]>
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 | 158 | { |
6629542e | 159 | int slot = *count; |
67cc2b81 | 160 | |
e4424fee | 161 | if (sh->ddf_layout) |
6629542e | 162 | (*count)++; |
d0dabf7e | 163 | if (idx == sh->pd_idx) |
67cc2b81 | 164 | return syndrome_disks; |
d0dabf7e | 165 | if (idx == sh->qd_idx) |
67cc2b81 | 166 | return syndrome_disks + 1; |
e4424fee | 167 | if (!sh->ddf_layout) |
6629542e | 168 | (*count)++; |
d0dabf7e N |
169 | return slot; |
170 | } | |
171 | ||
a4456856 DW |
172 | static void return_io(struct bio *return_bi) |
173 | { | |
174 | struct bio *bi = return_bi; | |
175 | while (bi) { | |
a4456856 DW |
176 | |
177 | return_bi = bi->bi_next; | |
178 | bi->bi_next = NULL; | |
179 | bi->bi_size = 0; | |
0e13fe23 | 180 | bio_endio(bi, 0); |
a4456856 DW |
181 | bi = return_bi; |
182 | } | |
183 | } | |
184 | ||
1da177e4 LT |
185 | static void print_raid5_conf (raid5_conf_t *conf); |
186 | ||
600aa109 DW |
187 | static int stripe_operations_active(struct stripe_head *sh) |
188 | { | |
189 | return sh->check_state || sh->reconstruct_state || | |
190 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
191 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
192 | } | |
193 | ||
858119e1 | 194 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
195 | { |
196 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
197 | BUG_ON(!list_empty(&sh->lru)); |
198 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 199 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 200 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 201 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
202 | blk_plug_device(conf->mddev->queue); |
203 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 204 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 205 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
206 | blk_plug_device(conf->mddev->queue); |
207 | } else { | |
72626685 | 208 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 209 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 210 | } |
1da177e4 LT |
211 | md_wakeup_thread(conf->mddev->thread); |
212 | } else { | |
600aa109 | 213 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
214 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
215 | atomic_dec(&conf->preread_active_stripes); | |
216 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
217 | md_wakeup_thread(conf->mddev->thread); | |
218 | } | |
1da177e4 | 219 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
220 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
221 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 222 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
223 | if (conf->retry_read_aligned) |
224 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 225 | } |
1da177e4 LT |
226 | } |
227 | } | |
228 | } | |
d0dabf7e | 229 | |
1da177e4 LT |
230 | static void release_stripe(struct stripe_head *sh) |
231 | { | |
232 | raid5_conf_t *conf = sh->raid_conf; | |
233 | unsigned long flags; | |
16a53ecc | 234 | |
1da177e4 LT |
235 | spin_lock_irqsave(&conf->device_lock, flags); |
236 | __release_stripe(conf, sh); | |
237 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
238 | } | |
239 | ||
fccddba0 | 240 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 241 | { |
45b4233c DW |
242 | pr_debug("remove_hash(), stripe %llu\n", |
243 | (unsigned long long)sh->sector); | |
1da177e4 | 244 | |
fccddba0 | 245 | hlist_del_init(&sh->hash); |
1da177e4 LT |
246 | } |
247 | ||
16a53ecc | 248 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 249 | { |
fccddba0 | 250 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 251 | |
45b4233c DW |
252 | pr_debug("insert_hash(), stripe %llu\n", |
253 | (unsigned long long)sh->sector); | |
1da177e4 LT |
254 | |
255 | CHECK_DEVLOCK(); | |
fccddba0 | 256 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
257 | } |
258 | ||
259 | ||
260 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
261 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
262 | { | |
263 | struct stripe_head *sh = NULL; | |
264 | struct list_head *first; | |
265 | ||
266 | CHECK_DEVLOCK(); | |
267 | if (list_empty(&conf->inactive_list)) | |
268 | goto out; | |
269 | first = conf->inactive_list.next; | |
270 | sh = list_entry(first, struct stripe_head, lru); | |
271 | list_del_init(first); | |
272 | remove_hash(sh); | |
273 | atomic_inc(&conf->active_stripes); | |
274 | out: | |
275 | return sh; | |
276 | } | |
277 | ||
278 | static void shrink_buffers(struct stripe_head *sh, int num) | |
279 | { | |
280 | struct page *p; | |
281 | int i; | |
282 | ||
283 | for (i=0; i<num ; i++) { | |
284 | p = sh->dev[i].page; | |
285 | if (!p) | |
286 | continue; | |
287 | sh->dev[i].page = NULL; | |
2d1f3b5d | 288 | put_page(p); |
1da177e4 LT |
289 | } |
290 | } | |
291 | ||
292 | static int grow_buffers(struct stripe_head *sh, int num) | |
293 | { | |
294 | int i; | |
295 | ||
296 | for (i=0; i<num; i++) { | |
297 | struct page *page; | |
298 | ||
299 | if (!(page = alloc_page(GFP_KERNEL))) { | |
300 | return 1; | |
301 | } | |
302 | sh->dev[i].page = page; | |
303 | } | |
304 | return 0; | |
305 | } | |
306 | ||
784052ec | 307 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
911d4ee8 N |
308 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
309 | struct stripe_head *sh); | |
1da177e4 | 310 | |
b5663ba4 | 311 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
312 | { |
313 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 314 | int i; |
1da177e4 | 315 | |
78bafebd ES |
316 | BUG_ON(atomic_read(&sh->count) != 0); |
317 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 318 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 319 | |
1da177e4 | 320 | CHECK_DEVLOCK(); |
45b4233c | 321 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
322 | (unsigned long long)sh->sector); |
323 | ||
324 | remove_hash(sh); | |
16a53ecc | 325 | |
86b42c71 | 326 | sh->generation = conf->generation - previous; |
b5663ba4 | 327 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 328 | sh->sector = sector; |
911d4ee8 | 329 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
330 | sh->state = 0; |
331 | ||
7ecaa1e6 N |
332 | |
333 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
334 | struct r5dev *dev = &sh->dev[i]; |
335 | ||
d84e0f10 | 336 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 337 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 338 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 339 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 340 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
341 | test_bit(R5_LOCKED, &dev->flags)); |
342 | BUG(); | |
343 | } | |
344 | dev->flags = 0; | |
784052ec | 345 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
346 | } |
347 | insert_hash(conf, sh); | |
348 | } | |
349 | ||
86b42c71 N |
350 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, |
351 | short generation) | |
1da177e4 LT |
352 | { |
353 | struct stripe_head *sh; | |
fccddba0 | 354 | struct hlist_node *hn; |
1da177e4 LT |
355 | |
356 | CHECK_DEVLOCK(); | |
45b4233c | 357 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 358 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 359 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 360 | return sh; |
45b4233c | 361 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
362 | return NULL; |
363 | } | |
364 | ||
365 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 366 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 367 | |
b5663ba4 N |
368 | static struct stripe_head * |
369 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
a8c906ca | 370 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
371 | { |
372 | struct stripe_head *sh; | |
373 | ||
45b4233c | 374 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
375 | |
376 | spin_lock_irq(&conf->device_lock); | |
377 | ||
378 | do { | |
72626685 | 379 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 380 | conf->quiesce == 0 || noquiesce, |
72626685 | 381 | conf->device_lock, /* nothing */); |
86b42c71 | 382 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
383 | if (!sh) { |
384 | if (!conf->inactive_blocked) | |
385 | sh = get_free_stripe(conf); | |
386 | if (noblock && sh == NULL) | |
387 | break; | |
388 | if (!sh) { | |
389 | conf->inactive_blocked = 1; | |
390 | wait_event_lock_irq(conf->wait_for_stripe, | |
391 | !list_empty(&conf->inactive_list) && | |
5036805b N |
392 | (atomic_read(&conf->active_stripes) |
393 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
394 | || !conf->inactive_blocked), |
395 | conf->device_lock, | |
f4370781 | 396 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
397 | ); |
398 | conf->inactive_blocked = 0; | |
399 | } else | |
b5663ba4 | 400 | init_stripe(sh, sector, previous); |
1da177e4 LT |
401 | } else { |
402 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
403 | BUG_ON(!list_empty(&sh->lru) |
404 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
405 | } else { |
406 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
407 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
408 | if (list_empty(&sh->lru) && |
409 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
410 | BUG(); |
411 | list_del_init(&sh->lru); | |
1da177e4 LT |
412 | } |
413 | } | |
414 | } while (sh == NULL); | |
415 | ||
416 | if (sh) | |
417 | atomic_inc(&sh->count); | |
418 | ||
419 | spin_unlock_irq(&conf->device_lock); | |
420 | return sh; | |
421 | } | |
422 | ||
6712ecf8 N |
423 | static void |
424 | raid5_end_read_request(struct bio *bi, int error); | |
425 | static void | |
426 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 427 | |
c4e5ac0a | 428 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
429 | { |
430 | raid5_conf_t *conf = sh->raid_conf; | |
431 | int i, disks = sh->disks; | |
432 | ||
433 | might_sleep(); | |
434 | ||
435 | for (i = disks; i--; ) { | |
436 | int rw; | |
437 | struct bio *bi; | |
438 | mdk_rdev_t *rdev; | |
439 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
440 | rw = WRITE; | |
441 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
442 | rw = READ; | |
443 | else | |
444 | continue; | |
445 | ||
446 | bi = &sh->dev[i].req; | |
447 | ||
448 | bi->bi_rw = rw; | |
449 | if (rw == WRITE) | |
450 | bi->bi_end_io = raid5_end_write_request; | |
451 | else | |
452 | bi->bi_end_io = raid5_end_read_request; | |
453 | ||
454 | rcu_read_lock(); | |
455 | rdev = rcu_dereference(conf->disks[i].rdev); | |
456 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
457 | rdev = NULL; | |
458 | if (rdev) | |
459 | atomic_inc(&rdev->nr_pending); | |
460 | rcu_read_unlock(); | |
461 | ||
462 | if (rdev) { | |
c4e5ac0a | 463 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
464 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
465 | ||
2b7497f0 DW |
466 | set_bit(STRIPE_IO_STARTED, &sh->state); |
467 | ||
91c00924 DW |
468 | bi->bi_bdev = rdev->bdev; |
469 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 470 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
471 | bi->bi_rw, i); |
472 | atomic_inc(&sh->count); | |
473 | bi->bi_sector = sh->sector + rdev->data_offset; | |
474 | bi->bi_flags = 1 << BIO_UPTODATE; | |
475 | bi->bi_vcnt = 1; | |
476 | bi->bi_max_vecs = 1; | |
477 | bi->bi_idx = 0; | |
478 | bi->bi_io_vec = &sh->dev[i].vec; | |
479 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
480 | bi->bi_io_vec[0].bv_offset = 0; | |
481 | bi->bi_size = STRIPE_SIZE; | |
482 | bi->bi_next = NULL; | |
483 | if (rw == WRITE && | |
484 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
485 | atomic_add(STRIPE_SECTORS, | |
486 | &rdev->corrected_errors); | |
487 | generic_make_request(bi); | |
488 | } else { | |
489 | if (rw == WRITE) | |
490 | set_bit(STRIPE_DEGRADED, &sh->state); | |
491 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
492 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
493 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
494 | set_bit(STRIPE_HANDLE, &sh->state); | |
495 | } | |
496 | } | |
497 | } | |
498 | ||
499 | static struct dma_async_tx_descriptor * | |
500 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
501 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
502 | { | |
503 | struct bio_vec *bvl; | |
504 | struct page *bio_page; | |
505 | int i; | |
506 | int page_offset; | |
a08abd8c | 507 | struct async_submit_ctl submit; |
0403e382 | 508 | enum async_tx_flags flags = 0; |
91c00924 DW |
509 | |
510 | if (bio->bi_sector >= sector) | |
511 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
512 | else | |
513 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 514 | |
0403e382 DW |
515 | if (frombio) |
516 | flags |= ASYNC_TX_FENCE; | |
517 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
518 | ||
91c00924 DW |
519 | bio_for_each_segment(bvl, bio, i) { |
520 | int len = bio_iovec_idx(bio, i)->bv_len; | |
521 | int clen; | |
522 | int b_offset = 0; | |
523 | ||
524 | if (page_offset < 0) { | |
525 | b_offset = -page_offset; | |
526 | page_offset += b_offset; | |
527 | len -= b_offset; | |
528 | } | |
529 | ||
530 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
531 | clen = STRIPE_SIZE - page_offset; | |
532 | else | |
533 | clen = len; | |
534 | ||
535 | if (clen > 0) { | |
536 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
537 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
538 | if (frombio) | |
539 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 540 | b_offset, clen, &submit); |
91c00924 DW |
541 | else |
542 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 543 | page_offset, clen, &submit); |
91c00924 | 544 | } |
a08abd8c DW |
545 | /* chain the operations */ |
546 | submit.depend_tx = tx; | |
547 | ||
91c00924 DW |
548 | if (clen < len) /* hit end of page */ |
549 | break; | |
550 | page_offset += len; | |
551 | } | |
552 | ||
553 | return tx; | |
554 | } | |
555 | ||
556 | static void ops_complete_biofill(void *stripe_head_ref) | |
557 | { | |
558 | struct stripe_head *sh = stripe_head_ref; | |
559 | struct bio *return_bi = NULL; | |
560 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 561 | int i; |
91c00924 | 562 | |
e46b272b | 563 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
564 | (unsigned long long)sh->sector); |
565 | ||
566 | /* clear completed biofills */ | |
83de75cc | 567 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
568 | for (i = sh->disks; i--; ) { |
569 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
570 | |
571 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
572 | /* and check if we need to reply to a read request, |
573 | * new R5_Wantfill requests are held off until | |
83de75cc | 574 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
575 | */ |
576 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 577 | struct bio *rbi, *rbi2; |
91c00924 | 578 | |
91c00924 DW |
579 | BUG_ON(!dev->read); |
580 | rbi = dev->read; | |
581 | dev->read = NULL; | |
582 | while (rbi && rbi->bi_sector < | |
583 | dev->sector + STRIPE_SECTORS) { | |
584 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 585 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
586 | rbi->bi_next = return_bi; |
587 | return_bi = rbi; | |
588 | } | |
91c00924 DW |
589 | rbi = rbi2; |
590 | } | |
591 | } | |
592 | } | |
83de75cc DW |
593 | spin_unlock_irq(&conf->device_lock); |
594 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
595 | |
596 | return_io(return_bi); | |
597 | ||
e4d84909 | 598 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
599 | release_stripe(sh); |
600 | } | |
601 | ||
602 | static void ops_run_biofill(struct stripe_head *sh) | |
603 | { | |
604 | struct dma_async_tx_descriptor *tx = NULL; | |
605 | raid5_conf_t *conf = sh->raid_conf; | |
a08abd8c | 606 | struct async_submit_ctl submit; |
91c00924 DW |
607 | int i; |
608 | ||
e46b272b | 609 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
610 | (unsigned long long)sh->sector); |
611 | ||
612 | for (i = sh->disks; i--; ) { | |
613 | struct r5dev *dev = &sh->dev[i]; | |
614 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
615 | struct bio *rbi; | |
616 | spin_lock_irq(&conf->device_lock); | |
617 | dev->read = rbi = dev->toread; | |
618 | dev->toread = NULL; | |
619 | spin_unlock_irq(&conf->device_lock); | |
620 | while (rbi && rbi->bi_sector < | |
621 | dev->sector + STRIPE_SECTORS) { | |
622 | tx = async_copy_data(0, rbi, dev->page, | |
623 | dev->sector, tx); | |
624 | rbi = r5_next_bio(rbi, dev->sector); | |
625 | } | |
626 | } | |
627 | } | |
628 | ||
629 | atomic_inc(&sh->count); | |
a08abd8c DW |
630 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
631 | async_trigger_callback(&submit); | |
91c00924 DW |
632 | } |
633 | ||
4e7d2c0a | 634 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 635 | { |
4e7d2c0a | 636 | struct r5dev *tgt; |
91c00924 | 637 | |
4e7d2c0a DW |
638 | if (target < 0) |
639 | return; | |
91c00924 | 640 | |
4e7d2c0a | 641 | tgt = &sh->dev[target]; |
91c00924 DW |
642 | set_bit(R5_UPTODATE, &tgt->flags); |
643 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
644 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
645 | } |
646 | ||
ac6b53b6 | 647 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
648 | { |
649 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 650 | |
e46b272b | 651 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
652 | (unsigned long long)sh->sector); |
653 | ||
ac6b53b6 | 654 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 655 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 656 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 657 | |
ecc65c9b DW |
658 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
659 | if (sh->check_state == check_state_compute_run) | |
660 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
661 | set_bit(STRIPE_HANDLE, &sh->state); |
662 | release_stripe(sh); | |
663 | } | |
664 | ||
d6f38f31 DW |
665 | /* return a pointer to the address conversion region of the scribble buffer */ |
666 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
667 | struct raid5_percpu *percpu) | |
668 | { | |
669 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
670 | } | |
671 | ||
672 | static struct dma_async_tx_descriptor * | |
673 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 674 | { |
91c00924 | 675 | int disks = sh->disks; |
d6f38f31 | 676 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
677 | int target = sh->ops.target; |
678 | struct r5dev *tgt = &sh->dev[target]; | |
679 | struct page *xor_dest = tgt->page; | |
680 | int count = 0; | |
681 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 682 | struct async_submit_ctl submit; |
91c00924 DW |
683 | int i; |
684 | ||
685 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 686 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
687 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
688 | ||
689 | for (i = disks; i--; ) | |
690 | if (i != target) | |
691 | xor_srcs[count++] = sh->dev[i].page; | |
692 | ||
693 | atomic_inc(&sh->count); | |
694 | ||
0403e382 | 695 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 696 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 697 | if (unlikely(count == 1)) |
a08abd8c | 698 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 699 | else |
a08abd8c | 700 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 701 | |
91c00924 DW |
702 | return tx; |
703 | } | |
704 | ||
ac6b53b6 DW |
705 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
706 | * @srcs - (struct page *) array of size sh->disks | |
707 | * @sh - stripe_head to parse | |
708 | * | |
709 | * Populates srcs in proper layout order for the stripe and returns the | |
710 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
711 | * destination buffer is recorded in srcs[count] and the Q destination | |
712 | * is recorded in srcs[count+1]]. | |
713 | */ | |
714 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
715 | { | |
716 | int disks = sh->disks; | |
717 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
718 | int d0_idx = raid6_d0(sh); | |
719 | int count; | |
720 | int i; | |
721 | ||
722 | for (i = 0; i < disks; i++) | |
5dd33c9a | 723 | srcs[i] = NULL; |
ac6b53b6 DW |
724 | |
725 | count = 0; | |
726 | i = d0_idx; | |
727 | do { | |
728 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
729 | ||
730 | srcs[slot] = sh->dev[i].page; | |
731 | i = raid6_next_disk(i, disks); | |
732 | } while (i != d0_idx); | |
ac6b53b6 | 733 | |
e4424fee | 734 | return syndrome_disks; |
ac6b53b6 DW |
735 | } |
736 | ||
737 | static struct dma_async_tx_descriptor * | |
738 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
739 | { | |
740 | int disks = sh->disks; | |
741 | struct page **blocks = percpu->scribble; | |
742 | int target; | |
743 | int qd_idx = sh->qd_idx; | |
744 | struct dma_async_tx_descriptor *tx; | |
745 | struct async_submit_ctl submit; | |
746 | struct r5dev *tgt; | |
747 | struct page *dest; | |
748 | int i; | |
749 | int count; | |
750 | ||
751 | if (sh->ops.target < 0) | |
752 | target = sh->ops.target2; | |
753 | else if (sh->ops.target2 < 0) | |
754 | target = sh->ops.target; | |
91c00924 | 755 | else |
ac6b53b6 DW |
756 | /* we should only have one valid target */ |
757 | BUG(); | |
758 | BUG_ON(target < 0); | |
759 | pr_debug("%s: stripe %llu block: %d\n", | |
760 | __func__, (unsigned long long)sh->sector, target); | |
761 | ||
762 | tgt = &sh->dev[target]; | |
763 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
764 | dest = tgt->page; | |
765 | ||
766 | atomic_inc(&sh->count); | |
767 | ||
768 | if (target == qd_idx) { | |
769 | count = set_syndrome_sources(blocks, sh); | |
770 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
771 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
772 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
773 | ops_complete_compute, sh, | |
ac6b53b6 DW |
774 | to_addr_conv(sh, percpu)); |
775 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
776 | } else { | |
777 | /* Compute any data- or p-drive using XOR */ | |
778 | count = 0; | |
779 | for (i = disks; i-- ; ) { | |
780 | if (i == target || i == qd_idx) | |
781 | continue; | |
782 | blocks[count++] = sh->dev[i].page; | |
783 | } | |
784 | ||
0403e382 DW |
785 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
786 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
787 | to_addr_conv(sh, percpu)); |
788 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
789 | } | |
91c00924 | 790 | |
91c00924 DW |
791 | return tx; |
792 | } | |
793 | ||
ac6b53b6 DW |
794 | static struct dma_async_tx_descriptor * |
795 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
796 | { | |
797 | int i, count, disks = sh->disks; | |
798 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
799 | int d0_idx = raid6_d0(sh); | |
800 | int faila = -1, failb = -1; | |
801 | int target = sh->ops.target; | |
802 | int target2 = sh->ops.target2; | |
803 | struct r5dev *tgt = &sh->dev[target]; | |
804 | struct r5dev *tgt2 = &sh->dev[target2]; | |
805 | struct dma_async_tx_descriptor *tx; | |
806 | struct page **blocks = percpu->scribble; | |
807 | struct async_submit_ctl submit; | |
808 | ||
809 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
810 | __func__, (unsigned long long)sh->sector, target, target2); | |
811 | BUG_ON(target < 0 || target2 < 0); | |
812 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
813 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
814 | ||
6c910a78 | 815 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
816 | * slot number conversion for 'faila' and 'failb' |
817 | */ | |
818 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 819 | blocks[i] = NULL; |
ac6b53b6 DW |
820 | count = 0; |
821 | i = d0_idx; | |
822 | do { | |
823 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
824 | ||
825 | blocks[slot] = sh->dev[i].page; | |
826 | ||
827 | if (i == target) | |
828 | faila = slot; | |
829 | if (i == target2) | |
830 | failb = slot; | |
831 | i = raid6_next_disk(i, disks); | |
832 | } while (i != d0_idx); | |
ac6b53b6 DW |
833 | |
834 | BUG_ON(faila == failb); | |
835 | if (failb < faila) | |
836 | swap(faila, failb); | |
837 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
838 | __func__, (unsigned long long)sh->sector, faila, failb); | |
839 | ||
840 | atomic_inc(&sh->count); | |
841 | ||
842 | if (failb == syndrome_disks+1) { | |
843 | /* Q disk is one of the missing disks */ | |
844 | if (faila == syndrome_disks) { | |
845 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
846 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
847 | ops_complete_compute, sh, | |
848 | to_addr_conv(sh, percpu)); | |
e4424fee | 849 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
850 | STRIPE_SIZE, &submit); |
851 | } else { | |
852 | struct page *dest; | |
853 | int data_target; | |
854 | int qd_idx = sh->qd_idx; | |
855 | ||
856 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
857 | if (target == qd_idx) | |
858 | data_target = target2; | |
859 | else | |
860 | data_target = target; | |
861 | ||
862 | count = 0; | |
863 | for (i = disks; i-- ; ) { | |
864 | if (i == data_target || i == qd_idx) | |
865 | continue; | |
866 | blocks[count++] = sh->dev[i].page; | |
867 | } | |
868 | dest = sh->dev[data_target].page; | |
0403e382 DW |
869 | init_async_submit(&submit, |
870 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
871 | NULL, NULL, NULL, | |
872 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
873 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
874 | &submit); | |
875 | ||
876 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
877 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
878 | ops_complete_compute, sh, | |
879 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
880 | return async_gen_syndrome(blocks, 0, count+2, |
881 | STRIPE_SIZE, &submit); | |
882 | } | |
ac6b53b6 | 883 | } else { |
6c910a78 DW |
884 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
885 | ops_complete_compute, sh, | |
886 | to_addr_conv(sh, percpu)); | |
887 | if (failb == syndrome_disks) { | |
888 | /* We're missing D+P. */ | |
889 | return async_raid6_datap_recov(syndrome_disks+2, | |
890 | STRIPE_SIZE, faila, | |
891 | blocks, &submit); | |
892 | } else { | |
893 | /* We're missing D+D. */ | |
894 | return async_raid6_2data_recov(syndrome_disks+2, | |
895 | STRIPE_SIZE, faila, failb, | |
896 | blocks, &submit); | |
897 | } | |
ac6b53b6 DW |
898 | } |
899 | } | |
900 | ||
901 | ||
91c00924 DW |
902 | static void ops_complete_prexor(void *stripe_head_ref) |
903 | { | |
904 | struct stripe_head *sh = stripe_head_ref; | |
905 | ||
e46b272b | 906 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 907 | (unsigned long long)sh->sector); |
91c00924 DW |
908 | } |
909 | ||
910 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
911 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
912 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 913 | { |
91c00924 | 914 | int disks = sh->disks; |
d6f38f31 | 915 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 916 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 917 | struct async_submit_ctl submit; |
91c00924 DW |
918 | |
919 | /* existing parity data subtracted */ | |
920 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
921 | ||
e46b272b | 922 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
923 | (unsigned long long)sh->sector); |
924 | ||
925 | for (i = disks; i--; ) { | |
926 | struct r5dev *dev = &sh->dev[i]; | |
927 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 928 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
929 | xor_srcs[count++] = dev->page; |
930 | } | |
931 | ||
0403e382 | 932 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 933 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 934 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
935 | |
936 | return tx; | |
