]>
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> |
91c00924 | 48 | #include <linux/async_tx.h> |
bff61975 | 49 | #include <linux/seq_file.h> |
43b2e5d8 | 50 | #include "md.h" |
bff61975 | 51 | #include "raid5.h" |
ef740c37 CH |
52 | #include "raid6.h" |
53 | #include "bitmap.h" | |
72626685 | 54 | |
1da177e4 LT |
55 | /* |
56 | * Stripe cache | |
57 | */ | |
58 | ||
59 | #define NR_STRIPES 256 | |
60 | #define STRIPE_SIZE PAGE_SIZE | |
61 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
62 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
63 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 64 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 65 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
66 | #define HASH_MASK (NR_HASH - 1) |
67 | ||
fccddba0 | 68 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
69 | |
70 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
71 | * order without overlap. There may be several bio's per stripe+device, and | |
72 | * a bio could span several devices. | |
73 | * When walking this list for a particular stripe+device, we must never proceed | |
74 | * beyond a bio that extends past this device, as the next bio might no longer | |
75 | * be valid. | |
76 | * This macro is used to determine the 'next' bio in the list, given the sector | |
77 | * of the current stripe+device | |
78 | */ | |
79 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
80 | /* | |
81 | * The following can be used to debug the driver | |
82 | */ | |
1da177e4 LT |
83 | #define RAID5_PARANOIA 1 |
84 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
85 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
86 | #else | |
87 | # define CHECK_DEVLOCK() | |
88 | #endif | |
89 | ||
45b4233c | 90 | #ifdef DEBUG |
1da177e4 LT |
91 | #define inline |
92 | #define __inline__ | |
93 | #endif | |
94 | ||
6be9d494 BS |
95 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
96 | ||
16a53ecc N |
97 | #if !RAID6_USE_EMPTY_ZERO_PAGE |
98 | /* In .bss so it's zeroed */ | |
99 | const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256))); | |
100 | #endif | |
101 | ||
960e739d | 102 | /* |
5b99c2ff JA |
103 | * We maintain a biased count of active stripes in the bottom 16 bits of |
104 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
105 | */ |
106 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
107 | { | |
5b99c2ff | 108 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
109 | } |
110 | ||
111 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
112 | { | |
5b99c2ff | 113 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
114 | } |
115 | ||
116 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
117 | { | |
118 | --bio->bi_phys_segments; | |
119 | return raid5_bi_phys_segments(bio); | |
120 | } | |
121 | ||
122 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
123 | { | |
124 | unsigned short val = raid5_bi_hw_segments(bio); | |
125 | ||
126 | --val; | |
5b99c2ff | 127 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
128 | return val; |
129 | } | |
130 | ||
131 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
132 | { | |
5b99c2ff | 133 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16); |
960e739d JA |
134 | } |
135 | ||
d0dabf7e N |
136 | /* Find first data disk in a raid6 stripe */ |
137 | static inline int raid6_d0(struct stripe_head *sh) | |
138 | { | |
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 | */ | |
155 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, int *count) | |
156 | { | |
157 | int slot; | |
158 | if (idx == sh->pd_idx) | |
159 | return sh->disks - 2; | |
160 | if (idx == sh->qd_idx) | |
161 | return sh->disks - 1; | |
162 | slot = (*count)++; | |
163 | return slot; | |
164 | } | |
165 | ||
a4456856 DW |
166 | static void return_io(struct bio *return_bi) |
167 | { | |
168 | struct bio *bi = return_bi; | |
169 | while (bi) { | |
a4456856 DW |
170 | |
171 | return_bi = bi->bi_next; | |
172 | bi->bi_next = NULL; | |
173 | bi->bi_size = 0; | |
0e13fe23 | 174 | bio_endio(bi, 0); |
a4456856 DW |
175 | bi = return_bi; |
176 | } | |
177 | } | |
178 | ||
1da177e4 LT |
179 | static void print_raid5_conf (raid5_conf_t *conf); |
180 | ||
600aa109 DW |
181 | static int stripe_operations_active(struct stripe_head *sh) |
182 | { | |
183 | return sh->check_state || sh->reconstruct_state || | |
184 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
185 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
186 | } | |
187 | ||
858119e1 | 188 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
189 | { |
190 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
191 | BUG_ON(!list_empty(&sh->lru)); |
192 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 193 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 194 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 195 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
196 | blk_plug_device(conf->mddev->queue); |
197 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 198 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 199 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
200 | blk_plug_device(conf->mddev->queue); |
201 | } else { | |
72626685 | 202 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 203 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 204 | } |
1da177e4 LT |
205 | md_wakeup_thread(conf->mddev->thread); |
206 | } else { | |
600aa109 | 207 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
208 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
209 | atomic_dec(&conf->preread_active_stripes); | |
210 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
211 | md_wakeup_thread(conf->mddev->thread); | |
212 | } | |
1da177e4 | 213 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
214 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
215 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 216 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
217 | if (conf->retry_read_aligned) |
218 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 219 | } |
1da177e4 LT |
220 | } |
221 | } | |
222 | } | |
d0dabf7e | 223 | |
1da177e4 LT |
224 | static void release_stripe(struct stripe_head *sh) |
225 | { | |
226 | raid5_conf_t *conf = sh->raid_conf; | |
227 | unsigned long flags; | |
16a53ecc | 228 | |
1da177e4 LT |
229 | spin_lock_irqsave(&conf->device_lock, flags); |
230 | __release_stripe(conf, sh); | |
231 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
232 | } | |
233 | ||
fccddba0 | 234 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 235 | { |
45b4233c DW |
236 | pr_debug("remove_hash(), stripe %llu\n", |
237 | (unsigned long long)sh->sector); | |
1da177e4 | 238 | |
fccddba0 | 239 | hlist_del_init(&sh->hash); |
1da177e4 LT |
240 | } |
241 | ||
16a53ecc | 242 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 243 | { |
fccddba0 | 244 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 245 | |
45b4233c DW |
246 | pr_debug("insert_hash(), stripe %llu\n", |
247 | (unsigned long long)sh->sector); | |
1da177e4 LT |
248 | |
249 | CHECK_DEVLOCK(); | |
fccddba0 | 250 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
251 | } |
252 | ||
253 | ||
254 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
255 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
256 | { | |
257 | struct stripe_head *sh = NULL; | |
258 | struct list_head *first; | |
259 | ||
260 | CHECK_DEVLOCK(); | |
261 | if (list_empty(&conf->inactive_list)) | |
262 | goto out; | |
263 | first = conf->inactive_list.next; | |
264 | sh = list_entry(first, struct stripe_head, lru); | |
265 | list_del_init(first); | |
266 | remove_hash(sh); | |
267 | atomic_inc(&conf->active_stripes); | |
268 | out: | |
269 | return sh; | |
270 | } | |
271 | ||
272 | static void shrink_buffers(struct stripe_head *sh, int num) | |
273 | { | |
274 | struct page *p; | |
275 | int i; | |
276 | ||
277 | for (i=0; i<num ; i++) { | |
278 | p = sh->dev[i].page; | |
279 | if (!p) | |
280 | continue; | |
281 | sh->dev[i].page = NULL; | |
2d1f3b5d | 282 | put_page(p); |
1da177e4 LT |
283 | } |
284 | } | |
285 | ||
286 | static int grow_buffers(struct stripe_head *sh, int num) | |
287 | { | |
288 | int i; | |
289 | ||
290 | for (i=0; i<num; i++) { | |
291 | struct page *page; | |
292 | ||
293 | if (!(page = alloc_page(GFP_KERNEL))) { | |
294 | return 1; | |
295 | } | |
296 | sh->dev[i].page = page; | |
297 | } | |
298 | return 0; | |
299 | } | |
300 | ||
d710e138 | 301 | static void raid5_build_block(struct stripe_head *sh, int i); |
911d4ee8 N |
302 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
303 | struct stripe_head *sh); | |
1da177e4 | 304 | |
b5663ba4 | 305 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
306 | { |
307 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 308 | int i; |
1da177e4 | 309 | |
78bafebd ES |
310 | BUG_ON(atomic_read(&sh->count) != 0); |
311 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 312 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 313 | |
1da177e4 | 314 | CHECK_DEVLOCK(); |
45b4233c | 315 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
316 | (unsigned long long)sh->sector); |
317 | ||
318 | remove_hash(sh); | |
16a53ecc | 319 | |
b5663ba4 | 320 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 321 | sh->sector = sector; |
911d4ee8 | 322 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
323 | sh->state = 0; |
324 | ||
7ecaa1e6 N |
325 | |
326 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
327 | struct r5dev *dev = &sh->dev[i]; |
328 | ||
d84e0f10 | 329 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 330 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 331 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 332 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 333 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
334 | test_bit(R5_LOCKED, &dev->flags)); |
335 | BUG(); | |
336 | } | |
337 | dev->flags = 0; | |
338 | raid5_build_block(sh, i); | |
339 | } | |
340 | insert_hash(conf, sh); | |
341 | } | |
342 | ||
7ecaa1e6 | 343 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, int disks) |
1da177e4 LT |
344 | { |
345 | struct stripe_head *sh; | |
fccddba0 | 346 | struct hlist_node *hn; |
1da177e4 LT |
347 | |
348 | CHECK_DEVLOCK(); | |
45b4233c | 349 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 350 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
7ecaa1e6 | 351 | if (sh->sector == sector && sh->disks == disks) |
1da177e4 | 352 | return sh; |
45b4233c | 353 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
354 | return NULL; |
355 | } | |
356 | ||
357 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 358 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 359 | |
b5663ba4 N |
360 | static struct stripe_head * |
361 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
362 | int previous, int noblock) | |
1da177e4 LT |
363 | { |
364 | struct stripe_head *sh; | |
b5663ba4 | 365 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 366 | |
45b4233c | 367 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
368 | |
369 | spin_lock_irq(&conf->device_lock); | |
370 | ||
371 | do { | |
72626685 N |
372 | wait_event_lock_irq(conf->wait_for_stripe, |
373 | conf->quiesce == 0, | |
374 | conf->device_lock, /* nothing */); | |
7ecaa1e6 | 375 | sh = __find_stripe(conf, sector, disks); |
1da177e4 LT |
376 | if (!sh) { |
377 | if (!conf->inactive_blocked) | |
378 | sh = get_free_stripe(conf); | |
379 | if (noblock && sh == NULL) | |
380 | break; | |
381 | if (!sh) { | |
382 | conf->inactive_blocked = 1; | |
383 | wait_event_lock_irq(conf->wait_for_stripe, | |
384 | !list_empty(&conf->inactive_list) && | |
5036805b N |
385 | (atomic_read(&conf->active_stripes) |
386 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
387 | || !conf->inactive_blocked), |
388 | conf->device_lock, | |
f4370781 | 389 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
390 | ); |
391 | conf->inactive_blocked = 0; | |
392 | } else | |
b5663ba4 | 393 | init_stripe(sh, sector, previous); |
1da177e4 LT |
394 | } else { |
395 | if (atomic_read(&sh->count)) { | |
78bafebd | 396 | BUG_ON(!list_empty(&sh->lru)); |
1da177e4 LT |
397 | } else { |
398 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
399 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
400 | if (list_empty(&sh->lru) && |
401 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
402 | BUG(); |
403 | list_del_init(&sh->lru); | |
1da177e4 LT |
404 | } |
405 | } | |
406 | } while (sh == NULL); | |
407 | ||
408 | if (sh) | |
409 | atomic_inc(&sh->count); | |
410 | ||
411 | spin_unlock_irq(&conf->device_lock); | |
412 | return sh; | |
413 | } | |
414 | ||
6712ecf8 N |
415 | static void |
416 | raid5_end_read_request(struct bio *bi, int error); | |
417 | static void | |
418 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 419 | |
c4e5ac0a | 420 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
421 | { |
422 | raid5_conf_t *conf = sh->raid_conf; | |
423 | int i, disks = sh->disks; | |
424 | ||
425 | might_sleep(); | |
426 | ||
427 | for (i = disks; i--; ) { | |
428 | int rw; | |
429 | struct bio *bi; | |
430 | mdk_rdev_t *rdev; | |
431 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
432 | rw = WRITE; | |
433 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
434 | rw = READ; | |
435 | else | |
436 | continue; | |
437 | ||
438 | bi = &sh->dev[i].req; | |
439 | ||
440 | bi->bi_rw = rw; | |
441 | if (rw == WRITE) | |
442 | bi->bi_end_io = raid5_end_write_request; | |
443 | else | |
444 | bi->bi_end_io = raid5_end_read_request; | |
445 | ||
446 | rcu_read_lock(); | |
447 | rdev = rcu_dereference(conf->disks[i].rdev); | |
448 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
449 | rdev = NULL; | |
450 | if (rdev) | |
451 | atomic_inc(&rdev->nr_pending); | |
452 | rcu_read_unlock(); | |
453 | ||
454 | if (rdev) { | |
c4e5ac0a | 455 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
456 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
457 | ||
2b7497f0 DW |
458 | set_bit(STRIPE_IO_STARTED, &sh->state); |
459 | ||
91c00924 DW |
460 | bi->bi_bdev = rdev->bdev; |
461 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 462 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
463 | bi->bi_rw, i); |
464 | atomic_inc(&sh->count); | |
465 | bi->bi_sector = sh->sector + rdev->data_offset; | |
466 | bi->bi_flags = 1 << BIO_UPTODATE; | |
467 | bi->bi_vcnt = 1; | |
468 | bi->bi_max_vecs = 1; | |
469 | bi->bi_idx = 0; | |
470 | bi->bi_io_vec = &sh->dev[i].vec; | |
471 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
472 | bi->bi_io_vec[0].bv_offset = 0; | |
473 | bi->bi_size = STRIPE_SIZE; | |
474 | bi->bi_next = NULL; | |
475 | if (rw == WRITE && | |
476 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
477 | atomic_add(STRIPE_SECTORS, | |
478 | &rdev->corrected_errors); | |
479 | generic_make_request(bi); | |
480 | } else { | |
481 | if (rw == WRITE) | |
482 | set_bit(STRIPE_DEGRADED, &sh->state); | |
483 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
484 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
485 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
486 | set_bit(STRIPE_HANDLE, &sh->state); | |
487 | } | |
488 | } | |
489 | } | |
490 | ||
491 | static struct dma_async_tx_descriptor * | |
492 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
493 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
494 | { | |
495 | struct bio_vec *bvl; | |
496 | struct page *bio_page; | |
497 | int i; | |
498 | int page_offset; | |
499 | ||
500 | if (bio->bi_sector >= sector) | |
501 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
502 | else | |
503 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
504 | bio_for_each_segment(bvl, bio, i) { | |
505 | int len = bio_iovec_idx(bio, i)->bv_len; | |
506 | int clen; | |
507 | int b_offset = 0; | |
508 | ||
509 | if (page_offset < 0) { | |
510 | b_offset = -page_offset; | |
511 | page_offset += b_offset; | |
512 | len -= b_offset; | |
513 | } | |
514 | ||
515 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
516 | clen = STRIPE_SIZE - page_offset; | |
517 | else | |
518 | clen = len; | |
519 | ||
520 | if (clen > 0) { | |
521 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
522 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
523 | if (frombio) | |
524 | tx = async_memcpy(page, bio_page, page_offset, | |
525 | b_offset, clen, | |
eb0645a8 | 526 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
527 | tx, NULL, NULL); |
528 | else | |
529 | tx = async_memcpy(bio_page, page, b_offset, | |
530 | page_offset, clen, | |
eb0645a8 | 531 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
532 | tx, NULL, NULL); |
533 | } | |
534 | if (clen < len) /* hit end of page */ | |
535 | break; | |
536 | page_offset += len; | |
537 | } | |
538 | ||
539 | return tx; | |
540 | } | |
541 | ||
542 | static void ops_complete_biofill(void *stripe_head_ref) | |
543 | { | |
544 | struct stripe_head *sh = stripe_head_ref; | |
545 | struct bio *return_bi = NULL; | |
546 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 547 | int i; |
91c00924 | 548 | |
e46b272b | 549 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
550 | (unsigned long long)sh->sector); |
551 | ||
552 | /* clear completed biofills */ | |
83de75cc | 553 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
554 | for (i = sh->disks; i--; ) { |
555 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
556 | |
557 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
558 | /* and check if we need to reply to a read request, |
559 | * new R5_Wantfill requests are held off until | |
83de75cc | 560 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
561 | */ |
562 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 563 | struct bio *rbi, *rbi2; |
91c00924 | 564 | |
91c00924 DW |
565 | BUG_ON(!dev->read); |
566 | rbi = dev->read; | |
567 | dev->read = NULL; | |
568 | while (rbi && rbi->bi_sector < | |
569 | dev->sector + STRIPE_SECTORS) { | |
570 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 571 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
572 | rbi->bi_next = return_bi; |
573 | return_bi = rbi; | |
574 | } | |
91c00924 DW |
575 | rbi = rbi2; |
576 | } | |
577 | } | |
578 | } | |
83de75cc DW |
579 | spin_unlock_irq(&conf->device_lock); |
580 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
581 | |
582 | return_io(return_bi); | |
583 | ||
e4d84909 | 584 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
585 | release_stripe(sh); |
586 | } | |
587 | ||
588 | static void ops_run_biofill(struct stripe_head *sh) | |
589 | { | |
590 | struct dma_async_tx_descriptor *tx = NULL; | |
591 | raid5_conf_t *conf = sh->raid_conf; | |
592 | int i; | |
593 | ||
e46b272b | 594 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
595 | (unsigned long long)sh->sector); |
596 | ||
597 | for (i = sh->disks; i--; ) { | |
598 | struct r5dev *dev = &sh->dev[i]; | |
599 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
600 | struct bio *rbi; | |
601 | spin_lock_irq(&conf->device_lock); | |
602 | dev->read = rbi = dev->toread; | |
603 | dev->toread = NULL; | |
604 | spin_unlock_irq(&conf->device_lock); | |
605 | while (rbi && rbi->bi_sector < | |
606 | dev->sector + STRIPE_SECTORS) { | |
607 | tx = async_copy_data(0, rbi, dev->page, | |
608 | dev->sector, tx); | |
609 | rbi = r5_next_bio(rbi, dev->sector); | |
610 | } | |
611 | } | |
612 | } | |
613 | ||
614 | atomic_inc(&sh->count); | |
615 | async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
616 | ops_complete_biofill, sh); | |
617 | } | |
618 | ||
619 | static void ops_complete_compute5(void *stripe_head_ref) | |
620 | { | |
621 | struct stripe_head *sh = stripe_head_ref; | |
622 | int target = sh->ops.target; | |
623 | struct r5dev *tgt = &sh->dev[target]; | |
624 | ||
e46b272b | 625 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
626 | (unsigned long long)sh->sector); |
627 | ||
628 | set_bit(R5_UPTODATE, &tgt->flags); | |
629 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
630 | clear_bit(R5_Wantcompute, &tgt->flags); | |
ecc65c9b DW |
631 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
632 | if (sh->check_state == check_state_compute_run) | |
633 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
634 | set_bit(STRIPE_HANDLE, &sh->state); |
635 | release_stripe(sh); | |
636 | } | |
637 | ||
7b3a871e | 638 | static struct dma_async_tx_descriptor *ops_run_compute5(struct stripe_head *sh) |
91c00924 DW |
639 | { |
640 | /* kernel stack size limits the total number of disks */ | |
641 | int disks = sh->disks; | |
642 | struct page *xor_srcs[disks]; | |
643 | int target = sh->ops.target; | |
644 | struct r5dev *tgt = &sh->dev[target]; | |
645 | struct page *xor_dest = tgt->page; | |
646 | int count = 0; | |
647 | struct dma_async_tx_descriptor *tx; | |
648 | int i; | |
649 | ||
650 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 651 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
652 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
653 | ||
654 | for (i = disks; i--; ) | |
655 | if (i != target) | |
656 | xor_srcs[count++] = sh->dev[i].page; | |
657 | ||
658 | atomic_inc(&sh->count); | |
659 | ||
660 | if (unlikely(count == 1)) | |
661 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
662 | 0, NULL, ops_complete_compute5, sh); | |
663 | else | |
664 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
665 | ASYNC_TX_XOR_ZERO_DST, NULL, | |
666 | ops_complete_compute5, sh); | |
667 | ||
91c00924 DW |
668 | return tx; |
669 | } | |
670 | ||
671 | static void ops_complete_prexor(void *stripe_head_ref) | |
672 | { | |
673 | struct stripe_head *sh = stripe_head_ref; | |
674 | ||
e46b272b | 675 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 676 | (unsigned long long)sh->sector); |
91c00924 DW |
677 | } |
678 | ||
679 | static struct dma_async_tx_descriptor * | |
680 | ops_run_prexor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) | |
681 | { | |
682 | /* kernel stack size limits the total number of disks */ | |
683 | int disks = sh->disks; | |
684 | struct page *xor_srcs[disks]; | |
685 | int count = 0, pd_idx = sh->pd_idx, i; | |
686 | ||
687 | /* existing parity data subtracted */ | |
688 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
689 | ||
e46b272b | 690 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
691 | (unsigned long long)sh->sector); |
692 | ||
693 | for (i = disks; i--; ) { | |
694 | struct r5dev *dev = &sh->dev[i]; | |
695 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 696 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
697 | xor_srcs[count++] = dev->page; |
698 | } | |
699 | ||
700 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
701 | ASYNC_TX_DEP_ACK | ASYNC_TX_XOR_DROP_DST, tx, | |
702 | ops_complete_prexor, sh); | |
703 | ||
704 | return tx; | |
705 | } | |
706 | ||
707 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 708 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
709 | { |
710 | int disks = sh->disks; | |
d8ee0728 | 711 | int i; |
91c00924 | 712 | |
e46b272b | 713 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
714 | (unsigned long long)sh->sector); |
715 | ||
716 | for (i = disks; i--; ) { | |
717 | struct r5dev *dev = &sh->dev[i]; | |
718 | struct bio *chosen; | |
91c00924 | 719 | |
d8ee0728 | 720 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
