]>
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 JA |
101 | */ |
102 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
103 | { | |
5b99c2ff | 104 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
105 | } |
106 | ||
107 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
108 | { | |
5b99c2ff | 109 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
110 | } |
111 | ||
112 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
113 | { | |
114 | --bio->bi_phys_segments; | |
115 | return raid5_bi_phys_segments(bio); | |
116 | } | |
117 | ||
118 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
119 | { | |
120 | unsigned short val = raid5_bi_hw_segments(bio); | |
121 | ||
122 | --val; | |
5b99c2ff | 123 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
124 | return val; |
125 | } | |
126 | ||
127 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
128 | { | |
9b2dc8b6 | 129 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16); |
960e739d JA |
130 | } |
131 | ||
d0dabf7e N |
132 | /* Find first data disk in a raid6 stripe */ |
133 | static inline int raid6_d0(struct stripe_head *sh) | |
134 | { | |
67cc2b81 N |
135 | if (sh->ddf_layout) |
136 | /* ddf always start from first device */ | |
137 | return 0; | |
138 | /* md starts just after Q block */ | |
d0dabf7e N |
139 | if (sh->qd_idx == sh->disks - 1) |
140 | return 0; | |
141 | else | |
142 | return sh->qd_idx + 1; | |
143 | } | |
16a53ecc N |
144 | static inline int raid6_next_disk(int disk, int raid_disks) |
145 | { | |
146 | disk++; | |
147 | return (disk < raid_disks) ? disk : 0; | |
148 | } | |
a4456856 | 149 | |
d0dabf7e N |
150 | /* When walking through the disks in a raid5, starting at raid6_d0, |
151 | * We need to map each disk to a 'slot', where the data disks are slot | |
152 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
153 | * is raid_disks-1. This help does that mapping. | |
154 | */ | |
67cc2b81 N |
155 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
156 | int *count, int syndrome_disks) | |
d0dabf7e | 157 | { |
6629542e | 158 | int slot = *count; |
67cc2b81 | 159 | |
e4424fee | 160 | if (sh->ddf_layout) |
6629542e | 161 | (*count)++; |
d0dabf7e | 162 | if (idx == sh->pd_idx) |
67cc2b81 | 163 | return syndrome_disks; |
d0dabf7e | 164 | if (idx == sh->qd_idx) |
67cc2b81 | 165 | return syndrome_disks + 1; |
e4424fee | 166 | if (!sh->ddf_layout) |
6629542e | 167 | (*count)++; |
d0dabf7e N |
168 | return slot; |
169 | } | |
170 | ||
a4456856 DW |
171 | static void return_io(struct bio *return_bi) |
172 | { | |
173 | struct bio *bi = return_bi; | |
174 | while (bi) { | |
a4456856 DW |
175 | |
176 | return_bi = bi->bi_next; | |
177 | bi->bi_next = NULL; | |
178 | bi->bi_size = 0; | |
0e13fe23 | 179 | bio_endio(bi, 0); |
a4456856 DW |
180 | bi = return_bi; |
181 | } | |
182 | } | |
183 | ||
d1688a6d | 184 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 185 | |
600aa109 DW |
186 | static int stripe_operations_active(struct stripe_head *sh) |
187 | { | |
188 | return sh->check_state || sh->reconstruct_state || | |
189 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
190 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
191 | } | |
192 | ||
d1688a6d | 193 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 LT |
194 | { |
195 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
196 | BUG_ON(!list_empty(&sh->lru)); |
197 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 198 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
482c0834 | 199 | if (test_bit(STRIPE_DELAYED, &sh->state)) |
1da177e4 | 200 | list_add_tail(&sh->lru, &conf->delayed_list); |
482c0834 N |
201 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
202 | sh->bm_seq - conf->seq_write > 0) | |
72626685 | 203 | list_add_tail(&sh->lru, &conf->bitmap_list); |
482c0834 | 204 | else { |
72626685 | 205 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 206 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 207 | } |
1da177e4 LT |
208 | md_wakeup_thread(conf->mddev->thread); |
209 | } else { | |
600aa109 | 210 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
211 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
212 | atomic_dec(&conf->preread_active_stripes); | |
213 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
214 | md_wakeup_thread(conf->mddev->thread); | |
215 | } | |
1da177e4 | 216 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
217 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
218 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 219 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
220 | if (conf->retry_read_aligned) |
221 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 222 | } |
1da177e4 LT |
223 | } |
224 | } | |
225 | } | |
d0dabf7e | 226 | |
1da177e4 LT |
227 | static void release_stripe(struct stripe_head *sh) |
228 | { | |
d1688a6d | 229 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 230 | unsigned long flags; |
16a53ecc | 231 | |
1da177e4 LT |
232 | spin_lock_irqsave(&conf->device_lock, flags); |
233 | __release_stripe(conf, sh); | |
234 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
235 | } | |
236 | ||
fccddba0 | 237 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 238 | { |
45b4233c DW |
239 | pr_debug("remove_hash(), stripe %llu\n", |
240 | (unsigned long long)sh->sector); | |
1da177e4 | 241 | |
fccddba0 | 242 | hlist_del_init(&sh->hash); |
1da177e4 LT |
243 | } |
244 | ||
d1688a6d | 245 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 246 | { |
fccddba0 | 247 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 248 | |
45b4233c DW |
249 | pr_debug("insert_hash(), stripe %llu\n", |
250 | (unsigned long long)sh->sector); | |
1da177e4 | 251 | |
fccddba0 | 252 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
253 | } |
254 | ||
255 | ||
256 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
d1688a6d | 257 | static struct stripe_head *get_free_stripe(struct r5conf *conf) |
1da177e4 LT |
258 | { |
259 | struct stripe_head *sh = NULL; | |
260 | struct list_head *first; | |
261 | ||
1da177e4 LT |
262 | if (list_empty(&conf->inactive_list)) |
263 | goto out; | |
264 | first = conf->inactive_list.next; | |
265 | sh = list_entry(first, struct stripe_head, lru); | |
266 | list_del_init(first); | |
267 | remove_hash(sh); | |
268 | atomic_inc(&conf->active_stripes); | |
269 | out: | |
270 | return sh; | |
271 | } | |
272 | ||
e4e11e38 | 273 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
274 | { |
275 | struct page *p; | |
276 | int i; | |
e4e11e38 | 277 | int num = sh->raid_conf->pool_size; |
1da177e4 | 278 | |
e4e11e38 | 279 | for (i = 0; i < num ; i++) { |
1da177e4 LT |
280 | p = sh->dev[i].page; |
281 | if (!p) | |
282 | continue; | |
283 | sh->dev[i].page = NULL; | |
2d1f3b5d | 284 | put_page(p); |
1da177e4 LT |
285 | } |
286 | } | |
287 | ||
e4e11e38 | 288 | static int grow_buffers(struct stripe_head *sh) |
1da177e4 LT |
289 | { |
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 | struct page *page; |
295 | ||
296 | if (!(page = alloc_page(GFP_KERNEL))) { | |
297 | return 1; | |
298 | } | |
299 | sh->dev[i].page = page; | |
300 | } | |
301 | return 0; | |
302 | } | |
303 | ||
784052ec | 304 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 305 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 306 | struct stripe_head *sh); |
1da177e4 | 307 | |
b5663ba4 | 308 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 309 | { |
d1688a6d | 310 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 311 | int i; |
1da177e4 | 312 | |
78bafebd ES |
313 | BUG_ON(atomic_read(&sh->count) != 0); |
314 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 315 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 316 | |
45b4233c | 317 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
318 | (unsigned long long)sh->sector); |
319 | ||
320 | remove_hash(sh); | |
16a53ecc | 321 | |
86b42c71 | 322 | sh->generation = conf->generation - previous; |
b5663ba4 | 323 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 324 | sh->sector = sector; |
911d4ee8 | 325 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
326 | sh->state = 0; |
327 | ||
7ecaa1e6 N |
328 | |
329 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
330 | struct r5dev *dev = &sh->dev[i]; |
331 | ||
d84e0f10 | 332 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 333 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 334 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 335 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 336 | dev->read, dev->towrite, dev->written, |
1da177e4 | 337 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 338 | WARN_ON(1); |
1da177e4 LT |
339 | } |
340 | dev->flags = 0; | |
784052ec | 341 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
342 | } |
343 | insert_hash(conf, sh); | |
344 | } | |
345 | ||
d1688a6d | 346 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 347 | short generation) |
1da177e4 LT |
348 | { |
349 | struct stripe_head *sh; | |
fccddba0 | 350 | struct hlist_node *hn; |
1da177e4 | 351 | |
45b4233c | 352 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 353 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 354 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 355 | return sh; |
45b4233c | 356 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
357 | return NULL; |
358 | } | |
359 | ||
674806d6 N |
360 | /* |
361 | * Need to check if array has failed when deciding whether to: | |
362 | * - start an array | |
363 | * - remove non-faulty devices | |
364 | * - add a spare | |
365 | * - allow a reshape | |
366 | * This determination is simple when no reshape is happening. | |
367 | * However if there is a reshape, we need to carefully check | |
368 | * both the before and after sections. | |
369 | * This is because some failed devices may only affect one | |
370 | * of the two sections, and some non-in_sync devices may | |
371 | * be insync in the section most affected by failed devices. | |
372 | */ | |
908f4fbd | 373 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 374 | { |
908f4fbd | 375 | int degraded, degraded2; |
674806d6 | 376 | int i; |
674806d6 N |
377 | |
378 | rcu_read_lock(); | |
379 | degraded = 0; | |
380 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 381 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
674806d6 N |
382 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
383 | degraded++; | |
384 | else if (test_bit(In_sync, &rdev->flags)) | |
385 | ; | |
386 | else | |
387 | /* not in-sync or faulty. | |
388 | * If the reshape increases the number of devices, | |
389 | * this is being recovered by the reshape, so | |
390 | * this 'previous' section is not in_sync. | |
391 | * If the number of devices is being reduced however, | |
392 | * the device can only be part of the array if | |
393 | * we are reverting a reshape, so this section will | |
394 | * be in-sync. | |
395 | */ | |
396 | if (conf->raid_disks >= conf->previous_raid_disks) | |
397 | degraded++; | |
398 | } | |
399 | rcu_read_unlock(); | |
908f4fbd N |
400 | if (conf->raid_disks == conf->previous_raid_disks) |
401 | return degraded; | |
674806d6 | 402 | rcu_read_lock(); |
908f4fbd | 403 | degraded2 = 0; |
674806d6 | 404 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 405 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
674806d6 | 406 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 407 | degraded2++; |
674806d6 N |
408 | else if (test_bit(In_sync, &rdev->flags)) |
409 | ; | |
410 | else | |
411 | /* not in-sync or faulty. | |
412 | * If reshape increases the number of devices, this | |
413 | * section has already been recovered, else it | |
414 | * almost certainly hasn't. | |
415 | */ | |
416 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 417 | degraded2++; |
674806d6 N |
418 | } |
419 | rcu_read_unlock(); | |
908f4fbd N |
420 | if (degraded2 > degraded) |
421 | return degraded2; | |
422 | return degraded; | |
423 | } | |
424 | ||
425 | static int has_failed(struct r5conf *conf) | |
426 | { | |
427 | int degraded; | |
428 | ||
429 | if (conf->mddev->reshape_position == MaxSector) | |
430 | return conf->mddev->degraded > conf->max_degraded; | |
431 | ||
432 | degraded = calc_degraded(conf); | |
674806d6 N |
433 | if (degraded > conf->max_degraded) |
434 | return 1; | |
435 | return 0; | |
436 | } | |
437 | ||
b5663ba4 | 438 | static struct stripe_head * |
d1688a6d | 439 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 440 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
441 | { |
442 | struct stripe_head *sh; | |
443 | ||
45b4233c | 444 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
445 | |
446 | spin_lock_irq(&conf->device_lock); | |
447 | ||
448 | do { | |
72626685 | 449 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 450 | conf->quiesce == 0 || noquiesce, |
72626685 | 451 | conf->device_lock, /* nothing */); |
86b42c71 | 452 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
453 | if (!sh) { |
454 | if (!conf->inactive_blocked) | |
455 | sh = get_free_stripe(conf); | |
456 | if (noblock && sh == NULL) | |
457 | break; | |
458 | if (!sh) { | |
459 | conf->inactive_blocked = 1; | |
460 | wait_event_lock_irq(conf->wait_for_stripe, | |
461 | !list_empty(&conf->inactive_list) && | |
5036805b N |
462 | (atomic_read(&conf->active_stripes) |
463 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
464 | || !conf->inactive_blocked), |
465 | conf->device_lock, | |
7c13edc8 | 466 | ); |
1da177e4 LT |
467 | conf->inactive_blocked = 0; |
468 | } else | |
b5663ba4 | 469 | init_stripe(sh, sector, previous); |
1da177e4 LT |
470 | } else { |
471 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
472 | BUG_ON(!list_empty(&sh->lru) |
473 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
474 | } else { |
475 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
476 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
477 | if (list_empty(&sh->lru) && |
478 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
479 | BUG(); |
480 | list_del_init(&sh->lru); | |
1da177e4 LT |
481 | } |
482 | } | |
483 | } while (sh == NULL); | |
484 | ||
485 | if (sh) | |
486 | atomic_inc(&sh->count); | |
487 | ||
488 | spin_unlock_irq(&conf->device_lock); | |
489 | return sh; | |
490 | } | |
491 | ||
6712ecf8 N |
492 | static void |
493 | raid5_end_read_request(struct bio *bi, int error); | |
494 | static void | |
495 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 496 | |
c4e5ac0a | 497 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 498 | { |
d1688a6d | 499 | struct r5conf *conf = sh->raid_conf; |
91c00924 DW |
500 | int i, disks = sh->disks; |
501 | ||
502 | might_sleep(); | |
503 | ||
504 | for (i = disks; i--; ) { | |
505 | int rw; | |
9a3e1101 | 506 | int replace_only = 0; |
977df362 N |
507 | struct bio *bi, *rbi; |
508 | struct md_rdev *rdev, *rrdev = NULL; | |
e9c7469b TH |
509 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
510 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
511 | rw = WRITE_FUA; | |
512 | else | |
513 | rw = WRITE; | |
514 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
91c00924 | 515 | rw = READ; |
9a3e1101 N |
516 | else if (test_and_clear_bit(R5_WantReplace, |
517 | &sh->dev[i].flags)) { | |
518 | rw = WRITE; | |
519 | replace_only = 1; | |
520 | } else | |
91c00924 DW |
521 | continue; |
522 | ||
523 | bi = &sh->dev[i].req; | |
977df362 | 524 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 DW |
525 | |
526 | bi->bi_rw = rw; | |
977df362 N |
527 | rbi->bi_rw = rw; |
528 | if (rw & WRITE) { | |
91c00924 | 529 | bi->bi_end_io = raid5_end_write_request; |
977df362 N |
530 | rbi->bi_end_io = raid5_end_write_request; |
531 | } else | |
91c00924 DW |
532 | bi->bi_end_io = raid5_end_read_request; |
533 | ||
534 | rcu_read_lock(); | |
977df362 | 535 | rdev = rcu_dereference(conf->disks[i].rdev); |
9a3e1101 N |
536 | rrdev = rcu_dereference(conf->disks[i].replacement); |
537 | if (rw & WRITE) { | |
538 | if (replace_only) | |
539 | rdev = NULL; | |
540 | } else { | |
541 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) | |
542 | rdev = rrdev; | |
543 | rrdev = NULL; | |
544 | } | |
977df362 | 545 | |
91c00924 DW |
546 | if (rdev && test_bit(Faulty, &rdev->flags)) |
547 | rdev = NULL; | |
548 | if (rdev) | |
549 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
550 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
551 | rrdev = NULL; | |
552 | if (rrdev) | |
553 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
554 | rcu_read_unlock(); |
555 | ||
73e92e51 | 556 | /* We have already checked bad blocks for reads. Now |
977df362 N |
557 | * need to check for writes. We never accept write errors |
558 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
559 | */ |
560 | while ((rw & WRITE) && rdev && | |
561 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
562 | sector_t first_bad; | |
563 | int bad_sectors; | |
564 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
565 | &first_bad, &bad_sectors); | |
566 | if (!bad) | |
567 | break; | |
568 | ||
569 | if (bad < 0) { | |
570 | set_bit(BlockedBadBlocks, &rdev->flags); | |
571 | if (!conf->mddev->external && | |
572 | conf->mddev->flags) { | |
573 | /* It is very unlikely, but we might | |
574 | * still need to write out the | |
575 | * bad block log - better give it | |
576 | * a chance*/ | |
577 | md_check_recovery(conf->mddev); | |
578 | } | |
579 | md_wait_for_blocked_rdev(rdev, conf->mddev); | |
580 | } else { | |
581 | /* Acknowledged bad block - skip the write */ | |
582 | rdev_dec_pending(rdev, conf->mddev); | |
583 | rdev = NULL; | |
584 | } | |
585 | } | |
586 | ||
91c00924 | 587 | if (rdev) { |
9a3e1101 N |
588 | if (s->syncing || s->expanding || s->expanded |
589 | || s->replacing) | |
91c00924 DW |
590 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
591 | ||
2b7497f0 DW |
592 | set_bit(STRIPE_IO_STARTED, &sh->state); |
593 | ||
91c00924 DW |
594 | bi->bi_bdev = rdev->bdev; |
595 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 596 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
597 | bi->bi_rw, i); |
598 | atomic_inc(&sh->count); | |
599 | bi->bi_sector = sh->sector + rdev->data_offset; | |
600 | bi->bi_flags = 1 << BIO_UPTODATE; | |
91c00924 | 601 | bi->bi_idx = 0; |
91c00924 DW |
602 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
603 | bi->bi_io_vec[0].bv_offset = 0; | |
604 | bi->bi_size = STRIPE_SIZE; | |
605 | bi->bi_next = NULL; | |
977df362 N |
606 | if (rrdev) |
607 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
91c00924 | 608 | generic_make_request(bi); |
977df362 N |
609 | } |
610 | if (rrdev) { | |
9a3e1101 N |
611 | if (s->syncing || s->expanding || s->expanded |
612 | || s->replacing) | |
977df362 N |
613 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
614 | ||
615 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
616 | ||
617 | rbi->bi_bdev = rrdev->bdev; | |
618 | pr_debug("%s: for %llu schedule op %ld on " | |
619 | "replacement disc %d\n", | |
620 | __func__, (unsigned long long)sh->sector, | |
621 | rbi->bi_rw, i); | |
622 | atomic_inc(&sh->count); | |
623 | rbi->bi_sector = sh->sector + rrdev->data_offset; | |
624 | rbi->bi_flags = 1 << BIO_UPTODATE; | |
625 | rbi->bi_idx = 0; | |
626 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
627 | rbi->bi_io_vec[0].bv_offset = 0; | |
628 | rbi->bi_size = STRIPE_SIZE; | |
629 | rbi->bi_next = NULL; | |
630 | generic_make_request(rbi); | |
631 | } | |
632 | if (!rdev && !rrdev) { | |
b062962e | 633 | if (rw & WRITE) |
91c00924 DW |
634 | set_bit(STRIPE_DEGRADED, &sh->state); |
635 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
636 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
637 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
638 | set_bit(STRIPE_HANDLE, &sh->state); | |
639 | } | |
640 | } | |
641 | } | |
642 | ||
643 | static struct dma_async_tx_descriptor * | |
644 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
645 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
646 | { | |
647 | struct bio_vec *bvl; | |
648 | struct page *bio_page; | |
649 | int i; | |
650 | int page_offset; | |
a08abd8c | 651 | struct async_submit_ctl submit; |
0403e382 | 652 | enum async_tx_flags flags = 0; |
91c00924 DW |
653 | |
654 | if (bio->bi_sector >= sector) | |
655 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
656 | else | |
657 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 658 | |
0403e382 DW |
659 | if (frombio) |
660 | flags |= ASYNC_TX_FENCE; | |
661 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
662 | ||
91c00924 | 663 | bio_for_each_segment(bvl, bio, i) { |
fcde9075 | 664 | int len = bvl->bv_len; |
91c00924 DW |
665 | int clen; |
666 | int b_offset = 0; | |
667 | ||
668 | if (page_offset < 0) { | |
669 | b_offset = -page_offset; | |
670 | page_offset += b_offset; | |
671 | len -= b_offset; | |
672 | } | |
673 | ||
674 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
675 | clen = STRIPE_SIZE - page_offset; | |
676 | else | |
677 | clen = len; | |
678 | ||
679 | if (clen > 0) { | |
fcde9075 NK |
680 | b_offset += bvl->bv_offset; |
681 | bio_page = bvl->bv_page; | |
91c00924 DW |
682 | if (frombio) |
683 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 684 | b_offset, clen, &submit); |
91c00924 DW |
685 | else |
686 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 687 | page_offset, clen, &submit); |
91c00924 | 688 | } |
a08abd8c DW |
689 | /* chain the operations */ |
690 | submit.depend_tx = tx; | |
691 | ||
91c00924 DW |
692 | if (clen < len) /* hit end of page */ |
693 | break; | |
694 | page_offset += len; | |
695 | } | |
696 | ||
697 | return tx; | |
698 | } | |
699 | ||
700 | static void ops_complete_biofill(void *stripe_head_ref) | |
701 | { | |
702 | struct stripe_head *sh = stripe_head_ref; | |
703 | struct bio *return_bi = NULL; | |
d1688a6d | 704 | struct r5conf *conf = sh->raid_conf; |
e4d84909 | 705 | int i; |
91c00924 | 706 | |
e46b272b | 707 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
708 | (unsigned long long)sh->sector); |
709 | ||
710 | /* clear completed biofills */ | |
83de75cc | 711 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
712 | for (i = sh->disks; i--; ) { |
713 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
714 | |
715 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
716 | /* and check if we need to reply to a read request, |
717 | * new R5_Wantfill requests are held off until | |
83de75cc | 718 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
719 | */ |
720 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 721 | struct bio *rbi, *rbi2; |
91c00924 | 722 | |
91c00924 DW |
723 | BUG_ON(!dev->read); |
724 | rbi = dev->read; | |
725 | dev->read = NULL; | |
726 | while (rbi && rbi->bi_sector < | |
727 | dev->sector + STRIPE_SECTORS) { | |
728 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 729 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
730 | rbi->bi_next = return_bi; |
731 | return_bi = rbi; | |
732 | } | |
91c00924 DW |
733 | rbi = rbi2; |
734 | } | |
735 | } | |
736 | } | |
83de75cc DW |
737 | spin_unlock_irq(&conf->device_lock); |
738 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
739 | |
740 | return_io(return_bi); | |
741 | ||
e4d84909 | 742 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
743 | release_stripe(sh); |
744 | } | |
745 | ||
746 | static void ops_run_biofill(struct stripe_head *sh) | |
747 | { | |
748 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 749 | struct r5conf *conf = sh->raid_conf; |
a08abd8c | 750 | struct async_submit_ctl submit; |
91c00924 DW |
751 | int i; |
752 | ||
e46b272b | 753 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
754 | (unsigned long long)sh->sector); |
755 | ||
756 | for (i = sh->disks; i--; ) { | |
757 | struct r5dev *dev = &sh->dev[i]; | |
758 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
759 | struct bio *rbi; | |
760 | spin_lock_irq(&conf->device_lock); | |
761 | dev->read = rbi = dev->toread; | |
762 | dev->toread = NULL; | |
763 | spin_unlock_irq(&conf->device_lock); | |
764 | while (rbi && rbi->bi_sector < | |
765 | dev->sector + STRIPE_SECTORS) { | |
766 | tx = async_copy_data(0, rbi, dev->page, | |
767 | dev->sector, tx); | |
768 | rbi = r5_next_bio(rbi, dev->sector); | |
769 | } | |
770 | } | |
771 | } | |
772 | ||
773 | atomic_inc(&sh->count); | |
a08abd8c DW |
774 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
775 | async_trigger_callback(&submit); | |
91c00924 DW |
776 | } |
777 | ||
4e7d2c0a | 778 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 779 | { |
4e7d2c0a | 780 | struct r5dev *tgt; |
91c00924 | 781 | |
4e7d2c0a DW |
782 | if (target < 0) |
783 | return; | |
91c00924 | 784 | |
4e7d2c0a | 785 | tgt = &sh->dev[target]; |
91c00924 DW |
786 | set_bit(R5_UPTODATE, &tgt->flags); |
787 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
788 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
789 | } |
790 | ||
ac6b53b6 | 791 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
792 | { |
793 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 794 | |
e46b272b | 795 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
796 | (unsigned long long)sh->sector); |
797 | ||
ac6b53b6 | 798 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 799 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 800 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 801 | |
ecc65c9b DW |
802 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
803 | if (sh->check_state == check_state_compute_run) | |
804 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
805 | set_bit(STRIPE_HANDLE, &sh->state); |
806 | release_stripe(sh); | |
807 | } | |
808 | ||
d6f38f31 DW |
809 | /* return a pointer to the address conversion region of the scribble buffer */ |
810 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
811 | struct raid5_percpu *percpu) | |
812 | { | |
813 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
814 | } | |
815 | ||
816 | static struct dma_async_tx_descriptor * | |
817 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 818 | { |
91c00924 | 819 | int disks = sh->disks; |
d6f38f31 | 820 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
821 | int target = sh->ops.target; |
822 | struct r5dev *tgt = &sh->dev[target]; | |
823 | struct page *xor_dest = tgt->page; | |
824 | int count = 0; | |
825 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 826 | struct async_submit_ctl submit; |
91c00924 DW |
827 | int i; |
828 | ||
829 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 830 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
831 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
832 | ||
833 | for (i = disks; i--; ) | |
834 | if (i != target) | |
835 | xor_srcs[count++] = sh->dev[i].page; | |
836 | ||
837 | atomic_inc(&sh->count); | |
838 | ||
0403e382 | 839 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 840 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 841 | if (unlikely(count == 1)) |
a08abd8c | 842 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 843 | else |
a08abd8c | 844 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 845 | |
91c00924 DW |
846 | return tx; |
847 | } | |
848 | ||
ac6b53b6 DW |
849 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
850 | * @srcs - (struct page *) array of size sh->disks | |
851 | * @sh - stripe_head to parse | |
852 | * | |
853 | * Populates srcs in proper layout order for the stripe and returns the | |
854 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
855 | * destination buffer is recorded in srcs[count] and the Q destination | |
856 | * is recorded in srcs[count+1]]. | |
857 | */ | |
858 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
859 | { | |
860 | int disks = sh->disks; | |
861 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
862 | int d0_idx = raid6_d0(sh); | |
863 | int count; | |
864 | int i; | |
865 | ||
866 | for (i = 0; i < disks; i++) | |
5dd33c9a | 867 | srcs[i] = NULL; |
ac6b53b6 DW |
868 | |
869 | count = 0; | |
870 | i = d0_idx; | |
871 | do { | |
872 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
873 | ||
874 | srcs[slot] = sh->dev[i].page; | |
875 | i = raid6_next_disk(i, disks); | |
876 | } while (i != d0_idx); | |
ac6b53b6 | 877 | |
e4424fee | 878 | return syndrome_disks; |
ac6b53b6 DW |
879 | } |
