]>
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. | |
30 | * conf->bm_write is the number of the last batch successfully written. | |
31 | * conf->bm_flush is the number of the last batch that was closed to | |
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 | |
35 | * the number of the batch it will be in. This is bm_flush+1. | |
36 | * When we are ready to do a write, if that batch hasn't been written yet, | |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
07a3b417 | 50 | #include <linux/async.h> |
bff61975 | 51 | #include <linux/seq_file.h> |
36d1c647 | 52 | #include <linux/cpu.h> |
43b2e5d8 | 53 | #include "md.h" |
bff61975 | 54 | #include "raid5.h" |
ef740c37 | 55 | #include "bitmap.h" |
72626685 | 56 | |
1da177e4 LT |
57 | /* |
58 | * Stripe cache | |
59 | */ | |
60 | ||
61 | #define NR_STRIPES 256 | |
62 | #define STRIPE_SIZE PAGE_SIZE | |
63 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
64 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
65 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 66 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 67 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
68 | #define HASH_MASK (NR_HASH - 1) |
69 | ||
fccddba0 | 70 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
71 | |
72 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
73 | * order without overlap. There may be several bio's per stripe+device, and | |
74 | * a bio could span several devices. | |
75 | * When walking this list for a particular stripe+device, we must never proceed | |
76 | * beyond a bio that extends past this device, as the next bio might no longer | |
77 | * be valid. | |
78 | * This macro is used to determine the 'next' bio in the list, given the sector | |
79 | * of the current stripe+device | |
80 | */ | |
81 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
82 | /* | |
83 | * The following can be used to debug the driver | |
84 | */ | |
1da177e4 LT |
85 | #define RAID5_PARANOIA 1 |
86 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
87 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
88 | #else | |
89 | # define CHECK_DEVLOCK() | |
90 | #endif | |
91 | ||
45b4233c | 92 | #ifdef DEBUG |
1da177e4 LT |
93 | #define inline |
94 | #define __inline__ | |
95 | #endif | |
96 | ||
6be9d494 BS |
97 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
98 | ||
960e739d | 99 | /* |
5b99c2ff JA |
100 | * We maintain a biased count of active stripes in the bottom 16 bits of |
101 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
102 | */ |
103 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
104 | { | |
5b99c2ff | 105 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
106 | } |
107 | ||
108 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
109 | { | |
5b99c2ff | 110 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
111 | } |
112 | ||
113 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
114 | { | |
115 | --bio->bi_phys_segments; | |
116 | return raid5_bi_phys_segments(bio); | |
117 | } | |
118 | ||
119 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
120 | { | |
121 | unsigned short val = raid5_bi_hw_segments(bio); | |
122 | ||
123 | --val; | |
5b99c2ff | 124 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
125 | return val; |
126 | } | |
127 | ||
128 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
129 | { | |
5b99c2ff | 130 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16); |
960e739d JA |
131 | } |
132 | ||
d0dabf7e N |
133 | /* Find first data disk in a raid6 stripe */ |
134 | static inline int raid6_d0(struct stripe_head *sh) | |
135 | { | |
67cc2b81 N |
136 | if (sh->ddf_layout) |
137 | /* ddf always start from first device */ | |
138 | return 0; | |
139 | /* md starts just after Q block */ | |
d0dabf7e N |
140 | if (sh->qd_idx == sh->disks - 1) |
141 | return 0; | |
142 | else | |
143 | return sh->qd_idx + 1; | |
144 | } | |
16a53ecc N |
145 | static inline int raid6_next_disk(int disk, int raid_disks) |
146 | { | |
147 | disk++; | |
148 | return (disk < raid_disks) ? disk : 0; | |
149 | } | |
a4456856 | 150 | |
d0dabf7e N |
151 | /* When walking through the disks in a raid5, starting at raid6_d0, |
152 | * We need to map each disk to a 'slot', where the data disks are slot | |
153 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
154 | * is raid_disks-1. This help does that mapping. | |
155 | */ | |
67cc2b81 N |
156 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
157 | int *count, int syndrome_disks) | |
d0dabf7e N |
158 | { |
159 | int slot; | |
67cc2b81 | 160 | |
d0dabf7e | 161 | if (idx == sh->pd_idx) |
67cc2b81 | 162 | return syndrome_disks; |
d0dabf7e | 163 | if (idx == sh->qd_idx) |
67cc2b81 | 164 | return syndrome_disks + 1; |
d0dabf7e N |
165 | slot = (*count)++; |
166 | return slot; | |
167 | } | |
168 | ||
a4456856 DW |
169 | static void return_io(struct bio *return_bi) |
170 | { | |
171 | struct bio *bi = return_bi; | |
172 | while (bi) { | |
a4456856 DW |
173 | |
174 | return_bi = bi->bi_next; | |
175 | bi->bi_next = NULL; | |
176 | bi->bi_size = 0; | |
0e13fe23 | 177 | bio_endio(bi, 0); |
a4456856 DW |
178 | bi = return_bi; |
179 | } | |
180 | } | |
181 | ||
1da177e4 LT |
182 | static void print_raid5_conf (raid5_conf_t *conf); |
183 | ||
600aa109 DW |
184 | static int stripe_operations_active(struct stripe_head *sh) |
185 | { | |
186 | return sh->check_state || sh->reconstruct_state || | |
187 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
188 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
189 | } | |
190 | ||
858119e1 | 191 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
192 | { |
193 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
194 | BUG_ON(!list_empty(&sh->lru)); |
195 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 196 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 197 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 198 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
199 | blk_plug_device(conf->mddev->queue); |
200 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 201 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 202 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
203 | blk_plug_device(conf->mddev->queue); |
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 | { | |
229 | raid5_conf_t *conf = sh->raid_conf; | |
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 | ||
16a53ecc | 245 | static inline void insert_hash(raid5_conf_t *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 LT |
251 | |
252 | CHECK_DEVLOCK(); | |
fccddba0 | 253 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
254 | } |
255 | ||
256 | ||
257 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
258 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
259 | { | |
260 | struct stripe_head *sh = NULL; | |
261 | struct list_head *first; | |
262 | ||
263 | CHECK_DEVLOCK(); | |
264 | if (list_empty(&conf->inactive_list)) | |
265 | goto out; | |
266 | first = conf->inactive_list.next; | |
267 | sh = list_entry(first, struct stripe_head, lru); | |
268 | list_del_init(first); | |
269 | remove_hash(sh); | |
270 | atomic_inc(&conf->active_stripes); | |
271 | out: | |
272 | return sh; | |
273 | } | |
274 | ||
275 | static void shrink_buffers(struct stripe_head *sh, int num) | |
276 | { | |
277 | struct page *p; | |
278 | int i; | |
279 | ||
280 | for (i=0; i<num ; i++) { | |
281 | p = sh->dev[i].page; | |
282 | if (!p) | |
283 | continue; | |
284 | sh->dev[i].page = NULL; | |
2d1f3b5d | 285 | put_page(p); |
1da177e4 LT |
286 | } |
287 | } | |
288 | ||
289 | static int grow_buffers(struct stripe_head *sh, int num) | |
290 | { | |
291 | int i; | |
292 | ||
293 | for (i=0; i<num; i++) { | |
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); |
911d4ee8 N |
305 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
306 | struct stripe_head *sh); | |
1da177e4 | 307 | |
b5663ba4 | 308 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
309 | { |
310 | raid5_conf_t *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 | |
1da177e4 | 317 | CHECK_DEVLOCK(); |
45b4233c | 318 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
319 | (unsigned long long)sh->sector); |
320 | ||
321 | remove_hash(sh); | |
16a53ecc | 322 | |
86b42c71 | 323 | sh->generation = conf->generation - previous; |
b5663ba4 | 324 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 325 | sh->sector = sector; |
911d4ee8 | 326 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
327 | sh->state = 0; |
328 | ||
7ecaa1e6 N |
329 | |
330 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
331 | struct r5dev *dev = &sh->dev[i]; |
332 | ||
d84e0f10 | 333 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 334 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 335 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 336 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 337 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
338 | test_bit(R5_LOCKED, &dev->flags)); |
339 | BUG(); | |
340 | } | |
341 | dev->flags = 0; | |
784052ec | 342 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
343 | } |
344 | insert_hash(conf, sh); | |
345 | } | |
346 | ||
86b42c71 N |
347 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, |
348 | short generation) | |
1da177e4 LT |
349 | { |
350 | struct stripe_head *sh; | |
fccddba0 | 351 | struct hlist_node *hn; |
1da177e4 LT |
352 | |
353 | CHECK_DEVLOCK(); | |
45b4233c | 354 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 355 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 356 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 357 | return sh; |
45b4233c | 358 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
359 | return NULL; |
360 | } | |
361 | ||
362 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 363 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 364 | |
b5663ba4 N |
365 | static struct stripe_head * |
366 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
a8c906ca | 367 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
368 | { |
369 | struct stripe_head *sh; | |
370 | ||
45b4233c | 371 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
372 | |
373 | spin_lock_irq(&conf->device_lock); | |
374 | ||
375 | do { | |
72626685 | 376 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 377 | conf->quiesce == 0 || noquiesce, |
72626685 | 378 | conf->device_lock, /* nothing */); |
86b42c71 | 379 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
380 | if (!sh) { |
381 | if (!conf->inactive_blocked) | |
382 | sh = get_free_stripe(conf); | |
383 | if (noblock && sh == NULL) | |
384 | break; | |
385 | if (!sh) { | |
386 | conf->inactive_blocked = 1; | |
387 | wait_event_lock_irq(conf->wait_for_stripe, | |
388 | !list_empty(&conf->inactive_list) && | |
5036805b N |
389 | (atomic_read(&conf->active_stripes) |
390 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
391 | || !conf->inactive_blocked), |
392 | conf->device_lock, | |
f4370781 | 393 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
394 | ); |
395 | conf->inactive_blocked = 0; | |
396 | } else | |
b5663ba4 | 397 | init_stripe(sh, sector, previous); |
1da177e4 LT |
398 | } else { |
399 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
400 | BUG_ON(!list_empty(&sh->lru) |
401 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
402 | } else { |
403 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
404 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
405 | if (list_empty(&sh->lru) && |
406 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
407 | BUG(); |
408 | list_del_init(&sh->lru); | |
1da177e4 LT |
409 | } |
410 | } | |
411 | } while (sh == NULL); | |
412 | ||
413 | if (sh) | |
414 | atomic_inc(&sh->count); | |
415 | ||
416 | spin_unlock_irq(&conf->device_lock); | |
417 | return sh; | |
418 | } | |
419 | ||
6712ecf8 N |
420 | static void |
421 | raid5_end_read_request(struct bio *bi, int error); | |
422 | static void | |
423 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 424 | |
c4e5ac0a | 425 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
426 | { |
427 | raid5_conf_t *conf = sh->raid_conf; | |
428 | int i, disks = sh->disks; | |
429 | ||
430 | might_sleep(); | |
431 | ||
432 | for (i = disks; i--; ) { | |
433 | int rw; | |
434 | struct bio *bi; | |
435 | mdk_rdev_t *rdev; | |
436 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
437 | rw = WRITE; | |
438 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
439 | rw = READ; | |
440 | else | |
441 | continue; | |
442 | ||
443 | bi = &sh->dev[i].req; | |
444 | ||
445 | bi->bi_rw = rw; | |
446 | if (rw == WRITE) | |
447 | bi->bi_end_io = raid5_end_write_request; | |
448 | else | |
449 | bi->bi_end_io = raid5_end_read_request; | |
450 | ||
451 | rcu_read_lock(); | |
452 | rdev = rcu_dereference(conf->disks[i].rdev); | |
453 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
454 | rdev = NULL; | |
455 | if (rdev) | |
456 | atomic_inc(&rdev->nr_pending); | |
457 | rcu_read_unlock(); | |
458 | ||
459 | if (rdev) { | |
c4e5ac0a | 460 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
461 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
462 | ||
2b7497f0 DW |
463 | set_bit(STRIPE_IO_STARTED, &sh->state); |
464 | ||
91c00924 DW |
465 | bi->bi_bdev = rdev->bdev; |
466 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 467 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
468 | bi->bi_rw, i); |
469 | atomic_inc(&sh->count); | |
470 | bi->bi_sector = sh->sector + rdev->data_offset; | |
471 | bi->bi_flags = 1 << BIO_UPTODATE; | |
472 | bi->bi_vcnt = 1; | |
473 | bi->bi_max_vecs = 1; | |
474 | bi->bi_idx = 0; | |
475 | bi->bi_io_vec = &sh->dev[i].vec; | |
476 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
477 | bi->bi_io_vec[0].bv_offset = 0; | |
478 | bi->bi_size = STRIPE_SIZE; | |
479 | bi->bi_next = NULL; | |
480 | if (rw == WRITE && | |
481 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
482 | atomic_add(STRIPE_SECTORS, | |
483 | &rdev->corrected_errors); | |
484 | generic_make_request(bi); | |
485 | } else { | |
486 | if (rw == WRITE) | |
487 | set_bit(STRIPE_DEGRADED, &sh->state); | |
488 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
489 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
490 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
491 | set_bit(STRIPE_HANDLE, &sh->state); | |
492 | } | |
493 | } | |
494 | } | |
495 | ||
496 | static struct dma_async_tx_descriptor * | |
497 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
498 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
499 | { | |
500 | struct bio_vec *bvl; | |
501 | struct page *bio_page; | |
502 | int i; | |
503 | int page_offset; | |
a08abd8c | 504 | struct async_submit_ctl submit; |
0403e382 | 505 | enum async_tx_flags flags = 0; |
91c00924 DW |
506 | |
507 | if (bio->bi_sector >= sector) | |
508 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
509 | else | |
510 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 511 | |
0403e382 DW |
512 | if (frombio) |
513 | flags |= ASYNC_TX_FENCE; | |
514 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
515 | ||
91c00924 DW |
516 | bio_for_each_segment(bvl, bio, i) { |
517 | int len = bio_iovec_idx(bio, i)->bv_len; | |
518 | int clen; | |
519 | int b_offset = 0; | |
520 | ||
521 | if (page_offset < 0) { | |
522 | b_offset = -page_offset; | |
523 | page_offset += b_offset; | |
524 | len -= b_offset; | |
525 | } | |
526 | ||
527 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
528 | clen = STRIPE_SIZE - page_offset; | |
529 | else | |
530 | clen = len; | |
531 | ||
532 | if (clen > 0) { | |
533 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
534 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
535 | if (frombio) | |
536 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 537 | b_offset, clen, &submit); |
91c00924 DW |
538 | else |
539 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 540 | page_offset, clen, &submit); |
91c00924 | 541 | } |
a08abd8c DW |
542 | /* chain the operations */ |
543 | submit.depend_tx = tx; | |
544 | ||
91c00924 DW |
545 | if (clen < len) /* hit end of page */ |
546 | break; | |
547 | page_offset += len; | |
548 | } | |
549 | ||
550 | return tx; | |
551 | } | |
552 | ||
553 | static void ops_complete_biofill(void *stripe_head_ref) | |
554 | { | |
555 | struct stripe_head *sh = stripe_head_ref; | |
556 | struct bio *return_bi = NULL; | |
557 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 558 | int i; |
91c00924 | 559 | |
e46b272b | 560 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
561 | (unsigned long long)sh->sector); |
562 | ||
563 | /* clear completed biofills */ | |
83de75cc | 564 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
565 | for (i = sh->disks; i--; ) { |
566 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
567 | |
568 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
569 | /* and check if we need to reply to a read request, |
570 | * new R5_Wantfill requests are held off until | |
83de75cc | 571 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
572 | */ |
573 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 574 | struct bio *rbi, *rbi2; |
91c00924 | 575 | |
91c00924 DW |
576 | BUG_ON(!dev->read); |
577 | rbi = dev->read; | |
578 | dev->read = NULL; | |
579 | while (rbi && rbi->bi_sector < | |
580 | dev->sector + STRIPE_SECTORS) { | |
581 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 582 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
583 | rbi->bi_next = return_bi; |
584 | return_bi = rbi; | |
585 | } | |
91c00924 DW |
586 | rbi = rbi2; |
587 | } | |
588 | } | |
589 | } | |
83de75cc DW |
590 | spin_unlock_irq(&conf->device_lock); |
591 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
592 | |
593 | return_io(return_bi); | |
594 | ||
e4d84909 | 595 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
596 | release_stripe(sh); |
597 | } | |
598 | ||
599 | static void ops_run_biofill(struct stripe_head *sh) | |
600 | { | |
601 | struct dma_async_tx_descriptor *tx = NULL; | |
602 | raid5_conf_t *conf = sh->raid_conf; | |
a08abd8c | 603 | struct async_submit_ctl submit; |
91c00924 DW |
604 | int i; |
605 | ||
e46b272b | 606 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
607 | (unsigned long long)sh->sector); |
608 | ||
609 | for (i = sh->disks; i--; ) { | |
610 | struct r5dev *dev = &sh->dev[i]; | |
611 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
612 | struct bio *rbi; | |
613 | spin_lock_irq(&conf->device_lock); | |
614 | dev->read = rbi = dev->toread; | |
615 | dev->toread = NULL; | |
616 | spin_unlock_irq(&conf->device_lock); | |
617 | while (rbi && rbi->bi_sector < | |
618 | dev->sector + STRIPE_SECTORS) { | |
619 | tx = async_copy_data(0, rbi, dev->page, | |
620 | dev->sector, tx); | |
621 | rbi = r5_next_bio(rbi, dev->sector); | |
622 | } | |
623 | } | |
624 | } | |
625 | ||
626 | atomic_inc(&sh->count); | |
a08abd8c DW |
627 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
628 | async_trigger_callback(&submit); | |
91c00924 DW |
629 | } |
630 | ||
4e7d2c0a | 631 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 632 | { |
4e7d2c0a | 633 | struct r5dev *tgt; |
91c00924 | 634 | |
4e7d2c0a DW |
635 | if (target < 0) |
636 | return; | |
91c00924 | 637 | |
4e7d2c0a | 638 | tgt = &sh->dev[target]; |
91c00924 DW |
639 | set_bit(R5_UPTODATE, &tgt->flags); |
640 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
641 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
642 | } |
643 | ||
ac6b53b6 | 644 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
645 | { |
646 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 647 | |
e46b272b | 648 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
649 | (unsigned long long)sh->sector); |
650 | ||
ac6b53b6 | 651 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 652 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 653 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 654 | |
ecc65c9b DW |
655 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
656 | if (sh->check_state == check_state_compute_run) | |
657 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
658 | set_bit(STRIPE_HANDLE, &sh->state); |
659 | release_stripe(sh); | |
660 | } | |
661 | ||
d6f38f31 DW |
662 | /* return a pointer to the address conversion region of the scribble buffer */ |
663 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
664 | struct raid5_percpu *percpu) | |
665 | { | |
666 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
667 | } | |
668 | ||
669 | static struct dma_async_tx_descriptor * | |
670 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 671 | { |
91c00924 | 672 | int disks = sh->disks; |
d6f38f31 | 673 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
674 | int target = sh->ops.target; |
675 | struct r5dev *tgt = &sh->dev[target]; | |
676 | struct page *xor_dest = tgt->page; | |
677 | int count = 0; | |
678 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 679 | struct async_submit_ctl submit; |
91c00924 DW |
680 | int i; |
681 | ||
682 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 683 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
684 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
685 | ||
686 | for (i = disks; i--; ) | |
687 | if (i != target) | |
688 | xor_srcs[count++] = sh->dev[i].page; | |
689 | ||
690 | atomic_inc(&sh->count); | |
691 | ||
0403e382 | 692 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 693 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 694 | if (unlikely(count == 1)) |
a08abd8c | 695 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 696 | else |
a08abd8c | 697 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 698 | |
91c00924 DW |
699 | return tx; |
700 | } | |
701 | ||
ac6b53b6 DW |
702 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
703 | * @srcs - (struct page *) array of size sh->disks | |
704 | * @sh - stripe_head to parse | |
705 | * | |
706 | * Populates srcs in proper layout order for the stripe and returns the | |
707 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
708 | * destination buffer is recorded in srcs[count] and the Q destination | |
709 | * is recorded in srcs[count+1]]. | |
710 | */ | |
711 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
712 | { | |
713 | int disks = sh->disks; | |
714 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
715 | int d0_idx = raid6_d0(sh); | |
716 | int count; | |
717 | int i; | |
718 | ||
719 | for (i = 0; i < disks; i++) | |
720 | srcs[i] = (void *)raid6_empty_zero_page; | |
721 | ||
722 | count = 0; | |
723 | i = d0_idx; | |
724 | do { | |
725 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
726 | ||
727 | srcs[slot] = sh->dev[i].page; | |
728 | i = raid6_next_disk(i, disks); | |
729 | } while (i != d0_idx); | |
730 | BUG_ON(count != syndrome_disks); | |
731 | ||
732 | return count; | |
733 | } | |
734 | ||
735 | static struct dma_async_tx_descriptor * | |
736 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
737 | { | |
738 | int disks = sh->disks; | |
739 | struct page **blocks = percpu->scribble; | |
740 | int target; | |
741 | int qd_idx = sh->qd_idx; | |
742 | struct dma_async_tx_descriptor *tx; | |
743 | struct async_submit_ctl submit; | |
744 | struct r5dev *tgt; | |
745 | struct page *dest; | |
746 | int i; | |
747 | int count; | |
748 | ||
749 | if (sh->ops.target < 0) | |
750 | target = sh->ops.target2; | |
751 | else if (sh->ops.target2 < 0) | |
752 | target = sh->ops.target; | |
91c00924 | 753 | else |
ac6b53b6 DW |
754 | /* we should only have one valid target */ |
755 | BUG(); | |
756 | BUG_ON(target < 0); | |
757 | pr_debug("%s: stripe %llu block: %d\n", | |
758 | __func__, (unsigned long long)sh->sector, target); | |
759 | ||
760 | tgt = &sh->dev[target]; | |
761 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
762 | dest = tgt->page; | |
763 | ||
764 | atomic_inc(&sh->count); | |
765 | ||
766 | if (target == qd_idx) { | |
767 | count = set_syndrome_sources(blocks, sh); | |
768 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
769 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
770 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
771 | ops_complete_compute, sh, | |
ac6b53b6 DW |
772 | to_addr_conv(sh, percpu)); |
773 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
774 | } else { | |
775 | /* Compute any data- or p-drive using XOR */ | |
776 | count = 0; | |
777 | for (i = disks; i-- ; ) { | |
778 | if (i == target || i == qd_idx) | |
779 | continue; | |
780 | blocks[count++] = sh->dev[i].page; | |
781 | } | |
782 | ||
0403e382 DW |
783 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
784 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
785 | to_addr_conv(sh, percpu)); |
786 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
787 | } | |
91c00924 | 788 | |
91c00924 DW |
789 | return tx; |
790 | } | |
791 | ||
ac6b53b6 DW |
792 | static struct dma_async_tx_descriptor * |
793 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
794 | { | |
795 | int i, count, disks = sh->disks; | |
796 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
797 | int d0_idx = raid6_d0(sh); | |
798 | int faila = -1, failb = -1; | |
799 | int target = sh->ops.target; | |
800 | int target2 = sh->ops.target2; | |
801 | struct r5dev *tgt = &sh->dev[target]; | |
802 | struct r5dev *tgt2 = &sh->dev[target2]; | |
803 | struct dma_async_tx_descriptor *tx; | |
804 | struct page **blocks = percpu->scribble; | |
805 | struct async_submit_ctl submit; | |
806 | ||
807 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
808 | __func__, (unsigned long long)sh->sector, target, target2); | |
809 | BUG_ON(target < 0 || target2 < 0); | |
810 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
811 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
812 | ||
6c910a78 | 813 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
814 | * slot number conversion for 'faila' and 'failb' |
815 | */ | |
816 | for (i = 0; i < disks ; i++) | |
817 | blocks[i] = (void *)raid6_empty_zero_page; | |
818 | count = 0; | |
819 | i = d0_idx; | |
820 | do { | |
821 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
822 | ||
823 | blocks[slot] = sh->dev[i].page; | |
824 | ||
825 | if (i == target) | |
826 | faila = slot; | |
827 | if (i == target2) | |
828 | failb = slot; | |
829 | i = raid6_next_disk(i, disks); | |
830 | } while (i != d0_idx); | |
831 | BUG_ON(count != syndrome_disks); | |
832 | ||
833 | BUG_ON(faila == failb); | |
834 | if (failb < faila) | |
835 | swap(faila, failb); | |
836 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
837 | __func__, (unsigned long long)sh->sector, faila, failb); | |
838 | ||
839 | atomic_inc(&sh->count); | |
840 | ||
841 | if (failb == syndrome_disks+1) { | |
842 | /* Q disk is one of the missing disks */ | |
843 | if (faila == syndrome_disks) { | |
844 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
845 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
846 | ops_complete_compute, sh, | |
847 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
848 | return async_gen_syndrome(blocks, 0, count+2, |
849 | STRIPE_SIZE, &submit); | |
850 | } else { | |
851 | struct page *dest; | |
852 | int data_target; | |
853 | int qd_idx = sh->qd_idx; | |
854 | ||
855 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
856 | if (target == qd_idx) | |
857 | data_target = target2; | |
858 | else | |
859 | data_target = target; | |
860 | ||
861 | count = 0; | |
862 | for (i = disks; i-- ; ) { | |
863 | if (i == data_target || i == qd_idx) | |
