]>
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> |
bff61975 | 50 | #include <linux/seq_file.h> |
36d1c647 | 51 | #include <linux/cpu.h> |
43b2e5d8 | 52 | #include "md.h" |
bff61975 | 53 | #include "raid5.h" |
ef740c37 | 54 | #include "bitmap.h" |
72626685 | 55 | |
1da177e4 LT |
56 | /* |
57 | * Stripe cache | |
58 | */ | |
59 | ||
60 | #define NR_STRIPES 256 | |
61 | #define STRIPE_SIZE PAGE_SIZE | |
62 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
63 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
64 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 65 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 66 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
67 | #define HASH_MASK (NR_HASH - 1) |
68 | ||
fccddba0 | 69 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
70 | |
71 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
72 | * order without overlap. There may be several bio's per stripe+device, and | |
73 | * a bio could span several devices. | |
74 | * When walking this list for a particular stripe+device, we must never proceed | |
75 | * beyond a bio that extends past this device, as the next bio might no longer | |
76 | * be valid. | |
77 | * This macro is used to determine the 'next' bio in the list, given the sector | |
78 | * of the current stripe+device | |
79 | */ | |
80 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
81 | /* | |
82 | * The following can be used to debug the driver | |
83 | */ | |
1da177e4 LT |
84 | #define RAID5_PARANOIA 1 |
85 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
86 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
87 | #else | |
88 | # define CHECK_DEVLOCK() | |
89 | #endif | |
90 | ||
45b4233c | 91 | #ifdef DEBUG |
1da177e4 LT |
92 | #define inline |
93 | #define __inline__ | |
94 | #endif | |
95 | ||
6be9d494 BS |
96 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
97 | ||
960e739d | 98 | /* |
5b99c2ff JA |
99 | * We maintain a biased count of active stripes in the bottom 16 bits of |
100 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
101 | */ |
102 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
103 | { | |
5b99c2ff | 104 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
105 | } |
106 | ||
107 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
108 | { | |
5b99c2ff | 109 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
110 | } |
111 | ||
112 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
113 | { | |
114 | --bio->bi_phys_segments; | |
115 | return raid5_bi_phys_segments(bio); | |
116 | } | |
117 | ||
118 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
119 | { | |
120 | unsigned short val = raid5_bi_hw_segments(bio); | |
121 | ||
122 | --val; | |
5b99c2ff | 123 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
124 | return val; |
125 | } | |
126 | ||
127 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
128 | { | |
5b99c2ff | 129 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16); |
960e739d JA |
130 | } |
131 | ||
d0dabf7e N |
132 | /* Find first data disk in a raid6 stripe */ |
133 | static inline int raid6_d0(struct stripe_head *sh) | |
134 | { | |
67cc2b81 N |
135 | if (sh->ddf_layout) |
136 | /* ddf always start from first device */ | |
137 | return 0; | |
138 | /* md starts just after Q block */ | |
d0dabf7e N |
139 | if (sh->qd_idx == sh->disks - 1) |
140 | return 0; | |
141 | else | |
142 | return sh->qd_idx + 1; | |
143 | } | |
16a53ecc N |
144 | static inline int raid6_next_disk(int disk, int raid_disks) |
145 | { | |
146 | disk++; | |
147 | return (disk < raid_disks) ? disk : 0; | |
148 | } | |
a4456856 | 149 | |
d0dabf7e N |
150 | /* When walking through the disks in a raid5, starting at raid6_d0, |
151 | * We need to map each disk to a 'slot', where the data disks are slot | |
152 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
153 | * is raid_disks-1. This help does that mapping. | |
154 | */ | |
67cc2b81 N |
155 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
156 | int *count, int syndrome_disks) | |
d0dabf7e N |
157 | { |
158 | int slot; | |
67cc2b81 | 159 | |
d0dabf7e | 160 | if (idx == sh->pd_idx) |
67cc2b81 | 161 | return syndrome_disks; |
d0dabf7e | 162 | if (idx == sh->qd_idx) |
67cc2b81 | 163 | return syndrome_disks + 1; |
d0dabf7e N |
164 | slot = (*count)++; |
165 | return slot; | |
166 | } | |
167 | ||
a4456856 DW |
168 | static void return_io(struct bio *return_bi) |
169 | { | |
170 | struct bio *bi = return_bi; | |
171 | while (bi) { | |
a4456856 DW |
172 | |
173 | return_bi = bi->bi_next; | |
174 | bi->bi_next = NULL; | |
175 | bi->bi_size = 0; | |
0e13fe23 | 176 | bio_endio(bi, 0); |
a4456856 DW |
177 | bi = return_bi; |
178 | } | |
179 | } | |
180 | ||
1da177e4 LT |
181 | static void print_raid5_conf (raid5_conf_t *conf); |
182 | ||
600aa109 DW |
183 | static int stripe_operations_active(struct stripe_head *sh) |
184 | { | |
185 | return sh->check_state || sh->reconstruct_state || | |
186 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
187 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
188 | } | |
189 | ||
858119e1 | 190 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
191 | { |
192 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
193 | BUG_ON(!list_empty(&sh->lru)); |
194 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 195 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 196 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 197 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
198 | blk_plug_device(conf->mddev->queue); |
199 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 200 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 201 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
202 | blk_plug_device(conf->mddev->queue); |
203 | } else { | |
72626685 | 204 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 205 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 206 | } |
1da177e4 LT |
207 | md_wakeup_thread(conf->mddev->thread); |
208 | } else { | |
600aa109 | 209 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
210 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
211 | atomic_dec(&conf->preread_active_stripes); | |
212 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
213 | md_wakeup_thread(conf->mddev->thread); | |
214 | } | |
1da177e4 | 215 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
216 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
217 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 218 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
219 | if (conf->retry_read_aligned) |
220 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 221 | } |
1da177e4 LT |
222 | } |
223 | } | |
224 | } | |
d0dabf7e | 225 | |
1da177e4 LT |
226 | static void release_stripe(struct stripe_head *sh) |
227 | { | |
228 | raid5_conf_t *conf = sh->raid_conf; | |
229 | unsigned long flags; | |
16a53ecc | 230 | |
1da177e4 LT |
231 | spin_lock_irqsave(&conf->device_lock, flags); |
232 | __release_stripe(conf, sh); | |
233 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
234 | } | |
235 | ||
fccddba0 | 236 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 237 | { |
45b4233c DW |
238 | pr_debug("remove_hash(), stripe %llu\n", |
239 | (unsigned long long)sh->sector); | |
1da177e4 | 240 | |
fccddba0 | 241 | hlist_del_init(&sh->hash); |
1da177e4 LT |
242 | } |
243 | ||
16a53ecc | 244 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 245 | { |
fccddba0 | 246 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 247 | |
45b4233c DW |
248 | pr_debug("insert_hash(), stripe %llu\n", |
249 | (unsigned long long)sh->sector); | |
1da177e4 LT |
250 | |
251 | CHECK_DEVLOCK(); | |
fccddba0 | 252 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
253 | } |
254 | ||
255 | ||
256 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
257 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
258 | { | |
259 | struct stripe_head *sh = NULL; | |
260 | struct list_head *first; | |
261 | ||
262 | CHECK_DEVLOCK(); | |
263 | if (list_empty(&conf->inactive_list)) | |
264 | goto out; | |
265 | first = conf->inactive_list.next; | |
266 | sh = list_entry(first, struct stripe_head, lru); | |
267 | list_del_init(first); | |
268 | remove_hash(sh); | |
269 | atomic_inc(&conf->active_stripes); | |
270 | out: | |
271 | return sh; | |
272 | } | |
273 | ||
274 | static void shrink_buffers(struct stripe_head *sh, int num) | |
275 | { | |
276 | struct page *p; | |
277 | int i; | |
278 | ||
279 | for (i=0; i<num ; i++) { | |
280 | p = sh->dev[i].page; | |
281 | if (!p) | |
282 | continue; | |
283 | sh->dev[i].page = NULL; | |
2d1f3b5d | 284 | put_page(p); |
1da177e4 LT |
285 | } |
286 | } | |
287 | ||
288 | static int grow_buffers(struct stripe_head *sh, int num) | |
289 | { | |
290 | int i; | |
291 | ||
292 | for (i=0; i<num; i++) { | |
293 | struct page *page; | |
294 | ||
295 | if (!(page = alloc_page(GFP_KERNEL))) { | |
296 | return 1; | |
297 | } | |
298 | sh->dev[i].page = page; | |
299 | } | |
300 | return 0; | |
301 | } | |
302 | ||
784052ec | 303 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
911d4ee8 N |
304 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
305 | struct stripe_head *sh); | |
1da177e4 | 306 | |
b5663ba4 | 307 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
308 | { |
309 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 310 | int i; |
1da177e4 | 311 | |
78bafebd ES |
312 | BUG_ON(atomic_read(&sh->count) != 0); |
313 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 314 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 315 | |
1da177e4 | 316 | CHECK_DEVLOCK(); |
45b4233c | 317 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
318 | (unsigned long long)sh->sector); |
319 | ||
320 | remove_hash(sh); | |
16a53ecc | 321 | |
86b42c71 | 322 | sh->generation = conf->generation - previous; |
b5663ba4 | 323 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 324 | sh->sector = sector; |
911d4ee8 | 325 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
326 | sh->state = 0; |
327 | ||
7ecaa1e6 N |
328 | |
329 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
330 | struct r5dev *dev = &sh->dev[i]; |
331 | ||
d84e0f10 | 332 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 333 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 334 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 335 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 336 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
337 | test_bit(R5_LOCKED, &dev->flags)); |
338 | BUG(); | |
339 | } | |
340 | dev->flags = 0; | |
784052ec | 341 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
342 | } |
343 | insert_hash(conf, sh); | |
344 | } | |
345 | ||
86b42c71 N |
346 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, |
347 | short generation) | |
1da177e4 LT |
348 | { |
349 | struct stripe_head *sh; | |
fccddba0 | 350 | struct hlist_node *hn; |
1da177e4 LT |
351 | |
352 | CHECK_DEVLOCK(); | |
45b4233c | 353 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 354 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 355 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 356 | return sh; |
45b4233c | 357 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
358 | return NULL; |
359 | } | |
360 | ||
361 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 362 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 363 | |
b5663ba4 N |
364 | static struct stripe_head * |
365 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
366 | int previous, int noblock) | |
1da177e4 LT |
367 | { |
368 | struct stripe_head *sh; | |
369 | ||
45b4233c | 370 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
371 | |
372 | spin_lock_irq(&conf->device_lock); | |
373 | ||
374 | do { | |
72626685 N |
375 | wait_event_lock_irq(conf->wait_for_stripe, |
376 | conf->quiesce == 0, | |
377 | conf->device_lock, /* nothing */); | |
86b42c71 | 378 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
379 | if (!sh) { |
380 | if (!conf->inactive_blocked) | |
381 | sh = get_free_stripe(conf); | |
382 | if (noblock && sh == NULL) | |
383 | break; | |
384 | if (!sh) { | |
385 | conf->inactive_blocked = 1; | |
386 | wait_event_lock_irq(conf->wait_for_stripe, | |
387 | !list_empty(&conf->inactive_list) && | |
5036805b N |
388 | (atomic_read(&conf->active_stripes) |
389 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
390 | || !conf->inactive_blocked), |
391 | conf->device_lock, | |
f4370781 | 392 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
393 | ); |
394 | conf->inactive_blocked = 0; | |
395 | } else | |
b5663ba4 | 396 | init_stripe(sh, sector, previous); |
1da177e4 LT |
397 | } else { |
398 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
399 | BUG_ON(!list_empty(&sh->lru) |
400 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
401 | } else { |
402 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
403 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
404 | if (list_empty(&sh->lru) && |
405 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
406 | BUG(); |
407 | list_del_init(&sh->lru); | |
1da177e4 LT |
408 | } |
409 | } | |
410 | } while (sh == NULL); | |
411 | ||
412 | if (sh) | |
413 | atomic_inc(&sh->count); | |
414 | ||
415 | spin_unlock_irq(&conf->device_lock); | |
416 | return sh; | |
417 | } | |
418 | ||
6712ecf8 N |
419 | static void |
420 | raid5_end_read_request(struct bio *bi, int error); | |
421 | static void | |
422 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 423 | |
c4e5ac0a | 424 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
425 | { |
426 | raid5_conf_t *conf = sh->raid_conf; | |
427 | int i, disks = sh->disks; | |
428 | ||
429 | might_sleep(); | |
430 | ||
431 | for (i = disks; i--; ) { | |
432 | int rw; | |
433 | struct bio *bi; | |
434 | mdk_rdev_t *rdev; | |
435 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
436 | rw = WRITE; | |
437 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
438 | rw = READ; | |
439 | else | |
440 | continue; | |
441 | ||
442 | bi = &sh->dev[i].req; | |
443 | ||
444 | bi->bi_rw = rw; | |
445 | if (rw == WRITE) | |
446 | bi->bi_end_io = raid5_end_write_request; | |
447 | else | |
448 | bi->bi_end_io = raid5_end_read_request; | |
449 | ||
450 | rcu_read_lock(); | |
451 | rdev = rcu_dereference(conf->disks[i].rdev); | |
452 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
453 | rdev = NULL; | |
454 | if (rdev) | |
455 | atomic_inc(&rdev->nr_pending); | |
456 | rcu_read_unlock(); | |
457 | ||
458 | if (rdev) { | |
c4e5ac0a | 459 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
460 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
461 | ||
2b7497f0 DW |
462 | set_bit(STRIPE_IO_STARTED, &sh->state); |
463 | ||
91c00924 DW |
464 | bi->bi_bdev = rdev->bdev; |
465 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 466 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
467 | bi->bi_rw, i); |
468 | atomic_inc(&sh->count); | |
469 | bi->bi_sector = sh->sector + rdev->data_offset; | |
470 | bi->bi_flags = 1 << BIO_UPTODATE; | |
471 | bi->bi_vcnt = 1; | |
472 | bi->bi_max_vecs = 1; | |
473 | bi->bi_idx = 0; | |
474 | bi->bi_io_vec = &sh->dev[i].vec; | |
475 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
476 | bi->bi_io_vec[0].bv_offset = 0; | |
477 | bi->bi_size = STRIPE_SIZE; | |
478 | bi->bi_next = NULL; | |
479 | if (rw == WRITE && | |
480 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
481 | atomic_add(STRIPE_SECTORS, | |
482 | &rdev->corrected_errors); | |
483 | generic_make_request(bi); | |
484 | } else { | |
485 | if (rw == WRITE) | |
486 | set_bit(STRIPE_DEGRADED, &sh->state); | |
487 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
488 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
489 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
490 | set_bit(STRIPE_HANDLE, &sh->state); | |
491 | } | |
492 | } | |
493 | } | |
494 | ||
495 | static struct dma_async_tx_descriptor * | |
496 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
497 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
498 | { | |
499 | struct bio_vec *bvl; | |
500 | struct page *bio_page; | |
501 | int i; | |
502 | int page_offset; | |
a08abd8c | 503 | struct async_submit_ctl submit; |
91c00924 DW |
504 | |
505 | if (bio->bi_sector >= sector) | |
506 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
507 | else | |
508 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c DW |
509 | |
510 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); | |
91c00924 DW |
511 | bio_for_each_segment(bvl, bio, i) { |
512 | int len = bio_iovec_idx(bio, i)->bv_len; | |
513 | int clen; | |
514 | int b_offset = 0; | |
515 | ||
516 | if (page_offset < 0) { | |
517 | b_offset = -page_offset; | |
518 | page_offset += b_offset; | |
519 | len -= b_offset; | |
520 | } | |
521 | ||
522 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
523 | clen = STRIPE_SIZE - page_offset; | |
524 | else | |
525 | clen = len; | |
526 | ||
527 | if (clen > 0) { | |
528 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
529 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
530 | if (frombio) | |
531 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 532 | b_offset, clen, &submit); |
91c00924 DW |
533 | else |
534 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 535 | page_offset, clen, &submit); |
91c00924 | 536 | } |
a08abd8c DW |
537 | /* chain the operations */ |
538 | submit.depend_tx = tx; | |
539 | ||
91c00924 DW |
540 | if (clen < len) /* hit end of page */ |
541 | break; | |
542 | page_offset += len; | |
543 | } | |
544 | ||
545 | return tx; | |
546 | } | |
547 | ||
548 | static void ops_complete_biofill(void *stripe_head_ref) | |
549 | { | |
550 | struct stripe_head *sh = stripe_head_ref; | |
551 | struct bio *return_bi = NULL; | |
552 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 553 | int i; |
91c00924 | 554 | |
e46b272b | 555 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
556 | (unsigned long long)sh->sector); |
557 | ||
558 | /* clear completed biofills */ | |
83de75cc | 559 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
560 | for (i = sh->disks; i--; ) { |
561 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
562 | |
563 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
564 | /* and check if we need to reply to a read request, |
565 | * new R5_Wantfill requests are held off until | |
83de75cc | 566 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
567 | */ |
568 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 569 | struct bio *rbi, *rbi2; |
91c00924 | 570 | |
91c00924 DW |
571 | BUG_ON(!dev->read); |
572 | rbi = dev->read; | |
573 | dev->read = NULL; | |
574 | while (rbi && rbi->bi_sector < | |
575 | dev->sector + STRIPE_SECTORS) { | |
576 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 577 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
578 | rbi->bi_next = return_bi; |
579 | return_bi = rbi; | |
580 | } | |
91c00924 DW |
581 | rbi = rbi2; |
582 | } | |
583 | } | |
584 | } | |
83de75cc DW |
585 | spin_unlock_irq(&conf->device_lock); |
586 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
587 | |
588 | return_io(return_bi); | |
589 | ||
e4d84909 | 590 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
591 | release_stripe(sh); |
592 | } | |
593 | ||
594 | static void ops_run_biofill(struct stripe_head *sh) | |
595 | { | |
596 | struct dma_async_tx_descriptor *tx = NULL; | |
597 | raid5_conf_t *conf = sh->raid_conf; | |
a08abd8c | 598 | struct async_submit_ctl submit; |
91c00924 DW |
599 | int i; |
600 | ||
e46b272b | 601 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
602 | (unsigned long long)sh->sector); |
603 | ||
604 | for (i = sh->disks; i--; ) { | |
605 | struct r5dev *dev = &sh->dev[i]; | |
606 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
607 | struct bio *rbi; | |
608 | spin_lock_irq(&conf->device_lock); | |
609 | dev->read = rbi = dev->toread; | |
610 | dev->toread = NULL; | |
611 | spin_unlock_irq(&conf->device_lock); | |
612 | while (rbi && rbi->bi_sector < | |
613 | dev->sector + STRIPE_SECTORS) { | |
614 | tx = async_copy_data(0, rbi, dev->page, | |
615 | dev->sector, tx); | |
616 | rbi = r5_next_bio(rbi, dev->sector); | |
617 | } | |
618 | } | |
619 | } | |
620 | ||
621 | atomic_inc(&sh->count); | |
a08abd8c DW |
622 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
623 | async_trigger_callback(&submit); | |
91c00924 DW |
624 | } |
625 | ||
4e7d2c0a DW |
626 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
627 | { | |
628 | struct r5dev *tgt; | |
629 | ||
630 | if (target < 0) | |
631 | return; | |
632 | ||
633 | tgt = &sh->dev[target]; | |
634 | set_bit(R5_UPTODATE, &tgt->flags); | |
635 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
636 | clear_bit(R5_Wantcompute, &tgt->flags); | |
637 | } | |
638 | ||
ac6b53b6 | 639 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
640 | { |
641 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 642 | |
e46b272b | 643 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
644 | (unsigned long long)sh->sector); |
645 | ||
ac6b53b6 | 646 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 647 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 648 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 649 | |
ecc65c9b DW |
650 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
651 | if (sh->check_state == check_state_compute_run) | |
652 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
653 | set_bit(STRIPE_HANDLE, &sh->state); |
654 | release_stripe(sh); | |
655 | } | |
656 | ||
d6f38f31 DW |
657 | /* return a pointer to the address conversion region of the scribble buffer */ |
658 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
659 | struct raid5_percpu *percpu) | |
660 | { | |
661 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
662 | } | |
663 | ||
664 | static struct dma_async_tx_descriptor * | |
665 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 666 | { |
91c00924 | 667 | int disks = sh->disks; |
d6f38f31 | 668 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
669 | int target = sh->ops.target; |
670 | struct r5dev *tgt = &sh->dev[target]; | |
671 | struct page *xor_dest = tgt->page; | |
672 | int count = 0; | |
673 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 674 | struct async_submit_ctl submit; |
91c00924 DW |
675 | int i; |
676 | ||
677 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 678 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
679 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
680 | ||
681 | for (i = disks; i--; ) | |
682 | if (i != target) | |
683 | xor_srcs[count++] = sh->dev[i].page; | |
684 | ||
685 | atomic_inc(&sh->count); | |
686 | ||
a08abd8c | 687 | init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 688 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 689 | if (unlikely(count == 1)) |
a08abd8c | 690 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 691 | else |
a08abd8c | 692 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 693 | |
91c00924 DW |
694 | return tx; |
695 | } | |
696 | ||
ac6b53b6 DW |
697 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
698 | * @srcs - (struct page *) array of size sh->disks | |
699 | * @sh - stripe_head to parse | |
700 | * | |
701 | * Populates srcs in proper layout order for the stripe and returns the | |
702 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
703 | * destination buffer is recorded in srcs[count] and the Q destination | |
704 | * is recorded in srcs[count+1]]. | |
705 | */ | |
706 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
707 | { | |
708 | int disks = sh->disks; | |
709 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
710 | int d0_idx = raid6_d0(sh); | |
711 | int count; | |
712 | int i; | |
713 | ||
714 | for (i = 0; i < disks; i++) | |
715 | srcs[i] = (void *)raid6_empty_zero_page; | |
716 | ||
717 | count = 0; | |
718 | i = d0_idx; | |
719 | do { | |
720 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
721 | ||
722 | srcs[slot] = sh->dev[i].page; | |
723 | i = raid6_next_disk(i, disks); | |
724 | } while (i != d0_idx); | |
725 | BUG_ON(count != syndrome_disks); | |
726 | ||
727 | return count; | |
728 | } | |
729 | ||
730 | static struct dma_async_tx_descriptor * | |
731 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
732 | { | |
733 | int disks = sh->disks; | |
734 | struct page **blocks = percpu->scribble; | |
735 | int target; | |
736 | int qd_idx = sh->qd_idx; | |
737 | struct dma_async_tx_descriptor *tx; | |
738 | struct async_submit_ctl submit; | |
739 | struct r5dev *tgt; | |
740 | struct page *dest; | |
741 | int i; | |
742 | int count; | |
743 | ||
744 | if (sh->ops.target < 0) | |
745 | target = sh->ops.target2; | |
746 | else if (sh->ops.target2 < 0) | |
747 | target = sh->ops.target; | |
748 | else | |
749 | /* we should only have one valid target */ | |
750 | BUG(); | |
751 | BUG_ON(target < 0); | |
752 | pr_debug("%s: stripe %llu block: %d\n", | |
753 | __func__, (unsigned long long)sh->sector, target); | |
754 | ||
755 | tgt = &sh->dev[target]; | |
756 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
757 | dest = tgt->page; | |
758 | ||
759 | atomic_inc(&sh->count); | |
760 | ||
761 | if (target == qd_idx) { | |
762 | count = set_syndrome_sources(blocks, sh); | |
763 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
764 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
765 | init_async_submit(&submit, 0, NULL, ops_complete_compute, sh, | |
766 | to_addr_conv(sh, percpu)); | |
767 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
768 | } else { | |
769 | /* Compute any data- or p-drive using XOR */ | |
770 | count = 0; | |
771 | for (i = disks; i-- ; ) { | |
772 | if (i == target || i == qd_idx) | |
773 | continue; | |
774 | blocks[count++] = sh->dev[i].page; | |
775 | } | |
776 | ||
777 | init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, | |
778 | ops_complete_compute, sh, | |
779 | to_addr_conv(sh, percpu)); | |
780 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
781 | } | |
782 | ||
783 | return tx; | |
784 | } | |
785 | ||
786 | static struct dma_async_tx_descriptor * | |
787 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
788 | { | |
789 | int i, count, disks = sh->disks; | |
790 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
791 | int d0_idx = raid6_d0(sh); | |
792 | int faila = -1, failb = -1; | |
793 | int target = sh->ops.target; | |
794 | int target2 = sh->ops.target2; | |
795 | struct r5dev *tgt = &sh->dev[target]; | |
796 | struct r5dev *tgt2 = &sh->dev[target2]; | |
797 | struct dma_async_tx_descriptor *tx; | |
798 | struct page **blocks = percpu->scribble; | |
799 | struct async_submit_ctl submit; | |
800 | ||
801 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
802 | __func__, (unsigned long long)sh->sector, target, target2); | |
803 | BUG_ON(target < 0 || target2 < 0); | |
804 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
805 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
806 | ||
807 | /* we need to open-code set_syndrome_sources to handle to the | |
808 | * slot number conversion for 'faila' and 'failb' | |
809 | */ | |
810 | for (i = 0; i < disks ; i++) | |
811 | blocks[i] = (void *)raid6_empty_zero_page; | |
812 | count = 0; | |
813 | i = d0_idx; | |
814 | do { | |
815 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
816 | ||
817 | blocks[slot] = sh->dev[i].page; | |
818 | ||
819 | if (i == target) | |
820 | faila = slot; | |
821 | if (i == target2) | |
822 | failb = slot; | |
823 | i = raid6_next_disk(i, disks); | |
824 | } while (i != d0_idx); | |
825 | BUG_ON(count != syndrome_disks); | |
826 | ||
827 | BUG_ON(faila == failb); | |
828 | if (failb < faila) | |
