xref: /linux/drivers/md/raid5.h (revision 765532c8aaac624b5f8687af6d319c6a1138a257)
1 #ifndef _RAID5_H
2 #define _RAID5_H
3 
4 #include <linux/raid/xor.h>
5 #include <linux/dmaengine.h>
6 
7 /*
8  *
9  * Each stripe contains one buffer per disc.  Each buffer can be in
10  * one of a number of states stored in "flags".  Changes between
11  * these states happen *almost* exclusively under a per-stripe
12  * spinlock.  Some very specific changes can happen in bi_end_io, and
13  * these are not protected by the spin lock.
14  *
15  * The flag bits that are used to represent these states are:
16  *   R5_UPTODATE and R5_LOCKED
17  *
18  * State Empty == !UPTODATE, !LOCK
19  *        We have no data, and there is no active request
20  * State Want == !UPTODATE, LOCK
21  *        A read request is being submitted for this block
22  * State Dirty == UPTODATE, LOCK
23  *        Some new data is in this buffer, and it is being written out
24  * State Clean == UPTODATE, !LOCK
25  *        We have valid data which is the same as on disc
26  *
27  * The possible state transitions are:
28  *
29  *  Empty -> Want   - on read or write to get old data for  parity calc
30  *  Empty -> Dirty  - on compute_parity to satisfy write/sync request.(RECONSTRUCT_WRITE)
31  *  Empty -> Clean  - on compute_block when computing a block for failed drive
32  *  Want  -> Empty  - on failed read
33  *  Want  -> Clean  - on successful completion of read request
34  *  Dirty -> Clean  - on successful completion of write request
35  *  Dirty -> Clean  - on failed write
36  *  Clean -> Dirty  - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
37  *
38  * The Want->Empty, Want->Clean, Dirty->Clean, transitions
39  * all happen in b_end_io at interrupt time.
40  * Each sets the Uptodate bit before releasing the Lock bit.
41  * This leaves one multi-stage transition:
42  *    Want->Dirty->Clean
43  * This is safe because thinking that a Clean buffer is actually dirty
44  * will at worst delay some action, and the stripe will be scheduled
45  * for attention after the transition is complete.
46  *
47  * There is one possibility that is not covered by these states.  That
48  * is if one drive has failed and there is a spare being rebuilt.  We
49  * can't distinguish between a clean block that has been generated
50  * from parity calculations, and a clean block that has been
51  * successfully written to the spare ( or to parity when resyncing).
52  * To distingush these states we have a stripe bit STRIPE_INSYNC that
53  * is set whenever a write is scheduled to the spare, or to the parity
54  * disc if there is no spare.  A sync request clears this bit, and
55  * when we find it set with no buffers locked, we know the sync is
56  * complete.
57  *
58  * Buffers for the md device that arrive via make_request are attached
59  * to the appropriate stripe in one of two lists linked on b_reqnext.
60  * One list (bh_read) for read requests, one (bh_write) for write.
61  * There should never be more than one buffer on the two lists
62  * together, but we are not guaranteed of that so we allow for more.
63  *
64  * If a buffer is on the read list when the associated cache buffer is
65  * Uptodate, the data is copied into the read buffer and it's b_end_io
66  * routine is called.  This may happen in the end_request routine only
67  * if the buffer has just successfully been read.  end_request should
68  * remove the buffers from the list and then set the Uptodate bit on
69  * the buffer.  Other threads may do this only if they first check
70  * that the Uptodate bit is set.  Once they have checked that they may
71  * take buffers off the read queue.
72  *
73  * When a buffer on the write list is committed for write it is copied
74  * into the cache buffer, which is then marked dirty, and moved onto a
75  * third list, the written list (bh_written).  Once both the parity
76  * block and the cached buffer are successfully written, any buffer on
77  * a written list can be returned with b_end_io.
78  *
79  * The write list and read list both act as fifos.  The read list is
80  * protected by the device_lock.  The write and written lists are
81  * protected by the stripe lock.  The device_lock, which can be
82  * claimed while the stipe lock is held, is only for list
83  * manipulations and will only be held for a very short time.  It can
84  * be claimed from interrupts.
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.
99  *  - stripes on the inactive_list never have their stripe_lock held.
100  *  - stripes have a reference counter. If count==0, they are on a list.
101  *  - If a stripe might need handling, STRIPE_HANDLE is set.