864 | continue; | |
865 | blocks[count++] = sh->dev[i].page; | |
866 | } | |
867 | dest = sh->dev[data_target].page; | |
0403e382 DW |
868 | init_async_submit(&submit, |
869 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
870 | NULL, NULL, NULL, | |
871 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
872 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
873 | &submit); | |
874 | ||
875 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
876 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
877 | ops_complete_compute, sh, | |
878 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
879 | return async_gen_syndrome(blocks, 0, count+2, |
880 | STRIPE_SIZE, &submit); | |
881 | } | |
ac6b53b6 | 882 | } else { |
6c910a78 DW |
883 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
884 | ops_complete_compute, sh, | |
885 | to_addr_conv(sh, percpu)); | |
886 | if (failb == syndrome_disks) { | |
887 | /* We're missing D+P. */ | |
888 | return async_raid6_datap_recov(syndrome_disks+2, | |
889 | STRIPE_SIZE, faila, | |
890 | blocks, &submit); | |
891 | } else { | |
892 | /* We're missing D+D. */ | |
893 | return async_raid6_2data_recov(syndrome_disks+2, | |
894 | STRIPE_SIZE, faila, failb, | |
895 | blocks, &submit); | |
896 | } | |
ac6b53b6 DW |
897 | } |
898 | } | |
899 | ||
900 | ||
91c00924 DW |
901 | static void ops_complete_prexor(void *stripe_head_ref) |
902 | { | |
903 | struct stripe_head *sh = stripe_head_ref; | |
904 | ||
e46b272b | 905 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 906 | (unsigned long long)sh->sector); |
91c00924 DW |
907 | } |
908 | ||
909 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
910 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
911 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 912 | { |
91c00924 | 913 | int disks = sh->disks; |
d6f38f31 | 914 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 915 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 916 | struct async_submit_ctl submit; |
91c00924 DW |
917 | |
918 | /* existing parity data subtracted */ | |
919 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
920 | ||
e46b272b | 921 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
922 | (unsigned long long)sh->sector); |
923 | ||
924 | for (i = disks; i--; ) { | |
925 | struct r5dev *dev = &sh->dev[i]; | |
926 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 927 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
928 | xor_srcs[count++] = dev->page; |
929 | } | |
930 | ||
0403e382 | 931 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 932 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 933 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
934 | |
935 | return tx; | |
936 | } | |
937 | ||
938 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 939 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
940 | { |
941 | int disks = sh->disks; | |
d8ee0728 | 942 | int i; |
91c00924 | 943 | |
e46b272b | 944 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
945 | (unsigned long long)sh->sector); |
946 | ||
947 | for (i = disks; i--; ) { | |
948 | struct r5dev *dev = &sh->dev[i]; | |
949 | struct bio *chosen; | |
91c00924 | 950 | |
d8ee0728 | 951 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
952 | struct bio *wbi; |
953 | ||
954 | spin_lock(&sh->lock); | |
955 | chosen = dev->towrite; | |
956 | dev->towrite = NULL; | |
957 | BUG_ON(dev->written); | |
958 | wbi = dev->written = chosen; | |
959 | spin_unlock(&sh->lock); | |
960 | ||
961 | while (wbi && wbi->bi_sector < | |
962 | dev->sector + STRIPE_SECTORS) { | |
963 | tx = async_copy_data(1, wbi, dev->page, | |
964 | dev->sector, tx); | |
965 | wbi = r5_next_bio(wbi, dev->sector); | |
966 | } | |
967 | } | |
968 | } | |
969 | ||
970 | return tx; | |
971 | } | |
972 | ||
ac6b53b6 | 973 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
974 | { |
975 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
976 | int disks = sh->disks; |
977 | int pd_idx = sh->pd_idx; | |
978 | int qd_idx = sh->qd_idx; | |
979 | int i; | |
91c00924 | 980 | |
e46b272b | 981 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
982 | (unsigned long long)sh->sector); |
983 | ||
984 | for (i = disks; i--; ) { | |
985 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 DW |
986 | |
987 | if (dev->written || i == pd_idx || i == qd_idx) | |
91c00924 DW |
988 | set_bit(R5_UPTODATE, &dev->flags); |
989 | } | |
990 | ||
d8ee0728 DW |
991 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
992 | sh->reconstruct_state = reconstruct_state_drain_result; | |
993 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
994 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
995 | else { | |
996 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
997 | sh->reconstruct_state = reconstruct_state_result; | |
998 | } | |
91c00924 DW |
999 | |
1000 | set_bit(STRIPE_HANDLE, &sh->state); | |
1001 | release_stripe(sh); | |
1002 | } | |
1003 | ||
1004 | static void | |
ac6b53b6 DW |
1005 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1006 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1007 | { |
91c00924 | 1008 | int disks = sh->disks; |
d6f38f31 | 1009 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1010 | struct async_submit_ctl submit; |
91c00924 DW |
1011 | int count = 0, pd_idx = sh->pd_idx, i; |
1012 | struct page *xor_dest; | |
d8ee0728 | 1013 | int prexor = 0; |
91c00924 | 1014 | unsigned long flags; |
91c00924 | 1015 | |
e46b272b | 1016 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1017 | (unsigned long long)sh->sector); |
1018 | ||
1019 | /* check if prexor is active which means only process blocks | |
1020 | * that are part of a read-modify-write (written) | |
1021 | */ | |
d8ee0728 DW |
1022 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1023 | prexor = 1; | |
91c00924 DW |
1024 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1025 | for (i = disks; i--; ) { | |
1026 | struct r5dev *dev = &sh->dev[i]; | |
1027 | if (dev->written) | |
1028 | xor_srcs[count++] = dev->page; | |
1029 | } | |
1030 | } else { | |
1031 | xor_dest = sh->dev[pd_idx].page; | |
1032 | for (i = disks; i--; ) { | |
1033 | struct r5dev *dev = &sh->dev[i]; | |
1034 | if (i != pd_idx) | |
1035 | xor_srcs[count++] = dev->page; | |
1036 | } | |
1037 | } | |
1038 | ||
91c00924 DW |
1039 | /* 1/ if we prexor'd then the dest is reused as a source |
1040 | * 2/ if we did not prexor then we are redoing the parity | |
1041 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1042 | * for the synchronous xor case | |
1043 | */ | |
88ba2aa5 | 1044 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1045 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1046 | ||
1047 | atomic_inc(&sh->count); | |
1048 | ||
ac6b53b6 | 1049 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1050 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1051 | if (unlikely(count == 1)) |
1052 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1053 | else | |
1054 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1055 | } |
1056 | ||
ac6b53b6 DW |
1057 | static void |
1058 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1059 | struct dma_async_tx_descriptor *tx) | |
1060 | { | |
1061 | struct async_submit_ctl submit; | |
1062 | struct page **blocks = percpu->scribble; | |
1063 | int count; | |
1064 | ||
1065 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1066 | ||
1067 | count = set_syndrome_sources(blocks, sh); | |
1068 | ||
1069 | atomic_inc(&sh->count); | |
1070 | ||
1071 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1072 | sh, to_addr_conv(sh, percpu)); | |
1073 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1074 | } |
1075 | ||
1076 | static void ops_complete_check(void *stripe_head_ref) | |
1077 | { | |
1078 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1079 | |
e46b272b | 1080 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1081 | (unsigned long long)sh->sector); |
1082 | ||
ecc65c9b | 1083 | sh->check_state = check_state_check_result; |
91c00924 DW |
1084 | set_bit(STRIPE_HANDLE, &sh->state); |
1085 | release_stripe(sh); | |
1086 | } | |
1087 | ||
ac6b53b6 | 1088 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1089 | { |
91c00924 | 1090 | int disks = sh->disks; |
ac6b53b6 DW |
1091 | int pd_idx = sh->pd_idx; |
1092 | int qd_idx = sh->qd_idx; | |
1093 | struct page *xor_dest; | |
d6f38f31 | 1094 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1095 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1096 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1097 | int count; |
1098 | int i; | |
91c00924 | 1099 | |
e46b272b | 1100 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1101 | (unsigned long long)sh->sector); |
1102 | ||
ac6b53b6 DW |
1103 | count = 0; |
1104 | xor_dest = sh->dev[pd_idx].page; | |
1105 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1106 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1107 | if (i == pd_idx || i == qd_idx) |
1108 | continue; | |
1109 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1110 | } |
1111 | ||
d6f38f31 DW |
1112 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1113 | to_addr_conv(sh, percpu)); | |
099f53cb | 1114 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1115 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1116 | |
91c00924 | 1117 | atomic_inc(&sh->count); |
a08abd8c DW |
1118 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1119 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1120 | } |
1121 | ||
ac6b53b6 DW |
1122 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1123 | { | |
1124 | struct page **srcs = percpu->scribble; | |
1125 | struct async_submit_ctl submit; | |
1126 | int count; | |
1127 | ||
1128 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1129 | (unsigned long long)sh->sector, checkp); | |
1130 | ||
1131 | count = set_syndrome_sources(srcs, sh); | |
1132 | if (!checkp) | |
1133 | srcs[count] = NULL; | |
91c00924 | 1134 | |
91c00924 | 1135 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1136 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1137 | sh, to_addr_conv(sh, percpu)); | |
1138 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1139 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1140 | } |
1141 | ||
ac6b53b6 | 1142 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1143 | { |
1144 | int overlap_clear = 0, i, disks = sh->disks; | |
1145 | struct dma_async_tx_descriptor *tx = NULL; | |
d6f38f31 | 1146 | raid5_conf_t *conf = sh->raid_conf; |
ac6b53b6 | 1147 | int level = conf->level; |
d6f38f31 DW |
1148 | struct raid5_percpu *percpu; |
1149 | unsigned long cpu; | |
91c00924 | 1150 | |
d6f38f31 DW |
1151 | cpu = get_cpu(); |
1152 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1153 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1154 | ops_run_biofill(sh); |
1155 | overlap_clear++; | |
1156 | } | |
1157 | ||
7b3a871e | 1158 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1159 | if (level < 6) |
1160 | tx = ops_run_compute5(sh, percpu); | |
1161 | else { | |
1162 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1163 | tx = ops_run_compute6_1(sh, percpu); | |
1164 | else | |
1165 | tx = ops_run_compute6_2(sh, percpu); | |
1166 | } | |
1167 | /* terminate the chain if reconstruct is not set to be run */ | |
1168 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1169 | async_tx_ack(tx); |
1170 | } | |
91c00924 | 1171 | |
600aa109 | 1172 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1173 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1174 | |
600aa109 | 1175 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1176 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1177 | overlap_clear++; |
1178 | } | |
1179 | ||
ac6b53b6 DW |
1180 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1181 | if (level < 6) | |
1182 | ops_run_reconstruct5(sh, percpu, tx); | |
1183 | else | |
1184 | ops_run_reconstruct6(sh, percpu, tx); | |
1185 | } | |
91c00924 | 1186 | |
ac6b53b6 DW |
1187 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1188 | if (sh->check_state == check_state_run) | |
1189 | ops_run_check_p(sh, percpu); | |
1190 | else if (sh->check_state == check_state_run_q) | |
1191 | ops_run_check_pq(sh, percpu, 0); | |
1192 | else if (sh->check_state == check_state_run_pq) | |
1193 | ops_run_check_pq(sh, percpu, 1); | |
1194 | else | |
1195 | BUG(); | |
1196 | } | |
91c00924 | 1197 | |
91c00924 DW |
1198 | if (overlap_clear) |
1199 | for (i = disks; i--; ) { | |
1200 | struct r5dev *dev = &sh->dev[i]; | |
1201 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1202 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1203 | } | |
d6f38f31 | 1204 | put_cpu(); |
91c00924 DW |
1205 | } |
1206 | ||
3f294f4f | 1207 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1208 | { |
1209 | struct stripe_head *sh; | |
3f294f4f N |
1210 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
1211 | if (!sh) | |
1212 | return 0; | |
1213 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
1214 | sh->raid_conf = conf; | |
1215 | spin_lock_init(&sh->lock); | |
1216 | ||
1217 | if (grow_buffers(sh, conf->raid_disks)) { | |
1218 | shrink_buffers(sh, conf->raid_disks); | |
1219 | kmem_cache_free(conf->slab_cache, sh); | |
1220 | return 0; | |
1221 | } | |
7ecaa1e6 | 1222 | sh->disks = conf->raid_disks; |
3f294f4f N |
1223 | /* we just created an active stripe so... */ |
1224 | atomic_set(&sh->count, 1); | |
1225 | atomic_inc(&conf->active_stripes); | |
1226 | INIT_LIST_HEAD(&sh->lru); | |
1227 | release_stripe(sh); | |
1228 | return 1; | |
1229 | } | |
1230 | ||
1231 | static int grow_stripes(raid5_conf_t *conf, int num) | |
1232 | { | |
e18b890b | 1233 | struct kmem_cache *sc; |
1da177e4 LT |
1234 | int devs = conf->raid_disks; |
1235 | ||
245f46c2 N |
1236 | sprintf(conf->cache_name[0], |
1237 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1238 | sprintf(conf->cache_name[1], | |
1239 | "raid%d-%s-alt", conf->level, mdname(conf->mddev)); | |
ad01c9e3 N |
1240 | conf->active_name = 0; |
1241 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1242 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1243 | 0, 0, NULL); |
1da177e4 LT |
1244 | if (!sc) |
1245 | return 1; | |
1246 | conf->slab_cache = sc; | |
ad01c9e3 | 1247 | conf->pool_size = devs; |
16a53ecc | 1248 | while (num--) |
3f294f4f | 1249 | if (!grow_one_stripe(conf)) |
1da177e4 | 1250 | return 1; |
1da177e4 LT |
1251 | return 0; |
1252 | } | |
29269553 | 1253 | |
d6f38f31 DW |
1254 | /** |
1255 | * scribble_len - return the required size of the scribble region | |
1256 | * @num - total number of disks in the array | |
1257 | * | |
1258 | * The size must be enough to contain: | |
1259 | * 1/ a struct page pointer for each device in the array +2 | |
1260 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1261 | * (dma_map_page()) or page (page_address()) address. | |
1262 | * | |
1263 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1264 | * calculate over all devices (not just the data blocks), using zeros in place | |
1265 | * of the P and Q blocks. | |
1266 | */ | |
1267 | static size_t scribble_len(int num) | |
1268 | { | |
1269 | size_t len; | |
1270 | ||
1271 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1272 | ||
1273 | return len; | |
1274 | } | |
1275 | ||
ad01c9e3 N |
1276 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
1277 | { | |
1278 | /* Make all the stripes able to hold 'newsize' devices. | |
1279 | * New slots in each stripe get 'page' set to a new page. | |
1280 | * | |
1281 | * This happens in stages: | |
1282 | * 1/ create a new kmem_cache and allocate the required number of | |
1283 | * stripe_heads. | |
1284 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1285 | * to the new stripe_heads. This will have the side effect of | |
1286 | * freezing the array as once all stripe_heads have been collected, | |
1287 | * no IO will be possible. Old stripe heads are freed once their | |
1288 | * pages have been transferred over, and the old kmem_cache is | |
1289 | * freed when all stripes are done. | |
1290 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1291 | * we simple return a failre status - no need to clean anything up. | |
1292 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1293 | * If this fails, we don't bother trying the shrink the | |
1294 | * stripe_heads down again, we just leave them as they are. | |
1295 | * As each stripe_head is processed the new one is released into | |
1296 | * active service. | |
1297 | * | |
1298 | * Once step2 is started, we cannot afford to wait for a write, | |
1299 | * so we use GFP_NOIO allocations. | |
1300 | */ | |
1301 | struct stripe_head *osh, *nsh; | |
1302 | LIST_HEAD(newstripes); | |
1303 | struct disk_info *ndisks; | |
d6f38f31 | 1304 | unsigned long cpu; |
b5470dc5 | 1305 | int err; |
e18b890b | 1306 | struct kmem_cache *sc; |
ad01c9e3 N |
1307 | int i; |
1308 | ||
1309 | if (newsize <= conf->pool_size) | |
1310 | return 0; /* never bother to shrink */ | |
1311 | ||
b5470dc5 DW |
1312 | err = md_allow_write(conf->mddev); |
1313 | if (err) | |
1314 | return err; | |
2a2275d6 | 1315 | |
ad01c9e3 N |
1316 | /* Step 1 */ |
1317 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1318 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1319 | 0, 0, NULL); |
ad01c9e3 N |
1320 | if (!sc) |
1321 | return -ENOMEM; | |
1322 | ||
1323 | for (i = conf->max_nr_stripes; i; i--) { | |
1324 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
1325 | if (!nsh) | |
1326 | break; | |
1327 | ||
1328 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
1329 | ||
1330 | nsh->raid_conf = conf; | |
1331 | spin_lock_init(&nsh->lock); | |
1332 | ||
1333 | list_add(&nsh->lru, &newstripes); | |
1334 | } | |
1335 | if (i) { | |
1336 | /* didn't get enough, give up */ | |
1337 | while (!list_empty(&newstripes)) { | |
1338 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1339 | list_del(&nsh->lru); | |
1340 | kmem_cache_free(sc, nsh); | |
1341 | } | |
1342 | kmem_cache_destroy(sc); | |
1343 | return -ENOMEM; | |
1344 | } | |
1345 | /* Step 2 - Must use GFP_NOIO now. | |
1346 | * OK, we have enough stripes, start collecting inactive | |
1347 | * stripes and copying them over | |
1348 | */ | |
1349 | list_for_each_entry(nsh, &newstripes, lru) { | |
1350 | spin_lock_irq(&conf->device_lock); | |
1351 | wait_event_lock_irq(conf->wait_for_stripe, | |
1352 | !list_empty(&conf->inactive_list), | |
1353 | conf->device_lock, | |
b3b46be3 | 1354 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1355 | ); |
1356 | osh = get_free_stripe(conf); | |
1357 | spin_unlock_irq(&conf->device_lock); | |
1358 | atomic_set(&nsh->count, 1); | |
1359 | for(i=0; i<conf->pool_size; i++) | |
1360 | nsh->dev[i].page = osh->dev[i].page; | |
1361 | for( ; i<newsize; i++) | |
1362 | nsh->dev[i].page = NULL; | |
1363 | kmem_cache_free(conf->slab_cache, osh); | |
1364 | } | |
1365 | kmem_cache_destroy(conf->slab_cache); | |
1366 | ||
1367 | /* Step 3. | |
1368 | * At this point, we are holding all the stripes so the array | |
1369 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1370 | * conf->disks and the scribble region |
ad01c9e3 N |
1371 | */ |
1372 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1373 | if (ndisks) { | |
1374 | for (i=0; i<conf->raid_disks; i++) | |
1375 | ndisks[i] = conf->disks[i]; | |
1376 | kfree(conf->disks); | |
1377 | conf->disks = ndisks; | |
1378 | } else | |
1379 | err = -ENOMEM; | |
1380 | ||
d6f38f31 DW |
1381 | get_online_cpus(); |
1382 | conf->scribble_len = scribble_len(newsize); | |
1383 | for_each_present_cpu(cpu) { | |
1384 | struct raid5_percpu *percpu; | |
1385 | void *scribble; | |
1386 | ||
1387 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1388 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1389 | ||
1390 | if (scribble) { | |
1391 | kfree(percpu->scribble); | |
1392 | percpu->scribble = scribble; | |
1393 | } else { | |
1394 | err = -ENOMEM; | |
1395 | break; | |
1396 | } | |
1397 | } | |
1398 | put_online_cpus(); | |
1399 | ||
ad01c9e3 N |
1400 | /* Step 4, return new stripes to service */ |
1401 | while(!list_empty(&newstripes)) { | |
1402 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1403 | list_del_init(&nsh->lru); | |
d6f38f31 | 1404 | |
ad01c9e3 N |
1405 | for (i=conf->raid_disks; i < newsize; i++) |
1406 | if (nsh->dev[i].page == NULL) { | |
1407 | struct page *p = alloc_page(GFP_NOIO); | |
1408 | nsh->dev[i].page = p; | |
1409 | if (!p) | |
1410 | err = -ENOMEM; | |
1411 | } | |
1412 | release_stripe(nsh); | |
1413 | } | |
1414 | /* critical section pass, GFP_NOIO no longer needed */ | |
1415 | ||
1416 | conf->slab_cache = sc; | |
1417 | conf->active_name = 1-conf->active_name; | |
1418 | conf->pool_size = newsize; | |
1419 | return err; | |
1420 | } | |
1da177e4 | 1421 | |
3f294f4f | 1422 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1423 | { |
1424 | struct stripe_head *sh; | |
1425 | ||
3f294f4f N |
1426 | spin_lock_irq(&conf->device_lock); |
1427 | sh = get_free_stripe(conf); | |
1428 | spin_unlock_irq(&conf->device_lock); | |
1429 | if (!sh) | |
1430 | return 0; | |
78bafebd | 1431 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1432 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1433 | kmem_cache_free(conf->slab_cache, sh); |
1434 | atomic_dec(&conf->active_stripes); | |
1435 | return 1; | |
1436 | } | |
1437 | ||
1438 | static void shrink_stripes(raid5_conf_t *conf) | |
1439 | { | |
1440 | while (drop_one_stripe(conf)) | |
1441 | ; | |
1442 | ||
29fc7e3e N |
1443 | if (conf->slab_cache) |
1444 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1445 | conf->slab_cache = NULL; |
1446 | } | |
1447 | ||
6712ecf8 | 1448 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1449 | { |
99c0fb5f | 1450 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1451 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1452 | int disks = sh->disks, i; |
1da177e4 | 1453 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1454 | char b[BDEVNAME_SIZE]; |
1455 | mdk_rdev_t *rdev; | |
1da177e4 | 1456 | |
1da177e4 LT |
1457 | |
1458 | for (i=0 ; i<disks; i++) | |
1459 | if (bi == &sh->dev[i].req) | |
1460 | break; | |
1461 | ||
45b4233c DW |
1462 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1463 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1464 | uptodate); |
1465 | if (i == disks) { | |
1466 | BUG(); | |
6712ecf8 | 1467 | return; |
1da177e4 LT |
1468 | } |
1469 | ||
1470 | if (uptodate) { | |
1da177e4 | 1471 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1472 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1473 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1474 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1475 | " (%lu sectors at %llu on %s)\n", | |
1476 | mdname(conf->mddev), STRIPE_SECTORS, | |
1477 | (unsigned long long)(sh->sector | |
1478 | + rdev->data_offset), | |
1479 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1480 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1481 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1482 | } | |
ba22dcbf N |
1483 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1484 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1485 | } else { |
d6950432 | 1486 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1487 | int retry = 0; |
d6950432 N |
1488 | rdev = conf->disks[i].rdev; |
1489 | ||
1da177e4 | 1490 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1491 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1492 | if (conf->mddev->degraded) |
6be9d494 BS |
1493 | printk_rl(KERN_WARNING |
1494 | "raid5:%s: read error not correctable " | |
1495 | "(sector %llu on %s).\n", | |
1496 | mdname(conf->mddev), | |
1497 | (unsigned long long)(sh->sector | |
1498 | + rdev->data_offset), | |
1499 | bdn); | |
ba22dcbf | 1500 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1501 | /* Oh, no!!! */ |
6be9d494 BS |
1502 | printk_rl(KERN_WARNING |
1503 | "raid5:%s: read error NOT corrected!! " | |
1504 | "(sector %llu on %s).\n", | |
1505 | mdname(conf->mddev), | |
1506 | (unsigned long long)(sh->sector | |
1507 | + rdev->data_offset), | |
1508 | bdn); | |
d6950432 | 1509 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1510 | > conf->max_nr_stripes) |
14f8d26b | 1511 | printk(KERN_WARNING |
d6950432 N |
1512 | "raid5:%s: Too many read errors, failing device %s.\n", |
1513 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1514 | else |
1515 | retry = 1; | |
1516 | if (retry) | |
1517 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1518 | else { | |
4e5314b5 N |
1519 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1520 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1521 | md_error(conf->mddev, rdev); |
ba22dcbf | 1522 | } |
1da177e4 LT |
1523 | } |
1524 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1525 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1526 | set_bit(STRIPE_HANDLE, &sh->state); | |
1527 | release_stripe(sh); | |
1da177e4 LT |
1528 | } |
1529 | ||
d710e138 | 1530 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1531 | { |
99c0fb5f | 1532 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1533 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1534 | int disks = sh->disks, i; |
1da177e4 LT |
1535 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1536 | ||
1da177e4 LT |
1537 | for (i=0 ; i<disks; i++) |
1538 | if (bi == &sh->dev[i].req) | |
1539 | break; | |
1540 | ||
45b4233c | 1541 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1542 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1543 | uptodate); | |
1544 | if (i == disks) { | |
1545 | BUG(); | |
6712ecf8 | 1546 | return; |
1da177e4 LT |
1547 | } |
1548 | ||
1da177e4 LT |
1549 | if (!uptodate) |
1550 | md_error(conf->mddev, conf->disks[i].rdev); | |
1551 | ||
1552 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1553 | ||
1554 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1555 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1556 | release_stripe(sh); |
1da177e4 LT |
1557 | } |
1558 | ||
1559 | ||
784052ec | 1560 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1561 | |
784052ec | 1562 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1563 | { |
1564 | struct r5dev *dev = &sh->dev[i]; | |
1565 | ||
1566 | bio_init(&dev->req); | |
1567 | dev->req.bi_io_vec = &dev->vec; | |
1568 | dev->req.bi_vcnt++; | |
1569 | dev->req.bi_max_vecs++; | |
1570 | dev->vec.bv_page = dev->page; | |
1571 | dev->vec.bv_len = STRIPE_SIZE; | |
1572 | dev->vec.bv_offset = 0; | |
1573 | ||
1574 | dev->req.bi_sector = sh->sector; | |
1575 | dev->req.bi_private = sh; | |
1576 | ||
1577 | dev->flags = 0; | |
784052ec | 1578 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1579 | } |
1580 | ||
1581 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1582 | { | |
1583 | char b[BDEVNAME_SIZE]; | |
1584 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1585 | pr_debug("raid5: error called\n"); |
1da177e4 | 1586 | |
b2d444d7 | 1587 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1588 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1589 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1590 | unsigned long flags; | |
1591 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1592 | mddev->degraded++; |
c04be0aa | 1593 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1594 | /* |
1595 | * if recovery was running, make sure it aborts. | |
1596 | */ | |
dfc70645 | 1597 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1598 | } |
b2d444d7 | 1599 | set_bit(Faulty, &rdev->flags); |