721 | struct bio *wbi; |
722 | ||
723 | spin_lock(&sh->lock); | |
724 | chosen = dev->towrite; | |
725 | dev->towrite = NULL; | |
726 | BUG_ON(dev->written); | |
727 | wbi = dev->written = chosen; | |
728 | spin_unlock(&sh->lock); | |
729 | ||
730 | while (wbi && wbi->bi_sector < | |
731 | dev->sector + STRIPE_SECTORS) { | |
732 | tx = async_copy_data(1, wbi, dev->page, | |
733 | dev->sector, tx); | |
734 | wbi = r5_next_bio(wbi, dev->sector); | |
735 | } | |
736 | } | |
737 | } | |
738 | ||
739 | return tx; | |
740 | } | |
741 | ||
742 | static void ops_complete_postxor(void *stripe_head_ref) | |
91c00924 DW |
743 | { |
744 | struct stripe_head *sh = stripe_head_ref; | |
745 | int disks = sh->disks, i, pd_idx = sh->pd_idx; | |
746 | ||
e46b272b | 747 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
748 | (unsigned long long)sh->sector); |
749 | ||
750 | for (i = disks; i--; ) { | |
751 | struct r5dev *dev = &sh->dev[i]; | |
752 | if (dev->written || i == pd_idx) | |
753 | set_bit(R5_UPTODATE, &dev->flags); | |
754 | } | |
755 | ||
d8ee0728 DW |
756 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
757 | sh->reconstruct_state = reconstruct_state_drain_result; | |
758 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
759 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
760 | else { | |
761 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
762 | sh->reconstruct_state = reconstruct_state_result; | |
763 | } | |
91c00924 DW |
764 | |
765 | set_bit(STRIPE_HANDLE, &sh->state); | |
766 | release_stripe(sh); | |
767 | } | |
768 | ||
769 | static void | |
d8ee0728 | 770 | ops_run_postxor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
771 | { |
772 | /* kernel stack size limits the total number of disks */ | |
773 | int disks = sh->disks; | |
774 | struct page *xor_srcs[disks]; | |
775 | ||
776 | int count = 0, pd_idx = sh->pd_idx, i; | |
777 | struct page *xor_dest; | |
d8ee0728 | 778 | int prexor = 0; |
91c00924 | 779 | unsigned long flags; |
91c00924 | 780 | |
e46b272b | 781 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
782 | (unsigned long long)sh->sector); |
783 | ||
784 | /* check if prexor is active which means only process blocks | |
785 | * that are part of a read-modify-write (written) | |
786 | */ | |
d8ee0728 DW |
787 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
788 | prexor = 1; | |
91c00924 DW |
789 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
790 | for (i = disks; i--; ) { | |
791 | struct r5dev *dev = &sh->dev[i]; | |
792 | if (dev->written) | |
793 | xor_srcs[count++] = dev->page; | |
794 | } | |
795 | } else { | |
796 | xor_dest = sh->dev[pd_idx].page; | |
797 | for (i = disks; i--; ) { | |
798 | struct r5dev *dev = &sh->dev[i]; | |
799 | if (i != pd_idx) | |
800 | xor_srcs[count++] = dev->page; | |
801 | } | |
802 | } | |
803 | ||
91c00924 DW |
804 | /* 1/ if we prexor'd then the dest is reused as a source |
805 | * 2/ if we did not prexor then we are redoing the parity | |
806 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
807 | * for the synchronous xor case | |
808 | */ | |
809 | flags = ASYNC_TX_DEP_ACK | ASYNC_TX_ACK | | |
810 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
811 | ||
812 | atomic_inc(&sh->count); | |
813 | ||
814 | if (unlikely(count == 1)) { | |
815 | flags &= ~(ASYNC_TX_XOR_DROP_DST | ASYNC_TX_XOR_ZERO_DST); | |
816 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
d8ee0728 | 817 | flags, tx, ops_complete_postxor, sh); |
91c00924 DW |
818 | } else |
819 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
d8ee0728 | 820 | flags, tx, ops_complete_postxor, sh); |
91c00924 DW |
821 | } |
822 | ||
823 | static void ops_complete_check(void *stripe_head_ref) | |
824 | { | |
825 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 826 | |
e46b272b | 827 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
828 | (unsigned long long)sh->sector); |
829 | ||
ecc65c9b | 830 | sh->check_state = check_state_check_result; |
91c00924 DW |
831 | set_bit(STRIPE_HANDLE, &sh->state); |
832 | release_stripe(sh); | |
833 | } | |
834 | ||
835 | static void ops_run_check(struct stripe_head *sh) | |
836 | { | |
837 | /* kernel stack size limits the total number of disks */ | |
838 | int disks = sh->disks; | |
839 | struct page *xor_srcs[disks]; | |
840 | struct dma_async_tx_descriptor *tx; | |
841 | ||
842 | int count = 0, pd_idx = sh->pd_idx, i; | |
843 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
844 | ||
e46b272b | 845 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
846 | (unsigned long long)sh->sector); |
847 | ||
848 | for (i = disks; i--; ) { | |
849 | struct r5dev *dev = &sh->dev[i]; | |
850 | if (i != pd_idx) | |
851 | xor_srcs[count++] = dev->page; | |
852 | } | |
853 | ||
854 | tx = async_xor_zero_sum(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
855 | &sh->ops.zero_sum_result, 0, NULL, NULL, NULL); | |
856 | ||
91c00924 DW |
857 | atomic_inc(&sh->count); |
858 | tx = async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
859 | ops_complete_check, sh); | |
860 | } | |
861 | ||
600aa109 | 862 | static void raid5_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
863 | { |
864 | int overlap_clear = 0, i, disks = sh->disks; | |
865 | struct dma_async_tx_descriptor *tx = NULL; | |
866 | ||
83de75cc | 867 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
868 | ops_run_biofill(sh); |
869 | overlap_clear++; | |
870 | } | |
871 | ||
7b3a871e DW |
872 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
873 | tx = ops_run_compute5(sh); | |
874 | /* terminate the chain if postxor is not set to be run */ | |
875 | if (tx && !test_bit(STRIPE_OP_POSTXOR, &ops_request)) | |
876 | async_tx_ack(tx); | |
877 | } | |
91c00924 | 878 | |
600aa109 | 879 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
91c00924 DW |
880 | tx = ops_run_prexor(sh, tx); |
881 | ||
600aa109 | 882 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 883 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
884 | overlap_clear++; |
885 | } | |
886 | ||
600aa109 | 887 | if (test_bit(STRIPE_OP_POSTXOR, &ops_request)) |
d8ee0728 | 888 | ops_run_postxor(sh, tx); |
91c00924 | 889 | |
ecc65c9b | 890 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) |
91c00924 DW |
891 | ops_run_check(sh); |
892 | ||
91c00924 DW |
893 | if (overlap_clear) |
894 | for (i = disks; i--; ) { | |
895 | struct r5dev *dev = &sh->dev[i]; | |
896 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
897 | wake_up(&sh->raid_conf->wait_for_overlap); | |
898 | } | |
899 | } | |
900 | ||
3f294f4f | 901 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
902 | { |
903 | struct stripe_head *sh; | |
3f294f4f N |
904 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
905 | if (!sh) | |
906 | return 0; | |
907 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
908 | sh->raid_conf = conf; | |
909 | spin_lock_init(&sh->lock); | |
910 | ||
911 | if (grow_buffers(sh, conf->raid_disks)) { | |
912 | shrink_buffers(sh, conf->raid_disks); | |
913 | kmem_cache_free(conf->slab_cache, sh); | |
914 | return 0; | |
915 | } | |
7ecaa1e6 | 916 | sh->disks = conf->raid_disks; |
3f294f4f N |
917 | /* we just created an active stripe so... */ |
918 | atomic_set(&sh->count, 1); | |
919 | atomic_inc(&conf->active_stripes); | |
920 | INIT_LIST_HEAD(&sh->lru); | |
921 | release_stripe(sh); | |
922 | return 1; | |
923 | } | |
924 | ||
925 | static int grow_stripes(raid5_conf_t *conf, int num) | |
926 | { | |
e18b890b | 927 | struct kmem_cache *sc; |
1da177e4 LT |
928 | int devs = conf->raid_disks; |
929 | ||
42b9bebe N |
930 | sprintf(conf->cache_name[0], "raid5-%s", mdname(conf->mddev)); |
931 | sprintf(conf->cache_name[1], "raid5-%s-alt", mdname(conf->mddev)); | |
ad01c9e3 N |
932 | conf->active_name = 0; |
933 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 934 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 935 | 0, 0, NULL); |
1da177e4 LT |
936 | if (!sc) |
937 | return 1; | |
938 | conf->slab_cache = sc; | |
ad01c9e3 | 939 | conf->pool_size = devs; |
16a53ecc | 940 | while (num--) |
3f294f4f | 941 | if (!grow_one_stripe(conf)) |
1da177e4 | 942 | return 1; |
1da177e4 LT |
943 | return 0; |
944 | } | |
29269553 N |
945 | |
946 | #ifdef CONFIG_MD_RAID5_RESHAPE | |
ad01c9e3 N |
947 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
948 | { | |
949 | /* Make all the stripes able to hold 'newsize' devices. | |
950 | * New slots in each stripe get 'page' set to a new page. | |
951 | * | |
952 | * This happens in stages: | |
953 | * 1/ create a new kmem_cache and allocate the required number of | |
954 | * stripe_heads. | |
955 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
956 | * to the new stripe_heads. This will have the side effect of | |
957 | * freezing the array as once all stripe_heads have been collected, | |
958 | * no IO will be possible. Old stripe heads are freed once their | |
959 | * pages have been transferred over, and the old kmem_cache is | |
960 | * freed when all stripes are done. | |
961 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
962 | * we simple return a failre status - no need to clean anything up. | |
963 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
964 | * If this fails, we don't bother trying the shrink the | |
965 | * stripe_heads down again, we just leave them as they are. | |
966 | * As each stripe_head is processed the new one is released into | |
967 | * active service. | |
968 | * | |
969 | * Once step2 is started, we cannot afford to wait for a write, | |
970 | * so we use GFP_NOIO allocations. | |
971 | */ | |
972 | struct stripe_head *osh, *nsh; | |
973 | LIST_HEAD(newstripes); | |
974 | struct disk_info *ndisks; | |
b5470dc5 | 975 | int err; |
e18b890b | 976 | struct kmem_cache *sc; |
ad01c9e3 N |
977 | int i; |
978 | ||
979 | if (newsize <= conf->pool_size) | |
980 | return 0; /* never bother to shrink */ | |
981 | ||
b5470dc5 DW |
982 | err = md_allow_write(conf->mddev); |
983 | if (err) | |
984 | return err; | |
2a2275d6 | 985 | |
ad01c9e3 N |
986 | /* Step 1 */ |
987 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
988 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 989 | 0, 0, NULL); |
ad01c9e3 N |
990 | if (!sc) |
991 | return -ENOMEM; | |
992 | ||
993 | for (i = conf->max_nr_stripes; i; i--) { | |
994 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
995 | if (!nsh) | |
996 | break; | |
997 | ||
998 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
999 | ||
1000 | nsh->raid_conf = conf; | |
1001 | spin_lock_init(&nsh->lock); | |
1002 | ||
1003 | list_add(&nsh->lru, &newstripes); | |
1004 | } | |
1005 | if (i) { | |
1006 | /* didn't get enough, give up */ | |
1007 | while (!list_empty(&newstripes)) { | |
1008 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1009 | list_del(&nsh->lru); | |
1010 | kmem_cache_free(sc, nsh); | |
1011 | } | |
1012 | kmem_cache_destroy(sc); | |
1013 | return -ENOMEM; | |
1014 | } | |
1015 | /* Step 2 - Must use GFP_NOIO now. | |
1016 | * OK, we have enough stripes, start collecting inactive | |
1017 | * stripes and copying them over | |
1018 | */ | |
1019 | list_for_each_entry(nsh, &newstripes, lru) { | |
1020 | spin_lock_irq(&conf->device_lock); | |
1021 | wait_event_lock_irq(conf->wait_for_stripe, | |
1022 | !list_empty(&conf->inactive_list), | |
1023 | conf->device_lock, | |
b3b46be3 | 1024 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1025 | ); |
1026 | osh = get_free_stripe(conf); | |
1027 | spin_unlock_irq(&conf->device_lock); | |
1028 | atomic_set(&nsh->count, 1); | |
1029 | for(i=0; i<conf->pool_size; i++) | |
1030 | nsh->dev[i].page = osh->dev[i].page; | |
1031 | for( ; i<newsize; i++) | |
1032 | nsh->dev[i].page = NULL; | |
1033 | kmem_cache_free(conf->slab_cache, osh); | |
1034 | } | |
1035 | kmem_cache_destroy(conf->slab_cache); | |
1036 | ||
1037 | /* Step 3. | |
1038 | * At this point, we are holding all the stripes so the array | |
1039 | * is completely stalled, so now is a good time to resize | |
1040 | * conf->disks. | |
1041 | */ | |
1042 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1043 | if (ndisks) { | |
1044 | for (i=0; i<conf->raid_disks; i++) | |
1045 | ndisks[i] = conf->disks[i]; | |
1046 | kfree(conf->disks); | |
1047 | conf->disks = ndisks; | |
1048 | } else | |
1049 | err = -ENOMEM; | |
1050 | ||
1051 | /* Step 4, return new stripes to service */ | |
1052 | while(!list_empty(&newstripes)) { | |
1053 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1054 | list_del_init(&nsh->lru); | |
1055 | for (i=conf->raid_disks; i < newsize; i++) | |
1056 | if (nsh->dev[i].page == NULL) { | |
1057 | struct page *p = alloc_page(GFP_NOIO); | |
1058 | nsh->dev[i].page = p; | |
1059 | if (!p) | |
1060 | err = -ENOMEM; | |
1061 | } | |
1062 | release_stripe(nsh); | |
1063 | } | |
1064 | /* critical section pass, GFP_NOIO no longer needed */ | |
1065 | ||
1066 | conf->slab_cache = sc; | |
1067 | conf->active_name = 1-conf->active_name; | |
1068 | conf->pool_size = newsize; | |
1069 | return err; | |
1070 | } | |
29269553 | 1071 | #endif |
1da177e4 | 1072 | |
3f294f4f | 1073 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1074 | { |
1075 | struct stripe_head *sh; | |
1076 | ||
3f294f4f N |
1077 | spin_lock_irq(&conf->device_lock); |
1078 | sh = get_free_stripe(conf); | |
1079 | spin_unlock_irq(&conf->device_lock); | |
1080 | if (!sh) | |
1081 | return 0; | |
78bafebd | 1082 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1083 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1084 | kmem_cache_free(conf->slab_cache, sh); |
1085 | atomic_dec(&conf->active_stripes); | |
1086 | return 1; | |
1087 | } | |
1088 | ||
1089 | static void shrink_stripes(raid5_conf_t *conf) | |
1090 | { | |
1091 | while (drop_one_stripe(conf)) | |
1092 | ; | |
1093 | ||
29fc7e3e N |
1094 | if (conf->slab_cache) |
1095 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1096 | conf->slab_cache = NULL; |
1097 | } | |
1098 | ||
6712ecf8 | 1099 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1100 | { |
99c0fb5f | 1101 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1102 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1103 | int disks = sh->disks, i; |
1da177e4 | 1104 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1105 | char b[BDEVNAME_SIZE]; |
1106 | mdk_rdev_t *rdev; | |
1da177e4 | 1107 | |
1da177e4 LT |
1108 | |
1109 | for (i=0 ; i<disks; i++) | |
1110 | if (bi == &sh->dev[i].req) | |
1111 | break; | |
1112 | ||
45b4233c DW |
1113 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1114 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1115 | uptodate); |
1116 | if (i == disks) { | |
1117 | BUG(); | |
6712ecf8 | 1118 | return; |
1da177e4 LT |
1119 | } |
1120 | ||
1121 | if (uptodate) { | |
1da177e4 | 1122 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1123 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1124 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1125 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1126 | " (%lu sectors at %llu on %s)\n", | |
1127 | mdname(conf->mddev), STRIPE_SECTORS, | |
1128 | (unsigned long long)(sh->sector | |
1129 | + rdev->data_offset), | |
1130 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1131 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1132 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1133 | } | |
ba22dcbf N |
1134 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1135 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1136 | } else { |
d6950432 | 1137 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1138 | int retry = 0; |
d6950432 N |
1139 | rdev = conf->disks[i].rdev; |
1140 | ||
1da177e4 | 1141 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1142 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1143 | if (conf->mddev->degraded) |
6be9d494 BS |
1144 | printk_rl(KERN_WARNING |
1145 | "raid5:%s: read error not correctable " | |
1146 | "(sector %llu on %s).\n", | |
1147 | mdname(conf->mddev), | |
1148 | (unsigned long long)(sh->sector | |
1149 | + rdev->data_offset), | |
1150 | bdn); | |
ba22dcbf | 1151 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1152 | /* Oh, no!!! */ |
6be9d494 BS |
1153 | printk_rl(KERN_WARNING |
1154 | "raid5:%s: read error NOT corrected!! " | |
1155 | "(sector %llu on %s).\n", | |
1156 | mdname(conf->mddev), | |
1157 | (unsigned long long)(sh->sector | |
1158 | + rdev->data_offset), | |
1159 | bdn); | |
d6950432 | 1160 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1161 | > conf->max_nr_stripes) |
14f8d26b | 1162 | printk(KERN_WARNING |
d6950432 N |
1163 | "raid5:%s: Too many read errors, failing device %s.\n", |
1164 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1165 | else |
1166 | retry = 1; | |
1167 | if (retry) | |
1168 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1169 | else { | |
4e5314b5 N |
1170 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1171 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1172 | md_error(conf->mddev, rdev); |
ba22dcbf | 1173 | } |
1da177e4 LT |
1174 | } |
1175 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1176 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1177 | set_bit(STRIPE_HANDLE, &sh->state); | |
1178 | release_stripe(sh); | |
1da177e4 LT |
1179 | } |
1180 | ||
d710e138 | 1181 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1182 | { |
99c0fb5f | 1183 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1184 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1185 | int disks = sh->disks, i; |
1da177e4 LT |
1186 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1187 | ||
1da177e4 LT |
1188 | for (i=0 ; i<disks; i++) |
1189 | if (bi == &sh->dev[i].req) | |
1190 | break; | |
1191 | ||
45b4233c | 1192 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1193 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1194 | uptodate); | |
1195 | if (i == disks) { | |
1196 | BUG(); | |
6712ecf8 | 1197 | return; |
1da177e4 LT |
1198 | } |
1199 | ||
1da177e4 LT |
1200 | if (!uptodate) |
1201 | md_error(conf->mddev, conf->disks[i].rdev); | |
1202 | ||
1203 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1204 | ||
1205 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1206 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1207 | release_stripe(sh); |
1da177e4 LT |
1208 | } |
1209 | ||
1210 | ||
1211 | static sector_t compute_blocknr(struct stripe_head *sh, int i); | |
1212 | ||
d710e138 | 1213 | static void raid5_build_block(struct stripe_head *sh, int i) |
1da177e4 LT |
1214 | { |
1215 | struct r5dev *dev = &sh->dev[i]; | |
1216 | ||
1217 | bio_init(&dev->req); | |
1218 | dev->req.bi_io_vec = &dev->vec; | |
1219 | dev->req.bi_vcnt++; | |
1220 | dev->req.bi_max_vecs++; | |
1221 | dev->vec.bv_page = dev->page; | |
1222 | dev->vec.bv_len = STRIPE_SIZE; | |
1223 | dev->vec.bv_offset = 0; | |
1224 | ||
1225 | dev->req.bi_sector = sh->sector; | |
1226 | dev->req.bi_private = sh; | |
1227 | ||
1228 | dev->flags = 0; | |
16a53ecc | 1229 | dev->sector = compute_blocknr(sh, i); |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1233 | { | |
1234 | char b[BDEVNAME_SIZE]; | |
1235 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1236 | pr_debug("raid5: error called\n"); |
1da177e4 | 1237 | |
b2d444d7 | 1238 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1239 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1240 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1241 | unsigned long flags; | |
1242 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1243 | mddev->degraded++; |
c04be0aa | 1244 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1245 | /* |
1246 | * if recovery was running, make sure it aborts. | |
1247 | */ | |
dfc70645 | 1248 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1249 | } |
b2d444d7 | 1250 | set_bit(Faulty, &rdev->flags); |
d710e138 N |
1251 | printk(KERN_ALERT |
1252 | "raid5: Disk failure on %s, disabling device.\n" | |
1253 | "raid5: Operation continuing on %d devices.\n", | |
1254 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); | |
1da177e4 | 1255 | } |
16a53ecc | 1256 | } |
1da177e4 LT |
1257 | |
1258 | /* | |
1259 | * Input: a 'big' sector number, | |
1260 | * Output: index of the data and parity disk, and the sector # in them. | |
1261 | */ | |
112bf897 | 1262 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1263 | int previous, int *dd_idx, |
1264 | struct stripe_head *sh) | |
1da177e4 LT |
1265 | { |
1266 | long stripe; | |
1267 | unsigned long chunk_number; | |
1268 | unsigned int chunk_offset; | |
911d4ee8 | 1269 | int pd_idx, qd_idx; |
1da177e4 LT |
1270 | sector_t new_sector; |
1271 | int sectors_per_chunk = conf->chunk_size >> 9; | |
112bf897 N |
1272 | int raid_disks = previous ? conf->previous_raid_disks |
1273 | : conf->raid_disks; | |
1274 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1275 | |
1276 | /* First compute the information on this sector */ | |
1277 | ||
1278 | /* | |
1279 | * Compute the chunk number and the sector offset inside the chunk | |
1280 | */ | |
1281 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1282 | chunk_number = r_sector; | |
1283 | BUG_ON(r_sector != chunk_number); | |
1284 | ||
1285 | /* | |
1286 | * Compute the stripe number | |
1287 | */ | |
1288 | stripe = chunk_number / data_disks; | |
1289 | ||
1290 | /* | |
1291 | * Compute the data disk and parity disk indexes inside the stripe | |
1292 | */ | |
1293 | *dd_idx = chunk_number % data_disks; | |
1294 | ||
1295 | /* | |
1296 | * Select the parity disk based on the user selected algorithm. | |
1297 | */ | |
911d4ee8 | 1298 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1299 | switch(conf->level) { |
1300 | case 4: | |
911d4ee8 | 1301 | pd_idx = data_disks; |
16a53ecc N |
1302 | break; |
1303 | case 5: | |
1304 | switch (conf->algorithm) { | |
1da177e4 | 1305 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1306 | pd_idx = data_disks - stripe % raid_disks; |
1307 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1308 | (*dd_idx)++; |
1309 | break; | |
1310 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1311 | pd_idx = stripe % raid_disks; |
1312 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1313 | (*dd_idx)++; |
1314 | break; | |
1315 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1316 | pd_idx = data_disks - stripe % raid_disks; |
1317 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 LT |
1318 | break; |
1319 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1320 | pd_idx = stripe % raid_disks; |
1321 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 | 1322 | break; |
99c0fb5f N |
1323 | case ALGORITHM_PARITY_0: |
1324 | pd_idx = 0; | |
1325 | (*dd_idx)++; | |
1326 | break; | |
1327 | case ALGORITHM_PARITY_N: | |
1328 | pd_idx = data_disks; | |
1329 | break; | |
1da177e4 | 1330 | default: |
14f8d26b | 1331 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 | 1332 | conf->algorithm); |
99c0fb5f | 1333 | BUG(); |
16a53ecc N |
1334 | } |
1335 | break; | |
1336 | case 6: | |
1337 | ||
16a53ecc N |
1338 | switch (conf->algorithm) { |
1339 | case ALGORITHM_LEFT_ASYMMETRIC: | |
911d4ee8 N |
1340 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1341 | qd_idx = pd_idx + 1; | |
1342 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1343 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1344 | qd_idx = 0; |
1345 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1346 | (*dd_idx) += 2; /* D D P Q D */ |