102  *  - When refcount reaches zero, then if STRIPE_HANDLE it is put on
103  *    handle_list else inactive_list
104  *
105  * This, combined with the fact that STRIPE_HANDLE is only ever
106  * cleared while a stripe has a non-zero count means that if the
107  * refcount is 0 and STRIPE_HANDLE is set, then it is on the
108  * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
109  * the stripe is on inactive_list.
110  *
111  * The possible transitions are:
112  *  activate an unhashed/inactive stripe (get_active_stripe())
113  *     lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
114  *  activate a hashed, possibly active stripe (get_active_stripe())
115  *     lockdev check-hash if(!cnt++)unlink-stripe unlockdev
116  *  attach a request to an active stripe (add_stripe_bh())
117  *     lockdev attach-buffer unlockdev
118  *  handle a stripe (handle_stripe())
119  *     lockstripe clrSTRIPE_HANDLE ...
120  *		(lockdev check-buffers unlockdev) ..
121  *		change-state ..
122  *		record io/ops needed unlockstripe schedule io/ops
123  *  release an active stripe (release_stripe())
124  *     lockdev if (!--cnt) { if  STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
125  *
126  * The refcount counts each thread that have activated the stripe,
127  * plus raid5d if it is handling it, plus one for each active request
128  * on a cached buffer, and plus one if the stripe is undergoing stripe
129  * operations.
130  *
131  * Stripe operations are performed outside the stripe lock,
132  * the stripe operations are:
133  * -copying data between the stripe cache and user application buffers
134  * -computing blocks to save a disk access, or to recover a missing block
135  * -updating the parity on a write operation (reconstruct write and
136  *  read-modify-write)
137  * -checking parity correctness
138  * -running i/o to disk
139  * These operations are carried out by raid5_run_ops which uses the async_tx
140  * api to (optionally) offload operations to dedicated hardware engines.
141  * When requesting an operation handle_stripe sets the pending bit for the
142  * operation and increments the count.  raid5_run_ops is then run whenever
143  * the count is non-zero.
144  * There are some critical dependencies between the operations that prevent some
145  * from being requested while another is in flight.
146  * 1/ Parity check operations destroy the in cache version of the parity block,
147  *    so we prevent parity dependent operations like writes and compute_blocks
148  *    from starting while a check is in progress.  Some dma engines can perform
149  *    the check without damaging the parity block, in these cases the parity
150  *    block is re-marked up to date (assuming the check was successful) and is
151  *    not re-read from disk.
152  * 2/ When a write operation is requested we immediately lock the affected
153  *    blocks, and mark them as not up to date.  This causes new read requests
154  *    to be held off, as well as parity checks and compute block operations.
155  * 3/ Once a compute block operation has been requested handle_stripe treats
156  *    that block as if it is up to date.  raid5_run_ops guaruntees that any
157  *    operation that is dependent on the compute block result is initiated after
158  *    the compute block completes.
159  */
160 
161 /*
162  * Operations state - intermediate states that are visible outside of sh->lock
163  * In general _idle indicates nothing is running, _run indicates a data
164  * processing operation is active, and _result means the data processing result
165  * is stable and can be acted upon.  For simple operations like biofill and
166  * compute that only have an _idle and _run state they are indicated with
167  * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
168  */
169 /**
170  * enum check_states - handles syncing / repairing a stripe
171  * @check_state_idle - check operations are quiesced
172  * @check_state_run - check operation is running
173  * @check_state_result - set outside lock when check result is valid
174  * @check_state_compute_run - check failed and we are repairing
175  * @check_state_compute_result - set outside lock when compute result is valid
176  */
177 enum check_states {
178 	check_state_idle = 0,
179 	check_state_run, /* xor parity check */
180 	check_state_run_q, /* q-parity check */
181 	check_state_run_pq, /* pq dual parity check */
182 	check_state_check_result,
183 	check_state_compute_run, /* parity repair */
184 	check_state_compute_result,
185 };
186 
187 /**
188  * enum reconstruct_states - handles writing or expanding a stripe
189  */
190 enum reconstruct_states {
191 	reconstruct_state_idle = 0,
192 	reconstruct_state_prexor_drain_run,	/* prexor-write */
