Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
32 | * new additions. |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
36 | * When we are ready to do a write, if that batch hasn't been written yet, |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
056075c7 | 50 | #include <linux/module.h> |
07a3b417 | 51 | #include <linux/async.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
36d1c647 | 53 | #include <linux/cpu.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
8bda470e | 55 | #include <linux/ratelimit.h> |
43b2e5d8 | 56 | #include "md.h" |
bff61975 | 57 | #include "raid5.h" |
54071b38 | 58 | #include "raid0.h" |
ef740c37 | 59 | #include "bitmap.h" |
72626685 | 60 | |
1da177e4 LT |
61 | /* |
62 | * Stripe cache | |
63 | */ | |
64 | ||
65 | #define NR_STRIPES 256 | |
66 | #define STRIPE_SIZE PAGE_SIZE | |
67 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
68 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
69 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 70 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 71 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
72 | #define HASH_MASK (NR_HASH - 1) |
73 | ||
d1688a6d | 74 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
75 | { |
76 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
77 | return &conf->stripe_hashtbl[hash]; | |
78 | } | |
1da177e4 LT |
79 | |
80 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
81 | * order without overlap. There may be several bio's per stripe+device, and | |
82 | * a bio could span several devices. | |
83 | * When walking this list for a particular stripe+device, we must never proceed | |
84 | * beyond a bio that extends past this device, as the next bio might no longer | |
85 | * be valid. | |
db298e19 | 86 | * This function is used to determine the 'next' bio in the list, given the sector |
1da177e4 LT |
87 | * of the current stripe+device |
88 | */ | |
db298e19 N |
89 | static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector) |
90 | { | |
91 | int sectors = bio->bi_size >> 9; | |
92 | if (bio->bi_sector + sectors < sector + STRIPE_SECTORS) | |
93 | return bio->bi_next; | |
94 | else | |
95 | return NULL; | |
96 | } | |
1da177e4 | 97 | |
960e739d | 98 | /* |
5b99c2ff JA |
99 | * We maintain a biased count of active stripes in the bottom 16 bits of |
100 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d | 101 | */ |
e7836bd6 | 102 | static inline int raid5_bi_processed_stripes(struct bio *bio) |
960e739d | 103 | { |
e7836bd6 SL |
104 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
105 | return (atomic_read(segments) >> 16) & 0xffff; | |
960e739d JA |
106 | } |
107 | ||
e7836bd6 | 108 | static inline int raid5_dec_bi_active_stripes(struct bio *bio) |
960e739d | 109 | { |
e7836bd6 SL |
110 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
111 | return atomic_sub_return(1, segments) & 0xffff; | |
960e739d JA |
112 | } |
113 | ||
e7836bd6 | 114 | static inline void raid5_inc_bi_active_stripes(struct bio *bio) |
960e739d | 115 | { |
e7836bd6 SL |
116 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
117 | atomic_inc(segments); | |
960e739d JA |
118 | } |
119 | ||
e7836bd6 SL |
120 | static inline void raid5_set_bi_processed_stripes(struct bio *bio, |
121 | unsigned int cnt) | |
960e739d | 122 | { |
e7836bd6 SL |
123 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
124 | int old, new; | |
960e739d | 125 | |
e7836bd6 SL |
126 | do { |
127 | old = atomic_read(segments); | |
128 | new = (old & 0xffff) | (cnt << 16); | |
129 | } while (atomic_cmpxchg(segments, old, new) != old); | |
960e739d JA |
130 | } |
131 | ||
e7836bd6 | 132 | static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt) |
960e739d | 133 | { |
e7836bd6 SL |
134 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
135 | atomic_set(segments, cnt); | |
960e739d JA |
136 | } |
137 | ||
d0dabf7e N |
138 | /* Find first data disk in a raid6 stripe */ |
139 | static inline int raid6_d0(struct stripe_head *sh) | |
140 | { | |
67cc2b81 N |
141 | if (sh->ddf_layout) |
142 | /* ddf always start from first device */ | |
143 | return 0; | |
144 | /* md starts just after Q block */ | |
d0dabf7e N |
145 | if (sh->qd_idx == sh->disks - 1) |
146 | return 0; | |
147 | else | |
148 | return sh->qd_idx + 1; | |
149 | } | |
16a53ecc N |
150 | static inline int raid6_next_disk(int disk, int raid_disks) |
151 | { | |
152 | disk++; | |
153 | return (disk < raid_disks) ? disk : 0; | |
154 | } | |
a4456856 | 155 | |
d0dabf7e N |
156 | /* When walking through the disks in a raid5, starting at raid6_d0, |
157 | * We need to map each disk to a 'slot', where the data disks are slot | |
158 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
159 | * is raid_disks-1. This help does that mapping. | |
160 | */ | |
67cc2b81 N |
161 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
162 | int *count, int syndrome_disks) | |
d0dabf7e | 163 | { |
6629542e | 164 | int slot = *count; |
67cc2b81 | 165 | |
e4424fee | 166 | if (sh->ddf_layout) |
6629542e | 167 | (*count)++; |
d0dabf7e | 168 | if (idx == sh->pd_idx) |
67cc2b81 | 169 | return syndrome_disks; |
d0dabf7e | 170 | if (idx == sh->qd_idx) |
67cc2b81 | 171 | return syndrome_disks + 1; |
e4424fee | 172 | if (!sh->ddf_layout) |
6629542e | 173 | (*count)++; |
d0dabf7e N |
174 | return slot; |
175 | } | |
176 | ||
a4456856 DW |
177 | static void return_io(struct bio *return_bi) |
178 | { | |
179 | struct bio *bi = return_bi; | |
180 | while (bi) { | |
a4456856 DW |
181 | |
182 | return_bi = bi->bi_next; | |
183 | bi->bi_next = NULL; | |
184 | bi->bi_size = 0; | |
0e13fe23 | 185 | bio_endio(bi, 0); |
a4456856 DW |
186 | bi = return_bi; |
187 | } | |
188 | } | |
189 | ||
d1688a6d | 190 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 191 | |
600aa109 DW |
192 | static int stripe_operations_active(struct stripe_head *sh) |
193 | { | |
194 | return sh->check_state || sh->reconstruct_state || | |
195 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
196 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
197 | } | |
198 | ||
4eb788df | 199 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 200 | { |
4eb788df SL |
201 | BUG_ON(!list_empty(&sh->lru)); |
202 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
203 | if (test_bit(STRIPE_HANDLE, &sh->state)) { | |
204 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
205 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
206 | list_add_tail(&sh->lru, &conf->delayed_list); | |
207 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
208 | sh->bm_seq - conf->seq_write > 0) | |
209 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
210 | else { | |
211 | clear_bit(STRIPE_DELAYED, &sh->state); | |
212 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
213 | list_add_tail(&sh->lru, &conf->handle_list); | |
214 | } | |
215 | md_wakeup_thread(conf->mddev->thread); | |
216 | } else { | |
217 | BUG_ON(stripe_operations_active(sh)); | |
218 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
219 | if (atomic_dec_return(&conf->preread_active_stripes) | |
220 | < IO_THRESHOLD) | |
221 | md_wakeup_thread(conf->mddev->thread); | |
222 | atomic_dec(&conf->active_stripes); | |
223 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { | |
224 | list_add_tail(&sh->lru, &conf->inactive_list); | |
225 | wake_up(&conf->wait_for_stripe); | |
226 | if (conf->retry_read_aligned) | |
227 | md_wakeup_thread(conf->mddev->thread); | |
1da177e4 LT |
228 | } |
229 | } | |
230 | } | |
d0dabf7e | 231 | |
4eb788df SL |
232 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh) |
233 | { | |
234 | if (atomic_dec_and_test(&sh->count)) | |
235 | do_release_stripe(conf, sh); | |
236 | } | |
237 | ||
1da177e4 LT |
238 | static void release_stripe(struct stripe_head *sh) |
239 | { | |
d1688a6d | 240 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 241 | unsigned long flags; |
16a53ecc | 242 | |
4eb788df SL |
243 | local_irq_save(flags); |
244 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { | |
245 | do_release_stripe(conf, sh); | |
246 | spin_unlock(&conf->device_lock); | |
247 | } | |
248 | local_irq_restore(flags); | |
1da177e4 LT |
249 | } |
250 | ||
fccddba0 | 251 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 252 | { |
45b4233c DW |
253 | pr_debug("remove_hash(), stripe %llu\n", |
254 | (unsigned long long)sh->sector); | |
1da177e4 | 255 | |
fccddba0 | 256 | hlist_del_init(&sh->hash); |
1da177e4 LT |
257 | } |
258 | ||
d1688a6d | 259 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 260 | { |
fccddba0 | 261 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 262 | |
45b4233c DW |
263 | pr_debug("insert_hash(), stripe %llu\n", |
264 | (unsigned long long)sh->sector); | |
1da177e4 | 265 | |
fccddba0 | 266 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
267 | } |
268 | ||
269 | ||
270 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
d1688a6d | 271 | static struct stripe_head *get_free_stripe(struct r5conf *conf) |
1da177e4 LT |
272 | { |
273 | struct stripe_head *sh = NULL; | |
274 | struct list_head *first; | |
275 | ||
1da177e4 LT |
276 | if (list_empty(&conf->inactive_list)) |
277 | goto out; | |
278 | first = conf->inactive_list.next; | |
279 | sh = list_entry(first, struct stripe_head, lru); | |
280 | list_del_init(first); | |
281 | remove_hash(sh); | |
282 | atomic_inc(&conf->active_stripes); | |
283 | out: | |
284 | return sh; | |
285 | } | |
286 | ||
e4e11e38 | 287 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
288 | { |
289 | struct page *p; | |
290 | int i; | |
e4e11e38 | 291 | int num = sh->raid_conf->pool_size; |
1da177e4 | 292 | |
e4e11e38 | 293 | for (i = 0; i < num ; i++) { |
1da177e4 LT |
294 | p = sh->dev[i].page; |
295 | if (!p) | |
296 | continue; | |
297 | sh->dev[i].page = NULL; | |
2d1f3b5d | 298 | put_page(p); |
1da177e4 LT |
299 | } |
300 | } | |
301 | ||
e4e11e38 | 302 | static int grow_buffers(struct stripe_head *sh) |
1da177e4 LT |
303 | { |
304 | int i; | |
e4e11e38 | 305 | int num = sh->raid_conf->pool_size; |
1da177e4 | 306 | |
e4e11e38 | 307 | for (i = 0; i < num; i++) { |
1da177e4 LT |
308 | struct page *page; |
309 | ||
310 | if (!(page = alloc_page(GFP_KERNEL))) { | |
311 | return 1; | |
312 | } | |
313 | sh->dev[i].page = page; | |
314 | } | |
315 | return 0; | |
316 | } | |
317 | ||
784052ec | 318 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 319 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 320 | struct stripe_head *sh); |
1da177e4 | 321 | |
b5663ba4 | 322 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 323 | { |
d1688a6d | 324 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 325 | int i; |
1da177e4 | 326 | |
78bafebd ES |
327 | BUG_ON(atomic_read(&sh->count) != 0); |
328 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 329 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 330 | |
45b4233c | 331 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
332 | (unsigned long long)sh->sector); |
333 | ||
334 | remove_hash(sh); | |
16a53ecc | 335 | |
86b42c71 | 336 | sh->generation = conf->generation - previous; |
b5663ba4 | 337 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 338 | sh->sector = sector; |
911d4ee8 | 339 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
340 | sh->state = 0; |
341 | ||
7ecaa1e6 N |
342 | |
343 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
344 | struct r5dev *dev = &sh->dev[i]; |
345 | ||
d84e0f10 | 346 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 347 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 348 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 349 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 350 | dev->read, dev->towrite, dev->written, |
1da177e4 | 351 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 352 | WARN_ON(1); |
1da177e4 LT |
353 | } |
354 | dev->flags = 0; | |
784052ec | 355 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
356 | } |
357 | insert_hash(conf, sh); | |
358 | } | |
359 | ||
d1688a6d | 360 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 361 | short generation) |
1da177e4 LT |
362 | { |
363 | struct stripe_head *sh; | |
fccddba0 | 364 | struct hlist_node *hn; |
1da177e4 | 365 | |
45b4233c | 366 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 367 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 368 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 369 | return sh; |
45b4233c | 370 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
371 | return NULL; |
372 | } | |
373 | ||
674806d6 N |
374 | /* |
375 | * Need to check if array has failed when deciding whether to: | |
376 | * - start an array | |
377 | * - remove non-faulty devices | |
378 | * - add a spare | |
379 | * - allow a reshape | |
380 | * This determination is simple when no reshape is happening. | |
381 | * However if there is a reshape, we need to carefully check | |
382 | * both the before and after sections. | |
383 | * This is because some failed devices may only affect one | |
384 | * of the two sections, and some non-in_sync devices may | |
385 | * be insync in the section most affected by failed devices. | |
386 | */ | |
908f4fbd | 387 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 388 | { |
908f4fbd | 389 | int degraded, degraded2; |
674806d6 | 390 | int i; |
674806d6 N |
391 | |
392 | rcu_read_lock(); | |
393 | degraded = 0; | |
394 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 395 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
396 | if (rdev && test_bit(Faulty, &rdev->flags)) |
397 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
398 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
399 | degraded++; | |
400 | else if (test_bit(In_sync, &rdev->flags)) | |
401 | ; | |
402 | else | |
403 | /* not in-sync or faulty. | |
404 | * If the reshape increases the number of devices, | |
405 | * this is being recovered by the reshape, so | |
406 | * this 'previous' section is not in_sync. | |
407 | * If the number of devices is being reduced however, | |
408 | * the device can only be part of the array if | |
409 | * we are reverting a reshape, so this section will | |
410 | * be in-sync. | |
411 | */ | |
412 | if (conf->raid_disks >= conf->previous_raid_disks) | |
413 | degraded++; | |
414 | } | |
415 | rcu_read_unlock(); | |
908f4fbd N |
416 | if (conf->raid_disks == conf->previous_raid_disks) |
417 | return degraded; | |
674806d6 | 418 | rcu_read_lock(); |
908f4fbd | 419 | degraded2 = 0; |
674806d6 | 420 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 421 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
422 | if (rdev && test_bit(Faulty, &rdev->flags)) |
423 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 424 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 425 | degraded2++; |
674806d6 N |
426 | else if (test_bit(In_sync, &rdev->flags)) |
427 | ; | |
428 | else | |
429 | /* not in-sync or faulty. | |
430 | * If reshape increases the number of devices, this | |
431 | * section has already been recovered, else it | |
432 | * almost certainly hasn't. | |
433 | */ | |
434 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 435 | degraded2++; |
674806d6 N |
436 | } |
437 | rcu_read_unlock(); | |
908f4fbd N |
438 | if (degraded2 > degraded) |
439 | return degraded2; | |
440 | return degraded; | |
441 | } | |
442 | ||
443 | static int has_failed(struct r5conf *conf) | |
444 | { | |
445 | int degraded; | |
446 | ||
447 | if (conf->mddev->reshape_position == MaxSector) | |
448 | return conf->mddev->degraded > conf->max_degraded; | |
449 | ||
450 | degraded = calc_degraded(conf); | |
674806d6 N |
451 | if (degraded > conf->max_degraded) |
452 | return 1; | |
453 | return 0; | |
454 | } | |
455 | ||
b5663ba4 | 456 | static struct stripe_head * |
d1688a6d | 457 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 458 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
459 | { |
460 | struct stripe_head *sh; | |
461 | ||
45b4233c | 462 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
463 | |
464 | spin_lock_irq(&conf->device_lock); | |
465 | ||
466 | do { | |
72626685 | 467 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 468 | conf->quiesce == 0 || noquiesce, |
72626685 | 469 | conf->device_lock, /* nothing */); |
86b42c71 | 470 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
471 | if (!sh) { |
472 | if (!conf->inactive_blocked) | |
473 | sh = get_free_stripe(conf); | |
474 | if (noblock && sh == NULL) | |
475 | break; | |
476 | if (!sh) { | |
477 | conf->inactive_blocked = 1; | |
478 | wait_event_lock_irq(conf->wait_for_stripe, | |
479 | !list_empty(&conf->inactive_list) && | |
5036805b N |
480 | (atomic_read(&conf->active_stripes) |
481 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
482 | || !conf->inactive_blocked), |
483 | conf->device_lock, | |
7c13edc8 | 484 | ); |
1da177e4 LT |
485 | conf->inactive_blocked = 0; |
486 | } else | |
b5663ba4 | 487 | init_stripe(sh, sector, previous); |
1da177e4 LT |
488 | } else { |
489 | if (atomic_read(&sh->count)) { | |
ab69ae12 | 490 | BUG_ON(!list_empty(&sh->lru) |
8811b596 SL |
491 | && !test_bit(STRIPE_EXPANDING, &sh->state) |
492 | && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)); | |
1da177e4 LT |
493 | } else { |
494 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
495 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
496 | if (list_empty(&sh->lru) && |
497 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
498 | BUG(); |
499 | list_del_init(&sh->lru); | |
1da177e4 LT |
500 | } |
501 | } | |
502 | } while (sh == NULL); | |
503 | ||
504 | if (sh) | |
505 | atomic_inc(&sh->count); | |
506 | ||
507 | spin_unlock_irq(&conf->device_lock); | |
508 | return sh; | |
509 | } | |
510 | ||
05616be5 N |
511 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
512 | * in this stripe_head. | |
513 | */ | |
514 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
515 | { | |
516 | sector_t progress = conf->reshape_progress; | |
517 | /* Need a memory barrier to make sure we see the value | |
518 | * of conf->generation, or ->data_offset that was set before | |
519 | * reshape_progress was updated. | |
520 | */ | |
521 | smp_rmb(); | |
522 | if (progress == MaxSector) | |
523 | return 0; | |
524 | if (sh->generation == conf->generation - 1) | |
525 | return 0; | |
526 | /* We are in a reshape, and this is a new-generation stripe, | |
527 | * so use new_data_offset. | |
528 | */ | |
529 | return 1; | |
530 | } | |
531 | ||
6712ecf8 N |
532 | static void |
533 | raid5_end_read_request(struct bio *bi, int error); | |
534 | static void | |
535 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 536 | |
c4e5ac0a | 537 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 538 | { |
d1688a6d | 539 | struct r5conf *conf = sh->raid_conf; |
91c00924 DW |
540 | int i, disks = sh->disks; |
541 | ||
542 | might_sleep(); | |
543 | ||
544 | for (i = disks; i--; ) { | |
545 | int rw; | |
9a3e1101 | 546 | int replace_only = 0; |
977df362 N |
547 | struct bio *bi, *rbi; |
548 | struct md_rdev *rdev, *rrdev = NULL; | |
e9c7469b TH |
549 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
550 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
551 | rw = WRITE_FUA; | |
552 | else | |
553 | rw = WRITE; | |
554 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
91c00924 | 555 | rw = READ; |
9a3e1101 N |
556 | else if (test_and_clear_bit(R5_WantReplace, |
557 | &sh->dev[i].flags)) { | |
558 | rw = WRITE; | |
559 | replace_only = 1; | |
560 | } else | |
91c00924 | 561 | continue; |
bc0934f0 SL |
562 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
563 | rw |= REQ_SYNC; | |
91c00924 DW |
564 | |
565 | bi = &sh->dev[i].req; | |
977df362 | 566 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 DW |
567 | |
568 | bi->bi_rw = rw; | |
977df362 N |
569 | rbi->bi_rw = rw; |
570 | if (rw & WRITE) { | |
91c00924 | 571 | bi->bi_end_io = raid5_end_write_request; |
977df362 N |
572 | rbi->bi_end_io = raid5_end_write_request; |
573 | } else | |
91c00924 DW |
574 | bi->bi_end_io = raid5_end_read_request; |
575 | ||
576 | rcu_read_lock(); | |
9a3e1101 | 577 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
578 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
579 | rdev = rcu_dereference(conf->disks[i].rdev); | |
580 | if (!rdev) { | |
581 | rdev = rrdev; | |
582 | rrdev = NULL; | |
583 | } | |
9a3e1101 N |
584 | if (rw & WRITE) { |
585 | if (replace_only) | |
586 | rdev = NULL; | |
dd054fce N |
587 | if (rdev == rrdev) |
588 | /* We raced and saw duplicates */ | |
589 | rrdev = NULL; | |
9a3e1101 | 590 | } else { |
dd054fce | 591 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev) |
9a3e1101 N |
592 | rdev = rrdev; |
593 | rrdev = NULL; | |
594 | } | |
977df362 | 595 | |
91c00924 DW |
596 | if (rdev && test_bit(Faulty, &rdev->flags)) |
597 | rdev = NULL; | |
598 | if (rdev) | |
599 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
600 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
601 | rrdev = NULL; | |
602 | if (rrdev) | |
603 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
604 | rcu_read_unlock(); |
605 | ||
73e92e51 | 606 | /* We have already checked bad blocks for reads. Now |
977df362 N |
607 | * need to check for writes. We never accept write errors |
608 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
609 | */ |
610 | while ((rw & WRITE) && rdev && | |
611 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
612 | sector_t first_bad; | |
613 | int bad_sectors; | |
614 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
615 | &first_bad, &bad_sectors); | |
616 | if (!bad) | |
617 | break; | |
618 | ||
619 | if (bad < 0) { | |
620 | set_bit(BlockedBadBlocks, &rdev->flags); | |
621 | if (!conf->mddev->external && | |
622 | conf->mddev->flags) { | |
623 | /* It is very unlikely, but we might | |
624 | * still need to write out the | |
625 | * bad block log - better give it | |
626 | * a chance*/ | |
627 | md_check_recovery(conf->mddev); | |
628 | } | |
1850753d | 629 | /* |
630 | * Because md_wait_for_blocked_rdev | |
631 | * will dec nr_pending, we must | |
632 | * increment it first. | |
633 | */ | |
634 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
635 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
636 | } else { | |
637 | /* Acknowledged bad block - skip the write */ | |
638 | rdev_dec_pending(rdev, conf->mddev); | |
639 | rdev = NULL; | |
640 | } | |
641 | } | |
642 | ||
91c00924 | 643 | if (rdev) { |
9a3e1101 N |
644 | if (s->syncing || s->expanding || s->expanded |
645 | || s->replacing) | |
91c00924 DW |
646 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
647 | ||
2b7497f0 DW |
648 | set_bit(STRIPE_IO_STARTED, &sh->state); |
649 | ||
91c00924 DW |
650 | bi->bi_bdev = rdev->bdev; |
651 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 652 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
653 | bi->bi_rw, i); |
654 | atomic_inc(&sh->count); | |
05616be5 N |
655 | if (use_new_offset(conf, sh)) |
656 | bi->bi_sector = (sh->sector | |
657 | + rdev->new_data_offset); | |
658 | else | |
659 | bi->bi_sector = (sh->sector | |
660 | + rdev->data_offset); | |
3f9e7c14 | 661 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
662 | bi->bi_rw |= REQ_FLUSH; | |
663 | ||
91c00924 | 664 | bi->bi_flags = 1 << BIO_UPTODATE; |
91c00924 | 665 | bi->bi_idx = 0; |
91c00924 DW |
666 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
667 | bi->bi_io_vec[0].bv_offset = 0; | |
668 | bi->bi_size = STRIPE_SIZE; | |
669 | bi->bi_next = NULL; | |
977df362 N |
670 | if (rrdev) |
671 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
91c00924 | 672 | generic_make_request(bi); |
977df362 N |
673 | } |
674 | if (rrdev) { | |
9a3e1101 N |
675 | if (s->syncing || s->expanding || s->expanded |
676 | || s->replacing) | |
977df362 N |
677 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
678 | ||
679 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
680 | ||
681 | rbi->bi_bdev = rrdev->bdev; | |
682 | pr_debug("%s: for %llu schedule op %ld on " | |
683 | "replacement disc %d\n", | |
684 | __func__, (unsigned long long)sh->sector, | |
685 | rbi->bi_rw, i); | |
686 | atomic_inc(&sh->count); | |
05616be5 N |
687 | if (use_new_offset(conf, sh)) |
688 | rbi->bi_sector = (sh->sector | |
689 | + rrdev->new_data_offset); | |
690 | else | |
691 | rbi->bi_sector = (sh->sector | |
692 | + rrdev->data_offset); | |
977df362 N |
693 | rbi->bi_flags = 1 << BIO_UPTODATE; |
694 | rbi->bi_idx = 0; | |
695 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
696 | rbi->bi_io_vec[0].bv_offset = 0; | |
697 | rbi->bi_size = STRIPE_SIZE; | |
698 | rbi->bi_next = NULL; | |
699 | generic_make_request(rbi); | |
700 | } | |
701 | if (!rdev && !rrdev) { | |
b062962e | 702 | if (rw & WRITE) |
91c00924 DW |
703 | set_bit(STRIPE_DEGRADED, &sh->state); |
704 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
705 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
706 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
707 | set_bit(STRIPE_HANDLE, &sh->state); | |
708 | } | |
709 | } | |
710 | } | |
711 | ||
712 | static struct dma_async_tx_descriptor * | |
713 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
714 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
715 | { | |
716 | struct bio_vec *bvl; | |
717 | struct page *bio_page; | |
718 | int i; | |
719 | int page_offset; | |
a08abd8c | 720 | struct async_submit_ctl submit; |
0403e382 | 721 | enum async_tx_flags flags = 0; |
91c00924 DW |
722 | |
723 | if (bio->bi_sector >= sector) | |
724 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
725 | else | |
726 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 727 | |
0403e382 DW |
728 | if (frombio) |
729 | flags |= ASYNC_TX_FENCE; | |
730 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
731 | ||
91c00924 | 732 | bio_for_each_segment(bvl, bio, i) { |
fcde9075 | 733 | int len = bvl->bv_len; |
91c00924 DW |
734 | int clen; |
735 | int b_offset = 0; | |
736 | ||
737 | if (page_offset < 0) { | |
738 | b_offset = -page_offset; | |
739 | page_offset += b_offset; | |
740 | len -= b_offset; | |
741 | } | |
742 | ||
743 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
744 | clen = STRIPE_SIZE - page_offset; | |
745 | else | |
746 | clen = len; | |
747 | ||
748 | if (clen > 0) { | |
fcde9075 NK |
749 | b_offset += bvl->bv_offset; |
750 | bio_page = bvl->bv_page; | |
91c00924 DW |
751 | if (frombio) |
752 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 753 | b_offset, clen, &submit); |
91c00924 DW |
754 | else |
755 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 756 | page_offset, clen, &submit); |
91c00924 | 757 | } |
a08abd8c DW |
758 | /* chain the operations */ |
759 | submit.depend_tx = tx; | |
760 | ||
91c00924 DW |
761 | if (clen < len) /* hit end of page */ |
762 | break; | |
763 | page_offset += len; | |
764 | } | |
765 | ||
766 | return tx; | |
767 | } | |
768 | ||
769 | static void ops_complete_biofill(void *stripe_head_ref) | |
770 | { | |
771 | struct stripe_head *sh = stripe_head_ref; | |
772 | struct bio *return_bi = NULL; | |
e4d84909 | 773 | int i; |
91c00924 | 774 | |
e46b272b | 775 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
776 | (unsigned long long)sh->sector); |
777 | ||
778 | /* clear completed biofills */ | |
779 | for (i = sh->disks; i--; ) { | |
780 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
781 | |
782 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
783 | /* and check if we need to reply to a read request, |
784 | * new R5_Wantfill requests are held off until | |
83de75cc | 785 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
786 | */ |
787 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 788 | struct bio *rbi, *rbi2; |
91c00924 | 789 | |
91c00924 DW |
790 | BUG_ON(!dev->read); |
791 | rbi = dev->read; | |
792 | dev->read = NULL; | |
793 | while (rbi && rbi->bi_sector < | |
794 | dev->sector + STRIPE_SECTORS) { | |
795 | rbi2 = r5_next_bio(rbi, dev->sector); | |
e7836bd6 | 796 | if (!raid5_dec_bi_active_stripes(rbi)) { |
91c00924 DW |
797 | rbi->bi_next = return_bi; |
798 | return_bi = rbi; | |
799 | } | |
91c00924 DW |
800 | rbi = rbi2; |
801 | } | |
802 | } | |
803 | } | |
83de75cc | 804 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 DW |
805 | |
806 | return_io(return_bi); | |
807 | ||
e4d84909 | 808 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
809 | release_stripe(sh); |
810 | } | |
811 | ||
812 | static void ops_run_biofill(struct stripe_head *sh) | |
813 | { | |
814 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 815 | struct async_submit_ctl submit; |
91c00924 DW |
816 | int i; |
817 | ||
e46b272b | 818 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
819 | (unsigned long long)sh->sector); |
820 | ||
821 | for (i = sh->disks; i--; ) { | |
822 | struct r5dev *dev = &sh->dev[i]; | |
823 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
824 | struct bio *rbi; | |
b17459c0 | 825 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
826 | dev->read = rbi = dev->toread; |
827 | dev->toread = NULL; | |
b17459c0 | 828 | spin_unlock_irq(&sh->stripe_lock); |
91c00924 DW |
829 | while (rbi && rbi->bi_sector < |
830 | dev->sector + STRIPE_SECTORS) { | |
831 | tx = async_copy_data(0, rbi, dev->page, | |
832 | dev->sector, tx); | |
833 | rbi = r5_next_bio(rbi, dev->sector); | |
834 | } | |
835 | } | |
836 | } | |
837 | ||
838 | atomic_inc(&sh->count); | |
a08abd8c DW |
839 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
840 | async_trigger_callback(&submit); | |
91c00924 DW |
841 | } |
842 | ||
4e7d2c0a | 843 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 844 | { |
4e7d2c0a | 845 | struct r5dev *tgt; |
91c00924 | 846 | |
4e7d2c0a DW |
847 | if (target < 0) |
848 | return; | |
91c00924 | 849 | |
4e7d2c0a | 850 | tgt = &sh->dev[target]; |
91c00924 DW |
851 | set_bit(R5_UPTODATE, &tgt->flags); |
852 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
853 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
854 | } |
855 | ||
ac6b53b6 | 856 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
857 | { |
858 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 859 | |
e46b272b | 860 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
861 | (unsigned long long)sh->sector); |
862 | ||
ac6b53b6 | 863 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 864 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 865 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 866 | |
