Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
32 | * new additions. |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
36 | * When we are ready to do a write, if that batch hasn't been written yet, |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
056075c7 | 50 | #include <linux/module.h> |
07a3b417 | 51 | #include <linux/async.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
36d1c647 | 53 | #include <linux/cpu.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
8bda470e | 55 | #include <linux/ratelimit.h> |
43b2e5d8 | 56 | #include "md.h" |
bff61975 | 57 | #include "raid5.h" |
54071b38 | 58 | #include "raid0.h" |
ef740c37 | 59 | #include "bitmap.h" |
72626685 | 60 | |
1da177e4 LT |
61 | /* |
62 | * Stripe cache | |
63 | */ | |
64 | ||
65 | #define NR_STRIPES 256 | |
66 | #define STRIPE_SIZE PAGE_SIZE | |
67 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
68 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
69 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 70 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 71 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
72 | #define HASH_MASK (NR_HASH - 1) |
73 | ||
d1688a6d | 74 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
75 | { |
76 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
77 | return &conf->stripe_hashtbl[hash]; | |
78 | } | |
1da177e4 LT |
79 | |
80 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
81 | * order without overlap. There may be several bio's per stripe+device, and | |
82 | * a bio could span several devices. | |
83 | * When walking this list for a particular stripe+device, we must never proceed | |
84 | * beyond a bio that extends past this device, as the next bio might no longer | |
85 | * be valid. | |
db298e19 | 86 | * This function is used to determine the 'next' bio in the list, given the sector |
1da177e4 LT |
87 | * of the current stripe+device |
88 | */ | |
db298e19 N |
89 | static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector) |
90 | { | |
91 | int sectors = bio->bi_size >> 9; | |
92 | if (bio->bi_sector + sectors < sector + STRIPE_SECTORS) | |
93 | return bio->bi_next; | |
94 | else | |
95 | return NULL; | |
96 | } | |
1da177e4 | 97 | |
960e739d | 98 | /* |
5b99c2ff JA |
99 | * We maintain a biased count of active stripes in the bottom 16 bits of |
100 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
101 | */ |
102 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
103 | { | |
5b99c2ff | 104 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
105 | } |
106 | ||
107 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
108 | { | |
5b99c2ff | 109 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
110 | } |
111 | ||
112 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
113 | { | |
114 | --bio->bi_phys_segments; | |
115 | return raid5_bi_phys_segments(bio); | |
116 | } | |
117 | ||
118 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
119 | { | |
120 | unsigned short val = raid5_bi_hw_segments(bio); | |
121 | ||
122 | --val; | |
5b99c2ff | 123 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
124 | return val; |
125 | } | |
126 | ||
127 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
128 | { | |
9b2dc8b6 | 129 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16); |
960e739d JA |
130 | } |
131 | ||
d0dabf7e N |
132 | /* Find first data disk in a raid6 stripe */ |
133 | static inline int raid6_d0(struct stripe_head *sh) | |
134 | { | |
67cc2b81 N |
135 | if (sh->ddf_layout) |
136 | /* ddf always start from first device */ | |
137 | return 0; | |
138 | /* md starts just after Q block */ | |
d0dabf7e N |
139 | if (sh->qd_idx == sh->disks - 1) |
140 | return 0; | |
141 | else | |
142 | return sh->qd_idx + 1; | |
143 | } | |
16a53ecc N |
144 | static inline int raid6_next_disk(int disk, int raid_disks) |
145 | { | |
146 | disk++; | |
147 | return (disk < raid_disks) ? disk : 0; | |
148 | } | |
a4456856 | 149 | |
d0dabf7e N |
150 | /* When walking through the disks in a raid5, starting at raid6_d0, |
151 | * We need to map each disk to a 'slot', where the data disks are slot | |
152 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
153 | * is raid_disks-1. This help does that mapping. | |
154 | */ | |
67cc2b81 N |
155 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
156 | int *count, int syndrome_disks) | |
d0dabf7e | 157 | { |
6629542e | 158 | int slot = *count; |
67cc2b81 | 159 | |
e4424fee | 160 | if (sh->ddf_layout) |
6629542e | 161 | (*count)++; |
d0dabf7e | 162 | if (idx == sh->pd_idx) |
67cc2b81 | 163 | return syndrome_disks; |
d0dabf7e | 164 | if (idx == sh->qd_idx) |
67cc2b81 | 165 | return syndrome_disks + 1; |
e4424fee | 166 | if (!sh->ddf_layout) |
6629542e | 167 | (*count)++; |
d0dabf7e N |
168 | return slot; |
169 | } | |
170 | ||
a4456856 DW |
171 | static void return_io(struct bio *return_bi) |
172 | { | |
173 | struct bio *bi = return_bi; | |
174 | while (bi) { | |
a4456856 DW |
175 | |
176 | return_bi = bi->bi_next; | |
177 | bi->bi_next = NULL; | |
178 | bi->bi_size = 0; | |
0e13fe23 | 179 | bio_endio(bi, 0); |
a4456856 DW |
180 | bi = return_bi; |
181 | } | |
182 | } | |
183 | ||
d1688a6d | 184 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 185 | |
600aa109 DW |
186 | static int stripe_operations_active(struct stripe_head *sh) |
187 | { | |
188 | return sh->check_state || sh->reconstruct_state || | |
189 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
190 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
191 | } | |
192 | ||
d1688a6d | 193 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 LT |
194 | { |
195 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
196 | BUG_ON(!list_empty(&sh->lru)); |
197 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 198 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
482c0834 | 199 | if (test_bit(STRIPE_DELAYED, &sh->state)) |
1da177e4 | 200 | list_add_tail(&sh->lru, &conf->delayed_list); |
482c0834 N |
201 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
202 | sh->bm_seq - conf->seq_write > 0) | |
72626685 | 203 | list_add_tail(&sh->lru, &conf->bitmap_list); |
482c0834 | 204 | else { |
72626685 | 205 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 206 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 207 | } |
1da177e4 LT |
208 | md_wakeup_thread(conf->mddev->thread); |
209 | } else { | |
600aa109 | 210 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
211 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
212 | atomic_dec(&conf->preread_active_stripes); | |
213 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
214 | md_wakeup_thread(conf->mddev->thread); | |
215 | } | |
1da177e4 | 216 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
217 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
218 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 219 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
220 | if (conf->retry_read_aligned) |
221 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 222 | } |
1da177e4 LT |
223 | } |
224 | } | |
225 | } | |
d0dabf7e | 226 | |
1da177e4 LT |
227 | static void release_stripe(struct stripe_head *sh) |
228 | { | |
d1688a6d | 229 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 230 | unsigned long flags; |
16a53ecc | 231 | |
1da177e4 LT |
232 | spin_lock_irqsave(&conf->device_lock, flags); |
233 | __release_stripe(conf, sh); | |
234 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
235 | } | |
236 | ||
fccddba0 | 237 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 238 | { |
45b4233c DW |
239 | pr_debug("remove_hash(), stripe %llu\n", |
240 | (unsigned long long)sh->sector); | |
1da177e4 | 241 | |
fccddba0 | 242 | hlist_del_init(&sh->hash); |
1da177e4 LT |
243 | } |
244 | ||
d1688a6d | 245 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 246 | { |
fccddba0 | 247 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 248 | |
45b4233c DW |
249 | pr_debug("insert_hash(), stripe %llu\n", |
250 | (unsigned long long)sh->sector); | |
1da177e4 | 251 | |
fccddba0 | 252 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
253 | } |
254 | ||
255 | ||
256 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
d1688a6d | 257 | static struct stripe_head *get_free_stripe(struct r5conf *conf) |
1da177e4 LT |
258 | { |
259 | struct stripe_head *sh = NULL; | |
260 | struct list_head *first; | |
261 | ||
1da177e4 LT |
262 | if (list_empty(&conf->inactive_list)) |
263 | goto out; | |
264 | first = conf->inactive_list.next; | |
265 | sh = list_entry(first, struct stripe_head, lru); | |
266 | list_del_init(first); | |
267 | remove_hash(sh); | |
268 | atomic_inc(&conf->active_stripes); | |
269 | out: | |
270 | return sh; | |
271 | } | |
272 | ||
e4e11e38 | 273 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
274 | { |
275 | struct page *p; | |
276 | int i; | |
e4e11e38 | 277 | int num = sh->raid_conf->pool_size; |
1da177e4 | 278 | |
e4e11e38 | 279 | for (i = 0; i < num ; i++) { |
1da177e4 LT |
280 | p = sh->dev[i].page; |
281 | if (!p) | |
282 | continue; | |
283 | sh->dev[i].page = NULL; | |
2d1f3b5d | 284 | put_page(p); |
1da177e4 LT |
285 | } |
286 | } | |
287 | ||
e4e11e38 | 288 | static int grow_buffers(struct stripe_head *sh) |
1da177e4 LT |
289 | { |
290 | int i; | |
e4e11e38 | 291 | int num = sh->raid_conf->pool_size; |
1da177e4 | 292 | |
e4e11e38 | 293 | for (i = 0; i < num; i++) { |
1da177e4 LT |
294 | struct page *page; |
295 | ||
296 | if (!(page = alloc_page(GFP_KERNEL))) { | |
297 | return 1; | |
298 | } | |
299 | sh->dev[i].page = page; | |
300 | } | |
301 | return 0; | |
302 | } | |
303 | ||
784052ec | 304 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 305 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 306 | struct stripe_head *sh); |
1da177e4 | 307 | |
b5663ba4 | 308 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 309 | { |
d1688a6d | 310 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 311 | int i; |
1da177e4 | 312 | |
78bafebd ES |
313 | BUG_ON(atomic_read(&sh->count) != 0); |
314 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 315 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 316 | |
45b4233c | 317 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
318 | (unsigned long long)sh->sector); |
319 | ||
320 | remove_hash(sh); | |
16a53ecc | 321 | |
86b42c71 | 322 | sh->generation = conf->generation - previous; |
b5663ba4 | 323 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 324 | sh->sector = sector; |
911d4ee8 | 325 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
326 | sh->state = 0; |
327 | ||
7ecaa1e6 N |
328 | |
329 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
330 | struct r5dev *dev = &sh->dev[i]; |
331 | ||
d84e0f10 | 332 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 333 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 334 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 335 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 336 | dev->read, dev->towrite, dev->written, |
1da177e4 | 337 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 338 | WARN_ON(1); |
1da177e4 LT |
339 | } |
340 | dev->flags = 0; | |
784052ec | 341 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
342 | } |
343 | insert_hash(conf, sh); | |
344 | } | |
345 | ||
d1688a6d | 346 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 347 | short generation) |
1da177e4 LT |
348 | { |
349 | struct stripe_head *sh; | |
fccddba0 | 350 | struct hlist_node *hn; |
1da177e4 | 351 | |
45b4233c | 352 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 353 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 354 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 355 | return sh; |
45b4233c | 356 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
357 | return NULL; |
358 | } | |
359 | ||
674806d6 N |
360 | /* |
361 | * Need to check if array has failed when deciding whether to: | |
362 | * - start an array | |
363 | * - remove non-faulty devices | |
364 | * - add a spare | |
365 | * - allow a reshape | |
366 | * This determination is simple when no reshape is happening. | |
367 | * However if there is a reshape, we need to carefully check | |
368 | * both the before and after sections. | |
369 | * This is because some failed devices may only affect one | |
370 | * of the two sections, and some non-in_sync devices may | |
371 | * be insync in the section most affected by failed devices. | |
372 | */ | |
908f4fbd | 373 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 374 | { |
908f4fbd | 375 | int degraded, degraded2; |
674806d6 | 376 | int i; |
674806d6 N |
377 | |
378 | rcu_read_lock(); | |
379 | degraded = 0; | |
380 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 381 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
674806d6 N |
382 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
383 | degraded++; | |
384 | else if (test_bit(In_sync, &rdev->flags)) | |
385 | ; | |
386 | else | |
387 | /* not in-sync or faulty. | |
388 | * If the reshape increases the number of devices, | |
389 | * this is being recovered by the reshape, so | |
390 | * this 'previous' section is not in_sync. | |
391 | * If the number of devices is being reduced however, | |
392 | * the device can only be part of the array if | |
393 | * we are reverting a reshape, so this section will | |
394 | * be in-sync. | |
395 | */ | |
396 | if (conf->raid_disks >= conf->previous_raid_disks) | |
397 | degraded++; | |
398 | } | |
399 | rcu_read_unlock(); | |
908f4fbd N |
400 | if (conf->raid_disks == conf->previous_raid_disks) |
401 | return degraded; | |
674806d6 | 402 | rcu_read_lock(); |
908f4fbd | 403 | degraded2 = 0; |
674806d6 | 404 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 405 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
674806d6 | 406 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 407 | degraded2++; |
674806d6 N |
408 | else if (test_bit(In_sync, &rdev->flags)) |
409 | ; | |
410 | else | |
411 | /* not in-sync or faulty. | |
412 | * If reshape increases the number of devices, this | |
413 | * section has already been recovered, else it | |
414 | * almost certainly hasn't. | |
415 | */ | |
416 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 417 | degraded2++; |
674806d6 N |
418 | } |
419 | rcu_read_unlock(); | |
908f4fbd N |
420 | if (degraded2 > degraded) |
421 | return degraded2; | |
422 | return degraded; | |
423 | } | |
424 | ||
425 | static int has_failed(struct r5conf *conf) | |
426 | { | |
427 | int degraded; | |
428 | ||
429 | if (conf->mddev->reshape_position == MaxSector) | |
430 | return conf->mddev->degraded > conf->max_degraded; | |
431 | ||
432 | degraded = calc_degraded(conf); | |
674806d6 N |
433 | if (degraded > conf->max_degraded) |
434 | return 1; | |
435 | return 0; | |
436 | } | |
437 | ||
b5663ba4 | 438 | static struct stripe_head * |
d1688a6d | 439 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 440 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
441 | { |
442 | struct stripe_head *sh; | |
443 | ||
45b4233c | 444 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
445 | |
446 | spin_lock_irq(&conf->device_lock); | |
447 | ||
448 | do { | |
72626685 | 449 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 450 | conf->quiesce == 0 || noquiesce, |
72626685 | 451 | conf->device_lock, /* nothing */); |
86b42c71 | 452 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
453 | if (!sh) { |
454 | if (!conf->inactive_blocked) | |
455 | sh = get_free_stripe(conf); | |
456 | if (noblock && sh == NULL) | |
457 | break; | |
458 | if (!sh) { | |
459 | conf->inactive_blocked = 1; | |
460 | wait_event_lock_irq(conf->wait_for_stripe, | |
461 | !list_empty(&conf->inactive_list) && | |
5036805b N |
462 | (atomic_read(&conf->active_stripes) |
463 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
464 | || !conf->inactive_blocked), |
465 | conf->device_lock, | |
7c13edc8 | 466 | ); |
1da177e4 LT |
467 | conf->inactive_blocked = 0; |
468 | } else | |
b5663ba4 | 469 | init_stripe(sh, sector, previous); |
1da177e4 LT |
470 | } else { |
471 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
472 | BUG_ON(!list_empty(&sh->lru) |
473 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
474 | } else { |
475 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
476 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
477 | if (list_empty(&sh->lru) && |
478 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
479 | BUG(); |
480 | list_del_init(&sh->lru); | |
1da177e4 LT |
481 | } |
482 | } | |
483 | } while (sh == NULL); | |
484 | ||
485 | if (sh) | |
486 | atomic_inc(&sh->count); | |
487 | ||
488 | spin_unlock_irq(&conf->device_lock); | |
489 | return sh; | |
490 | } | |
491 | ||
6712ecf8 N |
492 | static void |
493 | raid5_end_read_request(struct bio *bi, int error); | |
494 | static void | |
495 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 496 | |
c4e5ac0a | 497 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 498 | { |
d1688a6d | 499 | struct r5conf *conf = sh->raid_conf; |
91c00924 DW |
500 | int i, disks = sh->disks; |
501 | ||
502 | might_sleep(); | |
503 | ||
504 | for (i = disks; i--; ) { | |
505 | int rw; | |
9a3e1101 | 506 | int replace_only = 0; |
977df362 N |
507 | struct bio *bi, *rbi; |
508 | struct md_rdev *rdev, *rrdev = NULL; | |
e9c7469b TH |
509 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
510 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
511 | rw = WRITE_FUA; | |
512 | else | |
513 | rw = WRITE; | |
514 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
91c00924 | 515 | rw = READ; |
9a3e1101 N |
516 | else if (test_and_clear_bit(R5_WantReplace, |
517 | &sh->dev[i].flags)) { | |
518 | rw = WRITE; | |
519 | replace_only = 1; | |
520 | } else | |
91c00924 DW |
521 | continue; |
522 | ||
523 | bi = &sh->dev[i].req; | |
977df362 | 524 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 DW |
525 | |
526 | bi->bi_rw = rw; | |
977df362 N |
527 | rbi->bi_rw = rw; |
528 | if (rw & WRITE) { | |
91c00924 | 529 | bi->bi_end_io = raid5_end_write_request; |
977df362 N |
530 | rbi->bi_end_io = raid5_end_write_request; |
531 | } else | |
91c00924 DW |
532 | bi->bi_end_io = raid5_end_read_request; |
533 | ||
534 | rcu_read_lock(); | |
9a3e1101 | 535 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
536 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
537 | rdev = rcu_dereference(conf->disks[i].rdev); | |
538 | if (!rdev) { | |
539 | rdev = rrdev; | |
540 | rrdev = NULL; | |
541 | } | |
9a3e1101 N |
542 | if (rw & WRITE) { |
543 | if (replace_only) | |
544 | rdev = NULL; | |
dd054fce N |
545 | if (rdev == rrdev) |
546 | /* We raced and saw duplicates */ | |
547 | rrdev = NULL; | |
9a3e1101 | 548 | } else { |
dd054fce | 549 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev) |
9a3e1101 N |
550 | rdev = rrdev; |
551 | rrdev = NULL; | |
552 | } | |
977df362 | 553 | |
91c00924 DW |
554 | if (rdev && test_bit(Faulty, &rdev->flags)) |
555 | rdev = NULL; | |
556 | if (rdev) | |
557 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
558 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
559 | rrdev = NULL; | |
560 | if (rrdev) | |
561 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
562 | rcu_read_unlock(); |
563 | ||
73e92e51 | 564 | /* We have already checked bad blocks for reads. Now |
977df362 N |
565 | * need to check for writes. We never accept write errors |
566 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
567 | */ |
568 | while ((rw & WRITE) && rdev && | |
569 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
570 | sector_t first_bad; | |
571 | int bad_sectors; | |
572 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
573 | &first_bad, &bad_sectors); | |
574 | if (!bad) | |
575 | break; | |
576 | ||
577 | if (bad < 0) { | |
578 | set_bit(BlockedBadBlocks, &rdev->flags); | |
579 | if (!conf->mddev->external && | |
580 | conf->mddev->flags) { | |
581 | /* It is very unlikely, but we might | |
582 | * still need to write out the | |
583 | * bad block log - better give it | |
584 | * a chance*/ | |
585 | md_check_recovery(conf->mddev); | |
586 | } | |
587 | md_wait_for_blocked_rdev(rdev, conf->mddev); | |
588 | } else { | |
589 | /* Acknowledged bad block - skip the write */ | |
590 | rdev_dec_pending(rdev, conf->mddev); | |
591 | rdev = NULL; | |
592 | } | |
593 | } | |
594 | ||
91c00924 | 595 | if (rdev) { |
9a3e1101 N |
596 | if (s->syncing || s->expanding || s->expanded |
597 | || s->replacing) | |
91c00924 DW |
598 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
599 | ||
2b7497f0 DW |
600 | set_bit(STRIPE_IO_STARTED, &sh->state); |
601 | ||
91c00924 DW |
602 | bi->bi_bdev = rdev->bdev; |
603 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 604 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
605 | bi->bi_rw, i); |
606 | atomic_inc(&sh->count); | |
607 | bi->bi_sector = sh->sector + rdev->data_offset; | |
608 | bi->bi_flags = 1 << BIO_UPTODATE; | |
91c00924 | 609 | bi->bi_idx = 0; |
91c00924 DW |
610 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
611 | bi->bi_io_vec[0].bv_offset = 0; | |
612 | bi->bi_size = STRIPE_SIZE; | |
613 | bi->bi_next = NULL; | |
977df362 N |
614 | if (rrdev) |
615 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
91c00924 | 616 | generic_make_request(bi); |
977df362 N |
617 | } |
618 | if (rrdev) { | |
9a3e1101 N |
619 | if (s->syncing || s->expanding || s->expanded |
620 | || s->replacing) | |
977df362 N |
621 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
622 | ||
623 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
624 | ||
625 | rbi->bi_bdev = rrdev->bdev; | |
626 | pr_debug("%s: for %llu schedule op %ld on " | |
627 | "replacement disc %d\n", | |
628 | __func__, (unsigned long long)sh->sector, | |
629 | rbi->bi_rw, i); | |
630 | atomic_inc(&sh->count); | |
631 | rbi->bi_sector = sh->sector + rrdev->data_offset; | |
632 | rbi->bi_flags = 1 << BIO_UPTODATE; | |
633 | rbi->bi_idx = 0; | |
634 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
635 | rbi->bi_io_vec[0].bv_offset = 0; | |
636 | rbi->bi_size = STRIPE_SIZE; | |
637 | rbi->bi_next = NULL; | |
638 | generic_make_request(rbi); | |
639 | } | |
640 | if (!rdev && !rrdev) { | |
b062962e | 641 | if (rw & WRITE) |
91c00924 DW |
642 | set_bit(STRIPE_DEGRADED, &sh->state); |
643 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
644 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
645 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
646 | set_bit(STRIPE_HANDLE, &sh->state); | |
647 | } | |
648 | } | |
649 | } | |
650 | ||
651 | static struct dma_async_tx_descriptor * | |
652 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
653 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
654 | { | |
655 | struct bio_vec *bvl; | |
656 | struct page *bio_page; | |
657 | int i; | |
658 | int page_offset; | |
a08abd8c | 659 | struct async_submit_ctl submit; |
0403e382 | 660 | enum async_tx_flags flags = 0; |
91c00924 DW |
661 | |
662 | if (bio->bi_sector >= sector) | |
663 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
664 | else | |
665 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 666 | |
0403e382 DW |
667 | if (frombio) |
668 | flags |= ASYNC_TX_FENCE; | |
669 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
670 | ||
91c00924 | 671 | bio_for_each_segment(bvl, bio, i) { |
fcde9075 | 672 | int len = bvl->bv_len; |
91c00924 DW |
673 | int clen; |
674 | int b_offset = 0; | |
675 | ||
676 | if (page_offset < 0) { | |
677 | b_offset = -page_offset; | |
678 | page_offset += b_offset; | |
679 | len -= b_offset; | |
680 | } | |
681 | ||
682 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
683 | clen = STRIPE_SIZE - page_offset; | |
684 | else | |
685 | clen = len; | |
686 | ||
687 | if (clen > 0) { | |
fcde9075 NK |
688 | b_offset += bvl->bv_offset; |
689 | bio_page = bvl->bv_page; | |
91c00924 DW |
690 | if (frombio) |
691 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 692 | b_offset, clen, &submit); |
91c00924 DW |
693 | else |
694 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 695 | page_offset, clen, &submit); |
91c00924 | 696 | } |
a08abd8c DW |
697 | /* chain the operations */ |
698 | submit.depend_tx = tx; | |
699 | ||
91c00924 DW |
700 | if (clen < len) /* hit end of page */ |
701 | break; | |
702 | page_offset += len; | |
703 | } | |
704 | ||
705 | return tx; | |
706 | } | |
707 | ||
708 | static void ops_complete_biofill(void *stripe_head_ref) | |
709 | { | |
710 | struct stripe_head *sh = stripe_head_ref; | |
711 | struct bio *return_bi = NULL; | |
d1688a6d | 712 | struct r5conf *conf = sh->raid_conf; |
e4d84909 | 713 | int i; |
91c00924 | 714 | |
e46b272b | 715 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
716 | (unsigned long long)sh->sector); |
717 | ||
718 | /* clear completed biofills */ | |
83de75cc | 719 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
720 | for (i = sh->disks; i--; ) { |
721 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
722 | |
723 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
724 | /* and check if we need to reply to a read request, |
725 | * new R5_Wantfill requests are held off until | |
83de75cc | 726 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
727 | */ |
728 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 729 | struct bio *rbi, *rbi2; |
91c00924 | 730 | |
91c00924 DW |
731 | BUG_ON(!dev->read); |
732 | rbi = dev->read; | |
733 | dev->read = NULL; | |
734 | while (rbi && rbi->bi_sector < | |
735 | dev->sector + STRIPE_SECTORS) { | |
736 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 737 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
738 | rbi->bi_next = return_bi; |
739 | return_bi = rbi; | |
740 | } | |
91c00924 DW |
741 | rbi = rbi2; |
742 | } | |
743 | } | |
744 | } | |
83de75cc DW |
745 | spin_unlock_irq(&conf->device_lock); |
746 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
747 | |
748 | return_io(return_bi); | |
749 | ||
e4d84909 | 750 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
751 | release_stripe(sh); |
752 | } | |
753 | ||
754 | static void ops_run_biofill(struct stripe_head *sh) | |
755 | { | |
756 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 757 | struct r5conf *conf = sh->raid_conf; |
a08abd8c | 758 | struct async_submit_ctl submit; |
91c00924 DW |
759 | int i; |
760 | ||
e46b272b | 761 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
762 | (unsigned long long)sh->sector); |
763 | ||
764 | for (i = sh->disks; i--; ) { | |
765 | struct r5dev *dev = &sh->dev[i]; | |
766 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
767 | struct bio *rbi; | |
768 | spin_lock_irq(&conf->device_lock); | |
769 | dev->read = rbi = dev->toread; | |
770 | dev->toread = NULL; | |
771 | spin_unlock_irq(&conf->device_lock); | |
772 | while (rbi && rbi->bi_sector < | |
773 | dev->sector + STRIPE_SECTORS) { | |
774 | tx = async_copy_data(0, rbi, dev->page, | |
775 | dev->sector, tx); | |
776 | rbi = r5_next_bio(rbi, dev->sector); | |
777 | } | |
778 | } | |
779 | } | |
780 | ||
781 | atomic_inc(&sh->count); | |
a08abd8c DW |
782 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
783 | async_trigger_callback(&submit); | |
91c00924 DW |
784 | } |
785 | ||
4e7d2c0a | 786 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 787 | { |
4e7d2c0a | 788 | struct r5dev *tgt; |
91c00924 | 789 | |
4e7d2c0a DW |
790 | if (target < 0) |
791 | return; | |
91c00924 | 792 | |
4e7d2c0a | 793 | tgt = &sh->dev[target]; |
91c00924 DW |
794 | set_bit(R5_UPTODATE, &tgt->flags); |
795 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
796 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
797 | } |
798 | ||
ac6b53b6 | 799 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
800 | { |
801 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 802 | |
e46b272b | 803 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
804 | (unsigned long long)sh->sector); |
805 | ||
ac6b53b6 | 806 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 807 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 808 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 809 | |
ecc65c9b DW |
810 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
811 | if (sh->check_state == check_state_compute_run) | |
812 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
813 | set_bit(STRIPE_HANDLE, &sh->state); |
814 | release_stripe(sh); | |
815 | } | |
816 | ||
d6f38f31 DW |
817 | /* return a pointer to the address conversion region of the scribble buffer */ |
818 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
819 | struct raid5_percpu *percpu) | |
820 | { | |
821 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
822 | } | |
823 | ||
824 | static struct dma_async_tx_descriptor * | |
825 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 826 | { |
91c00924 | 827 | int disks = sh->disks; |
d6f38f31 | 828 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
829 | int target = sh->ops.target; |
830 | struct r5dev *tgt = &sh->dev[target]; | |
831 | struct page *xor_dest = tgt->page; | |
832 | int count = 0; | |
