]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
30 | * conf->bm_write is the number of the last batch successfully written. | |
31 | * conf->bm_flush is the number of the last batch that was closed to | |
32 | * new additions. | |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
35 | * the number of the batch it will be in. This is bm_flush+1. | |
36 | * When we are ready to do a write, if that batch hasn't been written yet, | |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
07a3b417 | 50 | #include <linux/async.h> |
bff61975 | 51 | #include <linux/seq_file.h> |
36d1c647 | 52 | #include <linux/cpu.h> |
5a0e3ad6 | 53 | #include <linux/slab.h> |
43b2e5d8 | 54 | #include "md.h" |
bff61975 | 55 | #include "raid5.h" |
54071b38 | 56 | #include "raid0.h" |
ef740c37 | 57 | #include "bitmap.h" |
72626685 | 58 | |
1da177e4 LT |
59 | /* |
60 | * Stripe cache | |
61 | */ | |
62 | ||
63 | #define NR_STRIPES 256 | |
64 | #define STRIPE_SIZE PAGE_SIZE | |
65 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
66 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
67 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 68 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 69 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
70 | #define HASH_MASK (NR_HASH - 1) |
71 | ||
fccddba0 | 72 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
73 | |
74 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
75 | * order without overlap. There may be several bio's per stripe+device, and | |
76 | * a bio could span several devices. | |
77 | * When walking this list for a particular stripe+device, we must never proceed | |
78 | * beyond a bio that extends past this device, as the next bio might no longer | |
79 | * be valid. | |
80 | * This macro is used to determine the 'next' bio in the list, given the sector | |
81 | * of the current stripe+device | |
82 | */ | |
83 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
84 | /* | |
85 | * The following can be used to debug the driver | |
86 | */ | |
1da177e4 LT |
87 | #define RAID5_PARANOIA 1 |
88 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
89 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
90 | #else | |
91 | # define CHECK_DEVLOCK() | |
92 | #endif | |
93 | ||
45b4233c | 94 | #ifdef DEBUG |
1da177e4 LT |
95 | #define inline |
96 | #define __inline__ | |
97 | #endif | |
98 | ||
6be9d494 BS |
99 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
100 | ||
960e739d | 101 | /* |
5b99c2ff JA |
102 | * We maintain a biased count of active stripes in the bottom 16 bits of |
103 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
104 | */ |
105 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
106 | { | |
5b99c2ff | 107 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
108 | } |
109 | ||
110 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
111 | { | |
5b99c2ff | 112 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
113 | } |
114 | ||
115 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
116 | { | |
117 | --bio->bi_phys_segments; | |
118 | return raid5_bi_phys_segments(bio); | |
119 | } | |
120 | ||
121 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
122 | { | |
123 | unsigned short val = raid5_bi_hw_segments(bio); | |
124 | ||
125 | --val; | |
5b99c2ff | 126 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
127 | return val; |
128 | } | |
129 | ||
130 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
131 | { | |
5b99c2ff | 132 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16); |
960e739d JA |
133 | } |
134 | ||
d0dabf7e N |
135 | /* Find first data disk in a raid6 stripe */ |
136 | static inline int raid6_d0(struct stripe_head *sh) | |
137 | { | |
67cc2b81 N |
138 | if (sh->ddf_layout) |
139 | /* ddf always start from first device */ | |
140 | return 0; | |
141 | /* md starts just after Q block */ | |
d0dabf7e N |
142 | if (sh->qd_idx == sh->disks - 1) |
143 | return 0; | |
144 | else | |
145 | return sh->qd_idx + 1; | |
146 | } | |
16a53ecc N |
147 | static inline int raid6_next_disk(int disk, int raid_disks) |
148 | { | |
149 | disk++; | |
150 | return (disk < raid_disks) ? disk : 0; | |
151 | } | |
a4456856 | 152 | |
d0dabf7e N |