829 | swap(faila, failb); | |
830 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
831 | __func__, (unsigned long long)sh->sector, faila, failb); | |
832 | ||
833 | atomic_inc(&sh->count); | |
834 | ||
835 | if (failb == syndrome_disks+1) { | |
836 | /* Q disk is one of the missing disks */ | |
837 | if (faila == syndrome_disks) { | |
838 | /* Missing P+Q, just recompute */ | |
839 | init_async_submit(&submit, 0, NULL, ops_complete_compute, | |
840 | sh, to_addr_conv(sh, percpu)); | |
841 | return async_gen_syndrome(blocks, 0, count+2, | |
842 | STRIPE_SIZE, &submit); | |
843 | } else { | |
844 | struct page *dest; | |
845 | int data_target; | |
846 | int qd_idx = sh->qd_idx; | |
847 | ||
848 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
849 | if (target == qd_idx) | |
850 | data_target = target2; | |
851 | else | |
852 | data_target = target; | |
853 | ||
854 | count = 0; | |
855 | for (i = disks; i-- ; ) { | |
856 | if (i == data_target || i == qd_idx) | |
857 | continue; | |
858 | blocks[count++] = sh->dev[i].page; | |
859 | } | |
860 | dest = sh->dev[data_target].page; | |
861 | init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, | |
862 | NULL, NULL, to_addr_conv(sh, percpu)); | |
863 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, | |
864 | &submit); | |
865 | ||
866 | count = set_syndrome_sources(blocks, sh); | |
867 | init_async_submit(&submit, 0, tx, ops_complete_compute, | |
868 | sh, to_addr_conv(sh, percpu)); | |
869 | return async_gen_syndrome(blocks, 0, count+2, | |
870 | STRIPE_SIZE, &submit); | |
871 | } | |
872 | } | |
873 | ||
874 | init_async_submit(&submit, 0, NULL, ops_complete_compute, sh, | |
875 | to_addr_conv(sh, percpu)); | |
876 | if (failb == syndrome_disks) { | |
877 | /* We're missing D+P. */ | |
878 | return async_raid6_datap_recov(syndrome_disks+2, STRIPE_SIZE, | |
879 | faila, blocks, &submit); | |
880 | } else { | |
881 | /* We're missing D+D. */ | |
882 | return async_raid6_2data_recov(syndrome_disks+2, STRIPE_SIZE, | |
883 | faila, failb, blocks, &submit); | |
884 | } | |
885 | } | |
886 | ||
887 | ||
91c00924 DW |
888 | static void ops_complete_prexor(void *stripe_head_ref) |
889 | { | |
890 | struct stripe_head *sh = stripe_head_ref; | |
891 | ||
e46b272b | 892 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 893 | (unsigned long long)sh->sector); |
91c00924 DW |
894 | } |
895 | ||
896 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
897 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
898 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 899 | { |
91c00924 | 900 | int disks = sh->disks; |
d6f38f31 | 901 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 902 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 903 | struct async_submit_ctl submit; |
91c00924 DW |
904 | |
905 | /* existing parity data subtracted */ | |
906 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
907 | ||
e46b272b | 908 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
909 | (unsigned long long)sh->sector); |
910 | ||
911 | for (i = disks; i--; ) { | |
912 | struct r5dev *dev = &sh->dev[i]; | |
913 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 914 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
915 | xor_srcs[count++] = dev->page; |
916 | } | |
917 | ||
a08abd8c | 918 | init_async_submit(&submit, ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 919 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 920 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
921 | |
922 | return tx; | |
923 | } | |
924 | ||
925 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 926 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
927 | { |
928 | int disks = sh->disks; | |
d8ee0728 | 929 | int i; |
91c00924 | 930 | |
e46b272b | 931 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
932 | (unsigned long long)sh->sector); |
933 | ||
934 | for (i = disks; i--; ) { | |
935 | struct r5dev *dev = &sh->dev[i]; | |
936 | struct bio *chosen; | |
91c00924 | 937 | |
d8ee0728 | 938 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
939 | struct bio *wbi; |
940 | ||
941 | spin_lock(&sh->lock); | |
942 | chosen = dev->towrite; | |
943 | dev->towrite = NULL; | |
944 | BUG_ON(dev->written); | |
945 | wbi = dev->written = chosen; | |
946 | spin_unlock(&sh->lock); | |
947 | ||
948 | while (wbi && wbi->bi_sector < | |
949 | dev->sector + STRIPE_SECTORS) { | |
950 | tx = async_copy_data(1, wbi, dev->page, | |
951 | dev->sector, tx); | |
952 | wbi = r5_next_bio(wbi, dev->sector); | |
953 | } | |
954 | } | |
955 | } | |
956 | ||
957 | return tx; | |
958 | } | |
959 | ||
ac6b53b6 | 960 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
961 | { |
962 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
963 | int disks = sh->disks; |
964 | int pd_idx = sh->pd_idx; | |
965 | int qd_idx = sh->qd_idx; | |
966 | int i; | |
91c00924 | 967 | |
e46b272b | 968 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
969 | (unsigned long long)sh->sector); |
970 | ||
971 | for (i = disks; i--; ) { | |
972 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 DW |
973 | |
974 | if (dev->written || i == pd_idx || i == qd_idx) | |
91c00924 DW |
975 | set_bit(R5_UPTODATE, &dev->flags); |
976 | } | |
977 | ||
d8ee0728 DW |
978 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
979 | sh->reconstruct_state = reconstruct_state_drain_result; | |
980 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
981 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
982 | else { | |
983 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
984 | sh->reconstruct_state = reconstruct_state_result; | |
985 | } | |
91c00924 DW |
986 | |
987 | set_bit(STRIPE_HANDLE, &sh->state); | |
988 | release_stripe(sh); | |
989 | } | |
990 | ||
991 | static void | |
ac6b53b6 DW |
992 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
993 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 994 | { |
91c00924 | 995 | int disks = sh->disks; |
d6f38f31 | 996 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 997 | struct async_submit_ctl submit; |
91c00924 DW |
998 | int count = 0, pd_idx = sh->pd_idx, i; |
999 | struct page *xor_dest; | |
d8ee0728 | 1000 | int prexor = 0; |
91c00924 | 1001 | unsigned long flags; |
91c00924 | 1002 | |
e46b272b | 1003 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1004 | (unsigned long long)sh->sector); |
1005 | ||
1006 | /* check if prexor is active which means only process blocks | |
1007 | * that are part of a read-modify-write (written) | |
1008 | */ | |
d8ee0728 DW |
1009 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1010 | prexor = 1; | |
91c00924 DW |
1011 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1012 | for (i = disks; i--; ) { | |
1013 | struct r5dev *dev = &sh->dev[i]; | |
1014 | if (dev->written) | |
1015 | xor_srcs[count++] = dev->page; | |
1016 | } | |
1017 | } else { | |
1018 | xor_dest = sh->dev[pd_idx].page; | |
1019 | for (i = disks; i--; ) { | |
1020 | struct r5dev *dev = &sh->dev[i]; | |
1021 | if (i != pd_idx) | |
1022 | xor_srcs[count++] = dev->page; | |
1023 | } | |
1024 | } | |
1025 | ||
91c00924 DW |
1026 | /* 1/ if we prexor'd then the dest is reused as a source |
1027 | * 2/ if we did not prexor then we are redoing the parity | |
1028 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1029 | * for the synchronous xor case | |
1030 | */ | |
88ba2aa5 | 1031 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1032 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1033 | ||
1034 | atomic_inc(&sh->count); | |
1035 | ||
ac6b53b6 | 1036 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1037 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1038 | if (unlikely(count == 1)) |
1039 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1040 | else | |
1041 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1042 | } |
1043 | ||
ac6b53b6 DW |
1044 | static void |
1045 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1046 | struct dma_async_tx_descriptor *tx) | |
1047 | { | |
1048 | struct async_submit_ctl submit; | |
1049 | struct page **blocks = percpu->scribble; | |
1050 | int count; | |
1051 | ||
1052 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1053 | ||
1054 | count = set_syndrome_sources(blocks, sh); | |
1055 | ||
1056 | atomic_inc(&sh->count); | |
1057 | ||
1058 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1059 | sh, to_addr_conv(sh, percpu)); | |
1060 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1061 | } | |
1062 | ||
91c00924 DW |
1063 | static void ops_complete_check(void *stripe_head_ref) |
1064 | { | |
1065 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1066 | |
e46b272b | 1067 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1068 | (unsigned long long)sh->sector); |
1069 | ||
ecc65c9b | 1070 | sh->check_state = check_state_check_result; |
91c00924 DW |
1071 | set_bit(STRIPE_HANDLE, &sh->state); |
1072 | release_stripe(sh); | |
1073 | } | |
1074 | ||
ac6b53b6 | 1075 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1076 | { |
91c00924 | 1077 | int disks = sh->disks; |
ac6b53b6 DW |
1078 | int pd_idx = sh->pd_idx; |
1079 | int qd_idx = sh->qd_idx; | |
1080 | struct page *xor_dest; | |
d6f38f31 | 1081 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1082 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1083 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1084 | int count; |
1085 | int i; | |
91c00924 | 1086 | |
e46b272b | 1087 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1088 | (unsigned long long)sh->sector); |
1089 | ||
ac6b53b6 DW |
1090 | count = 0; |
1091 | xor_dest = sh->dev[pd_idx].page; | |
1092 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1093 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1094 | if (i == pd_idx || i == qd_idx) |
1095 | continue; | |
1096 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1097 | } |
1098 | ||
d6f38f31 DW |
1099 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1100 | to_addr_conv(sh, percpu)); | |
099f53cb | 1101 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1102 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1103 | |
91c00924 | 1104 | atomic_inc(&sh->count); |
a08abd8c DW |
1105 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1106 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1107 | } |
1108 | ||
ac6b53b6 DW |
1109 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1110 | { | |
1111 | struct page **srcs = percpu->scribble; | |
1112 | struct async_submit_ctl submit; | |
1113 | int count; | |
1114 | ||
1115 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1116 | (unsigned long long)sh->sector, checkp); | |
1117 | ||
1118 | count = set_syndrome_sources(srcs, sh); | |
1119 | if (!checkp) | |
1120 | srcs[count] = NULL; | |
1121 | ||
1122 | atomic_inc(&sh->count); | |
1123 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, | |
1124 | sh, to_addr_conv(sh, percpu)); | |
1125 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1126 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
1127 | } | |
1128 | ||
1129 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
91c00924 DW |
1130 | { |
1131 | int overlap_clear = 0, i, disks = sh->disks; | |
1132 | struct dma_async_tx_descriptor *tx = NULL; | |
d6f38f31 | 1133 | raid5_conf_t *conf = sh->raid_conf; |
ac6b53b6 | 1134 | int level = conf->level; |
d6f38f31 DW |
1135 | struct raid5_percpu *percpu; |
1136 | unsigned long cpu; | |
91c00924 | 1137 | |
d6f38f31 DW |
1138 | cpu = get_cpu(); |
1139 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1140 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1141 | ops_run_biofill(sh); |
1142 | overlap_clear++; | |
1143 | } | |
1144 | ||
7b3a871e | 1145 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1146 | if (level < 6) |
1147 | tx = ops_run_compute5(sh, percpu); | |
1148 | else { | |
1149 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1150 | tx = ops_run_compute6_1(sh, percpu); | |
1151 | else | |
1152 | tx = ops_run_compute6_2(sh, percpu); | |
1153 | } | |
1154 | /* terminate the chain if reconstruct is not set to be run */ | |
1155 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1156 | async_tx_ack(tx); |
1157 | } | |
91c00924 | 1158 | |
600aa109 | 1159 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1160 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1161 | |
600aa109 | 1162 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1163 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1164 | overlap_clear++; |
1165 | } | |
1166 | ||
ac6b53b6 DW |
1167 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1168 | if (level < 6) | |
1169 | ops_run_reconstruct5(sh, percpu, tx); | |
1170 | else | |
1171 | ops_run_reconstruct6(sh, percpu, tx); | |
1172 | } | |
91c00924 | 1173 | |
ac6b53b6 DW |
1174 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1175 | if (sh->check_state == check_state_run) | |
1176 | ops_run_check_p(sh, percpu); | |
1177 | else if (sh->check_state == check_state_run_q) | |
1178 | ops_run_check_pq(sh, percpu, 0); | |
1179 | else if (sh->check_state == check_state_run_pq) | |
1180 | ops_run_check_pq(sh, percpu, 1); | |
1181 | else | |
1182 | BUG(); | |
1183 | } | |
91c00924 | 1184 | |
91c00924 DW |
1185 | if (overlap_clear) |
1186 | for (i = disks; i--; ) { | |
1187 | struct r5dev *dev = &sh->dev[i]; | |
1188 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1189 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1190 | } | |
d6f38f31 | 1191 | put_cpu(); |
91c00924 DW |
1192 | } |
1193 | ||
3f294f4f | 1194 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1195 | { |
1196 | struct stripe_head *sh; | |
3f294f4f N |
1197 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
1198 | if (!sh) | |
1199 | return 0; | |
1200 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
1201 | sh->raid_conf = conf; | |
1202 | spin_lock_init(&sh->lock); | |
1203 | ||
1204 | if (grow_buffers(sh, conf->raid_disks)) { | |
1205 | shrink_buffers(sh, conf->raid_disks); | |
1206 | kmem_cache_free(conf->slab_cache, sh); | |
1207 | return 0; | |
1208 | } | |
7ecaa1e6 | 1209 | sh->disks = conf->raid_disks; |
3f294f4f N |
1210 | /* we just created an active stripe so... */ |
1211 | atomic_set(&sh->count, 1); | |
1212 | atomic_inc(&conf->active_stripes); | |
1213 | INIT_LIST_HEAD(&sh->lru); | |
1214 | release_stripe(sh); | |
1215 | return 1; | |
1216 | } | |
1217 | ||
1218 | static int grow_stripes(raid5_conf_t *conf, int num) | |
1219 | { | |
e18b890b | 1220 | struct kmem_cache *sc; |
1da177e4 LT |
1221 | int devs = conf->raid_disks; |
1222 | ||
245f46c2 N |
1223 | sprintf(conf->cache_name[0], |
1224 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1225 | sprintf(conf->cache_name[1], | |
1226 | "raid%d-%s-alt", conf->level, mdname(conf->mddev)); | |
ad01c9e3 N |
1227 | conf->active_name = 0; |
1228 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1229 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1230 | 0, 0, NULL); |
1da177e4 LT |
1231 | if (!sc) |
1232 | return 1; | |
1233 | conf->slab_cache = sc; | |
ad01c9e3 | 1234 | conf->pool_size = devs; |
16a53ecc | 1235 | while (num--) |
3f294f4f | 1236 | if (!grow_one_stripe(conf)) |
1da177e4 | 1237 | return 1; |
1da177e4 LT |
1238 | return 0; |
1239 | } | |
29269553 | 1240 | |
d6f38f31 DW |
1241 | /** |
1242 | * scribble_len - return the required size of the scribble region | |
1243 | * @num - total number of disks in the array | |
1244 | * | |
1245 | * The size must be enough to contain: | |
1246 | * 1/ a struct page pointer for each device in the array +2 | |
1247 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1248 | * (dma_map_page()) or page (page_address()) address. | |
1249 | * | |
1250 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1251 | * calculate over all devices (not just the data blocks), using zeros in place | |
1252 | * of the P and Q blocks. | |
1253 | */ | |
1254 | static size_t scribble_len(int num) | |
1255 | { | |
1256 | size_t len; | |
1257 | ||
1258 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1259 | ||
1260 | return len; | |
1261 | } | |
1262 | ||
ad01c9e3 N |
1263 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
1264 | { | |
1265 | /* Make all the stripes able to hold 'newsize' devices. | |
1266 | * New slots in each stripe get 'page' set to a new page. | |
1267 | * | |
1268 | * This happens in stages: | |
1269 | * 1/ create a new kmem_cache and allocate the required number of | |
1270 | * stripe_heads. | |
1271 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1272 | * to the new stripe_heads. This will have the side effect of | |
1273 | * freezing the array as once all stripe_heads have been collected, | |
1274 | * no IO will be possible. Old stripe heads are freed once their | |
1275 | * pages have been transferred over, and the old kmem_cache is | |
1276 | * freed when all stripes are done. | |
1277 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1278 | * we simple return a failre status - no need to clean anything up. | |
1279 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1280 | * If this fails, we don't bother trying the shrink the | |
1281 | * stripe_heads down again, we just leave them as they are. | |
1282 | * As each stripe_head is processed the new one is released into | |
1283 | * active service. | |
1284 | * | |
1285 | * Once step2 is started, we cannot afford to wait for a write, | |
1286 | * so we use GFP_NOIO allocations. | |
1287 | */ | |
1288 | struct stripe_head *osh, *nsh; | |
1289 | LIST_HEAD(newstripes); | |
1290 | struct disk_info *ndisks; | |
d6f38f31 | 1291 | unsigned long cpu; |
b5470dc5 | 1292 | int err; |
e18b890b | 1293 | struct kmem_cache *sc; |
ad01c9e3 N |
1294 | int i; |
1295 | ||
1296 | if (newsize <= conf->pool_size) | |
1297 | return 0; /* never bother to shrink */ | |
1298 | ||
b5470dc5 DW |
1299 | err = md_allow_write(conf->mddev); |
1300 | if (err) | |
1301 | return err; | |
2a2275d6 | 1302 | |
ad01c9e3 N |
1303 | /* Step 1 */ |
1304 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1305 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1306 | 0, 0, NULL); |
ad01c9e3 N |
1307 | if (!sc) |
1308 | return -ENOMEM; | |
1309 | ||
1310 | for (i = conf->max_nr_stripes; i; i--) { | |
1311 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
1312 | if (!nsh) | |
1313 | break; | |
1314 | ||
1315 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
1316 | ||
1317 | nsh->raid_conf = conf; | |
1318 | spin_lock_init(&nsh->lock); | |
1319 | ||
1320 | list_add(&nsh->lru, &newstripes); | |
1321 | } | |
1322 | if (i) { | |
1323 | /* didn't get enough, give up */ | |
1324 | while (!list_empty(&newstripes)) { | |
1325 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1326 | list_del(&nsh->lru); | |
1327 | kmem_cache_free(sc, nsh); | |
1328 | } | |
1329 | kmem_cache_destroy(sc); | |
1330 | return -ENOMEM; | |
1331 | } | |
1332 | /* Step 2 - Must use GFP_NOIO now. | |
1333 | * OK, we have enough stripes, start collecting inactive | |
1334 | * stripes and copying them over | |
1335 | */ | |
1336 | list_for_each_entry(nsh, &newstripes, lru) { | |
1337 | spin_lock_irq(&conf->device_lock); | |
1338 | wait_event_lock_irq(conf->wait_for_stripe, | |
1339 | !list_empty(&conf->inactive_list), | |
1340 | conf->device_lock, | |
b3b46be3 | 1341 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1342 | ); |
1343 | osh = get_free_stripe(conf); | |
1344 | spin_unlock_irq(&conf->device_lock); | |
1345 | atomic_set(&nsh->count, 1); | |
1346 | for(i=0; i<conf->pool_size; i++) | |
1347 | nsh->dev[i].page = osh->dev[i].page; | |
1348 | for( ; i<newsize; i++) | |
1349 | nsh->dev[i].page = NULL; | |
1350 | kmem_cache_free(conf->slab_cache, osh); | |
1351 | } | |
1352 | kmem_cache_destroy(conf->slab_cache); | |
1353 | ||
1354 | /* Step 3. | |
1355 | * At this point, we are holding all the stripes so the array | |
1356 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1357 | * conf->disks and the scribble region |
ad01c9e3 N |
1358 | */ |
1359 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1360 | if (ndisks) { | |
1361 | for (i=0; i<conf->raid_disks; i++) | |
1362 | ndisks[i] = conf->disks[i]; | |
1363 | kfree(conf->disks); | |
1364 | conf->disks = ndisks; | |
1365 | } else | |
1366 | err = -ENOMEM; | |
1367 | ||
d6f38f31 DW |
1368 | get_online_cpus(); |
1369 | conf->scribble_len = scribble_len(newsize); | |
1370 | for_each_present_cpu(cpu) { | |
1371 | struct raid5_percpu *percpu; | |
1372 | void *scribble; | |
1373 | ||
1374 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1375 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1376 | ||
1377 | if (scribble) { | |
1378 | kfree(percpu->scribble); | |
1379 | percpu->scribble = scribble; | |
1380 | } else { | |
1381 | err = -ENOMEM; | |
1382 | break; | |
1383 | } | |
1384 | } | |
1385 | put_online_cpus(); | |
1386 | ||
ad01c9e3 N |
1387 | /* Step 4, return new stripes to service */ |
1388 | while(!list_empty(&newstripes)) { | |
1389 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1390 | list_del_init(&nsh->lru); | |
d6f38f31 | 1391 | |
ad01c9e3 N |
1392 | for (i=conf->raid_disks; i < newsize; i++) |
1393 | if (nsh->dev[i].page == NULL) { | |
1394 | struct page *p = alloc_page(GFP_NOIO); | |
1395 | nsh->dev[i].page = p; | |
1396 | if (!p) | |
1397 | err = -ENOMEM; | |
1398 | } | |
1399 | release_stripe(nsh); | |
1400 | } | |
1401 | /* critical section pass, GFP_NOIO no longer needed */ | |
1402 | ||
1403 | conf->slab_cache = sc; | |
1404 | conf->active_name = 1-conf->active_name; | |
1405 | conf->pool_size = newsize; | |
1406 | return err; | |
1407 | } | |
1da177e4 | 1408 | |
3f294f4f | 1409 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1410 | { |
1411 | struct stripe_head *sh; | |
1412 | ||
3f294f4f N |
1413 | spin_lock_irq(&conf->device_lock); |
1414 | sh = get_free_stripe(conf); | |
1415 | spin_unlock_irq(&conf->device_lock); | |
1416 | if (!sh) | |
1417 | return 0; | |
78bafebd | 1418 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1419 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1420 | kmem_cache_free(conf->slab_cache, sh); |
1421 | atomic_dec(&conf->active_stripes); | |
1422 | return 1; | |
1423 | } | |
1424 | ||
1425 | static void shrink_stripes(raid5_conf_t *conf) | |
1426 | { | |
1427 | while (drop_one_stripe(conf)) | |
1428 | ; | |
1429 | ||
29fc7e3e N |
1430 | if (conf->slab_cache) |
1431 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1432 | conf->slab_cache = NULL; |
1433 | } | |
1434 | ||
6712ecf8 | 1435 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1436 | { |
99c0fb5f | 1437 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1438 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1439 | int disks = sh->disks, i; |
1da177e4 | 1440 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1441 | char b[BDEVNAME_SIZE]; |
1442 | mdk_rdev_t *rdev; | |
1da177e4 | 1443 | |
1da177e4 LT |
1444 | |
1445 | for (i=0 ; i<disks; i++) | |
1446 | if (bi == &sh->dev[i].req) | |
1447 | break; | |
1448 | ||
45b4233c DW |
1449 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1450 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1451 | uptodate); |
1452 | if (i == disks) { | |
1453 | BUG(); | |
6712ecf8 | 1454 | return; |
1da177e4 LT |
1455 | } |
1456 | ||
1457 | if (uptodate) { | |
1da177e4 | 1458 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1459 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1460 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1461 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1462 | " (%lu sectors at %llu on %s)\n", | |
1463 | mdname(conf->mddev), STRIPE_SECTORS, | |
1464 | (unsigned long long)(sh->sector | |
1465 | + rdev->data_offset), | |
1466 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1467 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1468 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1469 | } | |
ba22dcbf N |
1470 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1471 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1472 | } else { |
d6950432 | 1473 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1474 | int retry = 0; |
d6950432 N |
1475 | rdev = conf->disks[i].rdev; |
1476 | ||
1da177e4 | 1477 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1478 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1479 | if (conf->mddev->degraded) |
6be9d494 BS |
1480 | printk_rl(KERN_WARNING |
1481 | "raid5:%s: read error not correctable " | |
1482 | "(sector %llu on %s).\n", | |
1483 | mdname(conf->mddev), | |
1484 | (unsigned long long)(sh->sector | |
1485 | + rdev->data_offset), | |
1486 | bdn); | |
ba22dcbf | 1487 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1488 | /* Oh, no!!! */ |
6be9d494 BS |
1489 | printk_rl(KERN_WARNING |
1490 | "raid5:%s: read error NOT corrected!! " | |
1491 | "(sector %llu on %s).\n", | |
1492 | mdname(conf->mddev), | |
1493 | (unsigned long long)(sh->sector | |
1494 | + rdev->data_offset), | |
1495 | bdn); | |
d6950432 | 1496 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1497 | > conf->max_nr_stripes) |
14f8d26b | 1498 | printk(KERN_WARNING |
d6950432 N |
1499 | "raid5:%s: Too many read errors, failing device %s.\n", |
1500 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1501 | else |
1502 | retry = 1; | |
1503 | if (retry) | |
1504 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1505 | else { | |
4e5314b5 N |
1506 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1507 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1508 | md_error(conf->mddev, rdev); |
ba22dcbf | 1509 | } |
1da177e4 LT |
1510 | } |
1511 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1512 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1513 | set_bit(STRIPE_HANDLE, &sh->state); | |
1514 | release_stripe(sh); | |
1da177e4 LT |
1515 | } |
1516 | ||
d710e138 | 1517 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1518 | { |
99c0fb5f | 1519 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1520 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1521 | int disks = sh->disks, i; |
1da177e4 LT |
1522 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1523 | ||
1da177e4 LT |
1524 | for (i=0 ; i<disks; i++) |
1525 | if (bi == &sh->dev[i].req) | |
1526 | break; | |
1527 | ||
45b4233c | 1528 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1529 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1530 | uptodate); | |
1531 | if (i == disks) { | |
1532 | BUG(); | |
6712ecf8 | 1533 | return; |
1da177e4 LT |
1534 | } |
1535 | ||
1da177e4 LT |
1536 | if (!uptodate) |
1537 | md_error(conf->mddev, conf->disks[i].rdev); | |
1538 | ||
1539 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1540 | ||
1541 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1542 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1543 | release_stripe(sh); |
1da177e4 LT |
1544 | } |
1545 | ||
1546 | ||
784052ec | 1547 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1548 | |
784052ec | 1549 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1550 | { |
1551 | struct r5dev *dev = &sh->dev[i]; | |
1552 | ||
1553 | bio_init(&dev->req); | |
1554 | dev->req.bi_io_vec = &dev->vec; | |
1555 | dev->req.bi_vcnt++; | |
1556 | dev->req.bi_max_vecs++; | |
1557 | dev->vec.bv_page = dev->page; | |
1558 | dev->vec.bv_len = STRIPE_SIZE; | |
1559 | dev->vec.bv_offset = 0; | |
1560 | ||
1561 | dev->req.bi_sector = sh->sector; | |
1562 | dev->req.bi_private = sh; | |
1563 | ||
1564 | dev->flags = 0; | |
784052ec | 1565 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1566 | } |
1567 | ||
1568 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1569 | { | |
1570 | char b[BDEVNAME_SIZE]; | |