d710e138 N |
1600 | printk(KERN_ALERT |
1601 | "raid5: Disk failure on %s, disabling device.\n" | |
1602 | "raid5: Operation continuing on %d devices.\n", | |
1603 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); | |
1da177e4 | 1604 | } |
16a53ecc | 1605 | } |
1da177e4 LT |
1606 | |
1607 | /* | |
1608 | * Input: a 'big' sector number, | |
1609 | * Output: index of the data and parity disk, and the sector # in them. | |
1610 | */ | |
112bf897 | 1611 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1612 | int previous, int *dd_idx, |
1613 | struct stripe_head *sh) | |
1da177e4 LT |
1614 | { |
1615 | long stripe; | |
1616 | unsigned long chunk_number; | |
1617 | unsigned int chunk_offset; | |
911d4ee8 | 1618 | int pd_idx, qd_idx; |
67cc2b81 | 1619 | int ddf_layout = 0; |
1da177e4 | 1620 | sector_t new_sector; |
e183eaed N |
1621 | int algorithm = previous ? conf->prev_algo |
1622 | : conf->algorithm; | |
09c9e5fa AN |
1623 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1624 | : conf->chunk_sectors; | |
112bf897 N |
1625 | int raid_disks = previous ? conf->previous_raid_disks |
1626 | : conf->raid_disks; | |
1627 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1628 | |
1629 | /* First compute the information on this sector */ | |
1630 | ||
1631 | /* | |
1632 | * Compute the chunk number and the sector offset inside the chunk | |
1633 | */ | |
1634 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1635 | chunk_number = r_sector; | |
1636 | BUG_ON(r_sector != chunk_number); | |
1637 | ||
1638 | /* | |
1639 | * Compute the stripe number | |
1640 | */ | |
1641 | stripe = chunk_number / data_disks; | |
1642 | ||
1643 | /* | |
1644 | * Compute the data disk and parity disk indexes inside the stripe | |
1645 | */ | |
1646 | *dd_idx = chunk_number % data_disks; | |
1647 | ||
1648 | /* | |
1649 | * Select the parity disk based on the user selected algorithm. | |
1650 | */ | |
911d4ee8 | 1651 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1652 | switch(conf->level) { |
1653 | case 4: | |
911d4ee8 | 1654 | pd_idx = data_disks; |
16a53ecc N |
1655 | break; |
1656 | case 5: | |
e183eaed | 1657 | switch (algorithm) { |
1da177e4 | 1658 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1659 | pd_idx = data_disks - stripe % raid_disks; |
1660 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1661 | (*dd_idx)++; |
1662 | break; | |
1663 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1664 | pd_idx = stripe % raid_disks; |
1665 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1666 | (*dd_idx)++; |
1667 | break; | |
1668 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1669 | pd_idx = data_disks - stripe % raid_disks; |
1670 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 LT |
1671 | break; |
1672 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1673 | pd_idx = stripe % raid_disks; |
1674 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 | 1675 | break; |
99c0fb5f N |
1676 | case ALGORITHM_PARITY_0: |
1677 | pd_idx = 0; | |
1678 | (*dd_idx)++; | |
1679 | break; | |
1680 | case ALGORITHM_PARITY_N: | |
1681 | pd_idx = data_disks; | |
1682 | break; | |
1da177e4 | 1683 | default: |
14f8d26b | 1684 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
e183eaed | 1685 | algorithm); |
99c0fb5f | 1686 | BUG(); |
16a53ecc N |
1687 | } |
1688 | break; | |
1689 | case 6: | |
1690 | ||
e183eaed | 1691 | switch (algorithm) { |
16a53ecc | 1692 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1693 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1694 | qd_idx = pd_idx + 1; | |
1695 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1696 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1697 | qd_idx = 0; |
1698 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1699 | (*dd_idx) += 2; /* D D P Q D */ |
1700 | break; | |
1701 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1702 | pd_idx = stripe % raid_disks; |
1703 | qd_idx = pd_idx + 1; | |
1704 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1705 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1706 | qd_idx = 0; |
1707 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1708 | (*dd_idx) += 2; /* D D P Q D */ |
1709 | break; | |
1710 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1711 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1712 | qd_idx = (pd_idx + 1) % raid_disks; | |
1713 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1714 | break; |
1715 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1716 | pd_idx = stripe % raid_disks; |
1717 | qd_idx = (pd_idx + 1) % raid_disks; | |
1718 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1719 | break; |
99c0fb5f N |
1720 | |
1721 | case ALGORITHM_PARITY_0: | |
1722 | pd_idx = 0; | |
1723 | qd_idx = 1; | |
1724 | (*dd_idx) += 2; | |
1725 | break; | |
1726 | case ALGORITHM_PARITY_N: | |
1727 | pd_idx = data_disks; | |
1728 | qd_idx = data_disks + 1; | |
1729 | break; | |
1730 | ||
1731 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1732 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1733 | * of blocks for computing Q is different. | |
1734 | */ | |
1735 | pd_idx = stripe % raid_disks; | |
1736 | qd_idx = pd_idx + 1; | |
1737 | if (pd_idx == raid_disks-1) { | |
1738 | (*dd_idx)++; /* Q D D D P */ | |
1739 | qd_idx = 0; | |
1740 | } else if (*dd_idx >= pd_idx) | |
1741 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1742 | ddf_layout = 1; |
99c0fb5f N |
1743 | break; |
1744 | ||
1745 | case ALGORITHM_ROTATING_N_RESTART: | |
1746 | /* Same a left_asymmetric, by first stripe is | |
1747 | * D D D P Q rather than | |
1748 | * Q D D D P | |
1749 | */ | |
1750 | pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
1751 | qd_idx = pd_idx + 1; | |
1752 | if (pd_idx == raid_disks-1) { | |
1753 | (*dd_idx)++; /* Q D D D P */ | |
1754 | qd_idx = 0; | |
1755 | } else if (*dd_idx >= pd_idx) | |
1756 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1757 | ddf_layout = 1; |
99c0fb5f N |
1758 | break; |
1759 | ||
1760 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1761 | /* Same as left_symmetric but Q is before P */ | |
1762 | pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1763 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; | |
1764 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1765 | ddf_layout = 1; |
99c0fb5f N |
1766 | break; |
1767 | ||
1768 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1769 | /* RAID5 left_asymmetric, with Q on last device */ | |
1770 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1771 | if (*dd_idx >= pd_idx) | |
1772 | (*dd_idx)++; | |
1773 | qd_idx = raid_disks - 1; | |
1774 | break; | |
1775 | ||
1776 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1777 | pd_idx = stripe % (raid_disks-1); | |
1778 | if (*dd_idx >= pd_idx) | |
1779 | (*dd_idx)++; | |
1780 | qd_idx = raid_disks - 1; | |
1781 | break; | |
1782 | ||
1783 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1784 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1785 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1786 | qd_idx = raid_disks - 1; | |
1787 | break; | |
1788 | ||
1789 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1790 | pd_idx = stripe % (raid_disks-1); | |
1791 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1792 | qd_idx = raid_disks - 1; | |
1793 | break; | |
1794 | ||
1795 | case ALGORITHM_PARITY_0_6: | |
1796 | pd_idx = 0; | |
1797 | (*dd_idx)++; | |
1798 | qd_idx = raid_disks - 1; | |
1799 | break; | |
1800 | ||
1801 | ||
16a53ecc | 1802 | default: |
d710e138 | 1803 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
e183eaed | 1804 | algorithm); |
99c0fb5f | 1805 | BUG(); |
16a53ecc N |
1806 | } |
1807 | break; | |
1da177e4 LT |
1808 | } |
1809 | ||
911d4ee8 N |
1810 | if (sh) { |
1811 | sh->pd_idx = pd_idx; | |
1812 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1813 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1814 | } |
1da177e4 LT |
1815 | /* |
1816 | * Finally, compute the new sector number | |
1817 | */ | |
1818 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1819 | return new_sector; | |
1820 | } | |
1821 | ||
1822 | ||
784052ec | 1823 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1824 | { |
1825 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1826 | int raid_disks = sh->disks; |
1827 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1828 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
1829 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1830 | : conf->chunk_sectors; | |
e183eaed N |
1831 | int algorithm = previous ? conf->prev_algo |
1832 | : conf->algorithm; | |
1da177e4 LT |
1833 | sector_t stripe; |
1834 | int chunk_offset; | |
911d4ee8 | 1835 | int chunk_number, dummy1, dd_idx = i; |
1da177e4 | 1836 | sector_t r_sector; |
911d4ee8 | 1837 | struct stripe_head sh2; |
1da177e4 | 1838 | |
16a53ecc | 1839 | |
1da177e4 LT |
1840 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1841 | stripe = new_sector; | |
1842 | BUG_ON(new_sector != stripe); | |
1843 | ||
16a53ecc N |
1844 | if (i == sh->pd_idx) |
1845 | return 0; | |
1846 | switch(conf->level) { | |
1847 | case 4: break; | |
1848 | case 5: | |
e183eaed | 1849 | switch (algorithm) { |
1da177e4 LT |
1850 | case ALGORITHM_LEFT_ASYMMETRIC: |
1851 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1852 | if (i > sh->pd_idx) | |
1853 | i--; | |
1854 | break; | |
1855 | case ALGORITHM_LEFT_SYMMETRIC: | |
1856 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1857 | if (i < sh->pd_idx) | |
1858 | i += raid_disks; | |
1859 | i -= (sh->pd_idx + 1); | |
1860 | break; | |
99c0fb5f N |
1861 | case ALGORITHM_PARITY_0: |
1862 | i -= 1; | |
1863 | break; | |
1864 | case ALGORITHM_PARITY_N: | |
1865 | break; | |
1da177e4 | 1866 | default: |
14f8d26b | 1867 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
e183eaed | 1868 | algorithm); |
99c0fb5f | 1869 | BUG(); |
16a53ecc N |
1870 | } |
1871 | break; | |
1872 | case 6: | |
d0dabf7e | 1873 | if (i == sh->qd_idx) |
16a53ecc | 1874 | return 0; /* It is the Q disk */ |
e183eaed | 1875 | switch (algorithm) { |
16a53ecc N |
1876 | case ALGORITHM_LEFT_ASYMMETRIC: |
1877 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1878 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1879 | case ALGORITHM_ROTATING_N_RESTART: | |
1880 | if (sh->pd_idx == raid_disks-1) | |
1881 | i--; /* Q D D D P */ | |
16a53ecc N |
1882 | else if (i > sh->pd_idx) |
1883 | i -= 2; /* D D P Q D */ | |
1884 | break; | |
1885 | case ALGORITHM_LEFT_SYMMETRIC: | |
1886 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1887 | if (sh->pd_idx == raid_disks-1) | |
1888 | i--; /* Q D D D P */ | |
1889 | else { | |
1890 | /* D D P Q D */ | |
1891 | if (i < sh->pd_idx) | |
1892 | i += raid_disks; | |
1893 | i -= (sh->pd_idx + 2); | |
1894 | } | |
1895 | break; | |
99c0fb5f N |
1896 | case ALGORITHM_PARITY_0: |
1897 | i -= 2; | |
1898 | break; | |
1899 | case ALGORITHM_PARITY_N: | |
1900 | break; | |
1901 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1902 | if (sh->pd_idx == 0) | |
1903 | i--; /* P D D D Q */ | |
1904 | else if (i > sh->pd_idx) | |
1905 | i -= 2; /* D D Q P D */ | |
1906 | break; | |
1907 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1908 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1909 | if (i > sh->pd_idx) | |
1910 | i--; | |
1911 | break; | |
1912 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1913 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1914 | if (i < sh->pd_idx) | |
1915 | i += data_disks + 1; | |
1916 | i -= (sh->pd_idx + 1); | |
1917 | break; | |
1918 | case ALGORITHM_PARITY_0_6: | |
1919 | i -= 1; | |
1920 | break; | |
16a53ecc | 1921 | default: |
d710e138 | 1922 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
e183eaed | 1923 | algorithm); |
99c0fb5f | 1924 | BUG(); |
16a53ecc N |
1925 | } |
1926 | break; | |
1da177e4 LT |
1927 | } |
1928 | ||
1929 | chunk_number = stripe * data_disks + i; | |
1930 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1931 | ||
112bf897 | 1932 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 1933 | previous, &dummy1, &sh2); |
911d4ee8 N |
1934 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
1935 | || sh2.qd_idx != sh->qd_idx) { | |
14f8d26b | 1936 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1937 | return 0; |
1938 | } | |
1939 | return r_sector; | |
1940 | } | |
1941 | ||
1942 | ||
600aa109 | 1943 | static void |
c0f7bddb | 1944 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 1945 | int rcw, int expand) |
e33129d8 DW |
1946 | { |
1947 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
c0f7bddb YT |
1948 | raid5_conf_t *conf = sh->raid_conf; |
1949 | int level = conf->level; | |
e33129d8 DW |
1950 | |
1951 | if (rcw) { | |
1952 | /* if we are not expanding this is a proper write request, and | |
1953 | * there will be bios with new data to be drained into the | |
1954 | * stripe cache | |
1955 | */ | |
1956 | if (!expand) { | |
600aa109 DW |
1957 | sh->reconstruct_state = reconstruct_state_drain_run; |
1958 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1959 | } else | |
1960 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 1961 | |
ac6b53b6 | 1962 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
1963 | |
1964 | for (i = disks; i--; ) { | |
1965 | struct r5dev *dev = &sh->dev[i]; | |
1966 | ||
1967 | if (dev->towrite) { | |
1968 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 1969 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
1970 | if (!expand) |
1971 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1972 | s->locked++; |
e33129d8 DW |
1973 | } |
1974 | } | |
c0f7bddb | 1975 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 1976 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 1977 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 1978 | } else { |
c0f7bddb | 1979 | BUG_ON(level == 6); |
e33129d8 DW |
1980 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
1981 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
1982 | ||
d8ee0728 | 1983 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
1984 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
1985 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 1986 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
1987 | |
1988 | for (i = disks; i--; ) { | |
1989 | struct r5dev *dev = &sh->dev[i]; | |
1990 | if (i == pd_idx) | |
1991 | continue; | |
1992 | ||
e33129d8 DW |
1993 | if (dev->towrite && |
1994 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
1995 | test_bit(R5_Wantcompute, &dev->flags))) { |
1996 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
1997 | set_bit(R5_LOCKED, &dev->flags); |
1998 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1999 | s->locked++; |
e33129d8 DW |
2000 | } |
2001 | } | |
2002 | } | |
2003 | ||
c0f7bddb | 2004 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2005 | * are in flight |
2006 | */ | |
2007 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2008 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2009 | s->locked++; |
e33129d8 | 2010 | |
c0f7bddb YT |
2011 | if (level == 6) { |
2012 | int qd_idx = sh->qd_idx; | |
2013 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2014 | ||
2015 | set_bit(R5_LOCKED, &dev->flags); | |
2016 | clear_bit(R5_UPTODATE, &dev->flags); | |
2017 | s->locked++; | |
2018 | } | |
2019 | ||
600aa109 | 2020 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2021 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2022 | s->locked, s->ops_request); |
e33129d8 | 2023 | } |
16a53ecc | 2024 | |
1da177e4 LT |
2025 | /* |
2026 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2027 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2028 | * The bi_next chain must be in order. |
2029 | */ | |
2030 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2031 | { | |
2032 | struct bio **bip; | |
2033 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 2034 | int firstwrite=0; |
1da177e4 | 2035 | |
45b4233c | 2036 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2037 | (unsigned long long)bi->bi_sector, |
2038 | (unsigned long long)sh->sector); | |
2039 | ||
2040 | ||
2041 | spin_lock(&sh->lock); | |
2042 | spin_lock_irq(&conf->device_lock); | |
72626685 | 2043 | if (forwrite) { |
1da177e4 | 2044 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2045 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2046 | firstwrite = 1; | |
2047 | } else | |
1da177e4 LT |
2048 | bip = &sh->dev[dd_idx].toread; |
2049 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2050 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2051 | goto overlap; | |
2052 | bip = & (*bip)->bi_next; | |
2053 | } | |
2054 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2055 | goto overlap; | |
2056 | ||
78bafebd | 2057 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2058 | if (*bip) |
2059 | bi->bi_next = *bip; | |
2060 | *bip = bi; | |
960e739d | 2061 | bi->bi_phys_segments++; |
1da177e4 LT |
2062 | spin_unlock_irq(&conf->device_lock); |
2063 | spin_unlock(&sh->lock); | |
2064 | ||
45b4233c | 2065 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
2066 | (unsigned long long)bi->bi_sector, |
2067 | (unsigned long long)sh->sector, dd_idx); | |
2068 | ||
72626685 | 2069 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
2070 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
2071 | STRIPE_SECTORS, 0); | |
ae3c20cc | 2072 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
2073 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
2074 | } | |
2075 | ||
1da177e4 LT |
2076 | if (forwrite) { |
2077 | /* check if page is covered */ | |
2078 | sector_t sector = sh->dev[dd_idx].sector; | |
2079 | for (bi=sh->dev[dd_idx].towrite; | |
2080 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2081 | bi && bi->bi_sector <= sector; | |
2082 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2083 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2084 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2085 | } | |
2086 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2087 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2088 | } | |
2089 | return 1; | |
2090 | ||
2091 | overlap: | |
2092 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2093 | spin_unlock_irq(&conf->device_lock); | |
2094 | spin_unlock(&sh->lock); | |
2095 | return 0; | |
2096 | } | |
2097 | ||
29269553 N |
2098 | static void end_reshape(raid5_conf_t *conf); |
2099 | ||
911d4ee8 N |
2100 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
2101 | struct stripe_head *sh) | |
ccfcc3c1 | 2102 | { |
784052ec | 2103 | int sectors_per_chunk = |
09c9e5fa | 2104 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2105 | int dd_idx; |
2d2063ce | 2106 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2107 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2108 | |
112bf897 N |
2109 | raid5_compute_sector(conf, |
2110 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2111 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2112 | previous, |
911d4ee8 | 2113 | &dd_idx, sh); |
ccfcc3c1 N |
2114 | } |
2115 | ||
a4456856 | 2116 | static void |
1fe797e6 | 2117 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2118 | struct stripe_head_state *s, int disks, |
2119 | struct bio **return_bi) | |
2120 | { | |
2121 | int i; | |
2122 | for (i = disks; i--; ) { | |
2123 | struct bio *bi; | |
2124 | int bitmap_end = 0; | |
2125 | ||
2126 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2127 | mdk_rdev_t *rdev; | |
2128 | rcu_read_lock(); | |
2129 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2130 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2131 | /* multiple read failures in one stripe */ | |
2132 | md_error(conf->mddev, rdev); | |
2133 | rcu_read_unlock(); | |
2134 | } | |
2135 | spin_lock_irq(&conf->device_lock); | |
2136 | /* fail all writes first */ | |
2137 | bi = sh->dev[i].towrite; | |
2138 | sh->dev[i].towrite = NULL; | |
2139 | if (bi) { | |
2140 | s->to_write--; | |
2141 | bitmap_end = 1; | |
2142 | } | |
2143 | ||
2144 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2145 | wake_up(&conf->wait_for_overlap); | |
2146 | ||
2147 | while (bi && bi->bi_sector < | |
2148 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2149 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2150 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2151 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2152 | md_write_end(conf->mddev); |
2153 | bi->bi_next = *return_bi; | |
2154 | *return_bi = bi; | |
2155 | } | |
2156 | bi = nextbi; | |
2157 | } | |
2158 | /* and fail all 'written' */ | |
2159 | bi = sh->dev[i].written; | |
2160 | sh->dev[i].written = NULL; | |
2161 | if (bi) bitmap_end = 1; | |
2162 | while (bi && bi->bi_sector < | |
2163 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2164 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2165 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2166 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2167 | md_write_end(conf->mddev); |
2168 | bi->bi_next = *return_bi; | |
2169 | *return_bi = bi; | |
2170 | } | |
2171 | bi = bi2; | |
2172 | } | |
2173 | ||
b5e98d65 DW |
2174 | /* fail any reads if this device is non-operational and |
2175 | * the data has not reached the cache yet. | |
2176 | */ | |
2177 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2178 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2179 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2180 | bi = sh->dev[i].toread; |
2181 | sh->dev[i].toread = NULL; | |
2182 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2183 | wake_up(&conf->wait_for_overlap); | |
2184 | if (bi) s->to_read--; | |
2185 | while (bi && bi->bi_sector < | |
2186 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2187 | struct bio *nextbi = | |
2188 | r5_next_bio(bi, sh->dev[i].sector); | |
2189 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2190 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2191 | bi->bi_next = *return_bi; |
2192 | *return_bi = bi; | |
2193 | } | |
2194 | bi = nextbi; | |
2195 | } | |
2196 | } | |
2197 | spin_unlock_irq(&conf->device_lock); | |
2198 | if (bitmap_end) | |
2199 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2200 | STRIPE_SECTORS, 0, 0); | |
2201 | } | |
2202 | ||
8b3e6cdc DW |
2203 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2204 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2205 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2206 | } |
2207 | ||
1fe797e6 DW |
2208 | /* fetch_block5 - checks the given member device to see if its data needs |
2209 | * to be read or computed to satisfy a request. | |
2210 | * | |
2211 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2212 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2213 | */ |
1fe797e6 DW |
2214 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2215 | int disk_idx, int disks) | |
f38e1219 DW |
2216 | { |
2217 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2218 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2219 | ||
f38e1219 DW |
2220 | /* is the data in this block needed, and can we get it? */ |
2221 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2222 | !test_bit(R5_UPTODATE, &dev->flags) && |
2223 | (dev->toread || | |
2224 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2225 | s->syncing || s->expanding || | |
2226 | (s->failed && | |
2227 | (failed_dev->toread || | |
2228 | (failed_dev->towrite && | |
2229 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2230 | /* We would like to get this block, possibly by computing it, |
2231 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2232 | */ |
2233 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2234 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2235 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2236 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2237 | set_bit(R5_Wantcompute, &dev->flags); |
2238 | sh->ops.target = disk_idx; | |
ac6b53b6 | 2239 | sh->ops.target2 = -1; |
f38e1219 | 2240 | s->req_compute = 1; |
f38e1219 | 2241 | /* Careful: from this point on 'uptodate' is in the eye |
ac6b53b6 | 2242 | * of raid_run_ops which services 'compute' operations |
f38e1219 DW |
2243 | * before writes. R5_Wantcompute flags a block that will |
2244 | * be R5_UPTODATE by the time it is needed for a | |
2245 | * subsequent operation. | |
2246 | */ | |
2247 | s->uptodate++; | |
1fe797e6 | 2248 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2249 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2250 | set_bit(R5_LOCKED, &dev->flags); |
2251 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2252 | s->locked++; |
2253 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2254 | s->syncing); | |
2255 | } | |
2256 | } | |
2257 | ||
1fe797e6 | 2258 | return 0; |
f38e1219 DW |
2259 | } |
2260 | ||
1fe797e6 DW |
2261 | /** |
2262 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2263 | */ | |
2264 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2265 | struct stripe_head_state *s, int disks) |
2266 | { | |
2267 | int i; | |
f38e1219 | 2268 | |
f38e1219 DW |
2269 | /* look for blocks to read/compute, skip this if a compute |
2270 | * is already in flight, or if the stripe contents are in the | |
2271 | * midst of changing due to a write | |
2272 | */ | |
976ea8d4 | 2273 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2274 | !sh->reconstruct_state) |
f38e1219 | 2275 | for (i = disks; i--; ) |
1fe797e6 | 2276 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2277 | break; |
a4456856 DW |
2278 | set_bit(STRIPE_HANDLE, &sh->state); |
2279 | } | |
2280 | ||
5599becc YT |
2281 | /* fetch_block6 - checks the given member device to see if its data needs |
2282 | * to be read or computed to satisfy a request. | |
2283 | * | |
2284 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2285 | * 0 to tell the loop in handle_stripe_fill6 to continue | |
2286 | */ | |
2287 | static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s, | |
2288 | struct r6_state *r6s, int disk_idx, int disks) | |
a4456856 | 2289 | { |
5599becc YT |
2290 | struct r5dev *dev = &sh->dev[disk_idx]; |
2291 | struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]], | |
2292 | &sh->dev[r6s->failed_num[1]] }; | |
2293 | ||
2294 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2295 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2296 | (dev->toread || | |
2297 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2298 | s->syncing || s->expanding || | |
2299 | (s->failed >= 1 && | |
2300 | (fdev[0]->toread || s->to_write)) || | |
2301 | (s->failed >= 2 && | |
2302 | (fdev[1]->toread || s->to_write)))) { | |
2303 | /* we would like to get this block, possibly by computing it, | |
2304 | * otherwise read it if the backing disk is insync | |
2305 | */ | |
2306 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2307 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2308 | if ((s->uptodate == disks - 1) && | |
2309 | (s->failed && (disk_idx == r6s->failed_num[0] || | |
2310 | disk_idx == r6s->failed_num[1]))) { | |
2311 | /* have disk failed, and we're requested to fetch it; | |
2312 | * do compute it | |
a4456856 | 2313 | */ |
5599becc YT |
2314 | pr_debug("Computing stripe %llu block %d\n", |
2315 | (unsigned long long)sh->sector, disk_idx); | |
2316 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2317 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2318 | set_bit(R5_Wantcompute, &dev->flags); | |
2319 | sh->ops.target = disk_idx; | |
2320 | sh->ops.target2 = -1; /* no 2nd target */ | |