153 | /* When walking through the disks in a raid5, starting at raid6_d0, |
154 | * We need to map each disk to a 'slot', where the data disks are slot | |
155 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
156 | * is raid_disks-1. This help does that mapping. | |
157 | */ | |
67cc2b81 N |
158 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
159 | int *count, int syndrome_disks) | |
d0dabf7e | 160 | { |
6629542e | 161 | int slot = *count; |
67cc2b81 | 162 | |
e4424fee | 163 | if (sh->ddf_layout) |
6629542e | 164 | (*count)++; |
d0dabf7e | 165 | if (idx == sh->pd_idx) |
67cc2b81 | 166 | return syndrome_disks; |
d0dabf7e | 167 | if (idx == sh->qd_idx) |
67cc2b81 | 168 | return syndrome_disks + 1; |
e4424fee | 169 | if (!sh->ddf_layout) |
6629542e | 170 | (*count)++; |
d0dabf7e N |
171 | return slot; |
172 | } | |
173 | ||
a4456856 DW |
174 | static void return_io(struct bio *return_bi) |
175 | { | |
176 | struct bio *bi = return_bi; | |
177 | while (bi) { | |
a4456856 DW |
178 | |
179 | return_bi = bi->bi_next; | |
180 | bi->bi_next = NULL; | |
181 | bi->bi_size = 0; | |
0e13fe23 | 182 | bio_endio(bi, 0); |
a4456856 DW |
183 | bi = return_bi; |
184 | } | |
185 | } | |
186 | ||
1da177e4 LT |
187 | static void print_raid5_conf (raid5_conf_t *conf); |
188 | ||
600aa109 DW |
189 | static int stripe_operations_active(struct stripe_head *sh) |
190 | { | |
191 | return sh->check_state || sh->reconstruct_state || | |
192 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
193 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
194 | } | |
195 | ||
858119e1 | 196 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
197 | { |
198 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
199 | BUG_ON(!list_empty(&sh->lru)); |
200 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 201 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 202 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 203 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
204 | blk_plug_device(conf->mddev->queue); |
205 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 206 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 207 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
208 | blk_plug_device(conf->mddev->queue); |
209 | } else { | |
72626685 | 210 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 211 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 212 | } |
1da177e4 LT |
213 | md_wakeup_thread(conf->mddev->thread); |
214 | } else { | |
600aa109 | 215 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
216 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
217 | atomic_dec(&conf->preread_active_stripes); | |
218 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
219 | md_wakeup_thread(conf->mddev->thread); | |
220 | } | |
1da177e4 | 221 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
222 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
223 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 224 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
225 | if (conf->retry_read_aligned) |
226 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 227 | } |
1da177e4 LT |
228 | } |
229 | } | |
230 | } | |
d0dabf7e | 231 | |
1da177e4 LT |
232 | static void release_stripe(struct stripe_head *sh) |
233 | { | |
234 | raid5_conf_t *conf = sh->raid_conf; | |
235 | unsigned long flags; | |
16a53ecc | 236 | |
1da177e4 LT |
237 | spin_lock_irqsave(&conf->device_lock, flags); |
238 | __release_stripe(conf, sh); | |
239 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
240 | } | |
241 | ||
fccddba0 | 242 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 243 | { |
45b4233c DW |
244 | pr_debug("remove_hash(), stripe %llu\n", |
245 | (unsigned long long)sh->sector); | |
1da177e4 | 246 | |
fccddba0 | 247 | hlist_del_init(&sh->hash); |
1da177e4 LT |
248 | } |
249 | ||
16a53ecc | 250 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 251 | { |
fccddba0 | 252 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 253 | |
45b4233c DW |
254 | pr_debug("insert_hash(), stripe %llu\n", |
255 | (unsigned long long)sh->sector); | |
1da177e4 LT |
256 | |
257 | CHECK_DEVLOCK(); | |
fccddba0 | 258 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
259 | } |
260 | ||
261 | ||