833 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 834 | struct async_submit_ctl submit; |
91c00924 DW |
835 | int i; |
836 | ||
837 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 838 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
839 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
840 | ||
841 | for (i = disks; i--; ) | |
842 | if (i != target) | |
843 | xor_srcs[count++] = sh->dev[i].page; | |
844 | ||
845 | atomic_inc(&sh->count); | |
846 | ||
0403e382 | 847 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 848 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 849 | if (unlikely(count == 1)) |
a08abd8c | 850 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 851 | else |
a08abd8c | 852 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 853 | |
91c00924 DW |
854 | return tx; |
855 | } | |
856 | ||
ac6b53b6 DW |
857 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
858 | * @srcs - (struct page *) array of size sh->disks | |
859 | * @sh - stripe_head to parse | |
860 | * | |
861 | * Populates srcs in proper layout order for the stripe and returns the | |
862 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
863 | * destination buffer is recorded in srcs[count] and the Q destination | |
864 | * is recorded in srcs[count+1]]. | |
865 | */ | |
866 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
867 | { | |
868 | int disks = sh->disks; | |
869 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
870 | int d0_idx = raid6_d0(sh); | |
871 | int count; | |
872 | int i; | |
873 | ||
874 | for (i = 0; i < disks; i++) | |
5dd33c9a | 875 | srcs[i] = NULL; |
ac6b53b6 DW |
876 | |
877 | count = 0; | |
878 | i = d0_idx; | |
879 | do { | |
880 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
881 | ||
882 | srcs[slot] = sh->dev[i].page; | |
883 | i = raid6_next_disk(i, disks); | |
884 | } while (i != d0_idx); | |
ac6b53b6 | 885 | |
e4424fee | 886 | return syndrome_disks; |
ac6b53b6 DW |
887 | } |
888 | ||
889 | static struct dma_async_tx_descriptor * | |
890 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
891 | { | |
892 | int disks = sh->disks; | |
893 | struct page **blocks = percpu->scribble; | |
894 | int target; | |
895 | int qd_idx = sh->qd_idx; | |
896 | struct dma_async_tx_descriptor *tx; | |
897 | struct async_submit_ctl submit; | |
898 | struct r5dev *tgt; | |
899 | struct page *dest; | |
900 | int i; | |
901 | int count; | |
902 | ||
903 | if (sh->ops.target < 0) | |
904 | target = sh->ops.target2; | |
905 | else if (sh->ops.target2 < 0) | |
906 | target = sh->ops.target; | |
91c00924 | 907 | else |
ac6b53b6 DW |
908 | /* we should only have one valid target */ |
909 | BUG(); | |
910 | BUG_ON(target < 0); | |
911 | pr_debug("%s: stripe %llu block: %d\n", | |
912 | __func__, (unsigned long long)sh->sector, target); | |
913 | ||
914 | tgt = &sh->dev[target]; | |
915 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
916 | dest = tgt->page; | |
917 | ||
918 | atomic_inc(&sh->count); | |
919 | ||
920 | if (target == qd_idx) { | |
921 | count = set_syndrome_sources(blocks, sh); | |
922 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
923 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
924 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
925 | ops_complete_compute, sh, | |
ac6b53b6 DW |
926 | to_addr_conv(sh, percpu)); |
927 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
928 | } else { | |
929 | /* Compute any data- or p-drive using XOR */ | |
930 | count = 0; | |
931 | for (i = disks; i-- ; ) { | |
932 | if (i == target || i == qd_idx) | |
933 | continue; | |
934 | blocks[count++] = sh->dev[i].page; | |
935 | } | |
936 | ||
0403e382 DW |
937 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
938 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
939 | to_addr_conv(sh, percpu)); |
940 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
941 | } | |
91c00924 | 942 | |
91c00924 DW |
943 | return tx; |
944 | } | |
945 | ||
ac6b53b6 DW |
946 | static struct dma_async_tx_descriptor * |
947 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
948 | { | |
949 | int i, count, disks = sh->disks; | |
950 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
951 | int d0_idx = raid6_d0(sh); | |
952 | int faila = -1, failb = -1; | |
953 | int target = sh->ops.target; | |
954 | int target2 = sh->ops.target2; | |
955 | struct r5dev *tgt = &sh->dev[target]; | |
956 | struct r5dev *tgt2 = &sh->dev[target2]; | |
957 | struct dma_async_tx_descriptor *tx; | |
958 | struct page **blocks = percpu->scribble; | |
959 | struct async_submit_ctl submit; | |
960 | ||
961 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
962 | __func__, (unsigned long long)sh->sector, target, target2); | |
963 | BUG_ON(target < 0 || target2 < 0); | |
964 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
965 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
966 | ||
6c910a78 | 967 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
968 | * slot number conversion for 'faila' and 'failb' |
969 | */ | |
970 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 971 | blocks[i] = NULL; |
ac6b53b6 DW |
972 | count = 0; |
973 | i = d0_idx; | |
974 | do { | |
975 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
976 | ||
977 | blocks[slot] = sh->dev[i].page; | |
978 | ||
979 | if (i == target) | |
980 | faila = slot; | |
981 | if (i == target2) | |
982 | failb = slot; | |
983 | i = raid6_next_disk(i, disks); | |
984 | } while (i != d0_idx); | |
ac6b53b6 DW |
985 | |
986 | BUG_ON(faila == failb); | |
987 | if (failb < faila) | |
988 | swap(faila, failb); | |
989 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
990 | __func__, (unsigned long long)sh->sector, faila, failb); | |
991 | ||
992 | atomic_inc(&sh->count); | |
993 | ||
994 | if (failb == syndrome_disks+1) { | |
995 | /* Q disk is one of the missing disks */ | |
996 | if (faila == syndrome_disks) { | |
997 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
998 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
999 | ops_complete_compute, sh, | |
1000 | to_addr_conv(sh, percpu)); | |
e4424fee | 1001 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1002 | STRIPE_SIZE, &submit); |
1003 | } else { | |
1004 | struct page *dest; | |
1005 | int data_target; | |
1006 | int qd_idx = sh->qd_idx; | |
1007 | ||
1008 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1009 | if (target == qd_idx) | |
1010 | data_target = target2; | |
1011 | else | |
1012 | data_target = target; | |
1013 | ||
1014 | count = 0; | |
1015 | for (i = disks; i-- ; ) { | |
1016 | if (i == data_target || i == qd_idx) | |
1017 | continue; | |
1018 | blocks[count++] = sh->dev[i].page; | |
1019 | } | |
1020 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1021 | init_async_submit(&submit, |
1022 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1023 | NULL, NULL, NULL, | |
1024 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1025 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1026 | &submit); | |
1027 | ||
1028 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
1029 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1030 | ops_complete_compute, sh, | |
1031 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
1032 | return async_gen_syndrome(blocks, 0, count+2, |
1033 | STRIPE_SIZE, &submit); | |
1034 | } | |
ac6b53b6 | 1035 | } else { |
6c910a78 DW |
1036 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1037 | ops_complete_compute, sh, | |
1038 | to_addr_conv(sh, percpu)); | |
1039 | if (failb == syndrome_disks) { | |
1040 | /* We're missing D+P. */ | |
1041 | return async_raid6_datap_recov(syndrome_disks+2, | |
1042 | STRIPE_SIZE, faila, | |
1043 | blocks, &submit); | |
1044 | } else { | |
1045 | /* We're missing D+D. */ | |
1046 | return async_raid6_2data_recov(syndrome_disks+2, | |
1047 | STRIPE_SIZE, faila, failb, | |
1048 | blocks, &submit); | |
1049 | } | |
ac6b53b6 DW |
1050 | } |
1051 | } | |
1052 | ||
1053 | ||
91c00924 DW |
1054 | static void ops_complete_prexor(void *stripe_head_ref) |
1055 | { | |
1056 | struct stripe_head *sh = stripe_head_ref; | |
1057 | ||
e46b272b | 1058 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1059 | (unsigned long long)sh->sector); |
91c00924 DW |
1060 | } |
1061 | ||
1062 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
1063 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
1064 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1065 | { |
91c00924 | 1066 | int disks = sh->disks; |
d6f38f31 | 1067 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1068 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1069 | struct async_submit_ctl submit; |
91c00924 DW |
1070 | |
1071 | /* existing parity data subtracted */ | |
1072 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1073 | ||
e46b272b | 1074 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1075 | (unsigned long long)sh->sector); |
1076 | ||
1077 | for (i = disks; i--; ) { | |
1078 | struct r5dev *dev = &sh->dev[i]; | |
1079 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1080 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1081 | xor_srcs[count++] = dev->page; |
1082 | } | |
1083 | ||
0403e382 | 1084 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 1085 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 1086 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1087 | |
1088 | return tx; | |
1089 | } | |
1090 | ||
1091 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1092 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1093 | { |
1094 | int disks = sh->disks; | |
d8ee0728 | 1095 | int i; |
91c00924 | 1096 | |
e46b272b | 1097 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1098 | (unsigned long long)sh->sector); |
1099 | ||
1100 | for (i = disks; i--; ) { | |
1101 | struct r5dev *dev = &sh->dev[i]; | |
1102 | struct bio *chosen; | |
91c00924 | 1103 | |
d8ee0728 | 1104 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1105 | struct bio *wbi; |
1106 | ||
cbe47ec5 | 1107 | spin_lock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1108 | chosen = dev->towrite; |
1109 | dev->towrite = NULL; | |
1110 | BUG_ON(dev->written); | |
1111 | wbi = dev->written = chosen; | |
cbe47ec5 | 1112 | spin_unlock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1113 | |
1114 | while (wbi && wbi->bi_sector < | |
1115 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1116 | if (wbi->bi_rw & REQ_FUA) |
1117 | set_bit(R5_WantFUA, &dev->flags); | |
91c00924 DW |
1118 | tx = async_copy_data(1, wbi, dev->page, |
1119 | dev->sector, tx); | |
1120 | wbi = r5_next_bio(wbi, dev->sector); | |
1121 | } | |
1122 | } | |
1123 | } | |
1124 | ||
1125 | return tx; | |
1126 | } | |
1127 | ||
ac6b53b6 | 1128 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1129 | { |
1130 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1131 | int disks = sh->disks; |
1132 | int pd_idx = sh->pd_idx; | |
1133 | int qd_idx = sh->qd_idx; | |
1134 | int i; | |
e9c7469b | 1135 | bool fua = false; |
91c00924 | 1136 | |
e46b272b | 1137 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1138 | (unsigned long long)sh->sector); |
1139 | ||
e9c7469b TH |
1140 | for (i = disks; i--; ) |
1141 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); | |
1142 | ||
91c00924 DW |
1143 | for (i = disks; i--; ) { |
1144 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1145 | |
e9c7469b | 1146 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1147 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1148 | if (fua) |
1149 | set_bit(R5_WantFUA, &dev->flags); | |
1150 | } | |
91c00924 DW |
1151 | } |
1152 | ||
d8ee0728 DW |
1153 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1154 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1155 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1156 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1157 | else { | |
1158 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1159 | sh->reconstruct_state = reconstruct_state_result; | |
1160 | } | |
91c00924 DW |
1161 | |
1162 | set_bit(STRIPE_HANDLE, &sh->state); | |
1163 | release_stripe(sh); | |
1164 | } | |
1165 | ||
1166 | static void | |
ac6b53b6 DW |
1167 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1168 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1169 | { |
91c00924 | 1170 | int disks = sh->disks; |
d6f38f31 | 1171 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1172 | struct async_submit_ctl submit; |
91c00924 DW |
1173 | int count = 0, pd_idx = sh->pd_idx, i; |
1174 | struct page *xor_dest; | |
d8ee0728 | 1175 | int prexor = 0; |
91c00924 | 1176 | unsigned long flags; |
91c00924 | 1177 | |
e46b272b | 1178 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1179 | (unsigned long long)sh->sector); |
1180 | ||
1181 | /* check if prexor is active which means only process blocks | |
1182 | * that are part of a read-modify-write (written) | |
1183 | */ | |
d8ee0728 DW |
1184 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1185 | prexor = 1; | |
91c00924 DW |
1186 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1187 | for (i = disks; i--; ) { | |
1188 | struct r5dev *dev = &sh->dev[i]; | |
1189 | if (dev->written) | |
1190 | xor_srcs[count++] = dev->page; | |
1191 | } | |
1192 | } else { | |
1193 | xor_dest = sh->dev[pd_idx].page; | |
1194 | for (i = disks; i--; ) { | |
1195 | struct r5dev *dev = &sh->dev[i]; | |
1196 | if (i != pd_idx) | |
1197 | xor_srcs[count++] = dev->page; | |
1198 | } | |
1199 | } | |
1200 | ||
91c00924 DW |
1201 | /* 1/ if we prexor'd then the dest is reused as a source |
1202 | * 2/ if we did not prexor then we are redoing the parity | |
1203 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1204 | * for the synchronous xor case | |
1205 | */ | |
88ba2aa5 | 1206 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1207 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1208 | ||
1209 | atomic_inc(&sh->count); | |
1210 | ||
ac6b53b6 | 1211 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1212 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1213 | if (unlikely(count == 1)) |
1214 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1215 | else | |
1216 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1217 | } |
1218 | ||
ac6b53b6 DW |
1219 | static void |
1220 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1221 | struct dma_async_tx_descriptor *tx) | |
1222 | { | |
1223 | struct async_submit_ctl submit; | |
1224 | struct page **blocks = percpu->scribble; | |
1225 | int count; | |
1226 | ||
1227 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1228 | ||
1229 | count = set_syndrome_sources(blocks, sh); | |
1230 | ||
1231 | atomic_inc(&sh->count); | |
1232 | ||
1233 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1234 | sh, to_addr_conv(sh, percpu)); | |
1235 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1236 | } |
1237 | ||
1238 | static void ops_complete_check(void *stripe_head_ref) | |
1239 | { | |
1240 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1241 | |
e46b272b | 1242 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1243 | (unsigned long long)sh->sector); |
1244 | ||
ecc65c9b | 1245 | sh->check_state = check_state_check_result; |
91c00924 DW |
1246 | set_bit(STRIPE_HANDLE, &sh->state); |
1247 | release_stripe(sh); | |
1248 | } | |
1249 | ||
ac6b53b6 | 1250 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1251 | { |
91c00924 | 1252 | int disks = sh->disks; |
ac6b53b6 DW |
1253 | int pd_idx = sh->pd_idx; |
1254 | int qd_idx = sh->qd_idx; | |
1255 | struct page *xor_dest; | |
d6f38f31 | 1256 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1257 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1258 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1259 | int count; |
1260 | int i; | |
91c00924 | 1261 | |
e46b272b | 1262 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1263 | (unsigned long long)sh->sector); |
1264 | ||
ac6b53b6 DW |
1265 | count = 0; |
1266 | xor_dest = sh->dev[pd_idx].page; | |
1267 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1268 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1269 | if (i == pd_idx || i == qd_idx) |
1270 | continue; | |
1271 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1272 | } |
1273 | ||
d6f38f31 DW |
1274 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1275 | to_addr_conv(sh, percpu)); | |
099f53cb | 1276 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1277 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1278 | |
91c00924 | 1279 | atomic_inc(&sh->count); |
a08abd8c DW |
1280 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1281 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1282 | } |
1283 | ||
ac6b53b6 DW |
1284 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1285 | { | |
1286 | struct page **srcs = percpu->scribble; | |
1287 | struct async_submit_ctl submit; | |
1288 | int count; | |
1289 | ||
1290 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1291 | (unsigned long long)sh->sector, checkp); | |
1292 | ||
1293 | count = set_syndrome_sources(srcs, sh); | |
1294 | if (!checkp) | |
1295 | srcs[count] = NULL; | |
91c00924 | 1296 | |
91c00924 | 1297 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1298 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1299 | sh, to_addr_conv(sh, percpu)); | |
1300 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1301 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1302 | } |
1303 | ||
417b8d4a | 1304 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1305 | { |
1306 | int overlap_clear = 0, i, disks = sh->disks; | |
1307 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1308 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1309 | int level = conf->level; |
d6f38f31 DW |
1310 | struct raid5_percpu *percpu; |
1311 | unsigned long cpu; | |
91c00924 | 1312 | |
d6f38f31 DW |
1313 | cpu = get_cpu(); |
1314 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1315 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1316 | ops_run_biofill(sh); |
1317 | overlap_clear++; | |
1318 | } | |
1319 | ||
7b3a871e | 1320 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1321 | if (level < 6) |
1322 | tx = ops_run_compute5(sh, percpu); | |
1323 | else { | |
1324 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1325 | tx = ops_run_compute6_1(sh, percpu); | |
1326 | else | |
1327 | tx = ops_run_compute6_2(sh, percpu); | |
1328 | } | |
1329 | /* terminate the chain if reconstruct is not set to be run */ | |
1330 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1331 | async_tx_ack(tx); |
1332 | } | |
91c00924 | 1333 | |
600aa109 | 1334 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1335 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1336 | |
600aa109 | 1337 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1338 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1339 | overlap_clear++; |
1340 | } | |
1341 | ||
ac6b53b6 DW |
1342 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1343 | if (level < 6) | |
1344 | ops_run_reconstruct5(sh, percpu, tx); | |
1345 | else | |
1346 | ops_run_reconstruct6(sh, percpu, tx); | |
1347 | } | |
91c00924 | 1348 | |
ac6b53b6 DW |
1349 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1350 | if (sh->check_state == check_state_run) | |
1351 | ops_run_check_p(sh, percpu); | |
1352 | else if (sh->check_state == check_state_run_q) | |
1353 | ops_run_check_pq(sh, percpu, 0); | |
1354 | else if (sh->check_state == check_state_run_pq) | |
1355 | ops_run_check_pq(sh, percpu, 1); | |
1356 | else | |
1357 | BUG(); | |
1358 | } | |
91c00924 | 1359 | |
91c00924 DW |
1360 | if (overlap_clear) |
1361 | for (i = disks; i--; ) { | |
1362 | struct r5dev *dev = &sh->dev[i]; | |
1363 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1364 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1365 | } | |
d6f38f31 | 1366 | put_cpu(); |
91c00924 DW |
1367 | } |
1368 | ||
417b8d4a DW |
1369 | #ifdef CONFIG_MULTICORE_RAID456 |
1370 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1371 | { | |
1372 | struct stripe_head *sh = param; | |
1373 | unsigned long ops_request = sh->ops.request; | |
1374 | ||
1375 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1376 | wake_up(&sh->ops.wait_for_ops); | |
1377 | ||
1378 | __raid_run_ops(sh, ops_request); | |
1379 | release_stripe(sh); | |
1380 | } | |
1381 | ||
1382 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1383 | { | |
1384 | /* since handle_stripe can be called outside of raid5d context | |
1385 | * we need to ensure sh->ops.request is de-staged before another | |
1386 | * request arrives | |
1387 | */ | |
1388 | wait_event(sh->ops.wait_for_ops, | |
1389 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1390 | sh->ops.request = ops_request; | |
1391 | ||
1392 | atomic_inc(&sh->count); | |
1393 | async_schedule(async_run_ops, sh); | |
1394 | } | |
1395 | #else | |
1396 | #define raid_run_ops __raid_run_ops | |
1397 | #endif | |
1398 | ||
d1688a6d | 1399 | static int grow_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1400 | { |
1401 | struct stripe_head *sh; | |
6ce32846 | 1402 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1403 | if (!sh) |
1404 | return 0; | |
6ce32846 | 1405 | |
3f294f4f | 1406 | sh->raid_conf = conf; |
417b8d4a DW |
1407 | #ifdef CONFIG_MULTICORE_RAID456 |
1408 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1409 | #endif | |
3f294f4f | 1410 | |
e4e11e38 N |
1411 | if (grow_buffers(sh)) { |
1412 | shrink_buffers(sh); | |
3f294f4f N |
1413 | kmem_cache_free(conf->slab_cache, sh); |
1414 | return 0; | |
1415 | } | |
1416 | /* we just created an active stripe so... */ | |
1417 | atomic_set(&sh->count, 1); | |
1418 | atomic_inc(&conf->active_stripes); | |
1419 | INIT_LIST_HEAD(&sh->lru); | |
1420 | release_stripe(sh); | |
1421 | return 1; | |
1422 | } | |
1423 | ||
d1688a6d | 1424 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 1425 | { |
e18b890b | 1426 | struct kmem_cache *sc; |
5e5e3e78 | 1427 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1428 | |
f4be6b43 N |
1429 | if (conf->mddev->gendisk) |
1430 | sprintf(conf->cache_name[0], | |
1431 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1432 | else | |
1433 | sprintf(conf->cache_name[0], | |
1434 | "raid%d-%p", conf->level, conf->mddev); | |
1435 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1436 | ||
ad01c9e3 N |
1437 | conf->active_name = 0; |
1438 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1439 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1440 | 0, 0, NULL); |
1da177e4 LT |
1441 | if (!sc) |
1442 | return 1; | |
1443 | conf->slab_cache = sc; | |
ad01c9e3 | 1444 | conf->pool_size = devs; |
16a53ecc | 1445 | while (num--) |
3f294f4f | 1446 | if (!grow_one_stripe(conf)) |
1da177e4 | 1447 | return 1; |
1da177e4 LT |
1448 | return 0; |
1449 | } | |
29269553 | 1450 | |
d6f38f31 DW |
1451 | /** |
1452 | * scribble_len - return the required size of the scribble region | |
1453 | * @num - total number of disks in the array | |
1454 | * | |
1455 | * The size must be enough to contain: | |
1456 | * 1/ a struct page pointer for each device in the array +2 | |
1457 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1458 | * (dma_map_page()) or page (page_address()) address. | |
1459 | * | |
1460 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1461 | * calculate over all devices (not just the data blocks), using zeros in place | |
1462 | * of the P and Q blocks. | |
1463 | */ | |
1464 | static size_t scribble_len(int num) | |
1465 | { | |
1466 | size_t len; | |
1467 | ||
1468 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1469 | ||
1470 | return len; | |
1471 | } | |
1472 | ||
d1688a6d | 1473 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
1474 | { |
1475 | /* Make all the stripes able to hold 'newsize' devices. | |
1476 | * New slots in each stripe get 'page' set to a new page. | |
1477 | * | |
1478 | * This happens in stages: | |
1479 | * 1/ create a new kmem_cache and allocate the required number of | |
1480 | * stripe_heads. | |
1481 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1482 | * to the new stripe_heads. This will have the side effect of | |
1483 | * freezing the array as once all stripe_heads have been collected, | |
1484 | * no IO will be possible. Old stripe heads are freed once their | |
1485 | * pages have been transferred over, and the old kmem_cache is | |
1486 | * freed when all stripes are done. | |
1487 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1488 | * we simple return a failre status - no need to clean anything up. | |
1489 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1490 | * If this fails, we don't bother trying the shrink the | |
1491 | * stripe_heads down again, we just leave them as they are. | |
1492 | * As each stripe_head is processed the new one is released into | |
1493 | * active service. | |
1494 | * | |
1495 | * Once step2 is started, we cannot afford to wait for a write, | |
1496 | * so we use GFP_NOIO allocations. | |
1497 | */ | |
1498 | struct stripe_head *osh, *nsh; | |
1499 | LIST_HEAD(newstripes); | |
1500 | struct disk_info *ndisks; | |
d6f38f31 | 1501 | unsigned long cpu; |
b5470dc5 | 1502 | int err; |
e18b890b | 1503 | struct kmem_cache *sc; |
ad01c9e3 N |
1504 | int i; |
1505 | ||
1506 | if (newsize <= conf->pool_size) | |
1507 | return 0; /* never bother to shrink */ | |
1508 | ||
b5470dc5 DW |
1509 | err = md_allow_write(conf->mddev); |
1510 | if (err) | |
1511 | return err; | |
2a2275d6 | 1512 | |
ad01c9e3 N |
1513 | /* Step 1 */ |
1514 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1515 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1516 | 0, 0, NULL); |
ad01c9e3 N |
1517 | if (!sc) |
1518 | return -ENOMEM; | |
1519 | ||
1520 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1521 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1522 | if (!nsh) |
1523 | break; | |
1524 | ||
ad01c9e3 | 1525 | nsh->raid_conf = conf; |
417b8d4a DW |
1526 | #ifdef CONFIG_MULTICORE_RAID456 |
1527 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1528 | #endif | |
ad01c9e3 N |
1529 | |
1530 | list_add(&nsh->lru, &newstripes); | |
1531 | } | |
1532 | if (i) { | |
1533 | /* didn't get enough, give up */ | |
1534 | while (!list_empty(&newstripes)) { | |
1535 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1536 | list_del(&nsh->lru); | |
1537 | kmem_cache_free(sc, nsh); | |
1538 | } | |
1539 | kmem_cache_destroy(sc); | |
1540 | return -ENOMEM; | |
1541 | } | |
1542 | /* Step 2 - Must use GFP_NOIO now. | |
1543 | * OK, we have enough stripes, start collecting inactive | |
1544 | * stripes and copying them over | |
1545 | */ | |
1546 | list_for_each_entry(nsh, &newstripes, lru) { | |
1547 | spin_lock_irq(&conf->device_lock); | |
1548 | wait_event_lock_irq(conf->wait_for_stripe, | |
1549 | !list_empty(&conf->inactive_list), | |
1550 | conf->device_lock, | |
482c0834 | 1551 | ); |
ad01c9e3 N |
1552 | osh = get_free_stripe(conf); |
1553 | spin_unlock_irq(&conf->device_lock); | |
1554 | atomic_set(&nsh->count, 1); | |
1555 | for(i=0; i<conf->pool_size; i++) | |
1556 | nsh->dev[i].page = osh->dev[i].page; | |
1557 | for( ; i<newsize; i++) | |
1558 | nsh->dev[i].page = NULL; | |
1559 | kmem_cache_free(conf->slab_cache, osh); | |
1560 | } | |
1561 | kmem_cache_destroy(conf->slab_cache); | |
1562 | ||
1563 | /* Step 3. | |
1564 | * At this point, we are holding all the stripes so the array | |
1565 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1566 | * conf->disks and the scribble region |
ad01c9e3 N |
1567 | */ |
1568 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1569 | if (ndisks) { | |
1570 | for (i=0; i<conf->raid_disks; i++) | |
1571 | ndisks[i] = conf->disks[i]; | |
1572 | kfree(conf->disks); | |
1573 | conf->disks = ndisks; | |
1574 | } else | |
1575 | err = -ENOMEM; | |
1576 | ||
d6f38f31 DW |
1577 | get_online_cpus(); |
1578 | conf->scribble_len = scribble_len(newsize); | |
1579 | for_each_present_cpu(cpu) { | |
1580 | struct raid5_percpu *percpu; | |
1581 | void *scribble; | |
1582 | ||
1583 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1584 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1585 | ||
1586 | if (scribble) { | |
1587 | kfree(percpu->scribble); | |
1588 | percpu->scribble = scribble; | |
1589 | } else { | |
1590 | err = -ENOMEM; | |
1591 | break; | |
1592 | } | |
1593 | } | |
1594 | put_online_cpus(); | |
1595 | ||
ad01c9e3 N |
1596 | /* Step 4, return new stripes to service */ |
1597 | while(!list_empty(&newstripes)) { | |
1598 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1599 | list_del_init(&nsh->lru); | |
d6f38f31 | 1600 | |
ad01c9e3 N |
1601 | for (i=conf->raid_disks; i < newsize; i++) |
1602 | if (nsh->dev[i].page == NULL) { | |
1603 | struct page *p = alloc_page(GFP_NOIO); | |
1604 | nsh->dev[i].page = p; | |
1605 | if (!p) | |
1606 | err = -ENOMEM; | |
1607 | } | |
1608 | release_stripe(nsh); | |
1609 | } | |
1610 | /* critical section pass, GFP_NOIO no longer needed */ | |
1611 | ||
1612 | conf->slab_cache = sc; | |
1613 | conf->active_name = 1-conf->active_name; | |
1614 | conf->pool_size = newsize; | |
1615 | return err; | |