2321 | s->req_compute = 1; | |
2322 | s->uptodate++; | |
2323 | return 1; | |
2324 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2325 | /* Computing 2-failure is *very* expensive; only | |
2326 | * do it if failed >= 2 | |
2327 | */ | |
2328 | int other; | |
2329 | for (other = disks; other--; ) { | |
2330 | if (other == disk_idx) | |
2331 | continue; | |
2332 | if (!test_bit(R5_UPTODATE, | |
2333 | &sh->dev[other].flags)) | |
2334 | break; | |
a4456856 | 2335 | } |
5599becc YT |
2336 | BUG_ON(other < 0); |
2337 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2338 | (unsigned long long)sh->sector, | |
2339 | disk_idx, other); | |
2340 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2341 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2342 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2343 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2344 | sh->ops.target = disk_idx; | |
2345 | sh->ops.target2 = other; | |
2346 | s->uptodate += 2; | |
2347 | s->req_compute = 1; | |
2348 | return 1; | |
2349 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2350 | set_bit(R5_LOCKED, &dev->flags); | |
2351 | set_bit(R5_Wantread, &dev->flags); | |
2352 | s->locked++; | |
2353 | pr_debug("Reading block %d (sync=%d)\n", | |
2354 | disk_idx, s->syncing); | |
a4456856 DW |
2355 | } |
2356 | } | |
5599becc YT |
2357 | |
2358 | return 0; | |
2359 | } | |
2360 | ||
2361 | /** | |
2362 | * handle_stripe_fill6 - read or compute data to satisfy pending requests. | |
2363 | */ | |
2364 | static void handle_stripe_fill6(struct stripe_head *sh, | |
2365 | struct stripe_head_state *s, struct r6_state *r6s, | |
2366 | int disks) | |
2367 | { | |
2368 | int i; | |
2369 | ||
2370 | /* look for blocks to read/compute, skip this if a compute | |
2371 | * is already in flight, or if the stripe contents are in the | |
2372 | * midst of changing due to a write | |
2373 | */ | |
2374 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2375 | !sh->reconstruct_state) | |
2376 | for (i = disks; i--; ) | |
2377 | if (fetch_block6(sh, s, r6s, i, disks)) | |
2378 | break; | |
a4456856 DW |
2379 | set_bit(STRIPE_HANDLE, &sh->state); |
2380 | } | |
2381 | ||
2382 | ||
1fe797e6 | 2383 | /* handle_stripe_clean_event |
a4456856 DW |
2384 | * any written block on an uptodate or failed drive can be returned. |
2385 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2386 | * never LOCKED, so we don't need to test 'failed' directly. | |
2387 | */ | |
1fe797e6 | 2388 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2389 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2390 | { | |
2391 | int i; | |
2392 | struct r5dev *dev; | |
2393 | ||
2394 | for (i = disks; i--; ) | |
2395 | if (sh->dev[i].written) { | |
2396 | dev = &sh->dev[i]; | |
2397 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2398 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2399 | /* We can return any write requests */ | |
2400 | struct bio *wbi, *wbi2; | |
2401 | int bitmap_end = 0; | |
45b4233c | 2402 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2403 | spin_lock_irq(&conf->device_lock); |
2404 | wbi = dev->written; | |
2405 | dev->written = NULL; | |
2406 | while (wbi && wbi->bi_sector < | |
2407 | dev->sector + STRIPE_SECTORS) { | |
2408 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2409 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2410 | md_write_end(conf->mddev); |
2411 | wbi->bi_next = *return_bi; | |
2412 | *return_bi = wbi; | |
2413 | } | |
2414 | wbi = wbi2; | |
2415 | } | |
2416 | if (dev->towrite == NULL) | |
2417 | bitmap_end = 1; | |
2418 | spin_unlock_irq(&conf->device_lock); | |
2419 | if (bitmap_end) | |
2420 | bitmap_endwrite(conf->mddev->bitmap, | |
2421 | sh->sector, | |
2422 | STRIPE_SECTORS, | |
2423 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2424 | 0); | |
2425 | } | |
2426 | } | |
8b3e6cdc DW |
2427 | |
2428 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2429 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2430 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2431 | } |
2432 | ||
1fe797e6 | 2433 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2434 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2435 | { | |
2436 | int rmw = 0, rcw = 0, i; | |
2437 | for (i = disks; i--; ) { | |
2438 | /* would I have to read this buffer for read_modify_write */ | |
2439 | struct r5dev *dev = &sh->dev[i]; | |
2440 | if ((dev->towrite || i == sh->pd_idx) && | |
2441 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2442 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2443 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2444 | if (test_bit(R5_Insync, &dev->flags)) |
2445 | rmw++; | |
2446 | else | |
2447 | rmw += 2*disks; /* cannot read it */ | |
2448 | } | |
2449 | /* Would I have to read this buffer for reconstruct_write */ | |
2450 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2451 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2452 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2453 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2454 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2455 | else |
2456 | rcw += 2*disks; | |
2457 | } | |
2458 | } | |
45b4233c | 2459 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2460 | (unsigned long long)sh->sector, rmw, rcw); |
2461 | set_bit(STRIPE_HANDLE, &sh->state); | |
2462 | if (rmw < rcw && rmw > 0) | |
2463 | /* prefer read-modify-write, but need to get some data */ | |
2464 | for (i = disks; i--; ) { | |
2465 | struct r5dev *dev = &sh->dev[i]; | |
2466 | if ((dev->towrite || i == sh->pd_idx) && | |
2467 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2468 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2469 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2470 | test_bit(R5_Insync, &dev->flags)) { |
2471 | if ( | |
2472 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2473 | pr_debug("Read_old block " |
a4456856 DW |
2474 | "%d for r-m-w\n", i); |
2475 | set_bit(R5_LOCKED, &dev->flags); | |
2476 | set_bit(R5_Wantread, &dev->flags); | |
2477 | s->locked++; | |
2478 | } else { | |
2479 | set_bit(STRIPE_DELAYED, &sh->state); | |
2480 | set_bit(STRIPE_HANDLE, &sh->state); | |
2481 | } | |
2482 | } | |
2483 | } | |
2484 | if (rcw <= rmw && rcw > 0) | |
2485 | /* want reconstruct write, but need to get some data */ | |
2486 | for (i = disks; i--; ) { | |
2487 | struct r5dev *dev = &sh->dev[i]; | |
2488 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2489 | i != sh->pd_idx && | |
2490 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2491 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2492 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2493 | test_bit(R5_Insync, &dev->flags)) { |
2494 | if ( | |
2495 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2496 | pr_debug("Read_old block " |
a4456856 DW |
2497 | "%d for Reconstruct\n", i); |
2498 | set_bit(R5_LOCKED, &dev->flags); | |
2499 | set_bit(R5_Wantread, &dev->flags); | |
2500 | s->locked++; | |
2501 | } else { | |
2502 | set_bit(STRIPE_DELAYED, &sh->state); | |
2503 | set_bit(STRIPE_HANDLE, &sh->state); | |
2504 | } | |
2505 | } | |
2506 | } | |
2507 | /* now if nothing is locked, and if we have enough data, | |
2508 | * we can start a write request | |
2509 | */ | |
f38e1219 DW |
2510 | /* since handle_stripe can be called at any time we need to handle the |
2511 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2512 | * subsequent call wants to start a write request. raid_run_ops only |
2513 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2514 | * simultaneously. If this is not the case then new writes need to be |
2515 | * held off until the compute completes. | |
2516 | */ | |
976ea8d4 DW |
2517 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2518 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2519 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2520 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2521 | } |
2522 | ||
1fe797e6 | 2523 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2524 | struct stripe_head *sh, struct stripe_head_state *s, |
2525 | struct r6_state *r6s, int disks) | |
2526 | { | |
a9b39a74 | 2527 | int rcw = 0, pd_idx = sh->pd_idx, i; |
34e04e87 | 2528 | int qd_idx = sh->qd_idx; |
a9b39a74 YT |
2529 | |
2530 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2531 | for (i = disks; i--; ) { |
2532 | struct r5dev *dev = &sh->dev[i]; | |
a9b39a74 YT |
2533 | /* check if we haven't enough data */ |
2534 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2535 | i != pd_idx && i != qd_idx && | |
2536 | !test_bit(R5_LOCKED, &dev->flags) && | |
2537 | !(test_bit(R5_UPTODATE, &dev->flags) || | |
2538 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2539 | rcw++; | |
2540 | if (!test_bit(R5_Insync, &dev->flags)) | |
2541 | continue; /* it's a failed drive */ | |
2542 | ||
2543 | if ( | |
2544 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2545 | pr_debug("Read_old stripe %llu " | |
2546 | "block %d for Reconstruct\n", | |
2547 | (unsigned long long)sh->sector, i); | |
2548 | set_bit(R5_LOCKED, &dev->flags); | |
2549 | set_bit(R5_Wantread, &dev->flags); | |
2550 | s->locked++; | |
2551 | } else { | |
2552 | pr_debug("Request delayed stripe %llu " | |
2553 | "block %d for Reconstruct\n", | |
2554 | (unsigned long long)sh->sector, i); | |
2555 | set_bit(STRIPE_DELAYED, &sh->state); | |
2556 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2557 | } |
2558 | } | |
2559 | } | |
a4456856 DW |
2560 | /* now if nothing is locked, and if we have enough data, we can start a |
2561 | * write request | |
2562 | */ | |
a9b39a74 YT |
2563 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2564 | s->locked == 0 && rcw == 0 && | |
a4456856 | 2565 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { |
a9b39a74 | 2566 | schedule_reconstruction(sh, s, 1, 0); |
a4456856 DW |
2567 | } |
2568 | } | |
2569 | ||
2570 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2571 | struct stripe_head_state *s, int disks) | |
2572 | { | |
ecc65c9b | 2573 | struct r5dev *dev = NULL; |
bd2ab670 | 2574 | |
a4456856 | 2575 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2576 | |
ecc65c9b DW |
2577 | switch (sh->check_state) { |
2578 | case check_state_idle: | |
2579 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2580 | if (s->failed == 0) { |
bd2ab670 | 2581 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2582 | sh->check_state = check_state_run; |
2583 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2584 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2585 | s->uptodate--; |
ecc65c9b | 2586 | break; |
bd2ab670 | 2587 | } |
ecc65c9b DW |
2588 | dev = &sh->dev[s->failed_num]; |
2589 | /* fall through */ | |
2590 | case check_state_compute_result: | |
2591 | sh->check_state = check_state_idle; | |
2592 | if (!dev) | |
2593 | dev = &sh->dev[sh->pd_idx]; | |
2594 | ||
2595 | /* check that a write has not made the stripe insync */ | |
2596 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2597 | break; | |
c8894419 | 2598 | |
a4456856 | 2599 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2600 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2601 | BUG_ON(s->uptodate != disks); | |
2602 | ||
2603 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2604 | s->locked++; |
a4456856 | 2605 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2606 | |
a4456856 | 2607 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2608 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2609 | break; |
2610 | case check_state_run: | |
2611 | break; /* we will be called again upon completion */ | |
2612 | case check_state_check_result: | |
2613 | sh->check_state = check_state_idle; | |
2614 | ||
2615 | /* if a failure occurred during the check operation, leave | |
2616 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2617 | */ | |
2618 | if (s->failed) | |
2619 | break; | |
2620 | ||
2621 | /* handle a successful check operation, if parity is correct | |
2622 | * we are done. Otherwise update the mismatch count and repair | |
2623 | * parity if !MD_RECOVERY_CHECK | |
2624 | */ | |
ad283ea4 | 2625 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2626 | /* parity is correct (on disc, |
2627 | * not in buffer any more) | |
2628 | */ | |
2629 | set_bit(STRIPE_INSYNC, &sh->state); | |
2630 | else { | |
2631 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2632 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2633 | /* don't try to repair!! */ | |
2634 | set_bit(STRIPE_INSYNC, &sh->state); | |
2635 | else { | |
2636 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2637 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2638 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2639 | set_bit(R5_Wantcompute, | |
2640 | &sh->dev[sh->pd_idx].flags); | |
2641 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2642 | sh->ops.target2 = -1; |
ecc65c9b DW |
2643 | s->uptodate++; |
2644 | } | |
2645 | } | |
2646 | break; | |
2647 | case check_state_compute_run: | |
2648 | break; | |
2649 | default: | |
2650 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2651 | __func__, sh->check_state, | |
2652 | (unsigned long long) sh->sector); | |
2653 | BUG(); | |
a4456856 DW |
2654 | } |
2655 | } | |
2656 | ||
2657 | ||
2658 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
36d1c647 DW |
2659 | struct stripe_head_state *s, |
2660 | struct r6_state *r6s, int disks) | |
a4456856 | 2661 | { |
a4456856 | 2662 | int pd_idx = sh->pd_idx; |
34e04e87 | 2663 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2664 | struct r5dev *dev; |
a4456856 DW |
2665 | |
2666 | set_bit(STRIPE_HANDLE, &sh->state); | |
2667 | ||
2668 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2669 | |
a4456856 DW |
2670 | /* Want to check and possibly repair P and Q. |
2671 | * However there could be one 'failed' device, in which | |
2672 | * case we can only check one of them, possibly using the | |
2673 | * other to generate missing data | |
2674 | */ | |
2675 | ||
d82dfee0 DW |
2676 | switch (sh->check_state) { |
2677 | case check_state_idle: | |
2678 | /* start a new check operation if there are < 2 failures */ | |
a4456856 | 2679 | if (s->failed == r6s->q_failed) { |
d82dfee0 | 2680 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2681 | * makes sense to check P (If anything else were failed, |
2682 | * we would have used P to recreate it). | |
2683 | */ | |
d82dfee0 | 2684 | sh->check_state = check_state_run; |
a4456856 DW |
2685 | } |
2686 | if (!r6s->q_failed && s->failed < 2) { | |
d82dfee0 | 2687 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2688 | * anything, so it makes sense to check it |
2689 | */ | |
d82dfee0 DW |
2690 | if (sh->check_state == check_state_run) |
2691 | sh->check_state = check_state_run_pq; | |
2692 | else | |
2693 | sh->check_state = check_state_run_q; | |
a4456856 | 2694 | } |
a4456856 | 2695 | |
d82dfee0 DW |
2696 | /* discard potentially stale zero_sum_result */ |
2697 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2698 | |
d82dfee0 DW |
2699 | if (sh->check_state == check_state_run) { |
2700 | /* async_xor_zero_sum destroys the contents of P */ | |
2701 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2702 | s->uptodate--; | |
a4456856 | 2703 | } |
d82dfee0 DW |
2704 | if (sh->check_state >= check_state_run && |
2705 | sh->check_state <= check_state_run_pq) { | |
2706 | /* async_syndrome_zero_sum preserves P and Q, so | |
2707 | * no need to mark them !uptodate here | |
2708 | */ | |
2709 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2710 | break; | |
a4456856 DW |
2711 | } |
2712 | ||
d82dfee0 DW |
2713 | /* we have 2-disk failure */ |
2714 | BUG_ON(s->failed != 2); | |
2715 | /* fall through */ | |
2716 | case check_state_compute_result: | |
2717 | sh->check_state = check_state_idle; | |
a4456856 | 2718 | |
d82dfee0 DW |
2719 | /* check that a write has not made the stripe insync */ |
2720 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2721 | break; | |
a4456856 DW |
2722 | |
2723 | /* now write out any block on a failed drive, | |
d82dfee0 | 2724 | * or P or Q if they were recomputed |
a4456856 | 2725 | */ |
d82dfee0 | 2726 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 DW |
2727 | if (s->failed == 2) { |
2728 | dev = &sh->dev[r6s->failed_num[1]]; | |
2729 | s->locked++; | |
2730 | set_bit(R5_LOCKED, &dev->flags); | |
2731 | set_bit(R5_Wantwrite, &dev->flags); | |
2732 | } | |
2733 | if (s->failed >= 1) { | |
2734 | dev = &sh->dev[r6s->failed_num[0]]; | |
2735 | s->locked++; | |
2736 | set_bit(R5_LOCKED, &dev->flags); | |
2737 | set_bit(R5_Wantwrite, &dev->flags); | |
2738 | } | |
d82dfee0 | 2739 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2740 | dev = &sh->dev[pd_idx]; |
2741 | s->locked++; | |
2742 | set_bit(R5_LOCKED, &dev->flags); | |
2743 | set_bit(R5_Wantwrite, &dev->flags); | |
2744 | } | |
d82dfee0 | 2745 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2746 | dev = &sh->dev[qd_idx]; |
2747 | s->locked++; | |
2748 | set_bit(R5_LOCKED, &dev->flags); | |
2749 | set_bit(R5_Wantwrite, &dev->flags); | |
2750 | } | |
2751 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2752 | ||
2753 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2754 | break; |
2755 | case check_state_run: | |
2756 | case check_state_run_q: | |
2757 | case check_state_run_pq: | |
2758 | break; /* we will be called again upon completion */ | |
2759 | case check_state_check_result: | |
2760 | sh->check_state = check_state_idle; | |
2761 | ||
2762 | /* handle a successful check operation, if parity is correct | |
2763 | * we are done. Otherwise update the mismatch count and repair | |
2764 | * parity if !MD_RECOVERY_CHECK | |
2765 | */ | |
2766 | if (sh->ops.zero_sum_result == 0) { | |
2767 | /* both parities are correct */ | |
2768 | if (!s->failed) | |
2769 | set_bit(STRIPE_INSYNC, &sh->state); | |
2770 | else { | |
2771 | /* in contrast to the raid5 case we can validate | |
2772 | * parity, but still have a failure to write | |
2773 | * back | |
2774 | */ | |
2775 | sh->check_state = check_state_compute_result; | |
2776 | /* Returning at this point means that we may go | |
2777 | * off and bring p and/or q uptodate again so | |
2778 | * we make sure to check zero_sum_result again | |
2779 | * to verify if p or q need writeback | |
2780 | */ | |
2781 | } | |
2782 | } else { | |
2783 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2784 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2785 | /* don't try to repair!! */ | |
2786 | set_bit(STRIPE_INSYNC, &sh->state); | |
2787 | else { | |
2788 | int *target = &sh->ops.target; | |
2789 | ||
2790 | sh->ops.target = -1; | |
2791 | sh->ops.target2 = -1; | |
2792 | sh->check_state = check_state_compute_run; | |
2793 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2794 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2795 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2796 | set_bit(R5_Wantcompute, | |
2797 | &sh->dev[pd_idx].flags); | |
2798 | *target = pd_idx; | |
2799 | target = &sh->ops.target2; | |
2800 | s->uptodate++; | |
2801 | } | |
2802 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2803 | set_bit(R5_Wantcompute, | |
2804 | &sh->dev[qd_idx].flags); | |
2805 | *target = qd_idx; | |
2806 | s->uptodate++; | |
2807 | } | |
2808 | } | |
2809 | } | |
2810 | break; | |
2811 | case check_state_compute_run: | |
2812 | break; | |
2813 | default: | |
2814 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2815 | __func__, sh->check_state, | |
2816 | (unsigned long long) sh->sector); | |
2817 | BUG(); | |
a4456856 DW |
2818 | } |
2819 | } | |
2820 | ||
2821 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2822 | struct r6_state *r6s) | |
2823 | { | |
2824 | int i; | |
2825 | ||
2826 | /* We have read all the blocks in this stripe and now we need to | |
2827 | * copy some of them into a target stripe for expand. | |
2828 | */ | |
f0a50d37 | 2829 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2830 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2831 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 2832 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 2833 | int dd_idx, j; |
a4456856 | 2834 | struct stripe_head *sh2; |
a08abd8c | 2835 | struct async_submit_ctl submit; |
a4456856 | 2836 | |
784052ec | 2837 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
2838 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2839 | &dd_idx, NULL); | |
a8c906ca | 2840 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
2841 | if (sh2 == NULL) |
2842 | /* so far only the early blocks of this stripe | |
2843 | * have been requested. When later blocks | |
2844 | * get requested, we will try again | |
2845 | */ | |
2846 | continue; | |
2847 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2848 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2849 | /* must have already done this block */ | |
2850 | release_stripe(sh2); | |
2851 | continue; | |
2852 | } | |
f0a50d37 DW |
2853 | |
2854 | /* place all the copies on one channel */ | |
a08abd8c | 2855 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 2856 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 2857 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 2858 | &submit); |
f0a50d37 | 2859 | |
a4456856 DW |
2860 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2861 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2862 | for (j = 0; j < conf->raid_disks; j++) | |
2863 | if (j != sh2->pd_idx && | |
d0dabf7e | 2864 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
2865 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2866 | break; | |
2867 | if (j == conf->raid_disks) { | |
2868 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2869 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2870 | } | |
2871 | release_stripe(sh2); | |
f0a50d37 | 2872 | |
a4456856 | 2873 | } |
a2e08551 N |
2874 | /* done submitting copies, wait for them to complete */ |
2875 | if (tx) { | |
2876 | async_tx_ack(tx); | |
2877 | dma_wait_for_async_tx(tx); | |
2878 | } | |
a4456856 | 2879 | } |
1da177e4 | 2880 | |
6bfe0b49 | 2881 | |
1da177e4 LT |
2882 | /* |
2883 | * handle_stripe - do things to a stripe. | |
2884 | * | |
2885 | * We lock the stripe and then examine the state of various bits | |
2886 | * to see what needs to be done. | |
2887 | * Possible results: | |
2888 | * return some read request which now have data | |
2889 | * return some write requests which are safely on disc | |
2890 | * schedule a read on some buffers | |
2891 | * schedule a write of some buffers | |
2892 | * return confirmation of parity correctness | |
2893 | * | |
1da177e4 LT |
2894 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2895 | * get BH_Lock set before the stripe lock is released. | |
2896 | * | |
2897 | */ | |
a4456856 | 2898 | |
1442577b | 2899 | static void handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2900 | { |
2901 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2902 | int disks = sh->disks, i; |
2903 | struct bio *return_bi = NULL; | |
2904 | struct stripe_head_state s; | |
1da177e4 | 2905 | struct r5dev *dev; |
6bfe0b49 | 2906 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2907 | int prexor; |
1da177e4 | 2908 | |
a4456856 | 2909 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
2910 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
2911 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
2912 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
2913 | sh->reconstruct_state); | |
1da177e4 LT |
2914 | |
2915 | spin_lock(&sh->lock); | |
2916 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2917 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2918 | ||
a4456856 DW |
2919 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2920 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2921 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 2922 | |
83de75cc | 2923 | /* Now to look around and see what can be done */ |
9910f16a | 2924 | rcu_read_lock(); |
1da177e4 LT |
2925 | for (i=disks; i--; ) { |
2926 | mdk_rdev_t *rdev; | |
a9f326eb N |
2927 | |
2928 | dev = &sh->dev[i]; | |
1da177e4 | 2929 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2930 | |
b5e98d65 DW |
2931 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2932 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2933 | dev->towrite, dev->written); | |
2934 | ||
2935 | /* maybe we can request a biofill operation | |
2936 | * | |
2937 | * new wantfill requests are only permitted while | |
83de75cc | 2938 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
2939 | */ |
2940 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 2941 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 2942 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
2943 | |
2944 | /* now count some things */ | |
a4456856 DW |
2945 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2946 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2947 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2948 | |
b5e98d65 DW |
2949 | if (test_bit(R5_Wantfill, &dev->flags)) |
2950 | s.to_fill++; | |
2951 | else if (dev->toread) | |
a4456856 | 2952 | s.to_read++; |
1da177e4 | 2953 | if (dev->towrite) { |
a4456856 | 2954 | s.to_write++; |
1da177e4 | 2955 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2956 | s.non_overwrite++; |
1da177e4 | 2957 | } |
a4456856 DW |
2958 | if (dev->written) |
2959 | s.written++; | |
9910f16a | 2960 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
2961 | if (blocked_rdev == NULL && |
2962 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
2963 | blocked_rdev = rdev; |
2964 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 2965 | } |
b2d444d7 | 2966 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 2967 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
2968 | clear_bit(R5_ReadError, &dev->flags); |
2969 | clear_bit(R5_ReWrite, &dev->flags); | |
2970 | } | |
b2d444d7 | 2971 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 2972 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
2973 | s.failed++; |
2974 | s.failed_num = i; | |
1da177e4 LT |
2975 | } else |
2976 | set_bit(R5_Insync, &dev->flags); | |
2977 | } | |
9910f16a | 2978 | rcu_read_unlock(); |
b5e98d65 | 2979 | |
6bfe0b49 | 2980 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
2981 | if (s.syncing || s.expanding || s.expanded || |
2982 | s.to_write || s.written) { | |
2983 | set_bit(STRIPE_HANDLE, &sh->state); | |
2984 | goto unlock; | |
2985 | } | |
2986 | /* There is nothing for the blocked_rdev to block */ | |
2987 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
2988 | blocked_rdev = NULL; | |
6bfe0b49 DW |
2989 | } |
2990 | ||
83de75cc DW |
2991 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
2992 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
2993 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
2994 | } | |
b5e98d65 | 2995 | |
45b4233c | 2996 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 2997 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
2998 | s.locked, s.uptodate, s.to_read, s.to_write, |
2999 | s.failed, s.failed_num); | |
1da177e4 LT |
3000 | /* check if the array has lost two devices and, if so, some requests might |
3001 | * need to be failed | |
3002 | */ | |
a4456856 | 3003 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3004 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3005 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
