4 #include <linux/raid/xor.h>
8 * Each stripe contains one buffer per disc. Each buffer can be in
9 * one of a number of states stored in "flags". Changes between
10 * these states happen *almost* exclusively under a per-stripe
11 * spinlock. Some very specific changes can happen in bi_end_io, and
12 * these are not protected by the spin lock.
14 * The flag bits that are used to represent these states are:
15 * R5_UPTODATE and R5_LOCKED
17 * State Empty == !UPTODATE, !LOCK
18 * We have no data, and there is no active request
19 * State Want == !UPTODATE, LOCK
20 * A read request is being submitted for this block
21 * State Dirty == UPTODATE, LOCK
22 * Some new data is in this buffer, and it is being written out
23 * State Clean == UPTODATE, !LOCK
24 * We have valid data which is the same as on disc
26 * The possible state transitions are:
28 * Empty -> Want - on read or write to get old data for parity calc
29 * Empty -> Dirty - on compute_parity to satisfy write/sync request.(RECONSTRUCT_WRITE)
30 * Empty -> Clean - on compute_block when computing a block for failed drive
31 * Want -> Empty - on failed read
32 * Want -> Clean - on successful completion of read request
33 * Dirty -> Clean - on successful completion of write request
34 * Dirty -> Clean - on failed write
35 * Clean -> Dirty - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
37 * The Want->Empty, Want->Clean, Dirty->Clean, transitions
38 * all happen in b_end_io at interrupt time.
39 * Each sets the Uptodate bit before releasing the Lock bit.
40 * This leaves one multi-stage transition:
42 * This is safe because thinking that a Clean buffer is actually dirty
43 * will at worst delay some action, and the stripe will be scheduled
44 * for attention after the transition is complete.
46 * There is one possibility that is not covered by these states. That
47 * is if one drive has failed and there is a spare being rebuilt. We
48 * can't distinguish between a clean block that has been generated
49 * from parity calculations, and a clean block that has been
50 * successfully written to the spare ( or to parity when resyncing).
51 * To distingush these states we have a stripe bit STRIPE_INSYNC that
52 * is set whenever a write is scheduled to the spare, or to the parity
53 * disc if there is no spare. A sync request clears this bit, and
54 * when we find it set with no buffers locked, we know the sync is
57 * Buffers for the md device that arrive via make_request are attached
58 * to the appropriate stripe in one of two lists linked on b_reqnext.
59 * One list (bh_read) for read requests, one (bh_write) for write.
60 * There should never be more than one buffer on the two lists
61 * together, but we are not guaranteed of that so we allow for more.
63 * If a buffer is on the read list when the associated cache buffer is
64 * Uptodate, the data is copied into the read buffer and it's b_end_io
65 * routine is called. This may happen in the end_request routine only
66 * if the buffer has just successfully been read. end_request should
67 * remove the buffers from the list and then set the Uptodate bit on
68 * the buffer. Other threads may do this only if they first check
69 * that the Uptodate bit is set. Once they have checked that they may
70 * take buffers off the read queue.
72 * When a buffer on the write list is committed for write it is copied
73 * into the cache buffer, which is then marked dirty, and moved onto a
74 * third list, the written list (bh_written). Once both the parity
75 * block and the cached buffer are successfully written, any buffer on
76 * a written list can be returned with b_end_io.
78 * The write list and read list both act as fifos. The read list is
79 * protected by the device_lock. The write and written lists are
80 * protected by the stripe lock. The device_lock, which can be
81 * claimed while the stipe lock is held, is only for list
82 * manipulations and will only be held for a very short time. It can
83 * be claimed from interrupts.
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.
96 * The inactive_list, handle_list and hash bucket lists are all protected by the
98 * - stripes on the inactive_list never have their stripe_lock held.
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
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.
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())
118 * lockstripe clrSTRIPE_HANDLE ...
119 * (lockdev check-buffers unlockdev) ..