193 	reconstruct_state_drain_run,		/* write */
194 	reconstruct_state_run,			/* expand */
195 	reconstruct_state_prexor_drain_result,
196 	reconstruct_state_drain_result,
197 	reconstruct_state_result,
198 };
199 
200 struct stripe_head {
201 	struct hlist_node	hash;
202 	struct list_head	lru;	      /* inactive_list or handle_list */
203 	struct raid5_private_data *raid_conf;
204 	short			generation;	/* increments with every
205 						 * reshape */
206 	sector_t		sector;		/* sector of this row */
207 	short			pd_idx;		/* parity disk index */
208 	short			qd_idx;		/* 'Q' disk index for raid6 */
209 	short			ddf_layout;/* use DDF ordering to calculate Q */
210 	unsigned long		state;		/* state flags */
211 	atomic_t		count;	      /* nr of active thread/requests */
212 	spinlock_t		lock;
213 	int			bm_seq;	/* sequence number for bitmap flushes */
214 	int			disks;		/* disks in stripe */
215 	enum check_states	check_state;
216 	enum reconstruct_states reconstruct_state;
217 	/**
218 	 * struct stripe_operations
219 	 * @target - STRIPE_OP_COMPUTE_BLK target
220 	 * @target2 - 2nd compute target in the raid6 case
221 	 * @zero_sum_result - P and Q verification flags
222 	 * @request - async service request flags for raid_run_ops
223 	 */
224 	struct stripe_operations {
225 		int 		     target, target2;
226 		enum sum_check_flags zero_sum_result;
227 		#ifdef CONFIG_MULTICORE_RAID456
228 		unsigned long	     request;
229 		wait_queue_head_t    wait_for_ops;
230 		#endif
231 	} ops;
232 	struct r5dev {
233 		struct bio	req;
234 		struct bio_vec	vec;
235 		struct page	*page;
236 		struct bio	*toread, *read, *towrite, *written;
237 		sector_t	sector;			/* sector of this page */
238 		unsigned long	flags;
239 	} dev[1]; /* allocated with extra space depending of RAID geometry */
240 };
241 
242 /* stripe_head_state - collects and tracks the dynamic state of a stripe_head
243  *     for handle_stripe.  It is only valid under spin_lock(sh->lock);
244  */
245 struct stripe_head_state {
246 	int syncing, expanding, expanded;
247 	int locked, uptodate, to_read, to_write, failed, written;
248 	int to_fill, compute, req_compute, non_overwrite;
249 	int failed_num;
250 	unsigned long ops_request;
251 };
252 
253 /* r6_state - extra state data only relevant to r6 */
254 struct r6_state {
255 	int p_failed, q_failed, failed_num[2];
256 };
257 
258 /* Flags */
259 #define	R5_UPTODATE	0	/* page contains current data */
260 #define	R5_LOCKED	1	/* IO has been submitted on "req" */
261 #define	R5_OVERWRITE	2	/* towrite covers whole page */
262 /* and some that are internal to handle_stripe */
263 #define	R5_Insync	3	/* rdev && rdev->in_sync at start */
264 #define	R5_Wantread	4	/* want to schedule a read */
265 #define	R5_Wantwrite	5
266 #define	R5_Overlap	7	/* There is a pending overlapping request on this block */
267 #define	R5_ReadError	8	/* seen a read error here recently */
268 #define	R5_ReWrite	9	/* have tried to over-write the readerror */
269 
270 #define	R5_Expanded	10	/* This block now has post-expand data */
271 #define	R5_Wantcompute	11 /* compute_block in progress treat as
272 				    * uptodate
273 				    */
274 #define	R5_Wantfill	12 /* dev->toread contains a bio that needs
275 				    * filling
276 				    */
277 #define R5_Wantdrain	13 /* dev->towrite needs to be drained */
278 #define R5_WantFUA	14	/* Write should be FUA */
279 /*
280  * Write method
281  */
282 #define RECONSTRUCT_WRITE	1
283 #define READ_MODIFY_WRITE	2
284 /* not a write method, but a compute_parity mode */
285 #define	CHECK_PARITY		3
286 /* Additional compute_parity mode -- updates the parity w/o LOCKING */
287 #define UPDATE_PARITY		4
288 
289 /*
290  * Stripe state
291  */
292 #define STRIPE_HANDLE		2
293 #define	STRIPE_SYNCING		3
294 #define	STRIPE_INSYNC		4
295 #define	STRIPE_PREREAD_ACTIVE	5
296 #define	STRIPE_DELAYED		6
297 #define	STRIPE_DEGRADED		7
298 #define	STRIPE_BIT_DELAY	8
299 #define	STRIPE_EXPANDING	9
300 #define	STRIPE_EXPAND_SOURCE	10
301 #define	STRIPE_EXPAND_READY	11
302 #define	STRIPE_IO_STARTED	12 /* do not count towards 'bypass_count' */
303 #define	STRIPE_FULL_WRITE	13 /* all blocks are set to be overwritten */
304 #define	STRIPE_BIOFILL_RUN	14
305 #define	STRIPE_COMPUTE_RUN	15
306 #define	STRIPE_OPS_REQ_PENDING	16
307 
308 /*
309  * Operation request flags
310  */
311 #define STRIPE_OP_BIOFILL	0
312 #define STRIPE_OP_COMPUTE_BLK	1
313 #define STRIPE_OP_PREXOR	2
314 #define STRIPE_OP_BIODRAIN	3
315 #define STRIPE_OP_RECONSTRUCT	4
316 #define STRIPE_OP_CHECK	5
317 
318 /*
319  * Plugging:
320  *
321  * To improve write throughput, we need to delay the handling of some
322  * stripes until there has been a chance that several write requests
323  * for the one stripe have all been collected.