ecc65c9b DW |
867 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
868 | if (sh->check_state == check_state_compute_run) | |
869 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
870 | set_bit(STRIPE_HANDLE, &sh->state); |
871 | release_stripe(sh); | |
872 | } | |
873 | ||
d6f38f31 DW |
874 | /* return a pointer to the address conversion region of the scribble buffer */ |
875 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
876 | struct raid5_percpu *percpu) | |
877 | { | |
878 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
879 | } | |
880 | ||
881 | static struct dma_async_tx_descriptor * | |
882 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 883 | { |
91c00924 | 884 | int disks = sh->disks; |
d6f38f31 | 885 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
886 | int target = sh->ops.target; |
887 | struct r5dev *tgt = &sh->dev[target]; | |
888 | struct page *xor_dest = tgt->page; | |
889 | int count = 0; | |
890 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 891 | struct async_submit_ctl submit; |
91c00924 DW |
892 | int i; |
893 | ||
894 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 895 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
896 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
897 | ||
898 | for (i = disks; i--; ) | |
899 | if (i != target) | |
900 | xor_srcs[count++] = sh->dev[i].page; | |
901 | ||
902 | atomic_inc(&sh->count); | |
903 | ||
0403e382 | 904 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 905 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 906 | if (unlikely(count == 1)) |
a08abd8c | 907 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 908 | else |
a08abd8c | 909 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 910 | |
91c00924 DW |
911 | return tx; |
912 | } | |
913 | ||
ac6b53b6 DW |
914 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
915 | * @srcs - (struct page *) array of size sh->disks | |
916 | * @sh - stripe_head to parse | |
917 | * | |
918 | * Populates srcs in proper layout order for the stripe and returns the | |
919 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
920 | * destination buffer is recorded in srcs[count] and the Q destination | |
921 | * is recorded in srcs[count+1]]. | |
922 | */ | |
923 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
924 | { | |
925 | int disks = sh->disks; | |
926 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
927 | int d0_idx = raid6_d0(sh); | |
928 | int count; | |
929 | int i; | |
930 | ||
931 | for (i = 0; i < disks; i++) | |
5dd33c9a | 932 | srcs[i] = NULL; |
ac6b53b6 DW |
933 | |
934 | count = 0; | |
935 | i = d0_idx; | |
936 | do { | |
937 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
938 | ||
939 | srcs[slot] = sh->dev[i].page; | |
940 | i = raid6_next_disk(i, disks); | |
941 | } while (i != d0_idx); | |
ac6b53b6 | 942 | |
e4424fee | 943 | return syndrome_disks; |
ac6b53b6 DW |
944 | } |
945 | ||
946 | static struct dma_async_tx_descriptor * | |
947 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
948 | { | |
949 | int disks = sh->disks; | |
950 | struct page **blocks = percpu->scribble; | |
951 | int target; | |
952 | int qd_idx = sh->qd_idx; | |
953 | struct dma_async_tx_descriptor *tx; | |
954 | struct async_submit_ctl submit; | |
955 | struct r5dev *tgt; | |
956 | struct page *dest; | |
957 | int i; | |
958 | int count; | |
959 | ||
960 | if (sh->ops.target < 0) | |
961 | target = sh->ops.target2; | |
962 | else if (sh->ops.target2 < 0) | |
963 | target = sh->ops.target; | |
91c00924 | 964 | else |
ac6b53b6 DW |
965 | /* we should only have one valid target */ |
966 | BUG(); | |
967 | BUG_ON(target < 0); | |
968 | pr_debug("%s: stripe %llu block: %d\n", | |
969 | __func__, (unsigned long long)sh->sector, target); | |
970 | ||
971 | tgt = &sh->dev[target]; | |
972 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
973 | dest = tgt->page; | |
974 | ||
975 | atomic_inc(&sh->count); | |
976 | ||
977 | if (target == qd_idx) { | |
978 | count = set_syndrome_sources(blocks, sh); | |
979 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
980 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
981 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
982 | ops_complete_compute, sh, | |
ac6b53b6 DW |
983 | to_addr_conv(sh, percpu)); |
984 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
985 | } else { | |
986 | /* Compute any data- or p-drive using XOR */ | |
987 | count = 0; | |
988 | for (i = disks; i-- ; ) { | |
989 | if (i == target || i == qd_idx) | |
990 | continue; | |
991 | blocks[count++] = sh->dev[i].page; | |
992 | } | |
993 | ||
0403e382 DW |
994 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
995 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
996 | to_addr_conv(sh, percpu)); |
997 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
998 | } | |
91c00924 | 999 | |
91c00924 DW |
1000 | return tx; |
1001 | } | |
1002 | ||
ac6b53b6 DW |
1003 | static struct dma_async_tx_descriptor * |
1004 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1005 | { | |
1006 | int i, count, disks = sh->disks; | |
1007 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1008 | int d0_idx = raid6_d0(sh); | |
1009 | int faila = -1, failb = -1; | |
1010 | int target = sh->ops.target; | |
1011 | int target2 = sh->ops.target2; | |
1012 | struct r5dev *tgt = &sh->dev[target]; | |
1013 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1014 | struct dma_async_tx_descriptor *tx; | |
1015 | struct page **blocks = percpu->scribble; | |
1016 | struct async_submit_ctl submit; | |
1017 | ||
1018 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
1019 | __func__, (unsigned long long)sh->sector, target, target2); | |
1020 | BUG_ON(target < 0 || target2 < 0); | |
1021 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1022 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1023 | ||
6c910a78 | 1024 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1025 | * slot number conversion for 'faila' and 'failb' |
1026 | */ | |
1027 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1028 | blocks[i] = NULL; |
ac6b53b6 DW |
1029 | count = 0; |
1030 | i = d0_idx; | |
1031 | do { | |
1032 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1033 | ||
1034 | blocks[slot] = sh->dev[i].page; | |
1035 | ||
1036 | if (i == target) | |
1037 | faila = slot; | |
1038 | if (i == target2) | |
1039 | failb = slot; | |
1040 | i = raid6_next_disk(i, disks); | |
1041 | } while (i != d0_idx); | |
ac6b53b6 DW |
1042 | |
1043 | BUG_ON(faila == failb); | |
1044 | if (failb < faila) | |
1045 | swap(faila, failb); | |
1046 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1047 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1048 | ||
1049 | atomic_inc(&sh->count); | |
1050 | ||
1051 | if (failb == syndrome_disks+1) { | |
1052 | /* Q disk is one of the missing disks */ | |
1053 | if (faila == syndrome_disks) { | |
1054 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1055 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1056 | ops_complete_compute, sh, | |
1057 | to_addr_conv(sh, percpu)); | |
e4424fee | 1058 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1059 | STRIPE_SIZE, &submit); |
1060 | } else { | |
1061 | struct page *dest; | |
1062 | int data_target; | |
1063 | int qd_idx = sh->qd_idx; | |
1064 | ||
1065 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1066 | if (target == qd_idx) | |
1067 | data_target = target2; | |
1068 | else | |
1069 | data_target = target; | |
1070 | ||
1071 | count = 0; | |
1072 | for (i = disks; i-- ; ) { | |
1073 | if (i == data_target || i == qd_idx) | |
1074 | continue; | |
1075 | blocks[count++] = sh->dev[i].page; | |
1076 | } | |
1077 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1078 | init_async_submit(&submit, |
1079 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1080 | NULL, NULL, NULL, | |
1081 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1082 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1083 | &submit); | |
1084 | ||
1085 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
1086 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1087 | ops_complete_compute, sh, | |
1088 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1089 | return async_gen_syndrome(blocks, 0, count+2, |
1090 | STRIPE_SIZE, &submit); | |
1091 | } | |
ac6b53b6 | 1092 | } else { |
6c910a78 DW |
1093 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1094 | ops_complete_compute, sh, | |
1095 | to_addr_conv(sh, percpu)); | |
1096 | if (failb == syndrome_disks) { | |
1097 | /* We're missing D+P. */ | |
1098 | return async_raid6_datap_recov(syndrome_disks+2, | |
1099 | STRIPE_SIZE, faila, | |
1100 | blocks, &submit); | |
1101 | } else { | |
1102 | /* We're missing D+D. */ | |
1103 | return async_raid6_2data_recov(syndrome_disks+2, | |
1104 | STRIPE_SIZE, faila, failb, | |
1105 | blocks, &submit); | |
1106 | } | |
ac6b53b6 DW |
1107 | } |
1108 | } | |
1109 | ||
1110 | ||
91c00924 DW |
1111 | static void ops_complete_prexor(void *stripe_head_ref) |
1112 | { | |
1113 | struct stripe_head *sh = stripe_head_ref; | |
1114 | ||
e46b272b | 1115 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1116 | (unsigned long long)sh->sector); |
91c00924 DW |
1117 | } |
1118 | ||
1119 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
1120 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
1121 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1122 | { |
91c00924 | 1123 | int disks = sh->disks; |
d6f38f31 | 1124 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1125 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1126 | struct async_submit_ctl submit; |
91c00924 DW |
1127 | |
1128 | /* existing parity data subtracted */ | |
1129 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1130 | ||
e46b272b | 1131 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1132 | (unsigned long long)sh->sector); |
1133 | ||
1134 | for (i = disks; i--; ) { | |
1135 | struct r5dev *dev = &sh->dev[i]; | |
1136 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1137 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1138 | xor_srcs[count++] = dev->page; |
1139 | } | |
1140 | ||
0403e382 | 1141 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 1142 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 1143 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1144 | |
1145 | return tx; | |
1146 | } | |
1147 | ||
1148 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1149 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1150 | { |
1151 | int disks = sh->disks; | |
d8ee0728 | 1152 | int i; |
91c00924 | 1153 | |
e46b272b | 1154 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1155 | (unsigned long long)sh->sector); |
1156 | ||
1157 | for (i = disks; i--; ) { | |
1158 | struct r5dev *dev = &sh->dev[i]; | |
1159 | struct bio *chosen; | |
91c00924 | 1160 | |
d8ee0728 | 1161 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1162 | struct bio *wbi; |
1163 | ||
b17459c0 | 1164 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1165 | chosen = dev->towrite; |
1166 | dev->towrite = NULL; | |
1167 | BUG_ON(dev->written); | |
1168 | wbi = dev->written = chosen; | |
b17459c0 | 1169 | spin_unlock_irq(&sh->stripe_lock); |
91c00924 DW |
1170 | |
1171 | while (wbi && wbi->bi_sector < | |
1172 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1173 | if (wbi->bi_rw & REQ_FUA) |
1174 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1175 | if (wbi->bi_rw & REQ_SYNC) |
1176 | set_bit(R5_SyncIO, &dev->flags); | |
91c00924 DW |
1177 | tx = async_copy_data(1, wbi, dev->page, |
1178 | dev->sector, tx); | |
1179 | wbi = r5_next_bio(wbi, dev->sector); | |
1180 | } | |
1181 | } | |
1182 | } | |
1183 | ||
1184 | return tx; | |
1185 | } | |
1186 | ||
ac6b53b6 | 1187 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1188 | { |
1189 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1190 | int disks = sh->disks; |
1191 | int pd_idx = sh->pd_idx; | |
1192 | int qd_idx = sh->qd_idx; | |
1193 | int i; | |
bc0934f0 | 1194 | bool fua = false, sync = false; |
91c00924 | 1195 | |
e46b272b | 1196 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1197 | (unsigned long long)sh->sector); |
1198 | ||
bc0934f0 | 1199 | for (i = disks; i--; ) { |
e9c7469b | 1200 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 SL |
1201 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
1202 | } | |
e9c7469b | 1203 | |
91c00924 DW |
1204 | for (i = disks; i--; ) { |
1205 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1206 | |
e9c7469b | 1207 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1208 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1209 | if (fua) |
1210 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1211 | if (sync) |
1212 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1213 | } |
91c00924 DW |
1214 | } |
1215 | ||
d8ee0728 DW |
1216 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1217 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1218 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1219 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1220 | else { | |
1221 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1222 | sh->reconstruct_state = reconstruct_state_result; | |
1223 | } | |
91c00924 DW |
1224 | |
1225 | set_bit(STRIPE_HANDLE, &sh->state); | |
1226 | release_stripe(sh); | |
1227 | } | |
1228 | ||
1229 | static void | |
ac6b53b6 DW |
1230 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1231 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1232 | { |
91c00924 | 1233 | int disks = sh->disks; |
d6f38f31 | 1234 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1235 | struct async_submit_ctl submit; |
91c00924 DW |
1236 | int count = 0, pd_idx = sh->pd_idx, i; |
1237 | struct page *xor_dest; | |
d8ee0728 | 1238 | int prexor = 0; |
91c00924 | 1239 | unsigned long flags; |
91c00924 | 1240 | |
e46b272b | 1241 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1242 | (unsigned long long)sh->sector); |
1243 | ||
1244 | /* check if prexor is active which means only process blocks | |
1245 | * that are part of a read-modify-write (written) | |
1246 | */ | |
d8ee0728 DW |
1247 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1248 | prexor = 1; | |
91c00924 DW |
1249 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1250 | for (i = disks; i--; ) { | |
1251 | struct r5dev *dev = &sh->dev[i]; | |
1252 | if (dev->written) | |
1253 | xor_srcs[count++] = dev->page; | |
1254 | } | |
1255 | } else { | |
1256 | xor_dest = sh->dev[pd_idx].page; | |
1257 | for (i = disks; i--; ) { | |
1258 | struct r5dev *dev = &sh->dev[i]; | |
1259 | if (i != pd_idx) | |
1260 | xor_srcs[count++] = dev->page; | |
1261 | } | |
1262 | } | |
1263 | ||
91c00924 DW |
1264 | /* 1/ if we prexor'd then the dest is reused as a source |
1265 | * 2/ if we did not prexor then we are redoing the parity | |
1266 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1267 | * for the synchronous xor case | |
1268 | */ | |
88ba2aa5 | 1269 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1270 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1271 | ||
1272 | atomic_inc(&sh->count); | |
1273 | ||
ac6b53b6 | 1274 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1275 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1276 | if (unlikely(count == 1)) |
1277 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1278 | else | |
1279 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1280 | } |
1281 | ||
ac6b53b6 DW |
1282 | static void |
1283 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1284 | struct dma_async_tx_descriptor *tx) | |
1285 | { | |
1286 | struct async_submit_ctl submit; | |
1287 | struct page **blocks = percpu->scribble; | |
1288 | int count; | |
1289 | ||
1290 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1291 | ||
1292 | count = set_syndrome_sources(blocks, sh); | |
1293 | ||
1294 | atomic_inc(&sh->count); | |
1295 | ||
1296 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1297 | sh, to_addr_conv(sh, percpu)); | |
1298 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1299 | } |
1300 | ||
1301 | static void ops_complete_check(void *stripe_head_ref) | |
1302 | { | |
1303 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1304 | |
e46b272b | 1305 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1306 | (unsigned long long)sh->sector); |
1307 | ||
ecc65c9b | 1308 | sh->check_state = check_state_check_result; |
91c00924 DW |
1309 | set_bit(STRIPE_HANDLE, &sh->state); |
1310 | release_stripe(sh); | |
1311 | } | |
1312 | ||
ac6b53b6 | 1313 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1314 | { |
91c00924 | 1315 | int disks = sh->disks; |
ac6b53b6 DW |
1316 | int pd_idx = sh->pd_idx; |
1317 | int qd_idx = sh->qd_idx; | |
1318 | struct page *xor_dest; | |
d6f38f31 | 1319 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1320 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1321 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1322 | int count; |
1323 | int i; | |
91c00924 | 1324 | |
e46b272b | 1325 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1326 | (unsigned long long)sh->sector); |
1327 | ||
ac6b53b6 DW |
1328 | count = 0; |
1329 | xor_dest = sh->dev[pd_idx].page; | |
1330 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1331 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1332 | if (i == pd_idx || i == qd_idx) |
1333 | continue; | |
1334 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1335 | } |
1336 | ||
d6f38f31 DW |
1337 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1338 | to_addr_conv(sh, percpu)); | |
099f53cb | 1339 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1340 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1341 | |
91c00924 | 1342 | atomic_inc(&sh->count); |
a08abd8c DW |
1343 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1344 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1345 | } |
1346 | ||
ac6b53b6 DW |
1347 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1348 | { | |
1349 | struct page **srcs = percpu->scribble; | |
1350 | struct async_submit_ctl submit; | |
1351 | int count; | |
1352 | ||
1353 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1354 | (unsigned long long)sh->sector, checkp); | |
1355 | ||
1356 | count = set_syndrome_sources(srcs, sh); | |
1357 | if (!checkp) | |
1358 | srcs[count] = NULL; | |
91c00924 | 1359 | |
91c00924 | 1360 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1361 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1362 | sh, to_addr_conv(sh, percpu)); | |
1363 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1364 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1365 | } |
1366 | ||
417b8d4a | 1367 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1368 | { |
1369 | int overlap_clear = 0, i, disks = sh->disks; | |
1370 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1371 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1372 | int level = conf->level; |
d6f38f31 DW |
1373 | struct raid5_percpu *percpu; |
1374 | unsigned long cpu; | |
91c00924 | 1375 | |
d6f38f31 DW |
1376 | cpu = get_cpu(); |
1377 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1378 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1379 | ops_run_biofill(sh); |
1380 | overlap_clear++; | |
1381 | } | |
1382 | ||
7b3a871e | 1383 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1384 | if (level < 6) |
1385 | tx = ops_run_compute5(sh, percpu); | |
1386 | else { | |
1387 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1388 | tx = ops_run_compute6_1(sh, percpu); | |
1389 | else | |
1390 | tx = ops_run_compute6_2(sh, percpu); | |
1391 | } | |
1392 | /* terminate the chain if reconstruct is not set to be run */ | |
1393 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1394 | async_tx_ack(tx); |
1395 | } | |
91c00924 | 1396 | |
600aa109 | 1397 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1398 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1399 | |
600aa109 | 1400 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1401 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1402 | overlap_clear++; |
1403 | } | |
1404 | ||
ac6b53b6 DW |
1405 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1406 | if (level < 6) | |
1407 | ops_run_reconstruct5(sh, percpu, tx); | |
1408 | else | |
1409 | ops_run_reconstruct6(sh, percpu, tx); | |
1410 | } | |
91c00924 | 1411 | |
ac6b53b6 DW |
1412 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1413 | if (sh->check_state == check_state_run) | |
1414 | ops_run_check_p(sh, percpu); | |
1415 | else if (sh->check_state == check_state_run_q) | |
1416 | ops_run_check_pq(sh, percpu, 0); | |
1417 | else if (sh->check_state == check_state_run_pq) | |
1418 | ops_run_check_pq(sh, percpu, 1); | |
1419 | else | |
1420 | BUG(); | |
1421 | } | |
91c00924 | 1422 | |
91c00924 DW |
1423 | if (overlap_clear) |
1424 | for (i = disks; i--; ) { | |
1425 | struct r5dev *dev = &sh->dev[i]; | |
1426 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1427 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1428 | } | |
d6f38f31 | 1429 | put_cpu(); |
91c00924 DW |
1430 | } |
1431 | ||
417b8d4a DW |
1432 | #ifdef CONFIG_MULTICORE_RAID456 |
1433 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1434 | { | |
1435 | struct stripe_head *sh = param; | |
1436 | unsigned long ops_request = sh->ops.request; | |
1437 | ||
1438 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1439 | wake_up(&sh->ops.wait_for_ops); | |
1440 | ||
1441 | __raid_run_ops(sh, ops_request); | |
1442 | release_stripe(sh); | |
1443 | } | |
1444 | ||
1445 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1446 | { | |
1447 | /* since handle_stripe can be called outside of raid5d context | |
1448 | * we need to ensure sh->ops.request is de-staged before another | |
1449 | * request arrives | |
1450 | */ | |
1451 | wait_event(sh->ops.wait_for_ops, | |
1452 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1453 | sh->ops.request = ops_request; | |
1454 | ||
1455 | atomic_inc(&sh->count); | |
1456 | async_schedule(async_run_ops, sh); | |
1457 | } | |
1458 | #else | |
1459 | #define raid_run_ops __raid_run_ops | |
1460 | #endif | |
1461 | ||
d1688a6d | 1462 | static int grow_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1463 | { |
1464 | struct stripe_head *sh; | |
6ce32846 | 1465 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1466 | if (!sh) |
1467 | return 0; | |
6ce32846 | 1468 | |
3f294f4f | 1469 | sh->raid_conf = conf; |
417b8d4a DW |
1470 | #ifdef CONFIG_MULTICORE_RAID456 |
1471 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1472 | #endif | |
3f294f4f | 1473 | |
b17459c0 SL |
1474 | spin_lock_init(&sh->stripe_lock); |
1475 | ||
e4e11e38 N |
1476 | if (grow_buffers(sh)) { |
1477 | shrink_buffers(sh); | |
3f294f4f N |
1478 | kmem_cache_free(conf->slab_cache, sh); |
1479 | return 0; | |
1480 | } | |
1481 | /* we just created an active stripe so... */ | |
1482 | atomic_set(&sh->count, 1); | |
1483 | atomic_inc(&conf->active_stripes); | |
1484 | INIT_LIST_HEAD(&sh->lru); | |
1485 | release_stripe(sh); | |
1486 | return 1; | |
1487 | } | |
1488 | ||
d1688a6d | 1489 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 1490 | { |
e18b890b | 1491 | struct kmem_cache *sc; |
5e5e3e78 | 1492 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1493 | |
f4be6b43 N |
1494 | if (conf->mddev->gendisk) |
1495 | sprintf(conf->cache_name[0], | |
1496 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1497 | else | |
1498 | sprintf(conf->cache_name[0], | |
1499 | "raid%d-%p", conf->level, conf->mddev); | |
1500 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1501 | ||
ad01c9e3 N |
1502 | conf->active_name = 0; |
1503 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1504 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1505 | 0, 0, NULL); |
1da177e4 LT |
1506 | if (!sc) |
1507 | return 1; | |
1508 | conf->slab_cache = sc; | |
ad01c9e3 | 1509 | conf->pool_size = devs; |
16a53ecc | 1510 | while (num--) |
3f294f4f | 1511 | if (!grow_one_stripe(conf)) |
1da177e4 | 1512 | return 1; |
1da177e4 LT |
1513 | return 0; |
1514 | } | |
29269553 | 1515 | |
d6f38f31 DW |
1516 | /** |
1517 | * scribble_len - return the required size of the scribble region | |
1518 | * @num - total number of disks in the array | |
1519 | * | |
1520 | * The size must be enough to contain: | |
1521 | * 1/ a struct page pointer for each device in the array +2 | |
1522 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1523 | * (dma_map_page()) or page (page_address()) address. | |
1524 | * | |
1525 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1526 | * calculate over all devices (not just the data blocks), using zeros in place | |
1527 | * of the P and Q blocks. | |
1528 | */ | |
1529 | static size_t scribble_len(int num) | |
1530 | { | |
1531 | size_t len; | |
1532 | ||
1533 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1534 | ||
1535 | return len; | |
1536 | } | |
1537 | ||
d1688a6d | 1538 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
1539 | { |
1540 | /* Make all the stripes able to hold 'newsize' devices. | |
1541 | * New slots in each stripe get 'page' set to a new page. | |
1542 | * | |
1543 | * This happens in stages: | |
1544 | * 1/ create a new kmem_cache and allocate the required number of | |
1545 | * stripe_heads. | |
1546 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1547 | * to the new stripe_heads. This will have the side effect of | |
1548 | * freezing the array as once all stripe_heads have been collected, | |
1549 | * no IO will be possible. Old stripe heads are freed once their | |
1550 | * pages have been transferred over, and the old kmem_cache is | |
1551 | * freed when all stripes are done. | |
1552 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1553 | * we simple return a failre status - no need to clean anything up. | |
1554 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1555 | * If this fails, we don't bother trying the shrink the | |
1556 | * stripe_heads down again, we just leave them as they are. | |
1557 | * As each stripe_head is processed the new one is released into | |
1558 | * active service. | |
1559 | * | |
1560 | * Once step2 is started, we cannot afford to wait for a write, | |
1561 | * so we use GFP_NOIO allocations. | |
1562 | */ | |
1563 | struct stripe_head *osh, *nsh; | |
1564 | LIST_HEAD(newstripes); | |
1565 | struct disk_info *ndisks; | |
d6f38f31 | 1566 | unsigned long cpu; |
b5470dc5 | 1567 | int err; |
e18b890b | 1568 | struct kmem_cache *sc; |
ad01c9e3 N |
1569 | int i; |
1570 | ||
1571 | if (newsize <= conf->pool_size) | |
1572 | return 0; /* never bother to shrink */ | |
1573 | ||
b5470dc5 DW |
1574 | err = md_allow_write(conf->mddev); |
1575 | if (err) | |
1576 | return err; | |
2a2275d6 | 1577 | |
ad01c9e3 N |
1578 | /* Step 1 */ |
1579 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1580 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1581 | 0, 0, NULL); |
ad01c9e3 N |
1582 | if (!sc) |
1583 | return -ENOMEM; | |
1584 | ||
1585 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1586 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1587 | if (!nsh) |
1588 | break; | |
1589 | ||
ad01c9e3 | 1590 | nsh->raid_conf = conf; |
417b8d4a DW |
1591 | #ifdef CONFIG_MULTICORE_RAID456 |
1592 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1593 | #endif | |
cb13ff69 | 1594 | spin_lock_init(&nsh->stripe_lock); |
ad01c9e3 N |
1595 | |
1596 | list_add(&nsh->lru, &newstripes); | |
1597 | } | |
1598 | if (i) { | |
1599 | /* didn't get enough, give up */ | |
1600 | while (!list_empty(&newstripes)) { | |
1601 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1602 | list_del(&nsh->lru); | |
1603 | kmem_cache_free(sc, nsh); | |
1604 | } | |
1605 | kmem_cache_destroy(sc); | |
1606 | return -ENOMEM; | |
1607 | } | |
1608 | /* Step 2 - Must use GFP_NOIO now. | |
1609 | * OK, we have enough stripes, start collecting inactive | |
1610 | * stripes and copying them over | |
1611 | */ | |
1612 | list_for_each_entry(nsh, &newstripes, lru) { | |
1613 | spin_lock_irq(&conf->device_lock); | |
1614 | wait_event_lock_irq(conf->wait_for_stripe, | |
1615 | !list_empty(&conf->inactive_list), | |
1616 | conf->device_lock, | |
482c0834 | 1617 | ); |
ad01c9e3 N |
1618 | osh = get_free_stripe(conf); |
1619 | spin_unlock_irq(&conf->device_lock); | |
1620 | atomic_set(&nsh->count, 1); | |
1621 | for(i=0; i<conf->pool_size; i++) | |
1622 | nsh->dev[i].page = osh->dev[i].page; | |
1623 | for( ; i<newsize; i++) | |
1624 | nsh->dev[i].page = NULL; | |
1625 | kmem_cache_free(conf->slab_cache, osh); | |
1626 | } | |
1627 | kmem_cache_destroy(conf->slab_cache); | |
1628 | ||
1629 | /* Step 3. | |
1630 | * At this point, we are holding all the stripes so the array | |
1631 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1632 | * conf->disks and the scribble region |
ad01c9e3 N |
1633 | */ |
1634 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1635 | if (ndisks) { | |
1636 | for (i=0; i<conf->raid_disks; i++) | |
1637 | ndisks[i] = conf->disks[i]; | |
1638 | kfree(conf->disks); | |
1639 | conf->disks = ndisks; | |
1640 | } else | |
1641 | err = -ENOMEM; | |
1642 | ||
d6f38f31 DW |
1643 | get_online_cpus(); |
1644 | conf->scribble_len = scribble_len(newsize); | |
1645 | for_each_present_cpu(cpu) { | |
1646 | struct raid5_percpu *percpu; | |
1647 | void *scribble; | |
1648 | ||
1649 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1650 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1651 | ||
1652 | if (scribble) { | |
1653 | kfree(percpu->scribble); | |
1654 | percpu->scribble = scribble; | |
1655 | } else { | |
1656 | err = -ENOMEM; | |
1657 | break; | |
1658 | } | |
1659 | } | |
1660 | put_online_cpus(); | |
1661 | ||
ad01c9e3 N |
1662 | /* Step 4, return new stripes to service */ |
1663 | while(!list_empty(&newstripes)) { | |
1664 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1665 | list_del_init(&nsh->lru); | |
d6f38f31 | 1666 | |
ad01c9e3 N |
1667 | for (i=conf->raid_disks; i < newsize; i++) |
1668 | if (nsh->dev[i].page == NULL) { | |
1669 | struct page *p = alloc_page(GFP_NOIO); | |
1670 | nsh->dev[i].page = p; | |
1671 | if (!p) | |
1672 | err = -ENOMEM; | |