1347 | break; | |
1348 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1349 | pd_idx = stripe % raid_disks; |
1350 | qd_idx = pd_idx + 1; | |
1351 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1352 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1353 | qd_idx = 0; |
1354 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1355 | (*dd_idx) += 2; /* D D P Q D */ |
1356 | break; | |
1357 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1358 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1359 | qd_idx = (pd_idx + 1) % raid_disks; | |
1360 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1361 | break; |
1362 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1363 | pd_idx = stripe % raid_disks; |
1364 | qd_idx = (pd_idx + 1) % raid_disks; | |
1365 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1366 | break; |
99c0fb5f N |
1367 | |
1368 | case ALGORITHM_PARITY_0: | |
1369 | pd_idx = 0; | |
1370 | qd_idx = 1; | |
1371 | (*dd_idx) += 2; | |
1372 | break; | |
1373 | case ALGORITHM_PARITY_N: | |
1374 | pd_idx = data_disks; | |
1375 | qd_idx = data_disks + 1; | |
1376 | break; | |
1377 | ||
1378 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1379 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1380 | * of blocks for computing Q is different. | |
1381 | */ | |
1382 | pd_idx = stripe % raid_disks; | |
1383 | qd_idx = pd_idx + 1; | |
1384 | if (pd_idx == raid_disks-1) { | |
1385 | (*dd_idx)++; /* Q D D D P */ | |
1386 | qd_idx = 0; | |
1387 | } else if (*dd_idx >= pd_idx) | |
1388 | (*dd_idx) += 2; /* D D P Q D */ | |
1389 | break; | |
1390 | ||
1391 | case ALGORITHM_ROTATING_N_RESTART: | |
1392 | /* Same a left_asymmetric, by first stripe is | |
1393 | * D D D P Q rather than | |
1394 | * Q D D D P | |
1395 | */ | |
1396 | pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
1397 | qd_idx = pd_idx + 1; | |
1398 | if (pd_idx == raid_disks-1) { | |
1399 | (*dd_idx)++; /* Q D D D P */ | |
1400 | qd_idx = 0; | |
1401 | } else if (*dd_idx >= pd_idx) | |
1402 | (*dd_idx) += 2; /* D D P Q D */ | |
1403 | break; | |
1404 | ||
1405 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1406 | /* Same as left_symmetric but Q is before P */ | |
1407 | pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1408 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; | |
1409 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1410 | break; | |
1411 | ||
1412 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1413 | /* RAID5 left_asymmetric, with Q on last device */ | |
1414 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1415 | if (*dd_idx >= pd_idx) | |
1416 | (*dd_idx)++; | |
1417 | qd_idx = raid_disks - 1; | |
1418 | break; | |
1419 | ||
1420 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1421 | pd_idx = stripe % (raid_disks-1); | |
1422 | if (*dd_idx >= pd_idx) | |
1423 | (*dd_idx)++; | |
1424 | qd_idx = raid_disks - 1; | |
1425 | break; | |
1426 | ||
1427 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1428 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1429 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1430 | qd_idx = raid_disks - 1; | |
1431 | break; | |
1432 | ||
1433 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1434 | pd_idx = stripe % (raid_disks-1); | |
1435 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1436 | qd_idx = raid_disks - 1; | |
1437 | break; | |
1438 | ||
1439 | case ALGORITHM_PARITY_0_6: | |
1440 | pd_idx = 0; | |
1441 | (*dd_idx)++; | |
1442 | qd_idx = raid_disks - 1; | |
1443 | break; | |
1444 | ||
1445 | ||
16a53ecc | 1446 | default: |
d710e138 N |
1447 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1448 | conf->algorithm); | |
99c0fb5f | 1449 | BUG(); |
16a53ecc N |
1450 | } |
1451 | break; | |
1da177e4 LT |
1452 | } |
1453 | ||
911d4ee8 N |
1454 | if (sh) { |
1455 | sh->pd_idx = pd_idx; | |
1456 | sh->qd_idx = qd_idx; | |
1457 | } | |
1da177e4 LT |
1458 | /* |
1459 | * Finally, compute the new sector number | |
1460 | */ | |
1461 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1462 | return new_sector; | |
1463 | } | |
1464 | ||
1465 | ||
1466 | static sector_t compute_blocknr(struct stripe_head *sh, int i) | |
1467 | { | |
1468 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1469 | int raid_disks = sh->disks; |
1470 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1471 | sector_t new_sector = sh->sector, check; |
1472 | int sectors_per_chunk = conf->chunk_size >> 9; | |
1473 | sector_t stripe; | |
1474 | int chunk_offset; | |
911d4ee8 | 1475 | int chunk_number, dummy1, dd_idx = i; |
1da177e4 | 1476 | sector_t r_sector; |
911d4ee8 | 1477 | struct stripe_head sh2; |
1da177e4 | 1478 | |
16a53ecc | 1479 | |
1da177e4 LT |
1480 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1481 | stripe = new_sector; | |
1482 | BUG_ON(new_sector != stripe); | |
1483 | ||
16a53ecc N |
1484 | if (i == sh->pd_idx) |
1485 | return 0; | |
1486 | switch(conf->level) { | |
1487 | case 4: break; | |
1488 | case 5: | |
1489 | switch (conf->algorithm) { | |
1da177e4 LT |
1490 | case ALGORITHM_LEFT_ASYMMETRIC: |
1491 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1492 | if (i > sh->pd_idx) | |
1493 | i--; | |
1494 | break; | |
1495 | case ALGORITHM_LEFT_SYMMETRIC: | |
1496 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1497 | if (i < sh->pd_idx) | |
1498 | i += raid_disks; | |
1499 | i -= (sh->pd_idx + 1); | |
1500 | break; | |
99c0fb5f N |
1501 | case ALGORITHM_PARITY_0: |
1502 | i -= 1; | |
1503 | break; | |
1504 | case ALGORITHM_PARITY_N: | |
1505 | break; | |
1da177e4 | 1506 | default: |
14f8d26b | 1507 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
16a53ecc | 1508 | conf->algorithm); |
99c0fb5f | 1509 | BUG(); |
16a53ecc N |
1510 | } |
1511 | break; | |
1512 | case 6: | |
d0dabf7e | 1513 | if (i == sh->qd_idx) |
16a53ecc N |
1514 | return 0; /* It is the Q disk */ |
1515 | switch (conf->algorithm) { | |
1516 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1517 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1518 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1519 | case ALGORITHM_ROTATING_N_RESTART: | |
1520 | if (sh->pd_idx == raid_disks-1) | |
1521 | i--; /* Q D D D P */ | |
16a53ecc N |
1522 | else if (i > sh->pd_idx) |
1523 | i -= 2; /* D D P Q D */ | |
1524 | break; | |
1525 | case ALGORITHM_LEFT_SYMMETRIC: | |
1526 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1527 | if (sh->pd_idx == raid_disks-1) | |
1528 | i--; /* Q D D D P */ | |
1529 | else { | |
1530 | /* D D P Q D */ | |
1531 | if (i < sh->pd_idx) | |
1532 | i += raid_disks; | |
1533 | i -= (sh->pd_idx + 2); | |
1534 | } | |
1535 | break; | |
99c0fb5f N |
1536 | case ALGORITHM_PARITY_0: |
1537 | i -= 2; | |
1538 | break; | |
1539 | case ALGORITHM_PARITY_N: | |
1540 | break; | |
1541 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1542 | if (sh->pd_idx == 0) | |
1543 | i--; /* P D D D Q */ | |
1544 | else if (i > sh->pd_idx) | |
1545 | i -= 2; /* D D Q P D */ | |
1546 | break; | |
1547 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1548 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1549 | if (i > sh->pd_idx) | |
1550 | i--; | |
1551 | break; | |
1552 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1553 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1554 | if (i < sh->pd_idx) | |
1555 | i += data_disks + 1; | |
1556 | i -= (sh->pd_idx + 1); | |
1557 | break; | |
1558 | case ALGORITHM_PARITY_0_6: | |
1559 | i -= 1; | |
1560 | break; | |
16a53ecc | 1561 | default: |
d710e138 N |
1562 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1563 | conf->algorithm); | |
99c0fb5f | 1564 | BUG(); |
16a53ecc N |
1565 | } |
1566 | break; | |
1da177e4 LT |
1567 | } |
1568 | ||
1569 | chunk_number = stripe * data_disks + i; | |
1570 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1571 | ||
112bf897 N |
1572 | check = raid5_compute_sector(conf, r_sector, |
1573 | (raid_disks != conf->raid_disks), | |
911d4ee8 N |
1574 | &dummy1, &sh2); |
1575 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx | |
1576 | || sh2.qd_idx != sh->qd_idx) { | |
14f8d26b | 1577 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1578 | return 0; |
1579 | } | |
1580 | return r_sector; | |
1581 | } | |
1582 | ||
1583 | ||
1584 | ||
1585 | /* | |
16a53ecc N |
1586 | * Copy data between a page in the stripe cache, and one or more bion |
1587 | * The page could align with the middle of the bio, or there could be | |
1588 | * several bion, each with several bio_vecs, which cover part of the page | |
1589 | * Multiple bion are linked together on bi_next. There may be extras | |
1590 | * at the end of this list. We ignore them. | |
1da177e4 LT |
1591 | */ |
1592 | static void copy_data(int frombio, struct bio *bio, | |
1593 | struct page *page, | |
1594 | sector_t sector) | |
1595 | { | |
1596 | char *pa = page_address(page); | |
1597 | struct bio_vec *bvl; | |
1598 | int i; | |
1599 | int page_offset; | |
1600 | ||
1601 | if (bio->bi_sector >= sector) | |
1602 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
1603 | else | |
1604 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
1605 | bio_for_each_segment(bvl, bio, i) { | |
1606 | int len = bio_iovec_idx(bio,i)->bv_len; | |
1607 | int clen; | |
1608 | int b_offset = 0; | |
1609 | ||
1610 | if (page_offset < 0) { | |
1611 | b_offset = -page_offset; | |
1612 | page_offset += b_offset; | |
1613 | len -= b_offset; | |
1614 | } | |
1615 | ||
1616 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1617 | clen = STRIPE_SIZE - page_offset; | |
1618 | else clen = len; | |
16a53ecc | 1619 | |
1da177e4 LT |
1620 | if (clen > 0) { |
1621 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
1622 | if (frombio) | |
1623 | memcpy(pa+page_offset, ba+b_offset, clen); | |
1624 | else | |
1625 | memcpy(ba+b_offset, pa+page_offset, clen); | |
1626 | __bio_kunmap_atomic(ba, KM_USER0); | |
1627 | } | |
1628 | if (clen < len) /* hit end of page */ | |
1629 | break; | |
1630 | page_offset += len; | |
1631 | } | |
1632 | } | |
1633 | ||
9bc89cd8 DW |
1634 | #define check_xor() do { \ |
1635 | if (count == MAX_XOR_BLOCKS) { \ | |
1636 | xor_blocks(count, STRIPE_SIZE, dest, ptr);\ | |
1637 | count = 0; \ | |
1638 | } \ | |
1da177e4 LT |
1639 | } while(0) |
1640 | ||
16a53ecc N |
1641 | static void compute_parity6(struct stripe_head *sh, int method) |
1642 | { | |
bff61975 | 1643 | raid5_conf_t *conf = sh->raid_conf; |
d0dabf7e | 1644 | int i, pd_idx, qd_idx, d0_idx, disks = sh->disks, count; |
16a53ecc N |
1645 | struct bio *chosen; |
1646 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1647 | void *ptrs[disks]; | |
1648 | ||
d0dabf7e N |
1649 | pd_idx = sh->pd_idx; |
1650 | qd_idx = sh->qd_idx; | |
1651 | d0_idx = raid6_d0(sh); | |
16a53ecc | 1652 | |
45b4233c | 1653 | pr_debug("compute_parity, stripe %llu, method %d\n", |
16a53ecc N |
1654 | (unsigned long long)sh->sector, method); |
1655 | ||
1656 | switch(method) { | |
1657 | case READ_MODIFY_WRITE: | |
1658 | BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ | |
1659 | case RECONSTRUCT_WRITE: | |
1660 | for (i= disks; i-- ;) | |
1661 | if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { | |
1662 | chosen = sh->dev[i].towrite; | |
1663 | sh->dev[i].towrite = NULL; | |
1664 | ||
1665 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1666 | wake_up(&conf->wait_for_overlap); | |
1667 | ||
52e5f9d1 | 1668 | BUG_ON(sh->dev[i].written); |
16a53ecc N |
1669 | sh->dev[i].written = chosen; |
1670 | } | |
1671 | break; | |
1672 | case CHECK_PARITY: | |
1673 | BUG(); /* Not implemented yet */ | |
1674 | } | |
1675 | ||
1676 | for (i = disks; i--;) | |
1677 | if (sh->dev[i].written) { | |
1678 | sector_t sector = sh->dev[i].sector; | |
1679 | struct bio *wbi = sh->dev[i].written; | |
1680 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
1681 | copy_data(1, wbi, sh->dev[i].page, sector); | |
1682 | wbi = r5_next_bio(wbi, sector); | |
1683 | } | |
1684 | ||
1685 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
1686 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
1687 | } | |
1688 | ||
d0dabf7e N |
1689 | /* Note that unlike RAID-5, the ordering of the disks matters greatly.*/ |
1690 | /* FIX: Is this ordering of drives even remotely optimal? */ | |
1691 | count = 0; | |
1692 | i = d0_idx; | |
1693 | do { | |
1694 | int slot = raid6_idx_to_slot(i, sh, &count); | |
1695 | ptrs[slot] = page_address(sh->dev[i].page); | |
1696 | if (slot < sh->disks - 2 && | |
1697 | !test_bit(R5_UPTODATE, &sh->dev[i].flags)) { | |
1698 | printk(KERN_ERR "block %d/%d not uptodate " | |
1699 | "on parity calc\n", i, count); | |
1700 | BUG(); | |
1701 | } | |
1702 | i = raid6_next_disk(i, disks); | |
1703 | } while (i != d0_idx); | |
1704 | BUG_ON(count+2 != disks); | |
16a53ecc N |
1705 | |
1706 | raid6_call.gen_syndrome(disks, STRIPE_SIZE, ptrs); | |
1707 | ||
1708 | switch(method) { | |
1709 | case RECONSTRUCT_WRITE: | |
1710 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1711 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1712 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1713 | set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); | |
1714 | break; | |
1715 | case UPDATE_PARITY: | |
1716 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1717 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1718 | break; | |
1719 | } | |
1720 | } | |
1721 | ||
1722 | ||
1723 | /* Compute one missing block */ | |
1724 | static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero) | |
1725 | { | |
f416885e | 1726 | int i, count, disks = sh->disks; |
9bc89cd8 | 1727 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
d0dabf7e | 1728 | int qd_idx = sh->qd_idx; |
16a53ecc | 1729 | |
45b4233c | 1730 | pr_debug("compute_block_1, stripe %llu, idx %d\n", |
16a53ecc N |
1731 | (unsigned long long)sh->sector, dd_idx); |
1732 | ||
1733 | if ( dd_idx == qd_idx ) { | |
1734 | /* We're actually computing the Q drive */ | |
1735 | compute_parity6(sh, UPDATE_PARITY); | |
1736 | } else { | |
9bc89cd8 DW |
1737 | dest = page_address(sh->dev[dd_idx].page); |
1738 | if (!nozero) memset(dest, 0, STRIPE_SIZE); | |
1739 | count = 0; | |
16a53ecc N |
1740 | for (i = disks ; i--; ) { |
1741 | if (i == dd_idx || i == qd_idx) | |
1742 | continue; | |
1743 | p = page_address(sh->dev[i].page); | |
1744 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1745 | ptr[count++] = p; | |
1746 | else | |
1747 | printk("compute_block() %d, stripe %llu, %d" | |
1748 | " not present\n", dd_idx, | |
1749 | (unsigned long long)sh->sector, i); | |
1750 | ||
1751 | check_xor(); | |
1752 | } | |
9bc89cd8 DW |
1753 | if (count) |
1754 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
16a53ecc N |
1755 | if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
1756 | else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
1757 | } | |
1758 | } | |
1759 | ||
1760 | /* Compute two missing blocks */ | |
1761 | static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) | |
1762 | { | |
f416885e | 1763 | int i, count, disks = sh->disks; |
d0dabf7e N |
1764 | int d0_idx = raid6_d0(sh); |
1765 | int faila = -1, failb = -1; | |
1766 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1767 | void *ptrs[disks]; | |
16a53ecc | 1768 | |
d0dabf7e N |
1769 | count = 0; |
1770 | i = d0_idx; | |
1771 | do { | |
1772 | int slot; | |
1773 | slot = raid6_idx_to_slot(i, sh, &count); | |
1774 | ptrs[slot] = page_address(sh->dev[i].page); | |
1775 | if (i == dd_idx1) | |
1776 | faila = slot; | |
1777 | if (i == dd_idx2) | |
1778 | failb = slot; | |
1779 | i = raid6_next_disk(i, disks); | |
1780 | } while (i != d0_idx); | |
1781 | BUG_ON(count+2 != disks); | |
16a53ecc N |
1782 | |
1783 | BUG_ON(faila == failb); | |
1784 | if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } | |
1785 | ||
45b4233c | 1786 | pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", |
d0dabf7e N |
1787 | (unsigned long long)sh->sector, dd_idx1, dd_idx2, |
1788 | faila, failb); | |
16a53ecc N |
1789 | |
1790 | if ( failb == disks-1 ) { | |
1791 | /* Q disk is one of the missing disks */ | |
1792 | if ( faila == disks-2 ) { | |
1793 | /* Missing P+Q, just recompute */ | |
1794 | compute_parity6(sh, UPDATE_PARITY); | |
1795 | return; | |
1796 | } else { | |
1797 | /* We're missing D+Q; recompute D from P */ | |
d0dabf7e N |
1798 | compute_block_1(sh, ((dd_idx1 == sh->qd_idx) ? |
1799 | dd_idx2 : dd_idx1), | |
1800 | 0); | |
16a53ecc N |
1801 | compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */ |
1802 | return; | |
1803 | } | |
1804 | } | |
1805 | ||
d0dabf7e N |
1806 | /* We're missing D+P or D+D; */ |
1807 | if (failb == disks-2) { | |
1808 | /* We're missing D+P. */ | |
1809 | raid6_datap_recov(disks, STRIPE_SIZE, faila, ptrs); | |
1810 | } else { | |
1811 | /* We're missing D+D. */ | |
1812 | raid6_2data_recov(disks, STRIPE_SIZE, faila, failb, ptrs); | |
16a53ecc | 1813 | } |
d0dabf7e N |
1814 | |
1815 | /* Both the above update both missing blocks */ | |
1816 | set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); | |
1817 | set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); | |
16a53ecc N |
1818 | } |
1819 | ||
600aa109 | 1820 | static void |
1fe797e6 | 1821 | schedule_reconstruction5(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 1822 | int rcw, int expand) |
e33129d8 DW |
1823 | { |
1824 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
e33129d8 DW |
1825 | |
1826 | if (rcw) { | |
1827 | /* if we are not expanding this is a proper write request, and | |
1828 | * there will be bios with new data to be drained into the | |
1829 | * stripe cache | |
1830 | */ | |
1831 | if (!expand) { | |
600aa109 DW |
1832 | sh->reconstruct_state = reconstruct_state_drain_run; |
1833 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1834 | } else | |
1835 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 1836 | |
600aa109 | 1837 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); |
e33129d8 DW |
1838 | |
1839 | for (i = disks; i--; ) { | |
1840 | struct r5dev *dev = &sh->dev[i]; | |
1841 | ||
1842 | if (dev->towrite) { | |
1843 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 1844 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
1845 | if (!expand) |
1846 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1847 | s->locked++; |
e33129d8 DW |
1848 | } |
1849 | } | |
600aa109 | 1850 | if (s->locked + 1 == disks) |
8b3e6cdc DW |
1851 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
1852 | atomic_inc(&sh->raid_conf->pending_full_writes); | |
e33129d8 DW |
1853 | } else { |
1854 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
1855 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
1856 | ||
d8ee0728 | 1857 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
1858 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
1859 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1860 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); | |
e33129d8 DW |
1861 | |
1862 | for (i = disks; i--; ) { | |
1863 | struct r5dev *dev = &sh->dev[i]; | |
1864 | if (i == pd_idx) | |
1865 | continue; | |
1866 | ||
e33129d8 DW |
1867 | if (dev->towrite && |
1868 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
1869 | test_bit(R5_Wantcompute, &dev->flags))) { |
1870 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
1871 | set_bit(R5_LOCKED, &dev->flags); |
1872 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1873 | s->locked++; |
e33129d8 DW |
1874 | } |
1875 | } | |
1876 | } | |
1877 | ||
1878 | /* keep the parity disk locked while asynchronous operations | |
1879 | * are in flight | |
1880 | */ | |
1881 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1882 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 1883 | s->locked++; |
e33129d8 | 1884 | |
600aa109 | 1885 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 1886 | __func__, (unsigned long long)sh->sector, |
600aa109 | 1887 | s->locked, s->ops_request); |
e33129d8 | 1888 | } |
16a53ecc | 1889 | |
1da177e4 LT |
1890 | /* |
1891 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 1892 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
1893 | * The bi_next chain must be in order. |
1894 | */ | |
1895 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
1896 | { | |
1897 | struct bio **bip; | |
1898 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 1899 | int firstwrite=0; |
1da177e4 | 1900 | |
45b4233c | 1901 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
1902 | (unsigned long long)bi->bi_sector, |
1903 | (unsigned long long)sh->sector); | |
1904 | ||
1905 | ||
1906 | spin_lock(&sh->lock); | |
1907 | spin_lock_irq(&conf->device_lock); | |
72626685 | 1908 | if (forwrite) { |
1da177e4 | 1909 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
1910 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
1911 | firstwrite = 1; | |
1912 | } else | |
1da177e4 LT |
1913 | bip = &sh->dev[dd_idx].toread; |
1914 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
1915 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
1916 | goto overlap; | |
1917 | bip = & (*bip)->bi_next; | |
1918 | } | |
1919 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
1920 | goto overlap; | |
1921 | ||
78bafebd | 1922 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
1923 | if (*bip) |
1924 | bi->bi_next = *bip; | |
1925 | *bip = bi; | |
960e739d | 1926 | bi->bi_phys_segments++; |
1da177e4 LT |
1927 | spin_unlock_irq(&conf->device_lock); |
1928 | spin_unlock(&sh->lock); | |
1929 | ||
45b4233c | 1930 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
1931 | (unsigned long long)bi->bi_sector, |
1932 | (unsigned long long)sh->sector, dd_idx); | |
1933 | ||
72626685 | 1934 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
1935 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
1936 | STRIPE_SECTORS, 0); | |
ae3c20cc | 1937 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
1938 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
1939 | } | |
1940 | ||
1da177e4 LT |
1941 | if (forwrite) { |
1942 | /* check if page is covered */ | |
1943 | sector_t sector = sh->dev[dd_idx].sector; | |
1944 | for (bi=sh->dev[dd_idx].towrite; | |
1945 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
1946 | bi && bi->bi_sector <= sector; | |
1947 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
1948 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
1949 | sector = bi->bi_sector + (bi->bi_size>>9); | |
1950 | } | |
1951 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
1952 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
1953 | } | |
1954 | return 1; | |
1955 | ||
1956 | overlap: | |
1957 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
1958 | spin_unlock_irq(&conf->device_lock); | |
1959 | spin_unlock(&sh->lock); | |
1960 | return 0; | |
1961 | } | |
1962 | ||
29269553 N |
1963 | static void end_reshape(raid5_conf_t *conf); |
1964 | ||
16a53ecc N |
1965 | static int page_is_zero(struct page *p) |
1966 | { | |
1967 | char *a = page_address(p); | |
1968 | return ((*(u32*)a) == 0 && | |
1969 | memcmp(a, a+4, STRIPE_SIZE-4)==0); | |
1970 | } | |
1971 | ||
911d4ee8 N |
1972 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
1973 | struct stripe_head *sh) | |
ccfcc3c1 N |
1974 | { |
1975 | int sectors_per_chunk = conf->chunk_size >> 9; | |
911d4ee8 | 1976 | int dd_idx; |