324  * In particular, any write request that would require pre-reading
325  * is put on a "delayed" queue until there are no stripes currently
326  * in a pre-read phase.  Further, if the "delayed" queue is empty when
327  * a stripe is put on it then we "plug" the queue and do not process it
328  * until an unplug call is made. (the unplug_io_fn() is called).
329  *
330  * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
331  * it to the count of prereading stripes.
332  * When write is initiated, or the stripe refcnt == 0 (just in case) we
333  * clear the PREREAD_ACTIVE flag and decrement the count
334  * Whenever the 'handle' queue is empty and the device is not plugged, we
335  * move any strips from delayed to handle and clear the DELAYED flag and set
336  * PREREAD_ACTIVE.
337  * In stripe_handle, if we find pre-reading is necessary, we do it if
338  * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
339  * HANDLE gets cleared if stripe_handle leave nothing locked.
340  */
341 
342 
343 struct disk_info {
344 	mdk_rdev_t	*rdev;
345 };
346 
347 struct raid5_private_data {
348 	struct hlist_head	*stripe_hashtbl;
349 	mddev_t			*mddev;
350 	struct disk_info	*spare;
351 	int			chunk_sectors;
352 	int			level, algorithm;
353 	int			max_degraded;
354 	int			raid_disks;
355 	int			max_nr_stripes;
356 
357 	/* reshape_progress is the leading edge of a 'reshape'
358 	 * It has value MaxSector when no reshape is happening
359 	 * If delta_disks < 0, it is the last sector we started work on,
360 	 * else is it the next sector to work on.
361 	 */
362 	sector_t		reshape_progress;
363 	/* reshape_safe is the trailing edge of a reshape.  We know that
364 	 * before (or after) this address, all reshape has completed.
365 	 */
366 	sector_t		reshape_safe;
367 	int			previous_raid_disks;
368 	int			prev_chunk_sectors;
369 	int			prev_algo;
370 	short			generation; /* increments with every reshape */
371 	unsigned long		reshape_checkpoint; /* Time we last updated
372 						     * metadata */
373 
374 	struct list_head	handle_list; /* stripes needing handling */
375 	struct list_head	hold_list; /* preread ready stripes */
376 	struct list_head	delayed_list; /* stripes that have plugged requests */
377 	struct list_head	bitmap_list; /* stripes delaying awaiting bitmap update */
378 	struct bio		*retry_read_aligned; /* currently retrying aligned bios   */
379 	struct bio		*retry_read_aligned_list; /* aligned bios retry list  */
380 	atomic_t		preread_active_stripes; /* stripes with scheduled io */
381 	atomic_t		active_aligned_reads;
382 	atomic_t		pending_full_writes; /* full write backlog */
383 	int			bypass_count; /* bypassed prereads */
384 	int			bypass_threshold; /* preread nice */
385 	struct list_head	*last_hold; /* detect hold_list promotions */
386 
387 	atomic_t		reshape_stripes; /* stripes with pending writes for reshape */
388 	/* unfortunately we need two cache names as we temporarily have
389 	 * two caches.
390 	 */
391 	int			active_name;
392 	char			cache_name[2][32];
393 	struct kmem_cache		*slab_cache; /* for allocating stripes */
394 
395 	int			seq_flush, seq_write;
396 	int			quiesce;
397 
398 	int			fullsync;  /* set to 1 if a full sync is needed,
399 					    * (fresh device added).
400 					    * Cleared when a sync completes.