262 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
263 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
264 | { | |
265 | struct stripe_head *sh = NULL; | |
266 | struct list_head *first; | |
267 | ||
268 | CHECK_DEVLOCK(); | |
269 | if (list_empty(&conf->inactive_list)) | |
270 | goto out; | |
271 | first = conf->inactive_list.next; | |
272 | sh = list_entry(first, struct stripe_head, lru); | |
273 | list_del_init(first); | |
274 | remove_hash(sh); | |
275 | atomic_inc(&conf->active_stripes); | |
276 | out: | |
277 | return sh; | |
278 | } | |
279 | ||
280 | static void shrink_buffers(struct stripe_head *sh, int num) | |
281 | { | |
282 | struct page *p; | |
283 | int i; | |
284 | ||
285 | for (i=0; i<num ; i++) { | |
286 | p = sh->dev[i].page; | |
287 | if (!p) | |
288 | continue; | |
289 | sh->dev[i].page = NULL; | |
2d1f3b5d | 290 | put_page(p); |
1da177e4 LT |
291 | } |
292 | } | |
293 | ||
294 | static int grow_buffers(struct stripe_head *sh, int num) | |
295 | { | |
296 | int i; | |
297 | ||
298 | for (i=0; i<num; i++) { | |
299 | struct page *page; | |
300 | ||
301 | if (!(page = alloc_page(GFP_KERNEL))) { | |
302 | return 1; | |
303 | } | |
304 | sh->dev[i].page = page; | |
305 | } | |
306 | return 0; | |
307 | } | |
308 | ||
784052ec | 309 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
911d4ee8 N |
310 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
311 | struct stripe_head *sh); | |
1da177e4 | 312 | |
b5663ba4 | 313 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
314 | { |
315 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 316 | int i; |
1da177e4 | 317 | |
78bafebd ES |
318 | BUG_ON(atomic_read(&sh->count) != 0); |
319 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 320 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 321 | |
1da177e4 | 322 | CHECK_DEVLOCK(); |
45b4233c | 323 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
324 | (unsigned long long)sh->sector); |
325 | ||
326 | remove_hash(sh); | |
16a53ecc | 327 | |
86b42c71 | 328 | sh->generation = conf->generation - previous; |
b5663ba4 | 329 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 330 | sh->sector = sector; |
911d4ee8 | 331 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
332 | sh->state = 0; |
333 | ||
7ecaa1e6 N |
334 | |
335 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
336 | struct r5dev *dev = &sh->dev[i]; |
337 | ||
d84e0f10 | 338 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 339 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 340 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 341 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 342 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
343 | test_bit(R5_LOCKED, &dev->flags)); |
344 | BUG(); | |
345 | } | |
346 | dev->flags = 0; | |
784052ec | 347 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
348 | } |
349 | insert_hash(conf, sh); | |
350 | } | |
351 | ||
86b42c71 N |
352 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, |
353 | short generation) | |
1da177e4 LT |
354 | { |
355 | struct stripe_head *sh; | |
fccddba0 | 356 | struct hlist_node *hn; |
1da177e4 LT |
357 | |
358 | CHECK_DEVLOCK(); | |
45b4233c | 359 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 360 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 361 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 362 | return sh; |
45b4233c | 363 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
364 | return NULL; |
365 | } | |
366 | ||
367 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 368 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 369 | |
b5663ba4 N |
370 | static struct stripe_head * |
371 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
a8c906ca | 372 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
373 | { |
374 | struct stripe_head *sh; | |
375 | ||
45b4233c | 376 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
377 | |
378 | spin_lock_irq(&conf->device_lock); | |
379 | ||
380 | do { | |
72626685 | 381 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 382 | conf->quiesce == 0 || noquiesce, |
72626685 | 383 | conf->device_lock, /* nothing */); |
86b42c71 | 384 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
385 | if (!sh) { |
386 | if (!conf->inactive_blocked) | |
