455  * There are three safe ways to access disk_info.rdev.
456  * 1/ when holding mddev->reconfig_mutex
457  * 2/ when resync/recovery/reshape is known to be happening - i.e. in code that
458  *    is called as part of performing resync/recovery/reshape.
459  * 3/ while holding rcu_read_lock(), use rcu_dereference to get the pointer
460  *    and if it is non-NULL, increment rdev->nr_pending before dropping the RCU
461  *    lock.
462  * When .rdev is set to NULL, the nr_pending count checked again and if
463  * it has been incremented, the pointer is put back in .rdev.
464  */
465 
466 struct disk_info {
467 	struct md_rdev	*rdev, *replacement;
468 	struct page	*extra_page; /* extra page to use in prexor */
469 };
470 
471 /*
472  * Stripe cache
473  */
474 
475 #define NR_STRIPES		256
476 #define STRIPE_SIZE		PAGE_SIZE
477 #define STRIPE_SHIFT		(PAGE_SHIFT - 9)
478 #define STRIPE_SECTORS		(STRIPE_SIZE>>9)
479 #define	IO_THRESHOLD		1
480 #define BYPASS_THRESHOLD	1
481 #define NR_HASH			(PAGE_SIZE / sizeof(struct hlist_head))
482 #define HASH_MASK		(NR_HASH - 1)
483 #define MAX_STRIPE_BATCH	8
484 
485 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
486  * order without overlap.  There may be several bio's per stripe+device, and
487  * a bio could span several devices.
488  * When walking this list for a particular stripe+device, we must never proceed
489  * beyond a bio that extends past this device, as the next bio might no longer
490  * be valid.
491  * This function is used to determine the 'next' bio in the list, given the
492  * sector of the current stripe+device
493  */
r5_next_bio(struct bio * bio,sector_t sector)494 static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
495 {
496 	int sectors = bio_sectors(bio);
497 
498 	if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS)
499 		return bio->bi_next;
500 	else
501 		return NULL;
502 }
503 
504 /* NOTE NR_STRIPE_HASH_LOCKS must remain below 64.
505  * This is because we sometimes take all the spinlocks
506  * and creating that much locking depth can cause
507  * problems.
508  */
509 #define NR_STRIPE_HASH_LOCKS 8
510 #define STRIPE_HASH_LOCKS_MASK (NR_STRIPE_HASH_LOCKS - 1)
511 
512 struct r5worker {
513 	struct work_struct work;
514 	struct r5worker_group *group;
515 	struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
516 	bool working;
517 };
518 
519 struct r5worker_group {
520 	struct list_head handle_list;
521 	struct list_head loprio_list;
522 	struct r5conf *conf;
523 	struct r5worker *workers;
524 	int stripes_cnt;
525 };
526 
527 /*
528  * r5c journal modes of the array: write-back or write-through.
529  * write-through mode has identical behavior as existing log only
530  * implementation.
531  */
532 enum r5c_journal_mode {
533 	R5C_JOURNAL_MODE_WRITE_THROUGH = 0,
534 	R5C_JOURNAL_MODE_WRITE_BACK = 1,
535 };
536 
537 enum r5_cache_state {
538 	R5_INACTIVE_BLOCKED,	/* release of inactive stripes blocked,
539 				 * waiting for 25% to be free
540 				 */
541 	R5_ALLOC_MORE,		/* It might help to allocate another
542 				 * stripe.
543 				 */
544 	R5_DID_ALLOC,		/* A stripe was allocated, don't allocate
545 				 * more until at least one has been
546 				 * released.  This avoids flooding
547 				 * the cache.
548 				 */
549 	R5C_LOG_TIGHT,		/* log device space tight, need to
550 				 * prioritize stripes at last_checkpoint
551 				 */
552 	R5C_LOG_CRITICAL,	/* log device is running out of space,
553 				 * only process stripes that are already
554 				 * occupying the log
555 				 */
556 	R5C_EXTRA_PAGE_IN_USE,	/* a stripe is using disk_info.extra_page
557 				 * for prexor
558 				 */
559 };
560 
561 #define PENDING_IO_MAX 512
562 #define PENDING_IO_ONE_FLUSH 128
563 struct r5pending_data {
564 	struct list_head sibling;
565 	sector_t sector; /* stripe sector */
566 	struct bio_list bios;
567 };
568 
569 struct r5conf {
570 	struct hlist_head	*stripe_hashtbl;
571 	/* only protect corresponding hash list and inactive_list */
572 	spinlock_t		hash_locks[NR_STRIPE_HASH_LOCKS];
573 	struct mddev		*mddev;
574 	int			chunk_sectors;
575 	int			level, algorithm, rmw_level;
576 	int			max_degraded;
577 	int			raid_disks;
578 	int			max_nr_stripes;
579 	int			min_nr_stripes;
580 
581 	/* reshape_progress is the leading edge of a 'reshape'
582 	 * It has value MaxSector when no reshape is happening
583 	 * If delta_disks < 0, it is the last sector we started work on,
584 	 * else is it the next sector to work on.