401 					    */
402 
403 	struct plug_handle	plug;
404 
405 	/* per cpu variables */
406 	struct raid5_percpu {
407 		struct page	*spare_page; /* Used when checking P/Q in raid6 */
408 		void		*scribble;   /* space for constructing buffer
409 					      * lists and performing address
410 					      * conversions
411 					      */
412 	} __percpu *percpu;
413 	size_t			scribble_len; /* size of scribble region must be
414 					       * associated with conf to handle
415 					       * cpu hotplug while reshaping
416 					       */
417 #ifdef CONFIG_HOTPLUG_CPU
418 	struct notifier_block	cpu_notify;
419 #endif
420 
421 	/*
422 	 * Free stripes pool
423 	 */
424 	atomic_t		active_stripes;
425 	struct list_head	inactive_list;
426 	wait_queue_head_t	wait_for_stripe;
427 	wait_queue_head_t	wait_for_overlap;
428 	int			inactive_blocked;	/* release of inactive stripes blocked,
429 							 * waiting for 25% to be free
430 							 */
431 	int			pool_size; /* number of disks in stripeheads in pool */
432 	spinlock_t		device_lock;
433 	struct disk_info	*disks;
434 
435 	/* When taking over an array from a different personality, we store
436 	 * the new thread here until we fully activate the array.
437 	 */
438 	struct mdk_thread_s	*thread;
439 };
440 
441 typedef struct raid5_private_data raid5_conf_t;
442 
443 /*
444  * Our supported algorithms
445  */
446 #define ALGORITHM_LEFT_ASYMMETRIC	0 /* Rotating Parity N with Data Restart */
447 #define ALGORITHM_RIGHT_ASYMMETRIC	1 /* Rotating Parity 0 with Data Restart */
448 #define ALGORITHM_LEFT_SYMMETRIC	2 /* Rotating Parity N with Data Continuation */
449 #define ALGORITHM_RIGHT_SYMMETRIC	3 /* Rotating Parity 0 with Data Continuation */
450 
451 /* Define non-rotating (raid4) algorithms.  These allow
452  * conversion of raid4 to raid5.
453  */
454 #define ALGORITHM_PARITY_0		4 /* P or P,Q are initial devices */
455 #define ALGORITHM_PARITY_N		5 /* P or P,Q are final devices. */
456 
457 /* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
458  * Firstly, the exact positioning of the parity block is slightly
459  * different between the 'LEFT_*' modes of md and the "_N_*" modes
460  * of DDF.
461  * Secondly, or order of datablocks over which the Q syndrome is computed
462  * is different.
463  * Consequently we have different layouts for DDF/raid6 than md/raid6.
464  * These layouts are from the DDFv1.2 spec.
465  * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
466  * leaves RLQ=3 as 'Vendor Specific'
467  */
468 
469 #define ALGORITHM_ROTATING_ZERO_RESTART	8 /* DDF PRL=6 RLQ=1 */
470 #define ALGORITHM_ROTATING_N_RESTART	9 /* DDF PRL=6 RLQ=2 */
471 #define ALGORITHM_ROTATING_N_CONTINUE	10 /*DDF PRL=6 RLQ=3 */
472 
473 
474 /* For every RAID5 algorithm we define a RAID6 algorithm
475  * with exactly the same layout for data and parity, and
476  * with the Q block always on the last device (N-1).
477  * This allows trivial conversion from RAID5 to RAID6
478  */
479 #define ALGORITHM_LEFT_ASYMMETRIC_6	16
480 #define ALGORITHM_RIGHT_ASYMMETRIC_6	17
481 #define ALGORITHM_LEFT_SYMMETRIC_6	18
482 #define ALGORITHM_RIGHT_SYMMETRIC_6	19
483 #define ALGORITHM_PARITY_0_6		20
484 #define ALGORITHM_PARITY_N_6		ALGORITHM_PARITY_N
485 
486 static inline int algorithm_valid_raid5(int layout)
487 {
488 	return (layout >= 0) &&
489 		(layout <= 5);
490 }
491 static inline int algorithm_valid_raid6(int layout)
492 {
493 	return (layout >= 0 && layout <= 5)
494 		||
495 		(layout >= 8 && layout <= 10)
496 		||
497 		(layout >= 16 && layout <= 20);
498 }
499 
500 static inline int algorithm_is_DDF(int layout)
501 {
502 	return layout >= 8 && layout <= 10;
503 }
504 
505 extern int md_raid5_congested(mddev_t *mddev, int bits);
506 extern void md_raid5_unplug_device(raid5_conf_t *conf);
507 extern int raid5_set_cache_size(mddev_t *mddev, int size);
508 #endif
509