1616 | } | |
1da177e4 | 1617 | |
d1688a6d | 1618 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
1619 | { |
1620 | struct stripe_head *sh; | |
1621 | ||
3f294f4f N |
1622 | spin_lock_irq(&conf->device_lock); |
1623 | sh = get_free_stripe(conf); | |
1624 | spin_unlock_irq(&conf->device_lock); | |
1625 | if (!sh) | |
1626 | return 0; | |
78bafebd | 1627 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1628 | shrink_buffers(sh); |
3f294f4f N |
1629 | kmem_cache_free(conf->slab_cache, sh); |
1630 | atomic_dec(&conf->active_stripes); | |
1631 | return 1; | |
1632 | } | |
1633 | ||
d1688a6d | 1634 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f N |
1635 | { |
1636 | while (drop_one_stripe(conf)) | |
1637 | ; | |
1638 | ||
29fc7e3e N |
1639 | if (conf->slab_cache) |
1640 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1641 | conf->slab_cache = NULL; |
1642 | } | |
1643 | ||
6712ecf8 | 1644 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1645 | { |
99c0fb5f | 1646 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1647 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1648 | int disks = sh->disks, i; |
1da177e4 | 1649 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 1650 | char b[BDEVNAME_SIZE]; |
dd054fce | 1651 | struct md_rdev *rdev = NULL; |
1da177e4 | 1652 | |
1da177e4 LT |
1653 | |
1654 | for (i=0 ; i<disks; i++) | |
1655 | if (bi == &sh->dev[i].req) | |
1656 | break; | |
1657 | ||
45b4233c DW |
1658 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1659 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1660 | uptodate); |
1661 | if (i == disks) { | |
1662 | BUG(); | |
6712ecf8 | 1663 | return; |
1da177e4 | 1664 | } |
14a75d3e | 1665 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
1666 | /* If replacement finished while this request was outstanding, |
1667 | * 'replacement' might be NULL already. | |
1668 | * In that case it moved down to 'rdev'. | |
1669 | * rdev is not removed until all requests are finished. | |
1670 | */ | |
14a75d3e | 1671 | rdev = conf->disks[i].replacement; |
dd054fce | 1672 | if (!rdev) |
14a75d3e | 1673 | rdev = conf->disks[i].rdev; |
1da177e4 LT |
1674 | |
1675 | if (uptodate) { | |
1da177e4 | 1676 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1677 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
1678 | /* Note that this cannot happen on a |
1679 | * replacement device. We just fail those on | |
1680 | * any error | |
1681 | */ | |
8bda470e CD |
1682 | printk_ratelimited( |
1683 | KERN_INFO | |
1684 | "md/raid:%s: read error corrected" | |
1685 | " (%lu sectors at %llu on %s)\n", | |
1686 | mdname(conf->mddev), STRIPE_SECTORS, | |
1687 | (unsigned long long)(sh->sector | |
1688 | + rdev->data_offset), | |
1689 | bdevname(rdev->bdev, b)); | |
ddd5115f | 1690 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
1691 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1692 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1693 | } | |
14a75d3e N |
1694 | if (atomic_read(&rdev->read_errors)) |
1695 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 1696 | } else { |
14a75d3e | 1697 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 1698 | int retry = 0; |
d6950432 | 1699 | |
1da177e4 | 1700 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1701 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
1702 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
1703 | printk_ratelimited( | |
1704 | KERN_WARNING | |
1705 | "md/raid:%s: read error on replacement device " | |
1706 | "(sector %llu on %s).\n", | |
1707 | mdname(conf->mddev), | |
1708 | (unsigned long long)(sh->sector | |
1709 | + rdev->data_offset), | |
1710 | bdn); | |
1711 | else if (conf->mddev->degraded >= conf->max_degraded) | |
8bda470e CD |
1712 | printk_ratelimited( |
1713 | KERN_WARNING | |
1714 | "md/raid:%s: read error not correctable " | |
1715 | "(sector %llu on %s).\n", | |
1716 | mdname(conf->mddev), | |
1717 | (unsigned long long)(sh->sector | |
1718 | + rdev->data_offset), | |
1719 | bdn); | |
ba22dcbf | 1720 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1721 | /* Oh, no!!! */ |
8bda470e CD |
1722 | printk_ratelimited( |
1723 | KERN_WARNING | |
1724 | "md/raid:%s: read error NOT corrected!! " | |
1725 | "(sector %llu on %s).\n", | |
1726 | mdname(conf->mddev), | |
1727 | (unsigned long long)(sh->sector | |
1728 | + rdev->data_offset), | |
1729 | bdn); | |
d6950432 | 1730 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1731 | > conf->max_nr_stripes) |
14f8d26b | 1732 | printk(KERN_WARNING |
0c55e022 | 1733 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1734 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1735 | else |
1736 | retry = 1; | |
1737 | if (retry) | |
1738 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1739 | else { | |
4e5314b5 N |
1740 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1741 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1742 | md_error(conf->mddev, rdev); |
ba22dcbf | 1743 | } |
1da177e4 | 1744 | } |
14a75d3e | 1745 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
1746 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1747 | set_bit(STRIPE_HANDLE, &sh->state); | |
1748 | release_stripe(sh); | |
1da177e4 LT |
1749 | } |
1750 | ||
d710e138 | 1751 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1752 | { |
99c0fb5f | 1753 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1754 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1755 | int disks = sh->disks, i; |
977df362 | 1756 | struct md_rdev *uninitialized_var(rdev); |
1da177e4 | 1757 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
1758 | sector_t first_bad; |
1759 | int bad_sectors; | |
977df362 | 1760 | int replacement = 0; |
1da177e4 | 1761 | |
977df362 N |
1762 | for (i = 0 ; i < disks; i++) { |
1763 | if (bi == &sh->dev[i].req) { | |
1764 | rdev = conf->disks[i].rdev; | |
1da177e4 | 1765 | break; |
977df362 N |
1766 | } |
1767 | if (bi == &sh->dev[i].rreq) { | |
1768 | rdev = conf->disks[i].replacement; | |
dd054fce N |
1769 | if (rdev) |
1770 | replacement = 1; | |
1771 | else | |
1772 | /* rdev was removed and 'replacement' | |
1773 | * replaced it. rdev is not removed | |
1774 | * until all requests are finished. | |
1775 | */ | |
1776 | rdev = conf->disks[i].rdev; | |
977df362 N |
1777 | break; |
1778 | } | |
1779 | } | |
45b4233c | 1780 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1781 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1782 | uptodate); | |
1783 | if (i == disks) { | |
1784 | BUG(); | |
6712ecf8 | 1785 | return; |
1da177e4 LT |
1786 | } |
1787 | ||
977df362 N |
1788 | if (replacement) { |
1789 | if (!uptodate) | |
1790 | md_error(conf->mddev, rdev); | |
1791 | else if (is_badblock(rdev, sh->sector, | |
1792 | STRIPE_SECTORS, | |
1793 | &first_bad, &bad_sectors)) | |
1794 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
1795 | } else { | |
1796 | if (!uptodate) { | |
1797 | set_bit(WriteErrorSeen, &rdev->flags); | |
1798 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
1799 | } else if (is_badblock(rdev, sh->sector, | |
1800 | STRIPE_SECTORS, | |
1801 | &first_bad, &bad_sectors)) | |
1802 | set_bit(R5_MadeGood, &sh->dev[i].flags); | |
1803 | } | |
1804 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 1805 | |
977df362 N |
1806 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
1807 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 1808 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 1809 | release_stripe(sh); |
1da177e4 LT |
1810 | } |
1811 | ||
784052ec | 1812 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1813 | |
784052ec | 1814 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1815 | { |
1816 | struct r5dev *dev = &sh->dev[i]; | |
1817 | ||
1818 | bio_init(&dev->req); | |
1819 | dev->req.bi_io_vec = &dev->vec; | |
1820 | dev->req.bi_vcnt++; | |
1821 | dev->req.bi_max_vecs++; | |
1da177e4 | 1822 | dev->req.bi_private = sh; |
995c4275 | 1823 | dev->vec.bv_page = dev->page; |
1da177e4 | 1824 | |
977df362 N |
1825 | bio_init(&dev->rreq); |
1826 | dev->rreq.bi_io_vec = &dev->rvec; | |
1827 | dev->rreq.bi_vcnt++; | |
1828 | dev->rreq.bi_max_vecs++; | |
1829 | dev->rreq.bi_private = sh; | |
1830 | dev->rvec.bv_page = dev->page; | |
1831 | ||
1da177e4 | 1832 | dev->flags = 0; |
784052ec | 1833 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1834 | } |
1835 | ||
fd01b88c | 1836 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1837 | { |
1838 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 1839 | struct r5conf *conf = mddev->private; |
908f4fbd | 1840 | unsigned long flags; |
0c55e022 | 1841 | pr_debug("raid456: error called\n"); |
1da177e4 | 1842 | |
908f4fbd N |
1843 | spin_lock_irqsave(&conf->device_lock, flags); |
1844 | clear_bit(In_sync, &rdev->flags); | |
1845 | mddev->degraded = calc_degraded(conf); | |
1846 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1847 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1848 | ||
de393cde | 1849 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
1850 | set_bit(Faulty, &rdev->flags); |
1851 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1852 | printk(KERN_ALERT | |
1853 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1854 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1855 | mdname(mddev), | |
1856 | bdevname(rdev->bdev, b), | |
1857 | mdname(mddev), | |
1858 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1859 | } |
1da177e4 LT |
1860 | |
1861 | /* | |
1862 | * Input: a 'big' sector number, | |
1863 | * Output: index of the data and parity disk, and the sector # in them. | |
1864 | */ | |
d1688a6d | 1865 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
1866 | int previous, int *dd_idx, |
1867 | struct stripe_head *sh) | |
1da177e4 | 1868 | { |
6e3b96ed | 1869 | sector_t stripe, stripe2; |
35f2a591 | 1870 | sector_t chunk_number; |
1da177e4 | 1871 | unsigned int chunk_offset; |
911d4ee8 | 1872 | int pd_idx, qd_idx; |
67cc2b81 | 1873 | int ddf_layout = 0; |
1da177e4 | 1874 | sector_t new_sector; |
e183eaed N |
1875 | int algorithm = previous ? conf->prev_algo |
1876 | : conf->algorithm; | |
09c9e5fa AN |
1877 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1878 | : conf->chunk_sectors; | |
112bf897 N |
1879 | int raid_disks = previous ? conf->previous_raid_disks |
1880 | : conf->raid_disks; | |
1881 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1882 | |
1883 | /* First compute the information on this sector */ | |
1884 | ||
1885 | /* | |
1886 | * Compute the chunk number and the sector offset inside the chunk | |
1887 | */ | |
1888 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1889 | chunk_number = r_sector; | |
1da177e4 LT |
1890 | |
1891 | /* | |
1892 | * Compute the stripe number | |
1893 | */ | |
35f2a591 N |
1894 | stripe = chunk_number; |
1895 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1896 | stripe2 = stripe; |
1da177e4 LT |
1897 | /* |
1898 | * Select the parity disk based on the user selected algorithm. | |
1899 | */ | |
84789554 | 1900 | pd_idx = qd_idx = -1; |
16a53ecc N |
1901 | switch(conf->level) { |
1902 | case 4: | |
911d4ee8 | 1903 | pd_idx = data_disks; |
16a53ecc N |
1904 | break; |
1905 | case 5: | |
e183eaed | 1906 | switch (algorithm) { |
1da177e4 | 1907 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1908 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1909 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1910 | (*dd_idx)++; |
1911 | break; | |
1912 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1913 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1914 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1915 | (*dd_idx)++; |
1916 | break; | |
1917 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1918 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1919 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1920 | break; |
1921 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1922 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1923 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 1924 | break; |
99c0fb5f N |
1925 | case ALGORITHM_PARITY_0: |
1926 | pd_idx = 0; | |
1927 | (*dd_idx)++; | |
1928 | break; | |
1929 | case ALGORITHM_PARITY_N: | |
1930 | pd_idx = data_disks; | |
1931 | break; | |
1da177e4 | 1932 | default: |
99c0fb5f | 1933 | BUG(); |
16a53ecc N |
1934 | } |
1935 | break; | |
1936 | case 6: | |
1937 | ||
e183eaed | 1938 | switch (algorithm) { |
16a53ecc | 1939 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1940 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1941 | qd_idx = pd_idx + 1; |
1942 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1943 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1944 | qd_idx = 0; |
1945 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1946 | (*dd_idx) += 2; /* D D P Q D */ |
1947 | break; | |
1948 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1949 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1950 | qd_idx = pd_idx + 1; |
1951 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1952 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1953 | qd_idx = 0; |
1954 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1955 | (*dd_idx) += 2; /* D D P Q D */ |
1956 | break; | |
1957 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1958 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1959 | qd_idx = (pd_idx + 1) % raid_disks; |
1960 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1961 | break; |
1962 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1963 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1964 | qd_idx = (pd_idx + 1) % raid_disks; |
1965 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1966 | break; |
99c0fb5f N |
1967 | |
1968 | case ALGORITHM_PARITY_0: | |
1969 | pd_idx = 0; | |
1970 | qd_idx = 1; | |
1971 | (*dd_idx) += 2; | |
1972 | break; | |
1973 | case ALGORITHM_PARITY_N: | |
1974 | pd_idx = data_disks; | |
1975 | qd_idx = data_disks + 1; | |
1976 | break; | |
1977 | ||
1978 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1979 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1980 | * of blocks for computing Q is different. | |
1981 | */ | |
6e3b96ed | 1982 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
1983 | qd_idx = pd_idx + 1; |
1984 | if (pd_idx == raid_disks-1) { | |
1985 | (*dd_idx)++; /* Q D D D P */ | |
1986 | qd_idx = 0; | |
1987 | } else if (*dd_idx >= pd_idx) | |
1988 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1989 | ddf_layout = 1; |
99c0fb5f N |
1990 | break; |
1991 | ||
1992 | case ALGORITHM_ROTATING_N_RESTART: | |
1993 | /* Same a left_asymmetric, by first stripe is | |
1994 | * D D D P Q rather than | |
1995 | * Q D D D P | |
1996 | */ | |
6e3b96ed N |
1997 | stripe2 += 1; |
1998 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
1999 | qd_idx = pd_idx + 1; |
2000 | if (pd_idx == raid_disks-1) { | |
2001 | (*dd_idx)++; /* Q D D D P */ | |
2002 | qd_idx = 0; | |
2003 | } else if (*dd_idx >= pd_idx) | |
2004 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2005 | ddf_layout = 1; |
99c0fb5f N |
2006 | break; |
2007 | ||
2008 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2009 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2010 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2011 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2012 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2013 | ddf_layout = 1; |
99c0fb5f N |
2014 | break; |
2015 | ||
2016 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2017 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2018 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2019 | if (*dd_idx >= pd_idx) |
2020 | (*dd_idx)++; | |
2021 | qd_idx = raid_disks - 1; | |
2022 | break; | |
2023 | ||
2024 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2025 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2026 | if (*dd_idx >= pd_idx) |
2027 | (*dd_idx)++; | |
2028 | qd_idx = raid_disks - 1; | |
2029 | break; | |
2030 | ||
2031 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2032 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2033 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2034 | qd_idx = raid_disks - 1; | |
2035 | break; | |
2036 | ||
2037 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2038 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2039 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2040 | qd_idx = raid_disks - 1; | |
2041 | break; | |
2042 | ||
2043 | case ALGORITHM_PARITY_0_6: | |
2044 | pd_idx = 0; | |
2045 | (*dd_idx)++; | |
2046 | qd_idx = raid_disks - 1; | |
2047 | break; | |
2048 | ||
16a53ecc | 2049 | default: |
99c0fb5f | 2050 | BUG(); |
16a53ecc N |
2051 | } |
2052 | break; | |
1da177e4 LT |
2053 | } |
2054 | ||
911d4ee8 N |
2055 | if (sh) { |
2056 | sh->pd_idx = pd_idx; | |
2057 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2058 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2059 | } |
1da177e4 LT |
2060 | /* |
2061 | * Finally, compute the new sector number | |
2062 | */ | |
2063 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2064 | return new_sector; | |
2065 | } | |
2066 | ||
2067 | ||
784052ec | 2068 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2069 | { |
d1688a6d | 2070 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2071 | int raid_disks = sh->disks; |
2072 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2073 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2074 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2075 | : conf->chunk_sectors; | |
e183eaed N |
2076 | int algorithm = previous ? conf->prev_algo |
2077 | : conf->algorithm; | |
1da177e4 LT |
2078 | sector_t stripe; |
2079 | int chunk_offset; | |
35f2a591 N |
2080 | sector_t chunk_number; |
2081 | int dummy1, dd_idx = i; | |
1da177e4 | 2082 | sector_t r_sector; |
911d4ee8 | 2083 | struct stripe_head sh2; |
1da177e4 | 2084 | |
16a53ecc | 2085 | |
1da177e4 LT |
2086 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
2087 | stripe = new_sector; | |
1da177e4 | 2088 | |
16a53ecc N |
2089 | if (i == sh->pd_idx) |
2090 | return 0; | |
2091 | switch(conf->level) { | |
2092 | case 4: break; | |
2093 | case 5: | |
e183eaed | 2094 | switch (algorithm) { |
1da177e4 LT |
2095 | case ALGORITHM_LEFT_ASYMMETRIC: |
2096 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2097 | if (i > sh->pd_idx) | |
2098 | i--; | |
2099 | break; | |
2100 | case ALGORITHM_LEFT_SYMMETRIC: | |
2101 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2102 | if (i < sh->pd_idx) | |
2103 | i += raid_disks; | |
2104 | i -= (sh->pd_idx + 1); | |
2105 | break; | |
99c0fb5f N |
2106 | case ALGORITHM_PARITY_0: |
2107 | i -= 1; | |
2108 | break; | |
2109 | case ALGORITHM_PARITY_N: | |
2110 | break; | |
1da177e4 | 2111 | default: |
99c0fb5f | 2112 | BUG(); |
16a53ecc N |
2113 | } |
2114 | break; | |
2115 | case 6: | |
d0dabf7e | 2116 | if (i == sh->qd_idx) |
16a53ecc | 2117 | return 0; /* It is the Q disk */ |
e183eaed | 2118 | switch (algorithm) { |
16a53ecc N |
2119 | case ALGORITHM_LEFT_ASYMMETRIC: |
2120 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2121 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2122 | case ALGORITHM_ROTATING_N_RESTART: | |
2123 | if (sh->pd_idx == raid_disks-1) | |
2124 | i--; /* Q D D D P */ | |
16a53ecc N |
2125 | else if (i > sh->pd_idx) |
2126 | i -= 2; /* D D P Q D */ | |
2127 | break; | |
2128 | case ALGORITHM_LEFT_SYMMETRIC: | |
2129 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2130 | if (sh->pd_idx == raid_disks-1) | |
2131 | i--; /* Q D D D P */ | |
2132 | else { | |
2133 | /* D D P Q D */ | |
2134 | if (i < sh->pd_idx) | |
2135 | i += raid_disks; | |
2136 | i -= (sh->pd_idx + 2); | |
2137 | } | |
2138 | break; | |
99c0fb5f N |
2139 | case ALGORITHM_PARITY_0: |
2140 | i -= 2; | |
2141 | break; | |
2142 | case ALGORITHM_PARITY_N: | |
2143 | break; | |
2144 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2145 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2146 | if (sh->pd_idx == 0) |
2147 | i--; /* P D D D Q */ | |
e4424fee N |
2148 | else { |
2149 | /* D D Q P D */ | |
2150 | if (i < sh->pd_idx) | |
2151 | i += raid_disks; | |
2152 | i -= (sh->pd_idx + 1); | |
2153 | } | |
99c0fb5f N |
2154 | break; |
2155 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2156 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2157 | if (i > sh->pd_idx) | |
2158 | i--; | |
2159 | break; | |
2160 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2161 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2162 | if (i < sh->pd_idx) | |
2163 | i += data_disks + 1; | |
2164 | i -= (sh->pd_idx + 1); | |
2165 | break; | |
2166 | case ALGORITHM_PARITY_0_6: | |
2167 | i -= 1; | |
2168 | break; | |
16a53ecc | 2169 | default: |
99c0fb5f | 2170 | BUG(); |
16a53ecc N |
2171 | } |
2172 | break; | |
1da177e4 LT |
2173 | } |
2174 | ||
2175 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2176 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2177 | |
112bf897 | 2178 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2179 | previous, &dummy1, &sh2); |
911d4ee8 N |
2180 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2181 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2182 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2183 | mdname(conf->mddev)); | |
1da177e4 LT |
2184 | return 0; |
2185 | } | |
2186 | return r_sector; | |
2187 | } | |
2188 | ||
2189 | ||
600aa109 | 2190 | static void |
c0f7bddb | 2191 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2192 | int rcw, int expand) |
e33129d8 DW |
2193 | { |
2194 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
d1688a6d | 2195 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2196 | int level = conf->level; |
e33129d8 DW |
2197 | |
2198 | if (rcw) { | |
2199 | /* if we are not expanding this is a proper write request, and | |
2200 | * there will be bios with new data to be drained into the | |
2201 | * stripe cache | |
2202 | */ | |
2203 | if (!expand) { | |
600aa109 DW |
2204 | sh->reconstruct_state = reconstruct_state_drain_run; |
2205 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2206 | } else | |
2207 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2208 | |
ac6b53b6 | 2209 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2210 | |
2211 | for (i = disks; i--; ) { | |
2212 | struct r5dev *dev = &sh->dev[i]; | |
2213 | ||
2214 | if (dev->towrite) { | |
2215 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2216 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2217 | if (!expand) |
2218 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2219 | s->locked++; |
e33129d8 DW |
2220 | } |
2221 | } | |
c0f7bddb | 2222 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2223 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2224 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2225 | } else { |
c0f7bddb | 2226 | BUG_ON(level == 6); |
e33129d8 DW |
2227 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2228 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2229 | ||
d8ee0728 | 2230 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2231 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2232 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2233 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2234 | |
2235 | for (i = disks; i--; ) { | |
2236 | struct r5dev *dev = &sh->dev[i]; | |
2237 | if (i == pd_idx) | |
2238 | continue; | |
2239 | ||
e33129d8 DW |
2240 | if (dev->towrite && |
2241 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2242 | test_bit(R5_Wantcompute, &dev->flags))) { |
2243 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2244 | set_bit(R5_LOCKED, &dev->flags); |
2245 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2246 | s->locked++; |
e33129d8 DW |
2247 | } |
2248 | } | |
2249 | } | |
2250 | ||
c0f7bddb | 2251 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2252 | * are in flight |
2253 | */ | |
2254 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2255 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2256 | s->locked++; |
e33129d8 | 2257 | |
c0f7bddb YT |
2258 | if (level == 6) { |
2259 | int qd_idx = sh->qd_idx; | |
2260 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2261 | ||
2262 | set_bit(R5_LOCKED, &dev->flags); | |
2263 | clear_bit(R5_UPTODATE, &dev->flags); | |
2264 | s->locked++; | |
2265 | } | |
2266 | ||
600aa109 | 2267 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2268 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2269 | s->locked, s->ops_request); |
e33129d8 | 2270 | } |
16a53ecc | 2271 | |
1da177e4 LT |
2272 | /* |
2273 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2274 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2275 | * The bi_next chain must be in order. |
2276 | */ | |
2277 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2278 | { | |
2279 | struct bio **bip; | |
d1688a6d | 2280 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2281 | int firstwrite=0; |
1da177e4 | 2282 | |
cbe47ec5 | 2283 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2284 | (unsigned long long)bi->bi_sector, |
2285 | (unsigned long long)sh->sector); | |
2286 | ||
2287 | ||
1da177e4 | 2288 | spin_lock_irq(&conf->device_lock); |
72626685 | 2289 | if (forwrite) { |
1da177e4 | 2290 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2291 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2292 | firstwrite = 1; | |
2293 | } else | |
1da177e4 LT |
2294 | bip = &sh->dev[dd_idx].toread; |
2295 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2296 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2297 | goto overlap; | |
2298 | bip = & (*bip)->bi_next; | |
2299 | } | |
2300 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2301 | goto overlap; | |
2302 | ||
78bafebd | 2303 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2304 | if (*bip) |
2305 | bi->bi_next = *bip; | |
2306 | *bip = bi; | |
960e739d | 2307 | bi->bi_phys_segments++; |
72626685 | 2308 | |
1da177e4 LT |
2309 | if (forwrite) { |
2310 | /* check if page is covered */ | |
2311 | sector_t sector = sh->dev[dd_idx].sector; | |
2312 | for (bi=sh->dev[dd_idx].towrite; | |
2313 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2314 | bi && bi->bi_sector <= sector; | |
2315 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2316 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2317 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2318 | } | |
2319 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2320 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2321 | } | |
cbe47ec5 | 2322 | spin_unlock_irq(&conf->device_lock); |
cbe47ec5 N |
2323 | |
2324 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2325 | (unsigned long long)(*bip)->bi_sector, | |
2326 | (unsigned long long)sh->sector, dd_idx); | |
2327 | ||
2328 | if (conf->mddev->bitmap && firstwrite) { | |
2329 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2330 | STRIPE_SECTORS, 0); | |
2331 | sh->bm_seq = conf->seq_flush+1; | |
2332 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2333 | } | |
1da177e4 LT |
2334 | return 1; |
2335 | ||
2336 | overlap: | |
2337 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2338 | spin_unlock_irq(&conf->device_lock); | |
1da177e4 LT |
2339 | return 0; |
2340 | } | |
2341 | ||
d1688a6d | 2342 | static void end_reshape(struct r5conf *conf); |
29269553 | 2343 | |
d1688a6d | 2344 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 2345 | struct stripe_head *sh) |
ccfcc3c1 | 2346 | { |
784052ec | 2347 | int sectors_per_chunk = |
09c9e5fa | 2348 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2349 | int dd_idx; |
2d2063ce | 2350 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2351 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2352 | |
112bf897 N |
2353 | raid5_compute_sector(conf, |
2354 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2355 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2356 | previous, |
911d4ee8 | 2357 | &dd_idx, sh); |
ccfcc3c1 N |
2358 | } |
2359 | ||
a4456856 | 2360 | static void |
d1688a6d | 2361 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2362 | struct stripe_head_state *s, int disks, |
2363 | struct bio **return_bi) | |
2364 | { | |
2365 | int i; | |
2366 | for (i = disks; i--; ) { | |
2367 | struct bio *bi; | |
2368 | int bitmap_end = 0; | |
2369 | ||
2370 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 2371 | struct md_rdev *rdev; |
a4456856 DW |
2372 | rcu_read_lock(); |
2373 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2374 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
2375 | atomic_inc(&rdev->nr_pending); |
2376 | else | |
2377 | rdev = NULL; | |
a4456856 | 2378 | rcu_read_unlock(); |
7f0da59b N |
2379 | if (rdev) { |
2380 | if (!rdev_set_badblocks( | |
2381 | rdev, | |
2382 | sh->sector, | |
2383 | STRIPE_SECTORS, 0)) | |
2384 | md_error(conf->mddev, rdev); | |
2385 | rdev_dec_pending(rdev, conf->mddev); | |
2386 | } | |
a4456856 DW |
2387 | } |
2388 | spin_lock_irq(&conf->device_lock); | |
2389 | /* fail all writes first */ | |
2390 | bi = sh->dev[i].towrite; | |
2391 | sh->dev[i].towrite = NULL; | |