121 * record io/ops needed unlockstripe schedule io/ops
122 * release an active stripe (release_stripe())
123 * lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
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
127 * on a cached buffer, and plus one if the stripe is undergoing stripe
130 * Stripe operations are performed outside the stripe lock,
131 * the stripe operations are:
132 * -copying data between the stripe cache and user application buffers
133 * -computing blocks to save a disk access, or to recover a missing block
134 * -updating the parity on a write operation (reconstruct write and
136 * -checking parity correctness
137 * -running i/o to disk
138 * These operations are carried out by raid5_run_ops which uses the async_tx
139 * api to (optionally) offload operations to dedicated hardware engines.
140 * When requesting an operation handle_stripe sets the pending bit for the
141 * operation and increments the count. raid5_run_ops is then run whenever
142 * the count is non-zero.
143 * There are some critical dependencies between the operations that prevent some
144 * from being requested while another is in flight.
145 * 1/ Parity check operations destroy the in cache version of the parity block,
146 * so we prevent parity dependent operations like writes and compute_blocks
147 * from starting while a check is in progress. Some dma engines can perform
148 * the check without damaging the parity block, in these cases the parity
149 * block is re-marked up to date (assuming the check was successful) and is
150 * not re-read from disk.
151 * 2/ When a write operation is requested we immediately lock the affected
152 * blocks, and mark them as not up to date. This causes new read requests
153 * to be held off, as well as parity checks and compute block operations.
154 * 3/ Once a compute block operation has been requested handle_stripe treats
155 * that block as if it is up to date. raid5_run_ops guaruntees that any
156 * operation that is dependent on the compute block result is initiated after
157 * the compute block completes.
161 * Operations state - intermediate states that are visible outside of sh->lock
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)
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
177 check_state_idle = 0,
178 check_state_run, /* parity check */
179 check_state_check_result,
180 check_state_compute_run, /* parity repair */
181 check_state_compute_result,
185 * enum reconstruct_states - handles writing or expanding a stripe
187 enum reconstruct_states {
188 reconstruct_state_idle = 0,
189 reconstruct_state_prexor_drain_run, /* prexor-write */
190 reconstruct_state_drain_run, /* write */
191 reconstruct_state_run, /* expand */
192 reconstruct_state_prexor_drain_result,
193 reconstruct_state_drain_result,
194 reconstruct_state_result,
198 struct hlist_node hash;
199 struct list_head lru; /* inactive_list or handle_list */
200 struct raid5_private_data *raid_conf;
201 short generation; /* increments with every
203 sector_t sector; /* sector of this row */
204 short pd_idx; /* parity disk index */
205 short qd_idx; /* 'Q' disk index for raid6 */
206 short ddf_layout;/* use DDF ordering to calculate Q */
207 unsigned long state; /* state flags */
208 atomic_t count; /* nr of active thread/requests */
210 int bm_seq; /* sequence number for bitmap flushes */
211 int disks; /* disks in stripe */
212 enum check_states check_state;
213 enum reconstruct_states reconstruct_state;
215 * @target - STRIPE_OP_COMPUTE_BLK target
217 struct stripe_operations {
225 struct bio *toread, *read, *towrite, *written;
226 sector_t sector; /* sector of this page */
228 } dev[1]; /* allocated with extra space depending of RAID geometry */
231 /* stripe_head_state - collects and tracks the dynamic state of a stripe_head
232 * for handle_stripe. It is only valid under spin_lock(sh->lock);