1673 | } | |
1674 | release_stripe(nsh); | |
1675 | } | |
1676 | /* critical section pass, GFP_NOIO no longer needed */ | |
1677 | ||
1678 | conf->slab_cache = sc; | |
1679 | conf->active_name = 1-conf->active_name; | |
1680 | conf->pool_size = newsize; | |
1681 | return err; | |
1682 | } | |
1da177e4 | 1683 | |
d1688a6d | 1684 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1685 | { |
1686 | struct stripe_head *sh; | |
1687 | ||
3f294f4f N |
1688 | spin_lock_irq(&conf->device_lock); |
1689 | sh = get_free_stripe(conf); | |
1690 | spin_unlock_irq(&conf->device_lock); | |
1691 | if (!sh) | |
1692 | return 0; | |
78bafebd | 1693 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1694 | shrink_buffers(sh); |
3f294f4f N |
1695 | kmem_cache_free(conf->slab_cache, sh); |
1696 | atomic_dec(&conf->active_stripes); | |
1697 | return 1; | |
1698 | } | |
1699 | ||
d1688a6d | 1700 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f N |
1701 | { |
1702 | while (drop_one_stripe(conf)) | |
1703 | ; | |
1704 | ||
29fc7e3e N |
1705 | if (conf->slab_cache) |
1706 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1707 | conf->slab_cache = NULL; |
1708 | } | |
1709 | ||
6712ecf8 | 1710 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1711 | { |
99c0fb5f | 1712 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1713 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1714 | int disks = sh->disks, i; |
1da177e4 | 1715 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 1716 | char b[BDEVNAME_SIZE]; |
dd054fce | 1717 | struct md_rdev *rdev = NULL; |
05616be5 | 1718 | sector_t s; |
1da177e4 LT |
1719 | |
1720 | for (i=0 ; i<disks; i++) | |
1721 | if (bi == &sh->dev[i].req) | |
1722 | break; | |
1723 | ||
45b4233c DW |
1724 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1725 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1726 | uptodate); |
1727 | if (i == disks) { | |
1728 | BUG(); | |
6712ecf8 | 1729 | return; |
1da177e4 | 1730 | } |
14a75d3e | 1731 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
1732 | /* If replacement finished while this request was outstanding, |
1733 | * 'replacement' might be NULL already. | |
1734 | * In that case it moved down to 'rdev'. | |
1735 | * rdev is not removed until all requests are finished. | |
1736 | */ | |
14a75d3e | 1737 | rdev = conf->disks[i].replacement; |
dd054fce | 1738 | if (!rdev) |
14a75d3e | 1739 | rdev = conf->disks[i].rdev; |
1da177e4 | 1740 | |
05616be5 N |
1741 | if (use_new_offset(conf, sh)) |
1742 | s = sh->sector + rdev->new_data_offset; | |
1743 | else | |
1744 | s = sh->sector + rdev->data_offset; | |
1da177e4 | 1745 | if (uptodate) { |
1da177e4 | 1746 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1747 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
1748 | /* Note that this cannot happen on a |
1749 | * replacement device. We just fail those on | |
1750 | * any error | |
1751 | */ | |
8bda470e CD |
1752 | printk_ratelimited( |
1753 | KERN_INFO | |
1754 | "md/raid:%s: read error corrected" | |
1755 | " (%lu sectors at %llu on %s)\n", | |
1756 | mdname(conf->mddev), STRIPE_SECTORS, | |
05616be5 | 1757 | (unsigned long long)s, |
8bda470e | 1758 | bdevname(rdev->bdev, b)); |
ddd5115f | 1759 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
1760 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1761 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 1762 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
1763 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
1764 | ||
14a75d3e N |
1765 | if (atomic_read(&rdev->read_errors)) |
1766 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 1767 | } else { |
14a75d3e | 1768 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 1769 | int retry = 0; |
2e8ac303 | 1770 | int set_bad = 0; |
d6950432 | 1771 | |
1da177e4 | 1772 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1773 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
1774 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
1775 | printk_ratelimited( | |
1776 | KERN_WARNING | |
1777 | "md/raid:%s: read error on replacement device " | |
1778 | "(sector %llu on %s).\n", | |
1779 | mdname(conf->mddev), | |
05616be5 | 1780 | (unsigned long long)s, |
14a75d3e | 1781 | bdn); |
2e8ac303 | 1782 | else if (conf->mddev->degraded >= conf->max_degraded) { |
1783 | set_bad = 1; | |
8bda470e CD |
1784 | printk_ratelimited( |
1785 | KERN_WARNING | |
1786 | "md/raid:%s: read error not correctable " | |
1787 | "(sector %llu on %s).\n", | |
1788 | mdname(conf->mddev), | |
05616be5 | 1789 | (unsigned long long)s, |
8bda470e | 1790 | bdn); |
2e8ac303 | 1791 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 1792 | /* Oh, no!!! */ |
2e8ac303 | 1793 | set_bad = 1; |
8bda470e CD |
1794 | printk_ratelimited( |
1795 | KERN_WARNING | |
1796 | "md/raid:%s: read error NOT corrected!! " | |
1797 | "(sector %llu on %s).\n", | |
1798 | mdname(conf->mddev), | |
05616be5 | 1799 | (unsigned long long)s, |
8bda470e | 1800 | bdn); |
2e8ac303 | 1801 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1802 | > conf->max_nr_stripes) |
14f8d26b | 1803 | printk(KERN_WARNING |
0c55e022 | 1804 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1805 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1806 | else |
1807 | retry = 1; | |
1808 | if (retry) | |
3f9e7c14 | 1809 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
1810 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1811 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
1812 | } else | |
1813 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 1814 | else { |
4e5314b5 N |
1815 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1816 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 1817 | if (!(set_bad |
1818 | && test_bit(In_sync, &rdev->flags) | |
1819 | && rdev_set_badblocks( | |
1820 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
1821 | md_error(conf->mddev, rdev); | |
ba22dcbf | 1822 | } |
1da177e4 | 1823 | } |
14a75d3e | 1824 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
1825 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1826 | set_bit(STRIPE_HANDLE, &sh->state); | |
1827 | release_stripe(sh); | |
1da177e4 LT |
1828 | } |
1829 | ||
d710e138 | 1830 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1831 | { |
99c0fb5f | 1832 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1833 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1834 | int disks = sh->disks, i; |
977df362 | 1835 | struct md_rdev *uninitialized_var(rdev); |
1da177e4 | 1836 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
1837 | sector_t first_bad; |
1838 | int bad_sectors; | |
977df362 | 1839 | int replacement = 0; |
1da177e4 | 1840 | |
977df362 N |
1841 | for (i = 0 ; i < disks; i++) { |
1842 | if (bi == &sh->dev[i].req) { | |
1843 | rdev = conf->disks[i].rdev; | |
1da177e4 | 1844 | break; |
977df362 N |
1845 | } |
1846 | if (bi == &sh->dev[i].rreq) { | |
1847 | rdev = conf->disks[i].replacement; | |
dd054fce N |
1848 | if (rdev) |
1849 | replacement = 1; | |
1850 | else | |
1851 | /* rdev was removed and 'replacement' | |
1852 | * replaced it. rdev is not removed | |
1853 | * until all requests are finished. | |
1854 | */ | |
1855 | rdev = conf->disks[i].rdev; | |
977df362 N |
1856 | break; |
1857 | } | |
1858 | } | |
45b4233c | 1859 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1860 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1861 | uptodate); | |
1862 | if (i == disks) { | |
1863 | BUG(); | |
6712ecf8 | 1864 | return; |
1da177e4 LT |
1865 | } |
1866 | ||
977df362 N |
1867 | if (replacement) { |
1868 | if (!uptodate) | |
1869 | md_error(conf->mddev, rdev); | |
1870 | else if (is_badblock(rdev, sh->sector, | |
1871 | STRIPE_SECTORS, | |
1872 | &first_bad, &bad_sectors)) | |
1873 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
1874 | } else { | |
1875 | if (!uptodate) { | |
1876 | set_bit(WriteErrorSeen, &rdev->flags); | |
1877 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
1878 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1879 | set_bit(MD_RECOVERY_NEEDED, | |
1880 | &rdev->mddev->recovery); | |
977df362 N |
1881 | } else if (is_badblock(rdev, sh->sector, |
1882 | STRIPE_SECTORS, | |
1883 | &first_bad, &bad_sectors)) | |
1884 | set_bit(R5_MadeGood, &sh->dev[i].flags); | |
1885 | } | |
1886 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 1887 | |
977df362 N |
1888 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
1889 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 1890 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 1891 | release_stripe(sh); |
1da177e4 LT |
1892 | } |
1893 | ||
784052ec | 1894 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1895 | |
784052ec | 1896 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1897 | { |
1898 | struct r5dev *dev = &sh->dev[i]; | |
1899 | ||
1900 | bio_init(&dev->req); | |
1901 | dev->req.bi_io_vec = &dev->vec; | |
1902 | dev->req.bi_vcnt++; | |
1903 | dev->req.bi_max_vecs++; | |
1da177e4 | 1904 | dev->req.bi_private = sh; |
995c4275 | 1905 | dev->vec.bv_page = dev->page; |
1da177e4 | 1906 | |
977df362 N |
1907 | bio_init(&dev->rreq); |
1908 | dev->rreq.bi_io_vec = &dev->rvec; | |
1909 | dev->rreq.bi_vcnt++; | |
1910 | dev->rreq.bi_max_vecs++; | |
1911 | dev->rreq.bi_private = sh; | |
1912 | dev->rvec.bv_page = dev->page; | |
1913 | ||
1da177e4 | 1914 | dev->flags = 0; |
784052ec | 1915 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1916 | } |
1917 | ||
fd01b88c | 1918 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1919 | { |
1920 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 1921 | struct r5conf *conf = mddev->private; |
908f4fbd | 1922 | unsigned long flags; |
0c55e022 | 1923 | pr_debug("raid456: error called\n"); |
1da177e4 | 1924 | |
908f4fbd N |
1925 | spin_lock_irqsave(&conf->device_lock, flags); |
1926 | clear_bit(In_sync, &rdev->flags); | |
1927 | mddev->degraded = calc_degraded(conf); | |
1928 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1929 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1930 | ||
de393cde | 1931 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
1932 | set_bit(Faulty, &rdev->flags); |
1933 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1934 | printk(KERN_ALERT | |
1935 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1936 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1937 | mdname(mddev), | |
1938 | bdevname(rdev->bdev, b), | |
1939 | mdname(mddev), | |
1940 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1941 | } |
1da177e4 LT |
1942 | |
1943 | /* | |
1944 | * Input: a 'big' sector number, | |
1945 | * Output: index of the data and parity disk, and the sector # in them. | |
1946 | */ | |
d1688a6d | 1947 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
1948 | int previous, int *dd_idx, |
1949 | struct stripe_head *sh) | |
1da177e4 | 1950 | { |
6e3b96ed | 1951 | sector_t stripe, stripe2; |
35f2a591 | 1952 | sector_t chunk_number; |
1da177e4 | 1953 | unsigned int chunk_offset; |
911d4ee8 | 1954 | int pd_idx, qd_idx; |
67cc2b81 | 1955 | int ddf_layout = 0; |
1da177e4 | 1956 | sector_t new_sector; |
e183eaed N |
1957 | int algorithm = previous ? conf->prev_algo |
1958 | : conf->algorithm; | |
09c9e5fa AN |
1959 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1960 | : conf->chunk_sectors; | |
112bf897 N |
1961 | int raid_disks = previous ? conf->previous_raid_disks |
1962 | : conf->raid_disks; | |
1963 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1964 | |
1965 | /* First compute the information on this sector */ | |
1966 | ||
1967 | /* | |
1968 | * Compute the chunk number and the sector offset inside the chunk | |
1969 | */ | |
1970 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1971 | chunk_number = r_sector; | |
1da177e4 LT |
1972 | |
1973 | /* | |
1974 | * Compute the stripe number | |
1975 | */ | |
35f2a591 N |
1976 | stripe = chunk_number; |
1977 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1978 | stripe2 = stripe; |
1da177e4 LT |
1979 | /* |
1980 | * Select the parity disk based on the user selected algorithm. | |
1981 | */ | |
84789554 | 1982 | pd_idx = qd_idx = -1; |
16a53ecc N |
1983 | switch(conf->level) { |
1984 | case 4: | |
911d4ee8 | 1985 | pd_idx = data_disks; |
16a53ecc N |
1986 | break; |
1987 | case 5: | |
e183eaed | 1988 | switch (algorithm) { |
1da177e4 | 1989 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1990 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1991 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1992 | (*dd_idx)++; |
1993 | break; | |
1994 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1995 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1996 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1997 | (*dd_idx)++; |
1998 | break; | |
1999 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2000 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2001 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2002 | break; |
2003 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2004 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2005 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2006 | break; |
99c0fb5f N |
2007 | case ALGORITHM_PARITY_0: |
2008 | pd_idx = 0; | |
2009 | (*dd_idx)++; | |
2010 | break; | |
2011 | case ALGORITHM_PARITY_N: | |
2012 | pd_idx = data_disks; | |
2013 | break; | |
1da177e4 | 2014 | default: |
99c0fb5f | 2015 | BUG(); |
16a53ecc N |
2016 | } |
2017 | break; | |
2018 | case 6: | |
2019 | ||
e183eaed | 2020 | switch (algorithm) { |
16a53ecc | 2021 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2022 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2023 | qd_idx = pd_idx + 1; |
2024 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2025 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2026 | qd_idx = 0; |
2027 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2028 | (*dd_idx) += 2; /* D D P Q D */ |
2029 | break; | |
2030 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2031 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2032 | qd_idx = pd_idx + 1; |
2033 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2034 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2035 | qd_idx = 0; |
2036 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2037 | (*dd_idx) += 2; /* D D P Q D */ |
2038 | break; | |
2039 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2040 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2041 | qd_idx = (pd_idx + 1) % raid_disks; |
2042 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2043 | break; |
2044 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2045 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2046 | qd_idx = (pd_idx + 1) % raid_disks; |
2047 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2048 | break; |
99c0fb5f N |
2049 | |
2050 | case ALGORITHM_PARITY_0: | |
2051 | pd_idx = 0; | |
2052 | qd_idx = 1; | |
2053 | (*dd_idx) += 2; | |
2054 | break; | |
2055 | case ALGORITHM_PARITY_N: | |
2056 | pd_idx = data_disks; | |
2057 | qd_idx = data_disks + 1; | |
2058 | break; | |
2059 | ||
2060 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2061 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2062 | * of blocks for computing Q is different. | |
2063 | */ | |
6e3b96ed | 2064 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2065 | qd_idx = pd_idx + 1; |
2066 | if (pd_idx == raid_disks-1) { | |
2067 | (*dd_idx)++; /* Q D D D P */ | |
2068 | qd_idx = 0; | |
2069 | } else if (*dd_idx >= pd_idx) | |
2070 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2071 | ddf_layout = 1; |
99c0fb5f N |
2072 | break; |
2073 | ||
2074 | case ALGORITHM_ROTATING_N_RESTART: | |
2075 | /* Same a left_asymmetric, by first stripe is | |
2076 | * D D D P Q rather than | |
2077 | * Q D D D P | |
2078 | */ | |
6e3b96ed N |
2079 | stripe2 += 1; |
2080 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2081 | qd_idx = pd_idx + 1; |
2082 | if (pd_idx == raid_disks-1) { | |
2083 | (*dd_idx)++; /* Q D D D P */ | |
2084 | qd_idx = 0; | |
2085 | } else if (*dd_idx >= pd_idx) | |
2086 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2087 | ddf_layout = 1; |
99c0fb5f N |
2088 | break; |
2089 | ||
2090 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2091 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2092 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2093 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2094 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2095 | ddf_layout = 1; |
99c0fb5f N |
2096 | break; |
2097 | ||
2098 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2099 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2100 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2101 | if (*dd_idx >= pd_idx) |
2102 | (*dd_idx)++; | |
2103 | qd_idx = raid_disks - 1; | |
2104 | break; | |
2105 | ||
2106 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2107 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2108 | if (*dd_idx >= pd_idx) |
2109 | (*dd_idx)++; | |
2110 | qd_idx = raid_disks - 1; | |
2111 | break; | |
2112 | ||
2113 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2114 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2115 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2116 | qd_idx = raid_disks - 1; | |
2117 | break; | |
2118 | ||
2119 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2120 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2121 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2122 | qd_idx = raid_disks - 1; | |
2123 | break; | |
2124 | ||
2125 | case ALGORITHM_PARITY_0_6: | |
2126 | pd_idx = 0; | |
2127 | (*dd_idx)++; | |
2128 | qd_idx = raid_disks - 1; | |
2129 | break; | |
2130 | ||
16a53ecc | 2131 | default: |
99c0fb5f | 2132 | BUG(); |
16a53ecc N |
2133 | } |
2134 | break; | |
1da177e4 LT |
2135 | } |
2136 | ||
911d4ee8 N |
2137 | if (sh) { |
2138 | sh->pd_idx = pd_idx; | |
2139 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2140 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2141 | } |
1da177e4 LT |
2142 | /* |
2143 | * Finally, compute the new sector number | |
2144 | */ | |
2145 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2146 | return new_sector; | |
2147 | } | |
2148 | ||
2149 | ||
784052ec | 2150 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2151 | { |
d1688a6d | 2152 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2153 | int raid_disks = sh->disks; |
2154 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2155 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2156 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2157 | : conf->chunk_sectors; | |
e183eaed N |
2158 | int algorithm = previous ? conf->prev_algo |
2159 | : conf->algorithm; | |
1da177e4 LT |
2160 | sector_t stripe; |
2161 | int chunk_offset; | |
35f2a591 N |
2162 | sector_t chunk_number; |
2163 | int dummy1, dd_idx = i; | |
1da177e4 | 2164 | sector_t r_sector; |
911d4ee8 | 2165 | struct stripe_head sh2; |
1da177e4 | 2166 | |
16a53ecc | 2167 | |
1da177e4 LT |
2168 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
2169 | stripe = new_sector; | |
1da177e4 | 2170 | |
16a53ecc N |
2171 | if (i == sh->pd_idx) |
2172 | return 0; | |
2173 | switch(conf->level) { | |
2174 | case 4: break; | |
2175 | case 5: | |
e183eaed | 2176 | switch (algorithm) { |
1da177e4 LT |
2177 | case ALGORITHM_LEFT_ASYMMETRIC: |
2178 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2179 | if (i > sh->pd_idx) | |
2180 | i--; | |
2181 | break; | |
2182 | case ALGORITHM_LEFT_SYMMETRIC: | |
2183 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2184 | if (i < sh->pd_idx) | |
2185 | i += raid_disks; | |
2186 | i -= (sh->pd_idx + 1); | |
2187 | break; | |
99c0fb5f N |
2188 | case ALGORITHM_PARITY_0: |
2189 | i -= 1; | |
2190 | break; | |
2191 | case ALGORITHM_PARITY_N: | |
2192 | break; | |
1da177e4 | 2193 | default: |
99c0fb5f | 2194 | BUG(); |
16a53ecc N |
2195 | } |
2196 | break; | |
2197 | case 6: | |
d0dabf7e | 2198 | if (i == sh->qd_idx) |
16a53ecc | 2199 | return 0; /* It is the Q disk */ |
e183eaed | 2200 | switch (algorithm) { |
16a53ecc N |
2201 | case ALGORITHM_LEFT_ASYMMETRIC: |
2202 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2203 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2204 | case ALGORITHM_ROTATING_N_RESTART: | |
2205 | if (sh->pd_idx == raid_disks-1) | |
2206 | i--; /* Q D D D P */ | |
16a53ecc N |
2207 | else if (i > sh->pd_idx) |
2208 | i -= 2; /* D D P Q D */ | |
2209 | break; | |
2210 | case ALGORITHM_LEFT_SYMMETRIC: | |
2211 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2212 | if (sh->pd_idx == raid_disks-1) | |
2213 | i--; /* Q D D D P */ | |
2214 | else { | |
2215 | /* D D P Q D */ | |
2216 | if (i < sh->pd_idx) | |
2217 | i += raid_disks; | |
2218 | i -= (sh->pd_idx + 2); | |
2219 | } | |
2220 | break; | |
99c0fb5f N |
2221 | case ALGORITHM_PARITY_0: |
2222 | i -= 2; | |
2223 | break; | |
2224 | case ALGORITHM_PARITY_N: | |
2225 | break; | |
2226 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2227 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2228 | if (sh->pd_idx == 0) |
2229 | i--; /* P D D D Q */ | |
e4424fee N |
2230 | else { |
2231 | /* D D Q P D */ | |
2232 | if (i < sh->pd_idx) | |
2233 | i += raid_disks; | |
2234 | i -= (sh->pd_idx + 1); | |
2235 | } | |
99c0fb5f N |
2236 | break; |
2237 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2238 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2239 | if (i > sh->pd_idx) | |
2240 | i--; | |
2241 | break; | |
2242 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2243 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2244 | if (i < sh->pd_idx) | |
2245 | i += data_disks + 1; | |
2246 | i -= (sh->pd_idx + 1); | |
2247 | break; | |
2248 | case ALGORITHM_PARITY_0_6: | |
2249 | i -= 1; | |
2250 | break; | |
16a53ecc | 2251 | default: |
99c0fb5f | 2252 | BUG(); |
16a53ecc N |
2253 | } |
2254 | break; | |
1da177e4 LT |
2255 | } |
2256 | ||
2257 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2258 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2259 | |
112bf897 | 2260 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2261 | previous, &dummy1, &sh2); |
911d4ee8 N |
2262 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2263 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2264 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2265 | mdname(conf->mddev)); | |
1da177e4 LT |
2266 | return 0; |
2267 | } | |
2268 | return r_sector; | |
2269 | } | |
2270 | ||
2271 | ||
600aa109 | 2272 | static void |
c0f7bddb | 2273 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2274 | int rcw, int expand) |
e33129d8 DW |
2275 | { |
2276 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
d1688a6d | 2277 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2278 | int level = conf->level; |
e33129d8 DW |
2279 | |
2280 | if (rcw) { | |
2281 | /* if we are not expanding this is a proper write request, and | |
2282 | * there will be bios with new data to be drained into the | |
2283 | * stripe cache | |
2284 | */ | |
2285 | if (!expand) { | |
600aa109 DW |
2286 | sh->reconstruct_state = reconstruct_state_drain_run; |
2287 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2288 | } else | |
2289 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2290 | |
ac6b53b6 | 2291 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2292 | |
2293 | for (i = disks; i--; ) { | |
2294 | struct r5dev *dev = &sh->dev[i]; | |
2295 | ||
2296 | if (dev->towrite) { | |
2297 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2298 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2299 | if (!expand) |
2300 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2301 | s->locked++; |
e33129d8 DW |
2302 | } |
2303 | } | |
c0f7bddb | 2304 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2305 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2306 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2307 | } else { |
c0f7bddb | 2308 | BUG_ON(level == 6); |
e33129d8 DW |
2309 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2310 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2311 | ||
d8ee0728 | 2312 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2313 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2314 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2315 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2316 | |
2317 | for (i = disks; i--; ) { | |
2318 | struct r5dev *dev = &sh->dev[i]; | |
2319 | if (i == pd_idx) | |
2320 | continue; | |
2321 | ||
e33129d8 DW |
2322 | if (dev->towrite && |
2323 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2324 | test_bit(R5_Wantcompute, &dev->flags))) { |
2325 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2326 | set_bit(R5_LOCKED, &dev->flags); |
2327 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2328 | s->locked++; |
e33129d8 DW |
2329 | } |
2330 | } | |
2331 | } | |
2332 | ||
c0f7bddb | 2333 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2334 | * are in flight |
2335 | */ | |
2336 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2337 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2338 | s->locked++; |
e33129d8 | 2339 | |
c0f7bddb YT |
2340 | if (level == 6) { |
2341 | int qd_idx = sh->qd_idx; | |
2342 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2343 | ||
2344 | set_bit(R5_LOCKED, &dev->flags); | |
2345 | clear_bit(R5_UPTODATE, &dev->flags); | |
2346 | s->locked++; | |
2347 | } | |
2348 | ||
600aa109 | 2349 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2350 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2351 | s->locked, s->ops_request); |
e33129d8 | 2352 | } |
16a53ecc | 2353 | |
1da177e4 LT |
2354 | /* |
2355 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2356 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2357 | * The bi_next chain must be in order. |
2358 | */ | |
2359 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2360 | { | |
2361 | struct bio **bip; | |
d1688a6d | 2362 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2363 | int firstwrite=0; |
1da177e4 | 2364 | |
cbe47ec5 | 2365 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2366 | (unsigned long long)bi->bi_sector, |
2367 | (unsigned long long)sh->sector); | |
2368 | ||
b17459c0 SL |
2369 | /* |
2370 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
2371 | * reference count to avoid race. The reference count should already be | |
2372 | * increased before this function is called (for example, in | |
2373 | * make_request()), so other bio sharing this stripe will not free the | |
2374 | * stripe. If a stripe is owned by one stripe, the stripe lock will | |
2375 | * protect it. | |
2376 | */ | |
2377 | spin_lock_irq(&sh->stripe_lock); | |
72626685 | 2378 | if (forwrite) { |
1da177e4 | 2379 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 2380 | if (*bip == NULL) |
72626685 N |
2381 | firstwrite = 1; |
2382 | } else | |
1da177e4 LT |
2383 | bip = &sh->dev[dd_idx].toread; |
2384 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2385 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2386 | goto overlap; | |
2387 | bip = & (*bip)->bi_next; | |
2388 | } | |
2389 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2390 | goto overlap; | |
2391 | ||
78bafebd | 2392 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2393 | if (*bip) |
2394 | bi->bi_next = *bip; | |
2395 | *bip = bi; | |
e7836bd6 | 2396 | raid5_inc_bi_active_stripes(bi); |
72626685 | 2397 | |
1da177e4 LT |
2398 | if (forwrite) { |
2399 | /* check if page is covered */ | |
2400 | sector_t sector = sh->dev[dd_idx].sector; | |
2401 | for (bi=sh->dev[dd_idx].towrite; | |
2402 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2403 | bi && bi->bi_sector <= sector; | |
2404 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2405 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2406 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2407 | } | |
2408 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2409 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2410 | } | |
b17459c0 | 2411 | spin_unlock_irq(&sh->stripe_lock); |
cbe47ec5 N |
2412 | |
2413 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2414 | (unsigned long long)(*bip)->bi_sector, | |
2415 | (unsigned long long)sh->sector, dd_idx); | |
2416 | ||
2417 | if (conf->mddev->bitmap && firstwrite) { | |
2418 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2419 | STRIPE_SECTORS, 0); | |
2420 | sh->bm_seq = conf->seq_flush+1; | |
2421 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2422 | } | |
1da177e4 LT |
2423 | return 1; |
2424 | ||
2425 | overlap: | |
2426 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 2427 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
2428 | return 0; |
2429 | } | |
2430 | ||
d1688a6d | 2431 | static void end_reshape(struct r5conf *conf); |
29269553 | 2432 | |
d1688a6d | 2433 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 2434 | struct stripe_head *sh) |
ccfcc3c1 | 2435 | { |
784052ec | 2436 | int sectors_per_chunk = |
09c9e5fa | 2437 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2438 | int dd_idx; |
2d2063ce | 2439 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2440 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2441 | |
112bf897 N |
2442 | raid5_compute_sector(conf, |
2443 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2444 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2445 | previous, |
911d4ee8 | 2446 | &dd_idx, sh); |
ccfcc3c1 N |
2447 | } |
2448 | ||
a4456856 | 2449 | static void |
d1688a6d | 2450 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2451 | struct stripe_head_state *s, int disks, |
2452 | struct bio **return_bi) | |
2453 | { | |
2454 | int i; | |
2455 | for (i = disks; i--; ) { | |
2456 | struct bio *bi; | |
2457 | int bitmap_end = 0; | |
2458 | ||
2459 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 2460 | struct md_rdev *rdev; |
a4456856 DW |
2461 | rcu_read_lock(); |
2462 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2463 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
2464 | atomic_inc(&rdev->nr_pending); |
2465 | else | |
2466 | rdev = NULL; | |
a4456856 | 2467 | rcu_read_unlock(); |
7f0da59b N |
2468 | if (rdev) { |
2469 | if (!rdev_set_badblocks( | |
2470 | rdev, | |
2471 | sh->sector, | |
2472 | STRIPE_SECTORS, 0)) | |