2392 | if (bi) { | |
2393 | s->to_write--; | |
2394 | bitmap_end = 1; | |
2395 | } | |
2396 | ||
2397 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2398 | wake_up(&conf->wait_for_overlap); | |
2399 | ||
2400 | while (bi && bi->bi_sector < | |
2401 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2402 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2403 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2404 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2405 | md_write_end(conf->mddev); |
2406 | bi->bi_next = *return_bi; | |
2407 | *return_bi = bi; | |
2408 | } | |
2409 | bi = nextbi; | |
2410 | } | |
2411 | /* and fail all 'written' */ | |
2412 | bi = sh->dev[i].written; | |
2413 | sh->dev[i].written = NULL; | |
2414 | if (bi) bitmap_end = 1; | |
2415 | while (bi && bi->bi_sector < | |
2416 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2417 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2418 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2419 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2420 | md_write_end(conf->mddev); |
2421 | bi->bi_next = *return_bi; | |
2422 | *return_bi = bi; | |
2423 | } | |
2424 | bi = bi2; | |
2425 | } | |
2426 | ||
b5e98d65 DW |
2427 | /* fail any reads if this device is non-operational and |
2428 | * the data has not reached the cache yet. | |
2429 | */ | |
2430 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2431 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2432 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2433 | bi = sh->dev[i].toread; |
2434 | sh->dev[i].toread = NULL; | |
2435 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2436 | wake_up(&conf->wait_for_overlap); | |
2437 | if (bi) s->to_read--; | |
2438 | while (bi && bi->bi_sector < | |
2439 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2440 | struct bio *nextbi = | |
2441 | r5_next_bio(bi, sh->dev[i].sector); | |
2442 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2443 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2444 | bi->bi_next = *return_bi; |
2445 | *return_bi = bi; | |
2446 | } | |
2447 | bi = nextbi; | |
2448 | } | |
2449 | } | |
2450 | spin_unlock_irq(&conf->device_lock); | |
2451 | if (bitmap_end) | |
2452 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2453 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
2454 | /* If we were in the middle of a write the parity block might |
2455 | * still be locked - so just clear all R5_LOCKED flags | |
2456 | */ | |
2457 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
2458 | } |
2459 | ||
8b3e6cdc DW |
2460 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2461 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2462 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2463 | } |
2464 | ||
7f0da59b | 2465 | static void |
d1688a6d | 2466 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
2467 | struct stripe_head_state *s) |
2468 | { | |
2469 | int abort = 0; | |
2470 | int i; | |
2471 | ||
2472 | md_done_sync(conf->mddev, STRIPE_SECTORS, 0); | |
2473 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2474 | s->syncing = 0; | |
9a3e1101 | 2475 | s->replacing = 0; |
7f0da59b | 2476 | /* There is nothing more to do for sync/check/repair. |
9a3e1101 | 2477 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
2478 | * non-sync devices, or abort the recovery |
2479 | */ | |
2480 | if (!test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) | |
2481 | return; | |
2482 | /* During recovery devices cannot be removed, so locking and | |
2483 | * refcounting of rdevs is not needed | |
2484 | */ | |
2485 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 2486 | struct md_rdev *rdev = conf->disks[i].rdev; |
9a3e1101 N |
2487 | if (rdev |
2488 | && !test_bit(Faulty, &rdev->flags) | |
2489 | && !test_bit(In_sync, &rdev->flags) | |
2490 | && !rdev_set_badblocks(rdev, sh->sector, | |
2491 | STRIPE_SECTORS, 0)) | |
2492 | abort = 1; | |
2493 | rdev = conf->disks[i].replacement; | |
2494 | if (rdev | |
2495 | && !test_bit(Faulty, &rdev->flags) | |
2496 | && !test_bit(In_sync, &rdev->flags) | |
2497 | && !rdev_set_badblocks(rdev, sh->sector, | |
2498 | STRIPE_SECTORS, 0)) | |
7f0da59b N |
2499 | abort = 1; |
2500 | } | |
2501 | if (abort) { | |
2502 | conf->recovery_disabled = conf->mddev->recovery_disabled; | |
2503 | set_bit(MD_RECOVERY_INTR, &conf->mddev->recovery); | |
2504 | } | |
2505 | } | |
2506 | ||
9a3e1101 N |
2507 | static int want_replace(struct stripe_head *sh, int disk_idx) |
2508 | { | |
2509 | struct md_rdev *rdev; | |
2510 | int rv = 0; | |
2511 | /* Doing recovery so rcu locking not required */ | |
2512 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
2513 | if (rdev | |
2514 | && !test_bit(Faulty, &rdev->flags) | |
2515 | && !test_bit(In_sync, &rdev->flags) | |
2516 | && (rdev->recovery_offset <= sh->sector | |
2517 | || rdev->mddev->recovery_cp <= sh->sector)) | |
2518 | rv = 1; | |
2519 | ||
2520 | return rv; | |
2521 | } | |
2522 | ||
93b3dbce | 2523 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
2524 | * to be read or computed to satisfy a request. |
2525 | * | |
2526 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 2527 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 2528 | */ |
93b3dbce N |
2529 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
2530 | int disk_idx, int disks) | |
a4456856 | 2531 | { |
5599becc | 2532 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2533 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2534 | &sh->dev[s->failed_num[1]] }; | |
5599becc | 2535 | |
93b3dbce | 2536 | /* is the data in this block needed, and can we get it? */ |
5599becc YT |
2537 | if (!test_bit(R5_LOCKED, &dev->flags) && |
2538 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2539 | (dev->toread || | |
2540 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2541 | s->syncing || s->expanding || | |
9a3e1101 | 2542 | (s->replacing && want_replace(sh, disk_idx)) || |
5d35e09c N |
2543 | (s->failed >= 1 && fdev[0]->toread) || |
2544 | (s->failed >= 2 && fdev[1]->toread) || | |
93b3dbce N |
2545 | (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite && |
2546 | !test_bit(R5_OVERWRITE, &fdev[0]->flags)) || | |
2547 | (sh->raid_conf->level == 6 && s->failed && s->to_write))) { | |
5599becc YT |
2548 | /* we would like to get this block, possibly by computing it, |
2549 | * otherwise read it if the backing disk is insync | |
2550 | */ | |
2551 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2552 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2553 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
2554 | (s->failed && (disk_idx == s->failed_num[0] || |
2555 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
2556 | /* have disk failed, and we're requested to fetch it; |
2557 | * do compute it | |
a4456856 | 2558 | */ |
5599becc YT |
2559 | pr_debug("Computing stripe %llu block %d\n", |
2560 | (unsigned long long)sh->sector, disk_idx); | |
2561 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2562 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2563 | set_bit(R5_Wantcompute, &dev->flags); | |
2564 | sh->ops.target = disk_idx; | |
2565 | sh->ops.target2 = -1; /* no 2nd target */ | |
2566 | s->req_compute = 1; | |
93b3dbce N |
2567 | /* Careful: from this point on 'uptodate' is in the eye |
2568 | * of raid_run_ops which services 'compute' operations | |
2569 | * before writes. R5_Wantcompute flags a block that will | |
2570 | * be R5_UPTODATE by the time it is needed for a | |
2571 | * subsequent operation. | |
2572 | */ | |
5599becc YT |
2573 | s->uptodate++; |
2574 | return 1; | |
2575 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2576 | /* Computing 2-failure is *very* expensive; only | |
2577 | * do it if failed >= 2 | |
2578 | */ | |
2579 | int other; | |
2580 | for (other = disks; other--; ) { | |
2581 | if (other == disk_idx) | |
2582 | continue; | |
2583 | if (!test_bit(R5_UPTODATE, | |
2584 | &sh->dev[other].flags)) | |
2585 | break; | |
a4456856 | 2586 | } |
5599becc YT |
2587 | BUG_ON(other < 0); |
2588 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2589 | (unsigned long long)sh->sector, | |
2590 | disk_idx, other); | |
2591 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2592 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2593 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2594 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2595 | sh->ops.target = disk_idx; | |
2596 | sh->ops.target2 = other; | |
2597 | s->uptodate += 2; | |
2598 | s->req_compute = 1; | |
2599 | return 1; | |
2600 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2601 | set_bit(R5_LOCKED, &dev->flags); | |
2602 | set_bit(R5_Wantread, &dev->flags); | |
2603 | s->locked++; | |
2604 | pr_debug("Reading block %d (sync=%d)\n", | |
2605 | disk_idx, s->syncing); | |
a4456856 DW |
2606 | } |
2607 | } | |
5599becc YT |
2608 | |
2609 | return 0; | |
2610 | } | |
2611 | ||
2612 | /** | |
93b3dbce | 2613 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 2614 | */ |
93b3dbce N |
2615 | static void handle_stripe_fill(struct stripe_head *sh, |
2616 | struct stripe_head_state *s, | |
2617 | int disks) | |
5599becc YT |
2618 | { |
2619 | int i; | |
2620 | ||
2621 | /* look for blocks to read/compute, skip this if a compute | |
2622 | * is already in flight, or if the stripe contents are in the | |
2623 | * midst of changing due to a write | |
2624 | */ | |
2625 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2626 | !sh->reconstruct_state) | |
2627 | for (i = disks; i--; ) | |
93b3dbce | 2628 | if (fetch_block(sh, s, i, disks)) |
5599becc | 2629 | break; |
a4456856 DW |
2630 | set_bit(STRIPE_HANDLE, &sh->state); |
2631 | } | |
2632 | ||
2633 | ||
1fe797e6 | 2634 | /* handle_stripe_clean_event |
a4456856 DW |
2635 | * any written block on an uptodate or failed drive can be returned. |
2636 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2637 | * never LOCKED, so we don't need to test 'failed' directly. | |
2638 | */ | |
d1688a6d | 2639 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
2640 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2641 | { | |
2642 | int i; | |
2643 | struct r5dev *dev; | |
2644 | ||
2645 | for (i = disks; i--; ) | |
2646 | if (sh->dev[i].written) { | |
2647 | dev = &sh->dev[i]; | |
2648 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2649 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2650 | /* We can return any write requests */ | |
2651 | struct bio *wbi, *wbi2; | |
2652 | int bitmap_end = 0; | |
45b4233c | 2653 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2654 | spin_lock_irq(&conf->device_lock); |
2655 | wbi = dev->written; | |
2656 | dev->written = NULL; | |
2657 | while (wbi && wbi->bi_sector < | |
2658 | dev->sector + STRIPE_SECTORS) { | |
2659 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2660 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2661 | md_write_end(conf->mddev); |
2662 | wbi->bi_next = *return_bi; | |
2663 | *return_bi = wbi; | |
2664 | } | |
2665 | wbi = wbi2; | |
2666 | } | |
2667 | if (dev->towrite == NULL) | |
2668 | bitmap_end = 1; | |
2669 | spin_unlock_irq(&conf->device_lock); | |
2670 | if (bitmap_end) | |
2671 | bitmap_endwrite(conf->mddev->bitmap, | |
2672 | sh->sector, | |
2673 | STRIPE_SECTORS, | |
2674 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2675 | 0); | |
2676 | } | |
2677 | } | |
8b3e6cdc DW |
2678 | |
2679 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2680 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2681 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2682 | } |
2683 | ||
d1688a6d | 2684 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
2685 | struct stripe_head *sh, |
2686 | struct stripe_head_state *s, | |
2687 | int disks) | |
a4456856 DW |
2688 | { |
2689 | int rmw = 0, rcw = 0, i; | |
c8ac1803 N |
2690 | if (conf->max_degraded == 2) { |
2691 | /* RAID6 requires 'rcw' in current implementation | |
2692 | * Calculate the real rcw later - for now fake it | |
2693 | * look like rcw is cheaper | |
2694 | */ | |
2695 | rcw = 1; rmw = 2; | |
2696 | } else for (i = disks; i--; ) { | |
a4456856 DW |
2697 | /* would I have to read this buffer for read_modify_write */ |
2698 | struct r5dev *dev = &sh->dev[i]; | |
2699 | if ((dev->towrite || i == sh->pd_idx) && | |
2700 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2701 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2702 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2703 | if (test_bit(R5_Insync, &dev->flags)) |
2704 | rmw++; | |
2705 | else | |
2706 | rmw += 2*disks; /* cannot read it */ | |
2707 | } | |
2708 | /* Would I have to read this buffer for reconstruct_write */ | |
2709 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2710 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2711 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2712 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2713 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2714 | else |
2715 | rcw += 2*disks; | |
2716 | } | |
2717 | } | |
45b4233c | 2718 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2719 | (unsigned long long)sh->sector, rmw, rcw); |
2720 | set_bit(STRIPE_HANDLE, &sh->state); | |
2721 | if (rmw < rcw && rmw > 0) | |
2722 | /* prefer read-modify-write, but need to get some data */ | |
2723 | for (i = disks; i--; ) { | |
2724 | struct r5dev *dev = &sh->dev[i]; | |
2725 | if ((dev->towrite || i == sh->pd_idx) && | |
2726 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2727 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2728 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2729 | test_bit(R5_Insync, &dev->flags)) { |
2730 | if ( | |
2731 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2732 | pr_debug("Read_old block " |
a4456856 DW |
2733 | "%d for r-m-w\n", i); |
2734 | set_bit(R5_LOCKED, &dev->flags); | |
2735 | set_bit(R5_Wantread, &dev->flags); | |
2736 | s->locked++; | |
2737 | } else { | |
2738 | set_bit(STRIPE_DELAYED, &sh->state); | |
2739 | set_bit(STRIPE_HANDLE, &sh->state); | |
2740 | } | |
2741 | } | |
2742 | } | |
c8ac1803 | 2743 | if (rcw <= rmw && rcw > 0) { |
a4456856 | 2744 | /* want reconstruct write, but need to get some data */ |
c8ac1803 | 2745 | rcw = 0; |
a4456856 DW |
2746 | for (i = disks; i--; ) { |
2747 | struct r5dev *dev = &sh->dev[i]; | |
2748 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 2749 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 2750 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 2751 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
2752 | test_bit(R5_Wantcompute, &dev->flags))) { |
2753 | rcw++; | |
2754 | if (!test_bit(R5_Insync, &dev->flags)) | |
2755 | continue; /* it's a failed drive */ | |
a4456856 DW |
2756 | if ( |
2757 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2758 | pr_debug("Read_old block " |
a4456856 DW |
2759 | "%d for Reconstruct\n", i); |
2760 | set_bit(R5_LOCKED, &dev->flags); | |
2761 | set_bit(R5_Wantread, &dev->flags); | |
2762 | s->locked++; | |
2763 | } else { | |
2764 | set_bit(STRIPE_DELAYED, &sh->state); | |
2765 | set_bit(STRIPE_HANDLE, &sh->state); | |
2766 | } | |
2767 | } | |
2768 | } | |
c8ac1803 | 2769 | } |
a4456856 DW |
2770 | /* now if nothing is locked, and if we have enough data, |
2771 | * we can start a write request | |
2772 | */ | |
f38e1219 DW |
2773 | /* since handle_stripe can be called at any time we need to handle the |
2774 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2775 | * subsequent call wants to start a write request. raid_run_ops only |
2776 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2777 | * simultaneously. If this is not the case then new writes need to be |
2778 | * held off until the compute completes. | |
2779 | */ | |
976ea8d4 DW |
2780 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2781 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2782 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2783 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2784 | } |
2785 | ||
d1688a6d | 2786 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2787 | struct stripe_head_state *s, int disks) |
2788 | { | |
ecc65c9b | 2789 | struct r5dev *dev = NULL; |
bd2ab670 | 2790 | |
a4456856 | 2791 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2792 | |
ecc65c9b DW |
2793 | switch (sh->check_state) { |
2794 | case check_state_idle: | |
2795 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2796 | if (s->failed == 0) { |
bd2ab670 | 2797 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2798 | sh->check_state = check_state_run; |
2799 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2800 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2801 | s->uptodate--; |
ecc65c9b | 2802 | break; |
bd2ab670 | 2803 | } |
f2b3b44d | 2804 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
2805 | /* fall through */ |
2806 | case check_state_compute_result: | |
2807 | sh->check_state = check_state_idle; | |
2808 | if (!dev) | |
2809 | dev = &sh->dev[sh->pd_idx]; | |
2810 | ||
2811 | /* check that a write has not made the stripe insync */ | |
2812 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2813 | break; | |
c8894419 | 2814 | |
a4456856 | 2815 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2816 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2817 | BUG_ON(s->uptodate != disks); | |
2818 | ||
2819 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2820 | s->locked++; |
a4456856 | 2821 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2822 | |
a4456856 | 2823 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2824 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2825 | break; |
2826 | case check_state_run: | |
2827 | break; /* we will be called again upon completion */ | |
2828 | case check_state_check_result: | |
2829 | sh->check_state = check_state_idle; | |
2830 | ||
2831 | /* if a failure occurred during the check operation, leave | |
2832 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2833 | */ | |
2834 | if (s->failed) | |
2835 | break; | |
2836 | ||
2837 | /* handle a successful check operation, if parity is correct | |
2838 | * we are done. Otherwise update the mismatch count and repair | |
2839 | * parity if !MD_RECOVERY_CHECK | |
2840 | */ | |
ad283ea4 | 2841 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2842 | /* parity is correct (on disc, |
2843 | * not in buffer any more) | |
2844 | */ | |
2845 | set_bit(STRIPE_INSYNC, &sh->state); | |
2846 | else { | |
2847 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2848 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2849 | /* don't try to repair!! */ | |
2850 | set_bit(STRIPE_INSYNC, &sh->state); | |
2851 | else { | |
2852 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2853 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2854 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2855 | set_bit(R5_Wantcompute, | |
2856 | &sh->dev[sh->pd_idx].flags); | |
2857 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2858 | sh->ops.target2 = -1; |
ecc65c9b DW |
2859 | s->uptodate++; |
2860 | } | |
2861 | } | |
2862 | break; | |
2863 | case check_state_compute_run: | |
2864 | break; | |
2865 | default: | |
2866 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2867 | __func__, sh->check_state, | |
2868 | (unsigned long long) sh->sector); | |
2869 | BUG(); | |
a4456856 DW |
2870 | } |
2871 | } | |
2872 | ||
2873 | ||
d1688a6d | 2874 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 2875 | struct stripe_head_state *s, |
f2b3b44d | 2876 | int disks) |
a4456856 | 2877 | { |
a4456856 | 2878 | int pd_idx = sh->pd_idx; |
34e04e87 | 2879 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2880 | struct r5dev *dev; |
a4456856 DW |
2881 | |
2882 | set_bit(STRIPE_HANDLE, &sh->state); | |
2883 | ||
2884 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2885 | |
a4456856 DW |
2886 | /* Want to check and possibly repair P and Q. |
2887 | * However there could be one 'failed' device, in which | |
2888 | * case we can only check one of them, possibly using the | |
2889 | * other to generate missing data | |
2890 | */ | |
2891 | ||
d82dfee0 DW |
2892 | switch (sh->check_state) { |
2893 | case check_state_idle: | |
2894 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 2895 | if (s->failed == s->q_failed) { |
d82dfee0 | 2896 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2897 | * makes sense to check P (If anything else were failed, |
2898 | * we would have used P to recreate it). | |
2899 | */ | |
d82dfee0 | 2900 | sh->check_state = check_state_run; |
a4456856 | 2901 | } |
f2b3b44d | 2902 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 2903 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2904 | * anything, so it makes sense to check it |
2905 | */ | |
d82dfee0 DW |
2906 | if (sh->check_state == check_state_run) |
2907 | sh->check_state = check_state_run_pq; | |
2908 | else | |
2909 | sh->check_state = check_state_run_q; | |
a4456856 | 2910 | } |
a4456856 | 2911 | |
d82dfee0 DW |
2912 | /* discard potentially stale zero_sum_result */ |
2913 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2914 | |
d82dfee0 DW |
2915 | if (sh->check_state == check_state_run) { |
2916 | /* async_xor_zero_sum destroys the contents of P */ | |
2917 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2918 | s->uptodate--; | |
a4456856 | 2919 | } |
d82dfee0 DW |
2920 | if (sh->check_state >= check_state_run && |
2921 | sh->check_state <= check_state_run_pq) { | |
2922 | /* async_syndrome_zero_sum preserves P and Q, so | |
2923 | * no need to mark them !uptodate here | |
2924 | */ | |
2925 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2926 | break; | |
a4456856 DW |
2927 | } |
2928 | ||
d82dfee0 DW |
2929 | /* we have 2-disk failure */ |
2930 | BUG_ON(s->failed != 2); | |
2931 | /* fall through */ | |
2932 | case check_state_compute_result: | |
2933 | sh->check_state = check_state_idle; | |
a4456856 | 2934 | |
d82dfee0 DW |
2935 | /* check that a write has not made the stripe insync */ |
2936 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2937 | break; | |
a4456856 DW |
2938 | |
2939 | /* now write out any block on a failed drive, | |
d82dfee0 | 2940 | * or P or Q if they were recomputed |
a4456856 | 2941 | */ |
d82dfee0 | 2942 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 2943 | if (s->failed == 2) { |
f2b3b44d | 2944 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
2945 | s->locked++; |
2946 | set_bit(R5_LOCKED, &dev->flags); | |
2947 | set_bit(R5_Wantwrite, &dev->flags); | |
2948 | } | |
2949 | if (s->failed >= 1) { | |
f2b3b44d | 2950 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
2951 | s->locked++; |
2952 | set_bit(R5_LOCKED, &dev->flags); | |
2953 | set_bit(R5_Wantwrite, &dev->flags); | |
2954 | } | |
d82dfee0 | 2955 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2956 | dev = &sh->dev[pd_idx]; |
2957 | s->locked++; | |
2958 | set_bit(R5_LOCKED, &dev->flags); | |
2959 | set_bit(R5_Wantwrite, &dev->flags); | |
2960 | } | |
d82dfee0 | 2961 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2962 | dev = &sh->dev[qd_idx]; |
2963 | s->locked++; | |
2964 | set_bit(R5_LOCKED, &dev->flags); | |
2965 | set_bit(R5_Wantwrite, &dev->flags); | |
2966 | } | |
2967 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2968 | ||
2969 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2970 | break; |
2971 | case check_state_run: | |
2972 | case check_state_run_q: | |
2973 | case check_state_run_pq: | |
2974 | break; /* we will be called again upon completion */ | |
2975 | case check_state_check_result: | |
2976 | sh->check_state = check_state_idle; | |
2977 | ||
2978 | /* handle a successful check operation, if parity is correct | |
2979 | * we are done. Otherwise update the mismatch count and repair | |
2980 | * parity if !MD_RECOVERY_CHECK | |
2981 | */ | |
2982 | if (sh->ops.zero_sum_result == 0) { | |
2983 | /* both parities are correct */ | |
2984 | if (!s->failed) | |
2985 | set_bit(STRIPE_INSYNC, &sh->state); | |
2986 | else { | |
2987 | /* in contrast to the raid5 case we can validate | |
2988 | * parity, but still have a failure to write | |
2989 | * back | |
2990 | */ | |
2991 | sh->check_state = check_state_compute_result; | |
2992 | /* Returning at this point means that we may go | |
2993 | * off and bring p and/or q uptodate again so | |
2994 | * we make sure to check zero_sum_result again | |
2995 | * to verify if p or q need writeback | |
2996 | */ | |
2997 | } | |
2998 | } else { | |
2999 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
3000 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
3001 | /* don't try to repair!! */ | |
3002 | set_bit(STRIPE_INSYNC, &sh->state); | |
3003 | else { | |
3004 | int *target = &sh->ops.target; | |
3005 | ||
3006 | sh->ops.target = -1; | |
3007 | sh->ops.target2 = -1; | |
3008 | sh->check_state = check_state_compute_run; | |
3009 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3010 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3011 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3012 | set_bit(R5_Wantcompute, | |
3013 | &sh->dev[pd_idx].flags); | |
3014 | *target = pd_idx; | |
3015 | target = &sh->ops.target2; | |
3016 | s->uptodate++; | |
3017 | } | |
3018 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3019 | set_bit(R5_Wantcompute, | |
3020 | &sh->dev[qd_idx].flags); | |
3021 | *target = qd_idx; | |
3022 | s->uptodate++; | |
3023 | } | |
3024 | } | |
3025 | } | |
3026 | break; | |
3027 | case check_state_compute_run: | |
3028 | break; | |
3029 | default: | |
3030 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3031 | __func__, sh->check_state, | |
3032 | (unsigned long long) sh->sector); | |
3033 | BUG(); | |
a4456856 DW |
3034 | } |
3035 | } | |
3036 | ||
d1688a6d | 3037 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3038 | { |
3039 | int i; | |
3040 | ||
3041 | /* We have read all the blocks in this stripe and now we need to | |
3042 | * copy some of them into a target stripe for expand. | |
3043 | */ | |
f0a50d37 | 3044 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
3045 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3046 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3047 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3048 | int dd_idx, j; |
a4456856 | 3049 | struct stripe_head *sh2; |
a08abd8c | 3050 | struct async_submit_ctl submit; |
a4456856 | 3051 | |
784052ec | 3052 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3053 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3054 | &dd_idx, NULL); | |
a8c906ca | 3055 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3056 | if (sh2 == NULL) |
3057 | /* so far only the early blocks of this stripe | |
3058 | * have been requested. When later blocks | |
3059 | * get requested, we will try again | |
3060 | */ | |
3061 | continue; | |
3062 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3063 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3064 | /* must have already done this block */ | |
3065 | release_stripe(sh2); | |
3066 | continue; | |
3067 | } | |
f0a50d37 DW |
3068 | |
3069 | /* place all the copies on one channel */ | |
a08abd8c | 3070 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3071 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3072 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3073 | &submit); |
f0a50d37 | 3074 | |
a4456856 DW |
3075 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3076 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3077 | for (j = 0; j < conf->raid_disks; j++) | |
3078 | if (j != sh2->pd_idx && | |
86c374ba | 3079 | j != sh2->qd_idx && |
a4456856 DW |
3080 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3081 | break; | |
3082 | if (j == conf->raid_disks) { | |
3083 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3084 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3085 | } | |
3086 | release_stripe(sh2); | |
f0a50d37 | 3087 | |
a4456856 | 3088 | } |
a2e08551 N |
3089 | /* done submitting copies, wait for them to complete */ |
3090 | if (tx) { | |
3091 | async_tx_ack(tx); | |
3092 | dma_wait_for_async_tx(tx); | |
3093 | } | |
a4456856 | 3094 | } |
1da177e4 LT |
3095 | |
3096 | /* | |
3097 | * handle_stripe - do things to a stripe. | |
3098 | * | |
9a3e1101 N |
3099 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3100 | * state of various bits to see what needs to be done. | |
1da177e4 | 3101 | * Possible results: |
9a3e1101 N |
3102 | * return some read requests which now have data |
3103 | * return some write requests which are safely on storage | |
1da177e4 LT |
3104 | * schedule a read on some buffers |
3105 | * schedule a write of some buffers | |
3106 | * return confirmation of parity correctness | |
3107 | * | |
1da177e4 | 3108 | */ |
a4456856 | 3109 | |
acfe726b | 3110 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3111 | { |
d1688a6d | 3112 | struct r5conf *conf = sh->raid_conf; |
f416885e | 3113 | int disks = sh->disks; |
474af965 N |
3114 | struct r5dev *dev; |
3115 | int i; | |