880 | ||
881 | static struct dma_async_tx_descriptor * | |
882 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
883 | { | |
884 | int disks = sh->disks; | |
885 | struct page **blocks = percpu->scribble; | |
886 | int target; | |
887 | int qd_idx = sh->qd_idx; | |
888 | struct dma_async_tx_descriptor *tx; | |
889 | struct async_submit_ctl submit; | |
890 | struct r5dev *tgt; | |
891 | struct page *dest; | |
892 | int i; | |
893 | int count; | |
894 | ||
895 | if (sh->ops.target < 0) | |
896 | target = sh->ops.target2; | |
897 | else if (sh->ops.target2 < 0) | |
898 | target = sh->ops.target; | |
91c00924 | 899 | else |
ac6b53b6 DW |
900 | /* we should only have one valid target */ |
901 | BUG(); | |
902 | BUG_ON(target < 0); | |
903 | pr_debug("%s: stripe %llu block: %d\n", | |
904 | __func__, (unsigned long long)sh->sector, target); | |
905 | ||
906 | tgt = &sh->dev[target]; | |
907 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
908 | dest = tgt->page; | |
909 | ||
910 | atomic_inc(&sh->count); | |
911 | ||
912 | if (target == qd_idx) { | |
913 | count = set_syndrome_sources(blocks, sh); | |
914 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
915 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
916 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
917 | ops_complete_compute, sh, | |
ac6b53b6 DW |
918 | to_addr_conv(sh, percpu)); |
919 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
920 | } else { | |
921 | /* Compute any data- or p-drive using XOR */ | |
922 | count = 0; | |
923 | for (i = disks; i-- ; ) { | |
924 | if (i == target || i == qd_idx) | |
925 | continue; | |
926 | blocks[count++] = sh->dev[i].page; | |
927 | } | |
928 | ||
0403e382 DW |
929 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
930 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
931 | to_addr_conv(sh, percpu)); |
932 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
933 | } | |
91c00924 | 934 | |
91c00924 DW |
935 | return tx; |
936 | } | |
937 | ||
ac6b53b6 DW |
938 | static struct dma_async_tx_descriptor * |
939 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
940 | { | |
941 | int i, count, disks = sh->disks; | |
942 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
943 | int d0_idx = raid6_d0(sh); | |
944 | int faila = -1, failb = -1; | |
945 | int target = sh->ops.target; | |
946 | int target2 = sh->ops.target2; | |
947 | struct r5dev *tgt = &sh->dev[target]; | |
948 | struct r5dev *tgt2 = &sh->dev[target2]; | |
949 | struct dma_async_tx_descriptor *tx; | |
950 | struct page **blocks = percpu->scribble; | |
951 | struct async_submit_ctl submit; | |
952 | ||
953 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
954 | __func__, (unsigned long long)sh->sector, target, target2); | |
955 | BUG_ON(target < 0 || target2 < 0); | |
956 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
957 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
958 | ||
6c910a78 | 959 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
960 | * slot number conversion for 'faila' and 'failb' |
961 | */ | |
962 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 963 | blocks[i] = NULL; |
ac6b53b6 DW |
964 | count = 0; |
965 | i = d0_idx; | |
966 | do { | |
967 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
968 | ||
969 | blocks[slot] = sh->dev[i].page; | |
970 | ||
971 | if (i == target) | |
972 | faila = slot; | |
973 | if (i == target2) | |
974 | failb = slot; | |
975 | i = raid6_next_disk(i, disks); | |
976 | } while (i != d0_idx); | |
ac6b53b6 DW |
977 | |
978 | BUG_ON(faila == failb); | |
979 | if (failb < faila) | |
980 | swap(faila, failb); | |
981 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
982 | __func__, (unsigned long long)sh->sector, faila, failb); | |
983 | ||
984 | atomic_inc(&sh->count); | |
985 | ||
986 | if (failb == syndrome_disks+1) { | |
987 | /* Q disk is one of the missing disks */ | |
988 | if (faila == syndrome_disks) { | |
989 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
990 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
991 | ops_complete_compute, sh, | |
992 | to_addr_conv(sh, percpu)); | |
e4424fee | 993 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
994 | STRIPE_SIZE, &submit); |
995 | } else { | |
996 | struct page *dest; | |
997 | int data_target; | |
998 | int qd_idx = sh->qd_idx; | |
999 | ||
1000 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1001 | if (target == qd_idx) | |
1002 | data_target = target2; | |
1003 | else | |
1004 | data_target = target; | |
1005 | ||
1006 | count = 0; | |
1007 | for (i = disks; i-- ; ) { | |
1008 | if (i == data_target || i == qd_idx) | |
1009 | continue; | |
1010 | blocks[count++] = sh->dev[i].page; | |
1011 | } | |
1012 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1013 | init_async_submit(&submit, |
1014 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1015 | NULL, NULL, NULL, | |
1016 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1017 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1018 | &submit); | |
1019 | ||
1020 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
1021 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1022 | ops_complete_compute, sh, | |
1023 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1024 | return async_gen_syndrome(blocks, 0, count+2, |
1025 | STRIPE_SIZE, &submit); | |
1026 | } | |
ac6b53b6 | 1027 | } else { |
6c910a78 DW |
1028 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1029 | ops_complete_compute, sh, | |
1030 | to_addr_conv(sh, percpu)); | |
1031 | if (failb == syndrome_disks) { | |
1032 | /* We're missing D+P. */ | |
1033 | return async_raid6_datap_recov(syndrome_disks+2, | |
1034 | STRIPE_SIZE, faila, | |
1035 | blocks, &submit); | |
1036 | } else { | |
1037 | /* We're missing D+D. */ | |
1038 | return async_raid6_2data_recov(syndrome_disks+2, | |
1039 | STRIPE_SIZE, faila, failb, | |
1040 | blocks, &submit); | |
1041 | } | |
ac6b53b6 DW |
1042 | } |
1043 | } | |
1044 | ||
1045 | ||
91c00924 DW |
1046 | static void ops_complete_prexor(void *stripe_head_ref) |
1047 | { | |
1048 | struct stripe_head *sh = stripe_head_ref; | |
1049 | ||
e46b272b | 1050 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1051 | (unsigned long long)sh->sector); |
91c00924 DW |
1052 | } |
1053 | ||
1054 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
1055 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
1056 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1057 | { |
91c00924 | 1058 | int disks = sh->disks; |
d6f38f31 | 1059 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1060 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1061 | struct async_submit_ctl submit; |
91c00924 DW |
1062 | |
1063 | /* existing parity data subtracted */ | |
1064 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1065 | ||
e46b272b | 1066 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1067 | (unsigned long long)sh->sector); |
1068 | ||
1069 | for (i = disks; i--; ) { | |
1070 | struct r5dev *dev = &sh->dev[i]; | |
1071 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1072 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1073 | xor_srcs[count++] = dev->page; |
1074 | } | |
1075 | ||
0403e382 | 1076 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 1077 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 1078 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1079 | |
1080 | return tx; | |
1081 | } | |
1082 | ||
1083 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1084 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1085 | { |
1086 | int disks = sh->disks; | |
d8ee0728 | 1087 | int i; |
91c00924 | 1088 | |
e46b272b | 1089 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1090 | (unsigned long long)sh->sector); |
1091 | ||
1092 | for (i = disks; i--; ) { | |
1093 | struct r5dev *dev = &sh->dev[i]; | |
1094 | struct bio *chosen; | |
91c00924 | 1095 | |
d8ee0728 | 1096 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1097 | struct bio *wbi; |
1098 | ||
cbe47ec5 | 1099 | spin_lock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1100 | chosen = dev->towrite; |
1101 | dev->towrite = NULL; | |
1102 | BUG_ON(dev->written); | |
1103 | wbi = dev->written = chosen; | |
cbe47ec5 | 1104 | spin_unlock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1105 | |
1106 | while (wbi && wbi->bi_sector < | |
1107 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1108 | if (wbi->bi_rw & REQ_FUA) |
1109 | set_bit(R5_WantFUA, &dev->flags); | |
91c00924 DW |
1110 | tx = async_copy_data(1, wbi, dev->page, |
1111 | dev->sector, tx); | |
1112 | wbi = r5_next_bio(wbi, dev->sector); | |
1113 | } | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | return tx; | |
1118 | } | |
1119 | ||
ac6b53b6 | 1120 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1121 | { |
1122 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1123 | int disks = sh->disks; |
1124 | int pd_idx = sh->pd_idx; | |
1125 | int qd_idx = sh->qd_idx; | |
1126 | int i; | |
e9c7469b | 1127 | bool fua = false; |
91c00924 | 1128 | |
e46b272b | 1129 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1130 | (unsigned long long)sh->sector); |
1131 | ||
e9c7469b TH |
1132 | for (i = disks; i--; ) |
1133 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); | |
1134 | ||
91c00924 DW |
1135 | for (i = disks; i--; ) { |
1136 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1137 | |
e9c7469b | 1138 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1139 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1140 | if (fua) |
1141 | set_bit(R5_WantFUA, &dev->flags); | |
1142 | } | |
91c00924 DW |
1143 | } |
1144 | ||
d8ee0728 DW |
1145 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1146 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1147 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1148 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1149 | else { | |
1150 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1151 | sh->reconstruct_state = reconstruct_state_result; | |
1152 | } | |
91c00924 DW |
1153 | |
1154 | set_bit(STRIPE_HANDLE, &sh->state); | |
1155 | release_stripe(sh); | |
1156 | } | |
1157 | ||
1158 | static void | |
ac6b53b6 DW |
1159 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1160 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1161 | { |
91c00924 | 1162 | int disks = sh->disks; |
d6f38f31 | 1163 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1164 | struct async_submit_ctl submit; |
91c00924 DW |
1165 | int count = 0, pd_idx = sh->pd_idx, i; |
1166 | struct page *xor_dest; | |
d8ee0728 | 1167 | int prexor = 0; |
91c00924 | 1168 | unsigned long flags; |
91c00924 | 1169 | |
e46b272b | 1170 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1171 | (unsigned long long)sh->sector); |
1172 | ||
1173 | /* check if prexor is active which means only process blocks | |
1174 | * that are part of a read-modify-write (written) | |
1175 | */ | |
d8ee0728 DW |
1176 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1177 | prexor = 1; | |
91c00924 DW |
1178 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1179 | for (i = disks; i--; ) { | |
1180 | struct r5dev *dev = &sh->dev[i]; | |
1181 | if (dev->written) | |
1182 | xor_srcs[count++] = dev->page; | |
1183 | } | |
1184 | } else { | |
1185 | xor_dest = sh->dev[pd_idx].page; | |
1186 | for (i = disks; i--; ) { | |
1187 | struct r5dev *dev = &sh->dev[i]; | |
1188 | if (i != pd_idx) | |
1189 | xor_srcs[count++] = dev->page; | |
1190 | } | |
1191 | } | |
1192 | ||
91c00924 DW |
1193 | /* 1/ if we prexor'd then the dest is reused as a source |
1194 | * 2/ if we did not prexor then we are redoing the parity | |
1195 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1196 | * for the synchronous xor case | |
1197 | */ | |
88ba2aa5 | 1198 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1199 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1200 | ||
1201 | atomic_inc(&sh->count); | |
1202 | ||
ac6b53b6 | 1203 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1204 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1205 | if (unlikely(count == 1)) |
1206 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1207 | else | |
1208 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1209 | } |
1210 | ||
ac6b53b6 DW |
1211 | static void |
1212 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1213 | struct dma_async_tx_descriptor *tx) | |
1214 | { | |
1215 | struct async_submit_ctl submit; | |
1216 | struct page **blocks = percpu->scribble; | |
1217 | int count; | |
1218 | ||
1219 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1220 | ||
1221 | count = set_syndrome_sources(blocks, sh); | |
1222 | ||
1223 | atomic_inc(&sh->count); | |
1224 | ||
1225 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1226 | sh, to_addr_conv(sh, percpu)); | |
1227 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1228 | } |
1229 | ||
1230 | static void ops_complete_check(void *stripe_head_ref) | |
1231 | { | |
1232 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1233 | |
e46b272b | 1234 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1235 | (unsigned long long)sh->sector); |
1236 | ||
ecc65c9b | 1237 | sh->check_state = check_state_check_result; |
91c00924 DW |
1238 | set_bit(STRIPE_HANDLE, &sh->state); |
1239 | release_stripe(sh); | |
1240 | } | |
1241 | ||
ac6b53b6 | 1242 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1243 | { |
91c00924 | 1244 | int disks = sh->disks; |
ac6b53b6 DW |
1245 | int pd_idx = sh->pd_idx; |
1246 | int qd_idx = sh->qd_idx; | |
1247 | struct page *xor_dest; | |
d6f38f31 | 1248 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1249 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1250 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1251 | int count; |
1252 | int i; | |
91c00924 | 1253 | |
e46b272b | 1254 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1255 | (unsigned long long)sh->sector); |
1256 | ||
ac6b53b6 DW |
1257 | count = 0; |
1258 | xor_dest = sh->dev[pd_idx].page; | |
1259 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1260 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1261 | if (i == pd_idx || i == qd_idx) |
1262 | continue; | |
1263 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1264 | } |
1265 | ||
d6f38f31 DW |
1266 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1267 | to_addr_conv(sh, percpu)); | |
099f53cb | 1268 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1269 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1270 | |
91c00924 | 1271 | atomic_inc(&sh->count); |
a08abd8c DW |
1272 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1273 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1274 | } |
1275 | ||
ac6b53b6 DW |
1276 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1277 | { | |
1278 | struct page **srcs = percpu->scribble; | |
1279 | struct async_submit_ctl submit; | |
1280 | int count; | |
1281 | ||
1282 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1283 | (unsigned long long)sh->sector, checkp); | |
1284 | ||
1285 | count = set_syndrome_sources(srcs, sh); | |
1286 | if (!checkp) | |
1287 | srcs[count] = NULL; | |
91c00924 | 1288 | |
91c00924 | 1289 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1290 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1291 | sh, to_addr_conv(sh, percpu)); | |
1292 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1293 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1294 | } |
1295 | ||
417b8d4a | 1296 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1297 | { |
1298 | int overlap_clear = 0, i, disks = sh->disks; | |
1299 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1300 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1301 | int level = conf->level; |
d6f38f31 DW |
1302 | struct raid5_percpu *percpu; |
1303 | unsigned long cpu; | |
91c00924 | 1304 | |
d6f38f31 DW |
1305 | cpu = get_cpu(); |
1306 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1307 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1308 | ops_run_biofill(sh); |
1309 | overlap_clear++; | |
1310 | } | |
1311 | ||
7b3a871e | 1312 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1313 | if (level < 6) |
1314 | tx = ops_run_compute5(sh, percpu); | |
1315 | else { | |
1316 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1317 | tx = ops_run_compute6_1(sh, percpu); | |
1318 | else | |
1319 | tx = ops_run_compute6_2(sh, percpu); | |
1320 | } | |
1321 | /* terminate the chain if reconstruct is not set to be run */ | |
1322 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1323 | async_tx_ack(tx); |
1324 | } | |
91c00924 | 1325 | |
600aa109 | 1326 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1327 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1328 | |
600aa109 | 1329 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1330 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1331 | overlap_clear++; |
1332 | } | |
1333 | ||
ac6b53b6 DW |
1334 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1335 | if (level < 6) | |
1336 | ops_run_reconstruct5(sh, percpu, tx); | |
1337 | else | |
1338 | ops_run_reconstruct6(sh, percpu, tx); | |
1339 | } | |
91c00924 | 1340 | |
ac6b53b6 DW |
1341 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1342 | if (sh->check_state == check_state_run) | |
1343 | ops_run_check_p(sh, percpu); | |
1344 | else if (sh->check_state == check_state_run_q) | |
1345 | ops_run_check_pq(sh, percpu, 0); | |
1346 | else if (sh->check_state == check_state_run_pq) | |
1347 | ops_run_check_pq(sh, percpu, 1); | |
1348 | else | |
1349 | BUG(); | |
1350 | } | |
91c00924 | 1351 | |
91c00924 DW |
1352 | if (overlap_clear) |
1353 | for (i = disks; i--; ) { | |
1354 | struct r5dev *dev = &sh->dev[i]; | |
1355 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1356 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1357 | } | |
d6f38f31 | 1358 | put_cpu(); |
91c00924 DW |
1359 | } |
1360 | ||
417b8d4a DW |
1361 | #ifdef CONFIG_MULTICORE_RAID456 |
1362 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1363 | { | |
1364 | struct stripe_head *sh = param; | |
1365 | unsigned long ops_request = sh->ops.request; | |
1366 | ||
1367 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1368 | wake_up(&sh->ops.wait_for_ops); | |
1369 | ||
1370 | __raid_run_ops(sh, ops_request); | |
1371 | release_stripe(sh); | |
1372 | } | |
1373 | ||
1374 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1375 | { | |
1376 | /* since handle_stripe can be called outside of raid5d context | |
1377 | * we need to ensure sh->ops.request is de-staged before another | |
1378 | * request arrives | |
1379 | */ | |
1380 | wait_event(sh->ops.wait_for_ops, | |
1381 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1382 | sh->ops.request = ops_request; | |
1383 | ||
1384 | atomic_inc(&sh->count); | |
1385 | async_schedule(async_run_ops, sh); | |
1386 | } | |
1387 | #else | |
1388 | #define raid_run_ops __raid_run_ops | |
1389 | #endif | |
1390 | ||
d1688a6d | 1391 | static int grow_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1392 | { |
1393 | struct stripe_head *sh; | |
6ce32846 | 1394 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1395 | if (!sh) |
1396 | return 0; | |
6ce32846 | 1397 | |
3f294f4f | 1398 | sh->raid_conf = conf; |
417b8d4a DW |
1399 | #ifdef CONFIG_MULTICORE_RAID456 |
1400 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1401 | #endif | |
3f294f4f | 1402 | |
e4e11e38 N |
1403 | if (grow_buffers(sh)) { |
1404 | shrink_buffers(sh); | |
3f294f4f N |
1405 | kmem_cache_free(conf->slab_cache, sh); |
1406 | return 0; | |
1407 | } | |
1408 | /* we just created an active stripe so... */ | |
1409 | atomic_set(&sh->count, 1); | |
1410 | atomic_inc(&conf->active_stripes); | |
1411 | INIT_LIST_HEAD(&sh->lru); | |
1412 | release_stripe(sh); | |
1413 | return 1; | |
1414 | } | |
1415 | ||
d1688a6d | 1416 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 1417 | { |
e18b890b | 1418 | struct kmem_cache *sc; |
5e5e3e78 | 1419 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1420 | |
f4be6b43 N |
1421 | if (conf->mddev->gendisk) |
1422 | sprintf(conf->cache_name[0], | |
1423 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1424 | else | |
1425 | sprintf(conf->cache_name[0], | |
1426 | "raid%d-%p", conf->level, conf->mddev); | |
1427 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1428 | ||
ad01c9e3 N |
1429 | conf->active_name = 0; |
1430 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1431 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1432 | 0, 0, NULL); |
1da177e4 LT |
1433 | if (!sc) |
1434 | return 1; | |
1435 | conf->slab_cache = sc; | |
ad01c9e3 | 1436 | conf->pool_size = devs; |
16a53ecc | 1437 | while (num--) |
3f294f4f | 1438 | if (!grow_one_stripe(conf)) |
1da177e4 | 1439 | return 1; |
1da177e4 LT |
1440 | return 0; |
1441 | } | |
29269553 | 1442 | |
d6f38f31 DW |
1443 | /** |
1444 | * scribble_len - return the required size of the scribble region | |
1445 | * @num - total number of disks in the array | |
1446 | * | |
1447 | * The size must be enough to contain: | |
1448 | * 1/ a struct page pointer for each device in the array +2 | |
1449 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1450 | * (dma_map_page()) or page (page_address()) address. | |
1451 | * | |
1452 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1453 | * calculate over all devices (not just the data blocks), using zeros in place | |
1454 | * of the P and Q blocks. | |
1455 | */ | |
1456 | static size_t scribble_len(int num) | |
1457 | { | |
1458 | size_t len; | |
1459 | ||
1460 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1461 | ||
1462 | return len; | |
1463 | } | |
1464 | ||
d1688a6d | 1465 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
1466 | { |
1467 | /* Make all the stripes able to hold 'newsize' devices. | |
1468 | * New slots in each stripe get 'page' set to a new page. | |
1469 | * | |
1470 | * This happens in stages: | |
1471 | * 1/ create a new kmem_cache and allocate the required number of | |
1472 | * stripe_heads. | |
1473 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1474 | * to the new stripe_heads. This will have the side effect of | |
1475 | * freezing the array as once all stripe_heads have been collected, | |
1476 | * no IO will be possible. Old stripe heads are freed once their | |
1477 | * pages have been transferred over, and the old kmem_cache is | |
1478 | * freed when all stripes are done. | |
1479 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1480 | * we simple return a failre status - no need to clean anything up. | |
1481 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1482 | * If this fails, we don't bother trying the shrink the | |
1483 | * stripe_heads down again, we just leave them as they are. | |
1484 | * As each stripe_head is processed the new one is released into | |
1485 | * active service. | |
1486 | * | |
1487 | * Once step2 is started, we cannot afford to wait for a write, | |
1488 | * so we use GFP_NOIO allocations. | |
1489 | */ | |
1490 | struct stripe_head *osh, *nsh; | |
1491 | LIST_HEAD(newstripes); | |
1492 | struct disk_info *ndisks; | |
d6f38f31 | 1493 | unsigned long cpu; |
b5470dc5 | 1494 | int err; |
e18b890b | 1495 | struct kmem_cache *sc; |
ad01c9e3 N |
1496 | int i; |
1497 | ||
1498 | if (newsize <= conf->pool_size) | |
1499 | return 0; /* never bother to shrink */ | |
1500 | ||
b5470dc5 DW |
1501 | err = md_allow_write(conf->mddev); |
1502 | if (err) | |
1503 | return err; | |
2a2275d6 | 1504 | |
ad01c9e3 N |
1505 | /* Step 1 */ |
1506 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1507 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1508 | 0, 0, NULL); |
ad01c9e3 N |
1509 | if (!sc) |
1510 | return -ENOMEM; | |
1511 | ||
1512 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1513 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1514 | if (!nsh) |
1515 | break; | |
1516 | ||
ad01c9e3 | 1517 | nsh->raid_conf = conf; |
417b8d4a DW |
1518 | #ifdef CONFIG_MULTICORE_RAID456 |
1519 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1520 | #endif | |
ad01c9e3 N |
1521 | |
1522 | list_add(&nsh->lru, &newstripes); | |
1523 | } | |
1524 | if (i) { | |
1525 | /* didn't get enough, give up */ | |
1526 | while (!list_empty(&newstripes)) { | |
1527 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1528 | list_del(&nsh->lru); | |
1529 | kmem_cache_free(sc, nsh); | |
1530 | } | |
1531 | kmem_cache_destroy(sc); | |
1532 | return -ENOMEM; | |
1533 | } | |
1534 | /* Step 2 - Must use GFP_NOIO now. | |
1535 | * OK, we have enough stripes, start collecting inactive | |
1536 | * stripes and copying them over | |
1537 | */ | |
1538 | list_for_each_entry(nsh, &newstripes, lru) { | |
1539 | spin_lock_irq(&conf->device_lock); | |
1540 | wait_event_lock_irq(conf->wait_for_stripe, | |
1541 | !list_empty(&conf->inactive_list), | |
1542 | conf->device_lock, | |
482c0834 | 1543 | ); |
ad01c9e3 N |
1544 | osh = get_free_stripe(conf); |
1545 | spin_unlock_irq(&conf->device_lock); | |
1546 | atomic_set(&nsh->count, 1); | |
1547 | for(i=0; i<conf->pool_size; i++) | |
1548 | nsh->dev[i].page = osh->dev[i].page; | |
1549 | for( ; i<newsize; i++) | |
1550 | nsh->dev[i].page = NULL; | |
1551 | kmem_cache_free(conf->slab_cache, osh); | |
1552 | } | |
1553 | kmem_cache_destroy(conf->slab_cache); | |
1554 | ||
1555 | /* Step 3. | |
1556 | * At this point, we are holding all the stripes so the array | |
1557 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1558 | * conf->disks and the scribble region |
ad01c9e3 N |
1559 | */ |
1560 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1561 | if (ndisks) { | |
1562 | for (i=0; i<conf->raid_disks; i++) | |
1563 | ndisks[i] = conf->disks[i]; | |
1564 | kfree(conf->disks); | |
1565 | conf->disks = ndisks; | |
1566 | } else | |
1567 | err = -ENOMEM; | |
1568 | ||
d6f38f31 DW |
1569 | get_online_cpus(); |
1570 | conf->scribble_len = scribble_len(newsize); | |
1571 | for_each_present_cpu(cpu) { | |
1572 | struct raid5_percpu *percpu; | |
1573 | void *scribble; | |
1574 | ||
1575 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1576 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1577 | ||
1578 | if (scribble) { | |
1579 | kfree(percpu->scribble); | |
1580 | percpu->scribble = scribble; | |
1581 | } else { | |
1582 | err = -ENOMEM; | |
1583 | break; | |
1584 | } | |
1585 | } | |
1586 | put_online_cpus(); | |
1587 | ||
ad01c9e3 N |
1588 | /* Step 4, return new stripes to service */ |
1589 | while(!list_empty(&newstripes)) { | |
1590 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1591 | list_del_init(&nsh->lru); | |
d6f38f31 | 1592 | |
ad01c9e3 N |
1593 | for (i=conf->raid_disks; i < newsize; i++) |
1594 | if (nsh->dev[i].page == NULL) { | |
1595 | struct page *p = alloc_page(GFP_NOIO); | |
1596 | nsh->dev[i].page = p; | |
1597 | if (!p) | |
1598 | err = -ENOMEM; | |
1599 | } | |
1600 | release_stripe(nsh); | |
1601 | } | |
1602 | /* critical section pass, GFP_NOIO no longer needed */ | |
1603 | ||
1604 | conf->slab_cache = sc; | |
1605 | conf->active_name = 1-conf->active_name; | |
1606 | conf->pool_size = newsize; | |
1607 | return err; | |
1608 | } | |
1da177e4 | 1609 | |
d1688a6d | 1610 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1611 | { |
1612 | struct stripe_head *sh; | |
1613 | ||
3f294f4f N |
1614 | spin_lock_irq(&conf->device_lock); |
1615 | sh = get_free_stripe(conf); | |
1616 | spin_unlock_irq(&conf->device_lock); | |
1617 | if (!sh) | |
1618 | return 0; | |
78bafebd | 1619 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1620 | shrink_buffers(sh); |
3f294f4f N |
1621 | kmem_cache_free(conf->slab_cache, sh); |
1622 | atomic_dec(&conf->active_stripes); | |
1623 | return 1; | |
1624 | } | |
1625 | ||
d1688a6d | 1626 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f N |
1627 | { |
1628 | while (drop_one_stripe(conf)) | |
1629 | ; | |
1630 | ||
29fc7e3e N |
1631 | if (conf->slab_cache) |
1632 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1633 | conf->slab_cache = NULL; |
1634 | } | |
1635 | ||
6712ecf8 | 1636 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1637 | { |
99c0fb5f | 1638 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1639 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1640 | int disks = sh->disks, i; |