2473 | md_error(conf->mddev, rdev); | |
2474 | rdev_dec_pending(rdev, conf->mddev); | |
2475 | } | |
a4456856 | 2476 | } |
b17459c0 | 2477 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
2478 | /* fail all writes first */ |
2479 | bi = sh->dev[i].towrite; | |
2480 | sh->dev[i].towrite = NULL; | |
b17459c0 | 2481 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
2482 | if (bi) { |
2483 | s->to_write--; | |
2484 | bitmap_end = 1; | |
2485 | } | |
2486 | ||
2487 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2488 | wake_up(&conf->wait_for_overlap); | |
2489 | ||
2490 | while (bi && bi->bi_sector < | |
2491 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2492 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2493 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2494 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2495 | md_write_end(conf->mddev); |
2496 | bi->bi_next = *return_bi; | |
2497 | *return_bi = bi; | |
2498 | } | |
2499 | bi = nextbi; | |
2500 | } | |
7eaf7e8e SL |
2501 | if (bitmap_end) |
2502 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2503 | STRIPE_SECTORS, 0, 0); | |
2504 | bitmap_end = 0; | |
a4456856 DW |
2505 | /* and fail all 'written' */ |
2506 | bi = sh->dev[i].written; | |
2507 | sh->dev[i].written = NULL; | |
2508 | if (bi) bitmap_end = 1; | |
2509 | while (bi && bi->bi_sector < | |
2510 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2511 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2512 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2513 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2514 | md_write_end(conf->mddev); |
2515 | bi->bi_next = *return_bi; | |
2516 | *return_bi = bi; | |
2517 | } | |
2518 | bi = bi2; | |
2519 | } | |
2520 | ||
b5e98d65 DW |
2521 | /* fail any reads if this device is non-operational and |
2522 | * the data has not reached the cache yet. | |
2523 | */ | |
2524 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2525 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2526 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2527 | bi = sh->dev[i].toread; |
2528 | sh->dev[i].toread = NULL; | |
2529 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2530 | wake_up(&conf->wait_for_overlap); | |
2531 | if (bi) s->to_read--; | |
2532 | while (bi && bi->bi_sector < | |
2533 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2534 | struct bio *nextbi = | |
2535 | r5_next_bio(bi, sh->dev[i].sector); | |
2536 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2537 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2538 | bi->bi_next = *return_bi; |
2539 | *return_bi = bi; | |
2540 | } | |
2541 | bi = nextbi; | |
2542 | } | |
2543 | } | |
a4456856 DW |
2544 | if (bitmap_end) |
2545 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2546 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
2547 | /* If we were in the middle of a write the parity block might |
2548 | * still be locked - so just clear all R5_LOCKED flags | |
2549 | */ | |
2550 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
2551 | } |
2552 | ||
8b3e6cdc DW |
2553 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2554 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2555 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2556 | } |
2557 | ||
7f0da59b | 2558 | static void |
d1688a6d | 2559 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
2560 | struct stripe_head_state *s) |
2561 | { | |
2562 | int abort = 0; | |
2563 | int i; | |
2564 | ||
7f0da59b N |
2565 | clear_bit(STRIPE_SYNCING, &sh->state); |
2566 | s->syncing = 0; | |
9a3e1101 | 2567 | s->replacing = 0; |
7f0da59b | 2568 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
2569 | * Don't even need to abort as that is handled elsewhere |
2570 | * if needed, and not always wanted e.g. if there is a known | |
2571 | * bad block here. | |
9a3e1101 | 2572 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
2573 | * non-sync devices, or abort the recovery |
2574 | */ | |
18b9837e N |
2575 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
2576 | /* During recovery devices cannot be removed, so | |
2577 | * locking and refcounting of rdevs is not needed | |
2578 | */ | |
2579 | for (i = 0; i < conf->raid_disks; i++) { | |
2580 | struct md_rdev *rdev = conf->disks[i].rdev; | |
2581 | if (rdev | |
2582 | && !test_bit(Faulty, &rdev->flags) | |
2583 | && !test_bit(In_sync, &rdev->flags) | |
2584 | && !rdev_set_badblocks(rdev, sh->sector, | |
2585 | STRIPE_SECTORS, 0)) | |
2586 | abort = 1; | |
2587 | rdev = conf->disks[i].replacement; | |
2588 | if (rdev | |
2589 | && !test_bit(Faulty, &rdev->flags) | |
2590 | && !test_bit(In_sync, &rdev->flags) | |
2591 | && !rdev_set_badblocks(rdev, sh->sector, | |
2592 | STRIPE_SECTORS, 0)) | |
2593 | abort = 1; | |
2594 | } | |
2595 | if (abort) | |
2596 | conf->recovery_disabled = | |
2597 | conf->mddev->recovery_disabled; | |
7f0da59b | 2598 | } |
18b9837e | 2599 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
2600 | } |
2601 | ||
9a3e1101 N |
2602 | static int want_replace(struct stripe_head *sh, int disk_idx) |
2603 | { | |
2604 | struct md_rdev *rdev; | |
2605 | int rv = 0; | |
2606 | /* Doing recovery so rcu locking not required */ | |
2607 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
2608 | if (rdev | |
2609 | && !test_bit(Faulty, &rdev->flags) | |
2610 | && !test_bit(In_sync, &rdev->flags) | |
2611 | && (rdev->recovery_offset <= sh->sector | |
2612 | || rdev->mddev->recovery_cp <= sh->sector)) | |
2613 | rv = 1; | |
2614 | ||
2615 | return rv; | |
2616 | } | |
2617 | ||
93b3dbce | 2618 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
2619 | * to be read or computed to satisfy a request. |
2620 | * | |
2621 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 2622 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 2623 | */ |
93b3dbce N |
2624 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
2625 | int disk_idx, int disks) | |
a4456856 | 2626 | { |
5599becc | 2627 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2628 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2629 | &sh->dev[s->failed_num[1]] }; | |
5599becc | 2630 | |
93b3dbce | 2631 | /* is the data in this block needed, and can we get it? */ |
5599becc YT |
2632 | if (!test_bit(R5_LOCKED, &dev->flags) && |
2633 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2634 | (dev->toread || | |
2635 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2636 | s->syncing || s->expanding || | |
9a3e1101 | 2637 | (s->replacing && want_replace(sh, disk_idx)) || |
5d35e09c N |
2638 | (s->failed >= 1 && fdev[0]->toread) || |
2639 | (s->failed >= 2 && fdev[1]->toread) || | |
93b3dbce N |
2640 | (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite && |
2641 | !test_bit(R5_OVERWRITE, &fdev[0]->flags)) || | |
2642 | (sh->raid_conf->level == 6 && s->failed && s->to_write))) { | |
5599becc YT |
2643 | /* we would like to get this block, possibly by computing it, |
2644 | * otherwise read it if the backing disk is insync | |
2645 | */ | |
2646 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2647 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2648 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
2649 | (s->failed && (disk_idx == s->failed_num[0] || |
2650 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
2651 | /* have disk failed, and we're requested to fetch it; |
2652 | * do compute it | |
a4456856 | 2653 | */ |
5599becc YT |
2654 | pr_debug("Computing stripe %llu block %d\n", |
2655 | (unsigned long long)sh->sector, disk_idx); | |
2656 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2657 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2658 | set_bit(R5_Wantcompute, &dev->flags); | |
2659 | sh->ops.target = disk_idx; | |
2660 | sh->ops.target2 = -1; /* no 2nd target */ | |
2661 | s->req_compute = 1; | |
93b3dbce N |
2662 | /* Careful: from this point on 'uptodate' is in the eye |
2663 | * of raid_run_ops which services 'compute' operations | |
2664 | * before writes. R5_Wantcompute flags a block that will | |
2665 | * be R5_UPTODATE by the time it is needed for a | |
2666 | * subsequent operation. | |
2667 | */ | |
5599becc YT |
2668 | s->uptodate++; |
2669 | return 1; | |
2670 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2671 | /* Computing 2-failure is *very* expensive; only | |
2672 | * do it if failed >= 2 | |
2673 | */ | |
2674 | int other; | |
2675 | for (other = disks; other--; ) { | |
2676 | if (other == disk_idx) | |
2677 | continue; | |
2678 | if (!test_bit(R5_UPTODATE, | |
2679 | &sh->dev[other].flags)) | |
2680 | break; | |
a4456856 | 2681 | } |
5599becc YT |
2682 | BUG_ON(other < 0); |
2683 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2684 | (unsigned long long)sh->sector, | |
2685 | disk_idx, other); | |
2686 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2687 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2688 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2689 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2690 | sh->ops.target = disk_idx; | |
2691 | sh->ops.target2 = other; | |
2692 | s->uptodate += 2; | |
2693 | s->req_compute = 1; | |
2694 | return 1; | |
2695 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2696 | set_bit(R5_LOCKED, &dev->flags); | |
2697 | set_bit(R5_Wantread, &dev->flags); | |
2698 | s->locked++; | |
2699 | pr_debug("Reading block %d (sync=%d)\n", | |
2700 | disk_idx, s->syncing); | |
a4456856 DW |
2701 | } |
2702 | } | |
5599becc YT |
2703 | |
2704 | return 0; | |
2705 | } | |
2706 | ||
2707 | /** | |
93b3dbce | 2708 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 2709 | */ |
93b3dbce N |
2710 | static void handle_stripe_fill(struct stripe_head *sh, |
2711 | struct stripe_head_state *s, | |
2712 | int disks) | |
5599becc YT |
2713 | { |
2714 | int i; | |
2715 | ||
2716 | /* look for blocks to read/compute, skip this if a compute | |
2717 | * is already in flight, or if the stripe contents are in the | |
2718 | * midst of changing due to a write | |
2719 | */ | |
2720 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2721 | !sh->reconstruct_state) | |
2722 | for (i = disks; i--; ) | |
93b3dbce | 2723 | if (fetch_block(sh, s, i, disks)) |
5599becc | 2724 | break; |
a4456856 DW |
2725 | set_bit(STRIPE_HANDLE, &sh->state); |
2726 | } | |
2727 | ||
2728 | ||
1fe797e6 | 2729 | /* handle_stripe_clean_event |
a4456856 DW |
2730 | * any written block on an uptodate or failed drive can be returned. |
2731 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2732 | * never LOCKED, so we don't need to test 'failed' directly. | |
2733 | */ | |
d1688a6d | 2734 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
2735 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2736 | { | |
2737 | int i; | |
2738 | struct r5dev *dev; | |
2739 | ||
2740 | for (i = disks; i--; ) | |
2741 | if (sh->dev[i].written) { | |
2742 | dev = &sh->dev[i]; | |
2743 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2744 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2745 | /* We can return any write requests */ | |
2746 | struct bio *wbi, *wbi2; | |
45b4233c | 2747 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2748 | wbi = dev->written; |
2749 | dev->written = NULL; | |
2750 | while (wbi && wbi->bi_sector < | |
2751 | dev->sector + STRIPE_SECTORS) { | |
2752 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 2753 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 DW |
2754 | md_write_end(conf->mddev); |
2755 | wbi->bi_next = *return_bi; | |
2756 | *return_bi = wbi; | |
2757 | } | |
2758 | wbi = wbi2; | |
2759 | } | |
7eaf7e8e SL |
2760 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
2761 | STRIPE_SECTORS, | |
a4456856 | 2762 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 2763 | 0); |
a4456856 DW |
2764 | } |
2765 | } | |
8b3e6cdc DW |
2766 | |
2767 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2768 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2769 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2770 | } |
2771 | ||
d1688a6d | 2772 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
2773 | struct stripe_head *sh, |
2774 | struct stripe_head_state *s, | |
2775 | int disks) | |
a4456856 DW |
2776 | { |
2777 | int rmw = 0, rcw = 0, i; | |
c8ac1803 N |
2778 | if (conf->max_degraded == 2) { |
2779 | /* RAID6 requires 'rcw' in current implementation | |
2780 | * Calculate the real rcw later - for now fake it | |
2781 | * look like rcw is cheaper | |
2782 | */ | |
2783 | rcw = 1; rmw = 2; | |
2784 | } else for (i = disks; i--; ) { | |
a4456856 DW |
2785 | /* would I have to read this buffer for read_modify_write */ |
2786 | struct r5dev *dev = &sh->dev[i]; | |
2787 | if ((dev->towrite || i == sh->pd_idx) && | |
2788 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2789 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2790 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2791 | if (test_bit(R5_Insync, &dev->flags)) |
2792 | rmw++; | |
2793 | else | |
2794 | rmw += 2*disks; /* cannot read it */ | |
2795 | } | |
2796 | /* Would I have to read this buffer for reconstruct_write */ | |
2797 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2798 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2799 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2800 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2801 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2802 | else |
2803 | rcw += 2*disks; | |
2804 | } | |
2805 | } | |
45b4233c | 2806 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2807 | (unsigned long long)sh->sector, rmw, rcw); |
2808 | set_bit(STRIPE_HANDLE, &sh->state); | |
2809 | if (rmw < rcw && rmw > 0) | |
2810 | /* prefer read-modify-write, but need to get some data */ | |
2811 | for (i = disks; i--; ) { | |
2812 | struct r5dev *dev = &sh->dev[i]; | |
2813 | if ((dev->towrite || i == sh->pd_idx) && | |
2814 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2815 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2816 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2817 | test_bit(R5_Insync, &dev->flags)) { |
2818 | if ( | |
2819 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2820 | pr_debug("Read_old block " |
a4456856 DW |
2821 | "%d for r-m-w\n", i); |
2822 | set_bit(R5_LOCKED, &dev->flags); | |
2823 | set_bit(R5_Wantread, &dev->flags); | |
2824 | s->locked++; | |
2825 | } else { | |
2826 | set_bit(STRIPE_DELAYED, &sh->state); | |
2827 | set_bit(STRIPE_HANDLE, &sh->state); | |
2828 | } | |
2829 | } | |
2830 | } | |
c8ac1803 | 2831 | if (rcw <= rmw && rcw > 0) { |
a4456856 | 2832 | /* want reconstruct write, but need to get some data */ |
c8ac1803 | 2833 | rcw = 0; |
a4456856 DW |
2834 | for (i = disks; i--; ) { |
2835 | struct r5dev *dev = &sh->dev[i]; | |
2836 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 2837 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 2838 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 2839 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
2840 | test_bit(R5_Wantcompute, &dev->flags))) { |
2841 | rcw++; | |
2842 | if (!test_bit(R5_Insync, &dev->flags)) | |
2843 | continue; /* it's a failed drive */ | |
a4456856 DW |
2844 | if ( |
2845 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2846 | pr_debug("Read_old block " |
a4456856 DW |
2847 | "%d for Reconstruct\n", i); |
2848 | set_bit(R5_LOCKED, &dev->flags); | |
2849 | set_bit(R5_Wantread, &dev->flags); | |
2850 | s->locked++; | |
2851 | } else { | |
2852 | set_bit(STRIPE_DELAYED, &sh->state); | |
2853 | set_bit(STRIPE_HANDLE, &sh->state); | |
2854 | } | |
2855 | } | |
2856 | } | |
c8ac1803 | 2857 | } |
a4456856 DW |
2858 | /* now if nothing is locked, and if we have enough data, |
2859 | * we can start a write request | |
2860 | */ | |
f38e1219 DW |
2861 | /* since handle_stripe can be called at any time we need to handle the |
2862 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2863 | * subsequent call wants to start a write request. raid_run_ops only |
2864 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2865 | * simultaneously. If this is not the case then new writes need to be |
2866 | * held off until the compute completes. | |
2867 | */ | |
976ea8d4 DW |
2868 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2869 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2870 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2871 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2872 | } |
2873 | ||
d1688a6d | 2874 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2875 | struct stripe_head_state *s, int disks) |
2876 | { | |
ecc65c9b | 2877 | struct r5dev *dev = NULL; |
bd2ab670 | 2878 | |
a4456856 | 2879 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2880 | |
ecc65c9b DW |
2881 | switch (sh->check_state) { |
2882 | case check_state_idle: | |
2883 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2884 | if (s->failed == 0) { |
bd2ab670 | 2885 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2886 | sh->check_state = check_state_run; |
2887 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2888 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2889 | s->uptodate--; |
ecc65c9b | 2890 | break; |
bd2ab670 | 2891 | } |
f2b3b44d | 2892 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
2893 | /* fall through */ |
2894 | case check_state_compute_result: | |
2895 | sh->check_state = check_state_idle; | |
2896 | if (!dev) | |
2897 | dev = &sh->dev[sh->pd_idx]; | |
2898 | ||
2899 | /* check that a write has not made the stripe insync */ | |
2900 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2901 | break; | |
c8894419 | 2902 | |
a4456856 | 2903 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2904 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2905 | BUG_ON(s->uptodate != disks); | |
2906 | ||
2907 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2908 | s->locked++; |
a4456856 | 2909 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2910 | |
a4456856 | 2911 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2912 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2913 | break; |
2914 | case check_state_run: | |
2915 | break; /* we will be called again upon completion */ | |
2916 | case check_state_check_result: | |
2917 | sh->check_state = check_state_idle; | |
2918 | ||
2919 | /* if a failure occurred during the check operation, leave | |
2920 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2921 | */ | |
2922 | if (s->failed) | |
2923 | break; | |
2924 | ||
2925 | /* handle a successful check operation, if parity is correct | |
2926 | * we are done. Otherwise update the mismatch count and repair | |
2927 | * parity if !MD_RECOVERY_CHECK | |
2928 | */ | |
ad283ea4 | 2929 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2930 | /* parity is correct (on disc, |
2931 | * not in buffer any more) | |
2932 | */ | |
2933 | set_bit(STRIPE_INSYNC, &sh->state); | |
2934 | else { | |
2935 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2936 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2937 | /* don't try to repair!! */ | |
2938 | set_bit(STRIPE_INSYNC, &sh->state); | |
2939 | else { | |
2940 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2941 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2942 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2943 | set_bit(R5_Wantcompute, | |
2944 | &sh->dev[sh->pd_idx].flags); | |
2945 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2946 | sh->ops.target2 = -1; |
ecc65c9b DW |
2947 | s->uptodate++; |
2948 | } | |
2949 | } | |
2950 | break; | |
2951 | case check_state_compute_run: | |
2952 | break; | |
2953 | default: | |
2954 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2955 | __func__, sh->check_state, | |
2956 | (unsigned long long) sh->sector); | |
2957 | BUG(); | |
a4456856 DW |
2958 | } |
2959 | } | |
2960 | ||
2961 | ||
d1688a6d | 2962 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 2963 | struct stripe_head_state *s, |
f2b3b44d | 2964 | int disks) |
a4456856 | 2965 | { |
a4456856 | 2966 | int pd_idx = sh->pd_idx; |
34e04e87 | 2967 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2968 | struct r5dev *dev; |
a4456856 DW |
2969 | |
2970 | set_bit(STRIPE_HANDLE, &sh->state); | |
2971 | ||
2972 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2973 | |
a4456856 DW |
2974 | /* Want to check and possibly repair P and Q. |
2975 | * However there could be one 'failed' device, in which | |
2976 | * case we can only check one of them, possibly using the | |
2977 | * other to generate missing data | |
2978 | */ | |
2979 | ||
d82dfee0 DW |
2980 | switch (sh->check_state) { |
2981 | case check_state_idle: | |
2982 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 2983 | if (s->failed == s->q_failed) { |
d82dfee0 | 2984 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2985 | * makes sense to check P (If anything else were failed, |
2986 | * we would have used P to recreate it). | |
2987 | */ | |
d82dfee0 | 2988 | sh->check_state = check_state_run; |
a4456856 | 2989 | } |
f2b3b44d | 2990 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 2991 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2992 | * anything, so it makes sense to check it |
2993 | */ | |
d82dfee0 DW |
2994 | if (sh->check_state == check_state_run) |
2995 | sh->check_state = check_state_run_pq; | |
2996 | else | |
2997 | sh->check_state = check_state_run_q; | |
a4456856 | 2998 | } |
a4456856 | 2999 | |
d82dfee0 DW |
3000 | /* discard potentially stale zero_sum_result */ |
3001 | sh->ops.zero_sum_result = 0; | |
a4456856 | 3002 | |
d82dfee0 DW |
3003 | if (sh->check_state == check_state_run) { |
3004 | /* async_xor_zero_sum destroys the contents of P */ | |
3005 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
3006 | s->uptodate--; | |
a4456856 | 3007 | } |
d82dfee0 DW |
3008 | if (sh->check_state >= check_state_run && |
3009 | sh->check_state <= check_state_run_pq) { | |
3010 | /* async_syndrome_zero_sum preserves P and Q, so | |
3011 | * no need to mark them !uptodate here | |
3012 | */ | |
3013 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
3014 | break; | |
a4456856 DW |
3015 | } |
3016 | ||
d82dfee0 DW |
3017 | /* we have 2-disk failure */ |
3018 | BUG_ON(s->failed != 2); | |
3019 | /* fall through */ | |
3020 | case check_state_compute_result: | |
3021 | sh->check_state = check_state_idle; | |
a4456856 | 3022 | |
d82dfee0 DW |
3023 | /* check that a write has not made the stripe insync */ |
3024 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3025 | break; | |
a4456856 DW |
3026 | |
3027 | /* now write out any block on a failed drive, | |
d82dfee0 | 3028 | * or P or Q if they were recomputed |
a4456856 | 3029 | */ |
d82dfee0 | 3030 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 3031 | if (s->failed == 2) { |
f2b3b44d | 3032 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
3033 | s->locked++; |
3034 | set_bit(R5_LOCKED, &dev->flags); | |
3035 | set_bit(R5_Wantwrite, &dev->flags); | |
3036 | } | |
3037 | if (s->failed >= 1) { | |
f2b3b44d | 3038 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
3039 | s->locked++; |
3040 | set_bit(R5_LOCKED, &dev->flags); | |
3041 | set_bit(R5_Wantwrite, &dev->flags); | |
3042 | } | |
d82dfee0 | 3043 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
3044 | dev = &sh->dev[pd_idx]; |
3045 | s->locked++; | |
3046 | set_bit(R5_LOCKED, &dev->flags); | |
3047 | set_bit(R5_Wantwrite, &dev->flags); | |
3048 | } | |
d82dfee0 | 3049 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
3050 | dev = &sh->dev[qd_idx]; |
3051 | s->locked++; | |
3052 | set_bit(R5_LOCKED, &dev->flags); | |
3053 | set_bit(R5_Wantwrite, &dev->flags); | |
3054 | } | |
3055 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
3056 | ||
3057 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
3058 | break; |
3059 | case check_state_run: | |
3060 | case check_state_run_q: | |
3061 | case check_state_run_pq: | |
3062 | break; /* we will be called again upon completion */ | |
3063 | case check_state_check_result: | |
3064 | sh->check_state = check_state_idle; | |
3065 | ||
3066 | /* handle a successful check operation, if parity is correct | |
3067 | * we are done. Otherwise update the mismatch count and repair | |
3068 | * parity if !MD_RECOVERY_CHECK | |
3069 | */ | |
3070 | if (sh->ops.zero_sum_result == 0) { | |
3071 | /* both parities are correct */ | |
3072 | if (!s->failed) | |
3073 | set_bit(STRIPE_INSYNC, &sh->state); | |
3074 | else { | |
3075 | /* in contrast to the raid5 case we can validate | |
3076 | * parity, but still have a failure to write | |
3077 | * back | |
3078 | */ | |
3079 | sh->check_state = check_state_compute_result; | |
3080 | /* Returning at this point means that we may go | |
3081 | * off and bring p and/or q uptodate again so | |
3082 | * we make sure to check zero_sum_result again | |
3083 | * to verify if p or q need writeback | |
3084 | */ | |
3085 | } | |
3086 | } else { | |
3087 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
3088 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
3089 | /* don't try to repair!! */ | |
3090 | set_bit(STRIPE_INSYNC, &sh->state); | |
3091 | else { | |
3092 | int *target = &sh->ops.target; | |
3093 | ||
3094 | sh->ops.target = -1; | |
3095 | sh->ops.target2 = -1; | |
3096 | sh->check_state = check_state_compute_run; | |
3097 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3098 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3099 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3100 | set_bit(R5_Wantcompute, | |
3101 | &sh->dev[pd_idx].flags); | |
3102 | *target = pd_idx; | |
3103 | target = &sh->ops.target2; | |
3104 | s->uptodate++; | |
3105 | } | |
3106 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3107 | set_bit(R5_Wantcompute, | |
3108 | &sh->dev[qd_idx].flags); | |
3109 | *target = qd_idx; | |
3110 | s->uptodate++; | |
3111 | } | |
3112 | } | |
3113 | } | |
3114 | break; | |
3115 | case check_state_compute_run: | |
3116 | break; | |
3117 | default: | |
3118 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3119 | __func__, sh->check_state, | |
3120 | (unsigned long long) sh->sector); | |
3121 | BUG(); | |
a4456856 DW |
3122 | } |
3123 | } | |
3124 | ||
d1688a6d | 3125 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3126 | { |
3127 | int i; | |
3128 | ||
3129 | /* We have read all the blocks in this stripe and now we need to | |
3130 | * copy some of them into a target stripe for expand. | |
3131 | */ | |
f0a50d37 | 3132 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
3133 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3134 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3135 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3136 | int dd_idx, j; |
a4456856 | 3137 | struct stripe_head *sh2; |
a08abd8c | 3138 | struct async_submit_ctl submit; |
a4456856 | 3139 | |
784052ec | 3140 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3141 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3142 | &dd_idx, NULL); | |
a8c906ca | 3143 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3144 | if (sh2 == NULL) |
3145 | /* so far only the early blocks of this stripe | |
3146 | * have been requested. When later blocks | |
3147 | * get requested, we will try again | |
3148 | */ | |
3149 | continue; | |
3150 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3151 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3152 | /* must have already done this block */ | |
3153 | release_stripe(sh2); | |
3154 | continue; | |
3155 | } | |
f0a50d37 DW |
3156 | |
3157 | /* place all the copies on one channel */ | |
a08abd8c | 3158 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3159 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3160 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3161 | &submit); |
f0a50d37 | 3162 | |
a4456856 DW |
3163 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3164 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3165 | for (j = 0; j < conf->raid_disks; j++) | |
3166 | if (j != sh2->pd_idx && | |
86c374ba | 3167 | j != sh2->qd_idx && |
a4456856 DW |
3168 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3169 | break; | |
3170 | if (j == conf->raid_disks) { | |
3171 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3172 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3173 | } | |
3174 | release_stripe(sh2); | |
f0a50d37 | 3175 | |
a4456856 | 3176 | } |
a2e08551 N |
3177 | /* done submitting copies, wait for them to complete */ |
3178 | if (tx) { | |
3179 | async_tx_ack(tx); | |
3180 | dma_wait_for_async_tx(tx); | |
3181 | } | |
a4456856 | 3182 | } |
1da177e4 LT |
3183 | |
3184 | /* | |
3185 | * handle_stripe - do things to a stripe. | |
3186 | * | |
9a3e1101 N |
3187 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3188 | * state of various bits to see what needs to be done. | |
1da177e4 | 3189 | * Possible results: |
9a3e1101 N |
3190 | * return some read requests which now have data |
3191 | * return some write requests which are safely on storage | |
1da177e4 LT |
3192 | * schedule a read on some buffers |
3193 | * schedule a write of some buffers | |
3194 | * return confirmation of parity correctness | |
3195 | * | |
1da177e4 | 3196 | */ |
a4456856 | 3197 | |
acfe726b | 3198 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3199 | { |
d1688a6d | 3200 | struct r5conf *conf = sh->raid_conf; |
f416885e | 3201 | int disks = sh->disks; |
474af965 N |
3202 | struct r5dev *dev; |
3203 | int i; | |
9a3e1101 | 3204 | int do_recovery = 0; |
1da177e4 | 3205 | |
acfe726b N |
3206 | memset(s, 0, sizeof(*s)); |
3207 | ||
acfe726b N |
3208 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3209 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
3210 | s->failed_num[0] = -1; | |
3211 | s->failed_num[1] = -1; | |
1da177e4 | 3212 | |
acfe726b | 3213 | /* Now to look around and see what can be done */ |
1da177e4 | 3214 | rcu_read_lock(); |
16a53ecc | 3215 | for (i=disks; i--; ) { |
3cb03002 | 3216 | struct md_rdev *rdev; |
31c176ec N |
3217 | sector_t first_bad; |
3218 | int bad_sectors; | |
3219 | int is_bad = 0; | |
acfe726b | 3220 | |
16a53ecc | 3221 | dev = &sh->dev[i]; |
1da177e4 | 3222 | |
45b4233c | 3223 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
3224 | i, dev->flags, |
3225 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
