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); |
674806d6 N |
396 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
397 | degraded++; | |
398 | else if (test_bit(In_sync, &rdev->flags)) | |
399 | ; | |
400 | else | |
401 | /* not in-sync or faulty. | |
402 | * If the reshape increases the number of devices, | |
403 | * this is being recovered by the reshape, so | |
404 | * this 'previous' section is not in_sync. | |
405 | * If the number of devices is being reduced however, | |
406 | * the device can only be part of the array if | |
407 | * we are reverting a reshape, so this section will | |
408 | * be in-sync. | |
409 | */ | |
410 | if (conf->raid_disks >= conf->previous_raid_disks) | |
411 | degraded++; | |
412 | } | |
413 | rcu_read_unlock(); | |
908f4fbd N |
414 | if (conf->raid_disks == conf->previous_raid_disks) |
415 | return degraded; | |
674806d6 | 416 | rcu_read_lock(); |
908f4fbd | 417 | degraded2 = 0; |
674806d6 | 418 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 419 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
674806d6 | 420 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 421 | degraded2++; |
674806d6 N |
422 | else if (test_bit(In_sync, &rdev->flags)) |
423 | ; | |
424 | else | |
425 | /* not in-sync or faulty. | |
426 | * If reshape increases the number of devices, this | |
427 | * section has already been recovered, else it | |
428 | * almost certainly hasn't. | |
429 | */ | |
430 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 431 | degraded2++; |
674806d6 N |
432 | } |
433 | rcu_read_unlock(); | |
908f4fbd N |
434 | if (degraded2 > degraded) |
435 | return degraded2; | |
436 | return degraded; | |
437 | } | |
438 | ||
439 | static int has_failed(struct r5conf *conf) | |
440 | { | |
441 | int degraded; | |
442 | ||
443 | if (conf->mddev->reshape_position == MaxSector) | |
444 | return conf->mddev->degraded > conf->max_degraded; | |
445 | ||
446 | degraded = calc_degraded(conf); | |
674806d6 N |
447 | if (degraded > conf->max_degraded) |
448 | return 1; | |
449 | return 0; | |
450 | } | |
451 | ||
b5663ba4 | 452 | static struct stripe_head * |
d1688a6d | 453 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 454 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
455 | { |
456 | struct stripe_head *sh; | |
457 | ||
45b4233c | 458 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
459 | |
460 | spin_lock_irq(&conf->device_lock); | |
461 | ||
462 | do { | |
72626685 | 463 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 464 | conf->quiesce == 0 || noquiesce, |
72626685 | 465 | conf->device_lock, /* nothing */); |
86b42c71 | 466 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
467 | if (!sh) { |
468 | if (!conf->inactive_blocked) | |
469 | sh = get_free_stripe(conf); | |
470 | if (noblock && sh == NULL) | |
471 | break; | |
472 | if (!sh) { | |
473 | conf->inactive_blocked = 1; | |
474 | wait_event_lock_irq(conf->wait_for_stripe, | |
475 | !list_empty(&conf->inactive_list) && | |
5036805b N |
476 | (atomic_read(&conf->active_stripes) |
477 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
478 | || !conf->inactive_blocked), |
479 | conf->device_lock, | |
7c13edc8 | 480 | ); |
1da177e4 LT |
481 | conf->inactive_blocked = 0; |
482 | } else | |
b5663ba4 | 483 | init_stripe(sh, sector, previous); |
1da177e4 LT |
484 | } else { |
485 | if (atomic_read(&sh->count)) { | |
ab69ae12 | 486 | BUG_ON(!list_empty(&sh->lru) |
8811b596 SL |
487 | && !test_bit(STRIPE_EXPANDING, &sh->state) |
488 | && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)); | |
1da177e4 LT |
489 | } else { |
490 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
491 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
492 | if (list_empty(&sh->lru) && |
493 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
494 | BUG(); |
495 | list_del_init(&sh->lru); | |
1da177e4 LT |
496 | } |
497 | } | |
498 | } while (sh == NULL); | |
499 | ||
500 | if (sh) | |
501 | atomic_inc(&sh->count); | |
502 | ||
503 | spin_unlock_irq(&conf->device_lock); | |
504 | return sh; | |
505 | } | |
506 | ||
05616be5 N |
507 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
508 | * in this stripe_head. | |
509 | */ | |
510 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
511 | { | |
512 | sector_t progress = conf->reshape_progress; | |
513 | /* Need a memory barrier to make sure we see the value | |
514 | * of conf->generation, or ->data_offset that was set before | |
515 | * reshape_progress was updated. | |
516 | */ | |
517 | smp_rmb(); | |
518 | if (progress == MaxSector) | |
519 | return 0; | |
520 | if (sh->generation == conf->generation - 1) | |
521 | return 0; | |
522 | /* We are in a reshape, and this is a new-generation stripe, | |
523 | * so use new_data_offset. | |
524 | */ | |
525 | return 1; | |
526 | } | |
527 | ||
6712ecf8 N |
528 | static void |
529 | raid5_end_read_request(struct bio *bi, int error); | |
530 | static void | |
531 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 532 | |
c4e5ac0a | 533 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 534 | { |
d1688a6d | 535 | struct r5conf *conf = sh->raid_conf; |
91c00924 DW |
536 | int i, disks = sh->disks; |
537 | ||
538 | might_sleep(); | |
539 | ||
540 | for (i = disks; i--; ) { | |
541 | int rw; | |
9a3e1101 | 542 | int replace_only = 0; |
977df362 N |
543 | struct bio *bi, *rbi; |
544 | struct md_rdev *rdev, *rrdev = NULL; | |
e9c7469b TH |
545 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
546 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
547 | rw = WRITE_FUA; | |
548 | else | |
549 | rw = WRITE; | |
550 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
91c00924 | 551 | rw = READ; |
9a3e1101 N |
552 | else if (test_and_clear_bit(R5_WantReplace, |
553 | &sh->dev[i].flags)) { | |
554 | rw = WRITE; | |
555 | replace_only = 1; | |
556 | } else | |
91c00924 | 557 | continue; |
bc0934f0 SL |
558 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
559 | rw |= REQ_SYNC; | |
91c00924 DW |
560 | |
561 | bi = &sh->dev[i].req; | |
977df362 | 562 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 DW |
563 | |
564 | bi->bi_rw = rw; | |
977df362 N |
565 | rbi->bi_rw = rw; |
566 | if (rw & WRITE) { | |
91c00924 | 567 | bi->bi_end_io = raid5_end_write_request; |
977df362 N |
568 | rbi->bi_end_io = raid5_end_write_request; |
569 | } else | |
91c00924 DW |
570 | bi->bi_end_io = raid5_end_read_request; |
571 | ||
572 | rcu_read_lock(); | |
9a3e1101 | 573 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
574 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
575 | rdev = rcu_dereference(conf->disks[i].rdev); | |
576 | if (!rdev) { | |
577 | rdev = rrdev; | |
578 | rrdev = NULL; | |
579 | } | |
9a3e1101 N |
580 | if (rw & WRITE) { |
581 | if (replace_only) | |
582 | rdev = NULL; | |
dd054fce N |
583 | if (rdev == rrdev) |
584 | /* We raced and saw duplicates */ | |
585 | rrdev = NULL; | |
9a3e1101 | 586 | } else { |
dd054fce | 587 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev) |
9a3e1101 N |
588 | rdev = rrdev; |
589 | rrdev = NULL; | |
590 | } | |
977df362 | 591 | |
91c00924 DW |
592 | if (rdev && test_bit(Faulty, &rdev->flags)) |
593 | rdev = NULL; | |
594 | if (rdev) | |
595 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
596 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
597 | rrdev = NULL; | |
598 | if (rrdev) | |
599 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
600 | rcu_read_unlock(); |
601 | ||
73e92e51 | 602 | /* We have already checked bad blocks for reads. Now |
977df362 N |
603 | * need to check for writes. We never accept write errors |
604 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
605 | */ |
606 | while ((rw & WRITE) && rdev && | |
607 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
608 | sector_t first_bad; | |
609 | int bad_sectors; | |
610 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
611 | &first_bad, &bad_sectors); | |
612 | if (!bad) | |
613 | break; | |
614 | ||
615 | if (bad < 0) { | |
616 | set_bit(BlockedBadBlocks, &rdev->flags); | |
617 | if (!conf->mddev->external && | |
618 | conf->mddev->flags) { | |
619 | /* It is very unlikely, but we might | |
620 | * still need to write out the | |
621 | * bad block log - better give it | |
622 | * a chance*/ | |
623 | md_check_recovery(conf->mddev); | |
624 | } | |
1850753d | 625 | /* |
626 | * Because md_wait_for_blocked_rdev | |
627 | * will dec nr_pending, we must | |
628 | * increment it first. | |
629 | */ | |
630 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
631 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
632 | } else { | |
633 | /* Acknowledged bad block - skip the write */ | |
634 | rdev_dec_pending(rdev, conf->mddev); | |
635 | rdev = NULL; | |
636 | } | |
637 | } | |
638 | ||
91c00924 | 639 | if (rdev) { |
9a3e1101 N |
640 | if (s->syncing || s->expanding || s->expanded |
641 | || s->replacing) | |
91c00924 DW |
642 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
643 | ||
2b7497f0 DW |
644 | set_bit(STRIPE_IO_STARTED, &sh->state); |
645 | ||
91c00924 DW |
646 | bi->bi_bdev = rdev->bdev; |
647 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 648 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
649 | bi->bi_rw, i); |
650 | atomic_inc(&sh->count); | |
05616be5 N |
651 | if (use_new_offset(conf, sh)) |
652 | bi->bi_sector = (sh->sector | |
653 | + rdev->new_data_offset); | |
654 | else | |
655 | bi->bi_sector = (sh->sector | |
656 | + rdev->data_offset); | |
3f9e7c14 | 657 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
658 | bi->bi_rw |= REQ_FLUSH; | |
659 | ||
91c00924 | 660 | bi->bi_flags = 1 << BIO_UPTODATE; |
91c00924 | 661 | bi->bi_idx = 0; |
91c00924 DW |
662 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
663 | bi->bi_io_vec[0].bv_offset = 0; | |
664 | bi->bi_size = STRIPE_SIZE; | |
665 | bi->bi_next = NULL; | |
977df362 N |
666 | if (rrdev) |
667 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
91c00924 | 668 | generic_make_request(bi); |
977df362 N |
669 | } |
670 | if (rrdev) { | |
9a3e1101 N |
671 | if (s->syncing || s->expanding || s->expanded |
672 | || s->replacing) | |
977df362 N |
673 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
674 | ||
675 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
676 | ||
677 | rbi->bi_bdev = rrdev->bdev; | |
678 | pr_debug("%s: for %llu schedule op %ld on " | |
679 | "replacement disc %d\n", | |
680 | __func__, (unsigned long long)sh->sector, | |
681 | rbi->bi_rw, i); | |
682 | atomic_inc(&sh->count); | |
05616be5 N |
683 | if (use_new_offset(conf, sh)) |
684 | rbi->bi_sector = (sh->sector | |
685 | + rrdev->new_data_offset); | |
686 | else | |
687 | rbi->bi_sector = (sh->sector | |
688 | + rrdev->data_offset); | |
977df362 N |
689 | rbi->bi_flags = 1 << BIO_UPTODATE; |
690 | rbi->bi_idx = 0; | |
691 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
692 | rbi->bi_io_vec[0].bv_offset = 0; | |
693 | rbi->bi_size = STRIPE_SIZE; | |
694 | rbi->bi_next = NULL; | |
695 | generic_make_request(rbi); | |
696 | } | |
697 | if (!rdev && !rrdev) { | |
b062962e | 698 | if (rw & WRITE) |
91c00924 DW |
699 | set_bit(STRIPE_DEGRADED, &sh->state); |
700 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
701 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
702 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
703 | set_bit(STRIPE_HANDLE, &sh->state); | |
704 | } | |
705 | } | |
706 | } | |
707 | ||
708 | static struct dma_async_tx_descriptor * | |
709 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
710 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
711 | { | |
712 | struct bio_vec *bvl; | |
713 | struct page *bio_page; | |
714 | int i; | |
715 | int page_offset; | |
a08abd8c | 716 | struct async_submit_ctl submit; |
0403e382 | 717 | enum async_tx_flags flags = 0; |
91c00924 DW |
718 | |
719 | if (bio->bi_sector >= sector) | |
720 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
721 | else | |
722 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 723 | |
0403e382 DW |
724 | if (frombio) |
725 | flags |= ASYNC_TX_FENCE; | |
726 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
727 | ||
91c00924 | 728 | bio_for_each_segment(bvl, bio, i) { |
fcde9075 | 729 | int len = bvl->bv_len; |
91c00924 DW |
730 | int clen; |
731 | int b_offset = 0; | |
732 | ||
733 | if (page_offset < 0) { | |
734 | b_offset = -page_offset; | |
735 | page_offset += b_offset; | |
736 | len -= b_offset; | |
737 | } | |
738 | ||
739 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
740 | clen = STRIPE_SIZE - page_offset; | |
741 | else | |
742 | clen = len; | |
743 | ||
744 | if (clen > 0) { | |
fcde9075 NK |
745 | b_offset += bvl->bv_offset; |
746 | bio_page = bvl->bv_page; | |
91c00924 DW |
747 | if (frombio) |
748 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 749 | b_offset, clen, &submit); |
91c00924 DW |
750 | else |
751 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 752 | page_offset, clen, &submit); |
91c00924 | 753 | } |
a08abd8c DW |
754 | /* chain the operations */ |
755 | submit.depend_tx = tx; | |
756 | ||
91c00924 DW |
757 | if (clen < len) /* hit end of page */ |
758 | break; | |
759 | page_offset += len; | |
760 | } | |
761 | ||
762 | return tx; | |
763 | } | |
764 | ||
765 | static void ops_complete_biofill(void *stripe_head_ref) | |
766 | { | |
767 | struct stripe_head *sh = stripe_head_ref; | |
768 | struct bio *return_bi = NULL; | |
e4d84909 | 769 | int i; |
91c00924 | 770 | |
e46b272b | 771 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
772 | (unsigned long long)sh->sector); |
773 | ||
774 | /* clear completed biofills */ | |
775 | for (i = sh->disks; i--; ) { | |
776 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
777 | |
778 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
779 | /* and check if we need to reply to a read request, |
780 | * new R5_Wantfill requests are held off until | |
83de75cc | 781 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
782 | */ |
783 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 784 | struct bio *rbi, *rbi2; |
91c00924 | 785 | |
91c00924 DW |
786 | BUG_ON(!dev->read); |
787 | rbi = dev->read; | |
788 | dev->read = NULL; | |
789 | while (rbi && rbi->bi_sector < | |
790 | dev->sector + STRIPE_SECTORS) { | |
791 | rbi2 = r5_next_bio(rbi, dev->sector); | |
e7836bd6 | 792 | if (!raid5_dec_bi_active_stripes(rbi)) { |
91c00924 DW |
793 | rbi->bi_next = return_bi; |
794 | return_bi = rbi; | |
795 | } | |
91c00924 DW |
796 | rbi = rbi2; |
797 | } | |
798 | } | |
799 | } | |
83de75cc | 800 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 DW |
801 | |
802 | return_io(return_bi); | |
803 | ||
e4d84909 | 804 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
805 | release_stripe(sh); |
806 | } | |
807 | ||
808 | static void ops_run_biofill(struct stripe_head *sh) | |
809 | { | |
810 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 811 | struct async_submit_ctl submit; |
91c00924 DW |
812 | int i; |
813 | ||
e46b272b | 814 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
815 | (unsigned long long)sh->sector); |
816 | ||
817 | for (i = sh->disks; i--; ) { | |
818 | struct r5dev *dev = &sh->dev[i]; | |
819 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
820 | struct bio *rbi; | |
b17459c0 | 821 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
822 | dev->read = rbi = dev->toread; |
823 | dev->toread = NULL; | |
b17459c0 | 824 | spin_unlock_irq(&sh->stripe_lock); |
91c00924 DW |
825 | while (rbi && rbi->bi_sector < |
826 | dev->sector + STRIPE_SECTORS) { | |
827 | tx = async_copy_data(0, rbi, dev->page, | |
828 | dev->sector, tx); | |
829 | rbi = r5_next_bio(rbi, dev->sector); | |
830 | } | |
831 | } | |
832 | } | |
833 | ||
834 | atomic_inc(&sh->count); | |
a08abd8c DW |
835 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
836 | async_trigger_callback(&submit); | |
91c00924 DW |
837 | } |
838 | ||
4e7d2c0a | 839 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 840 | { |
4e7d2c0a | 841 | struct r5dev *tgt; |
91c00924 | 842 | |
4e7d2c0a DW |
843 | if (target < 0) |
844 | return; | |
91c00924 | 845 | |
4e7d2c0a | 846 | tgt = &sh->dev[target]; |
91c00924 DW |
847 | set_bit(R5_UPTODATE, &tgt->flags); |
848 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
849 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
850 | } |
851 | ||
ac6b53b6 | 852 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
853 | { |
854 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 855 | |
e46b272b | 856 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
857 | (unsigned long long)sh->sector); |
858 | ||
ac6b53b6 | 859 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 860 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 861 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 862 | |
ecc65c9b DW |
863 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
864 | if (sh->check_state == check_state_compute_run) | |
865 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
866 | set_bit(STRIPE_HANDLE, &sh->state); |
867 | release_stripe(sh); | |
868 | } | |
869 | ||
d6f38f31 DW |
870 | /* return a pointer to the address conversion region of the scribble buffer */ |
871 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
872 | struct raid5_percpu *percpu) | |
873 | { | |
874 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
875 | } | |
876 | ||
877 | static struct dma_async_tx_descriptor * | |
878 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 879 | { |
91c00924 | 880 | int disks = sh->disks; |
d6f38f31 | 881 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
882 | int target = sh->ops.target; |
883 | struct r5dev *tgt = &sh->dev[target]; | |
884 | struct page *xor_dest = tgt->page; | |
885 | int count = 0; | |
886 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 887 | struct async_submit_ctl submit; |
91c00924 DW |
888 | int i; |
889 | ||
890 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 891 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
892 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
893 | ||
894 | for (i = disks; i--; ) | |
895 | if (i != target) | |
896 | xor_srcs[count++] = sh->dev[i].page; | |
897 | ||
898 | atomic_inc(&sh->count); | |
899 | ||
0403e382 | 900 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 901 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 902 | if (unlikely(count == 1)) |
a08abd8c | 903 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 904 | else |
a08abd8c | 905 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 906 | |
91c00924 DW |
907 | return tx; |
908 | } | |
909 | ||
ac6b53b6 DW |
910 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
911 | * @srcs - (struct page *) array of size sh->disks | |
912 | * @sh - stripe_head to parse | |
913 | * | |
914 | * Populates srcs in proper layout order for the stripe and returns the | |
915 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
916 | * destination buffer is recorded in srcs[count] and the Q destination | |
917 | * is recorded in srcs[count+1]]. | |
918 | */ | |
919 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
920 | { | |
921 | int disks = sh->disks; | |
922 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
923 | int d0_idx = raid6_d0(sh); | |
924 | int count; | |
925 | int i; | |
926 | ||
927 | for (i = 0; i < disks; i++) | |
5dd33c9a | 928 | srcs[i] = NULL; |
ac6b53b6 DW |
929 | |
930 | count = 0; | |
931 | i = d0_idx; | |
932 | do { | |
933 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
934 | ||
935 | srcs[slot] = sh->dev[i].page; | |
936 | i = raid6_next_disk(i, disks); | |
937 | } while (i != d0_idx); | |
ac6b53b6 | 938 | |
e4424fee | 939 | return syndrome_disks; |
ac6b53b6 DW |
940 | } |
941 | ||
942 | static struct dma_async_tx_descriptor * | |
943 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
944 | { | |
945 | int disks = sh->disks; | |
946 | struct page **blocks = percpu->scribble; | |
947 | int target; | |
948 | int qd_idx = sh->qd_idx; | |
949 | struct dma_async_tx_descriptor *tx; | |
950 | struct async_submit_ctl submit; | |
951 | struct r5dev *tgt; | |
952 | struct page *dest; | |
953 | int i; | |
954 | int count; | |
955 | ||
956 | if (sh->ops.target < 0) | |
957 | target = sh->ops.target2; | |
958 | else if (sh->ops.target2 < 0) | |
959 | target = sh->ops.target; | |
91c00924 | 960 | else |
ac6b53b6 DW |
961 | /* we should only have one valid target */ |
962 | BUG(); | |
963 | BUG_ON(target < 0); | |
964 | pr_debug("%s: stripe %llu block: %d\n", | |
965 | __func__, (unsigned long long)sh->sector, target); | |
966 | ||
967 | tgt = &sh->dev[target]; | |
968 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
969 | dest = tgt->page; | |
970 | ||
971 | atomic_inc(&sh->count); | |
972 | ||
973 | if (target == qd_idx) { | |
974 | count = set_syndrome_sources(blocks, sh); | |
975 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
976 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
977 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
978 | ops_complete_compute, sh, | |
ac6b53b6 DW |
979 | to_addr_conv(sh, percpu)); |
980 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
981 | } else { | |
982 | /* Compute any data- or p-drive using XOR */ | |
983 | count = 0; | |
984 | for (i = disks; i-- ; ) { | |
985 | if (i == target || i == qd_idx) | |
986 | continue; | |
987 | blocks[count++] = sh->dev[i].page; | |
988 | } | |
989 | ||
0403e382 DW |
990 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
991 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
992 | to_addr_conv(sh, percpu)); |
993 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
994 | } | |
91c00924 | 995 | |
91c00924 DW |
996 | return tx; |
997 | } | |
998 | ||
ac6b53b6 DW |
999 | static struct dma_async_tx_descriptor * |
1000 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1001 | { | |
1002 | int i, count, disks = sh->disks; | |
1003 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1004 | int d0_idx = raid6_d0(sh); | |
1005 | int faila = -1, failb = -1; | |
1006 | int target = sh->ops.target; | |
1007 | int target2 = sh->ops.target2; | |
1008 | struct r5dev *tgt = &sh->dev[target]; | |
1009 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1010 | struct dma_async_tx_descriptor *tx; | |
1011 | struct page **blocks = percpu->scribble; | |
1012 | struct async_submit_ctl submit; | |
1013 | ||
1014 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
1015 | __func__, (unsigned long long)sh->sector, target, target2); | |
1016 | BUG_ON(target < 0 || target2 < 0); | |
1017 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1018 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1019 | ||
6c910a78 | 1020 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1021 | * slot number conversion for 'faila' and 'failb' |
1022 | */ | |
1023 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1024 | blocks[i] = NULL; |
ac6b53b6 DW |
1025 | count = 0; |
1026 | i = d0_idx; | |
1027 | do { | |
1028 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1029 | ||
1030 | blocks[slot] = sh->dev[i].page; | |
1031 | ||
1032 | if (i == target) | |
1033 | faila = slot; | |
1034 | if (i == target2) | |
1035 | failb = slot; | |
1036 | i = raid6_next_disk(i, disks); | |
1037 | } while (i != d0_idx); | |
ac6b53b6 DW |
1038 | |
1039 | BUG_ON(faila == failb); | |
1040 | if (failb < faila) | |
1041 | swap(faila, failb); | |
1042 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1043 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1044 | ||
1045 | atomic_inc(&sh->count); | |
1046 | ||
1047 | if (failb == syndrome_disks+1) { | |
1048 | /* Q disk is one of the missing disks */ | |
1049 | if (faila == syndrome_disks) { | |
1050 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1051 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1052 | ops_complete_compute, sh, | |
1053 | to_addr_conv(sh, percpu)); | |
e4424fee | 1054 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1055 | STRIPE_SIZE, &submit); |
1056 | } else { | |
1057 | struct page *dest; | |
1058 | int data_target; | |
1059 | int qd_idx = sh->qd_idx; | |
1060 | ||
1061 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1062 | if (target == qd_idx) | |
1063 | data_target = target2; | |
1064 | else | |
1065 | data_target = target; | |
1066 | ||
1067 | count = 0; | |
1068 | for (i = disks; i-- ; ) { | |
1069 | if (i == data_target || i == qd_idx) | |
1070 | continue; | |
1071 | blocks[count++] = sh->dev[i].page; | |
1072 | } | |
1073 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1074 | init_async_submit(&submit, |
1075 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1076 | NULL, NULL, NULL, | |
1077 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1078 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1079 | &submit); | |
1080 | ||
1081 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
1082 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1083 | ops_complete_compute, sh, | |
1084 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1085 | return async_gen_syndrome(blocks, 0, count+2, |
1086 | STRIPE_SIZE, &submit); | |
1087 | } | |
ac6b53b6 | 1088 | } else { |
6c910a78 DW |
1089 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1090 | ops_complete_compute, sh, | |
1091 | to_addr_conv(sh, percpu)); | |
1092 | if (failb == syndrome_disks) { | |
1093 | /* We're missing D+P. */ | |
1094 | return async_raid6_datap_recov(syndrome_disks+2, | |
1095 | STRIPE_SIZE, faila, | |
1096 | blocks, &submit); | |
1097 | } else { | |
1098 | /* We're missing D+D. */ | |
1099 | return async_raid6_2data_recov(syndrome_disks+2, | |
1100 | STRIPE_SIZE, faila, failb, | |
1101 | blocks, &submit); | |
1102 | } | |
ac6b53b6 DW |
1103 | } |
1104 | } | |
1105 | ||
1106 | ||
91c00924 DW |
1107 | static void ops_complete_prexor(void *stripe_head_ref) |
1108 | { | |
1109 | struct stripe_head *sh = stripe_head_ref; | |
1110 | ||
e46b272b | 1111 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1112 | (unsigned long long)sh->sector); |
91c00924 DW |
1113 | } |
1114 | ||
1115 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
1116 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
1117 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1118 | { |
91c00924 | 1119 | int disks = sh->disks; |
d6f38f31 | 1120 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1121 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1122 | struct async_submit_ctl submit; |
91c00924 DW |
1123 | |
1124 | /* existing parity data subtracted */ | |
1125 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1126 | ||
e46b272b | 1127 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1128 | (unsigned long long)sh->sector); |
1129 | ||
1130 | for (i = disks; i--; ) { | |
1131 | struct r5dev *dev = &sh->dev[i]; | |
1132 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1133 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1134 | xor_srcs[count++] = dev->page; |
1135 | } | |
1136 | ||
0403e382 | 1137 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 1138 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 1139 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1140 | |
1141 | return tx; | |
1142 | } | |
1143 | ||
1144 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1145 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1146 | { |
1147 | int disks = sh->disks; | |
d8ee0728 | 1148 | int i; |
91c00924 | 1149 | |
e46b272b | 1150 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1151 | (unsigned long long)sh->sector); |
1152 | ||
1153 | for (i = disks; i--; ) { | |
1154 | struct r5dev *dev = &sh->dev[i]; | |
1155 | struct bio *chosen; | |
91c00924 | 1156 | |
d8ee0728 | 1157 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1158 | struct bio *wbi; |
1159 | ||
b17459c0 | 1160 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1161 | chosen = dev->towrite; |
1162 | dev->towrite = NULL; | |
1163 | BUG_ON(dev->written); | |
1164 | wbi = dev->written = chosen; | |
b17459c0 | 1165 | spin_unlock_irq(&sh->stripe_lock); |
91c00924 DW |
1166 | |
1167 | while (wbi && wbi->bi_sector < | |
1168 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1169 | if (wbi->bi_rw & REQ_FUA) |
1170 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1171 | if (wbi->bi_rw & REQ_SYNC) |
1172 | set_bit(R5_SyncIO, &dev->flags); | |
91c00924 DW |
1173 | tx = async_copy_data(1, wbi, dev->page, |
1174 | dev->sector, tx); | |
1175 | wbi = r5_next_bio(wbi, dev->sector); | |
1176 | } | |
1177 | } | |
1178 | } | |
1179 | ||
1180 | return tx; | |
1181 | } | |
1182 | ||