3006 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3007 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3008 | s.syncing = 0; |
1da177e4 LT |
3009 | } |
3010 | ||
3011 | /* might be able to return some write requests if the parity block | |
3012 | * is safe, or on a failed drive | |
3013 | */ | |
3014 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
3015 | if ( s.written && |
3016 | ((test_bit(R5_Insync, &dev->flags) && | |
3017 | !test_bit(R5_LOCKED, &dev->flags) && | |
3018 | test_bit(R5_UPTODATE, &dev->flags)) || | |
3019 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 3020 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
3021 | |
3022 | /* Now we might consider reading some blocks, either to check/generate | |
3023 | * parity, or to satisfy requests | |
3024 | * or to load a block that is being partially written. | |
3025 | */ | |
a4456856 | 3026 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 3027 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3028 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 3029 | |
e33129d8 DW |
3030 | /* Now we check to see if any write operations have recently |
3031 | * completed | |
3032 | */ | |
e0a115e5 | 3033 | prexor = 0; |
d8ee0728 | 3034 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 3035 | prexor = 1; |
d8ee0728 DW |
3036 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
3037 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 3038 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
3039 | |
3040 | /* All the 'written' buffers and the parity block are ready to | |
3041 | * be written back to disk | |
3042 | */ | |
3043 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3044 | for (i = disks; i--; ) { | |
3045 | dev = &sh->dev[i]; | |
3046 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3047 | (i == sh->pd_idx || dev->written)) { | |
3048 | pr_debug("Writing block %d\n", i); | |
3049 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
3050 | if (prexor) |
3051 | continue; | |
e33129d8 DW |
3052 | if (!test_bit(R5_Insync, &dev->flags) || |
3053 | (i == sh->pd_idx && s.failed == 0)) | |
3054 | set_bit(STRIPE_INSYNC, &sh->state); | |
3055 | } | |
3056 | } | |
3057 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
3058 | atomic_dec(&conf->preread_active_stripes); | |
3059 | if (atomic_read(&conf->preread_active_stripes) < | |
3060 | IO_THRESHOLD) | |
3061 | md_wakeup_thread(conf->mddev->thread); | |
3062 | } | |
3063 | } | |
3064 | ||
3065 | /* Now to consider new write requests and what else, if anything | |
3066 | * should be read. We do not handle new writes when: | |
3067 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3068 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3069 | * block. | |
3070 | */ | |
600aa109 | 3071 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 3072 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
3073 | |
3074 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
3075 | * Any reads will already have been scheduled, so we just see if enough |
3076 | * data is available. The parity check is held off while parity | |
3077 | * dependent operations are in flight. | |
1da177e4 | 3078 | */ |
ecc65c9b DW |
3079 | if (sh->check_state || |
3080 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 3081 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 3082 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 3083 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 3084 | |
a4456856 | 3085 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
3086 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3087 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3088 | } | |
4e5314b5 N |
3089 | |
3090 | /* If the failed drive is just a ReadError, then we might need to progress | |
3091 | * the repair/check process | |
3092 | */ | |
a4456856 DW |
3093 | if (s.failed == 1 && !conf->mddev->ro && |
3094 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
3095 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
3096 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 3097 | ) { |
a4456856 | 3098 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
3099 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
3100 | set_bit(R5_Wantwrite, &dev->flags); | |
3101 | set_bit(R5_ReWrite, &dev->flags); | |
3102 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3103 | s.locked++; |
4e5314b5 N |
3104 | } else { |
3105 | /* let's read it back */ | |
3106 | set_bit(R5_Wantread, &dev->flags); | |
3107 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3108 | s.locked++; |
4e5314b5 N |
3109 | } |
3110 | } | |
3111 | ||
600aa109 DW |
3112 | /* Finish reconstruct operations initiated by the expansion process */ |
3113 | if (sh->reconstruct_state == reconstruct_state_result) { | |
ab69ae12 | 3114 | struct stripe_head *sh2 |
a8c906ca | 3115 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3116 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3117 | /* sh cannot be written until sh2 has been read. | |
3118 | * so arrange for sh to be delayed a little | |
3119 | */ | |
3120 | set_bit(STRIPE_DELAYED, &sh->state); | |
3121 | set_bit(STRIPE_HANDLE, &sh->state); | |
3122 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3123 | &sh2->state)) | |
3124 | atomic_inc(&conf->preread_active_stripes); | |
3125 | release_stripe(sh2); | |
3126 | goto unlock; | |
3127 | } | |
3128 | if (sh2) | |
3129 | release_stripe(sh2); | |
3130 | ||
600aa109 | 3131 | sh->reconstruct_state = reconstruct_state_idle; |
f0a50d37 | 3132 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 3133 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 3134 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 3135 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 3136 | s.locked++; |
23397883 | 3137 | } |
f0a50d37 DW |
3138 | } |
3139 | ||
3140 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 3141 | !sh->reconstruct_state) { |
f0a50d37 DW |
3142 | /* Need to write out all blocks after computing parity */ |
3143 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3144 | stripe_set_idx(sh->sector, conf, 0, sh); |
c0f7bddb | 3145 | schedule_reconstruction(sh, &s, 1, 1); |
600aa109 | 3146 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 3147 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 3148 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
3149 | wake_up(&conf->wait_for_overlap); |
3150 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3151 | } | |
3152 | ||
0f94e87c | 3153 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3154 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3155 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 3156 | |
6bfe0b49 | 3157 | unlock: |
1da177e4 LT |
3158 | spin_unlock(&sh->lock); |
3159 | ||
6bfe0b49 DW |
3160 | /* wait for this device to become unblocked */ |
3161 | if (unlikely(blocked_rdev)) | |
3162 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3163 | ||
600aa109 | 3164 | if (s.ops_request) |
ac6b53b6 | 3165 | raid_run_ops(sh, s.ops_request); |
d84e0f10 | 3166 | |
c4e5ac0a | 3167 | ops_run_io(sh, &s); |
1da177e4 | 3168 | |
a4456856 | 3169 | return_io(return_bi); |
1da177e4 LT |
3170 | } |
3171 | ||
1442577b | 3172 | static void handle_stripe6(struct stripe_head *sh) |
1da177e4 | 3173 | { |
bff61975 | 3174 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 3175 | int disks = sh->disks; |
a4456856 | 3176 | struct bio *return_bi = NULL; |
34e04e87 | 3177 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx; |
a4456856 DW |
3178 | struct stripe_head_state s; |
3179 | struct r6_state r6s; | |
16a53ecc | 3180 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 3181 | mdk_rdev_t *blocked_rdev = NULL; |
1da177e4 | 3182 | |
45b4233c | 3183 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
6c0069c0 | 3184 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", |
a4456856 | 3185 | (unsigned long long)sh->sector, sh->state, |
6c0069c0 YT |
3186 | atomic_read(&sh->count), pd_idx, qd_idx, |
3187 | sh->check_state, sh->reconstruct_state); | |
a4456856 | 3188 | memset(&s, 0, sizeof(s)); |
72626685 | 3189 | |
16a53ecc N |
3190 | spin_lock(&sh->lock); |
3191 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3192 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3193 | ||
a4456856 DW |
3194 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3195 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3196 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 3197 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3198 | |
3199 | rcu_read_lock(); | |
16a53ecc N |
3200 | for (i=disks; i--; ) { |
3201 | mdk_rdev_t *rdev; | |
3202 | dev = &sh->dev[i]; | |
3203 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 3204 | |
45b4233c | 3205 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc | 3206 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
6c0069c0 YT |
3207 | /* maybe we can reply to a read |
3208 | * | |
3209 | * new wantfill requests are only permitted while | |
3210 | * ops_complete_biofill is guaranteed to be inactive | |
3211 | */ | |
3212 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3213 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3214 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3215 | |
16a53ecc | 3216 | /* now count some things */ |
a4456856 DW |
3217 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3218 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
2d6e4ecc DW |
3219 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
3220 | s.compute++; | |
3221 | BUG_ON(s.compute > 2); | |
3222 | } | |
1da177e4 | 3223 | |
6c0069c0 YT |
3224 | if (test_bit(R5_Wantfill, &dev->flags)) { |
3225 | s.to_fill++; | |
3226 | } else if (dev->toread) | |
a4456856 | 3227 | s.to_read++; |
16a53ecc | 3228 | if (dev->towrite) { |
a4456856 | 3229 | s.to_write++; |
16a53ecc | 3230 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3231 | s.non_overwrite++; |
16a53ecc | 3232 | } |
a4456856 DW |
3233 | if (dev->written) |
3234 | s.written++; | |
16a53ecc | 3235 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3236 | if (blocked_rdev == NULL && |
3237 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3238 | blocked_rdev = rdev; |
3239 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3240 | } |
16a53ecc N |
3241 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3242 | /* The ReadError flag will just be confusing now */ | |
3243 | clear_bit(R5_ReadError, &dev->flags); | |
3244 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3245 | } |
16a53ecc N |
3246 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3247 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3248 | if (s.failed < 2) |
3249 | r6s.failed_num[s.failed] = i; | |
3250 | s.failed++; | |
16a53ecc N |
3251 | } else |
3252 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3253 | } |
3254 | rcu_read_unlock(); | |
6bfe0b49 DW |
3255 | |
3256 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3257 | if (s.syncing || s.expanding || s.expanded || |
3258 | s.to_write || s.written) { | |
3259 | set_bit(STRIPE_HANDLE, &sh->state); | |
3260 | goto unlock; | |
3261 | } | |
3262 | /* There is nothing for the blocked_rdev to block */ | |
3263 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3264 | blocked_rdev = NULL; | |
6bfe0b49 | 3265 | } |
ac4090d2 | 3266 | |
6c0069c0 YT |
3267 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3268 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3269 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3270 | } | |
3271 | ||
45b4233c | 3272 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3273 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3274 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3275 | r6s.failed_num[0], r6s.failed_num[1]); | |
3276 | /* check if the array has lost >2 devices and, if so, some requests | |
3277 | * might need to be failed | |
16a53ecc | 3278 | */ |
a4456856 | 3279 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3280 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3281 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3282 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3283 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3284 | s.syncing = 0; |
16a53ecc N |
3285 | } |
3286 | ||
3287 | /* | |
3288 | * might be able to return some write requests if the parity blocks | |
3289 | * are safe, or on a failed drive | |
3290 | */ | |
3291 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3292 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3293 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
34e04e87 N |
3294 | qdev = &sh->dev[qd_idx]; |
3295 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx) | |
3296 | || (s.failed >= 2 && r6s.failed_num[1] == qd_idx); | |
a4456856 DW |
3297 | |
3298 | if ( s.written && | |
3299 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3300 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3301 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3302 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3303 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3304 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3305 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3306 | |
3307 | /* Now we might consider reading some blocks, either to check/generate | |
3308 | * parity, or to satisfy requests | |
3309 | * or to load a block that is being partially written. | |
3310 | */ | |
a4456856 | 3311 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
6c0069c0 | 3312 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3313 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc | 3314 | |
6c0069c0 YT |
3315 | /* Now we check to see if any write operations have recently |
3316 | * completed | |
3317 | */ | |
3318 | if (sh->reconstruct_state == reconstruct_state_drain_result) { | |
3319 | int qd_idx = sh->qd_idx; | |
3320 | ||
3321 | sh->reconstruct_state = reconstruct_state_idle; | |
3322 | /* All the 'written' buffers and the parity blocks are ready to | |
3323 | * be written back to disk | |
3324 | */ | |
3325 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3326 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags)); | |
3327 | for (i = disks; i--; ) { | |
3328 | dev = &sh->dev[i]; | |
3329 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3330 | (i == sh->pd_idx || i == qd_idx || | |
3331 | dev->written)) { | |
3332 | pr_debug("Writing block %d\n", i); | |
3333 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); | |
3334 | set_bit(R5_Wantwrite, &dev->flags); | |
3335 | if (!test_bit(R5_Insync, &dev->flags) || | |
3336 | ((i == sh->pd_idx || i == qd_idx) && | |
3337 | s.failed == 0)) | |
3338 | set_bit(STRIPE_INSYNC, &sh->state); | |
3339 | } | |
3340 | } | |
3341 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
3342 | atomic_dec(&conf->preread_active_stripes); | |
3343 | if (atomic_read(&conf->preread_active_stripes) < | |
3344 | IO_THRESHOLD) | |
3345 | md_wakeup_thread(conf->mddev->thread); | |
3346 | } | |
3347 | } | |
3348 | ||
a9b39a74 YT |
3349 | /* Now to consider new write requests and what else, if anything |
3350 | * should be read. We do not handle new writes when: | |
3351 | * 1/ A 'write' operation (copy+gen_syndrome) is already in flight. | |
3352 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3353 | * block. | |
3354 | */ | |
3355 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
1fe797e6 | 3356 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3357 | |
3358 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 | 3359 | * Any reads will already have been scheduled, so we just see if enough |
6c0069c0 YT |
3360 | * data is available. The parity check is held off while parity |
3361 | * dependent operations are in flight. | |
16a53ecc | 3362 | */ |
6c0069c0 YT |
3363 | if (sh->check_state || |
3364 | (s.syncing && s.locked == 0 && | |
3365 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3366 | !test_bit(STRIPE_INSYNC, &sh->state))) | |
36d1c647 | 3367 | handle_parity_checks6(conf, sh, &s, &r6s, disks); |
16a53ecc | 3368 | |
a4456856 | 3369 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3370 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3371 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3372 | } | |
3373 | ||
3374 | /* If the failed drives are just a ReadError, then we might need | |
3375 | * to progress the repair/check process | |
3376 | */ | |
a4456856 DW |
3377 | if (s.failed <= 2 && !conf->mddev->ro) |
3378 | for (i = 0; i < s.failed; i++) { | |
3379 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3380 | if (test_bit(R5_ReadError, &dev->flags) |
3381 | && !test_bit(R5_LOCKED, &dev->flags) | |
3382 | && test_bit(R5_UPTODATE, &dev->flags) | |
3383 | ) { | |
3384 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3385 | set_bit(R5_Wantwrite, &dev->flags); | |
3386 | set_bit(R5_ReWrite, &dev->flags); | |
3387 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3388 | s.locked++; |
16a53ecc N |
3389 | } else { |
3390 | /* let's read it back */ | |
3391 | set_bit(R5_Wantread, &dev->flags); | |
3392 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3393 | s.locked++; |
16a53ecc N |
3394 | } |
3395 | } | |
3396 | } | |
f416885e | 3397 | |
6c0069c0 YT |
3398 | /* Finish reconstruct operations initiated by the expansion process */ |
3399 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3400 | sh->reconstruct_state = reconstruct_state_idle; | |
3401 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3402 | for (i = conf->raid_disks; i--; ) { | |
3403 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3404 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3405 | s.locked++; | |
3406 | } | |
3407 | } | |
3408 | ||
3409 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
3410 | !sh->reconstruct_state) { | |
ab69ae12 | 3411 | struct stripe_head *sh2 |
a8c906ca | 3412 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3413 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3414 | /* sh cannot be written until sh2 has been read. | |
3415 | * so arrange for sh to be delayed a little | |
3416 | */ | |
3417 | set_bit(STRIPE_DELAYED, &sh->state); | |
3418 | set_bit(STRIPE_HANDLE, &sh->state); | |
3419 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3420 | &sh2->state)) | |
3421 | atomic_inc(&conf->preread_active_stripes); | |
3422 | release_stripe(sh2); | |
3423 | goto unlock; | |
3424 | } | |
3425 | if (sh2) | |
3426 | release_stripe(sh2); | |
3427 | ||
f416885e N |
3428 | /* Need to write out all blocks after computing P&Q */ |
3429 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3430 | stripe_set_idx(sh->sector, conf, 0, sh); |
6c0069c0 YT |
3431 | schedule_reconstruction(sh, &s, 1, 1); |
3432 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
f416885e N |
3433 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3434 | atomic_dec(&conf->reshape_stripes); | |
3435 | wake_up(&conf->wait_for_overlap); | |
3436 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3437 | } | |
3438 | ||
0f94e87c | 3439 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3440 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3441 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3442 | |
6bfe0b49 | 3443 | unlock: |
16a53ecc N |
3444 | spin_unlock(&sh->lock); |
3445 | ||
6bfe0b49 DW |
3446 | /* wait for this device to become unblocked */ |
3447 | if (unlikely(blocked_rdev)) | |
3448 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3449 | ||
6c0069c0 YT |
3450 | if (s.ops_request) |
3451 | raid_run_ops(sh, s.ops_request); | |
3452 | ||
f0e43bcd | 3453 | ops_run_io(sh, &s); |
16a53ecc | 3454 | |
f0e43bcd | 3455 | return_io(return_bi); |
16a53ecc N |
3456 | } |
3457 | ||
1442577b | 3458 | static void handle_stripe(struct stripe_head *sh) |
16a53ecc N |
3459 | { |
3460 | if (sh->raid_conf->level == 6) | |
1442577b | 3461 | handle_stripe6(sh); |
16a53ecc | 3462 | else |
1442577b | 3463 | handle_stripe5(sh); |
16a53ecc N |
3464 | } |
3465 | ||
16a53ecc N |
3466 | static void raid5_activate_delayed(raid5_conf_t *conf) |
3467 | { | |
3468 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3469 | while (!list_empty(&conf->delayed_list)) { | |
3470 | struct list_head *l = conf->delayed_list.next; | |
3471 | struct stripe_head *sh; | |
3472 | sh = list_entry(l, struct stripe_head, lru); | |
3473 | list_del_init(l); | |
3474 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3475 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3476 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3477 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3478 | } |
6ed3003c N |
3479 | } else |
3480 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3481 | } |
3482 | ||
3483 | static void activate_bit_delay(raid5_conf_t *conf) | |
3484 | { | |
3485 | /* device_lock is held */ | |
3486 | struct list_head head; | |
3487 | list_add(&head, &conf->bitmap_list); | |
3488 | list_del_init(&conf->bitmap_list); | |
3489 | while (!list_empty(&head)) { | |
3490 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3491 | list_del_init(&sh->lru); | |
3492 | atomic_inc(&sh->count); | |
3493 | __release_stripe(conf, sh); | |
3494 | } | |
3495 | } | |
3496 | ||
3497 | static void unplug_slaves(mddev_t *mddev) | |
3498 | { | |
070ec55d | 3499 | raid5_conf_t *conf = mddev->private; |
16a53ecc N |
3500 | int i; |
3501 | ||
3502 | rcu_read_lock(); | |
f001a70c | 3503 | for (i = 0; i < conf->raid_disks; i++) { |
16a53ecc N |
3504 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); |
3505 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3506 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3507 | |
3508 | atomic_inc(&rdev->nr_pending); | |
3509 | rcu_read_unlock(); | |
3510 | ||
2ad8b1ef | 3511 | blk_unplug(r_queue); |
16a53ecc N |
3512 | |
3513 | rdev_dec_pending(rdev, mddev); | |
3514 | rcu_read_lock(); | |
3515 | } | |
3516 | } | |
3517 | rcu_read_unlock(); | |
3518 | } | |
3519 | ||
165125e1 | 3520 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3521 | { |
3522 | mddev_t *mddev = q->queuedata; | |
070ec55d | 3523 | raid5_conf_t *conf = mddev->private; |
16a53ecc N |
3524 | unsigned long flags; |
3525 | ||
3526 | spin_lock_irqsave(&conf->device_lock, flags); | |
3527 | ||
3528 | if (blk_remove_plug(q)) { | |
3529 | conf->seq_flush++; | |
3530 | raid5_activate_delayed(conf); | |
72626685 | 3531 | } |
1da177e4 LT |
3532 | md_wakeup_thread(mddev->thread); |
3533 | ||
3534 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3535 | ||
3536 | unplug_slaves(mddev); | |
3537 | } | |
3538 | ||
f022b2fd N |
3539 | static int raid5_congested(void *data, int bits) |
3540 | { | |
3541 | mddev_t *mddev = data; | |
070ec55d | 3542 | raid5_conf_t *conf = mddev->private; |
f022b2fd N |
3543 | |
3544 | /* No difference between reads and writes. Just check | |
3545 | * how busy the stripe_cache is | |
3546 | */ | |
3fa841d7 N |
3547 | |
3548 | if (mddev_congested(mddev, bits)) | |
3549 | return 1; | |
f022b2fd N |
3550 | if (conf->inactive_blocked) |
3551 | return 1; | |
3552 | if (conf->quiesce) | |
3553 | return 1; | |
3554 | if (list_empty_careful(&conf->inactive_list)) | |
3555 | return 1; | |
3556 | ||
3557 | return 0; | |
3558 | } | |
3559 | ||
23032a0e RBJ |
3560 | /* We want read requests to align with chunks where possible, |
3561 | * but write requests don't need to. | |
3562 | */ | |
cc371e66 AK |
3563 | static int raid5_mergeable_bvec(struct request_queue *q, |
3564 | struct bvec_merge_data *bvm, | |
3565 | struct bio_vec *biovec) | |
23032a0e RBJ |
3566 | { |
3567 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3568 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3569 | int max; |
9d8f0363 | 3570 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3571 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3572 | |
cc371e66 | 3573 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3574 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3575 | ||
664e7c41 AN |
3576 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3577 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3578 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3579 | if (max < 0) max = 0; | |
3580 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3581 | return biovec->bv_len; | |
3582 | else | |
3583 | return max; | |
3584 | } | |
3585 | ||
f679623f RBJ |
3586 | |
3587 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3588 | { | |
3589 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3590 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3591 | unsigned int bio_sectors = bio->bi_size >> 9; |
3592 | ||
664e7c41 AN |
3593 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3594 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3595 | return chunk_sectors >= |
3596 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3597 | } | |
3598 | ||
46031f9a RBJ |
3599 | /* |
3600 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3601 | * later sampled by raid5d. | |
3602 | */ | |
3603 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3604 | { | |
3605 | unsigned long flags; | |
3606 | ||
3607 | spin_lock_irqsave(&conf->device_lock, flags); | |
3608 | ||
3609 | bi->bi_next = conf->retry_read_aligned_list; | |
3610 | conf->retry_read_aligned_list = bi; | |
3611 | ||
3612 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3613 | md_wakeup_thread(conf->mddev->thread); | |
3614 | } | |
3615 | ||
3616 | ||
3617 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3618 | { | |
3619 | struct bio *bi; | |
3620 | ||
3621 | bi = conf->retry_read_aligned; | |
3622 | if (bi) { | |
3623 | conf->retry_read_aligned = NULL; | |
3624 | return bi; | |
3625 | } | |
3626 | bi = conf->retry_read_aligned_list; | |
3627 | if(bi) { | |
387bb173 | 3628 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3629 | bi->bi_next = NULL; |
960e739d JA |
3630 | /* |
3631 | * this sets the active strip count to 1 and the processed | |
3632 | * strip count to zero (upper 8 bits) | |
3633 | */ | |
46031f9a | 3634 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3635 | } |
3636 | ||
3637 | return bi; | |
3638 | } | |
3639 | ||
3640 | ||
f679623f RBJ |
3641 | /* |
3642 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3643 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3644 | * first). | |
3645 | * If the read failed.. | |
3646 | */ | |
6712ecf8 | 3647 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3648 | { |
3649 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3650 | mddev_t *mddev; |
3651 | raid5_conf_t *conf; | |
3652 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3653 | mdk_rdev_t *rdev; | |
3654 | ||
f679623f | 3655 | bio_put(bi); |
46031f9a RBJ |
3656 | |
3657 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
070ec55d | 3658 | conf = mddev->private; |
46031f9a RBJ |
3659 | rdev = (void*)raid_bi->bi_next; |
3660 | raid_bi->bi_next = NULL; | |
3661 | ||
3662 | rdev_dec_pending(rdev, conf->mddev); | |
3663 | ||
3664 | if (!error && uptodate) { | |
6712ecf8 | 3665 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3666 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3667 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3668 | return; |
46031f9a RBJ |
3669 | } |
3670 | ||
3671 | ||
45b4233c | 3672 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3673 | |
3674 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3675 | } |
3676 | ||
387bb173 NB |
3677 | static int bio_fits_rdev(struct bio *bi) |
3678 | { | |
165125e1 | 3679 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3680 | |
ae03bf63 | 3681 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3682 | return 0; |
