|
1 #ifndef _RAID5_H |
|
2 #define _RAID5_H |
|
3 |
|
4 #include <linux/raid/md.h> |
|
5 #include <linux/raid/xor.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, /* 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 raid5_private_data *raid_conf; |
|
202 sector_t sector; /* sector of this row */ |
|
203 int pd_idx; /* parity disk index */ |
|
204 unsigned long state; /* state flags */ |
|
205 atomic_t count; /* nr of active thread/requests */ |
|
206 spinlock_t lock; |
|
207 int bm_seq; /* sequence number for bitmap flushes */ |
|
208 int disks; /* disks in stripe */ |
|
209 enum check_states check_state; |
|
210 enum reconstruct_states reconstruct_state; |
|
211 /* stripe_operations |
|
212 * @target - STRIPE_OP_COMPUTE_BLK target |
|
213 */ |
|
214 struct stripe_operations { |
|
215 int target; |
|
216 u32 zero_sum_result; |
|
217 } ops; |
|
218 struct r5dev { |
|
219 struct bio req; |
|
220 struct bio_vec vec; |
|
221 struct page *page; |
|
222 struct bio *toread, *read, *towrite, *written; |
|
223 sector_t sector; /* sector of this page */ |
|
224 unsigned long flags; |
|
225 } dev[1]; /* allocated with extra space depending of RAID geometry */ |
|
226 }; |
|
227 |
|
228 /* stripe_head_state - collects and tracks the dynamic state of a stripe_head |
|
229 * for handle_stripe. It is only valid under spin_lock(sh->lock); |
|
230 */ |
|
231 struct stripe_head_state { |
|
232 int syncing, expanding, expanded; |
|
233 int locked, uptodate, to_read, to_write, failed, written; |
|
234 int to_fill, compute, req_compute, non_overwrite; |
|
235 int failed_num; |
|
236 unsigned long ops_request; |
|
237 }; |
|
238 |
|
239 /* r6_state - extra state data only relevant to r6 */ |
|
240 struct r6_state { |
|
241 int p_failed, q_failed, qd_idx, failed_num[2]; |
|
242 }; |
|
243 |
|
244 /* Flags */ |
|
245 #define R5_UPTODATE 0 /* page contains current data */ |
|
246 #define R5_LOCKED 1 /* IO has been submitted on "req" */ |
|
247 #define R5_OVERWRITE 2 /* towrite covers whole page */ |
|
248 /* and some that are internal to handle_stripe */ |
|
249 #define R5_Insync 3 /* rdev && rdev->in_sync at start */ |
|
250 #define R5_Wantread 4 /* want to schedule a read */ |
|
251 #define R5_Wantwrite 5 |
|
252 #define R5_Overlap 7 /* There is a pending overlapping request on this block */ |
|
253 #define R5_ReadError 8 /* seen a read error here recently */ |
|
254 #define R5_ReWrite 9 /* have tried to over-write the readerror */ |
|
255 |
|
256 #define R5_Expanded 10 /* This block now has post-expand data */ |
|
257 #define R5_Wantcompute 11 /* compute_block in progress treat as |
|
258 * uptodate |
|
259 */ |
|
260 #define R5_Wantfill 12 /* dev->toread contains a bio that needs |
|
261 * filling |
|
262 */ |
|
263 #define R5_Wantdrain 13 /* dev->towrite needs to be drained */ |
|
264 /* |
|
265 * Write method |
|
266 */ |
|
267 #define RECONSTRUCT_WRITE 1 |
|
268 #define READ_MODIFY_WRITE 2 |
|
269 /* not a write method, but a compute_parity mode */ |
|
270 #define CHECK_PARITY 3 |
|
271 |
|
272 /* |
|
273 * Stripe state |
|
274 */ |
|
275 #define STRIPE_HANDLE 2 |
|
276 #define STRIPE_SYNCING 3 |
|
277 #define STRIPE_INSYNC 4 |
|
278 #define STRIPE_PREREAD_ACTIVE 5 |
|
279 #define STRIPE_DELAYED 6 |
|
280 #define STRIPE_DEGRADED 7 |
|
281 #define STRIPE_BIT_DELAY 8 |
|
282 #define STRIPE_EXPANDING 9 |
|
283 #define STRIPE_EXPAND_SOURCE 10 |
|
284 #define STRIPE_EXPAND_READY 11 |
|
285 #define STRIPE_IO_STARTED 12 /* do not count towards 'bypass_count' */ |
|
286 #define STRIPE_FULL_WRITE 13 /* all blocks are set to be overwritten */ |
|
287 #define STRIPE_BIOFILL_RUN 14 |
|
288 #define STRIPE_COMPUTE_RUN 15 |
|
289 /* |
|
290 * Operation request flags |
|
291 */ |
|
292 #define STRIPE_OP_BIOFILL 0 |
|
293 #define STRIPE_OP_COMPUTE_BLK 1 |
|
294 #define STRIPE_OP_PREXOR 2 |
|
295 #define STRIPE_OP_BIODRAIN 3 |
|
296 #define STRIPE_OP_POSTXOR 4 |
|
297 #define STRIPE_OP_CHECK 5 |
|
298 |
|
299 /* |
|
300 * Plugging: |
|
301 * |
|
302 * To improve write throughput, we need to delay the handling of some |
|