ac6b53b6 | 1183 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1184 | { |
1185 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1186 | int disks = sh->disks; |
1187 | int pd_idx = sh->pd_idx; | |
1188 | int qd_idx = sh->qd_idx; | |
1189 | int i; | |
bc0934f0 | 1190 | bool fua = false, sync = false; |
91c00924 | 1191 | |
e46b272b | 1192 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1193 | (unsigned long long)sh->sector); |
1194 | ||
bc0934f0 | 1195 | for (i = disks; i--; ) { |
e9c7469b | 1196 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 SL |
1197 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
1198 | } | |
e9c7469b | 1199 | |
91c00924 DW |
1200 | for (i = disks; i--; ) { |
1201 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1202 | |
e9c7469b | 1203 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1204 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1205 | if (fua) |
1206 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1207 | if (sync) |
1208 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1209 | } |
91c00924 DW |
1210 | } |
1211 | ||
d8ee0728 DW |
1212 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1213 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1214 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1215 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1216 | else { | |
1217 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1218 | sh->reconstruct_state = reconstruct_state_result; | |
1219 | } | |
91c00924 DW |
1220 | |
1221 | set_bit(STRIPE_HANDLE, &sh->state); | |
1222 | release_stripe(sh); | |
1223 | } | |
1224 | ||
1225 | static void | |
ac6b53b6 DW |
1226 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1227 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1228 | { |
91c00924 | 1229 | int disks = sh->disks; |
d6f38f31 | 1230 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1231 | struct async_submit_ctl submit; |
91c00924 DW |
1232 | int count = 0, pd_idx = sh->pd_idx, i; |
1233 | struct page *xor_dest; | |
d8ee0728 | 1234 | int prexor = 0; |
91c00924 | 1235 | unsigned long flags; |
91c00924 | 1236 | |
e46b272b | 1237 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1238 | (unsigned long long)sh->sector); |
1239 | ||
1240 | /* check if prexor is active which means only process blocks | |
1241 | * that are part of a read-modify-write (written) | |
1242 | */ | |
d8ee0728 DW |
1243 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1244 | prexor = 1; | |
91c00924 DW |
1245 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1246 | for (i = disks; i--; ) { | |
1247 | struct r5dev *dev = &sh->dev[i]; | |
1248 | if (dev->written) | |
1249 | xor_srcs[count++] = dev->page; | |
1250 | } | |
1251 | } else { | |
1252 | xor_dest = sh->dev[pd_idx].page; | |
1253 | for (i = disks; i--; ) { | |
1254 | struct r5dev *dev = &sh->dev[i]; | |
1255 | if (i != pd_idx) | |
1256 | xor_srcs[count++] = dev->page; | |
1257 | } | |
1258 | } | |
1259 | ||
91c00924 DW |
1260 | /* 1/ if we prexor'd then the dest is reused as a source |
1261 | * 2/ if we did not prexor then we are redoing the parity | |
1262 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1263 | * for the synchronous xor case | |
1264 | */ | |
88ba2aa5 | 1265 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1266 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1267 | ||
1268 | atomic_inc(&sh->count); | |
1269 | ||
ac6b53b6 | 1270 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1271 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1272 | if (unlikely(count == 1)) |
1273 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1274 | else | |
1275 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1276 | } |
1277 | ||
ac6b53b6 DW |
1278 | static void |
1279 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1280 | struct dma_async_tx_descriptor *tx) | |
1281 | { | |
1282 | struct async_submit_ctl submit; | |
1283 | struct page **blocks = percpu->scribble; | |
1284 | int count; | |
1285 | ||
1286 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1287 | ||
1288 | count = set_syndrome_sources(blocks, sh); | |
1289 | ||
1290 | atomic_inc(&sh->count); | |
1291 | ||
1292 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1293 | sh, to_addr_conv(sh, percpu)); | |
1294 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1295 | } |
1296 | ||
1297 | static void ops_complete_check(void *stripe_head_ref) | |
1298 | { | |
1299 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1300 | |
e46b272b | 1301 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1302 | (unsigned long long)sh->sector); |
1303 | ||
ecc65c9b | 1304 | sh->check_state = check_state_check_result; |
91c00924 DW |
1305 | set_bit(STRIPE_HANDLE, &sh->state); |
1306 | release_stripe(sh); | |
1307 | } | |
1308 | ||
ac6b53b6 | 1309 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1310 | { |
91c00924 | 1311 | int disks = sh->disks; |
ac6b53b6 DW |
1312 | int pd_idx = sh->pd_idx; |
1313 | int qd_idx = sh->qd_idx; | |
1314 | struct page *xor_dest; | |
d6f38f31 | 1315 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1316 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1317 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1318 | int count; |
1319 | int i; | |
91c00924 | 1320 | |
e46b272b | 1321 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1322 | (unsigned long long)sh->sector); |
1323 | ||
ac6b53b6 DW |
1324 | count = 0; |
1325 | xor_dest = sh->dev[pd_idx].page; | |
1326 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1327 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1328 | if (i == pd_idx || i == qd_idx) |
1329 | continue; | |
1330 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1331 | } |
1332 | ||
d6f38f31 DW |
1333 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1334 | to_addr_conv(sh, percpu)); | |
099f53cb | 1335 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1336 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1337 | |
91c00924 | 1338 | atomic_inc(&sh->count); |
a08abd8c DW |
1339 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1340 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1341 | } |
1342 | ||
ac6b53b6 DW |
1343 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1344 | { | |
1345 | struct page **srcs = percpu->scribble; | |
1346 | struct async_submit_ctl submit; | |
1347 | int count; | |
1348 | ||
1349 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1350 | (unsigned long long)sh->sector, checkp); | |
1351 | ||
1352 | count = set_syndrome_sources(srcs, sh); | |
1353 | if (!checkp) | |
1354 | srcs[count] = NULL; | |
91c00924 | 1355 | |
91c00924 | 1356 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1357 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1358 | sh, to_addr_conv(sh, percpu)); | |
1359 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1360 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1361 | } |
1362 | ||
417b8d4a | 1363 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1364 | { |
1365 | int overlap_clear = 0, i, disks = sh->disks; | |
1366 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1367 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1368 | int level = conf->level; |
d6f38f31 DW |
1369 | struct raid5_percpu *percpu; |
1370 | unsigned long cpu; | |
91c00924 | 1371 | |
d6f38f31 DW |
1372 | cpu = get_cpu(); |
1373 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1374 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1375 | ops_run_biofill(sh); |
1376 | overlap_clear++; | |
1377 | } | |
1378 | ||
7b3a871e | 1379 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1380 | if (level < 6) |
1381 | tx = ops_run_compute5(sh, percpu); | |
1382 | else { | |
1383 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1384 | tx = ops_run_compute6_1(sh, percpu); | |
1385 | else | |
1386 | tx = ops_run_compute6_2(sh, percpu); | |
1387 | } | |
1388 | /* terminate the chain if reconstruct is not set to be run */ | |
1389 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1390 | async_tx_ack(tx); |
1391 | } | |
91c00924 | 1392 | |
600aa109 | 1393 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1394 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1395 | |
600aa109 | 1396 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1397 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1398 | overlap_clear++; |
1399 | } | |
1400 | ||
ac6b53b6 DW |
1401 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1402 | if (level < 6) | |
1403 | ops_run_reconstruct5(sh, percpu, tx); | |
1404 | else | |
1405 | ops_run_reconstruct6(sh, percpu, tx); | |
1406 | } | |
91c00924 | 1407 | |
ac6b53b6 DW |
1408 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1409 | if (sh->check_state == check_state_run) | |
1410 | ops_run_check_p(sh, percpu); | |
1411 | else if (sh->check_state == check_state_run_q) | |
1412 | ops_run_check_pq(sh, percpu, 0); | |
1413 | else if (sh->check_state == check_state_run_pq) | |
1414 | ops_run_check_pq(sh, percpu, 1); | |
1415 | else | |
1416 | BUG(); | |
1417 | } | |
91c00924 | 1418 | |
91c00924 DW |
1419 | if (overlap_clear) |
1420 | for (i = disks; i--; ) { | |
1421 | struct r5dev *dev = &sh->dev[i]; | |
1422 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1423 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1424 | } | |
d6f38f31 | 1425 | put_cpu(); |
91c00924 DW |
1426 | } |
1427 | ||
417b8d4a DW |
1428 | #ifdef CONFIG_MULTICORE_RAID456 |
1429 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1430 | { | |
1431 | struct stripe_head *sh = param; | |
1432 | unsigned long ops_request = sh->ops.request; | |
1433 | ||
1434 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1435 | wake_up(&sh->ops.wait_for_ops); | |
1436 | ||
1437 | __raid_run_ops(sh, ops_request); | |
1438 | release_stripe(sh); | |
1439 | } | |
1440 | ||
1441 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1442 | { | |
1443 | /* since handle_stripe can be called outside of raid5d context | |
1444 | * we need to ensure sh->ops.request is de-staged before another | |
1445 | * request arrives | |
1446 | */ | |
1447 | wait_event(sh->ops.wait_for_ops, | |
1448 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1449 | sh->ops.request = ops_request; | |
1450 | ||
1451 | atomic_inc(&sh->count); | |
1452 | async_schedule(async_run_ops, sh); | |
1453 | } | |
1454 | #else | |
1455 | #define raid_run_ops __raid_run_ops | |
1456 | #endif | |
1457 | ||
d1688a6d | 1458 | static int grow_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1459 | { |
1460 | struct stripe_head *sh; | |
6ce32846 | 1461 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1462 | if (!sh) |
1463 | return 0; | |
6ce32846 | 1464 | |
3f294f4f | 1465 | sh->raid_conf = conf; |
417b8d4a DW |
1466 | #ifdef CONFIG_MULTICORE_RAID456 |
1467 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1468 | #endif | |
3f294f4f | 1469 | |
b17459c0 SL |
1470 | spin_lock_init(&sh->stripe_lock); |
1471 | ||
e4e11e38 N |
1472 | if (grow_buffers(sh)) { |
1473 | shrink_buffers(sh); | |
3f294f4f N |
1474 | kmem_cache_free(conf->slab_cache, sh); |
1475 | return 0; | |
1476 | } | |
1477 | /* we just created an active stripe so... */ | |
1478 | atomic_set(&sh->count, 1); | |
1479 | atomic_inc(&conf->active_stripes); | |
1480 | INIT_LIST_HEAD(&sh->lru); | |
1481 | release_stripe(sh); | |
1482 | return 1; | |
1483 | } | |
1484 | ||
d1688a6d | 1485 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 1486 | { |
e18b890b | 1487 | struct kmem_cache *sc; |
5e5e3e78 | 1488 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1489 | |
f4be6b43 N |
1490 | if (conf->mddev->gendisk) |
1491 | sprintf(conf->cache_name[0], | |
1492 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1493 | else | |
1494 | sprintf(conf->cache_name[0], | |
1495 | "raid%d-%p", conf->level, conf->mddev); | |
1496 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1497 | ||
ad01c9e3 N |
1498 | conf->active_name = 0; |
1499 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1500 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1501 | 0, 0, NULL); |
1da177e4 LT |
1502 | if (!sc) |
1503 | return 1; | |
1504 | conf->slab_cache = sc; | |
ad01c9e3 | 1505 | conf->pool_size = devs; |
16a53ecc | 1506 | while (num--) |
3f294f4f | 1507 | if (!grow_one_stripe(conf)) |
1da177e4 | 1508 | return 1; |
1da177e4 LT |
1509 | return 0; |
1510 | } | |
29269553 | 1511 | |
d6f38f31 DW |
1512 | /** |
1513 | * scribble_len - return the required size of the scribble region | |
1514 | * @num - total number of disks in the array | |
1515 | * | |
1516 | * The size must be enough to contain: | |
1517 | * 1/ a struct page pointer for each device in the array +2 | |
1518 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1519 | * (dma_map_page()) or page (page_address()) address. | |
1520 | * | |
1521 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1522 | * calculate over all devices (not just the data blocks), using zeros in place | |
1523 | * of the P and Q blocks. | |
1524 | */ | |
1525 | static size_t scribble_len(int num) | |
1526 | { | |
1527 | size_t len; | |
1528 | ||
1529 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1530 | ||
1531 | return len; | |
1532 | } | |
1533 | ||
d1688a6d | 1534 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
1535 | { |
1536 | /* Make all the stripes able to hold 'newsize' devices. | |
1537 | * New slots in each stripe get 'page' set to a new page. | |
1538 | * | |
1539 | * This happens in stages: | |
1540 | * 1/ create a new kmem_cache and allocate the required number of | |
1541 | * stripe_heads. | |
1542 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1543 | * to the new stripe_heads. This will have the side effect of | |
1544 | * freezing the array as once all stripe_heads have been collected, | |
1545 | * no IO will be possible. Old stripe heads are freed once their | |
1546 | * pages have been transferred over, and the old kmem_cache is | |
1547 | * freed when all stripes are done. | |
1548 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1549 | * we simple return a failre status - no need to clean anything up. | |
1550 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1551 | * If this fails, we don't bother trying the shrink the | |
1552 | * stripe_heads down again, we just leave them as they are. | |
1553 | * As each stripe_head is processed the new one is released into | |
1554 | * active service. | |
1555 | * | |
1556 | * Once step2 is started, we cannot afford to wait for a write, | |
1557 | * so we use GFP_NOIO allocations. | |
1558 | */ | |
1559 | struct stripe_head *osh, *nsh; | |
1560 | LIST_HEAD(newstripes); | |
1561 | struct disk_info *ndisks; | |
d6f38f31 | 1562 | unsigned long cpu; |
b5470dc5 | 1563 | int err; |
e18b890b | 1564 | struct kmem_cache *sc; |
ad01c9e3 N |
1565 | int i; |
1566 | ||
1567 | if (newsize <= conf->pool_size) | |
1568 | return 0; /* never bother to shrink */ | |
1569 | ||
b5470dc5 DW |
1570 | err = md_allow_write(conf->mddev); |
1571 | if (err) | |
1572 | return err; | |
2a2275d6 | 1573 | |
ad01c9e3 N |
1574 | /* Step 1 */ |
1575 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1576 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1577 | 0, 0, NULL); |
ad01c9e3 N |
1578 | if (!sc) |
1579 | return -ENOMEM; | |
1580 | ||
1581 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1582 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1583 | if (!nsh) |
1584 | break; | |
1585 | ||
ad01c9e3 | 1586 | nsh->raid_conf = conf; |
417b8d4a DW |
1587 | #ifdef CONFIG_MULTICORE_RAID456 |
1588 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1589 | #endif | |
ad01c9e3 N |
1590 | |
1591 | list_add(&nsh->lru, &newstripes); | |
1592 | } | |
1593 | if (i) { | |
1594 | /* didn't get enough, give up */ | |
1595 | while (!list_empty(&newstripes)) { | |
1596 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1597 | list_del(&nsh->lru); | |
1598 | kmem_cache_free(sc, nsh); | |
1599 | } | |
1600 | kmem_cache_destroy(sc); | |
1601 | return -ENOMEM; | |
1602 | } | |
1603 | /* Step 2 - Must use GFP_NOIO now. | |
1604 | * OK, we have enough stripes, start collecting inactive | |
1605 | * stripes and copying them over | |
1606 | */ | |
1607 | list_for_each_entry(nsh, &newstripes, lru) { | |
1608 | spin_lock_irq(&conf->device_lock); | |
1609 | wait_event_lock_irq(conf->wait_for_stripe, | |
1610 | !list_empty(&conf->inactive_list), | |
1611 | conf->device_lock, | |
482c0834 | 1612 | ); |
ad01c9e3 N |
1613 | osh = get_free_stripe(conf); |
1614 | spin_unlock_irq(&conf->device_lock); | |
1615 | atomic_set(&nsh->count, 1); | |
1616 | for(i=0; i<conf->pool_size; i++) | |
1617 | nsh->dev[i].page = osh->dev[i].page; | |
1618 | for( ; i<newsize; i++) | |
1619 | nsh->dev[i].page = NULL; | |
1620 | kmem_cache_free(conf->slab_cache, osh); | |
1621 | } | |
1622 | kmem_cache_destroy(conf->slab_cache); | |
1623 | ||
1624 | /* Step 3. | |
1625 | * At this point, we are holding all the stripes so the array | |
1626 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1627 | * conf->disks and the scribble region |
ad01c9e3 N |
1628 | */ |
1629 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1630 | if (ndisks) { | |
1631 | for (i=0; i<conf->raid_disks; i++) | |
1632 | ndisks[i] = conf->disks[i]; | |
1633 | kfree(conf->disks); | |
1634 | conf->disks = ndisks; | |
1635 | } else | |
1636 | err = -ENOMEM; | |
1637 | ||
d6f38f31 DW |
1638 | get_online_cpus(); |
1639 | conf->scribble_len = scribble_len(newsize); | |
1640 | for_each_present_cpu(cpu) { | |
1641 | struct raid5_percpu *percpu; | |
1642 | void *scribble; | |
1643 | ||
1644 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1645 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1646 | ||
1647 | if (scribble) { | |
1648 | kfree(percpu->scribble); | |
1649 | percpu->scribble = scribble; | |
1650 | } else { | |
1651 | err = -ENOMEM; | |
1652 | break; | |
1653 | } | |
1654 | } | |
1655 | put_online_cpus(); | |
1656 | ||
ad01c9e3 N |
1657 | /* Step 4, return new stripes to service */ |
1658 | while(!list_empty(&newstripes)) { | |
1659 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1660 | list_del_init(&nsh->lru); | |
d6f38f31 | 1661 | |
ad01c9e3 N |
1662 | for (i=conf->raid_disks; i < newsize; i++) |
1663 | if (nsh->dev[i].page == NULL) { | |
1664 | struct page *p = alloc_page(GFP_NOIO); | |
1665 | nsh->dev[i].page = p; | |
1666 | if (!p) | |
1667 | err = -ENOMEM; | |
1668 | } | |
1669 | release_stripe(nsh); | |
1670 | } | |
1671 | /* critical section pass, GFP_NOIO no longer needed */ | |
1672 | ||
1673 | conf->slab_cache = sc; | |
1674 | conf->active_name = 1-conf->active_name; | |
1675 | conf->pool_size = newsize; | |
1676 | return err; | |
1677 | } | |
1da177e4 | 1678 | |
d1688a6d | 1679 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1680 | { |
1681 | struct stripe_head *sh; | |
1682 | ||
3f294f4f N |
1683 | spin_lock_irq(&conf->device_lock); |
1684 | sh = get_free_stripe(conf); | |
1685 | spin_unlock_irq(&conf->device_lock); | |
1686 | if (!sh) | |
1687 | return 0; | |
78bafebd | 1688 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1689 | shrink_buffers(sh); |
3f294f4f N |
1690 | kmem_cache_free(conf->slab_cache, sh); |
1691 | atomic_dec(&conf->active_stripes); | |
1692 | return 1; | |
1693 | } | |
1694 | ||
d1688a6d | 1695 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f N |
1696 | { |
1697 | while (drop_one_stripe(conf)) | |
1698 | ; | |
1699 | ||
29fc7e3e N |
1700 | if (conf->slab_cache) |
1701 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1702 | conf->slab_cache = NULL; |
1703 | } | |
1704 | ||
6712ecf8 | 1705 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1706 | { |
99c0fb5f | 1707 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1708 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1709 | int disks = sh->disks, i; |
1da177e4 | 1710 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 1711 | char b[BDEVNAME_SIZE]; |
dd054fce | 1712 | struct md_rdev *rdev = NULL; |
05616be5 | 1713 | sector_t s; |
1da177e4 LT |
1714 | |
1715 | for (i=0 ; i<disks; i++) | |
1716 | if (bi == &sh->dev[i].req) | |
1717 | break; | |
1718 | ||
45b4233c DW |
1719 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1720 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1721 | uptodate); |
1722 | if (i == disks) { | |
1723 | BUG(); | |
6712ecf8 | 1724 | return; |
1da177e4 | 1725 | } |
14a75d3e | 1726 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
1727 | /* If replacement finished while this request was outstanding, |
1728 | * 'replacement' might be NULL already. | |
1729 | * In that case it moved down to 'rdev'. | |
1730 | * rdev is not removed until all requests are finished. | |
1731 | */ | |
14a75d3e | 1732 | rdev = conf->disks[i].replacement; |
dd054fce | 1733 | if (!rdev) |
14a75d3e | 1734 | rdev = conf->disks[i].rdev; |
1da177e4 | 1735 | |
05616be5 N |
1736 | if (use_new_offset(conf, sh)) |
1737 | s = sh->sector + rdev->new_data_offset; | |
1738 | else | |
1739 | s = sh->sector + rdev->data_offset; | |
1da177e4 | 1740 | if (uptodate) { |
1da177e4 | 1741 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1742 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
1743 | /* Note that this cannot happen on a |
1744 | * replacement device. We just fail those on | |
1745 | * any error | |
1746 | */ | |
8bda470e CD |
1747 | printk_ratelimited( |
1748 | KERN_INFO | |
1749 | "md/raid:%s: read error corrected" | |
1750 | " (%lu sectors at %llu on %s)\n", | |
1751 | mdname(conf->mddev), STRIPE_SECTORS, | |
05616be5 | 1752 | (unsigned long long)s, |
8bda470e | 1753 | bdevname(rdev->bdev, b)); |
ddd5115f | 1754 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
1755 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1756 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 1757 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
1758 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
1759 | ||
14a75d3e N |
1760 | if (atomic_read(&rdev->read_errors)) |
1761 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 1762 | } else { |
14a75d3e | 1763 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 1764 | int retry = 0; |
2e8ac303 | 1765 | int set_bad = 0; |
d6950432 | 1766 | |
1da177e4 | 1767 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1768 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
1769 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
1770 | printk_ratelimited( | |
1771 | KERN_WARNING | |
1772 | "md/raid:%s: read error on replacement device " | |
1773 | "(sector %llu on %s).\n", | |
1774 | mdname(conf->mddev), | |
05616be5 | 1775 | (unsigned long long)s, |
14a75d3e | 1776 | bdn); |
2e8ac303 | 1777 | else if (conf->mddev->degraded >= conf->max_degraded) { |
1778 | set_bad = 1; | |
8bda470e CD |
1779 | printk_ratelimited( |
1780 | KERN_WARNING | |
1781 | "md/raid:%s: read error not correctable " | |
1782 | "(sector %llu on %s).\n", | |
1783 | mdname(conf->mddev), | |
05616be5 | 1784 | (unsigned long long)s, |
8bda470e | 1785 | bdn); |
2e8ac303 | 1786 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 1787 | /* Oh, no!!! */ |
2e8ac303 | 1788 | set_bad = 1; |
8bda470e CD |
1789 | printk_ratelimited( |
1790 | KERN_WARNING | |
1791 | "md/raid:%s: read error NOT corrected!! " | |
1792 | "(sector %llu on %s).\n", | |
1793 | mdname(conf->mddev), | |
05616be5 | 1794 | (unsigned long long)s, |
8bda470e | 1795 | bdn); |
2e8ac303 | 1796 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1797 | > conf->max_nr_stripes) |
14f8d26b | 1798 | printk(KERN_WARNING |
0c55e022 | 1799 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1800 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1801 | else |
1802 | retry = 1; | |
1803 | if (retry) | |
3f9e7c14 | 1804 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
1805 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1806 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
1807 | } else | |
1808 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 1809 | else { |
4e5314b5 N |
1810 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1811 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 1812 | if (!(set_bad |
1813 | && test_bit(In_sync, &rdev->flags) | |
1814 | && rdev_set_badblocks( | |
1815 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
1816 | md_error(conf->mddev, rdev); | |
ba22dcbf | 1817 | } |
1da177e4 | 1818 | } |
14a75d3e | 1819 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
1820 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1821 | set_bit(STRIPE_HANDLE, &sh->state); | |
1822 | release_stripe(sh); | |
1da177e4 LT |
1823 | } |
1824 | ||
d710e138 | 1825 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1826 | { |
99c0fb5f | 1827 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1828 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1829 | int disks = sh->disks, i; |
977df362 | 1830 | struct md_rdev *uninitialized_var(rdev); |
1da177e4 | 1831 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
1832 | sector_t first_bad; |
1833 | int bad_sectors; | |
977df362 | 1834 | int replacement = 0; |
1da177e4 | 1835 | |
977df362 N |
1836 | for (i = 0 ; i < disks; i++) { |
1837 | if (bi == &sh->dev[i].req) { | |
1838 | rdev = conf->disks[i].rdev; | |
1da177e4 | 1839 | break; |
977df362 N |
1840 | } |
1841 | if (bi == &sh->dev[i].rreq) { | |
1842 | rdev = conf->disks[i].replacement; | |
dd054fce N |
1843 | if (rdev) |
1844 | replacement = 1; | |
1845 | else | |
1846 | /* rdev was removed and 'replacement' | |
1847 | * replaced it. rdev is not removed | |
1848 | * until all requests are finished. | |
1849 | */ | |
1850 | rdev = conf->disks[i].rdev; | |
977df362 N |
1851 | break; |
1852 | } | |
1853 | } | |
45b4233c | 1854 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1855 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1856 | uptodate); | |
1857 | if (i == disks) { | |
1858 | BUG(); | |
6712ecf8 | 1859 | return; |
1da177e4 LT |
1860 | } |
1861 | ||
977df362 N |
1862 | if (replacement) { |
1863 | if (!uptodate) | |
1864 | md_error(conf->mddev, rdev); | |
1865 | else if (is_badblock(rdev, sh->sector, | |
1866 | STRIPE_SECTORS, | |
1867 | &first_bad, &bad_sectors)) | |
1868 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
1869 | } else { | |
1870 | if (!uptodate) { | |
1871 | set_bit(WriteErrorSeen, &rdev->flags); | |
1872 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
1873 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1874 | set_bit(MD_RECOVERY_NEEDED, | |
1875 | &rdev->mddev->recovery); | |
977df362 N |
1876 | } else if (is_badblock(rdev, sh->sector, |
1877 | STRIPE_SECTORS, | |
1878 | &first_bad, &bad_sectors)) | |
1879 | set_bit(R5_MadeGood, &sh->dev[i].flags); | |
1880 | } | |
1881 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 1882 | |
977df362 N |
1883 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
1884 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 1885 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 1886 | release_stripe(sh); |
1da177e4 LT |
1887 | } |
1888 | ||
784052ec | 1889 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1890 | |
784052ec | 1891 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1892 | { |
1893 | struct r5dev *dev = &sh->dev[i]; | |
1894 | ||
1895 | bio_init(&dev->req); | |
1896 | dev->req.bi_io_vec = &dev->vec; | |
1897 | dev->req.bi_vcnt++; | |
1898 | dev->req.bi_max_vecs++; | |
1da177e4 | 1899 | dev->req.bi_private = sh; |
995c4275 | 1900 | dev->vec.bv_page = dev->page; |
1da177e4 | 1901 | |
977df362 N |
1902 | bio_init(&dev->rreq); |
1903 | dev->rreq.bi_io_vec = &dev->rvec; | |
1904 | dev->rreq.bi_vcnt++; | |
1905 | dev->rreq.bi_max_vecs++; | |
1906 | dev->rreq.bi_private = sh; | |
1907 | dev->rvec.bv_page = dev->page; | |
1908 | ||
1da177e4 | 1909 | dev->flags = 0; |
784052ec | 1910 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1911 | } |
1912 | ||
fd01b88c | 1913 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1914 | { |
1915 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 1916 | struct r5conf *conf = mddev->private; |
908f4fbd | 1917 | unsigned long flags; |
0c55e022 | 1918 | pr_debug("raid456: error called\n"); |
1da177e4 | 1919 | |
908f4fbd N |
1920 | spin_lock_irqsave(&conf->device_lock, flags); |
1921 | clear_bit(In_sync, &rdev->flags); | |
1922 | mddev->degraded = calc_degraded(conf); | |
1923 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1924 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1925 | ||
de393cde | 1926 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
1927 | set_bit(Faulty, &rdev->flags); |
1928 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1929 | printk(KERN_ALERT | |