2d2063ce | 1977 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 1978 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 1979 | |
112bf897 N |
1980 | raid5_compute_sector(conf, |
1981 | stripe * (disks - conf->max_degraded) | |
b875e531 | 1982 | *sectors_per_chunk + chunk_offset, |
112bf897 | 1983 | previous, |
911d4ee8 | 1984 | &dd_idx, sh); |
ccfcc3c1 N |
1985 | } |
1986 | ||
a4456856 | 1987 | static void |
1fe797e6 | 1988 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
1989 | struct stripe_head_state *s, int disks, |
1990 | struct bio **return_bi) | |
1991 | { | |
1992 | int i; | |
1993 | for (i = disks; i--; ) { | |
1994 | struct bio *bi; | |
1995 | int bitmap_end = 0; | |
1996 | ||
1997 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
1998 | mdk_rdev_t *rdev; | |
1999 | rcu_read_lock(); | |
2000 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2001 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2002 | /* multiple read failures in one stripe */ | |
2003 | md_error(conf->mddev, rdev); | |
2004 | rcu_read_unlock(); | |
2005 | } | |
2006 | spin_lock_irq(&conf->device_lock); | |
2007 | /* fail all writes first */ | |
2008 | bi = sh->dev[i].towrite; | |
2009 | sh->dev[i].towrite = NULL; | |
2010 | if (bi) { | |
2011 | s->to_write--; | |
2012 | bitmap_end = 1; | |
2013 | } | |
2014 | ||
2015 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2016 | wake_up(&conf->wait_for_overlap); | |
2017 | ||
2018 | while (bi && bi->bi_sector < | |
2019 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2020 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2021 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2022 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2023 | md_write_end(conf->mddev); |
2024 | bi->bi_next = *return_bi; | |
2025 | *return_bi = bi; | |
2026 | } | |
2027 | bi = nextbi; | |
2028 | } | |
2029 | /* and fail all 'written' */ | |
2030 | bi = sh->dev[i].written; | |
2031 | sh->dev[i].written = NULL; | |
2032 | if (bi) bitmap_end = 1; | |
2033 | while (bi && bi->bi_sector < | |
2034 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2035 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2036 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2037 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2038 | md_write_end(conf->mddev); |
2039 | bi->bi_next = *return_bi; | |
2040 | *return_bi = bi; | |
2041 | } | |
2042 | bi = bi2; | |
2043 | } | |
2044 | ||
b5e98d65 DW |
2045 | /* fail any reads if this device is non-operational and |
2046 | * the data has not reached the cache yet. | |
2047 | */ | |
2048 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2049 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2050 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2051 | bi = sh->dev[i].toread; |
2052 | sh->dev[i].toread = NULL; | |
2053 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2054 | wake_up(&conf->wait_for_overlap); | |
2055 | if (bi) s->to_read--; | |
2056 | while (bi && bi->bi_sector < | |
2057 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2058 | struct bio *nextbi = | |
2059 | r5_next_bio(bi, sh->dev[i].sector); | |
2060 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2061 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2062 | bi->bi_next = *return_bi; |
2063 | *return_bi = bi; | |
2064 | } | |
2065 | bi = nextbi; | |
2066 | } | |
2067 | } | |
2068 | spin_unlock_irq(&conf->device_lock); | |
2069 | if (bitmap_end) | |
2070 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2071 | STRIPE_SECTORS, 0, 0); | |
2072 | } | |
2073 | ||
8b3e6cdc DW |
2074 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2075 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2076 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2077 | } |
2078 | ||
1fe797e6 DW |
2079 | /* fetch_block5 - checks the given member device to see if its data needs |
2080 | * to be read or computed to satisfy a request. | |
2081 | * | |
2082 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2083 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2084 | */ |
1fe797e6 DW |
2085 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2086 | int disk_idx, int disks) | |
f38e1219 DW |
2087 | { |
2088 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2089 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2090 | ||
f38e1219 DW |
2091 | /* is the data in this block needed, and can we get it? */ |
2092 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2093 | !test_bit(R5_UPTODATE, &dev->flags) && |
2094 | (dev->toread || | |
2095 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2096 | s->syncing || s->expanding || | |
2097 | (s->failed && | |
2098 | (failed_dev->toread || | |
2099 | (failed_dev->towrite && | |
2100 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2101 | /* We would like to get this block, possibly by computing it, |
2102 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2103 | */ |
2104 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2105 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2106 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2107 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2108 | set_bit(R5_Wantcompute, &dev->flags); |
2109 | sh->ops.target = disk_idx; | |
2110 | s->req_compute = 1; | |
f38e1219 DW |
2111 | /* Careful: from this point on 'uptodate' is in the eye |
2112 | * of raid5_run_ops which services 'compute' operations | |
2113 | * before writes. R5_Wantcompute flags a block that will | |
2114 | * be R5_UPTODATE by the time it is needed for a | |
2115 | * subsequent operation. | |
2116 | */ | |
2117 | s->uptodate++; | |
1fe797e6 | 2118 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2119 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2120 | set_bit(R5_LOCKED, &dev->flags); |
2121 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2122 | s->locked++; |
2123 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2124 | s->syncing); | |
2125 | } | |
2126 | } | |
2127 | ||
1fe797e6 | 2128 | return 0; |
f38e1219 DW |
2129 | } |
2130 | ||
1fe797e6 DW |
2131 | /** |
2132 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2133 | */ | |
2134 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2135 | struct stripe_head_state *s, int disks) |
2136 | { | |
2137 | int i; | |
f38e1219 | 2138 | |
f38e1219 DW |
2139 | /* look for blocks to read/compute, skip this if a compute |
2140 | * is already in flight, or if the stripe contents are in the | |
2141 | * midst of changing due to a write | |
2142 | */ | |
976ea8d4 | 2143 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2144 | !sh->reconstruct_state) |
f38e1219 | 2145 | for (i = disks; i--; ) |
1fe797e6 | 2146 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2147 | break; |
a4456856 DW |
2148 | set_bit(STRIPE_HANDLE, &sh->state); |
2149 | } | |
2150 | ||
1fe797e6 | 2151 | static void handle_stripe_fill6(struct stripe_head *sh, |
a4456856 DW |
2152 | struct stripe_head_state *s, struct r6_state *r6s, |
2153 | int disks) | |
2154 | { | |
2155 | int i; | |
2156 | for (i = disks; i--; ) { | |
2157 | struct r5dev *dev = &sh->dev[i]; | |
2158 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2159 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2160 | (dev->toread || (dev->towrite && | |
2161 | !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2162 | s->syncing || s->expanding || | |
2163 | (s->failed >= 1 && | |
2164 | (sh->dev[r6s->failed_num[0]].toread || | |
2165 | s->to_write)) || | |
2166 | (s->failed >= 2 && | |
2167 | (sh->dev[r6s->failed_num[1]].toread || | |
2168 | s->to_write)))) { | |
2169 | /* we would like to get this block, possibly | |
2170 | * by computing it, but we might not be able to | |
2171 | */ | |
c337869d DW |
2172 | if ((s->uptodate == disks - 1) && |
2173 | (s->failed && (i == r6s->failed_num[0] || | |
2174 | i == r6s->failed_num[1]))) { | |
45b4233c | 2175 | pr_debug("Computing stripe %llu block %d\n", |
a4456856 DW |
2176 | (unsigned long long)sh->sector, i); |
2177 | compute_block_1(sh, i, 0); | |
2178 | s->uptodate++; | |
2179 | } else if ( s->uptodate == disks-2 && s->failed >= 2 ) { | |
2180 | /* Computing 2-failure is *very* expensive; only | |
2181 | * do it if failed >= 2 | |
2182 | */ | |
2183 | int other; | |
2184 | for (other = disks; other--; ) { | |
2185 | if (other == i) | |
2186 | continue; | |
2187 | if (!test_bit(R5_UPTODATE, | |
2188 | &sh->dev[other].flags)) | |
2189 | break; | |
2190 | } | |
2191 | BUG_ON(other < 0); | |
45b4233c | 2192 | pr_debug("Computing stripe %llu blocks %d,%d\n", |
a4456856 DW |
2193 | (unsigned long long)sh->sector, |
2194 | i, other); | |
2195 | compute_block_2(sh, i, other); | |
2196 | s->uptodate += 2; | |
2197 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2198 | set_bit(R5_LOCKED, &dev->flags); | |
2199 | set_bit(R5_Wantread, &dev->flags); | |
2200 | s->locked++; | |
45b4233c | 2201 | pr_debug("Reading block %d (sync=%d)\n", |
a4456856 DW |
2202 | i, s->syncing); |
2203 | } | |
2204 | } | |
2205 | } | |
2206 | set_bit(STRIPE_HANDLE, &sh->state); | |
2207 | } | |
2208 | ||
2209 | ||
1fe797e6 | 2210 | /* handle_stripe_clean_event |
a4456856 DW |
2211 | * any written block on an uptodate or failed drive can be returned. |
2212 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2213 | * never LOCKED, so we don't need to test 'failed' directly. | |
2214 | */ | |
1fe797e6 | 2215 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2216 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2217 | { | |
2218 | int i; | |
2219 | struct r5dev *dev; | |
2220 | ||
2221 | for (i = disks; i--; ) | |
2222 | if (sh->dev[i].written) { | |
2223 | dev = &sh->dev[i]; | |
2224 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2225 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2226 | /* We can return any write requests */ | |
2227 | struct bio *wbi, *wbi2; | |
2228 | int bitmap_end = 0; | |
45b4233c | 2229 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2230 | spin_lock_irq(&conf->device_lock); |
2231 | wbi = dev->written; | |
2232 | dev->written = NULL; | |
2233 | while (wbi && wbi->bi_sector < | |
2234 | dev->sector + STRIPE_SECTORS) { | |
2235 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2236 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2237 | md_write_end(conf->mddev); |
2238 | wbi->bi_next = *return_bi; | |
2239 | *return_bi = wbi; | |
2240 | } | |
2241 | wbi = wbi2; | |
2242 | } | |
2243 | if (dev->towrite == NULL) | |
2244 | bitmap_end = 1; | |
2245 | spin_unlock_irq(&conf->device_lock); | |
2246 | if (bitmap_end) | |
2247 | bitmap_endwrite(conf->mddev->bitmap, | |
2248 | sh->sector, | |
2249 | STRIPE_SECTORS, | |
2250 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2251 | 0); | |
2252 | } | |
2253 | } | |
8b3e6cdc DW |
2254 | |
2255 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2256 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2257 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2258 | } |
2259 | ||
1fe797e6 | 2260 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2261 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2262 | { | |
2263 | int rmw = 0, rcw = 0, i; | |
2264 | for (i = disks; i--; ) { | |
2265 | /* would I have to read this buffer for read_modify_write */ | |
2266 | struct r5dev *dev = &sh->dev[i]; | |
2267 | if ((dev->towrite || i == sh->pd_idx) && | |
2268 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2269 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2270 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2271 | if (test_bit(R5_Insync, &dev->flags)) |
2272 | rmw++; | |
2273 | else | |
2274 | rmw += 2*disks; /* cannot read it */ | |
2275 | } | |
2276 | /* Would I have to read this buffer for reconstruct_write */ | |
2277 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2278 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2279 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2280 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2281 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2282 | else |
2283 | rcw += 2*disks; | |
2284 | } | |
2285 | } | |
45b4233c | 2286 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2287 | (unsigned long long)sh->sector, rmw, rcw); |
2288 | set_bit(STRIPE_HANDLE, &sh->state); | |
2289 | if (rmw < rcw && rmw > 0) | |
2290 | /* prefer read-modify-write, but need to get some data */ | |
2291 | for (i = disks; i--; ) { | |
2292 | struct r5dev *dev = &sh->dev[i]; | |
2293 | if ((dev->towrite || i == sh->pd_idx) && | |
2294 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2295 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2296 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2297 | test_bit(R5_Insync, &dev->flags)) { |
2298 | if ( | |
2299 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2300 | pr_debug("Read_old block " |
a4456856 DW |
2301 | "%d for r-m-w\n", i); |
2302 | set_bit(R5_LOCKED, &dev->flags); | |
2303 | set_bit(R5_Wantread, &dev->flags); | |
2304 | s->locked++; | |
2305 | } else { | |
2306 | set_bit(STRIPE_DELAYED, &sh->state); | |
2307 | set_bit(STRIPE_HANDLE, &sh->state); | |
2308 | } | |
2309 | } | |
2310 | } | |
2311 | if (rcw <= rmw && rcw > 0) | |
2312 | /* want reconstruct write, but need to get some data */ | |
2313 | for (i = disks; i--; ) { | |
2314 | struct r5dev *dev = &sh->dev[i]; | |
2315 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2316 | i != sh->pd_idx && | |
2317 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2318 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2319 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2320 | test_bit(R5_Insync, &dev->flags)) { |
2321 | if ( | |
2322 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2323 | pr_debug("Read_old block " |
a4456856 DW |
2324 | "%d for Reconstruct\n", i); |
2325 | set_bit(R5_LOCKED, &dev->flags); | |
2326 | set_bit(R5_Wantread, &dev->flags); | |
2327 | s->locked++; | |
2328 | } else { | |
2329 | set_bit(STRIPE_DELAYED, &sh->state); | |
2330 | set_bit(STRIPE_HANDLE, &sh->state); | |
2331 | } | |
2332 | } | |
2333 | } | |
2334 | /* now if nothing is locked, and if we have enough data, | |
2335 | * we can start a write request | |
2336 | */ | |
f38e1219 DW |
2337 | /* since handle_stripe can be called at any time we need to handle the |
2338 | * case where a compute block operation has been submitted and then a | |
2339 | * subsequent call wants to start a write request. raid5_run_ops only | |
2340 | * handles the case where compute block and postxor are requested | |
2341 | * simultaneously. If this is not the case then new writes need to be | |
2342 | * held off until the compute completes. | |
2343 | */ | |
976ea8d4 DW |
2344 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2345 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2346 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
1fe797e6 | 2347 | schedule_reconstruction5(sh, s, rcw == 0, 0); |
a4456856 DW |
2348 | } |
2349 | ||
1fe797e6 | 2350 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2351 | struct stripe_head *sh, struct stripe_head_state *s, |
2352 | struct r6_state *r6s, int disks) | |
2353 | { | |
2354 | int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i; | |
2355 | int qd_idx = r6s->qd_idx; | |
2356 | for (i = disks; i--; ) { | |
2357 | struct r5dev *dev = &sh->dev[i]; | |
2358 | /* Would I have to read this buffer for reconstruct_write */ | |
2359 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2360 | && i != pd_idx && i != qd_idx | |
2361 | && (!test_bit(R5_LOCKED, &dev->flags) | |
2362 | ) && | |
2363 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
2364 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
2365 | else { | |
45b4233c | 2366 | pr_debug("raid6: must_compute: " |
a4456856 DW |
2367 | "disk %d flags=%#lx\n", i, dev->flags); |
2368 | must_compute++; | |
2369 | } | |
2370 | } | |
2371 | } | |
45b4233c | 2372 | pr_debug("for sector %llu, rcw=%d, must_compute=%d\n", |
a4456856 DW |
2373 | (unsigned long long)sh->sector, rcw, must_compute); |
2374 | set_bit(STRIPE_HANDLE, &sh->state); | |
2375 | ||
2376 | if (rcw > 0) | |
2377 | /* want reconstruct write, but need to get some data */ | |
2378 | for (i = disks; i--; ) { | |
2379 | struct r5dev *dev = &sh->dev[i]; | |
2380 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2381 | && !(s->failed == 0 && (i == pd_idx || i == qd_idx)) | |
2382 | && !test_bit(R5_LOCKED, &dev->flags) && | |
2383 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2384 | test_bit(R5_Insync, &dev->flags)) { | |
2385 | if ( | |
2386 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2387 | pr_debug("Read_old stripe %llu " |
a4456856 DW |
2388 | "block %d for Reconstruct\n", |
2389 | (unsigned long long)sh->sector, i); | |
2390 | set_bit(R5_LOCKED, &dev->flags); | |
2391 | set_bit(R5_Wantread, &dev->flags); | |
2392 | s->locked++; | |
2393 | } else { | |
45b4233c | 2394 | pr_debug("Request delayed stripe %llu " |
a4456856 DW |
2395 | "block %d for Reconstruct\n", |
2396 | (unsigned long long)sh->sector, i); | |
2397 | set_bit(STRIPE_DELAYED, &sh->state); | |
2398 | set_bit(STRIPE_HANDLE, &sh->state); | |
2399 | } | |
2400 | } | |
2401 | } | |
2402 | /* now if nothing is locked, and if we have enough data, we can start a | |
2403 | * write request | |
2404 | */ | |
2405 | if (s->locked == 0 && rcw == 0 && | |
2406 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
2407 | if (must_compute > 0) { | |
2408 | /* We have failed blocks and need to compute them */ | |
2409 | switch (s->failed) { | |
2410 | case 0: | |
2411 | BUG(); | |
2412 | case 1: | |
2413 | compute_block_1(sh, r6s->failed_num[0], 0); | |
2414 | break; | |
2415 | case 2: | |
2416 | compute_block_2(sh, r6s->failed_num[0], | |
2417 | r6s->failed_num[1]); | |
2418 | break; | |
2419 | default: /* This request should have been failed? */ | |
2420 | BUG(); | |
2421 | } | |
2422 | } | |
2423 | ||
45b4233c | 2424 | pr_debug("Computing parity for stripe %llu\n", |
a4456856 DW |
2425 | (unsigned long long)sh->sector); |
2426 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
2427 | /* now every locked buffer is ready to be written */ | |
2428 | for (i = disks; i--; ) | |
2429 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
45b4233c | 2430 | pr_debug("Writing stripe %llu block %d\n", |
a4456856 DW |
2431 | (unsigned long long)sh->sector, i); |
2432 | s->locked++; | |
2433 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
2434 | } | |
8b3e6cdc DW |
2435 | if (s->locked == disks) |
2436 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2437 | atomic_inc(&conf->pending_full_writes); | |
a4456856 DW |
2438 | /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */ |
2439 | set_bit(STRIPE_INSYNC, &sh->state); | |
2440 | ||
2441 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2442 | atomic_dec(&conf->preread_active_stripes); | |
2443 | if (atomic_read(&conf->preread_active_stripes) < | |
2444 | IO_THRESHOLD) | |
2445 | md_wakeup_thread(conf->mddev->thread); | |
2446 | } | |
2447 | } | |
2448 | } | |
2449 | ||
2450 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2451 | struct stripe_head_state *s, int disks) | |
2452 | { | |
ecc65c9b | 2453 | struct r5dev *dev = NULL; |
bd2ab670 | 2454 | |
a4456856 | 2455 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2456 | |
ecc65c9b DW |
2457 | switch (sh->check_state) { |
2458 | case check_state_idle: | |
2459 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2460 | if (s->failed == 0) { |
bd2ab670 | 2461 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2462 | sh->check_state = check_state_run; |
2463 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2464 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2465 | s->uptodate--; |
ecc65c9b | 2466 | break; |
bd2ab670 | 2467 | } |
ecc65c9b DW |
2468 | dev = &sh->dev[s->failed_num]; |
2469 | /* fall through */ | |
2470 | case check_state_compute_result: | |
2471 | sh->check_state = check_state_idle; | |
2472 | if (!dev) | |
2473 | dev = &sh->dev[sh->pd_idx]; | |
2474 | ||
2475 | /* check that a write has not made the stripe insync */ | |
2476 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2477 | break; | |
c8894419 | 2478 | |
a4456856 | 2479 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2480 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2481 | BUG_ON(s->uptodate != disks); | |
2482 | ||
2483 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2484 | s->locked++; |
a4456856 | 2485 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2486 | |
a4456856 | 2487 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2488 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2489 | break; |
2490 | case check_state_run: | |
2491 | break; /* we will be called again upon completion */ | |
2492 | case check_state_check_result: | |
2493 | sh->check_state = check_state_idle; | |
2494 | ||
2495 | /* if a failure occurred during the check operation, leave | |
2496 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2497 | */ | |
2498 | if (s->failed) | |
2499 | break; | |
2500 | ||
2501 | /* handle a successful check operation, if parity is correct | |
2502 | * we are done. Otherwise update the mismatch count and repair | |
2503 | * parity if !MD_RECOVERY_CHECK | |
2504 | */ | |
2505 | if (sh->ops.zero_sum_result == 0) | |
2506 | /* parity is correct (on disc, | |
2507 | * not in buffer any more) | |
2508 | */ | |
2509 | set_bit(STRIPE_INSYNC, &sh->state); | |
2510 | else { | |
2511 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2512 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2513 | /* don't try to repair!! */ | |
2514 | set_bit(STRIPE_INSYNC, &sh->state); | |
2515 | else { | |
2516 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2517 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2518 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2519 | set_bit(R5_Wantcompute, | |
2520 | &sh->dev[sh->pd_idx].flags); | |
2521 | sh->ops.target = sh->pd_idx; | |
2522 | s->uptodate++; | |
2523 | } | |
2524 | } | |
2525 | break; | |
2526 | case check_state_compute_run: | |
2527 | break; | |
2528 | default: | |
2529 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2530 | __func__, sh->check_state, | |
2531 | (unsigned long long) sh->sector); | |
2532 | BUG(); | |
a4456856 DW |
2533 | } |
2534 | } | |
2535 | ||
2536 | ||
2537 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
2538 | struct stripe_head_state *s, | |
2539 | struct r6_state *r6s, struct page *tmp_page, | |
2540 | int disks) | |
2541 | { | |
2542 | int update_p = 0, update_q = 0; | |
2543 | struct r5dev *dev; | |
2544 | int pd_idx = sh->pd_idx; | |
2545 | int qd_idx = r6s->qd_idx; | |
2546 | ||
2547 | set_bit(STRIPE_HANDLE, &sh->state); | |
2548 | ||
2549 | BUG_ON(s->failed > 2); | |
2550 | BUG_ON(s->uptodate < disks); | |
2551 | /* Want to check and possibly repair P and Q. | |
2552 | * However there could be one 'failed' device, in which | |
2553 | * case we can only check one of them, possibly using the | |
2554 | * other to generate missing data | |
2555 | */ | |
2556 | ||
2557 | /* If !tmp_page, we cannot do the calculations, | |
2558 | * but as we have set STRIPE_HANDLE, we will soon be called | |
2559 | * by stripe_handle with a tmp_page - just wait until then. | |
2560 | */ | |
2561 | if (tmp_page) { | |
2562 | if (s->failed == r6s->q_failed) { | |
2563 | /* The only possible failed device holds 'Q', so it | |