1571 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1572 | pr_debug("raid5: error called\n"); |
1da177e4 | 1573 | |
b2d444d7 | 1574 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1575 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1576 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1577 | unsigned long flags; | |
1578 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1579 | mddev->degraded++; |
c04be0aa | 1580 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1581 | /* |
1582 | * if recovery was running, make sure it aborts. | |
1583 | */ | |
dfc70645 | 1584 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1585 | } |
b2d444d7 | 1586 | set_bit(Faulty, &rdev->flags); |
d710e138 N |
1587 | printk(KERN_ALERT |
1588 | "raid5: Disk failure on %s, disabling device.\n" | |
1589 | "raid5: Operation continuing on %d devices.\n", | |
1590 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); | |
1da177e4 | 1591 | } |
16a53ecc | 1592 | } |
1da177e4 LT |
1593 | |
1594 | /* | |
1595 | * Input: a 'big' sector number, | |
1596 | * Output: index of the data and parity disk, and the sector # in them. | |
1597 | */ | |
112bf897 | 1598 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1599 | int previous, int *dd_idx, |
1600 | struct stripe_head *sh) | |
1da177e4 LT |
1601 | { |
1602 | long stripe; | |
1603 | unsigned long chunk_number; | |
1604 | unsigned int chunk_offset; | |
911d4ee8 | 1605 | int pd_idx, qd_idx; |
67cc2b81 | 1606 | int ddf_layout = 0; |
1da177e4 | 1607 | sector_t new_sector; |
e183eaed N |
1608 | int algorithm = previous ? conf->prev_algo |
1609 | : conf->algorithm; | |
784052ec N |
1610 | int sectors_per_chunk = previous ? (conf->prev_chunk >> 9) |
1611 | : (conf->chunk_size >> 9); | |
112bf897 N |
1612 | int raid_disks = previous ? conf->previous_raid_disks |
1613 | : conf->raid_disks; | |
1614 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1615 | |
1616 | /* First compute the information on this sector */ | |
1617 | ||
1618 | /* | |
1619 | * Compute the chunk number and the sector offset inside the chunk | |
1620 | */ | |
1621 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1622 | chunk_number = r_sector; | |
1623 | BUG_ON(r_sector != chunk_number); | |
1624 | ||
1625 | /* | |
1626 | * Compute the stripe number | |
1627 | */ | |
1628 | stripe = chunk_number / data_disks; | |
1629 | ||
1630 | /* | |
1631 | * Compute the data disk and parity disk indexes inside the stripe | |
1632 | */ | |
1633 | *dd_idx = chunk_number % data_disks; | |
1634 | ||
1635 | /* | |
1636 | * Select the parity disk based on the user selected algorithm. | |
1637 | */ | |
911d4ee8 | 1638 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1639 | switch(conf->level) { |
1640 | case 4: | |
911d4ee8 | 1641 | pd_idx = data_disks; |
16a53ecc N |
1642 | break; |
1643 | case 5: | |
e183eaed | 1644 | switch (algorithm) { |
1da177e4 | 1645 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1646 | pd_idx = data_disks - stripe % raid_disks; |
1647 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1648 | (*dd_idx)++; |
1649 | break; | |
1650 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1651 | pd_idx = stripe % raid_disks; |
1652 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1653 | (*dd_idx)++; |
1654 | break; | |
1655 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1656 | pd_idx = data_disks - stripe % raid_disks; |
1657 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 LT |
1658 | break; |
1659 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1660 | pd_idx = stripe % raid_disks; |
1661 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 | 1662 | break; |
99c0fb5f N |
1663 | case ALGORITHM_PARITY_0: |
1664 | pd_idx = 0; | |
1665 | (*dd_idx)++; | |
1666 | break; | |
1667 | case ALGORITHM_PARITY_N: | |
1668 | pd_idx = data_disks; | |
1669 | break; | |
1da177e4 | 1670 | default: |
14f8d26b | 1671 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
e183eaed | 1672 | algorithm); |
99c0fb5f | 1673 | BUG(); |
16a53ecc N |
1674 | } |
1675 | break; | |
1676 | case 6: | |
1677 | ||
e183eaed | 1678 | switch (algorithm) { |
16a53ecc | 1679 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1680 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1681 | qd_idx = pd_idx + 1; | |
1682 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1683 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1684 | qd_idx = 0; |
1685 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1686 | (*dd_idx) += 2; /* D D P Q D */ |
1687 | break; | |
1688 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1689 | pd_idx = stripe % raid_disks; |
1690 | qd_idx = pd_idx + 1; | |
1691 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1692 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1693 | qd_idx = 0; |
1694 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1695 | (*dd_idx) += 2; /* D D P Q D */ |
1696 | break; | |
1697 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1698 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1699 | qd_idx = (pd_idx + 1) % raid_disks; | |
1700 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1701 | break; |
1702 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1703 | pd_idx = stripe % raid_disks; |
1704 | qd_idx = (pd_idx + 1) % raid_disks; | |
1705 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1706 | break; |
99c0fb5f N |
1707 | |
1708 | case ALGORITHM_PARITY_0: | |
1709 | pd_idx = 0; | |
1710 | qd_idx = 1; | |
1711 | (*dd_idx) += 2; | |
1712 | break; | |
1713 | case ALGORITHM_PARITY_N: | |
1714 | pd_idx = data_disks; | |
1715 | qd_idx = data_disks + 1; | |
1716 | break; | |
1717 | ||
1718 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1719 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1720 | * of blocks for computing Q is different. | |
1721 | */ | |
1722 | pd_idx = stripe % raid_disks; | |
1723 | qd_idx = pd_idx + 1; | |
1724 | if (pd_idx == raid_disks-1) { | |
1725 | (*dd_idx)++; /* Q D D D P */ | |
1726 | qd_idx = 0; | |
1727 | } else if (*dd_idx >= pd_idx) | |
1728 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1729 | ddf_layout = 1; |
99c0fb5f N |
1730 | break; |
1731 | ||
1732 | case ALGORITHM_ROTATING_N_RESTART: | |
1733 | /* Same a left_asymmetric, by first stripe is | |
1734 | * D D D P Q rather than | |
1735 | * Q D D D P | |
1736 | */ | |
1737 | pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
1738 | qd_idx = pd_idx + 1; | |
1739 | if (pd_idx == raid_disks-1) { | |
1740 | (*dd_idx)++; /* Q D D D P */ | |
1741 | qd_idx = 0; | |
1742 | } else if (*dd_idx >= pd_idx) | |
1743 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1744 | ddf_layout = 1; |
99c0fb5f N |
1745 | break; |
1746 | ||
1747 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1748 | /* Same as left_symmetric but Q is before P */ | |
1749 | pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1750 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; | |
1751 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1752 | ddf_layout = 1; |
99c0fb5f N |
1753 | break; |
1754 | ||
1755 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1756 | /* RAID5 left_asymmetric, with Q on last device */ | |
1757 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1758 | if (*dd_idx >= pd_idx) | |
1759 | (*dd_idx)++; | |
1760 | qd_idx = raid_disks - 1; | |
1761 | break; | |
1762 | ||
1763 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1764 | pd_idx = stripe % (raid_disks-1); | |
1765 | if (*dd_idx >= pd_idx) | |
1766 | (*dd_idx)++; | |
1767 | qd_idx = raid_disks - 1; | |
1768 | break; | |
1769 | ||
1770 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1771 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1772 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1773 | qd_idx = raid_disks - 1; | |
1774 | break; | |
1775 | ||
1776 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1777 | pd_idx = stripe % (raid_disks-1); | |
1778 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1779 | qd_idx = raid_disks - 1; | |
1780 | break; | |
1781 | ||
1782 | case ALGORITHM_PARITY_0_6: | |
1783 | pd_idx = 0; | |
1784 | (*dd_idx)++; | |
1785 | qd_idx = raid_disks - 1; | |
1786 | break; | |
1787 | ||
1788 | ||
16a53ecc | 1789 | default: |
d710e138 | 1790 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
e183eaed | 1791 | algorithm); |
99c0fb5f | 1792 | BUG(); |
16a53ecc N |
1793 | } |
1794 | break; | |
1da177e4 LT |
1795 | } |
1796 | ||
911d4ee8 N |
1797 | if (sh) { |
1798 | sh->pd_idx = pd_idx; | |
1799 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1800 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1801 | } |
1da177e4 LT |
1802 | /* |
1803 | * Finally, compute the new sector number | |
1804 | */ | |
1805 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1806 | return new_sector; | |
1807 | } | |
1808 | ||
1809 | ||
784052ec | 1810 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1811 | { |
1812 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1813 | int raid_disks = sh->disks; |
1814 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1815 | sector_t new_sector = sh->sector, check; |
784052ec N |
1816 | int sectors_per_chunk = previous ? (conf->prev_chunk >> 9) |
1817 | : (conf->chunk_size >> 9); | |
e183eaed N |
1818 | int algorithm = previous ? conf->prev_algo |
1819 | : conf->algorithm; | |
1da177e4 LT |
1820 | sector_t stripe; |
1821 | int chunk_offset; | |
911d4ee8 | 1822 | int chunk_number, dummy1, dd_idx = i; |
1da177e4 | 1823 | sector_t r_sector; |
911d4ee8 | 1824 | struct stripe_head sh2; |
1da177e4 | 1825 | |
16a53ecc | 1826 | |
1da177e4 LT |
1827 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1828 | stripe = new_sector; | |
1829 | BUG_ON(new_sector != stripe); | |
1830 | ||
16a53ecc N |
1831 | if (i == sh->pd_idx) |
1832 | return 0; | |
1833 | switch(conf->level) { | |
1834 | case 4: break; | |
1835 | case 5: | |
e183eaed | 1836 | switch (algorithm) { |
1da177e4 LT |
1837 | case ALGORITHM_LEFT_ASYMMETRIC: |
1838 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1839 | if (i > sh->pd_idx) | |
1840 | i--; | |
1841 | break; | |
1842 | case ALGORITHM_LEFT_SYMMETRIC: | |
1843 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1844 | if (i < sh->pd_idx) | |
1845 | i += raid_disks; | |
1846 | i -= (sh->pd_idx + 1); | |
1847 | break; | |
99c0fb5f N |
1848 | case ALGORITHM_PARITY_0: |
1849 | i -= 1; | |
1850 | break; | |
1851 | case ALGORITHM_PARITY_N: | |
1852 | break; | |
1da177e4 | 1853 | default: |
14f8d26b | 1854 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
e183eaed | 1855 | algorithm); |
99c0fb5f | 1856 | BUG(); |
16a53ecc N |
1857 | } |
1858 | break; | |
1859 | case 6: | |
d0dabf7e | 1860 | if (i == sh->qd_idx) |
16a53ecc | 1861 | return 0; /* It is the Q disk */ |
e183eaed | 1862 | switch (algorithm) { |
16a53ecc N |
1863 | case ALGORITHM_LEFT_ASYMMETRIC: |
1864 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1865 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1866 | case ALGORITHM_ROTATING_N_RESTART: | |
1867 | if (sh->pd_idx == raid_disks-1) | |
1868 | i--; /* Q D D D P */ | |
16a53ecc N |
1869 | else if (i > sh->pd_idx) |
1870 | i -= 2; /* D D P Q D */ | |
1871 | break; | |
1872 | case ALGORITHM_LEFT_SYMMETRIC: | |
1873 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1874 | if (sh->pd_idx == raid_disks-1) | |
1875 | i--; /* Q D D D P */ | |
1876 | else { | |
1877 | /* D D P Q D */ | |
1878 | if (i < sh->pd_idx) | |
1879 | i += raid_disks; | |
1880 | i -= (sh->pd_idx + 2); | |
1881 | } | |
1882 | break; | |
99c0fb5f N |
1883 | case ALGORITHM_PARITY_0: |
1884 | i -= 2; | |
1885 | break; | |
1886 | case ALGORITHM_PARITY_N: | |
1887 | break; | |
1888 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1889 | if (sh->pd_idx == 0) | |
1890 | i--; /* P D D D Q */ | |
1891 | else if (i > sh->pd_idx) | |
1892 | i -= 2; /* D D Q P D */ | |
1893 | break; | |
1894 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1895 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1896 | if (i > sh->pd_idx) | |
1897 | i--; | |
1898 | break; | |
1899 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1900 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1901 | if (i < sh->pd_idx) | |
1902 | i += data_disks + 1; | |
1903 | i -= (sh->pd_idx + 1); | |
1904 | break; | |
1905 | case ALGORITHM_PARITY_0_6: | |
1906 | i -= 1; | |
1907 | break; | |
16a53ecc | 1908 | default: |
d710e138 | 1909 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
e183eaed | 1910 | algorithm); |
99c0fb5f | 1911 | BUG(); |
16a53ecc N |
1912 | } |
1913 | break; | |
1da177e4 LT |
1914 | } |
1915 | ||
1916 | chunk_number = stripe * data_disks + i; | |
1917 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1918 | ||
112bf897 | 1919 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 1920 | previous, &dummy1, &sh2); |
911d4ee8 N |
1921 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
1922 | || sh2.qd_idx != sh->qd_idx) { | |
14f8d26b | 1923 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1924 | return 0; |
1925 | } | |
1926 | return r_sector; | |
1927 | } | |
1928 | ||
1929 | ||
1930 | ||
1931 | /* | |
16a53ecc N |
1932 | * Copy data between a page in the stripe cache, and one or more bion |
1933 | * The page could align with the middle of the bio, or there could be | |
1934 | * several bion, each with several bio_vecs, which cover part of the page | |
1935 | * Multiple bion are linked together on bi_next. There may be extras | |
1936 | * at the end of this list. We ignore them. | |
1da177e4 LT |
1937 | */ |
1938 | static void copy_data(int frombio, struct bio *bio, | |
1939 | struct page *page, | |
1940 | sector_t sector) | |
1941 | { | |
1942 | char *pa = page_address(page); | |
1943 | struct bio_vec *bvl; | |
1944 | int i; | |
1945 | int page_offset; | |
1946 | ||
1947 | if (bio->bi_sector >= sector) | |
1948 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
1949 | else | |
1950 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
1951 | bio_for_each_segment(bvl, bio, i) { | |
1952 | int len = bio_iovec_idx(bio,i)->bv_len; | |
1953 | int clen; | |
1954 | int b_offset = 0; | |
1955 | ||
1956 | if (page_offset < 0) { | |
1957 | b_offset = -page_offset; | |
1958 | page_offset += b_offset; | |
1959 | len -= b_offset; | |
1960 | } | |
1961 | ||
1962 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1963 | clen = STRIPE_SIZE - page_offset; | |
1964 | else clen = len; | |
16a53ecc | 1965 | |
1da177e4 LT |
1966 | if (clen > 0) { |
1967 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
1968 | if (frombio) | |
1969 | memcpy(pa+page_offset, ba+b_offset, clen); | |
1970 | else | |
1971 | memcpy(ba+b_offset, pa+page_offset, clen); | |
1972 | __bio_kunmap_atomic(ba, KM_USER0); | |
1973 | } | |
1974 | if (clen < len) /* hit end of page */ | |
1975 | break; | |
1976 | page_offset += len; | |
1977 | } | |
1978 | } | |
1979 | ||
9bc89cd8 DW |
1980 | #define check_xor() do { \ |
1981 | if (count == MAX_XOR_BLOCKS) { \ | |
1982 | xor_blocks(count, STRIPE_SIZE, dest, ptr);\ | |
1983 | count = 0; \ | |
1984 | } \ | |
1da177e4 LT |
1985 | } while(0) |
1986 | ||
16a53ecc N |
1987 | static void compute_parity6(struct stripe_head *sh, int method) |
1988 | { | |
bff61975 | 1989 | raid5_conf_t *conf = sh->raid_conf; |
d0dabf7e | 1990 | int i, pd_idx, qd_idx, d0_idx, disks = sh->disks, count; |
67cc2b81 | 1991 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); |
16a53ecc N |
1992 | struct bio *chosen; |
1993 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
67cc2b81 | 1994 | void *ptrs[syndrome_disks+2]; |
16a53ecc | 1995 | |
d0dabf7e N |
1996 | pd_idx = sh->pd_idx; |
1997 | qd_idx = sh->qd_idx; | |
1998 | d0_idx = raid6_d0(sh); | |
16a53ecc | 1999 | |
45b4233c | 2000 | pr_debug("compute_parity, stripe %llu, method %d\n", |
16a53ecc N |
2001 | (unsigned long long)sh->sector, method); |
2002 | ||
2003 | switch(method) { | |
2004 | case READ_MODIFY_WRITE: | |
2005 | BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ | |
2006 | case RECONSTRUCT_WRITE: | |
2007 | for (i= disks; i-- ;) | |
2008 | if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { | |
2009 | chosen = sh->dev[i].towrite; | |
2010 | sh->dev[i].towrite = NULL; | |
2011 | ||
2012 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2013 | wake_up(&conf->wait_for_overlap); | |
2014 | ||
52e5f9d1 | 2015 | BUG_ON(sh->dev[i].written); |
16a53ecc N |
2016 | sh->dev[i].written = chosen; |
2017 | } | |
2018 | break; | |
2019 | case CHECK_PARITY: | |
2020 | BUG(); /* Not implemented yet */ | |
2021 | } | |
2022 | ||
2023 | for (i = disks; i--;) | |
2024 | if (sh->dev[i].written) { | |
2025 | sector_t sector = sh->dev[i].sector; | |
2026 | struct bio *wbi = sh->dev[i].written; | |
2027 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
2028 | copy_data(1, wbi, sh->dev[i].page, sector); | |
2029 | wbi = r5_next_bio(wbi, sector); | |
2030 | } | |
2031 | ||
2032 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
2033 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
2034 | } | |
2035 | ||
d0dabf7e | 2036 | /* Note that unlike RAID-5, the ordering of the disks matters greatly.*/ |
67cc2b81 N |
2037 | |
2038 | for (i = 0; i < disks; i++) | |
2039 | ptrs[i] = (void *)raid6_empty_zero_page; | |
2040 | ||
d0dabf7e N |
2041 | count = 0; |
2042 | i = d0_idx; | |
2043 | do { | |
67cc2b81 N |
2044 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); |
2045 | ||
d0dabf7e | 2046 | ptrs[slot] = page_address(sh->dev[i].page); |
67cc2b81 | 2047 | if (slot < syndrome_disks && |
d0dabf7e N |
2048 | !test_bit(R5_UPTODATE, &sh->dev[i].flags)) { |
2049 | printk(KERN_ERR "block %d/%d not uptodate " | |
2050 | "on parity calc\n", i, count); | |
2051 | BUG(); | |
2052 | } | |
67cc2b81 | 2053 | |
d0dabf7e N |
2054 | i = raid6_next_disk(i, disks); |
2055 | } while (i != d0_idx); | |
67cc2b81 | 2056 | BUG_ON(count != syndrome_disks); |
16a53ecc | 2057 | |
67cc2b81 | 2058 | raid6_call.gen_syndrome(syndrome_disks+2, STRIPE_SIZE, ptrs); |
16a53ecc N |
2059 | |
2060 | switch(method) { | |
2061 | case RECONSTRUCT_WRITE: | |
2062 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2063 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
2064 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2065 | set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); | |
2066 | break; | |
2067 | case UPDATE_PARITY: | |
2068 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2069 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
2070 | break; | |
2071 | } | |
2072 | } | |
2073 | ||
2074 | ||
2075 | /* Compute one missing block */ | |
2076 | static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero) | |
2077 | { | |
f416885e | 2078 | int i, count, disks = sh->disks; |
9bc89cd8 | 2079 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
d0dabf7e | 2080 | int qd_idx = sh->qd_idx; |
16a53ecc | 2081 | |
45b4233c | 2082 | pr_debug("compute_block_1, stripe %llu, idx %d\n", |
16a53ecc N |
2083 | (unsigned long long)sh->sector, dd_idx); |
2084 | ||
2085 | if ( dd_idx == qd_idx ) { | |
2086 | /* We're actually computing the Q drive */ | |
2087 | compute_parity6(sh, UPDATE_PARITY); | |
2088 | } else { | |
9bc89cd8 DW |
2089 | dest = page_address(sh->dev[dd_idx].page); |
2090 | if (!nozero) memset(dest, 0, STRIPE_SIZE); | |
2091 | count = 0; | |
16a53ecc N |
2092 | for (i = disks ; i--; ) { |
2093 | if (i == dd_idx || i == qd_idx) | |
2094 | continue; | |
2095 | p = page_address(sh->dev[i].page); | |
2096 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
2097 | ptr[count++] = p; | |
2098 | else | |
2099 | printk("compute_block() %d, stripe %llu, %d" | |
2100 | " not present\n", dd_idx, | |
2101 | (unsigned long long)sh->sector, i); | |
2102 | ||
2103 | check_xor(); | |
2104 | } | |
9bc89cd8 DW |
2105 | if (count) |
2106 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
16a53ecc N |
2107 | if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
2108 | else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
2109 | } | |
2110 | } | |
2111 | ||
2112 | /* Compute two missing blocks */ | |
2113 | static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) | |
2114 | { | |
f416885e | 2115 | int i, count, disks = sh->disks; |
67cc2b81 | 2116 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; |
d0dabf7e N |
2117 | int d0_idx = raid6_d0(sh); |
2118 | int faila = -1, failb = -1; | |
2119 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
67cc2b81 | 2120 | void *ptrs[syndrome_disks+2]; |
16a53ecc | 2121 | |
67cc2b81 N |
2122 | for (i = 0; i < disks ; i++) |
2123 | ptrs[i] = (void *)raid6_empty_zero_page; | |
d0dabf7e N |
2124 | count = 0; |
2125 | i = d0_idx; | |
2126 | do { | |
67cc2b81 N |
2127 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); |
2128 | ||
d0dabf7e | 2129 | ptrs[slot] = page_address(sh->dev[i].page); |
67cc2b81 | 2130 | |
d0dabf7e N |
2131 | if (i == dd_idx1) |
2132 | faila = slot; | |
2133 | if (i == dd_idx2) | |
2134 | failb = slot; | |
2135 | i = raid6_next_disk(i, disks); | |
2136 | } while (i != d0_idx); | |
67cc2b81 | 2137 | BUG_ON(count != syndrome_disks); |
16a53ecc N |
2138 | |
2139 | BUG_ON(faila == failb); | |
2140 | if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } | |
2141 | ||
45b4233c | 2142 | pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", |
d0dabf7e N |
2143 | (unsigned long long)sh->sector, dd_idx1, dd_idx2, |
2144 | faila, failb); | |
16a53ecc | 2145 | |
67cc2b81 | 2146 | if (failb == syndrome_disks+1) { |
16a53ecc | 2147 | /* Q disk is one of the missing disks */ |
67cc2b81 | 2148 | if (faila == syndrome_disks) { |
16a53ecc N |
2149 | /* Missing P+Q, just recompute */ |
2150 | compute_parity6(sh, UPDATE_PARITY); | |
2151 | return; | |
2152 | } else { | |
2153 | /* We're missing D+Q; recompute D from P */ | |
d0dabf7e N |
2154 | compute_block_1(sh, ((dd_idx1 == sh->qd_idx) ? |
2155 | dd_idx2 : dd_idx1), | |
2156 | 0); | |
16a53ecc N |
2157 | compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */ |
2158 | return; | |
2159 | } | |
2160 | } | |
2161 | ||
d0dabf7e | 2162 | /* We're missing D+P or D+D; */ |
67cc2b81 | 2163 | if (failb == syndrome_disks) { |
d0dabf7e | 2164 | /* We're missing D+P. */ |
67cc2b81 | 2165 | raid6_datap_recov(syndrome_disks+2, STRIPE_SIZE, faila, ptrs); |
d0dabf7e N |
2166 | } else { |
2167 | /* We're missing D+D. */ | |
67cc2b81 N |
2168 | raid6_2data_recov(syndrome_disks+2, STRIPE_SIZE, faila, failb, |
2169 | ptrs); | |
16a53ecc | 2170 | } |
d0dabf7e N |
2171 | |
2172 | /* Both the above update both missing blocks */ | |
2173 | set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); | |
2174 | set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); | |
16a53ecc N |
2175 | } |
2176 | ||
600aa109 | 2177 | static void |
1fe797e6 | 2178 | schedule_reconstruction5(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2179 | int rcw, int expand) |
e33129d8 DW |
2180 | { |
2181 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
e33129d8 DW |
2182 | |
2183 | if (rcw) { | |
2184 | /* if we are not expanding this is a proper write request, and | |
2185 | * there will be bios with new data to be drained into the | |
2186 | * stripe cache | |
2187 | */ | |
2188 | if (!expand) { | |
600aa109 DW |
2189 | sh->reconstruct_state = reconstruct_state_drain_run; |
2190 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2191 | } else | |
2192 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2193 | |
ac6b53b6 | 2194 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2195 | |
2196 | for (i = disks; i--; ) { | |
2197 | struct r5dev *dev = &sh->dev[i]; | |
2198 | ||
2199 | if (dev->towrite) { | |
2200 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2201 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2202 | if (!expand) |
2203 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2204 | s->locked++; |
e33129d8 DW |
2205 | } |
2206 | } | |
600aa109 | 2207 | if (s->locked + 1 == disks) |
8b3e6cdc DW |
2208 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
2209 | atomic_inc(&sh->raid_conf->pending_full_writes); | |
e33129d8 DW |
2210 | } else { |
2211 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
2212 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2213 | ||
d8ee0728 | 2214 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2215 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2216 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2217 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2218 | |
2219 | for (i = disks; i--; ) { | |
2220 | struct r5dev *dev = &sh->dev[i]; | |
2221 | if (i == pd_idx) | |
2222 | continue; | |
2223 | ||
e33129d8 DW |
2224 | if (dev->towrite && |
2225 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2226 | test_bit(R5_Wantcompute, &dev->flags))) { |
2227 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2228 | set_bit(R5_LOCKED, &dev->flags); |
2229 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2230 | s->locked++; |
e33129d8 DW |
2231 | } |
2232 | } | |
2233 | } | |
2234 | ||
2235 | /* keep the parity disk locked while asynchronous operations | |
2236 | * are in flight | |
2237 | */ | |
2238 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2239 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2240 | s->locked++; |
e33129d8 | 2241 | |
600aa109 | 2242 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2243 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2244 | s->locked, s->ops_request); |
e33129d8 | 2245 | } |
16a53ecc | 2246 | |
1da177e4 LT |
2247 | /* |
2248 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2249 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2250 | * The bi_next chain must be in order. |
2251 | */ | |
2252 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2253 | { | |
2254 | struct bio **bip; | |
2255 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 2256 | int firstwrite=0; |
1da177e4 | 2257 | |
45b4233c | 2258 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2259 | (unsigned long long)bi->bi_sector, |
2260 | (unsigned long long)sh->sector); | |
2261 | ||
2262 | ||
2263 | spin_lock(&sh->lock); | |
2264 | spin_lock_irq(&conf->device_lock); | |
72626685 | 2265 | if (forwrite) { |
1da177e4 | 2266 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2267 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2268 | firstwrite = 1; | |
2269 | } else | |
1da177e4 LT |
2270 | bip = &sh->dev[dd_idx].toread; |
2271 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2272 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2273 | goto overlap; | |
2274 | bip = & (*bip)->bi_next; | |
2275 | } | |
2276 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2277 | goto overlap; | |
2278 | ||
78bafebd | 2279 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2280 | if (*bip) |
2281 | bi->bi_next = *bip; | |
2282 | *bip = bi; | |
960e739d | 2283 | bi->bi_phys_segments++; |
1da177e4 LT |
2284 | spin_unlock_irq(&conf->device_lock); |
2285 | spin_unlock(&sh->lock); | |
2286 | ||
45b4233c | 2287 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
2288 | (unsigned long long)bi->bi_sector, |
2289 | (unsigned long long)sh->sector, dd_idx); | |
2290 | ||
72626685 | 2291 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
2292 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
2293 | STRIPE_SECTORS, 0); | |
ae3c20cc | 2294 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
2295 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
2296 | } | |
2297 | ||
1da177e4 LT |
2298 | if (forwrite) { |
2299 | /* check if page is covered */ | |
2300 | sector_t sector = sh->dev[dd_idx].sector; | |
2301 | for (bi=sh->dev[dd_idx].towrite; | |
2302 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2303 | bi && bi->bi_sector <= sector; | |