1da177e4 | 1641 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 1642 | char b[BDEVNAME_SIZE]; |
3cb03002 | 1643 | struct md_rdev *rdev; |
1da177e4 | 1644 | |
1da177e4 LT |
1645 | |
1646 | for (i=0 ; i<disks; i++) | |
1647 | if (bi == &sh->dev[i].req) | |
1648 | break; | |
1649 | ||
45b4233c DW |
1650 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1651 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1652 | uptodate); |
1653 | if (i == disks) { | |
1654 | BUG(); | |
6712ecf8 | 1655 | return; |
1da177e4 | 1656 | } |
14a75d3e N |
1657 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
1658 | rdev = conf->disks[i].replacement; | |
1659 | else | |
1660 | rdev = conf->disks[i].rdev; | |
1da177e4 LT |
1661 | |
1662 | if (uptodate) { | |
1da177e4 | 1663 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1664 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
1665 | /* Note that this cannot happen on a |
1666 | * replacement device. We just fail those on | |
1667 | * any error | |
1668 | */ | |
8bda470e CD |
1669 | printk_ratelimited( |
1670 | KERN_INFO | |
1671 | "md/raid:%s: read error corrected" | |
1672 | " (%lu sectors at %llu on %s)\n", | |
1673 | mdname(conf->mddev), STRIPE_SECTORS, | |
1674 | (unsigned long long)(sh->sector | |
1675 | + rdev->data_offset), | |
1676 | bdevname(rdev->bdev, b)); | |
ddd5115f | 1677 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
1678 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1679 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1680 | } | |
14a75d3e N |
1681 | if (atomic_read(&rdev->read_errors)) |
1682 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 1683 | } else { |
14a75d3e | 1684 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 1685 | int retry = 0; |
d6950432 | 1686 | |
1da177e4 | 1687 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1688 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
1689 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
1690 | printk_ratelimited( | |
1691 | KERN_WARNING | |
1692 | "md/raid:%s: read error on replacement device " | |
1693 | "(sector %llu on %s).\n", | |
1694 | mdname(conf->mddev), | |
1695 | (unsigned long long)(sh->sector | |
1696 | + rdev->data_offset), | |
1697 | bdn); | |
1698 | else if (conf->mddev->degraded >= conf->max_degraded) | |
8bda470e CD |
1699 | printk_ratelimited( |
1700 | KERN_WARNING | |
1701 | "md/raid:%s: read error not correctable " | |
1702 | "(sector %llu on %s).\n", | |
1703 | mdname(conf->mddev), | |
1704 | (unsigned long long)(sh->sector | |
1705 | + rdev->data_offset), | |
1706 | bdn); | |
ba22dcbf | 1707 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1708 | /* Oh, no!!! */ |
8bda470e CD |
1709 | printk_ratelimited( |
1710 | KERN_WARNING | |
1711 | "md/raid:%s: read error NOT corrected!! " | |
1712 | "(sector %llu on %s).\n", | |
1713 | mdname(conf->mddev), | |
1714 | (unsigned long long)(sh->sector | |
1715 | + rdev->data_offset), | |
1716 | bdn); | |
d6950432 | 1717 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1718 | > conf->max_nr_stripes) |
14f8d26b | 1719 | printk(KERN_WARNING |
0c55e022 | 1720 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1721 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1722 | else |
1723 | retry = 1; | |
1724 | if (retry) | |
1725 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1726 | else { | |
4e5314b5 N |
1727 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1728 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1729 | md_error(conf->mddev, rdev); |
ba22dcbf | 1730 | } |
1da177e4 | 1731 | } |
14a75d3e | 1732 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
1733 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1734 | set_bit(STRIPE_HANDLE, &sh->state); | |
1735 | release_stripe(sh); | |
1da177e4 LT |
1736 | } |
1737 | ||
d710e138 | 1738 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1739 | { |
99c0fb5f | 1740 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1741 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1742 | int disks = sh->disks, i; |
977df362 | 1743 | struct md_rdev *uninitialized_var(rdev); |
1da177e4 | 1744 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
1745 | sector_t first_bad; |
1746 | int bad_sectors; | |
977df362 | 1747 | int replacement = 0; |
1da177e4 | 1748 | |
977df362 N |
1749 | for (i = 0 ; i < disks; i++) { |
1750 | if (bi == &sh->dev[i].req) { | |
1751 | rdev = conf->disks[i].rdev; | |
1da177e4 | 1752 | break; |
977df362 N |
1753 | } |
1754 | if (bi == &sh->dev[i].rreq) { | |
1755 | rdev = conf->disks[i].replacement; | |
1756 | replacement = 1; | |
1757 | break; | |
1758 | } | |
1759 | } | |
45b4233c | 1760 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1761 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1762 | uptodate); | |
1763 | if (i == disks) { | |
1764 | BUG(); | |
6712ecf8 | 1765 | return; |
1da177e4 LT |
1766 | } |
1767 | ||
977df362 N |
1768 | if (replacement) { |
1769 | if (!uptodate) | |
1770 | md_error(conf->mddev, rdev); | |
1771 | else if (is_badblock(rdev, sh->sector, | |
1772 | STRIPE_SECTORS, | |
1773 | &first_bad, &bad_sectors)) | |
1774 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
1775 | } else { | |
1776 | if (!uptodate) { | |
1777 | set_bit(WriteErrorSeen, &rdev->flags); | |
1778 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
1779 | } else if (is_badblock(rdev, sh->sector, | |
1780 | STRIPE_SECTORS, | |
1781 | &first_bad, &bad_sectors)) | |
1782 | set_bit(R5_MadeGood, &sh->dev[i].flags); | |
1783 | } | |
1784 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 1785 | |
977df362 N |
1786 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
1787 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 1788 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 1789 | release_stripe(sh); |
1da177e4 LT |
1790 | } |
1791 | ||
784052ec | 1792 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1793 | |
784052ec | 1794 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1795 | { |
1796 | struct r5dev *dev = &sh->dev[i]; | |
1797 | ||
1798 | bio_init(&dev->req); | |
1799 | dev->req.bi_io_vec = &dev->vec; | |
1800 | dev->req.bi_vcnt++; | |
1801 | dev->req.bi_max_vecs++; | |
1da177e4 | 1802 | dev->req.bi_private = sh; |
995c4275 | 1803 | dev->vec.bv_page = dev->page; |
1da177e4 | 1804 | |
977df362 N |
1805 | bio_init(&dev->rreq); |
1806 | dev->rreq.bi_io_vec = &dev->rvec; | |
1807 | dev->rreq.bi_vcnt++; | |
1808 | dev->rreq.bi_max_vecs++; | |
1809 | dev->rreq.bi_private = sh; | |
1810 | dev->rvec.bv_page = dev->page; | |
1811 | ||
1da177e4 | 1812 | dev->flags = 0; |
784052ec | 1813 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1814 | } |
1815 | ||
fd01b88c | 1816 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1817 | { |
1818 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 1819 | struct r5conf *conf = mddev->private; |
908f4fbd | 1820 | unsigned long flags; |
0c55e022 | 1821 | pr_debug("raid456: error called\n"); |
1da177e4 | 1822 | |
908f4fbd N |
1823 | spin_lock_irqsave(&conf->device_lock, flags); |
1824 | clear_bit(In_sync, &rdev->flags); | |
1825 | mddev->degraded = calc_degraded(conf); | |
1826 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1827 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1828 | ||
de393cde | 1829 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
1830 | set_bit(Faulty, &rdev->flags); |
1831 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1832 | printk(KERN_ALERT | |
1833 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1834 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1835 | mdname(mddev), | |
1836 | bdevname(rdev->bdev, b), | |
1837 | mdname(mddev), | |
1838 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1839 | } |
1da177e4 LT |
1840 | |
1841 | /* | |
1842 | * Input: a 'big' sector number, | |
1843 | * Output: index of the data and parity disk, and the sector # in them. | |
1844 | */ | |
d1688a6d | 1845 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
1846 | int previous, int *dd_idx, |
1847 | struct stripe_head *sh) | |
1da177e4 | 1848 | { |
6e3b96ed | 1849 | sector_t stripe, stripe2; |
35f2a591 | 1850 | sector_t chunk_number; |
1da177e4 | 1851 | unsigned int chunk_offset; |
911d4ee8 | 1852 | int pd_idx, qd_idx; |
67cc2b81 | 1853 | int ddf_layout = 0; |
1da177e4 | 1854 | sector_t new_sector; |
e183eaed N |
1855 | int algorithm = previous ? conf->prev_algo |
1856 | : conf->algorithm; | |
09c9e5fa AN |
1857 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1858 | : conf->chunk_sectors; | |
112bf897 N |
1859 | int raid_disks = previous ? conf->previous_raid_disks |
1860 | : conf->raid_disks; | |
1861 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1862 | |
1863 | /* First compute the information on this sector */ | |
1864 | ||
1865 | /* | |
1866 | * Compute the chunk number and the sector offset inside the chunk | |
1867 | */ | |
1868 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1869 | chunk_number = r_sector; | |
1da177e4 LT |
1870 | |
1871 | /* | |
1872 | * Compute the stripe number | |
1873 | */ | |
35f2a591 N |
1874 | stripe = chunk_number; |
1875 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1876 | stripe2 = stripe; |
1da177e4 LT |
1877 | /* |
1878 | * Select the parity disk based on the user selected algorithm. | |
1879 | */ | |
84789554 | 1880 | pd_idx = qd_idx = -1; |
16a53ecc N |
1881 | switch(conf->level) { |
1882 | case 4: | |
911d4ee8 | 1883 | pd_idx = data_disks; |
16a53ecc N |
1884 | break; |
1885 | case 5: | |
e183eaed | 1886 | switch (algorithm) { |
1da177e4 | 1887 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1888 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1889 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1890 | (*dd_idx)++; |
1891 | break; | |
1892 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1893 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1894 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1895 | (*dd_idx)++; |
1896 | break; | |
1897 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1898 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1899 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1900 | break; |
1901 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1902 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1903 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 1904 | break; |
99c0fb5f N |
1905 | case ALGORITHM_PARITY_0: |
1906 | pd_idx = 0; | |
1907 | (*dd_idx)++; | |
1908 | break; | |
1909 | case ALGORITHM_PARITY_N: | |
1910 | pd_idx = data_disks; | |
1911 | break; | |
1da177e4 | 1912 | default: |
99c0fb5f | 1913 | BUG(); |
16a53ecc N |
1914 | } |
1915 | break; | |
1916 | case 6: | |
1917 | ||
e183eaed | 1918 | switch (algorithm) { |
16a53ecc | 1919 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1920 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1921 | qd_idx = pd_idx + 1; |
1922 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1923 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1924 | qd_idx = 0; |
1925 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1926 | (*dd_idx) += 2; /* D D P Q D */ |
1927 | break; | |
1928 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1929 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1930 | qd_idx = pd_idx + 1; |
1931 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1932 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1933 | qd_idx = 0; |
1934 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1935 | (*dd_idx) += 2; /* D D P Q D */ |
1936 | break; | |
1937 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1938 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1939 | qd_idx = (pd_idx + 1) % raid_disks; |
1940 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1941 | break; |
1942 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1943 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1944 | qd_idx = (pd_idx + 1) % raid_disks; |
1945 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1946 | break; |
99c0fb5f N |
1947 | |
1948 | case ALGORITHM_PARITY_0: | |
1949 | pd_idx = 0; | |
1950 | qd_idx = 1; | |
1951 | (*dd_idx) += 2; | |
1952 | break; | |
1953 | case ALGORITHM_PARITY_N: | |
1954 | pd_idx = data_disks; | |
1955 | qd_idx = data_disks + 1; | |
1956 | break; | |
1957 | ||
1958 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1959 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1960 | * of blocks for computing Q is different. | |
1961 | */ | |
6e3b96ed | 1962 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
1963 | qd_idx = pd_idx + 1; |
1964 | if (pd_idx == raid_disks-1) { | |
1965 | (*dd_idx)++; /* Q D D D P */ | |
1966 | qd_idx = 0; | |
1967 | } else if (*dd_idx >= pd_idx) | |
1968 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1969 | ddf_layout = 1; |
99c0fb5f N |
1970 | break; |
1971 | ||
1972 | case ALGORITHM_ROTATING_N_RESTART: | |
1973 | /* Same a left_asymmetric, by first stripe is | |
1974 | * D D D P Q rather than | |
1975 | * Q D D D P | |
1976 | */ | |
6e3b96ed N |
1977 | stripe2 += 1; |
1978 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
1979 | qd_idx = pd_idx + 1; |
1980 | if (pd_idx == raid_disks-1) { | |
1981 | (*dd_idx)++; /* Q D D D P */ | |
1982 | qd_idx = 0; | |
1983 | } else if (*dd_idx >= pd_idx) | |
1984 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1985 | ddf_layout = 1; |
99c0fb5f N |
1986 | break; |
1987 | ||
1988 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1989 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 1990 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
1991 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
1992 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1993 | ddf_layout = 1; |
99c0fb5f N |
1994 | break; |
1995 | ||
1996 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1997 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 1998 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1999 | if (*dd_idx >= pd_idx) |
2000 | (*dd_idx)++; | |
2001 | qd_idx = raid_disks - 1; | |
2002 | break; | |
2003 | ||
2004 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2005 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2006 | if (*dd_idx >= pd_idx) |
2007 | (*dd_idx)++; | |
2008 | qd_idx = raid_disks - 1; | |
2009 | break; | |
2010 | ||
2011 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2012 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2013 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2014 | qd_idx = raid_disks - 1; | |
2015 | break; | |
2016 | ||
2017 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2018 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2019 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2020 | qd_idx = raid_disks - 1; | |
2021 | break; | |
2022 | ||
2023 | case ALGORITHM_PARITY_0_6: | |
2024 | pd_idx = 0; | |
2025 | (*dd_idx)++; | |
2026 | qd_idx = raid_disks - 1; | |
2027 | break; | |
2028 | ||
16a53ecc | 2029 | default: |
99c0fb5f | 2030 | BUG(); |
16a53ecc N |
2031 | } |
2032 | break; | |
1da177e4 LT |
2033 | } |
2034 | ||
911d4ee8 N |
2035 | if (sh) { |
2036 | sh->pd_idx = pd_idx; | |
2037 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2038 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2039 | } |
1da177e4 LT |
2040 | /* |
2041 | * Finally, compute the new sector number | |
2042 | */ | |
2043 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2044 | return new_sector; | |
2045 | } | |
2046 | ||
2047 | ||
784052ec | 2048 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2049 | { |
d1688a6d | 2050 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2051 | int raid_disks = sh->disks; |
2052 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2053 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2054 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2055 | : conf->chunk_sectors; | |
e183eaed N |
2056 | int algorithm = previous ? conf->prev_algo |
2057 | : conf->algorithm; | |
1da177e4 LT |
2058 | sector_t stripe; |
2059 | int chunk_offset; | |
35f2a591 N |
2060 | sector_t chunk_number; |
2061 | int dummy1, dd_idx = i; | |
1da177e4 | 2062 | sector_t r_sector; |
911d4ee8 | 2063 | struct stripe_head sh2; |
1da177e4 | 2064 | |
16a53ecc | 2065 | |
1da177e4 LT |
2066 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
2067 | stripe = new_sector; | |
1da177e4 | 2068 | |
16a53ecc N |
2069 | if (i == sh->pd_idx) |
2070 | return 0; | |
2071 | switch(conf->level) { | |
2072 | case 4: break; | |
2073 | case 5: | |
e183eaed | 2074 | switch (algorithm) { |
1da177e4 LT |
2075 | case ALGORITHM_LEFT_ASYMMETRIC: |
2076 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2077 | if (i > sh->pd_idx) | |
2078 | i--; | |
2079 | break; | |
2080 | case ALGORITHM_LEFT_SYMMETRIC: | |
2081 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2082 | if (i < sh->pd_idx) | |
2083 | i += raid_disks; | |
2084 | i -= (sh->pd_idx + 1); | |
2085 | break; | |
99c0fb5f N |
2086 | case ALGORITHM_PARITY_0: |
2087 | i -= 1; | |
2088 | break; | |
2089 | case ALGORITHM_PARITY_N: | |
2090 | break; | |
1da177e4 | 2091 | default: |
99c0fb5f | 2092 | BUG(); |
16a53ecc N |
2093 | } |
2094 | break; | |
2095 | case 6: | |
d0dabf7e | 2096 | if (i == sh->qd_idx) |
16a53ecc | 2097 | return 0; /* It is the Q disk */ |
e183eaed | 2098 | switch (algorithm) { |
16a53ecc N |
2099 | case ALGORITHM_LEFT_ASYMMETRIC: |
2100 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2101 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2102 | case ALGORITHM_ROTATING_N_RESTART: | |
2103 | if (sh->pd_idx == raid_disks-1) | |
2104 | i--; /* Q D D D P */ | |
16a53ecc N |
2105 | else if (i > sh->pd_idx) |
2106 | i -= 2; /* D D P Q D */ | |
2107 | break; | |
2108 | case ALGORITHM_LEFT_SYMMETRIC: | |
2109 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2110 | if (sh->pd_idx == raid_disks-1) | |
2111 | i--; /* Q D D D P */ | |
2112 | else { | |
2113 | /* D D P Q D */ | |
2114 | if (i < sh->pd_idx) | |
2115 | i += raid_disks; | |
2116 | i -= (sh->pd_idx + 2); | |
2117 | } | |
2118 | break; | |
99c0fb5f N |
2119 | case ALGORITHM_PARITY_0: |
2120 | i -= 2; | |
2121 | break; | |
2122 | case ALGORITHM_PARITY_N: | |
2123 | break; | |
2124 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2125 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2126 | if (sh->pd_idx == 0) |
2127 | i--; /* P D D D Q */ | |
e4424fee N |
2128 | else { |
2129 | /* D D Q P D */ | |
2130 | if (i < sh->pd_idx) | |
2131 | i += raid_disks; | |
2132 | i -= (sh->pd_idx + 1); | |
2133 | } | |
99c0fb5f N |
2134 | break; |
2135 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2136 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2137 | if (i > sh->pd_idx) | |
2138 | i--; | |
2139 | break; | |
2140 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2141 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2142 | if (i < sh->pd_idx) | |
2143 | i += data_disks + 1; | |
2144 | i -= (sh->pd_idx + 1); | |
2145 | break; | |
2146 | case ALGORITHM_PARITY_0_6: | |
2147 | i -= 1; | |
2148 | break; | |
16a53ecc | 2149 | default: |
99c0fb5f | 2150 | BUG(); |
16a53ecc N |
2151 | } |
2152 | break; | |
1da177e4 LT |
2153 | } |
2154 | ||
2155 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2156 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2157 | |
112bf897 | 2158 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2159 | previous, &dummy1, &sh2); |
911d4ee8 N |
2160 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2161 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2162 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2163 | mdname(conf->mddev)); | |
1da177e4 LT |
2164 | return 0; |
2165 | } | |
2166 | return r_sector; | |
2167 | } | |
2168 | ||
2169 | ||
600aa109 | 2170 | static void |
c0f7bddb | 2171 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2172 | int rcw, int expand) |
e33129d8 DW |
2173 | { |
2174 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
d1688a6d | 2175 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2176 | int level = conf->level; |
e33129d8 DW |
2177 | |
2178 | if (rcw) { | |
2179 | /* if we are not expanding this is a proper write request, and | |
2180 | * there will be bios with new data to be drained into the | |
2181 | * stripe cache | |
2182 | */ | |
2183 | if (!expand) { | |
600aa109 DW |
2184 | sh->reconstruct_state = reconstruct_state_drain_run; |
2185 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2186 | } else | |
2187 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2188 | |
ac6b53b6 | 2189 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2190 | |
2191 | for (i = disks; i--; ) { | |
2192 | struct r5dev *dev = &sh->dev[i]; | |
2193 | ||
2194 | if (dev->towrite) { | |
2195 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2196 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2197 | if (!expand) |
2198 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2199 | s->locked++; |
e33129d8 DW |
2200 | } |
2201 | } | |
c0f7bddb | 2202 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2203 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2204 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2205 | } else { |
c0f7bddb | 2206 | BUG_ON(level == 6); |
e33129d8 DW |
2207 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2208 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2209 | ||
d8ee0728 | 2210 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2211 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2212 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2213 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2214 | |
2215 | for (i = disks; i--; ) { | |
2216 | struct r5dev *dev = &sh->dev[i]; | |
2217 | if (i == pd_idx) | |
2218 | continue; | |
2219 | ||
e33129d8 DW |
2220 | if (dev->towrite && |
2221 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2222 | test_bit(R5_Wantcompute, &dev->flags))) { |
2223 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2224 | set_bit(R5_LOCKED, &dev->flags); |
2225 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2226 | s->locked++; |
e33129d8 DW |
2227 | } |
2228 | } | |
2229 | } | |
2230 | ||
c0f7bddb | 2231 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2232 | * are in flight |
2233 | */ | |
2234 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2235 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2236 | s->locked++; |
e33129d8 | 2237 | |
c0f7bddb YT |
2238 | if (level == 6) { |
2239 | int qd_idx = sh->qd_idx; | |
2240 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2241 | ||
2242 | set_bit(R5_LOCKED, &dev->flags); | |
2243 | clear_bit(R5_UPTODATE, &dev->flags); | |
2244 | s->locked++; | |
2245 | } | |
2246 | ||
600aa109 | 2247 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2248 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2249 | s->locked, s->ops_request); |
e33129d8 | 2250 | } |
16a53ecc | 2251 | |
1da177e4 LT |
2252 | /* |
2253 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2254 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2255 | * The bi_next chain must be in order. |
2256 | */ | |
2257 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2258 | { | |
2259 | struct bio **bip; | |
d1688a6d | 2260 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2261 | int firstwrite=0; |
1da177e4 | 2262 | |
cbe47ec5 | 2263 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2264 | (unsigned long long)bi->bi_sector, |
2265 | (unsigned long long)sh->sector); | |
2266 | ||
2267 | ||
1da177e4 | 2268 | spin_lock_irq(&conf->device_lock); |
72626685 | 2269 | if (forwrite) { |
1da177e4 | 2270 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2271 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2272 | firstwrite = 1; | |
2273 | } else | |
1da177e4 LT |
2274 | bip = &sh->dev[dd_idx].toread; |
2275 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2276 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2277 | goto overlap; | |
2278 | bip = & (*bip)->bi_next; | |
2279 | } | |
2280 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2281 | goto overlap; | |
2282 | ||
78bafebd | 2283 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2284 | if (*bip) |
2285 | bi->bi_next = *bip; | |
2286 | *bip = bi; | |
960e739d | 2287 | bi->bi_phys_segments++; |
72626685 | 2288 | |
1da177e4 LT |
2289 | if (forwrite) { |
2290 | /* check if page is covered */ | |
2291 | sector_t sector = sh->dev[dd_idx].sector; | |
2292 | for (bi=sh->dev[dd_idx].towrite; | |
2293 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2294 | bi && bi->bi_sector <= sector; | |
2295 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2296 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2297 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2298 | } | |
2299 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2300 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2301 | } | |
cbe47ec5 | 2302 | spin_unlock_irq(&conf->device_lock); |
cbe47ec5 N |
2303 | |
2304 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2305 | (unsigned long long)(*bip)->bi_sector, | |
2306 | (unsigned long long)sh->sector, dd_idx); | |
2307 | ||
2308 | if (conf->mddev->bitmap && firstwrite) { | |
2309 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2310 | STRIPE_SECTORS, 0); | |
2311 | sh->bm_seq = conf->seq_flush+1; | |
2312 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2313 | } | |
1da177e4 LT |
2314 | return 1; |
2315 | ||
2316 | overlap: | |
2317 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2318 | spin_unlock_irq(&conf->device_lock); | |
1da177e4 LT |
2319 | return 0; |
2320 | } | |
2321 | ||
d1688a6d | 2322 | static void end_reshape(struct r5conf *conf); |
29269553 | 2323 | |
d1688a6d | 2324 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 2325 | struct stripe_head *sh) |
ccfcc3c1 | 2326 | { |
784052ec | 2327 | int sectors_per_chunk = |
09c9e5fa | 2328 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2329 | int dd_idx; |
2d2063ce | 2330 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2331 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2332 | |
112bf897 N |
2333 | raid5_compute_sector(conf, |
2334 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2335 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2336 | previous, |
911d4ee8 | 2337 | &dd_idx, sh); |
ccfcc3c1 N |
2338 | } |
2339 | ||
a4456856 | 2340 | static void |
d1688a6d | 2341 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2342 | struct stripe_head_state *s, int disks, |
2343 | struct bio **return_bi) | |
2344 | { | |
2345 | int i; | |
2346 | for (i = disks; i--; ) { | |
2347 | struct bio *bi; | |
2348 | int bitmap_end = 0; | |
2349 | ||
2350 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 2351 | struct md_rdev *rdev; |
a4456856 DW |
2352 | rcu_read_lock(); |
2353 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2354 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
2355 | atomic_inc(&rdev->nr_pending); |
2356 | else | |
2357 | rdev = NULL; | |
a4456856 | 2358 | rcu_read_unlock(); |
7f0da59b N |
2359 | if (rdev) { |
2360 | if (!rdev_set_badblocks( | |
2361 | rdev, | |
2362 | sh->sector, | |
2363 | STRIPE_SECTORS, 0)) | |
2364 | md_error(conf->mddev, rdev); | |
2365 | rdev_dec_pending(rdev, conf->mddev); | |
2366 | } | |
a4456856 DW |
2367 | } |
2368 | spin_lock_irq(&conf->device_lock); | |
2369 | /* fail all writes first */ | |
2370 | bi = sh->dev[i].towrite; | |
2371 | sh->dev[i].towrite = NULL; | |
2372 | if (bi) { | |
2373 | s->to_write--; | |
2374 | bitmap_end = 1; | |
2375 | } | |
2376 | ||
2377 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2378 | wake_up(&conf->wait_for_overlap); | |
2379 | ||
2380 | while (bi && bi->bi_sector < | |
2381 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2382 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2383 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2384 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2385 | md_write_end(conf->mddev); |
2386 | bi->bi_next = *return_bi; | |
2387 | *return_bi = bi; | |
2388 | } | |
2389 | bi = nextbi; | |
2390 | } | |
2391 | /* and fail all 'written' */ | |
2392 | bi = sh->dev[i].written; | |
2393 | sh->dev[i].written = NULL; | |
2394 | if (bi) bitmap_end = 1; | |
2395 | while (bi && bi->bi_sector < | |
