]>
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> |
07a3b417 | 50 | #include <linux/async.h> |
bff61975 | 51 | #include <linux/seq_file.h> |
36d1c647 | 52 | #include <linux/cpu.h> |
5a0e3ad6 | 53 | #include <linux/slab.h> |
8bda470e | 54 | #include <linux/ratelimit.h> |
43b2e5d8 | 55 | #include "md.h" |
bff61975 | 56 | #include "raid5.h" |
54071b38 | 57 | #include "raid0.h" |
ef740c37 | 58 | #include "bitmap.h" |
72626685 | 59 | |
1da177e4 LT |
60 | /* |
61 | * Stripe cache | |
62 | */ | |
63 | ||
64 | #define NR_STRIPES 256 | |
65 | #define STRIPE_SIZE PAGE_SIZE | |
66 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
67 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
68 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 69 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 70 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
71 | #define HASH_MASK (NR_HASH - 1) |
72 | ||
d1688a6d | 73 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
74 | { |
75 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
76 | return &conf->stripe_hashtbl[hash]; | |
77 | } | |
1da177e4 LT |
78 | |
79 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
80 | * order without overlap. There may be several bio's per stripe+device, and | |
81 | * a bio could span several devices. | |
82 | * When walking this list for a particular stripe+device, we must never proceed | |
83 | * beyond a bio that extends past this device, as the next bio might no longer | |
84 | * be valid. | |
db298e19 | 85 | * This function is used to determine the 'next' bio in the list, given the sector |
1da177e4 LT |
86 | * of the current stripe+device |
87 | */ | |
db298e19 N |
88 | static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector) |
89 | { | |
90 | int sectors = bio->bi_size >> 9; | |
91 | if (bio->bi_sector + sectors < sector + STRIPE_SECTORS) | |
92 | return bio->bi_next; | |
93 | else | |
94 | return NULL; | |
95 | } | |
1da177e4 | 96 | |
960e739d | 97 | /* |
5b99c2ff JA |
98 | * We maintain a biased count of active stripes in the bottom 16 bits of |
99 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
100 | */ |
101 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
102 | { | |
5b99c2ff | 103 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
104 | } |
105 | ||
106 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
107 | { | |
5b99c2ff | 108 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
109 | } |
110 | ||
111 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
112 | { | |
113 | --bio->bi_phys_segments; | |
114 | return raid5_bi_phys_segments(bio); | |
115 | } | |
116 | ||
117 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
118 | { | |
119 | unsigned short val = raid5_bi_hw_segments(bio); | |
120 | ||
121 | --val; | |
5b99c2ff | 122 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
123 | return val; |
124 | } | |
125 | ||
126 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
127 | { | |
9b2dc8b6 | 128 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16); |
960e739d JA |
129 | } |
130 | ||
d0dabf7e N |
131 | /* Find first data disk in a raid6 stripe */ |
132 | static inline int raid6_d0(struct stripe_head *sh) | |
133 | { | |
67cc2b81 N |
134 | if (sh->ddf_layout) |
135 | /* ddf always start from first device */ | |
136 | return 0; | |
137 | /* md starts just after Q block */ | |
d0dabf7e N |
138 | if (sh->qd_idx == sh->disks - 1) |
139 | return 0; | |
140 | else | |
141 | return sh->qd_idx + 1; | |
142 | } | |
16a53ecc N |
143 | static inline int raid6_next_disk(int disk, int raid_disks) |
144 | { | |
145 | disk++; | |
146 | return (disk < raid_disks) ? disk : 0; | |
147 | } | |
a4456856 | 148 | |
d0dabf7e N |
149 | /* When walking through the disks in a raid5, starting at raid6_d0, |
150 | * We need to map each disk to a 'slot', where the data disks are slot | |
151 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
152 | * is raid_disks-1. This help does that mapping. | |
153 | */ | |
67cc2b81 N |
154 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
155 | int *count, int syndrome_disks) | |
d0dabf7e | 156 | { |
6629542e | 157 | int slot = *count; |
67cc2b81 | 158 | |
e4424fee | 159 | if (sh->ddf_layout) |
6629542e | 160 | (*count)++; |
d0dabf7e | 161 | if (idx == sh->pd_idx) |
67cc2b81 | 162 | return syndrome_disks; |
d0dabf7e | 163 | if (idx == sh->qd_idx) |
67cc2b81 | 164 | return syndrome_disks + 1; |
e4424fee | 165 | if (!sh->ddf_layout) |
6629542e | 166 | (*count)++; |
d0dabf7e N |
167 | return slot; |
168 | } | |
169 | ||
a4456856 DW |
170 | static void return_io(struct bio *return_bi) |
171 | { | |
172 | struct bio *bi = return_bi; | |
173 | while (bi) { | |
a4456856 DW |
174 | |
175 | return_bi = bi->bi_next; | |
176 | bi->bi_next = NULL; | |
177 | bi->bi_size = 0; | |
0e13fe23 | 178 | bio_endio(bi, 0); |
a4456856 DW |
179 | bi = return_bi; |
180 | } | |
181 | } | |
182 | ||
d1688a6d | 183 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 184 | |
600aa109 DW |
185 | static int stripe_operations_active(struct stripe_head *sh) |
186 | { | |
187 | return sh->check_state || sh->reconstruct_state || | |
188 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
189 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
190 | } | |
191 | ||
d1688a6d | 192 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 LT |
193 | { |
194 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
195 | BUG_ON(!list_empty(&sh->lru)); |
196 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 197 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
482c0834 | 198 | if (test_bit(STRIPE_DELAYED, &sh->state)) |
1da177e4 | 199 | list_add_tail(&sh->lru, &conf->delayed_list); |
482c0834 N |
200 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
201 | sh->bm_seq - conf->seq_write > 0) | |
72626685 | 202 | list_add_tail(&sh->lru, &conf->bitmap_list); |
482c0834 | 203 | else { |
72626685 | 204 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 205 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 206 | } |
1da177e4 LT |
207 | md_wakeup_thread(conf->mddev->thread); |
208 | } else { | |
600aa109 | 209 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
210 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
211 | atomic_dec(&conf->preread_active_stripes); | |
212 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
213 | md_wakeup_thread(conf->mddev->thread); | |
214 | } | |
1da177e4 | 215 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
216 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
217 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 218 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
219 | if (conf->retry_read_aligned) |
220 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 221 | } |
1da177e4 LT |
222 | } |
223 | } | |
224 | } | |
d0dabf7e | 225 | |
1da177e4 LT |
226 | static void release_stripe(struct stripe_head *sh) |
227 | { | |
d1688a6d | 228 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 229 | unsigned long flags; |
16a53ecc | 230 | |
1da177e4 LT |
231 | spin_lock_irqsave(&conf->device_lock, flags); |
232 | __release_stripe(conf, sh); | |
233 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
234 | } | |
235 | ||
fccddba0 | 236 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 237 | { |
45b4233c DW |
238 | pr_debug("remove_hash(), stripe %llu\n", |
239 | (unsigned long long)sh->sector); | |
1da177e4 | 240 | |
fccddba0 | 241 | hlist_del_init(&sh->hash); |
1da177e4 LT |
242 | } |
243 | ||
d1688a6d | 244 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 245 | { |
fccddba0 | 246 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 247 | |
45b4233c DW |
248 | pr_debug("insert_hash(), stripe %llu\n", |
249 | (unsigned long long)sh->sector); | |
1da177e4 | 250 | |
fccddba0 | 251 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
252 | } |
253 | ||
254 | ||
255 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
d1688a6d | 256 | static struct stripe_head *get_free_stripe(struct r5conf *conf) |
1da177e4 LT |
257 | { |
258 | struct stripe_head *sh = NULL; | |
259 | struct list_head *first; | |
260 | ||
1da177e4 LT |
261 | if (list_empty(&conf->inactive_list)) |
262 | goto out; | |
263 | first = conf->inactive_list.next; | |
264 | sh = list_entry(first, struct stripe_head, lru); | |
265 | list_del_init(first); | |
266 | remove_hash(sh); | |
267 | atomic_inc(&conf->active_stripes); | |
268 | out: | |
269 | return sh; | |
270 | } | |
271 | ||
e4e11e38 | 272 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
273 | { |
274 | struct page *p; | |
275 | int i; | |
e4e11e38 | 276 | int num = sh->raid_conf->pool_size; |
1da177e4 | 277 | |
e4e11e38 | 278 | for (i = 0; i < num ; i++) { |
1da177e4 LT |
279 | p = sh->dev[i].page; |
280 | if (!p) | |
281 | continue; | |
282 | sh->dev[i].page = NULL; | |
2d1f3b5d | 283 | put_page(p); |
1da177e4 LT |
284 | } |
285 | } | |
286 | ||
e4e11e38 | 287 | static int grow_buffers(struct stripe_head *sh) |
1da177e4 LT |
288 | { |
289 | int i; | |
e4e11e38 | 290 | int num = sh->raid_conf->pool_size; |
1da177e4 | 291 | |
e4e11e38 | 292 | for (i = 0; i < num; i++) { |
1da177e4 LT |
293 | struct page *page; |
294 | ||
295 | if (!(page = alloc_page(GFP_KERNEL))) { | |
296 | return 1; | |
297 | } | |
298 | sh->dev[i].page = page; | |
299 | } | |
300 | return 0; | |
301 | } | |
302 | ||
784052ec | 303 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 304 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 305 | struct stripe_head *sh); |
1da177e4 | 306 | |
b5663ba4 | 307 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 308 | { |
d1688a6d | 309 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 310 | int i; |
1da177e4 | 311 | |
78bafebd ES |
312 | BUG_ON(atomic_read(&sh->count) != 0); |
313 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 314 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 315 | |
45b4233c | 316 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
317 | (unsigned long long)sh->sector); |
318 | ||
319 | remove_hash(sh); | |
16a53ecc | 320 | |
86b42c71 | 321 | sh->generation = conf->generation - previous; |
b5663ba4 | 322 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 323 | sh->sector = sector; |
911d4ee8 | 324 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
325 | sh->state = 0; |
326 | ||
7ecaa1e6 N |
327 | |
328 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
329 | struct r5dev *dev = &sh->dev[i]; |
330 | ||
d84e0f10 | 331 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 332 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 333 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 334 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 335 | dev->read, dev->towrite, dev->written, |
1da177e4 | 336 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 337 | WARN_ON(1); |
1da177e4 LT |
338 | } |
339 | dev->flags = 0; | |
784052ec | 340 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
341 | } |
342 | insert_hash(conf, sh); | |
343 | } | |
344 | ||
d1688a6d | 345 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 346 | short generation) |
1da177e4 LT |
347 | { |
348 | struct stripe_head *sh; | |
fccddba0 | 349 | struct hlist_node *hn; |
1da177e4 | 350 | |
45b4233c | 351 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 352 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 353 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 354 | return sh; |
45b4233c | 355 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
356 | return NULL; |
357 | } | |
358 | ||
674806d6 N |
359 | /* |
360 | * Need to check if array has failed when deciding whether to: | |
361 | * - start an array | |
362 | * - remove non-faulty devices | |
363 | * - add a spare | |
364 | * - allow a reshape | |
365 | * This determination is simple when no reshape is happening. | |
366 | * However if there is a reshape, we need to carefully check | |
367 | * both the before and after sections. | |
368 | * This is because some failed devices may only affect one | |
369 | * of the two sections, and some non-in_sync devices may | |
370 | * be insync in the section most affected by failed devices. | |
371 | */ | |
d1688a6d | 372 | static int has_failed(struct r5conf *conf) |
674806d6 N |
373 | { |
374 | int degraded; | |
375 | int i; | |
376 | if (conf->mddev->reshape_position == MaxSector) | |
377 | return conf->mddev->degraded > conf->max_degraded; | |
378 | ||
379 | rcu_read_lock(); | |
380 | degraded = 0; | |
381 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 382 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
674806d6 N |
383 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
384 | degraded++; | |
385 | else if (test_bit(In_sync, &rdev->flags)) | |
386 | ; | |
387 | else | |
388 | /* not in-sync or faulty. | |
389 | * If the reshape increases the number of devices, | |
390 | * this is being recovered by the reshape, so | |
391 | * this 'previous' section is not in_sync. | |
392 | * If the number of devices is being reduced however, | |
393 | * the device can only be part of the array if | |
394 | * we are reverting a reshape, so this section will | |
395 | * be in-sync. | |
396 | */ | |
397 | if (conf->raid_disks >= conf->previous_raid_disks) | |
398 | degraded++; | |
399 | } | |
400 | rcu_read_unlock(); | |
401 | if (degraded > conf->max_degraded) | |
402 | return 1; | |
403 | rcu_read_lock(); | |
404 | degraded = 0; | |
405 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 406 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
674806d6 N |
407 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
408 | degraded++; | |
409 | else if (test_bit(In_sync, &rdev->flags)) | |
410 | ; | |
411 | else | |
412 | /* not in-sync or faulty. | |
413 | * If reshape increases the number of devices, this | |
414 | * section has already been recovered, else it | |
415 | * almost certainly hasn't. | |
416 | */ | |
417 | if (conf->raid_disks <= conf->previous_raid_disks) | |
418 | degraded++; | |
419 | } | |
420 | rcu_read_unlock(); | |
421 | if (degraded > conf->max_degraded) | |
422 | return 1; | |
423 | return 0; | |
424 | } | |
425 | ||
b5663ba4 | 426 | static struct stripe_head * |
d1688a6d | 427 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 428 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
429 | { |
430 | struct stripe_head *sh; | |
431 | ||
45b4233c | 432 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
433 | |
434 | spin_lock_irq(&conf->device_lock); | |
435 | ||
436 | do { | |
72626685 | 437 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 438 | conf->quiesce == 0 || noquiesce, |
72626685 | 439 | conf->device_lock, /* nothing */); |
86b42c71 | 440 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
441 | if (!sh) { |
442 | if (!conf->inactive_blocked) | |
443 | sh = get_free_stripe(conf); | |
444 | if (noblock && sh == NULL) | |
445 | break; | |
446 | if (!sh) { | |
447 | conf->inactive_blocked = 1; | |
448 | wait_event_lock_irq(conf->wait_for_stripe, | |
449 | !list_empty(&conf->inactive_list) && | |
5036805b N |
450 | (atomic_read(&conf->active_stripes) |
451 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
452 | || !conf->inactive_blocked), |
453 | conf->device_lock, | |
7c13edc8 | 454 | ); |
1da177e4 LT |
455 | conf->inactive_blocked = 0; |
456 | } else | |
b5663ba4 | 457 | init_stripe(sh, sector, previous); |
1da177e4 LT |
458 | } else { |
459 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
460 | BUG_ON(!list_empty(&sh->lru) |
461 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
462 | } else { |
463 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
464 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
465 | if (list_empty(&sh->lru) && |
466 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
467 | BUG(); |
468 | list_del_init(&sh->lru); | |
1da177e4 LT |
469 | } |
470 | } | |
471 | } while (sh == NULL); | |
472 | ||
473 | if (sh) | |
474 | atomic_inc(&sh->count); | |
475 | ||
476 | spin_unlock_irq(&conf->device_lock); | |
477 | return sh; | |
478 | } | |
479 | ||
6712ecf8 N |
480 | static void |
481 | raid5_end_read_request(struct bio *bi, int error); | |
482 | static void | |
483 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 484 | |
c4e5ac0a | 485 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 486 | { |
d1688a6d | 487 | struct r5conf *conf = sh->raid_conf; |
91c00924 DW |
488 | int i, disks = sh->disks; |
489 | ||
490 | might_sleep(); | |
491 | ||
492 | for (i = disks; i--; ) { | |
493 | int rw; | |
494 | struct bio *bi; | |
3cb03002 | 495 | struct md_rdev *rdev; |
e9c7469b TH |
496 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
497 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
498 | rw = WRITE_FUA; | |
499 | else | |
500 | rw = WRITE; | |
501 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
91c00924 DW |
502 | rw = READ; |
503 | else | |
504 | continue; | |
505 | ||
506 | bi = &sh->dev[i].req; | |
507 | ||
508 | bi->bi_rw = rw; | |
b062962e | 509 | if (rw & WRITE) |
91c00924 DW |
510 | bi->bi_end_io = raid5_end_write_request; |
511 | else | |
512 | bi->bi_end_io = raid5_end_read_request; | |
513 | ||
514 | rcu_read_lock(); | |
515 | rdev = rcu_dereference(conf->disks[i].rdev); | |
516 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
517 | rdev = NULL; | |
518 | if (rdev) | |
519 | atomic_inc(&rdev->nr_pending); | |
520 | rcu_read_unlock(); | |
521 | ||
73e92e51 N |
522 | /* We have already checked bad blocks for reads. Now |
523 | * need to check for writes. | |
524 | */ | |
525 | while ((rw & WRITE) && rdev && | |
526 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
527 | sector_t first_bad; | |
528 | int bad_sectors; | |
529 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
530 | &first_bad, &bad_sectors); | |
531 | if (!bad) | |
532 | break; | |
533 | ||
534 | if (bad < 0) { | |
535 | set_bit(BlockedBadBlocks, &rdev->flags); | |
536 | if (!conf->mddev->external && | |
537 | conf->mddev->flags) { | |
538 | /* It is very unlikely, but we might | |
539 | * still need to write out the | |
540 | * bad block log - better give it | |
541 | * a chance*/ | |
542 | md_check_recovery(conf->mddev); | |
543 | } | |
544 | md_wait_for_blocked_rdev(rdev, conf->mddev); | |
545 | } else { | |
546 | /* Acknowledged bad block - skip the write */ | |
547 | rdev_dec_pending(rdev, conf->mddev); | |
548 | rdev = NULL; | |
549 | } | |
550 | } | |
551 | ||
91c00924 | 552 | if (rdev) { |
c4e5ac0a | 553 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
554 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
555 | ||
2b7497f0 DW |
556 | set_bit(STRIPE_IO_STARTED, &sh->state); |
557 | ||
91c00924 DW |
558 | bi->bi_bdev = rdev->bdev; |
559 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 560 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
561 | bi->bi_rw, i); |
562 | atomic_inc(&sh->count); | |
563 | bi->bi_sector = sh->sector + rdev->data_offset; | |
564 | bi->bi_flags = 1 << BIO_UPTODATE; | |
565 | bi->bi_vcnt = 1; | |
566 | bi->bi_max_vecs = 1; | |
567 | bi->bi_idx = 0; | |
568 | bi->bi_io_vec = &sh->dev[i].vec; | |
569 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
570 | bi->bi_io_vec[0].bv_offset = 0; | |
571 | bi->bi_size = STRIPE_SIZE; | |
572 | bi->bi_next = NULL; | |
91c00924 DW |
573 | generic_make_request(bi); |
574 | } else { | |
b062962e | 575 | if (rw & WRITE) |
91c00924 DW |
576 | set_bit(STRIPE_DEGRADED, &sh->state); |
577 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
578 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
579 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
580 | set_bit(STRIPE_HANDLE, &sh->state); | |
581 | } | |
582 | } | |
583 | } | |
584 | ||
585 | static struct dma_async_tx_descriptor * | |
586 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
587 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
588 | { | |
589 | struct bio_vec *bvl; | |
590 | struct page *bio_page; | |
591 | int i; | |
592 | int page_offset; | |
a08abd8c | 593 | struct async_submit_ctl submit; |
0403e382 | 594 | enum async_tx_flags flags = 0; |
91c00924 DW |
595 | |
596 | if (bio->bi_sector >= sector) | |
597 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
598 | else | |
599 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 600 | |
0403e382 DW |
601 | if (frombio) |
602 | flags |= ASYNC_TX_FENCE; | |
603 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
604 | ||
91c00924 | 605 | bio_for_each_segment(bvl, bio, i) { |
fcde9075 | 606 | int len = bvl->bv_len; |
91c00924 DW |
607 | int clen; |
608 | int b_offset = 0; | |
609 | ||
610 | if (page_offset < 0) { | |
611 | b_offset = -page_offset; | |
612 | page_offset += b_offset; | |
613 | len -= b_offset; | |
614 | } | |
615 | ||
616 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
617 | clen = STRIPE_SIZE - page_offset; | |
618 | else | |
619 | clen = len; | |
620 | ||
621 | if (clen > 0) { | |
fcde9075 NK |
622 | b_offset += bvl->bv_offset; |
623 | bio_page = bvl->bv_page; | |
91c00924 DW |
624 | if (frombio) |
625 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 626 | b_offset, clen, &submit); |
91c00924 DW |
627 | else |
628 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 629 | page_offset, clen, &submit); |
91c00924 | 630 | } |
a08abd8c DW |
631 | /* chain the operations */ |
632 | submit.depend_tx = tx; | |
633 | ||
91c00924 DW |
634 | if (clen < len) /* hit end of page */ |
635 | break; | |
636 | page_offset += len; | |
637 | } | |
638 | ||
639 | return tx; | |
640 | } | |
641 | ||
642 | static void ops_complete_biofill(void *stripe_head_ref) | |
643 | { | |
644 | struct stripe_head *sh = stripe_head_ref; | |
645 | struct bio *return_bi = NULL; | |
d1688a6d | 646 | struct r5conf *conf = sh->raid_conf; |
e4d84909 | 647 | int i; |
91c00924 | 648 | |
e46b272b | 649 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
650 | (unsigned long long)sh->sector); |
651 | ||
652 | /* clear completed biofills */ | |
83de75cc | 653 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
654 | for (i = sh->disks; i--; ) { |
655 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
656 | |
657 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
658 | /* and check if we need to reply to a read request, |
659 | * new R5_Wantfill requests are held off until | |
83de75cc | 660 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
661 | */ |
662 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 663 | struct bio *rbi, *rbi2; |
91c00924 | 664 | |
91c00924 DW |
665 | BUG_ON(!dev->read); |
666 | rbi = dev->read; | |
667 | dev->read = NULL; | |
668 | while (rbi && rbi->bi_sector < | |
669 | dev->sector + STRIPE_SECTORS) { | |
670 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 671 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
672 | rbi->bi_next = return_bi; |
673 | return_bi = rbi; | |
674 | } | |
91c00924 DW |
675 | rbi = rbi2; |
676 | } | |
677 | } | |
678 | } | |
83de75cc DW |
679 | spin_unlock_irq(&conf->device_lock); |
680 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
681 | |
682 | return_io(return_bi); | |
683 | ||
e4d84909 | 684 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
685 | release_stripe(sh); |
686 | } | |
687 | ||
688 | static void ops_run_biofill(struct stripe_head *sh) | |
689 | { | |
690 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 691 | struct r5conf *conf = sh->raid_conf; |
a08abd8c | 692 | struct async_submit_ctl submit; |
91c00924 DW |
693 | int i; |
694 | ||
e46b272b | 695 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
696 | (unsigned long long)sh->sector); |
697 | ||
698 | for (i = sh->disks; i--; ) { | |
699 | struct r5dev *dev = &sh->dev[i]; | |
700 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
701 | struct bio *rbi; | |
702 | spin_lock_irq(&conf->device_lock); | |
703 | dev->read = rbi = dev->toread; | |
704 | dev->toread = NULL; | |
705 | spin_unlock_irq(&conf->device_lock); | |
706 | while (rbi && rbi->bi_sector < | |
707 | dev->sector + STRIPE_SECTORS) { | |
708 | tx = async_copy_data(0, rbi, dev->page, | |
709 | dev->sector, tx); | |
710 | rbi = r5_next_bio(rbi, dev->sector); | |
711 | } | |
712 | } | |
713 | } | |
714 | ||
715 | atomic_inc(&sh->count); | |
a08abd8c DW |
716 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
717 | async_trigger_callback(&submit); | |
91c00924 DW |
718 | } |
719 | ||
4e7d2c0a | 720 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 721 | { |
4e7d2c0a | 722 | struct r5dev *tgt; |
91c00924 | 723 | |
4e7d2c0a DW |
724 | if (target < 0) |
725 | return; | |
91c00924 | 726 | |
4e7d2c0a | 727 | tgt = &sh->dev[target]; |
91c00924 DW |
728 | set_bit(R5_UPTODATE, &tgt->flags); |
729 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
730 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
731 | } |
732 | ||
ac6b53b6 | 733 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
734 | { |
735 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 736 | |
e46b272b | 737 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
738 | (unsigned long long)sh->sector); |
739 | ||
ac6b53b6 | 740 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 741 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 742 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 743 | |
ecc65c9b DW |
744 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
745 | if (sh->check_state == check_state_compute_run) | |
746 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
747 | set_bit(STRIPE_HANDLE, &sh->state); |
748 | release_stripe(sh); | |
749 | } | |
750 | ||
d6f38f31 DW |
751 | /* return a pointer to the address conversion region of the scribble buffer */ |
752 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
753 | struct raid5_percpu *percpu) | |
754 | { | |
755 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
756 | } | |
757 | ||
758 | static struct dma_async_tx_descriptor * | |
759 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 760 | { |
91c00924 | 761 | int disks = sh->disks; |
d6f38f31 | 762 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
763 | int target = sh->ops.target; |
764 | struct r5dev *tgt = &sh->dev[target]; | |
765 | struct page *xor_dest = tgt->page; | |
766 | int count = 0; | |
767 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 768 | struct async_submit_ctl submit; |
91c00924 DW |
769 | int i; |
770 | ||
771 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 772 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
773 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
774 | ||
775 | for (i = disks; i--; ) | |
776 | if (i != target) | |
777 | xor_srcs[count++] = sh->dev[i].page; | |
778 | ||
779 | atomic_inc(&sh->count); | |
780 | ||
0403e382 | 781 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 782 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 783 | if (unlikely(count == 1)) |
a08abd8c | 784 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 785 | else |
a08abd8c | 786 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 787 | |
91c00924 DW |
788 | return tx; |
789 | } | |
790 | ||
ac6b53b6 DW |
791 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
792 | * @srcs - (struct page *) array of size sh->disks | |
793 | * @sh - stripe_head to parse | |
794 | * | |
795 | * Populates srcs in proper layout order for the stripe and returns the | |
796 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
797 | * destination buffer is recorded in srcs[count] and the Q destination | |
798 | * is recorded in srcs[count+1]]. | |
799 | */ | |
800 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