2304 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2305 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2306 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2307 | } | |
2308 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2309 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2310 | } | |
2311 | return 1; | |
2312 | ||
2313 | overlap: | |
2314 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2315 | spin_unlock_irq(&conf->device_lock); | |
2316 | spin_unlock(&sh->lock); | |
2317 | return 0; | |
2318 | } | |
2319 | ||
29269553 N |
2320 | static void end_reshape(raid5_conf_t *conf); |
2321 | ||
16a53ecc N |
2322 | static int page_is_zero(struct page *p) |
2323 | { | |
2324 | char *a = page_address(p); | |
2325 | return ((*(u32*)a) == 0 && | |
2326 | memcmp(a, a+4, STRIPE_SIZE-4)==0); | |
2327 | } | |
2328 | ||
911d4ee8 N |
2329 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
2330 | struct stripe_head *sh) | |
ccfcc3c1 | 2331 | { |
784052ec N |
2332 | int sectors_per_chunk = |
2333 | previous ? (conf->prev_chunk >> 9) | |
2334 | : (conf->chunk_size >> 9); | |
911d4ee8 | 2335 | int dd_idx; |
2d2063ce | 2336 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2337 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2338 | |
112bf897 N |
2339 | raid5_compute_sector(conf, |
2340 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2341 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2342 | previous, |
911d4ee8 | 2343 | &dd_idx, sh); |
ccfcc3c1 N |
2344 | } |
2345 | ||
a4456856 | 2346 | static void |
1fe797e6 | 2347 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2348 | struct stripe_head_state *s, int disks, |
2349 | struct bio **return_bi) | |
2350 | { | |
2351 | int i; | |
2352 | for (i = disks; i--; ) { | |
2353 | struct bio *bi; | |
2354 | int bitmap_end = 0; | |
2355 | ||
2356 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2357 | mdk_rdev_t *rdev; | |
2358 | rcu_read_lock(); | |
2359 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2360 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2361 | /* multiple read failures in one stripe */ | |
2362 | md_error(conf->mddev, rdev); | |
2363 | rcu_read_unlock(); | |
2364 | } | |
2365 | spin_lock_irq(&conf->device_lock); | |
2366 | /* fail all writes first */ | |
2367 | bi = sh->dev[i].towrite; | |
2368 | sh->dev[i].towrite = NULL; | |
2369 | if (bi) { | |
2370 | s->to_write--; | |
2371 | bitmap_end = 1; | |
2372 | } | |
2373 | ||
2374 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2375 | wake_up(&conf->wait_for_overlap); | |
2376 | ||
2377 | while (bi && bi->bi_sector < | |
2378 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2379 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2380 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2381 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2382 | md_write_end(conf->mddev); |
2383 | bi->bi_next = *return_bi; | |
2384 | *return_bi = bi; | |
2385 | } | |
2386 | bi = nextbi; | |
2387 | } | |
2388 | /* and fail all 'written' */ | |
2389 | bi = sh->dev[i].written; | |
2390 | sh->dev[i].written = NULL; | |
2391 | if (bi) bitmap_end = 1; | |
2392 | while (bi && bi->bi_sector < | |
2393 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2394 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2395 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2396 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2397 | md_write_end(conf->mddev); |
2398 | bi->bi_next = *return_bi; | |
2399 | *return_bi = bi; | |
2400 | } | |
2401 | bi = bi2; | |
2402 | } | |
2403 | ||
b5e98d65 DW |
2404 | /* fail any reads if this device is non-operational and |
2405 | * the data has not reached the cache yet. | |
2406 | */ | |
2407 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2408 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2409 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2410 | bi = sh->dev[i].toread; |
2411 | sh->dev[i].toread = NULL; | |
2412 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2413 | wake_up(&conf->wait_for_overlap); | |
2414 | if (bi) s->to_read--; | |
2415 | while (bi && bi->bi_sector < | |
2416 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2417 | struct bio *nextbi = | |
2418 | r5_next_bio(bi, sh->dev[i].sector); | |
2419 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2420 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2421 | bi->bi_next = *return_bi; |
2422 | *return_bi = bi; | |
2423 | } | |
2424 | bi = nextbi; | |
2425 | } | |
2426 | } | |
2427 | spin_unlock_irq(&conf->device_lock); | |
2428 | if (bitmap_end) | |
2429 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2430 | STRIPE_SECTORS, 0, 0); | |
2431 | } | |
2432 | ||
8b3e6cdc DW |
2433 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2434 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2435 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2436 | } |
2437 | ||
1fe797e6 DW |
2438 | /* fetch_block5 - checks the given member device to see if its data needs |
2439 | * to be read or computed to satisfy a request. | |
2440 | * | |
2441 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2442 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2443 | */ |
1fe797e6 DW |
2444 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2445 | int disk_idx, int disks) | |
f38e1219 DW |
2446 | { |
2447 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2448 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2449 | ||
f38e1219 DW |
2450 | /* is the data in this block needed, and can we get it? */ |
2451 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2452 | !test_bit(R5_UPTODATE, &dev->flags) && |
2453 | (dev->toread || | |
2454 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2455 | s->syncing || s->expanding || | |
2456 | (s->failed && | |
2457 | (failed_dev->toread || | |
2458 | (failed_dev->towrite && | |
2459 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2460 | /* We would like to get this block, possibly by computing it, |
2461 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2462 | */ |
2463 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2464 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2465 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2466 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2467 | set_bit(R5_Wantcompute, &dev->flags); |
2468 | sh->ops.target = disk_idx; | |
ac6b53b6 | 2469 | sh->ops.target2 = -1; |
f38e1219 | 2470 | s->req_compute = 1; |
f38e1219 | 2471 | /* Careful: from this point on 'uptodate' is in the eye |
ac6b53b6 | 2472 | * of raid_run_ops which services 'compute' operations |
f38e1219 DW |
2473 | * before writes. R5_Wantcompute flags a block that will |
2474 | * be R5_UPTODATE by the time it is needed for a | |
2475 | * subsequent operation. | |
2476 | */ | |
2477 | s->uptodate++; | |
1fe797e6 | 2478 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2479 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2480 | set_bit(R5_LOCKED, &dev->flags); |
2481 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2482 | s->locked++; |
2483 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2484 | s->syncing); | |
2485 | } | |
2486 | } | |
2487 | ||
1fe797e6 | 2488 | return 0; |
f38e1219 DW |
2489 | } |
2490 | ||
1fe797e6 DW |
2491 | /** |
2492 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2493 | */ | |
2494 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2495 | struct stripe_head_state *s, int disks) |
2496 | { | |
2497 | int i; | |
f38e1219 | 2498 | |
f38e1219 DW |
2499 | /* look for blocks to read/compute, skip this if a compute |
2500 | * is already in flight, or if the stripe contents are in the | |
2501 | * midst of changing due to a write | |
2502 | */ | |
976ea8d4 | 2503 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2504 | !sh->reconstruct_state) |
f38e1219 | 2505 | for (i = disks; i--; ) |
1fe797e6 | 2506 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2507 | break; |
a4456856 DW |
2508 | set_bit(STRIPE_HANDLE, &sh->state); |
2509 | } | |
2510 | ||
1fe797e6 | 2511 | static void handle_stripe_fill6(struct stripe_head *sh, |
a4456856 DW |
2512 | struct stripe_head_state *s, struct r6_state *r6s, |
2513 | int disks) | |
2514 | { | |
2515 | int i; | |
2516 | for (i = disks; i--; ) { | |
2517 | struct r5dev *dev = &sh->dev[i]; | |
2518 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2519 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2520 | (dev->toread || (dev->towrite && | |
2521 | !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2522 | s->syncing || s->expanding || | |
2523 | (s->failed >= 1 && | |
2524 | (sh->dev[r6s->failed_num[0]].toread || | |
2525 | s->to_write)) || | |
2526 | (s->failed >= 2 && | |
2527 | (sh->dev[r6s->failed_num[1]].toread || | |
2528 | s->to_write)))) { | |
2529 | /* we would like to get this block, possibly | |
2530 | * by computing it, but we might not be able to | |
2531 | */ | |
c337869d DW |
2532 | if ((s->uptodate == disks - 1) && |
2533 | (s->failed && (i == r6s->failed_num[0] || | |
2534 | i == r6s->failed_num[1]))) { | |
45b4233c | 2535 | pr_debug("Computing stripe %llu block %d\n", |
a4456856 DW |
2536 | (unsigned long long)sh->sector, i); |
2537 | compute_block_1(sh, i, 0); | |
2538 | s->uptodate++; | |
2539 | } else if ( s->uptodate == disks-2 && s->failed >= 2 ) { | |
2540 | /* Computing 2-failure is *very* expensive; only | |
2541 | * do it if failed >= 2 | |
2542 | */ | |
2543 | int other; | |
2544 | for (other = disks; other--; ) { | |
2545 | if (other == i) | |
2546 | continue; | |
2547 | if (!test_bit(R5_UPTODATE, | |
2548 | &sh->dev[other].flags)) | |
2549 | break; | |
2550 | } | |
2551 | BUG_ON(other < 0); | |
45b4233c | 2552 | pr_debug("Computing stripe %llu blocks %d,%d\n", |
a4456856 DW |
2553 | (unsigned long long)sh->sector, |
2554 | i, other); | |
2555 | compute_block_2(sh, i, other); | |
2556 | s->uptodate += 2; | |
2557 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2558 | set_bit(R5_LOCKED, &dev->flags); | |
2559 | set_bit(R5_Wantread, &dev->flags); | |
2560 | s->locked++; | |
45b4233c | 2561 | pr_debug("Reading block %d (sync=%d)\n", |
a4456856 DW |
2562 | i, s->syncing); |
2563 | } | |
2564 | } | |
2565 | } | |
2566 | set_bit(STRIPE_HANDLE, &sh->state); | |
2567 | } | |
2568 | ||
2569 | ||
1fe797e6 | 2570 | /* handle_stripe_clean_event |
a4456856 DW |
2571 | * any written block on an uptodate or failed drive can be returned. |
2572 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2573 | * never LOCKED, so we don't need to test 'failed' directly. | |
2574 | */ | |
1fe797e6 | 2575 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2576 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2577 | { | |
2578 | int i; | |
2579 | struct r5dev *dev; | |
2580 | ||
2581 | for (i = disks; i--; ) | |
2582 | if (sh->dev[i].written) { | |
2583 | dev = &sh->dev[i]; | |
2584 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2585 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2586 | /* We can return any write requests */ | |
2587 | struct bio *wbi, *wbi2; | |
2588 | int bitmap_end = 0; | |
45b4233c | 2589 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2590 | spin_lock_irq(&conf->device_lock); |
2591 | wbi = dev->written; | |
2592 | dev->written = NULL; | |
2593 | while (wbi && wbi->bi_sector < | |
2594 | dev->sector + STRIPE_SECTORS) { | |
2595 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2596 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2597 | md_write_end(conf->mddev); |
2598 | wbi->bi_next = *return_bi; | |
2599 | *return_bi = wbi; | |
2600 | } | |
2601 | wbi = wbi2; | |
2602 | } | |
2603 | if (dev->towrite == NULL) | |
2604 | bitmap_end = 1; | |
2605 | spin_unlock_irq(&conf->device_lock); | |
2606 | if (bitmap_end) | |
2607 | bitmap_endwrite(conf->mddev->bitmap, | |
2608 | sh->sector, | |
2609 | STRIPE_SECTORS, | |
2610 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2611 | 0); | |
2612 | } | |
2613 | } | |
8b3e6cdc DW |
2614 | |
2615 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2616 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2617 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2618 | } |
2619 | ||
1fe797e6 | 2620 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2621 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2622 | { | |
2623 | int rmw = 0, rcw = 0, i; | |
2624 | for (i = disks; i--; ) { | |
2625 | /* would I have to read this buffer for read_modify_write */ | |
2626 | struct r5dev *dev = &sh->dev[i]; | |
2627 | if ((dev->towrite || i == sh->pd_idx) && | |
2628 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2629 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2630 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2631 | if (test_bit(R5_Insync, &dev->flags)) |
2632 | rmw++; | |
2633 | else | |
2634 | rmw += 2*disks; /* cannot read it */ | |
2635 | } | |
2636 | /* Would I have to read this buffer for reconstruct_write */ | |
2637 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2638 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2639 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2640 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2641 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2642 | else |
2643 | rcw += 2*disks; | |
2644 | } | |
2645 | } | |
45b4233c | 2646 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2647 | (unsigned long long)sh->sector, rmw, rcw); |
2648 | set_bit(STRIPE_HANDLE, &sh->state); | |
2649 | if (rmw < rcw && rmw > 0) | |
2650 | /* prefer read-modify-write, but need to get some data */ | |
2651 | for (i = disks; i--; ) { | |
2652 | struct r5dev *dev = &sh->dev[i]; | |
2653 | if ((dev->towrite || i == sh->pd_idx) && | |
2654 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2655 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2656 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2657 | test_bit(R5_Insync, &dev->flags)) { |
2658 | if ( | |
2659 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2660 | pr_debug("Read_old block " |
a4456856 DW |
2661 | "%d for r-m-w\n", i); |
2662 | set_bit(R5_LOCKED, &dev->flags); | |
2663 | set_bit(R5_Wantread, &dev->flags); | |
2664 | s->locked++; | |
2665 | } else { | |
2666 | set_bit(STRIPE_DELAYED, &sh->state); | |
2667 | set_bit(STRIPE_HANDLE, &sh->state); | |
2668 | } | |
2669 | } | |
2670 | } | |
2671 | if (rcw <= rmw && rcw > 0) | |
2672 | /* want reconstruct write, but need to get some data */ | |
2673 | for (i = disks; i--; ) { | |
2674 | struct r5dev *dev = &sh->dev[i]; | |
2675 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2676 | i != sh->pd_idx && | |
2677 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2678 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2679 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2680 | test_bit(R5_Insync, &dev->flags)) { |
2681 | if ( | |
2682 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2683 | pr_debug("Read_old block " |
a4456856 DW |
2684 | "%d for Reconstruct\n", i); |
2685 | set_bit(R5_LOCKED, &dev->flags); | |
2686 | set_bit(R5_Wantread, &dev->flags); | |
2687 | s->locked++; | |
2688 | } else { | |
2689 | set_bit(STRIPE_DELAYED, &sh->state); | |
2690 | set_bit(STRIPE_HANDLE, &sh->state); | |
2691 | } | |
2692 | } | |
2693 | } | |
2694 | /* now if nothing is locked, and if we have enough data, | |
2695 | * we can start a write request | |
2696 | */ | |
f38e1219 DW |
2697 | /* since handle_stripe can be called at any time we need to handle the |
2698 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2699 | * subsequent call wants to start a write request. raid_run_ops only |
2700 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2701 | * simultaneously. If this is not the case then new writes need to be |
2702 | * held off until the compute completes. | |
2703 | */ | |
976ea8d4 DW |
2704 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2705 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2706 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
1fe797e6 | 2707 | schedule_reconstruction5(sh, s, rcw == 0, 0); |
a4456856 DW |
2708 | } |
2709 | ||
1fe797e6 | 2710 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2711 | struct stripe_head *sh, struct stripe_head_state *s, |
2712 | struct r6_state *r6s, int disks) | |
2713 | { | |
2714 | int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i; | |
34e04e87 | 2715 | int qd_idx = sh->qd_idx; |
a4456856 DW |
2716 | for (i = disks; i--; ) { |
2717 | struct r5dev *dev = &sh->dev[i]; | |
2718 | /* Would I have to read this buffer for reconstruct_write */ | |
2719 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2720 | && i != pd_idx && i != qd_idx | |
2721 | && (!test_bit(R5_LOCKED, &dev->flags) | |
2722 | ) && | |
2723 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
2724 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
2725 | else { | |
45b4233c | 2726 | pr_debug("raid6: must_compute: " |
a4456856 DW |
2727 | "disk %d flags=%#lx\n", i, dev->flags); |
2728 | must_compute++; | |
2729 | } | |
2730 | } | |
2731 | } | |
45b4233c | 2732 | pr_debug("for sector %llu, rcw=%d, must_compute=%d\n", |
a4456856 DW |
2733 | (unsigned long long)sh->sector, rcw, must_compute); |
2734 | set_bit(STRIPE_HANDLE, &sh->state); | |
2735 | ||
2736 | if (rcw > 0) | |
2737 | /* want reconstruct write, but need to get some data */ | |
2738 | for (i = disks; i--; ) { | |
2739 | struct r5dev *dev = &sh->dev[i]; | |
2740 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2741 | && !(s->failed == 0 && (i == pd_idx || i == qd_idx)) | |
2742 | && !test_bit(R5_LOCKED, &dev->flags) && | |
2743 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2744 | test_bit(R5_Insync, &dev->flags)) { | |
2745 | if ( | |
2746 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2747 | pr_debug("Read_old stripe %llu " |
a4456856 DW |
2748 | "block %d for Reconstruct\n", |
2749 | (unsigned long long)sh->sector, i); | |
2750 | set_bit(R5_LOCKED, &dev->flags); | |
2751 | set_bit(R5_Wantread, &dev->flags); | |
2752 | s->locked++; | |
2753 | } else { | |
45b4233c | 2754 | pr_debug("Request delayed stripe %llu " |
a4456856 DW |
2755 | "block %d for Reconstruct\n", |
2756 | (unsigned long long)sh->sector, i); | |
2757 | set_bit(STRIPE_DELAYED, &sh->state); | |
2758 | set_bit(STRIPE_HANDLE, &sh->state); | |
2759 | } | |
2760 | } | |
2761 | } | |
2762 | /* now if nothing is locked, and if we have enough data, we can start a | |
2763 | * write request | |
2764 | */ | |
2765 | if (s->locked == 0 && rcw == 0 && | |
2766 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
2767 | if (must_compute > 0) { | |
2768 | /* We have failed blocks and need to compute them */ | |
2769 | switch (s->failed) { | |
2770 | case 0: | |
2771 | BUG(); | |
2772 | case 1: | |
2773 | compute_block_1(sh, r6s->failed_num[0], 0); | |
2774 | break; | |
2775 | case 2: | |
2776 | compute_block_2(sh, r6s->failed_num[0], | |
2777 | r6s->failed_num[1]); | |
2778 | break; | |
2779 | default: /* This request should have been failed? */ | |
2780 | BUG(); | |
2781 | } | |
2782 | } | |
2783 | ||
45b4233c | 2784 | pr_debug("Computing parity for stripe %llu\n", |
a4456856 DW |
2785 | (unsigned long long)sh->sector); |
2786 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
2787 | /* now every locked buffer is ready to be written */ | |
2788 | for (i = disks; i--; ) | |
2789 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
45b4233c | 2790 | pr_debug("Writing stripe %llu block %d\n", |
a4456856 DW |
2791 | (unsigned long long)sh->sector, i); |
2792 | s->locked++; | |
2793 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
2794 | } | |
8b3e6cdc DW |
2795 | if (s->locked == disks) |
2796 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2797 | atomic_inc(&conf->pending_full_writes); | |
a4456856 DW |
2798 | /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */ |
2799 | set_bit(STRIPE_INSYNC, &sh->state); | |
2800 | ||
2801 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2802 | atomic_dec(&conf->preread_active_stripes); | |
2803 | if (atomic_read(&conf->preread_active_stripes) < | |
2804 | IO_THRESHOLD) | |
2805 | md_wakeup_thread(conf->mddev->thread); | |
2806 | } | |
2807 | } | |
2808 | } | |
2809 | ||
2810 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2811 | struct stripe_head_state *s, int disks) | |
2812 | { | |
ecc65c9b | 2813 | struct r5dev *dev = NULL; |
bd2ab670 | 2814 | |
a4456856 | 2815 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2816 | |
ecc65c9b DW |
2817 | switch (sh->check_state) { |
2818 | case check_state_idle: | |
2819 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2820 | if (s->failed == 0) { |
bd2ab670 | 2821 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2822 | sh->check_state = check_state_run; |
2823 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2824 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2825 | s->uptodate--; |
ecc65c9b | 2826 | break; |
bd2ab670 | 2827 | } |
ecc65c9b DW |
2828 | dev = &sh->dev[s->failed_num]; |
2829 | /* fall through */ | |
2830 | case check_state_compute_result: | |
2831 | sh->check_state = check_state_idle; | |
2832 | if (!dev) | |
2833 | dev = &sh->dev[sh->pd_idx]; | |
2834 | ||
2835 | /* check that a write has not made the stripe insync */ | |
2836 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2837 | break; | |
c8894419 | 2838 | |
a4456856 | 2839 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2840 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2841 | BUG_ON(s->uptodate != disks); | |
2842 | ||
2843 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2844 | s->locked++; |
a4456856 | 2845 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2846 | |
a4456856 | 2847 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2848 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2849 | break; |
2850 | case check_state_run: | |
2851 | break; /* we will be called again upon completion */ | |
2852 | case check_state_check_result: | |
2853 | sh->check_state = check_state_idle; | |
2854 | ||
2855 | /* if a failure occurred during the check operation, leave | |
2856 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2857 | */ | |
2858 | if (s->failed) | |
2859 | break; | |
2860 | ||
2861 | /* handle a successful check operation, if parity is correct | |
2862 | * we are done. Otherwise update the mismatch count and repair | |
2863 | * parity if !MD_RECOVERY_CHECK | |
2864 | */ | |
ad283ea4 | 2865 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2866 | /* parity is correct (on disc, |
2867 | * not in buffer any more) | |
2868 | */ | |
2869 | set_bit(STRIPE_INSYNC, &sh->state); | |
2870 | else { | |
2871 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2872 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2873 | /* don't try to repair!! */ | |
2874 | set_bit(STRIPE_INSYNC, &sh->state); | |
2875 | else { | |
2876 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2877 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2878 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2879 | set_bit(R5_Wantcompute, | |
2880 | &sh->dev[sh->pd_idx].flags); | |
2881 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2882 | sh->ops.target2 = -1; |
ecc65c9b DW |
2883 | s->uptodate++; |
2884 | } | |
2885 | } | |
2886 | break; | |
2887 | case check_state_compute_run: | |
2888 | break; | |
2889 | default: | |
2890 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2891 | __func__, sh->check_state, | |
2892 | (unsigned long long) sh->sector); | |
2893 | BUG(); | |
a4456856 DW |
2894 | } |
2895 | } | |
2896 | ||
2897 | ||
2898 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
36d1c647 DW |
2899 | struct stripe_head_state *s, |
2900 | struct r6_state *r6s, int disks) | |
a4456856 DW |
2901 | { |
2902 | int update_p = 0, update_q = 0; | |
2903 | struct r5dev *dev; | |
2904 | int pd_idx = sh->pd_idx; | |
34e04e87 | 2905 | int qd_idx = sh->qd_idx; |
36d1c647 DW |
2906 | unsigned long cpu; |
2907 | struct page *tmp_page; | |
a4456856 DW |
2908 | |
2909 | set_bit(STRIPE_HANDLE, &sh->state); | |
2910 | ||
2911 | BUG_ON(s->failed > 2); | |
2912 | BUG_ON(s->uptodate < disks); | |
2913 | /* Want to check and possibly repair P and Q. | |
2914 | * However there could be one 'failed' device, in which | |
2915 | * case we can only check one of them, possibly using the | |
2916 | * other to generate missing data | |
2917 | */ | |
36d1c647 DW |
2918 | cpu = get_cpu(); |
2919 | tmp_page = per_cpu_ptr(conf->percpu, cpu)->spare_page; | |
2920 | if (s->failed == r6s->q_failed) { | |
2921 | /* The only possible failed device holds 'Q', so it | |
2922 | * makes sense to check P (If anything else were failed, | |
2923 | * we would have used P to recreate it). | |
2924 | */ | |
2925 | compute_block_1(sh, pd_idx, 1); | |
2926 | if (!page_is_zero(sh->dev[pd_idx].page)) { | |
2927 | compute_block_1(sh, pd_idx, 0); | |
2928 | update_p = 1; | |
a4456856 | 2929 | } |
36d1c647 DW |
2930 | } |
2931 | if (!r6s->q_failed && s->failed < 2) { | |
2932 | /* q is not failed, and we didn't use it to generate | |
2933 | * anything, so it makes sense to check it | |
2934 | */ | |
2935 | memcpy(page_address(tmp_page), | |
2936 | page_address(sh->dev[qd_idx].page), | |
2937 | STRIPE_SIZE); | |
2938 | compute_parity6(sh, UPDATE_PARITY); | |
2939 | if (memcmp(page_address(tmp_page), | |
2940 | page_address(sh->dev[qd_idx].page), | |
2941 | STRIPE_SIZE) != 0) { | |
2942 | clear_bit(STRIPE_INSYNC, &sh->state); | |
2943 | update_q = 1; | |
a4456856 | 2944 | } |
36d1c647 DW |
2945 | } |
2946 | put_cpu(); | |
a4456856 | 2947 | |
36d1c647 DW |
2948 | if (update_p || update_q) { |
2949 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2950 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2951 | /* don't try to repair!! */ | |
2952 | update_p = update_q = 0; | |
2953 | } | |
a4456856 | 2954 | |
36d1c647 DW |
2955 | /* now write out any block on a failed drive, |
2956 | * or P or Q if they need it | |
2957 | */ | |
a4456856 | 2958 | |
36d1c647 DW |
2959 | if (s->failed == 2) { |
2960 | dev = &sh->dev[r6s->failed_num[1]]; | |
2961 | s->locked++; | |
2962 | set_bit(R5_LOCKED, &dev->flags); | |
2963 | set_bit(R5_Wantwrite, &dev->flags); | |
2964 | } | |
2965 | if (s->failed >= 1) { | |
2966 | dev = &sh->dev[r6s->failed_num[0]]; | |
2967 | s->locked++; | |
2968 | set_bit(R5_LOCKED, &dev->flags); | |
2969 | set_bit(R5_Wantwrite, &dev->flags); | |
2970 | } | |
a4456856 | 2971 | |
36d1c647 DW |
2972 | if (update_p) { |
2973 | dev = &sh->dev[pd_idx]; | |
2974 | s->locked++; | |
2975 | set_bit(R5_LOCKED, &dev->flags); | |
2976 | set_bit(R5_Wantwrite, &dev->flags); | |
2977 | } | |
2978 | if (update_q) { | |
2979 | dev = &sh->dev[qd_idx]; | |
2980 | s->locked++; | |
2981 | set_bit(R5_LOCKED, &dev->flags); | |
2982 | set_bit(R5_Wantwrite, &dev->flags); | |
a4456856 | 2983 | } |
36d1c647 DW |
2984 | clear_bit(STRIPE_DEGRADED, &sh->state); |
2985 | ||
2986 | set_bit(STRIPE_INSYNC, &sh->state); | |
a4456856 DW |
2987 | } |
2988 | ||
2989 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2990 | struct r6_state *r6s) | |
2991 | { | |
2992 | int i; | |
2993 | ||
2994 | /* We have read all the blocks in this stripe and now we need to | |
2995 | * copy some of them into a target stripe for expand. | |
2996 | */ | |
f0a50d37 | 2997 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2998 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2999 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3000 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3001 | int dd_idx, j; |
a4456856 | 3002 | struct stripe_head *sh2; |
a08abd8c | 3003 | struct async_submit_ctl submit; |
a4456856 | 3004 | |