3683 | blk_recount_segments(q, bi); | |
ae03bf63 | 3684 | if (bi->bi_phys_segments > queue_max_phys_segments(q)) |
387bb173 NB |
3685 | return 0; |
3686 | ||
3687 | if (q->merge_bvec_fn) | |
3688 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3689 | * just just give up | |
3690 | */ | |
3691 | return 0; | |
3692 | ||
3693 | return 1; | |
3694 | } | |
3695 | ||
3696 | ||
165125e1 | 3697 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3698 | { |
3699 | mddev_t *mddev = q->queuedata; | |
070ec55d | 3700 | raid5_conf_t *conf = mddev->private; |
911d4ee8 | 3701 | unsigned int dd_idx; |
f679623f RBJ |
3702 | struct bio* align_bi; |
3703 | mdk_rdev_t *rdev; | |
3704 | ||
3705 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3706 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3707 | return 0; |
3708 | } | |
3709 | /* | |
99c0fb5f | 3710 | * use bio_clone to make a copy of the bio |
f679623f RBJ |
3711 | */ |
3712 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3713 | if (!align_bi) | |
3714 | return 0; | |
3715 | /* | |
3716 | * set bi_end_io to a new function, and set bi_private to the | |
3717 | * original bio. | |
3718 | */ | |
3719 | align_bi->bi_end_io = raid5_align_endio; | |
3720 | align_bi->bi_private = raid_bio; | |
3721 | /* | |
3722 | * compute position | |
3723 | */ | |
112bf897 N |
3724 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3725 | 0, | |
911d4ee8 | 3726 | &dd_idx, NULL); |
f679623f RBJ |
3727 | |
3728 | rcu_read_lock(); | |
3729 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3730 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3731 | atomic_inc(&rdev->nr_pending); |
3732 | rcu_read_unlock(); | |
46031f9a RBJ |
3733 | raid_bio->bi_next = (void*)rdev; |
3734 | align_bi->bi_bdev = rdev->bdev; | |
3735 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3736 | align_bi->bi_sector += rdev->data_offset; | |
3737 | ||
387bb173 NB |
3738 | if (!bio_fits_rdev(align_bi)) { |
3739 | /* too big in some way */ | |
3740 | bio_put(align_bi); | |
3741 | rdev_dec_pending(rdev, mddev); | |
3742 | return 0; | |
3743 | } | |
3744 | ||
46031f9a RBJ |
3745 | spin_lock_irq(&conf->device_lock); |
3746 | wait_event_lock_irq(conf->wait_for_stripe, | |
3747 | conf->quiesce == 0, | |
3748 | conf->device_lock, /* nothing */); | |
3749 | atomic_inc(&conf->active_aligned_reads); | |
3750 | spin_unlock_irq(&conf->device_lock); | |
3751 | ||
f679623f RBJ |
3752 | generic_make_request(align_bi); |
3753 | return 1; | |
3754 | } else { | |
3755 | rcu_read_unlock(); | |
46031f9a | 3756 | bio_put(align_bi); |
f679623f RBJ |
3757 | return 0; |
3758 | } | |
3759 | } | |
3760 | ||
8b3e6cdc DW |
3761 | /* __get_priority_stripe - get the next stripe to process |
3762 | * | |
3763 | * Full stripe writes are allowed to pass preread active stripes up until | |
3764 | * the bypass_threshold is exceeded. In general the bypass_count | |
3765 | * increments when the handle_list is handled before the hold_list; however, it | |
3766 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3767 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3768 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3769 | * handle_list. | |
3770 | */ | |
3771 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3772 | { | |
3773 | struct stripe_head *sh; | |
3774 | ||
3775 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3776 | __func__, | |
3777 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3778 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3779 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3780 | ||
3781 | if (!list_empty(&conf->handle_list)) { | |
3782 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3783 | ||
3784 | if (list_empty(&conf->hold_list)) | |
3785 | conf->bypass_count = 0; | |
3786 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3787 | if (conf->hold_list.next == conf->last_hold) | |
3788 | conf->bypass_count++; | |
3789 | else { | |
3790 | conf->last_hold = conf->hold_list.next; | |
3791 | conf->bypass_count -= conf->bypass_threshold; | |
3792 | if (conf->bypass_count < 0) | |
3793 | conf->bypass_count = 0; | |
3794 | } | |
3795 | } | |
3796 | } else if (!list_empty(&conf->hold_list) && | |
3797 | ((conf->bypass_threshold && | |
3798 | conf->bypass_count > conf->bypass_threshold) || | |
3799 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3800 | sh = list_entry(conf->hold_list.next, | |
3801 | typeof(*sh), lru); | |
3802 | conf->bypass_count -= conf->bypass_threshold; | |
3803 | if (conf->bypass_count < 0) | |
3804 | conf->bypass_count = 0; | |
3805 | } else | |
3806 | return NULL; | |
3807 | ||
3808 | list_del_init(&sh->lru); | |
3809 | atomic_inc(&sh->count); | |
3810 | BUG_ON(atomic_read(&sh->count) != 1); | |
3811 | return sh; | |
3812 | } | |
f679623f | 3813 | |
165125e1 | 3814 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3815 | { |
3816 | mddev_t *mddev = q->queuedata; | |
070ec55d | 3817 | raid5_conf_t *conf = mddev->private; |
911d4ee8 | 3818 | int dd_idx; |
1da177e4 LT |
3819 | sector_t new_sector; |
3820 | sector_t logical_sector, last_sector; | |
3821 | struct stripe_head *sh; | |
a362357b | 3822 | const int rw = bio_data_dir(bi); |
c9959059 | 3823 | int cpu, remaining; |
1da177e4 | 3824 | |
1f98a13f | 3825 | if (unlikely(bio_rw_flagged(bi, BIO_RW_BARRIER))) { |
6712ecf8 | 3826 | bio_endio(bi, -EOPNOTSUPP); |
e5dcdd80 N |
3827 | return 0; |
3828 | } | |
3829 | ||
3d310eb7 | 3830 | md_write_start(mddev, bi); |
06d91a5f | 3831 | |
074a7aca TH |
3832 | cpu = part_stat_lock(); |
3833 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
3834 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
3835 | bio_sectors(bi)); | |
3836 | part_stat_unlock(); | |
1da177e4 | 3837 | |
802ba064 | 3838 | if (rw == READ && |
52488615 RBJ |
3839 | mddev->reshape_position == MaxSector && |
3840 | chunk_aligned_read(q,bi)) | |
99c0fb5f | 3841 | return 0; |
52488615 | 3842 | |
1da177e4 LT |
3843 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3844 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3845 | bi->bi_next = NULL; | |
3846 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3847 | |
1da177e4 LT |
3848 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3849 | DEFINE_WAIT(w); | |
16a53ecc | 3850 | int disks, data_disks; |
b5663ba4 | 3851 | int previous; |
b578d55f | 3852 | |
7ecaa1e6 | 3853 | retry: |
b5663ba4 | 3854 | previous = 0; |
b0f9ec04 | 3855 | disks = conf->raid_disks; |
b578d55f | 3856 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3857 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3858 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3859 | * 64bit on a 32bit platform, and so it might be |
3860 | * possible to see a half-updated value | |
fef9c61f | 3861 | * Ofcourse reshape_progress could change after |
df8e7f76 N |
3862 | * the lock is dropped, so once we get a reference |
3863 | * to the stripe that we think it is, we will have | |
3864 | * to check again. | |
3865 | */ | |
7ecaa1e6 | 3866 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3867 | if (mddev->delta_disks < 0 |
3868 | ? logical_sector < conf->reshape_progress | |
3869 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3870 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3871 | previous = 1; |
3872 | } else { | |
fef9c61f N |
3873 | if (mddev->delta_disks < 0 |
3874 | ? logical_sector < conf->reshape_safe | |
3875 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3876 | spin_unlock_irq(&conf->device_lock); |
3877 | schedule(); | |
3878 | goto retry; | |
3879 | } | |
3880 | } | |
7ecaa1e6 N |
3881 | spin_unlock_irq(&conf->device_lock); |
3882 | } | |
16a53ecc N |
3883 | data_disks = disks - conf->max_degraded; |
3884 | ||
112bf897 N |
3885 | new_sector = raid5_compute_sector(conf, logical_sector, |
3886 | previous, | |
911d4ee8 | 3887 | &dd_idx, NULL); |
45b4233c | 3888 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3889 | (unsigned long long)new_sector, |
3890 | (unsigned long long)logical_sector); | |
3891 | ||
b5663ba4 | 3892 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3893 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3894 | if (sh) { |
b0f9ec04 | 3895 | if (unlikely(previous)) { |
7ecaa1e6 | 3896 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3897 | * stripe, so we must do the range check again. |
3898 | * Expansion could still move past after this | |
3899 | * test, but as we are holding a reference to | |
3900 | * 'sh', we know that if that happens, | |
3901 | * STRIPE_EXPANDING will get set and the expansion | |
3902 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3903 | */ |
3904 | int must_retry = 0; | |
3905 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
3906 | if (mddev->delta_disks < 0 |
3907 | ? logical_sector >= conf->reshape_progress | |
3908 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
3909 | /* mismatch, need to try again */ |
3910 | must_retry = 1; | |
3911 | spin_unlock_irq(&conf->device_lock); | |
3912 | if (must_retry) { | |
3913 | release_stripe(sh); | |
7a3ab908 | 3914 | schedule(); |
7ecaa1e6 N |
3915 | goto retry; |
3916 | } | |
3917 | } | |
e62e58a5 | 3918 | |
a5c308d4 N |
3919 | if (bio_data_dir(bi) == WRITE && |
3920 | logical_sector >= mddev->suspend_lo && | |
e464eafd N |
3921 | logical_sector < mddev->suspend_hi) { |
3922 | release_stripe(sh); | |
e62e58a5 N |
3923 | /* As the suspend_* range is controlled by |
3924 | * userspace, we want an interruptible | |
3925 | * wait. | |
3926 | */ | |
3927 | flush_signals(current); | |
3928 | prepare_to_wait(&conf->wait_for_overlap, | |
3929 | &w, TASK_INTERRUPTIBLE); | |
3930 | if (logical_sector >= mddev->suspend_lo && | |
3931 | logical_sector < mddev->suspend_hi) | |
3932 | schedule(); | |
e464eafd N |
3933 | goto retry; |
3934 | } | |
7ecaa1e6 N |
3935 | |
3936 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
3937 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
3938 | /* Stripe is busy expanding or | |
3939 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3940 | * and wait a while |
3941 | */ | |
3942 | raid5_unplug_device(mddev->queue); | |
3943 | release_stripe(sh); | |
3944 | schedule(); | |
3945 | goto retry; | |
3946 | } | |
3947 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3948 | set_bit(STRIPE_HANDLE, &sh->state); |
3949 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1da177e4 | 3950 | release_stripe(sh); |
1da177e4 LT |
3951 | } else { |
3952 | /* cannot get stripe for read-ahead, just give-up */ | |
3953 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3954 | finish_wait(&conf->wait_for_overlap, &w); | |
3955 | break; | |
3956 | } | |
3957 | ||
3958 | } | |
3959 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3960 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3961 | spin_unlock_irq(&conf->device_lock); |
3962 | if (remaining == 0) { | |
1da177e4 | 3963 | |
16a53ecc | 3964 | if ( rw == WRITE ) |
1da177e4 | 3965 | md_write_end(mddev); |
6712ecf8 | 3966 | |
0e13fe23 | 3967 | bio_endio(bi, 0); |
1da177e4 | 3968 | } |
1da177e4 LT |
3969 | return 0; |
3970 | } | |
3971 | ||
b522adcd DW |
3972 | static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks); |
3973 | ||
52c03291 | 3974 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3975 | { |
52c03291 N |
3976 | /* reshaping is quite different to recovery/resync so it is |
3977 | * handled quite separately ... here. | |
3978 | * | |
3979 | * On each call to sync_request, we gather one chunk worth of | |
3980 | * destination stripes and flag them as expanding. | |
3981 | * Then we find all the source stripes and request reads. | |
3982 | * As the reads complete, handle_stripe will copy the data | |
3983 | * into the destination stripe and release that stripe. | |
3984 | */ | |
1da177e4 LT |
3985 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
3986 | struct stripe_head *sh; | |
ccfcc3c1 | 3987 | sector_t first_sector, last_sector; |
f416885e N |
3988 | int raid_disks = conf->previous_raid_disks; |
3989 | int data_disks = raid_disks - conf->max_degraded; | |
3990 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3991 | int i; |
3992 | int dd_idx; | |
c8f517c4 | 3993 | sector_t writepos, readpos, safepos; |
ec32a2bd | 3994 | sector_t stripe_addr; |
7a661381 | 3995 | int reshape_sectors; |
ab69ae12 | 3996 | struct list_head stripes; |
52c03291 | 3997 | |
fef9c61f N |
3998 | if (sector_nr == 0) { |
3999 | /* If restarting in the middle, skip the initial sectors */ | |
4000 | if (mddev->delta_disks < 0 && | |
4001 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4002 | sector_nr = raid5_size(mddev, 0, 0) | |
4003 | - conf->reshape_progress; | |
a639755c | 4004 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
4005 | conf->reshape_progress > 0) |
4006 | sector_nr = conf->reshape_progress; | |
f416885e | 4007 | sector_div(sector_nr, new_data_disks); |
fef9c61f N |
4008 | if (sector_nr) { |
4009 | *skipped = 1; | |
4010 | return sector_nr; | |
4011 | } | |
52c03291 N |
4012 | } |
4013 | ||
7a661381 N |
4014 | /* We need to process a full chunk at a time. |
4015 | * If old and new chunk sizes differ, we need to process the | |
4016 | * largest of these | |
4017 | */ | |
664e7c41 AN |
4018 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4019 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4020 | else |
9d8f0363 | 4021 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4022 | |
52c03291 N |
4023 | /* we update the metadata when there is more than 3Meg |
4024 | * in the block range (that is rather arbitrary, should | |
4025 | * probably be time based) or when the data about to be | |
4026 | * copied would over-write the source of the data at | |
4027 | * the front of the range. | |
fef9c61f N |
4028 | * i.e. one new_stripe along from reshape_progress new_maps |
4029 | * to after where reshape_safe old_maps to | |
52c03291 | 4030 | */ |
fef9c61f | 4031 | writepos = conf->reshape_progress; |
f416885e | 4032 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4033 | readpos = conf->reshape_progress; |
4034 | sector_div(readpos, data_disks); | |
fef9c61f | 4035 | safepos = conf->reshape_safe; |
f416885e | 4036 | sector_div(safepos, data_disks); |
fef9c61f | 4037 | if (mddev->delta_disks < 0) { |
ed37d83e | 4038 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4039 | readpos += reshape_sectors; |
7a661381 | 4040 | safepos += reshape_sectors; |
fef9c61f | 4041 | } else { |
7a661381 | 4042 | writepos += reshape_sectors; |
ed37d83e N |
4043 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4044 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4045 | } |
52c03291 | 4046 | |
c8f517c4 N |
4047 | /* 'writepos' is the most advanced device address we might write. |
4048 | * 'readpos' is the least advanced device address we might read. | |
4049 | * 'safepos' is the least address recorded in the metadata as having | |
4050 | * been reshaped. | |
4051 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4052 | * ensure safety in the face of a crash - that must be done by userspace | |
4053 | * making a backup of the data. So in that case there is no particular | |
4054 | * rush to update metadata. | |
4055 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4056 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4057 | * we can be safe in the event of a crash. | |
4058 | * So we insist on updating metadata if safepos is behind writepos and | |
4059 | * readpos is beyond writepos. | |
4060 | * In any case, update the metadata every 10 seconds. | |
4061 | * Maybe that number should be configurable, but I'm not sure it is | |
4062 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4063 | */ | |
fef9c61f | 4064 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4065 | ? (safepos > writepos && readpos < writepos) |
4066 | : (safepos < writepos && readpos > writepos)) || | |
4067 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4068 | /* Cannot proceed until we've updated the superblock... */ |
4069 | wait_event(conf->wait_for_overlap, | |
4070 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4071 | mddev->reshape_position = conf->reshape_progress; |
acb180b0 | 4072 | mddev->curr_resync_completed = mddev->curr_resync; |
c8f517c4 | 4073 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4074 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4075 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4076 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4077 | kthread_should_stop()); |
4078 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4079 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4080 | spin_unlock_irq(&conf->device_lock); |
4081 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4082 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4083 | } |
4084 | ||
ec32a2bd N |
4085 | if (mddev->delta_disks < 0) { |
4086 | BUG_ON(conf->reshape_progress == 0); | |
4087 | stripe_addr = writepos; | |
4088 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4089 | ~((sector_t)reshape_sectors - 1)) |
4090 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4091 | != sector_nr); |
4092 | } else { | |
7a661381 | 4093 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4094 | stripe_addr = sector_nr; |
4095 | } | |
ab69ae12 | 4096 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4097 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4098 | int j; |
a9f326eb | 4099 | int skipped_disk = 0; |
a8c906ca | 4100 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4101 | set_bit(STRIPE_EXPANDING, &sh->state); |
4102 | atomic_inc(&conf->reshape_stripes); | |
4103 | /* If any of this stripe is beyond the end of the old | |
4104 | * array, then we need to zero those blocks | |
4105 | */ | |
4106 | for (j=sh->disks; j--;) { | |
4107 | sector_t s; | |
4108 | if (j == sh->pd_idx) | |
4109 | continue; | |
f416885e | 4110 | if (conf->level == 6 && |
d0dabf7e | 4111 | j == sh->qd_idx) |
f416885e | 4112 | continue; |
784052ec | 4113 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4114 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4115 | skipped_disk = 1; |
52c03291 N |
4116 | continue; |
4117 | } | |
4118 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4119 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4120 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4121 | } | |
a9f326eb | 4122 | if (!skipped_disk) { |
52c03291 N |
4123 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4124 | set_bit(STRIPE_HANDLE, &sh->state); | |
4125 | } | |
ab69ae12 | 4126 | list_add(&sh->lru, &stripes); |
52c03291 N |
4127 | } |
4128 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4129 | if (mddev->delta_disks < 0) |
7a661381 | 4130 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4131 | else |
7a661381 | 4132 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4133 | spin_unlock_irq(&conf->device_lock); |
4134 | /* Ok, those stripe are ready. We can start scheduling | |
4135 | * reads on the source stripes. | |
4136 | * The source stripes are determined by mapping the first and last | |
4137 | * block on the destination stripes. | |
4138 | */ | |
52c03291 | 4139 | first_sector = |
ec32a2bd | 4140 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4141 | 1, &dd_idx, NULL); |
52c03291 | 4142 | last_sector = |
0e6e0271 | 4143 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4144 | * new_data_disks - 1), |
911d4ee8 | 4145 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4146 | if (last_sector >= mddev->dev_sectors) |
4147 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4148 | while (first_sector <= last_sector) { |
a8c906ca | 4149 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4150 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4151 | set_bit(STRIPE_HANDLE, &sh->state); | |
4152 | release_stripe(sh); | |
4153 | first_sector += STRIPE_SECTORS; | |
4154 | } | |
ab69ae12 N |
4155 | /* Now that the sources are clearly marked, we can release |
4156 | * the destination stripes | |
4157 | */ | |
4158 | while (!list_empty(&stripes)) { | |
4159 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4160 | list_del_init(&sh->lru); | |
4161 | release_stripe(sh); | |
4162 | } | |
c6207277 N |
4163 | /* If this takes us to the resync_max point where we have to pause, |
4164 | * then we need to write out the superblock. | |
4165 | */ | |
7a661381 | 4166 | sector_nr += reshape_sectors; |
c03f6a19 N |
4167 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4168 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4169 | /* Cannot proceed until we've updated the superblock... */ |
4170 | wait_event(conf->wait_for_overlap, | |
4171 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4172 | mddev->reshape_position = conf->reshape_progress; |
48606a9f | 4173 | mddev->curr_resync_completed = mddev->curr_resync + reshape_sectors; |
c8f517c4 | 4174 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4175 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4176 | md_wakeup_thread(mddev->thread); | |
4177 | wait_event(mddev->sb_wait, | |
4178 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4179 | || kthread_should_stop()); | |
4180 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4181 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4182 | spin_unlock_irq(&conf->device_lock); |
4183 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4184 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4185 | } |
7a661381 | 4186 | return reshape_sectors; |
52c03291 N |
4187 | } |
4188 | ||
4189 | /* FIXME go_faster isn't used */ | |
4190 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
4191 | { | |
4192 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4193 | struct stripe_head *sh; | |
58c0fed4 | 4194 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 4195 | int sync_blocks; |
16a53ecc N |
4196 | int still_degraded = 0; |
4197 | int i; | |
1da177e4 | 4198 | |
72626685 | 4199 | if (sector_nr >= max_sector) { |
1da177e4 LT |
4200 | /* just being told to finish up .. nothing much to do */ |
4201 | unplug_slaves(mddev); | |
cea9c228 | 4202 | |
29269553 N |
4203 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4204 | end_reshape(conf); | |
4205 | return 0; | |
4206 | } | |
72626685 N |
4207 | |
4208 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4209 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4210 | &sync_blocks, 1); | |
16a53ecc | 4211 | else /* completed sync */ |
72626685 N |
4212 | conf->fullsync = 0; |
4213 | bitmap_close_sync(mddev->bitmap); | |
4214 | ||
1da177e4 LT |
4215 | return 0; |
4216 | } | |
ccfcc3c1 | 4217 | |
64bd660b N |
4218 | /* Allow raid5_quiesce to complete */ |
4219 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4220 | ||
52c03291 N |
4221 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4222 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4223 | |
c6207277 N |
4224 | /* No need to check resync_max as we never do more than one |
4225 | * stripe, and as resync_max will always be on a chunk boundary, | |
4226 | * if the check in md_do_sync didn't fire, there is no chance | |
4227 | * of overstepping resync_max here | |
4228 | */ | |
4229 | ||
16a53ecc | 4230 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4231 | * to resync, then assert that we are finished, because there is |
4232 | * nothing we can do. | |
4233 | */ | |
3285edf1 | 4234 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4235 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4236 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4237 | *skipped = 1; |
1da177e4 LT |
4238 | return rv; |
4239 | } | |
72626685 | 4240 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4241 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4242 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4243 | /* we can skip this block, and probably more */ | |
4244 | sync_blocks /= STRIPE_SECTORS; | |
4245 | *skipped = 1; | |
4246 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4247 | } | |
1da177e4 | 4248 | |
b47490c9 N |
4249 | |
4250 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
4251 | ||
a8c906ca | 4252 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4253 | if (sh == NULL) { |
a8c906ca | 4254 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4255 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4256 | * is trying to get access |
1da177e4 | 4257 | */ |
66c006a5 | 4258 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4259 | } |
16a53ecc N |
4260 | /* Need to check if array will still be degraded after recovery/resync |
4261 | * We don't need to check the 'failed' flag as when that gets set, | |
4262 | * recovery aborts. | |
4263 | */ | |
f001a70c | 4264 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4265 | if (conf->disks[i].rdev == NULL) |
4266 | still_degraded = 1; | |
4267 | ||
4268 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4269 | ||
4270 | spin_lock(&sh->lock); | |
1da177e4 LT |
4271 | set_bit(STRIPE_SYNCING, &sh->state); |
4272 | clear_bit(STRIPE_INSYNC, &sh->state); | |
4273 | spin_unlock(&sh->lock); | |
4274 | ||
1442577b | 4275 | handle_stripe(sh); |
1da177e4 LT |
4276 | release_stripe(sh); |
4277 | ||
4278 | return STRIPE_SECTORS; | |
4279 | } | |
4280 | ||
46031f9a RBJ |
4281 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
4282 | { | |
4283 | /* We may not be able to submit a whole bio at once as there | |
4284 | * may not be enough stripe_heads available. | |
4285 | * We cannot pre-allocate enough stripe_heads as we may need | |
4286 | * more than exist in the cache (if we allow ever large chunks). | |
4287 | * So we do one stripe head at a time and record in | |
4288 | * ->bi_hw_segments how many have been done. | |
4289 | * | |
4290 | * We *know* that this entire raid_bio is in one chunk, so | |
4291 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4292 | */ | |
4293 | struct stripe_head *sh; | |
911d4ee8 | 4294 | int dd_idx; |
46031f9a RBJ |
4295 | sector_t sector, logical_sector, last_sector; |
4296 | int scnt = 0; | |
4297 | int remaining; | |
4298 | int handled = 0; | |
4299 | ||
4300 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4301 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4302 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4303 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4304 | ||
4305 | for (; logical_sector < last_sector; | |
387bb173 NB |
4306 | logical_sector += STRIPE_SECTORS, |
4307 | sector += STRIPE_SECTORS, | |
4308 | scnt++) { | |
46031f9a | 4309 | |
960e739d | 4310 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4311 | /* already done this stripe */ |
4312 | continue; | |
4313 | ||
a8c906ca | 4314 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4315 | |
4316 | if (!sh) { | |
4317 | /* failed to get a stripe - must wait */ | |
960e739d | 4318 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4319 | conf->retry_read_aligned = raid_bio; |
4320 | return handled; | |
4321 | } | |
4322 | ||
4323 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4324 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4325 | release_stripe(sh); | |
960e739d | 4326 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4327 | conf->retry_read_aligned = raid_bio; |
4328 | return handled; | |
4329 | } | |
4330 | ||
36d1c647 | 4331 | handle_stripe(sh); |
46031f9a RBJ |
4332 | release_stripe(sh); |
4333 | handled++; | |
4334 | } | |
4335 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4336 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4337 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4338 | if (remaining == 0) |