2396 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2397 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2398 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2399 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2400 | md_write_end(conf->mddev); |
2401 | bi->bi_next = *return_bi; | |
2402 | *return_bi = bi; | |
2403 | } | |
2404 | bi = bi2; | |
2405 | } | |
2406 | ||
b5e98d65 DW |
2407 | /* fail any reads if this device is non-operational and |
2408 | * the data has not reached the cache yet. | |
2409 | */ | |
2410 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2411 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2412 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2413 | bi = sh->dev[i].toread; |
2414 | sh->dev[i].toread = NULL; | |
2415 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2416 | wake_up(&conf->wait_for_overlap); | |
2417 | if (bi) s->to_read--; | |
2418 | while (bi && bi->bi_sector < | |
2419 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2420 | struct bio *nextbi = | |
2421 | r5_next_bio(bi, sh->dev[i].sector); | |
2422 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2423 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2424 | bi->bi_next = *return_bi; |
2425 | *return_bi = bi; | |
2426 | } | |
2427 | bi = nextbi; | |
2428 | } | |
2429 | } | |
2430 | spin_unlock_irq(&conf->device_lock); | |
2431 | if (bitmap_end) | |
2432 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2433 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
2434 | /* If we were in the middle of a write the parity block might |
2435 | * still be locked - so just clear all R5_LOCKED flags | |
2436 | */ | |
2437 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
2438 | } |
2439 | ||
8b3e6cdc DW |
2440 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2441 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2442 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2443 | } |
2444 | ||
7f0da59b | 2445 | static void |
d1688a6d | 2446 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
2447 | struct stripe_head_state *s) |
2448 | { | |
2449 | int abort = 0; | |
2450 | int i; | |
2451 | ||
2452 | md_done_sync(conf->mddev, STRIPE_SECTORS, 0); | |
2453 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2454 | s->syncing = 0; | |
9a3e1101 | 2455 | s->replacing = 0; |
7f0da59b | 2456 | /* There is nothing more to do for sync/check/repair. |
9a3e1101 | 2457 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
2458 | * non-sync devices, or abort the recovery |
2459 | */ | |
2460 | if (!test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) | |
2461 | return; | |
2462 | /* During recovery devices cannot be removed, so locking and | |
2463 | * refcounting of rdevs is not needed | |
2464 | */ | |
2465 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 2466 | struct md_rdev *rdev = conf->disks[i].rdev; |
9a3e1101 N |
2467 | if (rdev |
2468 | && !test_bit(Faulty, &rdev->flags) | |
2469 | && !test_bit(In_sync, &rdev->flags) | |
2470 | && !rdev_set_badblocks(rdev, sh->sector, | |
2471 | STRIPE_SECTORS, 0)) | |
2472 | abort = 1; | |
2473 | rdev = conf->disks[i].replacement; | |
2474 | if (rdev | |
2475 | && !test_bit(Faulty, &rdev->flags) | |
2476 | && !test_bit(In_sync, &rdev->flags) | |
2477 | && !rdev_set_badblocks(rdev, sh->sector, | |
2478 | STRIPE_SECTORS, 0)) | |
7f0da59b N |
2479 | abort = 1; |
2480 | } | |
2481 | if (abort) { | |
2482 | conf->recovery_disabled = conf->mddev->recovery_disabled; | |
2483 | set_bit(MD_RECOVERY_INTR, &conf->mddev->recovery); | |
2484 | } | |
2485 | } | |
2486 | ||
9a3e1101 N |
2487 | static int want_replace(struct stripe_head *sh, int disk_idx) |
2488 | { | |
2489 | struct md_rdev *rdev; | |
2490 | int rv = 0; | |
2491 | /* Doing recovery so rcu locking not required */ | |
2492 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
2493 | if (rdev | |
2494 | && !test_bit(Faulty, &rdev->flags) | |
2495 | && !test_bit(In_sync, &rdev->flags) | |
2496 | && (rdev->recovery_offset <= sh->sector | |
2497 | || rdev->mddev->recovery_cp <= sh->sector)) | |
2498 | rv = 1; | |
2499 | ||
2500 | return rv; | |
2501 | } | |
2502 | ||
93b3dbce | 2503 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
2504 | * to be read or computed to satisfy a request. |
2505 | * | |
2506 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 2507 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 2508 | */ |
93b3dbce N |
2509 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
2510 | int disk_idx, int disks) | |
a4456856 | 2511 | { |
5599becc | 2512 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2513 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2514 | &sh->dev[s->failed_num[1]] }; | |
5599becc | 2515 | |
93b3dbce | 2516 | /* is the data in this block needed, and can we get it? */ |
5599becc YT |
2517 | if (!test_bit(R5_LOCKED, &dev->flags) && |
2518 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2519 | (dev->toread || | |
2520 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2521 | s->syncing || s->expanding || | |
9a3e1101 | 2522 | (s->replacing && want_replace(sh, disk_idx)) || |
5d35e09c N |
2523 | (s->failed >= 1 && fdev[0]->toread) || |
2524 | (s->failed >= 2 && fdev[1]->toread) || | |
93b3dbce N |
2525 | (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite && |
2526 | !test_bit(R5_OVERWRITE, &fdev[0]->flags)) || | |
2527 | (sh->raid_conf->level == 6 && s->failed && s->to_write))) { | |
5599becc YT |
2528 | /* we would like to get this block, possibly by computing it, |
2529 | * otherwise read it if the backing disk is insync | |
2530 | */ | |
2531 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2532 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2533 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
2534 | (s->failed && (disk_idx == s->failed_num[0] || |
2535 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
2536 | /* have disk failed, and we're requested to fetch it; |
2537 | * do compute it | |
a4456856 | 2538 | */ |
5599becc YT |
2539 | pr_debug("Computing stripe %llu block %d\n", |
2540 | (unsigned long long)sh->sector, disk_idx); | |
2541 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2542 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2543 | set_bit(R5_Wantcompute, &dev->flags); | |
2544 | sh->ops.target = disk_idx; | |
2545 | sh->ops.target2 = -1; /* no 2nd target */ | |
2546 | s->req_compute = 1; | |
93b3dbce N |
2547 | /* Careful: from this point on 'uptodate' is in the eye |
2548 | * of raid_run_ops which services 'compute' operations | |
2549 | * before writes. R5_Wantcompute flags a block that will | |
2550 | * be R5_UPTODATE by the time it is needed for a | |
2551 | * subsequent operation. | |
2552 | */ | |
5599becc YT |
2553 | s->uptodate++; |
2554 | return 1; | |
2555 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2556 | /* Computing 2-failure is *very* expensive; only | |
2557 | * do it if failed >= 2 | |
2558 | */ | |
2559 | int other; | |
2560 | for (other = disks; other--; ) { | |
2561 | if (other == disk_idx) | |
2562 | continue; | |
2563 | if (!test_bit(R5_UPTODATE, | |
2564 | &sh->dev[other].flags)) | |
2565 | break; | |
a4456856 | 2566 | } |
5599becc YT |
2567 | BUG_ON(other < 0); |
2568 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2569 | (unsigned long long)sh->sector, | |
2570 | disk_idx, other); | |
2571 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2572 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2573 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2574 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2575 | sh->ops.target = disk_idx; | |
2576 | sh->ops.target2 = other; | |
2577 | s->uptodate += 2; | |
2578 | s->req_compute = 1; | |
2579 | return 1; | |
2580 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2581 | set_bit(R5_LOCKED, &dev->flags); | |
2582 | set_bit(R5_Wantread, &dev->flags); | |
2583 | s->locked++; | |
2584 | pr_debug("Reading block %d (sync=%d)\n", | |
2585 | disk_idx, s->syncing); | |
a4456856 DW |
2586 | } |
2587 | } | |
5599becc YT |
2588 | |
2589 | return 0; | |
2590 | } | |
2591 | ||
2592 | /** | |
93b3dbce | 2593 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 2594 | */ |
93b3dbce N |
2595 | static void handle_stripe_fill(struct stripe_head *sh, |
2596 | struct stripe_head_state *s, | |
2597 | int disks) | |
5599becc YT |
2598 | { |
2599 | int i; | |
2600 | ||
2601 | /* look for blocks to read/compute, skip this if a compute | |
2602 | * is already in flight, or if the stripe contents are in the | |
2603 | * midst of changing due to a write | |
2604 | */ | |
2605 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2606 | !sh->reconstruct_state) | |
2607 | for (i = disks; i--; ) | |
93b3dbce | 2608 | if (fetch_block(sh, s, i, disks)) |
5599becc | 2609 | break; |
a4456856 DW |
2610 | set_bit(STRIPE_HANDLE, &sh->state); |
2611 | } | |
2612 | ||
2613 | ||
1fe797e6 | 2614 | /* handle_stripe_clean_event |
a4456856 DW |
2615 | * any written block on an uptodate or failed drive can be returned. |
2616 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2617 | * never LOCKED, so we don't need to test 'failed' directly. | |
2618 | */ | |
d1688a6d | 2619 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
2620 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2621 | { | |
2622 | int i; | |
2623 | struct r5dev *dev; | |
2624 | ||
2625 | for (i = disks; i--; ) | |
2626 | if (sh->dev[i].written) { | |
2627 | dev = &sh->dev[i]; | |
2628 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2629 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2630 | /* We can return any write requests */ | |
2631 | struct bio *wbi, *wbi2; | |
2632 | int bitmap_end = 0; | |
45b4233c | 2633 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2634 | spin_lock_irq(&conf->device_lock); |
2635 | wbi = dev->written; | |
2636 | dev->written = NULL; | |
2637 | while (wbi && wbi->bi_sector < | |
2638 | dev->sector + STRIPE_SECTORS) { | |
2639 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2640 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2641 | md_write_end(conf->mddev); |
2642 | wbi->bi_next = *return_bi; | |
2643 | *return_bi = wbi; | |
2644 | } | |
2645 | wbi = wbi2; | |
2646 | } | |
2647 | if (dev->towrite == NULL) | |
2648 | bitmap_end = 1; | |
2649 | spin_unlock_irq(&conf->device_lock); | |
2650 | if (bitmap_end) | |
2651 | bitmap_endwrite(conf->mddev->bitmap, | |
2652 | sh->sector, | |
2653 | STRIPE_SECTORS, | |
2654 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2655 | 0); | |
2656 | } | |
2657 | } | |
8b3e6cdc DW |
2658 | |
2659 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2660 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2661 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2662 | } |
2663 | ||
d1688a6d | 2664 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
2665 | struct stripe_head *sh, |
2666 | struct stripe_head_state *s, | |
2667 | int disks) | |
a4456856 DW |
2668 | { |
2669 | int rmw = 0, rcw = 0, i; | |
c8ac1803 N |
2670 | if (conf->max_degraded == 2) { |
2671 | /* RAID6 requires 'rcw' in current implementation | |
2672 | * Calculate the real rcw later - for now fake it | |
2673 | * look like rcw is cheaper | |
2674 | */ | |
2675 | rcw = 1; rmw = 2; | |
2676 | } else for (i = disks; i--; ) { | |
a4456856 DW |
2677 | /* would I have to read this buffer for read_modify_write */ |
2678 | struct r5dev *dev = &sh->dev[i]; | |
2679 | if ((dev->towrite || i == sh->pd_idx) && | |
2680 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2681 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2682 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2683 | if (test_bit(R5_Insync, &dev->flags)) |
2684 | rmw++; | |
2685 | else | |
2686 | rmw += 2*disks; /* cannot read it */ | |
2687 | } | |
2688 | /* Would I have to read this buffer for reconstruct_write */ | |
2689 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2690 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2691 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2692 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2693 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2694 | else |
2695 | rcw += 2*disks; | |
2696 | } | |
2697 | } | |
45b4233c | 2698 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2699 | (unsigned long long)sh->sector, rmw, rcw); |
2700 | set_bit(STRIPE_HANDLE, &sh->state); | |
2701 | if (rmw < rcw && rmw > 0) | |
2702 | /* prefer read-modify-write, but need to get some data */ | |
2703 | for (i = disks; i--; ) { | |
2704 | struct r5dev *dev = &sh->dev[i]; | |
2705 | if ((dev->towrite || i == sh->pd_idx) && | |
2706 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2707 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2708 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2709 | test_bit(R5_Insync, &dev->flags)) { |
2710 | if ( | |
2711 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2712 | pr_debug("Read_old block " |
a4456856 DW |
2713 | "%d for r-m-w\n", i); |
2714 | set_bit(R5_LOCKED, &dev->flags); | |
2715 | set_bit(R5_Wantread, &dev->flags); | |
2716 | s->locked++; | |
2717 | } else { | |
2718 | set_bit(STRIPE_DELAYED, &sh->state); | |
2719 | set_bit(STRIPE_HANDLE, &sh->state); | |
2720 | } | |
2721 | } | |
2722 | } | |
c8ac1803 | 2723 | if (rcw <= rmw && rcw > 0) { |
a4456856 | 2724 | /* want reconstruct write, but need to get some data */ |
c8ac1803 | 2725 | rcw = 0; |
a4456856 DW |
2726 | for (i = disks; i--; ) { |
2727 | struct r5dev *dev = &sh->dev[i]; | |
2728 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 2729 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 2730 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 2731 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
2732 | test_bit(R5_Wantcompute, &dev->flags))) { |
2733 | rcw++; | |
2734 | if (!test_bit(R5_Insync, &dev->flags)) | |
2735 | continue; /* it's a failed drive */ | |
a4456856 DW |
2736 | if ( |
2737 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2738 | pr_debug("Read_old block " |
a4456856 DW |
2739 | "%d for Reconstruct\n", i); |
2740 | set_bit(R5_LOCKED, &dev->flags); | |
2741 | set_bit(R5_Wantread, &dev->flags); | |
2742 | s->locked++; | |
2743 | } else { | |
2744 | set_bit(STRIPE_DELAYED, &sh->state); | |
2745 | set_bit(STRIPE_HANDLE, &sh->state); | |
2746 | } | |
2747 | } | |
2748 | } | |
c8ac1803 | 2749 | } |
a4456856 DW |
2750 | /* now if nothing is locked, and if we have enough data, |
2751 | * we can start a write request | |
2752 | */ | |
f38e1219 DW |
2753 | /* since handle_stripe can be called at any time we need to handle the |
2754 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2755 | * subsequent call wants to start a write request. raid_run_ops only |
2756 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2757 | * simultaneously. If this is not the case then new writes need to be |
2758 | * held off until the compute completes. | |
2759 | */ | |
976ea8d4 DW |
2760 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2761 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2762 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2763 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2764 | } |
2765 | ||
d1688a6d | 2766 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2767 | struct stripe_head_state *s, int disks) |
2768 | { | |
ecc65c9b | 2769 | struct r5dev *dev = NULL; |
bd2ab670 | 2770 | |
a4456856 | 2771 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2772 | |
ecc65c9b DW |
2773 | switch (sh->check_state) { |
2774 | case check_state_idle: | |
2775 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2776 | if (s->failed == 0) { |
bd2ab670 | 2777 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2778 | sh->check_state = check_state_run; |
2779 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2780 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2781 | s->uptodate--; |
ecc65c9b | 2782 | break; |
bd2ab670 | 2783 | } |
f2b3b44d | 2784 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
2785 | /* fall through */ |
2786 | case check_state_compute_result: | |
2787 | sh->check_state = check_state_idle; | |
2788 | if (!dev) | |
2789 | dev = &sh->dev[sh->pd_idx]; | |
2790 | ||
2791 | /* check that a write has not made the stripe insync */ | |
2792 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2793 | break; | |
c8894419 | 2794 | |
a4456856 | 2795 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2796 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2797 | BUG_ON(s->uptodate != disks); | |
2798 | ||
2799 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2800 | s->locked++; |
a4456856 | 2801 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2802 | |
a4456856 | 2803 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2804 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2805 | break; |
2806 | case check_state_run: | |
2807 | break; /* we will be called again upon completion */ | |
2808 | case check_state_check_result: | |
2809 | sh->check_state = check_state_idle; | |
2810 | ||
2811 | /* if a failure occurred during the check operation, leave | |
2812 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2813 | */ | |
2814 | if (s->failed) | |
2815 | break; | |
2816 | ||
2817 | /* handle a successful check operation, if parity is correct | |
2818 | * we are done. Otherwise update the mismatch count and repair | |
2819 | * parity if !MD_RECOVERY_CHECK | |
2820 | */ | |
ad283ea4 | 2821 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2822 | /* parity is correct (on disc, |
2823 | * not in buffer any more) | |
2824 | */ | |
2825 | set_bit(STRIPE_INSYNC, &sh->state); | |
2826 | else { | |
2827 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2828 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2829 | /* don't try to repair!! */ | |
2830 | set_bit(STRIPE_INSYNC, &sh->state); | |
2831 | else { | |
2832 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2833 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2834 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2835 | set_bit(R5_Wantcompute, | |
2836 | &sh->dev[sh->pd_idx].flags); | |
2837 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2838 | sh->ops.target2 = -1; |
ecc65c9b DW |
2839 | s->uptodate++; |
2840 | } | |
2841 | } | |
2842 | break; | |
2843 | case check_state_compute_run: | |
2844 | break; | |
2845 | default: | |
2846 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2847 | __func__, sh->check_state, | |
2848 | (unsigned long long) sh->sector); | |
2849 | BUG(); | |
a4456856 DW |
2850 | } |
2851 | } | |
2852 | ||
2853 | ||
d1688a6d | 2854 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 2855 | struct stripe_head_state *s, |
f2b3b44d | 2856 | int disks) |
a4456856 | 2857 | { |
a4456856 | 2858 | int pd_idx = sh->pd_idx; |
34e04e87 | 2859 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2860 | struct r5dev *dev; |
a4456856 DW |
2861 | |
2862 | set_bit(STRIPE_HANDLE, &sh->state); | |
2863 | ||
2864 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2865 | |
a4456856 DW |
2866 | /* Want to check and possibly repair P and Q. |
2867 | * However there could be one 'failed' device, in which | |
2868 | * case we can only check one of them, possibly using the | |
2869 | * other to generate missing data | |
2870 | */ | |
2871 | ||
d82dfee0 DW |
2872 | switch (sh->check_state) { |
2873 | case check_state_idle: | |
2874 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 2875 | if (s->failed == s->q_failed) { |
d82dfee0 | 2876 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2877 | * makes sense to check P (If anything else were failed, |
2878 | * we would have used P to recreate it). | |
2879 | */ | |
d82dfee0 | 2880 | sh->check_state = check_state_run; |
a4456856 | 2881 | } |
f2b3b44d | 2882 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 2883 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2884 | * anything, so it makes sense to check it |
2885 | */ | |
d82dfee0 DW |
2886 | if (sh->check_state == check_state_run) |
2887 | sh->check_state = check_state_run_pq; | |
2888 | else | |
2889 | sh->check_state = check_state_run_q; | |
a4456856 | 2890 | } |
a4456856 | 2891 | |
d82dfee0 DW |
2892 | /* discard potentially stale zero_sum_result */ |
2893 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2894 | |
d82dfee0 DW |
2895 | if (sh->check_state == check_state_run) { |
2896 | /* async_xor_zero_sum destroys the contents of P */ | |
2897 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2898 | s->uptodate--; | |
a4456856 | 2899 | } |
d82dfee0 DW |
2900 | if (sh->check_state >= check_state_run && |
2901 | sh->check_state <= check_state_run_pq) { | |
2902 | /* async_syndrome_zero_sum preserves P and Q, so | |
2903 | * no need to mark them !uptodate here | |
2904 | */ | |
2905 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2906 | break; | |
a4456856 DW |
2907 | } |
2908 | ||
d82dfee0 DW |
2909 | /* we have 2-disk failure */ |
2910 | BUG_ON(s->failed != 2); | |
2911 | /* fall through */ | |
2912 | case check_state_compute_result: | |
2913 | sh->check_state = check_state_idle; | |
a4456856 | 2914 | |
d82dfee0 DW |
2915 | /* check that a write has not made the stripe insync */ |
2916 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2917 | break; | |
a4456856 DW |
2918 | |
2919 | /* now write out any block on a failed drive, | |
d82dfee0 | 2920 | * or P or Q if they were recomputed |
a4456856 | 2921 | */ |
d82dfee0 | 2922 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 2923 | if (s->failed == 2) { |
f2b3b44d | 2924 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
2925 | s->locked++; |
2926 | set_bit(R5_LOCKED, &dev->flags); | |
2927 | set_bit(R5_Wantwrite, &dev->flags); | |
2928 | } | |
2929 | if (s->failed >= 1) { | |
f2b3b44d | 2930 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
2931 | s->locked++; |
2932 | set_bit(R5_LOCKED, &dev->flags); | |
2933 | set_bit(R5_Wantwrite, &dev->flags); | |
2934 | } | |
d82dfee0 | 2935 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2936 | dev = &sh->dev[pd_idx]; |
2937 | s->locked++; | |
2938 | set_bit(R5_LOCKED, &dev->flags); | |
2939 | set_bit(R5_Wantwrite, &dev->flags); | |
2940 | } | |
d82dfee0 | 2941 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2942 | dev = &sh->dev[qd_idx]; |
2943 | s->locked++; | |
2944 | set_bit(R5_LOCKED, &dev->flags); | |
2945 | set_bit(R5_Wantwrite, &dev->flags); | |
2946 | } | |
2947 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2948 | ||
2949 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2950 | break; |
2951 | case check_state_run: | |
2952 | case check_state_run_q: | |
2953 | case check_state_run_pq: | |
2954 | break; /* we will be called again upon completion */ | |
2955 | case check_state_check_result: | |
2956 | sh->check_state = check_state_idle; | |
2957 | ||
2958 | /* handle a successful check operation, if parity is correct | |
2959 | * we are done. Otherwise update the mismatch count and repair | |
2960 | * parity if !MD_RECOVERY_CHECK | |
2961 | */ | |
2962 | if (sh->ops.zero_sum_result == 0) { | |
2963 | /* both parities are correct */ | |
2964 | if (!s->failed) | |
2965 | set_bit(STRIPE_INSYNC, &sh->state); | |
2966 | else { | |
2967 | /* in contrast to the raid5 case we can validate | |
2968 | * parity, but still have a failure to write | |
2969 | * back | |
2970 | */ | |
2971 | sh->check_state = check_state_compute_result; | |
2972 | /* Returning at this point means that we may go | |
2973 | * off and bring p and/or q uptodate again so | |
2974 | * we make sure to check zero_sum_result again | |
2975 | * to verify if p or q need writeback | |
2976 | */ | |
2977 | } | |
2978 | } else { | |
2979 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2980 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2981 | /* don't try to repair!! */ | |
2982 | set_bit(STRIPE_INSYNC, &sh->state); | |
2983 | else { | |
2984 | int *target = &sh->ops.target; | |
2985 | ||
2986 | sh->ops.target = -1; | |
2987 | sh->ops.target2 = -1; | |
2988 | sh->check_state = check_state_compute_run; | |
2989 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2990 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2991 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2992 | set_bit(R5_Wantcompute, | |
2993 | &sh->dev[pd_idx].flags); | |
2994 | *target = pd_idx; | |
2995 | target = &sh->ops.target2; | |
2996 | s->uptodate++; | |
2997 | } | |
2998 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2999 | set_bit(R5_Wantcompute, | |
3000 | &sh->dev[qd_idx].flags); | |
3001 | *target = qd_idx; | |
3002 | s->uptodate++; | |
3003 | } | |
3004 | } | |
3005 | } | |
3006 | break; | |
3007 | case check_state_compute_run: | |
3008 | break; | |
3009 | default: | |
3010 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3011 | __func__, sh->check_state, | |
3012 | (unsigned long long) sh->sector); | |
3013 | BUG(); | |
a4456856 DW |
3014 | } |
3015 | } | |
3016 | ||
d1688a6d | 3017 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3018 | { |
3019 | int i; | |
3020 | ||
3021 | /* We have read all the blocks in this stripe and now we need to | |
3022 | * copy some of them into a target stripe for expand. | |
3023 | */ | |
f0a50d37 | 3024 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
3025 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3026 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3027 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3028 | int dd_idx, j; |
a4456856 | 3029 | struct stripe_head *sh2; |
a08abd8c | 3030 | struct async_submit_ctl submit; |
a4456856 | 3031 | |
784052ec | 3032 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3033 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3034 | &dd_idx, NULL); | |
a8c906ca | 3035 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3036 | if (sh2 == NULL) |
3037 | /* so far only the early blocks of this stripe | |
3038 | * have been requested. When later blocks | |
3039 | * get requested, we will try again | |
3040 | */ | |
3041 | continue; | |
3042 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3043 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3044 | /* must have already done this block */ | |
3045 | release_stripe(sh2); | |
3046 | continue; | |
3047 | } | |
f0a50d37 DW |
3048 | |
3049 | /* place all the copies on one channel */ | |
a08abd8c | 3050 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3051 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3052 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3053 | &submit); |
f0a50d37 | 3054 | |
a4456856 DW |
3055 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3056 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3057 | for (j = 0; j < conf->raid_disks; j++) | |
3058 | if (j != sh2->pd_idx && | |
86c374ba | 3059 | j != sh2->qd_idx && |
a4456856 DW |
3060 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3061 | break; | |
3062 | if (j == conf->raid_disks) { | |
3063 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3064 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3065 | } | |
3066 | release_stripe(sh2); | |
f0a50d37 | 3067 | |
a4456856 | 3068 | } |
a2e08551 N |
3069 | /* done submitting copies, wait for them to complete */ |
3070 | if (tx) { | |
3071 | async_tx_ack(tx); | |
3072 | dma_wait_for_async_tx(tx); | |
3073 | } | |
a4456856 | 3074 | } |
1da177e4 LT |
3075 | |
3076 | /* | |
3077 | * handle_stripe - do things to a stripe. | |
3078 | * | |
9a3e1101 N |
3079 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3080 | * state of various bits to see what needs to be done. | |
1da177e4 | 3081 | * Possible results: |
9a3e1101 N |
3082 | * return some read requests which now have data |
3083 | * return some write requests which are safely on storage | |
1da177e4 LT |
3084 | * schedule a read on some buffers |
3085 | * schedule a write of some buffers | |
3086 | * return confirmation of parity correctness | |
3087 | * | |
1da177e4 | 3088 | */ |
a4456856 | 3089 | |
acfe726b | 3090 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3091 | { |
d1688a6d | 3092 | struct r5conf *conf = sh->raid_conf; |
f416885e | 3093 | int disks = sh->disks; |
474af965 N |
3094 | struct r5dev *dev; |
3095 | int i; | |
9a3e1101 | 3096 | int do_recovery = 0; |
1da177e4 | 3097 | |
acfe726b N |
3098 | memset(s, 0, sizeof(*s)); |
3099 | ||
acfe726b N |