2564 | * makes sense to check P (If anything else were failed, | |
2565 | * we would have used P to recreate it). | |
2566 | */ | |
2567 | compute_block_1(sh, pd_idx, 1); | |
2568 | if (!page_is_zero(sh->dev[pd_idx].page)) { | |
2569 | compute_block_1(sh, pd_idx, 0); | |
2570 | update_p = 1; | |
2571 | } | |
2572 | } | |
2573 | if (!r6s->q_failed && s->failed < 2) { | |
2574 | /* q is not failed, and we didn't use it to generate | |
2575 | * anything, so it makes sense to check it | |
2576 | */ | |
2577 | memcpy(page_address(tmp_page), | |
2578 | page_address(sh->dev[qd_idx].page), | |
2579 | STRIPE_SIZE); | |
2580 | compute_parity6(sh, UPDATE_PARITY); | |
2581 | if (memcmp(page_address(tmp_page), | |
2582 | page_address(sh->dev[qd_idx].page), | |
2583 | STRIPE_SIZE) != 0) { | |
2584 | clear_bit(STRIPE_INSYNC, &sh->state); | |
2585 | update_q = 1; | |
2586 | } | |
2587 | } | |
2588 | if (update_p || update_q) { | |
2589 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2590 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2591 | /* don't try to repair!! */ | |
2592 | update_p = update_q = 0; | |
2593 | } | |
2594 | ||
2595 | /* now write out any block on a failed drive, | |
2596 | * or P or Q if they need it | |
2597 | */ | |
2598 | ||
2599 | if (s->failed == 2) { | |
2600 | dev = &sh->dev[r6s->failed_num[1]]; | |
2601 | s->locked++; | |
2602 | set_bit(R5_LOCKED, &dev->flags); | |
2603 | set_bit(R5_Wantwrite, &dev->flags); | |
2604 | } | |
2605 | if (s->failed >= 1) { | |
2606 | dev = &sh->dev[r6s->failed_num[0]]; | |
2607 | s->locked++; | |
2608 | set_bit(R5_LOCKED, &dev->flags); | |
2609 | set_bit(R5_Wantwrite, &dev->flags); | |
2610 | } | |
2611 | ||
2612 | if (update_p) { | |
2613 | dev = &sh->dev[pd_idx]; | |
2614 | s->locked++; | |
2615 | set_bit(R5_LOCKED, &dev->flags); | |
2616 | set_bit(R5_Wantwrite, &dev->flags); | |
2617 | } | |
2618 | if (update_q) { | |
2619 | dev = &sh->dev[qd_idx]; | |
2620 | s->locked++; | |
2621 | set_bit(R5_LOCKED, &dev->flags); | |
2622 | set_bit(R5_Wantwrite, &dev->flags); | |
2623 | } | |
2624 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2625 | ||
2626 | set_bit(STRIPE_INSYNC, &sh->state); | |
2627 | } | |
2628 | } | |
2629 | ||
2630 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2631 | struct r6_state *r6s) | |
2632 | { | |
2633 | int i; | |
2634 | ||
2635 | /* We have read all the blocks in this stripe and now we need to | |
2636 | * copy some of them into a target stripe for expand. | |
2637 | */ | |
f0a50d37 | 2638 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2639 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2640 | for (i = 0; i < sh->disks; i++) | |
a2e08551 | 2641 | if (i != sh->pd_idx && (!r6s || i != r6s->qd_idx)) { |
911d4ee8 | 2642 | int dd_idx, j; |
a4456856 DW |
2643 | struct stripe_head *sh2; |
2644 | ||
2645 | sector_t bn = compute_blocknr(sh, i); | |
911d4ee8 N |
2646 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2647 | &dd_idx, NULL); | |
b5663ba4 | 2648 | sh2 = get_active_stripe(conf, s, 0, 1); |
a4456856 DW |
2649 | if (sh2 == NULL) |
2650 | /* so far only the early blocks of this stripe | |
2651 | * have been requested. When later blocks | |
2652 | * get requested, we will try again | |
2653 | */ | |
2654 | continue; | |
2655 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2656 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2657 | /* must have already done this block */ | |
2658 | release_stripe(sh2); | |
2659 | continue; | |
2660 | } | |
f0a50d37 DW |
2661 | |
2662 | /* place all the copies on one channel */ | |
2663 | tx = async_memcpy(sh2->dev[dd_idx].page, | |
2664 | sh->dev[i].page, 0, 0, STRIPE_SIZE, | |
2665 | ASYNC_TX_DEP_ACK, tx, NULL, NULL); | |
2666 | ||
a4456856 DW |
2667 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2668 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2669 | for (j = 0; j < conf->raid_disks; j++) | |
2670 | if (j != sh2->pd_idx && | |
d0dabf7e | 2671 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
2672 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2673 | break; | |
2674 | if (j == conf->raid_disks) { | |
2675 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2676 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2677 | } | |
2678 | release_stripe(sh2); | |
f0a50d37 | 2679 | |
a4456856 | 2680 | } |
a2e08551 N |
2681 | /* done submitting copies, wait for them to complete */ |
2682 | if (tx) { | |
2683 | async_tx_ack(tx); | |
2684 | dma_wait_for_async_tx(tx); | |
2685 | } | |
a4456856 | 2686 | } |
1da177e4 | 2687 | |
6bfe0b49 | 2688 | |
1da177e4 LT |
2689 | /* |
2690 | * handle_stripe - do things to a stripe. | |
2691 | * | |
2692 | * We lock the stripe and then examine the state of various bits | |
2693 | * to see what needs to be done. | |
2694 | * Possible results: | |
2695 | * return some read request which now have data | |
2696 | * return some write requests which are safely on disc | |
2697 | * schedule a read on some buffers | |
2698 | * schedule a write of some buffers | |
2699 | * return confirmation of parity correctness | |
2700 | * | |
1da177e4 LT |
2701 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2702 | * get BH_Lock set before the stripe lock is released. | |
2703 | * | |
2704 | */ | |
a4456856 | 2705 | |
df10cfbc | 2706 | static bool handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2707 | { |
2708 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2709 | int disks = sh->disks, i; |
2710 | struct bio *return_bi = NULL; | |
2711 | struct stripe_head_state s; | |
1da177e4 | 2712 | struct r5dev *dev; |
6bfe0b49 | 2713 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2714 | int prexor; |
1da177e4 | 2715 | |
a4456856 | 2716 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
2717 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
2718 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
2719 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
2720 | sh->reconstruct_state); | |
1da177e4 LT |
2721 | |
2722 | spin_lock(&sh->lock); | |
2723 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2724 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2725 | ||
a4456856 DW |
2726 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2727 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2728 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 2729 | |
83de75cc | 2730 | /* Now to look around and see what can be done */ |
9910f16a | 2731 | rcu_read_lock(); |
1da177e4 LT |
2732 | for (i=disks; i--; ) { |
2733 | mdk_rdev_t *rdev; | |
a4456856 | 2734 | struct r5dev *dev = &sh->dev[i]; |
1da177e4 | 2735 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2736 | |
b5e98d65 DW |
2737 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2738 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2739 | dev->towrite, dev->written); | |
2740 | ||
2741 | /* maybe we can request a biofill operation | |
2742 | * | |
2743 | * new wantfill requests are only permitted while | |
83de75cc | 2744 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
2745 | */ |
2746 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 2747 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 2748 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
2749 | |
2750 | /* now count some things */ | |
a4456856 DW |
2751 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2752 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2753 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2754 | |
b5e98d65 DW |
2755 | if (test_bit(R5_Wantfill, &dev->flags)) |
2756 | s.to_fill++; | |
2757 | else if (dev->toread) | |
a4456856 | 2758 | s.to_read++; |
1da177e4 | 2759 | if (dev->towrite) { |
a4456856 | 2760 | s.to_write++; |
1da177e4 | 2761 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2762 | s.non_overwrite++; |
1da177e4 | 2763 | } |
a4456856 DW |
2764 | if (dev->written) |
2765 | s.written++; | |
9910f16a | 2766 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
2767 | if (blocked_rdev == NULL && |
2768 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
2769 | blocked_rdev = rdev; |
2770 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 2771 | } |
b2d444d7 | 2772 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 2773 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
2774 | clear_bit(R5_ReadError, &dev->flags); |
2775 | clear_bit(R5_ReWrite, &dev->flags); | |
2776 | } | |
b2d444d7 | 2777 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 2778 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
2779 | s.failed++; |
2780 | s.failed_num = i; | |
1da177e4 LT |
2781 | } else |
2782 | set_bit(R5_Insync, &dev->flags); | |
2783 | } | |
9910f16a | 2784 | rcu_read_unlock(); |
b5e98d65 | 2785 | |
6bfe0b49 | 2786 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
2787 | if (s.syncing || s.expanding || s.expanded || |
2788 | s.to_write || s.written) { | |
2789 | set_bit(STRIPE_HANDLE, &sh->state); | |
2790 | goto unlock; | |
2791 | } | |
2792 | /* There is nothing for the blocked_rdev to block */ | |
2793 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
2794 | blocked_rdev = NULL; | |
6bfe0b49 DW |
2795 | } |
2796 | ||
83de75cc DW |
2797 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
2798 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
2799 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
2800 | } | |
b5e98d65 | 2801 | |
45b4233c | 2802 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 2803 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
2804 | s.locked, s.uptodate, s.to_read, s.to_write, |
2805 | s.failed, s.failed_num); | |
1da177e4 LT |
2806 | /* check if the array has lost two devices and, if so, some requests might |
2807 | * need to be failed | |
2808 | */ | |
a4456856 | 2809 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 2810 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 2811 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
2812 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2813 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2814 | s.syncing = 0; |
1da177e4 LT |
2815 | } |
2816 | ||
2817 | /* might be able to return some write requests if the parity block | |
2818 | * is safe, or on a failed drive | |
2819 | */ | |
2820 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
2821 | if ( s.written && |
2822 | ((test_bit(R5_Insync, &dev->flags) && | |
2823 | !test_bit(R5_LOCKED, &dev->flags) && | |
2824 | test_bit(R5_UPTODATE, &dev->flags)) || | |
2825 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 2826 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
2827 | |
2828 | /* Now we might consider reading some blocks, either to check/generate | |
2829 | * parity, or to satisfy requests | |
2830 | * or to load a block that is being partially written. | |
2831 | */ | |
a4456856 | 2832 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 2833 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 2834 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 2835 | |
e33129d8 DW |
2836 | /* Now we check to see if any write operations have recently |
2837 | * completed | |
2838 | */ | |
e0a115e5 | 2839 | prexor = 0; |
d8ee0728 | 2840 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 2841 | prexor = 1; |
d8ee0728 DW |
2842 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
2843 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 2844 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
2845 | |
2846 | /* All the 'written' buffers and the parity block are ready to | |
2847 | * be written back to disk | |
2848 | */ | |
2849 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
2850 | for (i = disks; i--; ) { | |
2851 | dev = &sh->dev[i]; | |
2852 | if (test_bit(R5_LOCKED, &dev->flags) && | |
2853 | (i == sh->pd_idx || dev->written)) { | |
2854 | pr_debug("Writing block %d\n", i); | |
2855 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
2856 | if (prexor) |
2857 | continue; | |
e33129d8 DW |
2858 | if (!test_bit(R5_Insync, &dev->flags) || |
2859 | (i == sh->pd_idx && s.failed == 0)) | |
2860 | set_bit(STRIPE_INSYNC, &sh->state); | |
2861 | } | |
2862 | } | |
2863 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2864 | atomic_dec(&conf->preread_active_stripes); | |
2865 | if (atomic_read(&conf->preread_active_stripes) < | |
2866 | IO_THRESHOLD) | |
2867 | md_wakeup_thread(conf->mddev->thread); | |
2868 | } | |
2869 | } | |
2870 | ||
2871 | /* Now to consider new write requests and what else, if anything | |
2872 | * should be read. We do not handle new writes when: | |
2873 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
2874 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
2875 | * block. | |
2876 | */ | |
600aa109 | 2877 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 2878 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
2879 | |
2880 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
2881 | * Any reads will already have been scheduled, so we just see if enough |
2882 | * data is available. The parity check is held off while parity | |
2883 | * dependent operations are in flight. | |
1da177e4 | 2884 | */ |
ecc65c9b DW |
2885 | if (sh->check_state || |
2886 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 2887 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 2888 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 2889 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 2890 | |
a4456856 | 2891 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
2892 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2893 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2894 | } | |
4e5314b5 N |
2895 | |
2896 | /* If the failed drive is just a ReadError, then we might need to progress | |
2897 | * the repair/check process | |
2898 | */ | |
a4456856 DW |
2899 | if (s.failed == 1 && !conf->mddev->ro && |
2900 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
2901 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
2902 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 2903 | ) { |
a4456856 | 2904 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
2905 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
2906 | set_bit(R5_Wantwrite, &dev->flags); | |
2907 | set_bit(R5_ReWrite, &dev->flags); | |
2908 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2909 | s.locked++; |
4e5314b5 N |
2910 | } else { |
2911 | /* let's read it back */ | |
2912 | set_bit(R5_Wantread, &dev->flags); | |
2913 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2914 | s.locked++; |
4e5314b5 N |
2915 | } |
2916 | } | |
2917 | ||
600aa109 DW |
2918 | /* Finish reconstruct operations initiated by the expansion process */ |
2919 | if (sh->reconstruct_state == reconstruct_state_result) { | |
2920 | sh->reconstruct_state = reconstruct_state_idle; | |
f0a50d37 | 2921 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 2922 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 2923 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 2924 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 2925 | s.locked++; |
23397883 | 2926 | } |
f0a50d37 DW |
2927 | } |
2928 | ||
2929 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 2930 | !sh->reconstruct_state) { |
f0a50d37 DW |
2931 | /* Need to write out all blocks after computing parity */ |
2932 | sh->disks = conf->raid_disks; | |
911d4ee8 | 2933 | stripe_set_idx(sh->sector, conf, 0, sh); |
1fe797e6 | 2934 | schedule_reconstruction5(sh, &s, 1, 1); |
600aa109 | 2935 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 2936 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 2937 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
2938 | wake_up(&conf->wait_for_overlap); |
2939 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
2940 | } | |
2941 | ||
0f94e87c | 2942 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 2943 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 2944 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 2945 | |
6bfe0b49 | 2946 | unlock: |
1da177e4 LT |
2947 | spin_unlock(&sh->lock); |
2948 | ||
6bfe0b49 DW |
2949 | /* wait for this device to become unblocked */ |
2950 | if (unlikely(blocked_rdev)) | |
2951 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
2952 | ||
600aa109 DW |
2953 | if (s.ops_request) |
2954 | raid5_run_ops(sh, s.ops_request); | |
d84e0f10 | 2955 | |
c4e5ac0a | 2956 | ops_run_io(sh, &s); |
1da177e4 | 2957 | |
a4456856 | 2958 | return_io(return_bi); |
df10cfbc DW |
2959 | |
2960 | return blocked_rdev == NULL; | |
1da177e4 LT |
2961 | } |
2962 | ||
df10cfbc | 2963 | static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page) |
1da177e4 | 2964 | { |
bff61975 | 2965 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 2966 | int disks = sh->disks; |
a4456856 DW |
2967 | struct bio *return_bi = NULL; |
2968 | int i, pd_idx = sh->pd_idx; | |
2969 | struct stripe_head_state s; | |
2970 | struct r6_state r6s; | |
16a53ecc | 2971 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 2972 | mdk_rdev_t *blocked_rdev = NULL; |
1da177e4 | 2973 | |
d0dabf7e | 2974 | r6s.qd_idx = sh->qd_idx; |
45b4233c | 2975 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
a4456856 DW |
2976 | "pd_idx=%d, qd_idx=%d\n", |
2977 | (unsigned long long)sh->sector, sh->state, | |
2978 | atomic_read(&sh->count), pd_idx, r6s.qd_idx); | |
2979 | memset(&s, 0, sizeof(s)); | |
72626685 | 2980 | |
16a53ecc N |
2981 | spin_lock(&sh->lock); |
2982 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2983 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2984 | ||
a4456856 DW |
2985 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2986 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2987 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 2988 | /* Now to look around and see what can be done */ |
1da177e4 LT |
2989 | |
2990 | rcu_read_lock(); | |
16a53ecc N |
2991 | for (i=disks; i--; ) { |
2992 | mdk_rdev_t *rdev; | |
2993 | dev = &sh->dev[i]; | |
2994 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 2995 | |
45b4233c | 2996 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc N |
2997 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
2998 | /* maybe we can reply to a read */ | |
2999 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
3000 | struct bio *rbi, *rbi2; | |
45b4233c | 3001 | pr_debug("Return read for disc %d\n", i); |
16a53ecc N |
3002 | spin_lock_irq(&conf->device_lock); |
3003 | rbi = dev->toread; | |
3004 | dev->toread = NULL; | |
3005 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
3006 | wake_up(&conf->wait_for_overlap); | |
3007 | spin_unlock_irq(&conf->device_lock); | |
3008 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
3009 | copy_data(0, rbi, dev->page, dev->sector); | |
3010 | rbi2 = r5_next_bio(rbi, dev->sector); | |
3011 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3012 | if (!raid5_dec_bi_phys_segments(rbi)) { |
16a53ecc N |
3013 | rbi->bi_next = return_bi; |
3014 | return_bi = rbi; | |
3015 | } | |
3016 | spin_unlock_irq(&conf->device_lock); | |
3017 | rbi = rbi2; | |
3018 | } | |
3019 | } | |
1da177e4 | 3020 | |
16a53ecc | 3021 | /* now count some things */ |
a4456856 DW |
3022 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3023 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 3024 | |
16a53ecc | 3025 | |
a4456856 DW |
3026 | if (dev->toread) |
3027 | s.to_read++; | |
16a53ecc | 3028 | if (dev->towrite) { |
a4456856 | 3029 | s.to_write++; |
16a53ecc | 3030 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3031 | s.non_overwrite++; |
16a53ecc | 3032 | } |
a4456856 DW |
3033 | if (dev->written) |
3034 | s.written++; | |
16a53ecc | 3035 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3036 | if (blocked_rdev == NULL && |
3037 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3038 | blocked_rdev = rdev; |
3039 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3040 | } |
16a53ecc N |
3041 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3042 | /* The ReadError flag will just be confusing now */ | |
3043 | clear_bit(R5_ReadError, &dev->flags); | |
3044 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3045 | } |
16a53ecc N |
3046 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3047 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3048 | if (s.failed < 2) |
3049 | r6s.failed_num[s.failed] = i; | |
3050 | s.failed++; | |
16a53ecc N |
3051 | } else |
3052 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3053 | } |
3054 | rcu_read_unlock(); | |
6bfe0b49 DW |
3055 | |
3056 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3057 | if (s.syncing || s.expanding || s.expanded || |
3058 | s.to_write || s.written) { | |
3059 | set_bit(STRIPE_HANDLE, &sh->state); | |
3060 | goto unlock; | |
3061 | } | |
3062 | /* There is nothing for the blocked_rdev to block */ | |
3063 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3064 | blocked_rdev = NULL; | |
6bfe0b49 | 3065 | } |
ac4090d2 | 3066 | |
45b4233c | 3067 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3068 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3069 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3070 | r6s.failed_num[0], r6s.failed_num[1]); | |
3071 | /* check if the array has lost >2 devices and, if so, some requests | |
3072 | * might need to be failed | |
16a53ecc | 3073 | */ |
a4456856 | 3074 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3075 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3076 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3077 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3078 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3079 | s.syncing = 0; |
16a53ecc N |
3080 | } |
3081 | ||
3082 | /* | |
3083 | * might be able to return some write requests if the parity blocks | |
3084 | * are safe, or on a failed drive | |
3085 | */ | |
3086 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3087 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3088 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
3089 | qdev = &sh->dev[r6s.qd_idx]; | |
3090 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == r6s.qd_idx) | |
3091 | || (s.failed >= 2 && r6s.failed_num[1] == r6s.qd_idx); | |
3092 | ||
3093 | if ( s.written && | |
3094 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3095 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3096 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3097 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3098 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3099 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3100 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3101 | |
3102 | /* Now we might consider reading some blocks, either to check/generate | |
3103 | * parity, or to satisfy requests | |
3104 | * or to load a block that is being partially written. | |
3105 | */ | |
a4456856 DW |
3106 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
3107 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
1fe797e6 | 3108 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc N |
3109 | |
3110 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 | 3111 | if (s.to_write) |
1fe797e6 | 3112 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3113 | |
3114 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 DW |
3115 | * Any reads will already have been scheduled, so we just see if enough |
3116 | * data is available | |
