4 #include <linux/raid/xor.h>
5 #include <linux/dmaengine.h>
9 * Each stripe contains one buffer per device. Each buffer can be in
10 * one of a number of states stored in "flags". Changes between
11 * these states happen *almost* exclusively under the protection of the
12 * STRIPE_ACTIVE flag. Some very specific changes can happen in bi_end_io, and
13 * these are not protected by STRIPE_ACTIVE.
15 * The flag bits that are used to represent these states are:
16 * R5_UPTODATE and R5_LOCKED
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
27 * The possible state transitions are:
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.
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)
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:
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.
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
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.
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.
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.
79 * The write list and read list both act as fifos. The read list,
80 * write list and written list are protected by the device_lock.
81 * The device_lock is only for list manipulations and will only be
82 * held for a very short time. It can be claimed from interrupts.
85 * Stripes in the stripe cache can be on one of two lists (or on
86 * neither). The "inactive_list" contains stripes which are not
87 * currently being used for any request. They can freely be reused
88 * for another stripe. The "handle_list" contains stripes that need
89 * to be handled in some way. Both of these are fifo queues. Each
90 * stripe is also (potentially) linked to a hash bucket in the hash
91 * table so that it can be found by sector number. Stripes that are
92 * not hashed must be on the inactive_list, and will normally be at
93 * the front. All stripes start life this way.
95 * The inactive_list, handle_list and hash bucket lists are all protected by the
97 * - stripes have a reference counter. If count==0, they are on a list.
98 * - If a stripe might need handling, STRIPE_HANDLE is set.
99 * - When refcount reaches zero, then if STRIPE_HANDLE it is put on
100 * handle_list else inactive_list
102 * This, combined with the fact that STRIPE_HANDLE is only ever
103 * cleared while a stripe has a non-zero count means that if the
104 * refcount is 0 and STRIPE_HANDLE is set, then it is on the
105 * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
106 * the stripe is on inactive_list.
108 * The possible transitions are:
109 * activate an unhashed/inactive stripe (get_active_stripe())
110 * lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
111 * activate a hashed, possibly active stripe (get_active_stripe())
112 * lockdev check-hash if(!cnt++)unlink-stripe unlockdev
113 * attach a request to an active stripe (add_stripe_bh())
114 * lockdev attach-buffer unlockdev
115 * handle a stripe (handle_stripe())
116 * setSTRIPE_ACTIVE, clrSTRIPE_HANDLE ...
117 * (lockdev check-buffers unlockdev) ..
119 * record io/ops needed clearSTRIPE_ACTIVE schedule io/ops
120 * release an active stripe (release_stripe())
121 * lockdev if (!--cnt) { if STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
123 * The refcount counts each thread that have activated the stripe,
124 * plus raid5d if it is handling it, plus one for each active request
125 * on a cached buffer, and plus one if the stripe is undergoing stripe
128 * The stripe operations are:
129 * -copying data between the stripe cache and user application buffers
130 * -computing blocks to save a disk access, or to recover a missing block
131 * -updating the parity on a write operation (reconstruct write and
133 * -checking parity correctness
134 * -running i/o to disk
135 * These operations are carried out by raid5_run_ops which uses the async_tx
136 * api to (optionally) offload operations to dedicated hardware engines.
137 * When requesting an operation handle_stripe sets the pending bit for the
138 * operation and increments the count. raid5_run_ops is then run whenever
139 * the count is non-zero.
140 * There are some critical dependencies between the operations that prevent some
141 * from being requested while another is in flight.
142 * 1/ Parity check operations destroy the in cache version of the parity block,
143 * so we prevent parity dependent operations like writes and compute_blocks
144 * from starting while a check is in progress. Some dma engines can perform
145 * the check without damaging the parity block, in these cases the parity
146 * block is re-marked up to date (assuming the check was successful) and is
147 * not re-read from disk.