3100 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3101 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
3102 | s->failed_num[0] = -1; | |
3103 | s->failed_num[1] = -1; | |
1da177e4 | 3104 | |
acfe726b | 3105 | /* Now to look around and see what can be done */ |
1da177e4 | 3106 | rcu_read_lock(); |
c4c1663b | 3107 | spin_lock_irq(&conf->device_lock); |
16a53ecc | 3108 | for (i=disks; i--; ) { |
3cb03002 | 3109 | struct md_rdev *rdev; |
31c176ec N |
3110 | sector_t first_bad; |
3111 | int bad_sectors; | |
3112 | int is_bad = 0; | |
acfe726b | 3113 | |
16a53ecc | 3114 | dev = &sh->dev[i]; |
1da177e4 | 3115 | |
45b4233c | 3116 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
3117 | i, dev->flags, |
3118 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
3119 | /* maybe we can reply to a read |
3120 | * | |
3121 | * new wantfill requests are only permitted while | |
3122 | * ops_complete_biofill is guaranteed to be inactive | |
3123 | */ | |
3124 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3125 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3126 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3127 | |
16a53ecc | 3128 | /* now count some things */ |
cc94015a N |
3129 | if (test_bit(R5_LOCKED, &dev->flags)) |
3130 | s->locked++; | |
3131 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3132 | s->uptodate++; | |
2d6e4ecc | 3133 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
3134 | s->compute++; |
3135 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 3136 | } |
1da177e4 | 3137 | |
acfe726b | 3138 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 3139 | s->to_fill++; |
acfe726b | 3140 | else if (dev->toread) |
cc94015a | 3141 | s->to_read++; |
16a53ecc | 3142 | if (dev->towrite) { |
cc94015a | 3143 | s->to_write++; |
16a53ecc | 3144 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3145 | s->non_overwrite++; |
16a53ecc | 3146 | } |
a4456856 | 3147 | if (dev->written) |
cc94015a | 3148 | s->written++; |
14a75d3e N |
3149 | /* Prefer to use the replacement for reads, but only |
3150 | * if it is recovered enough and has no bad blocks. | |
3151 | */ | |
3152 | rdev = rcu_dereference(conf->disks[i].replacement); | |
3153 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
3154 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
3155 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3156 | &first_bad, &bad_sectors)) | |
3157 | set_bit(R5_ReadRepl, &dev->flags); | |
3158 | else { | |
9a3e1101 N |
3159 | if (rdev) |
3160 | set_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
3161 | rdev = rcu_dereference(conf->disks[i].rdev); |
3162 | clear_bit(R5_ReadRepl, &dev->flags); | |
3163 | } | |
9283d8c5 N |
3164 | if (rdev && test_bit(Faulty, &rdev->flags)) |
3165 | rdev = NULL; | |
31c176ec N |
3166 | if (rdev) { |
3167 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3168 | &first_bad, &bad_sectors); | |
3169 | if (s->blocked_rdev == NULL | |
3170 | && (test_bit(Blocked, &rdev->flags) | |
3171 | || is_bad < 0)) { | |
3172 | if (is_bad < 0) | |
3173 | set_bit(BlockedBadBlocks, | |
3174 | &rdev->flags); | |
3175 | s->blocked_rdev = rdev; | |
3176 | atomic_inc(&rdev->nr_pending); | |
3177 | } | |
6bfe0b49 | 3178 | } |
415e72d0 N |
3179 | clear_bit(R5_Insync, &dev->flags); |
3180 | if (!rdev) | |
3181 | /* Not in-sync */; | |
31c176ec N |
3182 | else if (is_bad) { |
3183 | /* also not in-sync */ | |
3184 | if (!test_bit(WriteErrorSeen, &rdev->flags)) { | |
3185 | /* treat as in-sync, but with a read error | |
3186 | * which we can now try to correct | |
3187 | */ | |
3188 | set_bit(R5_Insync, &dev->flags); | |
3189 | set_bit(R5_ReadError, &dev->flags); | |
3190 | } | |
3191 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 3192 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 3193 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 3194 | /* in sync if before recovery_offset */ |
30d7a483 N |
3195 | set_bit(R5_Insync, &dev->flags); |
3196 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
3197 | test_bit(R5_Expanded, &dev->flags)) | |
3198 | /* If we've reshaped into here, we assume it is Insync. | |
3199 | * We will shortly update recovery_offset to make | |
3200 | * it official. | |
3201 | */ | |
3202 | set_bit(R5_Insync, &dev->flags); | |
3203 | ||
5d8c71f9 | 3204 | if (rdev && test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
3205 | /* This flag does not apply to '.replacement' |
3206 | * only to .rdev, so make sure to check that*/ | |
3207 | struct md_rdev *rdev2 = rcu_dereference( | |
3208 | conf->disks[i].rdev); | |
3209 | if (rdev2 == rdev) | |
3210 | clear_bit(R5_Insync, &dev->flags); | |
3211 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 3212 | s->handle_bad_blocks = 1; |
14a75d3e | 3213 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
3214 | } else |
3215 | clear_bit(R5_WriteError, &dev->flags); | |
3216 | } | |
5d8c71f9 | 3217 | if (rdev && test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
3218 | /* This flag does not apply to '.replacement' |
3219 | * only to .rdev, so make sure to check that*/ | |
3220 | struct md_rdev *rdev2 = rcu_dereference( | |
3221 | conf->disks[i].rdev); | |
3222 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 3223 | s->handle_bad_blocks = 1; |
14a75d3e | 3224 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
3225 | } else |
3226 | clear_bit(R5_MadeGood, &dev->flags); | |
3227 | } | |
977df362 N |
3228 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
3229 | struct md_rdev *rdev2 = rcu_dereference( | |
3230 | conf->disks[i].replacement); | |
3231 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
3232 | s->handle_bad_blocks = 1; | |
3233 | atomic_inc(&rdev2->nr_pending); | |
3234 | } else | |
3235 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
3236 | } | |
415e72d0 | 3237 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
3238 | /* The ReadError flag will just be confusing now */ |
3239 | clear_bit(R5_ReadError, &dev->flags); | |
3240 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3241 | } |
415e72d0 N |
3242 | if (test_bit(R5_ReadError, &dev->flags)) |
3243 | clear_bit(R5_Insync, &dev->flags); | |
3244 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3245 | if (s->failed < 2) |
3246 | s->failed_num[s->failed] = i; | |
3247 | s->failed++; | |
9a3e1101 N |
3248 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
3249 | do_recovery = 1; | |
415e72d0 | 3250 | } |
1da177e4 | 3251 | } |
c4c1663b | 3252 | spin_unlock_irq(&conf->device_lock); |
9a3e1101 N |
3253 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
3254 | /* If there is a failed device being replaced, | |
3255 | * we must be recovering. | |
3256 | * else if we are after recovery_cp, we must be syncing | |
3257 | * else we can only be replacing | |
3258 | * sync and recovery both need to read all devices, and so | |
3259 | * use the same flag. | |
3260 | */ | |
3261 | if (do_recovery || | |
3262 | sh->sector >= conf->mddev->recovery_cp) | |
3263 | s->syncing = 1; | |
3264 | else | |
3265 | s->replacing = 1; | |
3266 | } | |
1da177e4 | 3267 | rcu_read_unlock(); |
cc94015a N |
3268 | } |
3269 | ||
3270 | static void handle_stripe(struct stripe_head *sh) | |
3271 | { | |
3272 | struct stripe_head_state s; | |
d1688a6d | 3273 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 3274 | int i; |
84789554 N |
3275 | int prexor; |
3276 | int disks = sh->disks; | |
474af965 | 3277 | struct r5dev *pdev, *qdev; |
cc94015a N |
3278 | |
3279 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 3280 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
3281 | /* already being handled, ensure it gets handled |
3282 | * again when current action finishes */ | |
3283 | set_bit(STRIPE_HANDLE, &sh->state); | |
3284 | return; | |
3285 | } | |
3286 | ||
3287 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3288 | set_bit(STRIPE_SYNCING, &sh->state); | |
3289 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3290 | } | |
3291 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3292 | ||
3293 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3294 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3295 | (unsigned long long)sh->sector, sh->state, | |
3296 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3297 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 3298 | |
acfe726b | 3299 | analyse_stripe(sh, &s); |
c5a31000 | 3300 | |
bc2607f3 N |
3301 | if (s.handle_bad_blocks) { |
3302 | set_bit(STRIPE_HANDLE, &sh->state); | |
3303 | goto finish; | |
3304 | } | |
3305 | ||
474af965 N |
3306 | if (unlikely(s.blocked_rdev)) { |
3307 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 3308 | s.replacing || s.to_write || s.written) { |
474af965 N |
3309 | set_bit(STRIPE_HANDLE, &sh->state); |
3310 | goto finish; | |
3311 | } | |
3312 | /* There is nothing for the blocked_rdev to block */ | |
3313 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
3314 | s.blocked_rdev = NULL; | |
3315 | } | |
3316 | ||
3317 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
3318 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3319 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3320 | } | |
3321 | ||
3322 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
3323 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
3324 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
3325 | s.failed_num[0], s.failed_num[1]); | |
3326 | /* check if the array has lost more than max_degraded devices and, | |
3327 | * if so, some requests might need to be failed. | |
3328 | */ | |
9a3f530f N |
3329 | if (s.failed > conf->max_degraded) { |
3330 | sh->check_state = 0; | |
3331 | sh->reconstruct_state = 0; | |
3332 | if (s.to_read+s.to_write+s.written) | |
3333 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 3334 | if (s.syncing + s.replacing) |
9a3f530f N |
3335 | handle_failed_sync(conf, sh, &s); |
3336 | } | |
474af965 N |
3337 | |
3338 | /* | |
3339 | * might be able to return some write requests if the parity blocks | |
3340 | * are safe, or on a failed drive | |
3341 | */ | |
3342 | pdev = &sh->dev[sh->pd_idx]; | |
3343 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
3344 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
3345 | qdev = &sh->dev[sh->qd_idx]; | |
3346 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
3347 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
3348 | || conf->level < 6; | |
3349 | ||
3350 | if (s.written && | |
3351 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
3352 | && !test_bit(R5_LOCKED, &pdev->flags) | |
3353 | && test_bit(R5_UPTODATE, &pdev->flags)))) && | |
3354 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
3355 | && !test_bit(R5_LOCKED, &qdev->flags) | |
3356 | && test_bit(R5_UPTODATE, &qdev->flags))))) | |
3357 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
3358 | ||
3359 | /* Now we might consider reading some blocks, either to check/generate | |
3360 | * parity, or to satisfy requests | |
3361 | * or to load a block that is being partially written. | |
3362 | */ | |
3363 | if (s.to_read || s.non_overwrite | |
3364 | || (conf->level == 6 && s.to_write && s.failed) | |
9a3e1101 N |
3365 | || (s.syncing && (s.uptodate + s.compute < disks)) |
3366 | || s.replacing | |
3367 | || s.expanding) | |
474af965 N |
3368 | handle_stripe_fill(sh, &s, disks); |
3369 | ||
84789554 N |
3370 | /* Now we check to see if any write operations have recently |
3371 | * completed | |
3372 | */ | |
3373 | prexor = 0; | |
3374 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
3375 | prexor = 1; | |
3376 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
3377 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
3378 | sh->reconstruct_state = reconstruct_state_idle; | |
3379 | ||
3380 | /* All the 'written' buffers and the parity block are ready to | |
3381 | * be written back to disk | |
3382 | */ | |
3383 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3384 | BUG_ON(sh->qd_idx >= 0 && | |
3385 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags)); | |
3386 | for (i = disks; i--; ) { | |
3387 | struct r5dev *dev = &sh->dev[i]; | |
3388 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3389 | (i == sh->pd_idx || i == sh->qd_idx || | |
3390 | dev->written)) { | |
3391 | pr_debug("Writing block %d\n", i); | |
3392 | set_bit(R5_Wantwrite, &dev->flags); | |
3393 | if (prexor) | |
3394 | continue; | |
3395 | if (!test_bit(R5_Insync, &dev->flags) || | |
3396 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
3397 | s.failed == 0)) | |
3398 | set_bit(STRIPE_INSYNC, &sh->state); | |
3399 | } | |
3400 | } | |
3401 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3402 | s.dec_preread_active = 1; | |
3403 | } | |
3404 | ||
3405 | /* Now to consider new write requests and what else, if anything | |
3406 | * should be read. We do not handle new writes when: | |
3407 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3408 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3409 | * block. | |
3410 | */ | |
3411 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
3412 | handle_stripe_dirtying(conf, sh, &s, disks); | |
3413 | ||
3414 | /* maybe we need to check and possibly fix the parity for this stripe | |
3415 | * Any reads will already have been scheduled, so we just see if enough | |
3416 | * data is available. The parity check is held off while parity | |
3417 | * dependent operations are in flight. | |
3418 | */ | |
3419 | if (sh->check_state || | |
3420 | (s.syncing && s.locked == 0 && | |
3421 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3422 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
3423 | if (conf->level == 6) | |
3424 | handle_parity_checks6(conf, sh, &s, disks); | |
3425 | else | |
3426 | handle_parity_checks5(conf, sh, &s, disks); | |
3427 | } | |
c5a31000 | 3428 | |
9a3e1101 N |
3429 | if (s.replacing && s.locked == 0 |
3430 | && !test_bit(STRIPE_INSYNC, &sh->state)) { | |
3431 | /* Write out to replacement devices where possible */ | |
3432 | for (i = 0; i < conf->raid_disks; i++) | |
3433 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags) && | |
3434 | test_bit(R5_NeedReplace, &sh->dev[i].flags)) { | |
3435 | set_bit(R5_WantReplace, &sh->dev[i].flags); | |
3436 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3437 | s.locked++; | |
3438 | } | |
3439 | set_bit(STRIPE_INSYNC, &sh->state); | |
3440 | } | |
3441 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
3442 | test_bit(STRIPE_INSYNC, &sh->state)) { | |
c5a31000 N |
3443 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
3444 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3445 | } | |
3446 | ||
3447 | /* If the failed drives are just a ReadError, then we might need | |
3448 | * to progress the repair/check process | |
3449 | */ | |
3450 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
3451 | for (i = 0; i < s.failed; i++) { | |
3452 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
3453 | if (test_bit(R5_ReadError, &dev->flags) | |
3454 | && !test_bit(R5_LOCKED, &dev->flags) | |
3455 | && test_bit(R5_UPTODATE, &dev->flags) | |
3456 | ) { | |
3457 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3458 | set_bit(R5_Wantwrite, &dev->flags); | |
3459 | set_bit(R5_ReWrite, &dev->flags); | |
3460 | set_bit(R5_LOCKED, &dev->flags); | |
3461 | s.locked++; | |
3462 | } else { | |
3463 | /* let's read it back */ | |
3464 | set_bit(R5_Wantread, &dev->flags); | |
3465 | set_bit(R5_LOCKED, &dev->flags); | |
3466 | s.locked++; | |
3467 | } | |
3468 | } | |
3469 | } | |
3470 | ||
3471 | ||
3687c061 N |
3472 | /* Finish reconstruct operations initiated by the expansion process */ |
3473 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3474 | struct stripe_head *sh_src | |
3475 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
3476 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
3477 | /* sh cannot be written until sh_src has been read. | |
3478 | * so arrange for sh to be delayed a little | |
3479 | */ | |
3480 | set_bit(STRIPE_DELAYED, &sh->state); | |
3481 | set_bit(STRIPE_HANDLE, &sh->state); | |
3482 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3483 | &sh_src->state)) | |
3484 | atomic_inc(&conf->preread_active_stripes); | |
3485 | release_stripe(sh_src); | |
3486 | goto finish; | |
3487 | } | |
3488 | if (sh_src) | |
3489 | release_stripe(sh_src); | |
3490 | ||
3491 | sh->reconstruct_state = reconstruct_state_idle; | |
3492 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3493 | for (i = conf->raid_disks; i--; ) { | |
3494 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3495 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3496 | s.locked++; | |
3497 | } | |
3498 | } | |
f416885e | 3499 | |
3687c061 N |
3500 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
3501 | !sh->reconstruct_state) { | |
3502 | /* Need to write out all blocks after computing parity */ | |
3503 | sh->disks = conf->raid_disks; | |
3504 | stripe_set_idx(sh->sector, conf, 0, sh); | |
3505 | schedule_reconstruction(sh, &s, 1, 1); | |
3506 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
3507 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
3508 | atomic_dec(&conf->reshape_stripes); | |
3509 | wake_up(&conf->wait_for_overlap); | |
3510 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3511 | } | |
3512 | ||
3513 | if (s.expanding && s.locked == 0 && | |
3514 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
3515 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 3516 | |
3687c061 | 3517 | finish: |
6bfe0b49 | 3518 | /* wait for this device to become unblocked */ |
43220aa0 | 3519 | if (conf->mddev->external && unlikely(s.blocked_rdev)) |
c5709ef6 | 3520 | md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev); |
6bfe0b49 | 3521 | |
bc2607f3 N |
3522 | if (s.handle_bad_blocks) |
3523 | for (i = disks; i--; ) { | |
3cb03002 | 3524 | struct md_rdev *rdev; |
bc2607f3 N |
3525 | struct r5dev *dev = &sh->dev[i]; |
3526 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
3527 | /* We own a safe reference to the rdev */ | |
3528 | rdev = conf->disks[i].rdev; | |
3529 | if (!rdev_set_badblocks(rdev, sh->sector, | |
3530 | STRIPE_SECTORS, 0)) | |
3531 | md_error(conf->mddev, rdev); | |
3532 | rdev_dec_pending(rdev, conf->mddev); | |
3533 | } | |
b84db560 N |
3534 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
3535 | rdev = conf->disks[i].rdev; | |
3536 | rdev_clear_badblocks(rdev, sh->sector, | |
3537 | STRIPE_SECTORS); | |
3538 | rdev_dec_pending(rdev, conf->mddev); | |
3539 | } | |
977df362 N |
3540 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
3541 | rdev = conf->disks[i].replacement; | |
3542 | rdev_clear_badblocks(rdev, sh->sector, | |
3543 | STRIPE_SECTORS); | |
3544 | rdev_dec_pending(rdev, conf->mddev); | |
3545 | } | |
bc2607f3 N |
3546 | } |
3547 | ||
6c0069c0 YT |
3548 | if (s.ops_request) |
3549 | raid_run_ops(sh, s.ops_request); | |
3550 | ||
f0e43bcd | 3551 | ops_run_io(sh, &s); |
16a53ecc | 3552 | |
c5709ef6 | 3553 | if (s.dec_preread_active) { |
729a1866 | 3554 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 3555 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3556 | * have actually been submitted. |
3557 | */ | |
3558 | atomic_dec(&conf->preread_active_stripes); | |
3559 | if (atomic_read(&conf->preread_active_stripes) < | |
3560 | IO_THRESHOLD) | |
3561 | md_wakeup_thread(conf->mddev->thread); | |
3562 | } | |
3563 | ||
c5709ef6 | 3564 | return_io(s.return_bi); |
16a53ecc | 3565 | |
257a4b42 | 3566 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
3567 | } |
3568 | ||
d1688a6d | 3569 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
3570 | { |
3571 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3572 | while (!list_empty(&conf->delayed_list)) { | |
3573 | struct list_head *l = conf->delayed_list.next; | |
3574 | struct stripe_head *sh; | |
3575 | sh = list_entry(l, struct stripe_head, lru); | |
3576 | list_del_init(l); | |
3577 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3578 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3579 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3580 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3581 | } |
482c0834 | 3582 | } |
16a53ecc N |
3583 | } |
3584 | ||
d1688a6d | 3585 | static void activate_bit_delay(struct r5conf *conf) |
16a53ecc N |
3586 | { |
3587 | /* device_lock is held */ | |
3588 | struct list_head head; | |
3589 | list_add(&head, &conf->bitmap_list); | |
3590 | list_del_init(&conf->bitmap_list); | |
3591 | while (!list_empty(&head)) { | |
3592 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3593 | list_del_init(&sh->lru); | |
3594 | atomic_inc(&sh->count); | |
3595 | __release_stripe(conf, sh); | |
3596 | } | |
3597 | } | |
3598 | ||
fd01b88c | 3599 | int md_raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 3600 | { |
d1688a6d | 3601 | struct r5conf *conf = mddev->private; |
f022b2fd N |
3602 | |
3603 | /* No difference between reads and writes. Just check | |
3604 | * how busy the stripe_cache is | |
3605 | */ | |
3fa841d7 | 3606 | |
f022b2fd N |
3607 | if (conf->inactive_blocked) |
3608 | return 1; | |
3609 | if (conf->quiesce) | |
3610 | return 1; | |
3611 | if (list_empty_careful(&conf->inactive_list)) | |
3612 | return 1; | |
3613 | ||
3614 | return 0; | |
3615 | } | |
11d8a6e3 N |
3616 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3617 | ||
3618 | static int raid5_congested(void *data, int bits) | |
3619 | { | |
fd01b88c | 3620 | struct mddev *mddev = data; |
11d8a6e3 N |
3621 | |
3622 | return mddev_congested(mddev, bits) || | |
3623 | md_raid5_congested(mddev, bits); | |
3624 | } | |
f022b2fd | 3625 | |
23032a0e RBJ |
3626 | /* We want read requests to align with chunks where possible, |
3627 | * but write requests don't need to. | |
3628 | */ | |
cc371e66 AK |
3629 | static int raid5_mergeable_bvec(struct request_queue *q, |
3630 | struct bvec_merge_data *bvm, | |
3631 | struct bio_vec *biovec) | |
23032a0e | 3632 | { |
fd01b88c | 3633 | struct mddev *mddev = q->queuedata; |
cc371e66 | 3634 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3635 | int max; |
9d8f0363 | 3636 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3637 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3638 | |
cc371e66 | 3639 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3640 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3641 | ||
664e7c41 AN |
3642 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3643 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3644 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3645 | if (max < 0) max = 0; | |
3646 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3647 | return biovec->bv_len; | |
3648 | else | |
3649 | return max; | |
3650 | } | |
3651 | ||
f679623f | 3652 | |
fd01b88c | 3653 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f RBJ |
3654 | { |
3655 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3656 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3657 | unsigned int bio_sectors = bio->bi_size >> 9; |
3658 | ||
664e7c41 AN |
3659 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3660 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3661 | return chunk_sectors >= |
3662 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3663 | } | |
3664 | ||
46031f9a RBJ |
3665 | /* |
3666 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3667 | * later sampled by raid5d. | |
3668 | */ | |
d1688a6d | 3669 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
3670 | { |
3671 | unsigned long flags; | |
3672 | ||
3673 | spin_lock_irqsave(&conf->device_lock, flags); | |
3674 | ||
3675 | bi->bi_next = conf->retry_read_aligned_list; | |
3676 | conf->retry_read_aligned_list = bi; | |
3677 | ||
3678 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3679 | md_wakeup_thread(conf->mddev->thread); | |
3680 | } | |
3681 | ||
3682 | ||
d1688a6d | 3683 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
3684 | { |
3685 | struct bio *bi; | |
3686 | ||
3687 | bi = conf->retry_read_aligned; | |
3688 | if (bi) { | |
3689 | conf->retry_read_aligned = NULL; | |
3690 | return bi; | |
3691 | } | |
3692 | bi = conf->retry_read_aligned_list; | |
3693 | if(bi) { | |
387bb173 | 3694 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3695 | bi->bi_next = NULL; |
960e739d JA |
3696 | /* |
3697 | * this sets the active strip count to 1 and the processed | |
3698 | * strip count to zero (upper 8 bits) | |
3699 | */ | |
46031f9a | 3700 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3701 | } |
3702 | ||
3703 | return bi; | |
3704 | } | |
3705 | ||
3706 | ||
f679623f RBJ |
3707 | /* |
3708 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3709 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3710 | * first). | |
3711 | * If the read failed.. | |
3712 | */ | |
6712ecf8 | 3713 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3714 | { |
3715 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 3716 | struct mddev *mddev; |
d1688a6d | 3717 | struct r5conf *conf; |
46031f9a | 3718 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 3719 | struct md_rdev *rdev; |
46031f9a | 3720 | |
f679623f | 3721 | bio_put(bi); |
46031f9a | 3722 | |
46031f9a RBJ |
3723 | rdev = (void*)raid_bi->bi_next; |
3724 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3725 | mddev = rdev->mddev; |
3726 | conf = mddev->private; | |
46031f9a RBJ |
3727 | |
3728 | rdev_dec_pending(rdev, conf->mddev); | |
3729 | ||
3730 | if (!error && uptodate) { | |
6712ecf8 | 3731 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3732 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3733 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3734 | return; |
46031f9a RBJ |
3735 | } |
3736 | ||
3737 | ||
45b4233c | 3738 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3739 | |
3740 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3741 | } |
3742 | ||
387bb173 NB |
3743 | static int bio_fits_rdev(struct bio *bi) |
3744 | { | |
165125e1 | 3745 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3746 | |
ae03bf63 | 3747 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3748 | return 0; |
3749 | blk_recount_segments(q, bi); | |
8a78362c | 3750 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3751 | return 0; |
3752 | ||
3753 | if (q->merge_bvec_fn) | |
3754 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3755 | * just just give up | |
3756 | */ | |
3757 | return 0; | |
3758 | ||
3759 | return 1; | |
3760 | } | |
3761 | ||
3762 | ||
fd01b88c | 3763 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 3764 | { |
d1688a6d | 3765 | struct r5conf *conf = mddev->private; |
8553fe7e | 3766 | int dd_idx; |
f679623f | 3767 | struct bio* align_bi; |
3cb03002 | 3768 | struct md_rdev *rdev; |
671488cc | 3769 | sector_t end_sector; |
f679623f RBJ |
3770 | |
3771 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3772 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3773 | return 0; |
3774 | } | |
3775 | /* | |
a167f663 | 3776 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3777 | */ |
a167f663 | 3778 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3779 | if (!align_bi) |
3780 | return 0; | |
3781 | /* | |
3782 | * set bi_end_io to a new function, and set bi_private to the | |
3783 | * original bio. | |
3784 | */ | |
3785 | align_bi->bi_end_io = raid5_align_endio; | |
3786 | align_bi->bi_private = raid_bio; | |
3787 | /* | |
3788 | * compute position | |
3789 | */ | |
112bf897 N |
3790 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3791 | 0, | |
911d4ee8 | 3792 | &dd_idx, NULL); |
f679623f | 3793 | |
671488cc | 3794 | end_sector = align_bi->bi_sector + (align_bi->bi_size >> 9); |
f679623f | 3795 | rcu_read_lock(); |
671488cc N |
3796 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
3797 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
3798 | rdev->recovery_offset < end_sector) { | |
3799 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3800 | if (rdev && | |
3801 | (test_bit(Faulty, &rdev->flags) || | |
3802 | !(test_bit(In_sync, &rdev->flags) || | |
3803 | rdev->recovery_offset >= end_sector))) | |
3804 | rdev = NULL; | |
3805 | } | |
3806 | if (rdev) { | |
31c176ec N |
3807 | sector_t first_bad; |
3808 | int bad_sectors; | |
3809 | ||
f679623f RBJ |
3810 | atomic_inc(&rdev->nr_pending); |
3811 | rcu_read_unlock(); | |
46031f9a RBJ |
3812 | raid_bio->bi_next = (void*)rdev; |