937 | } | |
938 | ||
939 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 940 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
941 | { |
942 | int disks = sh->disks; | |
d8ee0728 | 943 | int i; |
91c00924 | 944 | |
e46b272b | 945 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
946 | (unsigned long long)sh->sector); |
947 | ||
948 | for (i = disks; i--; ) { | |
949 | struct r5dev *dev = &sh->dev[i]; | |
950 | struct bio *chosen; | |
91c00924 | 951 | |
d8ee0728 | 952 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
953 | struct bio *wbi; |
954 | ||
955 | spin_lock(&sh->lock); | |
956 | chosen = dev->towrite; | |
957 | dev->towrite = NULL; | |
958 | BUG_ON(dev->written); | |
959 | wbi = dev->written = chosen; | |
960 | spin_unlock(&sh->lock); | |
961 | ||
962 | while (wbi && wbi->bi_sector < | |
963 | dev->sector + STRIPE_SECTORS) { | |
964 | tx = async_copy_data(1, wbi, dev->page, | |
965 | dev->sector, tx); | |
966 | wbi = r5_next_bio(wbi, dev->sector); | |
967 | } | |
968 | } | |
969 | } | |
970 | ||
971 | return tx; | |
972 | } | |
973 | ||
ac6b53b6 | 974 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
975 | { |
976 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
977 | int disks = sh->disks; |
978 | int pd_idx = sh->pd_idx; | |
979 | int qd_idx = sh->qd_idx; | |
980 | int i; | |
91c00924 | 981 | |
e46b272b | 982 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
983 | (unsigned long long)sh->sector); |
984 | ||
985 | for (i = disks; i--; ) { | |
986 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 DW |
987 | |
988 | if (dev->written || i == pd_idx || i == qd_idx) | |
91c00924 DW |
989 | set_bit(R5_UPTODATE, &dev->flags); |
990 | } | |
991 | ||
d8ee0728 DW |
992 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
993 | sh->reconstruct_state = reconstruct_state_drain_result; | |
994 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
995 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
996 | else { | |
997 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
998 | sh->reconstruct_state = reconstruct_state_result; | |
999 | } | |
91c00924 DW |
1000 | |
1001 | set_bit(STRIPE_HANDLE, &sh->state); | |
1002 | release_stripe(sh); | |
1003 | } | |
1004 | ||
1005 | static void | |
ac6b53b6 DW |
1006 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1007 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1008 | { |
91c00924 | 1009 | int disks = sh->disks; |
d6f38f31 | 1010 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1011 | struct async_submit_ctl submit; |
91c00924 DW |
1012 | int count = 0, pd_idx = sh->pd_idx, i; |
1013 | struct page *xor_dest; | |
d8ee0728 | 1014 | int prexor = 0; |
91c00924 | 1015 | unsigned long flags; |
91c00924 | 1016 | |
e46b272b | 1017 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1018 | (unsigned long long)sh->sector); |
1019 | ||
1020 | /* check if prexor is active which means only process blocks | |
1021 | * that are part of a read-modify-write (written) | |
1022 | */ | |
d8ee0728 DW |
1023 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1024 | prexor = 1; | |
91c00924 DW |
1025 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1026 | for (i = disks; i--; ) { | |
1027 | struct r5dev *dev = &sh->dev[i]; | |
1028 | if (dev->written) | |
1029 | xor_srcs[count++] = dev->page; | |
1030 | } | |
1031 | } else { | |
1032 | xor_dest = sh->dev[pd_idx].page; | |
1033 | for (i = disks; i--; ) { | |
1034 | struct r5dev *dev = &sh->dev[i]; | |
1035 | if (i != pd_idx) | |
1036 | xor_srcs[count++] = dev->page; | |
1037 | } | |
1038 | } | |
1039 | ||
91c00924 DW |
1040 | /* 1/ if we prexor'd then the dest is reused as a source |
1041 | * 2/ if we did not prexor then we are redoing the parity | |
1042 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1043 | * for the synchronous xor case | |
1044 | */ | |
88ba2aa5 | 1045 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1046 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1047 | ||
1048 | atomic_inc(&sh->count); | |
1049 | ||
ac6b53b6 | 1050 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1051 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1052 | if (unlikely(count == 1)) |
1053 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1054 | else | |
1055 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1056 | } |
1057 | ||
ac6b53b6 DW |
1058 | static void |
1059 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1060 | struct dma_async_tx_descriptor *tx) | |
1061 | { | |
1062 | struct async_submit_ctl submit; | |
1063 | struct page **blocks = percpu->scribble; | |
1064 | int count; | |
1065 | ||
1066 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1067 | ||
1068 | count = set_syndrome_sources(blocks, sh); | |
1069 | ||
1070 | atomic_inc(&sh->count); | |
1071 | ||
1072 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1073 | sh, to_addr_conv(sh, percpu)); | |
1074 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1075 | } |
1076 | ||
1077 | static void ops_complete_check(void *stripe_head_ref) | |
1078 | { | |
1079 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1080 | |
e46b272b | 1081 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1082 | (unsigned long long)sh->sector); |
1083 | ||
ecc65c9b | 1084 | sh->check_state = check_state_check_result; |
91c00924 DW |
1085 | set_bit(STRIPE_HANDLE, &sh->state); |
1086 | release_stripe(sh); | |
1087 | } | |
1088 | ||
ac6b53b6 | 1089 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1090 | { |
91c00924 | 1091 | int disks = sh->disks; |
ac6b53b6 DW |
1092 | int pd_idx = sh->pd_idx; |
1093 | int qd_idx = sh->qd_idx; | |
1094 | struct page *xor_dest; | |
d6f38f31 | 1095 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1096 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1097 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1098 | int count; |
1099 | int i; | |
91c00924 | 1100 | |
e46b272b | 1101 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1102 | (unsigned long long)sh->sector); |
1103 | ||
ac6b53b6 DW |
1104 | count = 0; |
1105 | xor_dest = sh->dev[pd_idx].page; | |
1106 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1107 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1108 | if (i == pd_idx || i == qd_idx) |
1109 | continue; | |
1110 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1111 | } |
1112 | ||
d6f38f31 DW |
1113 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1114 | to_addr_conv(sh, percpu)); | |
099f53cb | 1115 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1116 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1117 | |
91c00924 | 1118 | atomic_inc(&sh->count); |
a08abd8c DW |
1119 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1120 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1121 | } |
1122 | ||
ac6b53b6 DW |
1123 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1124 | { | |
1125 | struct page **srcs = percpu->scribble; | |
1126 | struct async_submit_ctl submit; | |
1127 | int count; | |
1128 | ||
1129 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1130 | (unsigned long long)sh->sector, checkp); | |
1131 | ||
1132 | count = set_syndrome_sources(srcs, sh); | |
1133 | if (!checkp) | |
1134 | srcs[count] = NULL; | |
91c00924 | 1135 | |
91c00924 | 1136 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1137 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1138 | sh, to_addr_conv(sh, percpu)); | |
1139 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1140 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1141 | } |
1142 | ||
417b8d4a | 1143 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1144 | { |
1145 | int overlap_clear = 0, i, disks = sh->disks; | |
1146 | struct dma_async_tx_descriptor *tx = NULL; | |
d6f38f31 | 1147 | raid5_conf_t *conf = sh->raid_conf; |
ac6b53b6 | 1148 | int level = conf->level; |
d6f38f31 DW |
1149 | struct raid5_percpu *percpu; |
1150 | unsigned long cpu; | |
91c00924 | 1151 | |
d6f38f31 DW |
1152 | cpu = get_cpu(); |
1153 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1154 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1155 | ops_run_biofill(sh); |
1156 | overlap_clear++; | |
1157 | } | |
1158 | ||
7b3a871e | 1159 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1160 | if (level < 6) |
1161 | tx = ops_run_compute5(sh, percpu); | |
1162 | else { | |
1163 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1164 | tx = ops_run_compute6_1(sh, percpu); | |
1165 | else | |
1166 | tx = ops_run_compute6_2(sh, percpu); | |
1167 | } | |
1168 | /* terminate the chain if reconstruct is not set to be run */ | |
1169 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1170 | async_tx_ack(tx); |
1171 | } | |
91c00924 | 1172 | |
600aa109 | 1173 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1174 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1175 | |
600aa109 | 1176 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1177 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1178 | overlap_clear++; |
1179 | } | |
1180 | ||
ac6b53b6 DW |
1181 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1182 | if (level < 6) | |
1183 | ops_run_reconstruct5(sh, percpu, tx); | |
1184 | else | |
1185 | ops_run_reconstruct6(sh, percpu, tx); | |
1186 | } | |
91c00924 | 1187 | |
ac6b53b6 DW |
1188 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1189 | if (sh->check_state == check_state_run) | |
1190 | ops_run_check_p(sh, percpu); | |
1191 | else if (sh->check_state == check_state_run_q) | |
1192 | ops_run_check_pq(sh, percpu, 0); | |
1193 | else if (sh->check_state == check_state_run_pq) | |
1194 | ops_run_check_pq(sh, percpu, 1); | |
1195 | else | |
1196 | BUG(); | |
1197 | } | |
91c00924 | 1198 | |
91c00924 DW |
1199 | if (overlap_clear) |
1200 | for (i = disks; i--; ) { | |
1201 | struct r5dev *dev = &sh->dev[i]; | |
1202 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1203 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1204 | } | |
d6f38f31 | 1205 | put_cpu(); |
91c00924 DW |
1206 | } |
1207 | ||
417b8d4a DW |
1208 | #ifdef CONFIG_MULTICORE_RAID456 |
1209 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1210 | { | |
1211 | struct stripe_head *sh = param; | |
1212 | unsigned long ops_request = sh->ops.request; | |
1213 | ||
1214 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1215 | wake_up(&sh->ops.wait_for_ops); | |
1216 | ||
1217 | __raid_run_ops(sh, ops_request); | |
1218 | release_stripe(sh); | |
1219 | } | |
1220 | ||
1221 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1222 | { | |
1223 | /* since handle_stripe can be called outside of raid5d context | |
1224 | * we need to ensure sh->ops.request is de-staged before another | |
1225 | * request arrives | |
1226 | */ | |
1227 | wait_event(sh->ops.wait_for_ops, | |
1228 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1229 | sh->ops.request = ops_request; | |
1230 | ||
1231 | atomic_inc(&sh->count); | |
1232 | async_schedule(async_run_ops, sh); | |
1233 | } | |
1234 | #else | |
1235 | #define raid_run_ops __raid_run_ops | |
1236 | #endif | |
1237 | ||
3f294f4f | 1238 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1239 | { |
1240 | struct stripe_head *sh; | |
5e5e3e78 | 1241 | int disks = max(conf->raid_disks, conf->previous_raid_disks); |
3f294f4f N |
1242 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
1243 | if (!sh) | |
1244 | return 0; | |
5e5e3e78 | 1245 | memset(sh, 0, sizeof(*sh) + (disks-1)*sizeof(struct r5dev)); |
3f294f4f N |
1246 | sh->raid_conf = conf; |
1247 | spin_lock_init(&sh->lock); | |
417b8d4a DW |
1248 | #ifdef CONFIG_MULTICORE_RAID456 |
1249 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1250 | #endif | |
3f294f4f | 1251 | |
5e5e3e78 N |
1252 | if (grow_buffers(sh, disks)) { |
1253 | shrink_buffers(sh, disks); | |
3f294f4f N |
1254 | kmem_cache_free(conf->slab_cache, sh); |
1255 | return 0; | |
1256 | } | |
1257 | /* we just created an active stripe so... */ | |
1258 | atomic_set(&sh->count, 1); | |
1259 | atomic_inc(&conf->active_stripes); | |
1260 | INIT_LIST_HEAD(&sh->lru); | |
1261 | release_stripe(sh); | |
1262 | return 1; | |
1263 | } | |
1264 | ||
1265 | static int grow_stripes(raid5_conf_t *conf, int num) | |
1266 | { | |
e18b890b | 1267 | struct kmem_cache *sc; |
5e5e3e78 | 1268 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1269 | |
245f46c2 N |
1270 | sprintf(conf->cache_name[0], |
1271 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1272 | sprintf(conf->cache_name[1], | |
1273 | "raid%d-%s-alt", conf->level, mdname(conf->mddev)); | |
ad01c9e3 N |
1274 | conf->active_name = 0; |
1275 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1276 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1277 | 0, 0, NULL); |
1da177e4 LT |
1278 | if (!sc) |
1279 | return 1; | |
1280 | conf->slab_cache = sc; | |
ad01c9e3 | 1281 | conf->pool_size = devs; |
16a53ecc | 1282 | while (num--) |
3f294f4f | 1283 | if (!grow_one_stripe(conf)) |
1da177e4 | 1284 | return 1; |
1da177e4 LT |
1285 | return 0; |
1286 | } | |
29269553 | 1287 | |
d6f38f31 DW |
1288 | /** |
1289 | * scribble_len - return the required size of the scribble region | |
1290 | * @num - total number of disks in the array | |
1291 | * | |
1292 | * The size must be enough to contain: | |
1293 | * 1/ a struct page pointer for each device in the array +2 | |
1294 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1295 | * (dma_map_page()) or page (page_address()) address. | |
1296 | * | |
1297 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1298 | * calculate over all devices (not just the data blocks), using zeros in place | |
1299 | * of the P and Q blocks. | |
1300 | */ | |
1301 | static size_t scribble_len(int num) | |
1302 | { | |
1303 | size_t len; | |
1304 | ||
1305 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1306 | ||
1307 | return len; | |
1308 | } | |
1309 | ||
ad01c9e3 N |
1310 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
1311 | { | |
1312 | /* Make all the stripes able to hold 'newsize' devices. | |
1313 | * New slots in each stripe get 'page' set to a new page. | |
1314 | * | |
1315 | * This happens in stages: | |
1316 | * 1/ create a new kmem_cache and allocate the required number of | |
1317 | * stripe_heads. | |
1318 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1319 | * to the new stripe_heads. This will have the side effect of | |
1320 | * freezing the array as once all stripe_heads have been collected, | |
1321 | * no IO will be possible. Old stripe heads are freed once their | |
1322 | * pages have been transferred over, and the old kmem_cache is | |
1323 | * freed when all stripes are done. | |
1324 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1325 | * we simple return a failre status - no need to clean anything up. | |
1326 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1327 | * If this fails, we don't bother trying the shrink the | |
1328 | * stripe_heads down again, we just leave them as they are. | |
1329 | * As each stripe_head is processed the new one is released into | |
1330 | * active service. | |
1331 | * | |
1332 | * Once step2 is started, we cannot afford to wait for a write, | |
1333 | * so we use GFP_NOIO allocations. | |
1334 | */ | |
1335 | struct stripe_head *osh, *nsh; | |
1336 | LIST_HEAD(newstripes); | |
1337 | struct disk_info *ndisks; | |
d6f38f31 | 1338 | unsigned long cpu; |
b5470dc5 | 1339 | int err; |
e18b890b | 1340 | struct kmem_cache *sc; |
ad01c9e3 N |
1341 | int i; |
1342 | ||
1343 | if (newsize <= conf->pool_size) | |
1344 | return 0; /* never bother to shrink */ | |
1345 | ||
b5470dc5 DW |
1346 | err = md_allow_write(conf->mddev); |
1347 | if (err) | |
1348 | return err; | |
2a2275d6 | 1349 | |
ad01c9e3 N |
1350 | /* Step 1 */ |
1351 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1352 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1353 | 0, 0, NULL); |
ad01c9e3 N |
1354 | if (!sc) |
1355 | return -ENOMEM; | |
1356 | ||
1357 | for (i = conf->max_nr_stripes; i; i--) { | |
1358 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
1359 | if (!nsh) | |
1360 | break; | |
1361 | ||
1362 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
1363 | ||
1364 | nsh->raid_conf = conf; | |
1365 | spin_lock_init(&nsh->lock); | |
417b8d4a DW |
1366 | #ifdef CONFIG_MULTICORE_RAID456 |
1367 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1368 | #endif | |
ad01c9e3 N |
1369 | |
1370 | list_add(&nsh->lru, &newstripes); | |
1371 | } | |
1372 | if (i) { | |
1373 | /* didn't get enough, give up */ | |
1374 | while (!list_empty(&newstripes)) { | |
1375 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1376 | list_del(&nsh->lru); | |
1377 | kmem_cache_free(sc, nsh); | |
1378 | } | |
1379 | kmem_cache_destroy(sc); | |
1380 | return -ENOMEM; | |
1381 | } | |
1382 | /* Step 2 - Must use GFP_NOIO now. | |
1383 | * OK, we have enough stripes, start collecting inactive | |
1384 | * stripes and copying them over | |
1385 | */ | |
1386 | list_for_each_entry(nsh, &newstripes, lru) { | |
1387 | spin_lock_irq(&conf->device_lock); | |
1388 | wait_event_lock_irq(conf->wait_for_stripe, | |
1389 | !list_empty(&conf->inactive_list), | |
1390 | conf->device_lock, | |
b3b46be3 | 1391 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1392 | ); |
1393 | osh = get_free_stripe(conf); | |
1394 | spin_unlock_irq(&conf->device_lock); | |
1395 | atomic_set(&nsh->count, 1); | |
1396 | for(i=0; i<conf->pool_size; i++) | |
1397 | nsh->dev[i].page = osh->dev[i].page; | |
1398 | for( ; i<newsize; i++) | |
1399 | nsh->dev[i].page = NULL; | |
1400 | kmem_cache_free(conf->slab_cache, osh); | |
1401 | } | |
1402 | kmem_cache_destroy(conf->slab_cache); | |
1403 | ||
1404 | /* Step 3. | |
1405 | * At this point, we are holding all the stripes so the array | |
1406 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1407 | * conf->disks and the scribble region |
ad01c9e3 N |
1408 | */ |
1409 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1410 | if (ndisks) { | |
1411 | for (i=0; i<conf->raid_disks; i++) | |
1412 | ndisks[i] = conf->disks[i]; | |
1413 | kfree(conf->disks); | |
1414 | conf->disks = ndisks; | |
1415 | } else | |
1416 | err = -ENOMEM; | |
1417 | ||
d6f38f31 DW |
1418 | get_online_cpus(); |
1419 | conf->scribble_len = scribble_len(newsize); | |
1420 | for_each_present_cpu(cpu) { | |
1421 | struct raid5_percpu *percpu; | |
1422 | void *scribble; | |
1423 | ||
1424 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1425 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1426 | ||
1427 | if (scribble) { | |
1428 | kfree(percpu->scribble); | |
1429 | percpu->scribble = scribble; | |
1430 | } else { | |
1431 | err = -ENOMEM; | |
1432 | break; | |
1433 | } | |
1434 | } | |
1435 | put_online_cpus(); | |
1436 | ||
ad01c9e3 N |
1437 | /* Step 4, return new stripes to service */ |
1438 | while(!list_empty(&newstripes)) { | |
1439 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1440 | list_del_init(&nsh->lru); | |
d6f38f31 | 1441 | |
ad01c9e3 N |
1442 | for (i=conf->raid_disks; i < newsize; i++) |
1443 | if (nsh->dev[i].page == NULL) { | |
1444 | struct page *p = alloc_page(GFP_NOIO); | |
1445 | nsh->dev[i].page = p; | |
1446 | if (!p) | |
1447 | err = -ENOMEM; | |
1448 | } | |
1449 | release_stripe(nsh); | |
1450 | } | |
1451 | /* critical section pass, GFP_NOIO no longer needed */ | |
1452 | ||
1453 | conf->slab_cache = sc; | |
1454 | conf->active_name = 1-conf->active_name; | |
1455 | conf->pool_size = newsize; | |
1456 | return err; | |
1457 | } | |
1da177e4 | 1458 | |
3f294f4f | 1459 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1460 | { |
1461 | struct stripe_head *sh; | |
1462 | ||
3f294f4f N |
1463 | spin_lock_irq(&conf->device_lock); |
1464 | sh = get_free_stripe(conf); | |
1465 | spin_unlock_irq(&conf->device_lock); | |
1466 | if (!sh) | |
1467 | return 0; | |
78bafebd | 1468 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1469 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1470 | kmem_cache_free(conf->slab_cache, sh); |
1471 | atomic_dec(&conf->active_stripes); | |
1472 | return 1; | |
1473 | } | |
1474 | ||
1475 | static void shrink_stripes(raid5_conf_t *conf) | |
1476 | { | |
1477 | while (drop_one_stripe(conf)) | |
1478 | ; | |
1479 | ||
29fc7e3e N |
1480 | if (conf->slab_cache) |
1481 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1482 | conf->slab_cache = NULL; |
1483 | } | |
1484 | ||
6712ecf8 | 1485 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1486 | { |
99c0fb5f | 1487 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1488 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1489 | int disks = sh->disks, i; |
1da177e4 | 1490 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1491 | char b[BDEVNAME_SIZE]; |
1492 | mdk_rdev_t *rdev; | |
1da177e4 | 1493 | |
1da177e4 LT |
1494 | |
1495 | for (i=0 ; i<disks; i++) | |
1496 | if (bi == &sh->dev[i].req) | |
1497 | break; | |
1498 | ||
45b4233c DW |
1499 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1500 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1501 | uptodate); |
1502 | if (i == disks) { | |
1503 | BUG(); | |
6712ecf8 | 1504 | return; |
1da177e4 LT |
1505 | } |
1506 | ||
1507 | if (uptodate) { | |
1da177e4 | 1508 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1509 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1510 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1511 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1512 | " (%lu sectors at %llu on %s)\n", | |
1513 | mdname(conf->mddev), STRIPE_SECTORS, | |
1514 | (unsigned long long)(sh->sector | |
1515 | + rdev->data_offset), | |
1516 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1517 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1518 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1519 | } | |
ba22dcbf N |
1520 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1521 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1522 | } else { |
d6950432 | 1523 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1524 | int retry = 0; |
d6950432 N |
1525 | rdev = conf->disks[i].rdev; |
1526 | ||
1da177e4 | 1527 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1528 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1529 | if (conf->mddev->degraded) |
6be9d494 BS |
1530 | printk_rl(KERN_WARNING |
1531 | "raid5:%s: read error not correctable " | |
1532 | "(sector %llu on %s).\n", | |
1533 | mdname(conf->mddev), | |
1534 | (unsigned long long)(sh->sector | |
1535 | + rdev->data_offset), | |
1536 | bdn); | |
ba22dcbf | 1537 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1538 | /* Oh, no!!! */ |
6be9d494 BS |
1539 | printk_rl(KERN_WARNING |
1540 | "raid5:%s: read error NOT corrected!! " | |
1541 | "(sector %llu on %s).\n", | |
1542 | mdname(conf->mddev), | |
1543 | (unsigned long long)(sh->sector | |
1544 | + rdev->data_offset), | |
1545 | bdn); | |
d6950432 | 1546 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1547 | > conf->max_nr_stripes) |
14f8d26b | 1548 | printk(KERN_WARNING |
d6950432 N |
1549 | "raid5:%s: Too many read errors, failing device %s.\n", |
1550 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1551 | else |
1552 | retry = 1; | |
1553 | if (retry) | |
1554 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1555 | else { | |
4e5314b5 N |
1556 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1557 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1558 | md_error(conf->mddev, rdev); |
ba22dcbf | 1559 | } |
1da177e4 LT |
1560 | } |
1561 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1562 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1563 | set_bit(STRIPE_HANDLE, &sh->state); | |
1564 | release_stripe(sh); | |
1da177e4 LT |
1565 | } |
1566 | ||
d710e138 | 1567 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1568 | { |
99c0fb5f | 1569 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1570 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1571 | int disks = sh->disks, i; |
1da177e4 LT |
1572 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1573 | ||
1da177e4 LT |
1574 | for (i=0 ; i<disks; i++) |
1575 | if (bi == &sh->dev[i].req) | |
1576 | break; | |
1577 | ||
45b4233c | 1578 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1579 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1580 | uptodate); | |
1581 | if (i == disks) { | |
1582 | BUG(); | |
6712ecf8 | 1583 | return; |
1da177e4 LT |
1584 | } |
1585 | ||
1da177e4 LT |
1586 | if (!uptodate) |
1587 | md_error(conf->mddev, conf->disks[i].rdev); | |
1588 | ||
1589 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1590 | ||
1591 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1592 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1593 | release_stripe(sh); |
1da177e4 LT |
1594 | } |
1595 | ||
1596 | ||
784052ec | 1597 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1598 | |
784052ec | 1599 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1600 | { |
1601 | struct r5dev *dev = &sh->dev[i]; | |
1602 | ||
1603 | bio_init(&dev->req); | |
1604 | dev->req.bi_io_vec = &dev->vec; | |
1605 | dev->req.bi_vcnt++; | |
1606 | dev->req.bi_max_vecs++; | |
1607 | dev->vec.bv_page = dev->page; | |
1608 | dev->vec.bv_len = STRIPE_SIZE; | |
1609 | dev->vec.bv_offset = 0; | |
1610 | ||
1611 | dev->req.bi_sector = sh->sector; | |
1612 | dev->req.bi_private = sh; | |
1613 | ||
1614 | dev->flags = 0; | |
784052ec | 1615 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1616 | } |
1617 | ||
1618 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1619 | { | |
1620 | char b[BDEVNAME_SIZE]; | |
1621 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1622 | pr_debug("raid5: error called\n"); |
1da177e4 | 1623 | |
b2d444d7 | 1624 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1625 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1626 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1627 | unsigned long flags; | |
1628 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1629 | mddev->degraded++; |
c04be0aa | 1630 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1631 | /* |
1632 | * if recovery was running, make sure it aborts. | |
1633 | */ | |
dfc70645 | 1634 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1635 | } |
b2d444d7 | 1636 | set_bit(Faulty, &rdev->flags); |
d710e138 N |
1637 | printk(KERN_ALERT |
1638 | "raid5: Disk failure on %s, disabling device.\n" | |
1639 | "raid5: Operation continuing on %d devices.\n", | |
1640 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); | |
1da177e4 | 1641 | } |
16a53ecc | 1642 | } |
1da177e4 LT |
1643 | |
1644 | /* | |
1645 | * Input: a 'big' sector number, | |
1646 | * Output: index of the data and parity disk, and the sector # in them. | |
1647 | */ | |
112bf897 | 1648 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1649 | int previous, int *dd_idx, |
1650 | struct stripe_head *sh) | |
1da177e4 LT |
1651 | { |
1652 | long stripe; | |
1653 | unsigned long chunk_number; | |
1654 | unsigned int chunk_offset; | |
911d4ee8 | 1655 | int pd_idx, qd_idx; |
67cc2b81 | 1656 | int ddf_layout = 0; |
1da177e4 | 1657 | sector_t new_sector; |
e183eaed N |
1658 | int algorithm = previous ? conf->prev_algo |
1659 | : conf->algorithm; | |
09c9e5fa AN |
1660 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1661 | : conf->chunk_sectors; | |
112bf897 N |
1662 | int raid_disks = previous ? conf->previous_raid_disks |
1663 | : conf->raid_disks; | |
1664 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1665 | |
1666 | /* First compute the information on this sector */ | |
1667 | ||
1668 | /* | |
1669 | * Compute the chunk number and the sector offset inside the chunk | |
1670 | */ | |
1671 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1672 | chunk_number = r_sector; | |
1673 | BUG_ON(r_sector != chunk_number); | |
1674 | ||
1675 | /* | |
1676 | * Compute the stripe number | |
1677 | */ | |
1678 | stripe = chunk_number / data_disks; | |