801 | { | |
802 | int disks = sh->disks; | |
803 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
804 | int d0_idx = raid6_d0(sh); | |
805 | int count; | |
806 | int i; | |
807 | ||
808 | for (i = 0; i < disks; i++) | |
5dd33c9a | 809 | srcs[i] = NULL; |
ac6b53b6 DW |
810 | |
811 | count = 0; | |
812 | i = d0_idx; | |
813 | do { | |
814 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
815 | ||
816 | srcs[slot] = sh->dev[i].page; | |
817 | i = raid6_next_disk(i, disks); | |
818 | } while (i != d0_idx); | |
ac6b53b6 | 819 | |
e4424fee | 820 | return syndrome_disks; |
ac6b53b6 DW |
821 | } |
822 | ||
823 | static struct dma_async_tx_descriptor * | |
824 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
825 | { | |
826 | int disks = sh->disks; | |
827 | struct page **blocks = percpu->scribble; | |
828 | int target; | |
829 | int qd_idx = sh->qd_idx; | |
830 | struct dma_async_tx_descriptor *tx; | |
831 | struct async_submit_ctl submit; | |
832 | struct r5dev *tgt; | |
833 | struct page *dest; | |
834 | int i; | |
835 | int count; | |
836 | ||
837 | if (sh->ops.target < 0) | |
838 | target = sh->ops.target2; | |
839 | else if (sh->ops.target2 < 0) | |
840 | target = sh->ops.target; | |
91c00924 | 841 | else |
ac6b53b6 DW |
842 | /* we should only have one valid target */ |
843 | BUG(); | |
844 | BUG_ON(target < 0); | |
845 | pr_debug("%s: stripe %llu block: %d\n", | |
846 | __func__, (unsigned long long)sh->sector, target); | |
847 | ||
848 | tgt = &sh->dev[target]; | |
849 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
850 | dest = tgt->page; | |
851 | ||
852 | atomic_inc(&sh->count); | |
853 | ||
854 | if (target == qd_idx) { | |
855 | count = set_syndrome_sources(blocks, sh); | |
856 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
857 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
858 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
859 | ops_complete_compute, sh, | |
ac6b53b6 DW |
860 | to_addr_conv(sh, percpu)); |
861 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
862 | } else { | |
863 | /* Compute any data- or p-drive using XOR */ | |
864 | count = 0; | |
865 | for (i = disks; i-- ; ) { | |
866 | if (i == target || i == qd_idx) | |
867 | continue; | |
868 | blocks[count++] = sh->dev[i].page; | |
869 | } | |
870 | ||
0403e382 DW |
871 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
872 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
873 | to_addr_conv(sh, percpu)); |
874 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
875 | } | |
91c00924 | 876 | |
91c00924 DW |
877 | return tx; |
878 | } | |
879 | ||
ac6b53b6 DW |
880 | static struct dma_async_tx_descriptor * |
881 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
882 | { | |
883 | int i, count, disks = sh->disks; | |
884 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
885 | int d0_idx = raid6_d0(sh); | |
886 | int faila = -1, failb = -1; | |
887 | int target = sh->ops.target; | |
888 | int target2 = sh->ops.target2; | |
889 | struct r5dev *tgt = &sh->dev[target]; | |
890 | struct r5dev *tgt2 = &sh->dev[target2]; | |
891 | struct dma_async_tx_descriptor *tx; | |
892 | struct page **blocks = percpu->scribble; | |
893 | struct async_submit_ctl submit; | |
894 | ||
895 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
896 | __func__, (unsigned long long)sh->sector, target, target2); | |
897 | BUG_ON(target < 0 || target2 < 0); | |
898 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
899 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
900 | ||
6c910a78 | 901 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
902 | * slot number conversion for 'faila' and 'failb' |
903 | */ | |
904 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 905 | blocks[i] = NULL; |
ac6b53b6 DW |
906 | count = 0; |
907 | i = d0_idx; | |
908 | do { | |
909 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
910 | ||
911 | blocks[slot] = sh->dev[i].page; | |
912 | ||
913 | if (i == target) | |
914 | faila = slot; | |
915 | if (i == target2) | |
916 | failb = slot; | |
917 | i = raid6_next_disk(i, disks); | |
918 | } while (i != d0_idx); | |
ac6b53b6 DW |
919 | |
920 | BUG_ON(faila == failb); | |
921 | if (failb < faila) | |
922 | swap(faila, failb); | |
923 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
924 | __func__, (unsigned long long)sh->sector, faila, failb); | |
925 | ||
926 | atomic_inc(&sh->count); | |
927 | ||
928 | if (failb == syndrome_disks+1) { | |
929 | /* Q disk is one of the missing disks */ | |
930 | if (faila == syndrome_disks) { | |
931 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
932 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
933 | ops_complete_compute, sh, | |
934 | to_addr_conv(sh, percpu)); | |
e4424fee | 935 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
936 | STRIPE_SIZE, &submit); |
937 | } else { | |
938 | struct page *dest; | |
939 | int data_target; | |
940 | int qd_idx = sh->qd_idx; | |
941 | ||
942 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
943 | if (target == qd_idx) | |
944 | data_target = target2; | |
945 | else | |
946 | data_target = target; | |
947 | ||
948 | count = 0; | |
949 | for (i = disks; i-- ; ) { | |
950 | if (i == data_target || i == qd_idx) | |
951 | continue; | |
952 | blocks[count++] = sh->dev[i].page; | |
953 | } | |
954 | dest = sh->dev[data_target].page; | |
0403e382 DW |
955 | init_async_submit(&submit, |
956 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
957 | NULL, NULL, NULL, | |
958 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
959 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
960 | &submit); | |
961 | ||
962 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
963 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
964 | ops_complete_compute, sh, | |
965 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
966 | return async_gen_syndrome(blocks, 0, count+2, |
967 | STRIPE_SIZE, &submit); | |
968 | } | |
ac6b53b6 | 969 | } else { |
6c910a78 DW |
970 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
971 | ops_complete_compute, sh, | |
972 | to_addr_conv(sh, percpu)); | |
973 | if (failb == syndrome_disks) { | |
974 | /* We're missing D+P. */ | |
975 | return async_raid6_datap_recov(syndrome_disks+2, | |
976 | STRIPE_SIZE, faila, | |
977 | blocks, &submit); | |
978 | } else { | |
979 | /* We're missing D+D. */ | |
980 | return async_raid6_2data_recov(syndrome_disks+2, | |
981 | STRIPE_SIZE, faila, failb, | |
982 | blocks, &submit); | |
983 | } | |
ac6b53b6 DW |
984 | } |
985 | } | |
986 | ||
987 | ||
91c00924 DW |
988 | static void ops_complete_prexor(void *stripe_head_ref) |
989 | { | |
990 | struct stripe_head *sh = stripe_head_ref; | |
991 | ||
e46b272b | 992 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 993 | (unsigned long long)sh->sector); |
91c00924 DW |
994 | } |
995 | ||
996 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
997 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
998 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 999 | { |
91c00924 | 1000 | int disks = sh->disks; |
d6f38f31 | 1001 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1002 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1003 | struct async_submit_ctl submit; |
91c00924 DW |
1004 | |
1005 | /* existing parity data subtracted */ | |
1006 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1007 | ||
e46b272b | 1008 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1009 | (unsigned long long)sh->sector); |
1010 | ||
1011 | for (i = disks; i--; ) { | |
1012 | struct r5dev *dev = &sh->dev[i]; | |
1013 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1014 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1015 | xor_srcs[count++] = dev->page; |
1016 | } | |
1017 | ||
0403e382 | 1018 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 1019 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 1020 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1021 | |
1022 | return tx; | |
1023 | } | |
1024 | ||
1025 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1026 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1027 | { |
1028 | int disks = sh->disks; | |
d8ee0728 | 1029 | int i; |
91c00924 | 1030 | |
e46b272b | 1031 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1032 | (unsigned long long)sh->sector); |
1033 | ||
1034 | for (i = disks; i--; ) { | |
1035 | struct r5dev *dev = &sh->dev[i]; | |
1036 | struct bio *chosen; | |
91c00924 | 1037 | |
d8ee0728 | 1038 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1039 | struct bio *wbi; |
1040 | ||
cbe47ec5 | 1041 | spin_lock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1042 | chosen = dev->towrite; |
1043 | dev->towrite = NULL; | |
1044 | BUG_ON(dev->written); | |
1045 | wbi = dev->written = chosen; | |
cbe47ec5 | 1046 | spin_unlock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1047 | |
1048 | while (wbi && wbi->bi_sector < | |
1049 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1050 | if (wbi->bi_rw & REQ_FUA) |
1051 | set_bit(R5_WantFUA, &dev->flags); | |
91c00924 DW |
1052 | tx = async_copy_data(1, wbi, dev->page, |
1053 | dev->sector, tx); | |
1054 | wbi = r5_next_bio(wbi, dev->sector); | |
1055 | } | |
1056 | } | |
1057 | } | |
1058 | ||
1059 | return tx; | |
1060 | } | |
1061 | ||
ac6b53b6 | 1062 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1063 | { |
1064 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1065 | int disks = sh->disks; |
1066 | int pd_idx = sh->pd_idx; | |
1067 | int qd_idx = sh->qd_idx; | |
1068 | int i; | |
e9c7469b | 1069 | bool fua = false; |
91c00924 | 1070 | |
e46b272b | 1071 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1072 | (unsigned long long)sh->sector); |
1073 | ||
e9c7469b TH |
1074 | for (i = disks; i--; ) |
1075 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); | |
1076 | ||
91c00924 DW |
1077 | for (i = disks; i--; ) { |
1078 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1079 | |
e9c7469b | 1080 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1081 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1082 | if (fua) |
1083 | set_bit(R5_WantFUA, &dev->flags); | |
1084 | } | |
91c00924 DW |
1085 | } |
1086 | ||
d8ee0728 DW |
1087 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1088 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1089 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1090 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1091 | else { | |
1092 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1093 | sh->reconstruct_state = reconstruct_state_result; | |
1094 | } | |
91c00924 DW |
1095 | |
1096 | set_bit(STRIPE_HANDLE, &sh->state); | |
1097 | release_stripe(sh); | |
1098 | } | |
1099 | ||
1100 | static void | |
ac6b53b6 DW |
1101 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1102 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1103 | { |
91c00924 | 1104 | int disks = sh->disks; |
d6f38f31 | 1105 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1106 | struct async_submit_ctl submit; |
91c00924 DW |
1107 | int count = 0, pd_idx = sh->pd_idx, i; |
1108 | struct page *xor_dest; | |
d8ee0728 | 1109 | int prexor = 0; |
91c00924 | 1110 | unsigned long flags; |
91c00924 | 1111 | |
e46b272b | 1112 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1113 | (unsigned long long)sh->sector); |
1114 | ||
1115 | /* check if prexor is active which means only process blocks | |
1116 | * that are part of a read-modify-write (written) | |
1117 | */ | |
d8ee0728 DW |
1118 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1119 | prexor = 1; | |
91c00924 DW |
1120 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1121 | for (i = disks; i--; ) { | |
1122 | struct r5dev *dev = &sh->dev[i]; | |
1123 | if (dev->written) | |
1124 | xor_srcs[count++] = dev->page; | |
1125 | } | |
1126 | } else { | |
1127 | xor_dest = sh->dev[pd_idx].page; | |
1128 | for (i = disks; i--; ) { | |
1129 | struct r5dev *dev = &sh->dev[i]; | |
1130 | if (i != pd_idx) | |
1131 | xor_srcs[count++] = dev->page; | |
1132 | } | |
1133 | } | |
1134 | ||
91c00924 DW |
1135 | /* 1/ if we prexor'd then the dest is reused as a source |
1136 | * 2/ if we did not prexor then we are redoing the parity | |
1137 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1138 | * for the synchronous xor case | |
1139 | */ | |
88ba2aa5 | 1140 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1141 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1142 | ||
1143 | atomic_inc(&sh->count); | |
1144 | ||
ac6b53b6 | 1145 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1146 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1147 | if (unlikely(count == 1)) |
1148 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1149 | else | |
1150 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1151 | } |
1152 | ||
ac6b53b6 DW |
1153 | static void |
1154 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1155 | struct dma_async_tx_descriptor *tx) | |
1156 | { | |
1157 | struct async_submit_ctl submit; | |
1158 | struct page **blocks = percpu->scribble; | |
1159 | int count; | |
1160 | ||
1161 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1162 | ||
1163 | count = set_syndrome_sources(blocks, sh); | |
1164 | ||
1165 | atomic_inc(&sh->count); | |
1166 | ||
1167 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1168 | sh, to_addr_conv(sh, percpu)); | |
1169 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1170 | } |
1171 | ||
1172 | static void ops_complete_check(void *stripe_head_ref) | |
1173 | { | |
1174 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1175 | |
e46b272b | 1176 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1177 | (unsigned long long)sh->sector); |
1178 | ||
ecc65c9b | 1179 | sh->check_state = check_state_check_result; |
91c00924 DW |
1180 | set_bit(STRIPE_HANDLE, &sh->state); |
1181 | release_stripe(sh); | |
1182 | } | |
1183 | ||
ac6b53b6 | 1184 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1185 | { |
91c00924 | 1186 | int disks = sh->disks; |
ac6b53b6 DW |
1187 | int pd_idx = sh->pd_idx; |
1188 | int qd_idx = sh->qd_idx; | |
1189 | struct page *xor_dest; | |
d6f38f31 | 1190 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1191 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1192 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1193 | int count; |
1194 | int i; | |
91c00924 | 1195 | |
e46b272b | 1196 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1197 | (unsigned long long)sh->sector); |
1198 | ||
ac6b53b6 DW |
1199 | count = 0; |
1200 | xor_dest = sh->dev[pd_idx].page; | |
1201 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1202 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1203 | if (i == pd_idx || i == qd_idx) |
1204 | continue; | |
1205 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1206 | } |
1207 | ||
d6f38f31 DW |
1208 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1209 | to_addr_conv(sh, percpu)); | |
099f53cb | 1210 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1211 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1212 | |
91c00924 | 1213 | atomic_inc(&sh->count); |
a08abd8c DW |
1214 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1215 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1216 | } |
1217 | ||
ac6b53b6 DW |
1218 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1219 | { | |
1220 | struct page **srcs = percpu->scribble; | |
1221 | struct async_submit_ctl submit; | |
1222 | int count; | |
1223 | ||
1224 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1225 | (unsigned long long)sh->sector, checkp); | |
1226 | ||
1227 | count = set_syndrome_sources(srcs, sh); | |
1228 | if (!checkp) | |
1229 | srcs[count] = NULL; | |
91c00924 | 1230 | |
91c00924 | 1231 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1232 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1233 | sh, to_addr_conv(sh, percpu)); | |
1234 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1235 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1236 | } |
1237 | ||
417b8d4a | 1238 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1239 | { |
1240 | int overlap_clear = 0, i, disks = sh->disks; | |
1241 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1242 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1243 | int level = conf->level; |
d6f38f31 DW |
1244 | struct raid5_percpu *percpu; |
1245 | unsigned long cpu; | |
91c00924 | 1246 | |
d6f38f31 DW |
1247 | cpu = get_cpu(); |
1248 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1249 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1250 | ops_run_biofill(sh); |
1251 | overlap_clear++; | |
1252 | } | |
1253 | ||
7b3a871e | 1254 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1255 | if (level < 6) |
1256 | tx = ops_run_compute5(sh, percpu); | |
1257 | else { | |
1258 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1259 | tx = ops_run_compute6_1(sh, percpu); | |
1260 | else | |
1261 | tx = ops_run_compute6_2(sh, percpu); | |
1262 | } | |
1263 | /* terminate the chain if reconstruct is not set to be run */ | |
1264 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1265 | async_tx_ack(tx); |
1266 | } | |
91c00924 | 1267 | |
600aa109 | 1268 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1269 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1270 | |
600aa109 | 1271 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1272 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1273 | overlap_clear++; |
1274 | } | |
1275 | ||
ac6b53b6 DW |
1276 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1277 | if (level < 6) | |
1278 | ops_run_reconstruct5(sh, percpu, tx); | |
1279 | else | |
1280 | ops_run_reconstruct6(sh, percpu, tx); | |
1281 | } | |
91c00924 | 1282 | |
ac6b53b6 DW |
1283 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1284 | if (sh->check_state == check_state_run) | |
1285 | ops_run_check_p(sh, percpu); | |
1286 | else if (sh->check_state == check_state_run_q) | |
1287 | ops_run_check_pq(sh, percpu, 0); | |
1288 | else if (sh->check_state == check_state_run_pq) | |
1289 | ops_run_check_pq(sh, percpu, 1); | |
1290 | else | |
1291 | BUG(); | |
1292 | } | |
91c00924 | 1293 | |
91c00924 DW |
1294 | if (overlap_clear) |
1295 | for (i = disks; i--; ) { | |
1296 | struct r5dev *dev = &sh->dev[i]; | |
1297 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1298 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1299 | } | |
d6f38f31 | 1300 | put_cpu(); |
91c00924 DW |
1301 | } |
1302 | ||
417b8d4a DW |
1303 | #ifdef CONFIG_MULTICORE_RAID456 |
1304 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1305 | { | |
1306 | struct stripe_head *sh = param; | |
1307 | unsigned long ops_request = sh->ops.request; | |
1308 | ||
1309 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1310 | wake_up(&sh->ops.wait_for_ops); | |
1311 | ||
1312 | __raid_run_ops(sh, ops_request); | |
1313 | release_stripe(sh); | |
1314 | } | |
1315 | ||
1316 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1317 | { | |
1318 | /* since handle_stripe can be called outside of raid5d context | |
1319 | * we need to ensure sh->ops.request is de-staged before another | |
1320 | * request arrives | |
1321 | */ | |
1322 | wait_event(sh->ops.wait_for_ops, | |
1323 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1324 | sh->ops.request = ops_request; | |
1325 | ||
1326 | atomic_inc(&sh->count); | |
1327 | async_schedule(async_run_ops, sh); | |
1328 | } | |
1329 | #else | |
1330 | #define raid_run_ops __raid_run_ops | |
1331 | #endif | |
1332 | ||
d1688a6d | 1333 | static int grow_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1334 | { |
1335 | struct stripe_head *sh; | |
6ce32846 | 1336 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1337 | if (!sh) |
1338 | return 0; | |
6ce32846 | 1339 | |
3f294f4f | 1340 | sh->raid_conf = conf; |
417b8d4a DW |
1341 | #ifdef CONFIG_MULTICORE_RAID456 |
1342 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1343 | #endif | |
3f294f4f | 1344 | |
e4e11e38 N |
1345 | if (grow_buffers(sh)) { |
1346 | shrink_buffers(sh); | |
3f294f4f N |
1347 | kmem_cache_free(conf->slab_cache, sh); |
1348 | return 0; | |
1349 | } | |
1350 | /* we just created an active stripe so... */ | |
1351 | atomic_set(&sh->count, 1); | |
1352 | atomic_inc(&conf->active_stripes); | |
1353 | INIT_LIST_HEAD(&sh->lru); | |
1354 | release_stripe(sh); | |
1355 | return 1; | |
1356 | } | |
1357 | ||
d1688a6d | 1358 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 1359 | { |
e18b890b | 1360 | struct kmem_cache *sc; |
5e5e3e78 | 1361 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1362 | |
f4be6b43 N |
1363 | if (conf->mddev->gendisk) |
1364 | sprintf(conf->cache_name[0], | |
1365 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1366 | else | |
1367 | sprintf(conf->cache_name[0], | |
1368 | "raid%d-%p", conf->level, conf->mddev); | |
1369 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1370 | ||
ad01c9e3 N |
1371 | conf->active_name = 0; |
1372 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1373 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1374 | 0, 0, NULL); |
1da177e4 LT |
1375 | if (!sc) |
1376 | return 1; | |
1377 | conf->slab_cache = sc; | |
ad01c9e3 | 1378 | conf->pool_size = devs; |
16a53ecc | 1379 | while (num--) |
3f294f4f | 1380 | if (!grow_one_stripe(conf)) |
1da177e4 | 1381 | return 1; |
1da177e4 LT |
1382 | return 0; |
1383 | } | |
29269553 | 1384 | |
d6f38f31 DW |
1385 | /** |
1386 | * scribble_len - return the required size of the scribble region | |
1387 | * @num - total number of disks in the array | |
1388 | * | |
1389 | * The size must be enough to contain: | |
1390 | * 1/ a struct page pointer for each device in the array +2 | |
1391 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1392 | * (dma_map_page()) or page (page_address()) address. | |
1393 | * | |
1394 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1395 | * calculate over all devices (not just the data blocks), using zeros in place | |
1396 | * of the P and Q blocks. | |
1397 | */ | |
1398 | static size_t scribble_len(int num) | |
1399 | { | |
1400 | size_t len; | |
1401 | ||
1402 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1403 | ||
1404 | return len; | |
1405 | } | |
1406 | ||
d1688a6d | 1407 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
1408 | { |
1409 | /* Make all the stripes able to hold 'newsize' devices. | |
1410 | * New slots in each stripe get 'page' set to a new page. | |
1411 | * | |
1412 | * This happens in stages: | |
1413 | * 1/ create a new kmem_cache and allocate the required number of | |
1414 | * stripe_heads. | |
1415 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1416 | * to the new stripe_heads. This will have the side effect of | |
1417 | * freezing the array as once all stripe_heads have been collected, | |
1418 | * no IO will be possible. Old stripe heads are freed once their | |
1419 | * pages have been transferred over, and the old kmem_cache is | |
1420 | * freed when all stripes are done. | |
1421 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1422 | * we simple return a failre status - no need to clean anything up. | |
1423 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1424 | * If this fails, we don't bother trying the shrink the | |
1425 | * stripe_heads down again, we just leave them as they are. | |
1426 | * As each stripe_head is processed the new one is released into | |
1427 | * active service. | |
1428 | * | |
1429 | * Once step2 is started, we cannot afford to wait for a write, | |
1430 | * so we use GFP_NOIO allocations. | |
1431 | */ | |
1432 | struct stripe_head *osh, *nsh; | |
1433 | LIST_HEAD(newstripes); | |
1434 | struct disk_info *ndisks; | |
d6f38f31 | 1435 | unsigned long cpu; |
b5470dc5 | 1436 | int err; |
e18b890b | 1437 | struct kmem_cache *sc; |
ad01c9e3 N |
1438 | int i; |
1439 | ||
1440 | if (newsize <= conf->pool_size) | |
1441 | return 0; /* never bother to shrink */ | |
1442 | ||
b5470dc5 DW |
1443 | err = md_allow_write(conf->mddev); |
1444 | if (err) | |
1445 | return err; | |
2a2275d6 | 1446 | |
ad01c9e3 N |
1447 | /* Step 1 */ |
1448 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1449 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1450 | 0, 0, NULL); |
ad01c9e3 N |
1451 | if (!sc) |
1452 | return -ENOMEM; | |
1453 | ||
1454 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1455 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1456 | if (!nsh) |
1457 | break; | |
1458 | ||
ad01c9e3 | 1459 | nsh->raid_conf = conf; |
417b8d4a DW |
1460 | #ifdef CONFIG_MULTICORE_RAID456 |
1461 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1462 | #endif | |
ad01c9e3 N |
1463 | |
1464 | list_add(&nsh->lru, &newstripes); | |
1465 | } | |
1466 | if (i) { | |
1467 | /* didn't get enough, give up */ | |
1468 | while (!list_empty(&newstripes)) { | |
1469 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1470 | list_del(&nsh->lru); | |
1471 | kmem_cache_free(sc, nsh); | |
1472 | } | |
1473 | kmem_cache_destroy(sc); | |
1474 | return -ENOMEM; | |
1475 | } | |
1476 | /* Step 2 - Must use GFP_NOIO now. | |
1477 | * OK, we have enough stripes, start collecting inactive | |
1478 | * stripes and copying them over | |
1479 | */ | |
1480 | list_for_each_entry(nsh, &newstripes, lru) { | |
1481 | spin_lock_irq(&conf->device_lock); | |
1482 | wait_event_lock_irq(conf->wait_for_stripe, | |
1483 | !list_empty(&conf->inactive_list), | |
1484 | conf->device_lock, | |
482c0834 | 1485 | ); |
ad01c9e3 N |
1486 | osh = get_free_stripe(conf); |
1487 | spin_unlock_irq(&conf->device_lock); | |
1488 | atomic_set(&nsh->count, 1); | |
1489 | for(i=0; i<conf->pool_size; i++) | |
1490 | nsh->dev[i].page = osh->dev[i].page; | |
1491 | for( ; i<newsize; i++) | |
1492 | nsh->dev[i].page = NULL; | |
1493 | kmem_cache_free(conf->slab_cache, osh); | |
1494 | } | |
1495 | kmem_cache_destroy(conf->slab_cache); | |
1496 | ||
1497 | /* Step 3. | |
1498 | * At this point, we are holding all the stripes so the array | |
1499 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1500 | * conf->disks and the scribble region |
ad01c9e3 N |
1501 | */ |
1502 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1503 | if (ndisks) { | |
1504 | for (i=0; i<conf->raid_disks; i++) | |
1505 | ndisks[i] = conf->disks[i]; | |
1506 | kfree(conf->disks); | |
1507 | conf->disks = ndisks; | |
1508 | } else | |
1509 | err = -ENOMEM; | |
1510 | ||
d6f38f31 DW |
1511 | get_online_cpus(); |
1512 | conf->scribble_len = scribble_len(newsize); | |
1513 | for_each_present_cpu(cpu) { | |
1514 | struct raid5_percpu *percpu; | |
1515 | void *scribble; | |
1516 | ||
1517 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1518 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1519 | ||
1520 | if (scribble) { | |
1521 | kfree(percpu->scribble); | |
1522 | percpu->scribble = scribble; | |
1523 | } else { | |
1524 | err = -ENOMEM; | |
1525 | break; | |
1526 | } | |
1527 | } | |
1528 | put_online_cpus(); | |
1529 | ||
ad01c9e3 N |
1530 | /* Step 4, return new stripes to service */ |
1531 | while(!list_empty(&newstripes)) { | |
1532 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1533 | list_del_init(&nsh->lru); | |
d6f38f31 | 1534 | |
ad01c9e3 N |
1535 | for (i=conf->raid_disks; i < newsize; i++) |
1536 | if (nsh->dev[i].page == NULL) { | |
1537 | struct page *p = alloc_page(GFP_NOIO); | |
1538 | nsh->dev[i].page = p; | |
1539 | if (!p) | |
1540 | err = -ENOMEM; | |
1541 | } | |
1542 | release_stripe(nsh); | |
1543 | } | |
1544 | /* critical section pass, GFP_NOIO no longer needed */ | |