1930 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1931 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1932 | mdname(mddev), | |
1933 | bdevname(rdev->bdev, b), | |
1934 | mdname(mddev), | |
1935 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1936 | } |
1da177e4 LT |
1937 | |
1938 | /* | |
1939 | * Input: a 'big' sector number, | |
1940 | * Output: index of the data and parity disk, and the sector # in them. | |
1941 | */ | |
d1688a6d | 1942 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
1943 | int previous, int *dd_idx, |
1944 | struct stripe_head *sh) | |
1da177e4 | 1945 | { |
6e3b96ed | 1946 | sector_t stripe, stripe2; |
35f2a591 | 1947 | sector_t chunk_number; |
1da177e4 | 1948 | unsigned int chunk_offset; |
911d4ee8 | 1949 | int pd_idx, qd_idx; |
67cc2b81 | 1950 | int ddf_layout = 0; |
1da177e4 | 1951 | sector_t new_sector; |
e183eaed N |
1952 | int algorithm = previous ? conf->prev_algo |
1953 | : conf->algorithm; | |
09c9e5fa AN |
1954 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1955 | : conf->chunk_sectors; | |
112bf897 N |
1956 | int raid_disks = previous ? conf->previous_raid_disks |
1957 | : conf->raid_disks; | |
1958 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1959 | |
1960 | /* First compute the information on this sector */ | |
1961 | ||
1962 | /* | |
1963 | * Compute the chunk number and the sector offset inside the chunk | |
1964 | */ | |
1965 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1966 | chunk_number = r_sector; | |
1da177e4 LT |
1967 | |
1968 | /* | |
1969 | * Compute the stripe number | |
1970 | */ | |
35f2a591 N |
1971 | stripe = chunk_number; |
1972 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1973 | stripe2 = stripe; |
1da177e4 LT |
1974 | /* |
1975 | * Select the parity disk based on the user selected algorithm. | |
1976 | */ | |
84789554 | 1977 | pd_idx = qd_idx = -1; |
16a53ecc N |
1978 | switch(conf->level) { |
1979 | case 4: | |
911d4ee8 | 1980 | pd_idx = data_disks; |
16a53ecc N |
1981 | break; |
1982 | case 5: | |
e183eaed | 1983 | switch (algorithm) { |
1da177e4 | 1984 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1985 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1986 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1987 | (*dd_idx)++; |
1988 | break; | |
1989 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1990 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1991 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1992 | (*dd_idx)++; |
1993 | break; | |
1994 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1995 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1996 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1997 | break; |
1998 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1999 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2000 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2001 | break; |
99c0fb5f N |
2002 | case ALGORITHM_PARITY_0: |
2003 | pd_idx = 0; | |
2004 | (*dd_idx)++; | |
2005 | break; | |
2006 | case ALGORITHM_PARITY_N: | |
2007 | pd_idx = data_disks; | |
2008 | break; | |
1da177e4 | 2009 | default: |
99c0fb5f | 2010 | BUG(); |
16a53ecc N |
2011 | } |
2012 | break; | |
2013 | case 6: | |
2014 | ||
e183eaed | 2015 | switch (algorithm) { |
16a53ecc | 2016 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2017 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2018 | qd_idx = pd_idx + 1; |
2019 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2020 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2021 | qd_idx = 0; |
2022 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2023 | (*dd_idx) += 2; /* D D P Q D */ |
2024 | break; | |
2025 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2026 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2027 | qd_idx = pd_idx + 1; |
2028 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2029 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2030 | qd_idx = 0; |
2031 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2032 | (*dd_idx) += 2; /* D D P Q D */ |
2033 | break; | |
2034 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2035 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2036 | qd_idx = (pd_idx + 1) % raid_disks; |
2037 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2038 | break; |
2039 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2040 | pd_idx = 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 | 2043 | break; |
99c0fb5f N |
2044 | |
2045 | case ALGORITHM_PARITY_0: | |
2046 | pd_idx = 0; | |
2047 | qd_idx = 1; | |
2048 | (*dd_idx) += 2; | |
2049 | break; | |
2050 | case ALGORITHM_PARITY_N: | |
2051 | pd_idx = data_disks; | |
2052 | qd_idx = data_disks + 1; | |
2053 | break; | |
2054 | ||
2055 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2056 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2057 | * of blocks for computing Q is different. | |
2058 | */ | |
6e3b96ed | 2059 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2060 | qd_idx = pd_idx + 1; |
2061 | if (pd_idx == raid_disks-1) { | |
2062 | (*dd_idx)++; /* Q D D D P */ | |
2063 | qd_idx = 0; | |
2064 | } else if (*dd_idx >= pd_idx) | |
2065 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2066 | ddf_layout = 1; |
99c0fb5f N |
2067 | break; |
2068 | ||
2069 | case ALGORITHM_ROTATING_N_RESTART: | |
2070 | /* Same a left_asymmetric, by first stripe is | |
2071 | * D D D P Q rather than | |
2072 | * Q D D D P | |
2073 | */ | |
6e3b96ed N |
2074 | stripe2 += 1; |
2075 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2076 | qd_idx = pd_idx + 1; |
2077 | if (pd_idx == raid_disks-1) { | |
2078 | (*dd_idx)++; /* Q D D D P */ | |
2079 | qd_idx = 0; | |
2080 | } else if (*dd_idx >= pd_idx) | |
2081 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2082 | ddf_layout = 1; |
99c0fb5f N |
2083 | break; |
2084 | ||
2085 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2086 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2087 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2088 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2089 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2090 | ddf_layout = 1; |
99c0fb5f N |
2091 | break; |
2092 | ||
2093 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2094 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2095 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2096 | if (*dd_idx >= pd_idx) |
2097 | (*dd_idx)++; | |
2098 | qd_idx = raid_disks - 1; | |
2099 | break; | |
2100 | ||
2101 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2102 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2103 | if (*dd_idx >= pd_idx) |
2104 | (*dd_idx)++; | |
2105 | qd_idx = raid_disks - 1; | |
2106 | break; | |
2107 | ||
2108 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2109 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2110 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2111 | qd_idx = raid_disks - 1; | |
2112 | break; | |
2113 | ||
2114 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2115 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2116 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2117 | qd_idx = raid_disks - 1; | |
2118 | break; | |
2119 | ||
2120 | case ALGORITHM_PARITY_0_6: | |
2121 | pd_idx = 0; | |
2122 | (*dd_idx)++; | |
2123 | qd_idx = raid_disks - 1; | |
2124 | break; | |
2125 | ||
16a53ecc | 2126 | default: |
99c0fb5f | 2127 | BUG(); |
16a53ecc N |
2128 | } |
2129 | break; | |
1da177e4 LT |
2130 | } |
2131 | ||
911d4ee8 N |
2132 | if (sh) { |
2133 | sh->pd_idx = pd_idx; | |
2134 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2135 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2136 | } |
1da177e4 LT |
2137 | /* |
2138 | * Finally, compute the new sector number | |
2139 | */ | |
2140 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2141 | return new_sector; | |
2142 | } | |
2143 | ||
2144 | ||
784052ec | 2145 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2146 | { |
d1688a6d | 2147 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2148 | int raid_disks = sh->disks; |
2149 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2150 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2151 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2152 | : conf->chunk_sectors; | |
e183eaed N |
2153 | int algorithm = previous ? conf->prev_algo |
2154 | : conf->algorithm; | |
1da177e4 LT |
2155 | sector_t stripe; |
2156 | int chunk_offset; | |
35f2a591 N |
2157 | sector_t chunk_number; |
2158 | int dummy1, dd_idx = i; | |
1da177e4 | 2159 | sector_t r_sector; |
911d4ee8 | 2160 | struct stripe_head sh2; |
1da177e4 | 2161 | |
16a53ecc | 2162 | |
1da177e4 LT |
2163 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
2164 | stripe = new_sector; | |
1da177e4 | 2165 | |
16a53ecc N |
2166 | if (i == sh->pd_idx) |
2167 | return 0; | |
2168 | switch(conf->level) { | |
2169 | case 4: break; | |
2170 | case 5: | |
e183eaed | 2171 | switch (algorithm) { |
1da177e4 LT |
2172 | case ALGORITHM_LEFT_ASYMMETRIC: |
2173 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2174 | if (i > sh->pd_idx) | |
2175 | i--; | |
2176 | break; | |
2177 | case ALGORITHM_LEFT_SYMMETRIC: | |
2178 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2179 | if (i < sh->pd_idx) | |
2180 | i += raid_disks; | |
2181 | i -= (sh->pd_idx + 1); | |
2182 | break; | |
99c0fb5f N |
2183 | case ALGORITHM_PARITY_0: |
2184 | i -= 1; | |
2185 | break; | |
2186 | case ALGORITHM_PARITY_N: | |
2187 | break; | |
1da177e4 | 2188 | default: |
99c0fb5f | 2189 | BUG(); |
16a53ecc N |
2190 | } |
2191 | break; | |
2192 | case 6: | |
d0dabf7e | 2193 | if (i == sh->qd_idx) |
16a53ecc | 2194 | return 0; /* It is the Q disk */ |
e183eaed | 2195 | switch (algorithm) { |
16a53ecc N |
2196 | case ALGORITHM_LEFT_ASYMMETRIC: |
2197 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2198 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2199 | case ALGORITHM_ROTATING_N_RESTART: | |
2200 | if (sh->pd_idx == raid_disks-1) | |
2201 | i--; /* Q D D D P */ | |
16a53ecc N |
2202 | else if (i > sh->pd_idx) |
2203 | i -= 2; /* D D P Q D */ | |
2204 | break; | |
2205 | case ALGORITHM_LEFT_SYMMETRIC: | |
2206 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2207 | if (sh->pd_idx == raid_disks-1) | |
2208 | i--; /* Q D D D P */ | |
2209 | else { | |
2210 | /* D D P Q D */ | |
2211 | if (i < sh->pd_idx) | |
2212 | i += raid_disks; | |
2213 | i -= (sh->pd_idx + 2); | |
2214 | } | |
2215 | break; | |
99c0fb5f N |
2216 | case ALGORITHM_PARITY_0: |
2217 | i -= 2; | |
2218 | break; | |
2219 | case ALGORITHM_PARITY_N: | |
2220 | break; | |
2221 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2222 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2223 | if (sh->pd_idx == 0) |
2224 | i--; /* P D D D Q */ | |
e4424fee N |
2225 | else { |
2226 | /* D D Q P D */ | |
2227 | if (i < sh->pd_idx) | |
2228 | i += raid_disks; | |
2229 | i -= (sh->pd_idx + 1); | |
2230 | } | |
99c0fb5f N |
2231 | break; |
2232 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2233 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2234 | if (i > sh->pd_idx) | |
2235 | i--; | |
2236 | break; | |
2237 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2238 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2239 | if (i < sh->pd_idx) | |
2240 | i += data_disks + 1; | |
2241 | i -= (sh->pd_idx + 1); | |
2242 | break; | |
2243 | case ALGORITHM_PARITY_0_6: | |
2244 | i -= 1; | |
2245 | break; | |
16a53ecc | 2246 | default: |
99c0fb5f | 2247 | BUG(); |
16a53ecc N |
2248 | } |
2249 | break; | |
1da177e4 LT |
2250 | } |
2251 | ||
2252 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2253 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2254 | |
112bf897 | 2255 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2256 | previous, &dummy1, &sh2); |
911d4ee8 N |
2257 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2258 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2259 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2260 | mdname(conf->mddev)); | |
1da177e4 LT |
2261 | return 0; |
2262 | } | |
2263 | return r_sector; | |
2264 | } | |
2265 | ||
2266 | ||
600aa109 | 2267 | static void |
c0f7bddb | 2268 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2269 | int rcw, int expand) |
e33129d8 DW |
2270 | { |
2271 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
d1688a6d | 2272 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2273 | int level = conf->level; |
e33129d8 DW |
2274 | |
2275 | if (rcw) { | |
2276 | /* if we are not expanding this is a proper write request, and | |
2277 | * there will be bios with new data to be drained into the | |
2278 | * stripe cache | |
2279 | */ | |
2280 | if (!expand) { | |
600aa109 DW |
2281 | sh->reconstruct_state = reconstruct_state_drain_run; |
2282 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2283 | } else | |
2284 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2285 | |
ac6b53b6 | 2286 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2287 | |
2288 | for (i = disks; i--; ) { | |
2289 | struct r5dev *dev = &sh->dev[i]; | |
2290 | ||
2291 | if (dev->towrite) { | |
2292 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2293 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2294 | if (!expand) |
2295 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2296 | s->locked++; |
e33129d8 DW |
2297 | } |
2298 | } | |
c0f7bddb | 2299 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2300 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2301 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2302 | } else { |
c0f7bddb | 2303 | BUG_ON(level == 6); |
e33129d8 DW |
2304 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2305 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2306 | ||
d8ee0728 | 2307 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2308 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2309 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2310 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2311 | |
2312 | for (i = disks; i--; ) { | |
2313 | struct r5dev *dev = &sh->dev[i]; | |
2314 | if (i == pd_idx) | |
2315 | continue; | |
2316 | ||
e33129d8 DW |
2317 | if (dev->towrite && |
2318 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2319 | test_bit(R5_Wantcompute, &dev->flags))) { |
2320 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2321 | set_bit(R5_LOCKED, &dev->flags); |
2322 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2323 | s->locked++; |
e33129d8 DW |
2324 | } |
2325 | } | |
2326 | } | |
2327 | ||
c0f7bddb | 2328 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2329 | * are in flight |
2330 | */ | |
2331 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2332 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2333 | s->locked++; |
e33129d8 | 2334 | |
c0f7bddb YT |
2335 | if (level == 6) { |
2336 | int qd_idx = sh->qd_idx; | |
2337 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2338 | ||
2339 | set_bit(R5_LOCKED, &dev->flags); | |
2340 | clear_bit(R5_UPTODATE, &dev->flags); | |
2341 | s->locked++; | |
2342 | } | |
2343 | ||
600aa109 | 2344 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2345 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2346 | s->locked, s->ops_request); |
e33129d8 | 2347 | } |
16a53ecc | 2348 | |
1da177e4 LT |
2349 | /* |
2350 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2351 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2352 | * The bi_next chain must be in order. |
2353 | */ | |
2354 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2355 | { | |
2356 | struct bio **bip; | |
d1688a6d | 2357 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2358 | int firstwrite=0; |
1da177e4 | 2359 | |
cbe47ec5 | 2360 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2361 | (unsigned long long)bi->bi_sector, |
2362 | (unsigned long long)sh->sector); | |
2363 | ||
b17459c0 SL |
2364 | /* |
2365 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
2366 | * reference count to avoid race. The reference count should already be | |
2367 | * increased before this function is called (for example, in | |
2368 | * make_request()), so other bio sharing this stripe will not free the | |
2369 | * stripe. If a stripe is owned by one stripe, the stripe lock will | |
2370 | * protect it. | |
2371 | */ | |
2372 | spin_lock_irq(&sh->stripe_lock); | |
72626685 | 2373 | if (forwrite) { |
1da177e4 | 2374 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 2375 | if (*bip == NULL) |
72626685 N |
2376 | firstwrite = 1; |
2377 | } else | |
1da177e4 LT |
2378 | bip = &sh->dev[dd_idx].toread; |
2379 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2380 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2381 | goto overlap; | |
2382 | bip = & (*bip)->bi_next; | |
2383 | } | |
2384 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2385 | goto overlap; | |
2386 | ||
78bafebd | 2387 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2388 | if (*bip) |
2389 | bi->bi_next = *bip; | |
2390 | *bip = bi; | |
e7836bd6 | 2391 | raid5_inc_bi_active_stripes(bi); |
72626685 | 2392 | |
1da177e4 LT |
2393 | if (forwrite) { |
2394 | /* check if page is covered */ | |
2395 | sector_t sector = sh->dev[dd_idx].sector; | |
2396 | for (bi=sh->dev[dd_idx].towrite; | |
2397 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2398 | bi && bi->bi_sector <= sector; | |
2399 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2400 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2401 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2402 | } | |
2403 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2404 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2405 | } | |
b17459c0 | 2406 | spin_unlock_irq(&sh->stripe_lock); |
cbe47ec5 N |
2407 | |
2408 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2409 | (unsigned long long)(*bip)->bi_sector, | |
2410 | (unsigned long long)sh->sector, dd_idx); | |
2411 | ||
2412 | if (conf->mddev->bitmap && firstwrite) { | |
2413 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2414 | STRIPE_SECTORS, 0); | |
2415 | sh->bm_seq = conf->seq_flush+1; | |
2416 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2417 | } | |
1da177e4 LT |
2418 | return 1; |
2419 | ||
2420 | overlap: | |
2421 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 2422 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
2423 | return 0; |
2424 | } | |
2425 | ||
d1688a6d | 2426 | static void end_reshape(struct r5conf *conf); |
29269553 | 2427 | |
d1688a6d | 2428 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 2429 | struct stripe_head *sh) |
ccfcc3c1 | 2430 | { |
784052ec | 2431 | int sectors_per_chunk = |
09c9e5fa | 2432 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2433 | int dd_idx; |
2d2063ce | 2434 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2435 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2436 | |
112bf897 N |
2437 | raid5_compute_sector(conf, |
2438 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2439 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2440 | previous, |
911d4ee8 | 2441 | &dd_idx, sh); |
ccfcc3c1 N |
2442 | } |
2443 | ||
a4456856 | 2444 | static void |
d1688a6d | 2445 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2446 | struct stripe_head_state *s, int disks, |
2447 | struct bio **return_bi) | |
2448 | { | |
2449 | int i; | |
2450 | for (i = disks; i--; ) { | |
2451 | struct bio *bi; | |
2452 | int bitmap_end = 0; | |
2453 | ||
2454 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 2455 | struct md_rdev *rdev; |
a4456856 DW |
2456 | rcu_read_lock(); |
2457 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2458 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
2459 | atomic_inc(&rdev->nr_pending); |
2460 | else | |
2461 | rdev = NULL; | |
a4456856 | 2462 | rcu_read_unlock(); |
7f0da59b N |
2463 | if (rdev) { |
2464 | if (!rdev_set_badblocks( | |
2465 | rdev, | |
2466 | sh->sector, | |
2467 | STRIPE_SECTORS, 0)) | |
2468 | md_error(conf->mddev, rdev); | |
2469 | rdev_dec_pending(rdev, conf->mddev); | |
2470 | } | |
a4456856 | 2471 | } |
b17459c0 | 2472 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
2473 | /* fail all writes first */ |
2474 | bi = sh->dev[i].towrite; | |
2475 | sh->dev[i].towrite = NULL; | |
b17459c0 | 2476 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
2477 | if (bi) { |
2478 | s->to_write--; | |
2479 | bitmap_end = 1; | |
2480 | } | |
2481 | ||
2482 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2483 | wake_up(&conf->wait_for_overlap); | |
2484 | ||
2485 | while (bi && bi->bi_sector < | |
2486 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2487 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2488 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2489 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2490 | md_write_end(conf->mddev); |
2491 | bi->bi_next = *return_bi; | |
2492 | *return_bi = bi; | |
2493 | } | |
2494 | bi = nextbi; | |
2495 | } | |
7eaf7e8e SL |
2496 | if (bitmap_end) |
2497 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2498 | STRIPE_SECTORS, 0, 0); | |
2499 | bitmap_end = 0; | |
a4456856 DW |
2500 | /* and fail all 'written' */ |
2501 | bi = sh->dev[i].written; | |
2502 | sh->dev[i].written = NULL; | |
2503 | if (bi) bitmap_end = 1; | |
2504 | while (bi && bi->bi_sector < | |
2505 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2506 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2507 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2508 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2509 | md_write_end(conf->mddev); |
2510 | bi->bi_next = *return_bi; | |
2511 | *return_bi = bi; | |
2512 | } | |
2513 | bi = bi2; | |
2514 | } | |
2515 | ||
b5e98d65 DW |
2516 | /* fail any reads if this device is non-operational and |
2517 | * the data has not reached the cache yet. | |
2518 | */ | |
2519 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2520 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2521 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2522 | bi = sh->dev[i].toread; |
2523 | sh->dev[i].toread = NULL; | |
2524 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2525 | wake_up(&conf->wait_for_overlap); | |
2526 | if (bi) s->to_read--; | |
2527 | while (bi && bi->bi_sector < | |
2528 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2529 | struct bio *nextbi = | |
2530 | r5_next_bio(bi, sh->dev[i].sector); | |
2531 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2532 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2533 | bi->bi_next = *return_bi; |
2534 | *return_bi = bi; | |
2535 | } | |
2536 | bi = nextbi; | |
2537 | } | |
2538 | } | |
a4456856 DW |
2539 | if (bitmap_end) |
2540 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2541 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
2542 | /* If we were in the middle of a write the parity block might |
2543 | * still be locked - so just clear all R5_LOCKED flags | |
2544 | */ | |
2545 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
2546 | } |
2547 | ||
8b3e6cdc DW |
2548 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2549 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2550 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2551 | } |
2552 | ||
7f0da59b | 2553 | static void |
d1688a6d | 2554 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
2555 | struct stripe_head_state *s) |
2556 | { | |
2557 | int abort = 0; | |
2558 | int i; | |
2559 | ||
7f0da59b N |
2560 | clear_bit(STRIPE_SYNCING, &sh->state); |
2561 | s->syncing = 0; | |
9a3e1101 | 2562 | s->replacing = 0; |
7f0da59b | 2563 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
2564 | * Don't even need to abort as that is handled elsewhere |
2565 | * if needed, and not always wanted e.g. if there is a known | |
2566 | * bad block here. | |
9a3e1101 | 2567 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
2568 | * non-sync devices, or abort the recovery |
2569 | */ | |
18b9837e N |
2570 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
2571 | /* During recovery devices cannot be removed, so | |
2572 | * locking and refcounting of rdevs is not needed | |
2573 | */ | |
2574 | for (i = 0; i < conf->raid_disks; i++) { | |
2575 | struct md_rdev *rdev = conf->disks[i].rdev; | |
2576 | if (rdev | |
2577 | && !test_bit(Faulty, &rdev->flags) | |
2578 | && !test_bit(In_sync, &rdev->flags) | |
2579 | && !rdev_set_badblocks(rdev, sh->sector, | |
2580 | STRIPE_SECTORS, 0)) | |
2581 | abort = 1; | |
2582 | rdev = conf->disks[i].replacement; | |
2583 | if (rdev | |
2584 | && !test_bit(Faulty, &rdev->flags) | |
2585 | && !test_bit(In_sync, &rdev->flags) | |
2586 | && !rdev_set_badblocks(rdev, sh->sector, | |
2587 | STRIPE_SECTORS, 0)) | |
2588 | abort = 1; | |
2589 | } | |
2590 | if (abort) | |
2591 | conf->recovery_disabled = | |
2592 | conf->mddev->recovery_disabled; | |
7f0da59b | 2593 | } |
18b9837e | 2594 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
2595 | } |
2596 | ||
9a3e1101 N |
2597 | static int want_replace(struct stripe_head *sh, int disk_idx) |
2598 | { | |
2599 | struct md_rdev *rdev; | |
2600 | int rv = 0; | |
2601 | /* Doing recovery so rcu locking not required */ | |
2602 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
2603 | if (rdev | |
2604 | && !test_bit(Faulty, &rdev->flags) | |
2605 | && !test_bit(In_sync, &rdev->flags) | |
2606 | && (rdev->recovery_offset <= sh->sector | |
2607 | || rdev->mddev->recovery_cp <= sh->sector)) | |
2608 | rv = 1; | |
2609 | ||
2610 | return rv; | |
2611 | } | |
2612 | ||
93b3dbce | 2613 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
2614 | * to be read or computed to satisfy a request. |
2615 | * | |
2616 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 2617 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 2618 | */ |
93b3dbce N |
2619 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
2620 | int disk_idx, int disks) | |
a4456856 | 2621 | { |
5599becc | 2622 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2623 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2624 | &sh->dev[s->failed_num[1]] }; | |
5599becc | 2625 | |
93b3dbce | 2626 | /* is the data in this block needed, and can we get it? */ |
5599becc YT |
2627 | if (!test_bit(R5_LOCKED, &dev->flags) && |
2628 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2629 | (dev->toread || | |
2630 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2631 | s->syncing || s->expanding || | |
9a3e1101 | 2632 | (s->replacing && want_replace(sh, disk_idx)) || |
5d35e09c N |
2633 | (s->failed >= 1 && fdev[0]->toread) || |
2634 | (s->failed >= 2 && fdev[1]->toread) || | |
93b3dbce N |
2635 | (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite && |
2636 | !test_bit(R5_OVERWRITE, &fdev[0]->flags)) || | |
2637 | (sh->raid_conf->level == 6 && s->failed && s->to_write))) { | |
5599becc YT |
2638 | /* we would like to get this block, possibly by computing it, |
2639 | * otherwise read it if the backing disk is insync | |
2640 | */ | |
2641 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2642 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2643 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
2644 | (s->failed && (disk_idx == s->failed_num[0] || |
2645 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
2646 | /* have disk failed, and we're requested to fetch it; |
2647 | * do compute it | |
a4456856 | 2648 | */ |
5599becc YT |
2649 | pr_debug("Computing stripe %llu block %d\n", |
2650 | (unsigned long long)sh->sector, disk_idx); | |
2651 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2652 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2653 | set_bit(R5_Wantcompute, &dev->flags); | |
2654 | sh->ops.target = disk_idx; | |
2655 | sh->ops.target2 = -1; /* no 2nd target */ | |
2656 | s->req_compute = 1; | |
93b3dbce N |
2657 | /* Careful: from this point on 'uptodate' is in the eye |
2658 | * of raid_run_ops which services 'compute' operations | |
2659 | * before writes. R5_Wantcompute flags a block that will | |
2660 | * be R5_UPTODATE by the time it is needed for a | |
2661 | * subsequent operation. | |
2662 | */ | |
5599becc YT |
2663 | s->uptodate++; |
2664 | return 1; | |
2665 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2666 | /* Computing 2-failure is *very* expensive; only | |
2667 | * do it if failed >= 2 | |
2668 | */ | |
2669 | int other; | |
2670 | for (other = disks; other--; ) { | |
2671 | if (other == disk_idx) | |
2672 | continue; | |
2673 | if (!test_bit(R5_UPTODATE, | |
2674 | &sh->dev[other].flags)) | |
2675 | break; | |
a4456856 | 2676 | } |
5599becc YT |