1679 | ||
1680 | /* | |
1681 | * Compute the data disk and parity disk indexes inside the stripe | |
1682 | */ | |
1683 | *dd_idx = chunk_number % data_disks; | |
1684 | ||
1685 | /* | |
1686 | * Select the parity disk based on the user selected algorithm. | |
1687 | */ | |
911d4ee8 | 1688 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1689 | switch(conf->level) { |
1690 | case 4: | |
911d4ee8 | 1691 | pd_idx = data_disks; |
16a53ecc N |
1692 | break; |
1693 | case 5: | |
e183eaed | 1694 | switch (algorithm) { |
1da177e4 | 1695 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1696 | pd_idx = data_disks - stripe % raid_disks; |
1697 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1698 | (*dd_idx)++; |
1699 | break; | |
1700 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1701 | pd_idx = stripe % raid_disks; |
1702 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1703 | (*dd_idx)++; |
1704 | break; | |
1705 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1706 | pd_idx = data_disks - stripe % raid_disks; |
1707 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 LT |
1708 | break; |
1709 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1710 | pd_idx = stripe % raid_disks; |
1711 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 | 1712 | break; |
99c0fb5f N |
1713 | case ALGORITHM_PARITY_0: |
1714 | pd_idx = 0; | |
1715 | (*dd_idx)++; | |
1716 | break; | |
1717 | case ALGORITHM_PARITY_N: | |
1718 | pd_idx = data_disks; | |
1719 | break; | |
1da177e4 | 1720 | default: |
14f8d26b | 1721 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
e183eaed | 1722 | algorithm); |
99c0fb5f | 1723 | BUG(); |
16a53ecc N |
1724 | } |
1725 | break; | |
1726 | case 6: | |
1727 | ||
e183eaed | 1728 | switch (algorithm) { |
16a53ecc | 1729 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1730 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1731 | qd_idx = pd_idx + 1; | |
1732 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1733 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1734 | qd_idx = 0; |
1735 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1736 | (*dd_idx) += 2; /* D D P Q D */ |
1737 | break; | |
1738 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1739 | pd_idx = stripe % raid_disks; |
1740 | qd_idx = pd_idx + 1; | |
1741 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1742 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1743 | qd_idx = 0; |
1744 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1745 | (*dd_idx) += 2; /* D D P Q D */ |
1746 | break; | |
1747 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1748 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1749 | qd_idx = (pd_idx + 1) % raid_disks; | |
1750 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1751 | break; |
1752 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1753 | pd_idx = stripe % raid_disks; |
1754 | qd_idx = (pd_idx + 1) % raid_disks; | |
1755 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1756 | break; |
99c0fb5f N |
1757 | |
1758 | case ALGORITHM_PARITY_0: | |
1759 | pd_idx = 0; | |
1760 | qd_idx = 1; | |
1761 | (*dd_idx) += 2; | |
1762 | break; | |
1763 | case ALGORITHM_PARITY_N: | |
1764 | pd_idx = data_disks; | |
1765 | qd_idx = data_disks + 1; | |
1766 | break; | |
1767 | ||
1768 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1769 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1770 | * of blocks for computing Q is different. | |
1771 | */ | |
1772 | pd_idx = stripe % raid_disks; | |
1773 | qd_idx = pd_idx + 1; | |
1774 | if (pd_idx == raid_disks-1) { | |
1775 | (*dd_idx)++; /* Q D D D P */ | |
1776 | qd_idx = 0; | |
1777 | } else if (*dd_idx >= pd_idx) | |
1778 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1779 | ddf_layout = 1; |
99c0fb5f N |
1780 | break; |
1781 | ||
1782 | case ALGORITHM_ROTATING_N_RESTART: | |
1783 | /* Same a left_asymmetric, by first stripe is | |
1784 | * D D D P Q rather than | |
1785 | * Q D D D P | |
1786 | */ | |
1787 | pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
1788 | qd_idx = pd_idx + 1; | |
1789 | if (pd_idx == raid_disks-1) { | |
1790 | (*dd_idx)++; /* Q D D D P */ | |
1791 | qd_idx = 0; | |
1792 | } else if (*dd_idx >= pd_idx) | |
1793 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1794 | ddf_layout = 1; |
99c0fb5f N |
1795 | break; |
1796 | ||
1797 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1798 | /* Same as left_symmetric but Q is before P */ | |
1799 | pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1800 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; | |
1801 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1802 | ddf_layout = 1; |
99c0fb5f N |
1803 | break; |
1804 | ||
1805 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1806 | /* RAID5 left_asymmetric, with Q on last device */ | |
1807 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1808 | if (*dd_idx >= pd_idx) | |
1809 | (*dd_idx)++; | |
1810 | qd_idx = raid_disks - 1; | |
1811 | break; | |
1812 | ||
1813 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1814 | pd_idx = stripe % (raid_disks-1); | |
1815 | if (*dd_idx >= pd_idx) | |
1816 | (*dd_idx)++; | |
1817 | qd_idx = raid_disks - 1; | |
1818 | break; | |
1819 | ||
1820 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1821 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1822 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1823 | qd_idx = raid_disks - 1; | |
1824 | break; | |
1825 | ||
1826 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1827 | pd_idx = stripe % (raid_disks-1); | |
1828 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1829 | qd_idx = raid_disks - 1; | |
1830 | break; | |
1831 | ||
1832 | case ALGORITHM_PARITY_0_6: | |
1833 | pd_idx = 0; | |
1834 | (*dd_idx)++; | |
1835 | qd_idx = raid_disks - 1; | |
1836 | break; | |
1837 | ||
1838 | ||
16a53ecc | 1839 | default: |
d710e138 | 1840 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
e183eaed | 1841 | algorithm); |
99c0fb5f | 1842 | BUG(); |
16a53ecc N |
1843 | } |
1844 | break; | |
1da177e4 LT |
1845 | } |
1846 | ||
911d4ee8 N |
1847 | if (sh) { |
1848 | sh->pd_idx = pd_idx; | |
1849 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1850 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1851 | } |
1da177e4 LT |
1852 | /* |
1853 | * Finally, compute the new sector number | |
1854 | */ | |
1855 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1856 | return new_sector; | |
1857 | } | |
1858 | ||
1859 | ||
784052ec | 1860 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1861 | { |
1862 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1863 | int raid_disks = sh->disks; |
1864 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1865 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
1866 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1867 | : conf->chunk_sectors; | |
e183eaed N |
1868 | int algorithm = previous ? conf->prev_algo |
1869 | : conf->algorithm; | |
1da177e4 LT |
1870 | sector_t stripe; |
1871 | int chunk_offset; | |
911d4ee8 | 1872 | int chunk_number, dummy1, dd_idx = i; |
1da177e4 | 1873 | sector_t r_sector; |
911d4ee8 | 1874 | struct stripe_head sh2; |
1da177e4 | 1875 | |
16a53ecc | 1876 | |
1da177e4 LT |
1877 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1878 | stripe = new_sector; | |
1879 | BUG_ON(new_sector != stripe); | |
1880 | ||
16a53ecc N |
1881 | if (i == sh->pd_idx) |
1882 | return 0; | |
1883 | switch(conf->level) { | |
1884 | case 4: break; | |
1885 | case 5: | |
e183eaed | 1886 | switch (algorithm) { |
1da177e4 LT |
1887 | case ALGORITHM_LEFT_ASYMMETRIC: |
1888 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1889 | if (i > sh->pd_idx) | |
1890 | i--; | |
1891 | break; | |
1892 | case ALGORITHM_LEFT_SYMMETRIC: | |
1893 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1894 | if (i < sh->pd_idx) | |
1895 | i += raid_disks; | |
1896 | i -= (sh->pd_idx + 1); | |
1897 | break; | |
99c0fb5f N |
1898 | case ALGORITHM_PARITY_0: |
1899 | i -= 1; | |
1900 | break; | |
1901 | case ALGORITHM_PARITY_N: | |
1902 | break; | |
1da177e4 | 1903 | default: |
14f8d26b | 1904 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
e183eaed | 1905 | algorithm); |
99c0fb5f | 1906 | BUG(); |
16a53ecc N |
1907 | } |
1908 | break; | |
1909 | case 6: | |
d0dabf7e | 1910 | if (i == sh->qd_idx) |
16a53ecc | 1911 | return 0; /* It is the Q disk */ |
e183eaed | 1912 | switch (algorithm) { |
16a53ecc N |
1913 | case ALGORITHM_LEFT_ASYMMETRIC: |
1914 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1915 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1916 | case ALGORITHM_ROTATING_N_RESTART: | |
1917 | if (sh->pd_idx == raid_disks-1) | |
1918 | i--; /* Q D D D P */ | |
16a53ecc N |
1919 | else if (i > sh->pd_idx) |
1920 | i -= 2; /* D D P Q D */ | |
1921 | break; | |
1922 | case ALGORITHM_LEFT_SYMMETRIC: | |
1923 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1924 | if (sh->pd_idx == raid_disks-1) | |
1925 | i--; /* Q D D D P */ | |
1926 | else { | |
1927 | /* D D P Q D */ | |
1928 | if (i < sh->pd_idx) | |
1929 | i += raid_disks; | |
1930 | i -= (sh->pd_idx + 2); | |
1931 | } | |
1932 | break; | |
99c0fb5f N |
1933 | case ALGORITHM_PARITY_0: |
1934 | i -= 2; | |
1935 | break; | |
1936 | case ALGORITHM_PARITY_N: | |
1937 | break; | |
1938 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 1939 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
1940 | if (sh->pd_idx == 0) |
1941 | i--; /* P D D D Q */ | |
e4424fee N |
1942 | else { |
1943 | /* D D Q P D */ | |
1944 | if (i < sh->pd_idx) | |
1945 | i += raid_disks; | |
1946 | i -= (sh->pd_idx + 1); | |
1947 | } | |
99c0fb5f N |
1948 | break; |
1949 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1950 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1951 | if (i > sh->pd_idx) | |
1952 | i--; | |
1953 | break; | |
1954 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1955 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1956 | if (i < sh->pd_idx) | |
1957 | i += data_disks + 1; | |
1958 | i -= (sh->pd_idx + 1); | |
1959 | break; | |
1960 | case ALGORITHM_PARITY_0_6: | |
1961 | i -= 1; | |
1962 | break; | |
16a53ecc | 1963 | default: |
d710e138 | 1964 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
e183eaed | 1965 | algorithm); |
99c0fb5f | 1966 | BUG(); |
16a53ecc N |
1967 | } |
1968 | break; | |
1da177e4 LT |
1969 | } |
1970 | ||
1971 | chunk_number = stripe * data_disks + i; | |
1972 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1973 | ||
112bf897 | 1974 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 1975 | previous, &dummy1, &sh2); |
911d4ee8 N |
1976 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
1977 | || sh2.qd_idx != sh->qd_idx) { | |
14f8d26b | 1978 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1979 | return 0; |
1980 | } | |
1981 | return r_sector; | |
1982 | } | |
1983 | ||
1984 | ||
600aa109 | 1985 | static void |
c0f7bddb | 1986 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 1987 | int rcw, int expand) |
e33129d8 DW |
1988 | { |
1989 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
c0f7bddb YT |
1990 | raid5_conf_t *conf = sh->raid_conf; |
1991 | int level = conf->level; | |
e33129d8 DW |
1992 | |
1993 | if (rcw) { | |
1994 | /* if we are not expanding this is a proper write request, and | |
1995 | * there will be bios with new data to be drained into the | |
1996 | * stripe cache | |
1997 | */ | |
1998 | if (!expand) { | |
600aa109 DW |
1999 | sh->reconstruct_state = reconstruct_state_drain_run; |
2000 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2001 | } else | |
2002 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2003 | |
ac6b53b6 | 2004 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2005 | |
2006 | for (i = disks; i--; ) { | |
2007 | struct r5dev *dev = &sh->dev[i]; | |
2008 | ||
2009 | if (dev->towrite) { | |
2010 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2011 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2012 | if (!expand) |
2013 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2014 | s->locked++; |
e33129d8 DW |
2015 | } |
2016 | } | |
c0f7bddb | 2017 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2018 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2019 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2020 | } else { |
c0f7bddb | 2021 | BUG_ON(level == 6); |
e33129d8 DW |
2022 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2023 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2024 | ||
d8ee0728 | 2025 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2026 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2027 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2028 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2029 | |
2030 | for (i = disks; i--; ) { | |
2031 | struct r5dev *dev = &sh->dev[i]; | |
2032 | if (i == pd_idx) | |
2033 | continue; | |
2034 | ||
e33129d8 DW |
2035 | if (dev->towrite && |
2036 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2037 | test_bit(R5_Wantcompute, &dev->flags))) { |
2038 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2039 | set_bit(R5_LOCKED, &dev->flags); |
2040 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2041 | s->locked++; |
e33129d8 DW |
2042 | } |
2043 | } | |
2044 | } | |
2045 | ||
c0f7bddb | 2046 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2047 | * are in flight |
2048 | */ | |
2049 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2050 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2051 | s->locked++; |
e33129d8 | 2052 | |
c0f7bddb YT |
2053 | if (level == 6) { |
2054 | int qd_idx = sh->qd_idx; | |
2055 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2056 | ||
2057 | set_bit(R5_LOCKED, &dev->flags); | |
2058 | clear_bit(R5_UPTODATE, &dev->flags); | |
2059 | s->locked++; | |
2060 | } | |
2061 | ||
600aa109 | 2062 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2063 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2064 | s->locked, s->ops_request); |
e33129d8 | 2065 | } |
16a53ecc | 2066 | |
1da177e4 LT |
2067 | /* |
2068 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2069 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2070 | * The bi_next chain must be in order. |
2071 | */ | |
2072 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2073 | { | |
2074 | struct bio **bip; | |
2075 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 2076 | int firstwrite=0; |
1da177e4 | 2077 | |
45b4233c | 2078 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2079 | (unsigned long long)bi->bi_sector, |
2080 | (unsigned long long)sh->sector); | |
2081 | ||
2082 | ||
2083 | spin_lock(&sh->lock); | |
2084 | spin_lock_irq(&conf->device_lock); | |
72626685 | 2085 | if (forwrite) { |
1da177e4 | 2086 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2087 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2088 | firstwrite = 1; | |
2089 | } else | |
1da177e4 LT |
2090 | bip = &sh->dev[dd_idx].toread; |
2091 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2092 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2093 | goto overlap; | |
2094 | bip = & (*bip)->bi_next; | |
2095 | } | |
2096 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2097 | goto overlap; | |
2098 | ||
78bafebd | 2099 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2100 | if (*bip) |
2101 | bi->bi_next = *bip; | |
2102 | *bip = bi; | |
960e739d | 2103 | bi->bi_phys_segments++; |
1da177e4 LT |
2104 | spin_unlock_irq(&conf->device_lock); |
2105 | spin_unlock(&sh->lock); | |
2106 | ||
45b4233c | 2107 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
2108 | (unsigned long long)bi->bi_sector, |
2109 | (unsigned long long)sh->sector, dd_idx); | |
2110 | ||
72626685 | 2111 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
2112 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
2113 | STRIPE_SECTORS, 0); | |
ae3c20cc | 2114 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
2115 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
2116 | } | |
2117 | ||
1da177e4 LT |
2118 | if (forwrite) { |
2119 | /* check if page is covered */ | |
2120 | sector_t sector = sh->dev[dd_idx].sector; | |
2121 | for (bi=sh->dev[dd_idx].towrite; | |
2122 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2123 | bi && bi->bi_sector <= sector; | |
2124 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2125 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2126 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2127 | } | |
2128 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2129 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2130 | } | |
2131 | return 1; | |
2132 | ||
2133 | overlap: | |
2134 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2135 | spin_unlock_irq(&conf->device_lock); | |
2136 | spin_unlock(&sh->lock); | |
2137 | return 0; | |
2138 | } | |
2139 | ||
29269553 N |
2140 | static void end_reshape(raid5_conf_t *conf); |
2141 | ||
911d4ee8 N |
2142 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
2143 | struct stripe_head *sh) | |
ccfcc3c1 | 2144 | { |
784052ec | 2145 | int sectors_per_chunk = |
09c9e5fa | 2146 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2147 | int dd_idx; |
2d2063ce | 2148 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2149 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2150 | |
112bf897 N |
2151 | raid5_compute_sector(conf, |
2152 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2153 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2154 | previous, |
911d4ee8 | 2155 | &dd_idx, sh); |
ccfcc3c1 N |
2156 | } |
2157 | ||
a4456856 | 2158 | static void |
1fe797e6 | 2159 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2160 | struct stripe_head_state *s, int disks, |
2161 | struct bio **return_bi) | |
2162 | { | |
2163 | int i; | |
2164 | for (i = disks; i--; ) { | |
2165 | struct bio *bi; | |
2166 | int bitmap_end = 0; | |
2167 | ||
2168 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2169 | mdk_rdev_t *rdev; | |
2170 | rcu_read_lock(); | |
2171 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2172 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2173 | /* multiple read failures in one stripe */ | |
2174 | md_error(conf->mddev, rdev); | |
2175 | rcu_read_unlock(); | |
2176 | } | |
2177 | spin_lock_irq(&conf->device_lock); | |
2178 | /* fail all writes first */ | |
2179 | bi = sh->dev[i].towrite; | |
2180 | sh->dev[i].towrite = NULL; | |
2181 | if (bi) { | |
2182 | s->to_write--; | |
2183 | bitmap_end = 1; | |
2184 | } | |
2185 | ||
2186 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2187 | wake_up(&conf->wait_for_overlap); | |
2188 | ||
2189 | while (bi && bi->bi_sector < | |
2190 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2191 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2192 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2193 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2194 | md_write_end(conf->mddev); |
2195 | bi->bi_next = *return_bi; | |
2196 | *return_bi = bi; | |
2197 | } | |
2198 | bi = nextbi; | |
2199 | } | |
2200 | /* and fail all 'written' */ | |
2201 | bi = sh->dev[i].written; | |
2202 | sh->dev[i].written = NULL; | |
2203 | if (bi) bitmap_end = 1; | |
2204 | while (bi && bi->bi_sector < | |
2205 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2206 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2207 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2208 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2209 | md_write_end(conf->mddev); |
2210 | bi->bi_next = *return_bi; | |
2211 | *return_bi = bi; | |
2212 | } | |
2213 | bi = bi2; | |
2214 | } | |
2215 | ||
b5e98d65 DW |
2216 | /* fail any reads if this device is non-operational and |
2217 | * the data has not reached the cache yet. | |
2218 | */ | |
2219 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2220 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2221 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2222 | bi = sh->dev[i].toread; |
2223 | sh->dev[i].toread = NULL; | |
2224 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2225 | wake_up(&conf->wait_for_overlap); | |
2226 | if (bi) s->to_read--; | |
2227 | while (bi && bi->bi_sector < | |
2228 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2229 | struct bio *nextbi = | |
2230 | r5_next_bio(bi, sh->dev[i].sector); | |
2231 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2232 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2233 | bi->bi_next = *return_bi; |
2234 | *return_bi = bi; | |
2235 | } | |
2236 | bi = nextbi; | |
2237 | } | |
2238 | } | |
2239 | spin_unlock_irq(&conf->device_lock); | |
2240 | if (bitmap_end) | |
2241 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2242 | STRIPE_SECTORS, 0, 0); | |
2243 | } | |
2244 | ||
8b3e6cdc DW |
2245 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2246 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2247 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2248 | } |
2249 | ||
1fe797e6 DW |
2250 | /* fetch_block5 - checks the given member device to see if its data needs |
2251 | * to be read or computed to satisfy a request. | |
2252 | * | |
2253 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2254 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2255 | */ |
1fe797e6 DW |
2256 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2257 | int disk_idx, int disks) | |
f38e1219 DW |
2258 | { |
2259 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2260 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2261 | ||
f38e1219 DW |
2262 | /* is the data in this block needed, and can we get it? */ |
2263 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2264 | !test_bit(R5_UPTODATE, &dev->flags) && |
2265 | (dev->toread || | |
2266 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2267 | s->syncing || s->expanding || | |
2268 | (s->failed && | |
2269 | (failed_dev->toread || | |
2270 | (failed_dev->towrite && | |
2271 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2272 | /* We would like to get this block, possibly by computing it, |
2273 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2274 | */ |
2275 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2276 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2277 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2278 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2279 | set_bit(R5_Wantcompute, &dev->flags); |
2280 | sh->ops.target = disk_idx; | |
ac6b53b6 | 2281 | sh->ops.target2 = -1; |
f38e1219 | 2282 | s->req_compute = 1; |
f38e1219 | 2283 | /* Careful: from this point on 'uptodate' is in the eye |
ac6b53b6 | 2284 | * of raid_run_ops which services 'compute' operations |
f38e1219 DW |
2285 | * before writes. R5_Wantcompute flags a block that will |
2286 | * be R5_UPTODATE by the time it is needed for a | |
2287 | * subsequent operation. | |
2288 | */ | |
2289 | s->uptodate++; | |
1fe797e6 | 2290 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2291 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2292 | set_bit(R5_LOCKED, &dev->flags); |
2293 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2294 | s->locked++; |
2295 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2296 | s->syncing); | |
2297 | } | |
2298 | } | |
2299 | ||
1fe797e6 | 2300 | return 0; |
f38e1219 DW |
2301 | } |
2302 | ||
1fe797e6 DW |
2303 | /** |
2304 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2305 | */ | |
2306 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2307 | struct stripe_head_state *s, int disks) |
2308 | { | |
2309 | int i; | |
f38e1219 | 2310 | |
f38e1219 DW |
2311 | /* look for blocks to read/compute, skip this if a compute |
2312 | * is already in flight, or if the stripe contents are in the | |
2313 | * midst of changing due to a write | |
2314 | */ | |
976ea8d4 | 2315 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2316 | !sh->reconstruct_state) |
f38e1219 | 2317 | for (i = disks; i--; ) |
1fe797e6 | 2318 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2319 | break; |
a4456856 DW |
2320 | set_bit(STRIPE_HANDLE, &sh->state); |
2321 | } | |
2322 | ||
5599becc YT |
2323 | /* fetch_block6 - checks the given member device to see if its data needs |
2324 | * to be read or computed to satisfy a request. | |
2325 | * | |
2326 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2327 | * 0 to tell the loop in handle_stripe_fill6 to continue | |
2328 | */ | |
2329 | static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s, | |
2330 | struct r6_state *r6s, int disk_idx, int disks) | |
a4456856 | 2331 | { |
5599becc YT |
2332 | struct r5dev *dev = &sh->dev[disk_idx]; |
2333 | struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]], | |
2334 | &sh->dev[r6s->failed_num[1]] }; | |
2335 | ||
2336 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2337 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2338 | (dev->toread || | |
2339 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2340 | s->syncing || s->expanding || | |
2341 | (s->failed >= 1 && | |
2342 | (fdev[0]->toread || s->to_write)) || | |
2343 | (s->failed >= 2 && | |
2344 | (fdev[1]->toread || s->to_write)))) { | |
2345 | /* we would like to get this block, possibly by computing it, | |
2346 | * otherwise read it if the backing disk is insync | |
2347 | */ | |
2348 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2349 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2350 | if ((s->uptodate == disks - 1) && | |
2351 | (s->failed && (disk_idx == r6s->failed_num[0] || | |
2352 | disk_idx == r6s->failed_num[1]))) { | |
2353 | /* have disk failed, and we're requested to fetch it; | |
2354 | * do compute it | |
a4456856 | 2355 | */ |
5599becc YT |
2356 | pr_debug("Computing stripe %llu block %d\n", |
2357 | (unsigned long long)sh->sector, disk_idx); | |
2358 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2359 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2360 | set_bit(R5_Wantcompute, &dev->flags); | |
2361 | sh->ops.target = disk_idx; | |
2362 | sh->ops.target2 = -1; /* no 2nd target */ | |
2363 | s->req_compute = 1; | |
2364 | s->uptodate++; | |
2365 | return 1; | |
2366 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2367 | /* Computing 2-failure is *very* expensive; only | |
2368 | * do it if failed >= 2 | |
2369 | */ | |
2370 | int other; | |
2371 | for (other = disks; other--; ) { | |
2372 | if (other == disk_idx) | |
2373 | continue; | |
2374 | if (!test_bit(R5_UPTODATE, | |
2375 | &sh->dev[other].flags)) | |
2376 | break; | |
a4456856 | 2377 | } |
5599becc YT |
2378 | BUG_ON(other < 0); |
2379 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2380 | (unsigned long long)sh->sector, | |
2381 | disk_idx, other); | |
2382 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2383 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2384 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2385 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2386 | sh->ops.target = disk_idx; | |
2387 | sh->ops.target2 = other; | |
2388 | s->uptodate += 2; | |
2389 | s->req_compute = 1; | |
2390 | return 1; | |
2391 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2392 | set_bit(R5_LOCKED, &dev->flags); | |
2393 | set_bit(R5_Wantread, &dev->flags); | |
2394 | s->locked++; | |
2395 | pr_debug("Reading block %d (sync=%d)\n", | |
2396 | disk_idx, s->syncing); | |
a4456856 DW |
2397 | } |
2398 | } | |
5599becc YT |
2399 | |
2400 | return 0; | |
2401 | } | |
2402 | ||
2403 | /** | |
2404 | * handle_stripe_fill6 - read or compute data to satisfy pending requests. | |
2405 | */ | |
2406 | static void handle_stripe_fill6(struct stripe_head *sh, | |
2407 | struct stripe_head_state *s, struct r6_state *r6s, | |
2408 | int disks) | |
2409 | { | |
2410 | int i; | |
2411 | ||
2412 | /* look for blocks to read/compute, skip this if a compute | |
2413 | * is already in flight, or if the stripe contents are in the | |