1545 | ||
1546 | conf->slab_cache = sc; | |
1547 | conf->active_name = 1-conf->active_name; | |
1548 | conf->pool_size = newsize; | |
1549 | return err; | |
1550 | } | |
1da177e4 | 1551 | |
d1688a6d | 1552 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1553 | { |
1554 | struct stripe_head *sh; | |
1555 | ||
3f294f4f N |
1556 | spin_lock_irq(&conf->device_lock); |
1557 | sh = get_free_stripe(conf); | |
1558 | spin_unlock_irq(&conf->device_lock); | |
1559 | if (!sh) | |
1560 | return 0; | |
78bafebd | 1561 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1562 | shrink_buffers(sh); |
3f294f4f N |
1563 | kmem_cache_free(conf->slab_cache, sh); |
1564 | atomic_dec(&conf->active_stripes); | |
1565 | return 1; | |
1566 | } | |
1567 | ||
d1688a6d | 1568 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f N |
1569 | { |
1570 | while (drop_one_stripe(conf)) | |
1571 | ; | |
1572 | ||
29fc7e3e N |
1573 | if (conf->slab_cache) |
1574 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1575 | conf->slab_cache = NULL; |
1576 | } | |
1577 | ||
6712ecf8 | 1578 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1579 | { |
99c0fb5f | 1580 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1581 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1582 | int disks = sh->disks, i; |
1da177e4 | 1583 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 1584 | char b[BDEVNAME_SIZE]; |
3cb03002 | 1585 | struct md_rdev *rdev; |
1da177e4 | 1586 | |
1da177e4 LT |
1587 | |
1588 | for (i=0 ; i<disks; i++) | |
1589 | if (bi == &sh->dev[i].req) | |
1590 | break; | |
1591 | ||
45b4233c DW |
1592 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1593 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1594 | uptodate); |
1595 | if (i == disks) { | |
1596 | BUG(); | |
6712ecf8 | 1597 | return; |
1da177e4 LT |
1598 | } |
1599 | ||
1600 | if (uptodate) { | |
1da177e4 | 1601 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1602 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1603 | rdev = conf->disks[i].rdev; |
8bda470e CD |
1604 | printk_ratelimited( |
1605 | KERN_INFO | |
1606 | "md/raid:%s: read error corrected" | |
1607 | " (%lu sectors at %llu on %s)\n", | |
1608 | mdname(conf->mddev), STRIPE_SECTORS, | |
1609 | (unsigned long long)(sh->sector | |
1610 | + rdev->data_offset), | |
1611 | bdevname(rdev->bdev, b)); | |
ddd5115f | 1612 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
1613 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1614 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1615 | } | |
ba22dcbf N |
1616 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1617 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1618 | } else { |
d6950432 | 1619 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1620 | int retry = 0; |
d6950432 N |
1621 | rdev = conf->disks[i].rdev; |
1622 | ||
1da177e4 | 1623 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1624 | atomic_inc(&rdev->read_errors); |
7b0bb536 | 1625 | if (conf->mddev->degraded >= conf->max_degraded) |
8bda470e CD |
1626 | printk_ratelimited( |
1627 | KERN_WARNING | |
1628 | "md/raid:%s: read error not correctable " | |
1629 | "(sector %llu on %s).\n", | |
1630 | mdname(conf->mddev), | |
1631 | (unsigned long long)(sh->sector | |
1632 | + rdev->data_offset), | |
1633 | bdn); | |
ba22dcbf | 1634 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1635 | /* Oh, no!!! */ |
8bda470e CD |
1636 | printk_ratelimited( |
1637 | KERN_WARNING | |
1638 | "md/raid:%s: read error NOT corrected!! " | |
1639 | "(sector %llu on %s).\n", | |
1640 | mdname(conf->mddev), | |
1641 | (unsigned long long)(sh->sector | |
1642 | + rdev->data_offset), | |
1643 | bdn); | |
d6950432 | 1644 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1645 | > conf->max_nr_stripes) |
14f8d26b | 1646 | printk(KERN_WARNING |
0c55e022 | 1647 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1648 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1649 | else |
1650 | retry = 1; | |
1651 | if (retry) | |
1652 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1653 | else { | |
4e5314b5 N |
1654 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1655 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1656 | md_error(conf->mddev, rdev); |
ba22dcbf | 1657 | } |
1da177e4 LT |
1658 | } |
1659 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1660 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1661 | set_bit(STRIPE_HANDLE, &sh->state); | |
1662 | release_stripe(sh); | |
1da177e4 LT |
1663 | } |
1664 | ||
d710e138 | 1665 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1666 | { |
99c0fb5f | 1667 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1668 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1669 | int disks = sh->disks, i; |
1da177e4 | 1670 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
1671 | sector_t first_bad; |
1672 | int bad_sectors; | |
1da177e4 | 1673 | |
1da177e4 LT |
1674 | for (i=0 ; i<disks; i++) |
1675 | if (bi == &sh->dev[i].req) | |
1676 | break; | |
1677 | ||
45b4233c | 1678 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1679 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1680 | uptodate); | |
1681 | if (i == disks) { | |
1682 | BUG(); | |
6712ecf8 | 1683 | return; |
1da177e4 LT |
1684 | } |
1685 | ||
bc2607f3 N |
1686 | if (!uptodate) { |
1687 | set_bit(WriteErrorSeen, &conf->disks[i].rdev->flags); | |
1688 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
b84db560 N |
1689 | } else if (is_badblock(conf->disks[i].rdev, sh->sector, STRIPE_SECTORS, |
1690 | &first_bad, &bad_sectors)) | |
1691 | set_bit(R5_MadeGood, &sh->dev[i].flags); | |
1da177e4 LT |
1692 | |
1693 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1694 | ||
1695 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1696 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1697 | release_stripe(sh); |
1da177e4 LT |
1698 | } |
1699 | ||
1700 | ||
784052ec | 1701 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1702 | |
784052ec | 1703 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1704 | { |
1705 | struct r5dev *dev = &sh->dev[i]; | |
1706 | ||
1707 | bio_init(&dev->req); | |
1708 | dev->req.bi_io_vec = &dev->vec; | |
1709 | dev->req.bi_vcnt++; | |
1710 | dev->req.bi_max_vecs++; | |
1711 | dev->vec.bv_page = dev->page; | |
1712 | dev->vec.bv_len = STRIPE_SIZE; | |
1713 | dev->vec.bv_offset = 0; | |
1714 | ||
1715 | dev->req.bi_sector = sh->sector; | |
1716 | dev->req.bi_private = sh; | |
1717 | ||
1718 | dev->flags = 0; | |
784052ec | 1719 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1720 | } |
1721 | ||
fd01b88c | 1722 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1723 | { |
1724 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 1725 | struct r5conf *conf = mddev->private; |
0c55e022 | 1726 | pr_debug("raid456: error called\n"); |
1da177e4 | 1727 | |
6f8d0c77 N |
1728 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1729 | unsigned long flags; | |
1730 | spin_lock_irqsave(&conf->device_lock, flags); | |
1731 | mddev->degraded++; | |
1732 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1733 | /* | |
1734 | * if recovery was running, make sure it aborts. | |
1735 | */ | |
1736 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1da177e4 | 1737 | } |
de393cde | 1738 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
1739 | set_bit(Faulty, &rdev->flags); |
1740 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1741 | printk(KERN_ALERT | |
1742 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1743 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1744 | mdname(mddev), | |
1745 | bdevname(rdev->bdev, b), | |
1746 | mdname(mddev), | |
1747 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1748 | } |
1da177e4 LT |
1749 | |
1750 | /* | |
1751 | * Input: a 'big' sector number, | |
1752 | * Output: index of the data and parity disk, and the sector # in them. | |
1753 | */ | |
d1688a6d | 1754 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
1755 | int previous, int *dd_idx, |
1756 | struct stripe_head *sh) | |
1da177e4 | 1757 | { |
6e3b96ed | 1758 | sector_t stripe, stripe2; |
35f2a591 | 1759 | sector_t chunk_number; |
1da177e4 | 1760 | unsigned int chunk_offset; |
911d4ee8 | 1761 | int pd_idx, qd_idx; |
67cc2b81 | 1762 | int ddf_layout = 0; |
1da177e4 | 1763 | sector_t new_sector; |
e183eaed N |
1764 | int algorithm = previous ? conf->prev_algo |
1765 | : conf->algorithm; | |
09c9e5fa AN |
1766 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1767 | : conf->chunk_sectors; | |
112bf897 N |
1768 | int raid_disks = previous ? conf->previous_raid_disks |
1769 | : conf->raid_disks; | |
1770 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1771 | |
1772 | /* First compute the information on this sector */ | |
1773 | ||
1774 | /* | |
1775 | * Compute the chunk number and the sector offset inside the chunk | |
1776 | */ | |
1777 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1778 | chunk_number = r_sector; | |
1da177e4 LT |
1779 | |
1780 | /* | |
1781 | * Compute the stripe number | |
1782 | */ | |
35f2a591 N |
1783 | stripe = chunk_number; |
1784 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1785 | stripe2 = stripe; |
1da177e4 LT |
1786 | /* |
1787 | * Select the parity disk based on the user selected algorithm. | |
1788 | */ | |
84789554 | 1789 | pd_idx = qd_idx = -1; |
16a53ecc N |
1790 | switch(conf->level) { |
1791 | case 4: | |
911d4ee8 | 1792 | pd_idx = data_disks; |
16a53ecc N |
1793 | break; |
1794 | case 5: | |
e183eaed | 1795 | switch (algorithm) { |
1da177e4 | 1796 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1797 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1798 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1799 | (*dd_idx)++; |
1800 | break; | |
1801 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1802 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1803 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1804 | (*dd_idx)++; |
1805 | break; | |
1806 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1807 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1808 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1809 | break; |
1810 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1811 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1812 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 1813 | break; |
99c0fb5f N |
1814 | case ALGORITHM_PARITY_0: |
1815 | pd_idx = 0; | |
1816 | (*dd_idx)++; | |
1817 | break; | |
1818 | case ALGORITHM_PARITY_N: | |
1819 | pd_idx = data_disks; | |
1820 | break; | |
1da177e4 | 1821 | default: |
99c0fb5f | 1822 | BUG(); |
16a53ecc N |
1823 | } |
1824 | break; | |
1825 | case 6: | |
1826 | ||
e183eaed | 1827 | switch (algorithm) { |
16a53ecc | 1828 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1829 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1830 | qd_idx = pd_idx + 1; |
1831 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1832 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1833 | qd_idx = 0; |
1834 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1835 | (*dd_idx) += 2; /* D D P Q D */ |
1836 | break; | |
1837 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1838 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1839 | qd_idx = pd_idx + 1; |
1840 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1841 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1842 | qd_idx = 0; |
1843 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1844 | (*dd_idx) += 2; /* D D P Q D */ |
1845 | break; | |
1846 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1847 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1848 | qd_idx = (pd_idx + 1) % raid_disks; |
1849 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1850 | break; |
1851 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1852 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1853 | qd_idx = (pd_idx + 1) % raid_disks; |
1854 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1855 | break; |
99c0fb5f N |
1856 | |
1857 | case ALGORITHM_PARITY_0: | |
1858 | pd_idx = 0; | |
1859 | qd_idx = 1; | |
1860 | (*dd_idx) += 2; | |
1861 | break; | |
1862 | case ALGORITHM_PARITY_N: | |
1863 | pd_idx = data_disks; | |
1864 | qd_idx = data_disks + 1; | |
1865 | break; | |
1866 | ||
1867 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1868 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1869 | * of blocks for computing Q is different. | |
1870 | */ | |
6e3b96ed | 1871 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
1872 | qd_idx = pd_idx + 1; |
1873 | if (pd_idx == raid_disks-1) { | |
1874 | (*dd_idx)++; /* Q D D D P */ | |
1875 | qd_idx = 0; | |
1876 | } else if (*dd_idx >= pd_idx) | |
1877 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1878 | ddf_layout = 1; |
99c0fb5f N |
1879 | break; |
1880 | ||
1881 | case ALGORITHM_ROTATING_N_RESTART: | |
1882 | /* Same a left_asymmetric, by first stripe is | |
1883 | * D D D P Q rather than | |
1884 | * Q D D D P | |
1885 | */ | |
6e3b96ed N |
1886 | stripe2 += 1; |
1887 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
1888 | qd_idx = pd_idx + 1; |
1889 | if (pd_idx == raid_disks-1) { | |
1890 | (*dd_idx)++; /* Q D D D P */ | |
1891 | qd_idx = 0; | |
1892 | } else if (*dd_idx >= pd_idx) | |
1893 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1894 | ddf_layout = 1; |
99c0fb5f N |
1895 | break; |
1896 | ||
1897 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1898 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 1899 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
1900 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
1901 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1902 | ddf_layout = 1; |
99c0fb5f N |
1903 | break; |
1904 | ||
1905 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1906 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 1907 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1908 | if (*dd_idx >= pd_idx) |
1909 | (*dd_idx)++; | |
1910 | qd_idx = raid_disks - 1; | |
1911 | break; | |
1912 | ||
1913 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 1914 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1915 | if (*dd_idx >= pd_idx) |
1916 | (*dd_idx)++; | |
1917 | qd_idx = raid_disks - 1; | |
1918 | break; | |
1919 | ||
1920 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 1921 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1922 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1923 | qd_idx = raid_disks - 1; | |
1924 | break; | |
1925 | ||
1926 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 1927 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1928 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1929 | qd_idx = raid_disks - 1; | |
1930 | break; | |
1931 | ||
1932 | case ALGORITHM_PARITY_0_6: | |
1933 | pd_idx = 0; | |
1934 | (*dd_idx)++; | |
1935 | qd_idx = raid_disks - 1; | |
1936 | break; | |
1937 | ||
16a53ecc | 1938 | default: |
99c0fb5f | 1939 | BUG(); |
16a53ecc N |
1940 | } |
1941 | break; | |
1da177e4 LT |
1942 | } |
1943 | ||
911d4ee8 N |
1944 | if (sh) { |
1945 | sh->pd_idx = pd_idx; | |
1946 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1947 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1948 | } |
1da177e4 LT |
1949 | /* |
1950 | * Finally, compute the new sector number | |
1951 | */ | |
1952 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1953 | return new_sector; | |
1954 | } | |
1955 | ||
1956 | ||
784052ec | 1957 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 1958 | { |
d1688a6d | 1959 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
1960 | int raid_disks = sh->disks; |
1961 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1962 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
1963 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1964 | : conf->chunk_sectors; | |
e183eaed N |
1965 | int algorithm = previous ? conf->prev_algo |
1966 | : conf->algorithm; | |
1da177e4 LT |
1967 | sector_t stripe; |
1968 | int chunk_offset; | |
35f2a591 N |
1969 | sector_t chunk_number; |
1970 | int dummy1, dd_idx = i; | |
1da177e4 | 1971 | sector_t r_sector; |
911d4ee8 | 1972 | struct stripe_head sh2; |
1da177e4 | 1973 | |
16a53ecc | 1974 | |
1da177e4 LT |
1975 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1976 | stripe = new_sector; | |
1da177e4 | 1977 | |
16a53ecc N |
1978 | if (i == sh->pd_idx) |
1979 | return 0; | |
1980 | switch(conf->level) { | |
1981 | case 4: break; | |
1982 | case 5: | |
e183eaed | 1983 | switch (algorithm) { |
1da177e4 LT |
1984 | case ALGORITHM_LEFT_ASYMMETRIC: |
1985 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1986 | if (i > sh->pd_idx) | |
1987 | i--; | |
1988 | break; | |
1989 | case ALGORITHM_LEFT_SYMMETRIC: | |
1990 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1991 | if (i < sh->pd_idx) | |
1992 | i += raid_disks; | |
1993 | i -= (sh->pd_idx + 1); | |
1994 | break; | |
99c0fb5f N |
1995 | case ALGORITHM_PARITY_0: |
1996 | i -= 1; | |
1997 | break; | |
1998 | case ALGORITHM_PARITY_N: | |
1999 | break; | |
1da177e4 | 2000 | default: |
99c0fb5f | 2001 | BUG(); |
16a53ecc N |
2002 | } |
2003 | break; | |
2004 | case 6: | |
d0dabf7e | 2005 | if (i == sh->qd_idx) |
16a53ecc | 2006 | return 0; /* It is the Q disk */ |
e183eaed | 2007 | switch (algorithm) { |
16a53ecc N |
2008 | case ALGORITHM_LEFT_ASYMMETRIC: |
2009 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2010 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2011 | case ALGORITHM_ROTATING_N_RESTART: | |
2012 | if (sh->pd_idx == raid_disks-1) | |
2013 | i--; /* Q D D D P */ | |
16a53ecc N |
2014 | else if (i > sh->pd_idx) |
2015 | i -= 2; /* D D P Q D */ | |
2016 | break; | |
2017 | case ALGORITHM_LEFT_SYMMETRIC: | |
2018 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2019 | if (sh->pd_idx == raid_disks-1) | |
2020 | i--; /* Q D D D P */ | |
2021 | else { | |
2022 | /* D D P Q D */ | |
2023 | if (i < sh->pd_idx) | |
2024 | i += raid_disks; | |
2025 | i -= (sh->pd_idx + 2); | |
2026 | } | |
2027 | break; | |
99c0fb5f N |
2028 | case ALGORITHM_PARITY_0: |
2029 | i -= 2; | |
2030 | break; | |
2031 | case ALGORITHM_PARITY_N: | |
2032 | break; | |
2033 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2034 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2035 | if (sh->pd_idx == 0) |
2036 | i--; /* P D D D Q */ | |
e4424fee N |
2037 | else { |
2038 | /* D D Q P D */ | |
2039 | if (i < sh->pd_idx) | |
2040 | i += raid_disks; | |
2041 | i -= (sh->pd_idx + 1); | |
2042 | } | |
99c0fb5f N |
2043 | break; |
2044 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2045 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2046 | if (i > sh->pd_idx) | |
2047 | i--; | |
2048 | break; | |
2049 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2050 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2051 | if (i < sh->pd_idx) | |
2052 | i += data_disks + 1; | |
2053 | i -= (sh->pd_idx + 1); | |
2054 | break; | |
2055 | case ALGORITHM_PARITY_0_6: | |
2056 | i -= 1; | |
2057 | break; | |
16a53ecc | 2058 | default: |
99c0fb5f | 2059 | BUG(); |
16a53ecc N |
2060 | } |
2061 | break; | |
1da177e4 LT |
2062 | } |
2063 | ||
2064 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2065 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2066 | |
112bf897 | 2067 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2068 | previous, &dummy1, &sh2); |
911d4ee8 N |
2069 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2070 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2071 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2072 | mdname(conf->mddev)); | |
1da177e4 LT |
2073 | return 0; |
2074 | } | |
2075 | return r_sector; | |
2076 | } | |
2077 | ||
2078 | ||
600aa109 | 2079 | static void |
c0f7bddb | 2080 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2081 | int rcw, int expand) |
e33129d8 DW |
2082 | { |
2083 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
d1688a6d | 2084 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2085 | int level = conf->level; |
e33129d8 DW |
2086 | |
2087 | if (rcw) { | |
2088 | /* if we are not expanding this is a proper write request, and | |
2089 | * there will be bios with new data to be drained into the | |
2090 | * stripe cache | |
2091 | */ | |
2092 | if (!expand) { | |
600aa109 DW |
2093 | sh->reconstruct_state = reconstruct_state_drain_run; |
2094 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2095 | } else | |
2096 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2097 | |
ac6b53b6 | 2098 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2099 | |
2100 | for (i = disks; i--; ) { | |
2101 | struct r5dev *dev = &sh->dev[i]; | |
2102 | ||
2103 | if (dev->towrite) { | |
2104 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2105 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2106 | if (!expand) |
2107 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2108 | s->locked++; |
e33129d8 DW |
2109 | } |
2110 | } | |
c0f7bddb | 2111 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2112 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2113 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2114 | } else { |
c0f7bddb | 2115 | BUG_ON(level == 6); |
e33129d8 DW |
2116 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2117 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2118 | ||
d8ee0728 | 2119 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2120 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2121 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2122 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2123 | |
2124 | for (i = disks; i--; ) { | |
2125 | struct r5dev *dev = &sh->dev[i]; | |
2126 | if (i == pd_idx) | |
2127 | continue; | |
2128 | ||
e33129d8 DW |
2129 | if (dev->towrite && |
2130 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2131 | test_bit(R5_Wantcompute, &dev->flags))) { |
2132 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2133 | set_bit(R5_LOCKED, &dev->flags); |
2134 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2135 | s->locked++; |
e33129d8 DW |
2136 | } |
2137 | } | |
2138 | } | |
2139 | ||
c0f7bddb | 2140 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2141 | * are in flight |
2142 | */ | |
2143 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2144 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2145 | s->locked++; |
e33129d8 | 2146 | |
c0f7bddb YT |
2147 | if (level == 6) { |
2148 | int qd_idx = sh->qd_idx; | |
2149 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2150 | ||
2151 | set_bit(R5_LOCKED, &dev->flags); | |
2152 | clear_bit(R5_UPTODATE, &dev->flags); | |
2153 | s->locked++; | |
2154 | } | |
2155 | ||
600aa109 | 2156 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2157 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2158 | s->locked, s->ops_request); |
e33129d8 | 2159 | } |
16a53ecc | 2160 | |
1da177e4 LT |
2161 | /* |
2162 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2163 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2164 | * The bi_next chain must be in order. |
2165 | */ | |
2166 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2167 | { | |
2168 | struct bio **bip; | |
d1688a6d | 2169 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2170 | int firstwrite=0; |
1da177e4 | 2171 | |
cbe47ec5 | 2172 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2173 | (unsigned long long)bi->bi_sector, |
2174 | (unsigned long long)sh->sector); | |
2175 | ||
2176 | ||
1da177e4 | 2177 | spin_lock_irq(&conf->device_lock); |
72626685 | 2178 | if (forwrite) { |
1da177e4 | 2179 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2180 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2181 | firstwrite = 1; | |
2182 | } else | |
1da177e4 LT |
2183 | bip = &sh->dev[dd_idx].toread; |
2184 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2185 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2186 | goto overlap; | |
2187 | bip = & (*bip)->bi_next; | |
2188 | } | |
2189 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2190 | goto overlap; | |
2191 | ||
78bafebd | 2192 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2193 | if (*bip) |
2194 | bi->bi_next = *bip; | |
2195 | *bip = bi; | |
960e739d | 2196 | bi->bi_phys_segments++; |
72626685 | 2197 | |
1da177e4 LT |
2198 | if (forwrite) { |
2199 | /* check if page is covered */ | |
2200 | sector_t sector = sh->dev[dd_idx].sector; | |
2201 | for (bi=sh->dev[dd_idx].towrite; | |
2202 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2203 | bi && bi->bi_sector <= sector; | |
2204 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2205 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2206 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2207 | } | |
2208 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2209 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2210 | } | |
cbe47ec5 | 2211 | spin_unlock_irq(&conf->device_lock); |
cbe47ec5 N |
2212 | |
2213 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2214 | (unsigned long long)(*bip)->bi_sector, | |
2215 | (unsigned long long)sh->sector, dd_idx); | |
2216 | ||
2217 | if (conf->mddev->bitmap && firstwrite) { | |
2218 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2219 | STRIPE_SECTORS, 0); | |
2220 | sh->bm_seq = conf->seq_flush+1; | |
2221 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2222 | } | |
1da177e4 LT |
2223 | return 1; |
2224 | ||
2225 | overlap: | |
2226 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2227 | spin_unlock_irq(&conf->device_lock); | |
1da177e4 LT |
2228 | return 0; |
2229 | } | |
2230 | ||
d1688a6d | 2231 | static void end_reshape(struct r5conf *conf); |
29269553 | 2232 | |
d1688a6d | 2233 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 2234 | struct stripe_head *sh) |
ccfcc3c1 | 2235 | { |
784052ec | 2236 | int sectors_per_chunk = |
09c9e5fa | 2237 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2238 | int dd_idx; |
2d2063ce | 2239 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2240 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2241 | |
112bf897 N |
2242 | raid5_compute_sector(conf, |
2243 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2244 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2245 | previous, |
911d4ee8 | 2246 | &dd_idx, sh); |
ccfcc3c1 N |
2247 | } |
2248 | ||
a4456856 | 2249 | static void |
d1688a6d | 2250 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2251 | struct stripe_head_state *s, int disks, |
2252 | struct bio **return_bi) | |
2253 | { | |
2254 | int i; | |
2255 | for (i = disks; i--; ) { | |
2256 | struct bio *bi; | |
2257 | int bitmap_end = 0; | |
2258 | ||
2259 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 2260 | struct md_rdev *rdev; |
a4456856 DW |
2261 | rcu_read_lock(); |
2262 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2263 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
2264 | atomic_inc(&rdev->nr_pending); |
2265 | else | |
2266 | rdev = NULL; | |
a4456856 | 2267 | rcu_read_unlock(); |
7f0da59b N |
2268 | if (rdev) { |
2269 | if (!rdev_set_badblocks( | |
2270 | rdev, | |
2271 | sh->sector, | |
2272 | STRIPE_SECTORS, 0)) | |
2273 | md_error(conf->mddev, rdev); | |
2274 | rdev_dec_pending(rdev, conf->mddev); | |
2275 | } | |
a4456856 DW |
2276 | } |
2277 | spin_lock_irq(&conf->device_lock); | |
2278 | /* fail all writes first */ | |
2279 | bi = sh->dev[i].towrite; | |
2280 | sh->dev[i].towrite = NULL; | |
2281 | if (bi) { | |
2282 | s->to_write--; | |
2283 | bitmap_end = 1; | |
2284 | } | |
2285 | ||