2677 | BUG_ON(other < 0); |
2678 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2679 | (unsigned long long)sh->sector, | |
2680 | disk_idx, other); | |
2681 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2682 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2683 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2684 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2685 | sh->ops.target = disk_idx; | |
2686 | sh->ops.target2 = other; | |
2687 | s->uptodate += 2; | |
2688 | s->req_compute = 1; | |
2689 | return 1; | |
2690 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2691 | set_bit(R5_LOCKED, &dev->flags); | |
2692 | set_bit(R5_Wantread, &dev->flags); | |
2693 | s->locked++; | |
2694 | pr_debug("Reading block %d (sync=%d)\n", | |
2695 | disk_idx, s->syncing); | |
a4456856 DW |
2696 | } |
2697 | } | |
5599becc YT |
2698 | |
2699 | return 0; | |
2700 | } | |
2701 | ||
2702 | /** | |
93b3dbce | 2703 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 2704 | */ |
93b3dbce N |
2705 | static void handle_stripe_fill(struct stripe_head *sh, |
2706 | struct stripe_head_state *s, | |
2707 | int disks) | |
5599becc YT |
2708 | { |
2709 | int i; | |
2710 | ||
2711 | /* look for blocks to read/compute, skip this if a compute | |
2712 | * is already in flight, or if the stripe contents are in the | |
2713 | * midst of changing due to a write | |
2714 | */ | |
2715 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2716 | !sh->reconstruct_state) | |
2717 | for (i = disks; i--; ) | |
93b3dbce | 2718 | if (fetch_block(sh, s, i, disks)) |
5599becc | 2719 | break; |
a4456856 DW |
2720 | set_bit(STRIPE_HANDLE, &sh->state); |
2721 | } | |
2722 | ||
2723 | ||
1fe797e6 | 2724 | /* handle_stripe_clean_event |
a4456856 DW |
2725 | * any written block on an uptodate or failed drive can be returned. |
2726 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2727 | * never LOCKED, so we don't need to test 'failed' directly. | |
2728 | */ | |
d1688a6d | 2729 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
2730 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2731 | { | |
2732 | int i; | |
2733 | struct r5dev *dev; | |
2734 | ||
2735 | for (i = disks; i--; ) | |
2736 | if (sh->dev[i].written) { | |
2737 | dev = &sh->dev[i]; | |
2738 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2739 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2740 | /* We can return any write requests */ | |
2741 | struct bio *wbi, *wbi2; | |
45b4233c | 2742 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2743 | wbi = dev->written; |
2744 | dev->written = NULL; | |
2745 | while (wbi && wbi->bi_sector < | |
2746 | dev->sector + STRIPE_SECTORS) { | |
2747 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 2748 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 DW |
2749 | md_write_end(conf->mddev); |
2750 | wbi->bi_next = *return_bi; | |
2751 | *return_bi = wbi; | |
2752 | } | |
2753 | wbi = wbi2; | |
2754 | } | |
7eaf7e8e SL |
2755 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
2756 | STRIPE_SECTORS, | |
a4456856 | 2757 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 2758 | 0); |
a4456856 DW |
2759 | } |
2760 | } | |
8b3e6cdc DW |
2761 | |
2762 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2763 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2764 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2765 | } |
2766 | ||
d1688a6d | 2767 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
2768 | struct stripe_head *sh, |
2769 | struct stripe_head_state *s, | |
2770 | int disks) | |
a4456856 DW |
2771 | { |
2772 | int rmw = 0, rcw = 0, i; | |
c8ac1803 N |
2773 | if (conf->max_degraded == 2) { |
2774 | /* RAID6 requires 'rcw' in current implementation | |
2775 | * Calculate the real rcw later - for now fake it | |
2776 | * look like rcw is cheaper | |
2777 | */ | |
2778 | rcw = 1; rmw = 2; | |
2779 | } else for (i = disks; i--; ) { | |
a4456856 DW |
2780 | /* would I have to read this buffer for read_modify_write */ |
2781 | struct r5dev *dev = &sh->dev[i]; | |
2782 | if ((dev->towrite || i == sh->pd_idx) && | |
2783 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2784 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2785 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2786 | if (test_bit(R5_Insync, &dev->flags)) |
2787 | rmw++; | |
2788 | else | |
2789 | rmw += 2*disks; /* cannot read it */ | |
2790 | } | |
2791 | /* Would I have to read this buffer for reconstruct_write */ | |
2792 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2793 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2794 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2795 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2796 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2797 | else |
2798 | rcw += 2*disks; | |
2799 | } | |
2800 | } | |
45b4233c | 2801 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2802 | (unsigned long long)sh->sector, rmw, rcw); |
2803 | set_bit(STRIPE_HANDLE, &sh->state); | |
2804 | if (rmw < rcw && rmw > 0) | |
2805 | /* prefer read-modify-write, but need to get some data */ | |
2806 | for (i = disks; i--; ) { | |
2807 | struct r5dev *dev = &sh->dev[i]; | |
2808 | if ((dev->towrite || i == sh->pd_idx) && | |
2809 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2810 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2811 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2812 | test_bit(R5_Insync, &dev->flags)) { |
2813 | if ( | |
2814 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2815 | pr_debug("Read_old block " |
a4456856 DW |
2816 | "%d for r-m-w\n", i); |
2817 | set_bit(R5_LOCKED, &dev->flags); | |
2818 | set_bit(R5_Wantread, &dev->flags); | |
2819 | s->locked++; | |
2820 | } else { | |
2821 | set_bit(STRIPE_DELAYED, &sh->state); | |
2822 | set_bit(STRIPE_HANDLE, &sh->state); | |
2823 | } | |
2824 | } | |
2825 | } | |
c8ac1803 | 2826 | if (rcw <= rmw && rcw > 0) { |
a4456856 | 2827 | /* want reconstruct write, but need to get some data */ |
c8ac1803 | 2828 | rcw = 0; |
a4456856 DW |
2829 | for (i = disks; i--; ) { |
2830 | struct r5dev *dev = &sh->dev[i]; | |
2831 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 2832 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 2833 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 2834 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
2835 | test_bit(R5_Wantcompute, &dev->flags))) { |
2836 | rcw++; | |
2837 | if (!test_bit(R5_Insync, &dev->flags)) | |
2838 | continue; /* it's a failed drive */ | |
a4456856 DW |
2839 | if ( |
2840 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2841 | pr_debug("Read_old block " |
a4456856 DW |
2842 | "%d for Reconstruct\n", i); |
2843 | set_bit(R5_LOCKED, &dev->flags); | |
2844 | set_bit(R5_Wantread, &dev->flags); | |
2845 | s->locked++; | |
2846 | } else { | |
2847 | set_bit(STRIPE_DELAYED, &sh->state); | |
2848 | set_bit(STRIPE_HANDLE, &sh->state); | |
2849 | } | |
2850 | } | |
2851 | } | |
c8ac1803 | 2852 | } |
a4456856 DW |
2853 | /* now if nothing is locked, and if we have enough data, |
2854 | * we can start a write request | |
2855 | */ | |
f38e1219 DW |
2856 | /* since handle_stripe can be called at any time we need to handle the |
2857 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2858 | * subsequent call wants to start a write request. raid_run_ops only |
2859 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2860 | * simultaneously. If this is not the case then new writes need to be |
2861 | * held off until the compute completes. | |
2862 | */ | |
976ea8d4 DW |
2863 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2864 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2865 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2866 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2867 | } |
2868 | ||
d1688a6d | 2869 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2870 | struct stripe_head_state *s, int disks) |
2871 | { | |
ecc65c9b | 2872 | struct r5dev *dev = NULL; |
bd2ab670 | 2873 | |
a4456856 | 2874 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2875 | |
ecc65c9b DW |
2876 | switch (sh->check_state) { |
2877 | case check_state_idle: | |
2878 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2879 | if (s->failed == 0) { |
bd2ab670 | 2880 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2881 | sh->check_state = check_state_run; |
2882 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2883 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2884 | s->uptodate--; |
ecc65c9b | 2885 | break; |
bd2ab670 | 2886 | } |
f2b3b44d | 2887 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
2888 | /* fall through */ |
2889 | case check_state_compute_result: | |
2890 | sh->check_state = check_state_idle; | |
2891 | if (!dev) | |
2892 | dev = &sh->dev[sh->pd_idx]; | |
2893 | ||
2894 | /* check that a write has not made the stripe insync */ | |
2895 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2896 | break; | |
c8894419 | 2897 | |
a4456856 | 2898 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2899 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2900 | BUG_ON(s->uptodate != disks); | |
2901 | ||
2902 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2903 | s->locked++; |
a4456856 | 2904 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2905 | |
a4456856 | 2906 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2907 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2908 | break; |
2909 | case check_state_run: | |
2910 | break; /* we will be called again upon completion */ | |
2911 | case check_state_check_result: | |
2912 | sh->check_state = check_state_idle; | |
2913 | ||
2914 | /* if a failure occurred during the check operation, leave | |
2915 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2916 | */ | |
2917 | if (s->failed) | |
2918 | break; | |
2919 | ||
2920 | /* handle a successful check operation, if parity is correct | |
2921 | * we are done. Otherwise update the mismatch count and repair | |
2922 | * parity if !MD_RECOVERY_CHECK | |
2923 | */ | |
ad283ea4 | 2924 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2925 | /* parity is correct (on disc, |
2926 | * not in buffer any more) | |
2927 | */ | |
2928 | set_bit(STRIPE_INSYNC, &sh->state); | |
2929 | else { | |
2930 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2931 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2932 | /* don't try to repair!! */ | |
2933 | set_bit(STRIPE_INSYNC, &sh->state); | |
2934 | else { | |
2935 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2936 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2937 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2938 | set_bit(R5_Wantcompute, | |
2939 | &sh->dev[sh->pd_idx].flags); | |
2940 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2941 | sh->ops.target2 = -1; |
ecc65c9b DW |
2942 | s->uptodate++; |
2943 | } | |
2944 | } | |
2945 | break; | |
2946 | case check_state_compute_run: | |
2947 | break; | |
2948 | default: | |
2949 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2950 | __func__, sh->check_state, | |
2951 | (unsigned long long) sh->sector); | |
2952 | BUG(); | |
a4456856 DW |
2953 | } |
2954 | } | |
2955 | ||
2956 | ||
d1688a6d | 2957 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 2958 | struct stripe_head_state *s, |
f2b3b44d | 2959 | int disks) |
a4456856 | 2960 | { |
a4456856 | 2961 | int pd_idx = sh->pd_idx; |
34e04e87 | 2962 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2963 | struct r5dev *dev; |
a4456856 DW |
2964 | |
2965 | set_bit(STRIPE_HANDLE, &sh->state); | |
2966 | ||
2967 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2968 | |
a4456856 DW |
2969 | /* Want to check and possibly repair P and Q. |
2970 | * However there could be one 'failed' device, in which | |
2971 | * case we can only check one of them, possibly using the | |
2972 | * other to generate missing data | |
2973 | */ | |
2974 | ||
d82dfee0 DW |
2975 | switch (sh->check_state) { |
2976 | case check_state_idle: | |
2977 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 2978 | if (s->failed == s->q_failed) { |
d82dfee0 | 2979 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2980 | * makes sense to check P (If anything else were failed, |
2981 | * we would have used P to recreate it). | |
2982 | */ | |
d82dfee0 | 2983 | sh->check_state = check_state_run; |
a4456856 | 2984 | } |
f2b3b44d | 2985 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 2986 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2987 | * anything, so it makes sense to check it |
2988 | */ | |
d82dfee0 DW |
2989 | if (sh->check_state == check_state_run) |
2990 | sh->check_state = check_state_run_pq; | |
2991 | else | |
2992 | sh->check_state = check_state_run_q; | |
a4456856 | 2993 | } |
a4456856 | 2994 | |
d82dfee0 DW |
2995 | /* discard potentially stale zero_sum_result */ |
2996 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2997 | |
d82dfee0 DW |
2998 | if (sh->check_state == check_state_run) { |
2999 | /* async_xor_zero_sum destroys the contents of P */ | |
3000 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
3001 | s->uptodate--; | |
a4456856 | 3002 | } |
d82dfee0 DW |
3003 | if (sh->check_state >= check_state_run && |
3004 | sh->check_state <= check_state_run_pq) { | |
3005 | /* async_syndrome_zero_sum preserves P and Q, so | |
3006 | * no need to mark them !uptodate here | |
3007 | */ | |
3008 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
3009 | break; | |
a4456856 DW |
3010 | } |
3011 | ||
d82dfee0 DW |
3012 | /* we have 2-disk failure */ |
3013 | BUG_ON(s->failed != 2); | |
3014 | /* fall through */ | |
3015 | case check_state_compute_result: | |
3016 | sh->check_state = check_state_idle; | |
a4456856 | 3017 | |
d82dfee0 DW |
3018 | /* check that a write has not made the stripe insync */ |
3019 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3020 | break; | |
a4456856 DW |
3021 | |
3022 | /* now write out any block on a failed drive, | |
d82dfee0 | 3023 | * or P or Q if they were recomputed |
a4456856 | 3024 | */ |
d82dfee0 | 3025 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 3026 | if (s->failed == 2) { |
f2b3b44d | 3027 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
3028 | s->locked++; |
3029 | set_bit(R5_LOCKED, &dev->flags); | |
3030 | set_bit(R5_Wantwrite, &dev->flags); | |
3031 | } | |
3032 | if (s->failed >= 1) { | |
f2b3b44d | 3033 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
3034 | s->locked++; |
3035 | set_bit(R5_LOCKED, &dev->flags); | |
3036 | set_bit(R5_Wantwrite, &dev->flags); | |
3037 | } | |
d82dfee0 | 3038 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
3039 | dev = &sh->dev[pd_idx]; |
3040 | s->locked++; | |
3041 | set_bit(R5_LOCKED, &dev->flags); | |
3042 | set_bit(R5_Wantwrite, &dev->flags); | |
3043 | } | |
d82dfee0 | 3044 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
3045 | dev = &sh->dev[qd_idx]; |
3046 | s->locked++; | |
3047 | set_bit(R5_LOCKED, &dev->flags); | |
3048 | set_bit(R5_Wantwrite, &dev->flags); | |
3049 | } | |
3050 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
3051 | ||
3052 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
3053 | break; |
3054 | case check_state_run: | |
3055 | case check_state_run_q: | |
3056 | case check_state_run_pq: | |
3057 | break; /* we will be called again upon completion */ | |
3058 | case check_state_check_result: | |
3059 | sh->check_state = check_state_idle; | |
3060 | ||
3061 | /* handle a successful check operation, if parity is correct | |
3062 | * we are done. Otherwise update the mismatch count and repair | |
3063 | * parity if !MD_RECOVERY_CHECK | |
3064 | */ | |
3065 | if (sh->ops.zero_sum_result == 0) { | |
3066 | /* both parities are correct */ | |
3067 | if (!s->failed) | |
3068 | set_bit(STRIPE_INSYNC, &sh->state); | |
3069 | else { | |
3070 | /* in contrast to the raid5 case we can validate | |
3071 | * parity, but still have a failure to write | |
3072 | * back | |
3073 | */ | |
3074 | sh->check_state = check_state_compute_result; | |
3075 | /* Returning at this point means that we may go | |
3076 | * off and bring p and/or q uptodate again so | |
3077 | * we make sure to check zero_sum_result again | |
3078 | * to verify if p or q need writeback | |
3079 | */ | |
3080 | } | |
3081 | } else { | |
3082 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
3083 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
3084 | /* don't try to repair!! */ | |
3085 | set_bit(STRIPE_INSYNC, &sh->state); | |
3086 | else { | |
3087 | int *target = &sh->ops.target; | |
3088 | ||
3089 | sh->ops.target = -1; | |
3090 | sh->ops.target2 = -1; | |
3091 | sh->check_state = check_state_compute_run; | |
3092 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3093 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3094 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3095 | set_bit(R5_Wantcompute, | |
3096 | &sh->dev[pd_idx].flags); | |
3097 | *target = pd_idx; | |
3098 | target = &sh->ops.target2; | |
3099 | s->uptodate++; | |
3100 | } | |
3101 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3102 | set_bit(R5_Wantcompute, | |
3103 | &sh->dev[qd_idx].flags); | |
3104 | *target = qd_idx; | |
3105 | s->uptodate++; | |
3106 | } | |
3107 | } | |
3108 | } | |
3109 | break; | |
3110 | case check_state_compute_run: | |
3111 | break; | |
3112 | default: | |
3113 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3114 | __func__, sh->check_state, | |
3115 | (unsigned long long) sh->sector); | |
3116 | BUG(); | |
a4456856 DW |
3117 | } |
3118 | } | |
3119 | ||
d1688a6d | 3120 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3121 | { |
3122 | int i; | |
3123 | ||
3124 | /* We have read all the blocks in this stripe and now we need to | |
3125 | * copy some of them into a target stripe for expand. | |
3126 | */ | |
f0a50d37 | 3127 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
3128 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3129 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3130 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3131 | int dd_idx, j; |
a4456856 | 3132 | struct stripe_head *sh2; |
a08abd8c | 3133 | struct async_submit_ctl submit; |
a4456856 | 3134 | |
784052ec | 3135 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3136 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3137 | &dd_idx, NULL); | |
a8c906ca | 3138 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3139 | if (sh2 == NULL) |
3140 | /* so far only the early blocks of this stripe | |
3141 | * have been requested. When later blocks | |
3142 | * get requested, we will try again | |
3143 | */ | |
3144 | continue; | |
3145 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3146 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3147 | /* must have already done this block */ | |
3148 | release_stripe(sh2); | |
3149 | continue; | |
3150 | } | |
f0a50d37 DW |
3151 | |
3152 | /* place all the copies on one channel */ | |
a08abd8c | 3153 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3154 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3155 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3156 | &submit); |
f0a50d37 | 3157 | |
a4456856 DW |
3158 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3159 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3160 | for (j = 0; j < conf->raid_disks; j++) | |
3161 | if (j != sh2->pd_idx && | |
86c374ba | 3162 | j != sh2->qd_idx && |
a4456856 DW |
3163 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3164 | break; | |
3165 | if (j == conf->raid_disks) { | |
3166 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3167 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3168 | } | |
3169 | release_stripe(sh2); | |
f0a50d37 | 3170 | |
a4456856 | 3171 | } |
a2e08551 N |
3172 | /* done submitting copies, wait for them to complete */ |
3173 | if (tx) { | |
3174 | async_tx_ack(tx); | |
3175 | dma_wait_for_async_tx(tx); | |
3176 | } | |
a4456856 | 3177 | } |
1da177e4 LT |
3178 | |
3179 | /* | |
3180 | * handle_stripe - do things to a stripe. | |
3181 | * | |
9a3e1101 N |
3182 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3183 | * state of various bits to see what needs to be done. | |
1da177e4 | 3184 | * Possible results: |
9a3e1101 N |
3185 | * return some read requests which now have data |
3186 | * return some write requests which are safely on storage | |
1da177e4 LT |
3187 | * schedule a read on some buffers |
3188 | * schedule a write of some buffers | |
3189 | * return confirmation of parity correctness | |
3190 | * | |
1da177e4 | 3191 | */ |
a4456856 | 3192 | |
acfe726b | 3193 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3194 | { |
d1688a6d | 3195 | struct r5conf *conf = sh->raid_conf; |
f416885e | 3196 | int disks = sh->disks; |
474af965 N |
3197 | struct r5dev *dev; |
3198 | int i; | |
9a3e1101 | 3199 | int do_recovery = 0; |
1da177e4 | 3200 | |
acfe726b N |
3201 | memset(s, 0, sizeof(*s)); |
3202 | ||
acfe726b N |
3203 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3204 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
3205 | s->failed_num[0] = -1; | |
3206 | s->failed_num[1] = -1; | |
1da177e4 | 3207 | |
acfe726b | 3208 | /* Now to look around and see what can be done */ |
1da177e4 | 3209 | rcu_read_lock(); |
16a53ecc | 3210 | for (i=disks; i--; ) { |
3cb03002 | 3211 | struct md_rdev *rdev; |
31c176ec N |
3212 | sector_t first_bad; |
3213 | int bad_sectors; | |
3214 | int is_bad = 0; | |
acfe726b | 3215 | |
16a53ecc | 3216 | dev = &sh->dev[i]; |
1da177e4 | 3217 | |
45b4233c | 3218 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
3219 | i, dev->flags, |
3220 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
3221 | /* maybe we can reply to a read |
3222 | * | |
3223 | * new wantfill requests are only permitted while | |
3224 | * ops_complete_biofill is guaranteed to be inactive | |
3225 | */ | |
3226 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3227 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3228 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3229 | |
16a53ecc | 3230 | /* now count some things */ |
cc94015a N |
3231 | if (test_bit(R5_LOCKED, &dev->flags)) |
3232 | s->locked++; | |
3233 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3234 | s->uptodate++; | |
2d6e4ecc | 3235 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
3236 | s->compute++; |
3237 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 3238 | } |
1da177e4 | 3239 | |
acfe726b | 3240 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 3241 | s->to_fill++; |
acfe726b | 3242 | else if (dev->toread) |
cc94015a | 3243 | s->to_read++; |
16a53ecc | 3244 | if (dev->towrite) { |
cc94015a | 3245 | s->to_write++; |
16a53ecc | 3246 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3247 | s->non_overwrite++; |
16a53ecc | 3248 | } |
a4456856 | 3249 | if (dev->written) |
cc94015a | 3250 | s->written++; |
14a75d3e N |
3251 | /* Prefer to use the replacement for reads, but only |
3252 | * if it is recovered enough and has no bad blocks. | |
3253 | */ | |
3254 | rdev = rcu_dereference(conf->disks[i].replacement); | |
3255 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
3256 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
3257 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3258 | &first_bad, &bad_sectors)) | |
3259 | set_bit(R5_ReadRepl, &dev->flags); | |
3260 | else { | |
9a3e1101 N |
3261 | if (rdev) |
3262 | set_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
3263 | rdev = rcu_dereference(conf->disks[i].rdev); |
3264 | clear_bit(R5_ReadRepl, &dev->flags); | |
3265 | } | |
9283d8c5 N |
3266 | if (rdev && test_bit(Faulty, &rdev->flags)) |
3267 | rdev = NULL; | |
31c176ec N |
3268 | if (rdev) { |
3269 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3270 | &first_bad, &bad_sectors); | |
3271 | if (s->blocked_rdev == NULL | |
3272 | && (test_bit(Blocked, &rdev->flags) | |
3273 | || is_bad < 0)) { | |
3274 | if (is_bad < 0) | |
3275 | set_bit(BlockedBadBlocks, | |
3276 | &rdev->flags); | |
3277 | s->blocked_rdev = rdev; | |
3278 | atomic_inc(&rdev->nr_pending); | |
3279 | } | |
6bfe0b49 | 3280 | } |
415e72d0 N |
3281 | clear_bit(R5_Insync, &dev->flags); |
3282 | if (!rdev) | |
3283 | /* Not in-sync */; | |
31c176ec N |
3284 | else if (is_bad) { |
3285 | /* also not in-sync */ | |
18b9837e N |
3286 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
3287 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
3288 | /* treat as in-sync, but with a read error |
3289 | * which we can now try to correct | |
3290 | */ | |
3291 | set_bit(R5_Insync, &dev->flags); | |
3292 | set_bit(R5_ReadError, &dev->flags); | |
3293 | } | |
3294 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 3295 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 3296 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 3297 | /* in sync if before recovery_offset */ |
30d7a483 N |
3298 | set_bit(R5_Insync, &dev->flags); |
3299 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
3300 | test_bit(R5_Expanded, &dev->flags)) | |
3301 | /* If we've reshaped into here, we assume it is Insync. | |
3302 | * We will shortly update recovery_offset to make | |
3303 | * it official. | |
3304 | */ | |
3305 | set_bit(R5_Insync, &dev->flags); | |
3306 | ||
5d8c71f9 | 3307 | if (rdev && test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
3308 | /* This flag does not apply to '.replacement' |
3309 | * only to .rdev, so make sure to check that*/ | |
3310 | struct md_rdev *rdev2 = rcu_dereference( | |
3311 | conf->disks[i].rdev); | |
3312 | if (rdev2 == rdev) | |
3313 | clear_bit(R5_Insync, &dev->flags); | |
3314 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 3315 | s->handle_bad_blocks = 1; |
14a75d3e | 3316 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
3317 | } else |
3318 | clear_bit(R5_WriteError, &dev->flags); | |
3319 | } | |
5d8c71f9 | 3320 | if (rdev && test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
3321 | /* This flag does not apply to '.replacement' |
3322 | * only to .rdev, so make sure to check that*/ | |
3323 | struct md_rdev *rdev2 = rcu_dereference( | |
3324 | conf->disks[i].rdev); | |
3325 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 3326 | s->handle_bad_blocks = 1; |
14a75d3e | 3327 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
3328 | } else |
3329 | clear_bit(R5_MadeGood, &dev->flags); | |
3330 | } | |
977df362 N |
3331 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
3332 | struct md_rdev *rdev2 = rcu_dereference( | |
3333 | conf->disks[i].replacement); | |
3334 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
3335 | s->handle_bad_blocks = 1; | |
3336 | atomic_inc(&rdev2->nr_pending); | |
3337 | } else | |
3338 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
3339 | } | |
415e72d0 | 3340 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
3341 | /* The ReadError flag will just be confusing now */ |
3342 | clear_bit(R5_ReadError, &dev->flags); | |
3343 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3344 | } |
415e72d0 N |
3345 | if (test_bit(R5_ReadError, &dev->flags)) |
3346 | clear_bit(R5_Insync, &dev->flags); | |
3347 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3348 | if (s->failed < 2) |
3349 | s->failed_num[s->failed] = i; | |
3350 | s->failed++; | |
9a3e1101 N |
3351 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
3352 | do_recovery = 1; | |
415e72d0 | 3353 | } |
1da177e4 | 3354 | } |
9a3e1101 N |
3355 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
3356 | /* If there is a failed device being replaced, | |
3357 | * we must be recovering. | |
3358 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 3359 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
3360 | * else we can only be replacing |
3361 | * sync and recovery both need to read all devices, and so | |
3362 | * use the same flag. | |
3363 | */ | |
3364 | if (do_recovery || | |
c6d2e084 | 3365 | sh->sector >= conf->mddev->recovery_cp || |
3366 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
3367 | s->syncing = 1; |
3368 | else | |
3369 | s->replacing = 1; | |
3370 | } | |
1da177e4 | 3371 | rcu_read_unlock(); |
cc94015a N |
3372 | } |
3373 | ||
3374 | static void handle_stripe(struct stripe_head *sh) | |
3375 | { | |
3376 | struct stripe_head_state s; | |
d1688a6d | 3377 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 3378 | int i; |
84789554 N |
3379 | int prexor; |
3380 | int disks = sh->disks; | |
474af965 | 3381 | struct r5dev *pdev, *qdev; |
cc94015a N |
3382 | |
3383 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 3384 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