2414 | * midst of changing due to a write | |
2415 | */ | |
2416 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2417 | !sh->reconstruct_state) | |
2418 | for (i = disks; i--; ) | |
2419 | if (fetch_block6(sh, s, r6s, i, disks)) | |
2420 | break; | |
a4456856 DW |
2421 | set_bit(STRIPE_HANDLE, &sh->state); |
2422 | } | |
2423 | ||
2424 | ||
1fe797e6 | 2425 | /* handle_stripe_clean_event |
a4456856 DW |
2426 | * any written block on an uptodate or failed drive can be returned. |
2427 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2428 | * never LOCKED, so we don't need to test 'failed' directly. | |
2429 | */ | |
1fe797e6 | 2430 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2431 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2432 | { | |
2433 | int i; | |
2434 | struct r5dev *dev; | |
2435 | ||
2436 | for (i = disks; i--; ) | |
2437 | if (sh->dev[i].written) { | |
2438 | dev = &sh->dev[i]; | |
2439 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2440 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2441 | /* We can return any write requests */ | |
2442 | struct bio *wbi, *wbi2; | |
2443 | int bitmap_end = 0; | |
45b4233c | 2444 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2445 | spin_lock_irq(&conf->device_lock); |
2446 | wbi = dev->written; | |
2447 | dev->written = NULL; | |
2448 | while (wbi && wbi->bi_sector < | |
2449 | dev->sector + STRIPE_SECTORS) { | |
2450 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2451 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2452 | md_write_end(conf->mddev); |
2453 | wbi->bi_next = *return_bi; | |
2454 | *return_bi = wbi; | |
2455 | } | |
2456 | wbi = wbi2; | |
2457 | } | |
2458 | if (dev->towrite == NULL) | |
2459 | bitmap_end = 1; | |
2460 | spin_unlock_irq(&conf->device_lock); | |
2461 | if (bitmap_end) | |
2462 | bitmap_endwrite(conf->mddev->bitmap, | |
2463 | sh->sector, | |
2464 | STRIPE_SECTORS, | |
2465 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2466 | 0); | |
2467 | } | |
2468 | } | |
8b3e6cdc DW |
2469 | |
2470 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2471 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2472 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2473 | } |
2474 | ||
1fe797e6 | 2475 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2476 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2477 | { | |
2478 | int rmw = 0, rcw = 0, i; | |
2479 | for (i = disks; i--; ) { | |
2480 | /* would I have to read this buffer for read_modify_write */ | |
2481 | struct r5dev *dev = &sh->dev[i]; | |
2482 | if ((dev->towrite || i == sh->pd_idx) && | |
2483 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2484 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2485 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2486 | if (test_bit(R5_Insync, &dev->flags)) |
2487 | rmw++; | |
2488 | else | |
2489 | rmw += 2*disks; /* cannot read it */ | |
2490 | } | |
2491 | /* Would I have to read this buffer for reconstruct_write */ | |
2492 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2493 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2494 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2495 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2496 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2497 | else |
2498 | rcw += 2*disks; | |
2499 | } | |
2500 | } | |
45b4233c | 2501 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2502 | (unsigned long long)sh->sector, rmw, rcw); |
2503 | set_bit(STRIPE_HANDLE, &sh->state); | |
2504 | if (rmw < rcw && rmw > 0) | |
2505 | /* prefer read-modify-write, but need to get some data */ | |
2506 | for (i = disks; i--; ) { | |
2507 | struct r5dev *dev = &sh->dev[i]; | |
2508 | if ((dev->towrite || i == sh->pd_idx) && | |
2509 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2510 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2511 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2512 | test_bit(R5_Insync, &dev->flags)) { |
2513 | if ( | |
2514 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2515 | pr_debug("Read_old block " |
a4456856 DW |
2516 | "%d for r-m-w\n", i); |
2517 | set_bit(R5_LOCKED, &dev->flags); | |
2518 | set_bit(R5_Wantread, &dev->flags); | |
2519 | s->locked++; | |
2520 | } else { | |
2521 | set_bit(STRIPE_DELAYED, &sh->state); | |
2522 | set_bit(STRIPE_HANDLE, &sh->state); | |
2523 | } | |
2524 | } | |
2525 | } | |
2526 | if (rcw <= rmw && rcw > 0) | |
2527 | /* want reconstruct write, but need to get some data */ | |
2528 | for (i = disks; i--; ) { | |
2529 | struct r5dev *dev = &sh->dev[i]; | |
2530 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2531 | i != sh->pd_idx && | |
2532 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2533 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2534 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2535 | test_bit(R5_Insync, &dev->flags)) { |
2536 | if ( | |
2537 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2538 | pr_debug("Read_old block " |
a4456856 DW |
2539 | "%d for Reconstruct\n", i); |
2540 | set_bit(R5_LOCKED, &dev->flags); | |
2541 | set_bit(R5_Wantread, &dev->flags); | |
2542 | s->locked++; | |
2543 | } else { | |
2544 | set_bit(STRIPE_DELAYED, &sh->state); | |
2545 | set_bit(STRIPE_HANDLE, &sh->state); | |
2546 | } | |
2547 | } | |
2548 | } | |
2549 | /* now if nothing is locked, and if we have enough data, | |
2550 | * we can start a write request | |
2551 | */ | |
f38e1219 DW |
2552 | /* since handle_stripe can be called at any time we need to handle the |
2553 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2554 | * subsequent call wants to start a write request. raid_run_ops only |
2555 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2556 | * simultaneously. If this is not the case then new writes need to be |
2557 | * held off until the compute completes. | |
2558 | */ | |
976ea8d4 DW |
2559 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2560 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2561 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2562 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2563 | } |
2564 | ||
1fe797e6 | 2565 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2566 | struct stripe_head *sh, struct stripe_head_state *s, |
2567 | struct r6_state *r6s, int disks) | |
2568 | { | |
a9b39a74 | 2569 | int rcw = 0, pd_idx = sh->pd_idx, i; |
34e04e87 | 2570 | int qd_idx = sh->qd_idx; |
a9b39a74 YT |
2571 | |
2572 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2573 | for (i = disks; i--; ) { |
2574 | struct r5dev *dev = &sh->dev[i]; | |
a9b39a74 YT |
2575 | /* check if we haven't enough data */ |
2576 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2577 | i != pd_idx && i != qd_idx && | |
2578 | !test_bit(R5_LOCKED, &dev->flags) && | |
2579 | !(test_bit(R5_UPTODATE, &dev->flags) || | |
2580 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2581 | rcw++; | |
2582 | if (!test_bit(R5_Insync, &dev->flags)) | |
2583 | continue; /* it's a failed drive */ | |
2584 | ||
2585 | if ( | |
2586 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2587 | pr_debug("Read_old stripe %llu " | |
2588 | "block %d for Reconstruct\n", | |
2589 | (unsigned long long)sh->sector, i); | |
2590 | set_bit(R5_LOCKED, &dev->flags); | |
2591 | set_bit(R5_Wantread, &dev->flags); | |
2592 | s->locked++; | |
2593 | } else { | |
2594 | pr_debug("Request delayed stripe %llu " | |
2595 | "block %d for Reconstruct\n", | |
2596 | (unsigned long long)sh->sector, i); | |
2597 | set_bit(STRIPE_DELAYED, &sh->state); | |
2598 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2599 | } |
2600 | } | |
2601 | } | |
a4456856 DW |
2602 | /* now if nothing is locked, and if we have enough data, we can start a |
2603 | * write request | |
2604 | */ | |
a9b39a74 YT |
2605 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2606 | s->locked == 0 && rcw == 0 && | |
a4456856 | 2607 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { |
a9b39a74 | 2608 | schedule_reconstruction(sh, s, 1, 0); |
a4456856 DW |
2609 | } |
2610 | } | |
2611 | ||
2612 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2613 | struct stripe_head_state *s, int disks) | |
2614 | { | |
ecc65c9b | 2615 | struct r5dev *dev = NULL; |
bd2ab670 | 2616 | |
a4456856 | 2617 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2618 | |
ecc65c9b DW |
2619 | switch (sh->check_state) { |
2620 | case check_state_idle: | |
2621 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2622 | if (s->failed == 0) { |
bd2ab670 | 2623 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2624 | sh->check_state = check_state_run; |
2625 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2626 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2627 | s->uptodate--; |
ecc65c9b | 2628 | break; |
bd2ab670 | 2629 | } |
ecc65c9b DW |
2630 | dev = &sh->dev[s->failed_num]; |
2631 | /* fall through */ | |
2632 | case check_state_compute_result: | |
2633 | sh->check_state = check_state_idle; | |
2634 | if (!dev) | |
2635 | dev = &sh->dev[sh->pd_idx]; | |
2636 | ||
2637 | /* check that a write has not made the stripe insync */ | |
2638 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2639 | break; | |
c8894419 | 2640 | |
a4456856 | 2641 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2642 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2643 | BUG_ON(s->uptodate != disks); | |
2644 | ||
2645 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2646 | s->locked++; |
a4456856 | 2647 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2648 | |
a4456856 | 2649 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2650 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2651 | break; |
2652 | case check_state_run: | |
2653 | break; /* we will be called again upon completion */ | |
2654 | case check_state_check_result: | |
2655 | sh->check_state = check_state_idle; | |
2656 | ||
2657 | /* if a failure occurred during the check operation, leave | |
2658 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2659 | */ | |
2660 | if (s->failed) | |
2661 | break; | |
2662 | ||
2663 | /* handle a successful check operation, if parity is correct | |
2664 | * we are done. Otherwise update the mismatch count and repair | |
2665 | * parity if !MD_RECOVERY_CHECK | |
2666 | */ | |
ad283ea4 | 2667 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2668 | /* parity is correct (on disc, |
2669 | * not in buffer any more) | |
2670 | */ | |
2671 | set_bit(STRIPE_INSYNC, &sh->state); | |
2672 | else { | |
2673 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2674 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2675 | /* don't try to repair!! */ | |
2676 | set_bit(STRIPE_INSYNC, &sh->state); | |
2677 | else { | |
2678 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2679 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2680 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2681 | set_bit(R5_Wantcompute, | |
2682 | &sh->dev[sh->pd_idx].flags); | |
2683 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2684 | sh->ops.target2 = -1; |
ecc65c9b DW |
2685 | s->uptodate++; |
2686 | } | |
2687 | } | |
2688 | break; | |
2689 | case check_state_compute_run: | |
2690 | break; | |
2691 | default: | |
2692 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2693 | __func__, sh->check_state, | |
2694 | (unsigned long long) sh->sector); | |
2695 | BUG(); | |
a4456856 DW |
2696 | } |
2697 | } | |
2698 | ||
2699 | ||
2700 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
36d1c647 DW |
2701 | struct stripe_head_state *s, |
2702 | struct r6_state *r6s, int disks) | |
a4456856 | 2703 | { |
a4456856 | 2704 | int pd_idx = sh->pd_idx; |
34e04e87 | 2705 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2706 | struct r5dev *dev; |
a4456856 DW |
2707 | |
2708 | set_bit(STRIPE_HANDLE, &sh->state); | |
2709 | ||
2710 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2711 | |
a4456856 DW |
2712 | /* Want to check and possibly repair P and Q. |
2713 | * However there could be one 'failed' device, in which | |
2714 | * case we can only check one of them, possibly using the | |
2715 | * other to generate missing data | |
2716 | */ | |
2717 | ||
d82dfee0 DW |
2718 | switch (sh->check_state) { |
2719 | case check_state_idle: | |
2720 | /* start a new check operation if there are < 2 failures */ | |
a4456856 | 2721 | if (s->failed == r6s->q_failed) { |
d82dfee0 | 2722 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2723 | * makes sense to check P (If anything else were failed, |
2724 | * we would have used P to recreate it). | |
2725 | */ | |
d82dfee0 | 2726 | sh->check_state = check_state_run; |
a4456856 DW |
2727 | } |
2728 | if (!r6s->q_failed && s->failed < 2) { | |
d82dfee0 | 2729 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2730 | * anything, so it makes sense to check it |
2731 | */ | |
d82dfee0 DW |
2732 | if (sh->check_state == check_state_run) |
2733 | sh->check_state = check_state_run_pq; | |
2734 | else | |
2735 | sh->check_state = check_state_run_q; | |
a4456856 | 2736 | } |
a4456856 | 2737 | |
d82dfee0 DW |
2738 | /* discard potentially stale zero_sum_result */ |
2739 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2740 | |
d82dfee0 DW |
2741 | if (sh->check_state == check_state_run) { |
2742 | /* async_xor_zero_sum destroys the contents of P */ | |
2743 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2744 | s->uptodate--; | |
a4456856 | 2745 | } |
d82dfee0 DW |
2746 | if (sh->check_state >= check_state_run && |
2747 | sh->check_state <= check_state_run_pq) { | |
2748 | /* async_syndrome_zero_sum preserves P and Q, so | |
2749 | * no need to mark them !uptodate here | |
2750 | */ | |
2751 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2752 | break; | |
a4456856 DW |
2753 | } |
2754 | ||
d82dfee0 DW |
2755 | /* we have 2-disk failure */ |
2756 | BUG_ON(s->failed != 2); | |
2757 | /* fall through */ | |
2758 | case check_state_compute_result: | |
2759 | sh->check_state = check_state_idle; | |
a4456856 | 2760 | |
d82dfee0 DW |
2761 | /* check that a write has not made the stripe insync */ |
2762 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2763 | break; | |
a4456856 DW |
2764 | |
2765 | /* now write out any block on a failed drive, | |
d82dfee0 | 2766 | * or P or Q if they were recomputed |
a4456856 | 2767 | */ |
d82dfee0 | 2768 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 DW |
2769 | if (s->failed == 2) { |
2770 | dev = &sh->dev[r6s->failed_num[1]]; | |
2771 | s->locked++; | |
2772 | set_bit(R5_LOCKED, &dev->flags); | |
2773 | set_bit(R5_Wantwrite, &dev->flags); | |
2774 | } | |
2775 | if (s->failed >= 1) { | |
2776 | dev = &sh->dev[r6s->failed_num[0]]; | |
2777 | s->locked++; | |
2778 | set_bit(R5_LOCKED, &dev->flags); | |
2779 | set_bit(R5_Wantwrite, &dev->flags); | |
2780 | } | |
d82dfee0 | 2781 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2782 | dev = &sh->dev[pd_idx]; |
2783 | s->locked++; | |
2784 | set_bit(R5_LOCKED, &dev->flags); | |
2785 | set_bit(R5_Wantwrite, &dev->flags); | |
2786 | } | |
d82dfee0 | 2787 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2788 | dev = &sh->dev[qd_idx]; |
2789 | s->locked++; | |
2790 | set_bit(R5_LOCKED, &dev->flags); | |
2791 | set_bit(R5_Wantwrite, &dev->flags); | |
2792 | } | |
2793 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2794 | ||
2795 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2796 | break; |
2797 | case check_state_run: | |
2798 | case check_state_run_q: | |
2799 | case check_state_run_pq: | |
2800 | break; /* we will be called again upon completion */ | |
2801 | case check_state_check_result: | |
2802 | sh->check_state = check_state_idle; | |
2803 | ||
2804 | /* handle a successful check operation, if parity is correct | |
2805 | * we are done. Otherwise update the mismatch count and repair | |
2806 | * parity if !MD_RECOVERY_CHECK | |
2807 | */ | |
2808 | if (sh->ops.zero_sum_result == 0) { | |
2809 | /* both parities are correct */ | |
2810 | if (!s->failed) | |
2811 | set_bit(STRIPE_INSYNC, &sh->state); | |
2812 | else { | |
2813 | /* in contrast to the raid5 case we can validate | |
2814 | * parity, but still have a failure to write | |
2815 | * back | |
2816 | */ | |
2817 | sh->check_state = check_state_compute_result; | |
2818 | /* Returning at this point means that we may go | |
2819 | * off and bring p and/or q uptodate again so | |
2820 | * we make sure to check zero_sum_result again | |
2821 | * to verify if p or q need writeback | |
2822 | */ | |
2823 | } | |
2824 | } else { | |
2825 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2826 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2827 | /* don't try to repair!! */ | |
2828 | set_bit(STRIPE_INSYNC, &sh->state); | |
2829 | else { | |
2830 | int *target = &sh->ops.target; | |
2831 | ||
2832 | sh->ops.target = -1; | |
2833 | sh->ops.target2 = -1; | |
2834 | sh->check_state = check_state_compute_run; | |
2835 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2836 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2837 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2838 | set_bit(R5_Wantcompute, | |
2839 | &sh->dev[pd_idx].flags); | |
2840 | *target = pd_idx; | |
2841 | target = &sh->ops.target2; | |
2842 | s->uptodate++; | |
2843 | } | |
2844 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2845 | set_bit(R5_Wantcompute, | |
2846 | &sh->dev[qd_idx].flags); | |
2847 | *target = qd_idx; | |
2848 | s->uptodate++; | |
2849 | } | |
2850 | } | |
2851 | } | |
2852 | break; | |
2853 | case check_state_compute_run: | |
2854 | break; | |
2855 | default: | |
2856 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2857 | __func__, sh->check_state, | |
2858 | (unsigned long long) sh->sector); | |
2859 | BUG(); | |
a4456856 DW |
2860 | } |
2861 | } | |
2862 | ||
2863 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2864 | struct r6_state *r6s) | |
2865 | { | |
2866 | int i; | |
2867 | ||
2868 | /* We have read all the blocks in this stripe and now we need to | |
2869 | * copy some of them into a target stripe for expand. | |
2870 | */ | |
f0a50d37 | 2871 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2872 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2873 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 2874 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 2875 | int dd_idx, j; |
a4456856 | 2876 | struct stripe_head *sh2; |
a08abd8c | 2877 | struct async_submit_ctl submit; |
a4456856 | 2878 | |
784052ec | 2879 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
2880 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2881 | &dd_idx, NULL); | |
a8c906ca | 2882 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
2883 | if (sh2 == NULL) |
2884 | /* so far only the early blocks of this stripe | |
2885 | * have been requested. When later blocks | |
2886 | * get requested, we will try again | |
2887 | */ | |
2888 | continue; | |
2889 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2890 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2891 | /* must have already done this block */ | |
2892 | release_stripe(sh2); | |
2893 | continue; | |
2894 | } | |
f0a50d37 DW |
2895 | |
2896 | /* place all the copies on one channel */ | |
a08abd8c | 2897 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 2898 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 2899 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 2900 | &submit); |
f0a50d37 | 2901 | |
a4456856 DW |
2902 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2903 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2904 | for (j = 0; j < conf->raid_disks; j++) | |
2905 | if (j != sh2->pd_idx && | |
d0dabf7e | 2906 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
2907 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2908 | break; | |
2909 | if (j == conf->raid_disks) { | |
2910 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2911 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2912 | } | |
2913 | release_stripe(sh2); | |
f0a50d37 | 2914 | |
a4456856 | 2915 | } |
a2e08551 N |
2916 | /* done submitting copies, wait for them to complete */ |
2917 | if (tx) { | |
2918 | async_tx_ack(tx); | |
2919 | dma_wait_for_async_tx(tx); | |
2920 | } | |
a4456856 | 2921 | } |
1da177e4 | 2922 | |
6bfe0b49 | 2923 | |
1da177e4 LT |
2924 | /* |
2925 | * handle_stripe - do things to a stripe. | |
2926 | * | |
2927 | * We lock the stripe and then examine the state of various bits | |
2928 | * to see what needs to be done. | |
2929 | * Possible results: | |
2930 | * return some read request which now have data | |
2931 | * return some write requests which are safely on disc | |
2932 | * schedule a read on some buffers | |
2933 | * schedule a write of some buffers | |
2934 | * return confirmation of parity correctness | |
2935 | * | |
1da177e4 LT |
2936 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2937 | * get BH_Lock set before the stripe lock is released. | |
2938 | * | |
2939 | */ | |
a4456856 | 2940 | |
1442577b | 2941 | static void handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2942 | { |
2943 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2944 | int disks = sh->disks, i; |
2945 | struct bio *return_bi = NULL; | |
2946 | struct stripe_head_state s; | |
1da177e4 | 2947 | struct r5dev *dev; |
6bfe0b49 | 2948 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2949 | int prexor; |
729a1866 | 2950 | int dec_preread_active = 0; |
1da177e4 | 2951 | |
a4456856 | 2952 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
2953 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
2954 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
2955 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
2956 | sh->reconstruct_state); | |
1da177e4 LT |
2957 | |
2958 | spin_lock(&sh->lock); | |
2959 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2960 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2961 | ||
a4456856 DW |
2962 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2963 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2964 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 2965 | |
83de75cc | 2966 | /* Now to look around and see what can be done */ |
9910f16a | 2967 | rcu_read_lock(); |
1da177e4 LT |
2968 | for (i=disks; i--; ) { |
2969 | mdk_rdev_t *rdev; | |
a9f326eb N |
2970 | |
2971 | dev = &sh->dev[i]; | |
1da177e4 | 2972 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2973 | |
b5e98d65 DW |
2974 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2975 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2976 | dev->towrite, dev->written); | |
2977 | ||
2978 | /* maybe we can request a biofill operation | |
2979 | * | |
2980 | * new wantfill requests are only permitted while | |
83de75cc | 2981 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
2982 | */ |
2983 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 2984 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 2985 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
2986 | |
2987 | /* now count some things */ | |
a4456856 DW |
2988 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2989 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2990 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2991 | |
b5e98d65 DW |
2992 | if (test_bit(R5_Wantfill, &dev->flags)) |
2993 | s.to_fill++; | |
2994 | else if (dev->toread) | |
a4456856 | 2995 | s.to_read++; |
1da177e4 | 2996 | if (dev->towrite) { |
a4456856 | 2997 | s.to_write++; |
1da177e4 | 2998 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2999 | s.non_overwrite++; |
1da177e4 | 3000 | } |
a4456856 DW |
3001 | if (dev->written) |
3002 | s.written++; | |
9910f16a | 3003 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3004 | if (blocked_rdev == NULL && |
3005 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3006 | blocked_rdev = rdev; |
3007 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3008 | } |
b2d444d7 | 3009 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 3010 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
3011 | clear_bit(R5_ReadError, &dev->flags); |
3012 | clear_bit(R5_ReWrite, &dev->flags); | |
3013 | } | |
b2d444d7 | 3014 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 3015 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
3016 | s.failed++; |
3017 | s.failed_num = i; | |
1da177e4 LT |
3018 | } else |
3019 | set_bit(R5_Insync, &dev->flags); | |
3020 | } | |
9910f16a | 3021 | rcu_read_unlock(); |
b5e98d65 | 3022 | |
6bfe0b49 | 3023 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
3024 | if (s.syncing || s.expanding || s.expanded || |
3025 | s.to_write || s.written) { | |
3026 | set_bit(STRIPE_HANDLE, &sh->state); | |
3027 | goto unlock; | |
3028 | } | |
3029 | /* There is nothing for the blocked_rdev to block */ | |
3030 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3031 | blocked_rdev = NULL; | |
6bfe0b49 DW |
3032 | } |
3033 | ||
83de75cc DW |
3034 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3035 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3036 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3037 | } | |
b5e98d65 | 3038 | |
45b4233c | 3039 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 3040 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
3041 | s.locked, s.uptodate, s.to_read, s.to_write, |
3042 | s.failed, s.failed_num); | |
1da177e4 LT |
3043 | /* check if the array has lost two devices and, if so, some requests might |
3044 | * need to be failed | |
3045 | */ | |
a4456856 | 3046 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3047 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3048 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
3049 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3050 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3051 | s.syncing = 0; |
1da177e4 LT |
3052 | } |
3053 | ||
3054 | /* might be able to return some write requests if the parity block | |
3055 | * is safe, or on a failed drive | |
3056 | */ | |
3057 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
3058 | if ( s.written && |
3059 | ((test_bit(R5_Insync, &dev->flags) && | |
3060 | !test_bit(R5_LOCKED, &dev->flags) && | |
3061 | test_bit(R5_UPTODATE, &dev->flags)) || | |
3062 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 3063 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
3064 | |
3065 | /* Now we might consider reading some blocks, either to check/generate | |
3066 | * parity, or to satisfy requests | |
3067 | * or to load a block that is being partially written. | |
3068 | */ | |
a4456856 | 3069 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 3070 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3071 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 3072 | |
e33129d8 DW |
3073 | /* Now we check to see if any write operations have recently |
3074 | * completed | |
3075 | */ | |
e0a115e5 | 3076 | prexor = 0; |
d8ee0728 | 3077 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 3078 | prexor = 1; |
d8ee0728 DW |
3079 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
3080 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 3081 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
3082 | |
3083 | /* All the 'written' buffers and the parity block are ready to | |
3084 | * be written back to disk | |
3085 | */ | |
3086 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3087 | for (i = disks; i--; ) { | |
3088 | dev = &sh->dev[i]; | |
3089 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3090 | (i == sh->pd_idx || dev->written)) { | |
3091 | pr_debug("Writing block %d\n", i); | |
3092 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
3093 | if (prexor) |
3094 | continue; | |
e33129d8 DW |
3095 | if (!test_bit(R5_Insync, &dev->flags) || |
3096 | (i == sh->pd_idx && s.failed == 0)) | |
3097 | set_bit(STRIPE_INSYNC, &sh->state); | |
3098 | } | |
3099 | } | |