2286 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2287 | wake_up(&conf->wait_for_overlap); | |
2288 | ||
2289 | while (bi && bi->bi_sector < | |
2290 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2291 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2292 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2293 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2294 | md_write_end(conf->mddev); |
2295 | bi->bi_next = *return_bi; | |
2296 | *return_bi = bi; | |
2297 | } | |
2298 | bi = nextbi; | |
2299 | } | |
2300 | /* and fail all 'written' */ | |
2301 | bi = sh->dev[i].written; | |
2302 | sh->dev[i].written = NULL; | |
2303 | if (bi) bitmap_end = 1; | |
2304 | while (bi && bi->bi_sector < | |
2305 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2306 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2307 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2308 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2309 | md_write_end(conf->mddev); |
2310 | bi->bi_next = *return_bi; | |
2311 | *return_bi = bi; | |
2312 | } | |
2313 | bi = bi2; | |
2314 | } | |
2315 | ||
b5e98d65 DW |
2316 | /* fail any reads if this device is non-operational and |
2317 | * the data has not reached the cache yet. | |
2318 | */ | |
2319 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2320 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2321 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2322 | bi = sh->dev[i].toread; |
2323 | sh->dev[i].toread = NULL; | |
2324 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2325 | wake_up(&conf->wait_for_overlap); | |
2326 | if (bi) s->to_read--; | |
2327 | while (bi && bi->bi_sector < | |
2328 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2329 | struct bio *nextbi = | |
2330 | r5_next_bio(bi, sh->dev[i].sector); | |
2331 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2332 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2333 | bi->bi_next = *return_bi; |
2334 | *return_bi = bi; | |
2335 | } | |
2336 | bi = nextbi; | |
2337 | } | |
2338 | } | |
2339 | spin_unlock_irq(&conf->device_lock); | |
2340 | if (bitmap_end) | |
2341 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2342 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
2343 | /* If we were in the middle of a write the parity block might |
2344 | * still be locked - so just clear all R5_LOCKED flags | |
2345 | */ | |
2346 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
2347 | } |
2348 | ||
8b3e6cdc DW |
2349 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2350 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2351 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2352 | } |
2353 | ||
7f0da59b | 2354 | static void |
d1688a6d | 2355 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
2356 | struct stripe_head_state *s) |
2357 | { | |
2358 | int abort = 0; | |
2359 | int i; | |
2360 | ||
2361 | md_done_sync(conf->mddev, STRIPE_SECTORS, 0); | |
2362 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2363 | s->syncing = 0; | |
2364 | /* There is nothing more to do for sync/check/repair. | |
2365 | * For recover we need to record a bad block on all | |
2366 | * non-sync devices, or abort the recovery | |
2367 | */ | |
2368 | if (!test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) | |
2369 | return; | |
2370 | /* During recovery devices cannot be removed, so locking and | |
2371 | * refcounting of rdevs is not needed | |
2372 | */ | |
2373 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 2374 | struct md_rdev *rdev = conf->disks[i].rdev; |
7f0da59b N |
2375 | if (!rdev |
2376 | || test_bit(Faulty, &rdev->flags) | |
2377 | || test_bit(In_sync, &rdev->flags)) | |
2378 | continue; | |
2379 | if (!rdev_set_badblocks(rdev, sh->sector, | |
2380 | STRIPE_SECTORS, 0)) | |
2381 | abort = 1; | |
2382 | } | |
2383 | if (abort) { | |
2384 | conf->recovery_disabled = conf->mddev->recovery_disabled; | |
2385 | set_bit(MD_RECOVERY_INTR, &conf->mddev->recovery); | |
2386 | } | |
2387 | } | |
2388 | ||
93b3dbce | 2389 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
2390 | * to be read or computed to satisfy a request. |
2391 | * | |
2392 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 2393 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 2394 | */ |
93b3dbce N |
2395 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
2396 | int disk_idx, int disks) | |
a4456856 | 2397 | { |
5599becc | 2398 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2399 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2400 | &sh->dev[s->failed_num[1]] }; | |
5599becc | 2401 | |
93b3dbce | 2402 | /* is the data in this block needed, and can we get it? */ |
5599becc YT |
2403 | if (!test_bit(R5_LOCKED, &dev->flags) && |
2404 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2405 | (dev->toread || | |
2406 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2407 | s->syncing || s->expanding || | |
5d35e09c N |
2408 | (s->failed >= 1 && fdev[0]->toread) || |
2409 | (s->failed >= 2 && fdev[1]->toread) || | |
93b3dbce N |
2410 | (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite && |
2411 | !test_bit(R5_OVERWRITE, &fdev[0]->flags)) || | |
2412 | (sh->raid_conf->level == 6 && s->failed && s->to_write))) { | |
5599becc YT |
2413 | /* we would like to get this block, possibly by computing it, |
2414 | * otherwise read it if the backing disk is insync | |
2415 | */ | |
2416 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2417 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2418 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
2419 | (s->failed && (disk_idx == s->failed_num[0] || |
2420 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
2421 | /* have disk failed, and we're requested to fetch it; |
2422 | * do compute it | |
a4456856 | 2423 | */ |
5599becc YT |
2424 | pr_debug("Computing stripe %llu block %d\n", |
2425 | (unsigned long long)sh->sector, disk_idx); | |
2426 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2427 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2428 | set_bit(R5_Wantcompute, &dev->flags); | |
2429 | sh->ops.target = disk_idx; | |
2430 | sh->ops.target2 = -1; /* no 2nd target */ | |
2431 | s->req_compute = 1; | |
93b3dbce N |
2432 | /* Careful: from this point on 'uptodate' is in the eye |
2433 | * of raid_run_ops which services 'compute' operations | |
2434 | * before writes. R5_Wantcompute flags a block that will | |
2435 | * be R5_UPTODATE by the time it is needed for a | |
2436 | * subsequent operation. | |
2437 | */ | |
5599becc YT |
2438 | s->uptodate++; |
2439 | return 1; | |
2440 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2441 | /* Computing 2-failure is *very* expensive; only | |
2442 | * do it if failed >= 2 | |
2443 | */ | |
2444 | int other; | |
2445 | for (other = disks; other--; ) { | |
2446 | if (other == disk_idx) | |
2447 | continue; | |
2448 | if (!test_bit(R5_UPTODATE, | |
2449 | &sh->dev[other].flags)) | |
2450 | break; | |
a4456856 | 2451 | } |
5599becc YT |
2452 | BUG_ON(other < 0); |
2453 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2454 | (unsigned long long)sh->sector, | |
2455 | disk_idx, other); | |
2456 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2457 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2458 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2459 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2460 | sh->ops.target = disk_idx; | |
2461 | sh->ops.target2 = other; | |
2462 | s->uptodate += 2; | |
2463 | s->req_compute = 1; | |
2464 | return 1; | |
2465 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2466 | set_bit(R5_LOCKED, &dev->flags); | |
2467 | set_bit(R5_Wantread, &dev->flags); | |
2468 | s->locked++; | |
2469 | pr_debug("Reading block %d (sync=%d)\n", | |
2470 | disk_idx, s->syncing); | |
a4456856 DW |
2471 | } |
2472 | } | |
5599becc YT |
2473 | |
2474 | return 0; | |
2475 | } | |
2476 | ||
2477 | /** | |
93b3dbce | 2478 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 2479 | */ |
93b3dbce N |
2480 | static void handle_stripe_fill(struct stripe_head *sh, |
2481 | struct stripe_head_state *s, | |
2482 | int disks) | |
5599becc YT |
2483 | { |
2484 | int i; | |
2485 | ||
2486 | /* look for blocks to read/compute, skip this if a compute | |
2487 | * is already in flight, or if the stripe contents are in the | |
2488 | * midst of changing due to a write | |
2489 | */ | |
2490 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2491 | !sh->reconstruct_state) | |
2492 | for (i = disks; i--; ) | |
93b3dbce | 2493 | if (fetch_block(sh, s, i, disks)) |
5599becc | 2494 | break; |
a4456856 DW |
2495 | set_bit(STRIPE_HANDLE, &sh->state); |
2496 | } | |
2497 | ||
2498 | ||
1fe797e6 | 2499 | /* handle_stripe_clean_event |
a4456856 DW |
2500 | * any written block on an uptodate or failed drive can be returned. |
2501 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2502 | * never LOCKED, so we don't need to test 'failed' directly. | |
2503 | */ | |
d1688a6d | 2504 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
2505 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2506 | { | |
2507 | int i; | |
2508 | struct r5dev *dev; | |
2509 | ||
2510 | for (i = disks; i--; ) | |
2511 | if (sh->dev[i].written) { | |
2512 | dev = &sh->dev[i]; | |
2513 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2514 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2515 | /* We can return any write requests */ | |
2516 | struct bio *wbi, *wbi2; | |
2517 | int bitmap_end = 0; | |
45b4233c | 2518 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2519 | spin_lock_irq(&conf->device_lock); |
2520 | wbi = dev->written; | |
2521 | dev->written = NULL; | |
2522 | while (wbi && wbi->bi_sector < | |
2523 | dev->sector + STRIPE_SECTORS) { | |
2524 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2525 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2526 | md_write_end(conf->mddev); |
2527 | wbi->bi_next = *return_bi; | |
2528 | *return_bi = wbi; | |
2529 | } | |
2530 | wbi = wbi2; | |
2531 | } | |
2532 | if (dev->towrite == NULL) | |
2533 | bitmap_end = 1; | |
2534 | spin_unlock_irq(&conf->device_lock); | |
2535 | if (bitmap_end) | |
2536 | bitmap_endwrite(conf->mddev->bitmap, | |
2537 | sh->sector, | |
2538 | STRIPE_SECTORS, | |
2539 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2540 | 0); | |
2541 | } | |
2542 | } | |
8b3e6cdc DW |
2543 | |
2544 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2545 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2546 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2547 | } |
2548 | ||
d1688a6d | 2549 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
2550 | struct stripe_head *sh, |
2551 | struct stripe_head_state *s, | |
2552 | int disks) | |
a4456856 DW |
2553 | { |
2554 | int rmw = 0, rcw = 0, i; | |
c8ac1803 N |
2555 | if (conf->max_degraded == 2) { |
2556 | /* RAID6 requires 'rcw' in current implementation | |
2557 | * Calculate the real rcw later - for now fake it | |
2558 | * look like rcw is cheaper | |
2559 | */ | |
2560 | rcw = 1; rmw = 2; | |
2561 | } else for (i = disks; i--; ) { | |
a4456856 DW |
2562 | /* would I have to read this buffer for read_modify_write */ |
2563 | struct r5dev *dev = &sh->dev[i]; | |
2564 | if ((dev->towrite || i == sh->pd_idx) && | |
2565 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2566 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2567 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2568 | if (test_bit(R5_Insync, &dev->flags)) |
2569 | rmw++; | |
2570 | else | |
2571 | rmw += 2*disks; /* cannot read it */ | |
2572 | } | |
2573 | /* Would I have to read this buffer for reconstruct_write */ | |
2574 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2575 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2576 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2577 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2578 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2579 | else |
2580 | rcw += 2*disks; | |
2581 | } | |
2582 | } | |
45b4233c | 2583 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2584 | (unsigned long long)sh->sector, rmw, rcw); |
2585 | set_bit(STRIPE_HANDLE, &sh->state); | |
2586 | if (rmw < rcw && rmw > 0) | |
2587 | /* prefer read-modify-write, but need to get some data */ | |
2588 | for (i = disks; i--; ) { | |
2589 | struct r5dev *dev = &sh->dev[i]; | |
2590 | if ((dev->towrite || i == sh->pd_idx) && | |
2591 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2592 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2593 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2594 | test_bit(R5_Insync, &dev->flags)) { |
2595 | if ( | |
2596 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2597 | pr_debug("Read_old block " |
a4456856 DW |
2598 | "%d for r-m-w\n", i); |
2599 | set_bit(R5_LOCKED, &dev->flags); | |
2600 | set_bit(R5_Wantread, &dev->flags); | |
2601 | s->locked++; | |
2602 | } else { | |
2603 | set_bit(STRIPE_DELAYED, &sh->state); | |
2604 | set_bit(STRIPE_HANDLE, &sh->state); | |
2605 | } | |
2606 | } | |
2607 | } | |
c8ac1803 | 2608 | if (rcw <= rmw && rcw > 0) { |
a4456856 | 2609 | /* want reconstruct write, but need to get some data */ |
c8ac1803 | 2610 | rcw = 0; |
a4456856 DW |
2611 | for (i = disks; i--; ) { |
2612 | struct r5dev *dev = &sh->dev[i]; | |
2613 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 2614 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 2615 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 2616 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
2617 | test_bit(R5_Wantcompute, &dev->flags))) { |
2618 | rcw++; | |
2619 | if (!test_bit(R5_Insync, &dev->flags)) | |
2620 | continue; /* it's a failed drive */ | |
a4456856 DW |
2621 | if ( |
2622 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2623 | pr_debug("Read_old block " |
a4456856 DW |
2624 | "%d for Reconstruct\n", i); |
2625 | set_bit(R5_LOCKED, &dev->flags); | |
2626 | set_bit(R5_Wantread, &dev->flags); | |
2627 | s->locked++; | |
2628 | } else { | |
2629 | set_bit(STRIPE_DELAYED, &sh->state); | |
2630 | set_bit(STRIPE_HANDLE, &sh->state); | |
2631 | } | |
2632 | } | |
2633 | } | |
c8ac1803 | 2634 | } |
a4456856 DW |
2635 | /* now if nothing is locked, and if we have enough data, |
2636 | * we can start a write request | |
2637 | */ | |
f38e1219 DW |
2638 | /* since handle_stripe can be called at any time we need to handle the |
2639 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2640 | * subsequent call wants to start a write request. raid_run_ops only |
2641 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2642 | * simultaneously. If this is not the case then new writes need to be |
2643 | * held off until the compute completes. | |
2644 | */ | |
976ea8d4 DW |
2645 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2646 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2647 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2648 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2649 | } |
2650 | ||
d1688a6d | 2651 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2652 | struct stripe_head_state *s, int disks) |
2653 | { | |
ecc65c9b | 2654 | struct r5dev *dev = NULL; |
bd2ab670 | 2655 | |
a4456856 | 2656 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2657 | |
ecc65c9b DW |
2658 | switch (sh->check_state) { |
2659 | case check_state_idle: | |
2660 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2661 | if (s->failed == 0) { |
bd2ab670 | 2662 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2663 | sh->check_state = check_state_run; |
2664 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2665 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2666 | s->uptodate--; |
ecc65c9b | 2667 | break; |
bd2ab670 | 2668 | } |
f2b3b44d | 2669 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
2670 | /* fall through */ |
2671 | case check_state_compute_result: | |
2672 | sh->check_state = check_state_idle; | |
2673 | if (!dev) | |
2674 | dev = &sh->dev[sh->pd_idx]; | |
2675 | ||
2676 | /* check that a write has not made the stripe insync */ | |
2677 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2678 | break; | |
c8894419 | 2679 | |
a4456856 | 2680 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2681 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2682 | BUG_ON(s->uptodate != disks); | |
2683 | ||
2684 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2685 | s->locked++; |
a4456856 | 2686 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2687 | |
a4456856 | 2688 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2689 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2690 | break; |
2691 | case check_state_run: | |
2692 | break; /* we will be called again upon completion */ | |
2693 | case check_state_check_result: | |
2694 | sh->check_state = check_state_idle; | |
2695 | ||
2696 | /* if a failure occurred during the check operation, leave | |
2697 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2698 | */ | |
2699 | if (s->failed) | |
2700 | break; | |
2701 | ||
2702 | /* handle a successful check operation, if parity is correct | |
2703 | * we are done. Otherwise update the mismatch count and repair | |
2704 | * parity if !MD_RECOVERY_CHECK | |
2705 | */ | |
ad283ea4 | 2706 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2707 | /* parity is correct (on disc, |
2708 | * not in buffer any more) | |
2709 | */ | |
2710 | set_bit(STRIPE_INSYNC, &sh->state); | |
2711 | else { | |
2712 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2713 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2714 | /* don't try to repair!! */ | |
2715 | set_bit(STRIPE_INSYNC, &sh->state); | |
2716 | else { | |
2717 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2718 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2719 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2720 | set_bit(R5_Wantcompute, | |
2721 | &sh->dev[sh->pd_idx].flags); | |
2722 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2723 | sh->ops.target2 = -1; |
ecc65c9b DW |
2724 | s->uptodate++; |
2725 | } | |
2726 | } | |
2727 | break; | |
2728 | case check_state_compute_run: | |
2729 | break; | |
2730 | default: | |
2731 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2732 | __func__, sh->check_state, | |
2733 | (unsigned long long) sh->sector); | |
2734 | BUG(); | |
a4456856 DW |
2735 | } |
2736 | } | |
2737 | ||
2738 | ||
d1688a6d | 2739 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 2740 | struct stripe_head_state *s, |
f2b3b44d | 2741 | int disks) |
a4456856 | 2742 | { |
a4456856 | 2743 | int pd_idx = sh->pd_idx; |
34e04e87 | 2744 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2745 | struct r5dev *dev; |
a4456856 DW |
2746 | |
2747 | set_bit(STRIPE_HANDLE, &sh->state); | |
2748 | ||
2749 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2750 | |
a4456856 DW |
2751 | /* Want to check and possibly repair P and Q. |
2752 | * However there could be one 'failed' device, in which | |
2753 | * case we can only check one of them, possibly using the | |
2754 | * other to generate missing data | |
2755 | */ | |
2756 | ||
d82dfee0 DW |
2757 | switch (sh->check_state) { |
2758 | case check_state_idle: | |
2759 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 2760 | if (s->failed == s->q_failed) { |
d82dfee0 | 2761 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2762 | * makes sense to check P (If anything else were failed, |
2763 | * we would have used P to recreate it). | |
2764 | */ | |
d82dfee0 | 2765 | sh->check_state = check_state_run; |
a4456856 | 2766 | } |
f2b3b44d | 2767 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 2768 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2769 | * anything, so it makes sense to check it |
2770 | */ | |
d82dfee0 DW |
2771 | if (sh->check_state == check_state_run) |
2772 | sh->check_state = check_state_run_pq; | |
2773 | else | |
2774 | sh->check_state = check_state_run_q; | |
a4456856 | 2775 | } |
a4456856 | 2776 | |
d82dfee0 DW |
2777 | /* discard potentially stale zero_sum_result */ |
2778 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2779 | |
d82dfee0 DW |
2780 | if (sh->check_state == check_state_run) { |
2781 | /* async_xor_zero_sum destroys the contents of P */ | |
2782 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2783 | s->uptodate--; | |
a4456856 | 2784 | } |
d82dfee0 DW |
2785 | if (sh->check_state >= check_state_run && |
2786 | sh->check_state <= check_state_run_pq) { | |
2787 | /* async_syndrome_zero_sum preserves P and Q, so | |
2788 | * no need to mark them !uptodate here | |
2789 | */ | |
2790 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2791 | break; | |
a4456856 DW |
2792 | } |
2793 | ||
d82dfee0 DW |
2794 | /* we have 2-disk failure */ |
2795 | BUG_ON(s->failed != 2); | |
2796 | /* fall through */ | |
2797 | case check_state_compute_result: | |
2798 | sh->check_state = check_state_idle; | |
a4456856 | 2799 | |
d82dfee0 DW |
2800 | /* check that a write has not made the stripe insync */ |
2801 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2802 | break; | |
a4456856 DW |
2803 | |
2804 | /* now write out any block on a failed drive, | |
d82dfee0 | 2805 | * or P or Q if they were recomputed |
a4456856 | 2806 | */ |
d82dfee0 | 2807 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 2808 | if (s->failed == 2) { |
f2b3b44d | 2809 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
2810 | s->locked++; |
2811 | set_bit(R5_LOCKED, &dev->flags); | |
2812 | set_bit(R5_Wantwrite, &dev->flags); | |
2813 | } | |
2814 | if (s->failed >= 1) { | |
f2b3b44d | 2815 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
2816 | s->locked++; |
2817 | set_bit(R5_LOCKED, &dev->flags); | |
2818 | set_bit(R5_Wantwrite, &dev->flags); | |
2819 | } | |
d82dfee0 | 2820 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2821 | dev = &sh->dev[pd_idx]; |
2822 | s->locked++; | |
2823 | set_bit(R5_LOCKED, &dev->flags); | |
2824 | set_bit(R5_Wantwrite, &dev->flags); | |
2825 | } | |
d82dfee0 | 2826 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2827 | dev = &sh->dev[qd_idx]; |
2828 | s->locked++; | |
2829 | set_bit(R5_LOCKED, &dev->flags); | |
2830 | set_bit(R5_Wantwrite, &dev->flags); | |
2831 | } | |
2832 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2833 | ||
2834 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2835 | break; |
2836 | case check_state_run: | |
2837 | case check_state_run_q: | |
2838 | case check_state_run_pq: | |
2839 | break; /* we will be called again upon completion */ | |
2840 | case check_state_check_result: | |
2841 | sh->check_state = check_state_idle; | |
2842 | ||
2843 | /* handle a successful check operation, if parity is correct | |
2844 | * we are done. Otherwise update the mismatch count and repair | |
2845 | * parity if !MD_RECOVERY_CHECK | |
2846 | */ | |
2847 | if (sh->ops.zero_sum_result == 0) { | |
2848 | /* both parities are correct */ | |
2849 | if (!s->failed) | |
2850 | set_bit(STRIPE_INSYNC, &sh->state); | |
2851 | else { | |
2852 | /* in contrast to the raid5 case we can validate | |
2853 | * parity, but still have a failure to write | |
2854 | * back | |
2855 | */ | |
2856 | sh->check_state = check_state_compute_result; | |
2857 | /* Returning at this point means that we may go | |
2858 | * off and bring p and/or q uptodate again so | |
2859 | * we make sure to check zero_sum_result again | |
2860 | * to verify if p or q need writeback | |
2861 | */ | |
2862 | } | |
2863 | } else { | |
2864 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2865 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2866 | /* don't try to repair!! */ | |
2867 | set_bit(STRIPE_INSYNC, &sh->state); | |
2868 | else { | |
2869 | int *target = &sh->ops.target; | |
2870 | ||
2871 | sh->ops.target = -1; | |
2872 | sh->ops.target2 = -1; | |
2873 | sh->check_state = check_state_compute_run; | |
2874 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2875 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2876 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2877 | set_bit(R5_Wantcompute, | |
2878 | &sh->dev[pd_idx].flags); | |
2879 | *target = pd_idx; | |
2880 | target = &sh->ops.target2; | |
2881 | s->uptodate++; | |
2882 | } | |
2883 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2884 | set_bit(R5_Wantcompute, | |
2885 | &sh->dev[qd_idx].flags); | |
2886 | *target = qd_idx; | |
2887 | s->uptodate++; | |
2888 | } | |
2889 | } | |
2890 | } | |
2891 | break; | |
2892 | case check_state_compute_run: | |
2893 | break; | |
2894 | default: | |
2895 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2896 | __func__, sh->check_state, | |
2897 | (unsigned long long) sh->sector); | |
2898 | BUG(); | |
a4456856 DW |
2899 | } |
2900 | } | |
2901 | ||
d1688a6d | 2902 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
2903 | { |
2904 | int i; | |
2905 | ||
2906 | /* We have read all the blocks in this stripe and now we need to | |
2907 | * copy some of them into a target stripe for expand. | |
2908 | */ | |
f0a50d37 | 2909 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2910 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2911 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 2912 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 2913 | int dd_idx, j; |
a4456856 | 2914 | struct stripe_head *sh2; |
a08abd8c | 2915 | struct async_submit_ctl submit; |
a4456856 | 2916 | |
784052ec | 2917 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
2918 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2919 | &dd_idx, NULL); | |
a8c906ca | 2920 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
2921 | if (sh2 == NULL) |
2922 | /* so far only the early blocks of this stripe | |
2923 | * have been requested. When later blocks | |
2924 | * get requested, we will try again | |
2925 | */ | |
2926 | continue; | |
2927 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2928 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2929 | /* must have already done this block */ | |
2930 | release_stripe(sh2); | |
2931 | continue; | |
2932 | } | |
f0a50d37 DW |
2933 | |
2934 | /* place all the copies on one channel */ | |
a08abd8c | 2935 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 2936 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 2937 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 2938 | &submit); |
f0a50d37 | 2939 | |
a4456856 DW |
2940 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2941 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2942 | for (j = 0; j < conf->raid_disks; j++) | |
2943 | if (j != sh2->pd_idx && | |
86c374ba | 2944 | j != sh2->qd_idx && |
a4456856 DW |
2945 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2946 | break; | |
2947 | if (j == conf->raid_disks) { | |
2948 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2949 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2950 | } | |
2951 | release_stripe(sh2); | |
f0a50d37 | 2952 | |
a4456856 | 2953 | } |
a2e08551 N |
2954 | /* done submitting copies, wait for them to complete */ |
2955 | if (tx) { | |
2956 | async_tx_ack(tx); | |
2957 | dma_wait_for_async_tx(tx); | |
2958 | } | |
a4456856 | 2959 | } |
1da177e4 | 2960 | |
6bfe0b49 | 2961 | |
1da177e4 LT |
2962 | /* |
2963 | * handle_stripe - do things to a stripe. | |
2964 | * | |
2965 | * We lock the stripe and then examine the state of various bits | |
2966 | * to see what needs to be done. | |
2967 | * Possible results: | |
2968 | * return some read request which now have data | |
2969 | * return some write requests which are safely on disc | |