3226 | /* maybe we can reply to a read |
3227 | * | |
3228 | * new wantfill requests are only permitted while | |
3229 | * ops_complete_biofill is guaranteed to be inactive | |
3230 | */ | |
3231 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3232 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3233 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3234 | |
16a53ecc | 3235 | /* now count some things */ |
cc94015a N |
3236 | if (test_bit(R5_LOCKED, &dev->flags)) |
3237 | s->locked++; | |
3238 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3239 | s->uptodate++; | |
2d6e4ecc | 3240 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
3241 | s->compute++; |
3242 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 3243 | } |
1da177e4 | 3244 | |
acfe726b | 3245 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 3246 | s->to_fill++; |
acfe726b | 3247 | else if (dev->toread) |
cc94015a | 3248 | s->to_read++; |
16a53ecc | 3249 | if (dev->towrite) { |
cc94015a | 3250 | s->to_write++; |
16a53ecc | 3251 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3252 | s->non_overwrite++; |
16a53ecc | 3253 | } |
a4456856 | 3254 | if (dev->written) |
cc94015a | 3255 | s->written++; |
14a75d3e N |
3256 | /* Prefer to use the replacement for reads, but only |
3257 | * if it is recovered enough and has no bad blocks. | |
3258 | */ | |
3259 | rdev = rcu_dereference(conf->disks[i].replacement); | |
3260 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
3261 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
3262 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3263 | &first_bad, &bad_sectors)) | |
3264 | set_bit(R5_ReadRepl, &dev->flags); | |
3265 | else { | |
9a3e1101 N |
3266 | if (rdev) |
3267 | set_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
3268 | rdev = rcu_dereference(conf->disks[i].rdev); |
3269 | clear_bit(R5_ReadRepl, &dev->flags); | |
3270 | } | |
9283d8c5 N |
3271 | if (rdev && test_bit(Faulty, &rdev->flags)) |
3272 | rdev = NULL; | |
31c176ec N |
3273 | if (rdev) { |
3274 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3275 | &first_bad, &bad_sectors); | |
3276 | if (s->blocked_rdev == NULL | |
3277 | && (test_bit(Blocked, &rdev->flags) | |
3278 | || is_bad < 0)) { | |
3279 | if (is_bad < 0) | |
3280 | set_bit(BlockedBadBlocks, | |
3281 | &rdev->flags); | |
3282 | s->blocked_rdev = rdev; | |
3283 | atomic_inc(&rdev->nr_pending); | |
3284 | } | |
6bfe0b49 | 3285 | } |
415e72d0 N |
3286 | clear_bit(R5_Insync, &dev->flags); |
3287 | if (!rdev) | |
3288 | /* Not in-sync */; | |
31c176ec N |
3289 | else if (is_bad) { |
3290 | /* also not in-sync */ | |
18b9837e N |
3291 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
3292 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
3293 | /* treat as in-sync, but with a read error |
3294 | * which we can now try to correct | |
3295 | */ | |
3296 | set_bit(R5_Insync, &dev->flags); | |
3297 | set_bit(R5_ReadError, &dev->flags); | |
3298 | } | |
3299 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 3300 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 3301 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 3302 | /* in sync if before recovery_offset */ |
30d7a483 N |
3303 | set_bit(R5_Insync, &dev->flags); |
3304 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
3305 | test_bit(R5_Expanded, &dev->flags)) | |
3306 | /* If we've reshaped into here, we assume it is Insync. | |
3307 | * We will shortly update recovery_offset to make | |
3308 | * it official. | |
3309 | */ | |
3310 | set_bit(R5_Insync, &dev->flags); | |
3311 | ||
5d8c71f9 | 3312 | if (rdev && test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
3313 | /* This flag does not apply to '.replacement' |
3314 | * only to .rdev, so make sure to check that*/ | |
3315 | struct md_rdev *rdev2 = rcu_dereference( | |
3316 | conf->disks[i].rdev); | |
3317 | if (rdev2 == rdev) | |
3318 | clear_bit(R5_Insync, &dev->flags); | |
3319 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 3320 | s->handle_bad_blocks = 1; |
14a75d3e | 3321 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
3322 | } else |
3323 | clear_bit(R5_WriteError, &dev->flags); | |
3324 | } | |
5d8c71f9 | 3325 | if (rdev && test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
3326 | /* This flag does not apply to '.replacement' |
3327 | * only to .rdev, so make sure to check that*/ | |
3328 | struct md_rdev *rdev2 = rcu_dereference( | |
3329 | conf->disks[i].rdev); | |
3330 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 3331 | s->handle_bad_blocks = 1; |
14a75d3e | 3332 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
3333 | } else |
3334 | clear_bit(R5_MadeGood, &dev->flags); | |
3335 | } | |
977df362 N |
3336 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
3337 | struct md_rdev *rdev2 = rcu_dereference( | |
3338 | conf->disks[i].replacement); | |
3339 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
3340 | s->handle_bad_blocks = 1; | |
3341 | atomic_inc(&rdev2->nr_pending); | |
3342 | } else | |
3343 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
3344 | } | |
415e72d0 | 3345 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
3346 | /* The ReadError flag will just be confusing now */ |
3347 | clear_bit(R5_ReadError, &dev->flags); | |
3348 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3349 | } |
415e72d0 N |
3350 | if (test_bit(R5_ReadError, &dev->flags)) |
3351 | clear_bit(R5_Insync, &dev->flags); | |
3352 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3353 | if (s->failed < 2) |
3354 | s->failed_num[s->failed] = i; | |
3355 | s->failed++; | |
9a3e1101 N |
3356 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
3357 | do_recovery = 1; | |
415e72d0 | 3358 | } |
1da177e4 | 3359 | } |
9a3e1101 N |
3360 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
3361 | /* If there is a failed device being replaced, | |
3362 | * we must be recovering. | |
3363 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 3364 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
3365 | * else we can only be replacing |
3366 | * sync and recovery both need to read all devices, and so | |
3367 | * use the same flag. | |
3368 | */ | |
3369 | if (do_recovery || | |
c6d2e084 | 3370 | sh->sector >= conf->mddev->recovery_cp || |
3371 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
3372 | s->syncing = 1; |
3373 | else | |
3374 | s->replacing = 1; | |
3375 | } | |
1da177e4 | 3376 | rcu_read_unlock(); |
cc94015a N |
3377 | } |
3378 | ||
3379 | static void handle_stripe(struct stripe_head *sh) | |
3380 | { | |
3381 | struct stripe_head_state s; | |
d1688a6d | 3382 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 3383 | int i; |
84789554 N |
3384 | int prexor; |
3385 | int disks = sh->disks; | |
474af965 | 3386 | struct r5dev *pdev, *qdev; |
cc94015a N |
3387 | |
3388 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 3389 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
3390 | /* already being handled, ensure it gets handled |
3391 | * again when current action finishes */ | |
3392 | set_bit(STRIPE_HANDLE, &sh->state); | |
3393 | return; | |
3394 | } | |
3395 | ||
3396 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3397 | set_bit(STRIPE_SYNCING, &sh->state); | |
3398 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3399 | } | |
3400 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3401 | ||
3402 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3403 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3404 | (unsigned long long)sh->sector, sh->state, | |
3405 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3406 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 3407 | |
acfe726b | 3408 | analyse_stripe(sh, &s); |
c5a31000 | 3409 | |
bc2607f3 N |
3410 | if (s.handle_bad_blocks) { |
3411 | set_bit(STRIPE_HANDLE, &sh->state); | |
3412 | goto finish; | |
3413 | } | |
3414 | ||
474af965 N |
3415 | if (unlikely(s.blocked_rdev)) { |
3416 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 3417 | s.replacing || s.to_write || s.written) { |
474af965 N |
3418 | set_bit(STRIPE_HANDLE, &sh->state); |
3419 | goto finish; | |
3420 | } | |
3421 | /* There is nothing for the blocked_rdev to block */ | |
3422 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
3423 | s.blocked_rdev = NULL; | |
3424 | } | |
3425 | ||
3426 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
3427 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3428 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3429 | } | |
3430 | ||
3431 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
3432 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
3433 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
3434 | s.failed_num[0], s.failed_num[1]); | |
3435 | /* check if the array has lost more than max_degraded devices and, | |
3436 | * if so, some requests might need to be failed. | |
3437 | */ | |
9a3f530f N |
3438 | if (s.failed > conf->max_degraded) { |
3439 | sh->check_state = 0; | |
3440 | sh->reconstruct_state = 0; | |
3441 | if (s.to_read+s.to_write+s.written) | |
3442 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 3443 | if (s.syncing + s.replacing) |
9a3f530f N |
3444 | handle_failed_sync(conf, sh, &s); |
3445 | } | |
474af965 N |
3446 | |
3447 | /* | |
3448 | * might be able to return some write requests if the parity blocks | |
3449 | * are safe, or on a failed drive | |
3450 | */ | |
3451 | pdev = &sh->dev[sh->pd_idx]; | |
3452 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
3453 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
3454 | qdev = &sh->dev[sh->qd_idx]; | |
3455 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
3456 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
3457 | || conf->level < 6; | |
3458 | ||
3459 | if (s.written && | |
3460 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
3461 | && !test_bit(R5_LOCKED, &pdev->flags) | |
3462 | && test_bit(R5_UPTODATE, &pdev->flags)))) && | |
3463 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
3464 | && !test_bit(R5_LOCKED, &qdev->flags) | |
3465 | && test_bit(R5_UPTODATE, &qdev->flags))))) | |
3466 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
3467 | ||
3468 | /* Now we might consider reading some blocks, either to check/generate | |
3469 | * parity, or to satisfy requests | |
3470 | * or to load a block that is being partially written. | |
3471 | */ | |
3472 | if (s.to_read || s.non_overwrite | |
3473 | || (conf->level == 6 && s.to_write && s.failed) | |
9a3e1101 N |
3474 | || (s.syncing && (s.uptodate + s.compute < disks)) |
3475 | || s.replacing | |
3476 | || s.expanding) | |
474af965 N |
3477 | handle_stripe_fill(sh, &s, disks); |
3478 | ||
84789554 N |
3479 | /* Now we check to see if any write operations have recently |
3480 | * completed | |
3481 | */ | |
3482 | prexor = 0; | |
3483 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
3484 | prexor = 1; | |
3485 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
3486 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
3487 | sh->reconstruct_state = reconstruct_state_idle; | |
3488 | ||
3489 | /* All the 'written' buffers and the parity block are ready to | |
3490 | * be written back to disk | |
3491 | */ | |
3492 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3493 | BUG_ON(sh->qd_idx >= 0 && | |
3494 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags)); | |
3495 | for (i = disks; i--; ) { | |
3496 | struct r5dev *dev = &sh->dev[i]; | |
3497 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3498 | (i == sh->pd_idx || i == sh->qd_idx || | |
3499 | dev->written)) { | |
3500 | pr_debug("Writing block %d\n", i); | |
3501 | set_bit(R5_Wantwrite, &dev->flags); | |
3502 | if (prexor) | |
3503 | continue; | |
3504 | if (!test_bit(R5_Insync, &dev->flags) || | |
3505 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
3506 | s.failed == 0)) | |
3507 | set_bit(STRIPE_INSYNC, &sh->state); | |
3508 | } | |
3509 | } | |
3510 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3511 | s.dec_preread_active = 1; | |
3512 | } | |
3513 | ||
3514 | /* Now to consider new write requests and what else, if anything | |
3515 | * should be read. We do not handle new writes when: | |
3516 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3517 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3518 | * block. | |
3519 | */ | |
3520 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
3521 | handle_stripe_dirtying(conf, sh, &s, disks); | |
3522 | ||
3523 | /* maybe we need to check and possibly fix the parity for this stripe | |
3524 | * Any reads will already have been scheduled, so we just see if enough | |
3525 | * data is available. The parity check is held off while parity | |
3526 | * dependent operations are in flight. | |
3527 | */ | |
3528 | if (sh->check_state || | |
3529 | (s.syncing && s.locked == 0 && | |
3530 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3531 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
3532 | if (conf->level == 6) | |
3533 | handle_parity_checks6(conf, sh, &s, disks); | |
3534 | else | |
3535 | handle_parity_checks5(conf, sh, &s, disks); | |
3536 | } | |
c5a31000 | 3537 | |
9a3e1101 N |
3538 | if (s.replacing && s.locked == 0 |
3539 | && !test_bit(STRIPE_INSYNC, &sh->state)) { | |
3540 | /* Write out to replacement devices where possible */ | |
3541 | for (i = 0; i < conf->raid_disks; i++) | |
3542 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags) && | |
3543 | test_bit(R5_NeedReplace, &sh->dev[i].flags)) { | |
3544 | set_bit(R5_WantReplace, &sh->dev[i].flags); | |
3545 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3546 | s.locked++; | |
3547 | } | |
3548 | set_bit(STRIPE_INSYNC, &sh->state); | |
3549 | } | |
3550 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
3551 | test_bit(STRIPE_INSYNC, &sh->state)) { | |
c5a31000 N |
3552 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
3553 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3554 | } | |
3555 | ||
3556 | /* If the failed drives are just a ReadError, then we might need | |
3557 | * to progress the repair/check process | |
3558 | */ | |
3559 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
3560 | for (i = 0; i < s.failed; i++) { | |
3561 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
3562 | if (test_bit(R5_ReadError, &dev->flags) | |
3563 | && !test_bit(R5_LOCKED, &dev->flags) | |
3564 | && test_bit(R5_UPTODATE, &dev->flags) | |
3565 | ) { | |
3566 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3567 | set_bit(R5_Wantwrite, &dev->flags); | |
3568 | set_bit(R5_ReWrite, &dev->flags); | |
3569 | set_bit(R5_LOCKED, &dev->flags); | |
3570 | s.locked++; | |
3571 | } else { | |
3572 | /* let's read it back */ | |
3573 | set_bit(R5_Wantread, &dev->flags); | |
3574 | set_bit(R5_LOCKED, &dev->flags); | |
3575 | s.locked++; | |
3576 | } | |
3577 | } | |
3578 | } | |
3579 | ||
3580 | ||
3687c061 N |
3581 | /* Finish reconstruct operations initiated by the expansion process */ |
3582 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3583 | struct stripe_head *sh_src | |
3584 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
3585 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
3586 | /* sh cannot be written until sh_src has been read. | |
3587 | * so arrange for sh to be delayed a little | |
3588 | */ | |
3589 | set_bit(STRIPE_DELAYED, &sh->state); | |
3590 | set_bit(STRIPE_HANDLE, &sh->state); | |
3591 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3592 | &sh_src->state)) | |
3593 | atomic_inc(&conf->preread_active_stripes); | |
3594 | release_stripe(sh_src); | |
3595 | goto finish; | |
3596 | } | |
3597 | if (sh_src) | |
3598 | release_stripe(sh_src); | |
3599 | ||
3600 | sh->reconstruct_state = reconstruct_state_idle; | |
3601 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3602 | for (i = conf->raid_disks; i--; ) { | |
3603 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3604 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3605 | s.locked++; | |
3606 | } | |
3607 | } | |
f416885e | 3608 | |
3687c061 N |
3609 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
3610 | !sh->reconstruct_state) { | |
3611 | /* Need to write out all blocks after computing parity */ | |
3612 | sh->disks = conf->raid_disks; | |
3613 | stripe_set_idx(sh->sector, conf, 0, sh); | |
3614 | schedule_reconstruction(sh, &s, 1, 1); | |
3615 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
3616 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
3617 | atomic_dec(&conf->reshape_stripes); | |
3618 | wake_up(&conf->wait_for_overlap); | |
3619 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3620 | } | |
3621 | ||
3622 | if (s.expanding && s.locked == 0 && | |
3623 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
3624 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 3625 | |
3687c061 | 3626 | finish: |
6bfe0b49 | 3627 | /* wait for this device to become unblocked */ |
5f066c63 N |
3628 | if (unlikely(s.blocked_rdev)) { |
3629 | if (conf->mddev->external) | |
3630 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
3631 | conf->mddev); | |
3632 | else | |
3633 | /* Internal metadata will immediately | |
3634 | * be written by raid5d, so we don't | |
3635 | * need to wait here. | |
3636 | */ | |
3637 | rdev_dec_pending(s.blocked_rdev, | |
3638 | conf->mddev); | |
3639 | } | |
6bfe0b49 | 3640 | |
bc2607f3 N |
3641 | if (s.handle_bad_blocks) |
3642 | for (i = disks; i--; ) { | |
3cb03002 | 3643 | struct md_rdev *rdev; |
bc2607f3 N |
3644 | struct r5dev *dev = &sh->dev[i]; |
3645 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
3646 | /* We own a safe reference to the rdev */ | |
3647 | rdev = conf->disks[i].rdev; | |
3648 | if (!rdev_set_badblocks(rdev, sh->sector, | |
3649 | STRIPE_SECTORS, 0)) | |
3650 | md_error(conf->mddev, rdev); | |
3651 | rdev_dec_pending(rdev, conf->mddev); | |
3652 | } | |
b84db560 N |
3653 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
3654 | rdev = conf->disks[i].rdev; | |
3655 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 3656 | STRIPE_SECTORS, 0); |
b84db560 N |
3657 | rdev_dec_pending(rdev, conf->mddev); |
3658 | } | |
977df362 N |
3659 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
3660 | rdev = conf->disks[i].replacement; | |
dd054fce N |
3661 | if (!rdev) |
3662 | /* rdev have been moved down */ | |
3663 | rdev = conf->disks[i].rdev; | |
977df362 | 3664 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 3665 | STRIPE_SECTORS, 0); |
977df362 N |
3666 | rdev_dec_pending(rdev, conf->mddev); |
3667 | } | |
bc2607f3 N |
3668 | } |
3669 | ||
6c0069c0 YT |
3670 | if (s.ops_request) |
3671 | raid_run_ops(sh, s.ops_request); | |
3672 | ||
f0e43bcd | 3673 | ops_run_io(sh, &s); |
16a53ecc | 3674 | |
c5709ef6 | 3675 | if (s.dec_preread_active) { |
729a1866 | 3676 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 3677 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3678 | * have actually been submitted. |
3679 | */ | |
3680 | atomic_dec(&conf->preread_active_stripes); | |
3681 | if (atomic_read(&conf->preread_active_stripes) < | |
3682 | IO_THRESHOLD) | |
3683 | md_wakeup_thread(conf->mddev->thread); | |
3684 | } | |
3685 | ||
c5709ef6 | 3686 | return_io(s.return_bi); |
16a53ecc | 3687 | |
257a4b42 | 3688 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
3689 | } |
3690 | ||
d1688a6d | 3691 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
3692 | { |
3693 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3694 | while (!list_empty(&conf->delayed_list)) { | |
3695 | struct list_head *l = conf->delayed_list.next; | |
3696 | struct stripe_head *sh; | |
3697 | sh = list_entry(l, struct stripe_head, lru); | |
3698 | list_del_init(l); | |
3699 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3700 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3701 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3702 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3703 | } |
482c0834 | 3704 | } |
16a53ecc N |
3705 | } |
3706 | ||
d1688a6d | 3707 | static void activate_bit_delay(struct r5conf *conf) |
16a53ecc N |
3708 | { |
3709 | /* device_lock is held */ | |
3710 | struct list_head head; | |
3711 | list_add(&head, &conf->bitmap_list); | |
3712 | list_del_init(&conf->bitmap_list); | |
3713 | while (!list_empty(&head)) { | |
3714 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3715 | list_del_init(&sh->lru); | |
3716 | atomic_inc(&sh->count); | |
3717 | __release_stripe(conf, sh); | |
3718 | } | |
3719 | } | |
3720 | ||
fd01b88c | 3721 | int md_raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 3722 | { |
d1688a6d | 3723 | struct r5conf *conf = mddev->private; |
f022b2fd N |
3724 | |
3725 | /* No difference between reads and writes. Just check | |
3726 | * how busy the stripe_cache is | |
3727 | */ | |
3fa841d7 | 3728 | |
f022b2fd N |
3729 | if (conf->inactive_blocked) |
3730 | return 1; | |
3731 | if (conf->quiesce) | |
3732 | return 1; | |
3733 | if (list_empty_careful(&conf->inactive_list)) | |
3734 | return 1; | |
3735 | ||
3736 | return 0; | |
3737 | } | |
11d8a6e3 N |
3738 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3739 | ||
3740 | static int raid5_congested(void *data, int bits) | |
3741 | { | |
fd01b88c | 3742 | struct mddev *mddev = data; |
11d8a6e3 N |
3743 | |
3744 | return mddev_congested(mddev, bits) || | |
3745 | md_raid5_congested(mddev, bits); | |
3746 | } | |
f022b2fd | 3747 | |
23032a0e RBJ |
3748 | /* We want read requests to align with chunks where possible, |
3749 | * but write requests don't need to. | |
3750 | */ | |
cc371e66 AK |
3751 | static int raid5_mergeable_bvec(struct request_queue *q, |
3752 | struct bvec_merge_data *bvm, | |
3753 | struct bio_vec *biovec) | |
23032a0e | 3754 | { |
fd01b88c | 3755 | struct mddev *mddev = q->queuedata; |
cc371e66 | 3756 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3757 | int max; |
9d8f0363 | 3758 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3759 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3760 | |
cc371e66 | 3761 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3762 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3763 | ||
664e7c41 AN |
3764 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3765 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3766 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3767 | if (max < 0) max = 0; | |
3768 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3769 | return biovec->bv_len; | |
3770 | else | |
3771 | return max; | |
3772 | } | |
3773 | ||
f679623f | 3774 | |
fd01b88c | 3775 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f RBJ |
3776 | { |
3777 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3778 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3779 | unsigned int bio_sectors = bio->bi_size >> 9; |
3780 | ||
664e7c41 AN |
3781 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3782 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3783 | return chunk_sectors >= |
3784 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3785 | } | |
3786 | ||
46031f9a RBJ |
3787 | /* |
3788 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3789 | * later sampled by raid5d. | |
3790 | */ | |
d1688a6d | 3791 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
3792 | { |
3793 | unsigned long flags; | |
3794 | ||
3795 | spin_lock_irqsave(&conf->device_lock, flags); | |
3796 | ||
3797 | bi->bi_next = conf->retry_read_aligned_list; | |
3798 | conf->retry_read_aligned_list = bi; | |
3799 | ||
3800 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3801 | md_wakeup_thread(conf->mddev->thread); | |
3802 | } | |
3803 | ||
3804 | ||
d1688a6d | 3805 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
3806 | { |
3807 | struct bio *bi; | |
3808 | ||
3809 | bi = conf->retry_read_aligned; | |
3810 | if (bi) { | |
3811 | conf->retry_read_aligned = NULL; | |
3812 | return bi; | |
3813 | } | |
3814 | bi = conf->retry_read_aligned_list; | |
3815 | if(bi) { | |
387bb173 | 3816 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3817 | bi->bi_next = NULL; |
960e739d JA |
3818 | /* |
3819 | * this sets the active strip count to 1 and the processed | |
3820 | * strip count to zero (upper 8 bits) | |
3821 | */ | |
e7836bd6 | 3822 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
3823 | } |
3824 | ||
3825 | return bi; | |
3826 | } | |
3827 | ||
3828 | ||
f679623f RBJ |
3829 | /* |
3830 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3831 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3832 | * first). | |
3833 | * If the read failed.. | |
3834 | */ | |
6712ecf8 | 3835 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3836 | { |
3837 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 3838 | struct mddev *mddev; |
d1688a6d | 3839 | struct r5conf *conf; |
46031f9a | 3840 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 3841 | struct md_rdev *rdev; |
46031f9a | 3842 | |
f679623f | 3843 | bio_put(bi); |
46031f9a | 3844 | |
46031f9a RBJ |
3845 | rdev = (void*)raid_bi->bi_next; |
3846 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3847 | mddev = rdev->mddev; |
3848 | conf = mddev->private; | |
46031f9a RBJ |
3849 | |
3850 | rdev_dec_pending(rdev, conf->mddev); | |
3851 | ||
3852 | if (!error && uptodate) { | |
6712ecf8 | 3853 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3854 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3855 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3856 | return; |
46031f9a RBJ |
3857 | } |
3858 | ||
3859 | ||
45b4233c | 3860 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3861 | |
3862 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3863 | } |
3864 | ||
387bb173 NB |
3865 | static int bio_fits_rdev(struct bio *bi) |
3866 | { | |
165125e1 | 3867 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3868 | |
ae03bf63 | 3869 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3870 | return 0; |
3871 | blk_recount_segments(q, bi); | |
8a78362c | 3872 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3873 | return 0; |
3874 | ||
3875 | if (q->merge_bvec_fn) | |
3876 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3877 | * just just give up | |
3878 | */ | |
3879 | return 0; | |
3880 | ||
3881 | return 1; | |
3882 | } | |
3883 | ||
3884 | ||
fd01b88c | 3885 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 3886 | { |
d1688a6d | 3887 | struct r5conf *conf = mddev->private; |
8553fe7e | 3888 | int dd_idx; |
f679623f | 3889 | struct bio* align_bi; |
3cb03002 | 3890 | struct md_rdev *rdev; |
671488cc | 3891 | sector_t end_sector; |
f679623f RBJ |
3892 | |
3893 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3894 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3895 | return 0; |
3896 | } | |
3897 | /* | |
a167f663 | 3898 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3899 | */ |
a167f663 | 3900 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3901 | if (!align_bi) |
3902 | return 0; | |
3903 | /* | |
3904 | * set bi_end_io to a new function, and set bi_private to the | |
3905 | * original bio. | |
3906 | */ | |
3907 | align_bi->bi_end_io = raid5_align_endio; | |
3908 | align_bi->bi_private = raid_bio; | |
3909 | /* | |
3910 | * compute position | |
3911 | */ | |
112bf897 N |
3912 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3913 | 0, | |
911d4ee8 | 3914 | &dd_idx, NULL); |
f679623f | 3915 | |
671488cc | 3916 | end_sector = align_bi->bi_sector + (align_bi->bi_size >> 9); |
f679623f | 3917 | rcu_read_lock(); |
671488cc N |
3918 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
3919 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
3920 | rdev->recovery_offset < end_sector) { | |
3921 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3922 | if (rdev && | |
3923 | (test_bit(Faulty, &rdev->flags) || | |
3924 | !(test_bit(In_sync, &rdev->flags) || | |
3925 | rdev->recovery_offset >= end_sector))) | |
3926 | rdev = NULL; | |
3927 | } | |
3928 | if (rdev) { | |
31c176ec N |
3929 | sector_t first_bad; |
3930 | int bad_sectors; | |
3931 | ||
f679623f RBJ |
3932 | atomic_inc(&rdev->nr_pending); |
3933 | rcu_read_unlock(); | |
46031f9a RBJ |
3934 | raid_bio->bi_next = (void*)rdev; |
3935 | align_bi->bi_bdev = rdev->bdev; | |
3936 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
46031f9a | 3937 | |
31c176ec N |
3938 | if (!bio_fits_rdev(align_bi) || |
3939 | is_badblock(rdev, align_bi->bi_sector, align_bi->bi_size>>9, | |
3940 | &first_bad, &bad_sectors)) { | |
3941 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
3942 | bio_put(align_bi); |
3943 | rdev_dec_pending(rdev, mddev); | |
3944 | return 0; | |
3945 | } | |
3946 | ||
6c0544e2 | 3947 | /* No reshape active, so we can trust rdev->data_offset */ |
3948 | align_bi->bi_sector += rdev->data_offset; | |
3949 | ||
46031f9a RBJ |
3950 | spin_lock_irq(&conf->device_lock); |
3951 | wait_event_lock_irq(conf->wait_for_stripe, | |
3952 | conf->quiesce == 0, | |
3953 | conf->device_lock, /* nothing */); | |
3954 | atomic_inc(&conf->active_aligned_reads); | |
3955 | spin_unlock_irq(&conf->device_lock); | |
3956 | ||
f679623f RBJ |
3957 | generic_make_request(align_bi); |
3958 | return 1; | |
3959 | } else { | |
3960 | rcu_read_unlock(); | |
46031f9a | 3961 | bio_put(align_bi); |
f679623f RBJ |
3962 | return 0; |
3963 | } | |
3964 | } | |
3965 | ||
8b3e6cdc DW |
3966 | /* __get_priority_stripe - get the next stripe to process |
3967 | * | |
3968 | * Full stripe writes are allowed to pass preread active stripes up until | |
3969 | * the bypass_threshold is exceeded. In general the bypass_count | |
3970 | * increments when the handle_list is handled before the hold_list; however, it | |
3971 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3972 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3973 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3974 | * handle_list. | |
3975 | */ | |
d1688a6d | 3976 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf) |
8b3e6cdc DW |
3977 | { |
3978 | struct stripe_head *sh; | |
3979 | ||
3980 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3981 | __func__, | |
3982 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3983 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3984 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3985 | ||
3986 | if (!list_empty(&conf->handle_list)) { | |
3987 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3988 | ||