234 struct stripe_head_state {
235 int syncing, expanding, expanded;
236 int locked, uptodate, to_read, to_write, failed, written;
237 int to_fill, compute, req_compute, non_overwrite;
239 unsigned long ops_request;
242 /* r6_state - extra state data only relevant to r6 */
244 int p_failed, q_failed, failed_num[2];
248 #define R5_UPTODATE 0 /* page contains current data */
249 #define R5_LOCKED 1 /* IO has been submitted on "req" */
250 #define R5_OVERWRITE 2 /* towrite covers whole page */
251 /* and some that are internal to handle_stripe */
252 #define R5_Insync 3 /* rdev && rdev->in_sync at start */
253 #define R5_Wantread 4 /* want to schedule a read */
254 #define R5_Wantwrite 5
255 #define R5_Overlap 7 /* There is a pending overlapping request on this block */
256 #define R5_ReadError 8 /* seen a read error here recently */
257 #define R5_ReWrite 9 /* have tried to over-write the readerror */
259 #define R5_Expanded 10 /* This block now has post-expand data */
260 #define R5_Wantcompute 11 /* compute_block in progress treat as
263 #define R5_Wantfill 12 /* dev->toread contains a bio that needs
266 #define R5_Wantdrain 13 /* dev->towrite needs to be drained */
270 #define RECONSTRUCT_WRITE 1
271 #define READ_MODIFY_WRITE 2
272 /* not a write method, but a compute_parity mode */
273 #define CHECK_PARITY 3
274 /* Additional compute_parity mode -- updates the parity w/o LOCKING */
275 #define UPDATE_PARITY 4
280 #define STRIPE_HANDLE 2
281 #define STRIPE_SYNCING 3
282 #define STRIPE_INSYNC 4
283 #define STRIPE_PREREAD_ACTIVE 5
284 #define STRIPE_DELAYED 6
285 #define STRIPE_DEGRADED 7
286 #define STRIPE_BIT_DELAY 8
287 #define STRIPE_EXPANDING 9
288 #define STRIPE_EXPAND_SOURCE 10
289 #define STRIPE_EXPAND_READY 11
290 #define STRIPE_IO_STARTED 12 /* do not count towards 'bypass_count' */
291 #define STRIPE_FULL_WRITE 13 /* all blocks are set to be overwritten */
292 #define STRIPE_BIOFILL_RUN 14
293 #define STRIPE_COMPUTE_RUN 15
295 * Operation request flags
297 #define STRIPE_OP_BIOFILL 0
298 #define STRIPE_OP_COMPUTE_BLK 1
299 #define STRIPE_OP_PREXOR 2
300 #define STRIPE_OP_BIODRAIN 3
301 #define STRIPE_OP_POSTXOR 4
302 #define STRIPE_OP_CHECK 5
307 * To improve write throughput, we need to delay the handling of some
308 * stripes until there has been a chance that several write requests
309 * for the one stripe have all been collected.
310 * In particular, any write request that would require pre-reading
311 * is put on a "delayed" queue until there are no stripes currently
312 * in a pre-read phase. Further, if the "delayed" queue is empty when
313 * a stripe is put on it then we "plug" the queue and do not process it
314 * until an unplug call is made. (the unplug_io_fn() is called).
316 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
317 * it to the count of prereading stripes.
318 * When write is initiated, or the stripe refcnt == 0 (just in case) we
319 * clear the PREREAD_ACTIVE flag and decrement the count
320 * Whenever the 'handle' queue is empty and the device is not plugged, we
321 * move any strips from delayed to handle and clear the DELAYED flag and set
323 * In stripe_handle, if we find pre-reading is necessary, we do it if
324 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
325 * HANDLE gets cleared if stripe_handle leave nothing locked.
333 struct raid5_private_data {
334 struct hlist_head *stripe_hashtbl;
336 struct disk_info *spare;
337 int chunk_size, level, algorithm;
342 /* reshape_progress is the leading edge of a 'reshape'
343 * It has value MaxSector when no reshape is happening
344 * If delta_disks < 0, it is the last sector we started work on,
345 * else is it the next sector to work on.
347 sector_t reshape_progress;
348 /* reshape_safe is the trailing edge of a reshape. We know that
349 * before (or after) this address, all reshape has completed.