16a53ecc | 3117 | */ |
a4456856 DW |
3118 | if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state)) |
3119 | handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks); | |
16a53ecc | 3120 | |
a4456856 | 3121 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3122 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3123 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3124 | } | |
3125 | ||
3126 | /* If the failed drives are just a ReadError, then we might need | |
3127 | * to progress the repair/check process | |
3128 | */ | |
a4456856 DW |
3129 | if (s.failed <= 2 && !conf->mddev->ro) |
3130 | for (i = 0; i < s.failed; i++) { | |
3131 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3132 | if (test_bit(R5_ReadError, &dev->flags) |
3133 | && !test_bit(R5_LOCKED, &dev->flags) | |
3134 | && test_bit(R5_UPTODATE, &dev->flags) | |
3135 | ) { | |
3136 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3137 | set_bit(R5_Wantwrite, &dev->flags); | |
3138 | set_bit(R5_ReWrite, &dev->flags); | |
3139 | set_bit(R5_LOCKED, &dev->flags); | |
3140 | } else { | |
3141 | /* let's read it back */ | |
3142 | set_bit(R5_Wantread, &dev->flags); | |
3143 | set_bit(R5_LOCKED, &dev->flags); | |
3144 | } | |
3145 | } | |
3146 | } | |
f416885e | 3147 | |
a4456856 | 3148 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
f416885e N |
3149 | /* Need to write out all blocks after computing P&Q */ |
3150 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3151 | stripe_set_idx(sh->sector, conf, 0, sh); |
f416885e N |
3152 | compute_parity6(sh, RECONSTRUCT_WRITE); |
3153 | for (i = conf->raid_disks ; i-- ; ) { | |
3154 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3155 | s.locked++; |
f416885e N |
3156 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
3157 | } | |
3158 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 3159 | } else if (s.expanded) { |
f416885e N |
3160 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3161 | atomic_dec(&conf->reshape_stripes); | |
3162 | wake_up(&conf->wait_for_overlap); | |
3163 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3164 | } | |
3165 | ||
0f94e87c | 3166 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3167 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3168 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3169 | |
6bfe0b49 | 3170 | unlock: |
16a53ecc N |
3171 | spin_unlock(&sh->lock); |
3172 | ||
6bfe0b49 DW |
3173 | /* wait for this device to become unblocked */ |
3174 | if (unlikely(blocked_rdev)) | |
3175 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3176 | ||
f0e43bcd | 3177 | ops_run_io(sh, &s); |
16a53ecc | 3178 | |
f0e43bcd | 3179 | return_io(return_bi); |
df10cfbc DW |
3180 | |
3181 | return blocked_rdev == NULL; | |
16a53ecc N |
3182 | } |
3183 | ||
df10cfbc DW |
3184 | /* returns true if the stripe was handled */ |
3185 | static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page) | |
16a53ecc N |
3186 | { |
3187 | if (sh->raid_conf->level == 6) | |
df10cfbc | 3188 | return handle_stripe6(sh, tmp_page); |
16a53ecc | 3189 | else |
df10cfbc | 3190 | return handle_stripe5(sh); |
16a53ecc N |
3191 | } |
3192 | ||
3193 | ||
3194 | ||
3195 | static void raid5_activate_delayed(raid5_conf_t *conf) | |
3196 | { | |
3197 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3198 | while (!list_empty(&conf->delayed_list)) { | |
3199 | struct list_head *l = conf->delayed_list.next; | |
3200 | struct stripe_head *sh; | |
3201 | sh = list_entry(l, struct stripe_head, lru); | |
3202 | list_del_init(l); | |
3203 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3204 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3205 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3206 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3207 | } |
6ed3003c N |
3208 | } else |
3209 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3210 | } |
3211 | ||
3212 | static void activate_bit_delay(raid5_conf_t *conf) | |
3213 | { | |
3214 | /* device_lock is held */ | |
3215 | struct list_head head; | |
3216 | list_add(&head, &conf->bitmap_list); | |
3217 | list_del_init(&conf->bitmap_list); | |
3218 | while (!list_empty(&head)) { | |
3219 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3220 | list_del_init(&sh->lru); | |
3221 | atomic_inc(&sh->count); | |
3222 | __release_stripe(conf, sh); | |
3223 | } | |
3224 | } | |
3225 | ||
3226 | static void unplug_slaves(mddev_t *mddev) | |
3227 | { | |
3228 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3229 | int i; | |
3230 | ||
3231 | rcu_read_lock(); | |
3232 | for (i=0; i<mddev->raid_disks; i++) { | |
3233 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
3234 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3235 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3236 | |
3237 | atomic_inc(&rdev->nr_pending); | |
3238 | rcu_read_unlock(); | |
3239 | ||
2ad8b1ef | 3240 | blk_unplug(r_queue); |
16a53ecc N |
3241 | |
3242 | rdev_dec_pending(rdev, mddev); | |
3243 | rcu_read_lock(); | |
3244 | } | |
3245 | } | |
3246 | rcu_read_unlock(); | |
3247 | } | |
3248 | ||
165125e1 | 3249 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3250 | { |
3251 | mddev_t *mddev = q->queuedata; | |
3252 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3253 | unsigned long flags; | |
3254 | ||
3255 | spin_lock_irqsave(&conf->device_lock, flags); | |
3256 | ||
3257 | if (blk_remove_plug(q)) { | |
3258 | conf->seq_flush++; | |
3259 | raid5_activate_delayed(conf); | |
72626685 | 3260 | } |
1da177e4 LT |
3261 | md_wakeup_thread(mddev->thread); |
3262 | ||
3263 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3264 | ||
3265 | unplug_slaves(mddev); | |
3266 | } | |
3267 | ||
f022b2fd N |
3268 | static int raid5_congested(void *data, int bits) |
3269 | { | |
3270 | mddev_t *mddev = data; | |
3271 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3272 | ||
3273 | /* No difference between reads and writes. Just check | |
3274 | * how busy the stripe_cache is | |
3275 | */ | |
3276 | if (conf->inactive_blocked) | |
3277 | return 1; | |
3278 | if (conf->quiesce) | |
3279 | return 1; | |
3280 | if (list_empty_careful(&conf->inactive_list)) | |
3281 | return 1; | |
3282 | ||
3283 | return 0; | |
3284 | } | |
3285 | ||
23032a0e RBJ |
3286 | /* We want read requests to align with chunks where possible, |
3287 | * but write requests don't need to. | |
3288 | */ | |
cc371e66 AK |
3289 | static int raid5_mergeable_bvec(struct request_queue *q, |
3290 | struct bvec_merge_data *bvm, | |
3291 | struct bio_vec *biovec) | |
23032a0e RBJ |
3292 | { |
3293 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3294 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e RBJ |
3295 | int max; |
3296 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
cc371e66 | 3297 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3298 | |
cc371e66 | 3299 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3300 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3301 | ||
3302 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
3303 | if (max < 0) max = 0; | |
3304 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3305 | return biovec->bv_len; | |
3306 | else | |
3307 | return max; | |
3308 | } | |
3309 | ||
f679623f RBJ |
3310 | |
3311 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3312 | { | |
3313 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
3314 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
3315 | unsigned int bio_sectors = bio->bi_size >> 9; | |
3316 | ||
3317 | return chunk_sectors >= | |
3318 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3319 | } | |
3320 | ||
46031f9a RBJ |
3321 | /* |
3322 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3323 | * later sampled by raid5d. | |
3324 | */ | |
3325 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3326 | { | |
3327 | unsigned long flags; | |
3328 | ||
3329 | spin_lock_irqsave(&conf->device_lock, flags); | |
3330 | ||
3331 | bi->bi_next = conf->retry_read_aligned_list; | |
3332 | conf->retry_read_aligned_list = bi; | |
3333 | ||
3334 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3335 | md_wakeup_thread(conf->mddev->thread); | |
3336 | } | |
3337 | ||
3338 | ||
3339 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3340 | { | |
3341 | struct bio *bi; | |
3342 | ||
3343 | bi = conf->retry_read_aligned; | |
3344 | if (bi) { | |
3345 | conf->retry_read_aligned = NULL; | |
3346 | return bi; | |
3347 | } | |
3348 | bi = conf->retry_read_aligned_list; | |
3349 | if(bi) { | |
387bb173 | 3350 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3351 | bi->bi_next = NULL; |
960e739d JA |
3352 | /* |
3353 | * this sets the active strip count to 1 and the processed | |
3354 | * strip count to zero (upper 8 bits) | |
3355 | */ | |
46031f9a | 3356 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3357 | } |
3358 | ||
3359 | return bi; | |
3360 | } | |
3361 | ||
3362 | ||
f679623f RBJ |
3363 | /* |
3364 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3365 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3366 | * first). | |
3367 | * If the read failed.. | |
3368 | */ | |
6712ecf8 | 3369 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3370 | { |
3371 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3372 | mddev_t *mddev; |
3373 | raid5_conf_t *conf; | |
3374 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3375 | mdk_rdev_t *rdev; | |
3376 | ||
f679623f | 3377 | bio_put(bi); |
46031f9a RBJ |
3378 | |
3379 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
3380 | conf = mddev_to_conf(mddev); | |
3381 | rdev = (void*)raid_bi->bi_next; | |
3382 | raid_bi->bi_next = NULL; | |
3383 | ||
3384 | rdev_dec_pending(rdev, conf->mddev); | |
3385 | ||
3386 | if (!error && uptodate) { | |
6712ecf8 | 3387 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3388 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3389 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3390 | return; |
46031f9a RBJ |
3391 | } |
3392 | ||
3393 | ||
45b4233c | 3394 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3395 | |
3396 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3397 | } |
3398 | ||
387bb173 NB |
3399 | static int bio_fits_rdev(struct bio *bi) |
3400 | { | |
165125e1 | 3401 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 NB |
3402 | |
3403 | if ((bi->bi_size>>9) > q->max_sectors) | |
3404 | return 0; | |
3405 | blk_recount_segments(q, bi); | |
960e739d | 3406 | if (bi->bi_phys_segments > q->max_phys_segments) |
387bb173 NB |
3407 | return 0; |
3408 | ||
3409 | if (q->merge_bvec_fn) | |
3410 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3411 | * just just give up | |
3412 | */ | |
3413 | return 0; | |
3414 | ||
3415 | return 1; | |
3416 | } | |
3417 | ||
3418 | ||
165125e1 | 3419 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3420 | { |
3421 | mddev_t *mddev = q->queuedata; | |
3422 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3423 | unsigned int dd_idx; |
f679623f RBJ |
3424 | struct bio* align_bi; |
3425 | mdk_rdev_t *rdev; | |
3426 | ||
3427 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3428 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3429 | return 0; |
3430 | } | |
3431 | /* | |
99c0fb5f | 3432 | * use bio_clone to make a copy of the bio |
f679623f RBJ |
3433 | */ |
3434 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3435 | if (!align_bi) | |
3436 | return 0; | |
3437 | /* | |
3438 | * set bi_end_io to a new function, and set bi_private to the | |
3439 | * original bio. | |
3440 | */ | |
3441 | align_bi->bi_end_io = raid5_align_endio; | |
3442 | align_bi->bi_private = raid_bio; | |
3443 | /* | |
3444 | * compute position | |
3445 | */ | |
112bf897 N |
3446 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3447 | 0, | |
911d4ee8 | 3448 | &dd_idx, NULL); |
f679623f RBJ |
3449 | |
3450 | rcu_read_lock(); | |
3451 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3452 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3453 | atomic_inc(&rdev->nr_pending); |
3454 | rcu_read_unlock(); | |
46031f9a RBJ |
3455 | raid_bio->bi_next = (void*)rdev; |
3456 | align_bi->bi_bdev = rdev->bdev; | |
3457 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3458 | align_bi->bi_sector += rdev->data_offset; | |
3459 | ||
387bb173 NB |
3460 | if (!bio_fits_rdev(align_bi)) { |
3461 | /* too big in some way */ | |
3462 | bio_put(align_bi); | |
3463 | rdev_dec_pending(rdev, mddev); | |
3464 | return 0; | |
3465 | } | |
3466 | ||
46031f9a RBJ |
3467 | spin_lock_irq(&conf->device_lock); |
3468 | wait_event_lock_irq(conf->wait_for_stripe, | |
3469 | conf->quiesce == 0, | |
3470 | conf->device_lock, /* nothing */); | |
3471 | atomic_inc(&conf->active_aligned_reads); | |
3472 | spin_unlock_irq(&conf->device_lock); | |
3473 | ||
f679623f RBJ |
3474 | generic_make_request(align_bi); |
3475 | return 1; | |
3476 | } else { | |
3477 | rcu_read_unlock(); | |
46031f9a | 3478 | bio_put(align_bi); |
f679623f RBJ |
3479 | return 0; |
3480 | } | |
3481 | } | |
3482 | ||
8b3e6cdc DW |
3483 | /* __get_priority_stripe - get the next stripe to process |
3484 | * | |
3485 | * Full stripe writes are allowed to pass preread active stripes up until | |
3486 | * the bypass_threshold is exceeded. In general the bypass_count | |
3487 | * increments when the handle_list is handled before the hold_list; however, it | |
3488 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3489 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3490 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3491 | * handle_list. | |
3492 | */ | |
3493 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3494 | { | |
3495 | struct stripe_head *sh; | |
3496 | ||
3497 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3498 | __func__, | |
3499 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3500 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3501 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3502 | ||
3503 | if (!list_empty(&conf->handle_list)) { | |
3504 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3505 | ||
3506 | if (list_empty(&conf->hold_list)) | |
3507 | conf->bypass_count = 0; | |
3508 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3509 | if (conf->hold_list.next == conf->last_hold) | |
3510 | conf->bypass_count++; | |
3511 | else { | |
3512 | conf->last_hold = conf->hold_list.next; | |
3513 | conf->bypass_count -= conf->bypass_threshold; | |
3514 | if (conf->bypass_count < 0) | |
3515 | conf->bypass_count = 0; | |
3516 | } | |
3517 | } | |
3518 | } else if (!list_empty(&conf->hold_list) && | |
3519 | ((conf->bypass_threshold && | |
3520 | conf->bypass_count > conf->bypass_threshold) || | |
3521 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3522 | sh = list_entry(conf->hold_list.next, | |
3523 | typeof(*sh), lru); | |
3524 | conf->bypass_count -= conf->bypass_threshold; | |
3525 | if (conf->bypass_count < 0) | |
3526 | conf->bypass_count = 0; | |
3527 | } else | |
3528 | return NULL; | |
3529 | ||
3530 | list_del_init(&sh->lru); | |
3531 | atomic_inc(&sh->count); | |
3532 | BUG_ON(atomic_read(&sh->count) != 1); | |
3533 | return sh; | |
3534 | } | |
f679623f | 3535 | |
165125e1 | 3536 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3537 | { |
3538 | mddev_t *mddev = q->queuedata; | |
3539 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3540 | int dd_idx; |
1da177e4 LT |
3541 | sector_t new_sector; |
3542 | sector_t logical_sector, last_sector; | |
3543 | struct stripe_head *sh; | |
a362357b | 3544 | const int rw = bio_data_dir(bi); |
c9959059 | 3545 | int cpu, remaining; |
1da177e4 | 3546 | |
e5dcdd80 | 3547 | if (unlikely(bio_barrier(bi))) { |
6712ecf8 | 3548 | bio_endio(bi, -EOPNOTSUPP); |
e5dcdd80 N |
3549 | return 0; |
3550 | } | |
3551 | ||
3d310eb7 | 3552 | md_write_start(mddev, bi); |
06d91a5f | 3553 | |
074a7aca TH |
3554 | cpu = part_stat_lock(); |
3555 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
3556 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
3557 | bio_sectors(bi)); | |
3558 | part_stat_unlock(); | |
1da177e4 | 3559 | |
802ba064 | 3560 | if (rw == READ && |
52488615 RBJ |
3561 | mddev->reshape_position == MaxSector && |
3562 | chunk_aligned_read(q,bi)) | |
99c0fb5f | 3563 | return 0; |
52488615 | 3564 | |
1da177e4 LT |
3565 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3566 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3567 | bi->bi_next = NULL; | |
3568 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3569 | |
1da177e4 LT |
3570 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3571 | DEFINE_WAIT(w); | |
16a53ecc | 3572 | int disks, data_disks; |
b5663ba4 | 3573 | int previous; |
b578d55f | 3574 | |
7ecaa1e6 | 3575 | retry: |
b5663ba4 | 3576 | previous = 0; |
b578d55f | 3577 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
7ecaa1e6 N |
3578 | if (likely(conf->expand_progress == MaxSector)) |
3579 | disks = conf->raid_disks; | |
3580 | else { | |
df8e7f76 N |
3581 | /* spinlock is needed as expand_progress may be |
3582 | * 64bit on a 32bit platform, and so it might be | |
3583 | * possible to see a half-updated value | |
3584 | * Ofcourse expand_progress could change after | |
3585 | * the lock is dropped, so once we get a reference | |
3586 | * to the stripe that we think it is, we will have | |
3587 | * to check again. | |
3588 | */ | |
7ecaa1e6 N |
3589 | spin_lock_irq(&conf->device_lock); |
3590 | disks = conf->raid_disks; | |
b5663ba4 | 3591 | if (logical_sector >= conf->expand_progress) { |
7ecaa1e6 | 3592 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3593 | previous = 1; |
3594 | } else { | |
b578d55f N |
3595 | if (logical_sector >= conf->expand_lo) { |
3596 | spin_unlock_irq(&conf->device_lock); | |
3597 | schedule(); | |
3598 | goto retry; | |
3599 | } | |
3600 | } | |
7ecaa1e6 N |
3601 | spin_unlock_irq(&conf->device_lock); |
3602 | } | |
16a53ecc N |
3603 | data_disks = disks - conf->max_degraded; |
3604 | ||
112bf897 N |
3605 | new_sector = raid5_compute_sector(conf, logical_sector, |
3606 | previous, | |
911d4ee8 | 3607 | &dd_idx, NULL); |
45b4233c | 3608 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3609 | (unsigned long long)new_sector, |
3610 | (unsigned long long)logical_sector); | |
3611 | ||
b5663ba4 N |
3612 | sh = get_active_stripe(conf, new_sector, previous, |
3613 | (bi->bi_rw&RWA_MASK)); | |
1da177e4 | 3614 | if (sh) { |
7ecaa1e6 N |
3615 | if (unlikely(conf->expand_progress != MaxSector)) { |
3616 | /* expansion might have moved on while waiting for a | |
df8e7f76 N |
3617 | * stripe, so we must do the range check again. |
3618 | * Expansion could still move past after this | |
3619 | * test, but as we are holding a reference to | |
3620 | * 'sh', we know that if that happens, | |
3621 | * STRIPE_EXPANDING will get set and the expansion | |
3622 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3623 | */ |
3624 | int must_retry = 0; | |
3625 | spin_lock_irq(&conf->device_lock); | |
3626 | if (logical_sector < conf->expand_progress && | |
3627 | disks == conf->previous_raid_disks) | |
3628 | /* mismatch, need to try again */ | |
3629 | must_retry = 1; | |
3630 | spin_unlock_irq(&conf->device_lock); | |
3631 | if (must_retry) { | |
3632 | release_stripe(sh); | |
3633 | goto retry; | |
3634 | } | |
3635 | } | |
e464eafd N |
3636 | /* FIXME what if we get a false positive because these |
3637 | * are being updated. | |
3638 | */ | |
3639 | if (logical_sector >= mddev->suspend_lo && | |
3640 | logical_sector < mddev->suspend_hi) { | |
3641 | release_stripe(sh); | |
3642 | schedule(); | |
3643 | goto retry; | |
3644 | } | |
7ecaa1e6 N |
3645 | |
3646 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
3647 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
3648 | /* Stripe is busy expanding or | |
3649 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3650 | * and wait a while |
3651 | */ | |
3652 | raid5_unplug_device(mddev->queue); | |
3653 | release_stripe(sh); | |
3654 | schedule(); | |
3655 | goto retry; | |
3656 | } | |
3657 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3658 | set_bit(STRIPE_HANDLE, &sh->state); |
3659 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1da177e4 | 3660 | release_stripe(sh); |
1da177e4 LT |
3661 | } else { |
3662 | /* cannot get stripe for read-ahead, just give-up */ | |
3663 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3664 | finish_wait(&conf->wait_for_overlap, &w); | |
3665 | break; | |
3666 | } | |
3667 | ||
3668 | } | |
3669 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3670 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3671 | spin_unlock_irq(&conf->device_lock); |
3672 | if (remaining == 0) { | |
1da177e4 | 3673 | |
16a53ecc | 3674 | if ( rw == WRITE ) |
1da177e4 | 3675 | md_write_end(mddev); |
6712ecf8 | 3676 | |
0e13fe23 | 3677 | bio_endio(bi, 0); |
1da177e4 | 3678 | } |
1da177e4 LT |
3679 | return 0; |
3680 | } | |
3681 | ||
52c03291 | 3682 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3683 | { |
52c03291 N |
3684 | /* reshaping is quite different to recovery/resync so it is |
3685 | * handled quite separately ... here. | |
3686 | * | |
3687 | * On each call to sync_request, we gather one chunk worth of | |
3688 | * destination stripes and flag them as expanding. | |
3689 | * Then we find all the source stripes and request reads. | |
3690 | * As the reads complete, handle_stripe will copy the data | |
3691 | * into the destination stripe and release that stripe. | |
3692 | */ | |
1da177e4 LT |
3693 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
3694 | struct stripe_head *sh; | |
ccfcc3c1 | 3695 | sector_t first_sector, last_sector; |
f416885e N |
3696 | int raid_disks = conf->previous_raid_disks; |
3697 | int data_disks = raid_disks - conf->max_degraded; | |
3698 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3699 | int i; |
3700 | int dd_idx; | |
3701 | sector_t writepos, safepos, gap; | |
3702 | ||
3703 | if (sector_nr == 0 && | |
3704 | conf->expand_progress != 0) { | |
3705 | /* restarting in the middle, skip the initial sectors */ | |
3706 | sector_nr = conf->expand_progress; | |
f416885e | 3707 | sector_div(sector_nr, new_data_disks); |
52c03291 N |
3708 | *skipped = 1; |
3709 | return sector_nr; | |
3710 | } | |
3711 | ||
3712 | /* we update the metadata when there is more than 3Meg | |
3713 | * in the block range (that is rather arbitrary, should | |
3714 | * probably be time based) or when the data about to be | |
3715 | * copied would over-write the source of the data at | |
3716 | * the front of the range. | |
3717 | * i.e. one new_stripe forward from expand_progress new_maps | |
3718 | * to after where expand_lo old_maps to | |
3719 | */ | |
3720 | writepos = conf->expand_progress + | |
f416885e N |
3721 | conf->chunk_size/512*(new_data_disks); |
3722 | sector_div(writepos, new_data_disks); | |
52c03291 | 3723 | safepos = conf->expand_lo; |
f416885e | 3724 | sector_div(safepos, data_disks); |
52c03291 N |
3725 | gap = conf->expand_progress - conf->expand_lo; |
3726 | ||
3727 | if (writepos >= safepos || | |
f416885e | 3728 | gap > (new_data_disks)*3000*2 /*3Meg*/) { |
52c03291 N |
3729 | /* Cannot proceed until we've updated the superblock... */ |
3730 | wait_event(conf->wait_for_overlap, | |
3731 | atomic_read(&conf->reshape_stripes)==0); | |
3732 | mddev->reshape_position = conf->expand_progress; | |