148 * 2/ When a write operation is requested we immediately lock the affected
149 * blocks, and mark them as not up to date. This causes new read requests
150 * to be held off, as well as parity checks and compute block operations.
151 * 3/ Once a compute block operation has been requested handle_stripe treats
152 * that block as if it is up to date. raid5_run_ops guaruntees that any
153 * operation that is dependent on the compute block result is initiated after
154 * the compute block completes.
158 * Operations state - intermediate states that are visible outside of
160 * In general _idle indicates nothing is running, _run indicates a data
161 * processing operation is active, and _result means the data processing result
162 * is stable and can be acted upon. For simple operations like biofill and
163 * compute that only have an _idle and _run state they are indicated with
164 * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
167 * enum check_states - handles syncing / repairing a stripe
168 * @check_state_idle - check operations are quiesced
169 * @check_state_run - check operation is running
170 * @check_state_result - set outside lock when check result is valid
171 * @check_state_compute_run - check failed and we are repairing
172 * @check_state_compute_result - set outside lock when compute result is valid
175 check_state_idle = 0,
176 check_state_run, /* xor parity check */
177 check_state_run_q, /* q-parity check */
178 check_state_run_pq, /* pq dual 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 r5conf *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 */
209 int bm_seq; /* sequence number for bitmap flushes */
210 int disks; /* disks in stripe */
211 enum check_states check_state;
212 enum reconstruct_states reconstruct_state;
213 spinlock_t stripe_lock;
215 * struct stripe_operations
216 * @target - STRIPE_OP_COMPUTE_BLK target
217 * @target2 - 2nd compute target in the raid6 case
218 * @zero_sum_result - P and Q verification flags
219 * @request - async service request flags for raid_run_ops
221 struct stripe_operations {
223 enum sum_check_flags zero_sum_result;
226 /* rreq and rvec are used for the replacement device when
227 * writing data to both devices.
229 struct bio req, rreq;
230 struct bio_vec vec, rvec;
232 struct bio *toread, *read, *towrite, *written;
233 sector_t sector; /* sector of this page */
235 } dev[1]; /* allocated with extra space depending of RAID geometry */
238 /* stripe_head_state - collects and tracks the dynamic state of a stripe_head
241 struct stripe_head_state {
242 /* 'syncing' means that we need to read all devices, either
243 * to check/correct parity, or to reconstruct a missing device.
244 * 'replacing' means we are replacing one or more drives and
245 * the source is valid at this point so we don't need to
246 * read all devices, just the replacement targets.
248 int syncing, expanding, expanded, replacing;
249 int locked, uptodate, to_read, to_write, failed, written;
250 int to_fill, compute, req_compute, non_overwrite;
252 int p_failed, q_failed;
253 int dec_preread_active;
254 unsigned long ops_request;
256 struct bio *return_bi;
257 struct md_rdev *blocked_rdev;
258 int handle_bad_blocks;
261 /* Flags for struct r5dev.flags */
263 R5_UPTODATE, /* page contains current data */
264 R5_LOCKED, /* IO has been submitted on "req" */
265 R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */
266 R5_OVERWRITE, /* towrite covers whole page */
267 /* and some that are internal to handle_stripe */
268 R5_Insync, /* rdev && rdev->in_sync at start */
269 R5_Wantread, /* want to schedule a read */
271 R5_Overlap, /* There is a pending overlapping request
273 R5_ReadNoMerge, /* prevent bio from merging in block-layer */
274 R5_ReadError, /* seen a read error here recently */
275 R5_ReWrite, /* have tried to over-write the readerror */
277 R5_Expanded, /* This block now has post-expand data */
278 R5_Wantcompute, /* compute_block in progress treat as
281 R5_Wantfill, /* dev->toread contains a bio that needs
284 R5_Wantdrain, /* dev->towrite needs to be drained */
285 R5_WantFUA, /* Write should be FUA */
286 R5_SyncIO, /* The IO is sync */
287 R5_WriteError, /* got a write error - need to record it */
288 R5_MadeGood, /* A bad block has been fixed by writing to it */
289 R5_ReadRepl, /* Will/did read from replacement rather than orig */
290 R5_MadeGoodRepl,/* A bad block on the replacement device has been
291 * fixed by writing to it */
292 R5_NeedReplace, /* This device has a replacement which is not
293 * up-to-date at this stripe. */
294 R5_WantReplace, /* We need to update the replacement, we have read
295 * data in, and now is a good time to write it out.