784052ec | 3005 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3006 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3007 | &dd_idx, NULL); | |
b5663ba4 | 3008 | sh2 = get_active_stripe(conf, s, 0, 1); |
a4456856 DW |
3009 | if (sh2 == NULL) |
3010 | /* so far only the early blocks of this stripe | |
3011 | * have been requested. When later blocks | |
3012 | * get requested, we will try again | |
3013 | */ | |
3014 | continue; | |
3015 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3016 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3017 | /* must have already done this block */ | |
3018 | release_stripe(sh2); | |
3019 | continue; | |
3020 | } | |
f0a50d37 DW |
3021 | |
3022 | /* place all the copies on one channel */ | |
a08abd8c | 3023 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3024 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3025 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3026 | &submit); |
f0a50d37 | 3027 | |
a4456856 DW |
3028 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3029 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3030 | for (j = 0; j < conf->raid_disks; j++) | |
3031 | if (j != sh2->pd_idx && | |
d0dabf7e | 3032 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
3033 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3034 | break; | |
3035 | if (j == conf->raid_disks) { | |
3036 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3037 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3038 | } | |
3039 | release_stripe(sh2); | |
f0a50d37 | 3040 | |
a4456856 | 3041 | } |
a2e08551 N |
3042 | /* done submitting copies, wait for them to complete */ |
3043 | if (tx) { | |
3044 | async_tx_ack(tx); | |
3045 | dma_wait_for_async_tx(tx); | |
3046 | } | |
a4456856 | 3047 | } |
1da177e4 | 3048 | |
6bfe0b49 | 3049 | |
1da177e4 LT |
3050 | /* |
3051 | * handle_stripe - do things to a stripe. | |
3052 | * | |
3053 | * We lock the stripe and then examine the state of various bits | |
3054 | * to see what needs to be done. | |
3055 | * Possible results: | |
3056 | * return some read request which now have data | |
3057 | * return some write requests which are safely on disc | |
3058 | * schedule a read on some buffers | |
3059 | * schedule a write of some buffers | |
3060 | * return confirmation of parity correctness | |
3061 | * | |
1da177e4 LT |
3062 | * buffers are taken off read_list or write_list, and bh_cache buffers |
3063 | * get BH_Lock set before the stripe lock is released. | |
3064 | * | |
3065 | */ | |
a4456856 | 3066 | |
df10cfbc | 3067 | static bool handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
3068 | { |
3069 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
3070 | int disks = sh->disks, i; |
3071 | struct bio *return_bi = NULL; | |
3072 | struct stripe_head_state s; | |
1da177e4 | 3073 | struct r5dev *dev; |
6bfe0b49 | 3074 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 3075 | int prexor; |
1da177e4 | 3076 | |
a4456856 | 3077 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
3078 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
3079 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
3080 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
3081 | sh->reconstruct_state); | |
1da177e4 LT |
3082 | |
3083 | spin_lock(&sh->lock); | |
3084 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3085 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3086 | ||
a4456856 DW |
3087 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3088 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3089 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 3090 | |
83de75cc | 3091 | /* Now to look around and see what can be done */ |
9910f16a | 3092 | rcu_read_lock(); |
1da177e4 LT |
3093 | for (i=disks; i--; ) { |
3094 | mdk_rdev_t *rdev; | |
a4456856 | 3095 | struct r5dev *dev = &sh->dev[i]; |
1da177e4 | 3096 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 3097 | |
b5e98d65 DW |
3098 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
3099 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
3100 | dev->towrite, dev->written); | |
3101 | ||
3102 | /* maybe we can request a biofill operation | |
3103 | * | |
3104 | * new wantfill requests are only permitted while | |
83de75cc | 3105 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
3106 | */ |
3107 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 3108 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 3109 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
3110 | |
3111 | /* now count some things */ | |
a4456856 DW |
3112 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3113 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 3114 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 3115 | |
b5e98d65 DW |
3116 | if (test_bit(R5_Wantfill, &dev->flags)) |
3117 | s.to_fill++; | |
3118 | else if (dev->toread) | |
a4456856 | 3119 | s.to_read++; |
1da177e4 | 3120 | if (dev->towrite) { |
a4456856 | 3121 | s.to_write++; |
1da177e4 | 3122 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3123 | s.non_overwrite++; |
1da177e4 | 3124 | } |
a4456856 DW |
3125 | if (dev->written) |
3126 | s.written++; | |
9910f16a | 3127 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3128 | if (blocked_rdev == NULL && |
3129 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3130 | blocked_rdev = rdev; |
3131 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3132 | } |
b2d444d7 | 3133 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 3134 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
3135 | clear_bit(R5_ReadError, &dev->flags); |
3136 | clear_bit(R5_ReWrite, &dev->flags); | |
3137 | } | |
b2d444d7 | 3138 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 3139 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
3140 | s.failed++; |
3141 | s.failed_num = i; | |
1da177e4 LT |
3142 | } else |
3143 | set_bit(R5_Insync, &dev->flags); | |
3144 | } | |
9910f16a | 3145 | rcu_read_unlock(); |
b5e98d65 | 3146 | |
6bfe0b49 | 3147 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
3148 | if (s.syncing || s.expanding || s.expanded || |
3149 | s.to_write || s.written) { | |
3150 | set_bit(STRIPE_HANDLE, &sh->state); | |
3151 | goto unlock; | |
3152 | } | |
3153 | /* There is nothing for the blocked_rdev to block */ | |
3154 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3155 | blocked_rdev = NULL; | |
6bfe0b49 DW |
3156 | } |
3157 | ||
83de75cc DW |
3158 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3159 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3160 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3161 | } | |
b5e98d65 | 3162 | |
45b4233c | 3163 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 3164 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
3165 | s.locked, s.uptodate, s.to_read, s.to_write, |
3166 | s.failed, s.failed_num); | |
1da177e4 LT |
3167 | /* check if the array has lost two devices and, if so, some requests might |
3168 | * need to be failed | |
3169 | */ | |
a4456856 | 3170 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3171 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3172 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
3173 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3174 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3175 | s.syncing = 0; |
1da177e4 LT |
3176 | } |
3177 | ||
3178 | /* might be able to return some write requests if the parity block | |
3179 | * is safe, or on a failed drive | |
3180 | */ | |
3181 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
3182 | if ( s.written && |
3183 | ((test_bit(R5_Insync, &dev->flags) && | |
3184 | !test_bit(R5_LOCKED, &dev->flags) && | |
3185 | test_bit(R5_UPTODATE, &dev->flags)) || | |
3186 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 3187 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
3188 | |
3189 | /* Now we might consider reading some blocks, either to check/generate | |
3190 | * parity, or to satisfy requests | |
3191 | * or to load a block that is being partially written. | |
3192 | */ | |
a4456856 | 3193 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 3194 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3195 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 3196 | |
e33129d8 DW |
3197 | /* Now we check to see if any write operations have recently |
3198 | * completed | |
3199 | */ | |
e0a115e5 | 3200 | prexor = 0; |
d8ee0728 | 3201 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 3202 | prexor = 1; |
d8ee0728 DW |
3203 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
3204 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 3205 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
3206 | |
3207 | /* All the 'written' buffers and the parity block are ready to | |
3208 | * be written back to disk | |
3209 | */ | |
3210 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3211 | for (i = disks; i--; ) { | |
3212 | dev = &sh->dev[i]; | |
3213 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3214 | (i == sh->pd_idx || dev->written)) { | |
3215 | pr_debug("Writing block %d\n", i); | |
3216 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
3217 | if (prexor) |
3218 | continue; | |
e33129d8 DW |
3219 | if (!test_bit(R5_Insync, &dev->flags) || |
3220 | (i == sh->pd_idx && s.failed == 0)) | |
3221 | set_bit(STRIPE_INSYNC, &sh->state); | |
3222 | } | |
3223 | } | |
3224 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
3225 | atomic_dec(&conf->preread_active_stripes); | |
3226 | if (atomic_read(&conf->preread_active_stripes) < | |
3227 | IO_THRESHOLD) | |
3228 | md_wakeup_thread(conf->mddev->thread); | |
3229 | } | |
3230 | } | |
3231 | ||
3232 | /* Now to consider new write requests and what else, if anything | |
3233 | * should be read. We do not handle new writes when: | |
3234 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3235 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3236 | * block. | |
3237 | */ | |
600aa109 | 3238 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 3239 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
3240 | |
3241 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
3242 | * Any reads will already have been scheduled, so we just see if enough |
3243 | * data is available. The parity check is held off while parity | |
3244 | * dependent operations are in flight. | |
1da177e4 | 3245 | */ |
ecc65c9b DW |
3246 | if (sh->check_state || |
3247 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 3248 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 3249 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 3250 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 3251 | |
a4456856 | 3252 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
3253 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3254 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3255 | } | |
4e5314b5 N |
3256 | |
3257 | /* If the failed drive is just a ReadError, then we might need to progress | |
3258 | * the repair/check process | |
3259 | */ | |
a4456856 DW |
3260 | if (s.failed == 1 && !conf->mddev->ro && |
3261 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
3262 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
3263 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 3264 | ) { |
a4456856 | 3265 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
3266 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
3267 | set_bit(R5_Wantwrite, &dev->flags); | |
3268 | set_bit(R5_ReWrite, &dev->flags); | |
3269 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3270 | s.locked++; |
4e5314b5 N |
3271 | } else { |
3272 | /* let's read it back */ | |
3273 | set_bit(R5_Wantread, &dev->flags); | |
3274 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3275 | s.locked++; |
4e5314b5 N |
3276 | } |
3277 | } | |
3278 | ||
600aa109 DW |
3279 | /* Finish reconstruct operations initiated by the expansion process */ |
3280 | if (sh->reconstruct_state == reconstruct_state_result) { | |
ab69ae12 N |
3281 | struct stripe_head *sh2 |
3282 | = get_active_stripe(conf, sh->sector, 1, 1); | |
3283 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { | |
3284 | /* sh cannot be written until sh2 has been read. | |
3285 | * so arrange for sh to be delayed a little | |
3286 | */ | |
3287 | set_bit(STRIPE_DELAYED, &sh->state); | |
3288 | set_bit(STRIPE_HANDLE, &sh->state); | |
3289 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3290 | &sh2->state)) | |
3291 | atomic_inc(&conf->preread_active_stripes); | |
3292 | release_stripe(sh2); | |
3293 | goto unlock; | |
3294 | } | |
3295 | if (sh2) | |
3296 | release_stripe(sh2); | |
3297 | ||
600aa109 | 3298 | sh->reconstruct_state = reconstruct_state_idle; |
f0a50d37 | 3299 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 3300 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 3301 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 3302 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 3303 | s.locked++; |
23397883 | 3304 | } |
f0a50d37 DW |
3305 | } |
3306 | ||
3307 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 3308 | !sh->reconstruct_state) { |
f0a50d37 DW |
3309 | /* Need to write out all blocks after computing parity */ |
3310 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3311 | stripe_set_idx(sh->sector, conf, 0, sh); |
1fe797e6 | 3312 | schedule_reconstruction5(sh, &s, 1, 1); |
600aa109 | 3313 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 3314 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 3315 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
3316 | wake_up(&conf->wait_for_overlap); |
3317 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3318 | } | |
3319 | ||
0f94e87c | 3320 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3321 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3322 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 3323 | |
6bfe0b49 | 3324 | unlock: |
1da177e4 LT |
3325 | spin_unlock(&sh->lock); |
3326 | ||
6bfe0b49 DW |
3327 | /* wait for this device to become unblocked */ |
3328 | if (unlikely(blocked_rdev)) | |
3329 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3330 | ||
600aa109 | 3331 | if (s.ops_request) |
ac6b53b6 | 3332 | raid_run_ops(sh, s.ops_request); |
d84e0f10 | 3333 | |
c4e5ac0a | 3334 | ops_run_io(sh, &s); |
1da177e4 | 3335 | |
a4456856 | 3336 | return_io(return_bi); |
df10cfbc DW |
3337 | |
3338 | return blocked_rdev == NULL; | |
1da177e4 LT |
3339 | } |
3340 | ||
36d1c647 | 3341 | static bool handle_stripe6(struct stripe_head *sh) |
1da177e4 | 3342 | { |
bff61975 | 3343 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 3344 | int disks = sh->disks; |
a4456856 | 3345 | struct bio *return_bi = NULL; |
34e04e87 | 3346 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx; |
a4456856 DW |
3347 | struct stripe_head_state s; |
3348 | struct r6_state r6s; | |
16a53ecc | 3349 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 3350 | mdk_rdev_t *blocked_rdev = NULL; |
1da177e4 | 3351 | |
45b4233c | 3352 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
a4456856 DW |
3353 | "pd_idx=%d, qd_idx=%d\n", |
3354 | (unsigned long long)sh->sector, sh->state, | |
34e04e87 | 3355 | atomic_read(&sh->count), pd_idx, qd_idx); |
a4456856 | 3356 | memset(&s, 0, sizeof(s)); |
72626685 | 3357 | |
16a53ecc N |
3358 | spin_lock(&sh->lock); |
3359 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3360 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3361 | ||
a4456856 DW |
3362 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3363 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3364 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 3365 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3366 | |
3367 | rcu_read_lock(); | |
16a53ecc N |
3368 | for (i=disks; i--; ) { |
3369 | mdk_rdev_t *rdev; | |
3370 | dev = &sh->dev[i]; | |
3371 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 3372 | |
45b4233c | 3373 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc N |
3374 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
3375 | /* maybe we can reply to a read */ | |
3376 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
3377 | struct bio *rbi, *rbi2; | |
45b4233c | 3378 | pr_debug("Return read for disc %d\n", i); |
16a53ecc N |
3379 | spin_lock_irq(&conf->device_lock); |
3380 | rbi = dev->toread; | |
3381 | dev->toread = NULL; | |
3382 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
3383 | wake_up(&conf->wait_for_overlap); | |
3384 | spin_unlock_irq(&conf->device_lock); | |
3385 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
3386 | copy_data(0, rbi, dev->page, dev->sector); | |
3387 | rbi2 = r5_next_bio(rbi, dev->sector); | |
3388 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3389 | if (!raid5_dec_bi_phys_segments(rbi)) { |
16a53ecc N |
3390 | rbi->bi_next = return_bi; |
3391 | return_bi = rbi; | |
3392 | } | |
3393 | spin_unlock_irq(&conf->device_lock); | |
3394 | rbi = rbi2; | |
3395 | } | |
3396 | } | |
1da177e4 | 3397 | |
16a53ecc | 3398 | /* now count some things */ |
a4456856 DW |
3399 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3400 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 3401 | |
16a53ecc | 3402 | |
a4456856 DW |
3403 | if (dev->toread) |
3404 | s.to_read++; | |
16a53ecc | 3405 | if (dev->towrite) { |
a4456856 | 3406 | s.to_write++; |
16a53ecc | 3407 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3408 | s.non_overwrite++; |
16a53ecc | 3409 | } |
a4456856 DW |
3410 | if (dev->written) |
3411 | s.written++; | |
16a53ecc | 3412 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3413 | if (blocked_rdev == NULL && |
3414 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3415 | blocked_rdev = rdev; |
3416 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3417 | } |
16a53ecc N |
3418 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3419 | /* The ReadError flag will just be confusing now */ | |
3420 | clear_bit(R5_ReadError, &dev->flags); | |
3421 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3422 | } |
16a53ecc N |
3423 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3424 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3425 | if (s.failed < 2) |
3426 | r6s.failed_num[s.failed] = i; | |
3427 | s.failed++; | |
16a53ecc N |
3428 | } else |
3429 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3430 | } |
3431 | rcu_read_unlock(); | |
6bfe0b49 DW |
3432 | |
3433 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3434 | if (s.syncing || s.expanding || s.expanded || |
3435 | s.to_write || s.written) { | |
3436 | set_bit(STRIPE_HANDLE, &sh->state); | |
3437 | goto unlock; | |
3438 | } | |
3439 | /* There is nothing for the blocked_rdev to block */ | |
3440 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3441 | blocked_rdev = NULL; | |
6bfe0b49 | 3442 | } |
ac4090d2 | 3443 | |
45b4233c | 3444 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3445 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3446 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3447 | r6s.failed_num[0], r6s.failed_num[1]); | |
3448 | /* check if the array has lost >2 devices and, if so, some requests | |
3449 | * might need to be failed | |
16a53ecc | 3450 | */ |
a4456856 | 3451 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3452 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3453 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3454 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3455 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3456 | s.syncing = 0; |
16a53ecc N |
3457 | } |
3458 | ||
3459 | /* | |
3460 | * might be able to return some write requests if the parity blocks | |
3461 | * are safe, or on a failed drive | |
3462 | */ | |
3463 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3464 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3465 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
34e04e87 N |
3466 | qdev = &sh->dev[qd_idx]; |
3467 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx) | |
3468 | || (s.failed >= 2 && r6s.failed_num[1] == qd_idx); | |
a4456856 DW |
3469 | |
3470 | if ( s.written && | |
3471 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3472 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3473 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3474 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3475 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3476 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3477 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3478 | |
3479 | /* Now we might consider reading some blocks, either to check/generate | |
3480 | * parity, or to satisfy requests | |
3481 | * or to load a block that is being partially written. | |
3482 | */ | |
a4456856 DW |
3483 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
3484 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
1fe797e6 | 3485 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc N |
3486 | |
3487 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 | 3488 | if (s.to_write) |
1fe797e6 | 3489 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3490 | |
3491 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 DW |
3492 | * Any reads will already have been scheduled, so we just see if enough |
3493 | * data is available | |
16a53ecc | 3494 | */ |
a4456856 | 3495 | if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state)) |
36d1c647 | 3496 | handle_parity_checks6(conf, sh, &s, &r6s, disks); |
16a53ecc | 3497 | |
a4456856 | 3498 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3499 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3500 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3501 | } | |
3502 | ||
3503 | /* If the failed drives are just a ReadError, then we might need | |
3504 | * to progress the repair/check process | |
3505 | */ | |
a4456856 DW |
3506 | if (s.failed <= 2 && !conf->mddev->ro) |
3507 | for (i = 0; i < s.failed; i++) { | |
3508 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3509 | if (test_bit(R5_ReadError, &dev->flags) |
3510 | && !test_bit(R5_LOCKED, &dev->flags) | |
3511 | && test_bit(R5_UPTODATE, &dev->flags) | |
3512 | ) { | |
3513 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3514 | set_bit(R5_Wantwrite, &dev->flags); | |
3515 | set_bit(R5_ReWrite, &dev->flags); | |
3516 | set_bit(R5_LOCKED, &dev->flags); | |
3517 | } else { | |
3518 | /* let's read it back */ | |
3519 | set_bit(R5_Wantread, &dev->flags); | |
3520 | set_bit(R5_LOCKED, &dev->flags); | |
3521 | } | |
3522 | } | |
3523 | } | |
f416885e | 3524 | |
a4456856 | 3525 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
ab69ae12 N |
3526 | struct stripe_head *sh2 |
3527 | = get_active_stripe(conf, sh->sector, 1, 1); | |
3528 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { | |
3529 | /* sh cannot be written until sh2 has been read. | |
3530 | * so arrange for sh to be delayed a little | |
3531 | */ | |
3532 | set_bit(STRIPE_DELAYED, &sh->state); | |
3533 | set_bit(STRIPE_HANDLE, &sh->state); | |
3534 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3535 | &sh2->state)) | |
3536 | atomic_inc(&conf->preread_active_stripes); | |
3537 | release_stripe(sh2); | |
3538 | goto unlock; | |
3539 | } | |
3540 | if (sh2) | |
3541 | release_stripe(sh2); | |
3542 | ||
f416885e N |
3543 | /* Need to write out all blocks after computing P&Q */ |
3544 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3545 | stripe_set_idx(sh->sector, conf, 0, sh); |
f416885e N |
3546 | compute_parity6(sh, RECONSTRUCT_WRITE); |
3547 | for (i = conf->raid_disks ; i-- ; ) { | |
3548 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3549 | s.locked++; |
f416885e N |
3550 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
3551 | } | |
3552 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 3553 | } else if (s.expanded) { |
f416885e N |
3554 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3555 | atomic_dec(&conf->reshape_stripes); | |
3556 | wake_up(&conf->wait_for_overlap); | |
3557 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3558 | } | |
3559 | ||
0f94e87c | 3560 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3561 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3562 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3563 | |
6bfe0b49 | 3564 | unlock: |
16a53ecc N |
3565 | spin_unlock(&sh->lock); |
3566 | ||
6bfe0b49 DW |
3567 | /* wait for this device to become unblocked */ |
3568 | if (unlikely(blocked_rdev)) | |
3569 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3570 | ||
f0e43bcd | 3571 | ops_run_io(sh, &s); |
16a53ecc | 3572 | |
f0e43bcd | 3573 | return_io(return_bi); |
df10cfbc DW |
3574 | |
3575 | return blocked_rdev == NULL; | |
16a53ecc N |
3576 | } |
3577 | ||
df10cfbc | 3578 | /* returns true if the stripe was handled */ |
36d1c647 | 3579 | static bool handle_stripe(struct stripe_head *sh) |
16a53ecc N |
3580 | { |
3581 | if (sh->raid_conf->level == 6) | |
36d1c647 | 3582 | return handle_stripe6(sh); |
16a53ecc | 3583 | else |
df10cfbc | 3584 | return handle_stripe5(sh); |
16a53ecc N |
3585 | } |
3586 | ||
16a53ecc N |
3587 | static void raid5_activate_delayed(raid5_conf_t *conf) |
3588 | { | |
3589 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3590 | while (!list_empty(&conf->delayed_list)) { | |
3591 | struct list_head *l = conf->delayed_list.next; | |
3592 | struct stripe_head *sh; | |
3593 | sh = list_entry(l, struct stripe_head, lru); | |
3594 | list_del_init(l); | |
3595 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3596 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3597 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3598 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3599 | } |
6ed3003c N |
3600 | } else |
3601 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3602 | } |
3603 | ||
3604 | static void activate_bit_delay(raid5_conf_t *conf) | |
3605 | { | |
3606 | /* device_lock is held */ | |
3607 | struct list_head head; | |
3608 | list_add(&head, &conf->bitmap_list); | |
3609 | list_del_init(&conf->bitmap_list); | |
3610 | while (!list_empty(&head)) { | |
3611 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3612 | list_del_init(&sh->lru); | |
3613 | atomic_inc(&sh->count); | |
3614 | __release_stripe(conf, sh); | |
3615 | } | |
3616 | } | |
3617 | ||
3618 | static void unplug_slaves(mddev_t *mddev) | |
3619 | { | |
3620 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3621 | int i; | |
3622 | ||
3623 | rcu_read_lock(); | |
3624 | for (i=0; i<mddev->raid_disks; i++) { | |
3625 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
3626 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3627 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3628 | |
3629 | atomic_inc(&rdev->nr_pending); | |
3630 | rcu_read_unlock(); | |
3631 | ||
2ad8b1ef | 3632 | blk_unplug(r_queue); |
16a53ecc N |
3633 | |
3634 | rdev_dec_pending(rdev, mddev); | |
3635 | rcu_read_lock(); | |
3636 | } | |
3637 | } | |
3638 | rcu_read_unlock(); | |
3639 | } | |
3640 | ||
165125e1 | 3641 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3642 | { |
3643 | mddev_t *mddev = q->queuedata; | |
3644 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3645 | unsigned long flags; | |
3646 | ||
3647 | spin_lock_irqsave(&conf->device_lock, flags); | |
3648 | ||
3649 | if (blk_remove_plug(q)) { | |
3650 | conf->seq_flush++; | |
3651 | raid5_activate_delayed(conf); | |
72626685 | 3652 | } |
1da177e4 LT |
3653 | md_wakeup_thread(mddev->thread); |
3654 | ||
3655 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3656 | ||
3657 | unplug_slaves(mddev); | |
3658 | } | |
3659 | ||
f022b2fd N |
3660 | static int raid5_congested(void *data, int bits) |
3661 | { | |
3662 | mddev_t *mddev = data; | |
3663 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3664 | ||