585 	 */
586 	sector_t		reshape_progress;
587 	/* reshape_safe is the trailing edge of a reshape.  We know that
588 	 * before (or after) this address, all reshape has completed.
589 	 */
590 	sector_t		reshape_safe;
591 	int			previous_raid_disks;
592 	int			prev_chunk_sectors;
593 	int			prev_algo;
594 	short			generation; /* increments with every reshape */
595 	seqcount_t		gen_lock;	/* lock against generation changes */
596 	unsigned long		reshape_checkpoint; /* Time we last updated
597 						     * metadata */
598 	long long		min_offset_diff; /* minimum difference between
599 						  * data_offset and
600 						  * new_data_offset across all
601 						  * devices.  May be negative,
602 						  * but is closest to zero.
603 						  */
604 
605 	struct list_head	handle_list; /* stripes needing handling */
606 	struct list_head	loprio_list; /* low priority stripes */
607 	struct list_head	hold_list; /* preread ready stripes */
608 	struct list_head	delayed_list; /* stripes that have plugged requests */
609 	struct list_head	bitmap_list; /* stripes delaying awaiting bitmap update */
610 	struct bio		*retry_read_aligned; /* currently retrying aligned bios   */
611 	unsigned int		retry_read_offset; /* sector offset into retry_read_aligned */
612 	struct bio		*retry_read_aligned_list; /* aligned bios retry list  */
613 	atomic_t		preread_active_stripes; /* stripes with scheduled io */
614 	atomic_t		active_aligned_reads;
615 	atomic_t		pending_full_writes; /* full write backlog */
616 	int			bypass_count; /* bypassed prereads */
617 	int			bypass_threshold; /* preread nice */
618 	int			skip_copy; /* Don't copy data from bio to stripe cache */
619 	struct list_head	*last_hold; /* detect hold_list promotions */
620 
621 	atomic_t		reshape_stripes; /* stripes with pending writes for reshape */
622 	/* unfortunately we need two cache names as we temporarily have
623 	 * two caches.
624 	 */
625 	int			active_name;
626 	char			cache_name[2][32];
627 	struct kmem_cache	*slab_cache; /* for allocating stripes */
628 	struct mutex		cache_size_mutex; /* Protect changes to cache size */
629 
630 	int			seq_flush, seq_write;
631 	int			quiesce;
632 
633 	int			fullsync;  /* set to 1 if a full sync is needed,
634 					    * (fresh device added).
635 					    * Cleared when a sync completes.
636 					    */
637 	int			recovery_disabled;
638 	/* per cpu variables */
639 	struct raid5_percpu {
640 		struct page	*spare_page; /* Used when checking P/Q in raid6 */
641 		struct flex_array *scribble;   /* space for constructing buffer
642 					      * lists and performing address
643 					      * conversions
644 					      */
645 	} __percpu *percpu;
646 	int scribble_disks;
647 	int scribble_sectors;
648 	struct hlist_node node;
649 
650 	/*
651 	 * Free stripes pool
652 	 */
653 	atomic_t		active_stripes;
654 	struct list_head	inactive_list[NR_STRIPE_HASH_LOCKS];
655 
656 	atomic_t		r5c_cached_full_stripes;
657 	struct list_head	r5c_full_stripe_list;
658 	atomic_t		r5c_cached_partial_stripes;
659 	struct list_head	r5c_partial_stripe_list;
660 	atomic_t		r5c_flushing_full_stripes;
661 	atomic_t		r5c_flushing_partial_stripes;
662 
663 	atomic_t		empty_inactive_list_nr;
664 	struct llist_head	released_stripes;
665 	wait_queue_head_t	wait_for_quiescent;
666 	wait_queue_head_t	wait_for_stripe;
667 	wait_queue_head_t	wait_for_overlap;
668 	unsigned long		cache_state;
669 	struct shrinker		shrinker;
670 	int			pool_size; /* number of disks in stripeheads in pool */
671 	spinlock_t		device_lock;
672 	struct disk_info	*disks;
673 	struct bio_set		bio_split;
674 
675 	/* When taking over an array from a different personality, we store
676 	 * the new thread here until we fully activate the array.