850b2b42 | 3733 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 3734 | md_wakeup_thread(mddev->thread); |
850b2b42 | 3735 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
3736 | kthread_should_stop()); |
3737 | spin_lock_irq(&conf->device_lock); | |
3738 | conf->expand_lo = mddev->reshape_position; | |
3739 | spin_unlock_irq(&conf->device_lock); | |
3740 | wake_up(&conf->wait_for_overlap); | |
3741 | } | |
3742 | ||
3743 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { | |
3744 | int j; | |
3745 | int skipped = 0; | |
b5663ba4 | 3746 | sh = get_active_stripe(conf, sector_nr+i, 0, 0); |
52c03291 N |
3747 | set_bit(STRIPE_EXPANDING, &sh->state); |
3748 | atomic_inc(&conf->reshape_stripes); | |
3749 | /* If any of this stripe is beyond the end of the old | |
3750 | * array, then we need to zero those blocks | |
3751 | */ | |
3752 | for (j=sh->disks; j--;) { | |
3753 | sector_t s; | |
3754 | if (j == sh->pd_idx) | |
3755 | continue; | |
f416885e | 3756 | if (conf->level == 6 && |
d0dabf7e | 3757 | j == sh->qd_idx) |
f416885e | 3758 | continue; |
52c03291 | 3759 | s = compute_blocknr(sh, j); |
f233ea5c | 3760 | if (s < mddev->array_sectors) { |
52c03291 N |
3761 | skipped = 1; |
3762 | continue; | |
3763 | } | |
3764 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
3765 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
3766 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
3767 | } | |
3768 | if (!skipped) { | |
3769 | set_bit(STRIPE_EXPAND_READY, &sh->state); | |
3770 | set_bit(STRIPE_HANDLE, &sh->state); | |
3771 | } | |
3772 | release_stripe(sh); | |
3773 | } | |
3774 | spin_lock_irq(&conf->device_lock); | |
6d3baf2e | 3775 | conf->expand_progress = (sector_nr + i) * new_data_disks; |
52c03291 N |
3776 | spin_unlock_irq(&conf->device_lock); |
3777 | /* Ok, those stripe are ready. We can start scheduling | |
3778 | * reads on the source stripes. | |
3779 | * The source stripes are determined by mapping the first and last | |
3780 | * block on the destination stripes. | |
3781 | */ | |
52c03291 | 3782 | first_sector = |
112bf897 | 3783 | raid5_compute_sector(conf, sector_nr*(new_data_disks), |
911d4ee8 | 3784 | 1, &dd_idx, NULL); |
52c03291 | 3785 | last_sector = |
112bf897 N |
3786 | raid5_compute_sector(conf, ((sector_nr+conf->chunk_size/512) |
3787 | *(new_data_disks) - 1), | |
911d4ee8 | 3788 | 1, &dd_idx, NULL); |
58c0fed4 AN |
3789 | if (last_sector >= mddev->dev_sectors) |
3790 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 3791 | while (first_sector <= last_sector) { |
b5663ba4 | 3792 | sh = get_active_stripe(conf, first_sector, 1, 0); |
52c03291 N |
3793 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3794 | set_bit(STRIPE_HANDLE, &sh->state); | |
3795 | release_stripe(sh); | |
3796 | first_sector += STRIPE_SECTORS; | |
3797 | } | |
c6207277 N |
3798 | /* If this takes us to the resync_max point where we have to pause, |
3799 | * then we need to write out the superblock. | |
3800 | */ | |
3801 | sector_nr += conf->chunk_size>>9; | |
3802 | if (sector_nr >= mddev->resync_max) { | |
3803 | /* Cannot proceed until we've updated the superblock... */ | |
3804 | wait_event(conf->wait_for_overlap, | |
3805 | atomic_read(&conf->reshape_stripes) == 0); | |
3806 | mddev->reshape_position = conf->expand_progress; | |
3807 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3808 | md_wakeup_thread(mddev->thread); | |
3809 | wait_event(mddev->sb_wait, | |
3810 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
3811 | || kthread_should_stop()); | |
3812 | spin_lock_irq(&conf->device_lock); | |
3813 | conf->expand_lo = mddev->reshape_position; | |
3814 | spin_unlock_irq(&conf->device_lock); | |
3815 | wake_up(&conf->wait_for_overlap); | |
3816 | } | |
52c03291 N |
3817 | return conf->chunk_size>>9; |
3818 | } | |
3819 | ||
3820 | /* FIXME go_faster isn't used */ | |
3821 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
3822 | { | |
3823 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3824 | struct stripe_head *sh; | |
58c0fed4 | 3825 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 3826 | int sync_blocks; |
16a53ecc N |
3827 | int still_degraded = 0; |
3828 | int i; | |
1da177e4 | 3829 | |
72626685 | 3830 | if (sector_nr >= max_sector) { |
1da177e4 LT |
3831 | /* just being told to finish up .. nothing much to do */ |
3832 | unplug_slaves(mddev); | |
29269553 N |
3833 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3834 | end_reshape(conf); | |
3835 | return 0; | |
3836 | } | |
72626685 N |
3837 | |
3838 | if (mddev->curr_resync < max_sector) /* aborted */ | |
3839 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3840 | &sync_blocks, 1); | |
16a53ecc | 3841 | else /* completed sync */ |
72626685 N |
3842 | conf->fullsync = 0; |
3843 | bitmap_close_sync(mddev->bitmap); | |
3844 | ||
1da177e4 LT |
3845 | return 0; |
3846 | } | |
ccfcc3c1 | 3847 | |
52c03291 N |
3848 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3849 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 3850 | |
c6207277 N |
3851 | /* No need to check resync_max as we never do more than one |
3852 | * stripe, and as resync_max will always be on a chunk boundary, | |
3853 | * if the check in md_do_sync didn't fire, there is no chance | |
3854 | * of overstepping resync_max here | |
3855 | */ | |
3856 | ||
16a53ecc | 3857 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
3858 | * to resync, then assert that we are finished, because there is |
3859 | * nothing we can do. | |
3860 | */ | |
3285edf1 | 3861 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 3862 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 3863 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 3864 | *skipped = 1; |
1da177e4 LT |
3865 | return rv; |
3866 | } | |
72626685 | 3867 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 3868 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
3869 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
3870 | /* we can skip this block, and probably more */ | |
3871 | sync_blocks /= STRIPE_SECTORS; | |
3872 | *skipped = 1; | |
3873 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
3874 | } | |
1da177e4 | 3875 | |
b47490c9 N |
3876 | |
3877 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
3878 | ||
b5663ba4 | 3879 | sh = get_active_stripe(conf, sector_nr, 0, 1); |
1da177e4 | 3880 | if (sh == NULL) { |
b5663ba4 | 3881 | sh = get_active_stripe(conf, sector_nr, 0, 0); |
1da177e4 | 3882 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 3883 | * is trying to get access |
1da177e4 | 3884 | */ |
66c006a5 | 3885 | schedule_timeout_uninterruptible(1); |
1da177e4 | 3886 | } |
16a53ecc N |
3887 | /* Need to check if array will still be degraded after recovery/resync |
3888 | * We don't need to check the 'failed' flag as when that gets set, | |
3889 | * recovery aborts. | |
3890 | */ | |
3891 | for (i=0; i<mddev->raid_disks; i++) | |
3892 | if (conf->disks[i].rdev == NULL) | |
3893 | still_degraded = 1; | |
3894 | ||
3895 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
3896 | ||
3897 | spin_lock(&sh->lock); | |
1da177e4 LT |
3898 | set_bit(STRIPE_SYNCING, &sh->state); |
3899 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3900 | spin_unlock(&sh->lock); | |
3901 | ||
df10cfbc DW |
3902 | /* wait for any blocked device to be handled */ |
3903 | while(unlikely(!handle_stripe(sh, NULL))) | |
3904 | ; | |
1da177e4 LT |
3905 | release_stripe(sh); |
3906 | ||
3907 | return STRIPE_SECTORS; | |
3908 | } | |
3909 | ||
46031f9a RBJ |
3910 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
3911 | { | |
3912 | /* We may not be able to submit a whole bio at once as there | |
3913 | * may not be enough stripe_heads available. | |
3914 | * We cannot pre-allocate enough stripe_heads as we may need | |
3915 | * more than exist in the cache (if we allow ever large chunks). | |
3916 | * So we do one stripe head at a time and record in | |
3917 | * ->bi_hw_segments how many have been done. | |
3918 | * | |
3919 | * We *know* that this entire raid_bio is in one chunk, so | |
3920 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
3921 | */ | |
3922 | struct stripe_head *sh; | |
911d4ee8 | 3923 | int dd_idx; |
46031f9a RBJ |
3924 | sector_t sector, logical_sector, last_sector; |
3925 | int scnt = 0; | |
3926 | int remaining; | |
3927 | int handled = 0; | |
3928 | ||
3929 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 3930 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 3931 | 0, &dd_idx, NULL); |
46031f9a RBJ |
3932 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
3933 | ||
3934 | for (; logical_sector < last_sector; | |
387bb173 NB |
3935 | logical_sector += STRIPE_SECTORS, |
3936 | sector += STRIPE_SECTORS, | |
3937 | scnt++) { | |
46031f9a | 3938 | |
960e739d | 3939 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
3940 | /* already done this stripe */ |
3941 | continue; | |
3942 | ||
b5663ba4 | 3943 | sh = get_active_stripe(conf, sector, 0, 1); |
46031f9a RBJ |
3944 | |
3945 | if (!sh) { | |
3946 | /* failed to get a stripe - must wait */ | |
960e739d | 3947 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
3948 | conf->retry_read_aligned = raid_bio; |
3949 | return handled; | |
3950 | } | |
3951 | ||
3952 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
3953 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
3954 | release_stripe(sh); | |
960e739d | 3955 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
3956 | conf->retry_read_aligned = raid_bio; |
3957 | return handled; | |
3958 | } | |
3959 | ||
46031f9a RBJ |
3960 | handle_stripe(sh, NULL); |
3961 | release_stripe(sh); | |
3962 | handled++; | |
3963 | } | |
3964 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3965 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 3966 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
3967 | if (remaining == 0) |
3968 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
3969 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3970 | wake_up(&conf->wait_for_stripe); | |
3971 | return handled; | |
3972 | } | |
3973 | ||
3974 | ||
3975 | ||
1da177e4 LT |
3976 | /* |
3977 | * This is our raid5 kernel thread. | |
3978 | * | |
3979 | * We scan the hash table for stripes which can be handled now. | |
3980 | * During the scan, completed stripes are saved for us by the interrupt | |
3981 | * handler, so that they will not have to wait for our next wakeup. | |
3982 | */ | |
6ed3003c | 3983 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
3984 | { |
3985 | struct stripe_head *sh; | |
3986 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3987 | int handled; | |
3988 | ||
45b4233c | 3989 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
3990 | |
3991 | md_check_recovery(mddev); | |
1da177e4 LT |
3992 | |
3993 | handled = 0; | |
3994 | spin_lock_irq(&conf->device_lock); | |
3995 | while (1) { | |
46031f9a | 3996 | struct bio *bio; |
1da177e4 | 3997 | |
ae3c20cc | 3998 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 3999 | int seq = conf->seq_flush; |
700e432d | 4000 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4001 | bitmap_unplug(mddev->bitmap); |
700e432d | 4002 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4003 | conf->seq_write = seq; |
4004 | activate_bit_delay(conf); | |
4005 | } | |
4006 | ||
46031f9a RBJ |
4007 | while ((bio = remove_bio_from_retry(conf))) { |
4008 | int ok; | |
4009 | spin_unlock_irq(&conf->device_lock); | |
4010 | ok = retry_aligned_read(conf, bio); | |
4011 | spin_lock_irq(&conf->device_lock); | |
4012 | if (!ok) | |
4013 | break; | |
4014 | handled++; | |
4015 | } | |
4016 | ||
8b3e6cdc DW |
4017 | sh = __get_priority_stripe(conf); |
4018 | ||
c9f21aaf | 4019 | if (!sh) |
1da177e4 | 4020 | break; |
1da177e4 LT |
4021 | spin_unlock_irq(&conf->device_lock); |
4022 | ||
4023 | handled++; | |
16a53ecc | 4024 | handle_stripe(sh, conf->spare_page); |
1da177e4 LT |
4025 | release_stripe(sh); |
4026 | ||
4027 | spin_lock_irq(&conf->device_lock); | |
4028 | } | |
45b4233c | 4029 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4030 | |
4031 | spin_unlock_irq(&conf->device_lock); | |
4032 | ||
c9f21aaf | 4033 | async_tx_issue_pending_all(); |
1da177e4 LT |
4034 | unplug_slaves(mddev); |
4035 | ||
45b4233c | 4036 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4037 | } |
4038 | ||
3f294f4f | 4039 | static ssize_t |
007583c9 | 4040 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4041 | { |
007583c9 | 4042 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4043 | if (conf) |
4044 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4045 | else | |
4046 | return 0; | |
3f294f4f N |
4047 | } |
4048 | ||
4049 | static ssize_t | |
007583c9 | 4050 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4051 | { |
007583c9 | 4052 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4ef197d8 | 4053 | unsigned long new; |
b5470dc5 DW |
4054 | int err; |
4055 | ||
3f294f4f N |
4056 | if (len >= PAGE_SIZE) |
4057 | return -EINVAL; | |
96de1e66 N |
4058 | if (!conf) |
4059 | return -ENODEV; | |
3f294f4f | 4060 | |
4ef197d8 | 4061 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4062 | return -EINVAL; |
4063 | if (new <= 16 || new > 32768) | |
4064 | return -EINVAL; | |
4065 | while (new < conf->max_nr_stripes) { | |
4066 | if (drop_one_stripe(conf)) | |
4067 | conf->max_nr_stripes--; | |
4068 | else | |
4069 | break; | |
4070 | } | |
b5470dc5 DW |
4071 | err = md_allow_write(mddev); |
4072 | if (err) | |
4073 | return err; | |
3f294f4f N |
4074 | while (new > conf->max_nr_stripes) { |
4075 | if (grow_one_stripe(conf)) | |
4076 | conf->max_nr_stripes++; | |
4077 | else break; | |
4078 | } | |
4079 | return len; | |
4080 | } | |
007583c9 | 4081 | |
96de1e66 N |
4082 | static struct md_sysfs_entry |
4083 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4084 | raid5_show_stripe_cache_size, | |
4085 | raid5_store_stripe_cache_size); | |
3f294f4f | 4086 | |
8b3e6cdc DW |
4087 | static ssize_t |
4088 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4089 | { | |
4090 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4091 | if (conf) | |
4092 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4093 | else | |
4094 | return 0; | |
4095 | } | |
4096 | ||
4097 | static ssize_t | |
4098 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4099 | { | |
4100 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4ef197d8 | 4101 | unsigned long new; |
8b3e6cdc DW |
4102 | if (len >= PAGE_SIZE) |
4103 | return -EINVAL; | |
4104 | if (!conf) | |
4105 | return -ENODEV; | |
4106 | ||
4ef197d8 | 4107 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4108 | return -EINVAL; |
4ef197d8 | 4109 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4110 | return -EINVAL; |
4111 | conf->bypass_threshold = new; | |
4112 | return len; | |
4113 | } | |
4114 | ||
4115 | static struct md_sysfs_entry | |
4116 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4117 | S_IRUGO | S_IWUSR, | |
4118 | raid5_show_preread_threshold, | |
4119 | raid5_store_preread_threshold); | |
4120 | ||
3f294f4f | 4121 | static ssize_t |
96de1e66 | 4122 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4123 | { |
007583c9 | 4124 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4125 | if (conf) |
4126 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4127 | else | |
4128 | return 0; | |
3f294f4f N |
4129 | } |
4130 | ||
96de1e66 N |
4131 | static struct md_sysfs_entry |
4132 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4133 | |
007583c9 | 4134 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4135 | &raid5_stripecache_size.attr, |
4136 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4137 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4138 | NULL, |
4139 | }; | |
007583c9 N |
4140 | static struct attribute_group raid5_attrs_group = { |
4141 | .name = NULL, | |
4142 | .attrs = raid5_attrs, | |
3f294f4f N |
4143 | }; |
4144 | ||
72626685 | 4145 | static int run(mddev_t *mddev) |
1da177e4 LT |
4146 | { |
4147 | raid5_conf_t *conf; | |
4148 | int raid_disk, memory; | |
4149 | mdk_rdev_t *rdev; | |
4150 | struct disk_info *disk; | |
02c2de8c | 4151 | int working_disks = 0; |
1da177e4 | 4152 | |
16a53ecc N |
4153 | if (mddev->level != 5 && mddev->level != 4 && mddev->level != 6) { |
4154 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", | |
14f8d26b | 4155 | mdname(mddev), mddev->level); |
1da177e4 LT |
4156 | return -EIO; |
4157 | } | |
99c0fb5f N |
4158 | if ((mddev->level == 5 && !algorithm_valid_raid5(mddev->layout)) || |
4159 | (mddev->level == 6 && !algorithm_valid_raid6(mddev->layout))) { | |
4160 | printk(KERN_ERR "raid5: %s: layout %d not supported\n", | |
4161 | mdname(mddev), mddev->layout); | |
4162 | return -EIO; | |
4163 | } | |
1da177e4 | 4164 | |
4bbf3771 N |
4165 | if (mddev->chunk_size < PAGE_SIZE) { |
4166 | printk(KERN_ERR "md/raid5: chunk_size must be at least " | |
4167 | "PAGE_SIZE but %d < %ld\n", | |
4168 | mddev->chunk_size, PAGE_SIZE); | |
4169 | return -EINVAL; | |
4170 | } | |
4171 | ||
f6705578 N |
4172 | if (mddev->reshape_position != MaxSector) { |
4173 | /* Check that we can continue the reshape. | |
4174 | * Currently only disks can change, it must | |
4175 | * increase, and we must be past the point where | |
4176 | * a stripe over-writes itself | |
4177 | */ | |
4178 | sector_t here_new, here_old; | |
4179 | int old_disks; | |
f416885e | 4180 | int max_degraded = (mddev->level == 5 ? 1 : 2); |
f6705578 N |
4181 | |
4182 | if (mddev->new_level != mddev->level || | |
4183 | mddev->new_layout != mddev->layout || | |
4184 | mddev->new_chunk != mddev->chunk_size) { | |
f416885e N |
4185 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4186 | "required - aborting.\n", | |
f6705578 N |
4187 | mdname(mddev)); |
4188 | return -EINVAL; | |
4189 | } | |
4190 | if (mddev->delta_disks <= 0) { | |
f416885e N |
4191 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4192 | "(reduce disks) required - aborting.\n", | |
f6705578 N |
4193 | mdname(mddev)); |
4194 | return -EINVAL; | |
4195 | } | |
4196 | old_disks = mddev->raid_disks - mddev->delta_disks; | |
4197 | /* reshape_position must be on a new-stripe boundary, and one | |
f416885e N |
4198 | * further up in new geometry must map after here in old |
4199 | * geometry. | |
f6705578 N |
4200 | */ |
4201 | here_new = mddev->reshape_position; | |
f416885e N |
4202 | if (sector_div(here_new, (mddev->chunk_size>>9)* |
4203 | (mddev->raid_disks - max_degraded))) { | |
4204 | printk(KERN_ERR "raid5: reshape_position not " | |
4205 | "on a stripe boundary\n"); | |
f6705578 N |
4206 | return -EINVAL; |
4207 | } | |
4208 | /* here_new is the stripe we will write to */ | |
4209 | here_old = mddev->reshape_position; | |
f416885e N |
4210 | sector_div(here_old, (mddev->chunk_size>>9)* |
4211 | (old_disks-max_degraded)); | |
4212 | /* here_old is the first stripe that we might need to read | |
4213 | * from */ | |
f6705578 N |
4214 | if (here_new >= here_old) { |
4215 | /* Reading from the same stripe as writing to - bad */ | |
f416885e N |
4216 | printk(KERN_ERR "raid5: reshape_position too early for " |
4217 | "auto-recovery - aborting.\n"); | |
f6705578 N |
4218 | return -EINVAL; |
4219 | } | |
4220 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4221 | /* OK, we should be able to continue; */ | |
4222 | } | |
4223 | ||
4224 | ||
b55e6bfc | 4225 | mddev->private = kzalloc(sizeof (raid5_conf_t), GFP_KERNEL); |
1da177e4 LT |
4226 | if ((conf = mddev->private) == NULL) |
4227 | goto abort; | |
f6705578 N |
4228 | if (mddev->reshape_position == MaxSector) { |
4229 | conf->previous_raid_disks = conf->raid_disks = mddev->raid_disks; | |
4230 | } else { | |
4231 | conf->raid_disks = mddev->raid_disks; | |
4232 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; | |
4233 | } | |
4234 | ||
4235 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
4236 | GFP_KERNEL); |
4237 | if (!conf->disks) | |
4238 | goto abort; | |
9ffae0cf | 4239 | |
1da177e4 LT |
4240 | conf->mddev = mddev; |
4241 | ||
fccddba0 | 4242 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4243 | goto abort; |
1da177e4 | 4244 | |
16a53ecc N |
4245 | if (mddev->level == 6) { |
4246 | conf->spare_page = alloc_page(GFP_KERNEL); | |
4247 | if (!conf->spare_page) | |
4248 | goto abort; | |
4249 | } | |
1da177e4 | 4250 | spin_lock_init(&conf->device_lock); |
e7e72bf6 | 4251 | mddev->queue->queue_lock = &conf->device_lock; |
1da177e4 LT |
4252 | init_waitqueue_head(&conf->wait_for_stripe); |
4253 | init_waitqueue_head(&conf->wait_for_overlap); | |
4254 | INIT_LIST_HEAD(&conf->handle_list); | |
8b3e6cdc | 4255 | INIT_LIST_HEAD(&conf->hold_list); |
1da177e4 | 4256 | INIT_LIST_HEAD(&conf->delayed_list); |
72626685 | 4257 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
4258 | INIT_LIST_HEAD(&conf->inactive_list); |
4259 | atomic_set(&conf->active_stripes, 0); | |
4260 | atomic_set(&conf->preread_active_stripes, 0); | |
46031f9a | 4261 | atomic_set(&conf->active_aligned_reads, 0); |
8b3e6cdc | 4262 | conf->bypass_threshold = BYPASS_THRESHOLD; |
1da177e4 | 4263 | |
45b4233c | 4264 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4265 | |
159ec1fc | 4266 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4267 | raid_disk = rdev->raid_disk; |
f6705578 | 4268 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
4269 | || raid_disk < 0) |
4270 | continue; | |
4271 | disk = conf->disks + raid_disk; | |
4272 | ||
4273 | disk->rdev = rdev; | |
4274 | ||
b2d444d7 | 4275 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4276 | char b[BDEVNAME_SIZE]; |
4277 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4278 | " disk %d\n", bdevname(rdev->bdev,b), | |
4279 | raid_disk); | |
02c2de8c | 4280 | working_disks++; |
8c2e870a NB |
4281 | } else |
4282 | /* Cannot rely on bitmap to complete recovery */ | |
4283 | conf->fullsync = 1; | |
1da177e4 LT |
4284 | } |
4285 | ||
1da177e4 | 4286 | /* |
16a53ecc | 4287 | * 0 for a fully functional array, 1 or 2 for a degraded array. |
1da177e4 | 4288 | */ |
02c2de8c | 4289 | mddev->degraded = conf->raid_disks - working_disks; |
1da177e4 LT |
4290 | conf->mddev = mddev; |
4291 | conf->chunk_size = mddev->chunk_size; | |
4292 | conf->level = mddev->level; | |
16a53ecc N |
4293 | if (conf->level == 6) |
4294 | conf->max_degraded = 2; | |
4295 | else | |
4296 | conf->max_degraded = 1; | |
1da177e4 LT |
4297 | conf->algorithm = mddev->layout; |
4298 | conf->max_nr_stripes = NR_STRIPES; | |
f6705578 | 4299 | conf->expand_progress = mddev->reshape_position; |
1da177e4 LT |
4300 | |
4301 | /* device size must be a multiple of chunk size */ | |
58c0fed4 AN |
4302 | mddev->dev_sectors &= ~(mddev->chunk_size / 512 - 1); |
4303 | mddev->resync_max_sectors = mddev->dev_sectors; | |
1da177e4 | 4304 | |
16a53ecc N |
4305 | if (conf->level == 6 && conf->raid_disks < 4) { |
4306 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4307 | mdname(mddev), conf->raid_disks); | |
4308 | goto abort; | |
4309 | } | |
1da177e4 LT |
4310 | if (!conf->chunk_size || conf->chunk_size % 4) { |
4311 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
4312 | conf->chunk_size, mdname(mddev)); | |
4313 | goto abort; | |
4314 | } | |
16a53ecc | 4315 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
4316 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
4317 | " (%d/%d failed)\n", | |
02c2de8c | 4318 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4319 | goto abort; |