729a1866 N |
3100 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3101 | dec_preread_active = 1; | |
e33129d8 DW |
3102 | } |
3103 | ||
3104 | /* Now to consider new write requests and what else, if anything | |
3105 | * should be read. We do not handle new writes when: | |
3106 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3107 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3108 | * block. | |
3109 | */ | |
600aa109 | 3110 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 3111 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
3112 | |
3113 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
3114 | * Any reads will already have been scheduled, so we just see if enough |
3115 | * data is available. The parity check is held off while parity | |
3116 | * dependent operations are in flight. | |
1da177e4 | 3117 | */ |
ecc65c9b DW |
3118 | if (sh->check_state || |
3119 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 3120 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 3121 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 3122 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 3123 | |
a4456856 | 3124 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
3125 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3126 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3127 | } | |
4e5314b5 N |
3128 | |
3129 | /* If the failed drive is just a ReadError, then we might need to progress | |
3130 | * the repair/check process | |
3131 | */ | |
a4456856 DW |
3132 | if (s.failed == 1 && !conf->mddev->ro && |
3133 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
3134 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
3135 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 3136 | ) { |
a4456856 | 3137 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
3138 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
3139 | set_bit(R5_Wantwrite, &dev->flags); | |
3140 | set_bit(R5_ReWrite, &dev->flags); | |
3141 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3142 | s.locked++; |
4e5314b5 N |
3143 | } else { |
3144 | /* let's read it back */ | |
3145 | set_bit(R5_Wantread, &dev->flags); | |
3146 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3147 | s.locked++; |
4e5314b5 N |
3148 | } |
3149 | } | |
3150 | ||
600aa109 DW |
3151 | /* Finish reconstruct operations initiated by the expansion process */ |
3152 | if (sh->reconstruct_state == reconstruct_state_result) { | |
ab69ae12 | 3153 | struct stripe_head *sh2 |
a8c906ca | 3154 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3155 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3156 | /* sh cannot be written until sh2 has been read. | |
3157 | * so arrange for sh to be delayed a little | |
3158 | */ | |
3159 | set_bit(STRIPE_DELAYED, &sh->state); | |
3160 | set_bit(STRIPE_HANDLE, &sh->state); | |
3161 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3162 | &sh2->state)) | |
3163 | atomic_inc(&conf->preread_active_stripes); | |
3164 | release_stripe(sh2); | |
3165 | goto unlock; | |
3166 | } | |
3167 | if (sh2) | |
3168 | release_stripe(sh2); | |
3169 | ||
600aa109 | 3170 | sh->reconstruct_state = reconstruct_state_idle; |
f0a50d37 | 3171 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 3172 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 3173 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 3174 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 3175 | s.locked++; |
23397883 | 3176 | } |
f0a50d37 DW |
3177 | } |
3178 | ||
3179 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 3180 | !sh->reconstruct_state) { |
f0a50d37 DW |
3181 | /* Need to write out all blocks after computing parity */ |
3182 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3183 | stripe_set_idx(sh->sector, conf, 0, sh); |
c0f7bddb | 3184 | schedule_reconstruction(sh, &s, 1, 1); |
600aa109 | 3185 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 3186 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 3187 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
3188 | wake_up(&conf->wait_for_overlap); |
3189 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3190 | } | |
3191 | ||
0f94e87c | 3192 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3193 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3194 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 3195 | |
6bfe0b49 | 3196 | unlock: |
1da177e4 LT |
3197 | spin_unlock(&sh->lock); |
3198 | ||
6bfe0b49 DW |
3199 | /* wait for this device to become unblocked */ |
3200 | if (unlikely(blocked_rdev)) | |
3201 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3202 | ||
600aa109 | 3203 | if (s.ops_request) |
ac6b53b6 | 3204 | raid_run_ops(sh, s.ops_request); |
d84e0f10 | 3205 | |
c4e5ac0a | 3206 | ops_run_io(sh, &s); |
1da177e4 | 3207 | |
729a1866 N |
3208 | if (dec_preread_active) { |
3209 | /* We delay this until after ops_run_io so that if make_request | |
3210 | * is waiting on a barrier, it won't continue until the writes | |
3211 | * have actually been submitted. | |
3212 | */ | |
3213 | atomic_dec(&conf->preread_active_stripes); | |
3214 | if (atomic_read(&conf->preread_active_stripes) < | |
3215 | IO_THRESHOLD) | |
3216 | md_wakeup_thread(conf->mddev->thread); | |
3217 | } | |
a4456856 | 3218 | return_io(return_bi); |
1da177e4 LT |
3219 | } |
3220 | ||
1442577b | 3221 | static void handle_stripe6(struct stripe_head *sh) |
1da177e4 | 3222 | { |
bff61975 | 3223 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 3224 | int disks = sh->disks; |
a4456856 | 3225 | struct bio *return_bi = NULL; |
34e04e87 | 3226 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx; |
a4456856 DW |
3227 | struct stripe_head_state s; |
3228 | struct r6_state r6s; | |
16a53ecc | 3229 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 3230 | mdk_rdev_t *blocked_rdev = NULL; |
729a1866 | 3231 | int dec_preread_active = 0; |
1da177e4 | 3232 | |
45b4233c | 3233 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
6c0069c0 | 3234 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", |
a4456856 | 3235 | (unsigned long long)sh->sector, sh->state, |
6c0069c0 YT |
3236 | atomic_read(&sh->count), pd_idx, qd_idx, |
3237 | sh->check_state, sh->reconstruct_state); | |
a4456856 | 3238 | memset(&s, 0, sizeof(s)); |
72626685 | 3239 | |
16a53ecc N |
3240 | spin_lock(&sh->lock); |
3241 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3242 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3243 | ||
a4456856 DW |
3244 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3245 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3246 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 3247 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3248 | |
3249 | rcu_read_lock(); | |
16a53ecc N |
3250 | for (i=disks; i--; ) { |
3251 | mdk_rdev_t *rdev; | |
3252 | dev = &sh->dev[i]; | |
3253 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 3254 | |
45b4233c | 3255 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc | 3256 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
6c0069c0 YT |
3257 | /* maybe we can reply to a read |
3258 | * | |
3259 | * new wantfill requests are only permitted while | |
3260 | * ops_complete_biofill is guaranteed to be inactive | |
3261 | */ | |
3262 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3263 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3264 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3265 | |
16a53ecc | 3266 | /* now count some things */ |
a4456856 DW |
3267 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3268 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
2d6e4ecc DW |
3269 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
3270 | s.compute++; | |
3271 | BUG_ON(s.compute > 2); | |
3272 | } | |
1da177e4 | 3273 | |
6c0069c0 YT |
3274 | if (test_bit(R5_Wantfill, &dev->flags)) { |
3275 | s.to_fill++; | |
3276 | } else if (dev->toread) | |
a4456856 | 3277 | s.to_read++; |
16a53ecc | 3278 | if (dev->towrite) { |
a4456856 | 3279 | s.to_write++; |
16a53ecc | 3280 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3281 | s.non_overwrite++; |
16a53ecc | 3282 | } |
a4456856 DW |
3283 | if (dev->written) |
3284 | s.written++; | |
16a53ecc | 3285 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3286 | if (blocked_rdev == NULL && |
3287 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3288 | blocked_rdev = rdev; |
3289 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3290 | } |
16a53ecc N |
3291 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3292 | /* The ReadError flag will just be confusing now */ | |
3293 | clear_bit(R5_ReadError, &dev->flags); | |
3294 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3295 | } |
16a53ecc N |
3296 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3297 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3298 | if (s.failed < 2) |
3299 | r6s.failed_num[s.failed] = i; | |
3300 | s.failed++; | |
16a53ecc N |
3301 | } else |
3302 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3303 | } |
3304 | rcu_read_unlock(); | |
6bfe0b49 DW |
3305 | |
3306 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3307 | if (s.syncing || s.expanding || s.expanded || |
3308 | s.to_write || s.written) { | |
3309 | set_bit(STRIPE_HANDLE, &sh->state); | |
3310 | goto unlock; | |
3311 | } | |
3312 | /* There is nothing for the blocked_rdev to block */ | |
3313 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3314 | blocked_rdev = NULL; | |
6bfe0b49 | 3315 | } |
ac4090d2 | 3316 | |
6c0069c0 YT |
3317 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3318 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3319 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3320 | } | |
3321 | ||
45b4233c | 3322 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3323 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3324 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3325 | r6s.failed_num[0], r6s.failed_num[1]); | |
3326 | /* check if the array has lost >2 devices and, if so, some requests | |
3327 | * might need to be failed | |
16a53ecc | 3328 | */ |
a4456856 | 3329 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3330 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3331 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3332 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3333 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3334 | s.syncing = 0; |
16a53ecc N |
3335 | } |
3336 | ||
3337 | /* | |
3338 | * might be able to return some write requests if the parity blocks | |
3339 | * are safe, or on a failed drive | |
3340 | */ | |
3341 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3342 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3343 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
34e04e87 N |
3344 | qdev = &sh->dev[qd_idx]; |
3345 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx) | |
3346 | || (s.failed >= 2 && r6s.failed_num[1] == qd_idx); | |
a4456856 DW |
3347 | |
3348 | if ( s.written && | |
3349 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3350 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3351 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3352 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3353 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3354 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3355 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3356 | |
3357 | /* Now we might consider reading some blocks, either to check/generate | |
3358 | * parity, or to satisfy requests | |
3359 | * or to load a block that is being partially written. | |
3360 | */ | |
a4456856 | 3361 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
6c0069c0 | 3362 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3363 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc | 3364 | |
6c0069c0 YT |
3365 | /* Now we check to see if any write operations have recently |
3366 | * completed | |
3367 | */ | |
3368 | if (sh->reconstruct_state == reconstruct_state_drain_result) { | |
6c0069c0 YT |
3369 | |
3370 | sh->reconstruct_state = reconstruct_state_idle; | |
3371 | /* All the 'written' buffers and the parity blocks are ready to | |
3372 | * be written back to disk | |
3373 | */ | |
3374 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3375 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags)); | |
3376 | for (i = disks; i--; ) { | |
3377 | dev = &sh->dev[i]; | |
3378 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3379 | (i == sh->pd_idx || i == qd_idx || | |
3380 | dev->written)) { | |
3381 | pr_debug("Writing block %d\n", i); | |
3382 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); | |
3383 | set_bit(R5_Wantwrite, &dev->flags); | |
3384 | if (!test_bit(R5_Insync, &dev->flags) || | |
3385 | ((i == sh->pd_idx || i == qd_idx) && | |
3386 | s.failed == 0)) | |
3387 | set_bit(STRIPE_INSYNC, &sh->state); | |
3388 | } | |
3389 | } | |
729a1866 N |
3390 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3391 | dec_preread_active = 1; | |
6c0069c0 YT |
3392 | } |
3393 | ||
a9b39a74 YT |
3394 | /* Now to consider new write requests and what else, if anything |
3395 | * should be read. We do not handle new writes when: | |
3396 | * 1/ A 'write' operation (copy+gen_syndrome) is already in flight. | |
3397 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3398 | * block. | |
3399 | */ | |
3400 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
1fe797e6 | 3401 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3402 | |
3403 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 | 3404 | * Any reads will already have been scheduled, so we just see if enough |
6c0069c0 YT |
3405 | * data is available. The parity check is held off while parity |
3406 | * dependent operations are in flight. | |
16a53ecc | 3407 | */ |
6c0069c0 YT |
3408 | if (sh->check_state || |
3409 | (s.syncing && s.locked == 0 && | |
3410 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3411 | !test_bit(STRIPE_INSYNC, &sh->state))) | |
36d1c647 | 3412 | handle_parity_checks6(conf, sh, &s, &r6s, disks); |
16a53ecc | 3413 | |
a4456856 | 3414 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3415 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3416 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3417 | } | |
3418 | ||
3419 | /* If the failed drives are just a ReadError, then we might need | |
3420 | * to progress the repair/check process | |
3421 | */ | |
a4456856 DW |
3422 | if (s.failed <= 2 && !conf->mddev->ro) |
3423 | for (i = 0; i < s.failed; i++) { | |
3424 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3425 | if (test_bit(R5_ReadError, &dev->flags) |
3426 | && !test_bit(R5_LOCKED, &dev->flags) | |
3427 | && test_bit(R5_UPTODATE, &dev->flags) | |
3428 | ) { | |
3429 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3430 | set_bit(R5_Wantwrite, &dev->flags); | |
3431 | set_bit(R5_ReWrite, &dev->flags); | |
3432 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3433 | s.locked++; |
16a53ecc N |
3434 | } else { |
3435 | /* let's read it back */ | |
3436 | set_bit(R5_Wantread, &dev->flags); | |
3437 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3438 | s.locked++; |
16a53ecc N |
3439 | } |
3440 | } | |
3441 | } | |
f416885e | 3442 | |
6c0069c0 YT |
3443 | /* Finish reconstruct operations initiated by the expansion process */ |
3444 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3445 | sh->reconstruct_state = reconstruct_state_idle; | |
3446 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3447 | for (i = conf->raid_disks; i--; ) { | |
3448 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3449 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3450 | s.locked++; | |
3451 | } | |
3452 | } | |
3453 | ||
3454 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
3455 | !sh->reconstruct_state) { | |
ab69ae12 | 3456 | struct stripe_head *sh2 |
a8c906ca | 3457 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3458 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3459 | /* sh cannot be written until sh2 has been read. | |
3460 | * so arrange for sh to be delayed a little | |
3461 | */ | |
3462 | set_bit(STRIPE_DELAYED, &sh->state); | |
3463 | set_bit(STRIPE_HANDLE, &sh->state); | |
3464 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3465 | &sh2->state)) | |
3466 | atomic_inc(&conf->preread_active_stripes); | |
3467 | release_stripe(sh2); | |
3468 | goto unlock; | |
3469 | } | |
3470 | if (sh2) | |
3471 | release_stripe(sh2); | |
3472 | ||
f416885e N |
3473 | /* Need to write out all blocks after computing P&Q */ |
3474 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3475 | stripe_set_idx(sh->sector, conf, 0, sh); |
6c0069c0 YT |
3476 | schedule_reconstruction(sh, &s, 1, 1); |
3477 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
f416885e N |
3478 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3479 | atomic_dec(&conf->reshape_stripes); | |
3480 | wake_up(&conf->wait_for_overlap); | |
3481 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3482 | } | |
3483 | ||
0f94e87c | 3484 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3485 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3486 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3487 | |
6bfe0b49 | 3488 | unlock: |
16a53ecc N |
3489 | spin_unlock(&sh->lock); |
3490 | ||
6bfe0b49 DW |
3491 | /* wait for this device to become unblocked */ |
3492 | if (unlikely(blocked_rdev)) | |
3493 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3494 | ||
6c0069c0 YT |
3495 | if (s.ops_request) |
3496 | raid_run_ops(sh, s.ops_request); | |
3497 | ||
f0e43bcd | 3498 | ops_run_io(sh, &s); |
16a53ecc | 3499 | |
729a1866 N |
3500 | |
3501 | if (dec_preread_active) { | |
3502 | /* We delay this until after ops_run_io so that if make_request | |
3503 | * is waiting on a barrier, it won't continue until the writes | |
3504 | * have actually been submitted. | |
3505 | */ | |
3506 | atomic_dec(&conf->preread_active_stripes); | |
3507 | if (atomic_read(&conf->preread_active_stripes) < | |
3508 | IO_THRESHOLD) | |
3509 | md_wakeup_thread(conf->mddev->thread); | |
3510 | } | |
3511 | ||
f0e43bcd | 3512 | return_io(return_bi); |
16a53ecc N |
3513 | } |
3514 | ||
1442577b | 3515 | static void handle_stripe(struct stripe_head *sh) |
16a53ecc N |
3516 | { |
3517 | if (sh->raid_conf->level == 6) | |
1442577b | 3518 | handle_stripe6(sh); |
16a53ecc | 3519 | else |
1442577b | 3520 | handle_stripe5(sh); |
16a53ecc N |
3521 | } |
3522 | ||
16a53ecc N |
3523 | static void raid5_activate_delayed(raid5_conf_t *conf) |
3524 | { | |
3525 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3526 | while (!list_empty(&conf->delayed_list)) { | |
3527 | struct list_head *l = conf->delayed_list.next; | |
3528 | struct stripe_head *sh; | |
3529 | sh = list_entry(l, struct stripe_head, lru); | |
3530 | list_del_init(l); | |
3531 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3532 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3533 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3534 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3535 | } |
6ed3003c N |
3536 | } else |
3537 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3538 | } |
3539 | ||
3540 | static void activate_bit_delay(raid5_conf_t *conf) | |
3541 | { | |
3542 | /* device_lock is held */ | |
3543 | struct list_head head; | |
3544 | list_add(&head, &conf->bitmap_list); | |
3545 | list_del_init(&conf->bitmap_list); | |
3546 | while (!list_empty(&head)) { | |
3547 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3548 | list_del_init(&sh->lru); | |
3549 | atomic_inc(&sh->count); | |
3550 | __release_stripe(conf, sh); | |
3551 | } | |
3552 | } | |
3553 | ||
3554 | static void unplug_slaves(mddev_t *mddev) | |
3555 | { | |
070ec55d | 3556 | raid5_conf_t *conf = mddev->private; |
16a53ecc | 3557 | int i; |
5e5e3e78 | 3558 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
16a53ecc N |
3559 | |
3560 | rcu_read_lock(); | |
5e5e3e78 | 3561 | for (i = 0; i < devs; i++) { |
16a53ecc N |
3562 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); |
3563 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3564 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3565 | |
3566 | atomic_inc(&rdev->nr_pending); | |
3567 | rcu_read_unlock(); | |
3568 | ||
2ad8b1ef | 3569 | blk_unplug(r_queue); |
16a53ecc N |
3570 | |
3571 | rdev_dec_pending(rdev, mddev); | |
3572 | rcu_read_lock(); | |
3573 | } | |
3574 | } | |
3575 | rcu_read_unlock(); | |
3576 | } | |
3577 | ||
165125e1 | 3578 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3579 | { |
3580 | mddev_t *mddev = q->queuedata; | |
070ec55d | 3581 | raid5_conf_t *conf = mddev->private; |
16a53ecc N |
3582 | unsigned long flags; |
3583 | ||
3584 | spin_lock_irqsave(&conf->device_lock, flags); | |
3585 | ||
3586 | if (blk_remove_plug(q)) { | |
3587 | conf->seq_flush++; | |
3588 | raid5_activate_delayed(conf); | |
72626685 | 3589 | } |
1da177e4 LT |
3590 | md_wakeup_thread(mddev->thread); |
3591 | ||
3592 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3593 | ||
3594 | unplug_slaves(mddev); | |
3595 | } | |
3596 | ||
f022b2fd N |
3597 | static int raid5_congested(void *data, int bits) |
3598 | { | |
3599 | mddev_t *mddev = data; | |
070ec55d | 3600 | raid5_conf_t *conf = mddev->private; |
f022b2fd N |
3601 | |
3602 | /* No difference between reads and writes. Just check | |
3603 | * how busy the stripe_cache is | |
3604 | */ | |
3fa841d7 N |
3605 | |
3606 | if (mddev_congested(mddev, bits)) | |
3607 | return 1; | |
f022b2fd N |
3608 | if (conf->inactive_blocked) |
3609 | return 1; | |
3610 | if (conf->quiesce) | |
3611 | return 1; | |
3612 | if (list_empty_careful(&conf->inactive_list)) | |
3613 | return 1; | |
3614 | ||
3615 | return 0; | |
3616 | } | |
3617 | ||
23032a0e RBJ |
3618 | /* We want read requests to align with chunks where possible, |
3619 | * but write requests don't need to. | |
3620 | */ | |
cc371e66 AK |
3621 | static int raid5_mergeable_bvec(struct request_queue *q, |
3622 | struct bvec_merge_data *bvm, | |
3623 | struct bio_vec *biovec) | |
23032a0e RBJ |
3624 | { |
3625 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3626 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3627 | int max; |
9d8f0363 | 3628 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3629 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3630 | |
cc371e66 | 3631 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3632 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3633 | ||
664e7c41 AN |
3634 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3635 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3636 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3637 | if (max < 0) max = 0; | |
3638 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3639 | return biovec->bv_len; | |
3640 | else | |
3641 | return max; | |
3642 | } | |
3643 | ||
f679623f RBJ |
3644 | |
3645 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3646 | { | |
3647 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3648 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3649 | unsigned int bio_sectors = bio->bi_size >> 9; |
3650 | ||
664e7c41 AN |
3651 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3652 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3653 | return chunk_sectors >= |
3654 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3655 | } | |
3656 | ||
46031f9a RBJ |
3657 | /* |
3658 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3659 | * later sampled by raid5d. | |
3660 | */ | |
3661 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3662 | { | |
3663 | unsigned long flags; | |
3664 | ||
3665 | spin_lock_irqsave(&conf->device_lock, flags); | |
3666 | ||
3667 | bi->bi_next = conf->retry_read_aligned_list; | |
3668 | conf->retry_read_aligned_list = bi; | |
3669 | ||
3670 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3671 | md_wakeup_thread(conf->mddev->thread); | |
3672 | } | |
3673 | ||
3674 | ||
3675 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3676 | { | |
3677 | struct bio *bi; | |
3678 | ||
3679 | bi = conf->retry_read_aligned; | |
3680 | if (bi) { | |
3681 | conf->retry_read_aligned = NULL; | |
3682 | return bi; | |
3683 | } | |
3684 | bi = conf->retry_read_aligned_list; | |
3685 | if(bi) { | |
387bb173 | 3686 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3687 | bi->bi_next = NULL; |
960e739d JA |
3688 | /* |
3689 | * this sets the active strip count to 1 and the processed | |
3690 | * strip count to zero (upper 8 bits) | |
3691 | */ | |
46031f9a | 3692 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3693 | } |
3694 | ||
3695 | return bi; | |
3696 | } | |
3697 | ||
3698 | ||
f679623f RBJ |
3699 | /* |
3700 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3701 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3702 | * first). | |
3703 | * If the read failed.. | |
3704 | */ | |
6712ecf8 | 3705 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3706 | { |
3707 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3708 | mddev_t *mddev; |
3709 | raid5_conf_t *conf; | |
3710 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3711 | mdk_rdev_t *rdev; | |
3712 | ||
f679623f | 3713 | bio_put(bi); |
46031f9a RBJ |
3714 | |
3715 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
070ec55d | 3716 | conf = mddev->private; |
46031f9a RBJ |
3717 | rdev = (void*)raid_bi->bi_next; |
3718 | raid_bi->bi_next = NULL; | |
3719 | ||
3720 | rdev_dec_pending(rdev, conf->mddev); | |
3721 | ||
3722 | if (!error && uptodate) { | |
6712ecf8 | 3723 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3724 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3725 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3726 | return; |
46031f9a RBJ |
3727 | } |
3728 | ||
3729 | ||
45b4233c | 3730 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3731 | |
3732 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3733 | } |
3734 | ||
387bb173 NB |
3735 | static int bio_fits_rdev(struct bio *bi) |
3736 | { | |
165125e1 | 3737 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3738 | |
ae03bf63 | 3739 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3740 | return 0; |
3741 | blk_recount_segments(q, bi); | |
8a78362c | 3742 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3743 | return 0; |
3744 | ||
3745 | if (q->merge_bvec_fn) | |
3746 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3747 | * just just give up | |
3748 | */ | |
3749 | return 0; | |
3750 | ||
3751 | return 1; | |
3752 | } | |
3753 | ||
3754 | ||
165125e1 | 3755 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3756 | { |
3757 | mddev_t *mddev = q->queuedata; | |
070ec55d | 3758 | raid5_conf_t *conf = mddev->private; |
8553fe7e | 3759 | int dd_idx; |
f679623f RBJ |
3760 | struct bio* align_bi; |
3761 | mdk_rdev_t *rdev; | |
3762 | ||
3763 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3764 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3765 | return 0; |
3766 | } | |
3767 | /* | |
99c0fb5f | 3768 | * use bio_clone to make a copy of the bio |
f679623f RBJ |
3769 | */ |
3770 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3771 | if (!align_bi) | |
3772 | return 0; | |
3773 | /* | |
3774 | * set bi_end_io to a new function, and set bi_private to the | |
3775 | * original bio. | |
3776 | */ | |
3777 | align_bi->bi_end_io = raid5_align_endio; | |
3778 | align_bi->bi_private = raid_bio; | |
3779 | /* | |
3780 | * compute position | |
3781 | */ | |
112bf897 N |
3782 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3783 | 0, | |
911d4ee8 | 3784 | &dd_idx, NULL); |
f679623f RBJ |
3785 | |
3786 | rcu_read_lock(); | |
3787 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3788 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3789 | atomic_inc(&rdev->nr_pending); |
3790 | rcu_read_unlock(); | |
46031f9a RBJ |
3791 | raid_bio->bi_next = (void*)rdev; |