3989 | if (list_empty(&conf->hold_list)) | |
3990 | conf->bypass_count = 0; | |
3991 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3992 | if (conf->hold_list.next == conf->last_hold) | |
3993 | conf->bypass_count++; | |
3994 | else { | |
3995 | conf->last_hold = conf->hold_list.next; | |
3996 | conf->bypass_count -= conf->bypass_threshold; | |
3997 | if (conf->bypass_count < 0) | |
3998 | conf->bypass_count = 0; | |
3999 | } | |
4000 | } | |
4001 | } else if (!list_empty(&conf->hold_list) && | |
4002 | ((conf->bypass_threshold && | |
4003 | conf->bypass_count > conf->bypass_threshold) || | |
4004 | atomic_read(&conf->pending_full_writes) == 0)) { | |
4005 | sh = list_entry(conf->hold_list.next, | |
4006 | typeof(*sh), lru); | |
4007 | conf->bypass_count -= conf->bypass_threshold; | |
4008 | if (conf->bypass_count < 0) | |
4009 | conf->bypass_count = 0; | |
4010 | } else | |
4011 | return NULL; | |
4012 | ||
4013 | list_del_init(&sh->lru); | |
4014 | atomic_inc(&sh->count); | |
4015 | BUG_ON(atomic_read(&sh->count) != 1); | |
4016 | return sh; | |
4017 | } | |
f679623f | 4018 | |
8811b596 SL |
4019 | struct raid5_plug_cb { |
4020 | struct blk_plug_cb cb; | |
4021 | struct list_head list; | |
4022 | }; | |
4023 | ||
4024 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
4025 | { | |
4026 | struct raid5_plug_cb *cb = container_of( | |
4027 | blk_cb, struct raid5_plug_cb, cb); | |
4028 | struct stripe_head *sh; | |
4029 | struct mddev *mddev = cb->cb.data; | |
4030 | struct r5conf *conf = mddev->private; | |
4031 | ||
4032 | if (cb->list.next && !list_empty(&cb->list)) { | |
4033 | spin_lock_irq(&conf->device_lock); | |
4034 | while (!list_empty(&cb->list)) { | |
4035 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
4036 | list_del_init(&sh->lru); | |
4037 | /* | |
4038 | * avoid race release_stripe_plug() sees | |
4039 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
4040 | * is still in our list | |
4041 | */ | |
4042 | smp_mb__before_clear_bit(); | |
4043 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); | |
4044 | __release_stripe(conf, sh); | |
4045 | } | |
4046 | spin_unlock_irq(&conf->device_lock); | |
4047 | } | |
4048 | kfree(cb); | |
4049 | } | |
4050 | ||
4051 | static void release_stripe_plug(struct mddev *mddev, | |
4052 | struct stripe_head *sh) | |
4053 | { | |
4054 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
4055 | raid5_unplug, mddev, | |
4056 | sizeof(struct raid5_plug_cb)); | |
4057 | struct raid5_plug_cb *cb; | |
4058 | ||
4059 | if (!blk_cb) { | |
4060 | release_stripe(sh); | |
4061 | return; | |
4062 | } | |
4063 | ||
4064 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
4065 | ||
4066 | if (cb->list.next == NULL) | |
4067 | INIT_LIST_HEAD(&cb->list); | |
4068 | ||
4069 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
4070 | list_add_tail(&sh->lru, &cb->list); | |
4071 | else | |
4072 | release_stripe(sh); | |
4073 | } | |
4074 | ||
b4fdcb02 | 4075 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 4076 | { |
d1688a6d | 4077 | struct r5conf *conf = mddev->private; |
911d4ee8 | 4078 | int dd_idx; |
1da177e4 LT |
4079 | sector_t new_sector; |
4080 | sector_t logical_sector, last_sector; | |
4081 | struct stripe_head *sh; | |
a362357b | 4082 | const int rw = bio_data_dir(bi); |
49077326 | 4083 | int remaining; |
1da177e4 | 4084 | |
e9c7469b TH |
4085 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
4086 | md_flush_request(mddev, bi); | |
5a7bbad2 | 4087 | return; |
e5dcdd80 N |
4088 | } |
4089 | ||
3d310eb7 | 4090 | md_write_start(mddev, bi); |
06d91a5f | 4091 | |
802ba064 | 4092 | if (rw == READ && |
52488615 | 4093 | mddev->reshape_position == MaxSector && |
21a52c6d | 4094 | chunk_aligned_read(mddev,bi)) |
5a7bbad2 | 4095 | return; |
52488615 | 4096 | |
1da177e4 LT |
4097 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
4098 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
4099 | bi->bi_next = NULL; | |
4100 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 4101 | |
1da177e4 LT |
4102 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
4103 | DEFINE_WAIT(w); | |
b5663ba4 | 4104 | int previous; |
b578d55f | 4105 | |
7ecaa1e6 | 4106 | retry: |
b5663ba4 | 4107 | previous = 0; |
b578d55f | 4108 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 4109 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 4110 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
4111 | * 64bit on a 32bit platform, and so it might be |
4112 | * possible to see a half-updated value | |
aeb878b0 | 4113 | * Of course reshape_progress could change after |
df8e7f76 N |
4114 | * the lock is dropped, so once we get a reference |
4115 | * to the stripe that we think it is, we will have | |
4116 | * to check again. | |
4117 | */ | |
7ecaa1e6 | 4118 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 4119 | if (mddev->reshape_backwards |
fef9c61f N |
4120 | ? logical_sector < conf->reshape_progress |
4121 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
4122 | previous = 1; |
4123 | } else { | |
2c810cdd | 4124 | if (mddev->reshape_backwards |
fef9c61f N |
4125 | ? logical_sector < conf->reshape_safe |
4126 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
4127 | spin_unlock_irq(&conf->device_lock); |
4128 | schedule(); | |
4129 | goto retry; | |
4130 | } | |
4131 | } | |
7ecaa1e6 N |
4132 | spin_unlock_irq(&conf->device_lock); |
4133 | } | |
16a53ecc | 4134 | |
112bf897 N |
4135 | new_sector = raid5_compute_sector(conf, logical_sector, |
4136 | previous, | |
911d4ee8 | 4137 | &dd_idx, NULL); |
0c55e022 | 4138 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
4139 | (unsigned long long)new_sector, |
4140 | (unsigned long long)logical_sector); | |
4141 | ||
b5663ba4 | 4142 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 4143 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 4144 | if (sh) { |
b0f9ec04 | 4145 | if (unlikely(previous)) { |
7ecaa1e6 | 4146 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
4147 | * stripe, so we must do the range check again. |
4148 | * Expansion could still move past after this | |
4149 | * test, but as we are holding a reference to | |
4150 | * 'sh', we know that if that happens, | |
4151 | * STRIPE_EXPANDING will get set and the expansion | |
4152 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
4153 | */ |
4154 | int must_retry = 0; | |
4155 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 4156 | if (mddev->reshape_backwards |
b0f9ec04 N |
4157 | ? logical_sector >= conf->reshape_progress |
4158 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
4159 | /* mismatch, need to try again */ |
4160 | must_retry = 1; | |
4161 | spin_unlock_irq(&conf->device_lock); | |
4162 | if (must_retry) { | |
4163 | release_stripe(sh); | |
7a3ab908 | 4164 | schedule(); |
7ecaa1e6 N |
4165 | goto retry; |
4166 | } | |
4167 | } | |
e62e58a5 | 4168 | |
ffd96e35 | 4169 | if (rw == WRITE && |
a5c308d4 | 4170 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
4171 | logical_sector < mddev->suspend_hi) { |
4172 | release_stripe(sh); | |
e62e58a5 N |
4173 | /* As the suspend_* range is controlled by |
4174 | * userspace, we want an interruptible | |
4175 | * wait. | |
4176 | */ | |
4177 | flush_signals(current); | |
4178 | prepare_to_wait(&conf->wait_for_overlap, | |
4179 | &w, TASK_INTERRUPTIBLE); | |
4180 | if (logical_sector >= mddev->suspend_lo && | |
4181 | logical_sector < mddev->suspend_hi) | |
4182 | schedule(); | |
e464eafd N |
4183 | goto retry; |
4184 | } | |
7ecaa1e6 N |
4185 | |
4186 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 4187 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
4188 | /* Stripe is busy expanding or |
4189 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4190 | * and wait a while |
4191 | */ | |
482c0834 | 4192 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4193 | release_stripe(sh); |
4194 | schedule(); | |
4195 | goto retry; | |
4196 | } | |
4197 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
4198 | set_bit(STRIPE_HANDLE, &sh->state); |
4199 | clear_bit(STRIPE_DELAYED, &sh->state); | |
a852d7b8 | 4200 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
4201 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4202 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 4203 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
4204 | } else { |
4205 | /* cannot get stripe for read-ahead, just give-up */ | |
4206 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4207 | finish_wait(&conf->wait_for_overlap, &w); | |
4208 | break; | |
4209 | } | |
1da177e4 | 4210 | } |
7c13edc8 | 4211 | |
e7836bd6 | 4212 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 4213 | if (remaining == 0) { |
1da177e4 | 4214 | |
16a53ecc | 4215 | if ( rw == WRITE ) |
1da177e4 | 4216 | md_write_end(mddev); |
6712ecf8 | 4217 | |
0e13fe23 | 4218 | bio_endio(bi, 0); |
1da177e4 | 4219 | } |
1da177e4 LT |
4220 | } |
4221 | ||
fd01b88c | 4222 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 4223 | |
fd01b88c | 4224 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4225 | { |
52c03291 N |
4226 | /* reshaping is quite different to recovery/resync so it is |
4227 | * handled quite separately ... here. | |
4228 | * | |
4229 | * On each call to sync_request, we gather one chunk worth of | |
4230 | * destination stripes and flag them as expanding. | |
4231 | * Then we find all the source stripes and request reads. | |
4232 | * As the reads complete, handle_stripe will copy the data | |
4233 | * into the destination stripe and release that stripe. | |
4234 | */ | |
d1688a6d | 4235 | struct r5conf *conf = mddev->private; |
1da177e4 | 4236 | struct stripe_head *sh; |
ccfcc3c1 | 4237 | sector_t first_sector, last_sector; |
f416885e N |
4238 | int raid_disks = conf->previous_raid_disks; |
4239 | int data_disks = raid_disks - conf->max_degraded; | |
4240 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4241 | int i; |
4242 | int dd_idx; | |
c8f517c4 | 4243 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4244 | sector_t stripe_addr; |
7a661381 | 4245 | int reshape_sectors; |
ab69ae12 | 4246 | struct list_head stripes; |
52c03291 | 4247 | |
fef9c61f N |
4248 | if (sector_nr == 0) { |
4249 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 4250 | if (mddev->reshape_backwards && |
fef9c61f N |
4251 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
4252 | sector_nr = raid5_size(mddev, 0, 0) | |
4253 | - conf->reshape_progress; | |
2c810cdd | 4254 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
4255 | conf->reshape_progress > 0) |
4256 | sector_nr = conf->reshape_progress; | |
f416885e | 4257 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4258 | if (sector_nr) { |
8dee7211 N |
4259 | mddev->curr_resync_completed = sector_nr; |
4260 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4261 | *skipped = 1; |
4262 | return sector_nr; | |
4263 | } | |
52c03291 N |
4264 | } |
4265 | ||
7a661381 N |
4266 | /* We need to process a full chunk at a time. |
4267 | * If old and new chunk sizes differ, we need to process the | |
4268 | * largest of these | |
4269 | */ | |
664e7c41 AN |
4270 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4271 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4272 | else |
9d8f0363 | 4273 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4274 | |
b5254dd5 N |
4275 | /* We update the metadata at least every 10 seconds, or when |
4276 | * the data about to be copied would over-write the source of | |
4277 | * the data at the front of the range. i.e. one new_stripe | |
4278 | * along from reshape_progress new_maps to after where | |
4279 | * reshape_safe old_maps to | |
52c03291 | 4280 | */ |
fef9c61f | 4281 | writepos = conf->reshape_progress; |
f416885e | 4282 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4283 | readpos = conf->reshape_progress; |
4284 | sector_div(readpos, data_disks); | |
fef9c61f | 4285 | safepos = conf->reshape_safe; |
f416885e | 4286 | sector_div(safepos, data_disks); |
2c810cdd | 4287 | if (mddev->reshape_backwards) { |
ed37d83e | 4288 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4289 | readpos += reshape_sectors; |
7a661381 | 4290 | safepos += reshape_sectors; |
fef9c61f | 4291 | } else { |
7a661381 | 4292 | writepos += reshape_sectors; |
ed37d83e N |
4293 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4294 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4295 | } |
52c03291 | 4296 | |
b5254dd5 N |
4297 | /* Having calculated the 'writepos' possibly use it |
4298 | * to set 'stripe_addr' which is where we will write to. | |
4299 | */ | |
4300 | if (mddev->reshape_backwards) { | |
4301 | BUG_ON(conf->reshape_progress == 0); | |
4302 | stripe_addr = writepos; | |
4303 | BUG_ON((mddev->dev_sectors & | |
4304 | ~((sector_t)reshape_sectors - 1)) | |
4305 | - reshape_sectors - stripe_addr | |
4306 | != sector_nr); | |
4307 | } else { | |
4308 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
4309 | stripe_addr = sector_nr; | |
4310 | } | |
4311 | ||
c8f517c4 N |
4312 | /* 'writepos' is the most advanced device address we might write. |
4313 | * 'readpos' is the least advanced device address we might read. | |
4314 | * 'safepos' is the least address recorded in the metadata as having | |
4315 | * been reshaped. | |
b5254dd5 N |
4316 | * If there is a min_offset_diff, these are adjusted either by |
4317 | * increasing the safepos/readpos if diff is negative, or | |
4318 | * increasing writepos if diff is positive. | |
4319 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
4320 | * ensure safety in the face of a crash - that must be done by userspace |
4321 | * making a backup of the data. So in that case there is no particular | |
4322 | * rush to update metadata. | |
4323 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4324 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4325 | * we can be safe in the event of a crash. | |
4326 | * So we insist on updating metadata if safepos is behind writepos and | |
4327 | * readpos is beyond writepos. | |
4328 | * In any case, update the metadata every 10 seconds. | |
4329 | * Maybe that number should be configurable, but I'm not sure it is | |
4330 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4331 | */ | |
b5254dd5 N |
4332 | if (conf->min_offset_diff < 0) { |
4333 | safepos += -conf->min_offset_diff; | |
4334 | readpos += -conf->min_offset_diff; | |
4335 | } else | |
4336 | writepos += conf->min_offset_diff; | |
4337 | ||
2c810cdd | 4338 | if ((mddev->reshape_backwards |
c8f517c4 N |
4339 | ? (safepos > writepos && readpos < writepos) |
4340 | : (safepos < writepos && readpos > writepos)) || | |
4341 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4342 | /* Cannot proceed until we've updated the superblock... */ |
4343 | wait_event(conf->wait_for_overlap, | |
4344 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4345 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4346 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4347 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4348 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4349 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4350 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4351 | kthread_should_stop()); |
4352 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4353 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4354 | spin_unlock_irq(&conf->device_lock); |
4355 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4356 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4357 | } |
4358 | ||
ab69ae12 | 4359 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4360 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4361 | int j; |
a9f326eb | 4362 | int skipped_disk = 0; |
a8c906ca | 4363 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4364 | set_bit(STRIPE_EXPANDING, &sh->state); |
4365 | atomic_inc(&conf->reshape_stripes); | |
4366 | /* If any of this stripe is beyond the end of the old | |
4367 | * array, then we need to zero those blocks | |
4368 | */ | |
4369 | for (j=sh->disks; j--;) { | |
4370 | sector_t s; | |
4371 | if (j == sh->pd_idx) | |
4372 | continue; | |
f416885e | 4373 | if (conf->level == 6 && |
d0dabf7e | 4374 | j == sh->qd_idx) |
f416885e | 4375 | continue; |
784052ec | 4376 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4377 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4378 | skipped_disk = 1; |
52c03291 N |
4379 | continue; |
4380 | } | |
4381 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4382 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4383 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4384 | } | |
a9f326eb | 4385 | if (!skipped_disk) { |
52c03291 N |
4386 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4387 | set_bit(STRIPE_HANDLE, &sh->state); | |
4388 | } | |
ab69ae12 | 4389 | list_add(&sh->lru, &stripes); |
52c03291 N |
4390 | } |
4391 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 4392 | if (mddev->reshape_backwards) |
7a661381 | 4393 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4394 | else |
7a661381 | 4395 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4396 | spin_unlock_irq(&conf->device_lock); |
4397 | /* Ok, those stripe are ready. We can start scheduling | |
4398 | * reads on the source stripes. | |
4399 | * The source stripes are determined by mapping the first and last | |
4400 | * block on the destination stripes. | |
4401 | */ | |
52c03291 | 4402 | first_sector = |
ec32a2bd | 4403 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4404 | 1, &dd_idx, NULL); |
52c03291 | 4405 | last_sector = |
0e6e0271 | 4406 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4407 | * new_data_disks - 1), |
911d4ee8 | 4408 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4409 | if (last_sector >= mddev->dev_sectors) |
4410 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4411 | while (first_sector <= last_sector) { |
a8c906ca | 4412 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4413 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4414 | set_bit(STRIPE_HANDLE, &sh->state); | |
4415 | release_stripe(sh); | |
4416 | first_sector += STRIPE_SECTORS; | |
4417 | } | |
ab69ae12 N |
4418 | /* Now that the sources are clearly marked, we can release |
4419 | * the destination stripes | |
4420 | */ | |
4421 | while (!list_empty(&stripes)) { | |
4422 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4423 | list_del_init(&sh->lru); | |
4424 | release_stripe(sh); | |
4425 | } | |
c6207277 N |
4426 | /* If this takes us to the resync_max point where we have to pause, |
4427 | * then we need to write out the superblock. | |
4428 | */ | |
7a661381 | 4429 | sector_nr += reshape_sectors; |
c03f6a19 N |
4430 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4431 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4432 | /* Cannot proceed until we've updated the superblock... */ |
4433 | wait_event(conf->wait_for_overlap, | |
4434 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4435 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4436 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4437 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4438 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4439 | md_wakeup_thread(mddev->thread); | |
4440 | wait_event(mddev->sb_wait, | |
4441 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4442 | || kthread_should_stop()); | |
4443 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4444 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4445 | spin_unlock_irq(&conf->device_lock); |
4446 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4447 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4448 | } |
7a661381 | 4449 | return reshape_sectors; |
52c03291 N |
4450 | } |
4451 | ||
4452 | /* FIXME go_faster isn't used */ | |
fd01b88c | 4453 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
52c03291 | 4454 | { |
d1688a6d | 4455 | struct r5conf *conf = mddev->private; |
52c03291 | 4456 | struct stripe_head *sh; |
58c0fed4 | 4457 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 4458 | sector_t sync_blocks; |
16a53ecc N |
4459 | int still_degraded = 0; |
4460 | int i; | |
1da177e4 | 4461 | |
72626685 | 4462 | if (sector_nr >= max_sector) { |
1da177e4 | 4463 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 4464 | |
29269553 N |
4465 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4466 | end_reshape(conf); | |
4467 | return 0; | |
4468 | } | |
72626685 N |
4469 | |
4470 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4471 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4472 | &sync_blocks, 1); | |
16a53ecc | 4473 | else /* completed sync */ |
72626685 N |
4474 | conf->fullsync = 0; |
4475 | bitmap_close_sync(mddev->bitmap); | |
4476 | ||
1da177e4 LT |
4477 | return 0; |
4478 | } | |
ccfcc3c1 | 4479 | |
64bd660b N |
4480 | /* Allow raid5_quiesce to complete */ |
4481 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4482 | ||
52c03291 N |
4483 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4484 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4485 | |
c6207277 N |
4486 | /* No need to check resync_max as we never do more than one |
4487 | * stripe, and as resync_max will always be on a chunk boundary, | |
4488 | * if the check in md_do_sync didn't fire, there is no chance | |
4489 | * of overstepping resync_max here | |
4490 | */ | |
4491 | ||
16a53ecc | 4492 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4493 | * to resync, then assert that we are finished, because there is |
4494 | * nothing we can do. | |
4495 | */ | |
3285edf1 | 4496 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4497 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4498 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4499 | *skipped = 1; |
1da177e4 LT |
4500 | return rv; |
4501 | } | |
72626685 | 4502 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4503 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4504 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4505 | /* we can skip this block, and probably more */ | |
4506 | sync_blocks /= STRIPE_SECTORS; | |
4507 | *skipped = 1; | |
4508 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4509 | } | |
1da177e4 | 4510 | |
b47490c9 N |
4511 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
4512 | ||
a8c906ca | 4513 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4514 | if (sh == NULL) { |
a8c906ca | 4515 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4516 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4517 | * is trying to get access |
1da177e4 | 4518 | */ |
66c006a5 | 4519 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4520 | } |
16a53ecc N |
4521 | /* Need to check if array will still be degraded after recovery/resync |
4522 | * We don't need to check the 'failed' flag as when that gets set, | |
4523 | * recovery aborts. | |
4524 | */ | |
f001a70c | 4525 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4526 | if (conf->disks[i].rdev == NULL) |
4527 | still_degraded = 1; | |
4528 | ||
4529 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4530 | ||
83206d66 | 4531 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4532 | |
1442577b | 4533 | handle_stripe(sh); |
1da177e4 LT |
4534 | release_stripe(sh); |
4535 | ||
4536 | return STRIPE_SECTORS; | |
4537 | } | |
4538 | ||
d1688a6d | 4539 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
4540 | { |
4541 | /* We may not be able to submit a whole bio at once as there | |
4542 | * may not be enough stripe_heads available. | |
4543 | * We cannot pre-allocate enough stripe_heads as we may need | |
4544 | * more than exist in the cache (if we allow ever large chunks). | |
4545 | * So we do one stripe head at a time and record in | |
4546 | * ->bi_hw_segments how many have been done. | |
4547 | * | |
4548 | * We *know* that this entire raid_bio is in one chunk, so | |
4549 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4550 | */ | |
4551 | struct stripe_head *sh; | |
911d4ee8 | 4552 | int dd_idx; |
46031f9a RBJ |
4553 | sector_t sector, logical_sector, last_sector; |
4554 | int scnt = 0; | |
4555 | int remaining; | |
4556 | int handled = 0; | |
4557 | ||
4558 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4559 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4560 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4561 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4562 | ||
4563 | for (; logical_sector < last_sector; | |
387bb173 NB |
4564 | logical_sector += STRIPE_SECTORS, |
4565 | sector += STRIPE_SECTORS, | |
4566 | scnt++) { | |
46031f9a | 4567 | |
e7836bd6 | 4568 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
4569 | /* already done this stripe */ |
4570 | continue; | |
4571 | ||
a8c906ca | 4572 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4573 | |
4574 | if (!sh) { | |
4575 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 4576 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
4577 | conf->retry_read_aligned = raid_bio; |
4578 | return handled; | |
4579 | } | |
4580 | ||
387bb173 NB |
4581 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4582 | release_stripe(sh); | |
e7836bd6 | 4583 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
4584 | conf->retry_read_aligned = raid_bio; |
4585 | return handled; | |
4586 | } | |
4587 | ||
3f9e7c14 | 4588 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 4589 | handle_stripe(sh); |
46031f9a RBJ |
4590 | release_stripe(sh); |
4591 | handled++; | |
4592 | } | |
e7836bd6 | 4593 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0e13fe23 NB |
4594 | if (remaining == 0) |
4595 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4596 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4597 | wake_up(&conf->wait_for_stripe); | |
4598 | return handled; | |
4599 | } | |
4600 | ||
46a06401 SL |
4601 | #define MAX_STRIPE_BATCH 8 |
4602 | static int handle_active_stripes(struct r5conf *conf) | |
4603 | { | |
4604 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
4605 | int i, batch_size = 0; | |
4606 | ||
4607 | while (batch_size < MAX_STRIPE_BATCH && | |
4608 | (sh = __get_priority_stripe(conf)) != NULL) | |
4609 | batch[batch_size++] = sh; | |
4610 | ||
4611 | if (batch_size == 0) | |
4612 | return batch_size; | |
4613 | spin_unlock_irq(&conf->device_lock); | |
4614 | ||
4615 | for (i = 0; i < batch_size; i++) | |
4616 | handle_stripe(batch[i]); | |
4617 | ||
4618 | cond_resched(); | |
4619 | ||
4620 | spin_lock_irq(&conf->device_lock); | |
4621 | for (i = 0; i < batch_size; i++) | |
4622 | __release_stripe(conf, batch[i]); | |
4623 | return batch_size; | |
4624 | } | |
46031f9a | 4625 | |
1da177e4 LT |
4626 | /* |
4627 | * This is our raid5 kernel thread. | |
4628 | * | |
4629 | * We scan the hash table for stripes which can be handled now. | |
4630 | * During the scan, completed stripes are saved for us by the interrupt | |
4631 | * handler, so that they will not have to wait for our next wakeup. | |
4632 | */ | |
fd01b88c | 4633 | static void raid5d(struct mddev *mddev) |
1da177e4 | 4634 | { |
d1688a6d | 4635 | struct r5conf *conf = mddev->private; |
1da177e4 | 4636 | int handled; |
e1dfa0a2 | 4637 | struct blk_plug plug; |
1da177e4 | 4638 | |
45b4233c | 4639 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4640 | |
4641 | md_check_recovery(mddev); | |
1da177e4 | 4642 | |
e1dfa0a2 | 4643 | blk_start_plug(&plug); |
1da177e4 LT |
4644 | handled = 0; |
4645 | spin_lock_irq(&conf->device_lock); | |
4646 | while (1) { | |
46031f9a | 4647 | struct bio *bio; |
46a06401 | 4648 | int batch_size; |
1da177e4 | 4649 | |
0021b7bc | 4650 | if ( |
7c13edc8 N |
4651 | !list_empty(&conf->bitmap_list)) { |
4652 | /* Now is a good time to flush some bitmap updates */ | |
4653 | conf->seq_flush++; | |
700e432d | 4654 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4655 | bitmap_unplug(mddev->bitmap); |
700e432d | 4656 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4657 | conf->seq_write = conf->seq_flush; |
72626685 N |
4658 | activate_bit_delay(conf); |
4659 | } | |
0021b7bc | 4660 | raid5_activate_delayed(conf); |
72626685 | 4661 | |
46031f9a RBJ |
4662 | while ((bio = remove_bio_from_retry(conf))) { |
4663 | int ok; | |
4664 | spin_unlock_irq(&conf->device_lock); | |
4665 | ok = retry_aligned_read(conf, bio); | |
4666 | spin_lock_irq(&conf->device_lock); | |
4667 | if (!ok) | |
4668 | break; | |
4669 | handled++; | |
4670 | } | |
4671 | ||
46a06401 SL |
4672 | batch_size = handle_active_stripes(conf); |
4673 | if (!batch_size) | |
1da177e4 | 4674 | break; |
46a06401 | 4675 | handled += batch_size; |
1da177e4 | 4676 | |
46a06401 SL |
4677 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) { |
4678 | spin_unlock_irq(&conf->device_lock); | |
de393cde | 4679 | md_check_recovery(mddev); |
46a06401 SL |
4680 | spin_lock_irq(&conf->device_lock); |
4681 | } | |
1da177e4 | 4682 | } |
45b4233c | 4683 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4684 | |
4685 | spin_unlock_irq(&conf->device_lock); | |
4686 | ||
c9f21aaf | 4687 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4688 | blk_finish_plug(&plug); |
1da177e4 | 4689 | |
45b4233c | 4690 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4691 | } |
4692 | ||
3f294f4f | 4693 | static ssize_t |
fd01b88c | 4694 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 4695 | { |
d1688a6d | 4696 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4697 | if (conf) |
4698 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4699 | else | |
4700 | return 0; | |
3f294f4f N |
4701 | } |
4702 | ||
c41d4ac4 | 4703 | int |
fd01b88c | 4704 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 4705 | { |
d1688a6d | 4706 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
4707 | int err; |
4708 | ||
c41d4ac4 | 4709 | if (size <= 16 || size > 32768) |
3f294f4f | 4710 | return -EINVAL; |
c41d4ac4 | 4711 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4712 | if (drop_one_stripe(conf)) |
4713 | conf->max_nr_stripes--; | |
4714 | else | |
4715 | break; | |
4716 | } | |
b5470dc5 DW |
4717 | err = md_allow_write(mddev); |
4718 | if (err) | |
4719 | return err; | |
c41d4ac4 | 4720 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4721 | if (grow_one_stripe(conf)) |
4722 | conf->max_nr_stripes++; | |
4723 | else break; | |
4724 | } | |
c41d4ac4 N |
4725 | return 0; |