4320 | } | |
4321 | ||
16a53ecc | 4322 | if (mddev->degraded > 0 && |
1da177e4 | 4323 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4324 | if (mddev->ok_start_degraded) |
4325 | printk(KERN_WARNING | |
4326 | "raid5: starting dirty degraded array: %s" | |
4327 | "- data corruption possible.\n", | |
4328 | mdname(mddev)); | |
4329 | else { | |
4330 | printk(KERN_ERR | |
4331 | "raid5: cannot start dirty degraded array for %s\n", | |
4332 | mdname(mddev)); | |
4333 | goto abort; | |
4334 | } | |
1da177e4 LT |
4335 | } |
4336 | ||
4337 | { | |
4338 | mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5"); | |
4339 | if (!mddev->thread) { | |
4340 | printk(KERN_ERR | |
4341 | "raid5: couldn't allocate thread for %s\n", | |
4342 | mdname(mddev)); | |
4343 | goto abort; | |
4344 | } | |
4345 | } | |
5036805b | 4346 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
1da177e4 LT |
4347 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4348 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
4349 | printk(KERN_ERR | |
4350 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4351 | shrink_stripes(conf); | |
4352 | md_unregister_thread(mddev->thread); | |
4353 | goto abort; | |
4354 | } else | |
4355 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4356 | memory, mdname(mddev)); | |
4357 | ||
4358 | if (mddev->degraded == 0) | |
4359 | printk("raid5: raid level %d set %s active with %d out of %d" | |
4360 | " devices, algorithm %d\n", conf->level, mdname(mddev), | |
4361 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
4362 | conf->algorithm); | |
4363 | else | |
4364 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
4365 | " out of %d devices, algorithm %d\n", conf->level, | |
4366 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
4367 | mddev->raid_disks, conf->algorithm); | |
4368 | ||
4369 | print_raid5_conf(conf); | |
4370 | ||
f6705578 N |
4371 | if (conf->expand_progress != MaxSector) { |
4372 | printk("...ok start reshape thread\n"); | |
b578d55f | 4373 | conf->expand_lo = conf->expand_progress; |
f6705578 N |
4374 | atomic_set(&conf->reshape_stripes, 0); |
4375 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4376 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4377 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4378 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4379 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4380 | "%s_reshape"); | |
f6705578 N |
4381 | } |
4382 | ||
1da177e4 | 4383 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 4384 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
4385 | */ |
4386 | { | |
16a53ecc N |
4387 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
4388 | int stripe = data_disks * | |
8932c2e0 | 4389 | (mddev->chunk_size / PAGE_SIZE); |
1da177e4 LT |
4390 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
4391 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4392 | } | |
4393 | ||
4394 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
4395 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
4396 | printk(KERN_WARNING | |
4397 | "raid5: failed to create sysfs attributes for %s\n", | |
4398 | mdname(mddev)); | |
7a5febe9 N |
4399 | |
4400 | mddev->queue->unplug_fn = raid5_unplug_device; | |
f022b2fd | 4401 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 4402 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 4403 | |
58c0fed4 AN |
4404 | mddev->array_sectors = mddev->dev_sectors * |
4405 | (conf->previous_raid_disks - conf->max_degraded); | |
7a5febe9 | 4406 | |
23032a0e RBJ |
4407 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
4408 | ||
1da177e4 LT |
4409 | return 0; |
4410 | abort: | |
4411 | if (conf) { | |
4412 | print_raid5_conf(conf); | |
16a53ecc | 4413 | safe_put_page(conf->spare_page); |
b55e6bfc | 4414 | kfree(conf->disks); |
fccddba0 | 4415 | kfree(conf->stripe_hashtbl); |
1da177e4 LT |
4416 | kfree(conf); |
4417 | } | |
4418 | mddev->private = NULL; | |
4419 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
4420 | return -EIO; | |
4421 | } | |
4422 | ||
4423 | ||
4424 | ||
3f294f4f | 4425 | static int stop(mddev_t *mddev) |
1da177e4 LT |
4426 | { |
4427 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4428 | ||
4429 | md_unregister_thread(mddev->thread); | |
4430 | mddev->thread = NULL; | |
4431 | shrink_stripes(conf); | |
fccddba0 | 4432 | kfree(conf->stripe_hashtbl); |
041ae52e | 4433 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4434 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 4435 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
b55e6bfc | 4436 | kfree(conf->disks); |
96de1e66 | 4437 | kfree(conf); |
1da177e4 LT |
4438 | mddev->private = NULL; |
4439 | return 0; | |
4440 | } | |
4441 | ||
45b4233c | 4442 | #ifdef DEBUG |
d710e138 | 4443 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
4444 | { |
4445 | int i; | |
4446 | ||
16a53ecc N |
4447 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
4448 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
4449 | seq_printf(seq, "sh %llu, count %d.\n", | |
4450 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
4451 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 4452 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
4453 | seq_printf(seq, "(cache%d: %p %ld) ", |
4454 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 4455 | } |
16a53ecc | 4456 | seq_printf(seq, "\n"); |
1da177e4 LT |
4457 | } |
4458 | ||
d710e138 | 4459 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
4460 | { |
4461 | struct stripe_head *sh; | |
fccddba0 | 4462 | struct hlist_node *hn; |
1da177e4 LT |
4463 | int i; |
4464 | ||
4465 | spin_lock_irq(&conf->device_lock); | |
4466 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 4467 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
4468 | if (sh->raid_conf != conf) |
4469 | continue; | |
16a53ecc | 4470 | print_sh(seq, sh); |
1da177e4 LT |
4471 | } |
4472 | } | |
4473 | spin_unlock_irq(&conf->device_lock); | |
4474 | } | |
4475 | #endif | |
4476 | ||
d710e138 | 4477 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
4478 | { |
4479 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4480 | int i; | |
4481 | ||
4482 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
02c2de8c | 4483 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
4484 | for (i = 0; i < conf->raid_disks; i++) |
4485 | seq_printf (seq, "%s", | |
4486 | conf->disks[i].rdev && | |
b2d444d7 | 4487 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 4488 | seq_printf (seq, "]"); |
45b4233c | 4489 | #ifdef DEBUG |
16a53ecc N |
4490 | seq_printf (seq, "\n"); |
4491 | printall(seq, conf); | |
1da177e4 LT |
4492 | #endif |
4493 | } | |
4494 | ||
4495 | static void print_raid5_conf (raid5_conf_t *conf) | |
4496 | { | |
4497 | int i; | |
4498 | struct disk_info *tmp; | |
4499 | ||
4500 | printk("RAID5 conf printout:\n"); | |
4501 | if (!conf) { | |
4502 | printk("(conf==NULL)\n"); | |
4503 | return; | |
4504 | } | |
02c2de8c N |
4505 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
4506 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
4507 | |
4508 | for (i = 0; i < conf->raid_disks; i++) { | |
4509 | char b[BDEVNAME_SIZE]; | |
4510 | tmp = conf->disks + i; | |
4511 | if (tmp->rdev) | |
4512 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 4513 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
4514 | bdevname(tmp->rdev->bdev,b)); |
4515 | } | |
4516 | } | |
4517 | ||
4518 | static int raid5_spare_active(mddev_t *mddev) | |
4519 | { | |
4520 | int i; | |
4521 | raid5_conf_t *conf = mddev->private; | |
4522 | struct disk_info *tmp; | |
4523 | ||
4524 | for (i = 0; i < conf->raid_disks; i++) { | |
4525 | tmp = conf->disks + i; | |
4526 | if (tmp->rdev | |
b2d444d7 | 4527 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
4528 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
4529 | unsigned long flags; | |
4530 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 4531 | mddev->degraded--; |
c04be0aa | 4532 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
4533 | } |
4534 | } | |
4535 | print_raid5_conf(conf); | |
4536 | return 0; | |
4537 | } | |
4538 | ||
4539 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
4540 | { | |
4541 | raid5_conf_t *conf = mddev->private; | |
4542 | int err = 0; | |
4543 | mdk_rdev_t *rdev; | |
4544 | struct disk_info *p = conf->disks + number; | |
4545 | ||
4546 | print_raid5_conf(conf); | |
4547 | rdev = p->rdev; | |
4548 | if (rdev) { | |
b2d444d7 | 4549 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
4550 | atomic_read(&rdev->nr_pending)) { |
4551 | err = -EBUSY; | |
4552 | goto abort; | |
4553 | } | |
dfc70645 N |
4554 | /* Only remove non-faulty devices if recovery |
4555 | * isn't possible. | |
4556 | */ | |
4557 | if (!test_bit(Faulty, &rdev->flags) && | |
4558 | mddev->degraded <= conf->max_degraded) { | |
4559 | err = -EBUSY; | |
4560 | goto abort; | |
4561 | } | |
1da177e4 | 4562 | p->rdev = NULL; |
fbd568a3 | 4563 | synchronize_rcu(); |
1da177e4 LT |
4564 | if (atomic_read(&rdev->nr_pending)) { |
4565 | /* lost the race, try later */ | |
4566 | err = -EBUSY; | |
4567 | p->rdev = rdev; | |
4568 | } | |
4569 | } | |
4570 | abort: | |
4571 | ||
4572 | print_raid5_conf(conf); | |
4573 | return err; | |
4574 | } | |
4575 | ||
4576 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
4577 | { | |
4578 | raid5_conf_t *conf = mddev->private; | |
199050ea | 4579 | int err = -EEXIST; |
1da177e4 LT |
4580 | int disk; |
4581 | struct disk_info *p; | |
6c2fce2e NB |
4582 | int first = 0; |
4583 | int last = conf->raid_disks - 1; | |
1da177e4 | 4584 | |
16a53ecc | 4585 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 4586 | /* no point adding a device */ |
199050ea | 4587 | return -EINVAL; |
1da177e4 | 4588 | |
6c2fce2e NB |
4589 | if (rdev->raid_disk >= 0) |
4590 | first = last = rdev->raid_disk; | |
1da177e4 LT |
4591 | |
4592 | /* | |
16a53ecc N |
4593 | * find the disk ... but prefer rdev->saved_raid_disk |
4594 | * if possible. | |
1da177e4 | 4595 | */ |
16a53ecc | 4596 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 4597 | rdev->saved_raid_disk >= first && |
16a53ecc N |
4598 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
4599 | disk = rdev->saved_raid_disk; | |
4600 | else | |
6c2fce2e NB |
4601 | disk = first; |
4602 | for ( ; disk <= last ; disk++) | |
1da177e4 | 4603 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 4604 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4605 | rdev->raid_disk = disk; |
199050ea | 4606 | err = 0; |
72626685 N |
4607 | if (rdev->saved_raid_disk != disk) |
4608 | conf->fullsync = 1; | |
d6065f7b | 4609 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
4610 | break; |
4611 | } | |
4612 | print_raid5_conf(conf); | |
199050ea | 4613 | return err; |
1da177e4 LT |
4614 | } |
4615 | ||
4616 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
4617 | { | |
4618 | /* no resync is happening, and there is enough space | |
4619 | * on all devices, so we can resize. | |
4620 | * We need to make sure resync covers any new space. | |
4621 | * If the array is shrinking we should possibly wait until | |
4622 | * any io in the removed space completes, but it hardly seems | |
4623 | * worth it. | |
4624 | */ | |
16a53ecc N |
4625 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4626 | ||
1da177e4 | 4627 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); |
f233ea5c AN |
4628 | mddev->array_sectors = sectors * (mddev->raid_disks |
4629 | - conf->max_degraded); | |
4630 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
44ce6294 | 4631 | mddev->changed = 1; |
58c0fed4 AN |
4632 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
4633 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
4634 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
4635 | } | |
58c0fed4 | 4636 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 4637 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
4638 | return 0; |
4639 | } | |
4640 | ||
29269553 | 4641 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f | 4642 | static int raid5_check_reshape(mddev_t *mddev) |
29269553 N |
4643 | { |
4644 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4645 | int err; | |
29269553 | 4646 | |
63c70c4f N |
4647 | if (mddev->delta_disks < 0 || |
4648 | mddev->new_level != mddev->level) | |
4649 | return -EINVAL; /* Cannot shrink array or change level yet */ | |
4650 | if (mddev->delta_disks == 0) | |
29269553 | 4651 | return 0; /* nothing to do */ |
dba034ee N |
4652 | if (mddev->bitmap) |
4653 | /* Cannot grow a bitmap yet */ | |
4654 | return -EBUSY; | |
29269553 N |
4655 | |
4656 | /* Can only proceed if there are plenty of stripe_heads. | |
4657 | * We need a minimum of one full stripe,, and for sensible progress | |
4658 | * it is best to have about 4 times that. | |
4659 | * If we require 4 times, then the default 256 4K stripe_heads will | |
4660 | * allow for chunk sizes up to 256K, which is probably OK. | |
4661 | * If the chunk size is greater, user-space should request more | |
4662 | * stripe_heads first. | |
4663 | */ | |
63c70c4f N |
4664 | if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes || |
4665 | (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) { | |
29269553 N |
4666 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", |
4667 | (mddev->chunk_size / STRIPE_SIZE)*4); | |
4668 | return -ENOSPC; | |
4669 | } | |
4670 | ||
63c70c4f N |
4671 | err = resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
4672 | if (err) | |
4673 | return err; | |
4674 | ||
b4c4c7b8 N |
4675 | if (mddev->degraded > conf->max_degraded) |
4676 | return -EINVAL; | |
63c70c4f N |
4677 | /* looks like we might be able to manage this */ |
4678 | return 0; | |
4679 | } | |
4680 | ||
4681 | static int raid5_start_reshape(mddev_t *mddev) | |
4682 | { | |
4683 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4684 | mdk_rdev_t *rdev; | |
63c70c4f N |
4685 | int spares = 0; |
4686 | int added_devices = 0; | |
c04be0aa | 4687 | unsigned long flags; |
63c70c4f | 4688 | |
f416885e | 4689 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
4690 | return -EBUSY; |
4691 | ||
159ec1fc | 4692 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4693 | if (rdev->raid_disk < 0 && |
4694 | !test_bit(Faulty, &rdev->flags)) | |
4695 | spares++; | |
63c70c4f | 4696 | |
f416885e | 4697 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
4698 | /* Not enough devices even to make a degraded array |
4699 | * of that size | |
4700 | */ | |
4701 | return -EINVAL; | |
4702 | ||
f6705578 | 4703 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
4704 | spin_lock_irq(&conf->device_lock); |
4705 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 4706 | conf->raid_disks += mddev->delta_disks; |
29269553 | 4707 | conf->expand_progress = 0; |
b578d55f | 4708 | conf->expand_lo = 0; |
29269553 N |
4709 | spin_unlock_irq(&conf->device_lock); |
4710 | ||
4711 | /* Add some new drives, as many as will fit. | |
4712 | * We know there are enough to make the newly sized array work. | |
4713 | */ | |
159ec1fc | 4714 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4715 | if (rdev->raid_disk < 0 && |
4716 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 4717 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 N |
4718 | char nm[20]; |
4719 | set_bit(In_sync, &rdev->flags); | |
29269553 | 4720 | added_devices++; |
5fd6c1dc | 4721 | rdev->recovery_offset = 0; |
29269553 | 4722 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
4723 | if (sysfs_create_link(&mddev->kobj, |
4724 | &rdev->kobj, nm)) | |
4725 | printk(KERN_WARNING | |
4726 | "raid5: failed to create " | |
4727 | " link %s for %s\n", | |
4728 | nm, mdname(mddev)); | |
29269553 N |
4729 | } else |
4730 | break; | |
4731 | } | |
4732 | ||
c04be0aa | 4733 | spin_lock_irqsave(&conf->device_lock, flags); |
63c70c4f | 4734 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) - added_devices; |
c04be0aa | 4735 | spin_unlock_irqrestore(&conf->device_lock, flags); |
63c70c4f | 4736 | mddev->raid_disks = conf->raid_disks; |
f6705578 | 4737 | mddev->reshape_position = 0; |
850b2b42 | 4738 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 4739 | |
29269553 N |
4740 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
4741 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4742 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4743 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4744 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4745 | "%s_reshape"); | |
4746 | if (!mddev->sync_thread) { | |
4747 | mddev->recovery = 0; | |
4748 | spin_lock_irq(&conf->device_lock); | |
4749 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
4750 | conf->expand_progress = MaxSector; | |
4751 | spin_unlock_irq(&conf->device_lock); | |
4752 | return -EAGAIN; | |
4753 | } | |
4754 | md_wakeup_thread(mddev->sync_thread); | |
4755 | md_new_event(mddev); | |
4756 | return 0; | |
4757 | } | |
4758 | #endif | |
4759 | ||
4760 | static void end_reshape(raid5_conf_t *conf) | |
4761 | { | |
4762 | struct block_device *bdev; | |
4763 | ||
f6705578 | 4764 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
58c0fed4 | 4765 | conf->mddev->array_sectors = conf->mddev->dev_sectors * |
f416885e | 4766 | (conf->raid_disks - conf->max_degraded); |
f233ea5c | 4767 | set_capacity(conf->mddev->gendisk, conf->mddev->array_sectors); |
44ce6294 | 4768 | conf->mddev->changed = 1; |
f6705578 N |
4769 | |
4770 | bdev = bdget_disk(conf->mddev->gendisk, 0); | |
4771 | if (bdev) { | |
4772 | mutex_lock(&bdev->bd_inode->i_mutex); | |
f233ea5c AN |
4773 | i_size_write(bdev->bd_inode, |
4774 | (loff_t)conf->mddev->array_sectors << 9); | |
f6705578 N |
4775 | mutex_unlock(&bdev->bd_inode->i_mutex); |
4776 | bdput(bdev); | |
4777 | } | |
4778 | spin_lock_irq(&conf->device_lock); | |
4779 | conf->expand_progress = MaxSector; | |
4780 | spin_unlock_irq(&conf->device_lock); | |
4781 | conf->mddev->reshape_position = MaxSector; | |
16a53ecc N |
4782 | |
4783 | /* read-ahead size must cover two whole stripes, which is | |
4784 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4785 | */ | |
4786 | { | |
4787 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
4788 | int stripe = data_disks * | |
4789 | (conf->mddev->chunk_size / PAGE_SIZE); | |
4790 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4791 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4792 | } | |
29269553 | 4793 | } |
29269553 N |
4794 | } |
4795 | ||
72626685 N |
4796 | static void raid5_quiesce(mddev_t *mddev, int state) |
4797 | { | |
4798 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4799 | ||
4800 | switch(state) { | |
e464eafd N |
4801 | case 2: /* resume for a suspend */ |
4802 | wake_up(&conf->wait_for_overlap); | |
4803 | break; | |
4804 | ||
72626685 N |
4805 | case 1: /* stop all writes */ |
4806 | spin_lock_irq(&conf->device_lock); | |
4807 | conf->quiesce = 1; | |
4808 | wait_event_lock_irq(conf->wait_for_stripe, | |
46031f9a RBJ |
4809 | atomic_read(&conf->active_stripes) == 0 && |
4810 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 N |
4811 | conf->device_lock, /* nothing */); |
4812 | spin_unlock_irq(&conf->device_lock); | |
4813 | break; | |
4814 | ||
4815 | case 0: /* re-enable writes */ | |
4816 | spin_lock_irq(&conf->device_lock); | |
4817 | conf->quiesce = 0; | |
4818 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 4819 | wake_up(&conf->wait_for_overlap); |
72626685 N |
4820 | spin_unlock_irq(&conf->device_lock); |
4821 | break; | |
4822 | } | |
72626685 | 4823 | } |
b15c2e57 | 4824 | |
16a53ecc N |
4825 | static struct mdk_personality raid6_personality = |
4826 | { | |
4827 | .name = "raid6", | |
4828 | .level = 6, | |
4829 | .owner = THIS_MODULE, | |
4830 | .make_request = make_request, | |
4831 | .run = run, | |
4832 | .stop = stop, | |
4833 | .status = status, | |
4834 | .error_handler = error, | |
4835 | .hot_add_disk = raid5_add_disk, | |
4836 | .hot_remove_disk= raid5_remove_disk, | |
4837 | .spare_active = raid5_spare_active, | |
4838 | .sync_request = sync_request, | |
4839 | .resize = raid5_resize, | |
f416885e N |
4840 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4841 | .check_reshape = raid5_check_reshape, | |
4842 | .start_reshape = raid5_start_reshape, | |
4843 | #endif | |
16a53ecc N |
4844 | .quiesce = raid5_quiesce, |
4845 | }; | |
2604b703 | 4846 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
4847 | { |
4848 | .name = "raid5", | |
2604b703 | 4849 | .level = 5, |
1da177e4 LT |
4850 | .owner = THIS_MODULE, |
4851 | .make_request = make_request, | |
4852 | .run = run, | |
4853 | .stop = stop, | |
4854 | .status = status, | |
4855 | .error_handler = error, | |
4856 | .hot_add_disk = raid5_add_disk, | |
4857 | .hot_remove_disk= raid5_remove_disk, | |
4858 | .spare_active = raid5_spare_active, | |
4859 | .sync_request = sync_request, | |
4860 | .resize = raid5_resize, | |
29269553 | 4861 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f N |
4862 | .check_reshape = raid5_check_reshape, |
4863 | .start_reshape = raid5_start_reshape, | |
29269553 | 4864 | #endif |
72626685 | 4865 | .quiesce = raid5_quiesce, |
1da177e4 LT |
4866 | }; |
4867 | ||
2604b703 | 4868 | static struct mdk_personality raid4_personality = |
1da177e4 | 4869 | { |
2604b703 N |
4870 | .name = "raid4", |
4871 | .level = 4, | |
4872 | .owner = THIS_MODULE, | |
4873 | .make_request = make_request, | |
4874 | .run = run, | |
4875 | .stop = stop, | |
4876 | .status = status, | |
4877 | .error_handler = error, | |
4878 | .hot_add_disk = raid5_add_disk, | |
4879 | .hot_remove_disk= raid5_remove_disk, | |
4880 | .spare_active = raid5_spare_active, | |
4881 | .sync_request = sync_request, | |
4882 | .resize = raid5_resize, | |
3d37890b N |
4883 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4884 | .check_reshape = raid5_check_reshape, | |
4885 | .start_reshape = raid5_start_reshape, | |
4886 | #endif | |
2604b703 N |
4887 | .quiesce = raid5_quiesce, |
4888 | }; | |
4889 | ||
4890 | static int __init raid5_init(void) | |
4891 | { | |
16a53ecc N |
4892 | int e; |
4893 | ||
4894 | e = raid6_select_algo(); | |
4895 | if ( e ) | |
4896 | return e; | |
4897 | register_md_personality(&raid6_personality); | |
2604b703 N |
4898 | register_md_personality(&raid5_personality); |
4899 | register_md_personality(&raid4_personality); | |
4900 | return 0; | |
1da177e4 LT |
4901 | } |
4902 | ||
2604b703 | 4903 | static void raid5_exit(void) |
1da177e4 | 4904 | { |
16a53ecc | 4905 | unregister_md_personality(&raid6_personality); |
2604b703 N |
4906 | unregister_md_personality(&raid5_personality); |
4907 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
4908 | } |
4909 | ||
4910 | module_init(raid5_init); | |
4911 | module_exit(raid5_exit); | |
4912 | MODULE_LICENSE("GPL"); | |
4913 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
4914 | MODULE_ALIAS("md-raid5"); |
4915 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
4916 | MODULE_ALIAS("md-level-5"); |
4917 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
4918 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
4919 | MODULE_ALIAS("md-raid6"); | |
4920 | MODULE_ALIAS("md-level-6"); | |
4921 | ||
4922 | /* This used to be two separate modules, they were: */ | |
4923 | MODULE_ALIAS("raid5"); | |
4924 | MODULE_ALIAS("raid6"); |