2970 | * schedule a read on some buffers | |
2971 | * schedule a write of some buffers | |
2972 | * return confirmation of parity correctness | |
2973 | * | |
1da177e4 LT |
2974 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2975 | * get BH_Lock set before the stripe lock is released. | |
2976 | * | |
2977 | */ | |
a4456856 | 2978 | |
acfe726b | 2979 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 2980 | { |
d1688a6d | 2981 | struct r5conf *conf = sh->raid_conf; |
f416885e | 2982 | int disks = sh->disks; |
474af965 N |
2983 | struct r5dev *dev; |
2984 | int i; | |
1da177e4 | 2985 | |
acfe726b N |
2986 | memset(s, 0, sizeof(*s)); |
2987 | ||
2988 | s->syncing = test_bit(STRIPE_SYNCING, &sh->state); | |
2989 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2990 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
2991 | s->failed_num[0] = -1; | |
2992 | s->failed_num[1] = -1; | |
1da177e4 | 2993 | |
acfe726b | 2994 | /* Now to look around and see what can be done */ |
1da177e4 | 2995 | rcu_read_lock(); |
c4c1663b | 2996 | spin_lock_irq(&conf->device_lock); |
16a53ecc | 2997 | for (i=disks; i--; ) { |
3cb03002 | 2998 | struct md_rdev *rdev; |
31c176ec N |
2999 | sector_t first_bad; |
3000 | int bad_sectors; | |
3001 | int is_bad = 0; | |
acfe726b | 3002 | |
16a53ecc | 3003 | dev = &sh->dev[i]; |
1da177e4 | 3004 | |
45b4233c | 3005 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc | 3006 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
6c0069c0 YT |
3007 | /* maybe we can reply to a read |
3008 | * | |
3009 | * new wantfill requests are only permitted while | |
3010 | * ops_complete_biofill is guaranteed to be inactive | |
3011 | */ | |
3012 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3013 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3014 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3015 | |
16a53ecc | 3016 | /* now count some things */ |
cc94015a N |
3017 | if (test_bit(R5_LOCKED, &dev->flags)) |
3018 | s->locked++; | |
3019 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3020 | s->uptodate++; | |
2d6e4ecc | 3021 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
3022 | s->compute++; |
3023 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 3024 | } |
1da177e4 | 3025 | |
acfe726b | 3026 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 3027 | s->to_fill++; |
acfe726b | 3028 | else if (dev->toread) |
cc94015a | 3029 | s->to_read++; |
16a53ecc | 3030 | if (dev->towrite) { |
cc94015a | 3031 | s->to_write++; |
16a53ecc | 3032 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3033 | s->non_overwrite++; |
16a53ecc | 3034 | } |
a4456856 | 3035 | if (dev->written) |
cc94015a | 3036 | s->written++; |
16a53ecc | 3037 | rdev = rcu_dereference(conf->disks[i].rdev); |
31c176ec N |
3038 | if (rdev) { |
3039 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3040 | &first_bad, &bad_sectors); | |
3041 | if (s->blocked_rdev == NULL | |
3042 | && (test_bit(Blocked, &rdev->flags) | |
3043 | || is_bad < 0)) { | |
3044 | if (is_bad < 0) | |
3045 | set_bit(BlockedBadBlocks, | |
3046 | &rdev->flags); | |
3047 | s->blocked_rdev = rdev; | |
3048 | atomic_inc(&rdev->nr_pending); | |
3049 | } | |
6bfe0b49 | 3050 | } |
415e72d0 N |
3051 | clear_bit(R5_Insync, &dev->flags); |
3052 | if (!rdev) | |
3053 | /* Not in-sync */; | |
31c176ec N |
3054 | else if (is_bad) { |
3055 | /* also not in-sync */ | |
3056 | if (!test_bit(WriteErrorSeen, &rdev->flags)) { | |
3057 | /* treat as in-sync, but with a read error | |
3058 | * which we can now try to correct | |
3059 | */ | |
3060 | set_bit(R5_Insync, &dev->flags); | |
3061 | set_bit(R5_ReadError, &dev->flags); | |
3062 | } | |
3063 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 3064 | set_bit(R5_Insync, &dev->flags); |
355840e7 | 3065 | else if (!test_bit(Faulty, &rdev->flags)) { |
415e72d0 N |
3066 | /* in sync if before recovery_offset */ |
3067 | if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) | |
3068 | set_bit(R5_Insync, &dev->flags); | |
3069 | } | |
bc2607f3 N |
3070 | if (test_bit(R5_WriteError, &dev->flags)) { |
3071 | clear_bit(R5_Insync, &dev->flags); | |
3072 | if (!test_bit(Faulty, &rdev->flags)) { | |
3073 | s->handle_bad_blocks = 1; | |
3074 | atomic_inc(&rdev->nr_pending); | |
3075 | } else | |
3076 | clear_bit(R5_WriteError, &dev->flags); | |
3077 | } | |
b84db560 N |
3078 | if (test_bit(R5_MadeGood, &dev->flags)) { |
3079 | if (!test_bit(Faulty, &rdev->flags)) { | |
3080 | s->handle_bad_blocks = 1; | |
3081 | atomic_inc(&rdev->nr_pending); | |
3082 | } else | |
3083 | clear_bit(R5_MadeGood, &dev->flags); | |
3084 | } | |
415e72d0 | 3085 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
3086 | /* The ReadError flag will just be confusing now */ |
3087 | clear_bit(R5_ReadError, &dev->flags); | |
3088 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3089 | } |
415e72d0 N |
3090 | if (test_bit(R5_ReadError, &dev->flags)) |
3091 | clear_bit(R5_Insync, &dev->flags); | |
3092 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3093 | if (s->failed < 2) |
3094 | s->failed_num[s->failed] = i; | |
3095 | s->failed++; | |
415e72d0 | 3096 | } |
1da177e4 | 3097 | } |
c4c1663b | 3098 | spin_unlock_irq(&conf->device_lock); |
1da177e4 | 3099 | rcu_read_unlock(); |
cc94015a N |
3100 | } |
3101 | ||
3102 | static void handle_stripe(struct stripe_head *sh) | |
3103 | { | |
3104 | struct stripe_head_state s; | |
d1688a6d | 3105 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 3106 | int i; |
84789554 N |
3107 | int prexor; |
3108 | int disks = sh->disks; | |
474af965 | 3109 | struct r5dev *pdev, *qdev; |
cc94015a N |
3110 | |
3111 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3112 | if (test_and_set_bit(STRIPE_ACTIVE, &sh->state)) { | |
3113 | /* already being handled, ensure it gets handled | |
3114 | * again when current action finishes */ | |
3115 | set_bit(STRIPE_HANDLE, &sh->state); | |
3116 | return; | |
3117 | } | |
3118 | ||
3119 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3120 | set_bit(STRIPE_SYNCING, &sh->state); | |
3121 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3122 | } | |
3123 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3124 | ||
3125 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3126 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3127 | (unsigned long long)sh->sector, sh->state, | |
3128 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3129 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 3130 | |
acfe726b | 3131 | analyse_stripe(sh, &s); |
c5a31000 | 3132 | |
bc2607f3 N |
3133 | if (s.handle_bad_blocks) { |
3134 | set_bit(STRIPE_HANDLE, &sh->state); | |
3135 | goto finish; | |
3136 | } | |
3137 | ||
474af965 N |
3138 | if (unlikely(s.blocked_rdev)) { |
3139 | if (s.syncing || s.expanding || s.expanded || | |
3140 | s.to_write || s.written) { | |
3141 | set_bit(STRIPE_HANDLE, &sh->state); | |
3142 | goto finish; | |
3143 | } | |
3144 | /* There is nothing for the blocked_rdev to block */ | |
3145 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
3146 | s.blocked_rdev = NULL; | |
3147 | } | |
3148 | ||
3149 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
3150 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3151 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3152 | } | |
3153 | ||
3154 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
3155 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
3156 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
3157 | s.failed_num[0], s.failed_num[1]); | |
3158 | /* check if the array has lost more than max_degraded devices and, | |
3159 | * if so, some requests might need to be failed. | |
3160 | */ | |
3161 | if (s.failed > conf->max_degraded && s.to_read+s.to_write+s.written) | |
3162 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
7f0da59b N |
3163 | if (s.failed > conf->max_degraded && s.syncing) |
3164 | handle_failed_sync(conf, sh, &s); | |
474af965 N |
3165 | |
3166 | /* | |
3167 | * might be able to return some write requests if the parity blocks | |
3168 | * are safe, or on a failed drive | |
3169 | */ | |
3170 | pdev = &sh->dev[sh->pd_idx]; | |
3171 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
3172 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
3173 | qdev = &sh->dev[sh->qd_idx]; | |
3174 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
3175 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
3176 | || conf->level < 6; | |
3177 | ||
3178 | if (s.written && | |
3179 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
3180 | && !test_bit(R5_LOCKED, &pdev->flags) | |
3181 | && test_bit(R5_UPTODATE, &pdev->flags)))) && | |
3182 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
3183 | && !test_bit(R5_LOCKED, &qdev->flags) | |
3184 | && test_bit(R5_UPTODATE, &qdev->flags))))) | |
3185 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
3186 | ||
3187 | /* Now we might consider reading some blocks, either to check/generate | |
3188 | * parity, or to satisfy requests | |
3189 | * or to load a block that is being partially written. | |
3190 | */ | |
3191 | if (s.to_read || s.non_overwrite | |
3192 | || (conf->level == 6 && s.to_write && s.failed) | |
3193 | || (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) | |
3194 | handle_stripe_fill(sh, &s, disks); | |
3195 | ||
84789554 N |
3196 | /* Now we check to see if any write operations have recently |
3197 | * completed | |
3198 | */ | |
3199 | prexor = 0; | |
3200 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
3201 | prexor = 1; | |
3202 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
3203 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
3204 | sh->reconstruct_state = reconstruct_state_idle; | |
3205 | ||
3206 | /* All the 'written' buffers and the parity block are ready to | |
3207 | * be written back to disk | |
3208 | */ | |
3209 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3210 | BUG_ON(sh->qd_idx >= 0 && | |
3211 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags)); | |
3212 | for (i = disks; i--; ) { | |
3213 | struct r5dev *dev = &sh->dev[i]; | |
3214 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3215 | (i == sh->pd_idx || i == sh->qd_idx || | |
3216 | dev->written)) { | |
3217 | pr_debug("Writing block %d\n", i); | |
3218 | set_bit(R5_Wantwrite, &dev->flags); | |
3219 | if (prexor) | |
3220 | continue; | |
3221 | if (!test_bit(R5_Insync, &dev->flags) || | |
3222 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
3223 | s.failed == 0)) | |
3224 | set_bit(STRIPE_INSYNC, &sh->state); | |
3225 | } | |
3226 | } | |
3227 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3228 | s.dec_preread_active = 1; | |
3229 | } | |
3230 | ||
3231 | /* Now to consider new write requests and what else, if anything | |
3232 | * should be read. We do not handle new writes when: | |
3233 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3234 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3235 | * block. | |
3236 | */ | |
3237 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
3238 | handle_stripe_dirtying(conf, sh, &s, disks); | |
3239 | ||
3240 | /* maybe we need to check and possibly fix the parity for this stripe | |
3241 | * Any reads will already have been scheduled, so we just see if enough | |
3242 | * data is available. The parity check is held off while parity | |
3243 | * dependent operations are in flight. | |
3244 | */ | |
3245 | if (sh->check_state || | |
3246 | (s.syncing && s.locked == 0 && | |
3247 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3248 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
3249 | if (conf->level == 6) | |
3250 | handle_parity_checks6(conf, sh, &s, disks); | |
3251 | else | |
3252 | handle_parity_checks5(conf, sh, &s, disks); | |
3253 | } | |
c5a31000 N |
3254 | |
3255 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { | |
3256 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3257 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3258 | } | |
3259 | ||
3260 | /* If the failed drives are just a ReadError, then we might need | |
3261 | * to progress the repair/check process | |
3262 | */ | |
3263 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
3264 | for (i = 0; i < s.failed; i++) { | |
3265 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
3266 | if (test_bit(R5_ReadError, &dev->flags) | |
3267 | && !test_bit(R5_LOCKED, &dev->flags) | |
3268 | && test_bit(R5_UPTODATE, &dev->flags) | |
3269 | ) { | |
3270 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3271 | set_bit(R5_Wantwrite, &dev->flags); | |
3272 | set_bit(R5_ReWrite, &dev->flags); | |
3273 | set_bit(R5_LOCKED, &dev->flags); | |
3274 | s.locked++; | |
3275 | } else { | |
3276 | /* let's read it back */ | |
3277 | set_bit(R5_Wantread, &dev->flags); | |
3278 | set_bit(R5_LOCKED, &dev->flags); | |
3279 | s.locked++; | |
3280 | } | |
3281 | } | |
3282 | } | |
3283 | ||
3284 | ||
3687c061 N |
3285 | /* Finish reconstruct operations initiated by the expansion process */ |
3286 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3287 | struct stripe_head *sh_src | |
3288 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
3289 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
3290 | /* sh cannot be written until sh_src has been read. | |
3291 | * so arrange for sh to be delayed a little | |
3292 | */ | |
3293 | set_bit(STRIPE_DELAYED, &sh->state); | |
3294 | set_bit(STRIPE_HANDLE, &sh->state); | |
3295 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3296 | &sh_src->state)) | |
3297 | atomic_inc(&conf->preread_active_stripes); | |
3298 | release_stripe(sh_src); | |
3299 | goto finish; | |
3300 | } | |
3301 | if (sh_src) | |
3302 | release_stripe(sh_src); | |
3303 | ||
3304 | sh->reconstruct_state = reconstruct_state_idle; | |
3305 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3306 | for (i = conf->raid_disks; i--; ) { | |
3307 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3308 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3309 | s.locked++; | |
3310 | } | |
3311 | } | |
f416885e | 3312 | |
3687c061 N |
3313 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
3314 | !sh->reconstruct_state) { | |
3315 | /* Need to write out all blocks after computing parity */ | |
3316 | sh->disks = conf->raid_disks; | |
3317 | stripe_set_idx(sh->sector, conf, 0, sh); | |
3318 | schedule_reconstruction(sh, &s, 1, 1); | |
3319 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
3320 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
3321 | atomic_dec(&conf->reshape_stripes); | |
3322 | wake_up(&conf->wait_for_overlap); | |
3323 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3324 | } | |
3325 | ||
3326 | if (s.expanding && s.locked == 0 && | |
3327 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
3328 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 3329 | |
3687c061 | 3330 | finish: |
6bfe0b49 | 3331 | /* wait for this device to become unblocked */ |
43220aa0 | 3332 | if (conf->mddev->external && unlikely(s.blocked_rdev)) |
c5709ef6 | 3333 | md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev); |
6bfe0b49 | 3334 | |
bc2607f3 N |
3335 | if (s.handle_bad_blocks) |
3336 | for (i = disks; i--; ) { | |
3cb03002 | 3337 | struct md_rdev *rdev; |
bc2607f3 N |
3338 | struct r5dev *dev = &sh->dev[i]; |
3339 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
3340 | /* We own a safe reference to the rdev */ | |
3341 | rdev = conf->disks[i].rdev; | |
3342 | if (!rdev_set_badblocks(rdev, sh->sector, | |
3343 | STRIPE_SECTORS, 0)) | |
3344 | md_error(conf->mddev, rdev); | |
3345 | rdev_dec_pending(rdev, conf->mddev); | |
3346 | } | |
b84db560 N |
3347 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
3348 | rdev = conf->disks[i].rdev; | |
3349 | rdev_clear_badblocks(rdev, sh->sector, | |
3350 | STRIPE_SECTORS); | |
3351 | rdev_dec_pending(rdev, conf->mddev); | |
3352 | } | |
bc2607f3 N |
3353 | } |
3354 | ||
6c0069c0 YT |
3355 | if (s.ops_request) |
3356 | raid_run_ops(sh, s.ops_request); | |
3357 | ||
f0e43bcd | 3358 | ops_run_io(sh, &s); |
16a53ecc | 3359 | |
c5709ef6 | 3360 | if (s.dec_preread_active) { |
729a1866 | 3361 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 3362 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3363 | * have actually been submitted. |
3364 | */ | |
3365 | atomic_dec(&conf->preread_active_stripes); | |
3366 | if (atomic_read(&conf->preread_active_stripes) < | |
3367 | IO_THRESHOLD) | |
3368 | md_wakeup_thread(conf->mddev->thread); | |
3369 | } | |
3370 | ||
c5709ef6 | 3371 | return_io(s.return_bi); |
16a53ecc | 3372 | |
c4c1663b | 3373 | clear_bit(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
3374 | } |
3375 | ||
d1688a6d | 3376 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
3377 | { |
3378 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3379 | while (!list_empty(&conf->delayed_list)) { | |
3380 | struct list_head *l = conf->delayed_list.next; | |
3381 | struct stripe_head *sh; | |
3382 | sh = list_entry(l, struct stripe_head, lru); | |
3383 | list_del_init(l); | |
3384 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3385 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3386 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3387 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3388 | } |
482c0834 | 3389 | } |
16a53ecc N |
3390 | } |
3391 | ||
d1688a6d | 3392 | static void activate_bit_delay(struct r5conf *conf) |
16a53ecc N |
3393 | { |
3394 | /* device_lock is held */ | |
3395 | struct list_head head; | |
3396 | list_add(&head, &conf->bitmap_list); | |
3397 | list_del_init(&conf->bitmap_list); | |
3398 | while (!list_empty(&head)) { | |
3399 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3400 | list_del_init(&sh->lru); | |
3401 | atomic_inc(&sh->count); | |
3402 | __release_stripe(conf, sh); | |
3403 | } | |
3404 | } | |
3405 | ||
fd01b88c | 3406 | int md_raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 3407 | { |
d1688a6d | 3408 | struct r5conf *conf = mddev->private; |
f022b2fd N |
3409 | |
3410 | /* No difference between reads and writes. Just check | |
3411 | * how busy the stripe_cache is | |
3412 | */ | |
3fa841d7 | 3413 | |
f022b2fd N |
3414 | if (conf->inactive_blocked) |
3415 | return 1; | |
3416 | if (conf->quiesce) | |
3417 | return 1; | |
3418 | if (list_empty_careful(&conf->inactive_list)) | |
3419 | return 1; | |
3420 | ||
3421 | return 0; | |
3422 | } | |
11d8a6e3 N |
3423 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3424 | ||
3425 | static int raid5_congested(void *data, int bits) | |
3426 | { | |
fd01b88c | 3427 | struct mddev *mddev = data; |
11d8a6e3 N |
3428 | |
3429 | return mddev_congested(mddev, bits) || | |
3430 | md_raid5_congested(mddev, bits); | |
3431 | } | |
f022b2fd | 3432 | |
23032a0e RBJ |
3433 | /* We want read requests to align with chunks where possible, |
3434 | * but write requests don't need to. | |
3435 | */ | |
cc371e66 AK |
3436 | static int raid5_mergeable_bvec(struct request_queue *q, |
3437 | struct bvec_merge_data *bvm, | |
3438 | struct bio_vec *biovec) | |
23032a0e | 3439 | { |
fd01b88c | 3440 | struct mddev *mddev = q->queuedata; |
cc371e66 | 3441 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3442 | int max; |
9d8f0363 | 3443 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3444 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3445 | |
cc371e66 | 3446 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3447 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3448 | ||
664e7c41 AN |
3449 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3450 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3451 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3452 | if (max < 0) max = 0; | |
3453 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3454 | return biovec->bv_len; | |
3455 | else | |
3456 | return max; | |
3457 | } | |
3458 | ||
f679623f | 3459 | |
fd01b88c | 3460 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f RBJ |
3461 | { |
3462 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3463 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3464 | unsigned int bio_sectors = bio->bi_size >> 9; |
3465 | ||
664e7c41 AN |
3466 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3467 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3468 | return chunk_sectors >= |
3469 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3470 | } | |
3471 | ||
46031f9a RBJ |
3472 | /* |
3473 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3474 | * later sampled by raid5d. | |
3475 | */ | |
d1688a6d | 3476 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
3477 | { |
3478 | unsigned long flags; | |
3479 | ||
3480 | spin_lock_irqsave(&conf->device_lock, flags); | |
3481 | ||
3482 | bi->bi_next = conf->retry_read_aligned_list; | |
3483 | conf->retry_read_aligned_list = bi; | |
3484 | ||
3485 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3486 | md_wakeup_thread(conf->mddev->thread); | |
3487 | } | |
3488 | ||
3489 | ||
d1688a6d | 3490 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
3491 | { |
3492 | struct bio *bi; | |
3493 | ||
3494 | bi = conf->retry_read_aligned; | |
3495 | if (bi) { | |
3496 | conf->retry_read_aligned = NULL; | |
3497 | return bi; | |
3498 | } | |
3499 | bi = conf->retry_read_aligned_list; | |
3500 | if(bi) { | |
387bb173 | 3501 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3502 | bi->bi_next = NULL; |
960e739d JA |
3503 | /* |
3504 | * this sets the active strip count to 1 and the processed | |
3505 | * strip count to zero (upper 8 bits) | |
3506 | */ | |
46031f9a | 3507 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3508 | } |
3509 | ||
3510 | return bi; | |
3511 | } | |
3512 | ||
3513 | ||
f679623f RBJ |
3514 | /* |
3515 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3516 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3517 | * first). | |
3518 | * If the read failed.. | |
3519 | */ | |
6712ecf8 | 3520 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3521 | { |
3522 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 3523 | struct mddev *mddev; |
d1688a6d | 3524 | struct r5conf *conf; |
46031f9a | 3525 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 3526 | struct md_rdev *rdev; |
46031f9a | 3527 | |
f679623f | 3528 | bio_put(bi); |
46031f9a | 3529 | |
46031f9a RBJ |
3530 | rdev = (void*)raid_bi->bi_next; |
3531 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3532 | mddev = rdev->mddev; |
3533 | conf = mddev->private; | |
46031f9a RBJ |
3534 | |
3535 | rdev_dec_pending(rdev, conf->mddev); | |
3536 | ||
3537 | if (!error && uptodate) { | |
6712ecf8 | 3538 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3539 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3540 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3541 | return; |
46031f9a RBJ |
3542 | } |
3543 | ||
3544 | ||
45b4233c | 3545 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3546 | |
3547 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3548 | } |
3549 | ||
387bb173 NB |
3550 | static int bio_fits_rdev(struct bio *bi) |
3551 | { | |
165125e1 | 3552 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3553 | |
ae03bf63 | 3554 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3555 | return 0; |
3556 | blk_recount_segments(q, bi); | |
8a78362c | 3557 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3558 | return 0; |
3559 | ||
3560 | if (q->merge_bvec_fn) | |
3561 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3562 | * just just give up | |
3563 | */ | |
3564 | return 0; | |
3565 | ||
3566 | return 1; | |
3567 | } | |
3568 | ||
3569 | ||
fd01b88c | 3570 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 3571 | { |
d1688a6d | 3572 | struct r5conf *conf = mddev->private; |
8553fe7e | 3573 | int dd_idx; |
f679623f | 3574 | struct bio* align_bi; |
3cb03002 | 3575 | struct md_rdev *rdev; |
f679623f RBJ |
3576 | |
3577 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3578 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3579 | return 0; |
3580 | } | |
3581 | /* | |
a167f663 | 3582 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3583 | */ |
a167f663 | 3584 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3585 | if (!align_bi) |
3586 | return 0; | |
3587 | /* | |
3588 | * set bi_end_io to a new function, and set bi_private to the | |
3589 | * original bio. | |
3590 | */ | |
3591 | align_bi->bi_end_io = raid5_align_endio; | |
3592 | align_bi->bi_private = raid_bio; | |
3593 | /* | |
3594 | * compute position | |
3595 | */ | |
112bf897 N |
3596 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3597 | 0, | |
911d4ee8 | 3598 | &dd_idx, NULL); |
f679623f RBJ |
3599 | |
3600 | rcu_read_lock(); | |
3601 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3602 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
31c176ec N |
3603 | sector_t first_bad; |
3604 | int bad_sectors; | |
3605 | ||
f679623f RBJ |
3606 | atomic_inc(&rdev->nr_pending); |
3607 | rcu_read_unlock(); | |
46031f9a RBJ |
3608 | raid_bio->bi_next = (void*)rdev; |
3609 | align_bi->bi_bdev = rdev->bdev; | |
3610 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3611 | align_bi->bi_sector += rdev->data_offset; | |
3612 | ||
31c176ec N |
3613 | if (!bio_fits_rdev(align_bi) || |
3614 | is_badblock(rdev, align_bi->bi_sector, align_bi->bi_size>>9, | |
3615 | &first_bad, &bad_sectors)) { | |
3616 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
3617 | bio_put(align_bi); |
3618 | rdev_dec_pending(rdev, mddev); | |
3619 | return 0; | |
3620 | } | |
3621 | ||
46031f9a RBJ |
3622 | spin_lock_irq(&conf->device_lock); |
3623 | wait_event_lock_irq(conf->wait_for_stripe, | |
3624 | conf->quiesce == 0, | |
3625 | conf->device_lock, /* nothing */); | |
3626 | atomic_inc(&conf->active_aligned_reads); | |
3627 | spin_unlock_irq(&conf->device_lock); | |
3628 | ||
f679623f RBJ |
3629 | generic_make_request(align_bi); |
3630 | return 1; | |
3631 | } else { | |
3632 | rcu_read_unlock(); | |
46031f9a | 3633 | bio_put(align_bi); |
f679623f RBJ |
3634 | return 0; |
3635 | } | |
3636 | } | |
3637 | ||
8b3e6cdc DW |
3638 | /* __get_priority_stripe - get the next stripe to process |
3639 | * | |
3640 | * Full stripe writes are allowed to pass preread active stripes up until | |
3641 | * the bypass_threshold is exceeded. In general the bypass_count | |
3642 | * increments when the handle_list is handled before the hold_list; however, it | |
3643 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3644 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3645 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3646 | * handle_list. | |
3647 | */ | |
d1688a6d | 3648 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf) |
8b3e6cdc DW |
3649 | { |
3650 | struct stripe_head *sh; | |
3651 | ||
3652 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3653 | __func__, | |
3654 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3655 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3656 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3657 | ||
3658 | if (!list_empty(&conf->handle_list)) { | |
3659 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3660 | ||
3661 | if (list_empty(&conf->hold_list)) | |
3662 | conf->bypass_count = 0; | |
3663 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3664 | if (conf->hold_list.next == conf->last_hold) | |
3665 | conf->bypass_count++; | |
3666 | else { | |
3667 | conf->last_hold = conf->hold_list.next; | |
3668 | conf->bypass_count -= conf->bypass_threshold; | |
3669 | if (conf->bypass_count < 0) | |
3670 | conf->bypass_count = 0; | |