3792 | align_bi->bi_bdev = rdev->bdev; | |
3793 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3794 | align_bi->bi_sector += rdev->data_offset; | |
3795 | ||
387bb173 NB |
3796 | if (!bio_fits_rdev(align_bi)) { |
3797 | /* too big in some way */ | |
3798 | bio_put(align_bi); | |
3799 | rdev_dec_pending(rdev, mddev); | |
3800 | return 0; | |
3801 | } | |
3802 | ||
46031f9a RBJ |
3803 | spin_lock_irq(&conf->device_lock); |
3804 | wait_event_lock_irq(conf->wait_for_stripe, | |
3805 | conf->quiesce == 0, | |
3806 | conf->device_lock, /* nothing */); | |
3807 | atomic_inc(&conf->active_aligned_reads); | |
3808 | spin_unlock_irq(&conf->device_lock); | |
3809 | ||
f679623f RBJ |
3810 | generic_make_request(align_bi); |
3811 | return 1; | |
3812 | } else { | |
3813 | rcu_read_unlock(); | |
46031f9a | 3814 | bio_put(align_bi); |
f679623f RBJ |
3815 | return 0; |
3816 | } | |
3817 | } | |
3818 | ||
8b3e6cdc DW |
3819 | /* __get_priority_stripe - get the next stripe to process |
3820 | * | |
3821 | * Full stripe writes are allowed to pass preread active stripes up until | |
3822 | * the bypass_threshold is exceeded. In general the bypass_count | |
3823 | * increments when the handle_list is handled before the hold_list; however, it | |
3824 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3825 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3826 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3827 | * handle_list. | |
3828 | */ | |
3829 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3830 | { | |
3831 | struct stripe_head *sh; | |
3832 | ||
3833 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3834 | __func__, | |
3835 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3836 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3837 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3838 | ||
3839 | if (!list_empty(&conf->handle_list)) { | |
3840 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3841 | ||
3842 | if (list_empty(&conf->hold_list)) | |
3843 | conf->bypass_count = 0; | |
3844 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3845 | if (conf->hold_list.next == conf->last_hold) | |
3846 | conf->bypass_count++; | |
3847 | else { | |
3848 | conf->last_hold = conf->hold_list.next; | |
3849 | conf->bypass_count -= conf->bypass_threshold; | |
3850 | if (conf->bypass_count < 0) | |
3851 | conf->bypass_count = 0; | |
3852 | } | |
3853 | } | |
3854 | } else if (!list_empty(&conf->hold_list) && | |
3855 | ((conf->bypass_threshold && | |
3856 | conf->bypass_count > conf->bypass_threshold) || | |
3857 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3858 | sh = list_entry(conf->hold_list.next, | |
3859 | typeof(*sh), lru); | |
3860 | conf->bypass_count -= conf->bypass_threshold; | |
3861 | if (conf->bypass_count < 0) | |
3862 | conf->bypass_count = 0; | |
3863 | } else | |
3864 | return NULL; | |
3865 | ||
3866 | list_del_init(&sh->lru); | |
3867 | atomic_inc(&sh->count); | |
3868 | BUG_ON(atomic_read(&sh->count) != 1); | |
3869 | return sh; | |
3870 | } | |
f679623f | 3871 | |
165125e1 | 3872 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3873 | { |
3874 | mddev_t *mddev = q->queuedata; | |
070ec55d | 3875 | raid5_conf_t *conf = mddev->private; |
911d4ee8 | 3876 | int dd_idx; |
1da177e4 LT |
3877 | sector_t new_sector; |
3878 | sector_t logical_sector, last_sector; | |
3879 | struct stripe_head *sh; | |
a362357b | 3880 | const int rw = bio_data_dir(bi); |
c9959059 | 3881 | int cpu, remaining; |
1da177e4 | 3882 | |
1f98a13f | 3883 | if (unlikely(bio_rw_flagged(bi, BIO_RW_BARRIER))) { |
a2826aa9 N |
3884 | /* Drain all pending writes. We only really need |
3885 | * to ensure they have been submitted, but this is | |
3886 | * easier. | |
3887 | */ | |
3888 | mddev->pers->quiesce(mddev, 1); | |
3889 | mddev->pers->quiesce(mddev, 0); | |
3890 | md_barrier_request(mddev, bi); | |
e5dcdd80 N |
3891 | return 0; |
3892 | } | |
3893 | ||
3d310eb7 | 3894 | md_write_start(mddev, bi); |
06d91a5f | 3895 | |
074a7aca TH |
3896 | cpu = part_stat_lock(); |
3897 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
3898 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
3899 | bio_sectors(bi)); | |
3900 | part_stat_unlock(); | |
1da177e4 | 3901 | |
802ba064 | 3902 | if (rw == READ && |
52488615 RBJ |
3903 | mddev->reshape_position == MaxSector && |
3904 | chunk_aligned_read(q,bi)) | |
99c0fb5f | 3905 | return 0; |
52488615 | 3906 | |
1da177e4 LT |
3907 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3908 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3909 | bi->bi_next = NULL; | |
3910 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3911 | |
1da177e4 LT |
3912 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3913 | DEFINE_WAIT(w); | |
16a53ecc | 3914 | int disks, data_disks; |
b5663ba4 | 3915 | int previous; |
b578d55f | 3916 | |
7ecaa1e6 | 3917 | retry: |
b5663ba4 | 3918 | previous = 0; |
b0f9ec04 | 3919 | disks = conf->raid_disks; |
b578d55f | 3920 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3921 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3922 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3923 | * 64bit on a 32bit platform, and so it might be |
3924 | * possible to see a half-updated value | |
fef9c61f | 3925 | * Ofcourse reshape_progress could change after |
df8e7f76 N |
3926 | * the lock is dropped, so once we get a reference |
3927 | * to the stripe that we think it is, we will have | |
3928 | * to check again. | |
3929 | */ | |
7ecaa1e6 | 3930 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3931 | if (mddev->delta_disks < 0 |
3932 | ? logical_sector < conf->reshape_progress | |
3933 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3934 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3935 | previous = 1; |
3936 | } else { | |
fef9c61f N |
3937 | if (mddev->delta_disks < 0 |
3938 | ? logical_sector < conf->reshape_safe | |
3939 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3940 | spin_unlock_irq(&conf->device_lock); |
3941 | schedule(); | |
3942 | goto retry; | |
3943 | } | |
3944 | } | |
7ecaa1e6 N |
3945 | spin_unlock_irq(&conf->device_lock); |
3946 | } | |
16a53ecc N |
3947 | data_disks = disks - conf->max_degraded; |
3948 | ||
112bf897 N |
3949 | new_sector = raid5_compute_sector(conf, logical_sector, |
3950 | previous, | |
911d4ee8 | 3951 | &dd_idx, NULL); |
45b4233c | 3952 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3953 | (unsigned long long)new_sector, |
3954 | (unsigned long long)logical_sector); | |
3955 | ||
b5663ba4 | 3956 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3957 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3958 | if (sh) { |
b0f9ec04 | 3959 | if (unlikely(previous)) { |
7ecaa1e6 | 3960 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3961 | * stripe, so we must do the range check again. |
3962 | * Expansion could still move past after this | |
3963 | * test, but as we are holding a reference to | |
3964 | * 'sh', we know that if that happens, | |
3965 | * STRIPE_EXPANDING will get set and the expansion | |
3966 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3967 | */ |
3968 | int must_retry = 0; | |
3969 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
3970 | if (mddev->delta_disks < 0 |
3971 | ? logical_sector >= conf->reshape_progress | |
3972 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
3973 | /* mismatch, need to try again */ |
3974 | must_retry = 1; | |
3975 | spin_unlock_irq(&conf->device_lock); | |
3976 | if (must_retry) { | |
3977 | release_stripe(sh); | |
7a3ab908 | 3978 | schedule(); |
7ecaa1e6 N |
3979 | goto retry; |
3980 | } | |
3981 | } | |
e62e58a5 | 3982 | |
a5c308d4 N |
3983 | if (bio_data_dir(bi) == WRITE && |
3984 | logical_sector >= mddev->suspend_lo && | |
e464eafd N |
3985 | logical_sector < mddev->suspend_hi) { |
3986 | release_stripe(sh); | |
e62e58a5 N |
3987 | /* As the suspend_* range is controlled by |
3988 | * userspace, we want an interruptible | |
3989 | * wait. | |
3990 | */ | |
3991 | flush_signals(current); | |
3992 | prepare_to_wait(&conf->wait_for_overlap, | |
3993 | &w, TASK_INTERRUPTIBLE); | |
3994 | if (logical_sector >= mddev->suspend_lo && | |
3995 | logical_sector < mddev->suspend_hi) | |
3996 | schedule(); | |
e464eafd N |
3997 | goto retry; |
3998 | } | |
7ecaa1e6 N |
3999 | |
4000 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
4001 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
4002 | /* Stripe is busy expanding or | |
4003 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4004 | * and wait a while |
4005 | */ | |
4006 | raid5_unplug_device(mddev->queue); | |
4007 | release_stripe(sh); | |
4008 | schedule(); | |
4009 | goto retry; | |
4010 | } | |
4011 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
4012 | set_bit(STRIPE_HANDLE, &sh->state); |
4013 | clear_bit(STRIPE_DELAYED, &sh->state); | |
729a1866 N |
4014 | if (mddev->barrier && |
4015 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4016 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 4017 | release_stripe(sh); |
1da177e4 LT |
4018 | } else { |
4019 | /* cannot get stripe for read-ahead, just give-up */ | |
4020 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4021 | finish_wait(&conf->wait_for_overlap, &w); | |
4022 | break; | |
4023 | } | |
4024 | ||
4025 | } | |
4026 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4027 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
4028 | spin_unlock_irq(&conf->device_lock); |
4029 | if (remaining == 0) { | |
1da177e4 | 4030 | |
16a53ecc | 4031 | if ( rw == WRITE ) |
1da177e4 | 4032 | md_write_end(mddev); |
6712ecf8 | 4033 | |
0e13fe23 | 4034 | bio_endio(bi, 0); |
1da177e4 | 4035 | } |
729a1866 N |
4036 | |
4037 | if (mddev->barrier) { | |
4038 | /* We need to wait for the stripes to all be handled. | |
4039 | * So: wait for preread_active_stripes to drop to 0. | |
4040 | */ | |
4041 | wait_event(mddev->thread->wqueue, | |
4042 | atomic_read(&conf->preread_active_stripes) == 0); | |
4043 | } | |
1da177e4 LT |
4044 | return 0; |
4045 | } | |
4046 | ||
b522adcd DW |
4047 | static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks); |
4048 | ||
52c03291 | 4049 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4050 | { |
52c03291 N |
4051 | /* reshaping is quite different to recovery/resync so it is |
4052 | * handled quite separately ... here. | |
4053 | * | |
4054 | * On each call to sync_request, we gather one chunk worth of | |
4055 | * destination stripes and flag them as expanding. | |
4056 | * Then we find all the source stripes and request reads. | |
4057 | * As the reads complete, handle_stripe will copy the data | |
4058 | * into the destination stripe and release that stripe. | |
4059 | */ | |
1da177e4 LT |
4060 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
4061 | struct stripe_head *sh; | |
ccfcc3c1 | 4062 | sector_t first_sector, last_sector; |
f416885e N |
4063 | int raid_disks = conf->previous_raid_disks; |
4064 | int data_disks = raid_disks - conf->max_degraded; | |
4065 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4066 | int i; |
4067 | int dd_idx; | |
c8f517c4 | 4068 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4069 | sector_t stripe_addr; |
7a661381 | 4070 | int reshape_sectors; |
ab69ae12 | 4071 | struct list_head stripes; |
52c03291 | 4072 | |
fef9c61f N |
4073 | if (sector_nr == 0) { |
4074 | /* If restarting in the middle, skip the initial sectors */ | |
4075 | if (mddev->delta_disks < 0 && | |
4076 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4077 | sector_nr = raid5_size(mddev, 0, 0) | |
4078 | - conf->reshape_progress; | |
a639755c | 4079 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
4080 | conf->reshape_progress > 0) |
4081 | sector_nr = conf->reshape_progress; | |
f416885e | 4082 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4083 | if (sector_nr) { |
8dee7211 N |
4084 | mddev->curr_resync_completed = sector_nr; |
4085 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4086 | *skipped = 1; |
4087 | return sector_nr; | |
4088 | } | |
52c03291 N |
4089 | } |
4090 | ||
7a661381 N |
4091 | /* We need to process a full chunk at a time. |
4092 | * If old and new chunk sizes differ, we need to process the | |
4093 | * largest of these | |
4094 | */ | |
664e7c41 AN |
4095 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4096 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4097 | else |
9d8f0363 | 4098 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4099 | |
52c03291 N |
4100 | /* we update the metadata when there is more than 3Meg |
4101 | * in the block range (that is rather arbitrary, should | |
4102 | * probably be time based) or when the data about to be | |
4103 | * copied would over-write the source of the data at | |
4104 | * the front of the range. | |
fef9c61f N |
4105 | * i.e. one new_stripe along from reshape_progress new_maps |
4106 | * to after where reshape_safe old_maps to | |
52c03291 | 4107 | */ |
fef9c61f | 4108 | writepos = conf->reshape_progress; |
f416885e | 4109 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4110 | readpos = conf->reshape_progress; |
4111 | sector_div(readpos, data_disks); | |
fef9c61f | 4112 | safepos = conf->reshape_safe; |
f416885e | 4113 | sector_div(safepos, data_disks); |
fef9c61f | 4114 | if (mddev->delta_disks < 0) { |
ed37d83e | 4115 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4116 | readpos += reshape_sectors; |
7a661381 | 4117 | safepos += reshape_sectors; |
fef9c61f | 4118 | } else { |
7a661381 | 4119 | writepos += reshape_sectors; |
ed37d83e N |
4120 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4121 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4122 | } |
52c03291 | 4123 | |
c8f517c4 N |
4124 | /* 'writepos' is the most advanced device address we might write. |
4125 | * 'readpos' is the least advanced device address we might read. | |
4126 | * 'safepos' is the least address recorded in the metadata as having | |
4127 | * been reshaped. | |
4128 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4129 | * ensure safety in the face of a crash - that must be done by userspace | |
4130 | * making a backup of the data. So in that case there is no particular | |
4131 | * rush to update metadata. | |
4132 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4133 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4134 | * we can be safe in the event of a crash. | |
4135 | * So we insist on updating metadata if safepos is behind writepos and | |
4136 | * readpos is beyond writepos. | |
4137 | * In any case, update the metadata every 10 seconds. | |
4138 | * Maybe that number should be configurable, but I'm not sure it is | |
4139 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4140 | */ | |
fef9c61f | 4141 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4142 | ? (safepos > writepos && readpos < writepos) |
4143 | : (safepos < writepos && readpos > writepos)) || | |
4144 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4145 | /* Cannot proceed until we've updated the superblock... */ |
4146 | wait_event(conf->wait_for_overlap, | |
4147 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4148 | mddev->reshape_position = conf->reshape_progress; |
acb180b0 | 4149 | mddev->curr_resync_completed = mddev->curr_resync; |
c8f517c4 | 4150 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4151 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4152 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4153 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4154 | kthread_should_stop()); |
4155 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4156 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4157 | spin_unlock_irq(&conf->device_lock); |
4158 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4159 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4160 | } |
4161 | ||
ec32a2bd N |
4162 | if (mddev->delta_disks < 0) { |
4163 | BUG_ON(conf->reshape_progress == 0); | |
4164 | stripe_addr = writepos; | |
4165 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4166 | ~((sector_t)reshape_sectors - 1)) |
4167 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4168 | != sector_nr); |
4169 | } else { | |
7a661381 | 4170 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4171 | stripe_addr = sector_nr; |
4172 | } | |
ab69ae12 | 4173 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4174 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4175 | int j; |
a9f326eb | 4176 | int skipped_disk = 0; |
a8c906ca | 4177 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4178 | set_bit(STRIPE_EXPANDING, &sh->state); |
4179 | atomic_inc(&conf->reshape_stripes); | |
4180 | /* If any of this stripe is beyond the end of the old | |
4181 | * array, then we need to zero those blocks | |
4182 | */ | |
4183 | for (j=sh->disks; j--;) { | |
4184 | sector_t s; | |
4185 | if (j == sh->pd_idx) | |
4186 | continue; | |
f416885e | 4187 | if (conf->level == 6 && |
d0dabf7e | 4188 | j == sh->qd_idx) |
f416885e | 4189 | continue; |
784052ec | 4190 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4191 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4192 | skipped_disk = 1; |
52c03291 N |
4193 | continue; |
4194 | } | |
4195 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4196 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4197 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4198 | } | |
a9f326eb | 4199 | if (!skipped_disk) { |
52c03291 N |
4200 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4201 | set_bit(STRIPE_HANDLE, &sh->state); | |
4202 | } | |
ab69ae12 | 4203 | list_add(&sh->lru, &stripes); |
52c03291 N |
4204 | } |
4205 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4206 | if (mddev->delta_disks < 0) |
7a661381 | 4207 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4208 | else |
7a661381 | 4209 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4210 | spin_unlock_irq(&conf->device_lock); |
4211 | /* Ok, those stripe are ready. We can start scheduling | |
4212 | * reads on the source stripes. | |
4213 | * The source stripes are determined by mapping the first and last | |
4214 | * block on the destination stripes. | |
4215 | */ | |
52c03291 | 4216 | first_sector = |
ec32a2bd | 4217 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4218 | 1, &dd_idx, NULL); |
52c03291 | 4219 | last_sector = |
0e6e0271 | 4220 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4221 | * new_data_disks - 1), |
911d4ee8 | 4222 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4223 | if (last_sector >= mddev->dev_sectors) |
4224 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4225 | while (first_sector <= last_sector) { |
a8c906ca | 4226 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4227 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4228 | set_bit(STRIPE_HANDLE, &sh->state); | |
4229 | release_stripe(sh); | |
4230 | first_sector += STRIPE_SECTORS; | |
4231 | } | |
ab69ae12 N |
4232 | /* Now that the sources are clearly marked, we can release |
4233 | * the destination stripes | |
4234 | */ | |
4235 | while (!list_empty(&stripes)) { | |
4236 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4237 | list_del_init(&sh->lru); | |
4238 | release_stripe(sh); | |
4239 | } | |
c6207277 N |
4240 | /* If this takes us to the resync_max point where we have to pause, |
4241 | * then we need to write out the superblock. | |
4242 | */ | |
7a661381 | 4243 | sector_nr += reshape_sectors; |
c03f6a19 N |
4244 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4245 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4246 | /* Cannot proceed until we've updated the superblock... */ |
4247 | wait_event(conf->wait_for_overlap, | |
4248 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4249 | mddev->reshape_position = conf->reshape_progress; |
48606a9f | 4250 | mddev->curr_resync_completed = mddev->curr_resync + reshape_sectors; |
c8f517c4 | 4251 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4252 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4253 | md_wakeup_thread(mddev->thread); | |
4254 | wait_event(mddev->sb_wait, | |
4255 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4256 | || kthread_should_stop()); | |
4257 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4258 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4259 | spin_unlock_irq(&conf->device_lock); |
4260 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4261 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4262 | } |
7a661381 | 4263 | return reshape_sectors; |
52c03291 N |
4264 | } |
4265 | ||
4266 | /* FIXME go_faster isn't used */ | |
4267 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
4268 | { | |
4269 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4270 | struct stripe_head *sh; | |
58c0fed4 | 4271 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 4272 | int sync_blocks; |
16a53ecc N |
4273 | int still_degraded = 0; |
4274 | int i; | |
1da177e4 | 4275 | |
72626685 | 4276 | if (sector_nr >= max_sector) { |
1da177e4 LT |
4277 | /* just being told to finish up .. nothing much to do */ |
4278 | unplug_slaves(mddev); | |
cea9c228 | 4279 | |
29269553 N |
4280 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4281 | end_reshape(conf); | |
4282 | return 0; | |
4283 | } | |
72626685 N |
4284 | |
4285 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4286 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4287 | &sync_blocks, 1); | |
16a53ecc | 4288 | else /* completed sync */ |
72626685 N |
4289 | conf->fullsync = 0; |
4290 | bitmap_close_sync(mddev->bitmap); | |
4291 | ||
1da177e4 LT |
4292 | return 0; |
4293 | } | |
ccfcc3c1 | 4294 | |
64bd660b N |
4295 | /* Allow raid5_quiesce to complete */ |
4296 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4297 | ||
52c03291 N |
4298 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4299 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4300 | |
c6207277 N |
4301 | /* No need to check resync_max as we never do more than one |
4302 | * stripe, and as resync_max will always be on a chunk boundary, | |
4303 | * if the check in md_do_sync didn't fire, there is no chance | |
4304 | * of overstepping resync_max here | |
4305 | */ | |
4306 | ||
16a53ecc | 4307 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4308 | * to resync, then assert that we are finished, because there is |
4309 | * nothing we can do. | |
4310 | */ | |
3285edf1 | 4311 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4312 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4313 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4314 | *skipped = 1; |
1da177e4 LT |
4315 | return rv; |
4316 | } | |
72626685 | 4317 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4318 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4319 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4320 | /* we can skip this block, and probably more */ | |
4321 | sync_blocks /= STRIPE_SECTORS; | |
4322 | *skipped = 1; | |
4323 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4324 | } | |
1da177e4 | 4325 | |
b47490c9 N |
4326 | |
4327 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
4328 | ||
a8c906ca | 4329 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4330 | if (sh == NULL) { |
a8c906ca | 4331 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4332 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4333 | * is trying to get access |
1da177e4 | 4334 | */ |
66c006a5 | 4335 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4336 | } |
16a53ecc N |
4337 | /* Need to check if array will still be degraded after recovery/resync |
4338 | * We don't need to check the 'failed' flag as when that gets set, | |
4339 | * recovery aborts. | |
4340 | */ | |
f001a70c | 4341 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4342 | if (conf->disks[i].rdev == NULL) |
4343 | still_degraded = 1; | |
4344 | ||
4345 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4346 | ||
4347 | spin_lock(&sh->lock); | |
1da177e4 LT |
4348 | set_bit(STRIPE_SYNCING, &sh->state); |
4349 | clear_bit(STRIPE_INSYNC, &sh->state); | |
4350 | spin_unlock(&sh->lock); | |
4351 | ||
1442577b | 4352 | handle_stripe(sh); |
1da177e4 LT |
4353 | release_stripe(sh); |
4354 | ||
4355 | return STRIPE_SECTORS; | |
4356 | } | |
4357 | ||
46031f9a RBJ |
4358 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
4359 | { | |
4360 | /* We may not be able to submit a whole bio at once as there | |
4361 | * may not be enough stripe_heads available. | |
4362 | * We cannot pre-allocate enough stripe_heads as we may need | |
4363 | * more than exist in the cache (if we allow ever large chunks). | |
4364 | * So we do one stripe head at a time and record in | |
4365 | * ->bi_hw_segments how many have been done. | |
4366 | * | |
4367 | * We *know* that this entire raid_bio is in one chunk, so | |
4368 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4369 | */ | |
4370 | struct stripe_head *sh; | |
911d4ee8 | 4371 | int dd_idx; |
46031f9a RBJ |
4372 | sector_t sector, logical_sector, last_sector; |
4373 | int scnt = 0; | |
4374 | int remaining; | |
4375 | int handled = 0; | |
4376 | ||
4377 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4378 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4379 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4380 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4381 | ||
4382 | for (; logical_sector < last_sector; | |
387bb173 NB |
4383 | logical_sector += STRIPE_SECTORS, |
4384 | sector += STRIPE_SECTORS, | |
4385 | scnt++) { | |
46031f9a | 4386 | |
960e739d | 4387 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4388 | /* already done this stripe */ |
4389 | continue; | |
4390 | ||
a8c906ca | 4391 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4392 | |
4393 | if (!sh) { | |
4394 | /* failed to get a stripe - must wait */ | |
960e739d | 4395 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4396 | conf->retry_read_aligned = raid_bio; |
4397 | return handled; | |
4398 | } | |
4399 | ||
4400 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4401 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4402 | release_stripe(sh); | |
960e739d | 4403 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4404 | conf->retry_read_aligned = raid_bio; |
4405 | return handled; | |
4406 | } | |
4407 | ||
36d1c647 | 4408 | handle_stripe(sh); |
46031f9a RBJ |
4409 | release_stripe(sh); |
4410 | handled++; | |
4411 | } | |
4412 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4413 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4414 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4415 | if (remaining == 0) |
4416 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4417 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4418 | wake_up(&conf->wait_for_stripe); | |
4419 | return handled; | |
4420 | } | |
4421 | ||
46031f9a | 4422 | |
1da177e4 LT |
4423 | /* |
4424 | * This is our raid5 kernel thread. | |
4425 | * | |
4426 | * We scan the hash table for stripes which can be handled now. | |
4427 | * During the scan, completed stripes are saved for us by the interrupt | |
4428 | * handler, so that they will not have to wait for our next wakeup. | |
4429 | */ | |
6ed3003c | 4430 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4431 | { |
4432 | struct stripe_head *sh; | |
070ec55d | 4433 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
4434 | int handled; |
4435 | ||
45b4233c | 4436 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4437 | |
4438 | md_check_recovery(mddev); | |
1da177e4 LT |
4439 | |
4440 | handled = 0; | |
4441 | spin_lock_irq(&conf->device_lock); | |
4442 | while (1) { | |
46031f9a | 4443 | struct bio *bio; |
1da177e4 | 4444 | |
ae3c20cc | 4445 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 4446 | int seq = conf->seq_flush; |
700e432d | 4447 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4448 | bitmap_unplug(mddev->bitmap); |