677 	 */
678 	struct md_thread	*thread;
679 	struct list_head	temp_inactive_list[NR_STRIPE_HASH_LOCKS];
680 	struct r5worker_group	*worker_groups;
681 	int			group_cnt;
682 	int			worker_cnt_per_group;
683 	struct r5l_log		*log;
684 	void			*log_private;
685 
686 	spinlock_t		pending_bios_lock;
687 	bool			batch_bio_dispatch;
688 	struct r5pending_data	*pending_data;
689 	struct list_head	free_list;
690 	struct list_head	pending_list;
691 	int			pending_data_cnt;
692 	struct r5pending_data	*next_pending_data;
693 };
694 
695 
696 /*
697  * Our supported algorithms
698  */
699 #define ALGORITHM_LEFT_ASYMMETRIC	0 /* Rotating Parity N with Data Restart */
700 #define ALGORITHM_RIGHT_ASYMMETRIC	1 /* Rotating Parity 0 with Data Restart */
701 #define ALGORITHM_LEFT_SYMMETRIC	2 /* Rotating Parity N with Data Continuation */
702 #define ALGORITHM_RIGHT_SYMMETRIC	3 /* Rotating Parity 0 with Data Continuation */
703 
704 /* Define non-rotating (raid4) algorithms.  These allow
705  * conversion of raid4 to raid5.
706  */
707 #define ALGORITHM_PARITY_0		4 /* P or P,Q are initial devices */
708 #define ALGORITHM_PARITY_N		5 /* P or P,Q are final devices. */
709 
710 /* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
711  * Firstly, the exact positioning of the parity block is slightly
712  * different between the 'LEFT_*' modes of md and the "_N_*" modes
713  * of DDF.
714  * Secondly, or order of datablocks over which the Q syndrome is computed
715  * is different.
716  * Consequently we have different layouts for DDF/raid6 than md/raid6.
717  * These layouts are from the DDFv1.2 spec.
718  * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
719  * leaves RLQ=3 as 'Vendor Specific'
720  */
721 
722 #define ALGORITHM_ROTATING_ZERO_RESTART	8 /* DDF PRL=6 RLQ=1 */
723 #define ALGORITHM_ROTATING_N_RESTART	9 /* DDF PRL=6 RLQ=2 */
724 #define ALGORITHM_ROTATING_N_CONTINUE	10 /*DDF PRL=6 RLQ=3 */
725 
726 /* For every RAID5 algorithm we define a RAID6 algorithm
727  * with exactly the same layout for data and parity, and
728  * with the Q block always on the last device (N-1).
729  * This allows trivial conversion from RAID5 to RAID6
730  */
731 #define ALGORITHM_LEFT_ASYMMETRIC_6	16
732 #define ALGORITHM_RIGHT_ASYMMETRIC_6	17
733 #define ALGORITHM_LEFT_SYMMETRIC_6	18
734 #define ALGORITHM_RIGHT_SYMMETRIC_6	19
735 #define ALGORITHM_PARITY_0_6		20
736 #define ALGORITHM_PARITY_N_6		ALGORITHM_PARITY_N
737 
algorithm_valid_raid5(int layout)738 static inline int algorithm_valid_raid5(int layout)
739 {
740 	return (layout >= 0) &&
741 		(layout <= 5);
742 }
algorithm_valid_raid6(int layout)743 static inline int algorithm_valid_raid6(int layout)
744 {
745 	return (layout >= 0 && layout <= 5)
746 		||
747 		(layout >= 8 && layout <= 10)
748 		||
749 		(layout >= 16 && layout <= 20);
750 }
751 
algorithm_is_DDF(int layout)752 static inline int algorithm_is_DDF(int layout)
753 {
754 	return layout >= 8 && layout <= 10;
755 }
756 
757 extern void md_raid5_kick_device(struct r5conf *conf);
758 extern int raid5_set_cache_size(struct mddev *mddev, int size);
759 extern sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous);
760 extern void raid5_release_stripe(struct stripe_head *sh);
761 extern sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
762 				     int previous, int *dd_idx,
763 				     struct stripe_head *sh);
764 extern struct stripe_head *
765 raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
766 			int previous, int noblock, int noquiesce);
767 extern int raid5_calc_degraded(struct r5conf *conf);
768 extern int r5c_journal_mode_set(struct mddev *mddev, int journal_mode);
769 #endif
770