9a3e1101 | 3116 | int do_recovery = 0; |
1da177e4 | 3117 | |
acfe726b N |
3118 | memset(s, 0, sizeof(*s)); |
3119 | ||
acfe726b N |
3120 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3121 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
3122 | s->failed_num[0] = -1; | |
3123 | s->failed_num[1] = -1; | |
1da177e4 | 3124 | |
acfe726b | 3125 | /* Now to look around and see what can be done */ |
1da177e4 | 3126 | rcu_read_lock(); |
c4c1663b | 3127 | spin_lock_irq(&conf->device_lock); |
16a53ecc | 3128 | for (i=disks; i--; ) { |
3cb03002 | 3129 | struct md_rdev *rdev; |
31c176ec N |
3130 | sector_t first_bad; |
3131 | int bad_sectors; | |
3132 | int is_bad = 0; | |
acfe726b | 3133 | |
16a53ecc | 3134 | dev = &sh->dev[i]; |
1da177e4 | 3135 | |
45b4233c | 3136 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
3137 | i, dev->flags, |
3138 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
3139 | /* maybe we can reply to a read |
3140 | * | |
3141 | * new wantfill requests are only permitted while | |
3142 | * ops_complete_biofill is guaranteed to be inactive | |
3143 | */ | |
3144 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3145 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3146 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3147 | |
16a53ecc | 3148 | /* now count some things */ |
cc94015a N |
3149 | if (test_bit(R5_LOCKED, &dev->flags)) |
3150 | s->locked++; | |
3151 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3152 | s->uptodate++; | |
2d6e4ecc | 3153 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
3154 | s->compute++; |
3155 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 3156 | } |
1da177e4 | 3157 | |
acfe726b | 3158 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 3159 | s->to_fill++; |
acfe726b | 3160 | else if (dev->toread) |
cc94015a | 3161 | s->to_read++; |
16a53ecc | 3162 | if (dev->towrite) { |
cc94015a | 3163 | s->to_write++; |
16a53ecc | 3164 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3165 | s->non_overwrite++; |
16a53ecc | 3166 | } |
a4456856 | 3167 | if (dev->written) |
cc94015a | 3168 | s->written++; |
14a75d3e N |
3169 | /* Prefer to use the replacement for reads, but only |
3170 | * if it is recovered enough and has no bad blocks. | |
3171 | */ | |
3172 | rdev = rcu_dereference(conf->disks[i].replacement); | |
3173 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
3174 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
3175 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3176 | &first_bad, &bad_sectors)) | |
3177 | set_bit(R5_ReadRepl, &dev->flags); | |
3178 | else { | |
9a3e1101 N |
3179 | if (rdev) |
3180 | set_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
3181 | rdev = rcu_dereference(conf->disks[i].rdev); |
3182 | clear_bit(R5_ReadRepl, &dev->flags); | |
3183 | } | |
9283d8c5 N |
3184 | if (rdev && test_bit(Faulty, &rdev->flags)) |
3185 | rdev = NULL; | |
31c176ec N |
3186 | if (rdev) { |
3187 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3188 | &first_bad, &bad_sectors); | |
3189 | if (s->blocked_rdev == NULL | |
3190 | && (test_bit(Blocked, &rdev->flags) | |
3191 | || is_bad < 0)) { | |
3192 | if (is_bad < 0) | |
3193 | set_bit(BlockedBadBlocks, | |
3194 | &rdev->flags); | |
3195 | s->blocked_rdev = rdev; | |
3196 | atomic_inc(&rdev->nr_pending); | |
3197 | } | |
6bfe0b49 | 3198 | } |
415e72d0 N |
3199 | clear_bit(R5_Insync, &dev->flags); |
3200 | if (!rdev) | |
3201 | /* Not in-sync */; | |
31c176ec N |
3202 | else if (is_bad) { |
3203 | /* also not in-sync */ | |
3204 | if (!test_bit(WriteErrorSeen, &rdev->flags)) { | |
3205 | /* treat as in-sync, but with a read error | |
3206 | * which we can now try to correct | |
3207 | */ | |
3208 | set_bit(R5_Insync, &dev->flags); | |
3209 | set_bit(R5_ReadError, &dev->flags); | |
3210 | } | |
3211 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 3212 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 3213 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 3214 | /* in sync if before recovery_offset */ |
30d7a483 N |
3215 | set_bit(R5_Insync, &dev->flags); |
3216 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
3217 | test_bit(R5_Expanded, &dev->flags)) | |
3218 | /* If we've reshaped into here, we assume it is Insync. | |
3219 | * We will shortly update recovery_offset to make | |
3220 | * it official. | |
3221 | */ | |
3222 | set_bit(R5_Insync, &dev->flags); | |
3223 | ||
5d8c71f9 | 3224 | if (rdev && test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
3225 | /* This flag does not apply to '.replacement' |
3226 | * only to .rdev, so make sure to check that*/ | |
3227 | struct md_rdev *rdev2 = rcu_dereference( | |
3228 | conf->disks[i].rdev); | |
3229 | if (rdev2 == rdev) | |
3230 | clear_bit(R5_Insync, &dev->flags); | |
3231 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 3232 | s->handle_bad_blocks = 1; |
14a75d3e | 3233 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
3234 | } else |
3235 | clear_bit(R5_WriteError, &dev->flags); | |
3236 | } | |
5d8c71f9 | 3237 | if (rdev && test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
3238 | /* This flag does not apply to '.replacement' |
3239 | * only to .rdev, so make sure to check that*/ | |
3240 | struct md_rdev *rdev2 = rcu_dereference( | |
3241 | conf->disks[i].rdev); | |
3242 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 3243 | s->handle_bad_blocks = 1; |
14a75d3e | 3244 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
3245 | } else |
3246 | clear_bit(R5_MadeGood, &dev->flags); | |
3247 | } | |
977df362 N |
3248 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
3249 | struct md_rdev *rdev2 = rcu_dereference( | |
3250 | conf->disks[i].replacement); | |
3251 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
3252 | s->handle_bad_blocks = 1; | |
3253 | atomic_inc(&rdev2->nr_pending); | |
3254 | } else | |
3255 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
3256 | } | |
415e72d0 | 3257 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
3258 | /* The ReadError flag will just be confusing now */ |
3259 | clear_bit(R5_ReadError, &dev->flags); | |
3260 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3261 | } |
415e72d0 N |
3262 | if (test_bit(R5_ReadError, &dev->flags)) |
3263 | clear_bit(R5_Insync, &dev->flags); | |
3264 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3265 | if (s->failed < 2) |
3266 | s->failed_num[s->failed] = i; | |
3267 | s->failed++; | |
9a3e1101 N |
3268 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
3269 | do_recovery = 1; | |
415e72d0 | 3270 | } |
1da177e4 | 3271 | } |
c4c1663b | 3272 | spin_unlock_irq(&conf->device_lock); |
9a3e1101 N |
3273 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
3274 | /* If there is a failed device being replaced, | |
3275 | * we must be recovering. | |
3276 | * else if we are after recovery_cp, we must be syncing | |
3277 | * else we can only be replacing | |
3278 | * sync and recovery both need to read all devices, and so | |
3279 | * use the same flag. | |
3280 | */ | |
3281 | if (do_recovery || | |
3282 | sh->sector >= conf->mddev->recovery_cp) | |
3283 | s->syncing = 1; | |
3284 | else | |
3285 | s->replacing = 1; | |
3286 | } | |
1da177e4 | 3287 | rcu_read_unlock(); |
cc94015a N |
3288 | } |
3289 | ||
3290 | static void handle_stripe(struct stripe_head *sh) | |
3291 | { | |
3292 | struct stripe_head_state s; | |
d1688a6d | 3293 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 3294 | int i; |
84789554 N |
3295 | int prexor; |
3296 | int disks = sh->disks; | |
474af965 | 3297 | struct r5dev *pdev, *qdev; |
cc94015a N |
3298 | |
3299 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 3300 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
3301 | /* already being handled, ensure it gets handled |
3302 | * again when current action finishes */ | |
3303 | set_bit(STRIPE_HANDLE, &sh->state); | |
3304 | return; | |
3305 | } | |
3306 | ||
3307 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3308 | set_bit(STRIPE_SYNCING, &sh->state); | |
3309 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3310 | } | |
3311 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3312 | ||
3313 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3314 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3315 | (unsigned long long)sh->sector, sh->state, | |
3316 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3317 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 3318 | |
acfe726b | 3319 | analyse_stripe(sh, &s); |
c5a31000 | 3320 | |
bc2607f3 N |
3321 | if (s.handle_bad_blocks) { |
3322 | set_bit(STRIPE_HANDLE, &sh->state); | |
3323 | goto finish; | |
3324 | } | |
3325 | ||
474af965 N |
3326 | if (unlikely(s.blocked_rdev)) { |
3327 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 3328 | s.replacing || s.to_write || s.written) { |
474af965 N |
3329 | set_bit(STRIPE_HANDLE, &sh->state); |
3330 | goto finish; | |
3331 | } | |
3332 | /* There is nothing for the blocked_rdev to block */ | |
3333 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
3334 | s.blocked_rdev = NULL; | |
3335 | } | |
3336 | ||
3337 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
3338 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3339 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3340 | } | |
3341 | ||
3342 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
3343 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
3344 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
3345 | s.failed_num[0], s.failed_num[1]); | |
3346 | /* check if the array has lost more than max_degraded devices and, | |
3347 | * if so, some requests might need to be failed. | |
3348 | */ | |
9a3f530f N |
3349 | if (s.failed > conf->max_degraded) { |
3350 | sh->check_state = 0; | |
3351 | sh->reconstruct_state = 0; | |
3352 | if (s.to_read+s.to_write+s.written) | |
3353 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 3354 | if (s.syncing + s.replacing) |
9a3f530f N |
3355 | handle_failed_sync(conf, sh, &s); |
3356 | } | |
474af965 N |
3357 | |
3358 | /* | |
3359 | * might be able to return some write requests if the parity blocks | |
3360 | * are safe, or on a failed drive | |
3361 | */ | |
3362 | pdev = &sh->dev[sh->pd_idx]; | |
3363 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
3364 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
3365 | qdev = &sh->dev[sh->qd_idx]; | |
3366 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
3367 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
3368 | || conf->level < 6; | |
3369 | ||
3370 | if (s.written && | |
3371 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
3372 | && !test_bit(R5_LOCKED, &pdev->flags) | |
3373 | && test_bit(R5_UPTODATE, &pdev->flags)))) && | |
3374 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
3375 | && !test_bit(R5_LOCKED, &qdev->flags) | |
3376 | && test_bit(R5_UPTODATE, &qdev->flags))))) | |
3377 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
3378 | ||
3379 | /* Now we might consider reading some blocks, either to check/generate | |
3380 | * parity, or to satisfy requests | |
3381 | * or to load a block that is being partially written. | |
3382 | */ | |
3383 | if (s.to_read || s.non_overwrite | |
3384 | || (conf->level == 6 && s.to_write && s.failed) | |
9a3e1101 N |
3385 | || (s.syncing && (s.uptodate + s.compute < disks)) |
3386 | || s.replacing | |
3387 | || s.expanding) | |
474af965 N |
3388 | handle_stripe_fill(sh, &s, disks); |
3389 | ||
84789554 N |
3390 | /* Now we check to see if any write operations have recently |
3391 | * completed | |
3392 | */ | |
3393 | prexor = 0; | |
3394 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
3395 | prexor = 1; | |
3396 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
3397 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
3398 | sh->reconstruct_state = reconstruct_state_idle; | |
3399 | ||
3400 | /* All the 'written' buffers and the parity block are ready to | |
3401 | * be written back to disk | |
3402 | */ | |
3403 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3404 | BUG_ON(sh->qd_idx >= 0 && | |
3405 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags)); | |
3406 | for (i = disks; i--; ) { | |
3407 | struct r5dev *dev = &sh->dev[i]; | |
3408 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3409 | (i == sh->pd_idx || i == sh->qd_idx || | |
3410 | dev->written)) { | |
3411 | pr_debug("Writing block %d\n", i); | |
3412 | set_bit(R5_Wantwrite, &dev->flags); | |
3413 | if (prexor) | |
3414 | continue; | |
3415 | if (!test_bit(R5_Insync, &dev->flags) || | |
3416 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
3417 | s.failed == 0)) | |
3418 | set_bit(STRIPE_INSYNC, &sh->state); | |
3419 | } | |
3420 | } | |
3421 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3422 | s.dec_preread_active = 1; | |
3423 | } | |
3424 | ||
3425 | /* Now to consider new write requests and what else, if anything | |
3426 | * should be read. We do not handle new writes when: | |
3427 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3428 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3429 | * block. | |
3430 | */ | |
3431 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
3432 | handle_stripe_dirtying(conf, sh, &s, disks); | |
3433 | ||
3434 | /* maybe we need to check and possibly fix the parity for this stripe | |
3435 | * Any reads will already have been scheduled, so we just see if enough | |
3436 | * data is available. The parity check is held off while parity | |
3437 | * dependent operations are in flight. | |
3438 | */ | |
3439 | if (sh->check_state || | |
3440 | (s.syncing && s.locked == 0 && | |
3441 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3442 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
3443 | if (conf->level == 6) | |
3444 | handle_parity_checks6(conf, sh, &s, disks); | |
3445 | else | |
3446 | handle_parity_checks5(conf, sh, &s, disks); | |
3447 | } | |
c5a31000 | 3448 | |
9a3e1101 N |
3449 | if (s.replacing && s.locked == 0 |
3450 | && !test_bit(STRIPE_INSYNC, &sh->state)) { | |
3451 | /* Write out to replacement devices where possible */ | |
3452 | for (i = 0; i < conf->raid_disks; i++) | |
3453 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags) && | |
3454 | test_bit(R5_NeedReplace, &sh->dev[i].flags)) { | |
3455 | set_bit(R5_WantReplace, &sh->dev[i].flags); | |
3456 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3457 | s.locked++; | |
3458 | } | |
3459 | set_bit(STRIPE_INSYNC, &sh->state); | |
3460 | } | |
3461 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
3462 | test_bit(STRIPE_INSYNC, &sh->state)) { | |
c5a31000 N |
3463 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
3464 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3465 | } | |
3466 | ||
3467 | /* If the failed drives are just a ReadError, then we might need | |
3468 | * to progress the repair/check process | |
3469 | */ | |
3470 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
3471 | for (i = 0; i < s.failed; i++) { | |
3472 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
3473 | if (test_bit(R5_ReadError, &dev->flags) | |
3474 | && !test_bit(R5_LOCKED, &dev->flags) | |
3475 | && test_bit(R5_UPTODATE, &dev->flags) | |
3476 | ) { | |
3477 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3478 | set_bit(R5_Wantwrite, &dev->flags); | |
3479 | set_bit(R5_ReWrite, &dev->flags); | |
3480 | set_bit(R5_LOCKED, &dev->flags); | |
3481 | s.locked++; | |
3482 | } else { | |
3483 | /* let's read it back */ | |
3484 | set_bit(R5_Wantread, &dev->flags); | |
3485 | set_bit(R5_LOCKED, &dev->flags); | |
3486 | s.locked++; | |
3487 | } | |
3488 | } | |
3489 | } | |
3490 | ||
3491 | ||
3687c061 N |
3492 | /* Finish reconstruct operations initiated by the expansion process */ |
3493 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3494 | struct stripe_head *sh_src | |
3495 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
3496 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
3497 | /* sh cannot be written until sh_src has been read. | |
3498 | * so arrange for sh to be delayed a little | |
3499 | */ | |
3500 | set_bit(STRIPE_DELAYED, &sh->state); | |
3501 | set_bit(STRIPE_HANDLE, &sh->state); | |
3502 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3503 | &sh_src->state)) | |
3504 | atomic_inc(&conf->preread_active_stripes); | |
3505 | release_stripe(sh_src); | |
3506 | goto finish; | |
3507 | } | |
3508 | if (sh_src) | |
3509 | release_stripe(sh_src); | |
3510 | ||
3511 | sh->reconstruct_state = reconstruct_state_idle; | |
3512 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3513 | for (i = conf->raid_disks; i--; ) { | |
3514 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3515 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3516 | s.locked++; | |
3517 | } | |
3518 | } | |
f416885e | 3519 | |
3687c061 N |
3520 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
3521 | !sh->reconstruct_state) { | |
3522 | /* Need to write out all blocks after computing parity */ | |
3523 | sh->disks = conf->raid_disks; | |
3524 | stripe_set_idx(sh->sector, conf, 0, sh); | |
3525 | schedule_reconstruction(sh, &s, 1, 1); | |
3526 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
3527 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
3528 | atomic_dec(&conf->reshape_stripes); | |
3529 | wake_up(&conf->wait_for_overlap); | |
3530 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3531 | } | |
3532 | ||
3533 | if (s.expanding && s.locked == 0 && | |
3534 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
3535 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 3536 | |
3687c061 | 3537 | finish: |
6bfe0b49 | 3538 | /* wait for this device to become unblocked */ |
43220aa0 | 3539 | if (conf->mddev->external && unlikely(s.blocked_rdev)) |
c5709ef6 | 3540 | md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev); |
6bfe0b49 | 3541 | |
bc2607f3 N |
3542 | if (s.handle_bad_blocks) |
3543 | for (i = disks; i--; ) { | |
3cb03002 | 3544 | struct md_rdev *rdev; |
bc2607f3 N |
3545 | struct r5dev *dev = &sh->dev[i]; |
3546 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
3547 | /* We own a safe reference to the rdev */ | |
3548 | rdev = conf->disks[i].rdev; | |
3549 | if (!rdev_set_badblocks(rdev, sh->sector, | |
3550 | STRIPE_SECTORS, 0)) | |
3551 | md_error(conf->mddev, rdev); | |
3552 | rdev_dec_pending(rdev, conf->mddev); | |
3553 | } | |
b84db560 N |
3554 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
3555 | rdev = conf->disks[i].rdev; | |
3556 | rdev_clear_badblocks(rdev, sh->sector, | |
3557 | STRIPE_SECTORS); | |
3558 | rdev_dec_pending(rdev, conf->mddev); | |
3559 | } | |
977df362 N |
3560 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
3561 | rdev = conf->disks[i].replacement; | |
dd054fce N |
3562 | if (!rdev) |
3563 | /* rdev have been moved down */ | |
3564 | rdev = conf->disks[i].rdev; | |
977df362 N |
3565 | rdev_clear_badblocks(rdev, sh->sector, |
3566 | STRIPE_SECTORS); | |
3567 | rdev_dec_pending(rdev, conf->mddev); | |
3568 | } | |
bc2607f3 N |
3569 | } |
3570 | ||
6c0069c0 YT |
3571 | if (s.ops_request) |
3572 | raid_run_ops(sh, s.ops_request); | |
3573 | ||
f0e43bcd | 3574 | ops_run_io(sh, &s); |
16a53ecc | 3575 | |
c5709ef6 | 3576 | if (s.dec_preread_active) { |
729a1866 | 3577 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 3578 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3579 | * have actually been submitted. |
3580 | */ | |
3581 | atomic_dec(&conf->preread_active_stripes); | |
3582 | if (atomic_read(&conf->preread_active_stripes) < | |
3583 | IO_THRESHOLD) | |
3584 | md_wakeup_thread(conf->mddev->thread); | |
3585 | } | |
3586 | ||
c5709ef6 | 3587 | return_io(s.return_bi); |
16a53ecc | 3588 | |
257a4b42 | 3589 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
3590 | } |
3591 | ||
d1688a6d | 3592 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
3593 | { |
3594 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3595 | while (!list_empty(&conf->delayed_list)) { | |
3596 | struct list_head *l = conf->delayed_list.next; | |
3597 | struct stripe_head *sh; | |
3598 | sh = list_entry(l, struct stripe_head, lru); | |
3599 | list_del_init(l); | |
3600 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3601 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3602 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3603 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3604 | } |
482c0834 | 3605 | } |
16a53ecc N |
3606 | } |
3607 | ||
d1688a6d | 3608 | static void activate_bit_delay(struct r5conf *conf) |
16a53ecc N |
3609 | { |
3610 | /* device_lock is held */ | |
3611 | struct list_head head; | |
3612 | list_add(&head, &conf->bitmap_list); | |
3613 | list_del_init(&conf->bitmap_list); | |
3614 | while (!list_empty(&head)) { | |
3615 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3616 | list_del_init(&sh->lru); | |
3617 | atomic_inc(&sh->count); | |
3618 | __release_stripe(conf, sh); | |
3619 | } | |
3620 | } | |
3621 | ||
fd01b88c | 3622 | int md_raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 3623 | { |
d1688a6d | 3624 | struct r5conf *conf = mddev->private; |
f022b2fd N |
3625 | |
3626 | /* No difference between reads and writes. Just check | |
3627 | * how busy the stripe_cache is | |
3628 | */ | |
3fa841d7 | 3629 | |
f022b2fd N |
3630 | if (conf->inactive_blocked) |
3631 | return 1; | |
3632 | if (conf->quiesce) | |
3633 | return 1; | |
3634 | if (list_empty_careful(&conf->inactive_list)) | |
3635 | return 1; | |
3636 | ||
3637 | return 0; | |
3638 | } | |
11d8a6e3 N |
3639 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3640 | ||
3641 | static int raid5_congested(void *data, int bits) | |
3642 | { | |
fd01b88c | 3643 | struct mddev *mddev = data; |
11d8a6e3 N |
3644 | |
3645 | return mddev_congested(mddev, bits) || | |
3646 | md_raid5_congested(mddev, bits); | |
3647 | } | |
f022b2fd | 3648 | |
23032a0e RBJ |
3649 | /* We want read requests to align with chunks where possible, |
3650 | * but write requests don't need to. | |
3651 | */ | |
cc371e66 AK |
3652 | static int raid5_mergeable_bvec(struct request_queue *q, |
3653 | struct bvec_merge_data *bvm, | |
3654 | struct bio_vec *biovec) | |
23032a0e | 3655 | { |
fd01b88c | 3656 | struct mddev *mddev = q->queuedata; |
cc371e66 | 3657 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3658 | int max; |
9d8f0363 | 3659 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3660 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3661 | |
cc371e66 | 3662 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3663 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3664 | ||
664e7c41 AN |
3665 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3666 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3667 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3668 | if (max < 0) max = 0; | |
3669 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3670 | return biovec->bv_len; | |
3671 | else | |
3672 | return max; | |
3673 | } | |
3674 | ||
f679623f | 3675 | |
fd01b88c | 3676 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f RBJ |
3677 | { |
3678 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3679 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3680 | unsigned int bio_sectors = bio->bi_size >> 9; |
3681 | ||
664e7c41 AN |
3682 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3683 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3684 | return chunk_sectors >= |
3685 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3686 | } | |
3687 | ||
46031f9a RBJ |
3688 | /* |
3689 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3690 | * later sampled by raid5d. | |
3691 | */ | |
d1688a6d | 3692 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
3693 | { |
3694 | unsigned long flags; | |
3695 | ||
3696 | spin_lock_irqsave(&conf->device_lock, flags); | |
3697 | ||
3698 | bi->bi_next = conf->retry_read_aligned_list; | |
3699 | conf->retry_read_aligned_list = bi; | |
3700 | ||
3701 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3702 | md_wakeup_thread(conf->mddev->thread); | |
3703 | } | |
3704 | ||
3705 | ||
d1688a6d | 3706 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
3707 | { |
3708 | struct bio *bi; | |
3709 | ||
3710 | bi = conf->retry_read_aligned; | |
3711 | if (bi) { | |
3712 | conf->retry_read_aligned = NULL; | |
3713 | return bi; | |
3714 | } | |
3715 | bi = conf->retry_read_aligned_list; | |
3716 | if(bi) { | |
387bb173 | 3717 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3718 | bi->bi_next = NULL; |
960e739d JA |
3719 | /* |
3720 | * this sets the active strip count to 1 and the processed | |
3721 | * strip count to zero (upper 8 bits) | |
3722 | */ | |
46031f9a | 3723 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3724 | } |
3725 | ||
3726 | return bi; | |
3727 | } | |
3728 | ||
3729 | ||
f679623f RBJ |
3730 | /* |
3731 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3732 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3733 | * first). | |
3734 | * If the read failed.. | |
3735 | */ | |
6712ecf8 | 3736 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3737 | { |
3738 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 3739 | struct mddev *mddev; |
d1688a6d | 3740 | struct r5conf *conf; |
46031f9a | 3741 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 3742 | struct md_rdev *rdev; |
46031f9a | 3743 | |
f679623f | 3744 | bio_put(bi); |
46031f9a | 3745 | |
46031f9a RBJ |
3746 | rdev = (void*)raid_bi->bi_next; |
3747 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3748 | mddev = rdev->mddev; |
3749 | conf = mddev->private; | |
46031f9a RBJ |
3750 | |
3751 | rdev_dec_pending(rdev, conf->mddev); | |
3752 | ||
3753 | if (!error && uptodate) { | |
6712ecf8 | 3754 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3755 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3756 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3757 | return; |
46031f9a RBJ |
3758 | } |
3759 | ||
3760 | ||
45b4233c | 3761 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3762 | |
3763 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3764 | } |
3765 | ||
387bb173 NB |
3766 | static int bio_fits_rdev(struct bio *bi) |
3767 | { | |
165125e1 | 3768 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3769 | |
ae03bf63 | 3770 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3771 | return 0; |
3772 | blk_recount_segments(q, bi); | |
8a78362c | 3773 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3774 | return 0; |
3775 | ||
3776 | if (q->merge_bvec_fn) | |
3777 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3778 | * just just give up | |
3779 | */ | |
3780 | return 0; | |
3781 | ||
3782 | return 1; | |
3783 | } | |
3784 | ||
3785 | ||
fd01b88c | 3786 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 3787 | { |
d1688a6d | 3788 | struct r5conf *conf = mddev->private; |
8553fe7e | 3789 | int dd_idx; |
f679623f | 3790 | struct bio* align_bi; |
3cb03002 | 3791 | struct md_rdev *rdev; |
671488cc | 3792 | sector_t end_sector; |
f679623f RBJ |
3793 | |
3794 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3795 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3796 | return 0; |
3797 | } | |
3798 | /* | |
a167f663 | 3799 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3800 | */ |
a167f663 | 3801 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3802 | if (!align_bi) |
3803 | return 0; | |
3804 | /* | |
3805 | * set bi_end_io to a new function, and set bi_private to the | |
3806 | * original bio. | |
3807 | */ | |
3808 | align_bi->bi_end_io = raid5_align_endio; | |
3809 | align_bi->bi_private = raid_bio; | |
3810 | /* | |
3811 | * compute position | |
3812 | */ | |
112bf897 N |
3813 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3814 | 0, | |
911d4ee8 | 3815 | &dd_idx, NULL); |
f679623f | 3816 | |
671488cc | 3817 | end_sector = align_bi->bi_sector + (align_bi->bi_size >> 9); |
f679623f | 3818 | rcu_read_lock(); |
671488cc N |
3819 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
3820 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
3821 | rdev->recovery_offset < end_sector) { | |
3822 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3823 | if (rdev && | |
3824 | (test_bit(Faulty, &rdev->flags) || | |
3825 | !(test_bit(In_sync, &rdev->flags) || | |
3826 | rdev->recovery_offset >= end_sector))) | |
3827 | rdev = NULL; | |
3828 | } | |
3829 | if (rdev) { | |
31c176ec N |
3830 | sector_t first_bad; |
3831 | int bad_sectors; | |
3832 | ||
f679623f RBJ |
3833 | atomic_inc(&rdev->nr_pending); |
3834 | rcu_read_unlock(); | |
46031f9a RBJ |
3835 | raid_bio->bi_next = (void*)rdev; |
3836 | align_bi->bi_bdev = rdev->bdev; | |
3837 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3838 | align_bi->bi_sector += rdev->data_offset; | |
3839 | ||
31c176ec N |
3840 | if (!bio_fits_rdev(align_bi) || |
3841 | is_badblock(rdev, align_bi->bi_sector, align_bi->bi_size>>9, | |
3842 | &first_bad, &bad_sectors)) { | |
3843 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
3844 | bio_put(align_bi); |
3845 | rdev_dec_pending(rdev, mddev); | |
3846 | return 0; | |
3847 | } | |