297 R5_Discard, /* Discard the stripe */
306 STRIPE_SYNC_REQUESTED,
309 STRIPE_PREREAD_ACTIVE,
314 STRIPE_EXPAND_SOURCE,
316 STRIPE_IO_STARTED, /* do not count towards 'bypass_count' */
317 STRIPE_FULL_WRITE, /* all blocks are set to be overwritten */
320 STRIPE_OPS_REQ_PENDING,
321 STRIPE_ON_UNPLUG_LIST,
326 * Operation request flags
330 STRIPE_OP_COMPUTE_BLK,
333 STRIPE_OP_RECONSTRUCT,
339 * To improve write throughput, we need to delay the handling of some
340 * stripes until there has been a chance that several write requests
341 * for the one stripe have all been collected.
342 * In particular, any write request that would require pre-reading
343 * is put on a "delayed" queue until there are no stripes currently
344 * in a pre-read phase. Further, if the "delayed" queue is empty when
345 * a stripe is put on it then we "plug" the queue and do not process it
346 * until an unplug call is made. (the unplug_io_fn() is called).
348 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
349 * it to the count of prereading stripes.
350 * When write is initiated, or the stripe refcnt == 0 (just in case) we
351 * clear the PREREAD_ACTIVE flag and decrement the count
352 * Whenever the 'handle' queue is empty and the device is not plugged, we
353 * move any strips from delayed to handle and clear the DELAYED flag and set
355 * In stripe_handle, if we find pre-reading is necessary, we do it if
356 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
357 * HANDLE gets cleared if stripe_handle leaves nothing locked.
362 struct md_rdev *rdev, *replacement;
366 struct hlist_head *stripe_hashtbl;
369 int level, algorithm;
374 /* reshape_progress is the leading edge of a 'reshape'
375 * It has value MaxSector when no reshape is happening
376 * If delta_disks < 0, it is the last sector we started work on,
377 * else is it the next sector to work on.
379 sector_t reshape_progress;
380 /* reshape_safe is the trailing edge of a reshape. We know that
381 * before (or after) this address, all reshape has completed.
383 sector_t reshape_safe;
384 int previous_raid_disks;
385 int prev_chunk_sectors;
387 short generation; /* increments with every reshape */
388 unsigned long reshape_checkpoint; /* Time we last updated
390 long long min_offset_diff; /* minimum difference between
392 * new_data_offset across all
393 * devices. May be negative,
394 * but is closest to zero.
397 struct list_head handle_list; /* stripes needing handling */
398 struct list_head hold_list; /* preread ready stripes */
399 struct list_head delayed_list; /* stripes that have plugged requests */
400 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */
401 struct bio *retry_read_aligned; /* currently retrying aligned bios */
402 struct bio *retry_read_aligned_list; /* aligned bios retry list */
403 atomic_t preread_active_stripes; /* stripes with scheduled io */
404 atomic_t active_aligned_reads;
405 atomic_t pending_full_writes; /* full write backlog */
406 int bypass_count; /* bypassed prereads */
407 int bypass_threshold; /* preread nice */
408 struct list_head *last_hold; /* detect hold_list promotions */
410 atomic_t reshape_stripes; /* stripes with pending writes for reshape */
411 /* unfortunately we need two cache names as we temporarily have
415 char cache_name[2][32];
416 struct kmem_cache *slab_cache; /* for allocating stripes */
418 int seq_flush, seq_write;
421 int fullsync; /* set to 1 if a full sync is needed,
422 * (fresh device added).