3665 | /* No difference between reads and writes. Just check | |
3666 | * how busy the stripe_cache is | |
3667 | */ | |
3668 | if (conf->inactive_blocked) | |
3669 | return 1; | |
3670 | if (conf->quiesce) | |
3671 | return 1; | |
3672 | if (list_empty_careful(&conf->inactive_list)) | |
3673 | return 1; | |
3674 | ||
3675 | return 0; | |
3676 | } | |
3677 | ||
23032a0e RBJ |
3678 | /* We want read requests to align with chunks where possible, |
3679 | * but write requests don't need to. | |
3680 | */ | |
cc371e66 AK |
3681 | static int raid5_mergeable_bvec(struct request_queue *q, |
3682 | struct bvec_merge_data *bvm, | |
3683 | struct bio_vec *biovec) | |
23032a0e RBJ |
3684 | { |
3685 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3686 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e RBJ |
3687 | int max; |
3688 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
cc371e66 | 3689 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3690 | |
cc371e66 | 3691 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3692 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3693 | ||
784052ec N |
3694 | if (mddev->new_chunk < mddev->chunk_size) |
3695 | chunk_sectors = mddev->new_chunk >> 9; | |
23032a0e RBJ |
3696 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3697 | if (max < 0) max = 0; | |
3698 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3699 | return biovec->bv_len; | |
3700 | else | |
3701 | return max; | |
3702 | } | |
3703 | ||
f679623f RBJ |
3704 | |
3705 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3706 | { | |
3707 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
3708 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
3709 | unsigned int bio_sectors = bio->bi_size >> 9; | |
3710 | ||
784052ec N |
3711 | if (mddev->new_chunk < mddev->chunk_size) |
3712 | chunk_sectors = mddev->new_chunk >> 9; | |
f679623f RBJ |
3713 | return chunk_sectors >= |
3714 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3715 | } | |
3716 | ||
46031f9a RBJ |
3717 | /* |
3718 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3719 | * later sampled by raid5d. | |
3720 | */ | |
3721 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3722 | { | |
3723 | unsigned long flags; | |
3724 | ||
3725 | spin_lock_irqsave(&conf->device_lock, flags); | |
3726 | ||
3727 | bi->bi_next = conf->retry_read_aligned_list; | |
3728 | conf->retry_read_aligned_list = bi; | |
3729 | ||
3730 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3731 | md_wakeup_thread(conf->mddev->thread); | |
3732 | } | |
3733 | ||
3734 | ||
3735 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3736 | { | |
3737 | struct bio *bi; | |
3738 | ||
3739 | bi = conf->retry_read_aligned; | |
3740 | if (bi) { | |
3741 | conf->retry_read_aligned = NULL; | |
3742 | return bi; | |
3743 | } | |
3744 | bi = conf->retry_read_aligned_list; | |
3745 | if(bi) { | |
387bb173 | 3746 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3747 | bi->bi_next = NULL; |
960e739d JA |
3748 | /* |
3749 | * this sets the active strip count to 1 and the processed | |
3750 | * strip count to zero (upper 8 bits) | |
3751 | */ | |
46031f9a | 3752 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3753 | } |
3754 | ||
3755 | return bi; | |
3756 | } | |
3757 | ||
3758 | ||
f679623f RBJ |
3759 | /* |
3760 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3761 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3762 | * first). | |
3763 | * If the read failed.. | |
3764 | */ | |
6712ecf8 | 3765 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3766 | { |
3767 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3768 | mddev_t *mddev; |
3769 | raid5_conf_t *conf; | |
3770 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3771 | mdk_rdev_t *rdev; | |
3772 | ||
f679623f | 3773 | bio_put(bi); |
46031f9a RBJ |
3774 | |
3775 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
3776 | conf = mddev_to_conf(mddev); | |
3777 | rdev = (void*)raid_bi->bi_next; | |
3778 | raid_bi->bi_next = NULL; | |
3779 | ||
3780 | rdev_dec_pending(rdev, conf->mddev); | |
3781 | ||
3782 | if (!error && uptodate) { | |
6712ecf8 | 3783 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3784 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3785 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3786 | return; |
46031f9a RBJ |
3787 | } |
3788 | ||
3789 | ||
45b4233c | 3790 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3791 | |
3792 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3793 | } |
3794 | ||
387bb173 NB |
3795 | static int bio_fits_rdev(struct bio *bi) |
3796 | { | |
165125e1 | 3797 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 NB |
3798 | |
3799 | if ((bi->bi_size>>9) > q->max_sectors) | |
3800 | return 0; | |
3801 | blk_recount_segments(q, bi); | |
960e739d | 3802 | if (bi->bi_phys_segments > q->max_phys_segments) |
387bb173 NB |
3803 | return 0; |
3804 | ||
3805 | if (q->merge_bvec_fn) | |
3806 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3807 | * just just give up | |
3808 | */ | |
3809 | return 0; | |
3810 | ||
3811 | return 1; | |
3812 | } | |
3813 | ||
3814 | ||
165125e1 | 3815 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3816 | { |
3817 | mddev_t *mddev = q->queuedata; | |
3818 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3819 | unsigned int dd_idx; |
f679623f RBJ |
3820 | struct bio* align_bi; |
3821 | mdk_rdev_t *rdev; | |
3822 | ||
3823 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3824 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3825 | return 0; |
3826 | } | |
3827 | /* | |
99c0fb5f | 3828 | * use bio_clone to make a copy of the bio |
f679623f RBJ |
3829 | */ |
3830 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3831 | if (!align_bi) | |
3832 | return 0; | |
3833 | /* | |
3834 | * set bi_end_io to a new function, and set bi_private to the | |
3835 | * original bio. | |
3836 | */ | |
3837 | align_bi->bi_end_io = raid5_align_endio; | |
3838 | align_bi->bi_private = raid_bio; | |
3839 | /* | |
3840 | * compute position | |
3841 | */ | |
112bf897 N |
3842 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3843 | 0, | |
911d4ee8 | 3844 | &dd_idx, NULL); |
f679623f RBJ |
3845 | |
3846 | rcu_read_lock(); | |
3847 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3848 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3849 | atomic_inc(&rdev->nr_pending); |
3850 | rcu_read_unlock(); | |
46031f9a RBJ |
3851 | raid_bio->bi_next = (void*)rdev; |
3852 | align_bi->bi_bdev = rdev->bdev; | |
3853 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3854 | align_bi->bi_sector += rdev->data_offset; | |
3855 | ||
387bb173 NB |
3856 | if (!bio_fits_rdev(align_bi)) { |
3857 | /* too big in some way */ | |
3858 | bio_put(align_bi); | |
3859 | rdev_dec_pending(rdev, mddev); | |
3860 | return 0; | |
3861 | } | |
3862 | ||
46031f9a RBJ |
3863 | spin_lock_irq(&conf->device_lock); |
3864 | wait_event_lock_irq(conf->wait_for_stripe, | |
3865 | conf->quiesce == 0, | |
3866 | conf->device_lock, /* nothing */); | |
3867 | atomic_inc(&conf->active_aligned_reads); | |
3868 | spin_unlock_irq(&conf->device_lock); | |
3869 | ||
f679623f RBJ |
3870 | generic_make_request(align_bi); |
3871 | return 1; | |
3872 | } else { | |
3873 | rcu_read_unlock(); | |
46031f9a | 3874 | bio_put(align_bi); |
f679623f RBJ |
3875 | return 0; |
3876 | } | |
3877 | } | |
3878 | ||
8b3e6cdc DW |
3879 | /* __get_priority_stripe - get the next stripe to process |
3880 | * | |
3881 | * Full stripe writes are allowed to pass preread active stripes up until | |
3882 | * the bypass_threshold is exceeded. In general the bypass_count | |
3883 | * increments when the handle_list is handled before the hold_list; however, it | |
3884 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3885 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3886 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3887 | * handle_list. | |
3888 | */ | |
3889 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3890 | { | |
3891 | struct stripe_head *sh; | |
3892 | ||
3893 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3894 | __func__, | |
3895 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3896 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3897 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3898 | ||
3899 | if (!list_empty(&conf->handle_list)) { | |
3900 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3901 | ||
3902 | if (list_empty(&conf->hold_list)) | |
3903 | conf->bypass_count = 0; | |
3904 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3905 | if (conf->hold_list.next == conf->last_hold) | |
3906 | conf->bypass_count++; | |
3907 | else { | |
3908 | conf->last_hold = conf->hold_list.next; | |
3909 | conf->bypass_count -= conf->bypass_threshold; | |
3910 | if (conf->bypass_count < 0) | |
3911 | conf->bypass_count = 0; | |
3912 | } | |
3913 | } | |
3914 | } else if (!list_empty(&conf->hold_list) && | |
3915 | ((conf->bypass_threshold && | |
3916 | conf->bypass_count > conf->bypass_threshold) || | |
3917 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3918 | sh = list_entry(conf->hold_list.next, | |
3919 | typeof(*sh), lru); | |
3920 | conf->bypass_count -= conf->bypass_threshold; | |
3921 | if (conf->bypass_count < 0) | |
3922 | conf->bypass_count = 0; | |
3923 | } else | |
3924 | return NULL; | |
3925 | ||
3926 | list_del_init(&sh->lru); | |
3927 | atomic_inc(&sh->count); | |
3928 | BUG_ON(atomic_read(&sh->count) != 1); | |
3929 | return sh; | |
3930 | } | |
f679623f | 3931 | |
165125e1 | 3932 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3933 | { |
3934 | mddev_t *mddev = q->queuedata; | |
3935 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3936 | int dd_idx; |
1da177e4 LT |
3937 | sector_t new_sector; |
3938 | sector_t logical_sector, last_sector; | |
3939 | struct stripe_head *sh; | |
a362357b | 3940 | const int rw = bio_data_dir(bi); |
c9959059 | 3941 | int cpu, remaining; |
1da177e4 | 3942 | |
e5dcdd80 | 3943 | if (unlikely(bio_barrier(bi))) { |
6712ecf8 | 3944 | bio_endio(bi, -EOPNOTSUPP); |
e5dcdd80 N |
3945 | return 0; |
3946 | } | |
3947 | ||
3d310eb7 | 3948 | md_write_start(mddev, bi); |
06d91a5f | 3949 | |
074a7aca TH |
3950 | cpu = part_stat_lock(); |
3951 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
3952 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
3953 | bio_sectors(bi)); | |
3954 | part_stat_unlock(); | |
1da177e4 | 3955 | |
802ba064 | 3956 | if (rw == READ && |
52488615 RBJ |
3957 | mddev->reshape_position == MaxSector && |
3958 | chunk_aligned_read(q,bi)) | |
99c0fb5f | 3959 | return 0; |
52488615 | 3960 | |
1da177e4 LT |
3961 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3962 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3963 | bi->bi_next = NULL; | |
3964 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3965 | |
1da177e4 LT |
3966 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3967 | DEFINE_WAIT(w); | |
16a53ecc | 3968 | int disks, data_disks; |
b5663ba4 | 3969 | int previous; |
b578d55f | 3970 | |
7ecaa1e6 | 3971 | retry: |
b5663ba4 | 3972 | previous = 0; |
b0f9ec04 | 3973 | disks = conf->raid_disks; |
b578d55f | 3974 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3975 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3976 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3977 | * 64bit on a 32bit platform, and so it might be |
3978 | * possible to see a half-updated value | |
fef9c61f | 3979 | * Ofcourse reshape_progress could change after |
df8e7f76 N |
3980 | * the lock is dropped, so once we get a reference |
3981 | * to the stripe that we think it is, we will have | |
3982 | * to check again. | |
3983 | */ | |
7ecaa1e6 | 3984 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3985 | if (mddev->delta_disks < 0 |
3986 | ? logical_sector < conf->reshape_progress | |
3987 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3988 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3989 | previous = 1; |
3990 | } else { | |
fef9c61f N |
3991 | if (mddev->delta_disks < 0 |
3992 | ? logical_sector < conf->reshape_safe | |
3993 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3994 | spin_unlock_irq(&conf->device_lock); |
3995 | schedule(); | |
3996 | goto retry; | |
3997 | } | |
3998 | } | |
7ecaa1e6 N |
3999 | spin_unlock_irq(&conf->device_lock); |
4000 | } | |
16a53ecc N |
4001 | data_disks = disks - conf->max_degraded; |
4002 | ||
112bf897 N |
4003 | new_sector = raid5_compute_sector(conf, logical_sector, |
4004 | previous, | |
911d4ee8 | 4005 | &dd_idx, NULL); |
45b4233c | 4006 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
4007 | (unsigned long long)new_sector, |
4008 | (unsigned long long)logical_sector); | |
4009 | ||
b5663ba4 N |
4010 | sh = get_active_stripe(conf, new_sector, previous, |
4011 | (bi->bi_rw&RWA_MASK)); | |
1da177e4 | 4012 | if (sh) { |
b0f9ec04 | 4013 | if (unlikely(previous)) { |
7ecaa1e6 | 4014 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
4015 | * stripe, so we must do the range check again. |
4016 | * Expansion could still move past after this | |
4017 | * test, but as we are holding a reference to | |
4018 | * 'sh', we know that if that happens, | |
4019 | * STRIPE_EXPANDING will get set and the expansion | |
4020 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
4021 | */ |
4022 | int must_retry = 0; | |
4023 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
4024 | if (mddev->delta_disks < 0 |
4025 | ? logical_sector >= conf->reshape_progress | |
4026 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
4027 | /* mismatch, need to try again */ |
4028 | must_retry = 1; | |
4029 | spin_unlock_irq(&conf->device_lock); | |
4030 | if (must_retry) { | |
4031 | release_stripe(sh); | |
4032 | goto retry; | |
4033 | } | |
4034 | } | |
e464eafd N |
4035 | /* FIXME what if we get a false positive because these |
4036 | * are being updated. | |
4037 | */ | |
4038 | if (logical_sector >= mddev->suspend_lo && | |
4039 | logical_sector < mddev->suspend_hi) { | |
4040 | release_stripe(sh); | |
4041 | schedule(); | |
4042 | goto retry; | |
4043 | } | |
7ecaa1e6 N |
4044 | |
4045 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
4046 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
4047 | /* Stripe is busy expanding or | |
4048 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4049 | * and wait a while |
4050 | */ | |
4051 | raid5_unplug_device(mddev->queue); | |
4052 | release_stripe(sh); | |
4053 | schedule(); | |
4054 | goto retry; | |
4055 | } | |
4056 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
4057 | set_bit(STRIPE_HANDLE, &sh->state); |
4058 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1da177e4 | 4059 | release_stripe(sh); |
1da177e4 LT |
4060 | } else { |
4061 | /* cannot get stripe for read-ahead, just give-up */ | |
4062 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4063 | finish_wait(&conf->wait_for_overlap, &w); | |
4064 | break; | |
4065 | } | |
4066 | ||
4067 | } | |
4068 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4069 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
4070 | spin_unlock_irq(&conf->device_lock); |
4071 | if (remaining == 0) { | |
1da177e4 | 4072 | |
16a53ecc | 4073 | if ( rw == WRITE ) |
1da177e4 | 4074 | md_write_end(mddev); |
6712ecf8 | 4075 | |
0e13fe23 | 4076 | bio_endio(bi, 0); |
1da177e4 | 4077 | } |
1da177e4 LT |
4078 | return 0; |
4079 | } | |
4080 | ||
b522adcd DW |
4081 | static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks); |
4082 | ||
52c03291 | 4083 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4084 | { |
52c03291 N |
4085 | /* reshaping is quite different to recovery/resync so it is |
4086 | * handled quite separately ... here. | |
4087 | * | |
4088 | * On each call to sync_request, we gather one chunk worth of | |
4089 | * destination stripes and flag them as expanding. | |
4090 | * Then we find all the source stripes and request reads. | |
4091 | * As the reads complete, handle_stripe will copy the data | |
4092 | * into the destination stripe and release that stripe. | |
4093 | */ | |
1da177e4 LT |
4094 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
4095 | struct stripe_head *sh; | |
ccfcc3c1 | 4096 | sector_t first_sector, last_sector; |
f416885e N |
4097 | int raid_disks = conf->previous_raid_disks; |
4098 | int data_disks = raid_disks - conf->max_degraded; | |
4099 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4100 | int i; |
4101 | int dd_idx; | |
c8f517c4 | 4102 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4103 | sector_t stripe_addr; |
7a661381 | 4104 | int reshape_sectors; |
ab69ae12 | 4105 | struct list_head stripes; |
52c03291 | 4106 | |
fef9c61f N |
4107 | if (sector_nr == 0) { |
4108 | /* If restarting in the middle, skip the initial sectors */ | |
4109 | if (mddev->delta_disks < 0 && | |
4110 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4111 | sector_nr = raid5_size(mddev, 0, 0) | |
4112 | - conf->reshape_progress; | |
4113 | } else if (mddev->delta_disks > 0 && | |
4114 | conf->reshape_progress > 0) | |
4115 | sector_nr = conf->reshape_progress; | |
f416885e | 4116 | sector_div(sector_nr, new_data_disks); |
fef9c61f N |
4117 | if (sector_nr) { |
4118 | *skipped = 1; | |
4119 | return sector_nr; | |
4120 | } | |
52c03291 N |
4121 | } |
4122 | ||
7a661381 N |
4123 | /* We need to process a full chunk at a time. |
4124 | * If old and new chunk sizes differ, we need to process the | |
4125 | * largest of these | |
4126 | */ | |
4127 | if (mddev->new_chunk > mddev->chunk_size) | |
4128 | reshape_sectors = mddev->new_chunk / 512; | |
4129 | else | |
4130 | reshape_sectors = mddev->chunk_size / 512; | |
4131 | ||
52c03291 N |
4132 | /* we update the metadata when there is more than 3Meg |
4133 | * in the block range (that is rather arbitrary, should | |
4134 | * probably be time based) or when the data about to be | |
4135 | * copied would over-write the source of the data at | |
4136 | * the front of the range. | |
fef9c61f N |
4137 | * i.e. one new_stripe along from reshape_progress new_maps |
4138 | * to after where reshape_safe old_maps to | |
52c03291 | 4139 | */ |
fef9c61f | 4140 | writepos = conf->reshape_progress; |
f416885e | 4141 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4142 | readpos = conf->reshape_progress; |
4143 | sector_div(readpos, data_disks); | |
fef9c61f | 4144 | safepos = conf->reshape_safe; |
f416885e | 4145 | sector_div(safepos, data_disks); |
fef9c61f | 4146 | if (mddev->delta_disks < 0) { |
7a661381 | 4147 | writepos -= reshape_sectors; |
c8f517c4 | 4148 | readpos += reshape_sectors; |
7a661381 | 4149 | safepos += reshape_sectors; |
fef9c61f | 4150 | } else { |
7a661381 | 4151 | writepos += reshape_sectors; |
c8f517c4 | 4152 | readpos -= reshape_sectors; |
7a661381 | 4153 | safepos -= reshape_sectors; |
fef9c61f | 4154 | } |
52c03291 | 4155 | |
c8f517c4 N |
4156 | /* 'writepos' is the most advanced device address we might write. |
4157 | * 'readpos' is the least advanced device address we might read. | |
4158 | * 'safepos' is the least address recorded in the metadata as having | |
4159 | * been reshaped. | |
4160 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4161 | * ensure safety in the face of a crash - that must be done by userspace | |
4162 | * making a backup of the data. So in that case there is no particular | |
4163 | * rush to update metadata. | |
4164 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4165 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4166 | * we can be safe in the event of a crash. | |
4167 | * So we insist on updating metadata if safepos is behind writepos and | |
4168 | * readpos is beyond writepos. | |
4169 | * In any case, update the metadata every 10 seconds. | |
4170 | * Maybe that number should be configurable, but I'm not sure it is | |
4171 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4172 | */ | |
fef9c61f | 4173 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4174 | ? (safepos > writepos && readpos < writepos) |
4175 | : (safepos < writepos && readpos > writepos)) || | |
4176 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4177 | /* Cannot proceed until we've updated the superblock... */ |
4178 | wait_event(conf->wait_for_overlap, | |
4179 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4180 | mddev->reshape_position = conf->reshape_progress; |
c8f517c4 | 4181 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4182 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4183 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4184 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4185 | kthread_should_stop()); |
4186 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4187 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4188 | spin_unlock_irq(&conf->device_lock); |
4189 | wake_up(&conf->wait_for_overlap); | |
4190 | } | |
4191 | ||
ec32a2bd N |
4192 | if (mddev->delta_disks < 0) { |
4193 | BUG_ON(conf->reshape_progress == 0); | |
4194 | stripe_addr = writepos; | |
4195 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4196 | ~((sector_t)reshape_sectors - 1)) |
4197 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4198 | != sector_nr); |
4199 | } else { | |
7a661381 | 4200 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4201 | stripe_addr = sector_nr; |
4202 | } | |
ab69ae12 | 4203 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4204 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 N |
4205 | int j; |
4206 | int skipped = 0; | |
ec32a2bd | 4207 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0); |
52c03291 N |
4208 | set_bit(STRIPE_EXPANDING, &sh->state); |
4209 | atomic_inc(&conf->reshape_stripes); | |
4210 | /* If any of this stripe is beyond the end of the old | |
4211 | * array, then we need to zero those blocks | |
4212 | */ | |
4213 | for (j=sh->disks; j--;) { | |
4214 | sector_t s; | |
4215 | if (j == sh->pd_idx) | |
4216 | continue; | |
f416885e | 4217 | if (conf->level == 6 && |
d0dabf7e | 4218 | j == sh->qd_idx) |
f416885e | 4219 | continue; |
784052ec | 4220 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4221 | if (s < raid5_size(mddev, 0, 0)) { |
52c03291 N |
4222 | skipped = 1; |
4223 | continue; | |
4224 | } | |
4225 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4226 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4227 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4228 | } | |
4229 | if (!skipped) { | |
4230 | set_bit(STRIPE_EXPAND_READY, &sh->state); | |
4231 | set_bit(STRIPE_HANDLE, &sh->state); | |
4232 | } | |
ab69ae12 | 4233 | list_add(&sh->lru, &stripes); |
52c03291 N |
4234 | } |
4235 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4236 | if (mddev->delta_disks < 0) |
7a661381 | 4237 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4238 | else |
7a661381 | 4239 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4240 | spin_unlock_irq(&conf->device_lock); |
4241 | /* Ok, those stripe are ready. We can start scheduling | |
4242 | * reads on the source stripes. | |
4243 | * The source stripes are determined by mapping the first and last | |
4244 | * block on the destination stripes. | |
4245 | */ | |
52c03291 | 4246 | first_sector = |
ec32a2bd | 4247 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4248 | 1, &dd_idx, NULL); |
52c03291 | 4249 | last_sector = |
ec32a2bd | 4250 | raid5_compute_sector(conf, ((stripe_addr+conf->chunk_size/512) |
112bf897 | 4251 | *(new_data_disks) - 1), |
911d4ee8 | 4252 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4253 | if (last_sector >= mddev->dev_sectors) |
4254 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4255 | while (first_sector <= last_sector) { |
b5663ba4 | 4256 | sh = get_active_stripe(conf, first_sector, 1, 0); |
52c03291 N |
4257 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4258 | set_bit(STRIPE_HANDLE, &sh->state); | |
4259 | release_stripe(sh); | |
4260 | first_sector += STRIPE_SECTORS; | |
4261 | } | |
ab69ae12 N |
4262 | /* Now that the sources are clearly marked, we can release |
4263 | * the destination stripes | |
4264 | */ | |
4265 | while (!list_empty(&stripes)) { | |
4266 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4267 | list_del_init(&sh->lru); | |
4268 | release_stripe(sh); | |
4269 | } | |
c6207277 N |
4270 | /* If this takes us to the resync_max point where we have to pause, |
4271 | * then we need to write out the superblock. | |
4272 | */ | |
7a661381 | 4273 | sector_nr += reshape_sectors; |
c6207277 N |
4274 | if (sector_nr >= mddev->resync_max) { |
4275 | /* Cannot proceed until we've updated the superblock... */ | |
4276 | wait_event(conf->wait_for_overlap, | |
4277 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4278 | mddev->reshape_position = conf->reshape_progress; |
c8f517c4 | 4279 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4280 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4281 | md_wakeup_thread(mddev->thread); | |
4282 | wait_event(mddev->sb_wait, | |
4283 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4284 | || kthread_should_stop()); | |
4285 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4286 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4287 | spin_unlock_irq(&conf->device_lock); |
4288 | wake_up(&conf->wait_for_overlap); | |
4289 | } | |
7a661381 | 4290 | return reshape_sectors; |
52c03291 N |
4291 | } |
4292 | ||
4293 | /* FIXME go_faster isn't used */ | |
4294 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
4295 | { | |
4296 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4297 | struct stripe_head *sh; | |
58c0fed4 | 4298 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 4299 | int sync_blocks; |
16a53ecc N |
4300 | int still_degraded = 0; |
4301 | int i; | |
1da177e4 | 4302 | |
72626685 | 4303 | if (sector_nr >= max_sector) { |
1da177e4 LT |
4304 | /* just being told to finish up .. nothing much to do */ |
4305 | unplug_slaves(mddev); | |
cea9c228 | 4306 | |
29269553 N |
4307 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4308 | end_reshape(conf); | |
4309 | return 0; | |
4310 | } | |
72626685 N |
4311 | |
4312 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4313 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4314 | &sync_blocks, 1); | |
16a53ecc | 4315 | else /* completed sync */ |
72626685 N |
4316 | conf->fullsync = 0; |
4317 | bitmap_close_sync(mddev->bitmap); | |
4318 | ||
1da177e4 LT |
4319 | return 0; |
4320 | } | |
ccfcc3c1 | 4321 | |
52c03291 N |
4322 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4323 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4324 | |
c6207277 N |
4325 | /* No need to check resync_max as we never do more than one |
4326 | * stripe, and as resync_max will always be on a chunk boundary, | |
4327 | * if the check in md_do_sync didn't fire, there is no chance | |
4328 | * of overstepping resync_max here | |
4329 | */ | |
4330 | ||
16a53ecc | 4331 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4332 | * to resync, then assert that we are finished, because there is |
4333 | * nothing we can do. | |
4334 | */ | |
3285edf1 | 4335 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4336 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4337 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4338 | *skipped = 1; |
1da177e4 LT |
4339 | return rv; |
4340 | } | |
72626685 | 4341 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4342 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4343 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4344 | /* we can skip this block, and probably more */ | |
4345 | sync_blocks /= STRIPE_SECTORS; | |
4346 | *skipped = 1; | |
4347 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4348 | } | |
1da177e4 | 4349 | |