3848 | ||
46031f9a RBJ |
3849 | spin_lock_irq(&conf->device_lock); |
3850 | wait_event_lock_irq(conf->wait_for_stripe, | |
3851 | conf->quiesce == 0, | |
3852 | conf->device_lock, /* nothing */); | |
3853 | atomic_inc(&conf->active_aligned_reads); | |
3854 | spin_unlock_irq(&conf->device_lock); | |
3855 | ||
f679623f RBJ |
3856 | generic_make_request(align_bi); |
3857 | return 1; | |
3858 | } else { | |
3859 | rcu_read_unlock(); | |
46031f9a | 3860 | bio_put(align_bi); |
f679623f RBJ |
3861 | return 0; |
3862 | } | |
3863 | } | |
3864 | ||
8b3e6cdc DW |
3865 | /* __get_priority_stripe - get the next stripe to process |
3866 | * | |
3867 | * Full stripe writes are allowed to pass preread active stripes up until | |
3868 | * the bypass_threshold is exceeded. In general the bypass_count | |
3869 | * increments when the handle_list is handled before the hold_list; however, it | |
3870 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3871 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3872 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3873 | * handle_list. | |
3874 | */ | |
d1688a6d | 3875 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf) |
8b3e6cdc DW |
3876 | { |
3877 | struct stripe_head *sh; | |
3878 | ||
3879 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3880 | __func__, | |
3881 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3882 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3883 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3884 | ||
3885 | if (!list_empty(&conf->handle_list)) { | |
3886 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3887 | ||
3888 | if (list_empty(&conf->hold_list)) | |
3889 | conf->bypass_count = 0; | |
3890 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3891 | if (conf->hold_list.next == conf->last_hold) | |
3892 | conf->bypass_count++; | |
3893 | else { | |
3894 | conf->last_hold = conf->hold_list.next; | |
3895 | conf->bypass_count -= conf->bypass_threshold; | |
3896 | if (conf->bypass_count < 0) | |
3897 | conf->bypass_count = 0; | |
3898 | } | |
3899 | } | |
3900 | } else if (!list_empty(&conf->hold_list) && | |
3901 | ((conf->bypass_threshold && | |
3902 | conf->bypass_count > conf->bypass_threshold) || | |
3903 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3904 | sh = list_entry(conf->hold_list.next, | |
3905 | typeof(*sh), lru); | |
3906 | conf->bypass_count -= conf->bypass_threshold; | |
3907 | if (conf->bypass_count < 0) | |
3908 | conf->bypass_count = 0; | |
3909 | } else | |
3910 | return NULL; | |
3911 | ||
3912 | list_del_init(&sh->lru); | |
3913 | atomic_inc(&sh->count); | |
3914 | BUG_ON(atomic_read(&sh->count) != 1); | |
3915 | return sh; | |
3916 | } | |
f679623f | 3917 | |
b4fdcb02 | 3918 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 3919 | { |
d1688a6d | 3920 | struct r5conf *conf = mddev->private; |
911d4ee8 | 3921 | int dd_idx; |
1da177e4 LT |
3922 | sector_t new_sector; |
3923 | sector_t logical_sector, last_sector; | |
3924 | struct stripe_head *sh; | |
a362357b | 3925 | const int rw = bio_data_dir(bi); |
49077326 | 3926 | int remaining; |
7c13edc8 | 3927 | int plugged; |
1da177e4 | 3928 | |
e9c7469b TH |
3929 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
3930 | md_flush_request(mddev, bi); | |
5a7bbad2 | 3931 | return; |
e5dcdd80 N |
3932 | } |
3933 | ||
3d310eb7 | 3934 | md_write_start(mddev, bi); |
06d91a5f | 3935 | |
802ba064 | 3936 | if (rw == READ && |
52488615 | 3937 | mddev->reshape_position == MaxSector && |
21a52c6d | 3938 | chunk_aligned_read(mddev,bi)) |
5a7bbad2 | 3939 | return; |
52488615 | 3940 | |
1da177e4 LT |
3941 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3942 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3943 | bi->bi_next = NULL; | |
3944 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3945 | |
7c13edc8 | 3946 | plugged = mddev_check_plugged(mddev); |
1da177e4 LT |
3947 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3948 | DEFINE_WAIT(w); | |
16a53ecc | 3949 | int disks, data_disks; |
b5663ba4 | 3950 | int previous; |
b578d55f | 3951 | |
7ecaa1e6 | 3952 | retry: |
b5663ba4 | 3953 | previous = 0; |
b0f9ec04 | 3954 | disks = conf->raid_disks; |
b578d55f | 3955 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3956 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3957 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3958 | * 64bit on a 32bit platform, and so it might be |
3959 | * possible to see a half-updated value | |
aeb878b0 | 3960 | * Of course reshape_progress could change after |
df8e7f76 N |
3961 | * the lock is dropped, so once we get a reference |
3962 | * to the stripe that we think it is, we will have | |
3963 | * to check again. | |
3964 | */ | |
7ecaa1e6 | 3965 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3966 | if (mddev->delta_disks < 0 |
3967 | ? logical_sector < conf->reshape_progress | |
3968 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3969 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3970 | previous = 1; |
3971 | } else { | |
fef9c61f N |
3972 | if (mddev->delta_disks < 0 |
3973 | ? logical_sector < conf->reshape_safe | |
3974 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3975 | spin_unlock_irq(&conf->device_lock); |
3976 | schedule(); | |
3977 | goto retry; | |
3978 | } | |
3979 | } | |
7ecaa1e6 N |
3980 | spin_unlock_irq(&conf->device_lock); |
3981 | } | |
16a53ecc N |
3982 | data_disks = disks - conf->max_degraded; |
3983 | ||
112bf897 N |
3984 | new_sector = raid5_compute_sector(conf, logical_sector, |
3985 | previous, | |
911d4ee8 | 3986 | &dd_idx, NULL); |
0c55e022 | 3987 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3988 | (unsigned long long)new_sector, |
3989 | (unsigned long long)logical_sector); | |
3990 | ||
b5663ba4 | 3991 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3992 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3993 | if (sh) { |
b0f9ec04 | 3994 | if (unlikely(previous)) { |
7ecaa1e6 | 3995 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3996 | * stripe, so we must do the range check again. |
3997 | * Expansion could still move past after this | |
3998 | * test, but as we are holding a reference to | |
3999 | * 'sh', we know that if that happens, | |
4000 | * STRIPE_EXPANDING will get set and the expansion | |
4001 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
4002 | */ |
4003 | int must_retry = 0; | |
4004 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
4005 | if (mddev->delta_disks < 0 |
4006 | ? logical_sector >= conf->reshape_progress | |
4007 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
4008 | /* mismatch, need to try again */ |
4009 | must_retry = 1; | |
4010 | spin_unlock_irq(&conf->device_lock); | |
4011 | if (must_retry) { | |
4012 | release_stripe(sh); | |
7a3ab908 | 4013 | schedule(); |
7ecaa1e6 N |
4014 | goto retry; |
4015 | } | |
4016 | } | |
e62e58a5 | 4017 | |
ffd96e35 | 4018 | if (rw == WRITE && |
a5c308d4 | 4019 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
4020 | logical_sector < mddev->suspend_hi) { |
4021 | release_stripe(sh); | |
e62e58a5 N |
4022 | /* As the suspend_* range is controlled by |
4023 | * userspace, we want an interruptible | |
4024 | * wait. | |
4025 | */ | |
4026 | flush_signals(current); | |
4027 | prepare_to_wait(&conf->wait_for_overlap, | |
4028 | &w, TASK_INTERRUPTIBLE); | |
4029 | if (logical_sector >= mddev->suspend_lo && | |
4030 | logical_sector < mddev->suspend_hi) | |
4031 | schedule(); | |
e464eafd N |
4032 | goto retry; |
4033 | } | |
7ecaa1e6 N |
4034 | |
4035 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 4036 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
4037 | /* Stripe is busy expanding or |
4038 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4039 | * and wait a while |
4040 | */ | |
482c0834 | 4041 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4042 | release_stripe(sh); |
4043 | schedule(); | |
4044 | goto retry; | |
4045 | } | |
4046 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
4047 | set_bit(STRIPE_HANDLE, &sh->state); |
4048 | clear_bit(STRIPE_DELAYED, &sh->state); | |
e9c7469b | 4049 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
4050 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4051 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 4052 | release_stripe(sh); |
1da177e4 LT |
4053 | } else { |
4054 | /* cannot get stripe for read-ahead, just give-up */ | |
4055 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4056 | finish_wait(&conf->wait_for_overlap, &w); | |
4057 | break; | |
4058 | } | |
4059 | ||
4060 | } | |
7c13edc8 N |
4061 | if (!plugged) |
4062 | md_wakeup_thread(mddev->thread); | |
4063 | ||
1da177e4 | 4064 | spin_lock_irq(&conf->device_lock); |
960e739d | 4065 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
4066 | spin_unlock_irq(&conf->device_lock); |
4067 | if (remaining == 0) { | |
1da177e4 | 4068 | |
16a53ecc | 4069 | if ( rw == WRITE ) |
1da177e4 | 4070 | md_write_end(mddev); |
6712ecf8 | 4071 | |
0e13fe23 | 4072 | bio_endio(bi, 0); |
1da177e4 | 4073 | } |
1da177e4 LT |
4074 | } |
4075 | ||
fd01b88c | 4076 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 4077 | |
fd01b88c | 4078 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4079 | { |
52c03291 N |
4080 | /* reshaping is quite different to recovery/resync so it is |
4081 | * handled quite separately ... here. | |
4082 | * | |
4083 | * On each call to sync_request, we gather one chunk worth of | |
4084 | * destination stripes and flag them as expanding. | |
4085 | * Then we find all the source stripes and request reads. | |
4086 | * As the reads complete, handle_stripe will copy the data | |
4087 | * into the destination stripe and release that stripe. | |
4088 | */ | |
d1688a6d | 4089 | struct r5conf *conf = mddev->private; |
1da177e4 | 4090 | struct stripe_head *sh; |
ccfcc3c1 | 4091 | sector_t first_sector, last_sector; |
f416885e N |
4092 | int raid_disks = conf->previous_raid_disks; |
4093 | int data_disks = raid_disks - conf->max_degraded; | |
4094 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4095 | int i; |
4096 | int dd_idx; | |
c8f517c4 | 4097 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4098 | sector_t stripe_addr; |
7a661381 | 4099 | int reshape_sectors; |
ab69ae12 | 4100 | struct list_head stripes; |
52c03291 | 4101 | |
fef9c61f N |
4102 | if (sector_nr == 0) { |
4103 | /* If restarting in the middle, skip the initial sectors */ | |
4104 | if (mddev->delta_disks < 0 && | |
4105 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4106 | sector_nr = raid5_size(mddev, 0, 0) | |
4107 | - conf->reshape_progress; | |
a639755c | 4108 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
4109 | conf->reshape_progress > 0) |
4110 | sector_nr = conf->reshape_progress; | |
f416885e | 4111 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4112 | if (sector_nr) { |
8dee7211 N |
4113 | mddev->curr_resync_completed = sector_nr; |
4114 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4115 | *skipped = 1; |
4116 | return sector_nr; | |
4117 | } | |
52c03291 N |
4118 | } |
4119 | ||
7a661381 N |
4120 | /* We need to process a full chunk at a time. |
4121 | * If old and new chunk sizes differ, we need to process the | |
4122 | * largest of these | |
4123 | */ | |
664e7c41 AN |
4124 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4125 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4126 | else |
9d8f0363 | 4127 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4128 | |
52c03291 N |
4129 | /* we update the metadata when there is more than 3Meg |
4130 | * in the block range (that is rather arbitrary, should | |
4131 | * probably be time based) or when the data about to be | |
4132 | * copied would over-write the source of the data at | |
4133 | * the front of the range. | |
fef9c61f N |
4134 | * i.e. one new_stripe along from reshape_progress new_maps |
4135 | * to after where reshape_safe old_maps to | |
52c03291 | 4136 | */ |
fef9c61f | 4137 | writepos = conf->reshape_progress; |
f416885e | 4138 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4139 | readpos = conf->reshape_progress; |
4140 | sector_div(readpos, data_disks); | |
fef9c61f | 4141 | safepos = conf->reshape_safe; |
f416885e | 4142 | sector_div(safepos, data_disks); |
fef9c61f | 4143 | if (mddev->delta_disks < 0) { |
ed37d83e | 4144 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4145 | readpos += reshape_sectors; |
7a661381 | 4146 | safepos += reshape_sectors; |
fef9c61f | 4147 | } else { |
7a661381 | 4148 | writepos += reshape_sectors; |
ed37d83e N |
4149 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4150 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4151 | } |
52c03291 | 4152 | |
c8f517c4 N |
4153 | /* 'writepos' is the most advanced device address we might write. |
4154 | * 'readpos' is the least advanced device address we might read. | |
4155 | * 'safepos' is the least address recorded in the metadata as having | |
4156 | * been reshaped. | |
4157 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4158 | * ensure safety in the face of a crash - that must be done by userspace | |
4159 | * making a backup of the data. So in that case there is no particular | |
4160 | * rush to update metadata. | |
4161 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4162 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4163 | * we can be safe in the event of a crash. | |
4164 | * So we insist on updating metadata if safepos is behind writepos and | |
4165 | * readpos is beyond writepos. | |
4166 | * In any case, update the metadata every 10 seconds. | |
4167 | * Maybe that number should be configurable, but I'm not sure it is | |
4168 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4169 | */ | |
fef9c61f | 4170 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4171 | ? (safepos > writepos && readpos < writepos) |
4172 | : (safepos < writepos && readpos > writepos)) || | |
4173 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4174 | /* Cannot proceed until we've updated the superblock... */ |
4175 | wait_event(conf->wait_for_overlap, | |
4176 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4177 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4178 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4179 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4180 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4181 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4182 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4183 | kthread_should_stop()); |
4184 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4185 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4186 | spin_unlock_irq(&conf->device_lock); |
4187 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4188 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4189 | } |
4190 | ||
ec32a2bd N |
4191 | if (mddev->delta_disks < 0) { |
4192 | BUG_ON(conf->reshape_progress == 0); | |
4193 | stripe_addr = writepos; | |
4194 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4195 | ~((sector_t)reshape_sectors - 1)) |
4196 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4197 | != sector_nr); |
4198 | } else { | |
7a661381 | 4199 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4200 | stripe_addr = sector_nr; |
4201 | } | |
ab69ae12 | 4202 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4203 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4204 | int j; |
a9f326eb | 4205 | int skipped_disk = 0; |
a8c906ca | 4206 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4207 | set_bit(STRIPE_EXPANDING, &sh->state); |
4208 | atomic_inc(&conf->reshape_stripes); | |
4209 | /* If any of this stripe is beyond the end of the old | |
4210 | * array, then we need to zero those blocks | |
4211 | */ | |
4212 | for (j=sh->disks; j--;) { | |
4213 | sector_t s; | |
4214 | if (j == sh->pd_idx) | |
4215 | continue; | |
f416885e | 4216 | if (conf->level == 6 && |
d0dabf7e | 4217 | j == sh->qd_idx) |
f416885e | 4218 | continue; |
784052ec | 4219 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4220 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4221 | skipped_disk = 1; |
52c03291 N |
4222 | continue; |
4223 | } | |
4224 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4225 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4226 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4227 | } | |
a9f326eb | 4228 | if (!skipped_disk) { |
52c03291 N |
4229 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4230 | set_bit(STRIPE_HANDLE, &sh->state); | |
4231 | } | |
ab69ae12 | 4232 | list_add(&sh->lru, &stripes); |
52c03291 N |
4233 | } |
4234 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4235 | if (mddev->delta_disks < 0) |
7a661381 | 4236 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4237 | else |
7a661381 | 4238 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4239 | spin_unlock_irq(&conf->device_lock); |
4240 | /* Ok, those stripe are ready. We can start scheduling | |
4241 | * reads on the source stripes. | |
4242 | * The source stripes are determined by mapping the first and last | |
4243 | * block on the destination stripes. | |
4244 | */ | |
52c03291 | 4245 | first_sector = |
ec32a2bd | 4246 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4247 | 1, &dd_idx, NULL); |
52c03291 | 4248 | last_sector = |
0e6e0271 | 4249 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4250 | * new_data_disks - 1), |
911d4ee8 | 4251 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4252 | if (last_sector >= mddev->dev_sectors) |
4253 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4254 | while (first_sector <= last_sector) { |
a8c906ca | 4255 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4256 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4257 | set_bit(STRIPE_HANDLE, &sh->state); | |
4258 | release_stripe(sh); | |
4259 | first_sector += STRIPE_SECTORS; | |
4260 | } | |
ab69ae12 N |
4261 | /* Now that the sources are clearly marked, we can release |
4262 | * the destination stripes | |
4263 | */ | |
4264 | while (!list_empty(&stripes)) { | |
4265 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4266 | list_del_init(&sh->lru); | |
4267 | release_stripe(sh); | |
4268 | } | |
c6207277 N |
4269 | /* If this takes us to the resync_max point where we have to pause, |
4270 | * then we need to write out the superblock. | |
4271 | */ | |
7a661381 | 4272 | sector_nr += reshape_sectors; |
c03f6a19 N |
4273 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4274 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4275 | /* Cannot proceed until we've updated the superblock... */ |
4276 | wait_event(conf->wait_for_overlap, | |
4277 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4278 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4279 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4280 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4281 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4282 | md_wakeup_thread(mddev->thread); | |
4283 | wait_event(mddev->sb_wait, | |
4284 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4285 | || kthread_should_stop()); | |
4286 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4287 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4288 | spin_unlock_irq(&conf->device_lock); |
4289 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4290 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4291 | } |
7a661381 | 4292 | return reshape_sectors; |
52c03291 N |
4293 | } |
4294 | ||
4295 | /* FIXME go_faster isn't used */ | |
fd01b88c | 4296 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
52c03291 | 4297 | { |
d1688a6d | 4298 | struct r5conf *conf = mddev->private; |
52c03291 | 4299 | struct stripe_head *sh; |
58c0fed4 | 4300 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 4301 | sector_t sync_blocks; |
16a53ecc N |
4302 | int still_degraded = 0; |
4303 | int i; | |
1da177e4 | 4304 | |
72626685 | 4305 | if (sector_nr >= max_sector) { |
1da177e4 | 4306 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 4307 | |
29269553 N |
4308 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4309 | end_reshape(conf); | |
4310 | return 0; | |
4311 | } | |
72626685 N |
4312 | |
4313 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4314 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4315 | &sync_blocks, 1); | |
16a53ecc | 4316 | else /* completed sync */ |
72626685 N |
4317 | conf->fullsync = 0; |
4318 | bitmap_close_sync(mddev->bitmap); | |
4319 | ||
1da177e4 LT |
4320 | return 0; |
4321 | } | |
ccfcc3c1 | 4322 | |
64bd660b N |
4323 | /* Allow raid5_quiesce to complete */ |
4324 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4325 | ||
52c03291 N |
4326 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4327 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4328 | |
c6207277 N |
4329 | /* No need to check resync_max as we never do more than one |
4330 | * stripe, and as resync_max will always be on a chunk boundary, | |
4331 | * if the check in md_do_sync didn't fire, there is no chance | |
4332 | * of overstepping resync_max here | |
4333 | */ | |
4334 | ||
16a53ecc | 4335 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4336 | * to resync, then assert that we are finished, because there is |
4337 | * nothing we can do. | |
4338 | */ | |
3285edf1 | 4339 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4340 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4341 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4342 | *skipped = 1; |
1da177e4 LT |
4343 | return rv; |
4344 | } | |
72626685 | 4345 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4346 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4347 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4348 | /* we can skip this block, and probably more */ | |
4349 | sync_blocks /= STRIPE_SECTORS; | |
4350 | *skipped = 1; | |
4351 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4352 | } | |
1da177e4 | 4353 | |
b47490c9 N |
4354 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
4355 | ||
a8c906ca | 4356 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4357 | if (sh == NULL) { |
a8c906ca | 4358 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4359 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4360 | * is trying to get access |
1da177e4 | 4361 | */ |
66c006a5 | 4362 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4363 | } |
16a53ecc N |
4364 | /* Need to check if array will still be degraded after recovery/resync |
4365 | * We don't need to check the 'failed' flag as when that gets set, | |
4366 | * recovery aborts. | |
4367 | */ | |
f001a70c | 4368 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4369 | if (conf->disks[i].rdev == NULL) |
4370 | still_degraded = 1; | |
4371 | ||
4372 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4373 | ||
83206d66 | 4374 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4375 | |
1442577b | 4376 | handle_stripe(sh); |
1da177e4 LT |
4377 | release_stripe(sh); |
4378 | ||
4379 | return STRIPE_SECTORS; | |
4380 | } | |
4381 | ||
d1688a6d | 4382 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
4383 | { |
4384 | /* We may not be able to submit a whole bio at once as there | |
4385 | * may not be enough stripe_heads available. | |
4386 | * We cannot pre-allocate enough stripe_heads as we may need | |
4387 | * more than exist in the cache (if we allow ever large chunks). | |
4388 | * So we do one stripe head at a time and record in | |
4389 | * ->bi_hw_segments how many have been done. | |
4390 | * | |
4391 | * We *know* that this entire raid_bio is in one chunk, so | |
4392 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4393 | */ | |
4394 | struct stripe_head *sh; | |
911d4ee8 | 4395 | int dd_idx; |
46031f9a RBJ |
4396 | sector_t sector, logical_sector, last_sector; |
4397 | int scnt = 0; | |
4398 | int remaining; | |
4399 | int handled = 0; | |
4400 | ||
4401 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4402 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4403 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4404 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4405 | ||
4406 | for (; logical_sector < last_sector; | |
387bb173 NB |
4407 | logical_sector += STRIPE_SECTORS, |
4408 | sector += STRIPE_SECTORS, | |
4409 | scnt++) { | |
46031f9a | 4410 | |
960e739d | 4411 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4412 | /* already done this stripe */ |
4413 | continue; | |
4414 | ||
a8c906ca | 4415 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4416 | |
4417 | if (!sh) { | |
4418 | /* failed to get a stripe - must wait */ | |
960e739d | 4419 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4420 | conf->retry_read_aligned = raid_bio; |
4421 | return handled; | |
4422 | } | |
4423 | ||
387bb173 NB |
4424 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4425 | release_stripe(sh); | |
960e739d | 4426 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4427 | conf->retry_read_aligned = raid_bio; |
4428 | return handled; | |
4429 | } | |
4430 | ||
36d1c647 | 4431 | handle_stripe(sh); |
46031f9a RBJ |
4432 | release_stripe(sh); |
4433 | handled++; | |
4434 | } | |
4435 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4436 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4437 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4438 | if (remaining == 0) |
4439 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4440 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4441 | wake_up(&conf->wait_for_stripe); | |
4442 | return handled; | |
4443 | } | |
4444 | ||
46031f9a | 4445 | |
1da177e4 LT |
4446 | /* |
4447 | * This is our raid5 kernel thread. | |
4448 | * | |
4449 | * We scan the hash table for stripes which can be handled now. | |
4450 | * During the scan, completed stripes are saved for us by the interrupt | |
4451 | * handler, so that they will not have to wait for our next wakeup. | |
4452 | */ | |
fd01b88c | 4453 | static void raid5d(struct mddev *mddev) |
1da177e4 LT |
4454 | { |
4455 | struct stripe_head *sh; | |
d1688a6d | 4456 | struct r5conf *conf = mddev->private; |
1da177e4 | 4457 | int handled; |
e1dfa0a2 | 4458 | struct blk_plug plug; |
1da177e4 | 4459 | |
45b4233c | 4460 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4461 | |
4462 | md_check_recovery(mddev); | |
1da177e4 | 4463 | |
e1dfa0a2 | 4464 | blk_start_plug(&plug); |
1da177e4 LT |
4465 | handled = 0; |
4466 | spin_lock_irq(&conf->device_lock); | |
4467 | while (1) { | |
46031f9a | 4468 | struct bio *bio; |
1da177e4 | 4469 | |
7c13edc8 N |
4470 | if (atomic_read(&mddev->plug_cnt) == 0 && |
4471 | !list_empty(&conf->bitmap_list)) { | |
4472 | /* Now is a good time to flush some bitmap updates */ | |
4473 | conf->seq_flush++; | |
700e432d | 4474 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4475 | bitmap_unplug(mddev->bitmap); |
700e432d | 4476 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4477 | conf->seq_write = conf->seq_flush; |
72626685 N |
4478 | activate_bit_delay(conf); |
4479 | } | |
7c13edc8 N |
4480 | if (atomic_read(&mddev->plug_cnt) == 0) |
4481 | raid5_activate_delayed(conf); | |
72626685 | 4482 | |
46031f9a RBJ |
4483 | while ((bio = remove_bio_from_retry(conf))) { |
4484 | int ok; | |
4485 | spin_unlock_irq(&conf->device_lock); | |
4486 | ok = retry_aligned_read(conf, bio); | |
4487 | spin_lock_irq(&conf->device_lock); | |
4488 | if (!ok) | |
4489 | break; | |
4490 | handled++; | |
4491 | } | |
4492 | ||
8b3e6cdc DW |
4493 | sh = __get_priority_stripe(conf); |
4494 | ||
c9f21aaf | 4495 | if (!sh) |
1da177e4 | 4496 | break; |
1da177e4 LT |
4497 | spin_unlock_irq(&conf->device_lock); |
4498 | ||
4499 | handled++; | |
417b8d4a DW |
4500 | handle_stripe(sh); |
4501 | release_stripe(sh); | |
4502 | cond_resched(); | |
1da177e4 | 4503 | |
de393cde N |
4504 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
4505 | md_check_recovery(mddev); | |
4506 | ||
1da177e4 LT |
4507 | spin_lock_irq(&conf->device_lock); |
4508 | } | |
45b4233c | 4509 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4510 | |
4511 | spin_unlock_irq(&conf->device_lock); | |
4512 | ||
c9f21aaf | 4513 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4514 | blk_finish_plug(&plug); |
1da177e4 | 4515 | |
45b4233c | 4516 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4517 | } |
4518 | ||
3f294f4f | 4519 | static ssize_t |
fd01b88c | 4520 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 4521 | { |
d1688a6d | 4522 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4523 | if (conf) |
4524 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4525 | else | |
4526 | return 0; | |
3f294f4f N |
4527 | } |
4528 | ||
c41d4ac4 | 4529 | int |
fd01b88c | 4530 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 4531 | { |
d1688a6d | 4532 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
4533 | int err; |
4534 | ||
c41d4ac4 | 4535 | if (size <= 16 || size > 32768) |
3f294f4f | 4536 | return -EINVAL; |
c41d4ac4 | 4537 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4538 | if (drop_one_stripe(conf)) |
4539 | conf->max_nr_stripes--; | |