4726 | } | |
4727 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4728 | ||
4729 | static ssize_t | |
fd01b88c | 4730 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 4731 | { |
d1688a6d | 4732 | struct r5conf *conf = mddev->private; |
c41d4ac4 N |
4733 | unsigned long new; |
4734 | int err; | |
4735 | ||
4736 | if (len >= PAGE_SIZE) | |
4737 | return -EINVAL; | |
4738 | if (!conf) | |
4739 | return -ENODEV; | |
4740 | ||
4741 | if (strict_strtoul(page, 10, &new)) | |
4742 | return -EINVAL; | |
4743 | err = raid5_set_cache_size(mddev, new); | |
4744 | if (err) | |
4745 | return err; | |
3f294f4f N |
4746 | return len; |
4747 | } | |
007583c9 | 4748 | |
96de1e66 N |
4749 | static struct md_sysfs_entry |
4750 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4751 | raid5_show_stripe_cache_size, | |
4752 | raid5_store_stripe_cache_size); | |
3f294f4f | 4753 | |
8b3e6cdc | 4754 | static ssize_t |
fd01b88c | 4755 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 4756 | { |
d1688a6d | 4757 | struct r5conf *conf = mddev->private; |
8b3e6cdc DW |
4758 | if (conf) |
4759 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4760 | else | |
4761 | return 0; | |
4762 | } | |
4763 | ||
4764 | static ssize_t | |
fd01b88c | 4765 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 4766 | { |
d1688a6d | 4767 | struct r5conf *conf = mddev->private; |
4ef197d8 | 4768 | unsigned long new; |
8b3e6cdc DW |
4769 | if (len >= PAGE_SIZE) |
4770 | return -EINVAL; | |
4771 | if (!conf) | |
4772 | return -ENODEV; | |
4773 | ||
4ef197d8 | 4774 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4775 | return -EINVAL; |
4ef197d8 | 4776 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4777 | return -EINVAL; |
4778 | conf->bypass_threshold = new; | |
4779 | return len; | |
4780 | } | |
4781 | ||
4782 | static struct md_sysfs_entry | |
4783 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4784 | S_IRUGO | S_IWUSR, | |
4785 | raid5_show_preread_threshold, | |
4786 | raid5_store_preread_threshold); | |
4787 | ||
3f294f4f | 4788 | static ssize_t |
fd01b88c | 4789 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 4790 | { |
d1688a6d | 4791 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4792 | if (conf) |
4793 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4794 | else | |
4795 | return 0; | |
3f294f4f N |
4796 | } |
4797 | ||
96de1e66 N |
4798 | static struct md_sysfs_entry |
4799 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4800 | |
007583c9 | 4801 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4802 | &raid5_stripecache_size.attr, |
4803 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4804 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4805 | NULL, |
4806 | }; | |
007583c9 N |
4807 | static struct attribute_group raid5_attrs_group = { |
4808 | .name = NULL, | |
4809 | .attrs = raid5_attrs, | |
3f294f4f N |
4810 | }; |
4811 | ||
80c3a6ce | 4812 | static sector_t |
fd01b88c | 4813 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 4814 | { |
d1688a6d | 4815 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
4816 | |
4817 | if (!sectors) | |
4818 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4819 | if (!raid_disks) |
7ec05478 | 4820 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4821 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4822 | |
9d8f0363 | 4823 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4824 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4825 | return sectors * (raid_disks - conf->max_degraded); |
4826 | } | |
4827 | ||
d1688a6d | 4828 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 DW |
4829 | { |
4830 | struct raid5_percpu *percpu; | |
4831 | unsigned long cpu; | |
4832 | ||
4833 | if (!conf->percpu) | |
4834 | return; | |
4835 | ||
4836 | get_online_cpus(); | |
4837 | for_each_possible_cpu(cpu) { | |
4838 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4839 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4840 | kfree(percpu->scribble); |
36d1c647 DW |
4841 | } |
4842 | #ifdef CONFIG_HOTPLUG_CPU | |
4843 | unregister_cpu_notifier(&conf->cpu_notify); | |
4844 | #endif | |
4845 | put_online_cpus(); | |
4846 | ||
4847 | free_percpu(conf->percpu); | |
4848 | } | |
4849 | ||
d1688a6d | 4850 | static void free_conf(struct r5conf *conf) |
95fc17aa DW |
4851 | { |
4852 | shrink_stripes(conf); | |
36d1c647 | 4853 | raid5_free_percpu(conf); |
95fc17aa DW |
4854 | kfree(conf->disks); |
4855 | kfree(conf->stripe_hashtbl); | |
4856 | kfree(conf); | |
4857 | } | |
4858 | ||
36d1c647 DW |
4859 | #ifdef CONFIG_HOTPLUG_CPU |
4860 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4861 | void *hcpu) | |
4862 | { | |
d1688a6d | 4863 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
4864 | long cpu = (long)hcpu; |
4865 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4866 | ||
4867 | switch (action) { | |
4868 | case CPU_UP_PREPARE: | |
4869 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4870 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4871 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4872 | if (!percpu->scribble) |
4873 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4874 | ||
4875 | if (!percpu->scribble || | |
4876 | (conf->level == 6 && !percpu->spare_page)) { | |
4877 | safe_put_page(percpu->spare_page); | |
4878 | kfree(percpu->scribble); | |
36d1c647 DW |
4879 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4880 | __func__, cpu); | |
55af6bb5 | 4881 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4882 | } |
4883 | break; | |
4884 | case CPU_DEAD: | |
4885 | case CPU_DEAD_FROZEN: | |
4886 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4887 | kfree(percpu->scribble); |
36d1c647 | 4888 | percpu->spare_page = NULL; |
d6f38f31 | 4889 | percpu->scribble = NULL; |
36d1c647 DW |
4890 | break; |
4891 | default: | |
4892 | break; | |
4893 | } | |
4894 | return NOTIFY_OK; | |
4895 | } | |
4896 | #endif | |
4897 | ||
d1688a6d | 4898 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
4899 | { |
4900 | unsigned long cpu; | |
4901 | struct page *spare_page; | |
a29d8b8e | 4902 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4903 | void *scribble; |
36d1c647 DW |
4904 | int err; |
4905 | ||
36d1c647 DW |
4906 | allcpus = alloc_percpu(struct raid5_percpu); |
4907 | if (!allcpus) | |
4908 | return -ENOMEM; | |
4909 | conf->percpu = allcpus; | |
4910 | ||
4911 | get_online_cpus(); | |
4912 | err = 0; | |
4913 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4914 | if (conf->level == 6) { |
4915 | spare_page = alloc_page(GFP_KERNEL); | |
4916 | if (!spare_page) { | |
4917 | err = -ENOMEM; | |
4918 | break; | |
4919 | } | |
4920 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4921 | } | |
5e5e3e78 | 4922 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4923 | if (!scribble) { |
36d1c647 DW |
4924 | err = -ENOMEM; |
4925 | break; | |
4926 | } | |
d6f38f31 | 4927 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4928 | } |
4929 | #ifdef CONFIG_HOTPLUG_CPU | |
4930 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4931 | conf->cpu_notify.priority = 0; | |
4932 | if (err == 0) | |
4933 | err = register_cpu_notifier(&conf->cpu_notify); | |
4934 | #endif | |
4935 | put_online_cpus(); | |
4936 | ||
4937 | return err; | |
4938 | } | |
4939 | ||
d1688a6d | 4940 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 4941 | { |
d1688a6d | 4942 | struct r5conf *conf; |
5e5e3e78 | 4943 | int raid_disk, memory, max_disks; |
3cb03002 | 4944 | struct md_rdev *rdev; |
1da177e4 | 4945 | struct disk_info *disk; |
0232605d | 4946 | char pers_name[6]; |
1da177e4 | 4947 | |
91adb564 N |
4948 | if (mddev->new_level != 5 |
4949 | && mddev->new_level != 4 | |
4950 | && mddev->new_level != 6) { | |
0c55e022 | 4951 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4952 | mdname(mddev), mddev->new_level); |
4953 | return ERR_PTR(-EIO); | |
1da177e4 | 4954 | } |
91adb564 N |
4955 | if ((mddev->new_level == 5 |
4956 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4957 | (mddev->new_level == 6 | |
4958 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4959 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4960 | mdname(mddev), mddev->new_layout); |
4961 | return ERR_PTR(-EIO); | |
99c0fb5f | 4962 | } |
91adb564 | 4963 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4964 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4965 | mdname(mddev), mddev->raid_disks); |
4966 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4967 | } |
4968 | ||
664e7c41 AN |
4969 | if (!mddev->new_chunk_sectors || |
4970 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4971 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4972 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4973 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4974 | return ERR_PTR(-EINVAL); |
f6705578 N |
4975 | } |
4976 | ||
d1688a6d | 4977 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 4978 | if (conf == NULL) |
1da177e4 | 4979 | goto abort; |
f5efd45a DW |
4980 | spin_lock_init(&conf->device_lock); |
4981 | init_waitqueue_head(&conf->wait_for_stripe); | |
4982 | init_waitqueue_head(&conf->wait_for_overlap); | |
4983 | INIT_LIST_HEAD(&conf->handle_list); | |
4984 | INIT_LIST_HEAD(&conf->hold_list); | |
4985 | INIT_LIST_HEAD(&conf->delayed_list); | |
4986 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4987 | INIT_LIST_HEAD(&conf->inactive_list); | |
4988 | atomic_set(&conf->active_stripes, 0); | |
4989 | atomic_set(&conf->preread_active_stripes, 0); | |
4990 | atomic_set(&conf->active_aligned_reads, 0); | |
4991 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 4992 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
4993 | |
4994 | conf->raid_disks = mddev->raid_disks; | |
4995 | if (mddev->reshape_position == MaxSector) | |
4996 | conf->previous_raid_disks = mddev->raid_disks; | |
4997 | else | |
f6705578 | 4998 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4999 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
5000 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 5001 | |
5e5e3e78 | 5002 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
5003 | GFP_KERNEL); |
5004 | if (!conf->disks) | |
5005 | goto abort; | |
9ffae0cf | 5006 | |
1da177e4 LT |
5007 | conf->mddev = mddev; |
5008 | ||
fccddba0 | 5009 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 5010 | goto abort; |
1da177e4 | 5011 | |
36d1c647 DW |
5012 | conf->level = mddev->new_level; |
5013 | if (raid5_alloc_percpu(conf) != 0) | |
5014 | goto abort; | |
5015 | ||
0c55e022 | 5016 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 5017 | |
dafb20fa | 5018 | rdev_for_each(rdev, mddev) { |
1da177e4 | 5019 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 5020 | if (raid_disk >= max_disks |
1da177e4 LT |
5021 | || raid_disk < 0) |
5022 | continue; | |
5023 | disk = conf->disks + raid_disk; | |
5024 | ||
17045f52 N |
5025 | if (test_bit(Replacement, &rdev->flags)) { |
5026 | if (disk->replacement) | |
5027 | goto abort; | |
5028 | disk->replacement = rdev; | |
5029 | } else { | |
5030 | if (disk->rdev) | |
5031 | goto abort; | |
5032 | disk->rdev = rdev; | |
5033 | } | |
1da177e4 | 5034 | |
b2d444d7 | 5035 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 5036 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
5037 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
5038 | " disk %d\n", | |
5039 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 5040 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
5041 | /* Cannot rely on bitmap to complete recovery */ |
5042 | conf->fullsync = 1; | |
1da177e4 LT |
5043 | } |
5044 | ||
09c9e5fa | 5045 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 5046 | conf->level = mddev->new_level; |
16a53ecc N |
5047 | if (conf->level == 6) |
5048 | conf->max_degraded = 2; | |
5049 | else | |
5050 | conf->max_degraded = 1; | |
91adb564 | 5051 | conf->algorithm = mddev->new_layout; |
1da177e4 | 5052 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 5053 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 5054 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 5055 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
5056 | conf->prev_algo = mddev->layout; |
5057 | } | |
1da177e4 | 5058 | |
91adb564 | 5059 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 5060 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
5061 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
5062 | printk(KERN_ERR | |
0c55e022 N |
5063 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
5064 | mdname(mddev), memory); | |
91adb564 N |
5065 | goto abort; |
5066 | } else | |
0c55e022 N |
5067 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
5068 | mdname(mddev), memory); | |
1da177e4 | 5069 | |
0232605d N |
5070 | sprintf(pers_name, "raid%d", mddev->new_level); |
5071 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 N |
5072 | if (!conf->thread) { |
5073 | printk(KERN_ERR | |
0c55e022 | 5074 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 5075 | mdname(mddev)); |
16a53ecc N |
5076 | goto abort; |
5077 | } | |
91adb564 N |
5078 | |
5079 | return conf; | |
5080 | ||
5081 | abort: | |
5082 | if (conf) { | |
95fc17aa | 5083 | free_conf(conf); |
91adb564 N |
5084 | return ERR_PTR(-EIO); |
5085 | } else | |
5086 | return ERR_PTR(-ENOMEM); | |
5087 | } | |
5088 | ||
c148ffdc N |
5089 | |
5090 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
5091 | { | |
5092 | switch (algo) { | |
5093 | case ALGORITHM_PARITY_0: | |
5094 | if (raid_disk < max_degraded) | |
5095 | return 1; | |
5096 | break; | |
5097 | case ALGORITHM_PARITY_N: | |
5098 | if (raid_disk >= raid_disks - max_degraded) | |
5099 | return 1; | |
5100 | break; | |
5101 | case ALGORITHM_PARITY_0_6: | |
5102 | if (raid_disk == 0 || | |
5103 | raid_disk == raid_disks - 1) | |
5104 | return 1; | |
5105 | break; | |
5106 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5107 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5108 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5109 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5110 | if (raid_disk == raid_disks - 1) | |
5111 | return 1; | |
5112 | } | |
5113 | return 0; | |
5114 | } | |
5115 | ||
fd01b88c | 5116 | static int run(struct mddev *mddev) |
91adb564 | 5117 | { |
d1688a6d | 5118 | struct r5conf *conf; |
9f7c2220 | 5119 | int working_disks = 0; |
c148ffdc | 5120 | int dirty_parity_disks = 0; |
3cb03002 | 5121 | struct md_rdev *rdev; |
c148ffdc | 5122 | sector_t reshape_offset = 0; |
17045f52 | 5123 | int i; |
b5254dd5 N |
5124 | long long min_offset_diff = 0; |
5125 | int first = 1; | |
91adb564 | 5126 | |
8c6ac868 | 5127 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 5128 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
5129 | " -- starting background reconstruction\n", |
5130 | mdname(mddev)); | |
b5254dd5 N |
5131 | |
5132 | rdev_for_each(rdev, mddev) { | |
5133 | long long diff; | |
5134 | if (rdev->raid_disk < 0) | |
5135 | continue; | |
5136 | diff = (rdev->new_data_offset - rdev->data_offset); | |
5137 | if (first) { | |
5138 | min_offset_diff = diff; | |
5139 | first = 0; | |
5140 | } else if (mddev->reshape_backwards && | |
5141 | diff < min_offset_diff) | |
5142 | min_offset_diff = diff; | |
5143 | else if (!mddev->reshape_backwards && | |
5144 | diff > min_offset_diff) | |
5145 | min_offset_diff = diff; | |
5146 | } | |
5147 | ||
91adb564 N |
5148 | if (mddev->reshape_position != MaxSector) { |
5149 | /* Check that we can continue the reshape. | |
b5254dd5 N |
5150 | * Difficulties arise if the stripe we would write to |
5151 | * next is at or after the stripe we would read from next. | |
5152 | * For a reshape that changes the number of devices, this | |
5153 | * is only possible for a very short time, and mdadm makes | |
5154 | * sure that time appears to have past before assembling | |
5155 | * the array. So we fail if that time hasn't passed. | |
5156 | * For a reshape that keeps the number of devices the same | |
5157 | * mdadm must be monitoring the reshape can keeping the | |
5158 | * critical areas read-only and backed up. It will start | |
5159 | * the array in read-only mode, so we check for that. | |
91adb564 N |
5160 | */ |
5161 | sector_t here_new, here_old; | |
5162 | int old_disks; | |
18b00334 | 5163 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 5164 | |
88ce4930 | 5165 | if (mddev->new_level != mddev->level) { |
0c55e022 | 5166 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
5167 | "required - aborting.\n", |
5168 | mdname(mddev)); | |
5169 | return -EINVAL; | |
5170 | } | |
91adb564 N |
5171 | old_disks = mddev->raid_disks - mddev->delta_disks; |
5172 | /* reshape_position must be on a new-stripe boundary, and one | |
5173 | * further up in new geometry must map after here in old | |
5174 | * geometry. | |
5175 | */ | |
5176 | here_new = mddev->reshape_position; | |
664e7c41 | 5177 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 5178 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
5179 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
5180 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
5181 | return -EINVAL; |
5182 | } | |
c148ffdc | 5183 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
5184 | /* here_new is the stripe we will write to */ |
5185 | here_old = mddev->reshape_position; | |
9d8f0363 | 5186 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
5187 | (old_disks-max_degraded)); |
5188 | /* here_old is the first stripe that we might need to read | |
5189 | * from */ | |
67ac6011 | 5190 | if (mddev->delta_disks == 0) { |
b5254dd5 N |
5191 | if ((here_new * mddev->new_chunk_sectors != |
5192 | here_old * mddev->chunk_sectors)) { | |
5193 | printk(KERN_ERR "md/raid:%s: reshape position is" | |
5194 | " confused - aborting\n", mdname(mddev)); | |
5195 | return -EINVAL; | |
5196 | } | |
67ac6011 | 5197 | /* We cannot be sure it is safe to start an in-place |
b5254dd5 | 5198 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
5199 | * and taking constant backups. |
5200 | * mdadm always starts a situation like this in | |
5201 | * readonly mode so it can take control before | |
5202 | * allowing any writes. So just check for that. | |
5203 | */ | |
b5254dd5 N |
5204 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
5205 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
5206 | /* not really in-place - so OK */; | |
5207 | else if (mddev->ro == 0) { | |
5208 | printk(KERN_ERR "md/raid:%s: in-place reshape " | |
5209 | "must be started in read-only mode " | |
5210 | "- aborting\n", | |
0c55e022 | 5211 | mdname(mddev)); |
67ac6011 N |
5212 | return -EINVAL; |
5213 | } | |
2c810cdd | 5214 | } else if (mddev->reshape_backwards |
b5254dd5 | 5215 | ? (here_new * mddev->new_chunk_sectors + min_offset_diff <= |
67ac6011 N |
5216 | here_old * mddev->chunk_sectors) |
5217 | : (here_new * mddev->new_chunk_sectors >= | |
b5254dd5 | 5218 | here_old * mddev->chunk_sectors + (-min_offset_diff))) { |
91adb564 | 5219 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
5220 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
5221 | "auto-recovery - aborting.\n", | |
5222 | mdname(mddev)); | |
91adb564 N |
5223 | return -EINVAL; |
5224 | } | |
0c55e022 N |
5225 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
5226 | mdname(mddev)); | |
91adb564 N |
5227 | /* OK, we should be able to continue; */ |
5228 | } else { | |
5229 | BUG_ON(mddev->level != mddev->new_level); | |
5230 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 5231 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 5232 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 5233 | } |
91adb564 | 5234 | |
245f46c2 N |
5235 | if (mddev->private == NULL) |
5236 | conf = setup_conf(mddev); | |
5237 | else | |
5238 | conf = mddev->private; | |
5239 | ||
91adb564 N |
5240 | if (IS_ERR(conf)) |
5241 | return PTR_ERR(conf); | |
5242 | ||
b5254dd5 | 5243 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
5244 | mddev->thread = conf->thread; |
5245 | conf->thread = NULL; | |
5246 | mddev->private = conf; | |
5247 | ||
17045f52 N |
5248 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
5249 | i++) { | |
5250 | rdev = conf->disks[i].rdev; | |
5251 | if (!rdev && conf->disks[i].replacement) { | |
5252 | /* The replacement is all we have yet */ | |
5253 | rdev = conf->disks[i].replacement; | |
5254 | conf->disks[i].replacement = NULL; | |
5255 | clear_bit(Replacement, &rdev->flags); | |
5256 | conf->disks[i].rdev = rdev; | |
5257 | } | |
5258 | if (!rdev) | |
c148ffdc | 5259 | continue; |
17045f52 N |
5260 | if (conf->disks[i].replacement && |
5261 | conf->reshape_progress != MaxSector) { | |
5262 | /* replacements and reshape simply do not mix. */ | |
5263 | printk(KERN_ERR "md: cannot handle concurrent " | |
5264 | "replacement and reshape.\n"); | |
5265 | goto abort; | |
5266 | } | |
2f115882 | 5267 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 5268 | working_disks++; |
2f115882 N |
5269 | continue; |
5270 | } | |
c148ffdc N |
5271 | /* This disc is not fully in-sync. However if it |
5272 | * just stored parity (beyond the recovery_offset), | |
5273 | * when we don't need to be concerned about the | |
5274 | * array being dirty. | |
5275 | * When reshape goes 'backwards', we never have | |
5276 | * partially completed devices, so we only need | |
5277 | * to worry about reshape going forwards. | |
5278 | */ | |
5279 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
5280 | if (mddev->major_version == 0 && | |
5281 | mddev->minor_version > 90) | |
5282 | rdev->recovery_offset = reshape_offset; | |
5283 | ||
c148ffdc N |
5284 | if (rdev->recovery_offset < reshape_offset) { |
5285 | /* We need to check old and new layout */ | |
5286 | if (!only_parity(rdev->raid_disk, | |
5287 | conf->algorithm, | |
5288 | conf->raid_disks, | |
5289 | conf->max_degraded)) | |
5290 | continue; | |
5291 | } | |
5292 | if (!only_parity(rdev->raid_disk, | |
5293 | conf->prev_algo, | |
5294 | conf->previous_raid_disks, | |
5295 | conf->max_degraded)) | |
5296 | continue; | |
5297 | dirty_parity_disks++; | |
5298 | } | |
91adb564 | 5299 | |
17045f52 N |
5300 | /* |
5301 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
5302 | */ | |
908f4fbd | 5303 | mddev->degraded = calc_degraded(conf); |
91adb564 | 5304 | |
674806d6 | 5305 | if (has_failed(conf)) { |
0c55e022 | 5306 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 5307 | " (%d/%d failed)\n", |
02c2de8c | 5308 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
5309 | goto abort; |
5310 | } | |
5311 | ||
91adb564 | 5312 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 5313 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
5314 | mddev->resync_max_sectors = mddev->dev_sectors; |
5315 | ||
c148ffdc | 5316 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 5317 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
5318 | if (mddev->ok_start_degraded) |
5319 | printk(KERN_WARNING | |
0c55e022 N |
5320 | "md/raid:%s: starting dirty degraded array" |
5321 | " - data corruption possible.\n", | |
6ff8d8ec N |
5322 | mdname(mddev)); |
5323 | else { | |
5324 | printk(KERN_ERR | |
0c55e022 | 5325 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
5326 | mdname(mddev)); |
5327 | goto abort; | |
5328 | } | |
1da177e4 LT |
5329 | } |
5330 | ||
1da177e4 | 5331 | if (mddev->degraded == 0) |
0c55e022 N |
5332 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
5333 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
5334 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
5335 | mddev->new_layout); | |
1da177e4 | 5336 | else |
0c55e022 N |
5337 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
5338 | " out of %d devices, algorithm %d\n", | |
5339 | mdname(mddev), conf->level, | |
5340 | mddev->raid_disks - mddev->degraded, | |
5341 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
5342 | |
5343 | print_raid5_conf(conf); | |
5344 | ||
fef9c61f | 5345 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 5346 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5347 | atomic_set(&conf->reshape_stripes, 0); |
5348 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5349 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5350 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5351 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5352 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5353 | "reshape"); |
f6705578 N |
5354 | } |
5355 | ||
1da177e4 LT |
5356 | |
5357 | /* Ok, everything is just fine now */ | |
a64c876f N |
5358 | if (mddev->to_remove == &raid5_attrs_group) |
5359 | mddev->to_remove = NULL; | |
00bcb4ac N |
5360 | else if (mddev->kobj.sd && |
5361 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 5362 | printk(KERN_WARNING |
4a5add49 | 5363 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 5364 | mdname(mddev)); |
4a5add49 | 5365 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5366 | |
4a5add49 | 5367 | if (mddev->queue) { |
9f7c2220 | 5368 | int chunk_size; |
4a5add49 N |
5369 | /* read-ahead size must cover two whole stripes, which |
5370 | * is 2 * (datadisks) * chunksize where 'n' is the | |
5371 | * number of raid devices | |
5372 | */ | |
5373 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
5374 | int stripe = data_disks * | |
5375 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
5376 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
5377 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 5378 | |
4a5add49 | 5379 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 5380 | |
11d8a6e3 N |
5381 | mddev->queue->backing_dev_info.congested_data = mddev; |
5382 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 5383 | |
9f7c2220 N |
5384 | chunk_size = mddev->chunk_sectors << 9; |
5385 | blk_queue_io_min(mddev->queue, chunk_size); | |
5386 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5387 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 5388 | |
05616be5 | 5389 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
5390 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
5391 | rdev->data_offset << 9); | |
05616be5 N |
5392 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
5393 | rdev->new_data_offset << 9); | |
5394 | } | |
9f7c2220 | 5395 | } |
23032a0e | 5396 | |
1da177e4 LT |
5397 | return 0; |
5398 | abort: | |
01f96c0a | 5399 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
5400 | print_raid5_conf(conf); |
5401 | free_conf(conf); | |
1da177e4 | 5402 | mddev->private = NULL; |
0c55e022 | 5403 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
5404 | return -EIO; |
5405 | } | |
5406 | ||
fd01b88c | 5407 | static int stop(struct mddev *mddev) |
1da177e4 | 5408 | { |
d1688a6d | 5409 | struct r5conf *conf = mddev->private; |
1da177e4 | 5410 | |
01f96c0a | 5411 | md_unregister_thread(&mddev->thread); |
11d8a6e3 N |
5412 | if (mddev->queue) |
5413 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 5414 | free_conf(conf); |
a64c876f N |
5415 | mddev->private = NULL; |
5416 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
5417 | return 0; |
5418 | } | |
5419 | ||
fd01b88c | 5420 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 5421 | { |
d1688a6d | 5422 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
5423 | int i; |
5424 | ||
9d8f0363 AN |
5425 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5426 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5427 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5428 | for (i = 0; i < conf->raid_disks; i++) |
5429 | seq_printf (seq, "%s", | |
5430 | conf->disks[i].rdev && | |
b2d444d7 | 5431 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5432 | seq_printf (seq, "]"); |
1da177e4 LT |
5433 | } |
5434 | ||
d1688a6d | 5435 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
5436 | { |
5437 | int i; | |
5438 | struct disk_info *tmp; | |
5439 | ||
0c55e022 | 5440 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
5441 | if (!conf) { |
5442 | printk("(conf==NULL)\n"); | |
5443 | return; | |
5444 | } | |
0c55e022 N |
5445 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
5446 | conf->raid_disks, | |
5447 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5448 | |
5449 | for (i = 0; i < conf->raid_disks; i++) { | |
5450 | char b[BDEVNAME_SIZE]; | |
5451 | tmp = conf->disks + i; | |
5452 | if (tmp->rdev) | |
0c55e022 N |
5453 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
5454 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
5455 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
5456 | } |
5457 | } | |
5458 | ||
fd01b88c | 5459 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
5460 | { |
5461 | int i; | |
d1688a6d | 5462 | struct r5conf *conf = mddev->private; |
1da177e4 | 5463 | struct disk_info *tmp; |
6b965620 N |
5464 | int count = 0; |
5465 | unsigned long flags; | |
1da177e4 LT |
5466 | |
5467 | for (i = 0; i < conf->raid_disks; i++) { | |
5468 | tmp = conf->disks + i; | |
dd054fce N |
5469 | if (tmp->replacement |
5470 | && tmp->replacement->recovery_offset == MaxSector | |
5471 | && !test_bit(Faulty, &tmp->replacement->flags) | |
5472 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
5473 | /* Replacement has just become active. */ | |
5474 | if (!tmp->rdev | |
5475 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
5476 | count++; | |
5477 | if (tmp->rdev) { | |
5478 | /* Replaced device not technically faulty, | |
5479 | * but we need to be sure it gets removed | |
5480 | * and never re-added. | |
5481 | */ | |
5482 | set_bit(Faulty, &tmp->rdev->flags); | |
5483 | sysfs_notify_dirent_safe( | |
5484 | tmp->rdev->sysfs_state); | |
5485 | } | |
5486 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
5487 | } else if (tmp->rdev | |
70fffd0b | 5488 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 5489 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 5490 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 5491 | count++; |
43c73ca4 | 5492 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
5493 | } |
5494 | } | |
6b965620 | 5495 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5496 | mddev->degraded = calc_degraded(conf); |
6b965620 | 5497 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 5498 | print_raid5_conf(conf); |
6b965620 | 5499 | return count; |
1da177e4 LT |
5500 | } |
5501 | ||
b8321b68 | 5502 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5503 | { |