3385 | /* already being handled, ensure it gets handled |
3386 | * again when current action finishes */ | |
3387 | set_bit(STRIPE_HANDLE, &sh->state); | |
3388 | return; | |
3389 | } | |
3390 | ||
3391 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3392 | set_bit(STRIPE_SYNCING, &sh->state); | |
3393 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3394 | } | |
3395 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3396 | ||
3397 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3398 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3399 | (unsigned long long)sh->sector, sh->state, | |
3400 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3401 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 3402 | |
acfe726b | 3403 | analyse_stripe(sh, &s); |
c5a31000 | 3404 | |
bc2607f3 N |
3405 | if (s.handle_bad_blocks) { |
3406 | set_bit(STRIPE_HANDLE, &sh->state); | |
3407 | goto finish; | |
3408 | } | |
3409 | ||
474af965 N |
3410 | if (unlikely(s.blocked_rdev)) { |
3411 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 3412 | s.replacing || s.to_write || s.written) { |
474af965 N |
3413 | set_bit(STRIPE_HANDLE, &sh->state); |
3414 | goto finish; | |
3415 | } | |
3416 | /* There is nothing for the blocked_rdev to block */ | |
3417 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
3418 | s.blocked_rdev = NULL; | |
3419 | } | |
3420 | ||
3421 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
3422 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3423 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3424 | } | |
3425 | ||
3426 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
3427 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
3428 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
3429 | s.failed_num[0], s.failed_num[1]); | |
3430 | /* check if the array has lost more than max_degraded devices and, | |
3431 | * if so, some requests might need to be failed. | |
3432 | */ | |
9a3f530f N |
3433 | if (s.failed > conf->max_degraded) { |
3434 | sh->check_state = 0; | |
3435 | sh->reconstruct_state = 0; | |
3436 | if (s.to_read+s.to_write+s.written) | |
3437 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 3438 | if (s.syncing + s.replacing) |
9a3f530f N |
3439 | handle_failed_sync(conf, sh, &s); |
3440 | } | |
474af965 N |
3441 | |
3442 | /* | |
3443 | * might be able to return some write requests if the parity blocks | |
3444 | * are safe, or on a failed drive | |
3445 | */ | |
3446 | pdev = &sh->dev[sh->pd_idx]; | |
3447 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
3448 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
3449 | qdev = &sh->dev[sh->qd_idx]; | |
3450 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
3451 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
3452 | || conf->level < 6; | |
3453 | ||
3454 | if (s.written && | |
3455 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
3456 | && !test_bit(R5_LOCKED, &pdev->flags) | |
3457 | && test_bit(R5_UPTODATE, &pdev->flags)))) && | |
3458 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
3459 | && !test_bit(R5_LOCKED, &qdev->flags) | |
3460 | && test_bit(R5_UPTODATE, &qdev->flags))))) | |
3461 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
3462 | ||
3463 | /* Now we might consider reading some blocks, either to check/generate | |
3464 | * parity, or to satisfy requests | |
3465 | * or to load a block that is being partially written. | |
3466 | */ | |
3467 | if (s.to_read || s.non_overwrite | |
3468 | || (conf->level == 6 && s.to_write && s.failed) | |
9a3e1101 N |
3469 | || (s.syncing && (s.uptodate + s.compute < disks)) |
3470 | || s.replacing | |
3471 | || s.expanding) | |
474af965 N |
3472 | handle_stripe_fill(sh, &s, disks); |
3473 | ||
84789554 N |
3474 | /* Now we check to see if any write operations have recently |
3475 | * completed | |
3476 | */ | |
3477 | prexor = 0; | |
3478 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
3479 | prexor = 1; | |
3480 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
3481 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
3482 | sh->reconstruct_state = reconstruct_state_idle; | |
3483 | ||
3484 | /* All the 'written' buffers and the parity block are ready to | |
3485 | * be written back to disk | |
3486 | */ | |
3487 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3488 | BUG_ON(sh->qd_idx >= 0 && | |
3489 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags)); | |
3490 | for (i = disks; i--; ) { | |
3491 | struct r5dev *dev = &sh->dev[i]; | |
3492 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3493 | (i == sh->pd_idx || i == sh->qd_idx || | |
3494 | dev->written)) { | |
3495 | pr_debug("Writing block %d\n", i); | |
3496 | set_bit(R5_Wantwrite, &dev->flags); | |
3497 | if (prexor) | |
3498 | continue; | |
3499 | if (!test_bit(R5_Insync, &dev->flags) || | |
3500 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
3501 | s.failed == 0)) | |
3502 | set_bit(STRIPE_INSYNC, &sh->state); | |
3503 | } | |
3504 | } | |
3505 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3506 | s.dec_preread_active = 1; | |
3507 | } | |
3508 | ||
3509 | /* Now to consider new write requests and what else, if anything | |
3510 | * should be read. We do not handle new writes when: | |
3511 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3512 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3513 | * block. | |
3514 | */ | |
3515 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
3516 | handle_stripe_dirtying(conf, sh, &s, disks); | |
3517 | ||
3518 | /* maybe we need to check and possibly fix the parity for this stripe | |
3519 | * Any reads will already have been scheduled, so we just see if enough | |
3520 | * data is available. The parity check is held off while parity | |
3521 | * dependent operations are in flight. | |
3522 | */ | |
3523 | if (sh->check_state || | |
3524 | (s.syncing && s.locked == 0 && | |
3525 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3526 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
3527 | if (conf->level == 6) | |
3528 | handle_parity_checks6(conf, sh, &s, disks); | |
3529 | else | |
3530 | handle_parity_checks5(conf, sh, &s, disks); | |
3531 | } | |
c5a31000 | 3532 | |
9a3e1101 N |
3533 | if (s.replacing && s.locked == 0 |
3534 | && !test_bit(STRIPE_INSYNC, &sh->state)) { | |
3535 | /* Write out to replacement devices where possible */ | |
3536 | for (i = 0; i < conf->raid_disks; i++) | |
3537 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags) && | |
3538 | test_bit(R5_NeedReplace, &sh->dev[i].flags)) { | |
3539 | set_bit(R5_WantReplace, &sh->dev[i].flags); | |
3540 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3541 | s.locked++; | |
3542 | } | |
3543 | set_bit(STRIPE_INSYNC, &sh->state); | |
3544 | } | |
3545 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
3546 | test_bit(STRIPE_INSYNC, &sh->state)) { | |
c5a31000 N |
3547 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
3548 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3549 | } | |
3550 | ||
3551 | /* If the failed drives are just a ReadError, then we might need | |
3552 | * to progress the repair/check process | |
3553 | */ | |
3554 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
3555 | for (i = 0; i < s.failed; i++) { | |
3556 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
3557 | if (test_bit(R5_ReadError, &dev->flags) | |
3558 | && !test_bit(R5_LOCKED, &dev->flags) | |
3559 | && test_bit(R5_UPTODATE, &dev->flags) | |
3560 | ) { | |
3561 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3562 | set_bit(R5_Wantwrite, &dev->flags); | |
3563 | set_bit(R5_ReWrite, &dev->flags); | |
3564 | set_bit(R5_LOCKED, &dev->flags); | |
3565 | s.locked++; | |
3566 | } else { | |
3567 | /* let's read it back */ | |
3568 | set_bit(R5_Wantread, &dev->flags); | |
3569 | set_bit(R5_LOCKED, &dev->flags); | |
3570 | s.locked++; | |
3571 | } | |
3572 | } | |
3573 | } | |
3574 | ||
3575 | ||
3687c061 N |
3576 | /* Finish reconstruct operations initiated by the expansion process */ |
3577 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3578 | struct stripe_head *sh_src | |
3579 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
3580 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
3581 | /* sh cannot be written until sh_src has been read. | |
3582 | * so arrange for sh to be delayed a little | |
3583 | */ | |
3584 | set_bit(STRIPE_DELAYED, &sh->state); | |
3585 | set_bit(STRIPE_HANDLE, &sh->state); | |
3586 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3587 | &sh_src->state)) | |
3588 | atomic_inc(&conf->preread_active_stripes); | |
3589 | release_stripe(sh_src); | |
3590 | goto finish; | |
3591 | } | |
3592 | if (sh_src) | |
3593 | release_stripe(sh_src); | |
3594 | ||
3595 | sh->reconstruct_state = reconstruct_state_idle; | |
3596 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3597 | for (i = conf->raid_disks; i--; ) { | |
3598 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3599 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3600 | s.locked++; | |
3601 | } | |
3602 | } | |
f416885e | 3603 | |
3687c061 N |
3604 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
3605 | !sh->reconstruct_state) { | |
3606 | /* Need to write out all blocks after computing parity */ | |
3607 | sh->disks = conf->raid_disks; | |
3608 | stripe_set_idx(sh->sector, conf, 0, sh); | |
3609 | schedule_reconstruction(sh, &s, 1, 1); | |
3610 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
3611 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
3612 | atomic_dec(&conf->reshape_stripes); | |
3613 | wake_up(&conf->wait_for_overlap); | |
3614 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3615 | } | |
3616 | ||
3617 | if (s.expanding && s.locked == 0 && | |
3618 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
3619 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 3620 | |
3687c061 | 3621 | finish: |
6bfe0b49 | 3622 | /* wait for this device to become unblocked */ |
5f066c63 N |
3623 | if (unlikely(s.blocked_rdev)) { |
3624 | if (conf->mddev->external) | |
3625 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
3626 | conf->mddev); | |
3627 | else | |
3628 | /* Internal metadata will immediately | |
3629 | * be written by raid5d, so we don't | |
3630 | * need to wait here. | |
3631 | */ | |
3632 | rdev_dec_pending(s.blocked_rdev, | |
3633 | conf->mddev); | |
3634 | } | |
6bfe0b49 | 3635 | |
bc2607f3 N |
3636 | if (s.handle_bad_blocks) |
3637 | for (i = disks; i--; ) { | |
3cb03002 | 3638 | struct md_rdev *rdev; |
bc2607f3 N |
3639 | struct r5dev *dev = &sh->dev[i]; |
3640 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
3641 | /* We own a safe reference to the rdev */ | |
3642 | rdev = conf->disks[i].rdev; | |
3643 | if (!rdev_set_badblocks(rdev, sh->sector, | |
3644 | STRIPE_SECTORS, 0)) | |
3645 | md_error(conf->mddev, rdev); | |
3646 | rdev_dec_pending(rdev, conf->mddev); | |
3647 | } | |
b84db560 N |
3648 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
3649 | rdev = conf->disks[i].rdev; | |
3650 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 3651 | STRIPE_SECTORS, 0); |
b84db560 N |
3652 | rdev_dec_pending(rdev, conf->mddev); |
3653 | } | |
977df362 N |
3654 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
3655 | rdev = conf->disks[i].replacement; | |
dd054fce N |
3656 | if (!rdev) |
3657 | /* rdev have been moved down */ | |
3658 | rdev = conf->disks[i].rdev; | |
977df362 | 3659 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 3660 | STRIPE_SECTORS, 0); |
977df362 N |
3661 | rdev_dec_pending(rdev, conf->mddev); |
3662 | } | |
bc2607f3 N |
3663 | } |
3664 | ||
6c0069c0 YT |
3665 | if (s.ops_request) |
3666 | raid_run_ops(sh, s.ops_request); | |
3667 | ||
f0e43bcd | 3668 | ops_run_io(sh, &s); |
16a53ecc | 3669 | |
c5709ef6 | 3670 | if (s.dec_preread_active) { |
729a1866 | 3671 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 3672 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3673 | * have actually been submitted. |
3674 | */ | |
3675 | atomic_dec(&conf->preread_active_stripes); | |
3676 | if (atomic_read(&conf->preread_active_stripes) < | |
3677 | IO_THRESHOLD) | |
3678 | md_wakeup_thread(conf->mddev->thread); | |
3679 | } | |
3680 | ||
c5709ef6 | 3681 | return_io(s.return_bi); |
16a53ecc | 3682 | |
257a4b42 | 3683 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
3684 | } |
3685 | ||
d1688a6d | 3686 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
3687 | { |
3688 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3689 | while (!list_empty(&conf->delayed_list)) { | |
3690 | struct list_head *l = conf->delayed_list.next; | |
3691 | struct stripe_head *sh; | |
3692 | sh = list_entry(l, struct stripe_head, lru); | |
3693 | list_del_init(l); | |
3694 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3695 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3696 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3697 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3698 | } |
482c0834 | 3699 | } |
16a53ecc N |
3700 | } |
3701 | ||
d1688a6d | 3702 | static void activate_bit_delay(struct r5conf *conf) |
16a53ecc N |
3703 | { |
3704 | /* device_lock is held */ | |
3705 | struct list_head head; | |
3706 | list_add(&head, &conf->bitmap_list); | |
3707 | list_del_init(&conf->bitmap_list); | |
3708 | while (!list_empty(&head)) { | |
3709 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3710 | list_del_init(&sh->lru); | |
3711 | atomic_inc(&sh->count); | |
3712 | __release_stripe(conf, sh); | |
3713 | } | |
3714 | } | |
3715 | ||
fd01b88c | 3716 | int md_raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 3717 | { |
d1688a6d | 3718 | struct r5conf *conf = mddev->private; |
f022b2fd N |
3719 | |
3720 | /* No difference between reads and writes. Just check | |
3721 | * how busy the stripe_cache is | |
3722 | */ | |
3fa841d7 | 3723 | |
f022b2fd N |
3724 | if (conf->inactive_blocked) |
3725 | return 1; | |
3726 | if (conf->quiesce) | |
3727 | return 1; | |
3728 | if (list_empty_careful(&conf->inactive_list)) | |
3729 | return 1; | |
3730 | ||
3731 | return 0; | |
3732 | } | |
11d8a6e3 N |
3733 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3734 | ||
3735 | static int raid5_congested(void *data, int bits) | |
3736 | { | |
fd01b88c | 3737 | struct mddev *mddev = data; |
11d8a6e3 N |
3738 | |
3739 | return mddev_congested(mddev, bits) || | |
3740 | md_raid5_congested(mddev, bits); | |
3741 | } | |
f022b2fd | 3742 | |
23032a0e RBJ |
3743 | /* We want read requests to align with chunks where possible, |
3744 | * but write requests don't need to. | |
3745 | */ | |
cc371e66 AK |
3746 | static int raid5_mergeable_bvec(struct request_queue *q, |
3747 | struct bvec_merge_data *bvm, | |
3748 | struct bio_vec *biovec) | |
23032a0e | 3749 | { |
fd01b88c | 3750 | struct mddev *mddev = q->queuedata; |
cc371e66 | 3751 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3752 | int max; |
9d8f0363 | 3753 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3754 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3755 | |
cc371e66 | 3756 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3757 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3758 | ||
664e7c41 AN |
3759 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3760 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3761 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3762 | if (max < 0) max = 0; | |
3763 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3764 | return biovec->bv_len; | |
3765 | else | |
3766 | return max; | |
3767 | } | |
3768 | ||
f679623f | 3769 | |
fd01b88c | 3770 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f RBJ |
3771 | { |
3772 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3773 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3774 | unsigned int bio_sectors = bio->bi_size >> 9; |
3775 | ||
664e7c41 AN |
3776 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3777 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3778 | return chunk_sectors >= |
3779 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3780 | } | |
3781 | ||
46031f9a RBJ |
3782 | /* |
3783 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3784 | * later sampled by raid5d. | |
3785 | */ | |
d1688a6d | 3786 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
3787 | { |
3788 | unsigned long flags; | |
3789 | ||
3790 | spin_lock_irqsave(&conf->device_lock, flags); | |
3791 | ||
3792 | bi->bi_next = conf->retry_read_aligned_list; | |
3793 | conf->retry_read_aligned_list = bi; | |
3794 | ||
3795 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3796 | md_wakeup_thread(conf->mddev->thread); | |
3797 | } | |
3798 | ||
3799 | ||
d1688a6d | 3800 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
3801 | { |
3802 | struct bio *bi; | |
3803 | ||
3804 | bi = conf->retry_read_aligned; | |
3805 | if (bi) { | |
3806 | conf->retry_read_aligned = NULL; | |
3807 | return bi; | |
3808 | } | |
3809 | bi = conf->retry_read_aligned_list; | |
3810 | if(bi) { | |
387bb173 | 3811 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3812 | bi->bi_next = NULL; |
960e739d JA |
3813 | /* |
3814 | * this sets the active strip count to 1 and the processed | |
3815 | * strip count to zero (upper 8 bits) | |
3816 | */ | |
e7836bd6 | 3817 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
3818 | } |
3819 | ||
3820 | return bi; | |
3821 | } | |
3822 | ||
3823 | ||
f679623f RBJ |
3824 | /* |
3825 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3826 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3827 | * first). | |
3828 | * If the read failed.. | |
3829 | */ | |
6712ecf8 | 3830 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3831 | { |
3832 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 3833 | struct mddev *mddev; |
d1688a6d | 3834 | struct r5conf *conf; |
46031f9a | 3835 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 3836 | struct md_rdev *rdev; |
46031f9a | 3837 | |
f679623f | 3838 | bio_put(bi); |
46031f9a | 3839 | |
46031f9a RBJ |
3840 | rdev = (void*)raid_bi->bi_next; |
3841 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3842 | mddev = rdev->mddev; |
3843 | conf = mddev->private; | |
46031f9a RBJ |
3844 | |
3845 | rdev_dec_pending(rdev, conf->mddev); | |
3846 | ||
3847 | if (!error && uptodate) { | |
6712ecf8 | 3848 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3849 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3850 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3851 | return; |
46031f9a RBJ |
3852 | } |
3853 | ||
3854 | ||
45b4233c | 3855 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3856 | |
3857 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3858 | } |
3859 | ||
387bb173 NB |
3860 | static int bio_fits_rdev(struct bio *bi) |
3861 | { | |
165125e1 | 3862 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3863 | |
ae03bf63 | 3864 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3865 | return 0; |
3866 | blk_recount_segments(q, bi); | |
8a78362c | 3867 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3868 | return 0; |
3869 | ||
3870 | if (q->merge_bvec_fn) | |
3871 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3872 | * just just give up | |
3873 | */ | |
3874 | return 0; | |
3875 | ||
3876 | return 1; | |
3877 | } | |
3878 | ||
3879 | ||
fd01b88c | 3880 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 3881 | { |
d1688a6d | 3882 | struct r5conf *conf = mddev->private; |
8553fe7e | 3883 | int dd_idx; |
f679623f | 3884 | struct bio* align_bi; |
3cb03002 | 3885 | struct md_rdev *rdev; |
671488cc | 3886 | sector_t end_sector; |
f679623f RBJ |
3887 | |
3888 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3889 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3890 | return 0; |
3891 | } | |
3892 | /* | |
a167f663 | 3893 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3894 | */ |
a167f663 | 3895 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3896 | if (!align_bi) |
3897 | return 0; | |
3898 | /* | |
3899 | * set bi_end_io to a new function, and set bi_private to the | |
3900 | * original bio. | |
3901 | */ | |
3902 | align_bi->bi_end_io = raid5_align_endio; | |
3903 | align_bi->bi_private = raid_bio; | |
3904 | /* | |
3905 | * compute position | |
3906 | */ | |
112bf897 N |
3907 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3908 | 0, | |
911d4ee8 | 3909 | &dd_idx, NULL); |
f679623f | 3910 | |
671488cc | 3911 | end_sector = align_bi->bi_sector + (align_bi->bi_size >> 9); |
f679623f | 3912 | rcu_read_lock(); |
671488cc N |
3913 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
3914 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
3915 | rdev->recovery_offset < end_sector) { | |
3916 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3917 | if (rdev && | |
3918 | (test_bit(Faulty, &rdev->flags) || | |
3919 | !(test_bit(In_sync, &rdev->flags) || | |
3920 | rdev->recovery_offset >= end_sector))) | |
3921 | rdev = NULL; | |
3922 | } | |
3923 | if (rdev) { | |
31c176ec N |
3924 | sector_t first_bad; |
3925 | int bad_sectors; | |
3926 | ||
f679623f RBJ |
3927 | atomic_inc(&rdev->nr_pending); |
3928 | rcu_read_unlock(); | |
46031f9a RBJ |
3929 | raid_bio->bi_next = (void*)rdev; |
3930 | align_bi->bi_bdev = rdev->bdev; | |
3931 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
46031f9a | 3932 | |
31c176ec N |
3933 | if (!bio_fits_rdev(align_bi) || |
3934 | is_badblock(rdev, align_bi->bi_sector, align_bi->bi_size>>9, | |
3935 | &first_bad, &bad_sectors)) { | |
3936 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
3937 | bio_put(align_bi); |
3938 | rdev_dec_pending(rdev, mddev); | |
3939 | return 0; | |
3940 | } | |
3941 | ||
6c0544e2 | 3942 | /* No reshape active, so we can trust rdev->data_offset */ |
3943 | align_bi->bi_sector += rdev->data_offset; | |
3944 | ||
46031f9a RBJ |
3945 | spin_lock_irq(&conf->device_lock); |
3946 | wait_event_lock_irq(conf->wait_for_stripe, | |
3947 | conf->quiesce == 0, | |
3948 | conf->device_lock, /* nothing */); | |
3949 | atomic_inc(&conf->active_aligned_reads); | |
3950 | spin_unlock_irq(&conf->device_lock); | |
3951 | ||
f679623f RBJ |
3952 | generic_make_request(align_bi); |
3953 | return 1; | |
3954 | } else { | |
3955 | rcu_read_unlock(); | |
46031f9a | 3956 | bio_put(align_bi); |
f679623f RBJ |
3957 | return 0; |
3958 | } | |
3959 | } | |
3960 | ||
8b3e6cdc DW |
3961 | /* __get_priority_stripe - get the next stripe to process |
3962 | * | |
3963 | * Full stripe writes are allowed to pass preread active stripes up until | |
3964 | * the bypass_threshold is exceeded. In general the bypass_count | |
3965 | * increments when the handle_list is handled before the hold_list; however, it | |
3966 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3967 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3968 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3969 | * handle_list. | |
3970 | */ | |
d1688a6d | 3971 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf) |
8b3e6cdc DW |
3972 | { |
3973 | struct stripe_head *sh; | |
3974 | ||
3975 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3976 | __func__, | |
3977 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3978 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3979 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3980 | ||
3981 | if (!list_empty(&conf->handle_list)) { | |
3982 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3983 | ||
3984 | if (list_empty(&conf->hold_list)) | |
3985 | conf->bypass_count = 0; | |
3986 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3987 | if (conf->hold_list.next == conf->last_hold) | |
3988 | conf->bypass_count++; | |
3989 | else { | |
3990 | conf->last_hold = conf->hold_list.next; | |
3991 | conf->bypass_count -= conf->bypass_threshold; | |
3992 | if (conf->bypass_count < 0) | |
3993 | conf->bypass_count = 0; | |
3994 | } | |
3995 | } | |
3996 | } else if (!list_empty(&conf->hold_list) && | |
3997 | ((conf->bypass_threshold && | |
3998 | conf->bypass_count > conf->bypass_threshold) || | |
3999 | atomic_read(&conf->pending_full_writes) == 0)) { | |
4000 | sh = list_entry(conf->hold_list.next, | |
4001 | typeof(*sh), lru); | |
4002 | conf->bypass_count -= conf->bypass_threshold; | |
4003 | if (conf->bypass_count < 0) | |
4004 | conf->bypass_count = 0; | |
4005 | } else | |
4006 | return NULL; | |
4007 | ||
4008 | list_del_init(&sh->lru); | |
4009 | atomic_inc(&sh->count); | |
4010 | BUG_ON(atomic_read(&sh->count) != 1); | |
4011 | return sh; | |
4012 | } | |
f679623f | 4013 | |
8811b596 SL |
4014 | struct raid5_plug_cb { |
4015 | struct blk_plug_cb cb; | |
4016 | struct list_head list; | |
4017 | }; | |
4018 | ||
4019 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
4020 | { | |
4021 | struct raid5_plug_cb *cb = container_of( | |
4022 | blk_cb, struct raid5_plug_cb, cb); | |
4023 | struct stripe_head *sh; | |
4024 | struct mddev *mddev = cb->cb.data; | |
4025 | struct r5conf *conf = mddev->private; | |
4026 | ||
4027 | if (cb->list.next && !list_empty(&cb->list)) { | |
4028 | spin_lock_irq(&conf->device_lock); | |
4029 | while (!list_empty(&cb->list)) { | |
4030 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
4031 | list_del_init(&sh->lru); | |
4032 | /* | |
4033 | * avoid race release_stripe_plug() sees | |
4034 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
4035 | * is still in our list | |
4036 | */ | |
4037 | smp_mb__before_clear_bit(); | |
4038 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); | |
4039 | __release_stripe(conf, sh); | |
4040 | } | |
4041 | spin_unlock_irq(&conf->device_lock); | |
4042 | } | |
4043 | kfree(cb); | |
4044 | } | |
4045 | ||
4046 | static void release_stripe_plug(struct mddev *mddev, | |
4047 | struct stripe_head *sh) | |
4048 | { | |
4049 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
4050 | raid5_unplug, mddev, | |
4051 | sizeof(struct raid5_plug_cb)); | |
4052 | struct raid5_plug_cb *cb; | |
4053 | ||
4054 | if (!blk_cb) { | |
4055 | release_stripe(sh); | |
4056 | return; | |
4057 | } | |
4058 | ||
4059 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
4060 | ||
4061 | if (cb->list.next == NULL) | |
4062 | INIT_LIST_HEAD(&cb->list); | |
4063 | ||
4064 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
4065 | list_add_tail(&sh->lru, &cb->list); | |
4066 | else | |
4067 | release_stripe(sh); | |
4068 | } | |
4069 | ||
b4fdcb02 | 4070 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 4071 | { |
d1688a6d | 4072 | struct r5conf *conf = mddev->private; |
911d4ee8 | 4073 | int dd_idx; |
1da177e4 LT |
4074 | sector_t new_sector; |
4075 | sector_t logical_sector, last_sector; | |
4076 | struct stripe_head *sh; | |
a362357b | 4077 | const int rw = bio_data_dir(bi); |
49077326 | 4078 | int remaining; |
1da177e4 | 4079 | |
e9c7469b TH |
4080 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
4081 | md_flush_request(mddev, bi); | |
5a7bbad2 | 4082 | return; |
e5dcdd80 N |
4083 | } |
4084 | ||
3d310eb7 | 4085 | md_write_start(mddev, bi); |
06d91a5f | 4086 | |
802ba064 | 4087 | if (rw == READ && |
52488615 | 4088 | mddev->reshape_position == MaxSector && |
21a52c6d | 4089 | chunk_aligned_read(mddev,bi)) |
5a7bbad2 | 4090 | return; |
52488615 | 4091 | |
1da177e4 LT |
4092 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
4093 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
4094 | bi->bi_next = NULL; | |
4095 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 4096 | |
1da177e4 LT |
4097 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
4098 | DEFINE_WAIT(w); | |
b5663ba4 | 4099 | int previous; |
b578d55f | 4100 | |
7ecaa1e6 | 4101 | retry: |
b5663ba4 | 4102 | previous = 0; |
b578d55f | 4103 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 4104 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 4105 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
4106 | * 64bit on a 32bit platform, and so it might be |
4107 | * possible to see a half-updated value | |
aeb878b0 | 4108 | * Of course reshape_progress could change after |
df8e7f76 N |
4109 | * the lock is dropped, so once we get a reference |
4110 | * to the stripe that we think it is, we will have | |
4111 | * to check again. | |
4112 | */ | |
7ecaa1e6 | 4113 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 4114 | if (mddev->reshape_backwards |
fef9c61f N |
4115 | ? logical_sector < conf->reshape_progress |
4116 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
4117 | previous = 1; |
4118 | } else { | |
2c810cdd | 4119 | if (mddev->reshape_backwards |
fef9c61f N |
4120 | ? logical_sector < conf->reshape_safe |