387 | sh = get_free_stripe(conf); | |
388 | if (noblock && sh == NULL) | |
389 | break; | |
390 | if (!sh) { | |
391 | conf->inactive_blocked = 1; | |
392 | wait_event_lock_irq(conf->wait_for_stripe, | |
393 | !list_empty(&conf->inactive_list) && | |
5036805b N |
394 | (atomic_read(&conf->active_stripes) |
395 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
396 | || !conf->inactive_blocked), |
397 | conf->device_lock, | |
f4370781 | 398 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
399 | ); |
400 | conf->inactive_blocked = 0; | |
401 | } else | |
b5663ba4 | 402 | init_stripe(sh, sector, previous); |
1da177e4 LT |
403 | } else { |
404 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
405 | BUG_ON(!list_empty(&sh->lru) |
406 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
407 | } else { |
408 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
409 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
410 | if (list_empty(&sh->lru) && |
411 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
412 | BUG(); |
413 | list_del_init(&sh->lru); | |
1da177e4 LT |
414 | } |
415 | } | |
416 | } while (sh == NULL); | |
417 | ||
418 | if (sh) | |
419 | atomic_inc(&sh->count); | |
420 | ||
421 | spin_unlock_irq(&conf->device_lock); | |
422 | return sh; | |
423 | } | |
424 | ||
6712ecf8 N |
425 | static void |
426 | raid5_end_read_request(struct bio *bi, int error); | |
427 | static void | |
428 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 429 | |
c4e5ac0a | 430 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
431 | { |
432 | raid5_conf_t *conf = sh->raid_conf; | |
433 | int i, disks = sh->disks; | |
434 | ||
435 | might_sleep(); | |
436 | ||
437 | for (i = disks; i--; ) { | |
438 | int rw; | |
439 | struct bio *bi; | |
440 | mdk_rdev_t *rdev; | |
441 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
442 | rw = WRITE; | |
443 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
444 | rw = READ; | |
445 | else | |
446 | continue; | |
447 | ||
448 | bi = &sh->dev[i].req; | |
449 | ||
450 | bi->bi_rw = rw; | |
451 | if (rw == WRITE) | |
452 | bi->bi_end_io = raid5_end_write_request; | |
453 | else | |
454 | bi->bi_end_io = raid5_end_read_request; | |
455 | ||
456 | rcu_read_lock(); | |
457 | rdev = rcu_dereference(conf->disks[i].rdev); | |
458 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
459 | rdev = NULL; | |
460 | if (rdev) | |
461 | atomic_inc(&rdev->nr_pending); | |
462 | rcu_read_unlock(); | |
463 | ||
464 | if (rdev) { | |
c4e5ac0a | 465 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
466 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
467 | ||
2b7497f0 DW |
468 | set_bit(STRIPE_IO_STARTED, &sh->state); |
469 | ||
91c00924 DW |
470 | bi->bi_bdev = rdev->bdev; |
471 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 472 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
473 | bi->bi_rw, i); |
474 | atomic_inc(&sh->count); | |
475 | bi->bi_sector = sh->sector + rdev->data_offset; | |
476 | bi->bi_flags = 1 << BIO_UPTODATE; | |
477 | bi->bi_vcnt = 1; | |
478 | bi->bi_max_vecs = 1; | |
479 | bi->bi_idx = 0; | |
480 | bi->bi_io_vec = &sh->dev[i].vec; | |
481 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
482 | bi->bi_io_vec[0].bv_offset = 0; | |
483 | bi->bi_size = STRIPE_SIZE; | |
484 | bi->bi_next = NULL; | |
485 | if (rw == WRITE && | |
486 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
487 | atomic_add(STRIPE_SECTORS, | |
488 | &rdev->corrected_errors); | |
489 | generic_make_request(bi); | |
490 | } else { | |
491 | if (rw == WRITE) | |
492 | set_bit(STRIPE_DEGRADED, &sh->state); | |
493 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
494 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
495 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
496 | set_bit(STRIPE_HANDLE, &sh->state); | |
497 | } | |
498 | } | |
499 | } | |
500 | ||
501 | static struct dma_async_tx_descriptor * | |
502 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
503 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
504 | { | |
505 | struct bio_vec *bvl; | |
506 | struct page *bio_page; | |
507 | int i; | |
508 | int page_offset; | |
a08abd8c | 509 | struct async_submit_ctl submit; |