303 * stripes until there has been a chance that several write requests |
|
304 * for the one stripe have all been collected. |
|
305 * In particular, any write request that would require pre-reading |
|
306 * is put on a "delayed" queue until there are no stripes currently |
|
307 * in a pre-read phase. Further, if the "delayed" queue is empty when |
|
308 * a stripe is put on it then we "plug" the queue and do not process it |
|
309 * until an unplug call is made. (the unplug_io_fn() is called). |
|
310 * |
|
311 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add |
|
312 * it to the count of prereading stripes. |
|
313 * When write is initiated, or the stripe refcnt == 0 (just in case) we |
|
314 * clear the PREREAD_ACTIVE flag and decrement the count |
|
315 * Whenever the 'handle' queue is empty and the device is not plugged, we |
|
316 * move any strips from delayed to handle and clear the DELAYED flag and set |
|
317 * PREREAD_ACTIVE. |
|
318 * In stripe_handle, if we find pre-reading is necessary, we do it if |
|
319 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue. |
|
320 * HANDLE gets cleared if stripe_handle leave nothing locked. |
|
321 */ |
|
322 |
|
323 |
|
324 struct disk_info { |
|
325 mdk_rdev_t *rdev; |
|
326 }; |
|
327 |
|
328 struct raid5_private_data { |
|
329 struct hlist_head *stripe_hashtbl; |
|
330 mddev_t *mddev; |
|
331 struct disk_info *spare; |
|
332 int chunk_size, level, algorithm; |
|
333 int max_degraded; |
|
334 int raid_disks; |
|
335 int max_nr_stripes; |
|
336 |
|
337 /* used during an expand */ |
|
338 sector_t expand_progress; /* MaxSector when no expand happening */ |
|
339 sector_t expand_lo; /* from here up to expand_progress it out-of-bounds |
|
340 * as we haven't flushed the metadata yet |
|
341 */ |
|
342 int previous_raid_disks; |
|
343 |
|
344 struct list_head handle_list; /* stripes needing handling */ |
|
345 struct list_head hold_list; /* preread ready stripes */ |
|
346 struct list_head delayed_list; /* stripes that have plugged requests */ |
|
347 struct list_head bitmap_list; /* stripes delaying awaiting bitmap update */ |
|
348 struct bio *retry_read_aligned; /* currently retrying aligned bios */ |
|
349 struct bio *retry_read_aligned_list; /* aligned bios retry list */ |
|
350 atomic_t preread_active_stripes; /* stripes with scheduled io */ |
|
351 atomic_t active_aligned_reads; |
|
352 atomic_t pending_full_writes; /* full write backlog */ |
|
353 int bypass_count; /* bypassed prereads */ |
|
354 int bypass_threshold; /* preread nice */ |
|
355 struct list_head *last_hold; /* detect hold_list promotions */ |
|
356 |
|
357 atomic_t reshape_stripes; /* stripes with pending writes for reshape */ |
|
358 /* unfortunately we need two cache names as we temporarily have |
|
359 * two caches. |
|
360 */ |
|
361 int active_name; |
|
362 char cache_name[2][20]; |
|
363 struct kmem_cache *slab_cache; /* for allocating stripes */ |
|
364 |
|
365 int seq_flush, seq_write; |
|
366 int quiesce; |
|
367 |
|
368 int fullsync; /* set to 1 if a full sync is needed, |
|
369 * (fresh device added). |
|
370 * Cleared when a sync completes. |
|
371 */ |
|
372 |
|
373 struct page *spare_page; /* Used when checking P/Q in raid6 */ |
|
374 |
|
375 /* |
|
376 * Free stripes pool |
|
377 */ |
|
378 atomic_t active_stripes; |
|
379 struct list_head inactive_list; |
|
380 wait_queue_head_t wait_for_stripe; |
|
381 wait_queue_head_t wait_for_overlap; |
|
382 int inactive_blocked; /* release of inactive stripes blocked, |
|
383 * waiting for 25% to be free |
|
384 */ |
|
385 int pool_size; /* number of disks in stripeheads in pool */ |
|
386 spinlock_t device_lock; |
|
387 struct disk_info *disks; |
|
388 }; |
|
389 |
|
390 typedef struct raid5_private_data raid5_conf_t; |
|
391 |
|
392 #define mddev_to_conf(mddev) ((raid5_conf_t *) mddev->private) |
|
393 |
|
394 /* |
|
395 * Our supported algorithms |
|
396 */ |
|
397 #define ALGORITHM_LEFT_ASYMMETRIC 0 |
|
398 #define ALGORITHM_RIGHT_ASYMMETRIC 1 |
|
399 #define ALGORITHM_LEFT_SYMMETRIC 2 |
|
400 #define ALGORITHM_RIGHT_SYMMETRIC 3 |
|
401 |
|
402 #endif |