351 sector_t reshape_safe;
352 int previous_raid_disks;
353 int prev_chunk, prev_algo;
354 short generation; /* increments with every reshape */
355 unsigned long reshape_checkpoint; /* Time we last updated
358 struct list_head handle_list; /* stripes needing handling */
359 struct list_head hold_list; /* preread ready stripes */
360 struct list_head delayed_list; /* stripes that have plugged requests */
361 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
362 struct bio *retry_read_aligned; /* currently retrying aligned bios */
363 struct bio *retry_read_aligned_list; /* aligned bios retry list */
364 atomic_t preread_active_stripes; /* stripes with scheduled io */
365 atomic_t active_aligned_reads;
366 atomic_t pending_full_writes; /* full write backlog */
367 int bypass_count; /* bypassed prereads */
368 int bypass_threshold; /* preread nice */
369 struct list_head *last_hold; /* detect hold_list promotions */
371 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
372 /* unfortunately we need two cache names as we temporarily have
376 char cache_name[2][20];
377 struct kmem_cache *slab_cache; /* for allocating stripes */
379 int seq_flush, seq_write;
382 int fullsync; /* set to 1 if a full sync is needed,
383 * (fresh device added).
384 * Cleared when a sync completes.
387 struct page *spare_page; /* Used when checking P/Q in raid6 */
392 atomic_t active_stripes;
393 struct list_head inactive_list;
394 wait_queue_head_t wait_for_stripe;
395 wait_queue_head_t wait_for_overlap;
396 int inactive_blocked; /* release of inactive stripes blocked,
397 * waiting for 25% to be free
399 int pool_size; /* number of disks in stripeheads in pool */
400 spinlock_t device_lock;
401 struct disk_info *disks;
403 /* When taking over an array from a different personality, we store
404 * the new thread here until we fully activate the array.
406 struct mdk_thread_s *thread;
409 typedef struct raid5_private_data raid5_conf_t;
412 * Our supported algorithms
414 #define ALGORITHM_LEFT_ASYMMETRIC 0 /* Rotating Parity N with Data Restart */
415 #define ALGORITHM_RIGHT_ASYMMETRIC 1 /* Rotating Parity 0 with Data Restart */
416 #define ALGORITHM_LEFT_SYMMETRIC 2 /* Rotating Parity N with Data Continuation */
417 #define ALGORITHM_RIGHT_SYMMETRIC 3 /* Rotating Parity 0 with Data Continuation */
419 /* Define non-rotating (raid4) algorithms. These allow
420 * conversion of raid4 to raid5.
422 #define ALGORITHM_PARITY_0 4 /* P or P,Q are initial devices */
423 #define ALGORITHM_PARITY_N 5 /* P or P,Q are final devices. */
425 /* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
426 * Firstly, the exact positioning of the parity block is slightly
427 * different between the 'LEFT_*' modes of md and the "_N_*" modes
429 * Secondly, or order of datablocks over which the Q syndrome is computed
431 * Consequently we have different layouts for DDF/raid6 than md/raid6.
432 * These layouts are from the DDFv1.2 spec.
433 * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
434 * leaves RLQ=3 as 'Vendor Specific'
437 #define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
438 #define ALGORITHM_ROTATING_N_RESTART 9 /* DDF PRL=6 RLQ=2 */
439 #define ALGORITHM_ROTATING_N_CONTINUE 10 /*DDF PRL=6 RLQ=3 */
442 /* For every RAID5 algorithm we define a RAID6 algorithm
443 * with exactly the same layout for data and parity, and
444 * with the Q block always on the last device (N-1).
445 * This allows trivial conversion from RAID5 to RAID6
447 #define ALGORITHM_LEFT_ASYMMETRIC_6 16
448 #define ALGORITHM_RIGHT_ASYMMETRIC_6 17
449 #define ALGORITHM_LEFT_SYMMETRIC_6 18
450 #define ALGORITHM_RIGHT_SYMMETRIC_6 19
451 #define ALGORITHM_PARITY_0_6 20
452 #define ALGORITHM_PARITY_N_6 ALGORITHM_PARITY_N
454 static inline int algorithm_valid_raid5(int layout)
456 return (layout >= 0) &&
459 static inline int algorithm_valid_raid6(int layout)
461 return (layout >= 0 && layout <= 5)
463 (layout == 8 || layout == 10)
465 (layout >= 16 && layout <= 20);
468 static inline int algorithm_is_DDF(int layout)
470 return layout >= 8 && layout <= 10;