700e432d | 4449 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4450 | conf->seq_write = seq; |
4451 | activate_bit_delay(conf); | |
4452 | } | |
4453 | ||
46031f9a RBJ |
4454 | while ((bio = remove_bio_from_retry(conf))) { |
4455 | int ok; | |
4456 | spin_unlock_irq(&conf->device_lock); | |
4457 | ok = retry_aligned_read(conf, bio); | |
4458 | spin_lock_irq(&conf->device_lock); | |
4459 | if (!ok) | |
4460 | break; | |
4461 | handled++; | |
4462 | } | |
4463 | ||
8b3e6cdc DW |
4464 | sh = __get_priority_stripe(conf); |
4465 | ||
c9f21aaf | 4466 | if (!sh) |
1da177e4 | 4467 | break; |
1da177e4 LT |
4468 | spin_unlock_irq(&conf->device_lock); |
4469 | ||
4470 | handled++; | |
417b8d4a DW |
4471 | handle_stripe(sh); |
4472 | release_stripe(sh); | |
4473 | cond_resched(); | |
1da177e4 LT |
4474 | |
4475 | spin_lock_irq(&conf->device_lock); | |
4476 | } | |
45b4233c | 4477 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4478 | |
4479 | spin_unlock_irq(&conf->device_lock); | |
4480 | ||
c9f21aaf | 4481 | async_tx_issue_pending_all(); |
1da177e4 LT |
4482 | unplug_slaves(mddev); |
4483 | ||
45b4233c | 4484 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4485 | } |
4486 | ||
3f294f4f | 4487 | static ssize_t |
007583c9 | 4488 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4489 | { |
070ec55d | 4490 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4491 | if (conf) |
4492 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4493 | else | |
4494 | return 0; | |
3f294f4f N |
4495 | } |
4496 | ||
4497 | static ssize_t | |
007583c9 | 4498 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4499 | { |
070ec55d | 4500 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4501 | unsigned long new; |
b5470dc5 DW |
4502 | int err; |
4503 | ||
3f294f4f N |
4504 | if (len >= PAGE_SIZE) |
4505 | return -EINVAL; | |
96de1e66 N |
4506 | if (!conf) |
4507 | return -ENODEV; | |
3f294f4f | 4508 | |
4ef197d8 | 4509 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4510 | return -EINVAL; |
4511 | if (new <= 16 || new > 32768) | |
4512 | return -EINVAL; | |
4513 | while (new < conf->max_nr_stripes) { | |
4514 | if (drop_one_stripe(conf)) | |
4515 | conf->max_nr_stripes--; | |
4516 | else | |
4517 | break; | |
4518 | } | |
b5470dc5 DW |
4519 | err = md_allow_write(mddev); |
4520 | if (err) | |
4521 | return err; | |
3f294f4f N |
4522 | while (new > conf->max_nr_stripes) { |
4523 | if (grow_one_stripe(conf)) | |
4524 | conf->max_nr_stripes++; | |
4525 | else break; | |
4526 | } | |
4527 | return len; | |
4528 | } | |
007583c9 | 4529 | |
96de1e66 N |
4530 | static struct md_sysfs_entry |
4531 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4532 | raid5_show_stripe_cache_size, | |
4533 | raid5_store_stripe_cache_size); | |
3f294f4f | 4534 | |
8b3e6cdc DW |
4535 | static ssize_t |
4536 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4537 | { | |
070ec55d | 4538 | raid5_conf_t *conf = mddev->private; |
8b3e6cdc DW |
4539 | if (conf) |
4540 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4541 | else | |
4542 | return 0; | |
4543 | } | |
4544 | ||
4545 | static ssize_t | |
4546 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4547 | { | |
070ec55d | 4548 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4549 | unsigned long new; |
8b3e6cdc DW |
4550 | if (len >= PAGE_SIZE) |
4551 | return -EINVAL; | |
4552 | if (!conf) | |
4553 | return -ENODEV; | |
4554 | ||
4ef197d8 | 4555 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4556 | return -EINVAL; |
4ef197d8 | 4557 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4558 | return -EINVAL; |
4559 | conf->bypass_threshold = new; | |
4560 | return len; | |
4561 | } | |
4562 | ||
4563 | static struct md_sysfs_entry | |
4564 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4565 | S_IRUGO | S_IWUSR, | |
4566 | raid5_show_preread_threshold, | |
4567 | raid5_store_preread_threshold); | |
4568 | ||
3f294f4f | 4569 | static ssize_t |
96de1e66 | 4570 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4571 | { |
070ec55d | 4572 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4573 | if (conf) |
4574 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4575 | else | |
4576 | return 0; | |
3f294f4f N |
4577 | } |
4578 | ||
96de1e66 N |
4579 | static struct md_sysfs_entry |
4580 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4581 | |
007583c9 | 4582 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4583 | &raid5_stripecache_size.attr, |
4584 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4585 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4586 | NULL, |
4587 | }; | |
007583c9 N |
4588 | static struct attribute_group raid5_attrs_group = { |
4589 | .name = NULL, | |
4590 | .attrs = raid5_attrs, | |
3f294f4f N |
4591 | }; |
4592 | ||
80c3a6ce DW |
4593 | static sector_t |
4594 | raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
4595 | { | |
070ec55d | 4596 | raid5_conf_t *conf = mddev->private; |
80c3a6ce DW |
4597 | |
4598 | if (!sectors) | |
4599 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4600 | if (!raid_disks) |
7ec05478 | 4601 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4602 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4603 | |
9d8f0363 | 4604 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4605 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4606 | return sectors * (raid_disks - conf->max_degraded); |
4607 | } | |
4608 | ||
36d1c647 DW |
4609 | static void raid5_free_percpu(raid5_conf_t *conf) |
4610 | { | |
4611 | struct raid5_percpu *percpu; | |
4612 | unsigned long cpu; | |
4613 | ||
4614 | if (!conf->percpu) | |
4615 | return; | |
4616 | ||
4617 | get_online_cpus(); | |
4618 | for_each_possible_cpu(cpu) { | |
4619 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4620 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4621 | kfree(percpu->scribble); |
36d1c647 DW |
4622 | } |
4623 | #ifdef CONFIG_HOTPLUG_CPU | |
4624 | unregister_cpu_notifier(&conf->cpu_notify); | |
4625 | #endif | |
4626 | put_online_cpus(); | |
4627 | ||
4628 | free_percpu(conf->percpu); | |
4629 | } | |
4630 | ||
95fc17aa DW |
4631 | static void free_conf(raid5_conf_t *conf) |
4632 | { | |
4633 | shrink_stripes(conf); | |
36d1c647 | 4634 | raid5_free_percpu(conf); |
95fc17aa DW |
4635 | kfree(conf->disks); |
4636 | kfree(conf->stripe_hashtbl); | |
4637 | kfree(conf); | |
4638 | } | |
4639 | ||
36d1c647 DW |
4640 | #ifdef CONFIG_HOTPLUG_CPU |
4641 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4642 | void *hcpu) | |
4643 | { | |
4644 | raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify); | |
4645 | long cpu = (long)hcpu; | |
4646 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4647 | ||
4648 | switch (action) { | |
4649 | case CPU_UP_PREPARE: | |
4650 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4651 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4652 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4653 | if (!percpu->scribble) |
4654 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4655 | ||
4656 | if (!percpu->scribble || | |
4657 | (conf->level == 6 && !percpu->spare_page)) { | |
4658 | safe_put_page(percpu->spare_page); | |
4659 | kfree(percpu->scribble); | |
36d1c647 DW |
4660 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4661 | __func__, cpu); | |
4662 | return NOTIFY_BAD; | |
4663 | } | |
4664 | break; | |
4665 | case CPU_DEAD: | |
4666 | case CPU_DEAD_FROZEN: | |
4667 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4668 | kfree(percpu->scribble); |
36d1c647 | 4669 | percpu->spare_page = NULL; |
d6f38f31 | 4670 | percpu->scribble = NULL; |
36d1c647 DW |
4671 | break; |
4672 | default: | |
4673 | break; | |
4674 | } | |
4675 | return NOTIFY_OK; | |
4676 | } | |
4677 | #endif | |
4678 | ||
4679 | static int raid5_alloc_percpu(raid5_conf_t *conf) | |
4680 | { | |
4681 | unsigned long cpu; | |
4682 | struct page *spare_page; | |
4683 | struct raid5_percpu *allcpus; | |
d6f38f31 | 4684 | void *scribble; |
36d1c647 DW |
4685 | int err; |
4686 | ||
36d1c647 DW |
4687 | allcpus = alloc_percpu(struct raid5_percpu); |
4688 | if (!allcpus) | |
4689 | return -ENOMEM; | |
4690 | conf->percpu = allcpus; | |
4691 | ||
4692 | get_online_cpus(); | |
4693 | err = 0; | |
4694 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4695 | if (conf->level == 6) { |
4696 | spare_page = alloc_page(GFP_KERNEL); | |
4697 | if (!spare_page) { | |
4698 | err = -ENOMEM; | |
4699 | break; | |
4700 | } | |
4701 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4702 | } | |
5e5e3e78 | 4703 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4704 | if (!scribble) { |
36d1c647 DW |
4705 | err = -ENOMEM; |
4706 | break; | |
4707 | } | |
d6f38f31 | 4708 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4709 | } |
4710 | #ifdef CONFIG_HOTPLUG_CPU | |
4711 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4712 | conf->cpu_notify.priority = 0; | |
4713 | if (err == 0) | |
4714 | err = register_cpu_notifier(&conf->cpu_notify); | |
4715 | #endif | |
4716 | put_online_cpus(); | |
4717 | ||
4718 | return err; | |
4719 | } | |
4720 | ||
91adb564 | 4721 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4722 | { |
4723 | raid5_conf_t *conf; | |
5e5e3e78 | 4724 | int raid_disk, memory, max_disks; |
1da177e4 LT |
4725 | mdk_rdev_t *rdev; |
4726 | struct disk_info *disk; | |
1da177e4 | 4727 | |
91adb564 N |
4728 | if (mddev->new_level != 5 |
4729 | && mddev->new_level != 4 | |
4730 | && mddev->new_level != 6) { | |
16a53ecc | 4731 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4732 | mdname(mddev), mddev->new_level); |
4733 | return ERR_PTR(-EIO); | |
1da177e4 | 4734 | } |
91adb564 N |
4735 | if ((mddev->new_level == 5 |
4736 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4737 | (mddev->new_level == 6 | |
4738 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
99c0fb5f | 4739 | printk(KERN_ERR "raid5: %s: layout %d not supported\n", |
91adb564 N |
4740 | mdname(mddev), mddev->new_layout); |
4741 | return ERR_PTR(-EIO); | |
99c0fb5f | 4742 | } |
91adb564 N |
4743 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
4744 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4745 | mdname(mddev), mddev->raid_disks); | |
4746 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4747 | } |
4748 | ||
664e7c41 AN |
4749 | if (!mddev->new_chunk_sectors || |
4750 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4751 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
91adb564 | 4752 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", |
664e7c41 | 4753 | mddev->new_chunk_sectors << 9, mdname(mddev)); |
91adb564 | 4754 | return ERR_PTR(-EINVAL); |
f6705578 N |
4755 | } |
4756 | ||
91adb564 N |
4757 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4758 | if (conf == NULL) | |
1da177e4 | 4759 | goto abort; |
f5efd45a DW |
4760 | spin_lock_init(&conf->device_lock); |
4761 | init_waitqueue_head(&conf->wait_for_stripe); | |
4762 | init_waitqueue_head(&conf->wait_for_overlap); | |
4763 | INIT_LIST_HEAD(&conf->handle_list); | |
4764 | INIT_LIST_HEAD(&conf->hold_list); | |
4765 | INIT_LIST_HEAD(&conf->delayed_list); | |
4766 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4767 | INIT_LIST_HEAD(&conf->inactive_list); | |
4768 | atomic_set(&conf->active_stripes, 0); | |
4769 | atomic_set(&conf->preread_active_stripes, 0); | |
4770 | atomic_set(&conf->active_aligned_reads, 0); | |
4771 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
91adb564 N |
4772 | |
4773 | conf->raid_disks = mddev->raid_disks; | |
4774 | if (mddev->reshape_position == MaxSector) | |
4775 | conf->previous_raid_disks = mddev->raid_disks; | |
4776 | else | |
f6705578 | 4777 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4778 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4779 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4780 | |
5e5e3e78 | 4781 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4782 | GFP_KERNEL); |
4783 | if (!conf->disks) | |
4784 | goto abort; | |
9ffae0cf | 4785 | |
1da177e4 LT |
4786 | conf->mddev = mddev; |
4787 | ||
fccddba0 | 4788 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4789 | goto abort; |
1da177e4 | 4790 | |
36d1c647 DW |
4791 | conf->level = mddev->new_level; |
4792 | if (raid5_alloc_percpu(conf) != 0) | |
4793 | goto abort; | |
4794 | ||
45b4233c | 4795 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4796 | |
159ec1fc | 4797 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4798 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4799 | if (raid_disk >= max_disks |
1da177e4 LT |
4800 | || raid_disk < 0) |
4801 | continue; | |
4802 | disk = conf->disks + raid_disk; | |
4803 | ||
4804 | disk->rdev = rdev; | |
4805 | ||
b2d444d7 | 4806 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4807 | char b[BDEVNAME_SIZE]; |
4808 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4809 | " disk %d\n", bdevname(rdev->bdev,b), | |
4810 | raid_disk); | |
8c2e870a NB |
4811 | } else |
4812 | /* Cannot rely on bitmap to complete recovery */ | |
4813 | conf->fullsync = 1; | |
1da177e4 LT |
4814 | } |
4815 | ||
09c9e5fa | 4816 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4817 | conf->level = mddev->new_level; |
16a53ecc N |
4818 | if (conf->level == 6) |
4819 | conf->max_degraded = 2; | |
4820 | else | |
4821 | conf->max_degraded = 1; | |
91adb564 | 4822 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4823 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4824 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4825 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4826 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4827 | conf->prev_algo = mddev->layout; |
4828 | } | |
1da177e4 | 4829 | |
91adb564 | 4830 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4831 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4832 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4833 | printk(KERN_ERR | |
4834 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4835 | goto abort; | |
4836 | } else | |
4837 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4838 | memory, mdname(mddev)); | |
1da177e4 | 4839 | |
0da3c619 | 4840 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4841 | if (!conf->thread) { |
4842 | printk(KERN_ERR | |
4843 | "raid5: couldn't allocate thread for %s\n", | |
4844 | mdname(mddev)); | |
16a53ecc N |
4845 | goto abort; |
4846 | } | |
91adb564 N |
4847 | |
4848 | return conf; | |
4849 | ||
4850 | abort: | |
4851 | if (conf) { | |
95fc17aa | 4852 | free_conf(conf); |
91adb564 N |
4853 | return ERR_PTR(-EIO); |
4854 | } else | |
4855 | return ERR_PTR(-ENOMEM); | |
4856 | } | |
4857 | ||
c148ffdc N |
4858 | |
4859 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4860 | { | |
4861 | switch (algo) { | |
4862 | case ALGORITHM_PARITY_0: | |
4863 | if (raid_disk < max_degraded) | |
4864 | return 1; | |
4865 | break; | |
4866 | case ALGORITHM_PARITY_N: | |
4867 | if (raid_disk >= raid_disks - max_degraded) | |
4868 | return 1; | |
4869 | break; | |
4870 | case ALGORITHM_PARITY_0_6: | |
4871 | if (raid_disk == 0 || | |
4872 | raid_disk == raid_disks - 1) | |
4873 | return 1; | |
4874 | break; | |
4875 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4876 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4877 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4878 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4879 | if (raid_disk == raid_disks - 1) | |
4880 | return 1; | |
4881 | } | |
4882 | return 0; | |
4883 | } | |
4884 | ||
91adb564 N |
4885 | static int run(mddev_t *mddev) |
4886 | { | |
4887 | raid5_conf_t *conf; | |
8f6c2e4b | 4888 | int working_disks = 0, chunk_size; |
c148ffdc | 4889 | int dirty_parity_disks = 0; |
91adb564 | 4890 | mdk_rdev_t *rdev; |
c148ffdc | 4891 | sector_t reshape_offset = 0; |
91adb564 | 4892 | |
8c6ac868 AN |
4893 | if (mddev->recovery_cp != MaxSector) |
4894 | printk(KERN_NOTICE "raid5: %s is not clean" | |
4895 | " -- starting background reconstruction\n", | |
4896 | mdname(mddev)); | |
91adb564 N |
4897 | if (mddev->reshape_position != MaxSector) { |
4898 | /* Check that we can continue the reshape. | |
4899 | * Currently only disks can change, it must | |
4900 | * increase, and we must be past the point where | |
4901 | * a stripe over-writes itself | |
4902 | */ | |
4903 | sector_t here_new, here_old; | |
4904 | int old_disks; | |
18b00334 | 4905 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4906 | |
88ce4930 | 4907 | if (mddev->new_level != mddev->level) { |
91adb564 N |
4908 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4909 | "required - aborting.\n", | |
4910 | mdname(mddev)); | |
4911 | return -EINVAL; | |
4912 | } | |
91adb564 N |
4913 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4914 | /* reshape_position must be on a new-stripe boundary, and one | |
4915 | * further up in new geometry must map after here in old | |
4916 | * geometry. | |
4917 | */ | |
4918 | here_new = mddev->reshape_position; | |
664e7c41 | 4919 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 N |
4920 | (mddev->raid_disks - max_degraded))) { |
4921 | printk(KERN_ERR "raid5: reshape_position not " | |
4922 | "on a stripe boundary\n"); | |
4923 | return -EINVAL; | |
4924 | } | |
c148ffdc | 4925 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4926 | /* here_new is the stripe we will write to */ |
4927 | here_old = mddev->reshape_position; | |
9d8f0363 | 4928 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4929 | (old_disks-max_degraded)); |
4930 | /* here_old is the first stripe that we might need to read | |
4931 | * from */ | |
67ac6011 N |
4932 | if (mddev->delta_disks == 0) { |
4933 | /* We cannot be sure it is safe to start an in-place | |
4934 | * reshape. It is only safe if user-space if monitoring | |
4935 | * and taking constant backups. | |
4936 | * mdadm always starts a situation like this in | |
4937 | * readonly mode so it can take control before | |
4938 | * allowing any writes. So just check for that. | |
4939 | */ | |
4940 | if ((here_new * mddev->new_chunk_sectors != | |
4941 | here_old * mddev->chunk_sectors) || | |
4942 | mddev->ro == 0) { | |
4943 | printk(KERN_ERR "raid5: in-place reshape must be started" | |
4944 | " in read-only mode - aborting\n"); | |
4945 | return -EINVAL; | |
4946 | } | |
4947 | } else if (mddev->delta_disks < 0 | |
4948 | ? (here_new * mddev->new_chunk_sectors <= | |
4949 | here_old * mddev->chunk_sectors) | |
4950 | : (here_new * mddev->new_chunk_sectors >= | |
4951 | here_old * mddev->chunk_sectors)) { | |
91adb564 N |
4952 | /* Reading from the same stripe as writing to - bad */ |
4953 | printk(KERN_ERR "raid5: reshape_position too early for " | |
4954 | "auto-recovery - aborting.\n"); | |
4955 | return -EINVAL; | |
4956 | } | |
4957 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4958 | /* OK, we should be able to continue; */ | |
4959 | } else { | |
4960 | BUG_ON(mddev->level != mddev->new_level); | |
4961 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 4962 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 4963 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 4964 | } |
91adb564 | 4965 | |
245f46c2 N |
4966 | if (mddev->private == NULL) |
4967 | conf = setup_conf(mddev); | |
4968 | else | |
4969 | conf = mddev->private; | |
4970 | ||
91adb564 N |
4971 | if (IS_ERR(conf)) |
4972 | return PTR_ERR(conf); | |
4973 | ||
4974 | mddev->thread = conf->thread; | |
4975 | conf->thread = NULL; | |
4976 | mddev->private = conf; | |
4977 | ||
4978 | /* | |
4979 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4980 | */ | |
c148ffdc N |
4981 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
4982 | if (rdev->raid_disk < 0) | |
4983 | continue; | |
4984 | if (test_bit(In_sync, &rdev->flags)) | |
91adb564 | 4985 | working_disks++; |
c148ffdc N |
4986 | /* This disc is not fully in-sync. However if it |
4987 | * just stored parity (beyond the recovery_offset), | |
4988 | * when we don't need to be concerned about the | |
4989 | * array being dirty. | |
4990 | * When reshape goes 'backwards', we never have | |
4991 | * partially completed devices, so we only need | |
4992 | * to worry about reshape going forwards. | |
4993 | */ | |
4994 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
4995 | if (mddev->major_version == 0 && | |
4996 | mddev->minor_version > 90) | |
4997 | rdev->recovery_offset = reshape_offset; | |
4998 | ||
4999 | printk("%d: w=%d pa=%d pr=%d m=%d a=%d r=%d op1=%d op2=%d\n", | |
5000 | rdev->raid_disk, working_disks, conf->prev_algo, | |
5001 | conf->previous_raid_disks, conf->max_degraded, | |
5002 | conf->algorithm, conf->raid_disks, | |
5003 | only_parity(rdev->raid_disk, | |
5004 | conf->prev_algo, | |
5005 | conf->previous_raid_disks, | |
5006 | conf->max_degraded), | |
5007 | only_parity(rdev->raid_disk, | |
5008 | conf->algorithm, | |
5009 | conf->raid_disks, | |
5010 | conf->max_degraded)); | |
5011 | if (rdev->recovery_offset < reshape_offset) { | |
5012 | /* We need to check old and new layout */ | |
5013 | if (!only_parity(rdev->raid_disk, | |
5014 | conf->algorithm, | |
5015 | conf->raid_disks, | |
5016 | conf->max_degraded)) | |
5017 | continue; | |
5018 | } | |
5019 | if (!only_parity(rdev->raid_disk, | |
5020 | conf->prev_algo, | |
5021 | conf->previous_raid_disks, | |
5022 | conf->max_degraded)) | |
5023 | continue; | |
5024 | dirty_parity_disks++; | |
5025 | } | |
91adb564 | 5026 | |
5e5e3e78 N |
5027 | mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks) |
5028 | - working_disks); | |
91adb564 | 5029 | |
16a53ecc | 5030 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
5031 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
5032 | " (%d/%d failed)\n", | |
02c2de8c | 5033 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
5034 | goto abort; |
5035 | } | |
5036 | ||
91adb564 | 5037 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 5038 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
5039 | mddev->resync_max_sectors = mddev->dev_sectors; |
5040 | ||
c148ffdc | 5041 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 5042 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
5043 | if (mddev->ok_start_degraded) |
5044 | printk(KERN_WARNING | |
5045 | "raid5: starting dirty degraded array: %s" | |
5046 | "- data corruption possible.\n", | |
5047 | mdname(mddev)); | |
5048 | else { | |
5049 | printk(KERN_ERR | |
5050 | "raid5: cannot start dirty degraded array for %s\n", | |
5051 | mdname(mddev)); | |
5052 | goto abort; | |
5053 | } | |
1da177e4 LT |
5054 | } |
5055 | ||
1da177e4 LT |
5056 | if (mddev->degraded == 0) |
5057 | printk("raid5: raid level %d set %s active with %d out of %d" | |
e183eaed N |
5058 | " devices, algorithm %d\n", conf->level, mdname(mddev), |
5059 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
5060 | mddev->new_layout); | |
1da177e4 LT |
5061 | else |
5062 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
5063 | " out of %d devices, algorithm %d\n", conf->level, | |
5064 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
e183eaed | 5065 | mddev->raid_disks, mddev->new_layout); |
1da177e4 LT |
5066 | |
5067 | print_raid5_conf(conf); | |
5068 | ||
fef9c61f | 5069 | if (conf->reshape_progress != MaxSector) { |
f6705578 | 5070 | printk("...ok start reshape thread\n"); |
fef9c61f | 5071 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5072 | atomic_set(&conf->reshape_stripes, 0); |
5073 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5074 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5075 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5076 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5077 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5078 | "reshape"); |
f6705578 N |
5079 | } |
5080 | ||
1da177e4 | 5081 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 5082 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
5083 | */ |
5084 | { | |
16a53ecc N |
5085 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
5086 | int stripe = data_disks * | |
9d8f0363 | 5087 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
1da177e4 LT |
5088 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5089 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5090 | } | |
5091 | ||
5092 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
5093 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
5094 | printk(KERN_WARNING | |
5095 | "raid5: failed to create sysfs attributes for %s\n", | |
5096 | mdname(mddev)); | |
7a5febe9 | 5097 | |
91adb564 N |
5098 | mddev->queue->queue_lock = &conf->device_lock; |
5099 | ||
7a5febe9 | 5100 | mddev->queue->unplug_fn = raid5_unplug_device; |
f022b2fd | 5101 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 5102 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 5103 | |
1f403624 | 5104 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5105 | |
23032a0e | 5106 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
8f6c2e4b MP |
5107 | chunk_size = mddev->chunk_sectors << 9; |
5108 | blk_queue_io_min(mddev->queue, chunk_size); | |
5109 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5110 | (conf->raid_disks - conf->max_degraded)); | |
5111 | ||
5112 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5113 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
5114 | rdev->data_offset << 9); | |
23032a0e | 5115 | |
1da177e4 LT |
5116 | return 0; |
5117 | abort: | |
e0cf8f04 | 5118 | md_unregister_thread(mddev->thread); |
91adb564 | 5119 | mddev->thread = NULL; |
1da177e4 LT |
5120 | if (conf) { |
5121 | print_raid5_conf(conf); | |
95fc17aa | 5122 | free_conf(conf); |
1da177e4 LT |
5123 | } |
5124 | mddev->private = NULL; | |
5125 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
5126 | return -EIO; | |
5127 | } | |
5128 | ||
5129 | ||
5130 | ||
3f294f4f | 5131 | static int stop(mddev_t *mddev) |
1da177e4 LT |
5132 | { |
5133 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
5134 | ||
5135 | md_unregister_thread(mddev->thread); | |
5136 | mddev->thread = NULL; | |
041ae52e | 5137 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 5138 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
95fc17aa | 5139 | free_conf(conf); |
ef286f6f | 5140 | mddev->private = &raid5_attrs_group; |
1da177e4 LT |
5141 | return 0; |
5142 | } | |
5143 | ||
45b4233c | 5144 | #ifdef DEBUG |
d710e138 | 5145 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
5146 | { |
5147 | int i; | |
5148 | ||
16a53ecc N |
5149 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
5150 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
5151 | seq_printf(seq, "sh %llu, count %d.\n", | |
5152 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
5153 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 5154 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
5155 | seq_printf(seq, "(cache%d: %p %ld) ", |
5156 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 5157 | } |
16a53ecc | 5158 | seq_printf(seq, "\n"); |
1da177e4 LT |
5159 | } |
5160 | ||
d710e138 | 5161 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
5162 | { |
5163 | struct stripe_head *sh; | |
fccddba0 | 5164 | struct hlist_node *hn; |
1da177e4 LT |
5165 | int i; |
5166 | ||
5167 | spin_lock_irq(&conf->device_lock); | |
5168 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 5169 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
5170 | if (sh->raid_conf != conf) |
5171 | continue; | |
16a53ecc | 5172 | print_sh(seq, sh); |
1da177e4 LT |
5173 | } |
5174 | } | |
5175 | spin_unlock_irq(&conf->device_lock); | |
5176 | } | |
5177 | #endif | |
5178 | ||
d710e138 | 5179 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
5180 | { |
5181 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
5182 | int i; | |
5183 | ||
9d8f0363 AN |
5184 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5185 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5186 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5187 | for (i = 0; i < conf->raid_disks; i++) |