4540 | else | |
4541 | break; | |
4542 | } | |
b5470dc5 DW |
4543 | err = md_allow_write(mddev); |
4544 | if (err) | |
4545 | return err; | |
c41d4ac4 | 4546 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4547 | if (grow_one_stripe(conf)) |
4548 | conf->max_nr_stripes++; | |
4549 | else break; | |
4550 | } | |
c41d4ac4 N |
4551 | return 0; |
4552 | } | |
4553 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4554 | ||
4555 | static ssize_t | |
fd01b88c | 4556 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 4557 | { |
d1688a6d | 4558 | struct r5conf *conf = mddev->private; |
c41d4ac4 N |
4559 | unsigned long new; |
4560 | int err; | |
4561 | ||
4562 | if (len >= PAGE_SIZE) | |
4563 | return -EINVAL; | |
4564 | if (!conf) | |
4565 | return -ENODEV; | |
4566 | ||
4567 | if (strict_strtoul(page, 10, &new)) | |
4568 | return -EINVAL; | |
4569 | err = raid5_set_cache_size(mddev, new); | |
4570 | if (err) | |
4571 | return err; | |
3f294f4f N |
4572 | return len; |
4573 | } | |
007583c9 | 4574 | |
96de1e66 N |
4575 | static struct md_sysfs_entry |
4576 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4577 | raid5_show_stripe_cache_size, | |
4578 | raid5_store_stripe_cache_size); | |
3f294f4f | 4579 | |
8b3e6cdc | 4580 | static ssize_t |
fd01b88c | 4581 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 4582 | { |
d1688a6d | 4583 | struct r5conf *conf = mddev->private; |
8b3e6cdc DW |
4584 | if (conf) |
4585 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4586 | else | |
4587 | return 0; | |
4588 | } | |
4589 | ||
4590 | static ssize_t | |
fd01b88c | 4591 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 4592 | { |
d1688a6d | 4593 | struct r5conf *conf = mddev->private; |
4ef197d8 | 4594 | unsigned long new; |
8b3e6cdc DW |
4595 | if (len >= PAGE_SIZE) |
4596 | return -EINVAL; | |
4597 | if (!conf) | |
4598 | return -ENODEV; | |
4599 | ||
4ef197d8 | 4600 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4601 | return -EINVAL; |
4ef197d8 | 4602 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4603 | return -EINVAL; |
4604 | conf->bypass_threshold = new; | |
4605 | return len; | |
4606 | } | |
4607 | ||
4608 | static struct md_sysfs_entry | |
4609 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4610 | S_IRUGO | S_IWUSR, | |
4611 | raid5_show_preread_threshold, | |
4612 | raid5_store_preread_threshold); | |
4613 | ||
3f294f4f | 4614 | static ssize_t |
fd01b88c | 4615 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 4616 | { |
d1688a6d | 4617 | struct r5conf *conf = mddev->private; |
96de1e66 N |
4618 | if (conf) |
4619 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4620 | else | |
4621 | return 0; | |
3f294f4f N |
4622 | } |
4623 | ||
96de1e66 N |
4624 | static struct md_sysfs_entry |
4625 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4626 | |
007583c9 | 4627 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4628 | &raid5_stripecache_size.attr, |
4629 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4630 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4631 | NULL, |
4632 | }; | |
007583c9 N |
4633 | static struct attribute_group raid5_attrs_group = { |
4634 | .name = NULL, | |
4635 | .attrs = raid5_attrs, | |
3f294f4f N |
4636 | }; |
4637 | ||
80c3a6ce | 4638 | static sector_t |
fd01b88c | 4639 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 4640 | { |
d1688a6d | 4641 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
4642 | |
4643 | if (!sectors) | |
4644 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4645 | if (!raid_disks) |
7ec05478 | 4646 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4647 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4648 | |
9d8f0363 | 4649 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4650 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4651 | return sectors * (raid_disks - conf->max_degraded); |
4652 | } | |
4653 | ||
d1688a6d | 4654 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 DW |
4655 | { |
4656 | struct raid5_percpu *percpu; | |
4657 | unsigned long cpu; | |
4658 | ||
4659 | if (!conf->percpu) | |
4660 | return; | |
4661 | ||
4662 | get_online_cpus(); | |
4663 | for_each_possible_cpu(cpu) { | |
4664 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4665 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4666 | kfree(percpu->scribble); |
36d1c647 DW |
4667 | } |
4668 | #ifdef CONFIG_HOTPLUG_CPU | |
4669 | unregister_cpu_notifier(&conf->cpu_notify); | |
4670 | #endif | |
4671 | put_online_cpus(); | |
4672 | ||
4673 | free_percpu(conf->percpu); | |
4674 | } | |
4675 | ||
d1688a6d | 4676 | static void free_conf(struct r5conf *conf) |
95fc17aa DW |
4677 | { |
4678 | shrink_stripes(conf); | |
36d1c647 | 4679 | raid5_free_percpu(conf); |
95fc17aa DW |
4680 | kfree(conf->disks); |
4681 | kfree(conf->stripe_hashtbl); | |
4682 | kfree(conf); | |
4683 | } | |
4684 | ||
36d1c647 DW |
4685 | #ifdef CONFIG_HOTPLUG_CPU |
4686 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4687 | void *hcpu) | |
4688 | { | |
d1688a6d | 4689 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
4690 | long cpu = (long)hcpu; |
4691 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4692 | ||
4693 | switch (action) { | |
4694 | case CPU_UP_PREPARE: | |
4695 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4696 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4697 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4698 | if (!percpu->scribble) |
4699 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4700 | ||
4701 | if (!percpu->scribble || | |
4702 | (conf->level == 6 && !percpu->spare_page)) { | |
4703 | safe_put_page(percpu->spare_page); | |
4704 | kfree(percpu->scribble); | |
36d1c647 DW |
4705 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4706 | __func__, cpu); | |
55af6bb5 | 4707 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4708 | } |
4709 | break; | |
4710 | case CPU_DEAD: | |
4711 | case CPU_DEAD_FROZEN: | |
4712 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4713 | kfree(percpu->scribble); |
36d1c647 | 4714 | percpu->spare_page = NULL; |
d6f38f31 | 4715 | percpu->scribble = NULL; |
36d1c647 DW |
4716 | break; |
4717 | default: | |
4718 | break; | |
4719 | } | |
4720 | return NOTIFY_OK; | |
4721 | } | |
4722 | #endif | |
4723 | ||
d1688a6d | 4724 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
4725 | { |
4726 | unsigned long cpu; | |
4727 | struct page *spare_page; | |
a29d8b8e | 4728 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4729 | void *scribble; |
36d1c647 DW |
4730 | int err; |
4731 | ||
36d1c647 DW |
4732 | allcpus = alloc_percpu(struct raid5_percpu); |
4733 | if (!allcpus) | |
4734 | return -ENOMEM; | |
4735 | conf->percpu = allcpus; | |
4736 | ||
4737 | get_online_cpus(); | |
4738 | err = 0; | |
4739 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4740 | if (conf->level == 6) { |
4741 | spare_page = alloc_page(GFP_KERNEL); | |
4742 | if (!spare_page) { | |
4743 | err = -ENOMEM; | |
4744 | break; | |
4745 | } | |
4746 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4747 | } | |
5e5e3e78 | 4748 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4749 | if (!scribble) { |
36d1c647 DW |
4750 | err = -ENOMEM; |
4751 | break; | |
4752 | } | |
d6f38f31 | 4753 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4754 | } |
4755 | #ifdef CONFIG_HOTPLUG_CPU | |
4756 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4757 | conf->cpu_notify.priority = 0; | |
4758 | if (err == 0) | |
4759 | err = register_cpu_notifier(&conf->cpu_notify); | |
4760 | #endif | |
4761 | put_online_cpus(); | |
4762 | ||
4763 | return err; | |
4764 | } | |
4765 | ||
d1688a6d | 4766 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 4767 | { |
d1688a6d | 4768 | struct r5conf *conf; |
5e5e3e78 | 4769 | int raid_disk, memory, max_disks; |
3cb03002 | 4770 | struct md_rdev *rdev; |
1da177e4 | 4771 | struct disk_info *disk; |
1da177e4 | 4772 | |
91adb564 N |
4773 | if (mddev->new_level != 5 |
4774 | && mddev->new_level != 4 | |
4775 | && mddev->new_level != 6) { | |
0c55e022 | 4776 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4777 | mdname(mddev), mddev->new_level); |
4778 | return ERR_PTR(-EIO); | |
1da177e4 | 4779 | } |
91adb564 N |
4780 | if ((mddev->new_level == 5 |
4781 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4782 | (mddev->new_level == 6 | |
4783 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4784 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4785 | mdname(mddev), mddev->new_layout); |
4786 | return ERR_PTR(-EIO); | |
99c0fb5f | 4787 | } |
91adb564 | 4788 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4789 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4790 | mdname(mddev), mddev->raid_disks); |
4791 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4792 | } |
4793 | ||
664e7c41 AN |
4794 | if (!mddev->new_chunk_sectors || |
4795 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4796 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4797 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4798 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4799 | return ERR_PTR(-EINVAL); |
f6705578 N |
4800 | } |
4801 | ||
d1688a6d | 4802 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 4803 | if (conf == NULL) |
1da177e4 | 4804 | goto abort; |
f5efd45a DW |
4805 | spin_lock_init(&conf->device_lock); |
4806 | init_waitqueue_head(&conf->wait_for_stripe); | |
4807 | init_waitqueue_head(&conf->wait_for_overlap); | |
4808 | INIT_LIST_HEAD(&conf->handle_list); | |
4809 | INIT_LIST_HEAD(&conf->hold_list); | |
4810 | INIT_LIST_HEAD(&conf->delayed_list); | |
4811 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4812 | INIT_LIST_HEAD(&conf->inactive_list); | |
4813 | atomic_set(&conf->active_stripes, 0); | |
4814 | atomic_set(&conf->preread_active_stripes, 0); | |
4815 | atomic_set(&conf->active_aligned_reads, 0); | |
4816 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 4817 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
4818 | |
4819 | conf->raid_disks = mddev->raid_disks; | |
4820 | if (mddev->reshape_position == MaxSector) | |
4821 | conf->previous_raid_disks = mddev->raid_disks; | |
4822 | else | |
f6705578 | 4823 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4824 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4825 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4826 | |
5e5e3e78 | 4827 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4828 | GFP_KERNEL); |
4829 | if (!conf->disks) | |
4830 | goto abort; | |
9ffae0cf | 4831 | |
1da177e4 LT |
4832 | conf->mddev = mddev; |
4833 | ||
fccddba0 | 4834 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4835 | goto abort; |
1da177e4 | 4836 | |
36d1c647 DW |
4837 | conf->level = mddev->new_level; |
4838 | if (raid5_alloc_percpu(conf) != 0) | |
4839 | goto abort; | |
4840 | ||
0c55e022 | 4841 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4842 | |
159ec1fc | 4843 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4844 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4845 | if (raid_disk >= max_disks |
1da177e4 LT |
4846 | || raid_disk < 0) |
4847 | continue; | |
4848 | disk = conf->disks + raid_disk; | |
4849 | ||
17045f52 N |
4850 | if (test_bit(Replacement, &rdev->flags)) { |
4851 | if (disk->replacement) | |
4852 | goto abort; | |
4853 | disk->replacement = rdev; | |
4854 | } else { | |
4855 | if (disk->rdev) | |
4856 | goto abort; | |
4857 | disk->rdev = rdev; | |
4858 | } | |
1da177e4 | 4859 | |
b2d444d7 | 4860 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 4861 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
4862 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
4863 | " disk %d\n", | |
4864 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 4865 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
4866 | /* Cannot rely on bitmap to complete recovery */ |
4867 | conf->fullsync = 1; | |
1da177e4 LT |
4868 | } |
4869 | ||
09c9e5fa | 4870 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4871 | conf->level = mddev->new_level; |
16a53ecc N |
4872 | if (conf->level == 6) |
4873 | conf->max_degraded = 2; | |
4874 | else | |
4875 | conf->max_degraded = 1; | |
91adb564 | 4876 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4877 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4878 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4879 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4880 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4881 | conf->prev_algo = mddev->layout; |
4882 | } | |
1da177e4 | 4883 | |
91adb564 | 4884 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4885 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4886 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4887 | printk(KERN_ERR | |
0c55e022 N |
4888 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
4889 | mdname(mddev), memory); | |
91adb564 N |
4890 | goto abort; |
4891 | } else | |
0c55e022 N |
4892 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
4893 | mdname(mddev), memory); | |
1da177e4 | 4894 | |
0da3c619 | 4895 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4896 | if (!conf->thread) { |
4897 | printk(KERN_ERR | |
0c55e022 | 4898 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 4899 | mdname(mddev)); |
16a53ecc N |
4900 | goto abort; |
4901 | } | |
91adb564 N |
4902 | |
4903 | return conf; | |
4904 | ||
4905 | abort: | |
4906 | if (conf) { | |
95fc17aa | 4907 | free_conf(conf); |
91adb564 N |
4908 | return ERR_PTR(-EIO); |
4909 | } else | |
4910 | return ERR_PTR(-ENOMEM); | |
4911 | } | |
4912 | ||
c148ffdc N |
4913 | |
4914 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4915 | { | |
4916 | switch (algo) { | |
4917 | case ALGORITHM_PARITY_0: | |
4918 | if (raid_disk < max_degraded) | |
4919 | return 1; | |
4920 | break; | |
4921 | case ALGORITHM_PARITY_N: | |
4922 | if (raid_disk >= raid_disks - max_degraded) | |
4923 | return 1; | |
4924 | break; | |
4925 | case ALGORITHM_PARITY_0_6: | |
4926 | if (raid_disk == 0 || | |
4927 | raid_disk == raid_disks - 1) | |
4928 | return 1; | |
4929 | break; | |
4930 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4931 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4932 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4933 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4934 | if (raid_disk == raid_disks - 1) | |
4935 | return 1; | |
4936 | } | |
4937 | return 0; | |
4938 | } | |
4939 | ||
fd01b88c | 4940 | static int run(struct mddev *mddev) |
91adb564 | 4941 | { |
d1688a6d | 4942 | struct r5conf *conf; |
9f7c2220 | 4943 | int working_disks = 0; |
c148ffdc | 4944 | int dirty_parity_disks = 0; |
3cb03002 | 4945 | struct md_rdev *rdev; |
c148ffdc | 4946 | sector_t reshape_offset = 0; |
17045f52 | 4947 | int i; |
91adb564 | 4948 | |
8c6ac868 | 4949 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 4950 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
4951 | " -- starting background reconstruction\n", |
4952 | mdname(mddev)); | |
91adb564 N |
4953 | if (mddev->reshape_position != MaxSector) { |
4954 | /* Check that we can continue the reshape. | |
4955 | * Currently only disks can change, it must | |
4956 | * increase, and we must be past the point where | |
4957 | * a stripe over-writes itself | |
4958 | */ | |
4959 | sector_t here_new, here_old; | |
4960 | int old_disks; | |
18b00334 | 4961 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4962 | |
88ce4930 | 4963 | if (mddev->new_level != mddev->level) { |
0c55e022 | 4964 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
4965 | "required - aborting.\n", |
4966 | mdname(mddev)); | |
4967 | return -EINVAL; | |
4968 | } | |
91adb564 N |
4969 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4970 | /* reshape_position must be on a new-stripe boundary, and one | |
4971 | * further up in new geometry must map after here in old | |
4972 | * geometry. | |
4973 | */ | |
4974 | here_new = mddev->reshape_position; | |
664e7c41 | 4975 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 4976 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
4977 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
4978 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
4979 | return -EINVAL; |
4980 | } | |
c148ffdc | 4981 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4982 | /* here_new is the stripe we will write to */ |
4983 | here_old = mddev->reshape_position; | |
9d8f0363 | 4984 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4985 | (old_disks-max_degraded)); |
4986 | /* here_old is the first stripe that we might need to read | |
4987 | * from */ | |
67ac6011 N |
4988 | if (mddev->delta_disks == 0) { |
4989 | /* We cannot be sure it is safe to start an in-place | |
4990 | * reshape. It is only safe if user-space if monitoring | |
4991 | * and taking constant backups. | |
4992 | * mdadm always starts a situation like this in | |
4993 | * readonly mode so it can take control before | |
4994 | * allowing any writes. So just check for that. | |
4995 | */ | |
4996 | if ((here_new * mddev->new_chunk_sectors != | |
4997 | here_old * mddev->chunk_sectors) || | |
4998 | mddev->ro == 0) { | |
0c55e022 N |
4999 | printk(KERN_ERR "md/raid:%s: in-place reshape must be started" |
5000 | " in read-only mode - aborting\n", | |
5001 | mdname(mddev)); | |
67ac6011 N |
5002 | return -EINVAL; |
5003 | } | |
5004 | } else if (mddev->delta_disks < 0 | |
5005 | ? (here_new * mddev->new_chunk_sectors <= | |
5006 | here_old * mddev->chunk_sectors) | |
5007 | : (here_new * mddev->new_chunk_sectors >= | |
5008 | here_old * mddev->chunk_sectors)) { | |
91adb564 | 5009 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
5010 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
5011 | "auto-recovery - aborting.\n", | |
5012 | mdname(mddev)); | |
91adb564 N |
5013 | return -EINVAL; |
5014 | } | |
0c55e022 N |
5015 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
5016 | mdname(mddev)); | |
91adb564 N |
5017 | /* OK, we should be able to continue; */ |
5018 | } else { | |
5019 | BUG_ON(mddev->level != mddev->new_level); | |
5020 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 5021 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 5022 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 5023 | } |
91adb564 | 5024 | |
245f46c2 N |
5025 | if (mddev->private == NULL) |
5026 | conf = setup_conf(mddev); | |
5027 | else | |
5028 | conf = mddev->private; | |
5029 | ||
91adb564 N |
5030 | if (IS_ERR(conf)) |
5031 | return PTR_ERR(conf); | |
5032 | ||
5033 | mddev->thread = conf->thread; | |
5034 | conf->thread = NULL; | |
5035 | mddev->private = conf; | |
5036 | ||
17045f52 N |
5037 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
5038 | i++) { | |
5039 | rdev = conf->disks[i].rdev; | |
5040 | if (!rdev && conf->disks[i].replacement) { | |
5041 | /* The replacement is all we have yet */ | |
5042 | rdev = conf->disks[i].replacement; | |
5043 | conf->disks[i].replacement = NULL; | |
5044 | clear_bit(Replacement, &rdev->flags); | |
5045 | conf->disks[i].rdev = rdev; | |
5046 | } | |
5047 | if (!rdev) | |
c148ffdc | 5048 | continue; |
17045f52 N |
5049 | if (conf->disks[i].replacement && |
5050 | conf->reshape_progress != MaxSector) { | |
5051 | /* replacements and reshape simply do not mix. */ | |
5052 | printk(KERN_ERR "md: cannot handle concurrent " | |
5053 | "replacement and reshape.\n"); | |
5054 | goto abort; | |
5055 | } | |
2f115882 | 5056 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 5057 | working_disks++; |
2f115882 N |
5058 | continue; |
5059 | } | |
c148ffdc N |
5060 | /* This disc is not fully in-sync. However if it |
5061 | * just stored parity (beyond the recovery_offset), | |
5062 | * when we don't need to be concerned about the | |
5063 | * array being dirty. | |
5064 | * When reshape goes 'backwards', we never have | |
5065 | * partially completed devices, so we only need | |
5066 | * to worry about reshape going forwards. | |
5067 | */ | |
5068 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
5069 | if (mddev->major_version == 0 && | |
5070 | mddev->minor_version > 90) | |
5071 | rdev->recovery_offset = reshape_offset; | |
5072 | ||
c148ffdc N |
5073 | if (rdev->recovery_offset < reshape_offset) { |
5074 | /* We need to check old and new layout */ | |
5075 | if (!only_parity(rdev->raid_disk, | |
5076 | conf->algorithm, | |
5077 | conf->raid_disks, | |
5078 | conf->max_degraded)) | |
5079 | continue; | |
5080 | } | |
5081 | if (!only_parity(rdev->raid_disk, | |
5082 | conf->prev_algo, | |
5083 | conf->previous_raid_disks, | |
5084 | conf->max_degraded)) | |
5085 | continue; | |
5086 | dirty_parity_disks++; | |
5087 | } | |
91adb564 | 5088 | |
17045f52 N |
5089 | /* |
5090 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
5091 | */ | |
908f4fbd | 5092 | mddev->degraded = calc_degraded(conf); |
91adb564 | 5093 | |
674806d6 | 5094 | if (has_failed(conf)) { |
0c55e022 | 5095 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 5096 | " (%d/%d failed)\n", |
02c2de8c | 5097 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
5098 | goto abort; |
5099 | } | |
5100 | ||
91adb564 | 5101 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 5102 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
5103 | mddev->resync_max_sectors = mddev->dev_sectors; |
5104 | ||
c148ffdc | 5105 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 5106 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
5107 | if (mddev->ok_start_degraded) |
5108 | printk(KERN_WARNING | |
0c55e022 N |
5109 | "md/raid:%s: starting dirty degraded array" |
5110 | " - data corruption possible.\n", | |
6ff8d8ec N |
5111 | mdname(mddev)); |
5112 | else { | |
5113 | printk(KERN_ERR | |
0c55e022 | 5114 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
5115 | mdname(mddev)); |
5116 | goto abort; | |
5117 | } | |
1da177e4 LT |
5118 | } |
5119 | ||
1da177e4 | 5120 | if (mddev->degraded == 0) |
0c55e022 N |
5121 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
5122 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
5123 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
5124 | mddev->new_layout); | |
1da177e4 | 5125 | else |
0c55e022 N |
5126 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
5127 | " out of %d devices, algorithm %d\n", | |
5128 | mdname(mddev), conf->level, | |
5129 | mddev->raid_disks - mddev->degraded, | |
5130 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
5131 | |
5132 | print_raid5_conf(conf); | |
5133 | ||
fef9c61f | 5134 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 5135 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5136 | atomic_set(&conf->reshape_stripes, 0); |
5137 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5138 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5139 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5140 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5141 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5142 | "reshape"); |
f6705578 N |
5143 | } |
5144 | ||
1da177e4 LT |
5145 | |
5146 | /* Ok, everything is just fine now */ | |
a64c876f N |
5147 | if (mddev->to_remove == &raid5_attrs_group) |
5148 | mddev->to_remove = NULL; | |
00bcb4ac N |
5149 | else if (mddev->kobj.sd && |
5150 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 5151 | printk(KERN_WARNING |
4a5add49 | 5152 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 5153 | mdname(mddev)); |
4a5add49 | 5154 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5155 | |
4a5add49 | 5156 | if (mddev->queue) { |
9f7c2220 | 5157 | int chunk_size; |
4a5add49 N |
5158 | /* read-ahead size must cover two whole stripes, which |
5159 | * is 2 * (datadisks) * chunksize where 'n' is the | |
5160 | * number of raid devices | |
5161 | */ | |
5162 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
5163 | int stripe = data_disks * | |
5164 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
5165 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
5166 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 5167 | |
4a5add49 | 5168 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 5169 | |
11d8a6e3 N |
5170 | mddev->queue->backing_dev_info.congested_data = mddev; |
5171 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 5172 | |
9f7c2220 N |
5173 | chunk_size = mddev->chunk_sectors << 9; |
5174 | blk_queue_io_min(mddev->queue, chunk_size); | |
5175 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5176 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 5177 | |
9f7c2220 N |
5178 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5179 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
5180 | rdev->data_offset << 9); | |
5181 | } | |
23032a0e | 5182 | |
1da177e4 LT |
5183 | return 0; |
5184 | abort: | |
01f96c0a | 5185 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
5186 | print_raid5_conf(conf); |
5187 | free_conf(conf); | |
1da177e4 | 5188 | mddev->private = NULL; |
0c55e022 | 5189 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
5190 | return -EIO; |
5191 | } | |
5192 | ||
fd01b88c | 5193 | static int stop(struct mddev *mddev) |
1da177e4 | 5194 | { |
d1688a6d | 5195 | struct r5conf *conf = mddev->private; |
1da177e4 | 5196 | |
01f96c0a | 5197 | md_unregister_thread(&mddev->thread); |
11d8a6e3 N |
5198 | if (mddev->queue) |
5199 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 5200 | free_conf(conf); |
a64c876f N |
5201 | mddev->private = NULL; |
5202 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
5203 | return 0; |
5204 | } | |
5205 | ||
fd01b88c | 5206 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 5207 | { |
d1688a6d | 5208 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
5209 | int i; |
5210 | ||
9d8f0363 AN |
5211 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5212 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5213 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5214 | for (i = 0; i < conf->raid_disks; i++) |
5215 | seq_printf (seq, "%s", | |
5216 | conf->disks[i].rdev && | |
b2d444d7 | 5217 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5218 | seq_printf (seq, "]"); |
1da177e4 LT |
5219 | } |
5220 | ||
d1688a6d | 5221 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
5222 | { |
5223 | int i; | |
5224 | struct disk_info *tmp; | |
5225 | ||
0c55e022 | 5226 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
5227 | if (!conf) { |
5228 | printk("(conf==NULL)\n"); | |
5229 | return; | |
5230 | } | |
0c55e022 N |
5231 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
5232 | conf->raid_disks, | |
5233 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5234 | |
5235 | for (i = 0; i < conf->raid_disks; i++) { | |
5236 | char b[BDEVNAME_SIZE]; | |
5237 | tmp = conf->disks + i; | |
5238 | if (tmp->rdev) | |
0c55e022 N |
5239 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
5240 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
5241 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
5242 | } |
5243 | } | |
5244 | ||
fd01b88c | 5245 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
5246 | { |
5247 | int i; | |
d1688a6d | 5248 | struct r5conf *conf = mddev->private; |
1da177e4 | 5249 | struct disk_info *tmp; |
6b965620 N |
5250 | int count = 0; |
5251 | unsigned long flags; | |
1da177e4 LT |
5252 | |
5253 | for (i = 0; i < conf->raid_disks; i++) { | |
5254 | tmp = conf->disks + i; | |
dd054fce N |
5255 | if (tmp->replacement |
5256 | && tmp->replacement->recovery_offset == MaxSector | |
5257 | && !test_bit(Faulty, &tmp->replacement->flags) | |
5258 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
5259 | /* Replacement has just become active. */ | |
5260 | if (!tmp->rdev | |
5261 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
5262 | count++; | |
5263 | if (tmp->rdev) { | |
5264 | /* Replaced device not technically faulty, | |
5265 | * but we need to be sure it gets removed | |
5266 | * and never re-added. | |
5267 | */ | |
5268 | set_bit(Faulty, &tmp->rdev->flags); | |
5269 | sysfs_notify_dirent_safe( | |
5270 | tmp->rdev->sysfs_state); | |
5271 | } | |
5272 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
5273 | } else if (tmp->rdev | |
70fffd0b | 5274 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 5275 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 5276 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 5277 | count++; |
43c73ca4 | 5278 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
5279 | } |
5280 | } | |
6b965620 | 5281 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5282 | mddev->degraded = calc_degraded(conf); |
6b965620 | 5283 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 5284 | print_raid5_conf(conf); |
6b965620 | 5285 | return count; |
1da177e4 LT |
5286 | } |
5287 | ||
b8321b68 | 5288 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5289 | { |
d1688a6d | 5290 | struct r5conf *conf = mddev->private; |
1da177e4 | 5291 | int err = 0; |
b8321b68 | 5292 | int number = rdev->raid_disk; |
657e3e4d | 5293 | struct md_rdev **rdevp; |
1da177e4 LT |
5294 | struct disk_info *p = conf->disks + number; |
5295 | ||
5296 | print_raid5_conf(conf); | |
657e3e4d N |