4121 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
4122 | spin_unlock_irq(&conf->device_lock); |
4123 | schedule(); | |
4124 | goto retry; | |
4125 | } | |
4126 | } | |
7ecaa1e6 N |
4127 | spin_unlock_irq(&conf->device_lock); |
4128 | } | |
16a53ecc | 4129 | |
112bf897 N |
4130 | new_sector = raid5_compute_sector(conf, logical_sector, |
4131 | previous, | |
911d4ee8 | 4132 | &dd_idx, NULL); |
0c55e022 | 4133 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
4134 | (unsigned long long)new_sector, |
4135 | (unsigned long long)logical_sector); | |
4136 | ||
b5663ba4 | 4137 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 4138 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 4139 | if (sh) { |
b0f9ec04 | 4140 | if (unlikely(previous)) { |
7ecaa1e6 | 4141 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
4142 | * stripe, so we must do the range check again. |
4143 | * Expansion could still move past after this | |
4144 | * test, but as we are holding a reference to | |
4145 | * 'sh', we know that if that happens, | |
4146 | * STRIPE_EXPANDING will get set and the expansion | |
4147 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
4148 | */ |
4149 | int must_retry = 0; | |
4150 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 4151 | if (mddev->reshape_backwards |
b0f9ec04 N |
4152 | ? logical_sector >= conf->reshape_progress |
4153 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
4154 | /* mismatch, need to try again */ |
4155 | must_retry = 1; | |
4156 | spin_unlock_irq(&conf->device_lock); | |
4157 | if (must_retry) { | |
4158 | release_stripe(sh); | |
7a3ab908 | 4159 | schedule(); |
7ecaa1e6 N |
4160 | goto retry; |
4161 | } | |
4162 | } | |
e62e58a5 | 4163 | |
ffd96e35 | 4164 | if (rw == WRITE && |
a5c308d4 | 4165 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
4166 | logical_sector < mddev->suspend_hi) { |
4167 | release_stripe(sh); | |
e62e58a5 N |
4168 | /* As the suspend_* range is controlled by |
4169 | * userspace, we want an interruptible | |
4170 | * wait. | |
4171 | */ | |
4172 | flush_signals(current); | |
4173 | prepare_to_wait(&conf->wait_for_overlap, | |
4174 | &w, TASK_INTERRUPTIBLE); | |
4175 | if (logical_sector >= mddev->suspend_lo && | |
4176 | logical_sector < mddev->suspend_hi) | |
4177 | schedule(); | |
e464eafd N |
4178 | goto retry; |
4179 | } | |
7ecaa1e6 N |
4180 | |
4181 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 4182 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
4183 | /* Stripe is busy expanding or |
4184 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4185 | * and wait a while |
4186 | */ | |
482c0834 | 4187 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4188 | release_stripe(sh); |
4189 | schedule(); | |
4190 | goto retry; | |
4191 | } | |
4192 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
4193 | set_bit(STRIPE_HANDLE, &sh->state); |
4194 | clear_bit(STRIPE_DELAYED, &sh->state); | |
895e3c5c | 4195 | if ((bi->bi_rw & REQ_NOIDLE) && |
729a1866 N |
4196 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4197 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 4198 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
4199 | } else { |
4200 | /* cannot get stripe for read-ahead, just give-up */ | |
4201 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4202 | finish_wait(&conf->wait_for_overlap, &w); | |
4203 | break; | |
4204 | } | |
1da177e4 | 4205 | } |
7c13edc8 | 4206 | |
e7836bd6 | 4207 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 4208 | if (remaining == 0) { |
1da177e4 | 4209 | |
16a53ecc | 4210 | if ( rw == WRITE ) |
1da177e4 | 4211 | md_write_end(mddev); |
6712ecf8 | 4212 | |
0e13fe23 | 4213 | bio_endio(bi, 0); |
1da177e4 | 4214 | } |
1da177e4 LT |
4215 | } |
4216 | ||
fd01b88c | 4217 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 4218 | |
fd01b88c | 4219 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4220 | { |
52c03291 N |
4221 | /* reshaping is quite different to recovery/resync so it is |
4222 | * handled quite separately ... here. | |
4223 | * | |
4224 | * On each call to sync_request, we gather one chunk worth of | |
4225 | * destination stripes and flag them as expanding. | |
4226 | * Then we find all the source stripes and request reads. | |
4227 | * As the reads complete, handle_stripe will copy the data | |
4228 | * into the destination stripe and release that stripe. | |
4229 | */ | |
d1688a6d | 4230 | struct r5conf *conf = mddev->private; |
1da177e4 | 4231 | struct stripe_head *sh; |
ccfcc3c1 | 4232 | sector_t first_sector, last_sector; |
f416885e N |
4233 | int raid_disks = conf->previous_raid_disks; |
4234 | int data_disks = raid_disks - conf->max_degraded; | |
4235 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4236 | int i; |
4237 | int dd_idx; | |
c8f517c4 | 4238 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4239 | sector_t stripe_addr; |
7a661381 | 4240 | int reshape_sectors; |
ab69ae12 | 4241 | struct list_head stripes; |
52c03291 | 4242 | |
fef9c61f N |
4243 | if (sector_nr == 0) { |
4244 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 4245 | if (mddev->reshape_backwards && |
fef9c61f N |
4246 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
4247 | sector_nr = raid5_size(mddev, 0, 0) | |
4248 | - conf->reshape_progress; | |
2c810cdd | 4249 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
4250 | conf->reshape_progress > 0) |
4251 | sector_nr = conf->reshape_progress; | |
f416885e | 4252 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4253 | if (sector_nr) { |
8dee7211 N |
4254 | mddev->curr_resync_completed = sector_nr; |
4255 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4256 | *skipped = 1; |
4257 | return sector_nr; | |
4258 | } | |
52c03291 N |
4259 | } |
4260 | ||
7a661381 N |
4261 | /* We need to process a full chunk at a time. |
4262 | * If old and new chunk sizes differ, we need to process the | |
4263 | * largest of these | |
4264 | */ | |
664e7c41 AN |
4265 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4266 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4267 | else |
9d8f0363 | 4268 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4269 | |
b5254dd5 N |
4270 | /* We update the metadata at least every 10 seconds, or when |
4271 | * the data about to be copied would over-write the source of | |
4272 | * the data at the front of the range. i.e. one new_stripe | |
4273 | * along from reshape_progress new_maps to after where | |
4274 | * reshape_safe old_maps to | |
52c03291 | 4275 | */ |
fef9c61f | 4276 | writepos = conf->reshape_progress; |
f416885e | 4277 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4278 | readpos = conf->reshape_progress; |
4279 | sector_div(readpos, data_disks); | |
fef9c61f | 4280 | safepos = conf->reshape_safe; |
f416885e | 4281 | sector_div(safepos, data_disks); |
2c810cdd | 4282 | if (mddev->reshape_backwards) { |
ed37d83e | 4283 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4284 | readpos += reshape_sectors; |
7a661381 | 4285 | safepos += reshape_sectors; |
fef9c61f | 4286 | } else { |
7a661381 | 4287 | writepos += reshape_sectors; |
ed37d83e N |
4288 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4289 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4290 | } |
52c03291 | 4291 | |
b5254dd5 N |
4292 | /* Having calculated the 'writepos' possibly use it |
4293 | * to set 'stripe_addr' which is where we will write to. | |
4294 | */ | |
4295 | if (mddev->reshape_backwards) { | |
4296 | BUG_ON(conf->reshape_progress == 0); | |
4297 | stripe_addr = writepos; | |
4298 | BUG_ON((mddev->dev_sectors & | |
4299 | ~((sector_t)reshape_sectors - 1)) | |
4300 | - reshape_sectors - stripe_addr | |
4301 | != sector_nr); | |
4302 | } else { | |
4303 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
4304 | stripe_addr = sector_nr; | |
4305 | } | |
4306 | ||
c8f517c4 N |
4307 | /* 'writepos' is the most advanced device address we might write. |
4308 | * 'readpos' is the least advanced device address we might read. | |
4309 | * 'safepos' is the least address recorded in the metadata as having | |
4310 | * been reshaped. | |
b5254dd5 N |
4311 | * If there is a min_offset_diff, these are adjusted either by |
4312 | * increasing the safepos/readpos if diff is negative, or | |
4313 | * increasing writepos if diff is positive. | |
4314 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
4315 | * ensure safety in the face of a crash - that must be done by userspace |
4316 | * making a backup of the data. So in that case there is no particular | |
4317 | * rush to update metadata. | |
4318 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4319 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4320 | * we can be safe in the event of a crash. | |
4321 | * So we insist on updating metadata if safepos is behind writepos and | |
4322 | * readpos is beyond writepos. | |
4323 | * In any case, update the metadata every 10 seconds. | |
4324 | * Maybe that number should be configurable, but I'm not sure it is | |
4325 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4326 | */ | |
b5254dd5 N |
4327 | if (conf->min_offset_diff < 0) { |
4328 | safepos += -conf->min_offset_diff; | |
4329 | readpos += -conf->min_offset_diff; | |
4330 | } else | |
4331 | writepos += conf->min_offset_diff; | |
4332 | ||
2c810cdd | 4333 | if ((mddev->reshape_backwards |
c8f517c4 N |
4334 | ? (safepos > writepos && readpos < writepos) |
4335 | : (safepos < writepos && readpos > writepos)) || | |
4336 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4337 | /* Cannot proceed until we've updated the superblock... */ |
4338 | wait_event(conf->wait_for_overlap, | |
4339 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4340 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4341 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4342 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4343 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4344 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4345 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4346 | kthread_should_stop()); |
4347 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4348 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4349 | spin_unlock_irq(&conf->device_lock); |
4350 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4351 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4352 | } |
4353 | ||
ab69ae12 | 4354 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4355 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4356 | int j; |
a9f326eb | 4357 | int skipped_disk = 0; |
a8c906ca | 4358 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4359 | set_bit(STRIPE_EXPANDING, &sh->state); |
4360 | atomic_inc(&conf->reshape_stripes); | |
4361 | /* If any of this stripe is beyond the end of the old | |
4362 | * array, then we need to zero those blocks | |
4363 | */ | |
4364 | for (j=sh->disks; j--;) { | |
4365 | sector_t s; | |
4366 | if (j == sh->pd_idx) | |
4367 | continue; | |
f416885e | 4368 | if (conf->level == 6 && |
d0dabf7e | 4369 | j == sh->qd_idx) |
f416885e | 4370 | continue; |
784052ec | 4371 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4372 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4373 | skipped_disk = 1; |
52c03291 N |
4374 | continue; |
4375 | } | |
4376 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4377 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4378 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4379 | } | |
a9f326eb | 4380 | if (!skipped_disk) { |
52c03291 N |
4381 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4382 | set_bit(STRIPE_HANDLE, &sh->state); | |
4383 | } | |
ab69ae12 | 4384 | list_add(&sh->lru, &stripes); |
52c03291 N |
4385 | } |
4386 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 4387 | if (mddev->reshape_backwards) |
7a661381 | 4388 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4389 | else |
7a661381 | 4390 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4391 | spin_unlock_irq(&conf->device_lock); |
4392 | /* Ok, those stripe are ready. We can start scheduling | |
4393 | * reads on the source stripes. | |
4394 | * The source stripes are determined by mapping the first and last | |
4395 | * block on the destination stripes. | |
4396 | */ | |
52c03291 | 4397 | first_sector = |
ec32a2bd | 4398 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4399 | 1, &dd_idx, NULL); |
52c03291 | 4400 | last_sector = |
0e6e0271 | 4401 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4402 | * new_data_disks - 1), |
911d4ee8 | 4403 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4404 | if (last_sector >= mddev->dev_sectors) |
4405 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4406 | while (first_sector <= last_sector) { |
a8c906ca | 4407 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4408 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4409 | set_bit(STRIPE_HANDLE, &sh->state); | |
4410 | release_stripe(sh); | |
4411 | first_sector += STRIPE_SECTORS; | |
4412 | } | |
ab69ae12 N |
4413 | /* Now that the sources are clearly marked, we can release |
4414 | * the destination stripes | |
4415 | */ | |
4416 | while (!list_empty(&stripes)) { | |
4417 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4418 | list_del_init(&sh->lru); | |
4419 | release_stripe(sh); | |
4420 | } | |
c6207277 N |
4421 | /* If this takes us to the resync_max point where we have to pause, |
4422 | * then we need to write out the superblock. | |
4423 | */ | |
7a661381 | 4424 | sector_nr += reshape_sectors; |
c03f6a19 N |
4425 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4426 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4427 | /* Cannot proceed until we've updated the superblock... */ |
4428 | wait_event(conf->wait_for_overlap, | |
4429 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4430 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4431 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4432 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4433 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4434 | md_wakeup_thread(mddev->thread); | |
4435 | wait_event(mddev->sb_wait, | |
4436 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4437 | || kthread_should_stop()); | |
4438 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4439 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4440 | spin_unlock_irq(&conf->device_lock); |
4441 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4442 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4443 | } |
7a661381 | 4444 | return reshape_sectors; |
52c03291 N |
4445 | } |
4446 | ||
4447 | /* FIXME go_faster isn't used */ | |
fd01b88c | 4448 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
52c03291 | 4449 | { |
d1688a6d | 4450 | struct r5conf *conf = mddev->private; |
52c03291 | 4451 | struct stripe_head *sh; |
58c0fed4 | 4452 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 4453 | sector_t sync_blocks; |
16a53ecc N |
4454 | int still_degraded = 0; |
4455 | int i; | |
1da177e4 | 4456 | |
72626685 | 4457 | if (sector_nr >= max_sector) { |
1da177e4 | 4458 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 4459 | |
29269553 N |
4460 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4461 | end_reshape(conf); | |
4462 | return 0; | |
4463 | } | |
72626685 N |
4464 | |
4465 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4466 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4467 | &sync_blocks, 1); | |
16a53ecc | 4468 | else /* completed sync */ |
72626685 N |
4469 | conf->fullsync = 0; |
4470 | bitmap_close_sync(mddev->bitmap); | |
4471 | ||
1da177e4 LT |
4472 | return 0; |
4473 | } | |
ccfcc3c1 | 4474 | |
64bd660b N |
4475 | /* Allow raid5_quiesce to complete */ |
4476 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4477 | ||
52c03291 N |
4478 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4479 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4480 | |
c6207277 N |
4481 | /* No need to check resync_max as we never do more than one |
4482 | * stripe, and as resync_max will always be on a chunk boundary, | |
4483 | * if the check in md_do_sync didn't fire, there is no chance | |
4484 | * of overstepping resync_max here | |
4485 | */ | |
4486 | ||
16a53ecc | 4487 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4488 | * to resync, then assert that we are finished, because there is |
4489 | * nothing we can do. | |
4490 | */ | |
3285edf1 | 4491 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4492 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4493 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4494 | *skipped = 1; |
1da177e4 LT |
4495 | return rv; |
4496 | } | |
72626685 | 4497 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4498 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4499 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4500 | /* we can skip this block, and probably more */ | |
4501 | sync_blocks /= STRIPE_SECTORS; | |
4502 | *skipped = 1; | |
4503 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4504 | } | |
1da177e4 | 4505 | |
b47490c9 N |
4506 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
4507 | ||
a8c906ca | 4508 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4509 | if (sh == NULL) { |
a8c906ca | 4510 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4511 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4512 | * is trying to get access |
1da177e4 | 4513 | */ |
66c006a5 | 4514 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4515 | } |
16a53ecc N |
4516 | /* Need to check if array will still be degraded after recovery/resync |
4517 | * We don't need to check the 'failed' flag as when that gets set, | |
4518 | * recovery aborts. | |
4519 | */ | |
f001a70c | 4520 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4521 | if (conf->disks[i].rdev == NULL) |
4522 | still_degraded = 1; | |
4523 | ||
4524 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4525 | ||
83206d66 | 4526 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4527 | |
1442577b | 4528 | handle_stripe(sh); |
1da177e4 LT |
4529 | release_stripe(sh); |
4530 | ||
4531 | return STRIPE_SECTORS; | |
4532 | } | |
4533 | ||
d1688a6d | 4534 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
4535 | { |
4536 | /* We may not be able to submit a whole bio at once as there | |
4537 | * may not be enough stripe_heads available. | |
4538 | * We cannot pre-allocate enough stripe_heads as we may need | |
4539 | * more than exist in the cache (if we allow ever large chunks). | |
4540 | * So we do one stripe head at a time and record in | |
4541 | * ->bi_hw_segments how many have been done. | |
4542 | * | |
4543 | * We *know* that this entire raid_bio is in one chunk, so | |
4544 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4545 | */ | |
4546 | struct stripe_head *sh; | |
911d4ee8 | 4547 | int dd_idx; |
46031f9a RBJ |
4548 | sector_t sector, logical_sector, last_sector; |
4549 | int scnt = 0; | |
4550 | int remaining; | |
4551 | int handled = 0; | |
4552 | ||
4553 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4554 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4555 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4556 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4557 | ||
4558 | for (; logical_sector < last_sector; | |
387bb173 NB |
4559 | logical_sector += STRIPE_SECTORS, |
4560 | sector += STRIPE_SECTORS, | |
4561 | scnt++) { | |
46031f9a | 4562 | |
e7836bd6 | 4563 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
4564 | /* already done this stripe */ |
4565 | continue; | |
4566 | ||
a8c906ca | 4567 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4568 | |
4569 | if (!sh) { | |
4570 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 4571 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
4572 | conf->retry_read_aligned = raid_bio; |
4573 | return handled; | |
4574 | } | |
4575 | ||
387bb173 NB |
4576 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4577 | release_stripe(sh); | |
e7836bd6 | 4578 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
4579 | conf->retry_read_aligned = raid_bio; |
4580 | return handled; | |
4581 | } | |
4582 | ||
3f9e7c14 | 4583 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 4584 | handle_stripe(sh); |
46031f9a RBJ |
4585 | release_stripe(sh); |
4586 | handled++; | |
4587 | } | |
e7836bd6 | 4588 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0e13fe23 NB |
4589 | if (remaining == 0) |
4590 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4591 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4592 | wake_up(&conf->wait_for_stripe); | |
4593 | return handled; | |
4594 | } | |
4595 | ||
46a06401 SL |
4596 | #define MAX_STRIPE_BATCH 8 |
4597 | static int handle_active_stripes(struct r5conf *conf) | |
4598 | { | |
4599 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
4600 | int i, batch_size = 0; | |
4601 | ||
4602 | while (batch_size < MAX_STRIPE_BATCH && | |
4603 | (sh = __get_priority_stripe(conf)) != NULL) | |
4604 | batch[batch_size++] = sh; | |
4605 | ||
4606 | if (batch_size == 0) | |
4607 | return batch_size; | |
4608 | spin_unlock_irq(&conf->device_lock); | |
4609 | ||
4610 | for (i = 0; i < batch_size; i++) | |
4611 | handle_stripe(batch[i]); | |
4612 | ||
4613 | cond_resched(); | |
4614 | ||
4615 | spin_lock_irq(&conf->device_lock); | |
4616 | for (i = 0; i < batch_size; i++) | |
4617 | __release_stripe(conf, batch[i]); | |
4618 | return batch_size; | |
4619 | } | |
46031f9a | 4620 | |
1da177e4 LT |
4621 | /* |
4622 | * This is our raid5 kernel thread. | |
4623 | * | |
4624 | * We scan the hash table for stripes which can be handled now. | |
4625 | * During the scan, completed stripes are saved for us by the interrupt | |
4626 | * handler, so that they will not have to wait for our next wakeup. | |
4627 | */ | |
4ed8731d | 4628 | static void raid5d(struct md_thread *thread) |
1da177e4 | 4629 | { |
4ed8731d | 4630 | struct mddev *mddev = thread->mddev; |
d1688a6d | 4631 | struct r5conf *conf = mddev->private; |
1da177e4 | 4632 | int handled; |
e1dfa0a2 | 4633 | struct blk_plug plug; |
1da177e4 | 4634 | |
45b4233c | 4635 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4636 | |
4637 | md_check_recovery(mddev); | |
1da177e4 | 4638 | |
e1dfa0a2 | 4639 | blk_start_plug(&plug); |
1da177e4 LT |
4640 | handled = 0; |
4641 | spin_lock_irq(&conf->device_lock); | |
4642 | while (1) { | |
46031f9a | 4643 | struct bio *bio; |
46a06401 | 4644 | int batch_size; |
1da177e4 | 4645 | |
0021b7bc | 4646 | if ( |
7c13edc8 N |
4647 | !list_empty(&conf->bitmap_list)) { |
4648 | /* Now is a good time to flush some bitmap updates */ | |
4649 | conf->seq_flush++; | |
700e432d | 4650 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4651 | bitmap_unplug(mddev->bitmap); |
700e432d | 4652 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4653 | conf->seq_write = conf->seq_flush; |
72626685 N |
4654 | activate_bit_delay(conf); |
4655 | } | |
0021b7bc | 4656 | raid5_activate_delayed(conf); |
72626685 | 4657 | |
46031f9a RBJ |
4658 | while ((bio = remove_bio_from_retry(conf))) { |
4659 | int ok; | |
4660 | spin_unlock_irq(&conf->device_lock); | |
4661 | ok = retry_aligned_read(conf, bio); | |
4662 | spin_lock_irq(&conf->device_lock); | |
4663 | if (!ok) | |
4664 | break; | |
4665 | handled++; | |
4666 | } | |
4667 | ||
46a06401 SL |
4668 | batch_size = handle_active_stripes(conf); |
4669 | if (!batch_size) | |
1da177e4 | 4670 | break; |
46a06401 | 4671 | handled += batch_size; |
1da177e4 | 4672 | |
46a06401 SL |
4673 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) { |
4674 | spin_unlock_irq(&conf->device_lock); | |
de393cde | 4675 | md_check_recovery(mddev); |
46a06401 SL |
4676 | spin_lock_irq(&conf->device_lock); |
4677 | } | |
1da177e4 | 4678 | } |
45b4233c | 4679 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4680 | |
4681 | spin_unlock_irq(&conf->device_lock); | |
4682 | ||
c9f21aaf | 4683 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4684 | blk_finish_plug(&plug); |
1da177e4 | 4685 | |
45b4233c | 4686 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4687 | } |
4688 | ||
3f294f4f | 4689 | static ssize_t |
fd01b88c | 4690 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 4691 | { |
d1688a6d | 4692 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4693 | if (conf) |
4694 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4695 | else | |
4696 | return 0; | |
3f294f4f N |
4697 | } |
4698 | ||
c41d4ac4 | 4699 | int |
fd01b88c | 4700 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 4701 | { |
d1688a6d | 4702 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
4703 | int err; |
4704 | ||
c41d4ac4 | 4705 | if (size <= 16 || size > 32768) |
3f294f4f | 4706 | return -EINVAL; |
c41d4ac4 | 4707 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4708 | if (drop_one_stripe(conf)) |
4709 | conf->max_nr_stripes--; | |
4710 | else | |
4711 | break; | |
4712 | } | |
b5470dc5 DW |
4713 | err = md_allow_write(mddev); |
4714 | if (err) | |
4715 | return err; | |
c41d4ac4 | 4716 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4717 | if (grow_one_stripe(conf)) |
4718 | conf->max_nr_stripes++; | |
4719 | else break; | |
4720 | } | |
c41d4ac4 N |
4721 | return 0; |
4722 | } | |
4723 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4724 | ||
4725 | static ssize_t | |
fd01b88c | 4726 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 4727 | { |
d1688a6d | 4728 | struct r5conf *conf = mddev->private; |
c41d4ac4 N |
4729 | unsigned long new; |
4730 | int err; | |
4731 | ||
4732 | if (len >= PAGE_SIZE) | |
4733 | return -EINVAL; | |
4734 | if (!conf) | |
4735 | return -ENODEV; | |
4736 | ||
4737 | if (strict_strtoul(page, 10, &new)) | |
4738 | return -EINVAL; | |
4739 | err = raid5_set_cache_size(mddev, new); | |
4740 | if (err) | |
4741 | return err; | |
3f294f4f N |
4742 | return len; |
4743 | } | |
007583c9 | 4744 | |
96de1e66 N |
4745 | static struct md_sysfs_entry |
4746 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4747 | raid5_show_stripe_cache_size, | |
4748 | raid5_store_stripe_cache_size); | |
3f294f4f | 4749 | |
8b3e6cdc | 4750 | static ssize_t |
fd01b88c | 4751 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 4752 | { |
d1688a6d | 4753 | struct r5conf *conf = mddev->private; |
8b3e6cdc DW |
4754 | if (conf) |
4755 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4756 | else | |
4757 | return 0; | |
4758 | } | |
4759 | ||
4760 | static ssize_t | |
fd01b88c | 4761 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 4762 | { |
d1688a6d | 4763 | struct r5conf *conf = mddev->private; |
4ef197d8 | 4764 | unsigned long new; |
8b3e6cdc DW |
4765 | if (len >= PAGE_SIZE) |
4766 | return -EINVAL; | |
4767 | if (!conf) | |
4768 | return -ENODEV; | |
4769 | ||
4ef197d8 | 4770 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4771 | return -EINVAL; |
4ef197d8 | 4772 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4773 | return -EINVAL; |
4774 | conf->bypass_threshold = new; | |
4775 | return len; | |
4776 | } | |
4777 | ||
4778 | static struct md_sysfs_entry | |
4779 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4780 | S_IRUGO | S_IWUSR, | |
4781 | raid5_show_preread_threshold, | |
4782 | raid5_store_preread_threshold); | |
4783 | ||
3f294f4f | 4784 | static ssize_t |
fd01b88c | 4785 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 4786 | { |
d1688a6d | 4787 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4788 | if (conf) |
4789 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4790 | else | |
4791 | return 0; | |
3f294f4f N |
4792 | } |
4793 | ||
96de1e66 N |
4794 | static struct md_sysfs_entry |
4795 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4796 | |
007583c9 | 4797 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4798 | &raid5_stripecache_size.attr, |
4799 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4800 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4801 | NULL, |
4802 | }; | |
007583c9 N |
4803 | static struct attribute_group raid5_attrs_group = { |
4804 | .name = NULL, | |
4805 | .attrs = raid5_attrs, | |
3f294f4f N |
4806 | }; |
4807 | ||
80c3a6ce | 4808 | static sector_t |
fd01b88c | 4809 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 4810 | { |
d1688a6d | 4811 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
4812 | |
4813 | if (!sectors) | |
4814 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4815 | if (!raid_disks) |
7ec05478 | 4816 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4817 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4818 | |