3671 | } | |
3672 | } | |
3673 | } else if (!list_empty(&conf->hold_list) && | |
3674 | ((conf->bypass_threshold && | |
3675 | conf->bypass_count > conf->bypass_threshold) || | |
3676 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3677 | sh = list_entry(conf->hold_list.next, | |
3678 | typeof(*sh), lru); | |
3679 | conf->bypass_count -= conf->bypass_threshold; | |
3680 | if (conf->bypass_count < 0) | |
3681 | conf->bypass_count = 0; | |
3682 | } else | |
3683 | return NULL; | |
3684 | ||
3685 | list_del_init(&sh->lru); | |
3686 | atomic_inc(&sh->count); | |
3687 | BUG_ON(atomic_read(&sh->count) != 1); | |
3688 | return sh; | |
3689 | } | |
f679623f | 3690 | |
fd01b88c | 3691 | static int make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 3692 | { |
d1688a6d | 3693 | struct r5conf *conf = mddev->private; |
911d4ee8 | 3694 | int dd_idx; |
1da177e4 LT |
3695 | sector_t new_sector; |
3696 | sector_t logical_sector, last_sector; | |
3697 | struct stripe_head *sh; | |
a362357b | 3698 | const int rw = bio_data_dir(bi); |
49077326 | 3699 | int remaining; |
7c13edc8 | 3700 | int plugged; |
1da177e4 | 3701 | |
e9c7469b TH |
3702 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
3703 | md_flush_request(mddev, bi); | |
e5dcdd80 N |
3704 | return 0; |
3705 | } | |
3706 | ||
3d310eb7 | 3707 | md_write_start(mddev, bi); |
06d91a5f | 3708 | |
802ba064 | 3709 | if (rw == READ && |
52488615 | 3710 | mddev->reshape_position == MaxSector && |
21a52c6d | 3711 | chunk_aligned_read(mddev,bi)) |
99c0fb5f | 3712 | return 0; |
52488615 | 3713 | |
1da177e4 LT |
3714 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3715 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3716 | bi->bi_next = NULL; | |
3717 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3718 | |
7c13edc8 | 3719 | plugged = mddev_check_plugged(mddev); |
1da177e4 LT |
3720 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3721 | DEFINE_WAIT(w); | |
16a53ecc | 3722 | int disks, data_disks; |
b5663ba4 | 3723 | int previous; |
b578d55f | 3724 | |
7ecaa1e6 | 3725 | retry: |
b5663ba4 | 3726 | previous = 0; |
b0f9ec04 | 3727 | disks = conf->raid_disks; |
b578d55f | 3728 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3729 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3730 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3731 | * 64bit on a 32bit platform, and so it might be |
3732 | * possible to see a half-updated value | |
aeb878b0 | 3733 | * Of course reshape_progress could change after |
df8e7f76 N |
3734 | * the lock is dropped, so once we get a reference |
3735 | * to the stripe that we think it is, we will have | |
3736 | * to check again. | |
3737 | */ | |
7ecaa1e6 | 3738 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3739 | if (mddev->delta_disks < 0 |
3740 | ? logical_sector < conf->reshape_progress | |
3741 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3742 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3743 | previous = 1; |
3744 | } else { | |
fef9c61f N |
3745 | if (mddev->delta_disks < 0 |
3746 | ? logical_sector < conf->reshape_safe | |
3747 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3748 | spin_unlock_irq(&conf->device_lock); |
3749 | schedule(); | |
3750 | goto retry; | |
3751 | } | |
3752 | } | |
7ecaa1e6 N |
3753 | spin_unlock_irq(&conf->device_lock); |
3754 | } | |
16a53ecc N |
3755 | data_disks = disks - conf->max_degraded; |
3756 | ||
112bf897 N |
3757 | new_sector = raid5_compute_sector(conf, logical_sector, |
3758 | previous, | |
911d4ee8 | 3759 | &dd_idx, NULL); |
0c55e022 | 3760 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3761 | (unsigned long long)new_sector, |
3762 | (unsigned long long)logical_sector); | |
3763 | ||
b5663ba4 | 3764 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3765 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3766 | if (sh) { |
b0f9ec04 | 3767 | if (unlikely(previous)) { |
7ecaa1e6 | 3768 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3769 | * stripe, so we must do the range check again. |
3770 | * Expansion could still move past after this | |
3771 | * test, but as we are holding a reference to | |
3772 | * 'sh', we know that if that happens, | |
3773 | * STRIPE_EXPANDING will get set and the expansion | |
3774 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3775 | */ |
3776 | int must_retry = 0; | |
3777 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
3778 | if (mddev->delta_disks < 0 |
3779 | ? logical_sector >= conf->reshape_progress | |
3780 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
3781 | /* mismatch, need to try again */ |
3782 | must_retry = 1; | |
3783 | spin_unlock_irq(&conf->device_lock); | |
3784 | if (must_retry) { | |
3785 | release_stripe(sh); | |
7a3ab908 | 3786 | schedule(); |
7ecaa1e6 N |
3787 | goto retry; |
3788 | } | |
3789 | } | |
e62e58a5 | 3790 | |
ffd96e35 | 3791 | if (rw == WRITE && |
a5c308d4 | 3792 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
3793 | logical_sector < mddev->suspend_hi) { |
3794 | release_stripe(sh); | |
e62e58a5 N |
3795 | /* As the suspend_* range is controlled by |
3796 | * userspace, we want an interruptible | |
3797 | * wait. | |
3798 | */ | |
3799 | flush_signals(current); | |
3800 | prepare_to_wait(&conf->wait_for_overlap, | |
3801 | &w, TASK_INTERRUPTIBLE); | |
3802 | if (logical_sector >= mddev->suspend_lo && | |
3803 | logical_sector < mddev->suspend_hi) | |
3804 | schedule(); | |
e464eafd N |
3805 | goto retry; |
3806 | } | |
7ecaa1e6 N |
3807 | |
3808 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 3809 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
3810 | /* Stripe is busy expanding or |
3811 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3812 | * and wait a while |
3813 | */ | |
482c0834 | 3814 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
3815 | release_stripe(sh); |
3816 | schedule(); | |
3817 | goto retry; | |
3818 | } | |
3819 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3820 | set_bit(STRIPE_HANDLE, &sh->state); |
3821 | clear_bit(STRIPE_DELAYED, &sh->state); | |
e9c7469b | 3822 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
3823 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3824 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 3825 | release_stripe(sh); |
1da177e4 LT |
3826 | } else { |
3827 | /* cannot get stripe for read-ahead, just give-up */ | |
3828 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3829 | finish_wait(&conf->wait_for_overlap, &w); | |
3830 | break; | |
3831 | } | |
3832 | ||
3833 | } | |
7c13edc8 N |
3834 | if (!plugged) |
3835 | md_wakeup_thread(mddev->thread); | |
3836 | ||
1da177e4 | 3837 | spin_lock_irq(&conf->device_lock); |
960e739d | 3838 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3839 | spin_unlock_irq(&conf->device_lock); |
3840 | if (remaining == 0) { | |
1da177e4 | 3841 | |
16a53ecc | 3842 | if ( rw == WRITE ) |
1da177e4 | 3843 | md_write_end(mddev); |
6712ecf8 | 3844 | |
0e13fe23 | 3845 | bio_endio(bi, 0); |
1da177e4 | 3846 | } |
729a1866 | 3847 | |
1da177e4 LT |
3848 | return 0; |
3849 | } | |
3850 | ||
fd01b88c | 3851 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 3852 | |
fd01b88c | 3853 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3854 | { |
52c03291 N |
3855 | /* reshaping is quite different to recovery/resync so it is |
3856 | * handled quite separately ... here. | |
3857 | * | |
3858 | * On each call to sync_request, we gather one chunk worth of | |
3859 | * destination stripes and flag them as expanding. | |
3860 | * Then we find all the source stripes and request reads. | |
3861 | * As the reads complete, handle_stripe will copy the data | |
3862 | * into the destination stripe and release that stripe. | |
3863 | */ | |
d1688a6d | 3864 | struct r5conf *conf = mddev->private; |
1da177e4 | 3865 | struct stripe_head *sh; |
ccfcc3c1 | 3866 | sector_t first_sector, last_sector; |
f416885e N |
3867 | int raid_disks = conf->previous_raid_disks; |
3868 | int data_disks = raid_disks - conf->max_degraded; | |
3869 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3870 | int i; |
3871 | int dd_idx; | |
c8f517c4 | 3872 | sector_t writepos, readpos, safepos; |
ec32a2bd | 3873 | sector_t stripe_addr; |
7a661381 | 3874 | int reshape_sectors; |
ab69ae12 | 3875 | struct list_head stripes; |
52c03291 | 3876 | |
fef9c61f N |
3877 | if (sector_nr == 0) { |
3878 | /* If restarting in the middle, skip the initial sectors */ | |
3879 | if (mddev->delta_disks < 0 && | |
3880 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
3881 | sector_nr = raid5_size(mddev, 0, 0) | |
3882 | - conf->reshape_progress; | |
a639755c | 3883 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
3884 | conf->reshape_progress > 0) |
3885 | sector_nr = conf->reshape_progress; | |
f416885e | 3886 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 3887 | if (sector_nr) { |
8dee7211 N |
3888 | mddev->curr_resync_completed = sector_nr; |
3889 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
3890 | *skipped = 1; |
3891 | return sector_nr; | |
3892 | } | |
52c03291 N |
3893 | } |
3894 | ||
7a661381 N |
3895 | /* We need to process a full chunk at a time. |
3896 | * If old and new chunk sizes differ, we need to process the | |
3897 | * largest of these | |
3898 | */ | |
664e7c41 AN |
3899 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
3900 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 3901 | else |
9d8f0363 | 3902 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 3903 | |
52c03291 N |
3904 | /* we update the metadata when there is more than 3Meg |
3905 | * in the block range (that is rather arbitrary, should | |
3906 | * probably be time based) or when the data about to be | |
3907 | * copied would over-write the source of the data at | |
3908 | * the front of the range. | |
fef9c61f N |
3909 | * i.e. one new_stripe along from reshape_progress new_maps |
3910 | * to after where reshape_safe old_maps to | |
52c03291 | 3911 | */ |
fef9c61f | 3912 | writepos = conf->reshape_progress; |
f416885e | 3913 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
3914 | readpos = conf->reshape_progress; |
3915 | sector_div(readpos, data_disks); | |
fef9c61f | 3916 | safepos = conf->reshape_safe; |
f416885e | 3917 | sector_div(safepos, data_disks); |
fef9c61f | 3918 | if (mddev->delta_disks < 0) { |
ed37d83e | 3919 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 3920 | readpos += reshape_sectors; |
7a661381 | 3921 | safepos += reshape_sectors; |
fef9c61f | 3922 | } else { |
7a661381 | 3923 | writepos += reshape_sectors; |
ed37d83e N |
3924 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
3925 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 3926 | } |
52c03291 | 3927 | |
c8f517c4 N |
3928 | /* 'writepos' is the most advanced device address we might write. |
3929 | * 'readpos' is the least advanced device address we might read. | |
3930 | * 'safepos' is the least address recorded in the metadata as having | |
3931 | * been reshaped. | |
3932 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
3933 | * ensure safety in the face of a crash - that must be done by userspace | |
3934 | * making a backup of the data. So in that case there is no particular | |
3935 | * rush to update metadata. | |
3936 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
3937 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
3938 | * we can be safe in the event of a crash. | |
3939 | * So we insist on updating metadata if safepos is behind writepos and | |
3940 | * readpos is beyond writepos. | |
3941 | * In any case, update the metadata every 10 seconds. | |
3942 | * Maybe that number should be configurable, but I'm not sure it is | |
3943 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
3944 | */ | |
fef9c61f | 3945 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
3946 | ? (safepos > writepos && readpos < writepos) |
3947 | : (safepos < writepos && readpos > writepos)) || | |
3948 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
3949 | /* Cannot proceed until we've updated the superblock... */ |
3950 | wait_event(conf->wait_for_overlap, | |
3951 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 3952 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 3953 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 3954 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 3955 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 3956 | md_wakeup_thread(mddev->thread); |
850b2b42 | 3957 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
3958 | kthread_should_stop()); |
3959 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 3960 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
3961 | spin_unlock_irq(&conf->device_lock); |
3962 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 3963 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
3964 | } |
3965 | ||
ec32a2bd N |
3966 | if (mddev->delta_disks < 0) { |
3967 | BUG_ON(conf->reshape_progress == 0); | |
3968 | stripe_addr = writepos; | |
3969 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
3970 | ~((sector_t)reshape_sectors - 1)) |
3971 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
3972 | != sector_nr); |
3973 | } else { | |
7a661381 | 3974 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
3975 | stripe_addr = sector_nr; |
3976 | } | |
ab69ae12 | 3977 | INIT_LIST_HEAD(&stripes); |
7a661381 | 3978 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 3979 | int j; |
a9f326eb | 3980 | int skipped_disk = 0; |
a8c906ca | 3981 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
3982 | set_bit(STRIPE_EXPANDING, &sh->state); |
3983 | atomic_inc(&conf->reshape_stripes); | |
3984 | /* If any of this stripe is beyond the end of the old | |
3985 | * array, then we need to zero those blocks | |
3986 | */ | |
3987 | for (j=sh->disks; j--;) { | |
3988 | sector_t s; | |
3989 | if (j == sh->pd_idx) | |
3990 | continue; | |
f416885e | 3991 | if (conf->level == 6 && |
d0dabf7e | 3992 | j == sh->qd_idx) |
f416885e | 3993 | continue; |
784052ec | 3994 | s = compute_blocknr(sh, j, 0); |
b522adcd | 3995 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 3996 | skipped_disk = 1; |
52c03291 N |
3997 | continue; |
3998 | } | |
3999 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4000 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4001 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4002 | } | |
a9f326eb | 4003 | if (!skipped_disk) { |
52c03291 N |
4004 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4005 | set_bit(STRIPE_HANDLE, &sh->state); | |
4006 | } | |
ab69ae12 | 4007 | list_add(&sh->lru, &stripes); |
52c03291 N |
4008 | } |
4009 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4010 | if (mddev->delta_disks < 0) |
7a661381 | 4011 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4012 | else |
7a661381 | 4013 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4014 | spin_unlock_irq(&conf->device_lock); |
4015 | /* Ok, those stripe are ready. We can start scheduling | |
4016 | * reads on the source stripes. | |
4017 | * The source stripes are determined by mapping the first and last | |
4018 | * block on the destination stripes. | |
4019 | */ | |
52c03291 | 4020 | first_sector = |
ec32a2bd | 4021 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4022 | 1, &dd_idx, NULL); |
52c03291 | 4023 | last_sector = |
0e6e0271 | 4024 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4025 | * new_data_disks - 1), |
911d4ee8 | 4026 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4027 | if (last_sector >= mddev->dev_sectors) |
4028 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4029 | while (first_sector <= last_sector) { |
a8c906ca | 4030 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4031 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4032 | set_bit(STRIPE_HANDLE, &sh->state); | |
4033 | release_stripe(sh); | |
4034 | first_sector += STRIPE_SECTORS; | |
4035 | } | |
ab69ae12 N |
4036 | /* Now that the sources are clearly marked, we can release |
4037 | * the destination stripes | |
4038 | */ | |
4039 | while (!list_empty(&stripes)) { | |
4040 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4041 | list_del_init(&sh->lru); | |
4042 | release_stripe(sh); | |
4043 | } | |
c6207277 N |
4044 | /* If this takes us to the resync_max point where we have to pause, |
4045 | * then we need to write out the superblock. | |
4046 | */ | |
7a661381 | 4047 | sector_nr += reshape_sectors; |
c03f6a19 N |
4048 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4049 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4050 | /* Cannot proceed until we've updated the superblock... */ |
4051 | wait_event(conf->wait_for_overlap, | |
4052 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4053 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4054 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4055 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4056 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4057 | md_wakeup_thread(mddev->thread); | |
4058 | wait_event(mddev->sb_wait, | |
4059 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4060 | || kthread_should_stop()); | |
4061 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4062 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4063 | spin_unlock_irq(&conf->device_lock); |
4064 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4065 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4066 | } |
7a661381 | 4067 | return reshape_sectors; |
52c03291 N |
4068 | } |
4069 | ||
4070 | /* FIXME go_faster isn't used */ | |
fd01b88c | 4071 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
52c03291 | 4072 | { |
d1688a6d | 4073 | struct r5conf *conf = mddev->private; |
52c03291 | 4074 | struct stripe_head *sh; |
58c0fed4 | 4075 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 4076 | sector_t sync_blocks; |
16a53ecc N |
4077 | int still_degraded = 0; |
4078 | int i; | |
1da177e4 | 4079 | |
72626685 | 4080 | if (sector_nr >= max_sector) { |
1da177e4 | 4081 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 4082 | |
29269553 N |
4083 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4084 | end_reshape(conf); | |
4085 | return 0; | |
4086 | } | |
72626685 N |
4087 | |
4088 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4089 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4090 | &sync_blocks, 1); | |
16a53ecc | 4091 | else /* completed sync */ |
72626685 N |
4092 | conf->fullsync = 0; |
4093 | bitmap_close_sync(mddev->bitmap); | |
4094 | ||
1da177e4 LT |
4095 | return 0; |
4096 | } | |
ccfcc3c1 | 4097 | |
64bd660b N |
4098 | /* Allow raid5_quiesce to complete */ |
4099 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4100 | ||
52c03291 N |
4101 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4102 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4103 | |
c6207277 N |
4104 | /* No need to check resync_max as we never do more than one |
4105 | * stripe, and as resync_max will always be on a chunk boundary, | |
4106 | * if the check in md_do_sync didn't fire, there is no chance | |
4107 | * of overstepping resync_max here | |
4108 | */ | |
4109 | ||
16a53ecc | 4110 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4111 | * to resync, then assert that we are finished, because there is |
4112 | * nothing we can do. | |
4113 | */ | |
3285edf1 | 4114 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4115 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4116 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4117 | *skipped = 1; |
1da177e4 LT |
4118 | return rv; |
4119 | } | |
72626685 | 4120 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4121 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4122 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4123 | /* we can skip this block, and probably more */ | |
4124 | sync_blocks /= STRIPE_SECTORS; | |
4125 | *skipped = 1; | |
4126 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4127 | } | |
1da177e4 | 4128 | |
b47490c9 N |
4129 | |
4130 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
4131 | ||
a8c906ca | 4132 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4133 | if (sh == NULL) { |
a8c906ca | 4134 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4135 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4136 | * is trying to get access |
1da177e4 | 4137 | */ |
66c006a5 | 4138 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4139 | } |
16a53ecc N |
4140 | /* Need to check if array will still be degraded after recovery/resync |
4141 | * We don't need to check the 'failed' flag as when that gets set, | |
4142 | * recovery aborts. | |
4143 | */ | |
f001a70c | 4144 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4145 | if (conf->disks[i].rdev == NULL) |
4146 | still_degraded = 1; | |
4147 | ||
4148 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4149 | ||
83206d66 | 4150 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4151 | |
1442577b | 4152 | handle_stripe(sh); |
1da177e4 LT |
4153 | release_stripe(sh); |
4154 | ||
4155 | return STRIPE_SECTORS; | |
4156 | } | |
4157 | ||
d1688a6d | 4158 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
4159 | { |
4160 | /* We may not be able to submit a whole bio at once as there | |
4161 | * may not be enough stripe_heads available. | |
4162 | * We cannot pre-allocate enough stripe_heads as we may need | |
4163 | * more than exist in the cache (if we allow ever large chunks). | |
4164 | * So we do one stripe head at a time and record in | |
4165 | * ->bi_hw_segments how many have been done. | |
4166 | * | |
4167 | * We *know* that this entire raid_bio is in one chunk, so | |
4168 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4169 | */ | |
4170 | struct stripe_head *sh; | |
911d4ee8 | 4171 | int dd_idx; |
46031f9a RBJ |
4172 | sector_t sector, logical_sector, last_sector; |
4173 | int scnt = 0; | |
4174 | int remaining; | |
4175 | int handled = 0; | |
4176 | ||
4177 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4178 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4179 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4180 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4181 | ||
4182 | for (; logical_sector < last_sector; | |
387bb173 NB |
4183 | logical_sector += STRIPE_SECTORS, |
4184 | sector += STRIPE_SECTORS, | |
4185 | scnt++) { | |
46031f9a | 4186 | |
960e739d | 4187 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4188 | /* already done this stripe */ |
4189 | continue; | |
4190 | ||
a8c906ca | 4191 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4192 | |
4193 | if (!sh) { | |
4194 | /* failed to get a stripe - must wait */ | |
960e739d | 4195 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4196 | conf->retry_read_aligned = raid_bio; |
4197 | return handled; | |
4198 | } | |
4199 | ||
4200 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4201 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4202 | release_stripe(sh); | |
960e739d | 4203 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4204 | conf->retry_read_aligned = raid_bio; |
4205 | return handled; | |
4206 | } | |
4207 | ||
36d1c647 | 4208 | handle_stripe(sh); |
46031f9a RBJ |
4209 | release_stripe(sh); |
4210 | handled++; | |
4211 | } | |
4212 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4213 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4214 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4215 | if (remaining == 0) |
4216 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4217 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4218 | wake_up(&conf->wait_for_stripe); | |
4219 | return handled; | |
4220 | } | |
4221 | ||
46031f9a | 4222 | |
1da177e4 LT |
4223 | /* |
4224 | * This is our raid5 kernel thread. | |
4225 | * | |
4226 | * We scan the hash table for stripes which can be handled now. | |
4227 | * During the scan, completed stripes are saved for us by the interrupt | |
4228 | * handler, so that they will not have to wait for our next wakeup. | |
4229 | */ | |
fd01b88c | 4230 | static void raid5d(struct mddev *mddev) |
1da177e4 LT |
4231 | { |
4232 | struct stripe_head *sh; | |
d1688a6d | 4233 | struct r5conf *conf = mddev->private; |
1da177e4 | 4234 | int handled; |
e1dfa0a2 | 4235 | struct blk_plug plug; |
1da177e4 | 4236 | |
45b4233c | 4237 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4238 | |
4239 | md_check_recovery(mddev); | |
1da177e4 | 4240 | |
e1dfa0a2 | 4241 | blk_start_plug(&plug); |
1da177e4 LT |
4242 | handled = 0; |
4243 | spin_lock_irq(&conf->device_lock); | |
4244 | while (1) { | |
46031f9a | 4245 | struct bio *bio; |
1da177e4 | 4246 | |
7c13edc8 N |
4247 | if (atomic_read(&mddev->plug_cnt) == 0 && |
4248 | !list_empty(&conf->bitmap_list)) { | |
4249 | /* Now is a good time to flush some bitmap updates */ | |
4250 | conf->seq_flush++; | |
700e432d | 4251 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4252 | bitmap_unplug(mddev->bitmap); |
700e432d | 4253 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4254 | conf->seq_write = conf->seq_flush; |
72626685 N |
4255 | activate_bit_delay(conf); |
4256 | } | |
7c13edc8 N |
4257 | if (atomic_read(&mddev->plug_cnt) == 0) |
4258 | raid5_activate_delayed(conf); | |
72626685 | 4259 | |
46031f9a RBJ |
4260 | while ((bio = remove_bio_from_retry(conf))) { |
4261 | int ok; | |
4262 | spin_unlock_irq(&conf->device_lock); | |
4263 | ok = retry_aligned_read(conf, bio); | |
4264 | spin_lock_irq(&conf->device_lock); | |
4265 | if (!ok) | |
4266 | break; | |
4267 | handled++; | |
4268 | } | |
4269 | ||
8b3e6cdc DW |
4270 | sh = __get_priority_stripe(conf); |
4271 | ||
c9f21aaf | 4272 | if (!sh) |
1da177e4 | 4273 | break; |
1da177e4 LT |
4274 | spin_unlock_irq(&conf->device_lock); |
4275 | ||
4276 | handled++; | |
417b8d4a DW |
4277 | handle_stripe(sh); |
4278 | release_stripe(sh); | |
4279 | cond_resched(); | |
1da177e4 | 4280 | |
de393cde N |
4281 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
4282 | md_check_recovery(mddev); | |
4283 | ||
1da177e4 LT |
4284 | spin_lock_irq(&conf->device_lock); |
4285 | } | |
45b4233c | 4286 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4287 | |
4288 | spin_unlock_irq(&conf->device_lock); | |
4289 | ||
c9f21aaf | 4290 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4291 | blk_finish_plug(&plug); |
1da177e4 | 4292 | |
45b4233c | 4293 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4294 | } |
4295 | ||
3f294f4f | 4296 | static ssize_t |
fd01b88c | 4297 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 4298 | { |
d1688a6d | 4299 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4300 | if (conf) |
4301 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4302 | else | |
4303 | return 0; | |
3f294f4f N |
4304 | } |
4305 | ||
c41d4ac4 | 4306 | int |
fd01b88c | 4307 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 4308 | { |
d1688a6d | 4309 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
4310 | int err; |
4311 | ||
c41d4ac4 | 4312 | if (size <= 16 || size > 32768) |
3f294f4f | 4313 | return -EINVAL; |
c41d4ac4 | 4314 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4315 | if (drop_one_stripe(conf)) |
4316 | conf->max_nr_stripes--; | |
4317 | else | |
4318 | break; | |
4319 | } | |
b5470dc5 DW |
4320 | err = md_allow_write(mddev); |
4321 | if (err) | |
4322 | return err; | |
c41d4ac4 | 4323 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4324 | if (grow_one_stripe(conf)) |
4325 | conf->max_nr_stripes++; | |
4326 | else break; | |
4327 | } | |
c41d4ac4 N |
4328 | return 0; |
4329 | } | |