4339 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4340 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4341 | wake_up(&conf->wait_for_stripe); | |
4342 | return handled; | |
4343 | } | |
4344 | ||
07a3b417 DW |
4345 | #ifdef CONFIG_MULTICORE_RAID456 |
4346 | static void __process_stripe(void *param, async_cookie_t cookie) | |
4347 | { | |
4348 | struct stripe_head *sh = param; | |
4349 | ||
4350 | handle_stripe(sh); | |
4351 | release_stripe(sh); | |
4352 | } | |
4353 | ||
4354 | static void process_stripe(struct stripe_head *sh, struct list_head *domain) | |
4355 | { | |
4356 | async_schedule_domain(__process_stripe, sh, domain); | |
4357 | } | |
4358 | ||
4359 | static void synchronize_stripe_processing(struct list_head *domain) | |
4360 | { | |
4361 | async_synchronize_full_domain(domain); | |
4362 | } | |
4363 | #else | |
4364 | static void process_stripe(struct stripe_head *sh, struct list_head *domain) | |
4365 | { | |
4366 | handle_stripe(sh); | |
4367 | release_stripe(sh); | |
4368 | cond_resched(); | |
4369 | } | |
4370 | ||
4371 | static void synchronize_stripe_processing(struct list_head *domain) | |
4372 | { | |
4373 | } | |
4374 | #endif | |
46031f9a RBJ |
4375 | |
4376 | ||
1da177e4 LT |
4377 | /* |
4378 | * This is our raid5 kernel thread. | |
4379 | * | |
4380 | * We scan the hash table for stripes which can be handled now. | |
4381 | * During the scan, completed stripes are saved for us by the interrupt | |
4382 | * handler, so that they will not have to wait for our next wakeup. | |
4383 | */ | |
6ed3003c | 4384 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4385 | { |
4386 | struct stripe_head *sh; | |
070ec55d | 4387 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 4388 | int handled; |
07a3b417 | 4389 | LIST_HEAD(raid_domain); |
1da177e4 | 4390 | |
45b4233c | 4391 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4392 | |
4393 | md_check_recovery(mddev); | |
1da177e4 LT |
4394 | |
4395 | handled = 0; | |
4396 | spin_lock_irq(&conf->device_lock); | |
4397 | while (1) { | |
46031f9a | 4398 | struct bio *bio; |
1da177e4 | 4399 | |
ae3c20cc | 4400 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 4401 | int seq = conf->seq_flush; |
700e432d | 4402 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4403 | bitmap_unplug(mddev->bitmap); |
700e432d | 4404 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4405 | conf->seq_write = seq; |
4406 | activate_bit_delay(conf); | |
4407 | } | |
4408 | ||
46031f9a RBJ |
4409 | while ((bio = remove_bio_from_retry(conf))) { |
4410 | int ok; | |
4411 | spin_unlock_irq(&conf->device_lock); | |
4412 | ok = retry_aligned_read(conf, bio); | |
4413 | spin_lock_irq(&conf->device_lock); | |
4414 | if (!ok) | |
4415 | break; | |
4416 | handled++; | |
4417 | } | |
4418 | ||
8b3e6cdc DW |
4419 | sh = __get_priority_stripe(conf); |
4420 | ||
c9f21aaf | 4421 | if (!sh) |
1da177e4 | 4422 | break; |
1da177e4 LT |
4423 | spin_unlock_irq(&conf->device_lock); |
4424 | ||
4425 | handled++; | |
07a3b417 | 4426 | process_stripe(sh, &raid_domain); |
1da177e4 LT |
4427 | |
4428 | spin_lock_irq(&conf->device_lock); | |
4429 | } | |
45b4233c | 4430 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4431 | |
4432 | spin_unlock_irq(&conf->device_lock); | |
4433 | ||
07a3b417 | 4434 | synchronize_stripe_processing(&raid_domain); |
c9f21aaf | 4435 | async_tx_issue_pending_all(); |
1da177e4 LT |
4436 | unplug_slaves(mddev); |
4437 | ||
45b4233c | 4438 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4439 | } |
4440 | ||
3f294f4f | 4441 | static ssize_t |
007583c9 | 4442 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4443 | { |
070ec55d | 4444 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4445 | if (conf) |
4446 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4447 | else | |
4448 | return 0; | |
3f294f4f N |
4449 | } |
4450 | ||
4451 | static ssize_t | |
007583c9 | 4452 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4453 | { |
070ec55d | 4454 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4455 | unsigned long new; |
b5470dc5 DW |
4456 | int err; |
4457 | ||
3f294f4f N |
4458 | if (len >= PAGE_SIZE) |
4459 | return -EINVAL; | |
96de1e66 N |
4460 | if (!conf) |
4461 | return -ENODEV; | |
3f294f4f | 4462 | |
4ef197d8 | 4463 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4464 | return -EINVAL; |
4465 | if (new <= 16 || new > 32768) | |
4466 | return -EINVAL; | |
4467 | while (new < conf->max_nr_stripes) { | |
4468 | if (drop_one_stripe(conf)) | |
4469 | conf->max_nr_stripes--; | |
4470 | else | |
4471 | break; | |
4472 | } | |
b5470dc5 DW |
4473 | err = md_allow_write(mddev); |
4474 | if (err) | |
4475 | return err; | |
3f294f4f N |
4476 | while (new > conf->max_nr_stripes) { |
4477 | if (grow_one_stripe(conf)) | |
4478 | conf->max_nr_stripes++; | |
4479 | else break; | |
4480 | } | |
4481 | return len; | |
4482 | } | |
007583c9 | 4483 | |
96de1e66 N |
4484 | static struct md_sysfs_entry |
4485 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4486 | raid5_show_stripe_cache_size, | |
4487 | raid5_store_stripe_cache_size); | |
3f294f4f | 4488 | |
8b3e6cdc DW |
4489 | static ssize_t |
4490 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4491 | { | |
070ec55d | 4492 | raid5_conf_t *conf = mddev->private; |
8b3e6cdc DW |
4493 | if (conf) |
4494 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4495 | else | |
4496 | return 0; | |
4497 | } | |
4498 | ||
4499 | static ssize_t | |
4500 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4501 | { | |
070ec55d | 4502 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4503 | unsigned long new; |
8b3e6cdc DW |
4504 | if (len >= PAGE_SIZE) |
4505 | return -EINVAL; | |
4506 | if (!conf) | |
4507 | return -ENODEV; | |
4508 | ||
4ef197d8 | 4509 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4510 | return -EINVAL; |
4ef197d8 | 4511 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4512 | return -EINVAL; |
4513 | conf->bypass_threshold = new; | |
4514 | return len; | |
4515 | } | |
4516 | ||
4517 | static struct md_sysfs_entry | |
4518 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4519 | S_IRUGO | S_IWUSR, | |
4520 | raid5_show_preread_threshold, | |
4521 | raid5_store_preread_threshold); | |
4522 | ||
3f294f4f | 4523 | static ssize_t |
96de1e66 | 4524 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4525 | { |
070ec55d | 4526 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4527 | if (conf) |
4528 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4529 | else | |
4530 | return 0; | |
3f294f4f N |
4531 | } |
4532 | ||
96de1e66 N |
4533 | static struct md_sysfs_entry |
4534 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4535 | |
007583c9 | 4536 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4537 | &raid5_stripecache_size.attr, |
4538 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4539 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4540 | NULL, |
4541 | }; | |
007583c9 N |
4542 | static struct attribute_group raid5_attrs_group = { |
4543 | .name = NULL, | |
4544 | .attrs = raid5_attrs, | |
3f294f4f N |
4545 | }; |
4546 | ||
80c3a6ce DW |
4547 | static sector_t |
4548 | raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
4549 | { | |
070ec55d | 4550 | raid5_conf_t *conf = mddev->private; |
80c3a6ce DW |
4551 | |
4552 | if (!sectors) | |
4553 | sectors = mddev->dev_sectors; | |
7ec05478 N |
4554 | if (!raid_disks) { |
4555 | /* size is defined by the smallest of previous and new size */ | |
4556 | if (conf->raid_disks < conf->previous_raid_disks) | |
4557 | raid_disks = conf->raid_disks; | |
4558 | else | |
4559 | raid_disks = conf->previous_raid_disks; | |
4560 | } | |
80c3a6ce | 4561 | |
9d8f0363 | 4562 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4563 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4564 | return sectors * (raid_disks - conf->max_degraded); |
4565 | } | |
4566 | ||
36d1c647 DW |
4567 | static void raid5_free_percpu(raid5_conf_t *conf) |
4568 | { | |
4569 | struct raid5_percpu *percpu; | |
4570 | unsigned long cpu; | |
4571 | ||
4572 | if (!conf->percpu) | |
4573 | return; | |
4574 | ||
4575 | get_online_cpus(); | |
4576 | for_each_possible_cpu(cpu) { | |
4577 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4578 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4579 | kfree(percpu->scribble); |
36d1c647 DW |
4580 | } |
4581 | #ifdef CONFIG_HOTPLUG_CPU | |
4582 | unregister_cpu_notifier(&conf->cpu_notify); | |
4583 | #endif | |
4584 | put_online_cpus(); | |
4585 | ||
4586 | free_percpu(conf->percpu); | |
4587 | } | |
4588 | ||
95fc17aa DW |
4589 | static void free_conf(raid5_conf_t *conf) |
4590 | { | |
4591 | shrink_stripes(conf); | |
36d1c647 | 4592 | raid5_free_percpu(conf); |
95fc17aa DW |
4593 | kfree(conf->disks); |
4594 | kfree(conf->stripe_hashtbl); | |
4595 | kfree(conf); | |
4596 | } | |
4597 | ||
36d1c647 DW |
4598 | #ifdef CONFIG_HOTPLUG_CPU |
4599 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4600 | void *hcpu) | |
4601 | { | |
4602 | raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify); | |
4603 | long cpu = (long)hcpu; | |
4604 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4605 | ||
4606 | switch (action) { | |
4607 | case CPU_UP_PREPARE: | |
4608 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4609 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4610 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4611 | if (!percpu->scribble) |
4612 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4613 | ||
4614 | if (!percpu->scribble || | |
4615 | (conf->level == 6 && !percpu->spare_page)) { | |
4616 | safe_put_page(percpu->spare_page); | |
4617 | kfree(percpu->scribble); | |
36d1c647 DW |
4618 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4619 | __func__, cpu); | |
4620 | return NOTIFY_BAD; | |
4621 | } | |
4622 | break; | |
4623 | case CPU_DEAD: | |
4624 | case CPU_DEAD_FROZEN: | |
4625 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4626 | kfree(percpu->scribble); |
36d1c647 | 4627 | percpu->spare_page = NULL; |
d6f38f31 | 4628 | percpu->scribble = NULL; |
36d1c647 DW |
4629 | break; |
4630 | default: | |
4631 | break; | |
4632 | } | |
4633 | return NOTIFY_OK; | |
4634 | } | |
4635 | #endif | |
4636 | ||
4637 | static int raid5_alloc_percpu(raid5_conf_t *conf) | |
4638 | { | |
4639 | unsigned long cpu; | |
4640 | struct page *spare_page; | |
4641 | struct raid5_percpu *allcpus; | |
d6f38f31 | 4642 | void *scribble; |
36d1c647 DW |
4643 | int err; |
4644 | ||
36d1c647 DW |
4645 | allcpus = alloc_percpu(struct raid5_percpu); |
4646 | if (!allcpus) | |
4647 | return -ENOMEM; | |
4648 | conf->percpu = allcpus; | |
4649 | ||
4650 | get_online_cpus(); | |
4651 | err = 0; | |
4652 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4653 | if (conf->level == 6) { |
4654 | spare_page = alloc_page(GFP_KERNEL); | |
4655 | if (!spare_page) { | |
4656 | err = -ENOMEM; | |
4657 | break; | |
4658 | } | |
4659 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4660 | } | |
4661 | scribble = kmalloc(scribble_len(conf->raid_disks), GFP_KERNEL); | |
4662 | if (!scribble) { | |
36d1c647 DW |
4663 | err = -ENOMEM; |
4664 | break; | |
4665 | } | |
d6f38f31 | 4666 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4667 | } |
4668 | #ifdef CONFIG_HOTPLUG_CPU | |
4669 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4670 | conf->cpu_notify.priority = 0; | |
4671 | if (err == 0) | |
4672 | err = register_cpu_notifier(&conf->cpu_notify); | |
4673 | #endif | |
4674 | put_online_cpus(); | |
4675 | ||
4676 | return err; | |
4677 | } | |
4678 | ||
91adb564 | 4679 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4680 | { |
4681 | raid5_conf_t *conf; | |
4682 | int raid_disk, memory; | |
4683 | mdk_rdev_t *rdev; | |
4684 | struct disk_info *disk; | |
1da177e4 | 4685 | |
91adb564 N |
4686 | if (mddev->new_level != 5 |
4687 | && mddev->new_level != 4 | |
4688 | && mddev->new_level != 6) { | |
16a53ecc | 4689 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4690 | mdname(mddev), mddev->new_level); |
4691 | return ERR_PTR(-EIO); | |
1da177e4 | 4692 | } |
91adb564 N |
4693 | if ((mddev->new_level == 5 |
4694 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4695 | (mddev->new_level == 6 | |
4696 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
99c0fb5f | 4697 | printk(KERN_ERR "raid5: %s: layout %d not supported\n", |
91adb564 N |
4698 | mdname(mddev), mddev->new_layout); |
4699 | return ERR_PTR(-EIO); | |
99c0fb5f | 4700 | } |
91adb564 N |
4701 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
4702 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4703 | mdname(mddev), mddev->raid_disks); | |
4704 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4705 | } |
4706 | ||
664e7c41 AN |
4707 | if (!mddev->new_chunk_sectors || |
4708 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4709 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
91adb564 | 4710 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", |
664e7c41 | 4711 | mddev->new_chunk_sectors << 9, mdname(mddev)); |
91adb564 | 4712 | return ERR_PTR(-EINVAL); |
f6705578 N |
4713 | } |
4714 | ||
91adb564 N |
4715 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4716 | if (conf == NULL) | |
1da177e4 | 4717 | goto abort; |
f5efd45a DW |
4718 | spin_lock_init(&conf->device_lock); |
4719 | init_waitqueue_head(&conf->wait_for_stripe); | |
4720 | init_waitqueue_head(&conf->wait_for_overlap); | |
4721 | INIT_LIST_HEAD(&conf->handle_list); | |
4722 | INIT_LIST_HEAD(&conf->hold_list); | |
4723 | INIT_LIST_HEAD(&conf->delayed_list); | |
4724 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4725 | INIT_LIST_HEAD(&conf->inactive_list); | |
4726 | atomic_set(&conf->active_stripes, 0); | |
4727 | atomic_set(&conf->preread_active_stripes, 0); | |
4728 | atomic_set(&conf->active_aligned_reads, 0); | |
4729 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
91adb564 N |
4730 | |
4731 | conf->raid_disks = mddev->raid_disks; | |
d6f38f31 | 4732 | conf->scribble_len = scribble_len(conf->raid_disks); |
91adb564 N |
4733 | if (mddev->reshape_position == MaxSector) |
4734 | conf->previous_raid_disks = mddev->raid_disks; | |
4735 | else | |
f6705578 | 4736 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
f6705578 N |
4737 | |
4738 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
4739 | GFP_KERNEL); |
4740 | if (!conf->disks) | |
4741 | goto abort; | |
9ffae0cf | 4742 | |
1da177e4 LT |
4743 | conf->mddev = mddev; |
4744 | ||
fccddba0 | 4745 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4746 | goto abort; |
1da177e4 | 4747 | |
36d1c647 DW |
4748 | conf->level = mddev->new_level; |
4749 | if (raid5_alloc_percpu(conf) != 0) | |
4750 | goto abort; | |
4751 | ||
45b4233c | 4752 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4753 | |
159ec1fc | 4754 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4755 | raid_disk = rdev->raid_disk; |
f6705578 | 4756 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
4757 | || raid_disk < 0) |
4758 | continue; | |
4759 | disk = conf->disks + raid_disk; | |
4760 | ||
4761 | disk->rdev = rdev; | |
4762 | ||
b2d444d7 | 4763 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4764 | char b[BDEVNAME_SIZE]; |
4765 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4766 | " disk %d\n", bdevname(rdev->bdev,b), | |
4767 | raid_disk); | |
8c2e870a NB |
4768 | } else |
4769 | /* Cannot rely on bitmap to complete recovery */ | |
4770 | conf->fullsync = 1; | |
1da177e4 LT |
4771 | } |
4772 | ||
09c9e5fa | 4773 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4774 | conf->level = mddev->new_level; |
16a53ecc N |
4775 | if (conf->level == 6) |
4776 | conf->max_degraded = 2; | |
4777 | else | |
4778 | conf->max_degraded = 1; | |
91adb564 | 4779 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4780 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4781 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4782 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4783 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4784 | conf->prev_algo = mddev->layout; |
4785 | } | |
1da177e4 | 4786 | |
91adb564 N |
4787 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
4788 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; | |
4789 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
4790 | printk(KERN_ERR | |
4791 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4792 | goto abort; | |
4793 | } else | |
4794 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4795 | memory, mdname(mddev)); | |
1da177e4 | 4796 | |
0da3c619 | 4797 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4798 | if (!conf->thread) { |
4799 | printk(KERN_ERR | |
4800 | "raid5: couldn't allocate thread for %s\n", | |
4801 | mdname(mddev)); | |
16a53ecc N |
4802 | goto abort; |
4803 | } | |
91adb564 N |
4804 | |
4805 | return conf; | |
4806 | ||
4807 | abort: | |
4808 | if (conf) { | |
95fc17aa | 4809 | free_conf(conf); |
91adb564 N |
4810 | return ERR_PTR(-EIO); |
4811 | } else | |
4812 | return ERR_PTR(-ENOMEM); | |
4813 | } | |
4814 | ||
4815 | static int run(mddev_t *mddev) | |
4816 | { | |
4817 | raid5_conf_t *conf; | |
8f6c2e4b | 4818 | int working_disks = 0, chunk_size; |
91adb564 N |
4819 | mdk_rdev_t *rdev; |
4820 | ||
8c6ac868 AN |
4821 | if (mddev->recovery_cp != MaxSector) |
4822 | printk(KERN_NOTICE "raid5: %s is not clean" | |
4823 | " -- starting background reconstruction\n", | |
4824 | mdname(mddev)); | |
91adb564 N |
4825 | if (mddev->reshape_position != MaxSector) { |
4826 | /* Check that we can continue the reshape. | |
4827 | * Currently only disks can change, it must | |
4828 | * increase, and we must be past the point where | |
4829 | * a stripe over-writes itself | |
4830 | */ | |
4831 | sector_t here_new, here_old; | |
4832 | int old_disks; | |
18b00334 | 4833 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4834 | |
88ce4930 | 4835 | if (mddev->new_level != mddev->level) { |
91adb564 N |
4836 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4837 | "required - aborting.\n", | |
4838 | mdname(mddev)); | |
4839 | return -EINVAL; | |
4840 | } | |
91adb564 N |
4841 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4842 | /* reshape_position must be on a new-stripe boundary, and one | |
4843 | * further up in new geometry must map after here in old | |
4844 | * geometry. | |
4845 | */ | |
4846 | here_new = mddev->reshape_position; | |
664e7c41 | 4847 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 N |
4848 | (mddev->raid_disks - max_degraded))) { |
4849 | printk(KERN_ERR "raid5: reshape_position not " | |
4850 | "on a stripe boundary\n"); | |
4851 | return -EINVAL; | |
4852 | } | |
4853 | /* here_new is the stripe we will write to */ | |
4854 | here_old = mddev->reshape_position; | |
9d8f0363 | 4855 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4856 | (old_disks-max_degraded)); |
4857 | /* here_old is the first stripe that we might need to read | |
4858 | * from */ | |
67ac6011 N |
4859 | if (mddev->delta_disks == 0) { |
4860 | /* We cannot be sure it is safe to start an in-place | |
4861 | * reshape. It is only safe if user-space if monitoring | |
4862 | * and taking constant backups. | |
4863 | * mdadm always starts a situation like this in | |
4864 | * readonly mode so it can take control before | |
4865 | * allowing any writes. So just check for that. | |
4866 | */ | |
4867 | if ((here_new * mddev->new_chunk_sectors != | |
4868 | here_old * mddev->chunk_sectors) || | |
4869 | mddev->ro == 0) { | |
4870 | printk(KERN_ERR "raid5: in-place reshape must be started" | |
4871 | " in read-only mode - aborting\n"); | |
4872 | return -EINVAL; | |
4873 | } | |
4874 | } else if (mddev->delta_disks < 0 | |
4875 | ? (here_new * mddev->new_chunk_sectors <= | |
4876 | here_old * mddev->chunk_sectors) | |
4877 | : (here_new * mddev->new_chunk_sectors >= | |
4878 | here_old * mddev->chunk_sectors)) { | |
91adb564 N |
4879 | /* Reading from the same stripe as writing to - bad */ |
4880 | printk(KERN_ERR "raid5: reshape_position too early for " | |
4881 | "auto-recovery - aborting.\n"); | |
4882 | return -EINVAL; | |
4883 | } | |
4884 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4885 | /* OK, we should be able to continue; */ | |
4886 | } else { | |
4887 | BUG_ON(mddev->level != mddev->new_level); | |
4888 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 4889 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 4890 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 4891 | } |
91adb564 | 4892 | |
245f46c2 N |
4893 | if (mddev->private == NULL) |
4894 | conf = setup_conf(mddev); | |
4895 | else | |
4896 | conf = mddev->private; | |
4897 | ||
91adb564 N |
4898 | if (IS_ERR(conf)) |
4899 | return PTR_ERR(conf); | |
4900 | ||
4901 | mddev->thread = conf->thread; | |
4902 | conf->thread = NULL; | |
4903 | mddev->private = conf; | |
4904 | ||
4905 | /* | |
4906 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4907 | */ | |
4908 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4909 | if (rdev->raid_disk >= 0 && | |
4910 | test_bit(In_sync, &rdev->flags)) | |
4911 | working_disks++; | |
4912 | ||
4913 | mddev->degraded = conf->raid_disks - working_disks; | |
4914 | ||
16a53ecc | 4915 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
4916 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
4917 | " (%d/%d failed)\n", | |
02c2de8c | 4918 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4919 | goto abort; |
4920 | } | |
4921 | ||
91adb564 | 4922 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 4923 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
4924 | mddev->resync_max_sectors = mddev->dev_sectors; |
4925 | ||
16a53ecc | 4926 | if (mddev->degraded > 0 && |
1da177e4 | 4927 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4928 | if (mddev->ok_start_degraded) |
4929 | printk(KERN_WARNING | |
4930 | "raid5: starting dirty degraded array: %s" | |
4931 | "- data corruption possible.\n", | |
4932 | mdname(mddev)); | |
4933 | else { | |
4934 | printk(KERN_ERR | |
4935 | "raid5: cannot start dirty degraded array for %s\n", | |
4936 | mdname(mddev)); | |
4937 | goto abort; | |
4938 | } | |
1da177e4 LT |
4939 | } |
4940 | ||
1da177e4 LT |
4941 | if (mddev->degraded == 0) |
4942 | printk("raid5: raid level %d set %s active with %d out of %d" | |
e183eaed N |
4943 | " devices, algorithm %d\n", conf->level, mdname(mddev), |
4944 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
4945 | mddev->new_layout); | |
1da177e4 LT |
4946 | else |
4947 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
4948 | " out of %d devices, algorithm %d\n", conf->level, | |
4949 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
e183eaed | 4950 | mddev->raid_disks, mddev->new_layout); |
1da177e4 LT |
4951 | |
4952 | print_raid5_conf(conf); | |
4953 | ||
fef9c61f | 4954 | if (conf->reshape_progress != MaxSector) { |
f6705578 | 4955 | printk("...ok start reshape thread\n"); |
fef9c61f | 4956 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
4957 | atomic_set(&conf->reshape_stripes, 0); |
4958 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4959 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4960 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4961 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4962 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 4963 | "reshape"); |
f6705578 N |
4964 | } |
4965 | ||
1da177e4 | 4966 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 4967 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
4968 | */ |
4969 | { | |
16a53ecc N |
4970 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
4971 | int stripe = data_disks * | |
9d8f0363 | 4972 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
1da177e4 LT |
4973 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
4974 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4975 | } | |
4976 | ||
4977 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
4978 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
4979 | printk(KERN_WARNING | |
4980 | "raid5: failed to create sysfs attributes for %s\n", | |
4981 | mdname(mddev)); | |
7a5febe9 | 4982 | |
91adb564 N |
4983 | mddev->queue->queue_lock = &conf->device_lock; |
4984 | ||
7a5febe9 | 4985 | mddev->queue->unplug_fn = raid5_unplug_device; |
f022b2fd | 4986 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 4987 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 4988 | |
1f403624 | 4989 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 4990 | |
23032a0e | 4991 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
8f6c2e4b MP |
4992 | chunk_size = mddev->chunk_sectors << 9; |
4993 | blk_queue_io_min(mddev->queue, chunk_size); | |
4994 | blk_queue_io_opt(mddev->queue, chunk_size * | |
4995 | (conf->raid_disks - conf->max_degraded)); | |
4996 | ||
4997 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4998 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
4999 | rdev->data_offset << 9); | |
23032a0e | 5000 | |
1da177e4 LT |
5001 | return 0; |
5002 | abort: | |
e0cf8f04 | 5003 | md_unregister_thread(mddev->thread); |
91adb564 | 5004 | mddev->thread = NULL; |
1da177e4 LT |
5005 | if (conf) { |
5006 | print_raid5_conf(conf); | |
95fc17aa | 5007 | free_conf(conf); |
1da177e4 LT |
5008 | } |
5009 | mddev->private = NULL; | |
5010 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
5011 | return -EIO; | |
5012 | } | |
5013 | ||
5014 | ||
5015 | ||
3f294f4f | 5016 | static int stop(mddev_t *mddev) |
1da177e4 LT |
5017 | { |
5018 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
5019 | ||
5020 | md_unregister_thread(mddev->thread); | |
5021 | mddev->thread = NULL; | |
041ae52e | 5022 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 5023 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 5024 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
95fc17aa | 5025 | free_conf(conf); |
1da177e4 LT |
5026 | mddev->private = NULL; |
5027 | return 0; | |
5028 | } | |
5029 | ||
45b4233c | 5030 | #ifdef DEBUG |
d710e138 | 5031 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
5032 | { |
5033 | int i; | |
5034 | ||
16a53ecc N |
5035 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
5036 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
5037 | seq_printf(seq, "sh %llu, count %d.\n", | |
5038 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
5039 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 5040 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
5041 | seq_printf(seq, "(cache%d: %p %ld) ", |
5042 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 5043 | } |
16a53ecc | 5044 | seq_printf(seq, "\n"); |
1da177e4 LT |
5045 | } |
5046 | ||
d710e138 | 5047 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
5048 | { |
5049 | struct stripe_head *sh; | |
fccddba0 | 5050 | struct hlist_node *hn; |
1da177e4 LT |
5051 | int i; |
5052 | ||
5053 | spin_lock_irq(&conf->device_lock); | |
5054 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 5055 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
5056 | if (sh->raid_conf != conf) |
5057 | continue; | |
16a53ecc | 5058 | print_sh(seq, sh); |
1da177e4 LT |
5059 | } |
5060 | } | |
5061 | spin_unlock_irq(&conf->device_lock); | |
5062 | } | |
5063 | #endif | |
5064 | ||
d710e138 | 5065 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
5066 | { |
5067 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
5068 | int i; | |
5069 | ||
9d8f0363 AN |