3813 | align_bi->bi_bdev = rdev->bdev; | |
3814 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3815 | align_bi->bi_sector += rdev->data_offset; | |
3816 | ||
31c176ec N |
3817 | if (!bio_fits_rdev(align_bi) || |
3818 | is_badblock(rdev, align_bi->bi_sector, align_bi->bi_size>>9, | |
3819 | &first_bad, &bad_sectors)) { | |
3820 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
3821 | bio_put(align_bi); |
3822 | rdev_dec_pending(rdev, mddev); | |
3823 | return 0; | |
3824 | } | |
3825 | ||
46031f9a RBJ |
3826 | spin_lock_irq(&conf->device_lock); |
3827 | wait_event_lock_irq(conf->wait_for_stripe, | |
3828 | conf->quiesce == 0, | |
3829 | conf->device_lock, /* nothing */); | |
3830 | atomic_inc(&conf->active_aligned_reads); | |
3831 | spin_unlock_irq(&conf->device_lock); | |
3832 | ||
f679623f RBJ |
3833 | generic_make_request(align_bi); |
3834 | return 1; | |
3835 | } else { | |
3836 | rcu_read_unlock(); | |
46031f9a | 3837 | bio_put(align_bi); |
f679623f RBJ |
3838 | return 0; |
3839 | } | |
3840 | } | |
3841 | ||
8b3e6cdc DW |
3842 | /* __get_priority_stripe - get the next stripe to process |
3843 | * | |
3844 | * Full stripe writes are allowed to pass preread active stripes up until | |
3845 | * the bypass_threshold is exceeded. In general the bypass_count | |
3846 | * increments when the handle_list is handled before the hold_list; however, it | |
3847 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3848 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3849 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3850 | * handle_list. | |
3851 | */ | |
d1688a6d | 3852 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf) |
8b3e6cdc DW |
3853 | { |
3854 | struct stripe_head *sh; | |
3855 | ||
3856 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3857 | __func__, | |
3858 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3859 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3860 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3861 | ||
3862 | if (!list_empty(&conf->handle_list)) { | |
3863 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3864 | ||
3865 | if (list_empty(&conf->hold_list)) | |
3866 | conf->bypass_count = 0; | |
3867 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3868 | if (conf->hold_list.next == conf->last_hold) | |
3869 | conf->bypass_count++; | |
3870 | else { | |
3871 | conf->last_hold = conf->hold_list.next; | |
3872 | conf->bypass_count -= conf->bypass_threshold; | |
3873 | if (conf->bypass_count < 0) | |
3874 | conf->bypass_count = 0; | |
3875 | } | |
3876 | } | |
3877 | } else if (!list_empty(&conf->hold_list) && | |
3878 | ((conf->bypass_threshold && | |
3879 | conf->bypass_count > conf->bypass_threshold) || | |
3880 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3881 | sh = list_entry(conf->hold_list.next, | |
3882 | typeof(*sh), lru); | |
3883 | conf->bypass_count -= conf->bypass_threshold; | |
3884 | if (conf->bypass_count < 0) | |
3885 | conf->bypass_count = 0; | |
3886 | } else | |
3887 | return NULL; | |
3888 | ||
3889 | list_del_init(&sh->lru); | |
3890 | atomic_inc(&sh->count); | |
3891 | BUG_ON(atomic_read(&sh->count) != 1); | |
3892 | return sh; | |
3893 | } | |
f679623f | 3894 | |
b4fdcb02 | 3895 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 3896 | { |
d1688a6d | 3897 | struct r5conf *conf = mddev->private; |
911d4ee8 | 3898 | int dd_idx; |
1da177e4 LT |
3899 | sector_t new_sector; |
3900 | sector_t logical_sector, last_sector; | |
3901 | struct stripe_head *sh; | |
a362357b | 3902 | const int rw = bio_data_dir(bi); |
49077326 | 3903 | int remaining; |
7c13edc8 | 3904 | int plugged; |
1da177e4 | 3905 | |
e9c7469b TH |
3906 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
3907 | md_flush_request(mddev, bi); | |
5a7bbad2 | 3908 | return; |
e5dcdd80 N |
3909 | } |
3910 | ||
3d310eb7 | 3911 | md_write_start(mddev, bi); |
06d91a5f | 3912 | |
802ba064 | 3913 | if (rw == READ && |
52488615 | 3914 | mddev->reshape_position == MaxSector && |
21a52c6d | 3915 | chunk_aligned_read(mddev,bi)) |
5a7bbad2 | 3916 | return; |
52488615 | 3917 | |
1da177e4 LT |
3918 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3919 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3920 | bi->bi_next = NULL; | |
3921 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3922 | |
7c13edc8 | 3923 | plugged = mddev_check_plugged(mddev); |
1da177e4 LT |
3924 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3925 | DEFINE_WAIT(w); | |
16a53ecc | 3926 | int disks, data_disks; |
b5663ba4 | 3927 | int previous; |
b578d55f | 3928 | |
7ecaa1e6 | 3929 | retry: |
b5663ba4 | 3930 | previous = 0; |
b0f9ec04 | 3931 | disks = conf->raid_disks; |
b578d55f | 3932 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3933 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3934 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3935 | * 64bit on a 32bit platform, and so it might be |
3936 | * possible to see a half-updated value | |
aeb878b0 | 3937 | * Of course reshape_progress could change after |
df8e7f76 N |
3938 | * the lock is dropped, so once we get a reference |
3939 | * to the stripe that we think it is, we will have | |
3940 | * to check again. | |
3941 | */ | |
7ecaa1e6 | 3942 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3943 | if (mddev->delta_disks < 0 |
3944 | ? logical_sector < conf->reshape_progress | |
3945 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3946 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3947 | previous = 1; |
3948 | } else { | |
fef9c61f N |
3949 | if (mddev->delta_disks < 0 |
3950 | ? logical_sector < conf->reshape_safe | |
3951 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3952 | spin_unlock_irq(&conf->device_lock); |
3953 | schedule(); | |
3954 | goto retry; | |
3955 | } | |
3956 | } | |
7ecaa1e6 N |
3957 | spin_unlock_irq(&conf->device_lock); |
3958 | } | |
16a53ecc N |
3959 | data_disks = disks - conf->max_degraded; |
3960 | ||
112bf897 N |
3961 | new_sector = raid5_compute_sector(conf, logical_sector, |
3962 | previous, | |
911d4ee8 | 3963 | &dd_idx, NULL); |
0c55e022 | 3964 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3965 | (unsigned long long)new_sector, |
3966 | (unsigned long long)logical_sector); | |
3967 | ||
b5663ba4 | 3968 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3969 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3970 | if (sh) { |
b0f9ec04 | 3971 | if (unlikely(previous)) { |
7ecaa1e6 | 3972 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3973 | * stripe, so we must do the range check again. |
3974 | * Expansion could still move past after this | |
3975 | * test, but as we are holding a reference to | |
3976 | * 'sh', we know that if that happens, | |
3977 | * STRIPE_EXPANDING will get set and the expansion | |
3978 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3979 | */ |
3980 | int must_retry = 0; | |
3981 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
3982 | if (mddev->delta_disks < 0 |
3983 | ? logical_sector >= conf->reshape_progress | |
3984 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
3985 | /* mismatch, need to try again */ |
3986 | must_retry = 1; | |
3987 | spin_unlock_irq(&conf->device_lock); | |
3988 | if (must_retry) { | |
3989 | release_stripe(sh); | |
7a3ab908 | 3990 | schedule(); |
7ecaa1e6 N |
3991 | goto retry; |
3992 | } | |
3993 | } | |
e62e58a5 | 3994 | |
ffd96e35 | 3995 | if (rw == WRITE && |
a5c308d4 | 3996 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
3997 | logical_sector < mddev->suspend_hi) { |
3998 | release_stripe(sh); | |
e62e58a5 N |
3999 | /* As the suspend_* range is controlled by |
4000 | * userspace, we want an interruptible | |
4001 | * wait. | |
4002 | */ | |
4003 | flush_signals(current); | |
4004 | prepare_to_wait(&conf->wait_for_overlap, | |
4005 | &w, TASK_INTERRUPTIBLE); | |
4006 | if (logical_sector >= mddev->suspend_lo && | |
4007 | logical_sector < mddev->suspend_hi) | |
4008 | schedule(); | |
e464eafd N |
4009 | goto retry; |
4010 | } | |
7ecaa1e6 N |
4011 | |
4012 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 4013 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
4014 | /* Stripe is busy expanding or |
4015 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4016 | * and wait a while |
4017 | */ | |
482c0834 | 4018 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4019 | release_stripe(sh); |
4020 | schedule(); | |
4021 | goto retry; | |
4022 | } | |
4023 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
4024 | set_bit(STRIPE_HANDLE, &sh->state); |
4025 | clear_bit(STRIPE_DELAYED, &sh->state); | |
e9c7469b | 4026 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
4027 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4028 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 4029 | release_stripe(sh); |
1da177e4 LT |
4030 | } else { |
4031 | /* cannot get stripe for read-ahead, just give-up */ | |
4032 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4033 | finish_wait(&conf->wait_for_overlap, &w); | |
4034 | break; | |
4035 | } | |
4036 | ||
4037 | } | |
7c13edc8 N |
4038 | if (!plugged) |
4039 | md_wakeup_thread(mddev->thread); | |
4040 | ||
1da177e4 | 4041 | spin_lock_irq(&conf->device_lock); |
960e739d | 4042 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
4043 | spin_unlock_irq(&conf->device_lock); |
4044 | if (remaining == 0) { | |
1da177e4 | 4045 | |
16a53ecc | 4046 | if ( rw == WRITE ) |
1da177e4 | 4047 | md_write_end(mddev); |
6712ecf8 | 4048 | |
0e13fe23 | 4049 | bio_endio(bi, 0); |
1da177e4 | 4050 | } |
1da177e4 LT |
4051 | } |
4052 | ||
fd01b88c | 4053 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 4054 | |
fd01b88c | 4055 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4056 | { |
52c03291 N |
4057 | /* reshaping is quite different to recovery/resync so it is |
4058 | * handled quite separately ... here. | |
4059 | * | |
4060 | * On each call to sync_request, we gather one chunk worth of | |
4061 | * destination stripes and flag them as expanding. | |
4062 | * Then we find all the source stripes and request reads. | |
4063 | * As the reads complete, handle_stripe will copy the data | |
4064 | * into the destination stripe and release that stripe. | |
4065 | */ | |
d1688a6d | 4066 | struct r5conf *conf = mddev->private; |
1da177e4 | 4067 | struct stripe_head *sh; |
ccfcc3c1 | 4068 | sector_t first_sector, last_sector; |
f416885e N |
4069 | int raid_disks = conf->previous_raid_disks; |
4070 | int data_disks = raid_disks - conf->max_degraded; | |
4071 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4072 | int i; |
4073 | int dd_idx; | |
c8f517c4 | 4074 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4075 | sector_t stripe_addr; |
7a661381 | 4076 | int reshape_sectors; |
ab69ae12 | 4077 | struct list_head stripes; |
52c03291 | 4078 | |
fef9c61f N |
4079 | if (sector_nr == 0) { |
4080 | /* If restarting in the middle, skip the initial sectors */ | |
4081 | if (mddev->delta_disks < 0 && | |
4082 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4083 | sector_nr = raid5_size(mddev, 0, 0) | |
4084 | - conf->reshape_progress; | |
a639755c | 4085 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
4086 | conf->reshape_progress > 0) |
4087 | sector_nr = conf->reshape_progress; | |
f416885e | 4088 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4089 | if (sector_nr) { |
8dee7211 N |
4090 | mddev->curr_resync_completed = sector_nr; |
4091 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4092 | *skipped = 1; |
4093 | return sector_nr; | |
4094 | } | |
52c03291 N |
4095 | } |
4096 | ||
7a661381 N |
4097 | /* We need to process a full chunk at a time. |
4098 | * If old and new chunk sizes differ, we need to process the | |
4099 | * largest of these | |
4100 | */ | |
664e7c41 AN |
4101 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4102 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4103 | else |
9d8f0363 | 4104 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4105 | |
52c03291 N |
4106 | /* we update the metadata when there is more than 3Meg |
4107 | * in the block range (that is rather arbitrary, should | |
4108 | * probably be time based) or when the data about to be | |
4109 | * copied would over-write the source of the data at | |
4110 | * the front of the range. | |
fef9c61f N |
4111 | * i.e. one new_stripe along from reshape_progress new_maps |
4112 | * to after where reshape_safe old_maps to | |
52c03291 | 4113 | */ |
fef9c61f | 4114 | writepos = conf->reshape_progress; |
f416885e | 4115 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4116 | readpos = conf->reshape_progress; |
4117 | sector_div(readpos, data_disks); | |
fef9c61f | 4118 | safepos = conf->reshape_safe; |
f416885e | 4119 | sector_div(safepos, data_disks); |
fef9c61f | 4120 | if (mddev->delta_disks < 0) { |
ed37d83e | 4121 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4122 | readpos += reshape_sectors; |
7a661381 | 4123 | safepos += reshape_sectors; |
fef9c61f | 4124 | } else { |
7a661381 | 4125 | writepos += reshape_sectors; |
ed37d83e N |
4126 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4127 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4128 | } |
52c03291 | 4129 | |
c8f517c4 N |
4130 | /* 'writepos' is the most advanced device address we might write. |
4131 | * 'readpos' is the least advanced device address we might read. | |
4132 | * 'safepos' is the least address recorded in the metadata as having | |
4133 | * been reshaped. | |
4134 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4135 | * ensure safety in the face of a crash - that must be done by userspace | |
4136 | * making a backup of the data. So in that case there is no particular | |
4137 | * rush to update metadata. | |
4138 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4139 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4140 | * we can be safe in the event of a crash. | |
4141 | * So we insist on updating metadata if safepos is behind writepos and | |
4142 | * readpos is beyond writepos. | |
4143 | * In any case, update the metadata every 10 seconds. | |
4144 | * Maybe that number should be configurable, but I'm not sure it is | |
4145 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4146 | */ | |
fef9c61f | 4147 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4148 | ? (safepos > writepos && readpos < writepos) |
4149 | : (safepos < writepos && readpos > writepos)) || | |
4150 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4151 | /* Cannot proceed until we've updated the superblock... */ |
4152 | wait_event(conf->wait_for_overlap, | |
4153 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4154 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4155 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4156 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4157 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4158 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4159 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4160 | kthread_should_stop()); |
4161 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4162 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4163 | spin_unlock_irq(&conf->device_lock); |
4164 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4165 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4166 | } |
4167 | ||
ec32a2bd N |
4168 | if (mddev->delta_disks < 0) { |
4169 | BUG_ON(conf->reshape_progress == 0); | |
4170 | stripe_addr = writepos; | |
4171 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4172 | ~((sector_t)reshape_sectors - 1)) |
4173 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4174 | != sector_nr); |
4175 | } else { | |
7a661381 | 4176 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4177 | stripe_addr = sector_nr; |
4178 | } | |
ab69ae12 | 4179 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4180 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4181 | int j; |
a9f326eb | 4182 | int skipped_disk = 0; |
a8c906ca | 4183 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4184 | set_bit(STRIPE_EXPANDING, &sh->state); |
4185 | atomic_inc(&conf->reshape_stripes); | |
4186 | /* If any of this stripe is beyond the end of the old | |
4187 | * array, then we need to zero those blocks | |
4188 | */ | |
4189 | for (j=sh->disks; j--;) { | |
4190 | sector_t s; | |
4191 | if (j == sh->pd_idx) | |
4192 | continue; | |
f416885e | 4193 | if (conf->level == 6 && |
d0dabf7e | 4194 | j == sh->qd_idx) |
f416885e | 4195 | continue; |
784052ec | 4196 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4197 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4198 | skipped_disk = 1; |
52c03291 N |
4199 | continue; |
4200 | } | |
4201 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4202 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4203 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4204 | } | |
a9f326eb | 4205 | if (!skipped_disk) { |
52c03291 N |
4206 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4207 | set_bit(STRIPE_HANDLE, &sh->state); | |
4208 | } | |
ab69ae12 | 4209 | list_add(&sh->lru, &stripes); |
52c03291 N |
4210 | } |
4211 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4212 | if (mddev->delta_disks < 0) |
7a661381 | 4213 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4214 | else |
7a661381 | 4215 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4216 | spin_unlock_irq(&conf->device_lock); |
4217 | /* Ok, those stripe are ready. We can start scheduling | |
4218 | * reads on the source stripes. | |
4219 | * The source stripes are determined by mapping the first and last | |
4220 | * block on the destination stripes. | |
4221 | */ | |
52c03291 | 4222 | first_sector = |
ec32a2bd | 4223 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4224 | 1, &dd_idx, NULL); |
52c03291 | 4225 | last_sector = |
0e6e0271 | 4226 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4227 | * new_data_disks - 1), |
911d4ee8 | 4228 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4229 | if (last_sector >= mddev->dev_sectors) |
4230 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4231 | while (first_sector <= last_sector) { |
a8c906ca | 4232 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4233 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4234 | set_bit(STRIPE_HANDLE, &sh->state); | |
4235 | release_stripe(sh); | |
4236 | first_sector += STRIPE_SECTORS; | |
4237 | } | |
ab69ae12 N |
4238 | /* Now that the sources are clearly marked, we can release |
4239 | * the destination stripes | |
4240 | */ | |
4241 | while (!list_empty(&stripes)) { | |
4242 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4243 | list_del_init(&sh->lru); | |
4244 | release_stripe(sh); | |
4245 | } | |
c6207277 N |
4246 | /* If this takes us to the resync_max point where we have to pause, |
4247 | * then we need to write out the superblock. | |
4248 | */ | |
7a661381 | 4249 | sector_nr += reshape_sectors; |
c03f6a19 N |
4250 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4251 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4252 | /* Cannot proceed until we've updated the superblock... */ |
4253 | wait_event(conf->wait_for_overlap, | |
4254 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4255 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4256 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4257 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4258 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4259 | md_wakeup_thread(mddev->thread); | |
4260 | wait_event(mddev->sb_wait, | |
4261 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4262 | || kthread_should_stop()); | |
4263 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4264 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4265 | spin_unlock_irq(&conf->device_lock); |
4266 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4267 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4268 | } |
7a661381 | 4269 | return reshape_sectors; |
52c03291 N |
4270 | } |
4271 | ||
4272 | /* FIXME go_faster isn't used */ | |
fd01b88c | 4273 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
52c03291 | 4274 | { |
d1688a6d | 4275 | struct r5conf *conf = mddev->private; |
52c03291 | 4276 | struct stripe_head *sh; |
58c0fed4 | 4277 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 4278 | sector_t sync_blocks; |
16a53ecc N |
4279 | int still_degraded = 0; |
4280 | int i; | |
1da177e4 | 4281 | |
72626685 | 4282 | if (sector_nr >= max_sector) { |
1da177e4 | 4283 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 4284 | |
29269553 N |
4285 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4286 | end_reshape(conf); | |
4287 | return 0; | |
4288 | } | |
72626685 N |
4289 | |
4290 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4291 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4292 | &sync_blocks, 1); | |
16a53ecc | 4293 | else /* completed sync */ |
72626685 N |
4294 | conf->fullsync = 0; |
4295 | bitmap_close_sync(mddev->bitmap); | |
4296 | ||
1da177e4 LT |
4297 | return 0; |
4298 | } | |
ccfcc3c1 | 4299 | |
64bd660b N |
4300 | /* Allow raid5_quiesce to complete */ |
4301 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4302 | ||
52c03291 N |
4303 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4304 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4305 | |
c6207277 N |
4306 | /* No need to check resync_max as we never do more than one |
4307 | * stripe, and as resync_max will always be on a chunk boundary, | |
4308 | * if the check in md_do_sync didn't fire, there is no chance | |
4309 | * of overstepping resync_max here | |
4310 | */ | |
4311 | ||
16a53ecc | 4312 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4313 | * to resync, then assert that we are finished, because there is |
4314 | * nothing we can do. | |
4315 | */ | |
3285edf1 | 4316 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4317 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4318 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4319 | *skipped = 1; |
1da177e4 LT |
4320 | return rv; |
4321 | } | |
72626685 | 4322 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4323 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4324 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4325 | /* we can skip this block, and probably more */ | |
4326 | sync_blocks /= STRIPE_SECTORS; | |
4327 | *skipped = 1; | |
4328 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4329 | } | |
1da177e4 | 4330 | |
b47490c9 N |
4331 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
4332 | ||
a8c906ca | 4333 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4334 | if (sh == NULL) { |
a8c906ca | 4335 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4336 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4337 | * is trying to get access |
1da177e4 | 4338 | */ |
66c006a5 | 4339 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4340 | } |
16a53ecc N |
4341 | /* Need to check if array will still be degraded after recovery/resync |
4342 | * We don't need to check the 'failed' flag as when that gets set, | |
4343 | * recovery aborts. | |
4344 | */ | |
f001a70c | 4345 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4346 | if (conf->disks[i].rdev == NULL) |
4347 | still_degraded = 1; | |
4348 | ||
4349 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4350 | ||
83206d66 | 4351 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4352 | |
1442577b | 4353 | handle_stripe(sh); |
1da177e4 LT |
4354 | release_stripe(sh); |
4355 | ||
4356 | return STRIPE_SECTORS; | |
4357 | } | |
4358 | ||
d1688a6d | 4359 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
4360 | { |
4361 | /* We may not be able to submit a whole bio at once as there | |
4362 | * may not be enough stripe_heads available. | |
4363 | * We cannot pre-allocate enough stripe_heads as we may need | |
4364 | * more than exist in the cache (if we allow ever large chunks). | |
4365 | * So we do one stripe head at a time and record in | |
4366 | * ->bi_hw_segments how many have been done. | |
4367 | * | |
4368 | * We *know* that this entire raid_bio is in one chunk, so | |
4369 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4370 | */ | |
4371 | struct stripe_head *sh; | |
911d4ee8 | 4372 | int dd_idx; |
46031f9a RBJ |
4373 | sector_t sector, logical_sector, last_sector; |
4374 | int scnt = 0; | |
4375 | int remaining; | |
4376 | int handled = 0; | |
4377 | ||
4378 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4379 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4380 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4381 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4382 | ||
4383 | for (; logical_sector < last_sector; | |
387bb173 NB |
4384 | logical_sector += STRIPE_SECTORS, |
4385 | sector += STRIPE_SECTORS, | |
4386 | scnt++) { | |
46031f9a | 4387 | |
960e739d | 4388 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4389 | /* already done this stripe */ |
4390 | continue; | |
4391 | ||
a8c906ca | 4392 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4393 | |
4394 | if (!sh) { | |
4395 | /* failed to get a stripe - must wait */ | |
960e739d | 4396 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4397 | conf->retry_read_aligned = raid_bio; |
4398 | return handled; | |
4399 | } | |
4400 | ||
387bb173 NB |
4401 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4402 | release_stripe(sh); | |
960e739d | 4403 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4404 | conf->retry_read_aligned = raid_bio; |
4405 | return handled; | |
4406 | } | |
4407 | ||
36d1c647 | 4408 | handle_stripe(sh); |
46031f9a RBJ |
4409 | release_stripe(sh); |
4410 | handled++; | |
4411 | } | |
4412 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4413 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4414 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4415 | if (remaining == 0) |
4416 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4417 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4418 | wake_up(&conf->wait_for_stripe); | |
4419 | return handled; | |
4420 | } | |
4421 | ||
46031f9a | 4422 | |
1da177e4 LT |
4423 | /* |
4424 | * This is our raid5 kernel thread. | |
4425 | * | |
4426 | * We scan the hash table for stripes which can be handled now. | |
4427 | * During the scan, completed stripes are saved for us by the interrupt | |
4428 | * handler, so that they will not have to wait for our next wakeup. | |
4429 | */ | |
fd01b88c | 4430 | static void raid5d(struct mddev *mddev) |
1da177e4 LT |
4431 | { |
4432 | struct stripe_head *sh; | |
d1688a6d | 4433 | struct r5conf *conf = mddev->private; |
1da177e4 | 4434 | int handled; |
e1dfa0a2 | 4435 | struct blk_plug plug; |
1da177e4 | 4436 | |
45b4233c | 4437 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4438 | |
4439 | md_check_recovery(mddev); | |
1da177e4 | 4440 | |
e1dfa0a2 | 4441 | blk_start_plug(&plug); |
1da177e4 LT |
4442 | handled = 0; |
4443 | spin_lock_irq(&conf->device_lock); | |
4444 | while (1) { | |
46031f9a | 4445 | struct bio *bio; |
1da177e4 | 4446 | |
7c13edc8 N |
4447 | if (atomic_read(&mddev->plug_cnt) == 0 && |
4448 | !list_empty(&conf->bitmap_list)) { | |
4449 | /* Now is a good time to flush some bitmap updates */ | |
4450 | conf->seq_flush++; | |
700e432d | 4451 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4452 | bitmap_unplug(mddev->bitmap); |
700e432d | 4453 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4454 | conf->seq_write = conf->seq_flush; |
72626685 N |
4455 | activate_bit_delay(conf); |
4456 | } | |
7c13edc8 N |
4457 | if (atomic_read(&mddev->plug_cnt) == 0) |
4458 | raid5_activate_delayed(conf); | |
72626685 | 4459 | |
46031f9a RBJ |
4460 | while ((bio = remove_bio_from_retry(conf))) { |
4461 | int ok; | |
4462 | spin_unlock_irq(&conf->device_lock); | |
4463 | ok = retry_aligned_read(conf, bio); | |
4464 | spin_lock_irq(&conf->device_lock); | |
4465 | if (!ok) | |
4466 | break; | |
4467 | handled++; | |
4468 | } | |
4469 | ||
8b3e6cdc DW |
4470 | sh = __get_priority_stripe(conf); |
4471 | ||
c9f21aaf | 4472 | if (!sh) |
1da177e4 | 4473 | break; |
1da177e4 LT |
4474 | spin_unlock_irq(&conf->device_lock); |
4475 | ||
4476 | handled++; | |
417b8d4a DW |
4477 | handle_stripe(sh); |
4478 | release_stripe(sh); | |
4479 | cond_resched(); | |
1da177e4 | 4480 | |
de393cde N |