d1688a6d | 5504 | struct r5conf *conf = mddev->private; |
1da177e4 | 5505 | int err = 0; |
b8321b68 | 5506 | int number = rdev->raid_disk; |
657e3e4d | 5507 | struct md_rdev **rdevp; |
1da177e4 LT |
5508 | struct disk_info *p = conf->disks + number; |
5509 | ||
5510 | print_raid5_conf(conf); | |
657e3e4d N |
5511 | if (rdev == p->rdev) |
5512 | rdevp = &p->rdev; | |
5513 | else if (rdev == p->replacement) | |
5514 | rdevp = &p->replacement; | |
5515 | else | |
5516 | return 0; | |
5517 | ||
5518 | if (number >= conf->raid_disks && | |
5519 | conf->reshape_progress == MaxSector) | |
5520 | clear_bit(In_sync, &rdev->flags); | |
5521 | ||
5522 | if (test_bit(In_sync, &rdev->flags) || | |
5523 | atomic_read(&rdev->nr_pending)) { | |
5524 | err = -EBUSY; | |
5525 | goto abort; | |
5526 | } | |
5527 | /* Only remove non-faulty devices if recovery | |
5528 | * isn't possible. | |
5529 | */ | |
5530 | if (!test_bit(Faulty, &rdev->flags) && | |
5531 | mddev->recovery_disabled != conf->recovery_disabled && | |
5532 | !has_failed(conf) && | |
dd054fce | 5533 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
5534 | number < conf->raid_disks) { |
5535 | err = -EBUSY; | |
5536 | goto abort; | |
5537 | } | |
5538 | *rdevp = NULL; | |
5539 | synchronize_rcu(); | |
5540 | if (atomic_read(&rdev->nr_pending)) { | |
5541 | /* lost the race, try later */ | |
5542 | err = -EBUSY; | |
5543 | *rdevp = rdev; | |
dd054fce N |
5544 | } else if (p->replacement) { |
5545 | /* We must have just cleared 'rdev' */ | |
5546 | p->rdev = p->replacement; | |
5547 | clear_bit(Replacement, &p->replacement->flags); | |
5548 | smp_mb(); /* Make sure other CPUs may see both as identical | |
5549 | * but will never see neither - if they are careful | |
5550 | */ | |
5551 | p->replacement = NULL; | |
5552 | clear_bit(WantReplacement, &rdev->flags); | |
5553 | } else | |
5554 | /* We might have just removed the Replacement as faulty- | |
5555 | * clear the bit just in case | |
5556 | */ | |
5557 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
5558 | abort: |
5559 | ||
5560 | print_raid5_conf(conf); | |
5561 | return err; | |
5562 | } | |
5563 | ||
fd01b88c | 5564 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5565 | { |
d1688a6d | 5566 | struct r5conf *conf = mddev->private; |
199050ea | 5567 | int err = -EEXIST; |
1da177e4 LT |
5568 | int disk; |
5569 | struct disk_info *p; | |
6c2fce2e NB |
5570 | int first = 0; |
5571 | int last = conf->raid_disks - 1; | |
1da177e4 | 5572 | |
7f0da59b N |
5573 | if (mddev->recovery_disabled == conf->recovery_disabled) |
5574 | return -EBUSY; | |
5575 | ||
dc10c643 | 5576 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 5577 | /* no point adding a device */ |
199050ea | 5578 | return -EINVAL; |
1da177e4 | 5579 | |
6c2fce2e NB |
5580 | if (rdev->raid_disk >= 0) |
5581 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5582 | |
5583 | /* | |
16a53ecc N |
5584 | * find the disk ... but prefer rdev->saved_raid_disk |
5585 | * if possible. | |
1da177e4 | 5586 | */ |
16a53ecc | 5587 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5588 | rdev->saved_raid_disk >= first && |
16a53ecc | 5589 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
5590 | first = rdev->saved_raid_disk; |
5591 | ||
5592 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
5593 | p = conf->disks + disk; |
5594 | if (p->rdev == NULL) { | |
b2d444d7 | 5595 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5596 | rdev->raid_disk = disk; |
199050ea | 5597 | err = 0; |
72626685 N |
5598 | if (rdev->saved_raid_disk != disk) |
5599 | conf->fullsync = 1; | |
d6065f7b | 5600 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 5601 | goto out; |
1da177e4 | 5602 | } |
5cfb22a1 N |
5603 | } |
5604 | for (disk = first; disk <= last; disk++) { | |
5605 | p = conf->disks + disk; | |
7bfec5f3 N |
5606 | if (test_bit(WantReplacement, &p->rdev->flags) && |
5607 | p->replacement == NULL) { | |
5608 | clear_bit(In_sync, &rdev->flags); | |
5609 | set_bit(Replacement, &rdev->flags); | |
5610 | rdev->raid_disk = disk; | |
5611 | err = 0; | |
5612 | conf->fullsync = 1; | |
5613 | rcu_assign_pointer(p->replacement, rdev); | |
5614 | break; | |
5615 | } | |
5616 | } | |
5cfb22a1 | 5617 | out: |
1da177e4 | 5618 | print_raid5_conf(conf); |
199050ea | 5619 | return err; |
1da177e4 LT |
5620 | } |
5621 | ||
fd01b88c | 5622 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
5623 | { |
5624 | /* no resync is happening, and there is enough space | |
5625 | * on all devices, so we can resize. | |
5626 | * We need to make sure resync covers any new space. | |
5627 | * If the array is shrinking we should possibly wait until | |
5628 | * any io in the removed space completes, but it hardly seems | |
5629 | * worth it. | |
5630 | */ | |
a4a6125a | 5631 | sector_t newsize; |
9d8f0363 | 5632 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
a4a6125a N |
5633 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
5634 | if (mddev->external_size && | |
5635 | mddev->array_sectors > newsize) | |
b522adcd | 5636 | return -EINVAL; |
a4a6125a N |
5637 | if (mddev->bitmap) { |
5638 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
5639 | if (ret) | |
5640 | return ret; | |
5641 | } | |
5642 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 5643 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5644 | revalidate_disk(mddev->gendisk); |
b098636c N |
5645 | if (sectors > mddev->dev_sectors && |
5646 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5647 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5648 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5649 | } | |
58c0fed4 | 5650 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5651 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5652 | return 0; |
5653 | } | |
5654 | ||
fd01b88c | 5655 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
5656 | { |
5657 | /* Can only proceed if there are plenty of stripe_heads. | |
5658 | * We need a minimum of one full stripe,, and for sensible progress | |
5659 | * it is best to have about 4 times that. | |
5660 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5661 | * allow for chunk sizes up to 256K, which is probably OK. | |
5662 | * If the chunk size is greater, user-space should request more | |
5663 | * stripe_heads first. | |
5664 | */ | |
d1688a6d | 5665 | struct r5conf *conf = mddev->private; |
01ee22b4 N |
5666 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
5667 | > conf->max_nr_stripes || | |
5668 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5669 | > conf->max_nr_stripes) { | |
0c55e022 N |
5670 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5671 | mdname(mddev), | |
01ee22b4 N |
5672 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5673 | / STRIPE_SIZE)*4); | |
5674 | return 0; | |
5675 | } | |
5676 | return 1; | |
5677 | } | |
5678 | ||
fd01b88c | 5679 | static int check_reshape(struct mddev *mddev) |
29269553 | 5680 | { |
d1688a6d | 5681 | struct r5conf *conf = mddev->private; |
29269553 | 5682 | |
88ce4930 N |
5683 | if (mddev->delta_disks == 0 && |
5684 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5685 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5686 | return 0; /* nothing to do */ |
674806d6 | 5687 | if (has_failed(conf)) |
ec32a2bd N |
5688 | return -EINVAL; |
5689 | if (mddev->delta_disks < 0) { | |
5690 | /* We might be able to shrink, but the devices must | |
5691 | * be made bigger first. | |
5692 | * For raid6, 4 is the minimum size. | |
5693 | * Otherwise 2 is the minimum | |
5694 | */ | |
5695 | int min = 2; | |
5696 | if (mddev->level == 6) | |
5697 | min = 4; | |
5698 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5699 | return -EINVAL; | |
5700 | } | |
29269553 | 5701 | |
01ee22b4 | 5702 | if (!check_stripe_cache(mddev)) |
29269553 | 5703 | return -ENOSPC; |
29269553 | 5704 | |
ec32a2bd | 5705 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5706 | } |
5707 | ||
fd01b88c | 5708 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 5709 | { |
d1688a6d | 5710 | struct r5conf *conf = mddev->private; |
3cb03002 | 5711 | struct md_rdev *rdev; |
63c70c4f | 5712 | int spares = 0; |
c04be0aa | 5713 | unsigned long flags; |
63c70c4f | 5714 | |
f416885e | 5715 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5716 | return -EBUSY; |
5717 | ||
01ee22b4 N |
5718 | if (!check_stripe_cache(mddev)) |
5719 | return -ENOSPC; | |
5720 | ||
30b67645 N |
5721 | if (has_failed(conf)) |
5722 | return -EINVAL; | |
5723 | ||
c6563a8c | 5724 | rdev_for_each(rdev, mddev) { |
469518a3 N |
5725 | if (!test_bit(In_sync, &rdev->flags) |
5726 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5727 | spares++; |
c6563a8c | 5728 | } |
63c70c4f | 5729 | |
f416885e | 5730 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5731 | /* Not enough devices even to make a degraded array |
5732 | * of that size | |
5733 | */ | |
5734 | return -EINVAL; | |
5735 | ||
ec32a2bd N |
5736 | /* Refuse to reduce size of the array. Any reductions in |
5737 | * array size must be through explicit setting of array_size | |
5738 | * attribute. | |
5739 | */ | |
5740 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5741 | < mddev->array_sectors) { | |
0c55e022 | 5742 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5743 | "before number of disks\n", mdname(mddev)); |
5744 | return -EINVAL; | |
5745 | } | |
5746 | ||
f6705578 | 5747 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5748 | spin_lock_irq(&conf->device_lock); |
5749 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5750 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5751 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5752 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5753 | conf->prev_algo = conf->algorithm; |
5754 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
5755 | conf->generation++; |
5756 | /* Code that selects data_offset needs to see the generation update | |
5757 | * if reshape_progress has been set - so a memory barrier needed. | |
5758 | */ | |
5759 | smp_mb(); | |
2c810cdd | 5760 | if (mddev->reshape_backwards) |
fef9c61f N |
5761 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
5762 | else | |
5763 | conf->reshape_progress = 0; | |
5764 | conf->reshape_safe = conf->reshape_progress; | |
29269553 N |
5765 | spin_unlock_irq(&conf->device_lock); |
5766 | ||
5767 | /* Add some new drives, as many as will fit. | |
5768 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5769 | * Don't add devices if we are reducing the number of |
5770 | * devices in the array. This is because it is not possible | |
5771 | * to correctly record the "partially reconstructed" state of | |
5772 | * such devices during the reshape and confusion could result. | |
29269553 | 5773 | */ |
87a8dec9 | 5774 | if (mddev->delta_disks >= 0) { |
dafb20fa | 5775 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
5776 | if (rdev->raid_disk < 0 && |
5777 | !test_bit(Faulty, &rdev->flags)) { | |
5778 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 5779 | if (rdev->raid_disk |
9d4c7d87 | 5780 | >= conf->previous_raid_disks) |
87a8dec9 | 5781 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 5782 | else |
87a8dec9 | 5783 | rdev->recovery_offset = 0; |
36fad858 NK |
5784 | |
5785 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 5786 | /* Failure here is OK */; |
50da0840 | 5787 | } |
87a8dec9 N |
5788 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5789 | && !test_bit(Faulty, &rdev->flags)) { | |
5790 | /* This is a spare that was manually added */ | |
5791 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 5792 | } |
29269553 | 5793 | |
87a8dec9 N |
5794 | /* When a reshape changes the number of devices, |
5795 | * ->degraded is measured against the larger of the | |
5796 | * pre and post number of devices. | |
5797 | */ | |
ec32a2bd | 5798 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5799 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
5800 | spin_unlock_irqrestore(&conf->device_lock, flags); |
5801 | } | |
63c70c4f | 5802 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5803 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5804 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5805 | |
29269553 N |
5806 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5807 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5808 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5809 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5810 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5811 | "reshape"); |
29269553 N |
5812 | if (!mddev->sync_thread) { |
5813 | mddev->recovery = 0; | |
5814 | spin_lock_irq(&conf->device_lock); | |
5815 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
05616be5 N |
5816 | rdev_for_each(rdev, mddev) |
5817 | rdev->new_data_offset = rdev->data_offset; | |
5818 | smp_wmb(); | |
fef9c61f | 5819 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 5820 | mddev->reshape_position = MaxSector; |
29269553 N |
5821 | spin_unlock_irq(&conf->device_lock); |
5822 | return -EAGAIN; | |
5823 | } | |
c8f517c4 | 5824 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5825 | md_wakeup_thread(mddev->sync_thread); |
5826 | md_new_event(mddev); | |
5827 | return 0; | |
5828 | } | |
29269553 | 5829 | |
ec32a2bd N |
5830 | /* This is called from the reshape thread and should make any |
5831 | * changes needed in 'conf' | |
5832 | */ | |
d1688a6d | 5833 | static void end_reshape(struct r5conf *conf) |
29269553 | 5834 | { |
29269553 | 5835 | |
f6705578 | 5836 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 5837 | struct md_rdev *rdev; |
f6705578 | 5838 | |
f6705578 | 5839 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5840 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
5841 | rdev_for_each(rdev, conf->mddev) |
5842 | rdev->data_offset = rdev->new_data_offset; | |
5843 | smp_wmb(); | |
fef9c61f | 5844 | conf->reshape_progress = MaxSector; |
f6705578 | 5845 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5846 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5847 | |
5848 | /* read-ahead size must cover two whole stripes, which is | |
5849 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5850 | */ | |
4a5add49 | 5851 | if (conf->mddev->queue) { |
cea9c228 | 5852 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5853 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5854 | / PAGE_SIZE); |
16a53ecc N |
5855 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5856 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5857 | } | |
29269553 | 5858 | } |
29269553 N |
5859 | } |
5860 | ||
ec32a2bd N |
5861 | /* This is called from the raid5d thread with mddev_lock held. |
5862 | * It makes config changes to the device. | |
5863 | */ | |
fd01b88c | 5864 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 5865 | { |
d1688a6d | 5866 | struct r5conf *conf = mddev->private; |
cea9c228 N |
5867 | |
5868 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5869 | ||
ec32a2bd N |
5870 | if (mddev->delta_disks > 0) { |
5871 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5872 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5873 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5874 | } else { |
5875 | int d; | |
908f4fbd N |
5876 | spin_lock_irq(&conf->device_lock); |
5877 | mddev->degraded = calc_degraded(conf); | |
5878 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
5879 | for (d = conf->raid_disks ; |
5880 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 5881 | d++) { |
3cb03002 | 5882 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
5883 | if (rdev) |
5884 | clear_bit(In_sync, &rdev->flags); | |
5885 | rdev = conf->disks[d].replacement; | |
5886 | if (rdev) | |
5887 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 5888 | } |
cea9c228 | 5889 | } |
88ce4930 | 5890 | mddev->layout = conf->algorithm; |
09c9e5fa | 5891 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5892 | mddev->reshape_position = MaxSector; |
5893 | mddev->delta_disks = 0; | |
2c810cdd | 5894 | mddev->reshape_backwards = 0; |
cea9c228 N |
5895 | } |
5896 | } | |
5897 | ||
fd01b88c | 5898 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 5899 | { |
d1688a6d | 5900 | struct r5conf *conf = mddev->private; |
72626685 N |
5901 | |
5902 | switch(state) { | |
e464eafd N |
5903 | case 2: /* resume for a suspend */ |
5904 | wake_up(&conf->wait_for_overlap); | |
5905 | break; | |
5906 | ||
72626685 N |
5907 | case 1: /* stop all writes */ |
5908 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5909 | /* '2' tells resync/reshape to pause so that all |
5910 | * active stripes can drain | |
5911 | */ | |
5912 | conf->quiesce = 2; | |
72626685 | 5913 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5914 | atomic_read(&conf->active_stripes) == 0 && |
5915 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5916 | conf->device_lock, /* nothing */); |
64bd660b | 5917 | conf->quiesce = 1; |
72626685 | 5918 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5919 | /* allow reshape to continue */ |
5920 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5921 | break; |
5922 | ||
5923 | case 0: /* re-enable writes */ | |
5924 | spin_lock_irq(&conf->device_lock); | |
5925 | conf->quiesce = 0; | |
5926 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5927 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5928 | spin_unlock_irq(&conf->device_lock); |
5929 | break; | |
5930 | } | |
72626685 | 5931 | } |
b15c2e57 | 5932 | |
d562b0c4 | 5933 | |
fd01b88c | 5934 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 5935 | { |
e373ab10 | 5936 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 5937 | sector_t sectors; |
54071b38 | 5938 | |
f1b29bca | 5939 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 5940 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
5941 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5942 | mdname(mddev)); | |
f1b29bca DW |
5943 | return ERR_PTR(-EINVAL); |
5944 | } | |
5945 | ||
e373ab10 N |
5946 | sectors = raid0_conf->strip_zone[0].zone_end; |
5947 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 5948 | mddev->dev_sectors = sectors; |
f1b29bca | 5949 | mddev->new_level = level; |
54071b38 TM |
5950 | mddev->new_layout = ALGORITHM_PARITY_N; |
5951 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5952 | mddev->raid_disks += 1; | |
5953 | mddev->delta_disks = 1; | |
5954 | /* make sure it will be not marked as dirty */ | |
5955 | mddev->recovery_cp = MaxSector; | |
5956 | ||
5957 | return setup_conf(mddev); | |
5958 | } | |
5959 | ||
5960 | ||
fd01b88c | 5961 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
5962 | { |
5963 | int chunksect; | |
5964 | ||
5965 | if (mddev->raid_disks != 2 || | |
5966 | mddev->degraded > 1) | |
5967 | return ERR_PTR(-EINVAL); | |
5968 | ||
5969 | /* Should check if there are write-behind devices? */ | |
5970 | ||
5971 | chunksect = 64*2; /* 64K by default */ | |
5972 | ||
5973 | /* The array must be an exact multiple of chunksize */ | |
5974 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5975 | chunksect >>= 1; | |
5976 | ||
5977 | if ((chunksect<<9) < STRIPE_SIZE) | |
5978 | /* array size does not allow a suitable chunk size */ | |
5979 | return ERR_PTR(-EINVAL); | |
5980 | ||
5981 | mddev->new_level = 5; | |
5982 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5983 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5984 | |
5985 | return setup_conf(mddev); | |
5986 | } | |
5987 | ||
fd01b88c | 5988 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
5989 | { |
5990 | int new_layout; | |
5991 | ||
5992 | switch (mddev->layout) { | |
5993 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5994 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5995 | break; | |
5996 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5997 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5998 | break; | |
5999 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6000 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
6001 | break; | |
6002 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6003 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
6004 | break; | |
6005 | case ALGORITHM_PARITY_0_6: | |
6006 | new_layout = ALGORITHM_PARITY_0; | |
6007 | break; | |
6008 | case ALGORITHM_PARITY_N: | |
6009 | new_layout = ALGORITHM_PARITY_N; | |
6010 | break; | |
6011 | default: | |
6012 | return ERR_PTR(-EINVAL); | |
6013 | } | |
6014 | mddev->new_level = 5; | |
6015 | mddev->new_layout = new_layout; | |
6016 | mddev->delta_disks = -1; | |
6017 | mddev->raid_disks -= 1; | |
6018 | return setup_conf(mddev); | |
6019 | } | |
6020 | ||
d562b0c4 | 6021 | |
fd01b88c | 6022 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 6023 | { |
88ce4930 N |
6024 | /* For a 2-drive array, the layout and chunk size can be changed |
6025 | * immediately as not restriping is needed. | |
6026 | * For larger arrays we record the new value - after validation | |
6027 | * to be used by a reshape pass. | |
b3546035 | 6028 | */ |
d1688a6d | 6029 | struct r5conf *conf = mddev->private; |
597a711b | 6030 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 6031 | |
597a711b | 6032 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
6033 | return -EINVAL; |
6034 | if (new_chunk > 0) { | |
0ba459d2 | 6035 | if (!is_power_of_2(new_chunk)) |
b3546035 | 6036 | return -EINVAL; |
597a711b | 6037 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 6038 | return -EINVAL; |
597a711b | 6039 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
6040 | /* not factor of array size */ |
6041 | return -EINVAL; | |
6042 | } | |
6043 | ||
6044 | /* They look valid */ | |
6045 | ||
88ce4930 | 6046 | if (mddev->raid_disks == 2) { |
597a711b N |
6047 | /* can make the change immediately */ |
6048 | if (mddev->new_layout >= 0) { | |
6049 | conf->algorithm = mddev->new_layout; | |
6050 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
6051 | } |
6052 | if (new_chunk > 0) { | |
597a711b N |
6053 | conf->chunk_sectors = new_chunk ; |
6054 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
6055 | } |
6056 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
6057 | md_wakeup_thread(mddev->thread); | |
b3546035 | 6058 | } |
50ac168a | 6059 | return check_reshape(mddev); |
88ce4930 N |
6060 | } |
6061 | ||
fd01b88c | 6062 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 6063 | { |
597a711b | 6064 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 6065 | |
597a711b | 6066 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 6067 | return -EINVAL; |
b3546035 | 6068 | if (new_chunk > 0) { |
0ba459d2 | 6069 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 6070 | return -EINVAL; |
597a711b | 6071 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 6072 | return -EINVAL; |
597a711b | 6073 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
6074 | /* not factor of array size */ |
6075 | return -EINVAL; | |
b3546035 | 6076 | } |
88ce4930 N |
6077 | |
6078 | /* They look valid */ | |
50ac168a | 6079 | return check_reshape(mddev); |
b3546035 N |
6080 | } |
6081 | ||
fd01b88c | 6082 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
6083 | { |
6084 | /* raid5 can take over: | |
f1b29bca | 6085 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
6086 | * raid1 - if there are two drives. We need to know the chunk size |
6087 | * raid4 - trivial - just use a raid4 layout. | |
6088 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 6089 | */ |
f1b29bca DW |
6090 | if (mddev->level == 0) |
6091 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
6092 | if (mddev->level == 1) |
6093 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
6094 | if (mddev->level == 4) { |
6095 | mddev->new_layout = ALGORITHM_PARITY_N; | |
6096 | mddev->new_level = 5; | |
6097 | return setup_conf(mddev); | |
6098 | } | |
fc9739c6 N |
6099 | if (mddev->level == 6) |
6100 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
6101 | |
6102 | return ERR_PTR(-EINVAL); | |
6103 | } | |
6104 | ||
fd01b88c | 6105 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 6106 | { |
f1b29bca DW |
6107 | /* raid4 can take over: |
6108 | * raid0 - if there is only one strip zone | |
6109 | * raid5 - if layout is right | |
a78d38a1 | 6110 | */ |
f1b29bca DW |
6111 | if (mddev->level == 0) |
6112 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
6113 | if (mddev->level == 5 && |
6114 | mddev->layout == ALGORITHM_PARITY_N) { | |
6115 | mddev->new_layout = 0; | |
6116 | mddev->new_level = 4; | |
6117 | return setup_conf(mddev); | |
6118 | } | |
6119 | return ERR_PTR(-EINVAL); | |
6120 | } | |
d562b0c4 | 6121 | |
84fc4b56 | 6122 | static struct md_personality raid5_personality; |
245f46c2 | 6123 | |
fd01b88c | 6124 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
6125 | { |
6126 | /* Currently can only take over a raid5. We map the | |
6127 | * personality to an equivalent raid6 personality | |
6128 | * with the Q block at the end. | |
6129 | */ | |
6130 | int new_layout; | |
6131 | ||
6132 | if (mddev->pers != &raid5_personality) | |
6133 | return ERR_PTR(-EINVAL); | |
6134 | if (mddev->degraded > 1) | |
6135 | return ERR_PTR(-EINVAL); | |
6136 | if (mddev->raid_disks > 253) | |
6137 | return ERR_PTR(-EINVAL); | |
6138 | if (mddev->raid_disks < 3) | |
6139 | return ERR_PTR(-EINVAL); | |
6140 | ||
6141 | switch (mddev->layout) { | |
6142 | case ALGORITHM_LEFT_ASYMMETRIC: | |
6143 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
6144 | break; | |
6145 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6146 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
6147 | break; | |
6148 | case ALGORITHM_LEFT_SYMMETRIC: | |
6149 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
6150 | break; | |
6151 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6152 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
6153 | break; | |
6154 | case ALGORITHM_PARITY_0: | |
6155 | new_layout = ALGORITHM_PARITY_0_6; | |
6156 | break; | |
6157 | case ALGORITHM_PARITY_N: | |
6158 | new_layout = ALGORITHM_PARITY_N; | |
6159 | break; | |
6160 | default: | |
6161 | return ERR_PTR(-EINVAL); | |
6162 | } | |
6163 | mddev->new_level = 6; | |
6164 | mddev->new_layout = new_layout; | |
6165 | mddev->delta_disks = 1; | |
6166 | mddev->raid_disks += 1; | |
6167 | return setup_conf(mddev); | |
6168 | } | |
6169 | ||
6170 | ||
84fc4b56 | 6171 | static struct md_personality raid6_personality = |
16a53ecc N |
6172 | { |
6173 | .name = "raid6", | |
6174 | .level = 6, | |
6175 | .owner = THIS_MODULE, | |
6176 | .make_request = make_request, | |
6177 | .run = run, | |
6178 | .stop = stop, | |
6179 | .status = status, | |
6180 | .error_handler = error, | |
6181 | .hot_add_disk = raid5_add_disk, | |
6182 | .hot_remove_disk= raid5_remove_disk, | |
6183 | .spare_active = raid5_spare_active, | |
6184 | .sync_request = sync_request, | |
6185 | .resize = raid5_resize, | |
80c3a6ce | 6186 | .size = raid5_size, |
50ac168a | 6187 | .check_reshape = raid6_check_reshape, |
f416885e | 6188 | .start_reshape = raid5_start_reshape, |
cea9c228 | 6189 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 6190 | .quiesce = raid5_quiesce, |
245f46c2 | 6191 | .takeover = raid6_takeover, |
16a53ecc | 6192 | }; |
84fc4b56 | 6193 | static struct md_personality raid5_personality = |
1da177e4 LT |
6194 | { |
6195 | .name = "raid5", | |
2604b703 | 6196 | .level = 5, |
1da177e4 LT |
6197 | .owner = THIS_MODULE, |
6198 | .make_request = make_request, | |
6199 | .run = run, | |
6200 | .stop = stop, | |
6201 | .status = status, | |
6202 | .error_handler = error, | |
6203 | .hot_add_disk = raid5_add_disk, | |
6204 | .hot_remove_disk= raid5_remove_disk, | |
6205 | .spare_active = raid5_spare_active, | |
6206 | .sync_request = sync_request, | |
6207 | .resize = raid5_resize, | |
80c3a6ce | 6208 | .size = raid5_size, |
63c70c4f N |
6209 | .check_reshape = raid5_check_reshape, |
6210 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 6211 | .finish_reshape = raid5_finish_reshape, |
72626685 | 6212 | .quiesce = raid5_quiesce, |
d562b0c4 | 6213 | .takeover = raid5_takeover, |
1da177e4 LT |
6214 | }; |
6215 | ||
84fc4b56 | 6216 | static struct md_personality raid4_personality = |
1da177e4 | 6217 | { |
2604b703 N |
6218 | .name = "raid4", |
6219 | .level = 4, | |
6220 | .owner = THIS_MODULE, | |
6221 | .make_request = make_request, | |
6222 | .run = run, | |
6223 | .stop = stop, | |
6224 | .status = status, | |
6225 | .error_handler = error, | |
6226 | .hot_add_disk = raid5_add_disk, | |
6227 | .hot_remove_disk= raid5_remove_disk, | |
6228 | .spare_active = raid5_spare_active, | |
6229 | .sync_request = sync_request, | |
6230 | .resize = raid5_resize, | |
80c3a6ce | 6231 | .size = raid5_size, |
3d37890b N |
6232 | .check_reshape = raid5_check_reshape, |
6233 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 6234 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 6235 | .quiesce = raid5_quiesce, |
a78d38a1 | 6236 | .takeover = raid4_takeover, |
2604b703 N |
6237 | }; |
6238 | ||
6239 | static int __init raid5_init(void) | |
6240 | { | |
16a53ecc | 6241 | register_md_personality(&raid6_personality); |
2604b703 N |
6242 | register_md_personality(&raid5_personality); |
6243 | register_md_personality(&raid4_personality); | |
6244 | return 0; | |
1da177e4 LT |
6245 | } |
6246 | ||
2604b703 | 6247 | static void raid5_exit(void) |
1da177e4 | 6248 | { |
16a53ecc | 6249 | unregister_md_personality(&raid6_personality); |
2604b703 N |
6250 | unregister_md_personality(&raid5_personality); |
6251 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
6252 | } |
6253 | ||
6254 | module_init(raid5_init); | |
6255 | module_exit(raid5_exit); | |
6256 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 6257 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 6258 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
6259 | MODULE_ALIAS("md-raid5"); |
6260 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
6261 | MODULE_ALIAS("md-level-5"); |
6262 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
6263 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
6264 | MODULE_ALIAS("md-raid6"); | |
6265 | MODULE_ALIAS("md-level-6"); | |
6266 | ||
6267 | /* This used to be two separate modules, they were: */ | |
6268 | MODULE_ALIAS("raid5"); | |
6269 | MODULE_ALIAS("raid6"); |