9d8f0363 | 4819 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4820 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4821 | return sectors * (raid_disks - conf->max_degraded); |
4822 | } | |
4823 | ||
d1688a6d | 4824 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 DW |
4825 | { |
4826 | struct raid5_percpu *percpu; | |
4827 | unsigned long cpu; | |
4828 | ||
4829 | if (!conf->percpu) | |
4830 | return; | |
4831 | ||
4832 | get_online_cpus(); | |
4833 | for_each_possible_cpu(cpu) { | |
4834 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4835 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4836 | kfree(percpu->scribble); |
36d1c647 DW |
4837 | } |
4838 | #ifdef CONFIG_HOTPLUG_CPU | |
4839 | unregister_cpu_notifier(&conf->cpu_notify); | |
4840 | #endif | |
4841 | put_online_cpus(); | |
4842 | ||
4843 | free_percpu(conf->percpu); | |
4844 | } | |
4845 | ||
d1688a6d | 4846 | static void free_conf(struct r5conf *conf) |
95fc17aa DW |
4847 | { |
4848 | shrink_stripes(conf); | |
36d1c647 | 4849 | raid5_free_percpu(conf); |
95fc17aa DW |
4850 | kfree(conf->disks); |
4851 | kfree(conf->stripe_hashtbl); | |
4852 | kfree(conf); | |
4853 | } | |
4854 | ||
36d1c647 DW |
4855 | #ifdef CONFIG_HOTPLUG_CPU |
4856 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4857 | void *hcpu) | |
4858 | { | |
d1688a6d | 4859 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
4860 | long cpu = (long)hcpu; |
4861 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4862 | ||
4863 | switch (action) { | |
4864 | case CPU_UP_PREPARE: | |
4865 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4866 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4867 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4868 | if (!percpu->scribble) |
4869 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4870 | ||
4871 | if (!percpu->scribble || | |
4872 | (conf->level == 6 && !percpu->spare_page)) { | |
4873 | safe_put_page(percpu->spare_page); | |
4874 | kfree(percpu->scribble); | |
36d1c647 DW |
4875 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4876 | __func__, cpu); | |
55af6bb5 | 4877 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4878 | } |
4879 | break; | |
4880 | case CPU_DEAD: | |
4881 | case CPU_DEAD_FROZEN: | |
4882 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4883 | kfree(percpu->scribble); |
36d1c647 | 4884 | percpu->spare_page = NULL; |
d6f38f31 | 4885 | percpu->scribble = NULL; |
36d1c647 DW |
4886 | break; |
4887 | default: | |
4888 | break; | |
4889 | } | |
4890 | return NOTIFY_OK; | |
4891 | } | |
4892 | #endif | |
4893 | ||
d1688a6d | 4894 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
4895 | { |
4896 | unsigned long cpu; | |
4897 | struct page *spare_page; | |
a29d8b8e | 4898 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4899 | void *scribble; |
36d1c647 DW |
4900 | int err; |
4901 | ||
36d1c647 DW |
4902 | allcpus = alloc_percpu(struct raid5_percpu); |
4903 | if (!allcpus) | |
4904 | return -ENOMEM; | |
4905 | conf->percpu = allcpus; | |
4906 | ||
4907 | get_online_cpus(); | |
4908 | err = 0; | |
4909 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4910 | if (conf->level == 6) { |
4911 | spare_page = alloc_page(GFP_KERNEL); | |
4912 | if (!spare_page) { | |
4913 | err = -ENOMEM; | |
4914 | break; | |
4915 | } | |
4916 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4917 | } | |
5e5e3e78 | 4918 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4919 | if (!scribble) { |
36d1c647 DW |
4920 | err = -ENOMEM; |
4921 | break; | |
4922 | } | |
d6f38f31 | 4923 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4924 | } |
4925 | #ifdef CONFIG_HOTPLUG_CPU | |
4926 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4927 | conf->cpu_notify.priority = 0; | |
4928 | if (err == 0) | |
4929 | err = register_cpu_notifier(&conf->cpu_notify); | |
4930 | #endif | |
4931 | put_online_cpus(); | |
4932 | ||
4933 | return err; | |
4934 | } | |
4935 | ||
d1688a6d | 4936 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 4937 | { |
d1688a6d | 4938 | struct r5conf *conf; |
5e5e3e78 | 4939 | int raid_disk, memory, max_disks; |
3cb03002 | 4940 | struct md_rdev *rdev; |
1da177e4 | 4941 | struct disk_info *disk; |
0232605d | 4942 | char pers_name[6]; |
1da177e4 | 4943 | |
91adb564 N |
4944 | if (mddev->new_level != 5 |
4945 | && mddev->new_level != 4 | |
4946 | && mddev->new_level != 6) { | |
0c55e022 | 4947 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4948 | mdname(mddev), mddev->new_level); |
4949 | return ERR_PTR(-EIO); | |
1da177e4 | 4950 | } |
91adb564 N |
4951 | if ((mddev->new_level == 5 |
4952 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4953 | (mddev->new_level == 6 | |
4954 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4955 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4956 | mdname(mddev), mddev->new_layout); |
4957 | return ERR_PTR(-EIO); | |
99c0fb5f | 4958 | } |
91adb564 | 4959 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4960 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4961 | mdname(mddev), mddev->raid_disks); |
4962 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4963 | } |
4964 | ||
664e7c41 AN |
4965 | if (!mddev->new_chunk_sectors || |
4966 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4967 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4968 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4969 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4970 | return ERR_PTR(-EINVAL); |
f6705578 N |
4971 | } |
4972 | ||
d1688a6d | 4973 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 4974 | if (conf == NULL) |
1da177e4 | 4975 | goto abort; |
f5efd45a DW |
4976 | spin_lock_init(&conf->device_lock); |
4977 | init_waitqueue_head(&conf->wait_for_stripe); | |
4978 | init_waitqueue_head(&conf->wait_for_overlap); | |
4979 | INIT_LIST_HEAD(&conf->handle_list); | |
4980 | INIT_LIST_HEAD(&conf->hold_list); | |
4981 | INIT_LIST_HEAD(&conf->delayed_list); | |
4982 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4983 | INIT_LIST_HEAD(&conf->inactive_list); | |
4984 | atomic_set(&conf->active_stripes, 0); | |
4985 | atomic_set(&conf->preread_active_stripes, 0); | |
4986 | atomic_set(&conf->active_aligned_reads, 0); | |
4987 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 4988 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
4989 | |
4990 | conf->raid_disks = mddev->raid_disks; | |
4991 | if (mddev->reshape_position == MaxSector) | |
4992 | conf->previous_raid_disks = mddev->raid_disks; | |
4993 | else | |
f6705578 | 4994 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4995 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4996 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4997 | |
5e5e3e78 | 4998 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4999 | GFP_KERNEL); |
5000 | if (!conf->disks) | |
5001 | goto abort; | |
9ffae0cf | 5002 | |
1da177e4 LT |
5003 | conf->mddev = mddev; |
5004 | ||
fccddba0 | 5005 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 5006 | goto abort; |
1da177e4 | 5007 | |
36d1c647 DW |
5008 | conf->level = mddev->new_level; |
5009 | if (raid5_alloc_percpu(conf) != 0) | |
5010 | goto abort; | |
5011 | ||
0c55e022 | 5012 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 5013 | |
dafb20fa | 5014 | rdev_for_each(rdev, mddev) { |
1da177e4 | 5015 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 5016 | if (raid_disk >= max_disks |
1da177e4 LT |
5017 | || raid_disk < 0) |
5018 | continue; | |
5019 | disk = conf->disks + raid_disk; | |
5020 | ||
17045f52 N |
5021 | if (test_bit(Replacement, &rdev->flags)) { |
5022 | if (disk->replacement) | |
5023 | goto abort; | |
5024 | disk->replacement = rdev; | |
5025 | } else { | |
5026 | if (disk->rdev) | |
5027 | goto abort; | |
5028 | disk->rdev = rdev; | |
5029 | } | |
1da177e4 | 5030 | |
b2d444d7 | 5031 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 5032 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
5033 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
5034 | " disk %d\n", | |
5035 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 5036 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
5037 | /* Cannot rely on bitmap to complete recovery */ |
5038 | conf->fullsync = 1; | |
1da177e4 LT |
5039 | } |
5040 | ||
09c9e5fa | 5041 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 5042 | conf->level = mddev->new_level; |
16a53ecc N |
5043 | if (conf->level == 6) |
5044 | conf->max_degraded = 2; | |
5045 | else | |
5046 | conf->max_degraded = 1; | |
91adb564 | 5047 | conf->algorithm = mddev->new_layout; |
1da177e4 | 5048 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 5049 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 5050 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 5051 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
5052 | conf->prev_algo = mddev->layout; |
5053 | } | |
1da177e4 | 5054 | |
91adb564 | 5055 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 5056 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
5057 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
5058 | printk(KERN_ERR | |
0c55e022 N |
5059 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
5060 | mdname(mddev), memory); | |
91adb564 N |
5061 | goto abort; |
5062 | } else | |
0c55e022 N |
5063 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
5064 | mdname(mddev), memory); | |
1da177e4 | 5065 | |
0232605d N |
5066 | sprintf(pers_name, "raid%d", mddev->new_level); |
5067 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 N |
5068 | if (!conf->thread) { |
5069 | printk(KERN_ERR | |
0c55e022 | 5070 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 5071 | mdname(mddev)); |
16a53ecc N |
5072 | goto abort; |
5073 | } | |
91adb564 N |
5074 | |
5075 | return conf; | |
5076 | ||
5077 | abort: | |
5078 | if (conf) { | |
95fc17aa | 5079 | free_conf(conf); |
91adb564 N |
5080 | return ERR_PTR(-EIO); |
5081 | } else | |
5082 | return ERR_PTR(-ENOMEM); | |
5083 | } | |
5084 | ||
c148ffdc N |
5085 | |
5086 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
5087 | { | |
5088 | switch (algo) { | |
5089 | case ALGORITHM_PARITY_0: | |
5090 | if (raid_disk < max_degraded) | |
5091 | return 1; | |
5092 | break; | |
5093 | case ALGORITHM_PARITY_N: | |
5094 | if (raid_disk >= raid_disks - max_degraded) | |
5095 | return 1; | |
5096 | break; | |
5097 | case ALGORITHM_PARITY_0_6: | |
5098 | if (raid_disk == 0 || | |
5099 | raid_disk == raid_disks - 1) | |
5100 | return 1; | |
5101 | break; | |
5102 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5103 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5104 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5105 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5106 | if (raid_disk == raid_disks - 1) | |
5107 | return 1; | |
5108 | } | |
5109 | return 0; | |
5110 | } | |
5111 | ||
fd01b88c | 5112 | static int run(struct mddev *mddev) |
91adb564 | 5113 | { |
d1688a6d | 5114 | struct r5conf *conf; |
9f7c2220 | 5115 | int working_disks = 0; |
c148ffdc | 5116 | int dirty_parity_disks = 0; |
3cb03002 | 5117 | struct md_rdev *rdev; |
c148ffdc | 5118 | sector_t reshape_offset = 0; |
17045f52 | 5119 | int i; |
b5254dd5 N |
5120 | long long min_offset_diff = 0; |
5121 | int first = 1; | |
91adb564 | 5122 | |
8c6ac868 | 5123 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 5124 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
5125 | " -- starting background reconstruction\n", |
5126 | mdname(mddev)); | |
b5254dd5 N |
5127 | |
5128 | rdev_for_each(rdev, mddev) { | |
5129 | long long diff; | |
5130 | if (rdev->raid_disk < 0) | |
5131 | continue; | |
5132 | diff = (rdev->new_data_offset - rdev->data_offset); | |
5133 | if (first) { | |
5134 | min_offset_diff = diff; | |
5135 | first = 0; | |
5136 | } else if (mddev->reshape_backwards && | |
5137 | diff < min_offset_diff) | |
5138 | min_offset_diff = diff; | |
5139 | else if (!mddev->reshape_backwards && | |
5140 | diff > min_offset_diff) | |
5141 | min_offset_diff = diff; | |
5142 | } | |
5143 | ||
91adb564 N |
5144 | if (mddev->reshape_position != MaxSector) { |
5145 | /* Check that we can continue the reshape. | |
b5254dd5 N |
5146 | * Difficulties arise if the stripe we would write to |
5147 | * next is at or after the stripe we would read from next. | |
5148 | * For a reshape that changes the number of devices, this | |
5149 | * is only possible for a very short time, and mdadm makes | |
5150 | * sure that time appears to have past before assembling | |
5151 | * the array. So we fail if that time hasn't passed. | |
5152 | * For a reshape that keeps the number of devices the same | |
5153 | * mdadm must be monitoring the reshape can keeping the | |
5154 | * critical areas read-only and backed up. It will start | |
5155 | * the array in read-only mode, so we check for that. | |
91adb564 N |
5156 | */ |
5157 | sector_t here_new, here_old; | |
5158 | int old_disks; | |
18b00334 | 5159 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 5160 | |
88ce4930 | 5161 | if (mddev->new_level != mddev->level) { |
0c55e022 | 5162 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
5163 | "required - aborting.\n", |
5164 | mdname(mddev)); | |
5165 | return -EINVAL; | |
5166 | } | |
91adb564 N |
5167 | old_disks = mddev->raid_disks - mddev->delta_disks; |
5168 | /* reshape_position must be on a new-stripe boundary, and one | |
5169 | * further up in new geometry must map after here in old | |
5170 | * geometry. | |
5171 | */ | |
5172 | here_new = mddev->reshape_position; | |
664e7c41 | 5173 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 5174 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
5175 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
5176 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
5177 | return -EINVAL; |
5178 | } | |
c148ffdc | 5179 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
5180 | /* here_new is the stripe we will write to */ |
5181 | here_old = mddev->reshape_position; | |
9d8f0363 | 5182 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
5183 | (old_disks-max_degraded)); |
5184 | /* here_old is the first stripe that we might need to read | |
5185 | * from */ | |
67ac6011 | 5186 | if (mddev->delta_disks == 0) { |
b5254dd5 N |
5187 | if ((here_new * mddev->new_chunk_sectors != |
5188 | here_old * mddev->chunk_sectors)) { | |
5189 | printk(KERN_ERR "md/raid:%s: reshape position is" | |
5190 | " confused - aborting\n", mdname(mddev)); | |
5191 | return -EINVAL; | |
5192 | } | |
67ac6011 | 5193 | /* We cannot be sure it is safe to start an in-place |
b5254dd5 | 5194 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
5195 | * and taking constant backups. |
5196 | * mdadm always starts a situation like this in | |
5197 | * readonly mode so it can take control before | |
5198 | * allowing any writes. So just check for that. | |
5199 | */ | |
b5254dd5 N |
5200 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
5201 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
5202 | /* not really in-place - so OK */; | |
5203 | else if (mddev->ro == 0) { | |
5204 | printk(KERN_ERR "md/raid:%s: in-place reshape " | |
5205 | "must be started in read-only mode " | |
5206 | "- aborting\n", | |
0c55e022 | 5207 | mdname(mddev)); |
67ac6011 N |
5208 | return -EINVAL; |
5209 | } | |
2c810cdd | 5210 | } else if (mddev->reshape_backwards |
b5254dd5 | 5211 | ? (here_new * mddev->new_chunk_sectors + min_offset_diff <= |
67ac6011 N |
5212 | here_old * mddev->chunk_sectors) |
5213 | : (here_new * mddev->new_chunk_sectors >= | |
b5254dd5 | 5214 | here_old * mddev->chunk_sectors + (-min_offset_diff))) { |
91adb564 | 5215 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
5216 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
5217 | "auto-recovery - aborting.\n", | |
5218 | mdname(mddev)); | |
91adb564 N |
5219 | return -EINVAL; |
5220 | } | |
0c55e022 N |
5221 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
5222 | mdname(mddev)); | |
91adb564 N |
5223 | /* OK, we should be able to continue; */ |
5224 | } else { | |
5225 | BUG_ON(mddev->level != mddev->new_level); | |
5226 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 5227 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 5228 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 5229 | } |
91adb564 | 5230 | |
245f46c2 N |
5231 | if (mddev->private == NULL) |
5232 | conf = setup_conf(mddev); | |
5233 | else | |
5234 | conf = mddev->private; | |
5235 | ||
91adb564 N |
5236 | if (IS_ERR(conf)) |
5237 | return PTR_ERR(conf); | |
5238 | ||
b5254dd5 | 5239 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
5240 | mddev->thread = conf->thread; |
5241 | conf->thread = NULL; | |
5242 | mddev->private = conf; | |
5243 | ||
17045f52 N |
5244 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
5245 | i++) { | |
5246 | rdev = conf->disks[i].rdev; | |
5247 | if (!rdev && conf->disks[i].replacement) { | |
5248 | /* The replacement is all we have yet */ | |
5249 | rdev = conf->disks[i].replacement; | |
5250 | conf->disks[i].replacement = NULL; | |
5251 | clear_bit(Replacement, &rdev->flags); | |
5252 | conf->disks[i].rdev = rdev; | |
5253 | } | |
5254 | if (!rdev) | |
c148ffdc | 5255 | continue; |
17045f52 N |
5256 | if (conf->disks[i].replacement && |
5257 | conf->reshape_progress != MaxSector) { | |
5258 | /* replacements and reshape simply do not mix. */ | |
5259 | printk(KERN_ERR "md: cannot handle concurrent " | |
5260 | "replacement and reshape.\n"); | |
5261 | goto abort; | |
5262 | } | |
2f115882 | 5263 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 5264 | working_disks++; |
2f115882 N |
5265 | continue; |
5266 | } | |
c148ffdc N |
5267 | /* This disc is not fully in-sync. However if it |
5268 | * just stored parity (beyond the recovery_offset), | |
5269 | * when we don't need to be concerned about the | |
5270 | * array being dirty. | |
5271 | * When reshape goes 'backwards', we never have | |
5272 | * partially completed devices, so we only need | |
5273 | * to worry about reshape going forwards. | |
5274 | */ | |
5275 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
5276 | if (mddev->major_version == 0 && | |
5277 | mddev->minor_version > 90) | |
5278 | rdev->recovery_offset = reshape_offset; | |
5279 | ||
c148ffdc N |
5280 | if (rdev->recovery_offset < reshape_offset) { |
5281 | /* We need to check old and new layout */ | |
5282 | if (!only_parity(rdev->raid_disk, | |
5283 | conf->algorithm, | |
5284 | conf->raid_disks, | |
5285 | conf->max_degraded)) | |
5286 | continue; | |
5287 | } | |
5288 | if (!only_parity(rdev->raid_disk, | |
5289 | conf->prev_algo, | |
5290 | conf->previous_raid_disks, | |
5291 | conf->max_degraded)) | |
5292 | continue; | |
5293 | dirty_parity_disks++; | |
5294 | } | |
91adb564 | 5295 | |
17045f52 N |
5296 | /* |
5297 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
5298 | */ | |
908f4fbd | 5299 | mddev->degraded = calc_degraded(conf); |
91adb564 | 5300 | |
674806d6 | 5301 | if (has_failed(conf)) { |
0c55e022 | 5302 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 5303 | " (%d/%d failed)\n", |
02c2de8c | 5304 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
5305 | goto abort; |
5306 | } | |
5307 | ||
91adb564 | 5308 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 5309 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
5310 | mddev->resync_max_sectors = mddev->dev_sectors; |
5311 | ||
c148ffdc | 5312 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 5313 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
5314 | if (mddev->ok_start_degraded) |
5315 | printk(KERN_WARNING | |
0c55e022 N |
5316 | "md/raid:%s: starting dirty degraded array" |
5317 | " - data corruption possible.\n", | |
6ff8d8ec N |
5318 | mdname(mddev)); |
5319 | else { | |
5320 | printk(KERN_ERR | |
0c55e022 | 5321 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
5322 | mdname(mddev)); |
5323 | goto abort; | |
5324 | } | |
1da177e4 LT |
5325 | } |
5326 | ||
1da177e4 | 5327 | if (mddev->degraded == 0) |
0c55e022 N |
5328 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
5329 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
5330 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
5331 | mddev->new_layout); | |
1da177e4 | 5332 | else |
0c55e022 N |
5333 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
5334 | " out of %d devices, algorithm %d\n", | |
5335 | mdname(mddev), conf->level, | |
5336 | mddev->raid_disks - mddev->degraded, | |
5337 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
5338 | |
5339 | print_raid5_conf(conf); | |
5340 | ||
fef9c61f | 5341 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 5342 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5343 | atomic_set(&conf->reshape_stripes, 0); |
5344 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5345 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5346 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5347 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5348 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5349 | "reshape"); |
f6705578 N |
5350 | } |
5351 | ||
1da177e4 LT |
5352 | |
5353 | /* Ok, everything is just fine now */ | |
a64c876f N |
5354 | if (mddev->to_remove == &raid5_attrs_group) |
5355 | mddev->to_remove = NULL; | |
00bcb4ac N |
5356 | else if (mddev->kobj.sd && |
5357 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 5358 | printk(KERN_WARNING |
4a5add49 | 5359 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 5360 | mdname(mddev)); |
4a5add49 | 5361 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5362 | |
4a5add49 | 5363 | if (mddev->queue) { |
9f7c2220 | 5364 | int chunk_size; |
4a5add49 N |
5365 | /* read-ahead size must cover two whole stripes, which |
5366 | * is 2 * (datadisks) * chunksize where 'n' is the | |
5367 | * number of raid devices | |
5368 | */ | |
5369 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
5370 | int stripe = data_disks * | |
5371 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
5372 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
5373 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 5374 | |
4a5add49 | 5375 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 5376 | |
11d8a6e3 N |
5377 | mddev->queue->backing_dev_info.congested_data = mddev; |
5378 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 5379 | |
9f7c2220 N |
5380 | chunk_size = mddev->chunk_sectors << 9; |
5381 | blk_queue_io_min(mddev->queue, chunk_size); | |
5382 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5383 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 5384 | |
05616be5 | 5385 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
5386 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
5387 | rdev->data_offset << 9); | |
05616be5 N |
5388 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
5389 | rdev->new_data_offset << 9); | |
5390 | } | |
9f7c2220 | 5391 | } |
23032a0e | 5392 | |
1da177e4 LT |
5393 | return 0; |
5394 | abort: | |
01f96c0a | 5395 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
5396 | print_raid5_conf(conf); |
5397 | free_conf(conf); | |
1da177e4 | 5398 | mddev->private = NULL; |
0c55e022 | 5399 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
5400 | return -EIO; |
5401 | } | |
5402 | ||
fd01b88c | 5403 | static int stop(struct mddev *mddev) |
1da177e4 | 5404 | { |
d1688a6d | 5405 | struct r5conf *conf = mddev->private; |
1da177e4 | 5406 | |
01f96c0a | 5407 | md_unregister_thread(&mddev->thread); |
11d8a6e3 N |
5408 | if (mddev->queue) |
5409 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 5410 | free_conf(conf); |
a64c876f N |
5411 | mddev->private = NULL; |
5412 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
5413 | return 0; |
5414 | } | |
5415 | ||
fd01b88c | 5416 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 5417 | { |
d1688a6d | 5418 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
5419 | int i; |
5420 | ||
9d8f0363 AN |
5421 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5422 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5423 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5424 | for (i = 0; i < conf->raid_disks; i++) |
5425 | seq_printf (seq, "%s", | |
5426 | conf->disks[i].rdev && | |
b2d444d7 | 5427 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5428 | seq_printf (seq, "]"); |
1da177e4 LT |
5429 | } |
5430 | ||
d1688a6d | 5431 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
5432 | { |
5433 | int i; | |
5434 | struct disk_info *tmp; | |
5435 | ||
0c55e022 | 5436 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
5437 | if (!conf) { |
5438 | printk("(conf==NULL)\n"); | |
5439 | return; | |
5440 | } | |
0c55e022 N |
5441 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
5442 | conf->raid_disks, | |
5443 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5444 | |
5445 | for (i = 0; i < conf->raid_disks; i++) { | |
5446 | char b[BDEVNAME_SIZE]; | |
5447 | tmp = conf->disks + i; | |
5448 | if (tmp->rdev) | |
0c55e022 N |
5449 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
5450 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
5451 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
5452 | } |
5453 | } | |
5454 | ||
fd01b88c | 5455 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
5456 | { |
5457 | int i; | |
d1688a6d | 5458 | struct r5conf *conf = mddev->private; |
1da177e4 | 5459 | struct disk_info *tmp; |
6b965620 N |
5460 | int count = 0; |
5461 | unsigned long flags; | |
1da177e4 LT |
5462 | |
5463 | for (i = 0; i < conf->raid_disks; i++) { | |
5464 | tmp = conf->disks + i; | |
dd054fce N |
5465 | if (tmp->replacement |
5466 | && tmp->replacement->recovery_offset == MaxSector | |
5467 | && !test_bit(Faulty, &tmp->replacement->flags) | |
5468 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
5469 | /* Replacement has just become active. */ | |
5470 | if (!tmp->rdev | |
5471 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
5472 | count++; | |
5473 | if (tmp->rdev) { | |
5474 | /* Replaced device not technically faulty, | |
5475 | * but we need to be sure it gets removed | |
5476 | * and never re-added. | |
5477 | */ | |
5478 | set_bit(Faulty, &tmp->rdev->flags); | |
5479 | sysfs_notify_dirent_safe( | |
5480 | tmp->rdev->sysfs_state); | |
5481 | } | |
5482 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
5483 | } else if (tmp->rdev | |
70fffd0b | 5484 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 5485 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 5486 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 5487 | count++; |
43c73ca4 | 5488 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
5489 | } |
5490 | } | |
6b965620 | 5491 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5492 | mddev->degraded = calc_degraded(conf); |
6b965620 | 5493 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 5494 | print_raid5_conf(conf); |
6b965620 | 5495 | return count; |
1da177e4 LT |
5496 | } |
5497 | ||
b8321b68 | 5498 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5499 | { |
d1688a6d | 5500 | struct r5conf *conf = mddev->private; |
1da177e4 | 5501 | int err = 0; |
b8321b68 | 5502 | int number = rdev->raid_disk; |
657e3e4d | 5503 | struct md_rdev **rdevp; |
1da177e4 LT |
5504 | struct disk_info *p = conf->disks + number; |
5505 | ||
5506 | print_raid5_conf(conf); | |
657e3e4d N |
5507 | if (rdev == p->rdev) |
5508 | rdevp = &p->rdev; | |
5509 | else if (rdev == p->replacement) | |
5510 | rdevp = &p->replacement; | |
5511 | else | |
5512 | return 0; | |
5513 | ||
5514 | if (number >= conf->raid_disks && | |
5515 | conf->reshape_progress == MaxSector) | |
5516 | clear_bit(In_sync, &rdev->flags); | |
5517 | ||
5518 | if (test_bit(In_sync, &rdev->flags) || | |
5519 | atomic_read(&rdev->nr_pending)) { | |
5520 | err = -EBUSY; | |
5521 | goto abort; | |
5522 | } | |
5523 | /* Only remove non-faulty devices if recovery | |
5524 | * isn't possible. | |
5525 | */ | |
5526 | if (!test_bit(Faulty, &rdev->flags) && | |
5527 | mddev->recovery_disabled != conf->recovery_disabled && | |
5528 | !has_failed(conf) && | |
dd054fce | 5529 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
5530 | number < conf->raid_disks) { |
5531 | err = -EBUSY; | |
5532 | goto abort; | |
5533 | } | |
5534 | *rdevp = NULL; | |
5535 | synchronize_rcu(); | |
5536 | if (atomic_read(&rdev->nr_pending)) { | |
5537 | /* lost the race, try later */ | |
5538 | err = -EBUSY; | |
5539 | *rdevp = rdev; | |
dd054fce N |
5540 | } else if (p->replacement) { |
5541 | /* We must have just cleared 'rdev' */ | |
5542 | p->rdev = p->replacement; | |
5543 | clear_bit(Replacement, &p->replacement->flags); | |
5544 | smp_mb(); /* Make sure other CPUs may see both as identical | |