b47490c9 N |
4350 | |
4351 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
4352 | ||
b5663ba4 | 4353 | sh = get_active_stripe(conf, sector_nr, 0, 1); |
1da177e4 | 4354 | if (sh == NULL) { |
b5663ba4 | 4355 | sh = get_active_stripe(conf, sector_nr, 0, 0); |
1da177e4 | 4356 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4357 | * is trying to get access |
1da177e4 | 4358 | */ |
66c006a5 | 4359 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4360 | } |
16a53ecc N |
4361 | /* Need to check if array will still be degraded after recovery/resync |
4362 | * We don't need to check the 'failed' flag as when that gets set, | |
4363 | * recovery aborts. | |
4364 | */ | |
4365 | for (i=0; i<mddev->raid_disks; i++) | |
4366 | if (conf->disks[i].rdev == NULL) | |
4367 | still_degraded = 1; | |
4368 | ||
4369 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4370 | ||
4371 | spin_lock(&sh->lock); | |
1da177e4 LT |
4372 | set_bit(STRIPE_SYNCING, &sh->state); |
4373 | clear_bit(STRIPE_INSYNC, &sh->state); | |
4374 | spin_unlock(&sh->lock); | |
4375 | ||
df10cfbc | 4376 | /* wait for any blocked device to be handled */ |
36d1c647 | 4377 | while (unlikely(!handle_stripe(sh))) |
df10cfbc | 4378 | ; |
1da177e4 LT |
4379 | release_stripe(sh); |
4380 | ||
4381 | return STRIPE_SECTORS; | |
4382 | } | |
4383 | ||
46031f9a RBJ |
4384 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
4385 | { | |
4386 | /* We may not be able to submit a whole bio at once as there | |
4387 | * may not be enough stripe_heads available. | |
4388 | * We cannot pre-allocate enough stripe_heads as we may need | |
4389 | * more than exist in the cache (if we allow ever large chunks). | |
4390 | * So we do one stripe head at a time and record in | |
4391 | * ->bi_hw_segments how many have been done. | |
4392 | * | |
4393 | * We *know* that this entire raid_bio is in one chunk, so | |
4394 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4395 | */ | |
4396 | struct stripe_head *sh; | |
911d4ee8 | 4397 | int dd_idx; |
46031f9a RBJ |
4398 | sector_t sector, logical_sector, last_sector; |
4399 | int scnt = 0; | |
4400 | int remaining; | |
4401 | int handled = 0; | |
4402 | ||
4403 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4404 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4405 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4406 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4407 | ||
4408 | for (; logical_sector < last_sector; | |
387bb173 NB |
4409 | logical_sector += STRIPE_SECTORS, |
4410 | sector += STRIPE_SECTORS, | |
4411 | scnt++) { | |
46031f9a | 4412 | |
960e739d | 4413 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4414 | /* already done this stripe */ |
4415 | continue; | |
4416 | ||
b5663ba4 | 4417 | sh = get_active_stripe(conf, sector, 0, 1); |
46031f9a RBJ |
4418 | |
4419 | if (!sh) { | |
4420 | /* failed to get a stripe - must wait */ | |
960e739d | 4421 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4422 | conf->retry_read_aligned = raid_bio; |
4423 | return handled; | |
4424 | } | |
4425 | ||
4426 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4427 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4428 | release_stripe(sh); | |
960e739d | 4429 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4430 | conf->retry_read_aligned = raid_bio; |
4431 | return handled; | |
4432 | } | |
4433 | ||
36d1c647 | 4434 | handle_stripe(sh); |
46031f9a RBJ |
4435 | release_stripe(sh); |
4436 | handled++; | |
4437 | } | |
4438 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4439 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4440 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4441 | if (remaining == 0) |
4442 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4443 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4444 | wake_up(&conf->wait_for_stripe); | |
4445 | return handled; | |
4446 | } | |
4447 | ||
4448 | ||
4449 | ||
1da177e4 LT |
4450 | /* |
4451 | * This is our raid5 kernel thread. | |
4452 | * | |
4453 | * We scan the hash table for stripes which can be handled now. | |
4454 | * During the scan, completed stripes are saved for us by the interrupt | |
4455 | * handler, so that they will not have to wait for our next wakeup. | |
4456 | */ | |
6ed3003c | 4457 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4458 | { |
4459 | struct stripe_head *sh; | |
4460 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4461 | int handled; | |
4462 | ||
45b4233c | 4463 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4464 | |
4465 | md_check_recovery(mddev); | |
1da177e4 LT |
4466 | |
4467 | handled = 0; | |
4468 | spin_lock_irq(&conf->device_lock); | |
4469 | while (1) { | |
46031f9a | 4470 | struct bio *bio; |
1da177e4 | 4471 | |
ae3c20cc | 4472 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 4473 | int seq = conf->seq_flush; |
700e432d | 4474 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4475 | bitmap_unplug(mddev->bitmap); |
700e432d | 4476 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4477 | conf->seq_write = seq; |
4478 | activate_bit_delay(conf); | |
4479 | } | |
4480 | ||
46031f9a RBJ |
4481 | while ((bio = remove_bio_from_retry(conf))) { |
4482 | int ok; | |
4483 | spin_unlock_irq(&conf->device_lock); | |
4484 | ok = retry_aligned_read(conf, bio); | |
4485 | spin_lock_irq(&conf->device_lock); | |
4486 | if (!ok) | |
4487 | break; | |
4488 | handled++; | |
4489 | } | |
4490 | ||
8b3e6cdc DW |
4491 | sh = __get_priority_stripe(conf); |
4492 | ||
c9f21aaf | 4493 | if (!sh) |
1da177e4 | 4494 | break; |
1da177e4 LT |
4495 | spin_unlock_irq(&conf->device_lock); |
4496 | ||
4497 | handled++; | |
36d1c647 | 4498 | handle_stripe(sh); |
1da177e4 LT |
4499 | release_stripe(sh); |
4500 | ||
4501 | spin_lock_irq(&conf->device_lock); | |
4502 | } | |
45b4233c | 4503 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4504 | |
4505 | spin_unlock_irq(&conf->device_lock); | |
4506 | ||
c9f21aaf | 4507 | async_tx_issue_pending_all(); |
1da177e4 LT |
4508 | unplug_slaves(mddev); |
4509 | ||
45b4233c | 4510 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4511 | } |
4512 | ||
3f294f4f | 4513 | static ssize_t |
007583c9 | 4514 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4515 | { |
007583c9 | 4516 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4517 | if (conf) |
4518 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4519 | else | |
4520 | return 0; | |
3f294f4f N |
4521 | } |
4522 | ||
4523 | static ssize_t | |
007583c9 | 4524 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4525 | { |
007583c9 | 4526 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4ef197d8 | 4527 | unsigned long new; |
b5470dc5 DW |
4528 | int err; |
4529 | ||
3f294f4f N |
4530 | if (len >= PAGE_SIZE) |
4531 | return -EINVAL; | |
96de1e66 N |
4532 | if (!conf) |
4533 | return -ENODEV; | |
3f294f4f | 4534 | |
4ef197d8 | 4535 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4536 | return -EINVAL; |
4537 | if (new <= 16 || new > 32768) | |
4538 | return -EINVAL; | |
4539 | while (new < conf->max_nr_stripes) { | |
4540 | if (drop_one_stripe(conf)) | |
4541 | conf->max_nr_stripes--; | |
4542 | else | |
4543 | break; | |
4544 | } | |
b5470dc5 DW |
4545 | err = md_allow_write(mddev); |
4546 | if (err) | |
4547 | return err; | |
3f294f4f N |
4548 | while (new > conf->max_nr_stripes) { |
4549 | if (grow_one_stripe(conf)) | |
4550 | conf->max_nr_stripes++; | |
4551 | else break; | |
4552 | } | |
4553 | return len; | |
4554 | } | |
007583c9 | 4555 | |
96de1e66 N |
4556 | static struct md_sysfs_entry |
4557 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4558 | raid5_show_stripe_cache_size, | |
4559 | raid5_store_stripe_cache_size); | |
3f294f4f | 4560 | |
8b3e6cdc DW |
4561 | static ssize_t |
4562 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4563 | { | |
4564 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4565 | if (conf) | |
4566 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4567 | else | |
4568 | return 0; | |
4569 | } | |
4570 | ||
4571 | static ssize_t | |
4572 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4573 | { | |
4574 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4ef197d8 | 4575 | unsigned long new; |
8b3e6cdc DW |
4576 | if (len >= PAGE_SIZE) |
4577 | return -EINVAL; | |
4578 | if (!conf) | |
4579 | return -ENODEV; | |
4580 | ||
4ef197d8 | 4581 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4582 | return -EINVAL; |
4ef197d8 | 4583 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4584 | return -EINVAL; |
4585 | conf->bypass_threshold = new; | |
4586 | return len; | |
4587 | } | |
4588 | ||
4589 | static struct md_sysfs_entry | |
4590 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4591 | S_IRUGO | S_IWUSR, | |
4592 | raid5_show_preread_threshold, | |
4593 | raid5_store_preread_threshold); | |
4594 | ||
3f294f4f | 4595 | static ssize_t |
96de1e66 | 4596 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4597 | { |
007583c9 | 4598 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4599 | if (conf) |
4600 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4601 | else | |
4602 | return 0; | |
3f294f4f N |
4603 | } |
4604 | ||
96de1e66 N |
4605 | static struct md_sysfs_entry |
4606 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4607 | |
007583c9 | 4608 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4609 | &raid5_stripecache_size.attr, |
4610 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4611 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4612 | NULL, |
4613 | }; | |
007583c9 N |
4614 | static struct attribute_group raid5_attrs_group = { |
4615 | .name = NULL, | |
4616 | .attrs = raid5_attrs, | |
3f294f4f N |
4617 | }; |
4618 | ||
80c3a6ce DW |
4619 | static sector_t |
4620 | raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
4621 | { | |
4622 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4623 | ||
4624 | if (!sectors) | |
4625 | sectors = mddev->dev_sectors; | |
7ec05478 N |
4626 | if (!raid_disks) { |
4627 | /* size is defined by the smallest of previous and new size */ | |
4628 | if (conf->raid_disks < conf->previous_raid_disks) | |
4629 | raid_disks = conf->raid_disks; | |
4630 | else | |
4631 | raid_disks = conf->previous_raid_disks; | |
4632 | } | |
80c3a6ce DW |
4633 | |
4634 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); | |
784052ec | 4635 | sectors &= ~((sector_t)mddev->new_chunk/512 - 1); |
80c3a6ce DW |
4636 | return sectors * (raid_disks - conf->max_degraded); |
4637 | } | |
4638 | ||
36d1c647 DW |
4639 | static void raid5_free_percpu(raid5_conf_t *conf) |
4640 | { | |
4641 | struct raid5_percpu *percpu; | |
4642 | unsigned long cpu; | |
4643 | ||
4644 | if (!conf->percpu) | |
4645 | return; | |
4646 | ||
4647 | get_online_cpus(); | |
4648 | for_each_possible_cpu(cpu) { | |
4649 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4650 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4651 | kfree(percpu->scribble); |
36d1c647 DW |
4652 | } |
4653 | #ifdef CONFIG_HOTPLUG_CPU | |
4654 | unregister_cpu_notifier(&conf->cpu_notify); | |
4655 | #endif | |
4656 | put_online_cpus(); | |
4657 | ||
4658 | free_percpu(conf->percpu); | |
4659 | } | |
4660 | ||
a11034b4 DW |
4661 | static void free_conf(raid5_conf_t *conf) |
4662 | { | |
4663 | shrink_stripes(conf); | |
36d1c647 | 4664 | raid5_free_percpu(conf); |
a11034b4 DW |
4665 | kfree(conf->disks); |
4666 | kfree(conf->stripe_hashtbl); | |
4667 | kfree(conf); | |
4668 | } | |
4669 | ||
36d1c647 DW |
4670 | #ifdef CONFIG_HOTPLUG_CPU |
4671 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4672 | void *hcpu) | |
4673 | { | |
4674 | raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify); | |
4675 | long cpu = (long)hcpu; | |
4676 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4677 | ||
4678 | switch (action) { | |
4679 | case CPU_UP_PREPARE: | |
4680 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4681 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4682 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4683 | if (!percpu->scribble) |
4684 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4685 | ||
4686 | if (!percpu->scribble || | |
4687 | (conf->level == 6 && !percpu->spare_page)) { | |
4688 | safe_put_page(percpu->spare_page); | |
4689 | kfree(percpu->scribble); | |
36d1c647 DW |
4690 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4691 | __func__, cpu); | |
4692 | return NOTIFY_BAD; | |
4693 | } | |
4694 | break; | |
4695 | case CPU_DEAD: | |
4696 | case CPU_DEAD_FROZEN: | |
4697 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4698 | kfree(percpu->scribble); |
36d1c647 | 4699 | percpu->spare_page = NULL; |
d6f38f31 | 4700 | percpu->scribble = NULL; |
36d1c647 DW |
4701 | break; |
4702 | default: | |
4703 | break; | |
4704 | } | |
4705 | return NOTIFY_OK; | |
4706 | } | |
4707 | #endif | |
4708 | ||
4709 | static int raid5_alloc_percpu(raid5_conf_t *conf) | |
4710 | { | |
4711 | unsigned long cpu; | |
4712 | struct page *spare_page; | |
4713 | struct raid5_percpu *allcpus; | |
d6f38f31 | 4714 | void *scribble; |
36d1c647 DW |
4715 | int err; |
4716 | ||
36d1c647 DW |
4717 | allcpus = alloc_percpu(struct raid5_percpu); |
4718 | if (!allcpus) | |
4719 | return -ENOMEM; | |
4720 | conf->percpu = allcpus; | |
4721 | ||
4722 | get_online_cpus(); | |
4723 | err = 0; | |
4724 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4725 | if (conf->level == 6) { |
4726 | spare_page = alloc_page(GFP_KERNEL); | |
4727 | if (!spare_page) { | |
4728 | err = -ENOMEM; | |
4729 | break; | |
4730 | } | |
4731 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4732 | } | |
4733 | scribble = kmalloc(scribble_len(conf->raid_disks), GFP_KERNEL); | |
4734 | if (!scribble) { | |
36d1c647 DW |
4735 | err = -ENOMEM; |
4736 | break; | |
4737 | } | |
d6f38f31 | 4738 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4739 | } |
4740 | #ifdef CONFIG_HOTPLUG_CPU | |
4741 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4742 | conf->cpu_notify.priority = 0; | |
4743 | if (err == 0) | |
4744 | err = register_cpu_notifier(&conf->cpu_notify); | |
4745 | #endif | |
4746 | put_online_cpus(); | |
4747 | ||
4748 | return err; | |
4749 | } | |
4750 | ||
91adb564 | 4751 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4752 | { |
4753 | raid5_conf_t *conf; | |
4754 | int raid_disk, memory; | |
4755 | mdk_rdev_t *rdev; | |
4756 | struct disk_info *disk; | |
1da177e4 | 4757 | |
91adb564 N |
4758 | if (mddev->new_level != 5 |
4759 | && mddev->new_level != 4 | |
4760 | && mddev->new_level != 6) { | |
16a53ecc | 4761 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4762 | mdname(mddev), mddev->new_level); |
4763 | return ERR_PTR(-EIO); | |
1da177e4 | 4764 | } |
91adb564 N |
4765 | if ((mddev->new_level == 5 |
4766 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4767 | (mddev->new_level == 6 | |
4768 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
99c0fb5f | 4769 | printk(KERN_ERR "raid5: %s: layout %d not supported\n", |
91adb564 N |
4770 | mdname(mddev), mddev->new_layout); |
4771 | return ERR_PTR(-EIO); | |
99c0fb5f | 4772 | } |
91adb564 N |
4773 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
4774 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4775 | mdname(mddev), mddev->raid_disks); | |
4776 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4777 | } |
4778 | ||
91adb564 N |
4779 | if (!mddev->new_chunk || mddev->new_chunk % PAGE_SIZE) { |
4780 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
4781 | mddev->new_chunk, mdname(mddev)); | |
4782 | return ERR_PTR(-EINVAL); | |
f6705578 N |
4783 | } |
4784 | ||
91adb564 N |
4785 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4786 | if (conf == NULL) | |
1da177e4 | 4787 | goto abort; |
91adb564 N |
4788 | |
4789 | conf->raid_disks = mddev->raid_disks; | |
d6f38f31 | 4790 | conf->scribble_len = scribble_len(conf->raid_disks); |
91adb564 N |
4791 | if (mddev->reshape_position == MaxSector) |
4792 | conf->previous_raid_disks = mddev->raid_disks; | |
4793 | else | |
f6705578 | 4794 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
f6705578 N |
4795 | |
4796 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
4797 | GFP_KERNEL); |
4798 | if (!conf->disks) | |
4799 | goto abort; | |
9ffae0cf | 4800 | |
1da177e4 LT |
4801 | conf->mddev = mddev; |
4802 | ||
fccddba0 | 4803 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4804 | goto abort; |
1da177e4 | 4805 | |
36d1c647 DW |
4806 | conf->level = mddev->new_level; |
4807 | if (raid5_alloc_percpu(conf) != 0) | |
4808 | goto abort; | |
4809 | ||
1da177e4 LT |
4810 | spin_lock_init(&conf->device_lock); |
4811 | init_waitqueue_head(&conf->wait_for_stripe); | |
4812 | init_waitqueue_head(&conf->wait_for_overlap); | |
4813 | INIT_LIST_HEAD(&conf->handle_list); | |
8b3e6cdc | 4814 | INIT_LIST_HEAD(&conf->hold_list); |
1da177e4 | 4815 | INIT_LIST_HEAD(&conf->delayed_list); |
72626685 | 4816 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
4817 | INIT_LIST_HEAD(&conf->inactive_list); |
4818 | atomic_set(&conf->active_stripes, 0); | |
4819 | atomic_set(&conf->preread_active_stripes, 0); | |
46031f9a | 4820 | atomic_set(&conf->active_aligned_reads, 0); |
8b3e6cdc | 4821 | conf->bypass_threshold = BYPASS_THRESHOLD; |
1da177e4 | 4822 | |
45b4233c | 4823 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4824 | |
159ec1fc | 4825 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4826 | raid_disk = rdev->raid_disk; |
f6705578 | 4827 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
4828 | || raid_disk < 0) |
4829 | continue; | |
4830 | disk = conf->disks + raid_disk; | |
4831 | ||
4832 | disk->rdev = rdev; | |
4833 | ||
b2d444d7 | 4834 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4835 | char b[BDEVNAME_SIZE]; |
4836 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4837 | " disk %d\n", bdevname(rdev->bdev,b), | |
4838 | raid_disk); | |
8c2e870a NB |
4839 | } else |
4840 | /* Cannot rely on bitmap to complete recovery */ | |
4841 | conf->fullsync = 1; | |
1da177e4 LT |
4842 | } |
4843 | ||
91adb564 | 4844 | conf->chunk_size = mddev->new_chunk; |
16a53ecc N |
4845 | if (conf->level == 6) |
4846 | conf->max_degraded = 2; | |
4847 | else | |
4848 | conf->max_degraded = 1; | |
91adb564 | 4849 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4850 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4851 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4852 | if (conf->reshape_progress != MaxSector) { |
784052ec | 4853 | conf->prev_chunk = mddev->chunk_size; |
e183eaed N |
4854 | conf->prev_algo = mddev->layout; |
4855 | } | |
1da177e4 | 4856 | |
91adb564 N |
4857 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
4858 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; | |
4859 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
4860 | printk(KERN_ERR | |
4861 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4862 | goto abort; | |
4863 | } else | |
4864 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4865 | memory, mdname(mddev)); | |
1da177e4 | 4866 | |
91adb564 N |
4867 | conf->thread = md_register_thread(raid5d, mddev, "%s_raid5"); |
4868 | if (!conf->thread) { | |
4869 | printk(KERN_ERR | |
4870 | "raid5: couldn't allocate thread for %s\n", | |
4871 | mdname(mddev)); | |
16a53ecc N |
4872 | goto abort; |
4873 | } | |
91adb564 N |
4874 | |
4875 | return conf; | |
4876 | ||
4877 | abort: | |
4878 | if (conf) { | |
a11034b4 | 4879 | free_conf(conf); |
91adb564 N |
4880 | return ERR_PTR(-EIO); |
4881 | } else | |
4882 | return ERR_PTR(-ENOMEM); | |
4883 | } | |
4884 | ||
4885 | static int run(mddev_t *mddev) | |
4886 | { | |
4887 | raid5_conf_t *conf; | |
4888 | int working_disks = 0; | |
4889 | mdk_rdev_t *rdev; | |
4890 | ||
4891 | if (mddev->reshape_position != MaxSector) { | |
4892 | /* Check that we can continue the reshape. | |
4893 | * Currently only disks can change, it must | |
4894 | * increase, and we must be past the point where | |
4895 | * a stripe over-writes itself | |
4896 | */ | |
4897 | sector_t here_new, here_old; | |
4898 | int old_disks; | |
18b00334 | 4899 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4900 | |
88ce4930 | 4901 | if (mddev->new_level != mddev->level) { |
91adb564 N |
4902 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4903 | "required - aborting.\n", | |
4904 | mdname(mddev)); | |
4905 | return -EINVAL; | |
4906 | } | |
91adb564 N |
4907 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4908 | /* reshape_position must be on a new-stripe boundary, and one | |
4909 | * further up in new geometry must map after here in old | |
4910 | * geometry. | |
4911 | */ | |
4912 | here_new = mddev->reshape_position; | |
784052ec | 4913 | if (sector_div(here_new, (mddev->new_chunk>>9)* |
91adb564 N |
4914 | (mddev->raid_disks - max_degraded))) { |
4915 | printk(KERN_ERR "raid5: reshape_position not " | |
4916 | "on a stripe boundary\n"); | |
4917 | return -EINVAL; | |
4918 | } | |
4919 | /* here_new is the stripe we will write to */ | |
4920 | here_old = mddev->reshape_position; | |
4921 | sector_div(here_old, (mddev->chunk_size>>9)* | |
4922 | (old_disks-max_degraded)); | |
4923 | /* here_old is the first stripe that we might need to read | |
4924 | * from */ | |
4925 | if (here_new >= here_old) { | |
4926 | /* Reading from the same stripe as writing to - bad */ | |
4927 | printk(KERN_ERR "raid5: reshape_position too early for " | |
4928 | "auto-recovery - aborting.\n"); | |
4929 | return -EINVAL; | |
4930 | } | |
4931 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4932 | /* OK, we should be able to continue; */ | |
4933 | } else { | |
4934 | BUG_ON(mddev->level != mddev->new_level); | |
4935 | BUG_ON(mddev->layout != mddev->new_layout); | |
4936 | BUG_ON(mddev->chunk_size != mddev->new_chunk); | |
4937 | BUG_ON(mddev->delta_disks != 0); | |
1da177e4 | 4938 | } |
91adb564 | 4939 | |
245f46c2 N |
4940 | if (mddev->private == NULL) |
4941 | conf = setup_conf(mddev); | |
4942 | else | |
4943 | conf = mddev->private; | |
4944 | ||
91adb564 N |
4945 | if (IS_ERR(conf)) |
4946 | return PTR_ERR(conf); | |
4947 | ||
4948 | mddev->thread = conf->thread; | |
4949 | conf->thread = NULL; | |
4950 | mddev->private = conf; | |
4951 | ||
4952 | /* | |
4953 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4954 | */ | |
4955 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4956 | if (rdev->raid_disk >= 0 && | |
4957 | test_bit(In_sync, &rdev->flags)) | |
4958 | working_disks++; | |
4959 | ||
4960 | mddev->degraded = conf->raid_disks - working_disks; | |
4961 | ||
16a53ecc | 4962 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
4963 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
4964 | " (%d/%d failed)\n", | |
02c2de8c | 4965 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4966 | goto abort; |
4967 | } | |
4968 | ||
91adb564 N |
4969 | /* device size must be a multiple of chunk size */ |
4970 | mddev->dev_sectors &= ~(mddev->chunk_size / 512 - 1); | |
4971 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4972 | ||
16a53ecc | 4973 | if (mddev->degraded > 0 && |
1da177e4 | 4974 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4975 | if (mddev->ok_start_degraded) |
4976 | printk(KERN_WARNING | |
4977 | "raid5: starting dirty degraded array: %s" | |
4978 | "- data corruption possible.\n", | |
4979 | mdname(mddev)); | |
4980 | else { | |
4981 | printk(KERN_ERR | |
4982 | "raid5: cannot start dirty degraded array for %s\n", | |
4983 | mdname(mddev)); | |
4984 | goto abort; | |
4985 | } | |
1da177e4 LT |
4986 | } |
4987 | ||
1da177e4 LT |
4988 | if (mddev->degraded == 0) |
4989 | printk("raid5: raid level %d set %s active with %d out of %d" | |
e183eaed N |
4990 | " devices, algorithm %d\n", conf->level, mdname(mddev), |
4991 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
4992 | mddev->new_layout); | |
1da177e4 LT |
4993 | else |
4994 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
4995 | " out of %d devices, algorithm %d\n", conf->level, | |
4996 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
e183eaed | 4997 | mddev->raid_disks, mddev->new_layout); |
1da177e4 LT |
4998 | |
4999 | print_raid5_conf(conf); | |
5000 | ||
fef9c61f | 5001 | if (conf->reshape_progress != MaxSector) { |
f6705578 | 5002 | printk("...ok start reshape thread\n"); |
fef9c61f | 5003 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5004 | atomic_set(&conf->reshape_stripes, 0); |
5005 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5006 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5007 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5008 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5009 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
5010 | "%s_reshape"); | |
f6705578 N |
5011 | } |
5012 | ||
1da177e4 | 5013 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 5014 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
5015 | */ |
5016 | { | |
16a53ecc N |
5017 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
5018 | int stripe = data_disks * | |
8932c2e0 | 5019 | (mddev->chunk_size / PAGE_SIZE); |
1da177e4 LT |
5020 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5021 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5022 | } | |
5023 | ||
5024 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
5025 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
5026 | printk(KERN_WARNING | |
5027 | "raid5: failed to create sysfs attributes for %s\n", | |
5028 | mdname(mddev)); | |
7a5febe9 | 5029 | |
91adb564 N |
5030 | mddev->queue->queue_lock = &conf->device_lock; |
5031 | ||
7a5febe9 | 5032 | mddev->queue->unplug_fn = raid5_unplug_device; |
f022b2fd | 5033 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 5034 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 5035 | |
1f403624 | 5036 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5037 | |
23032a0e RBJ |
5038 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
5039 | ||
1da177e4 LT |
5040 | return 0; |
5041 | abort: | |
e0cf8f04 | 5042 | md_unregister_thread(mddev->thread); |
91adb564 | 5043 | mddev->thread = NULL; |
1da177e4 LT |
5044 | if (conf) { |
5045 | print_raid5_conf(conf); | |
a11034b4 | 5046 | free_conf(conf); |
1da177e4 LT |
5047 | } |
5048 | mddev->private = NULL; | |
5049 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
5050 | return -EIO; | |
5051 | } | |
5052 | ||
5053 | ||
5054 | ||
3f294f4f | 5055 | static int stop(mddev_t *mddev) |
1da177e4 LT |
5056 | { |
5057 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
5058 | ||
5059 | md_unregister_thread(mddev->thread); | |
5060 | mddev->thread = NULL; | |
041ae52e | 5061 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 5062 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 5063 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
a11034b4 | 5064 | free_conf(conf); |
1da177e4 LT |
5065 | mddev->private = NULL; |
5066 | return 0; | |
5067 | } | |
5068 | ||
45b4233c | 5069 | #ifdef DEBUG |
d710e138 | 5070 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
5071 | { |
5072 | int i; | |
5073 | ||
16a53ecc N |
5074 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
5075 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
5076 | seq_printf(seq, "sh %llu, count %d.\n", | |
5077 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
5078 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 5079 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
5080 | seq_printf(seq, "(cache%d: %p %ld) ", |
5081 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 5082 | } |
16a53ecc | 5083 | seq_printf(seq, "\n"); |
1da177e4 LT |
5084 | } |
5085 | ||
d710e138 | 5086 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
5087 | { |
5088 | struct stripe_head *sh; | |