4330 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4331 | ||
4332 | static ssize_t | |
fd01b88c | 4333 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 4334 | { |
d1688a6d | 4335 | struct r5conf *conf = mddev->private; |
c41d4ac4 N |
4336 | unsigned long new; |
4337 | int err; | |
4338 | ||
4339 | if (len >= PAGE_SIZE) | |
4340 | return -EINVAL; | |
4341 | if (!conf) | |
4342 | return -ENODEV; | |
4343 | ||
4344 | if (strict_strtoul(page, 10, &new)) | |
4345 | return -EINVAL; | |
4346 | err = raid5_set_cache_size(mddev, new); | |
4347 | if (err) | |
4348 | return err; | |
3f294f4f N |
4349 | return len; |
4350 | } | |
007583c9 | 4351 | |
96de1e66 N |
4352 | static struct md_sysfs_entry |
4353 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4354 | raid5_show_stripe_cache_size, | |
4355 | raid5_store_stripe_cache_size); | |
3f294f4f | 4356 | |
8b3e6cdc | 4357 | static ssize_t |
fd01b88c | 4358 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 4359 | { |
d1688a6d | 4360 | struct r5conf *conf = mddev->private; |
8b3e6cdc DW |
4361 | if (conf) |
4362 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4363 | else | |
4364 | return 0; | |
4365 | } | |
4366 | ||
4367 | static ssize_t | |
fd01b88c | 4368 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 4369 | { |
d1688a6d | 4370 | struct r5conf *conf = mddev->private; |
4ef197d8 | 4371 | unsigned long new; |
8b3e6cdc DW |
4372 | if (len >= PAGE_SIZE) |
4373 | return -EINVAL; | |
4374 | if (!conf) | |
4375 | return -ENODEV; | |
4376 | ||
4ef197d8 | 4377 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4378 | return -EINVAL; |
4ef197d8 | 4379 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4380 | return -EINVAL; |
4381 | conf->bypass_threshold = new; | |
4382 | return len; | |
4383 | } | |
4384 | ||
4385 | static struct md_sysfs_entry | |
4386 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4387 | S_IRUGO | S_IWUSR, | |
4388 | raid5_show_preread_threshold, | |
4389 | raid5_store_preread_threshold); | |
4390 | ||
3f294f4f | 4391 | static ssize_t |
fd01b88c | 4392 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 4393 | { |
d1688a6d | 4394 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4395 | if (conf) |
4396 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4397 | else | |
4398 | return 0; | |
3f294f4f N |
4399 | } |
4400 | ||
96de1e66 N |
4401 | static struct md_sysfs_entry |
4402 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4403 | |
007583c9 | 4404 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4405 | &raid5_stripecache_size.attr, |
4406 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4407 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4408 | NULL, |
4409 | }; | |
007583c9 N |
4410 | static struct attribute_group raid5_attrs_group = { |
4411 | .name = NULL, | |
4412 | .attrs = raid5_attrs, | |
3f294f4f N |
4413 | }; |
4414 | ||
80c3a6ce | 4415 | static sector_t |
fd01b88c | 4416 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 4417 | { |
d1688a6d | 4418 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
4419 | |
4420 | if (!sectors) | |
4421 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4422 | if (!raid_disks) |
7ec05478 | 4423 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4424 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4425 | |
9d8f0363 | 4426 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4427 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4428 | return sectors * (raid_disks - conf->max_degraded); |
4429 | } | |
4430 | ||
d1688a6d | 4431 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 DW |
4432 | { |
4433 | struct raid5_percpu *percpu; | |
4434 | unsigned long cpu; | |
4435 | ||
4436 | if (!conf->percpu) | |
4437 | return; | |
4438 | ||
4439 | get_online_cpus(); | |
4440 | for_each_possible_cpu(cpu) { | |
4441 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4442 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4443 | kfree(percpu->scribble); |
36d1c647 DW |
4444 | } |
4445 | #ifdef CONFIG_HOTPLUG_CPU | |
4446 | unregister_cpu_notifier(&conf->cpu_notify); | |
4447 | #endif | |
4448 | put_online_cpus(); | |
4449 | ||
4450 | free_percpu(conf->percpu); | |
4451 | } | |
4452 | ||
d1688a6d | 4453 | static void free_conf(struct r5conf *conf) |
95fc17aa DW |
4454 | { |
4455 | shrink_stripes(conf); | |
36d1c647 | 4456 | raid5_free_percpu(conf); |
95fc17aa DW |
4457 | kfree(conf->disks); |
4458 | kfree(conf->stripe_hashtbl); | |
4459 | kfree(conf); | |
4460 | } | |
4461 | ||
36d1c647 DW |
4462 | #ifdef CONFIG_HOTPLUG_CPU |
4463 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4464 | void *hcpu) | |
4465 | { | |
d1688a6d | 4466 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
4467 | long cpu = (long)hcpu; |
4468 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4469 | ||
4470 | switch (action) { | |
4471 | case CPU_UP_PREPARE: | |
4472 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4473 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4474 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4475 | if (!percpu->scribble) |
4476 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4477 | ||
4478 | if (!percpu->scribble || | |
4479 | (conf->level == 6 && !percpu->spare_page)) { | |
4480 | safe_put_page(percpu->spare_page); | |
4481 | kfree(percpu->scribble); | |
36d1c647 DW |
4482 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4483 | __func__, cpu); | |
55af6bb5 | 4484 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4485 | } |
4486 | break; | |
4487 | case CPU_DEAD: | |
4488 | case CPU_DEAD_FROZEN: | |
4489 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4490 | kfree(percpu->scribble); |
36d1c647 | 4491 | percpu->spare_page = NULL; |
d6f38f31 | 4492 | percpu->scribble = NULL; |
36d1c647 DW |
4493 | break; |
4494 | default: | |
4495 | break; | |
4496 | } | |
4497 | return NOTIFY_OK; | |
4498 | } | |
4499 | #endif | |
4500 | ||
d1688a6d | 4501 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
4502 | { |
4503 | unsigned long cpu; | |
4504 | struct page *spare_page; | |
a29d8b8e | 4505 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4506 | void *scribble; |
36d1c647 DW |
4507 | int err; |
4508 | ||
36d1c647 DW |
4509 | allcpus = alloc_percpu(struct raid5_percpu); |
4510 | if (!allcpus) | |
4511 | return -ENOMEM; | |
4512 | conf->percpu = allcpus; | |
4513 | ||
4514 | get_online_cpus(); | |
4515 | err = 0; | |
4516 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4517 | if (conf->level == 6) { |
4518 | spare_page = alloc_page(GFP_KERNEL); | |
4519 | if (!spare_page) { | |
4520 | err = -ENOMEM; | |
4521 | break; | |
4522 | } | |
4523 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4524 | } | |
5e5e3e78 | 4525 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4526 | if (!scribble) { |
36d1c647 DW |
4527 | err = -ENOMEM; |
4528 | break; | |
4529 | } | |
d6f38f31 | 4530 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4531 | } |
4532 | #ifdef CONFIG_HOTPLUG_CPU | |
4533 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4534 | conf->cpu_notify.priority = 0; | |
4535 | if (err == 0) | |
4536 | err = register_cpu_notifier(&conf->cpu_notify); | |
4537 | #endif | |
4538 | put_online_cpus(); | |
4539 | ||
4540 | return err; | |
4541 | } | |
4542 | ||
d1688a6d | 4543 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 4544 | { |
d1688a6d | 4545 | struct r5conf *conf; |
5e5e3e78 | 4546 | int raid_disk, memory, max_disks; |
3cb03002 | 4547 | struct md_rdev *rdev; |
1da177e4 | 4548 | struct disk_info *disk; |
1da177e4 | 4549 | |
91adb564 N |
4550 | if (mddev->new_level != 5 |
4551 | && mddev->new_level != 4 | |
4552 | && mddev->new_level != 6) { | |
0c55e022 | 4553 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4554 | mdname(mddev), mddev->new_level); |
4555 | return ERR_PTR(-EIO); | |
1da177e4 | 4556 | } |
91adb564 N |
4557 | if ((mddev->new_level == 5 |
4558 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4559 | (mddev->new_level == 6 | |
4560 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4561 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4562 | mdname(mddev), mddev->new_layout); |
4563 | return ERR_PTR(-EIO); | |
99c0fb5f | 4564 | } |
91adb564 | 4565 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4566 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4567 | mdname(mddev), mddev->raid_disks); |
4568 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4569 | } |
4570 | ||
664e7c41 AN |
4571 | if (!mddev->new_chunk_sectors || |
4572 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4573 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4574 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4575 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4576 | return ERR_PTR(-EINVAL); |
f6705578 N |
4577 | } |
4578 | ||
d1688a6d | 4579 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 4580 | if (conf == NULL) |
1da177e4 | 4581 | goto abort; |
f5efd45a DW |
4582 | spin_lock_init(&conf->device_lock); |
4583 | init_waitqueue_head(&conf->wait_for_stripe); | |
4584 | init_waitqueue_head(&conf->wait_for_overlap); | |
4585 | INIT_LIST_HEAD(&conf->handle_list); | |
4586 | INIT_LIST_HEAD(&conf->hold_list); | |
4587 | INIT_LIST_HEAD(&conf->delayed_list); | |
4588 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4589 | INIT_LIST_HEAD(&conf->inactive_list); | |
4590 | atomic_set(&conf->active_stripes, 0); | |
4591 | atomic_set(&conf->preread_active_stripes, 0); | |
4592 | atomic_set(&conf->active_aligned_reads, 0); | |
4593 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 4594 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
4595 | |
4596 | conf->raid_disks = mddev->raid_disks; | |
4597 | if (mddev->reshape_position == MaxSector) | |
4598 | conf->previous_raid_disks = mddev->raid_disks; | |
4599 | else | |
f6705578 | 4600 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4601 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4602 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4603 | |
5e5e3e78 | 4604 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4605 | GFP_KERNEL); |
4606 | if (!conf->disks) | |
4607 | goto abort; | |
9ffae0cf | 4608 | |
1da177e4 LT |
4609 | conf->mddev = mddev; |
4610 | ||
fccddba0 | 4611 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4612 | goto abort; |
1da177e4 | 4613 | |
36d1c647 DW |
4614 | conf->level = mddev->new_level; |
4615 | if (raid5_alloc_percpu(conf) != 0) | |
4616 | goto abort; | |
4617 | ||
0c55e022 | 4618 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4619 | |
159ec1fc | 4620 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4621 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4622 | if (raid_disk >= max_disks |
1da177e4 LT |
4623 | || raid_disk < 0) |
4624 | continue; | |
4625 | disk = conf->disks + raid_disk; | |
4626 | ||
4627 | disk->rdev = rdev; | |
4628 | ||
b2d444d7 | 4629 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 4630 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
4631 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
4632 | " disk %d\n", | |
4633 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 4634 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
4635 | /* Cannot rely on bitmap to complete recovery */ |
4636 | conf->fullsync = 1; | |
1da177e4 LT |
4637 | } |
4638 | ||
09c9e5fa | 4639 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4640 | conf->level = mddev->new_level; |
16a53ecc N |
4641 | if (conf->level == 6) |
4642 | conf->max_degraded = 2; | |
4643 | else | |
4644 | conf->max_degraded = 1; | |
91adb564 | 4645 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4646 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4647 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4648 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4649 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4650 | conf->prev_algo = mddev->layout; |
4651 | } | |
1da177e4 | 4652 | |
91adb564 | 4653 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4654 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4655 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4656 | printk(KERN_ERR | |
0c55e022 N |
4657 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
4658 | mdname(mddev), memory); | |
91adb564 N |
4659 | goto abort; |
4660 | } else | |
0c55e022 N |
4661 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
4662 | mdname(mddev), memory); | |
1da177e4 | 4663 | |
0da3c619 | 4664 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4665 | if (!conf->thread) { |
4666 | printk(KERN_ERR | |
0c55e022 | 4667 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 4668 | mdname(mddev)); |
16a53ecc N |
4669 | goto abort; |
4670 | } | |
91adb564 N |
4671 | |
4672 | return conf; | |
4673 | ||
4674 | abort: | |
4675 | if (conf) { | |
95fc17aa | 4676 | free_conf(conf); |
91adb564 N |
4677 | return ERR_PTR(-EIO); |
4678 | } else | |
4679 | return ERR_PTR(-ENOMEM); | |
4680 | } | |
4681 | ||
c148ffdc N |
4682 | |
4683 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4684 | { | |
4685 | switch (algo) { | |
4686 | case ALGORITHM_PARITY_0: | |
4687 | if (raid_disk < max_degraded) | |
4688 | return 1; | |
4689 | break; | |
4690 | case ALGORITHM_PARITY_N: | |
4691 | if (raid_disk >= raid_disks - max_degraded) | |
4692 | return 1; | |
4693 | break; | |
4694 | case ALGORITHM_PARITY_0_6: | |
4695 | if (raid_disk == 0 || | |
4696 | raid_disk == raid_disks - 1) | |
4697 | return 1; | |
4698 | break; | |
4699 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4700 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4701 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4702 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4703 | if (raid_disk == raid_disks - 1) | |
4704 | return 1; | |
4705 | } | |
4706 | return 0; | |
4707 | } | |
4708 | ||
fd01b88c | 4709 | static int run(struct mddev *mddev) |
91adb564 | 4710 | { |
d1688a6d | 4711 | struct r5conf *conf; |
9f7c2220 | 4712 | int working_disks = 0; |
c148ffdc | 4713 | int dirty_parity_disks = 0; |
3cb03002 | 4714 | struct md_rdev *rdev; |
c148ffdc | 4715 | sector_t reshape_offset = 0; |
91adb564 | 4716 | |
8c6ac868 | 4717 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 4718 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
4719 | " -- starting background reconstruction\n", |
4720 | mdname(mddev)); | |
91adb564 N |
4721 | if (mddev->reshape_position != MaxSector) { |
4722 | /* Check that we can continue the reshape. | |
4723 | * Currently only disks can change, it must | |
4724 | * increase, and we must be past the point where | |
4725 | * a stripe over-writes itself | |
4726 | */ | |
4727 | sector_t here_new, here_old; | |
4728 | int old_disks; | |
18b00334 | 4729 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4730 | |
88ce4930 | 4731 | if (mddev->new_level != mddev->level) { |
0c55e022 | 4732 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
4733 | "required - aborting.\n", |
4734 | mdname(mddev)); | |
4735 | return -EINVAL; | |
4736 | } | |
91adb564 N |
4737 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4738 | /* reshape_position must be on a new-stripe boundary, and one | |
4739 | * further up in new geometry must map after here in old | |
4740 | * geometry. | |
4741 | */ | |
4742 | here_new = mddev->reshape_position; | |
664e7c41 | 4743 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 4744 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
4745 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
4746 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
4747 | return -EINVAL; |
4748 | } | |
c148ffdc | 4749 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4750 | /* here_new is the stripe we will write to */ |
4751 | here_old = mddev->reshape_position; | |
9d8f0363 | 4752 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4753 | (old_disks-max_degraded)); |
4754 | /* here_old is the first stripe that we might need to read | |
4755 | * from */ | |
67ac6011 N |
4756 | if (mddev->delta_disks == 0) { |
4757 | /* We cannot be sure it is safe to start an in-place | |
4758 | * reshape. It is only safe if user-space if monitoring | |
4759 | * and taking constant backups. | |
4760 | * mdadm always starts a situation like this in | |
4761 | * readonly mode so it can take control before | |
4762 | * allowing any writes. So just check for that. | |
4763 | */ | |
4764 | if ((here_new * mddev->new_chunk_sectors != | |
4765 | here_old * mddev->chunk_sectors) || | |
4766 | mddev->ro == 0) { | |
0c55e022 N |
4767 | printk(KERN_ERR "md/raid:%s: in-place reshape must be started" |
4768 | " in read-only mode - aborting\n", | |
4769 | mdname(mddev)); | |
67ac6011 N |
4770 | return -EINVAL; |
4771 | } | |
4772 | } else if (mddev->delta_disks < 0 | |
4773 | ? (here_new * mddev->new_chunk_sectors <= | |
4774 | here_old * mddev->chunk_sectors) | |
4775 | : (here_new * mddev->new_chunk_sectors >= | |
4776 | here_old * mddev->chunk_sectors)) { | |
91adb564 | 4777 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
4778 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
4779 | "auto-recovery - aborting.\n", | |
4780 | mdname(mddev)); | |
91adb564 N |
4781 | return -EINVAL; |
4782 | } | |
0c55e022 N |
4783 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
4784 | mdname(mddev)); | |
91adb564 N |
4785 | /* OK, we should be able to continue; */ |
4786 | } else { | |
4787 | BUG_ON(mddev->level != mddev->new_level); | |
4788 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 4789 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 4790 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 4791 | } |
91adb564 | 4792 | |
245f46c2 N |
4793 | if (mddev->private == NULL) |
4794 | conf = setup_conf(mddev); | |
4795 | else | |
4796 | conf = mddev->private; | |
4797 | ||
91adb564 N |
4798 | if (IS_ERR(conf)) |
4799 | return PTR_ERR(conf); | |
4800 | ||
4801 | mddev->thread = conf->thread; | |
4802 | conf->thread = NULL; | |
4803 | mddev->private = conf; | |
4804 | ||
4805 | /* | |
4806 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4807 | */ | |
c148ffdc N |
4808 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
4809 | if (rdev->raid_disk < 0) | |
4810 | continue; | |
2f115882 | 4811 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 4812 | working_disks++; |
2f115882 N |
4813 | continue; |
4814 | } | |
c148ffdc N |
4815 | /* This disc is not fully in-sync. However if it |
4816 | * just stored parity (beyond the recovery_offset), | |
4817 | * when we don't need to be concerned about the | |
4818 | * array being dirty. | |
4819 | * When reshape goes 'backwards', we never have | |
4820 | * partially completed devices, so we only need | |
4821 | * to worry about reshape going forwards. | |
4822 | */ | |
4823 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
4824 | if (mddev->major_version == 0 && | |
4825 | mddev->minor_version > 90) | |
4826 | rdev->recovery_offset = reshape_offset; | |
4827 | ||
c148ffdc N |
4828 | if (rdev->recovery_offset < reshape_offset) { |
4829 | /* We need to check old and new layout */ | |
4830 | if (!only_parity(rdev->raid_disk, | |
4831 | conf->algorithm, | |
4832 | conf->raid_disks, | |
4833 | conf->max_degraded)) | |
4834 | continue; | |
4835 | } | |
4836 | if (!only_parity(rdev->raid_disk, | |
4837 | conf->prev_algo, | |
4838 | conf->previous_raid_disks, | |
4839 | conf->max_degraded)) | |
4840 | continue; | |
4841 | dirty_parity_disks++; | |
4842 | } | |
91adb564 | 4843 | |
5e5e3e78 N |
4844 | mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks) |
4845 | - working_disks); | |
91adb564 | 4846 | |
674806d6 | 4847 | if (has_failed(conf)) { |
0c55e022 | 4848 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 4849 | " (%d/%d failed)\n", |
02c2de8c | 4850 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4851 | goto abort; |
4852 | } | |
4853 | ||
91adb564 | 4854 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 4855 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
4856 | mddev->resync_max_sectors = mddev->dev_sectors; |
4857 | ||
c148ffdc | 4858 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 4859 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4860 | if (mddev->ok_start_degraded) |
4861 | printk(KERN_WARNING | |
0c55e022 N |
4862 | "md/raid:%s: starting dirty degraded array" |
4863 | " - data corruption possible.\n", | |
6ff8d8ec N |
4864 | mdname(mddev)); |
4865 | else { | |
4866 | printk(KERN_ERR | |
0c55e022 | 4867 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
4868 | mdname(mddev)); |
4869 | goto abort; | |
4870 | } | |
1da177e4 LT |
4871 | } |
4872 | ||
1da177e4 | 4873 | if (mddev->degraded == 0) |
0c55e022 N |
4874 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
4875 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
4876 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
4877 | mddev->new_layout); | |
1da177e4 | 4878 | else |
0c55e022 N |
4879 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
4880 | " out of %d devices, algorithm %d\n", | |
4881 | mdname(mddev), conf->level, | |
4882 | mddev->raid_disks - mddev->degraded, | |
4883 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
4884 | |
4885 | print_raid5_conf(conf); | |
4886 | ||
fef9c61f | 4887 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 4888 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
4889 | atomic_set(&conf->reshape_stripes, 0); |
4890 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4891 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4892 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4893 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4894 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 4895 | "reshape"); |
f6705578 N |
4896 | } |
4897 | ||
1da177e4 LT |
4898 | |
4899 | /* Ok, everything is just fine now */ | |
a64c876f N |
4900 | if (mddev->to_remove == &raid5_attrs_group) |
4901 | mddev->to_remove = NULL; | |
00bcb4ac N |
4902 | else if (mddev->kobj.sd && |
4903 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 4904 | printk(KERN_WARNING |
4a5add49 | 4905 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 4906 | mdname(mddev)); |
4a5add49 | 4907 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 4908 | |
4a5add49 | 4909 | if (mddev->queue) { |
9f7c2220 | 4910 | int chunk_size; |
4a5add49 N |
4911 | /* read-ahead size must cover two whole stripes, which |
4912 | * is 2 * (datadisks) * chunksize where 'n' is the | |
4913 | * number of raid devices | |
4914 | */ | |
4915 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
4916 | int stripe = data_disks * | |
4917 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
4918 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4919 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 4920 | |
4a5add49 | 4921 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 4922 | |
11d8a6e3 N |
4923 | mddev->queue->backing_dev_info.congested_data = mddev; |
4924 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 4925 | |
9f7c2220 N |
4926 | chunk_size = mddev->chunk_sectors << 9; |
4927 | blk_queue_io_min(mddev->queue, chunk_size); | |
4928 | blk_queue_io_opt(mddev->queue, chunk_size * | |
4929 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 4930 | |
9f7c2220 N |
4931 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4932 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
4933 | rdev->data_offset << 9); | |
4934 | } | |
23032a0e | 4935 | |
1da177e4 LT |
4936 | return 0; |
4937 | abort: | |
01f96c0a | 4938 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
4939 | print_raid5_conf(conf); |
4940 | free_conf(conf); | |
1da177e4 | 4941 | mddev->private = NULL; |
0c55e022 | 4942 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
4943 | return -EIO; |
4944 | } | |
4945 | ||
fd01b88c | 4946 | static int stop(struct mddev *mddev) |
1da177e4 | 4947 | { |
d1688a6d | 4948 | struct r5conf *conf = mddev->private; |
1da177e4 | 4949 | |
01f96c0a | 4950 | md_unregister_thread(&mddev->thread); |
11d8a6e3 N |
4951 | if (mddev->queue) |
4952 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 4953 | free_conf(conf); |
a64c876f N |
4954 | mddev->private = NULL; |
4955 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
4956 | return 0; |
4957 | } | |
4958 | ||
fd01b88c | 4959 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 4960 | { |
d1688a6d | 4961 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
4962 | int i; |
4963 | ||
9d8f0363 AN |
4964 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
4965 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 4966 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
4967 | for (i = 0; i < conf->raid_disks; i++) |
4968 | seq_printf (seq, "%s", | |
4969 | conf->disks[i].rdev && | |
b2d444d7 | 4970 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 4971 | seq_printf (seq, "]"); |
1da177e4 LT |
4972 | } |
4973 | ||
d1688a6d | 4974 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
4975 | { |
4976 | int i; | |
4977 | struct disk_info *tmp; | |
4978 | ||
0c55e022 | 4979 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
4980 | if (!conf) { |
4981 | printk("(conf==NULL)\n"); | |
4982 | return; | |
4983 | } | |
0c55e022 N |
4984 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
4985 | conf->raid_disks, | |
4986 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
4987 | |
4988 | for (i = 0; i < conf->raid_disks; i++) { | |
4989 | char b[BDEVNAME_SIZE]; | |
4990 | tmp = conf->disks + i; | |
4991 | if (tmp->rdev) | |
0c55e022 N |
4992 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
4993 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
4994 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
4995 | } |
4996 | } | |
4997 | ||
fd01b88c | 4998 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
4999 | { |
5000 | int i; | |
d1688a6d | 5001 | struct r5conf *conf = mddev->private; |
1da177e4 | 5002 | struct disk_info *tmp; |
6b965620 N |
5003 | int count = 0; |
5004 | unsigned long flags; | |
1da177e4 LT |
5005 | |
5006 | for (i = 0; i < conf->raid_disks; i++) { | |
5007 | tmp = conf->disks + i; | |
5008 | if (tmp->rdev | |
70fffd0b | 5009 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 5010 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 5011 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 5012 | count++; |
43c73ca4 | 5013 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
5014 | } |
5015 | } | |
6b965620 N |
5016 | spin_lock_irqsave(&conf->device_lock, flags); |
5017 | mddev->degraded -= count; | |
5018 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 5019 | print_raid5_conf(conf); |
6b965620 | 5020 | return count; |
1da177e4 LT |
5021 | } |
5022 | ||
fd01b88c | 5023 | static int raid5_remove_disk(struct mddev *mddev, int number) |
1da177e4 | 5024 | { |
d1688a6d | 5025 | struct r5conf *conf = mddev->private; |
1da177e4 | 5026 | int err = 0; |
3cb03002 | 5027 | struct md_rdev *rdev; |
1da177e4 LT |
5028 | struct disk_info *p = conf->disks + number; |
5029 | ||
5030 | print_raid5_conf(conf); | |
5031 | rdev = p->rdev; | |
5032 | if (rdev) { | |
ec32a2bd N |
5033 | if (number >= conf->raid_disks && |
5034 | conf->reshape_progress == MaxSector) | |
5035 | clear_bit(In_sync, &rdev->flags); | |
5036 | ||
b2d444d7 | 5037 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
5038 | atomic_read(&rdev->nr_pending)) { |
5039 | err = -EBUSY; | |
5040 | goto abort; | |
5041 | } | |
dfc70645 N |
5042 | /* Only remove non-faulty devices if recovery |
5043 | * isn't possible. | |
5044 | */ | |
5045 | if (!test_bit(Faulty, &rdev->flags) && | |