5297 | if (rdev == p->rdev) |
5298 | rdevp = &p->rdev; | |
5299 | else if (rdev == p->replacement) | |
5300 | rdevp = &p->replacement; | |
5301 | else | |
5302 | return 0; | |
5303 | ||
5304 | if (number >= conf->raid_disks && | |
5305 | conf->reshape_progress == MaxSector) | |
5306 | clear_bit(In_sync, &rdev->flags); | |
5307 | ||
5308 | if (test_bit(In_sync, &rdev->flags) || | |
5309 | atomic_read(&rdev->nr_pending)) { | |
5310 | err = -EBUSY; | |
5311 | goto abort; | |
5312 | } | |
5313 | /* Only remove non-faulty devices if recovery | |
5314 | * isn't possible. | |
5315 | */ | |
5316 | if (!test_bit(Faulty, &rdev->flags) && | |
5317 | mddev->recovery_disabled != conf->recovery_disabled && | |
5318 | !has_failed(conf) && | |
dd054fce | 5319 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
5320 | number < conf->raid_disks) { |
5321 | err = -EBUSY; | |
5322 | goto abort; | |
5323 | } | |
5324 | *rdevp = NULL; | |
5325 | synchronize_rcu(); | |
5326 | if (atomic_read(&rdev->nr_pending)) { | |
5327 | /* lost the race, try later */ | |
5328 | err = -EBUSY; | |
5329 | *rdevp = rdev; | |
dd054fce N |
5330 | } else if (p->replacement) { |
5331 | /* We must have just cleared 'rdev' */ | |
5332 | p->rdev = p->replacement; | |
5333 | clear_bit(Replacement, &p->replacement->flags); | |
5334 | smp_mb(); /* Make sure other CPUs may see both as identical | |
5335 | * but will never see neither - if they are careful | |
5336 | */ | |
5337 | p->replacement = NULL; | |
5338 | clear_bit(WantReplacement, &rdev->flags); | |
5339 | } else | |
5340 | /* We might have just removed the Replacement as faulty- | |
5341 | * clear the bit just in case | |
5342 | */ | |
5343 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
5344 | abort: |
5345 | ||
5346 | print_raid5_conf(conf); | |
5347 | return err; | |
5348 | } | |
5349 | ||
fd01b88c | 5350 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 5351 | { |
d1688a6d | 5352 | struct r5conf *conf = mddev->private; |
199050ea | 5353 | int err = -EEXIST; |
1da177e4 LT |
5354 | int disk; |
5355 | struct disk_info *p; | |
6c2fce2e NB |
5356 | int first = 0; |
5357 | int last = conf->raid_disks - 1; | |
1da177e4 | 5358 | |
7f0da59b N |
5359 | if (mddev->recovery_disabled == conf->recovery_disabled) |
5360 | return -EBUSY; | |
5361 | ||
674806d6 | 5362 | if (has_failed(conf)) |
1da177e4 | 5363 | /* no point adding a device */ |
199050ea | 5364 | return -EINVAL; |
1da177e4 | 5365 | |
6c2fce2e NB |
5366 | if (rdev->raid_disk >= 0) |
5367 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5368 | |
5369 | /* | |
16a53ecc N |
5370 | * find the disk ... but prefer rdev->saved_raid_disk |
5371 | * if possible. | |
1da177e4 | 5372 | */ |
16a53ecc | 5373 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5374 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5375 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5376 | disk = rdev->saved_raid_disk; | |
5377 | else | |
6c2fce2e | 5378 | disk = first; |
7bfec5f3 N |
5379 | for ( ; disk <= last ; disk++) { |
5380 | p = conf->disks + disk; | |
5381 | if (p->rdev == NULL) { | |
b2d444d7 | 5382 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5383 | rdev->raid_disk = disk; |
199050ea | 5384 | err = 0; |
72626685 N |
5385 | if (rdev->saved_raid_disk != disk) |
5386 | conf->fullsync = 1; | |
d6065f7b | 5387 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5388 | break; |
5389 | } | |
7bfec5f3 N |
5390 | if (test_bit(WantReplacement, &p->rdev->flags) && |
5391 | p->replacement == NULL) { | |
5392 | clear_bit(In_sync, &rdev->flags); | |
5393 | set_bit(Replacement, &rdev->flags); | |
5394 | rdev->raid_disk = disk; | |
5395 | err = 0; | |
5396 | conf->fullsync = 1; | |
5397 | rcu_assign_pointer(p->replacement, rdev); | |
5398 | break; | |
5399 | } | |
5400 | } | |
1da177e4 | 5401 | print_raid5_conf(conf); |
199050ea | 5402 | return err; |
1da177e4 LT |
5403 | } |
5404 | ||
fd01b88c | 5405 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
5406 | { |
5407 | /* no resync is happening, and there is enough space | |
5408 | * on all devices, so we can resize. | |
5409 | * We need to make sure resync covers any new space. | |
5410 | * If the array is shrinking we should possibly wait until | |
5411 | * any io in the removed space completes, but it hardly seems | |
5412 | * worth it. | |
5413 | */ | |
9d8f0363 | 5414 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5415 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5416 | mddev->raid_disks)); | |
b522adcd DW |
5417 | if (mddev->array_sectors > |
5418 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5419 | return -EINVAL; | |
f233ea5c | 5420 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5421 | revalidate_disk(mddev->gendisk); |
b098636c N |
5422 | if (sectors > mddev->dev_sectors && |
5423 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5424 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5425 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5426 | } | |
58c0fed4 | 5427 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5428 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5429 | return 0; |
5430 | } | |
5431 | ||
fd01b88c | 5432 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
5433 | { |
5434 | /* Can only proceed if there are plenty of stripe_heads. | |
5435 | * We need a minimum of one full stripe,, and for sensible progress | |
5436 | * it is best to have about 4 times that. | |
5437 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5438 | * allow for chunk sizes up to 256K, which is probably OK. | |
5439 | * If the chunk size is greater, user-space should request more | |
5440 | * stripe_heads first. | |
5441 | */ | |
d1688a6d | 5442 | struct r5conf *conf = mddev->private; |
01ee22b4 N |
5443 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
5444 | > conf->max_nr_stripes || | |
5445 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5446 | > conf->max_nr_stripes) { | |
0c55e022 N |
5447 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5448 | mdname(mddev), | |
01ee22b4 N |
5449 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5450 | / STRIPE_SIZE)*4); | |
5451 | return 0; | |
5452 | } | |
5453 | return 1; | |
5454 | } | |
5455 | ||
fd01b88c | 5456 | static int check_reshape(struct mddev *mddev) |
29269553 | 5457 | { |
d1688a6d | 5458 | struct r5conf *conf = mddev->private; |
29269553 | 5459 | |
88ce4930 N |
5460 | if (mddev->delta_disks == 0 && |
5461 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5462 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5463 | return 0; /* nothing to do */ |
dba034ee N |
5464 | if (mddev->bitmap) |
5465 | /* Cannot grow a bitmap yet */ | |
5466 | return -EBUSY; | |
674806d6 | 5467 | if (has_failed(conf)) |
ec32a2bd N |
5468 | return -EINVAL; |
5469 | if (mddev->delta_disks < 0) { | |
5470 | /* We might be able to shrink, but the devices must | |
5471 | * be made bigger first. | |
5472 | * For raid6, 4 is the minimum size. | |
5473 | * Otherwise 2 is the minimum | |
5474 | */ | |
5475 | int min = 2; | |
5476 | if (mddev->level == 6) | |
5477 | min = 4; | |
5478 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5479 | return -EINVAL; | |
5480 | } | |
29269553 | 5481 | |
01ee22b4 | 5482 | if (!check_stripe_cache(mddev)) |
29269553 | 5483 | return -ENOSPC; |
29269553 | 5484 | |
ec32a2bd | 5485 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5486 | } |
5487 | ||
fd01b88c | 5488 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 5489 | { |
d1688a6d | 5490 | struct r5conf *conf = mddev->private; |
3cb03002 | 5491 | struct md_rdev *rdev; |
63c70c4f | 5492 | int spares = 0; |
c04be0aa | 5493 | unsigned long flags; |
63c70c4f | 5494 | |
f416885e | 5495 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5496 | return -EBUSY; |
5497 | ||
01ee22b4 N |
5498 | if (!check_stripe_cache(mddev)) |
5499 | return -ENOSPC; | |
5500 | ||
159ec1fc | 5501 | list_for_each_entry(rdev, &mddev->disks, same_set) |
469518a3 N |
5502 | if (!test_bit(In_sync, &rdev->flags) |
5503 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5504 | spares++; |
63c70c4f | 5505 | |
f416885e | 5506 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5507 | /* Not enough devices even to make a degraded array |
5508 | * of that size | |
5509 | */ | |
5510 | return -EINVAL; | |
5511 | ||
ec32a2bd N |
5512 | /* Refuse to reduce size of the array. Any reductions in |
5513 | * array size must be through explicit setting of array_size | |
5514 | * attribute. | |
5515 | */ | |
5516 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5517 | < mddev->array_sectors) { | |
0c55e022 | 5518 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5519 | "before number of disks\n", mdname(mddev)); |
5520 | return -EINVAL; | |
5521 | } | |
5522 | ||
f6705578 | 5523 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5524 | spin_lock_irq(&conf->device_lock); |
5525 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5526 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5527 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5528 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5529 | conf->prev_algo = conf->algorithm; |
5530 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5531 | if (mddev->delta_disks < 0) |
5532 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5533 | else | |
5534 | conf->reshape_progress = 0; | |
5535 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5536 | conf->generation++; |
29269553 N |
5537 | spin_unlock_irq(&conf->device_lock); |
5538 | ||
5539 | /* Add some new drives, as many as will fit. | |
5540 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5541 | * Don't add devices if we are reducing the number of |
5542 | * devices in the array. This is because it is not possible | |
5543 | * to correctly record the "partially reconstructed" state of | |
5544 | * such devices during the reshape and confusion could result. | |
29269553 | 5545 | */ |
87a8dec9 N |
5546 | if (mddev->delta_disks >= 0) { |
5547 | int added_devices = 0; | |
5548 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5549 | if (rdev->raid_disk < 0 && | |
5550 | !test_bit(Faulty, &rdev->flags)) { | |
5551 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 N |
5552 | if (rdev->raid_disk |
5553 | >= conf->previous_raid_disks) { | |
5554 | set_bit(In_sync, &rdev->flags); | |
5555 | added_devices++; | |
5556 | } else | |
5557 | rdev->recovery_offset = 0; | |
36fad858 NK |
5558 | |
5559 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 5560 | /* Failure here is OK */; |
50da0840 | 5561 | } |
87a8dec9 N |
5562 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5563 | && !test_bit(Faulty, &rdev->flags)) { | |
5564 | /* This is a spare that was manually added */ | |
5565 | set_bit(In_sync, &rdev->flags); | |
5566 | added_devices++; | |
5567 | } | |
29269553 | 5568 | |
87a8dec9 N |
5569 | /* When a reshape changes the number of devices, |
5570 | * ->degraded is measured against the larger of the | |
5571 | * pre and post number of devices. | |
5572 | */ | |
ec32a2bd | 5573 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 5574 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
5575 | spin_unlock_irqrestore(&conf->device_lock, flags); |
5576 | } | |
63c70c4f | 5577 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5578 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5579 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5580 | |
29269553 N |
5581 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5582 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5583 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5584 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5585 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5586 | "reshape"); |
29269553 N |
5587 | if (!mddev->sync_thread) { |
5588 | mddev->recovery = 0; | |
5589 | spin_lock_irq(&conf->device_lock); | |
5590 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5591 | conf->reshape_progress = MaxSector; |
29269553 N |
5592 | spin_unlock_irq(&conf->device_lock); |
5593 | return -EAGAIN; | |
5594 | } | |
c8f517c4 | 5595 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5596 | md_wakeup_thread(mddev->sync_thread); |
5597 | md_new_event(mddev); | |
5598 | return 0; | |
5599 | } | |
29269553 | 5600 | |
ec32a2bd N |
5601 | /* This is called from the reshape thread and should make any |
5602 | * changes needed in 'conf' | |
5603 | */ | |
d1688a6d | 5604 | static void end_reshape(struct r5conf *conf) |
29269553 | 5605 | { |
29269553 | 5606 | |
f6705578 | 5607 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5608 | |
f6705578 | 5609 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5610 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5611 | conf->reshape_progress = MaxSector; |
f6705578 | 5612 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5613 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5614 | |
5615 | /* read-ahead size must cover two whole stripes, which is | |
5616 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5617 | */ | |
4a5add49 | 5618 | if (conf->mddev->queue) { |
cea9c228 | 5619 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5620 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5621 | / PAGE_SIZE); |
16a53ecc N |
5622 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5623 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5624 | } | |
29269553 | 5625 | } |
29269553 N |
5626 | } |
5627 | ||
ec32a2bd N |
5628 | /* This is called from the raid5d thread with mddev_lock held. |
5629 | * It makes config changes to the device. | |
5630 | */ | |
fd01b88c | 5631 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 5632 | { |
d1688a6d | 5633 | struct r5conf *conf = mddev->private; |
cea9c228 N |
5634 | |
5635 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5636 | ||
ec32a2bd N |
5637 | if (mddev->delta_disks > 0) { |
5638 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5639 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5640 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5641 | } else { |
5642 | int d; | |
908f4fbd N |
5643 | spin_lock_irq(&conf->device_lock); |
5644 | mddev->degraded = calc_degraded(conf); | |
5645 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
5646 | for (d = conf->raid_disks ; |
5647 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 5648 | d++) { |
3cb03002 | 5649 | struct md_rdev *rdev = conf->disks[d].rdev; |
b8321b68 N |
5650 | if (rdev && |
5651 | raid5_remove_disk(mddev, rdev) == 0) { | |
36fad858 | 5652 | sysfs_unlink_rdev(mddev, rdev); |
1a67dde0 N |
5653 | rdev->raid_disk = -1; |
5654 | } | |
5655 | } | |
cea9c228 | 5656 | } |
88ce4930 | 5657 | mddev->layout = conf->algorithm; |
09c9e5fa | 5658 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5659 | mddev->reshape_position = MaxSector; |
5660 | mddev->delta_disks = 0; | |
cea9c228 N |
5661 | } |
5662 | } | |
5663 | ||
fd01b88c | 5664 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 5665 | { |
d1688a6d | 5666 | struct r5conf *conf = mddev->private; |
72626685 N |
5667 | |
5668 | switch(state) { | |
e464eafd N |
5669 | case 2: /* resume for a suspend */ |
5670 | wake_up(&conf->wait_for_overlap); | |
5671 | break; | |
5672 | ||
72626685 N |
5673 | case 1: /* stop all writes */ |
5674 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5675 | /* '2' tells resync/reshape to pause so that all |
5676 | * active stripes can drain | |
5677 | */ | |
5678 | conf->quiesce = 2; | |
72626685 | 5679 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5680 | atomic_read(&conf->active_stripes) == 0 && |
5681 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5682 | conf->device_lock, /* nothing */); |
64bd660b | 5683 | conf->quiesce = 1; |
72626685 | 5684 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5685 | /* allow reshape to continue */ |
5686 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5687 | break; |
5688 | ||
5689 | case 0: /* re-enable writes */ | |
5690 | spin_lock_irq(&conf->device_lock); | |
5691 | conf->quiesce = 0; | |
5692 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5693 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5694 | spin_unlock_irq(&conf->device_lock); |
5695 | break; | |
5696 | } | |
72626685 | 5697 | } |
b15c2e57 | 5698 | |
d562b0c4 | 5699 | |
fd01b88c | 5700 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 5701 | { |
e373ab10 | 5702 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 5703 | sector_t sectors; |
54071b38 | 5704 | |
f1b29bca | 5705 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 5706 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
5707 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5708 | mdname(mddev)); | |
f1b29bca DW |
5709 | return ERR_PTR(-EINVAL); |
5710 | } | |
5711 | ||
e373ab10 N |
5712 | sectors = raid0_conf->strip_zone[0].zone_end; |
5713 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 5714 | mddev->dev_sectors = sectors; |
f1b29bca | 5715 | mddev->new_level = level; |
54071b38 TM |
5716 | mddev->new_layout = ALGORITHM_PARITY_N; |
5717 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5718 | mddev->raid_disks += 1; | |
5719 | mddev->delta_disks = 1; | |
5720 | /* make sure it will be not marked as dirty */ | |
5721 | mddev->recovery_cp = MaxSector; | |
5722 | ||
5723 | return setup_conf(mddev); | |
5724 | } | |
5725 | ||
5726 | ||
fd01b88c | 5727 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
5728 | { |
5729 | int chunksect; | |
5730 | ||
5731 | if (mddev->raid_disks != 2 || | |
5732 | mddev->degraded > 1) | |
5733 | return ERR_PTR(-EINVAL); | |
5734 | ||
5735 | /* Should check if there are write-behind devices? */ | |
5736 | ||
5737 | chunksect = 64*2; /* 64K by default */ | |
5738 | ||
5739 | /* The array must be an exact multiple of chunksize */ | |
5740 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5741 | chunksect >>= 1; | |
5742 | ||
5743 | if ((chunksect<<9) < STRIPE_SIZE) | |
5744 | /* array size does not allow a suitable chunk size */ | |
5745 | return ERR_PTR(-EINVAL); | |
5746 | ||
5747 | mddev->new_level = 5; | |
5748 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5749 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5750 | |
5751 | return setup_conf(mddev); | |
5752 | } | |
5753 | ||
fd01b88c | 5754 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
5755 | { |
5756 | int new_layout; | |
5757 | ||
5758 | switch (mddev->layout) { | |
5759 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5760 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5761 | break; | |
5762 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5763 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5764 | break; | |
5765 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5766 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5767 | break; | |
5768 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5769 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5770 | break; | |
5771 | case ALGORITHM_PARITY_0_6: | |
5772 | new_layout = ALGORITHM_PARITY_0; | |
5773 | break; | |
5774 | case ALGORITHM_PARITY_N: | |
5775 | new_layout = ALGORITHM_PARITY_N; | |
5776 | break; | |
5777 | default: | |
5778 | return ERR_PTR(-EINVAL); | |
5779 | } | |
5780 | mddev->new_level = 5; | |
5781 | mddev->new_layout = new_layout; | |
5782 | mddev->delta_disks = -1; | |
5783 | mddev->raid_disks -= 1; | |
5784 | return setup_conf(mddev); | |
5785 | } | |
5786 | ||
d562b0c4 | 5787 | |
fd01b88c | 5788 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 5789 | { |
88ce4930 N |
5790 | /* For a 2-drive array, the layout and chunk size can be changed |
5791 | * immediately as not restriping is needed. | |
5792 | * For larger arrays we record the new value - after validation | |
5793 | * to be used by a reshape pass. | |
b3546035 | 5794 | */ |
d1688a6d | 5795 | struct r5conf *conf = mddev->private; |
597a711b | 5796 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5797 | |
597a711b | 5798 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5799 | return -EINVAL; |
5800 | if (new_chunk > 0) { | |
0ba459d2 | 5801 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5802 | return -EINVAL; |
597a711b | 5803 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5804 | return -EINVAL; |
597a711b | 5805 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5806 | /* not factor of array size */ |
5807 | return -EINVAL; | |
5808 | } | |
5809 | ||
5810 | /* They look valid */ | |
5811 | ||
88ce4930 | 5812 | if (mddev->raid_disks == 2) { |
597a711b N |
5813 | /* can make the change immediately */ |
5814 | if (mddev->new_layout >= 0) { | |
5815 | conf->algorithm = mddev->new_layout; | |
5816 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5817 | } |
5818 | if (new_chunk > 0) { | |
597a711b N |
5819 | conf->chunk_sectors = new_chunk ; |
5820 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5821 | } |
5822 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5823 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5824 | } |
50ac168a | 5825 | return check_reshape(mddev); |
88ce4930 N |
5826 | } |
5827 | ||
fd01b88c | 5828 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 5829 | { |
597a711b | 5830 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5831 | |
597a711b | 5832 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5833 | return -EINVAL; |
b3546035 | 5834 | if (new_chunk > 0) { |
0ba459d2 | 5835 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5836 | return -EINVAL; |
597a711b | 5837 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5838 | return -EINVAL; |
597a711b | 5839 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5840 | /* not factor of array size */ |
5841 | return -EINVAL; | |
b3546035 | 5842 | } |
88ce4930 N |
5843 | |
5844 | /* They look valid */ | |
50ac168a | 5845 | return check_reshape(mddev); |
b3546035 N |
5846 | } |
5847 | ||
fd01b88c | 5848 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
5849 | { |
5850 | /* raid5 can take over: | |
f1b29bca | 5851 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
5852 | * raid1 - if there are two drives. We need to know the chunk size |
5853 | * raid4 - trivial - just use a raid4 layout. | |
5854 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 5855 | */ |
f1b29bca DW |
5856 | if (mddev->level == 0) |
5857 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
5858 | if (mddev->level == 1) |
5859 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5860 | if (mddev->level == 4) { |
5861 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5862 | mddev->new_level = 5; | |
5863 | return setup_conf(mddev); | |
5864 | } | |
fc9739c6 N |
5865 | if (mddev->level == 6) |
5866 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5867 | |
5868 | return ERR_PTR(-EINVAL); | |
5869 | } | |
5870 | ||
fd01b88c | 5871 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 5872 | { |
f1b29bca DW |
5873 | /* raid4 can take over: |
5874 | * raid0 - if there is only one strip zone | |
5875 | * raid5 - if layout is right | |
a78d38a1 | 5876 | */ |
f1b29bca DW |
5877 | if (mddev->level == 0) |
5878 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
5879 | if (mddev->level == 5 && |
5880 | mddev->layout == ALGORITHM_PARITY_N) { | |
5881 | mddev->new_layout = 0; | |
5882 | mddev->new_level = 4; | |
5883 | return setup_conf(mddev); | |
5884 | } | |
5885 | return ERR_PTR(-EINVAL); | |
5886 | } | |
d562b0c4 | 5887 | |
84fc4b56 | 5888 | static struct md_personality raid5_personality; |
245f46c2 | 5889 | |
fd01b88c | 5890 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
5891 | { |
5892 | /* Currently can only take over a raid5. We map the | |
5893 | * personality to an equivalent raid6 personality | |
5894 | * with the Q block at the end. | |
5895 | */ | |
5896 | int new_layout; | |
5897 | ||
5898 | if (mddev->pers != &raid5_personality) | |
5899 | return ERR_PTR(-EINVAL); | |
5900 | if (mddev->degraded > 1) | |
5901 | return ERR_PTR(-EINVAL); | |
5902 | if (mddev->raid_disks > 253) | |
5903 | return ERR_PTR(-EINVAL); | |
5904 | if (mddev->raid_disks < 3) | |
5905 | return ERR_PTR(-EINVAL); | |
5906 | ||
5907 | switch (mddev->layout) { | |
5908 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5909 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5910 | break; | |
5911 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5912 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5913 | break; | |
5914 | case ALGORITHM_LEFT_SYMMETRIC: | |
5915 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5916 | break; | |
5917 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5918 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5919 | break; | |
5920 | case ALGORITHM_PARITY_0: | |
5921 | new_layout = ALGORITHM_PARITY_0_6; | |
5922 | break; | |
5923 | case ALGORITHM_PARITY_N: | |
5924 | new_layout = ALGORITHM_PARITY_N; | |
5925 | break; | |
5926 | default: | |
5927 | return ERR_PTR(-EINVAL); | |
5928 | } | |
5929 | mddev->new_level = 6; | |
5930 | mddev->new_layout = new_layout; | |
5931 | mddev->delta_disks = 1; | |
5932 | mddev->raid_disks += 1; | |
5933 | return setup_conf(mddev); | |
5934 | } | |
5935 | ||
5936 | ||
84fc4b56 | 5937 | static struct md_personality raid6_personality = |
16a53ecc N |
5938 | { |
5939 | .name = "raid6", | |
5940 | .level = 6, | |
5941 | .owner = THIS_MODULE, | |
5942 | .make_request = make_request, | |
5943 | .run = run, | |
5944 | .stop = stop, | |
5945 | .status = status, | |
5946 | .error_handler = error, | |
5947 | .hot_add_disk = raid5_add_disk, | |
5948 | .hot_remove_disk= raid5_remove_disk, | |
5949 | .spare_active = raid5_spare_active, | |
5950 | .sync_request = sync_request, | |
5951 | .resize = raid5_resize, | |
80c3a6ce | 5952 | .size = raid5_size, |
50ac168a | 5953 | .check_reshape = raid6_check_reshape, |
f416885e | 5954 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5955 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5956 | .quiesce = raid5_quiesce, |
245f46c2 | 5957 | .takeover = raid6_takeover, |
16a53ecc | 5958 | }; |
84fc4b56 | 5959 | static struct md_personality raid5_personality = |
1da177e4 LT |
5960 | { |
5961 | .name = "raid5", | |
2604b703 | 5962 | .level = 5, |
1da177e4 LT |
5963 | .owner = THIS_MODULE, |
5964 | .make_request = make_request, | |
5965 | .run = run, | |
5966 | .stop = stop, | |
5967 | .status = status, | |
5968 | .error_handler = error, | |
5969 | .hot_add_disk = raid5_add_disk, | |
5970 | .hot_remove_disk= raid5_remove_disk, | |
5971 | .spare_active = raid5_spare_active, | |
5972 | .sync_request = sync_request, | |
5973 | .resize = raid5_resize, | |
80c3a6ce | 5974 | .size = raid5_size, |
63c70c4f N |
5975 | .check_reshape = raid5_check_reshape, |
5976 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5977 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5978 | .quiesce = raid5_quiesce, |
d562b0c4 | 5979 | .takeover = raid5_takeover, |
1da177e4 LT |
5980 | }; |
5981 | ||
84fc4b56 | 5982 | static struct md_personality raid4_personality = |
1da177e4 | 5983 | { |
2604b703 N |
5984 | .name = "raid4", |
5985 | .level = 4, | |
5986 | .owner = THIS_MODULE, | |
5987 | .make_request = make_request, | |
5988 | .run = run, | |
5989 | .stop = stop, | |
5990 | .status = status, | |
5991 | .error_handler = error, | |
5992 | .hot_add_disk = raid5_add_disk, | |
5993 | .hot_remove_disk= raid5_remove_disk, | |
5994 | .spare_active = raid5_spare_active, | |
5995 | .sync_request = sync_request, | |
5996 | .resize = raid5_resize, | |
80c3a6ce | 5997 | .size = raid5_size, |
3d37890b N |
5998 | .check_reshape = raid5_check_reshape, |
5999 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 6000 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 6001 | .quiesce = raid5_quiesce, |
a78d38a1 | 6002 | .takeover = raid4_takeover, |
2604b703 N |
6003 | }; |
6004 | ||
6005 | static int __init raid5_init(void) | |
6006 | { | |
16a53ecc | 6007 | register_md_personality(&raid6_personality); |
2604b703 N |
6008 | register_md_personality(&raid5_personality); |
6009 | register_md_personality(&raid4_personality); | |
6010 | return 0; | |
1da177e4 LT |
6011 | } |
6012 | ||
2604b703 | 6013 | static void raid5_exit(void) |
1da177e4 | 6014 | { |
16a53ecc | 6015 | unregister_md_personality(&raid6_personality); |
2604b703 N |
6016 | unregister_md_personality(&raid5_personality); |
6017 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
6018 | } |
6019 | ||
6020 | module_init(raid5_init); | |
6021 | module_exit(raid5_exit); | |
6022 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 6023 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 6024 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
6025 | MODULE_ALIAS("md-raid5"); |
6026 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
6027 | MODULE_ALIAS("md-level-5"); |
6028 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
6029 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
6030 | MODULE_ALIAS("md-raid6"); | |
6031 | MODULE_ALIAS("md-level-6"); | |
6032 | ||
6033 | /* This used to be two separate modules, they were: */ | |
6034 | MODULE_ALIAS("raid5"); | |
6035 | MODULE_ALIAS("raid6"); |