fccddba0 | 5089 | struct hlist_node *hn; |
1da177e4 LT |
5090 | int i; |
5091 | ||
5092 | spin_lock_irq(&conf->device_lock); | |
5093 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 5094 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
5095 | if (sh->raid_conf != conf) |
5096 | continue; | |
16a53ecc | 5097 | print_sh(seq, sh); |
1da177e4 LT |
5098 | } |
5099 | } | |
5100 | spin_unlock_irq(&conf->device_lock); | |
5101 | } | |
5102 | #endif | |
5103 | ||
d710e138 | 5104 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
5105 | { |
5106 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
5107 | int i; | |
5108 | ||
5109 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
02c2de8c | 5110 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5111 | for (i = 0; i < conf->raid_disks; i++) |
5112 | seq_printf (seq, "%s", | |
5113 | conf->disks[i].rdev && | |
b2d444d7 | 5114 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5115 | seq_printf (seq, "]"); |
45b4233c | 5116 | #ifdef DEBUG |
16a53ecc N |
5117 | seq_printf (seq, "\n"); |
5118 | printall(seq, conf); | |
1da177e4 LT |
5119 | #endif |
5120 | } | |
5121 | ||
5122 | static void print_raid5_conf (raid5_conf_t *conf) | |
5123 | { | |
5124 | int i; | |
5125 | struct disk_info *tmp; | |
5126 | ||
5127 | printk("RAID5 conf printout:\n"); | |
5128 | if (!conf) { | |
5129 | printk("(conf==NULL)\n"); | |
5130 | return; | |
5131 | } | |
02c2de8c N |
5132 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
5133 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5134 | |
5135 | for (i = 0; i < conf->raid_disks; i++) { | |
5136 | char b[BDEVNAME_SIZE]; | |
5137 | tmp = conf->disks + i; | |
5138 | if (tmp->rdev) | |
5139 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 5140 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
5141 | bdevname(tmp->rdev->bdev,b)); |
5142 | } | |
5143 | } | |
5144 | ||
5145 | static int raid5_spare_active(mddev_t *mddev) | |
5146 | { | |
5147 | int i; | |
5148 | raid5_conf_t *conf = mddev->private; | |
5149 | struct disk_info *tmp; | |
5150 | ||
5151 | for (i = 0; i < conf->raid_disks; i++) { | |
5152 | tmp = conf->disks + i; | |
5153 | if (tmp->rdev | |
b2d444d7 | 5154 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
5155 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
5156 | unsigned long flags; | |
5157 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 5158 | mddev->degraded--; |
c04be0aa | 5159 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
5160 | } |
5161 | } | |
5162 | print_raid5_conf(conf); | |
5163 | return 0; | |
5164 | } | |
5165 | ||
5166 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
5167 | { | |
5168 | raid5_conf_t *conf = mddev->private; | |
5169 | int err = 0; | |
5170 | mdk_rdev_t *rdev; | |
5171 | struct disk_info *p = conf->disks + number; | |
5172 | ||
5173 | print_raid5_conf(conf); | |
5174 | rdev = p->rdev; | |
5175 | if (rdev) { | |
ec32a2bd N |
5176 | if (number >= conf->raid_disks && |
5177 | conf->reshape_progress == MaxSector) | |
5178 | clear_bit(In_sync, &rdev->flags); | |
5179 | ||
b2d444d7 | 5180 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
5181 | atomic_read(&rdev->nr_pending)) { |
5182 | err = -EBUSY; | |
5183 | goto abort; | |
5184 | } | |
dfc70645 N |
5185 | /* Only remove non-faulty devices if recovery |
5186 | * isn't possible. | |
5187 | */ | |
5188 | if (!test_bit(Faulty, &rdev->flags) && | |
ec32a2bd N |
5189 | mddev->degraded <= conf->max_degraded && |
5190 | number < conf->raid_disks) { | |
dfc70645 N |
5191 | err = -EBUSY; |
5192 | goto abort; | |
5193 | } | |
1da177e4 | 5194 | p->rdev = NULL; |
fbd568a3 | 5195 | synchronize_rcu(); |
1da177e4 LT |
5196 | if (atomic_read(&rdev->nr_pending)) { |
5197 | /* lost the race, try later */ | |
5198 | err = -EBUSY; | |
5199 | p->rdev = rdev; | |
5200 | } | |
5201 | } | |
5202 | abort: | |
5203 | ||
5204 | print_raid5_conf(conf); | |
5205 | return err; | |
5206 | } | |
5207 | ||
5208 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
5209 | { | |
5210 | raid5_conf_t *conf = mddev->private; | |
199050ea | 5211 | int err = -EEXIST; |
1da177e4 LT |
5212 | int disk; |
5213 | struct disk_info *p; | |
6c2fce2e NB |
5214 | int first = 0; |
5215 | int last = conf->raid_disks - 1; | |
1da177e4 | 5216 | |
16a53ecc | 5217 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 5218 | /* no point adding a device */ |
199050ea | 5219 | return -EINVAL; |
1da177e4 | 5220 | |
6c2fce2e NB |
5221 | if (rdev->raid_disk >= 0) |
5222 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5223 | |
5224 | /* | |
16a53ecc N |
5225 | * find the disk ... but prefer rdev->saved_raid_disk |
5226 | * if possible. | |
1da177e4 | 5227 | */ |
16a53ecc | 5228 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5229 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5230 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5231 | disk = rdev->saved_raid_disk; | |
5232 | else | |
6c2fce2e NB |
5233 | disk = first; |
5234 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5235 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5236 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5237 | rdev->raid_disk = disk; |
199050ea | 5238 | err = 0; |
72626685 N |
5239 | if (rdev->saved_raid_disk != disk) |
5240 | conf->fullsync = 1; | |
d6065f7b | 5241 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5242 | break; |
5243 | } | |
5244 | print_raid5_conf(conf); | |
199050ea | 5245 | return err; |
1da177e4 LT |
5246 | } |
5247 | ||
5248 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
5249 | { | |
5250 | /* no resync is happening, and there is enough space | |
5251 | * on all devices, so we can resize. | |
5252 | * We need to make sure resync covers any new space. | |
5253 | * If the array is shrinking we should possibly wait until | |
5254 | * any io in the removed space completes, but it hardly seems | |
5255 | * worth it. | |
5256 | */ | |
5257 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); | |
1f403624 DW |
5258 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5259 | mddev->raid_disks)); | |
b522adcd DW |
5260 | if (mddev->array_sectors > |
5261 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5262 | return -EINVAL; | |
f233ea5c | 5263 | set_capacity(mddev->gendisk, mddev->array_sectors); |
44ce6294 | 5264 | mddev->changed = 1; |
58c0fed4 AN |
5265 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
5266 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
5267 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5268 | } | |
58c0fed4 | 5269 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5270 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5271 | return 0; |
5272 | } | |
5273 | ||
63c70c4f | 5274 | static int raid5_check_reshape(mddev_t *mddev) |
29269553 N |
5275 | { |
5276 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
29269553 | 5277 | |
88ce4930 N |
5278 | if (mddev->delta_disks == 0 && |
5279 | mddev->new_layout == mddev->layout && | |
5280 | mddev->new_chunk == mddev->chunk_size) | |
5281 | return -EINVAL; /* nothing to do */ | |
dba034ee N |
5282 | if (mddev->bitmap) |
5283 | /* Cannot grow a bitmap yet */ | |
5284 | return -EBUSY; | |
ec32a2bd N |
5285 | if (mddev->degraded > conf->max_degraded) |
5286 | return -EINVAL; | |
5287 | if (mddev->delta_disks < 0) { | |
5288 | /* We might be able to shrink, but the devices must | |
5289 | * be made bigger first. | |
5290 | * For raid6, 4 is the minimum size. | |
5291 | * Otherwise 2 is the minimum | |
5292 | */ | |
5293 | int min = 2; | |
5294 | if (mddev->level == 6) | |
5295 | min = 4; | |
5296 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5297 | return -EINVAL; | |
5298 | } | |
29269553 N |
5299 | |
5300 | /* Can only proceed if there are plenty of stripe_heads. | |
5301 | * We need a minimum of one full stripe,, and for sensible progress | |
5302 | * it is best to have about 4 times that. | |
5303 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5304 | * allow for chunk sizes up to 256K, which is probably OK. | |
5305 | * If the chunk size is greater, user-space should request more | |
5306 | * stripe_heads first. | |
5307 | */ | |
63c70c4f N |
5308 | if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes || |
5309 | (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) { | |
29269553 | 5310 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", |
784052ec N |
5311 | (max(mddev->chunk_size, mddev->new_chunk) |
5312 | / STRIPE_SIZE)*4); | |
29269553 N |
5313 | return -ENOSPC; |
5314 | } | |
5315 | ||
ec32a2bd | 5316 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5317 | } |
5318 | ||
5319 | static int raid5_start_reshape(mddev_t *mddev) | |
5320 | { | |
5321 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
5322 | mdk_rdev_t *rdev; | |
63c70c4f N |
5323 | int spares = 0; |
5324 | int added_devices = 0; | |
c04be0aa | 5325 | unsigned long flags; |
63c70c4f | 5326 | |
f416885e | 5327 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5328 | return -EBUSY; |
5329 | ||
159ec1fc | 5330 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5331 | if (rdev->raid_disk < 0 && |
5332 | !test_bit(Faulty, &rdev->flags)) | |
5333 | spares++; | |
63c70c4f | 5334 | |
f416885e | 5335 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5336 | /* Not enough devices even to make a degraded array |
5337 | * of that size | |
5338 | */ | |
5339 | return -EINVAL; | |
5340 | ||
ec32a2bd N |
5341 | /* Refuse to reduce size of the array. Any reductions in |
5342 | * array size must be through explicit setting of array_size | |
5343 | * attribute. | |
5344 | */ | |
5345 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5346 | < mddev->array_sectors) { | |
5347 | printk(KERN_ERR "md: %s: array size must be reduced " | |
5348 | "before number of disks\n", mdname(mddev)); | |
5349 | return -EINVAL; | |
5350 | } | |
5351 | ||
f6705578 | 5352 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5353 | spin_lock_irq(&conf->device_lock); |
5354 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5355 | conf->raid_disks += mddev->delta_disks; |
88ce4930 N |
5356 | conf->prev_chunk = conf->chunk_size; |
5357 | conf->chunk_size = mddev->new_chunk; | |
5358 | conf->prev_algo = conf->algorithm; | |
5359 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5360 | if (mddev->delta_disks < 0) |
5361 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5362 | else | |
5363 | conf->reshape_progress = 0; | |
5364 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5365 | conf->generation++; |
29269553 N |
5366 | spin_unlock_irq(&conf->device_lock); |
5367 | ||
5368 | /* Add some new drives, as many as will fit. | |
5369 | * We know there are enough to make the newly sized array work. | |
5370 | */ | |
159ec1fc | 5371 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5372 | if (rdev->raid_disk < 0 && |
5373 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 5374 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 N |
5375 | char nm[20]; |
5376 | set_bit(In_sync, &rdev->flags); | |
29269553 | 5377 | added_devices++; |
5fd6c1dc | 5378 | rdev->recovery_offset = 0; |
29269553 | 5379 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
5380 | if (sysfs_create_link(&mddev->kobj, |
5381 | &rdev->kobj, nm)) | |
5382 | printk(KERN_WARNING | |
5383 | "raid5: failed to create " | |
5384 | " link %s for %s\n", | |
5385 | nm, mdname(mddev)); | |
29269553 N |
5386 | } else |
5387 | break; | |
5388 | } | |
5389 | ||
ec32a2bd N |
5390 | if (mddev->delta_disks > 0) { |
5391 | spin_lock_irqsave(&conf->device_lock, flags); | |
5392 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) | |
5393 | - added_devices; | |
5394 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5395 | } | |
63c70c4f | 5396 | mddev->raid_disks = conf->raid_disks; |
f6705578 | 5397 | mddev->reshape_position = 0; |
850b2b42 | 5398 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5399 | |
29269553 N |
5400 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5401 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5402 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5403 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5404 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
5405 | "%s_reshape"); | |
5406 | if (!mddev->sync_thread) { | |
5407 | mddev->recovery = 0; | |
5408 | spin_lock_irq(&conf->device_lock); | |
5409 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5410 | conf->reshape_progress = MaxSector; |
29269553 N |
5411 | spin_unlock_irq(&conf->device_lock); |
5412 | return -EAGAIN; | |
5413 | } | |
c8f517c4 | 5414 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5415 | md_wakeup_thread(mddev->sync_thread); |
5416 | md_new_event(mddev); | |
5417 | return 0; | |
5418 | } | |
29269553 | 5419 | |
ec32a2bd N |
5420 | /* This is called from the reshape thread and should make any |
5421 | * changes needed in 'conf' | |
5422 | */ | |
29269553 N |
5423 | static void end_reshape(raid5_conf_t *conf) |
5424 | { | |
29269553 | 5425 | |
f6705578 | 5426 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5427 | |
f6705578 | 5428 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5429 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5430 | conf->reshape_progress = MaxSector; |
f6705578 | 5431 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5432 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5433 | |
5434 | /* read-ahead size must cover two whole stripes, which is | |
5435 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5436 | */ | |
5437 | { | |
cea9c228 N |
5438 | int data_disks = conf->raid_disks - conf->max_degraded; |
5439 | int stripe = data_disks * (conf->chunk_size | |
5440 | / PAGE_SIZE); | |
16a53ecc N |
5441 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5442 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5443 | } | |
29269553 | 5444 | } |
29269553 N |
5445 | } |
5446 | ||
ec32a2bd N |
5447 | /* This is called from the raid5d thread with mddev_lock held. |
5448 | * It makes config changes to the device. | |
5449 | */ | |
cea9c228 N |
5450 | static void raid5_finish_reshape(mddev_t *mddev) |
5451 | { | |
5452 | struct block_device *bdev; | |
88ce4930 | 5453 | raid5_conf_t *conf = mddev_to_conf(mddev); |
cea9c228 N |
5454 | |
5455 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5456 | ||
ec32a2bd N |
5457 | if (mddev->delta_disks > 0) { |
5458 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5459 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
5460 | mddev->changed = 1; | |
5461 | ||
5462 | bdev = bdget_disk(mddev->gendisk, 0); | |
5463 | if (bdev) { | |
5464 | mutex_lock(&bdev->bd_inode->i_mutex); | |
5465 | i_size_write(bdev->bd_inode, | |
5466 | (loff_t)mddev->array_sectors << 9); | |
5467 | mutex_unlock(&bdev->bd_inode->i_mutex); | |
5468 | bdput(bdev); | |
5469 | } | |
5470 | } else { | |
5471 | int d; | |
ec32a2bd N |
5472 | mddev->degraded = conf->raid_disks; |
5473 | for (d = 0; d < conf->raid_disks ; d++) | |
5474 | if (conf->disks[d].rdev && | |
5475 | test_bit(In_sync, | |
5476 | &conf->disks[d].rdev->flags)) | |
5477 | mddev->degraded--; | |
5478 | for (d = conf->raid_disks ; | |
5479 | d < conf->raid_disks - mddev->delta_disks; | |
5480 | d++) | |
5481 | raid5_remove_disk(mddev, d); | |
cea9c228 | 5482 | } |
88ce4930 N |
5483 | mddev->layout = conf->algorithm; |
5484 | mddev->chunk_size = conf->chunk_size; | |
ec32a2bd N |
5485 | mddev->reshape_position = MaxSector; |
5486 | mddev->delta_disks = 0; | |
cea9c228 N |
5487 | } |
5488 | } | |
5489 | ||
72626685 N |
5490 | static void raid5_quiesce(mddev_t *mddev, int state) |
5491 | { | |
5492 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
5493 | ||
5494 | switch(state) { | |
e464eafd N |
5495 | case 2: /* resume for a suspend */ |
5496 | wake_up(&conf->wait_for_overlap); | |
5497 | break; | |
5498 | ||
72626685 N |
5499 | case 1: /* stop all writes */ |
5500 | spin_lock_irq(&conf->device_lock); | |
5501 | conf->quiesce = 1; | |
5502 | wait_event_lock_irq(conf->wait_for_stripe, | |
46031f9a RBJ |
5503 | atomic_read(&conf->active_stripes) == 0 && |
5504 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 N |
5505 | conf->device_lock, /* nothing */); |
5506 | spin_unlock_irq(&conf->device_lock); | |
5507 | break; | |
5508 | ||
5509 | case 0: /* re-enable writes */ | |
5510 | spin_lock_irq(&conf->device_lock); | |
5511 | conf->quiesce = 0; | |
5512 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5513 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5514 | spin_unlock_irq(&conf->device_lock); |
5515 | break; | |
5516 | } | |
72626685 | 5517 | } |
b15c2e57 | 5518 | |
d562b0c4 N |
5519 | |
5520 | static void *raid5_takeover_raid1(mddev_t *mddev) | |
5521 | { | |
5522 | int chunksect; | |
5523 | ||
5524 | if (mddev->raid_disks != 2 || | |
5525 | mddev->degraded > 1) | |
5526 | return ERR_PTR(-EINVAL); | |
5527 | ||
5528 | /* Should check if there are write-behind devices? */ | |
5529 | ||
5530 | chunksect = 64*2; /* 64K by default */ | |
5531 | ||
5532 | /* The array must be an exact multiple of chunksize */ | |
5533 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5534 | chunksect >>= 1; | |
5535 | ||
5536 | if ((chunksect<<9) < STRIPE_SIZE) | |
5537 | /* array size does not allow a suitable chunk size */ | |
5538 | return ERR_PTR(-EINVAL); | |
5539 | ||
5540 | mddev->new_level = 5; | |
5541 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5542 | mddev->new_chunk = chunksect << 9; | |
5543 | ||
5544 | return setup_conf(mddev); | |
5545 | } | |
5546 | ||
fc9739c6 N |
5547 | static void *raid5_takeover_raid6(mddev_t *mddev) |
5548 | { | |
5549 | int new_layout; | |
5550 | ||
5551 | switch (mddev->layout) { | |
5552 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5553 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5554 | break; | |
5555 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5556 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5557 | break; | |
5558 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5559 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5560 | break; | |
5561 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5562 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5563 | break; | |
5564 | case ALGORITHM_PARITY_0_6: | |
5565 | new_layout = ALGORITHM_PARITY_0; | |
5566 | break; | |
5567 | case ALGORITHM_PARITY_N: | |
5568 | new_layout = ALGORITHM_PARITY_N; | |
5569 | break; | |
5570 | default: | |
5571 | return ERR_PTR(-EINVAL); | |
5572 | } | |
5573 | mddev->new_level = 5; | |
5574 | mddev->new_layout = new_layout; | |
5575 | mddev->delta_disks = -1; | |
5576 | mddev->raid_disks -= 1; | |
5577 | return setup_conf(mddev); | |
5578 | } | |
5579 | ||
d562b0c4 | 5580 | |
b3546035 N |
5581 | static int raid5_reconfig(mddev_t *mddev, int new_layout, int new_chunk) |
5582 | { | |
88ce4930 N |
5583 | /* For a 2-drive array, the layout and chunk size can be changed |
5584 | * immediately as not restriping is needed. | |
5585 | * For larger arrays we record the new value - after validation | |
5586 | * to be used by a reshape pass. | |
b3546035 N |
5587 | */ |
5588 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
5589 | ||
5590 | if (new_layout >= 0 && !algorithm_valid_raid5(new_layout)) | |
5591 | return -EINVAL; | |
5592 | if (new_chunk > 0) { | |
5593 | if (new_chunk & (new_chunk-1)) | |
5594 | /* not a power of 2 */ | |
5595 | return -EINVAL; | |
5596 | if (new_chunk < PAGE_SIZE) | |
5597 | return -EINVAL; | |
5598 | if (mddev->array_sectors & ((new_chunk>>9)-1)) | |
5599 | /* not factor of array size */ | |
5600 | return -EINVAL; | |
5601 | } | |
5602 | ||
5603 | /* They look valid */ | |
5604 | ||
88ce4930 | 5605 | if (mddev->raid_disks == 2) { |
b3546035 | 5606 | |
88ce4930 N |
5607 | if (new_layout >= 0) { |
5608 | conf->algorithm = new_layout; | |
5609 | mddev->layout = mddev->new_layout = new_layout; | |
5610 | } | |
5611 | if (new_chunk > 0) { | |
5612 | conf->chunk_size = new_chunk; | |
5613 | mddev->chunk_size = mddev->new_chunk = new_chunk; | |
5614 | } | |
5615 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5616 | md_wakeup_thread(mddev->thread); | |
5617 | } else { | |
5618 | if (new_layout >= 0) | |
5619 | mddev->new_layout = new_layout; | |
5620 | if (new_chunk > 0) | |
5621 | mddev->new_chunk = new_chunk; | |
b3546035 | 5622 | } |
88ce4930 N |
5623 | return 0; |
5624 | } | |
5625 | ||
5626 | static int raid6_reconfig(mddev_t *mddev, int new_layout, int new_chunk) | |
5627 | { | |
5628 | if (new_layout >= 0 && !algorithm_valid_raid6(new_layout)) | |
5629 | return -EINVAL; | |
b3546035 | 5630 | if (new_chunk > 0) { |
88ce4930 N |
5631 | if (new_chunk & (new_chunk-1)) |
5632 | /* not a power of 2 */ | |
5633 | return -EINVAL; | |
5634 | if (new_chunk < PAGE_SIZE) | |
5635 | return -EINVAL; | |
5636 | if (mddev->array_sectors & ((new_chunk>>9)-1)) | |
5637 | /* not factor of array size */ | |
5638 | return -EINVAL; | |
b3546035 | 5639 | } |
88ce4930 N |
5640 | |
5641 | /* They look valid */ | |
5642 | ||
5643 | if (new_layout >= 0) | |
5644 | mddev->new_layout = new_layout; | |
5645 | if (new_chunk > 0) | |
5646 | mddev->new_chunk = new_chunk; | |
5647 | ||
b3546035 N |
5648 | return 0; |
5649 | } | |
5650 | ||
d562b0c4 N |
5651 | static void *raid5_takeover(mddev_t *mddev) |
5652 | { | |
5653 | /* raid5 can take over: | |
5654 | * raid0 - if all devices are the same - make it a raid4 layout | |
5655 | * raid1 - if there are two drives. We need to know the chunk size | |
5656 | * raid4 - trivial - just use a raid4 layout. | |
5657 | * raid6 - Providing it is a *_6 layout | |
5658 | * | |
5659 | * For now, just do raid1 | |
5660 | */ | |
5661 | ||
5662 | if (mddev->level == 1) | |
5663 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5664 | if (mddev->level == 4) { |
5665 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5666 | mddev->new_level = 5; | |
5667 | return setup_conf(mddev); | |
5668 | } | |
fc9739c6 N |
5669 | if (mddev->level == 6) |
5670 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5671 | |
5672 | return ERR_PTR(-EINVAL); | |
5673 | } | |
5674 | ||
5675 | ||
245f46c2 N |
5676 | static struct mdk_personality raid5_personality; |
5677 | ||
5678 | static void *raid6_takeover(mddev_t *mddev) | |
5679 | { | |
5680 | /* Currently can only take over a raid5. We map the | |
5681 | * personality to an equivalent raid6 personality | |
5682 | * with the Q block at the end. | |
5683 | */ | |
5684 | int new_layout; | |
5685 | ||
5686 | if (mddev->pers != &raid5_personality) | |
5687 | return ERR_PTR(-EINVAL); | |
5688 | if (mddev->degraded > 1) | |
5689 | return ERR_PTR(-EINVAL); | |
5690 | if (mddev->raid_disks > 253) | |
5691 | return ERR_PTR(-EINVAL); | |
5692 | if (mddev->raid_disks < 3) | |
5693 | return ERR_PTR(-EINVAL); | |
5694 | ||
5695 | switch (mddev->layout) { | |
5696 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5697 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5698 | break; | |
5699 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5700 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5701 | break; | |
5702 | case ALGORITHM_LEFT_SYMMETRIC: | |
5703 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5704 | break; | |
5705 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5706 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5707 | break; | |
5708 | case ALGORITHM_PARITY_0: | |
5709 | new_layout = ALGORITHM_PARITY_0_6; | |
5710 | break; | |
5711 | case ALGORITHM_PARITY_N: | |
5712 | new_layout = ALGORITHM_PARITY_N; | |
5713 | break; | |
5714 | default: | |
5715 | return ERR_PTR(-EINVAL); | |
5716 | } | |
5717 | mddev->new_level = 6; | |
5718 | mddev->new_layout = new_layout; | |
5719 | mddev->delta_disks = 1; | |
5720 | mddev->raid_disks += 1; | |
5721 | return setup_conf(mddev); | |
5722 | } | |
5723 | ||
5724 | ||
16a53ecc N |
5725 | static struct mdk_personality raid6_personality = |
5726 | { | |
5727 | .name = "raid6", | |
5728 | .level = 6, | |
5729 | .owner = THIS_MODULE, | |
5730 | .make_request = make_request, | |
5731 | .run = run, | |
5732 | .stop = stop, | |
5733 | .status = status, | |
5734 | .error_handler = error, | |
5735 | .hot_add_disk = raid5_add_disk, | |
5736 | .hot_remove_disk= raid5_remove_disk, | |
5737 | .spare_active = raid5_spare_active, | |
5738 | .sync_request = sync_request, | |
5739 | .resize = raid5_resize, | |
80c3a6ce | 5740 | .size = raid5_size, |
f416885e N |
5741 | .check_reshape = raid5_check_reshape, |
5742 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5743 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5744 | .quiesce = raid5_quiesce, |
245f46c2 | 5745 | .takeover = raid6_takeover, |
88ce4930 | 5746 | .reconfig = raid6_reconfig, |
16a53ecc | 5747 | }; |
2604b703 | 5748 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
5749 | { |
5750 | .name = "raid5", | |
2604b703 | 5751 | .level = 5, |
1da177e4 LT |
5752 | .owner = THIS_MODULE, |
5753 | .make_request = make_request, | |
5754 | .run = run, | |
5755 | .stop = stop, | |
5756 | .status = status, | |
5757 | .error_handler = error, | |
5758 | .hot_add_disk = raid5_add_disk, | |
5759 | .hot_remove_disk= raid5_remove_disk, | |
5760 | .spare_active = raid5_spare_active, | |
5761 | .sync_request = sync_request, | |
5762 | .resize = raid5_resize, | |
80c3a6ce | 5763 | .size = raid5_size, |
63c70c4f N |
5764 | .check_reshape = raid5_check_reshape, |
5765 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5766 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5767 | .quiesce = raid5_quiesce, |
d562b0c4 | 5768 | .takeover = raid5_takeover, |
b3546035 | 5769 | .reconfig = raid5_reconfig, |
1da177e4 LT |
5770 | }; |
5771 | ||
2604b703 | 5772 | static struct mdk_personality raid4_personality = |
1da177e4 | 5773 | { |
2604b703 N |
5774 | .name = "raid4", |
5775 | .level = 4, | |
5776 | .owner = THIS_MODULE, | |
5777 | .make_request = make_request, | |
5778 | .run = run, | |
5779 | .stop = stop, | |
5780 | .status = status, | |
5781 | .error_handler = error, | |
5782 | .hot_add_disk = raid5_add_disk, | |
5783 | .hot_remove_disk= raid5_remove_disk, | |
5784 | .spare_active = raid5_spare_active, | |
5785 | .sync_request = sync_request, | |
5786 | .resize = raid5_resize, | |
80c3a6ce | 5787 | .size = raid5_size, |
3d37890b N |
5788 | .check_reshape = raid5_check_reshape, |
5789 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5790 | .finish_reshape = raid5_finish_reshape, |
2604b703 N |
5791 | .quiesce = raid5_quiesce, |
5792 | }; | |
5793 | ||
5794 | static int __init raid5_init(void) | |
5795 | { | |
16a53ecc | 5796 | register_md_personality(&raid6_personality); |
2604b703 N |
5797 | register_md_personality(&raid5_personality); |
5798 | register_md_personality(&raid4_personality); | |
5799 | return 0; | |
1da177e4 LT |
5800 | } |
5801 | ||
2604b703 | 5802 | static void raid5_exit(void) |
1da177e4 | 5803 | { |
16a53ecc | 5804 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5805 | unregister_md_personality(&raid5_personality); |
5806 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5807 | } |
5808 | ||
5809 | module_init(raid5_init); | |
5810 | module_exit(raid5_exit); | |
5811 | MODULE_LICENSE("GPL"); | |
5812 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
5813 | MODULE_ALIAS("md-raid5"); |
5814 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5815 | MODULE_ALIAS("md-level-5"); |
5816 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5817 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5818 | MODULE_ALIAS("md-raid6"); | |
5819 | MODULE_ALIAS("md-level-6"); | |
5820 | ||
5821 | /* This used to be two separate modules, they were: */ | |
5822 | MODULE_ALIAS("raid5"); | |
5823 | MODULE_ALIAS("raid6"); |