7f0da59b | 5046 | mddev->recovery_disabled != conf->recovery_disabled && |
674806d6 | 5047 | !has_failed(conf) && |
ec32a2bd | 5048 | number < conf->raid_disks) { |
dfc70645 N |
5049 | err = -EBUSY; |
5050 | goto abort; | |
5051 | } | |
1da177e4 | 5052 | p->rdev = NULL; |
fbd568a3 | 5053 | synchronize_rcu(); |
1da177e4 LT |
5054 | if (atomic_read(&rdev->nr_pending)) { |
5055 | /* lost the race, try later */ | |
5056 | err = -EBUSY; | |
5057 | p->rdev = rdev; | |
5058 | } | |
5059 | } | |
5060 | abort: | |
5061 | ||
5062 | print_raid5_conf(conf); | |
5063 | return err; | |
5064 | } | |
5065 | ||
fd01b88c | 5066 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5067 | { |
d1688a6d | 5068 | struct r5conf *conf = mddev->private; |
199050ea | 5069 | int err = -EEXIST; |
1da177e4 LT |
5070 | int disk; |
5071 | struct disk_info *p; | |
6c2fce2e NB |
5072 | int first = 0; |
5073 | int last = conf->raid_disks - 1; | |
1da177e4 | 5074 | |
7f0da59b N |
5075 | if (mddev->recovery_disabled == conf->recovery_disabled) |
5076 | return -EBUSY; | |
5077 | ||
674806d6 | 5078 | if (has_failed(conf)) |
1da177e4 | 5079 | /* no point adding a device */ |
199050ea | 5080 | return -EINVAL; |
1da177e4 | 5081 | |
6c2fce2e NB |
5082 | if (rdev->raid_disk >= 0) |
5083 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5084 | |
5085 | /* | |
16a53ecc N |
5086 | * find the disk ... but prefer rdev->saved_raid_disk |
5087 | * if possible. | |
1da177e4 | 5088 | */ |
16a53ecc | 5089 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5090 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5091 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5092 | disk = rdev->saved_raid_disk; | |
5093 | else | |
6c2fce2e NB |
5094 | disk = first; |
5095 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5096 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5097 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5098 | rdev->raid_disk = disk; |
199050ea | 5099 | err = 0; |
72626685 N |
5100 | if (rdev->saved_raid_disk != disk) |
5101 | conf->fullsync = 1; | |
d6065f7b | 5102 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5103 | break; |
5104 | } | |
5105 | print_raid5_conf(conf); | |
199050ea | 5106 | return err; |
1da177e4 LT |
5107 | } |
5108 | ||
fd01b88c | 5109 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
5110 | { |
5111 | /* no resync is happening, and there is enough space | |
5112 | * on all devices, so we can resize. | |
5113 | * We need to make sure resync covers any new space. | |
5114 | * If the array is shrinking we should possibly wait until | |
5115 | * any io in the removed space completes, but it hardly seems | |
5116 | * worth it. | |
5117 | */ | |
9d8f0363 | 5118 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5119 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5120 | mddev->raid_disks)); | |
b522adcd DW |
5121 | if (mddev->array_sectors > |
5122 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5123 | return -EINVAL; | |
f233ea5c | 5124 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5125 | revalidate_disk(mddev->gendisk); |
b098636c N |
5126 | if (sectors > mddev->dev_sectors && |
5127 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5128 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5129 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5130 | } | |
58c0fed4 | 5131 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5132 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5133 | return 0; |
5134 | } | |
5135 | ||
fd01b88c | 5136 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
5137 | { |
5138 | /* Can only proceed if there are plenty of stripe_heads. | |
5139 | * We need a minimum of one full stripe,, and for sensible progress | |
5140 | * it is best to have about 4 times that. | |
5141 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5142 | * allow for chunk sizes up to 256K, which is probably OK. | |
5143 | * If the chunk size is greater, user-space should request more | |
5144 | * stripe_heads first. | |
5145 | */ | |
d1688a6d | 5146 | struct r5conf *conf = mddev->private; |
01ee22b4 N |
5147 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
5148 | > conf->max_nr_stripes || | |
5149 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5150 | > conf->max_nr_stripes) { | |
0c55e022 N |
5151 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5152 | mdname(mddev), | |
01ee22b4 N |
5153 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5154 | / STRIPE_SIZE)*4); | |
5155 | return 0; | |
5156 | } | |
5157 | return 1; | |
5158 | } | |
5159 | ||
fd01b88c | 5160 | static int check_reshape(struct mddev *mddev) |
29269553 | 5161 | { |
d1688a6d | 5162 | struct r5conf *conf = mddev->private; |
29269553 | 5163 | |
88ce4930 N |
5164 | if (mddev->delta_disks == 0 && |
5165 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5166 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5167 | return 0; /* nothing to do */ |
dba034ee N |
5168 | if (mddev->bitmap) |
5169 | /* Cannot grow a bitmap yet */ | |
5170 | return -EBUSY; | |
674806d6 | 5171 | if (has_failed(conf)) |
ec32a2bd N |
5172 | return -EINVAL; |
5173 | if (mddev->delta_disks < 0) { | |
5174 | /* We might be able to shrink, but the devices must | |
5175 | * be made bigger first. | |
5176 | * For raid6, 4 is the minimum size. | |
5177 | * Otherwise 2 is the minimum | |
5178 | */ | |
5179 | int min = 2; | |
5180 | if (mddev->level == 6) | |
5181 | min = 4; | |
5182 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5183 | return -EINVAL; | |
5184 | } | |
29269553 | 5185 | |
01ee22b4 | 5186 | if (!check_stripe_cache(mddev)) |
29269553 | 5187 | return -ENOSPC; |
29269553 | 5188 | |
ec32a2bd | 5189 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5190 | } |
5191 | ||
fd01b88c | 5192 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 5193 | { |
d1688a6d | 5194 | struct r5conf *conf = mddev->private; |
3cb03002 | 5195 | struct md_rdev *rdev; |
63c70c4f | 5196 | int spares = 0; |
c04be0aa | 5197 | unsigned long flags; |
63c70c4f | 5198 | |
f416885e | 5199 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5200 | return -EBUSY; |
5201 | ||
01ee22b4 N |
5202 | if (!check_stripe_cache(mddev)) |
5203 | return -ENOSPC; | |
5204 | ||
159ec1fc | 5205 | list_for_each_entry(rdev, &mddev->disks, same_set) |
469518a3 N |
5206 | if (!test_bit(In_sync, &rdev->flags) |
5207 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5208 | spares++; |
63c70c4f | 5209 | |
f416885e | 5210 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5211 | /* Not enough devices even to make a degraded array |
5212 | * of that size | |
5213 | */ | |
5214 | return -EINVAL; | |
5215 | ||
ec32a2bd N |
5216 | /* Refuse to reduce size of the array. Any reductions in |
5217 | * array size must be through explicit setting of array_size | |
5218 | * attribute. | |
5219 | */ | |
5220 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5221 | < mddev->array_sectors) { | |
0c55e022 | 5222 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5223 | "before number of disks\n", mdname(mddev)); |
5224 | return -EINVAL; | |
5225 | } | |
5226 | ||
f6705578 | 5227 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5228 | spin_lock_irq(&conf->device_lock); |
5229 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5230 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5231 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5232 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5233 | conf->prev_algo = conf->algorithm; |
5234 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5235 | if (mddev->delta_disks < 0) |
5236 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5237 | else | |
5238 | conf->reshape_progress = 0; | |
5239 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5240 | conf->generation++; |
29269553 N |
5241 | spin_unlock_irq(&conf->device_lock); |
5242 | ||
5243 | /* Add some new drives, as many as will fit. | |
5244 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5245 | * Don't add devices if we are reducing the number of |
5246 | * devices in the array. This is because it is not possible | |
5247 | * to correctly record the "partially reconstructed" state of | |
5248 | * such devices during the reshape and confusion could result. | |
29269553 | 5249 | */ |
87a8dec9 N |
5250 | if (mddev->delta_disks >= 0) { |
5251 | int added_devices = 0; | |
5252 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5253 | if (rdev->raid_disk < 0 && | |
5254 | !test_bit(Faulty, &rdev->flags)) { | |
5255 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 N |
5256 | if (rdev->raid_disk |
5257 | >= conf->previous_raid_disks) { | |
5258 | set_bit(In_sync, &rdev->flags); | |
5259 | added_devices++; | |
5260 | } else | |
5261 | rdev->recovery_offset = 0; | |
36fad858 NK |
5262 | |
5263 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 5264 | /* Failure here is OK */; |
50da0840 | 5265 | } |
87a8dec9 N |
5266 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5267 | && !test_bit(Faulty, &rdev->flags)) { | |
5268 | /* This is a spare that was manually added */ | |
5269 | set_bit(In_sync, &rdev->flags); | |
5270 | added_devices++; | |
5271 | } | |
29269553 | 5272 | |
87a8dec9 N |
5273 | /* When a reshape changes the number of devices, |
5274 | * ->degraded is measured against the larger of the | |
5275 | * pre and post number of devices. | |
5276 | */ | |
ec32a2bd | 5277 | spin_lock_irqsave(&conf->device_lock, flags); |
9eb07c25 | 5278 | mddev->degraded += (conf->raid_disks - conf->previous_raid_disks) |
ec32a2bd N |
5279 | - added_devices; |
5280 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5281 | } | |
63c70c4f | 5282 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5283 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5284 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5285 | |
29269553 N |
5286 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5287 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5288 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5289 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5290 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5291 | "reshape"); |
29269553 N |
5292 | if (!mddev->sync_thread) { |
5293 | mddev->recovery = 0; | |
5294 | spin_lock_irq(&conf->device_lock); | |
5295 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5296 | conf->reshape_progress = MaxSector; |
29269553 N |
5297 | spin_unlock_irq(&conf->device_lock); |
5298 | return -EAGAIN; | |
5299 | } | |
c8f517c4 | 5300 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5301 | md_wakeup_thread(mddev->sync_thread); |
5302 | md_new_event(mddev); | |
5303 | return 0; | |
5304 | } | |
29269553 | 5305 | |
ec32a2bd N |
5306 | /* This is called from the reshape thread and should make any |
5307 | * changes needed in 'conf' | |
5308 | */ | |
d1688a6d | 5309 | static void end_reshape(struct r5conf *conf) |
29269553 | 5310 | { |
29269553 | 5311 | |
f6705578 | 5312 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5313 | |
f6705578 | 5314 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5315 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5316 | conf->reshape_progress = MaxSector; |
f6705578 | 5317 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5318 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5319 | |
5320 | /* read-ahead size must cover two whole stripes, which is | |
5321 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5322 | */ | |
4a5add49 | 5323 | if (conf->mddev->queue) { |
cea9c228 | 5324 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5325 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5326 | / PAGE_SIZE); |
16a53ecc N |
5327 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5328 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5329 | } | |
29269553 | 5330 | } |
29269553 N |
5331 | } |
5332 | ||
ec32a2bd N |
5333 | /* This is called from the raid5d thread with mddev_lock held. |
5334 | * It makes config changes to the device. | |
5335 | */ | |
fd01b88c | 5336 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 5337 | { |
d1688a6d | 5338 | struct r5conf *conf = mddev->private; |
cea9c228 N |
5339 | |
5340 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5341 | ||
ec32a2bd N |
5342 | if (mddev->delta_disks > 0) { |
5343 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5344 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5345 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5346 | } else { |
5347 | int d; | |
ec32a2bd N |
5348 | mddev->degraded = conf->raid_disks; |
5349 | for (d = 0; d < conf->raid_disks ; d++) | |
5350 | if (conf->disks[d].rdev && | |
5351 | test_bit(In_sync, | |
5352 | &conf->disks[d].rdev->flags)) | |
5353 | mddev->degraded--; | |
5354 | for (d = conf->raid_disks ; | |
5355 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 5356 | d++) { |
3cb03002 | 5357 | struct md_rdev *rdev = conf->disks[d].rdev; |
1a67dde0 | 5358 | if (rdev && raid5_remove_disk(mddev, d) == 0) { |
36fad858 | 5359 | sysfs_unlink_rdev(mddev, rdev); |
1a67dde0 N |
5360 | rdev->raid_disk = -1; |
5361 | } | |
5362 | } | |
cea9c228 | 5363 | } |
88ce4930 | 5364 | mddev->layout = conf->algorithm; |
09c9e5fa | 5365 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5366 | mddev->reshape_position = MaxSector; |
5367 | mddev->delta_disks = 0; | |
cea9c228 N |
5368 | } |
5369 | } | |
5370 | ||
fd01b88c | 5371 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 5372 | { |
d1688a6d | 5373 | struct r5conf *conf = mddev->private; |
72626685 N |
5374 | |
5375 | switch(state) { | |
e464eafd N |
5376 | case 2: /* resume for a suspend */ |
5377 | wake_up(&conf->wait_for_overlap); | |
5378 | break; | |
5379 | ||
72626685 N |
5380 | case 1: /* stop all writes */ |
5381 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5382 | /* '2' tells resync/reshape to pause so that all |
5383 | * active stripes can drain | |
5384 | */ | |
5385 | conf->quiesce = 2; | |
72626685 | 5386 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5387 | atomic_read(&conf->active_stripes) == 0 && |
5388 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5389 | conf->device_lock, /* nothing */); |
64bd660b | 5390 | conf->quiesce = 1; |
72626685 | 5391 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5392 | /* allow reshape to continue */ |
5393 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5394 | break; |
5395 | ||
5396 | case 0: /* re-enable writes */ | |
5397 | spin_lock_irq(&conf->device_lock); | |
5398 | conf->quiesce = 0; | |
5399 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5400 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5401 | spin_unlock_irq(&conf->device_lock); |
5402 | break; | |
5403 | } | |
72626685 | 5404 | } |
b15c2e57 | 5405 | |
d562b0c4 | 5406 | |
fd01b88c | 5407 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 5408 | { |
e373ab10 | 5409 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 5410 | sector_t sectors; |
54071b38 | 5411 | |
f1b29bca | 5412 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 5413 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
5414 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5415 | mdname(mddev)); | |
f1b29bca DW |
5416 | return ERR_PTR(-EINVAL); |
5417 | } | |
5418 | ||
e373ab10 N |
5419 | sectors = raid0_conf->strip_zone[0].zone_end; |
5420 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 5421 | mddev->dev_sectors = sectors; |
f1b29bca | 5422 | mddev->new_level = level; |
54071b38 TM |
5423 | mddev->new_layout = ALGORITHM_PARITY_N; |
5424 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5425 | mddev->raid_disks += 1; | |
5426 | mddev->delta_disks = 1; | |
5427 | /* make sure it will be not marked as dirty */ | |
5428 | mddev->recovery_cp = MaxSector; | |
5429 | ||
5430 | return setup_conf(mddev); | |
5431 | } | |
5432 | ||
5433 | ||
fd01b88c | 5434 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
5435 | { |
5436 | int chunksect; | |
5437 | ||
5438 | if (mddev->raid_disks != 2 || | |
5439 | mddev->degraded > 1) | |
5440 | return ERR_PTR(-EINVAL); | |
5441 | ||
5442 | /* Should check if there are write-behind devices? */ | |
5443 | ||
5444 | chunksect = 64*2; /* 64K by default */ | |
5445 | ||
5446 | /* The array must be an exact multiple of chunksize */ | |
5447 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5448 | chunksect >>= 1; | |
5449 | ||
5450 | if ((chunksect<<9) < STRIPE_SIZE) | |
5451 | /* array size does not allow a suitable chunk size */ | |
5452 | return ERR_PTR(-EINVAL); | |
5453 | ||
5454 | mddev->new_level = 5; | |
5455 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5456 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5457 | |
5458 | return setup_conf(mddev); | |
5459 | } | |
5460 | ||
fd01b88c | 5461 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
5462 | { |
5463 | int new_layout; | |
5464 | ||
5465 | switch (mddev->layout) { | |
5466 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5467 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5468 | break; | |
5469 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5470 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5471 | break; | |
5472 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5473 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5474 | break; | |
5475 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5476 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5477 | break; | |
5478 | case ALGORITHM_PARITY_0_6: | |
5479 | new_layout = ALGORITHM_PARITY_0; | |
5480 | break; | |
5481 | case ALGORITHM_PARITY_N: | |
5482 | new_layout = ALGORITHM_PARITY_N; | |
5483 | break; | |
5484 | default: | |
5485 | return ERR_PTR(-EINVAL); | |
5486 | } | |
5487 | mddev->new_level = 5; | |
5488 | mddev->new_layout = new_layout; | |
5489 | mddev->delta_disks = -1; | |
5490 | mddev->raid_disks -= 1; | |
5491 | return setup_conf(mddev); | |
5492 | } | |
5493 | ||
d562b0c4 | 5494 | |
fd01b88c | 5495 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 5496 | { |
88ce4930 N |
5497 | /* For a 2-drive array, the layout and chunk size can be changed |
5498 | * immediately as not restriping is needed. | |
5499 | * For larger arrays we record the new value - after validation | |
5500 | * to be used by a reshape pass. | |
b3546035 | 5501 | */ |
d1688a6d | 5502 | struct r5conf *conf = mddev->private; |
597a711b | 5503 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5504 | |
597a711b | 5505 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5506 | return -EINVAL; |
5507 | if (new_chunk > 0) { | |
0ba459d2 | 5508 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5509 | return -EINVAL; |
597a711b | 5510 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5511 | return -EINVAL; |
597a711b | 5512 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5513 | /* not factor of array size */ |
5514 | return -EINVAL; | |
5515 | } | |
5516 | ||
5517 | /* They look valid */ | |
5518 | ||
88ce4930 | 5519 | if (mddev->raid_disks == 2) { |
597a711b N |
5520 | /* can make the change immediately */ |
5521 | if (mddev->new_layout >= 0) { | |
5522 | conf->algorithm = mddev->new_layout; | |
5523 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5524 | } |
5525 | if (new_chunk > 0) { | |
597a711b N |
5526 | conf->chunk_sectors = new_chunk ; |
5527 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5528 | } |
5529 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5530 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5531 | } |
50ac168a | 5532 | return check_reshape(mddev); |
88ce4930 N |
5533 | } |
5534 | ||
fd01b88c | 5535 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 5536 | { |
597a711b | 5537 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5538 | |
597a711b | 5539 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5540 | return -EINVAL; |
b3546035 | 5541 | if (new_chunk > 0) { |
0ba459d2 | 5542 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5543 | return -EINVAL; |
597a711b | 5544 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5545 | return -EINVAL; |
597a711b | 5546 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5547 | /* not factor of array size */ |
5548 | return -EINVAL; | |
b3546035 | 5549 | } |
88ce4930 N |
5550 | |
5551 | /* They look valid */ | |
50ac168a | 5552 | return check_reshape(mddev); |
b3546035 N |
5553 | } |
5554 | ||
fd01b88c | 5555 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
5556 | { |
5557 | /* raid5 can take over: | |
f1b29bca | 5558 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
5559 | * raid1 - if there are two drives. We need to know the chunk size |
5560 | * raid4 - trivial - just use a raid4 layout. | |
5561 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 5562 | */ |
f1b29bca DW |
5563 | if (mddev->level == 0) |
5564 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
5565 | if (mddev->level == 1) |
5566 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5567 | if (mddev->level == 4) { |
5568 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5569 | mddev->new_level = 5; | |
5570 | return setup_conf(mddev); | |
5571 | } | |
fc9739c6 N |
5572 | if (mddev->level == 6) |
5573 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5574 | |
5575 | return ERR_PTR(-EINVAL); | |
5576 | } | |
5577 | ||
fd01b88c | 5578 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 5579 | { |
f1b29bca DW |
5580 | /* raid4 can take over: |
5581 | * raid0 - if there is only one strip zone | |
5582 | * raid5 - if layout is right | |
a78d38a1 | 5583 | */ |
f1b29bca DW |
5584 | if (mddev->level == 0) |
5585 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
5586 | if (mddev->level == 5 && |
5587 | mddev->layout == ALGORITHM_PARITY_N) { | |
5588 | mddev->new_layout = 0; | |
5589 | mddev->new_level = 4; | |
5590 | return setup_conf(mddev); | |
5591 | } | |
5592 | return ERR_PTR(-EINVAL); | |
5593 | } | |
d562b0c4 | 5594 | |
84fc4b56 | 5595 | static struct md_personality raid5_personality; |
245f46c2 | 5596 | |
fd01b88c | 5597 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
5598 | { |
5599 | /* Currently can only take over a raid5. We map the | |
5600 | * personality to an equivalent raid6 personality | |
5601 | * with the Q block at the end. | |
5602 | */ | |
5603 | int new_layout; | |
5604 | ||
5605 | if (mddev->pers != &raid5_personality) | |
5606 | return ERR_PTR(-EINVAL); | |
5607 | if (mddev->degraded > 1) | |
5608 | return ERR_PTR(-EINVAL); | |
5609 | if (mddev->raid_disks > 253) | |
5610 | return ERR_PTR(-EINVAL); | |
5611 | if (mddev->raid_disks < 3) | |
5612 | return ERR_PTR(-EINVAL); | |
5613 | ||
5614 | switch (mddev->layout) { | |
5615 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5616 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5617 | break; | |
5618 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5619 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5620 | break; | |
5621 | case ALGORITHM_LEFT_SYMMETRIC: | |
5622 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5623 | break; | |
5624 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5625 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5626 | break; | |
5627 | case ALGORITHM_PARITY_0: | |
5628 | new_layout = ALGORITHM_PARITY_0_6; | |
5629 | break; | |
5630 | case ALGORITHM_PARITY_N: | |
5631 | new_layout = ALGORITHM_PARITY_N; | |
5632 | break; | |
5633 | default: | |
5634 | return ERR_PTR(-EINVAL); | |
5635 | } | |
5636 | mddev->new_level = 6; | |
5637 | mddev->new_layout = new_layout; | |
5638 | mddev->delta_disks = 1; | |
5639 | mddev->raid_disks += 1; | |
5640 | return setup_conf(mddev); | |
5641 | } | |
5642 | ||
5643 | ||
84fc4b56 | 5644 | static struct md_personality raid6_personality = |
16a53ecc N |
5645 | { |
5646 | .name = "raid6", | |
5647 | .level = 6, | |
5648 | .owner = THIS_MODULE, | |
5649 | .make_request = make_request, | |
5650 | .run = run, | |
5651 | .stop = stop, | |
5652 | .status = status, | |
5653 | .error_handler = error, | |
5654 | .hot_add_disk = raid5_add_disk, | |
5655 | .hot_remove_disk= raid5_remove_disk, | |
5656 | .spare_active = raid5_spare_active, | |
5657 | .sync_request = sync_request, | |
5658 | .resize = raid5_resize, | |
80c3a6ce | 5659 | .size = raid5_size, |
50ac168a | 5660 | .check_reshape = raid6_check_reshape, |
f416885e | 5661 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5662 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5663 | .quiesce = raid5_quiesce, |
245f46c2 | 5664 | .takeover = raid6_takeover, |
16a53ecc | 5665 | }; |
84fc4b56 | 5666 | static struct md_personality raid5_personality = |
1da177e4 LT |
5667 | { |
5668 | .name = "raid5", | |
2604b703 | 5669 | .level = 5, |
1da177e4 LT |
5670 | .owner = THIS_MODULE, |
5671 | .make_request = make_request, | |
5672 | .run = run, | |
5673 | .stop = stop, | |
5674 | .status = status, | |
5675 | .error_handler = error, | |
5676 | .hot_add_disk = raid5_add_disk, | |
5677 | .hot_remove_disk= raid5_remove_disk, | |
5678 | .spare_active = raid5_spare_active, | |
5679 | .sync_request = sync_request, | |
5680 | .resize = raid5_resize, | |
80c3a6ce | 5681 | .size = raid5_size, |
63c70c4f N |
5682 | .check_reshape = raid5_check_reshape, |
5683 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5684 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5685 | .quiesce = raid5_quiesce, |
d562b0c4 | 5686 | .takeover = raid5_takeover, |
1da177e4 LT |
5687 | }; |
5688 | ||
84fc4b56 | 5689 | static struct md_personality raid4_personality = |
1da177e4 | 5690 | { |
2604b703 N |
5691 | .name = "raid4", |
5692 | .level = 4, | |
5693 | .owner = THIS_MODULE, | |
5694 | .make_request = make_request, | |
5695 | .run = run, | |
5696 | .stop = stop, | |
5697 | .status = status, | |
5698 | .error_handler = error, | |
5699 | .hot_add_disk = raid5_add_disk, | |
5700 | .hot_remove_disk= raid5_remove_disk, | |
5701 | .spare_active = raid5_spare_active, | |
5702 | .sync_request = sync_request, | |
5703 | .resize = raid5_resize, | |
80c3a6ce | 5704 | .size = raid5_size, |
3d37890b N |
5705 | .check_reshape = raid5_check_reshape, |
5706 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5707 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 5708 | .quiesce = raid5_quiesce, |
a78d38a1 | 5709 | .takeover = raid4_takeover, |
2604b703 N |
5710 | }; |
5711 | ||
5712 | static int __init raid5_init(void) | |
5713 | { | |
16a53ecc | 5714 | register_md_personality(&raid6_personality); |
2604b703 N |
5715 | register_md_personality(&raid5_personality); |
5716 | register_md_personality(&raid4_personality); | |
5717 | return 0; | |
1da177e4 LT |
5718 | } |
5719 | ||
2604b703 | 5720 | static void raid5_exit(void) |
1da177e4 | 5721 | { |
16a53ecc | 5722 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5723 | unregister_md_personality(&raid5_personality); |
5724 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5725 | } |
5726 | ||
5727 | module_init(raid5_init); | |
5728 | module_exit(raid5_exit); | |
5729 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5730 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 5731 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
5732 | MODULE_ALIAS("md-raid5"); |
5733 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5734 | MODULE_ALIAS("md-level-5"); |
5735 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5736 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5737 | MODULE_ALIAS("md-raid6"); | |
5738 | MODULE_ALIAS("md-level-6"); | |
5739 | ||
5740 | /* This used to be two separate modules, they were: */ | |
5741 | MODULE_ALIAS("raid5"); | |
5742 | MODULE_ALIAS("raid6"); |