5070 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5071 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5072 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5073 | for (i = 0; i < conf->raid_disks; i++) |
5074 | seq_printf (seq, "%s", | |
5075 | conf->disks[i].rdev && | |
b2d444d7 | 5076 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5077 | seq_printf (seq, "]"); |
45b4233c | 5078 | #ifdef DEBUG |
16a53ecc N |
5079 | seq_printf (seq, "\n"); |
5080 | printall(seq, conf); | |
1da177e4 LT |
5081 | #endif |
5082 | } | |
5083 | ||
5084 | static void print_raid5_conf (raid5_conf_t *conf) | |
5085 | { | |
5086 | int i; | |
5087 | struct disk_info *tmp; | |
5088 | ||
5089 | printk("RAID5 conf printout:\n"); | |
5090 | if (!conf) { | |
5091 | printk("(conf==NULL)\n"); | |
5092 | return; | |
5093 | } | |
02c2de8c N |
5094 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
5095 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5096 | |
5097 | for (i = 0; i < conf->raid_disks; i++) { | |
5098 | char b[BDEVNAME_SIZE]; | |
5099 | tmp = conf->disks + i; | |
5100 | if (tmp->rdev) | |
5101 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 5102 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
5103 | bdevname(tmp->rdev->bdev,b)); |
5104 | } | |
5105 | } | |
5106 | ||
5107 | static int raid5_spare_active(mddev_t *mddev) | |
5108 | { | |
5109 | int i; | |
5110 | raid5_conf_t *conf = mddev->private; | |
5111 | struct disk_info *tmp; | |
5112 | ||
5113 | for (i = 0; i < conf->raid_disks; i++) { | |
5114 | tmp = conf->disks + i; | |
5115 | if (tmp->rdev | |
b2d444d7 | 5116 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
5117 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
5118 | unsigned long flags; | |
5119 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 5120 | mddev->degraded--; |
c04be0aa | 5121 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
5122 | } |
5123 | } | |
5124 | print_raid5_conf(conf); | |
5125 | return 0; | |
5126 | } | |
5127 | ||
5128 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
5129 | { | |
5130 | raid5_conf_t *conf = mddev->private; | |
5131 | int err = 0; | |
5132 | mdk_rdev_t *rdev; | |
5133 | struct disk_info *p = conf->disks + number; | |
5134 | ||
5135 | print_raid5_conf(conf); | |
5136 | rdev = p->rdev; | |
5137 | if (rdev) { | |
ec32a2bd N |
5138 | if (number >= conf->raid_disks && |
5139 | conf->reshape_progress == MaxSector) | |
5140 | clear_bit(In_sync, &rdev->flags); | |
5141 | ||
b2d444d7 | 5142 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
5143 | atomic_read(&rdev->nr_pending)) { |
5144 | err = -EBUSY; | |
5145 | goto abort; | |
5146 | } | |
dfc70645 N |
5147 | /* Only remove non-faulty devices if recovery |
5148 | * isn't possible. | |
5149 | */ | |
5150 | if (!test_bit(Faulty, &rdev->flags) && | |
ec32a2bd N |
5151 | mddev->degraded <= conf->max_degraded && |
5152 | number < conf->raid_disks) { | |
dfc70645 N |
5153 | err = -EBUSY; |
5154 | goto abort; | |
5155 | } | |
1da177e4 | 5156 | p->rdev = NULL; |
fbd568a3 | 5157 | synchronize_rcu(); |
1da177e4 LT |
5158 | if (atomic_read(&rdev->nr_pending)) { |
5159 | /* lost the race, try later */ | |
5160 | err = -EBUSY; | |
5161 | p->rdev = rdev; | |
5162 | } | |
5163 | } | |
5164 | abort: | |
5165 | ||
5166 | print_raid5_conf(conf); | |
5167 | return err; | |
5168 | } | |
5169 | ||
5170 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
5171 | { | |
5172 | raid5_conf_t *conf = mddev->private; | |
199050ea | 5173 | int err = -EEXIST; |
1da177e4 LT |
5174 | int disk; |
5175 | struct disk_info *p; | |
6c2fce2e NB |
5176 | int first = 0; |
5177 | int last = conf->raid_disks - 1; | |
1da177e4 | 5178 | |
16a53ecc | 5179 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 5180 | /* no point adding a device */ |
199050ea | 5181 | return -EINVAL; |
1da177e4 | 5182 | |
6c2fce2e NB |
5183 | if (rdev->raid_disk >= 0) |
5184 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5185 | |
5186 | /* | |
16a53ecc N |
5187 | * find the disk ... but prefer rdev->saved_raid_disk |
5188 | * if possible. | |
1da177e4 | 5189 | */ |
16a53ecc | 5190 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5191 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5192 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5193 | disk = rdev->saved_raid_disk; | |
5194 | else | |
6c2fce2e NB |
5195 | disk = first; |
5196 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5197 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5198 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5199 | rdev->raid_disk = disk; |
199050ea | 5200 | err = 0; |
72626685 N |
5201 | if (rdev->saved_raid_disk != disk) |
5202 | conf->fullsync = 1; | |
d6065f7b | 5203 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5204 | break; |
5205 | } | |
5206 | print_raid5_conf(conf); | |
199050ea | 5207 | return err; |
1da177e4 LT |
5208 | } |
5209 | ||
5210 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
5211 | { | |
5212 | /* no resync is happening, and there is enough space | |
5213 | * on all devices, so we can resize. | |
5214 | * We need to make sure resync covers any new space. | |
5215 | * If the array is shrinking we should possibly wait until | |
5216 | * any io in the removed space completes, but it hardly seems | |
5217 | * worth it. | |
5218 | */ | |
9d8f0363 | 5219 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5220 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5221 | mddev->raid_disks)); | |
b522adcd DW |
5222 | if (mddev->array_sectors > |
5223 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5224 | return -EINVAL; | |
f233ea5c | 5225 | set_capacity(mddev->gendisk, mddev->array_sectors); |
44ce6294 | 5226 | mddev->changed = 1; |
449aad3e | 5227 | revalidate_disk(mddev->gendisk); |
58c0fed4 AN |
5228 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
5229 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
5230 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5231 | } | |
58c0fed4 | 5232 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5233 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5234 | return 0; |
5235 | } | |
5236 | ||
01ee22b4 N |
5237 | static int check_stripe_cache(mddev_t *mddev) |
5238 | { | |
5239 | /* Can only proceed if there are plenty of stripe_heads. | |
5240 | * We need a minimum of one full stripe,, and for sensible progress | |
5241 | * it is best to have about 4 times that. | |
5242 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5243 | * allow for chunk sizes up to 256K, which is probably OK. | |
5244 | * If the chunk size is greater, user-space should request more | |
5245 | * stripe_heads first. | |
5246 | */ | |
5247 | raid5_conf_t *conf = mddev->private; | |
5248 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5249 | > conf->max_nr_stripes || | |
5250 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5251 | > conf->max_nr_stripes) { | |
5252 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", | |
5253 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
5254 | / STRIPE_SIZE)*4); | |
5255 | return 0; | |
5256 | } | |
5257 | return 1; | |
5258 | } | |
5259 | ||
50ac168a | 5260 | static int check_reshape(mddev_t *mddev) |
29269553 | 5261 | { |
070ec55d | 5262 | raid5_conf_t *conf = mddev->private; |
29269553 | 5263 | |
88ce4930 N |
5264 | if (mddev->delta_disks == 0 && |
5265 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5266 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5267 | return 0; /* nothing to do */ |
dba034ee N |
5268 | if (mddev->bitmap) |
5269 | /* Cannot grow a bitmap yet */ | |
5270 | return -EBUSY; | |
ec32a2bd N |
5271 | if (mddev->degraded > conf->max_degraded) |
5272 | return -EINVAL; | |
5273 | if (mddev->delta_disks < 0) { | |
5274 | /* We might be able to shrink, but the devices must | |
5275 | * be made bigger first. | |
5276 | * For raid6, 4 is the minimum size. | |
5277 | * Otherwise 2 is the minimum | |
5278 | */ | |
5279 | int min = 2; | |
5280 | if (mddev->level == 6) | |
5281 | min = 4; | |
5282 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5283 | return -EINVAL; | |
5284 | } | |
29269553 | 5285 | |
01ee22b4 | 5286 | if (!check_stripe_cache(mddev)) |
29269553 | 5287 | return -ENOSPC; |
29269553 | 5288 | |
ec32a2bd | 5289 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5290 | } |
5291 | ||
5292 | static int raid5_start_reshape(mddev_t *mddev) | |
5293 | { | |
070ec55d | 5294 | raid5_conf_t *conf = mddev->private; |
63c70c4f | 5295 | mdk_rdev_t *rdev; |
63c70c4f N |
5296 | int spares = 0; |
5297 | int added_devices = 0; | |
c04be0aa | 5298 | unsigned long flags; |
63c70c4f | 5299 | |
f416885e | 5300 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5301 | return -EBUSY; |
5302 | ||
01ee22b4 N |
5303 | if (!check_stripe_cache(mddev)) |
5304 | return -ENOSPC; | |
5305 | ||
159ec1fc | 5306 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5307 | if (rdev->raid_disk < 0 && |
5308 | !test_bit(Faulty, &rdev->flags)) | |
5309 | spares++; | |
63c70c4f | 5310 | |
f416885e | 5311 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5312 | /* Not enough devices even to make a degraded array |
5313 | * of that size | |
5314 | */ | |
5315 | return -EINVAL; | |
5316 | ||
ec32a2bd N |
5317 | /* Refuse to reduce size of the array. Any reductions in |
5318 | * array size must be through explicit setting of array_size | |
5319 | * attribute. | |
5320 | */ | |
5321 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5322 | < mddev->array_sectors) { | |
5323 | printk(KERN_ERR "md: %s: array size must be reduced " | |
5324 | "before number of disks\n", mdname(mddev)); | |
5325 | return -EINVAL; | |
5326 | } | |
5327 | ||
f6705578 | 5328 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5329 | spin_lock_irq(&conf->device_lock); |
5330 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5331 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5332 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5333 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5334 | conf->prev_algo = conf->algorithm; |
5335 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5336 | if (mddev->delta_disks < 0) |
5337 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5338 | else | |
5339 | conf->reshape_progress = 0; | |
5340 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5341 | conf->generation++; |
29269553 N |
5342 | spin_unlock_irq(&conf->device_lock); |
5343 | ||
5344 | /* Add some new drives, as many as will fit. | |
5345 | * We know there are enough to make the newly sized array work. | |
5346 | */ | |
159ec1fc | 5347 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5348 | if (rdev->raid_disk < 0 && |
5349 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 5350 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 N |
5351 | char nm[20]; |
5352 | set_bit(In_sync, &rdev->flags); | |
29269553 | 5353 | added_devices++; |
5fd6c1dc | 5354 | rdev->recovery_offset = 0; |
29269553 | 5355 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
5356 | if (sysfs_create_link(&mddev->kobj, |
5357 | &rdev->kobj, nm)) | |
5358 | printk(KERN_WARNING | |
5359 | "raid5: failed to create " | |
5360 | " link %s for %s\n", | |
5361 | nm, mdname(mddev)); | |
29269553 N |
5362 | } else |
5363 | break; | |
5364 | } | |
5365 | ||
ec32a2bd N |
5366 | if (mddev->delta_disks > 0) { |
5367 | spin_lock_irqsave(&conf->device_lock, flags); | |
5368 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) | |
5369 | - added_devices; | |
5370 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5371 | } | |
63c70c4f | 5372 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5373 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5374 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5375 | |
29269553 N |
5376 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5377 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5378 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5379 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5380 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5381 | "reshape"); |
29269553 N |
5382 | if (!mddev->sync_thread) { |
5383 | mddev->recovery = 0; | |
5384 | spin_lock_irq(&conf->device_lock); | |
5385 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5386 | conf->reshape_progress = MaxSector; |
29269553 N |
5387 | spin_unlock_irq(&conf->device_lock); |
5388 | return -EAGAIN; | |
5389 | } | |
c8f517c4 | 5390 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5391 | md_wakeup_thread(mddev->sync_thread); |
5392 | md_new_event(mddev); | |
5393 | return 0; | |
5394 | } | |
29269553 | 5395 | |
ec32a2bd N |
5396 | /* This is called from the reshape thread and should make any |
5397 | * changes needed in 'conf' | |
5398 | */ | |
29269553 N |
5399 | static void end_reshape(raid5_conf_t *conf) |
5400 | { | |
29269553 | 5401 | |
f6705578 | 5402 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5403 | |
f6705578 | 5404 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5405 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5406 | conf->reshape_progress = MaxSector; |
f6705578 | 5407 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5408 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5409 | |
5410 | /* read-ahead size must cover two whole stripes, which is | |
5411 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5412 | */ | |
5413 | { | |
cea9c228 | 5414 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5415 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5416 | / PAGE_SIZE); |
16a53ecc N |
5417 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5418 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5419 | } | |
29269553 | 5420 | } |
29269553 N |
5421 | } |
5422 | ||
ec32a2bd N |
5423 | /* This is called from the raid5d thread with mddev_lock held. |
5424 | * It makes config changes to the device. | |
5425 | */ | |
cea9c228 N |
5426 | static void raid5_finish_reshape(mddev_t *mddev) |
5427 | { | |
070ec55d | 5428 | raid5_conf_t *conf = mddev->private; |
cea9c228 N |
5429 | |
5430 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5431 | ||
ec32a2bd N |
5432 | if (mddev->delta_disks > 0) { |
5433 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5434 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
5435 | mddev->changed = 1; | |
449aad3e | 5436 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5437 | } else { |
5438 | int d; | |
ec32a2bd N |
5439 | mddev->degraded = conf->raid_disks; |
5440 | for (d = 0; d < conf->raid_disks ; d++) | |
5441 | if (conf->disks[d].rdev && | |
5442 | test_bit(In_sync, | |
5443 | &conf->disks[d].rdev->flags)) | |
5444 | mddev->degraded--; | |
5445 | for (d = conf->raid_disks ; | |
5446 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 N |
5447 | d++) { |
5448 | mdk_rdev_t *rdev = conf->disks[d].rdev; | |
5449 | if (rdev && raid5_remove_disk(mddev, d) == 0) { | |
5450 | char nm[20]; | |
5451 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5452 | sysfs_remove_link(&mddev->kobj, nm); | |
5453 | rdev->raid_disk = -1; | |
5454 | } | |
5455 | } | |
cea9c228 | 5456 | } |
88ce4930 | 5457 | mddev->layout = conf->algorithm; |
09c9e5fa | 5458 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5459 | mddev->reshape_position = MaxSector; |
5460 | mddev->delta_disks = 0; | |
cea9c228 N |
5461 | } |
5462 | } | |
5463 | ||
72626685 N |
5464 | static void raid5_quiesce(mddev_t *mddev, int state) |
5465 | { | |
070ec55d | 5466 | raid5_conf_t *conf = mddev->private; |
72626685 N |
5467 | |
5468 | switch(state) { | |
e464eafd N |
5469 | case 2: /* resume for a suspend */ |
5470 | wake_up(&conf->wait_for_overlap); | |
5471 | break; | |
5472 | ||
72626685 N |
5473 | case 1: /* stop all writes */ |
5474 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5475 | /* '2' tells resync/reshape to pause so that all |
5476 | * active stripes can drain | |
5477 | */ | |
5478 | conf->quiesce = 2; | |
72626685 | 5479 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5480 | atomic_read(&conf->active_stripes) == 0 && |
5481 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5482 | conf->device_lock, /* nothing */); |
64bd660b | 5483 | conf->quiesce = 1; |
72626685 | 5484 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5485 | /* allow reshape to continue */ |
5486 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5487 | break; |
5488 | ||
5489 | case 0: /* re-enable writes */ | |
5490 | spin_lock_irq(&conf->device_lock); | |
5491 | conf->quiesce = 0; | |
5492 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5493 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5494 | spin_unlock_irq(&conf->device_lock); |
5495 | break; | |
5496 | } | |
72626685 | 5497 | } |
b15c2e57 | 5498 | |
d562b0c4 N |
5499 | |
5500 | static void *raid5_takeover_raid1(mddev_t *mddev) | |
5501 | { | |
5502 | int chunksect; | |
5503 | ||
5504 | if (mddev->raid_disks != 2 || | |
5505 | mddev->degraded > 1) | |
5506 | return ERR_PTR(-EINVAL); | |
5507 | ||
5508 | /* Should check if there are write-behind devices? */ | |
5509 | ||
5510 | chunksect = 64*2; /* 64K by default */ | |
5511 | ||
5512 | /* The array must be an exact multiple of chunksize */ | |
5513 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5514 | chunksect >>= 1; | |
5515 | ||
5516 | if ((chunksect<<9) < STRIPE_SIZE) | |
5517 | /* array size does not allow a suitable chunk size */ | |
5518 | return ERR_PTR(-EINVAL); | |
5519 | ||
5520 | mddev->new_level = 5; | |
5521 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5522 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5523 | |
5524 | return setup_conf(mddev); | |
5525 | } | |
5526 | ||
fc9739c6 N |
5527 | static void *raid5_takeover_raid6(mddev_t *mddev) |
5528 | { | |
5529 | int new_layout; | |
5530 | ||
5531 | switch (mddev->layout) { | |
5532 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5533 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5534 | break; | |
5535 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5536 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5537 | break; | |
5538 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5539 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5540 | break; | |
5541 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5542 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5543 | break; | |
5544 | case ALGORITHM_PARITY_0_6: | |
5545 | new_layout = ALGORITHM_PARITY_0; | |
5546 | break; | |
5547 | case ALGORITHM_PARITY_N: | |
5548 | new_layout = ALGORITHM_PARITY_N; | |
5549 | break; | |
5550 | default: | |
5551 | return ERR_PTR(-EINVAL); | |
5552 | } | |
5553 | mddev->new_level = 5; | |
5554 | mddev->new_layout = new_layout; | |
5555 | mddev->delta_disks = -1; | |
5556 | mddev->raid_disks -= 1; | |
5557 | return setup_conf(mddev); | |
5558 | } | |
5559 | ||
d562b0c4 | 5560 | |
50ac168a | 5561 | static int raid5_check_reshape(mddev_t *mddev) |
b3546035 | 5562 | { |
88ce4930 N |
5563 | /* For a 2-drive array, the layout and chunk size can be changed |
5564 | * immediately as not restriping is needed. | |
5565 | * For larger arrays we record the new value - after validation | |
5566 | * to be used by a reshape pass. | |
b3546035 | 5567 | */ |
070ec55d | 5568 | raid5_conf_t *conf = mddev->private; |
597a711b | 5569 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5570 | |
597a711b | 5571 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5572 | return -EINVAL; |
5573 | if (new_chunk > 0) { | |
0ba459d2 | 5574 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5575 | return -EINVAL; |
597a711b | 5576 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5577 | return -EINVAL; |
597a711b | 5578 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5579 | /* not factor of array size */ |
5580 | return -EINVAL; | |
5581 | } | |
5582 | ||
5583 | /* They look valid */ | |
5584 | ||
88ce4930 | 5585 | if (mddev->raid_disks == 2) { |
597a711b N |
5586 | /* can make the change immediately */ |
5587 | if (mddev->new_layout >= 0) { | |
5588 | conf->algorithm = mddev->new_layout; | |
5589 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5590 | } |
5591 | if (new_chunk > 0) { | |
597a711b N |
5592 | conf->chunk_sectors = new_chunk ; |
5593 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5594 | } |
5595 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5596 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5597 | } |
50ac168a | 5598 | return check_reshape(mddev); |
88ce4930 N |
5599 | } |
5600 | ||
50ac168a | 5601 | static int raid6_check_reshape(mddev_t *mddev) |
88ce4930 | 5602 | { |
597a711b | 5603 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5604 | |
597a711b | 5605 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5606 | return -EINVAL; |
b3546035 | 5607 | if (new_chunk > 0) { |
0ba459d2 | 5608 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5609 | return -EINVAL; |
597a711b | 5610 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5611 | return -EINVAL; |
597a711b | 5612 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5613 | /* not factor of array size */ |
5614 | return -EINVAL; | |
b3546035 | 5615 | } |
88ce4930 N |
5616 | |
5617 | /* They look valid */ | |
50ac168a | 5618 | return check_reshape(mddev); |
b3546035 N |
5619 | } |
5620 | ||
d562b0c4 N |
5621 | static void *raid5_takeover(mddev_t *mddev) |
5622 | { | |
5623 | /* raid5 can take over: | |
5624 | * raid0 - if all devices are the same - make it a raid4 layout | |
5625 | * raid1 - if there are two drives. We need to know the chunk size | |
5626 | * raid4 - trivial - just use a raid4 layout. | |
5627 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 N |
5628 | */ |
5629 | ||
5630 | if (mddev->level == 1) | |
5631 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5632 | if (mddev->level == 4) { |
5633 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5634 | mddev->new_level = 5; | |
5635 | return setup_conf(mddev); | |
5636 | } | |
fc9739c6 N |
5637 | if (mddev->level == 6) |
5638 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5639 | |
5640 | return ERR_PTR(-EINVAL); | |
5641 | } | |
5642 | ||
5643 | ||
245f46c2 N |
5644 | static struct mdk_personality raid5_personality; |
5645 | ||
5646 | static void *raid6_takeover(mddev_t *mddev) | |
5647 | { | |
5648 | /* Currently can only take over a raid5. We map the | |
5649 | * personality to an equivalent raid6 personality | |
5650 | * with the Q block at the end. | |
5651 | */ | |
5652 | int new_layout; | |
5653 | ||
5654 | if (mddev->pers != &raid5_personality) | |
5655 | return ERR_PTR(-EINVAL); | |
5656 | if (mddev->degraded > 1) | |
5657 | return ERR_PTR(-EINVAL); | |
5658 | if (mddev->raid_disks > 253) | |
5659 | return ERR_PTR(-EINVAL); | |
5660 | if (mddev->raid_disks < 3) | |
5661 | return ERR_PTR(-EINVAL); | |
5662 | ||
5663 | switch (mddev->layout) { | |
5664 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5665 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5666 | break; | |
5667 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5668 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5669 | break; | |
5670 | case ALGORITHM_LEFT_SYMMETRIC: | |
5671 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5672 | break; | |
5673 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5674 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5675 | break; | |
5676 | case ALGORITHM_PARITY_0: | |
5677 | new_layout = ALGORITHM_PARITY_0_6; | |
5678 | break; | |
5679 | case ALGORITHM_PARITY_N: | |
5680 | new_layout = ALGORITHM_PARITY_N; | |
5681 | break; | |
5682 | default: | |
5683 | return ERR_PTR(-EINVAL); | |
5684 | } | |
5685 | mddev->new_level = 6; | |
5686 | mddev->new_layout = new_layout; | |
5687 | mddev->delta_disks = 1; | |
5688 | mddev->raid_disks += 1; | |
5689 | return setup_conf(mddev); | |
5690 | } | |
5691 | ||
5692 | ||
16a53ecc N |
5693 | static struct mdk_personality raid6_personality = |
5694 | { | |
5695 | .name = "raid6", | |
5696 | .level = 6, | |
5697 | .owner = THIS_MODULE, | |
5698 | .make_request = make_request, | |
5699 | .run = run, | |
5700 | .stop = stop, | |
5701 | .status = status, | |
5702 | .error_handler = error, | |
5703 | .hot_add_disk = raid5_add_disk, | |
5704 | .hot_remove_disk= raid5_remove_disk, | |
5705 | .spare_active = raid5_spare_active, | |
5706 | .sync_request = sync_request, | |
5707 | .resize = raid5_resize, | |
80c3a6ce | 5708 | .size = raid5_size, |
50ac168a | 5709 | .check_reshape = raid6_check_reshape, |
f416885e | 5710 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5711 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5712 | .quiesce = raid5_quiesce, |
245f46c2 | 5713 | .takeover = raid6_takeover, |
16a53ecc | 5714 | }; |
2604b703 | 5715 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
5716 | { |
5717 | .name = "raid5", | |
2604b703 | 5718 | .level = 5, |
1da177e4 LT |
5719 | .owner = THIS_MODULE, |
5720 | .make_request = make_request, | |
5721 | .run = run, | |
5722 | .stop = stop, | |
5723 | .status = status, | |
5724 | .error_handler = error, | |
5725 | .hot_add_disk = raid5_add_disk, | |
5726 | .hot_remove_disk= raid5_remove_disk, | |
5727 | .spare_active = raid5_spare_active, | |
5728 | .sync_request = sync_request, | |
5729 | .resize = raid5_resize, | |
80c3a6ce | 5730 | .size = raid5_size, |
63c70c4f N |
5731 | .check_reshape = raid5_check_reshape, |
5732 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5733 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5734 | .quiesce = raid5_quiesce, |
d562b0c4 | 5735 | .takeover = raid5_takeover, |
1da177e4 LT |
5736 | }; |
5737 | ||
2604b703 | 5738 | static struct mdk_personality raid4_personality = |
1da177e4 | 5739 | { |
2604b703 N |
5740 | .name = "raid4", |
5741 | .level = 4, | |
5742 | .owner = THIS_MODULE, | |
5743 | .make_request = make_request, | |
5744 | .run = run, | |
5745 | .stop = stop, | |
5746 | .status = status, | |
5747 | .error_handler = error, | |
5748 | .hot_add_disk = raid5_add_disk, | |
5749 | .hot_remove_disk= raid5_remove_disk, | |
5750 | .spare_active = raid5_spare_active, | |
5751 | .sync_request = sync_request, | |
5752 | .resize = raid5_resize, | |
80c3a6ce | 5753 | .size = raid5_size, |
3d37890b N |
5754 | .check_reshape = raid5_check_reshape, |
5755 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5756 | .finish_reshape = raid5_finish_reshape, |
2604b703 N |
5757 | .quiesce = raid5_quiesce, |
5758 | }; | |
5759 | ||
5760 | static int __init raid5_init(void) | |
5761 | { | |
16a53ecc | 5762 | register_md_personality(&raid6_personality); |
2604b703 N |
5763 | register_md_personality(&raid5_personality); |
5764 | register_md_personality(&raid4_personality); | |
5765 | return 0; | |
1da177e4 LT |
5766 | } |
5767 | ||
2604b703 | 5768 | static void raid5_exit(void) |
1da177e4 | 5769 | { |
16a53ecc | 5770 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5771 | unregister_md_personality(&raid5_personality); |
5772 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5773 | } |
5774 | ||
5775 | module_init(raid5_init); | |
5776 | module_exit(raid5_exit); | |
5777 | MODULE_LICENSE("GPL"); | |
5778 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
5779 | MODULE_ALIAS("md-raid5"); |
5780 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5781 | MODULE_ALIAS("md-level-5"); |
5782 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5783 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5784 | MODULE_ALIAS("md-raid6"); | |
5785 | MODULE_ALIAS("md-level-6"); | |
5786 | ||
5787 | /* This used to be two separate modules, they were: */ | |
5788 | MODULE_ALIAS("raid5"); | |
5789 | MODULE_ALIAS("raid6"); |