423 * Cleared when a sync completes.
425 int recovery_disabled;
426 /* per cpu variables */
427 struct raid5_percpu {
428 struct page *spare_page; /* Used when checking P/Q in raid6 */
429 void *scribble; /* space for constructing buffer
430 * lists and performing address
434 size_t scribble_len; /* size of scribble region must be
435 * associated with conf to handle
436 * cpu hotplug while reshaping
438 #ifdef CONFIG_HOTPLUG_CPU
439 struct notifier_block cpu_notify;
445 atomic_t active_stripes;
446 struct list_head inactive_list;
447 wait_queue_head_t wait_for_stripe;
448 wait_queue_head_t wait_for_overlap;
449 int inactive_blocked; /* release of inactive stripes blocked,
450 * waiting for 25% to be free
452 int pool_size; /* number of disks in stripeheads in pool */
453 spinlock_t device_lock;
454 struct disk_info *disks;
456 /* When taking over an array from a different personality, we store
457 * the new thread here until we fully activate the array.
459 struct md_thread *thread;
463 * Our supported algorithms
465 #define ALGORITHM_LEFT_ASYMMETRIC 0 /* Rotating Parity N with Data Restart */
466 #define ALGORITHM_RIGHT_ASYMMETRIC 1 /* Rotating Parity 0 with Data Restart */
467 #define ALGORITHM_LEFT_SYMMETRIC 2 /* Rotating Parity N with Data Continuation */
468 #define ALGORITHM_RIGHT_SYMMETRIC 3 /* Rotating Parity 0 with Data Continuation */
470 /* Define non-rotating (raid4) algorithms. These allow
471 * conversion of raid4 to raid5.
473 #define ALGORITHM_PARITY_0 4 /* P or P,Q are initial devices */
474 #define ALGORITHM_PARITY_N 5 /* P or P,Q are final devices. */
476 /* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
477 * Firstly, the exact positioning of the parity block is slightly
478 * different between the 'LEFT_*' modes of md and the "_N_*" modes
480 * Secondly, or order of datablocks over which the Q syndrome is computed
482 * Consequently we have different layouts for DDF/raid6 than md/raid6.
483 * These layouts are from the DDFv1.2 spec.
484 * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
485 * leaves RLQ=3 as 'Vendor Specific'
488 #define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
489 #define ALGORITHM_ROTATING_N_RESTART 9 /* DDF PRL=6 RLQ=2 */
490 #define ALGORITHM_ROTATING_N_CONTINUE 10 /*DDF PRL=6 RLQ=3 */
493 /* For every RAID5 algorithm we define a RAID6 algorithm
494 * with exactly the same layout for data and parity, and
495 * with the Q block always on the last device (N-1).
496 * This allows trivial conversion from RAID5 to RAID6
498 #define ALGORITHM_LEFT_ASYMMETRIC_6 16
499 #define ALGORITHM_RIGHT_ASYMMETRIC_6 17
500 #define ALGORITHM_LEFT_SYMMETRIC_6 18
501 #define ALGORITHM_RIGHT_SYMMETRIC_6 19
502 #define ALGORITHM_PARITY_0_6 20
503 #define ALGORITHM_PARITY_N_6 ALGORITHM_PARITY_N
505 static inline int algorithm_valid_raid5(int layout)
507 return (layout >= 0) &&
510 static inline int algorithm_valid_raid6(int layout)
512 return (layout >= 0 && layout <= 5)
514 (layout >= 8 && layout <= 10)
516 (layout >= 16 && layout <= 20);
519 static inline int algorithm_is_DDF(int layout)
521 return layout >= 8 && layout <= 10;
524 extern int md_raid5_congested(struct mddev *mddev, int bits);
525 extern void md_raid5_kick_device(struct r5conf *conf);
526 extern int raid5_set_cache_size(struct mddev *mddev, int size);