4481 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
4482 | md_check_recovery(mddev); | |
4483 | ||
1da177e4 LT |
4484 | spin_lock_irq(&conf->device_lock); |
4485 | } | |
45b4233c | 4486 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4487 | |
4488 | spin_unlock_irq(&conf->device_lock); | |
4489 | ||
c9f21aaf | 4490 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4491 | blk_finish_plug(&plug); |
1da177e4 | 4492 | |
45b4233c | 4493 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4494 | } |
4495 | ||
3f294f4f | 4496 | static ssize_t |
fd01b88c | 4497 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 4498 | { |
d1688a6d | 4499 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4500 | if (conf) |
4501 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4502 | else | |
4503 | return 0; | |
3f294f4f N |
4504 | } |
4505 | ||
c41d4ac4 | 4506 | int |
fd01b88c | 4507 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 4508 | { |
d1688a6d | 4509 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
4510 | int err; |
4511 | ||
c41d4ac4 | 4512 | if (size <= 16 || size > 32768) |
3f294f4f | 4513 | return -EINVAL; |
c41d4ac4 | 4514 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4515 | if (drop_one_stripe(conf)) |
4516 | conf->max_nr_stripes--; | |
4517 | else | |
4518 | break; | |
4519 | } | |
b5470dc5 DW |
4520 | err = md_allow_write(mddev); |
4521 | if (err) | |
4522 | return err; | |
c41d4ac4 | 4523 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4524 | if (grow_one_stripe(conf)) |
4525 | conf->max_nr_stripes++; | |
4526 | else break; | |
4527 | } | |
c41d4ac4 N |
4528 | return 0; |
4529 | } | |
4530 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4531 | ||
4532 | static ssize_t | |
fd01b88c | 4533 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 4534 | { |
d1688a6d | 4535 | struct r5conf *conf = mddev->private; |
c41d4ac4 N |
4536 | unsigned long new; |
4537 | int err; | |
4538 | ||
4539 | if (len >= PAGE_SIZE) | |
4540 | return -EINVAL; | |
4541 | if (!conf) | |
4542 | return -ENODEV; | |
4543 | ||
4544 | if (strict_strtoul(page, 10, &new)) | |
4545 | return -EINVAL; | |
4546 | err = raid5_set_cache_size(mddev, new); | |
4547 | if (err) | |
4548 | return err; | |
3f294f4f N |
4549 | return len; |
4550 | } | |
007583c9 | 4551 | |
96de1e66 N |
4552 | static struct md_sysfs_entry |
4553 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4554 | raid5_show_stripe_cache_size, | |
4555 | raid5_store_stripe_cache_size); | |
3f294f4f | 4556 | |
8b3e6cdc | 4557 | static ssize_t |
fd01b88c | 4558 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 4559 | { |
d1688a6d | 4560 | struct r5conf *conf = mddev->private; |
8b3e6cdc DW |
4561 | if (conf) |
4562 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4563 | else | |
4564 | return 0; | |
4565 | } | |
4566 | ||
4567 | static ssize_t | |
fd01b88c | 4568 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 4569 | { |
d1688a6d | 4570 | struct r5conf *conf = mddev->private; |
4ef197d8 | 4571 | unsigned long new; |
8b3e6cdc DW |
4572 | if (len >= PAGE_SIZE) |
4573 | return -EINVAL; | |
4574 | if (!conf) | |
4575 | return -ENODEV; | |
4576 | ||
4ef197d8 | 4577 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4578 | return -EINVAL; |
4ef197d8 | 4579 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4580 | return -EINVAL; |
4581 | conf->bypass_threshold = new; | |
4582 | return len; | |
4583 | } | |
4584 | ||
4585 | static struct md_sysfs_entry | |
4586 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4587 | S_IRUGO | S_IWUSR, | |
4588 | raid5_show_preread_threshold, | |
4589 | raid5_store_preread_threshold); | |
4590 | ||
3f294f4f | 4591 | static ssize_t |
fd01b88c | 4592 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 4593 | { |
d1688a6d | 4594 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4595 | if (conf) |
4596 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4597 | else | |
4598 | return 0; | |
3f294f4f N |
4599 | } |
4600 | ||
96de1e66 N |
4601 | static struct md_sysfs_entry |
4602 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4603 | |
007583c9 | 4604 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4605 | &raid5_stripecache_size.attr, |
4606 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4607 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4608 | NULL, |
4609 | }; | |
007583c9 N |
4610 | static struct attribute_group raid5_attrs_group = { |
4611 | .name = NULL, | |
4612 | .attrs = raid5_attrs, | |
3f294f4f N |
4613 | }; |
4614 | ||
80c3a6ce | 4615 | static sector_t |
fd01b88c | 4616 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 4617 | { |
d1688a6d | 4618 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
4619 | |
4620 | if (!sectors) | |
4621 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4622 | if (!raid_disks) |
7ec05478 | 4623 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4624 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4625 | |
9d8f0363 | 4626 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4627 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4628 | return sectors * (raid_disks - conf->max_degraded); |
4629 | } | |
4630 | ||
d1688a6d | 4631 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 DW |
4632 | { |
4633 | struct raid5_percpu *percpu; | |
4634 | unsigned long cpu; | |
4635 | ||
4636 | if (!conf->percpu) | |
4637 | return; | |
4638 | ||
4639 | get_online_cpus(); | |
4640 | for_each_possible_cpu(cpu) { | |
4641 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4642 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4643 | kfree(percpu->scribble); |
36d1c647 DW |
4644 | } |
4645 | #ifdef CONFIG_HOTPLUG_CPU | |
4646 | unregister_cpu_notifier(&conf->cpu_notify); | |
4647 | #endif | |
4648 | put_online_cpus(); | |
4649 | ||
4650 | free_percpu(conf->percpu); | |
4651 | } | |
4652 | ||
d1688a6d | 4653 | static void free_conf(struct r5conf *conf) |
95fc17aa DW |
4654 | { |
4655 | shrink_stripes(conf); | |
36d1c647 | 4656 | raid5_free_percpu(conf); |
95fc17aa DW |
4657 | kfree(conf->disks); |
4658 | kfree(conf->stripe_hashtbl); | |
4659 | kfree(conf); | |
4660 | } | |
4661 | ||
36d1c647 DW |
4662 | #ifdef CONFIG_HOTPLUG_CPU |
4663 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4664 | void *hcpu) | |
4665 | { | |
d1688a6d | 4666 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
4667 | long cpu = (long)hcpu; |
4668 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4669 | ||
4670 | switch (action) { | |
4671 | case CPU_UP_PREPARE: | |
4672 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4673 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4674 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4675 | if (!percpu->scribble) |
4676 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4677 | ||
4678 | if (!percpu->scribble || | |
4679 | (conf->level == 6 && !percpu->spare_page)) { | |
4680 | safe_put_page(percpu->spare_page); | |
4681 | kfree(percpu->scribble); | |
36d1c647 DW |
4682 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4683 | __func__, cpu); | |
55af6bb5 | 4684 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4685 | } |
4686 | break; | |
4687 | case CPU_DEAD: | |
4688 | case CPU_DEAD_FROZEN: | |
4689 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4690 | kfree(percpu->scribble); |
36d1c647 | 4691 | percpu->spare_page = NULL; |
d6f38f31 | 4692 | percpu->scribble = NULL; |
36d1c647 DW |
4693 | break; |
4694 | default: | |
4695 | break; | |
4696 | } | |
4697 | return NOTIFY_OK; | |
4698 | } | |
4699 | #endif | |
4700 | ||
d1688a6d | 4701 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
4702 | { |
4703 | unsigned long cpu; | |
4704 | struct page *spare_page; | |
a29d8b8e | 4705 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4706 | void *scribble; |
36d1c647 DW |
4707 | int err; |
4708 | ||
36d1c647 DW |
4709 | allcpus = alloc_percpu(struct raid5_percpu); |
4710 | if (!allcpus) | |
4711 | return -ENOMEM; | |
4712 | conf->percpu = allcpus; | |
4713 | ||
4714 | get_online_cpus(); | |
4715 | err = 0; | |
4716 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4717 | if (conf->level == 6) { |
4718 | spare_page = alloc_page(GFP_KERNEL); | |
4719 | if (!spare_page) { | |
4720 | err = -ENOMEM; | |
4721 | break; | |
4722 | } | |
4723 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4724 | } | |
5e5e3e78 | 4725 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4726 | if (!scribble) { |
36d1c647 DW |
4727 | err = -ENOMEM; |
4728 | break; | |
4729 | } | |
d6f38f31 | 4730 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4731 | } |
4732 | #ifdef CONFIG_HOTPLUG_CPU | |
4733 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4734 | conf->cpu_notify.priority = 0; | |
4735 | if (err == 0) | |
4736 | err = register_cpu_notifier(&conf->cpu_notify); | |
4737 | #endif | |
4738 | put_online_cpus(); | |
4739 | ||
4740 | return err; | |
4741 | } | |
4742 | ||
d1688a6d | 4743 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 4744 | { |
d1688a6d | 4745 | struct r5conf *conf; |
5e5e3e78 | 4746 | int raid_disk, memory, max_disks; |
3cb03002 | 4747 | struct md_rdev *rdev; |
1da177e4 | 4748 | struct disk_info *disk; |
1da177e4 | 4749 | |
91adb564 N |
4750 | if (mddev->new_level != 5 |
4751 | && mddev->new_level != 4 | |
4752 | && mddev->new_level != 6) { | |
0c55e022 | 4753 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4754 | mdname(mddev), mddev->new_level); |
4755 | return ERR_PTR(-EIO); | |
1da177e4 | 4756 | } |
91adb564 N |
4757 | if ((mddev->new_level == 5 |
4758 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4759 | (mddev->new_level == 6 | |
4760 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4761 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4762 | mdname(mddev), mddev->new_layout); |
4763 | return ERR_PTR(-EIO); | |
99c0fb5f | 4764 | } |
91adb564 | 4765 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4766 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4767 | mdname(mddev), mddev->raid_disks); |
4768 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4769 | } |
4770 | ||
664e7c41 AN |
4771 | if (!mddev->new_chunk_sectors || |
4772 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4773 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4774 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4775 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4776 | return ERR_PTR(-EINVAL); |
f6705578 N |
4777 | } |
4778 | ||
d1688a6d | 4779 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 4780 | if (conf == NULL) |
1da177e4 | 4781 | goto abort; |
f5efd45a DW |
4782 | spin_lock_init(&conf->device_lock); |
4783 | init_waitqueue_head(&conf->wait_for_stripe); | |
4784 | init_waitqueue_head(&conf->wait_for_overlap); | |
4785 | INIT_LIST_HEAD(&conf->handle_list); | |
4786 | INIT_LIST_HEAD(&conf->hold_list); | |
4787 | INIT_LIST_HEAD(&conf->delayed_list); | |
4788 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4789 | INIT_LIST_HEAD(&conf->inactive_list); | |
4790 | atomic_set(&conf->active_stripes, 0); | |
4791 | atomic_set(&conf->preread_active_stripes, 0); | |
4792 | atomic_set(&conf->active_aligned_reads, 0); | |
4793 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 4794 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
4795 | |
4796 | conf->raid_disks = mddev->raid_disks; | |
4797 | if (mddev->reshape_position == MaxSector) | |
4798 | conf->previous_raid_disks = mddev->raid_disks; | |
4799 | else | |
f6705578 | 4800 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4801 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4802 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4803 | |
5e5e3e78 | 4804 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4805 | GFP_KERNEL); |
4806 | if (!conf->disks) | |
4807 | goto abort; | |
9ffae0cf | 4808 | |
1da177e4 LT |
4809 | conf->mddev = mddev; |
4810 | ||
fccddba0 | 4811 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4812 | goto abort; |
1da177e4 | 4813 | |
36d1c647 DW |
4814 | conf->level = mddev->new_level; |
4815 | if (raid5_alloc_percpu(conf) != 0) | |
4816 | goto abort; | |
4817 | ||
0c55e022 | 4818 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4819 | |
159ec1fc | 4820 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4821 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4822 | if (raid_disk >= max_disks |
1da177e4 LT |
4823 | || raid_disk < 0) |
4824 | continue; | |
4825 | disk = conf->disks + raid_disk; | |
4826 | ||
4827 | disk->rdev = rdev; | |
4828 | ||
b2d444d7 | 4829 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 4830 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
4831 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
4832 | " disk %d\n", | |
4833 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 4834 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
4835 | /* Cannot rely on bitmap to complete recovery */ |
4836 | conf->fullsync = 1; | |
1da177e4 LT |
4837 | } |
4838 | ||
09c9e5fa | 4839 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4840 | conf->level = mddev->new_level; |
16a53ecc N |
4841 | if (conf->level == 6) |
4842 | conf->max_degraded = 2; | |
4843 | else | |
4844 | conf->max_degraded = 1; | |
91adb564 | 4845 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4846 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4847 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4848 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4849 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4850 | conf->prev_algo = mddev->layout; |
4851 | } | |
1da177e4 | 4852 | |
91adb564 | 4853 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4854 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4855 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4856 | printk(KERN_ERR | |
0c55e022 N |
4857 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
4858 | mdname(mddev), memory); | |
91adb564 N |
4859 | goto abort; |
4860 | } else | |
0c55e022 N |
4861 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
4862 | mdname(mddev), memory); | |
1da177e4 | 4863 | |
0da3c619 | 4864 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4865 | if (!conf->thread) { |
4866 | printk(KERN_ERR | |
0c55e022 | 4867 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 4868 | mdname(mddev)); |
16a53ecc N |
4869 | goto abort; |
4870 | } | |
91adb564 N |
4871 | |
4872 | return conf; | |
4873 | ||
4874 | abort: | |
4875 | if (conf) { | |
95fc17aa | 4876 | free_conf(conf); |
91adb564 N |
4877 | return ERR_PTR(-EIO); |
4878 | } else | |
4879 | return ERR_PTR(-ENOMEM); | |
4880 | } | |
4881 | ||
c148ffdc N |
4882 | |
4883 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4884 | { | |
4885 | switch (algo) { | |
4886 | case ALGORITHM_PARITY_0: | |
4887 | if (raid_disk < max_degraded) | |
4888 | return 1; | |
4889 | break; | |
4890 | case ALGORITHM_PARITY_N: | |
4891 | if (raid_disk >= raid_disks - max_degraded) | |
4892 | return 1; | |
4893 | break; | |
4894 | case ALGORITHM_PARITY_0_6: | |
4895 | if (raid_disk == 0 || | |
4896 | raid_disk == raid_disks - 1) | |
4897 | return 1; | |
4898 | break; | |
4899 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4900 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4901 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4902 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4903 | if (raid_disk == raid_disks - 1) | |
4904 | return 1; | |
4905 | } | |
4906 | return 0; | |
4907 | } | |
4908 | ||
fd01b88c | 4909 | static int run(struct mddev *mddev) |
91adb564 | 4910 | { |
d1688a6d | 4911 | struct r5conf *conf; |
9f7c2220 | 4912 | int working_disks = 0; |
c148ffdc | 4913 | int dirty_parity_disks = 0; |
3cb03002 | 4914 | struct md_rdev *rdev; |
c148ffdc | 4915 | sector_t reshape_offset = 0; |
91adb564 | 4916 | |
8c6ac868 | 4917 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 4918 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
4919 | " -- starting background reconstruction\n", |
4920 | mdname(mddev)); | |
91adb564 N |
4921 | if (mddev->reshape_position != MaxSector) { |
4922 | /* Check that we can continue the reshape. | |
4923 | * Currently only disks can change, it must | |
4924 | * increase, and we must be past the point where | |
4925 | * a stripe over-writes itself | |
4926 | */ | |
4927 | sector_t here_new, here_old; | |
4928 | int old_disks; | |
18b00334 | 4929 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4930 | |
88ce4930 | 4931 | if (mddev->new_level != mddev->level) { |
0c55e022 | 4932 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
4933 | "required - aborting.\n", |
4934 | mdname(mddev)); | |
4935 | return -EINVAL; | |
4936 | } | |
91adb564 N |
4937 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4938 | /* reshape_position must be on a new-stripe boundary, and one | |
4939 | * further up in new geometry must map after here in old | |
4940 | * geometry. | |
4941 | */ | |
4942 | here_new = mddev->reshape_position; | |
664e7c41 | 4943 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 4944 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
4945 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
4946 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
4947 | return -EINVAL; |
4948 | } | |
c148ffdc | 4949 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4950 | /* here_new is the stripe we will write to */ |
4951 | here_old = mddev->reshape_position; | |
9d8f0363 | 4952 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4953 | (old_disks-max_degraded)); |
4954 | /* here_old is the first stripe that we might need to read | |
4955 | * from */ | |
67ac6011 N |
4956 | if (mddev->delta_disks == 0) { |
4957 | /* We cannot be sure it is safe to start an in-place | |
4958 | * reshape. It is only safe if user-space if monitoring | |
4959 | * and taking constant backups. | |
4960 | * mdadm always starts a situation like this in | |
4961 | * readonly mode so it can take control before | |
4962 | * allowing any writes. So just check for that. | |
4963 | */ | |
4964 | if ((here_new * mddev->new_chunk_sectors != | |
4965 | here_old * mddev->chunk_sectors) || | |
4966 | mddev->ro == 0) { | |
0c55e022 N |
4967 | printk(KERN_ERR "md/raid:%s: in-place reshape must be started" |
4968 | " in read-only mode - aborting\n", | |
4969 | mdname(mddev)); | |
67ac6011 N |
4970 | return -EINVAL; |
4971 | } | |
4972 | } else if (mddev->delta_disks < 0 | |
4973 | ? (here_new * mddev->new_chunk_sectors <= | |
4974 | here_old * mddev->chunk_sectors) | |
4975 | : (here_new * mddev->new_chunk_sectors >= | |
4976 | here_old * mddev->chunk_sectors)) { | |
91adb564 | 4977 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
4978 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
4979 | "auto-recovery - aborting.\n", | |
4980 | mdname(mddev)); | |
91adb564 N |
4981 | return -EINVAL; |
4982 | } | |
0c55e022 N |
4983 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
4984 | mdname(mddev)); | |
91adb564 N |
4985 | /* OK, we should be able to continue; */ |
4986 | } else { | |
4987 | BUG_ON(mddev->level != mddev->new_level); | |
4988 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 4989 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 4990 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 4991 | } |
91adb564 | 4992 | |
245f46c2 N |
4993 | if (mddev->private == NULL) |
4994 | conf = setup_conf(mddev); | |
4995 | else | |
4996 | conf = mddev->private; | |
4997 | ||
91adb564 N |
4998 | if (IS_ERR(conf)) |
4999 | return PTR_ERR(conf); | |
5000 | ||
5001 | mddev->thread = conf->thread; | |
5002 | conf->thread = NULL; | |
5003 | mddev->private = conf; | |
5004 | ||
5005 | /* | |
5006 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
5007 | */ | |
c148ffdc N |
5008 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
5009 | if (rdev->raid_disk < 0) | |
5010 | continue; | |
2f115882 | 5011 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 5012 | working_disks++; |
2f115882 N |
5013 | continue; |
5014 | } | |
c148ffdc N |
5015 | /* This disc is not fully in-sync. However if it |
5016 | * just stored parity (beyond the recovery_offset), | |
5017 | * when we don't need to be concerned about the | |
5018 | * array being dirty. | |
5019 | * When reshape goes 'backwards', we never have | |
5020 | * partially completed devices, so we only need | |
5021 | * to worry about reshape going forwards. | |
5022 | */ | |
5023 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
5024 | if (mddev->major_version == 0 && | |
5025 | mddev->minor_version > 90) | |
5026 | rdev->recovery_offset = reshape_offset; | |
5027 | ||
c148ffdc N |
5028 | if (rdev->recovery_offset < reshape_offset) { |
5029 | /* We need to check old and new layout */ | |
5030 | if (!only_parity(rdev->raid_disk, | |
5031 | conf->algorithm, | |
5032 | conf->raid_disks, | |
5033 | conf->max_degraded)) | |
5034 | continue; | |
5035 | } | |
5036 | if (!only_parity(rdev->raid_disk, | |
5037 | conf->prev_algo, | |
5038 | conf->previous_raid_disks, | |
5039 | conf->max_degraded)) | |
5040 | continue; | |
5041 | dirty_parity_disks++; | |
5042 | } | |
91adb564 | 5043 | |
908f4fbd | 5044 | mddev->degraded = calc_degraded(conf); |
91adb564 | 5045 | |
674806d6 | 5046 | if (has_failed(conf)) { |
0c55e022 | 5047 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 5048 | " (%d/%d failed)\n", |
02c2de8c | 5049 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
5050 | goto abort; |
5051 | } | |
5052 | ||
91adb564 | 5053 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 5054 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
5055 | mddev->resync_max_sectors = mddev->dev_sectors; |
5056 | ||
c148ffdc | 5057 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 5058 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
5059 | if (mddev->ok_start_degraded) |
5060 | printk(KERN_WARNING | |
0c55e022 N |
5061 | "md/raid:%s: starting dirty degraded array" |
5062 | " - data corruption possible.\n", | |
6ff8d8ec N |
5063 | mdname(mddev)); |
5064 | else { | |
5065 | printk(KERN_ERR | |
0c55e022 | 5066 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
5067 | mdname(mddev)); |
5068 | goto abort; | |
5069 | } | |
1da177e4 LT |
5070 | } |
5071 | ||
1da177e4 | 5072 | if (mddev->degraded == 0) |
0c55e022 N |
5073 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
5074 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
5075 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
5076 | mddev->new_layout); | |
1da177e4 | 5077 | else |
0c55e022 N |
5078 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
5079 | " out of %d devices, algorithm %d\n", | |
5080 | mdname(mddev), conf->level, | |
5081 | mddev->raid_disks - mddev->degraded, | |
5082 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
5083 | |
5084 | print_raid5_conf(conf); | |
5085 | ||
fef9c61f | 5086 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 5087 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5088 | atomic_set(&conf->reshape_stripes, 0); |
5089 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5090 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5091 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5092 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5093 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5094 | "reshape"); |
f6705578 N |
5095 | } |
5096 | ||
1da177e4 LT |
5097 | |
5098 | /* Ok, everything is just fine now */ | |
a64c876f N |
5099 | if (mddev->to_remove == &raid5_attrs_group) |
5100 | mddev->to_remove = NULL; | |
00bcb4ac N |
5101 | else if (mddev->kobj.sd && |
5102 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 5103 | printk(KERN_WARNING |
4a5add49 | 5104 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 5105 | mdname(mddev)); |
4a5add49 | 5106 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5107 | |
4a5add49 | 5108 | if (mddev->queue) { |
9f7c2220 | 5109 | int chunk_size; |
4a5add49 N |
5110 | /* read-ahead size must cover two whole stripes, which |
5111 | * is 2 * (datadisks) * chunksize where 'n' is the | |
5112 | * number of raid devices | |
5113 | */ | |
5114 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
5115 | int stripe = data_disks * | |
5116 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
5117 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
5118 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 5119 | |
4a5add49 | 5120 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 5121 | |
11d8a6e3 N |
5122 | mddev->queue->backing_dev_info.congested_data = mddev; |
5123 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 5124 | |
9f7c2220 N |
5125 | chunk_size = mddev->chunk_sectors << 9; |
5126 | blk_queue_io_min(mddev->queue, chunk_size); | |
5127 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5128 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 5129 | |
9f7c2220 N |
5130 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5131 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
5132 | rdev->data_offset << 9); | |
5133 | } | |
23032a0e | 5134 | |
1da177e4 LT |
5135 | return 0; |
5136 | abort: | |
01f96c0a | 5137 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
5138 | print_raid5_conf(conf); |
5139 | free_conf(conf); | |
1da177e4 | 5140 | mddev->private = NULL; |
0c55e022 | 5141 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
5142 | return -EIO; |
5143 | } | |
5144 | ||
fd01b88c | 5145 | static int stop(struct mddev *mddev) |
1da177e4 | 5146 | { |
d1688a6d | 5147 | struct r5conf *conf = mddev->private; |
1da177e4 | 5148 | |
01f96c0a | 5149 | md_unregister_thread(&mddev->thread); |
11d8a6e3 N |
5150 | if (mddev->queue) |
5151 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 5152 | free_conf(conf); |
a64c876f N |
5153 | mddev->private = NULL; |
5154 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
5155 | return 0; |
5156 | } | |
5157 | ||
fd01b88c | 5158 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 5159 | { |
d1688a6d | 5160 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
5161 | int i; |
5162 | ||
9d8f0363 AN |
5163 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5164 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5165 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5166 | for (i = 0; i < conf->raid_disks; i++) |
5167 | seq_printf (seq, "%s", | |
5168 | conf->disks[i].rdev && | |
b2d444d7 | 5169 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5170 | seq_printf (seq, "]"); |
1da177e4 LT |
5171 | } |
5172 | ||
d1688a6d | 5173 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
5174 | { |
5175 | int i; | |
5176 | struct disk_info *tmp; | |
5177 | ||
0c55e022 | 5178 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
5179 | if (!conf) { |
5180 | printk("(conf==NULL)\n"); | |
5181 | return; | |
5182 | } | |
0c55e022 N |
5183 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
5184 | conf->raid_disks, | |
5185 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5186 | |
5187 | for (i = 0; i < conf->raid_disks; i++) { | |
5188 | char b[BDEVNAME_SIZE]; | |
5189 | tmp = conf->disks + i; | |
5190 | if (tmp->rdev) | |
0c55e022 N |
5191 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
5192 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
5193 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
5194 | } |
5195 | } | |
5196 | ||
fd01b88c | 5197 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
5198 | { |
5199 | int i; | |
d1688a6d | 5200 | struct r5conf *conf = mddev->private; |
1da177e4 | 5201 | struct disk_info *tmp; |
6b965620 N |
5202 | int count = 0; |
5203 | unsigned long flags; | |
1da177e4 LT |
5204 | |
5205 | for (i = 0; i < conf->raid_disks; i++) { | |
5206 | tmp = conf->disks + i; | |
5207 | if (tmp->rdev | |
70fffd0b | 5208 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 5209 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 5210 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 5211 | count++; |
43c73ca4 | 5212 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
5213 | } |
5214 | } | |
6b965620 | 5215 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5216 | mddev->degraded = calc_degraded(conf); |
6b965620 | 5217 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 5218 | print_raid5_conf(conf); |
6b965620 | 5219 | return count; |
1da177e4 LT |
5220 | } |
5221 | ||
b8321b68 | 5222 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5223 | { |