5188 | seq_printf (seq, "%s", | |
5189 | conf->disks[i].rdev && | |
b2d444d7 | 5190 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5191 | seq_printf (seq, "]"); |
45b4233c | 5192 | #ifdef DEBUG |
16a53ecc N |
5193 | seq_printf (seq, "\n"); |
5194 | printall(seq, conf); | |
1da177e4 LT |
5195 | #endif |
5196 | } | |
5197 | ||
5198 | static void print_raid5_conf (raid5_conf_t *conf) | |
5199 | { | |
5200 | int i; | |
5201 | struct disk_info *tmp; | |
5202 | ||
5203 | printk("RAID5 conf printout:\n"); | |
5204 | if (!conf) { | |
5205 | printk("(conf==NULL)\n"); | |
5206 | return; | |
5207 | } | |
02c2de8c N |
5208 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
5209 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5210 | |
5211 | for (i = 0; i < conf->raid_disks; i++) { | |
5212 | char b[BDEVNAME_SIZE]; | |
5213 | tmp = conf->disks + i; | |
5214 | if (tmp->rdev) | |
5215 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 5216 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
5217 | bdevname(tmp->rdev->bdev,b)); |
5218 | } | |
5219 | } | |
5220 | ||
5221 | static int raid5_spare_active(mddev_t *mddev) | |
5222 | { | |
5223 | int i; | |
5224 | raid5_conf_t *conf = mddev->private; | |
5225 | struct disk_info *tmp; | |
5226 | ||
5227 | for (i = 0; i < conf->raid_disks; i++) { | |
5228 | tmp = conf->disks + i; | |
5229 | if (tmp->rdev | |
b2d444d7 | 5230 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
5231 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
5232 | unsigned long flags; | |
5233 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 5234 | mddev->degraded--; |
c04be0aa | 5235 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
5236 | } |
5237 | } | |
5238 | print_raid5_conf(conf); | |
5239 | return 0; | |
5240 | } | |
5241 | ||
5242 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
5243 | { | |
5244 | raid5_conf_t *conf = mddev->private; | |
5245 | int err = 0; | |
5246 | mdk_rdev_t *rdev; | |
5247 | struct disk_info *p = conf->disks + number; | |
5248 | ||
5249 | print_raid5_conf(conf); | |
5250 | rdev = p->rdev; | |
5251 | if (rdev) { | |
ec32a2bd N |
5252 | if (number >= conf->raid_disks && |
5253 | conf->reshape_progress == MaxSector) | |
5254 | clear_bit(In_sync, &rdev->flags); | |
5255 | ||
b2d444d7 | 5256 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
5257 | atomic_read(&rdev->nr_pending)) { |
5258 | err = -EBUSY; | |
5259 | goto abort; | |
5260 | } | |
dfc70645 N |
5261 | /* Only remove non-faulty devices if recovery |
5262 | * isn't possible. | |
5263 | */ | |
5264 | if (!test_bit(Faulty, &rdev->flags) && | |
ec32a2bd N |
5265 | mddev->degraded <= conf->max_degraded && |
5266 | number < conf->raid_disks) { | |
dfc70645 N |
5267 | err = -EBUSY; |
5268 | goto abort; | |
5269 | } | |
1da177e4 | 5270 | p->rdev = NULL; |
fbd568a3 | 5271 | synchronize_rcu(); |
1da177e4 LT |
5272 | if (atomic_read(&rdev->nr_pending)) { |
5273 | /* lost the race, try later */ | |
5274 | err = -EBUSY; | |
5275 | p->rdev = rdev; | |
5276 | } | |
5277 | } | |
5278 | abort: | |
5279 | ||
5280 | print_raid5_conf(conf); | |
5281 | return err; | |
5282 | } | |
5283 | ||
5284 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
5285 | { | |
5286 | raid5_conf_t *conf = mddev->private; | |
199050ea | 5287 | int err = -EEXIST; |
1da177e4 LT |
5288 | int disk; |
5289 | struct disk_info *p; | |
6c2fce2e NB |
5290 | int first = 0; |
5291 | int last = conf->raid_disks - 1; | |
1da177e4 | 5292 | |
16a53ecc | 5293 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 5294 | /* no point adding a device */ |
199050ea | 5295 | return -EINVAL; |
1da177e4 | 5296 | |
6c2fce2e NB |
5297 | if (rdev->raid_disk >= 0) |
5298 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5299 | |
5300 | /* | |
16a53ecc N |
5301 | * find the disk ... but prefer rdev->saved_raid_disk |
5302 | * if possible. | |
1da177e4 | 5303 | */ |
16a53ecc | 5304 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5305 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5306 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5307 | disk = rdev->saved_raid_disk; | |
5308 | else | |
6c2fce2e NB |
5309 | disk = first; |
5310 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5311 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5312 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5313 | rdev->raid_disk = disk; |
199050ea | 5314 | err = 0; |
72626685 N |
5315 | if (rdev->saved_raid_disk != disk) |
5316 | conf->fullsync = 1; | |
d6065f7b | 5317 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5318 | break; |
5319 | } | |
5320 | print_raid5_conf(conf); | |
199050ea | 5321 | return err; |
1da177e4 LT |
5322 | } |
5323 | ||
5324 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
5325 | { | |
5326 | /* no resync is happening, and there is enough space | |
5327 | * on all devices, so we can resize. | |
5328 | * We need to make sure resync covers any new space. | |
5329 | * If the array is shrinking we should possibly wait until | |
5330 | * any io in the removed space completes, but it hardly seems | |
5331 | * worth it. | |
5332 | */ | |
9d8f0363 | 5333 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5334 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5335 | mddev->raid_disks)); | |
b522adcd DW |
5336 | if (mddev->array_sectors > |
5337 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5338 | return -EINVAL; | |
f233ea5c | 5339 | set_capacity(mddev->gendisk, mddev->array_sectors); |
44ce6294 | 5340 | mddev->changed = 1; |
449aad3e | 5341 | revalidate_disk(mddev->gendisk); |
58c0fed4 AN |
5342 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
5343 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
5344 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5345 | } | |
58c0fed4 | 5346 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5347 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5348 | return 0; |
5349 | } | |
5350 | ||
01ee22b4 N |
5351 | static int check_stripe_cache(mddev_t *mddev) |
5352 | { | |
5353 | /* Can only proceed if there are plenty of stripe_heads. | |
5354 | * We need a minimum of one full stripe,, and for sensible progress | |
5355 | * it is best to have about 4 times that. | |
5356 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5357 | * allow for chunk sizes up to 256K, which is probably OK. | |
5358 | * If the chunk size is greater, user-space should request more | |
5359 | * stripe_heads first. | |
5360 | */ | |
5361 | raid5_conf_t *conf = mddev->private; | |
5362 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5363 | > conf->max_nr_stripes || | |
5364 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5365 | > conf->max_nr_stripes) { | |
5366 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", | |
5367 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
5368 | / STRIPE_SIZE)*4); | |
5369 | return 0; | |
5370 | } | |
5371 | return 1; | |
5372 | } | |
5373 | ||
50ac168a | 5374 | static int check_reshape(mddev_t *mddev) |
29269553 | 5375 | { |
070ec55d | 5376 | raid5_conf_t *conf = mddev->private; |
29269553 | 5377 | |
88ce4930 N |
5378 | if (mddev->delta_disks == 0 && |
5379 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5380 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5381 | return 0; /* nothing to do */ |
dba034ee N |
5382 | if (mddev->bitmap) |
5383 | /* Cannot grow a bitmap yet */ | |
5384 | return -EBUSY; | |
ec32a2bd N |
5385 | if (mddev->degraded > conf->max_degraded) |
5386 | return -EINVAL; | |
5387 | if (mddev->delta_disks < 0) { | |
5388 | /* We might be able to shrink, but the devices must | |
5389 | * be made bigger first. | |
5390 | * For raid6, 4 is the minimum size. | |
5391 | * Otherwise 2 is the minimum | |
5392 | */ | |
5393 | int min = 2; | |
5394 | if (mddev->level == 6) | |
5395 | min = 4; | |
5396 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5397 | return -EINVAL; | |
5398 | } | |
29269553 | 5399 | |
01ee22b4 | 5400 | if (!check_stripe_cache(mddev)) |
29269553 | 5401 | return -ENOSPC; |
29269553 | 5402 | |
ec32a2bd | 5403 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5404 | } |
5405 | ||
5406 | static int raid5_start_reshape(mddev_t *mddev) | |
5407 | { | |
070ec55d | 5408 | raid5_conf_t *conf = mddev->private; |
63c70c4f | 5409 | mdk_rdev_t *rdev; |
63c70c4f N |
5410 | int spares = 0; |
5411 | int added_devices = 0; | |
c04be0aa | 5412 | unsigned long flags; |
63c70c4f | 5413 | |
f416885e | 5414 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5415 | return -EBUSY; |
5416 | ||
01ee22b4 N |
5417 | if (!check_stripe_cache(mddev)) |
5418 | return -ENOSPC; | |
5419 | ||
159ec1fc | 5420 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5421 | if (rdev->raid_disk < 0 && |
5422 | !test_bit(Faulty, &rdev->flags)) | |
5423 | spares++; | |
63c70c4f | 5424 | |
f416885e | 5425 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5426 | /* Not enough devices even to make a degraded array |
5427 | * of that size | |
5428 | */ | |
5429 | return -EINVAL; | |
5430 | ||
ec32a2bd N |
5431 | /* Refuse to reduce size of the array. Any reductions in |
5432 | * array size must be through explicit setting of array_size | |
5433 | * attribute. | |
5434 | */ | |
5435 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5436 | < mddev->array_sectors) { | |
5437 | printk(KERN_ERR "md: %s: array size must be reduced " | |
5438 | "before number of disks\n", mdname(mddev)); | |
5439 | return -EINVAL; | |
5440 | } | |
5441 | ||
f6705578 | 5442 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5443 | spin_lock_irq(&conf->device_lock); |
5444 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5445 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5446 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5447 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5448 | conf->prev_algo = conf->algorithm; |
5449 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5450 | if (mddev->delta_disks < 0) |
5451 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5452 | else | |
5453 | conf->reshape_progress = 0; | |
5454 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5455 | conf->generation++; |
29269553 N |
5456 | spin_unlock_irq(&conf->device_lock); |
5457 | ||
5458 | /* Add some new drives, as many as will fit. | |
5459 | * We know there are enough to make the newly sized array work. | |
5460 | */ | |
159ec1fc | 5461 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5462 | if (rdev->raid_disk < 0 && |
5463 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 5464 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 | 5465 | char nm[20]; |
9eb07c25 | 5466 | if (rdev->raid_disk >= conf->previous_raid_disks) { |
7ef90146 | 5467 | set_bit(In_sync, &rdev->flags); |
9eb07c25 N |
5468 | added_devices++; |
5469 | } else | |
7ef90146 | 5470 | rdev->recovery_offset = 0; |
29269553 | 5471 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
5472 | if (sysfs_create_link(&mddev->kobj, |
5473 | &rdev->kobj, nm)) | |
5474 | printk(KERN_WARNING | |
5475 | "raid5: failed to create " | |
5476 | " link %s for %s\n", | |
5477 | nm, mdname(mddev)); | |
29269553 N |
5478 | } else |
5479 | break; | |
5480 | } | |
5481 | ||
9eb07c25 N |
5482 | /* When a reshape changes the number of devices, ->degraded |
5483 | * is measured against the large of the pre and post number of | |
5484 | * devices.*/ | |
ec32a2bd N |
5485 | if (mddev->delta_disks > 0) { |
5486 | spin_lock_irqsave(&conf->device_lock, flags); | |
9eb07c25 | 5487 | mddev->degraded += (conf->raid_disks - conf->previous_raid_disks) |
ec32a2bd N |
5488 | - added_devices; |
5489 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5490 | } | |
63c70c4f | 5491 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5492 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5493 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5494 | |
29269553 N |
5495 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5496 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5497 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5498 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5499 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5500 | "reshape"); |
29269553 N |
5501 | if (!mddev->sync_thread) { |
5502 | mddev->recovery = 0; | |
5503 | spin_lock_irq(&conf->device_lock); | |
5504 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5505 | conf->reshape_progress = MaxSector; |
29269553 N |
5506 | spin_unlock_irq(&conf->device_lock); |
5507 | return -EAGAIN; | |
5508 | } | |
c8f517c4 | 5509 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5510 | md_wakeup_thread(mddev->sync_thread); |
5511 | md_new_event(mddev); | |
5512 | return 0; | |
5513 | } | |
29269553 | 5514 | |
ec32a2bd N |
5515 | /* This is called from the reshape thread and should make any |
5516 | * changes needed in 'conf' | |
5517 | */ | |
29269553 N |
5518 | static void end_reshape(raid5_conf_t *conf) |
5519 | { | |
29269553 | 5520 | |
f6705578 | 5521 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5522 | |
f6705578 | 5523 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5524 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5525 | conf->reshape_progress = MaxSector; |
f6705578 | 5526 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5527 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5528 | |
5529 | /* read-ahead size must cover two whole stripes, which is | |
5530 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5531 | */ | |
5532 | { | |
cea9c228 | 5533 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5534 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5535 | / PAGE_SIZE); |
16a53ecc N |
5536 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5537 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5538 | } | |
29269553 | 5539 | } |
29269553 N |
5540 | } |
5541 | ||
ec32a2bd N |
5542 | /* This is called from the raid5d thread with mddev_lock held. |
5543 | * It makes config changes to the device. | |
5544 | */ | |
cea9c228 N |
5545 | static void raid5_finish_reshape(mddev_t *mddev) |
5546 | { | |
070ec55d | 5547 | raid5_conf_t *conf = mddev->private; |
cea9c228 N |
5548 | |
5549 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5550 | ||
ec32a2bd N |
5551 | if (mddev->delta_disks > 0) { |
5552 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5553 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
5554 | mddev->changed = 1; | |
449aad3e | 5555 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5556 | } else { |
5557 | int d; | |
ec32a2bd N |
5558 | mddev->degraded = conf->raid_disks; |
5559 | for (d = 0; d < conf->raid_disks ; d++) | |
5560 | if (conf->disks[d].rdev && | |
5561 | test_bit(In_sync, | |
5562 | &conf->disks[d].rdev->flags)) | |
5563 | mddev->degraded--; | |
5564 | for (d = conf->raid_disks ; | |
5565 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 N |
5566 | d++) { |
5567 | mdk_rdev_t *rdev = conf->disks[d].rdev; | |
5568 | if (rdev && raid5_remove_disk(mddev, d) == 0) { | |
5569 | char nm[20]; | |
5570 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5571 | sysfs_remove_link(&mddev->kobj, nm); | |
5572 | rdev->raid_disk = -1; | |
5573 | } | |
5574 | } | |
cea9c228 | 5575 | } |
88ce4930 | 5576 | mddev->layout = conf->algorithm; |
09c9e5fa | 5577 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5578 | mddev->reshape_position = MaxSector; |
5579 | mddev->delta_disks = 0; | |
cea9c228 N |
5580 | } |
5581 | } | |
5582 | ||
72626685 N |
5583 | static void raid5_quiesce(mddev_t *mddev, int state) |
5584 | { | |
070ec55d | 5585 | raid5_conf_t *conf = mddev->private; |
72626685 N |
5586 | |
5587 | switch(state) { | |
e464eafd N |
5588 | case 2: /* resume for a suspend */ |
5589 | wake_up(&conf->wait_for_overlap); | |
5590 | break; | |
5591 | ||
72626685 N |
5592 | case 1: /* stop all writes */ |
5593 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5594 | /* '2' tells resync/reshape to pause so that all |
5595 | * active stripes can drain | |
5596 | */ | |
5597 | conf->quiesce = 2; | |
72626685 | 5598 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5599 | atomic_read(&conf->active_stripes) == 0 && |
5600 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5601 | conf->device_lock, /* nothing */); |
64bd660b | 5602 | conf->quiesce = 1; |
72626685 | 5603 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5604 | /* allow reshape to continue */ |
5605 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5606 | break; |
5607 | ||
5608 | case 0: /* re-enable writes */ | |
5609 | spin_lock_irq(&conf->device_lock); | |
5610 | conf->quiesce = 0; | |
5611 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5612 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5613 | spin_unlock_irq(&conf->device_lock); |
5614 | break; | |
5615 | } | |
72626685 | 5616 | } |
b15c2e57 | 5617 | |
d562b0c4 N |
5618 | |
5619 | static void *raid5_takeover_raid1(mddev_t *mddev) | |
5620 | { | |
5621 | int chunksect; | |
5622 | ||
5623 | if (mddev->raid_disks != 2 || | |
5624 | mddev->degraded > 1) | |
5625 | return ERR_PTR(-EINVAL); | |
5626 | ||
5627 | /* Should check if there are write-behind devices? */ | |
5628 | ||
5629 | chunksect = 64*2; /* 64K by default */ | |
5630 | ||
5631 | /* The array must be an exact multiple of chunksize */ | |
5632 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5633 | chunksect >>= 1; | |
5634 | ||
5635 | if ((chunksect<<9) < STRIPE_SIZE) | |
5636 | /* array size does not allow a suitable chunk size */ | |
5637 | return ERR_PTR(-EINVAL); | |
5638 | ||
5639 | mddev->new_level = 5; | |
5640 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5641 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5642 | |
5643 | return setup_conf(mddev); | |
5644 | } | |
5645 | ||
fc9739c6 N |
5646 | static void *raid5_takeover_raid6(mddev_t *mddev) |
5647 | { | |
5648 | int new_layout; | |
5649 | ||
5650 | switch (mddev->layout) { | |
5651 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5652 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5653 | break; | |
5654 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5655 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5656 | break; | |
5657 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5658 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5659 | break; | |
5660 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5661 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5662 | break; | |
5663 | case ALGORITHM_PARITY_0_6: | |
5664 | new_layout = ALGORITHM_PARITY_0; | |
5665 | break; | |
5666 | case ALGORITHM_PARITY_N: | |
5667 | new_layout = ALGORITHM_PARITY_N; | |
5668 | break; | |
5669 | default: | |
5670 | return ERR_PTR(-EINVAL); | |
5671 | } | |
5672 | mddev->new_level = 5; | |
5673 | mddev->new_layout = new_layout; | |
5674 | mddev->delta_disks = -1; | |
5675 | mddev->raid_disks -= 1; | |
5676 | return setup_conf(mddev); | |
5677 | } | |
5678 | ||
d562b0c4 | 5679 | |
50ac168a | 5680 | static int raid5_check_reshape(mddev_t *mddev) |
b3546035 | 5681 | { |
88ce4930 N |
5682 | /* For a 2-drive array, the layout and chunk size can be changed |
5683 | * immediately as not restriping is needed. | |
5684 | * For larger arrays we record the new value - after validation | |
5685 | * to be used by a reshape pass. | |
b3546035 | 5686 | */ |
070ec55d | 5687 | raid5_conf_t *conf = mddev->private; |
597a711b | 5688 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5689 | |
597a711b | 5690 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5691 | return -EINVAL; |
5692 | if (new_chunk > 0) { | |
0ba459d2 | 5693 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5694 | return -EINVAL; |
597a711b | 5695 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5696 | return -EINVAL; |
597a711b | 5697 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5698 | /* not factor of array size */ |
5699 | return -EINVAL; | |
5700 | } | |
5701 | ||
5702 | /* They look valid */ | |
5703 | ||
88ce4930 | 5704 | if (mddev->raid_disks == 2) { |
597a711b N |
5705 | /* can make the change immediately */ |
5706 | if (mddev->new_layout >= 0) { | |
5707 | conf->algorithm = mddev->new_layout; | |
5708 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5709 | } |
5710 | if (new_chunk > 0) { | |
597a711b N |
5711 | conf->chunk_sectors = new_chunk ; |
5712 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5713 | } |
5714 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5715 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5716 | } |
50ac168a | 5717 | return check_reshape(mddev); |
88ce4930 N |
5718 | } |
5719 | ||
50ac168a | 5720 | static int raid6_check_reshape(mddev_t *mddev) |
88ce4930 | 5721 | { |
597a711b | 5722 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5723 | |
597a711b | 5724 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5725 | return -EINVAL; |
b3546035 | 5726 | if (new_chunk > 0) { |
0ba459d2 | 5727 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5728 | return -EINVAL; |
597a711b | 5729 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5730 | return -EINVAL; |
597a711b | 5731 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5732 | /* not factor of array size */ |
5733 | return -EINVAL; | |
b3546035 | 5734 | } |
88ce4930 N |
5735 | |
5736 | /* They look valid */ | |
50ac168a | 5737 | return check_reshape(mddev); |
b3546035 N |
5738 | } |
5739 | ||
d562b0c4 N |
5740 | static void *raid5_takeover(mddev_t *mddev) |
5741 | { | |
5742 | /* raid5 can take over: | |
5743 | * raid0 - if all devices are the same - make it a raid4 layout | |
5744 | * raid1 - if there are two drives. We need to know the chunk size | |
5745 | * raid4 - trivial - just use a raid4 layout. | |
5746 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 N |
5747 | */ |
5748 | ||
5749 | if (mddev->level == 1) | |
5750 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5751 | if (mddev->level == 4) { |
5752 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5753 | mddev->new_level = 5; | |
5754 | return setup_conf(mddev); | |
5755 | } | |
fc9739c6 N |
5756 | if (mddev->level == 6) |
5757 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5758 | |
5759 | return ERR_PTR(-EINVAL); | |
5760 | } | |
5761 | ||
5762 | ||
245f46c2 N |
5763 | static struct mdk_personality raid5_personality; |
5764 | ||
5765 | static void *raid6_takeover(mddev_t *mddev) | |
5766 | { | |
5767 | /* Currently can only take over a raid5. We map the | |
5768 | * personality to an equivalent raid6 personality | |
5769 | * with the Q block at the end. | |
5770 | */ | |
5771 | int new_layout; | |
5772 | ||
5773 | if (mddev->pers != &raid5_personality) | |
5774 | return ERR_PTR(-EINVAL); | |
5775 | if (mddev->degraded > 1) | |
5776 | return ERR_PTR(-EINVAL); | |
5777 | if (mddev->raid_disks > 253) | |
5778 | return ERR_PTR(-EINVAL); | |
5779 | if (mddev->raid_disks < 3) | |
5780 | return ERR_PTR(-EINVAL); | |
5781 | ||
5782 | switch (mddev->layout) { | |
5783 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5784 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5785 | break; | |
5786 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5787 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5788 | break; | |
5789 | case ALGORITHM_LEFT_SYMMETRIC: | |
5790 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5791 | break; | |
5792 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5793 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5794 | break; | |
5795 | case ALGORITHM_PARITY_0: | |
5796 | new_layout = ALGORITHM_PARITY_0_6; | |
5797 | break; | |
5798 | case ALGORITHM_PARITY_N: | |
5799 | new_layout = ALGORITHM_PARITY_N; | |
5800 | break; | |
5801 | default: | |
5802 | return ERR_PTR(-EINVAL); | |
5803 | } | |
5804 | mddev->new_level = 6; | |
5805 | mddev->new_layout = new_layout; | |
5806 | mddev->delta_disks = 1; | |
5807 | mddev->raid_disks += 1; | |
5808 | return setup_conf(mddev); | |
5809 | } | |
5810 | ||
5811 | ||
16a53ecc N |
5812 | static struct mdk_personality raid6_personality = |
5813 | { | |
5814 | .name = "raid6", | |
5815 | .level = 6, | |
5816 | .owner = THIS_MODULE, | |
5817 | .make_request = make_request, | |
5818 | .run = run, | |
5819 | .stop = stop, | |
5820 | .status = status, | |
5821 | .error_handler = error, | |
5822 | .hot_add_disk = raid5_add_disk, | |
5823 | .hot_remove_disk= raid5_remove_disk, | |
5824 | .spare_active = raid5_spare_active, | |
5825 | .sync_request = sync_request, | |
5826 | .resize = raid5_resize, | |
80c3a6ce | 5827 | .size = raid5_size, |
50ac168a | 5828 | .check_reshape = raid6_check_reshape, |
f416885e | 5829 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5830 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5831 | .quiesce = raid5_quiesce, |
245f46c2 | 5832 | .takeover = raid6_takeover, |
16a53ecc | 5833 | }; |
2604b703 | 5834 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
5835 | { |
5836 | .name = "raid5", | |
2604b703 | 5837 | .level = 5, |
1da177e4 LT |
5838 | .owner = THIS_MODULE, |
5839 | .make_request = make_request, | |
5840 | .run = run, | |
5841 | .stop = stop, | |
5842 | .status = status, | |
5843 | .error_handler = error, | |
5844 | .hot_add_disk = raid5_add_disk, | |
5845 | .hot_remove_disk= raid5_remove_disk, | |
5846 | .spare_active = raid5_spare_active, | |
5847 | .sync_request = sync_request, | |
5848 | .resize = raid5_resize, | |
80c3a6ce | 5849 | .size = raid5_size, |
63c70c4f N |
5850 | .check_reshape = raid5_check_reshape, |
5851 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5852 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5853 | .quiesce = raid5_quiesce, |
d562b0c4 | 5854 | .takeover = raid5_takeover, |
1da177e4 LT |
5855 | }; |
5856 | ||
2604b703 | 5857 | static struct mdk_personality raid4_personality = |
1da177e4 | 5858 | { |
2604b703 N |
5859 | .name = "raid4", |
5860 | .level = 4, | |
5861 | .owner = THIS_MODULE, | |
5862 | .make_request = make_request, | |
5863 | .run = run, | |
5864 | .stop = stop, | |
5865 | .status = status, | |
5866 | .error_handler = error, | |
5867 | .hot_add_disk = raid5_add_disk, | |
5868 | .hot_remove_disk= raid5_remove_disk, | |
5869 | .spare_active = raid5_spare_active, | |
5870 | .sync_request = sync_request, | |
5871 | .resize = raid5_resize, | |
80c3a6ce | 5872 | .size = raid5_size, |
3d37890b N |
5873 | .check_reshape = raid5_check_reshape, |
5874 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5875 | .finish_reshape = raid5_finish_reshape, |
2604b703 N |
5876 | .quiesce = raid5_quiesce, |
5877 | }; | |
5878 | ||
5879 | static int __init raid5_init(void) | |
5880 | { | |
16a53ecc | 5881 | register_md_personality(&raid6_personality); |
2604b703 N |
5882 | register_md_personality(&raid5_personality); |
5883 | register_md_personality(&raid4_personality); | |
5884 | return 0; | |
1da177e4 LT |
5885 | } |
5886 | ||
2604b703 | 5887 | static void raid5_exit(void) |
1da177e4 | 5888 | { |
16a53ecc | 5889 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5890 | unregister_md_personality(&raid5_personality); |
5891 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5892 | } |
5893 | ||
5894 | module_init(raid5_init); | |
5895 | module_exit(raid5_exit); | |
5896 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5897 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 5898 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
5899 | MODULE_ALIAS("md-raid5"); |
5900 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5901 | MODULE_ALIAS("md-level-5"); |
5902 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5903 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5904 | MODULE_ALIAS("md-raid6"); | |
5905 | MODULE_ALIAS("md-level-6"); | |
5906 | ||
5907 | /* This used to be two separate modules, they were: */ | |
5908 | MODULE_ALIAS("raid5"); | |
5909 | MODULE_ALIAS("raid6"); |