5545 | * but will never see neither - if they are careful | |
5546 | */ | |
5547 | p->replacement = NULL; | |
5548 | clear_bit(WantReplacement, &rdev->flags); | |
5549 | } else | |
5550 | /* We might have just removed the Replacement as faulty- | |
5551 | * clear the bit just in case | |
5552 | */ | |
5553 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
5554 | abort: |
5555 | ||
5556 | print_raid5_conf(conf); | |
5557 | return err; | |
5558 | } | |
5559 | ||
fd01b88c | 5560 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5561 | { |
d1688a6d | 5562 | struct r5conf *conf = mddev->private; |
199050ea | 5563 | int err = -EEXIST; |
1da177e4 LT |
5564 | int disk; |
5565 | struct disk_info *p; | |
6c2fce2e NB |
5566 | int first = 0; |
5567 | int last = conf->raid_disks - 1; | |
1da177e4 | 5568 | |
7f0da59b N |
5569 | if (mddev->recovery_disabled == conf->recovery_disabled) |
5570 | return -EBUSY; | |
5571 | ||
dc10c643 | 5572 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 5573 | /* no point adding a device */ |
199050ea | 5574 | return -EINVAL; |
1da177e4 | 5575 | |
6c2fce2e NB |
5576 | if (rdev->raid_disk >= 0) |
5577 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5578 | |
5579 | /* | |
16a53ecc N |
5580 | * find the disk ... but prefer rdev->saved_raid_disk |
5581 | * if possible. | |
1da177e4 | 5582 | */ |
16a53ecc | 5583 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5584 | rdev->saved_raid_disk >= first && |
16a53ecc | 5585 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
5586 | first = rdev->saved_raid_disk; |
5587 | ||
5588 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
5589 | p = conf->disks + disk; |
5590 | if (p->rdev == NULL) { | |
b2d444d7 | 5591 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5592 | rdev->raid_disk = disk; |
199050ea | 5593 | err = 0; |
72626685 N |
5594 | if (rdev->saved_raid_disk != disk) |
5595 | conf->fullsync = 1; | |
d6065f7b | 5596 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 5597 | goto out; |
1da177e4 | 5598 | } |
5cfb22a1 N |
5599 | } |
5600 | for (disk = first; disk <= last; disk++) { | |
5601 | p = conf->disks + disk; | |
7bfec5f3 N |
5602 | if (test_bit(WantReplacement, &p->rdev->flags) && |
5603 | p->replacement == NULL) { | |
5604 | clear_bit(In_sync, &rdev->flags); | |
5605 | set_bit(Replacement, &rdev->flags); | |
5606 | rdev->raid_disk = disk; | |
5607 | err = 0; | |
5608 | conf->fullsync = 1; | |
5609 | rcu_assign_pointer(p->replacement, rdev); | |
5610 | break; | |
5611 | } | |
5612 | } | |
5cfb22a1 | 5613 | out: |
1da177e4 | 5614 | print_raid5_conf(conf); |
199050ea | 5615 | return err; |
1da177e4 LT |
5616 | } |
5617 | ||
fd01b88c | 5618 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
5619 | { |
5620 | /* no resync is happening, and there is enough space | |
5621 | * on all devices, so we can resize. | |
5622 | * We need to make sure resync covers any new space. | |
5623 | * If the array is shrinking we should possibly wait until | |
5624 | * any io in the removed space completes, but it hardly seems | |
5625 | * worth it. | |
5626 | */ | |
a4a6125a | 5627 | sector_t newsize; |
9d8f0363 | 5628 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
a4a6125a N |
5629 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
5630 | if (mddev->external_size && | |
5631 | mddev->array_sectors > newsize) | |
b522adcd | 5632 | return -EINVAL; |
a4a6125a N |
5633 | if (mddev->bitmap) { |
5634 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
5635 | if (ret) | |
5636 | return ret; | |
5637 | } | |
5638 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 5639 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5640 | revalidate_disk(mddev->gendisk); |
b098636c N |
5641 | if (sectors > mddev->dev_sectors && |
5642 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5643 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5644 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5645 | } | |
58c0fed4 | 5646 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5647 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5648 | return 0; |
5649 | } | |
5650 | ||
fd01b88c | 5651 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
5652 | { |
5653 | /* Can only proceed if there are plenty of stripe_heads. | |
5654 | * We need a minimum of one full stripe,, and for sensible progress | |
5655 | * it is best to have about 4 times that. | |
5656 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5657 | * allow for chunk sizes up to 256K, which is probably OK. | |
5658 | * If the chunk size is greater, user-space should request more | |
5659 | * stripe_heads first. | |
5660 | */ | |
d1688a6d | 5661 | struct r5conf *conf = mddev->private; |
01ee22b4 N |
5662 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
5663 | > conf->max_nr_stripes || | |
5664 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5665 | > conf->max_nr_stripes) { | |
0c55e022 N |
5666 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5667 | mdname(mddev), | |
01ee22b4 N |
5668 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5669 | / STRIPE_SIZE)*4); | |
5670 | return 0; | |
5671 | } | |
5672 | return 1; | |
5673 | } | |
5674 | ||
fd01b88c | 5675 | static int check_reshape(struct mddev *mddev) |
29269553 | 5676 | { |
d1688a6d | 5677 | struct r5conf *conf = mddev->private; |
29269553 | 5678 | |
88ce4930 N |
5679 | if (mddev->delta_disks == 0 && |
5680 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5681 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5682 | return 0; /* nothing to do */ |
674806d6 | 5683 | if (has_failed(conf)) |
ec32a2bd N |
5684 | return -EINVAL; |
5685 | if (mddev->delta_disks < 0) { | |
5686 | /* We might be able to shrink, but the devices must | |
5687 | * be made bigger first. | |
5688 | * For raid6, 4 is the minimum size. | |
5689 | * Otherwise 2 is the minimum | |
5690 | */ | |
5691 | int min = 2; | |
5692 | if (mddev->level == 6) | |
5693 | min = 4; | |
5694 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5695 | return -EINVAL; | |
5696 | } | |
29269553 | 5697 | |
01ee22b4 | 5698 | if (!check_stripe_cache(mddev)) |
29269553 | 5699 | return -ENOSPC; |
29269553 | 5700 | |
ec32a2bd | 5701 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5702 | } |
5703 | ||
fd01b88c | 5704 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 5705 | { |
d1688a6d | 5706 | struct r5conf *conf = mddev->private; |
3cb03002 | 5707 | struct md_rdev *rdev; |
63c70c4f | 5708 | int spares = 0; |
c04be0aa | 5709 | unsigned long flags; |
63c70c4f | 5710 | |
f416885e | 5711 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5712 | return -EBUSY; |
5713 | ||
01ee22b4 N |
5714 | if (!check_stripe_cache(mddev)) |
5715 | return -ENOSPC; | |
5716 | ||
30b67645 N |
5717 | if (has_failed(conf)) |
5718 | return -EINVAL; | |
5719 | ||
c6563a8c | 5720 | rdev_for_each(rdev, mddev) { |
469518a3 N |
5721 | if (!test_bit(In_sync, &rdev->flags) |
5722 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5723 | spares++; |
c6563a8c | 5724 | } |
63c70c4f | 5725 | |
f416885e | 5726 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5727 | /* Not enough devices even to make a degraded array |
5728 | * of that size | |
5729 | */ | |
5730 | return -EINVAL; | |
5731 | ||
ec32a2bd N |
5732 | /* Refuse to reduce size of the array. Any reductions in |
5733 | * array size must be through explicit setting of array_size | |
5734 | * attribute. | |
5735 | */ | |
5736 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5737 | < mddev->array_sectors) { | |
0c55e022 | 5738 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5739 | "before number of disks\n", mdname(mddev)); |
5740 | return -EINVAL; | |
5741 | } | |
5742 | ||
f6705578 | 5743 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5744 | spin_lock_irq(&conf->device_lock); |
5745 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5746 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5747 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5748 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5749 | conf->prev_algo = conf->algorithm; |
5750 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
5751 | conf->generation++; |
5752 | /* Code that selects data_offset needs to see the generation update | |
5753 | * if reshape_progress has been set - so a memory barrier needed. | |
5754 | */ | |
5755 | smp_mb(); | |
2c810cdd | 5756 | if (mddev->reshape_backwards) |
fef9c61f N |
5757 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
5758 | else | |
5759 | conf->reshape_progress = 0; | |
5760 | conf->reshape_safe = conf->reshape_progress; | |
29269553 N |
5761 | spin_unlock_irq(&conf->device_lock); |
5762 | ||
5763 | /* Add some new drives, as many as will fit. | |
5764 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5765 | * Don't add devices if we are reducing the number of |
5766 | * devices in the array. This is because it is not possible | |
5767 | * to correctly record the "partially reconstructed" state of | |
5768 | * such devices during the reshape and confusion could result. | |
29269553 | 5769 | */ |
87a8dec9 | 5770 | if (mddev->delta_disks >= 0) { |
dafb20fa | 5771 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
5772 | if (rdev->raid_disk < 0 && |
5773 | !test_bit(Faulty, &rdev->flags)) { | |
5774 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 5775 | if (rdev->raid_disk |
9d4c7d87 | 5776 | >= conf->previous_raid_disks) |
87a8dec9 | 5777 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 5778 | else |
87a8dec9 | 5779 | rdev->recovery_offset = 0; |
36fad858 NK |
5780 | |
5781 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 5782 | /* Failure here is OK */; |
50da0840 | 5783 | } |
87a8dec9 N |
5784 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5785 | && !test_bit(Faulty, &rdev->flags)) { | |
5786 | /* This is a spare that was manually added */ | |
5787 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 5788 | } |
29269553 | 5789 | |
87a8dec9 N |
5790 | /* When a reshape changes the number of devices, |
5791 | * ->degraded is measured against the larger of the | |
5792 | * pre and post number of devices. | |
5793 | */ | |
ec32a2bd | 5794 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5795 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
5796 | spin_unlock_irqrestore(&conf->device_lock, flags); |
5797 | } | |
63c70c4f | 5798 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5799 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5800 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5801 | |
29269553 N |
5802 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5803 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5804 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5805 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5806 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5807 | "reshape"); |
29269553 N |
5808 | if (!mddev->sync_thread) { |
5809 | mddev->recovery = 0; | |
5810 | spin_lock_irq(&conf->device_lock); | |
5811 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
05616be5 N |
5812 | rdev_for_each(rdev, mddev) |
5813 | rdev->new_data_offset = rdev->data_offset; | |
5814 | smp_wmb(); | |
fef9c61f | 5815 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 5816 | mddev->reshape_position = MaxSector; |
29269553 N |
5817 | spin_unlock_irq(&conf->device_lock); |
5818 | return -EAGAIN; | |
5819 | } | |
c8f517c4 | 5820 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5821 | md_wakeup_thread(mddev->sync_thread); |
5822 | md_new_event(mddev); | |
5823 | return 0; | |
5824 | } | |
29269553 | 5825 | |
ec32a2bd N |
5826 | /* This is called from the reshape thread and should make any |
5827 | * changes needed in 'conf' | |
5828 | */ | |
d1688a6d | 5829 | static void end_reshape(struct r5conf *conf) |
29269553 | 5830 | { |
29269553 | 5831 | |
f6705578 | 5832 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 5833 | struct md_rdev *rdev; |
f6705578 | 5834 | |
f6705578 | 5835 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5836 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
5837 | rdev_for_each(rdev, conf->mddev) |
5838 | rdev->data_offset = rdev->new_data_offset; | |
5839 | smp_wmb(); | |
fef9c61f | 5840 | conf->reshape_progress = MaxSector; |
f6705578 | 5841 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5842 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5843 | |
5844 | /* read-ahead size must cover two whole stripes, which is | |
5845 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5846 | */ | |
4a5add49 | 5847 | if (conf->mddev->queue) { |
cea9c228 | 5848 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5849 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5850 | / PAGE_SIZE); |
16a53ecc N |
5851 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5852 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5853 | } | |
29269553 | 5854 | } |
29269553 N |
5855 | } |
5856 | ||
ec32a2bd N |
5857 | /* This is called from the raid5d thread with mddev_lock held. |
5858 | * It makes config changes to the device. | |
5859 | */ | |
fd01b88c | 5860 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 5861 | { |
d1688a6d | 5862 | struct r5conf *conf = mddev->private; |
cea9c228 N |
5863 | |
5864 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5865 | ||
ec32a2bd N |
5866 | if (mddev->delta_disks > 0) { |
5867 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5868 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5869 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5870 | } else { |
5871 | int d; | |
908f4fbd N |
5872 | spin_lock_irq(&conf->device_lock); |
5873 | mddev->degraded = calc_degraded(conf); | |
5874 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
5875 | for (d = conf->raid_disks ; |
5876 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 5877 | d++) { |
3cb03002 | 5878 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
5879 | if (rdev) |
5880 | clear_bit(In_sync, &rdev->flags); | |
5881 | rdev = conf->disks[d].replacement; | |
5882 | if (rdev) | |
5883 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 5884 | } |
cea9c228 | 5885 | } |
88ce4930 | 5886 | mddev->layout = conf->algorithm; |
09c9e5fa | 5887 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5888 | mddev->reshape_position = MaxSector; |
5889 | mddev->delta_disks = 0; | |
2c810cdd | 5890 | mddev->reshape_backwards = 0; |
cea9c228 N |
5891 | } |
5892 | } | |
5893 | ||
fd01b88c | 5894 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 5895 | { |
d1688a6d | 5896 | struct r5conf *conf = mddev->private; |
72626685 N |
5897 | |
5898 | switch(state) { | |
e464eafd N |
5899 | case 2: /* resume for a suspend */ |
5900 | wake_up(&conf->wait_for_overlap); | |
5901 | break; | |
5902 | ||
72626685 N |
5903 | case 1: /* stop all writes */ |
5904 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5905 | /* '2' tells resync/reshape to pause so that all |
5906 | * active stripes can drain | |
5907 | */ | |
5908 | conf->quiesce = 2; | |
72626685 | 5909 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5910 | atomic_read(&conf->active_stripes) == 0 && |
5911 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5912 | conf->device_lock, /* nothing */); |
64bd660b | 5913 | conf->quiesce = 1; |
72626685 | 5914 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5915 | /* allow reshape to continue */ |
5916 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5917 | break; |
5918 | ||
5919 | case 0: /* re-enable writes */ | |
5920 | spin_lock_irq(&conf->device_lock); | |
5921 | conf->quiesce = 0; | |
5922 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5923 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5924 | spin_unlock_irq(&conf->device_lock); |
5925 | break; | |
5926 | } | |
72626685 | 5927 | } |
b15c2e57 | 5928 | |
d562b0c4 | 5929 | |
fd01b88c | 5930 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 5931 | { |
e373ab10 | 5932 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 5933 | sector_t sectors; |
54071b38 | 5934 | |
f1b29bca | 5935 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 5936 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
5937 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5938 | mdname(mddev)); | |
f1b29bca DW |
5939 | return ERR_PTR(-EINVAL); |
5940 | } | |
5941 | ||
e373ab10 N |
5942 | sectors = raid0_conf->strip_zone[0].zone_end; |
5943 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 5944 | mddev->dev_sectors = sectors; |
f1b29bca | 5945 | mddev->new_level = level; |
54071b38 TM |
5946 | mddev->new_layout = ALGORITHM_PARITY_N; |
5947 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5948 | mddev->raid_disks += 1; | |
5949 | mddev->delta_disks = 1; | |
5950 | /* make sure it will be not marked as dirty */ | |
5951 | mddev->recovery_cp = MaxSector; | |
5952 | ||
5953 | return setup_conf(mddev); | |
5954 | } | |
5955 | ||
5956 | ||
fd01b88c | 5957 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
5958 | { |
5959 | int chunksect; | |
5960 | ||
5961 | if (mddev->raid_disks != 2 || | |
5962 | mddev->degraded > 1) | |
5963 | return ERR_PTR(-EINVAL); | |
5964 | ||
5965 | /* Should check if there are write-behind devices? */ | |
5966 | ||
5967 | chunksect = 64*2; /* 64K by default */ | |
5968 | ||
5969 | /* The array must be an exact multiple of chunksize */ | |
5970 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5971 | chunksect >>= 1; | |
5972 | ||
5973 | if ((chunksect<<9) < STRIPE_SIZE) | |
5974 | /* array size does not allow a suitable chunk size */ | |
5975 | return ERR_PTR(-EINVAL); | |
5976 | ||
5977 | mddev->new_level = 5; | |
5978 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5979 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5980 | |
5981 | return setup_conf(mddev); | |
5982 | } | |
5983 | ||
fd01b88c | 5984 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
5985 | { |
5986 | int new_layout; | |
5987 | ||
5988 | switch (mddev->layout) { | |
5989 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5990 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5991 | break; | |
5992 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5993 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5994 | break; | |
5995 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5996 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5997 | break; | |
5998 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5999 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
6000 | break; | |
6001 | case ALGORITHM_PARITY_0_6: | |
6002 | new_layout = ALGORITHM_PARITY_0; | |
6003 | break; | |
6004 | case ALGORITHM_PARITY_N: | |
6005 | new_layout = ALGORITHM_PARITY_N; | |
6006 | break; | |
6007 | default: | |
6008 | return ERR_PTR(-EINVAL); | |
6009 | } | |
6010 | mddev->new_level = 5; | |
6011 | mddev->new_layout = new_layout; | |
6012 | mddev->delta_disks = -1; | |
6013 | mddev->raid_disks -= 1; | |
6014 | return setup_conf(mddev); | |
6015 | } | |
6016 | ||
d562b0c4 | 6017 | |
fd01b88c | 6018 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 6019 | { |
88ce4930 N |
6020 | /* For a 2-drive array, the layout and chunk size can be changed |
6021 | * immediately as not restriping is needed. | |
6022 | * For larger arrays we record the new value - after validation | |
6023 | * to be used by a reshape pass. | |
b3546035 | 6024 | */ |
d1688a6d | 6025 | struct r5conf *conf = mddev->private; |
597a711b | 6026 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 6027 | |
597a711b | 6028 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
6029 | return -EINVAL; |
6030 | if (new_chunk > 0) { | |
0ba459d2 | 6031 | if (!is_power_of_2(new_chunk)) |
b3546035 | 6032 | return -EINVAL; |
597a711b | 6033 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 6034 | return -EINVAL; |
597a711b | 6035 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
6036 | /* not factor of array size */ |
6037 | return -EINVAL; | |
6038 | } | |
6039 | ||
6040 | /* They look valid */ | |
6041 | ||
88ce4930 | 6042 | if (mddev->raid_disks == 2) { |
597a711b N |
6043 | /* can make the change immediately */ |
6044 | if (mddev->new_layout >= 0) { | |
6045 | conf->algorithm = mddev->new_layout; | |
6046 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
6047 | } |
6048 | if (new_chunk > 0) { | |
597a711b N |
6049 | conf->chunk_sectors = new_chunk ; |
6050 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
6051 | } |
6052 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
6053 | md_wakeup_thread(mddev->thread); | |
b3546035 | 6054 | } |
50ac168a | 6055 | return check_reshape(mddev); |
88ce4930 N |
6056 | } |
6057 | ||
fd01b88c | 6058 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 6059 | { |
597a711b | 6060 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 6061 | |
597a711b | 6062 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 6063 | return -EINVAL; |
b3546035 | 6064 | if (new_chunk > 0) { |
0ba459d2 | 6065 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 6066 | return -EINVAL; |
597a711b | 6067 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 6068 | return -EINVAL; |
597a711b | 6069 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
6070 | /* not factor of array size */ |
6071 | return -EINVAL; | |
b3546035 | 6072 | } |
88ce4930 N |
6073 | |
6074 | /* They look valid */ | |
50ac168a | 6075 | return check_reshape(mddev); |
b3546035 N |
6076 | } |
6077 | ||
fd01b88c | 6078 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
6079 | { |
6080 | /* raid5 can take over: | |
f1b29bca | 6081 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
6082 | * raid1 - if there are two drives. We need to know the chunk size |
6083 | * raid4 - trivial - just use a raid4 layout. | |
6084 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 6085 | */ |
f1b29bca DW |
6086 | if (mddev->level == 0) |
6087 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
6088 | if (mddev->level == 1) |
6089 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
6090 | if (mddev->level == 4) { |
6091 | mddev->new_layout = ALGORITHM_PARITY_N; | |
6092 | mddev->new_level = 5; | |
6093 | return setup_conf(mddev); | |
6094 | } | |
fc9739c6 N |
6095 | if (mddev->level == 6) |
6096 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
6097 | |
6098 | return ERR_PTR(-EINVAL); | |
6099 | } | |
6100 | ||
fd01b88c | 6101 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 6102 | { |
f1b29bca DW |
6103 | /* raid4 can take over: |
6104 | * raid0 - if there is only one strip zone | |
6105 | * raid5 - if layout is right | |
a78d38a1 | 6106 | */ |
f1b29bca DW |
6107 | if (mddev->level == 0) |
6108 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
6109 | if (mddev->level == 5 && |
6110 | mddev->layout == ALGORITHM_PARITY_N) { | |
6111 | mddev->new_layout = 0; | |
6112 | mddev->new_level = 4; | |
6113 | return setup_conf(mddev); | |
6114 | } | |
6115 | return ERR_PTR(-EINVAL); | |
6116 | } | |
d562b0c4 | 6117 | |
84fc4b56 | 6118 | static struct md_personality raid5_personality; |
245f46c2 | 6119 | |
fd01b88c | 6120 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
6121 | { |
6122 | /* Currently can only take over a raid5. We map the | |
6123 | * personality to an equivalent raid6 personality | |
6124 | * with the Q block at the end. | |
6125 | */ | |
6126 | int new_layout; | |
6127 | ||
6128 | if (mddev->pers != &raid5_personality) | |
6129 | return ERR_PTR(-EINVAL); | |
6130 | if (mddev->degraded > 1) | |
6131 | return ERR_PTR(-EINVAL); | |
6132 | if (mddev->raid_disks > 253) | |
6133 | return ERR_PTR(-EINVAL); | |
6134 | if (mddev->raid_disks < 3) | |
6135 | return ERR_PTR(-EINVAL); | |
6136 | ||
6137 | switch (mddev->layout) { | |
6138 | case ALGORITHM_LEFT_ASYMMETRIC: | |
6139 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
6140 | break; | |
6141 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6142 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
6143 | break; | |
6144 | case ALGORITHM_LEFT_SYMMETRIC: | |
6145 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
6146 | break; | |
6147 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6148 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
6149 | break; | |
6150 | case ALGORITHM_PARITY_0: | |
6151 | new_layout = ALGORITHM_PARITY_0_6; | |
6152 | break; | |
6153 | case ALGORITHM_PARITY_N: | |
6154 | new_layout = ALGORITHM_PARITY_N; | |
6155 | break; | |
6156 | default: | |
6157 | return ERR_PTR(-EINVAL); | |
6158 | } | |
6159 | mddev->new_level = 6; | |
6160 | mddev->new_layout = new_layout; | |
6161 | mddev->delta_disks = 1; | |
6162 | mddev->raid_disks += 1; | |
6163 | return setup_conf(mddev); | |
6164 | } | |
6165 | ||
6166 | ||
84fc4b56 | 6167 | static struct md_personality raid6_personality = |
16a53ecc N |
6168 | { |
6169 | .name = "raid6", | |
6170 | .level = 6, | |
6171 | .owner = THIS_MODULE, | |
6172 | .make_request = make_request, | |
6173 | .run = run, | |
6174 | .stop = stop, | |
6175 | .status = status, | |
6176 | .error_handler = error, | |
6177 | .hot_add_disk = raid5_add_disk, | |
6178 | .hot_remove_disk= raid5_remove_disk, | |
6179 | .spare_active = raid5_spare_active, | |
6180 | .sync_request = sync_request, | |
6181 | .resize = raid5_resize, | |
80c3a6ce | 6182 | .size = raid5_size, |
50ac168a | 6183 | .check_reshape = raid6_check_reshape, |
f416885e | 6184 | .start_reshape = raid5_start_reshape, |
cea9c228 | 6185 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 6186 | .quiesce = raid5_quiesce, |
245f46c2 | 6187 | .takeover = raid6_takeover, |
16a53ecc | 6188 | }; |
84fc4b56 | 6189 | static struct md_personality raid5_personality = |
1da177e4 LT |
6190 | { |
6191 | .name = "raid5", | |
2604b703 | 6192 | .level = 5, |
1da177e4 LT |
6193 | .owner = THIS_MODULE, |
6194 | .make_request = make_request, | |
6195 | .run = run, | |
6196 | .stop = stop, | |
6197 | .status = status, | |
6198 | .error_handler = error, | |
6199 | .hot_add_disk = raid5_add_disk, | |
6200 | .hot_remove_disk= raid5_remove_disk, | |
6201 | .spare_active = raid5_spare_active, | |
6202 | .sync_request = sync_request, | |
6203 | .resize = raid5_resize, | |
80c3a6ce | 6204 | .size = raid5_size, |
63c70c4f N |
6205 | .check_reshape = raid5_check_reshape, |
6206 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 6207 | .finish_reshape = raid5_finish_reshape, |
72626685 | 6208 | .quiesce = raid5_quiesce, |
d562b0c4 | 6209 | .takeover = raid5_takeover, |
1da177e4 LT |
6210 | }; |
6211 | ||
84fc4b56 | 6212 | static struct md_personality raid4_personality = |
1da177e4 | 6213 | { |
2604b703 N |
6214 | .name = "raid4", |
6215 | .level = 4, | |
6216 | .owner = THIS_MODULE, | |
6217 | .make_request = make_request, | |
6218 | .run = run, | |
6219 | .stop = stop, | |
6220 | .status = status, | |
6221 | .error_handler = error, | |
6222 | .hot_add_disk = raid5_add_disk, | |
6223 | .hot_remove_disk= raid5_remove_disk, | |
6224 | .spare_active = raid5_spare_active, | |
6225 | .sync_request = sync_request, | |
6226 | .resize = raid5_resize, | |
80c3a6ce | 6227 | .size = raid5_size, |
3d37890b N |
6228 | .check_reshape = raid5_check_reshape, |
6229 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 6230 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 6231 | .quiesce = raid5_quiesce, |
a78d38a1 | 6232 | .takeover = raid4_takeover, |
2604b703 N |
6233 | }; |
6234 | ||
6235 | static int __init raid5_init(void) | |
6236 | { | |
16a53ecc | 6237 | register_md_personality(&raid6_personality); |
2604b703 N |
6238 | register_md_personality(&raid5_personality); |
6239 | register_md_personality(&raid4_personality); | |
6240 | return 0; | |
1da177e4 LT |
6241 | } |
6242 | ||
2604b703 | 6243 | static void raid5_exit(void) |
1da177e4 | 6244 | { |
16a53ecc | 6245 | unregister_md_personality(&raid6_personality); |
2604b703 N |
6246 | unregister_md_personality(&raid5_personality); |
6247 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
6248 | } |
6249 | ||
6250 | module_init(raid5_init); | |
6251 | module_exit(raid5_exit); | |
6252 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 6253 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 6254 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
6255 | MODULE_ALIAS("md-raid5"); |
6256 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
6257 | MODULE_ALIAS("md-level-5"); |
6258 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
6259 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
6260 | MODULE_ALIAS("md-raid6"); | |
6261 | MODULE_ALIAS("md-level-6"); | |
6262 | ||
6263 | /* This used to be two separate modules, they were: */ | |
6264 | MODULE_ALIAS("raid5"); | |
6265 | MODULE_ALIAS("raid6"); |