d1688a6d | 5224 | struct r5conf *conf = mddev->private; |
1da177e4 | 5225 | int err = 0; |
b8321b68 | 5226 | int number = rdev->raid_disk; |
657e3e4d | 5227 | struct md_rdev **rdevp; |
1da177e4 LT |
5228 | struct disk_info *p = conf->disks + number; |
5229 | ||
5230 | print_raid5_conf(conf); | |
657e3e4d N |
5231 | if (rdev == p->rdev) |
5232 | rdevp = &p->rdev; | |
5233 | else if (rdev == p->replacement) | |
5234 | rdevp = &p->replacement; | |
5235 | else | |
5236 | return 0; | |
5237 | ||
5238 | if (number >= conf->raid_disks && | |
5239 | conf->reshape_progress == MaxSector) | |
5240 | clear_bit(In_sync, &rdev->flags); | |
5241 | ||
5242 | if (test_bit(In_sync, &rdev->flags) || | |
5243 | atomic_read(&rdev->nr_pending)) { | |
5244 | err = -EBUSY; | |
5245 | goto abort; | |
5246 | } | |
5247 | /* Only remove non-faulty devices if recovery | |
5248 | * isn't possible. | |
5249 | */ | |
5250 | if (!test_bit(Faulty, &rdev->flags) && | |
5251 | mddev->recovery_disabled != conf->recovery_disabled && | |
5252 | !has_failed(conf) && | |
5253 | number < conf->raid_disks) { | |
5254 | err = -EBUSY; | |
5255 | goto abort; | |
5256 | } | |
5257 | *rdevp = NULL; | |
5258 | synchronize_rcu(); | |
5259 | if (atomic_read(&rdev->nr_pending)) { | |
5260 | /* lost the race, try later */ | |
5261 | err = -EBUSY; | |
5262 | *rdevp = rdev; | |
1da177e4 LT |
5263 | } |
5264 | abort: | |
5265 | ||
5266 | print_raid5_conf(conf); | |
5267 | return err; | |
5268 | } | |
5269 | ||
fd01b88c | 5270 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5271 | { |
d1688a6d | 5272 | struct r5conf *conf = mddev->private; |
199050ea | 5273 | int err = -EEXIST; |
1da177e4 LT |
5274 | int disk; |
5275 | struct disk_info *p; | |
6c2fce2e NB |
5276 | int first = 0; |
5277 | int last = conf->raid_disks - 1; | |
1da177e4 | 5278 | |
7f0da59b N |
5279 | if (mddev->recovery_disabled == conf->recovery_disabled) |
5280 | return -EBUSY; | |
5281 | ||
674806d6 | 5282 | if (has_failed(conf)) |
1da177e4 | 5283 | /* no point adding a device */ |
199050ea | 5284 | return -EINVAL; |
1da177e4 | 5285 | |
6c2fce2e NB |
5286 | if (rdev->raid_disk >= 0) |
5287 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5288 | |
5289 | /* | |
16a53ecc N |
5290 | * find the disk ... but prefer rdev->saved_raid_disk |
5291 | * if possible. | |
1da177e4 | 5292 | */ |
16a53ecc | 5293 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5294 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5295 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5296 | disk = rdev->saved_raid_disk; | |
5297 | else | |
6c2fce2e NB |
5298 | disk = first; |
5299 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5300 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5301 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5302 | rdev->raid_disk = disk; |
199050ea | 5303 | err = 0; |
72626685 N |
5304 | if (rdev->saved_raid_disk != disk) |
5305 | conf->fullsync = 1; | |
d6065f7b | 5306 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5307 | break; |
5308 | } | |
5309 | print_raid5_conf(conf); | |
199050ea | 5310 | return err; |
1da177e4 LT |
5311 | } |
5312 | ||
fd01b88c | 5313 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
5314 | { |
5315 | /* no resync is happening, and there is enough space | |
5316 | * on all devices, so we can resize. | |
5317 | * We need to make sure resync covers any new space. | |
5318 | * If the array is shrinking we should possibly wait until | |
5319 | * any io in the removed space completes, but it hardly seems | |
5320 | * worth it. | |
5321 | */ | |
9d8f0363 | 5322 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5323 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5324 | mddev->raid_disks)); | |
b522adcd DW |
5325 | if (mddev->array_sectors > |
5326 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5327 | return -EINVAL; | |
f233ea5c | 5328 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5329 | revalidate_disk(mddev->gendisk); |
b098636c N |
5330 | if (sectors > mddev->dev_sectors && |
5331 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5332 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5333 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5334 | } | |
58c0fed4 | 5335 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5336 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5337 | return 0; |
5338 | } | |
5339 | ||
fd01b88c | 5340 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
5341 | { |
5342 | /* Can only proceed if there are plenty of stripe_heads. | |
5343 | * We need a minimum of one full stripe,, and for sensible progress | |
5344 | * it is best to have about 4 times that. | |
5345 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5346 | * allow for chunk sizes up to 256K, which is probably OK. | |
5347 | * If the chunk size is greater, user-space should request more | |
5348 | * stripe_heads first. | |
5349 | */ | |
d1688a6d | 5350 | struct r5conf *conf = mddev->private; |
01ee22b4 N |
5351 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
5352 | > conf->max_nr_stripes || | |
5353 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5354 | > conf->max_nr_stripes) { | |
0c55e022 N |
5355 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5356 | mdname(mddev), | |
01ee22b4 N |
5357 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5358 | / STRIPE_SIZE)*4); | |
5359 | return 0; | |
5360 | } | |
5361 | return 1; | |
5362 | } | |
5363 | ||
fd01b88c | 5364 | static int check_reshape(struct mddev *mddev) |
29269553 | 5365 | { |
d1688a6d | 5366 | struct r5conf *conf = mddev->private; |
29269553 | 5367 | |
88ce4930 N |
5368 | if (mddev->delta_disks == 0 && |
5369 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5370 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5371 | return 0; /* nothing to do */ |
dba034ee N |
5372 | if (mddev->bitmap) |
5373 | /* Cannot grow a bitmap yet */ | |
5374 | return -EBUSY; | |
674806d6 | 5375 | if (has_failed(conf)) |
ec32a2bd N |
5376 | return -EINVAL; |
5377 | if (mddev->delta_disks < 0) { | |
5378 | /* We might be able to shrink, but the devices must | |
5379 | * be made bigger first. | |
5380 | * For raid6, 4 is the minimum size. | |
5381 | * Otherwise 2 is the minimum | |
5382 | */ | |
5383 | int min = 2; | |
5384 | if (mddev->level == 6) | |
5385 | min = 4; | |
5386 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5387 | return -EINVAL; | |
5388 | } | |
29269553 | 5389 | |
01ee22b4 | 5390 | if (!check_stripe_cache(mddev)) |
29269553 | 5391 | return -ENOSPC; |
29269553 | 5392 | |
ec32a2bd | 5393 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5394 | } |
5395 | ||
fd01b88c | 5396 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 5397 | { |
d1688a6d | 5398 | struct r5conf *conf = mddev->private; |
3cb03002 | 5399 | struct md_rdev *rdev; |
63c70c4f | 5400 | int spares = 0; |
c04be0aa | 5401 | unsigned long flags; |
63c70c4f | 5402 | |
f416885e | 5403 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5404 | return -EBUSY; |
5405 | ||
01ee22b4 N |
5406 | if (!check_stripe_cache(mddev)) |
5407 | return -ENOSPC; | |
5408 | ||
159ec1fc | 5409 | list_for_each_entry(rdev, &mddev->disks, same_set) |
469518a3 N |
5410 | if (!test_bit(In_sync, &rdev->flags) |
5411 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5412 | spares++; |
63c70c4f | 5413 | |
f416885e | 5414 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5415 | /* Not enough devices even to make a degraded array |
5416 | * of that size | |
5417 | */ | |
5418 | return -EINVAL; | |
5419 | ||
ec32a2bd N |
5420 | /* Refuse to reduce size of the array. Any reductions in |
5421 | * array size must be through explicit setting of array_size | |
5422 | * attribute. | |
5423 | */ | |
5424 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5425 | < mddev->array_sectors) { | |
0c55e022 | 5426 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5427 | "before number of disks\n", mdname(mddev)); |
5428 | return -EINVAL; | |
5429 | } | |
5430 | ||
f6705578 | 5431 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5432 | spin_lock_irq(&conf->device_lock); |
5433 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5434 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5435 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5436 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5437 | conf->prev_algo = conf->algorithm; |
5438 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5439 | if (mddev->delta_disks < 0) |
5440 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5441 | else | |
5442 | conf->reshape_progress = 0; | |
5443 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5444 | conf->generation++; |
29269553 N |
5445 | spin_unlock_irq(&conf->device_lock); |
5446 | ||
5447 | /* Add some new drives, as many as will fit. | |
5448 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5449 | * Don't add devices if we are reducing the number of |
5450 | * devices in the array. This is because it is not possible | |
5451 | * to correctly record the "partially reconstructed" state of | |
5452 | * such devices during the reshape and confusion could result. | |
29269553 | 5453 | */ |
87a8dec9 N |
5454 | if (mddev->delta_disks >= 0) { |
5455 | int added_devices = 0; | |
5456 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5457 | if (rdev->raid_disk < 0 && | |
5458 | !test_bit(Faulty, &rdev->flags)) { | |
5459 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 N |
5460 | if (rdev->raid_disk |
5461 | >= conf->previous_raid_disks) { | |
5462 | set_bit(In_sync, &rdev->flags); | |
5463 | added_devices++; | |
5464 | } else | |
5465 | rdev->recovery_offset = 0; | |
36fad858 NK |
5466 | |
5467 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 5468 | /* Failure here is OK */; |
50da0840 | 5469 | } |
87a8dec9 N |
5470 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5471 | && !test_bit(Faulty, &rdev->flags)) { | |
5472 | /* This is a spare that was manually added */ | |
5473 | set_bit(In_sync, &rdev->flags); | |
5474 | added_devices++; | |
5475 | } | |
29269553 | 5476 | |
87a8dec9 N |
5477 | /* When a reshape changes the number of devices, |
5478 | * ->degraded is measured against the larger of the | |
5479 | * pre and post number of devices. | |
5480 | */ | |
ec32a2bd | 5481 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5482 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
5483 | spin_unlock_irqrestore(&conf->device_lock, flags); |
5484 | } | |
63c70c4f | 5485 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5486 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5487 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5488 | |
29269553 N |
5489 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5490 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5491 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5492 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5493 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5494 | "reshape"); |
29269553 N |
5495 | if (!mddev->sync_thread) { |
5496 | mddev->recovery = 0; | |
5497 | spin_lock_irq(&conf->device_lock); | |
5498 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5499 | conf->reshape_progress = MaxSector; |
29269553 N |
5500 | spin_unlock_irq(&conf->device_lock); |
5501 | return -EAGAIN; | |
5502 | } | |
c8f517c4 | 5503 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5504 | md_wakeup_thread(mddev->sync_thread); |
5505 | md_new_event(mddev); | |
5506 | return 0; | |
5507 | } | |
29269553 | 5508 | |
ec32a2bd N |
5509 | /* This is called from the reshape thread and should make any |
5510 | * changes needed in 'conf' | |
5511 | */ | |
d1688a6d | 5512 | static void end_reshape(struct r5conf *conf) |
29269553 | 5513 | { |
29269553 | 5514 | |
f6705578 | 5515 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5516 | |
f6705578 | 5517 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5518 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5519 | conf->reshape_progress = MaxSector; |
f6705578 | 5520 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5521 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5522 | |
5523 | /* read-ahead size must cover two whole stripes, which is | |
5524 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5525 | */ | |
4a5add49 | 5526 | if (conf->mddev->queue) { |
cea9c228 | 5527 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5528 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5529 | / PAGE_SIZE); |
16a53ecc N |
5530 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5531 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5532 | } | |
29269553 | 5533 | } |
29269553 N |
5534 | } |
5535 | ||
ec32a2bd N |
5536 | /* This is called from the raid5d thread with mddev_lock held. |
5537 | * It makes config changes to the device. | |
5538 | */ | |
fd01b88c | 5539 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 5540 | { |
d1688a6d | 5541 | struct r5conf *conf = mddev->private; |
cea9c228 N |
5542 | |
5543 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5544 | ||
ec32a2bd N |
5545 | if (mddev->delta_disks > 0) { |
5546 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5547 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5548 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5549 | } else { |
5550 | int d; | |
908f4fbd N |
5551 | spin_lock_irq(&conf->device_lock); |
5552 | mddev->degraded = calc_degraded(conf); | |
5553 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
5554 | for (d = conf->raid_disks ; |
5555 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 5556 | d++) { |
3cb03002 | 5557 | struct md_rdev *rdev = conf->disks[d].rdev; |
b8321b68 N |
5558 | if (rdev && |
5559 | raid5_remove_disk(mddev, rdev) == 0) { | |
36fad858 | 5560 | sysfs_unlink_rdev(mddev, rdev); |
1a67dde0 N |
5561 | rdev->raid_disk = -1; |
5562 | } | |
5563 | } | |
cea9c228 | 5564 | } |
88ce4930 | 5565 | mddev->layout = conf->algorithm; |
09c9e5fa | 5566 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5567 | mddev->reshape_position = MaxSector; |
5568 | mddev->delta_disks = 0; | |
cea9c228 N |
5569 | } |
5570 | } | |
5571 | ||
fd01b88c | 5572 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 5573 | { |
d1688a6d | 5574 | struct r5conf *conf = mddev->private; |
72626685 N |
5575 | |
5576 | switch(state) { | |
e464eafd N |
5577 | case 2: /* resume for a suspend */ |
5578 | wake_up(&conf->wait_for_overlap); | |
5579 | break; | |
5580 | ||
72626685 N |
5581 | case 1: /* stop all writes */ |
5582 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5583 | /* '2' tells resync/reshape to pause so that all |
5584 | * active stripes can drain | |
5585 | */ | |
5586 | conf->quiesce = 2; | |
72626685 | 5587 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5588 | atomic_read(&conf->active_stripes) == 0 && |
5589 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5590 | conf->device_lock, /* nothing */); |
64bd660b | 5591 | conf->quiesce = 1; |
72626685 | 5592 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5593 | /* allow reshape to continue */ |
5594 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5595 | break; |
5596 | ||
5597 | case 0: /* re-enable writes */ | |
5598 | spin_lock_irq(&conf->device_lock); | |
5599 | conf->quiesce = 0; | |
5600 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5601 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5602 | spin_unlock_irq(&conf->device_lock); |
5603 | break; | |
5604 | } | |
72626685 | 5605 | } |
b15c2e57 | 5606 | |
d562b0c4 | 5607 | |
fd01b88c | 5608 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 5609 | { |
e373ab10 | 5610 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 5611 | sector_t sectors; |
54071b38 | 5612 | |
f1b29bca | 5613 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 5614 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
5615 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5616 | mdname(mddev)); | |
f1b29bca DW |
5617 | return ERR_PTR(-EINVAL); |
5618 | } | |
5619 | ||
e373ab10 N |
5620 | sectors = raid0_conf->strip_zone[0].zone_end; |
5621 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 5622 | mddev->dev_sectors = sectors; |
f1b29bca | 5623 | mddev->new_level = level; |
54071b38 TM |
5624 | mddev->new_layout = ALGORITHM_PARITY_N; |
5625 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5626 | mddev->raid_disks += 1; | |
5627 | mddev->delta_disks = 1; | |
5628 | /* make sure it will be not marked as dirty */ | |
5629 | mddev->recovery_cp = MaxSector; | |
5630 | ||
5631 | return setup_conf(mddev); | |
5632 | } | |
5633 | ||
5634 | ||
fd01b88c | 5635 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
5636 | { |
5637 | int chunksect; | |
5638 | ||
5639 | if (mddev->raid_disks != 2 || | |
5640 | mddev->degraded > 1) | |
5641 | return ERR_PTR(-EINVAL); | |
5642 | ||
5643 | /* Should check if there are write-behind devices? */ | |
5644 | ||
5645 | chunksect = 64*2; /* 64K by default */ | |
5646 | ||
5647 | /* The array must be an exact multiple of chunksize */ | |
5648 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5649 | chunksect >>= 1; | |
5650 | ||
5651 | if ((chunksect<<9) < STRIPE_SIZE) | |
5652 | /* array size does not allow a suitable chunk size */ | |
5653 | return ERR_PTR(-EINVAL); | |
5654 | ||
5655 | mddev->new_level = 5; | |
5656 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5657 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5658 | |
5659 | return setup_conf(mddev); | |
5660 | } | |
5661 | ||
fd01b88c | 5662 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
5663 | { |
5664 | int new_layout; | |
5665 | ||
5666 | switch (mddev->layout) { | |
5667 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5668 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5669 | break; | |
5670 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5671 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5672 | break; | |
5673 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5674 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5675 | break; | |
5676 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5677 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5678 | break; | |
5679 | case ALGORITHM_PARITY_0_6: | |
5680 | new_layout = ALGORITHM_PARITY_0; | |
5681 | break; | |
5682 | case ALGORITHM_PARITY_N: | |
5683 | new_layout = ALGORITHM_PARITY_N; | |
5684 | break; | |
5685 | default: | |
5686 | return ERR_PTR(-EINVAL); | |
5687 | } | |
5688 | mddev->new_level = 5; | |
5689 | mddev->new_layout = new_layout; | |
5690 | mddev->delta_disks = -1; | |
5691 | mddev->raid_disks -= 1; | |
5692 | return setup_conf(mddev); | |
5693 | } | |
5694 | ||
d562b0c4 | 5695 | |
fd01b88c | 5696 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 5697 | { |
88ce4930 N |
5698 | /* For a 2-drive array, the layout and chunk size can be changed |
5699 | * immediately as not restriping is needed. | |
5700 | * For larger arrays we record the new value - after validation | |
5701 | * to be used by a reshape pass. | |
b3546035 | 5702 | */ |
d1688a6d | 5703 | struct r5conf *conf = mddev->private; |
597a711b | 5704 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5705 | |
597a711b | 5706 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5707 | return -EINVAL; |
5708 | if (new_chunk > 0) { | |
0ba459d2 | 5709 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5710 | return -EINVAL; |
597a711b | 5711 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5712 | return -EINVAL; |
597a711b | 5713 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5714 | /* not factor of array size */ |
5715 | return -EINVAL; | |
5716 | } | |
5717 | ||
5718 | /* They look valid */ | |
5719 | ||
88ce4930 | 5720 | if (mddev->raid_disks == 2) { |
597a711b N |
5721 | /* can make the change immediately */ |
5722 | if (mddev->new_layout >= 0) { | |
5723 | conf->algorithm = mddev->new_layout; | |
5724 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5725 | } |
5726 | if (new_chunk > 0) { | |
597a711b N |
5727 | conf->chunk_sectors = new_chunk ; |
5728 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5729 | } |
5730 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5731 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5732 | } |
50ac168a | 5733 | return check_reshape(mddev); |
88ce4930 N |
5734 | } |
5735 | ||
fd01b88c | 5736 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 5737 | { |
597a711b | 5738 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5739 | |
597a711b | 5740 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5741 | return -EINVAL; |
b3546035 | 5742 | if (new_chunk > 0) { |
0ba459d2 | 5743 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5744 | return -EINVAL; |
597a711b | 5745 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5746 | return -EINVAL; |
597a711b | 5747 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5748 | /* not factor of array size */ |
5749 | return -EINVAL; | |
b3546035 | 5750 | } |
88ce4930 N |
5751 | |
5752 | /* They look valid */ | |
50ac168a | 5753 | return check_reshape(mddev); |
b3546035 N |
5754 | } |
5755 | ||
fd01b88c | 5756 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
5757 | { |
5758 | /* raid5 can take over: | |
f1b29bca | 5759 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
5760 | * raid1 - if there are two drives. We need to know the chunk size |
5761 | * raid4 - trivial - just use a raid4 layout. | |
5762 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 5763 | */ |
f1b29bca DW |
5764 | if (mddev->level == 0) |
5765 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
5766 | if (mddev->level == 1) |
5767 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5768 | if (mddev->level == 4) { |
5769 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5770 | mddev->new_level = 5; | |
5771 | return setup_conf(mddev); | |
5772 | } | |
fc9739c6 N |
5773 | if (mddev->level == 6) |
5774 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5775 | |
5776 | return ERR_PTR(-EINVAL); | |
5777 | } | |
5778 | ||
fd01b88c | 5779 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 5780 | { |
f1b29bca DW |
5781 | /* raid4 can take over: |
5782 | * raid0 - if there is only one strip zone | |
5783 | * raid5 - if layout is right | |
a78d38a1 | 5784 | */ |
f1b29bca DW |
5785 | if (mddev->level == 0) |
5786 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
5787 | if (mddev->level == 5 && |
5788 | mddev->layout == ALGORITHM_PARITY_N) { | |
5789 | mddev->new_layout = 0; | |
5790 | mddev->new_level = 4; | |
5791 | return setup_conf(mddev); | |
5792 | } | |
5793 | return ERR_PTR(-EINVAL); | |
5794 | } | |
d562b0c4 | 5795 | |
84fc4b56 | 5796 | static struct md_personality raid5_personality; |
245f46c2 | 5797 | |
fd01b88c | 5798 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
5799 | { |
5800 | /* Currently can only take over a raid5. We map the | |
5801 | * personality to an equivalent raid6 personality | |
5802 | * with the Q block at the end. | |
5803 | */ | |
5804 | int new_layout; | |
5805 | ||
5806 | if (mddev->pers != &raid5_personality) | |
5807 | return ERR_PTR(-EINVAL); | |
5808 | if (mddev->degraded > 1) | |
5809 | return ERR_PTR(-EINVAL); | |
5810 | if (mddev->raid_disks > 253) | |
5811 | return ERR_PTR(-EINVAL); | |
5812 | if (mddev->raid_disks < 3) | |
5813 | return ERR_PTR(-EINVAL); | |
5814 | ||
5815 | switch (mddev->layout) { | |
5816 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5817 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5818 | break; | |
5819 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5820 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5821 | break; | |
5822 | case ALGORITHM_LEFT_SYMMETRIC: | |
5823 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5824 | break; | |
5825 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5826 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5827 | break; | |
5828 | case ALGORITHM_PARITY_0: | |
5829 | new_layout = ALGORITHM_PARITY_0_6; | |
5830 | break; | |
5831 | case ALGORITHM_PARITY_N: | |
5832 | new_layout = ALGORITHM_PARITY_N; | |
5833 | break; | |
5834 | default: | |
5835 | return ERR_PTR(-EINVAL); | |
5836 | } | |
5837 | mddev->new_level = 6; | |
5838 | mddev->new_layout = new_layout; | |
5839 | mddev->delta_disks = 1; | |
5840 | mddev->raid_disks += 1; | |
5841 | return setup_conf(mddev); | |
5842 | } | |
5843 | ||
5844 | ||
84fc4b56 | 5845 | static struct md_personality raid6_personality = |
16a53ecc N |
5846 | { |
5847 | .name = "raid6", | |
5848 | .level = 6, | |
5849 | .owner = THIS_MODULE, | |
5850 | .make_request = make_request, | |
5851 | .run = run, | |
5852 | .stop = stop, | |
5853 | .status = status, | |
5854 | .error_handler = error, | |
5855 | .hot_add_disk = raid5_add_disk, | |
5856 | .hot_remove_disk= raid5_remove_disk, | |
5857 | .spare_active = raid5_spare_active, | |
5858 | .sync_request = sync_request, | |
5859 | .resize = raid5_resize, | |
80c3a6ce | 5860 | .size = raid5_size, |
50ac168a | 5861 | .check_reshape = raid6_check_reshape, |
f416885e | 5862 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5863 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5864 | .quiesce = raid5_quiesce, |
245f46c2 | 5865 | .takeover = raid6_takeover, |
16a53ecc | 5866 | }; |
84fc4b56 | 5867 | static struct md_personality raid5_personality = |
1da177e4 LT |
5868 | { |
5869 | .name = "raid5", | |
2604b703 | 5870 | .level = 5, |
1da177e4 LT |
5871 | .owner = THIS_MODULE, |
5872 | .make_request = make_request, | |
5873 | .run = run, | |
5874 | .stop = stop, | |
5875 | .status = status, | |
5876 | .error_handler = error, | |
5877 | .hot_add_disk = raid5_add_disk, | |
5878 | .hot_remove_disk= raid5_remove_disk, | |
5879 | .spare_active = raid5_spare_active, | |
5880 | .sync_request = sync_request, | |
5881 | .resize = raid5_resize, | |
80c3a6ce | 5882 | .size = raid5_size, |
63c70c4f N |
5883 | .check_reshape = raid5_check_reshape, |
5884 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5885 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5886 | .quiesce = raid5_quiesce, |
d562b0c4 | 5887 | .takeover = raid5_takeover, |
1da177e4 LT |
5888 | }; |
5889 | ||
84fc4b56 | 5890 | static struct md_personality raid4_personality = |
1da177e4 | 5891 | { |
2604b703 N |
5892 | .name = "raid4", |
5893 | .level = 4, | |
5894 | .owner = THIS_MODULE, | |
5895 | .make_request = make_request, | |
5896 | .run = run, | |
5897 | .stop = stop, | |
5898 | .status = status, | |
5899 | .error_handler = error, | |
5900 | .hot_add_disk = raid5_add_disk, | |
5901 | .hot_remove_disk= raid5_remove_disk, | |
5902 | .spare_active = raid5_spare_active, | |
5903 | .sync_request = sync_request, | |
5904 | .resize = raid5_resize, | |
80c3a6ce | 5905 | .size = raid5_size, |
3d37890b N |
5906 | .check_reshape = raid5_check_reshape, |
5907 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5908 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 5909 | .quiesce = raid5_quiesce, |
a78d38a1 | 5910 | .takeover = raid4_takeover, |
2604b703 N |
5911 | }; |
5912 | ||
5913 | static int __init raid5_init(void) | |
5914 | { | |
16a53ecc | 5915 | register_md_personality(&raid6_personality); |
2604b703 N |
5916 | register_md_personality(&raid5_personality); |
5917 | register_md_personality(&raid4_personality); | |
5918 | return 0; | |
1da177e4 LT |
5919 | } |
5920 | ||
2604b703 | 5921 | static void raid5_exit(void) |
1da177e4 | 5922 | { |
16a53ecc | 5923 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5924 | unregister_md_personality(&raid5_personality); |
5925 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5926 | } |
5927 | ||
5928 | module_init(raid5_init); | |
5929 | module_exit(raid5_exit); | |
5930 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5931 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 5932 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
5933 | MODULE_ALIAS("md-raid5"); |
5934 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5935 | MODULE_ALIAS("md-level-5"); |
5936 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5937 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5938 | MODULE_ALIAS("md-raid6"); | |
5939 | MODULE_ALIAS("md-level-6"); | |
5940 | ||
5941 | /* This used to be two separate modules, they were: */ | |
5942 | MODULE_ALIAS("raid5"); | |
5943 | MODULE_ALIAS("raid6"); |