0403e382 | 510 | enum async_tx_flags flags = 0; |
91c00924 DW |
511 | |
512 | if (bio->bi_sector >= sector) | |
513 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
514 | else | |
515 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 516 | |
0403e382 DW |
517 | if (frombio) |
518 | flags |= ASYNC_TX_FENCE; | |
519 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
520 | ||
91c00924 DW |
521 | bio_for_each_segment(bvl, bio, i) { |
522 | int len = bio_iovec_idx(bio, i)->bv_len; | |
523 | int clen; | |
524 | int b_offset = 0; | |
525 | ||
526 | if (page_offset < 0) { | |
527 | b_offset = -page_offset; | |
528 | page_offset += b_offset; | |
529 | len -= b_offset; | |
530 | } | |
531 | ||
532 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
533 | clen = STRIPE_SIZE - page_offset; | |
534 | else | |
535 | clen = len; | |
536 | ||
537 | if (clen > 0) { | |
538 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
539 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
540 | if (frombio) | |
541 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 542 | b_offset, clen, &submit); |
91c00924 DW |
543 | else |
544 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 545 | page_offset, clen, &submit); |
91c00924 | 546 | } |
a08abd8c DW |
547 | /* chain the operations */ |
548 | submit.depend_tx = tx; | |
549 | ||
91c00924 DW |
550 | if (clen < len) /* hit end of page */ |
551 | break; | |
552 | page_offset += len; | |
553 | } | |
554 | ||
555 | return tx; | |
556 | } | |
557 | ||
558 | static void ops_complete_biofill(void *stripe_head_ref) | |
559 | { | |
560 | struct stripe_head *sh = stripe_head_ref; | |
561 | struct bio *return_bi = NULL; | |
562 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 563 | int i; |
91c00924 | 564 | |
e46b272b | 565 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
566 | (unsigned long long)sh->sector); |
567 | ||
568 | /* clear completed biofills */ | |
83de75cc | 569 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
570 | for (i = sh->disks; i--; ) { |
571 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
572 | |
573 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
574 | /* and check if we need to reply to a read request, |
575 | * new R5_Wantfill requests are held off until | |
83de75cc | 576 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
577 | */ |
578 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 579 | struct bio *rbi, *rbi2; |
91c00924 | 580 | |
91c00924 DW |
581 | BUG_ON(!dev->read); |
582 | rbi = dev->read; | |
583 | dev->read = NULL; | |
584 | while (rbi && rbi->bi_sector < | |
585 | dev->sector + STRIPE_SECTORS) { | |
586 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 587 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
588 | rbi->bi_next = return_bi; |
589 | return_bi = rbi; | |
590 | } | |
91c00924 DW |
591 | rbi = rbi2; |
592 | } | |
593 | } | |
594 | } | |
83de75cc DW |
595 | spin_unlock_irq(&conf->device_lock); |
596 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
597 | |
598 | return_io(return_bi); | |
599 | ||
e4d84909 | 600 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
601 | release_stripe(sh); |
602 | } | |
603 | ||
604 | static void ops_run_biofill(struct stripe_head *sh) | |
605 | { | |
606 | struct dma_async_tx_descriptor *tx = NULL; | |
607 | raid5_conf_t *conf = sh->raid_conf; | |
a08abd8c | 608 | struct async_submit_ctl submit; |
91c00924 DW |
609 | int i; |
610 | ||
e46b272b | 611 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
612 | (unsigned long long)sh->sector); |
613 | ||
614 | for (i = sh->disks; i--; ) { | |
615 | struct r5dev *dev = &sh->dev[i]; | |
616 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
617 | struct bio *rbi; | |
618 | spin_lock_irq(&conf->device_lock); | |
619 | dev->read = rbi = dev->toread; | |
620 | dev->toread = NULL; | |
621 | spin_unlock_irq(&conf->device_lock); | |
622 | while (rbi && rbi->bi_sector < | |
623 | dev->sector + STRIPE_SECTORS) { | |
624 | tx = async_copy_data(0, rbi, dev->page, | |
625 | dev->sector, tx); | |
626 | rbi = r5_next_bio(rbi, dev->sector); | |
627 | } | |
628 | } | |
629 | } | |
630 | ||
631 | atomic_inc(&sh->count); | |
a08abd8c DW |
632 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
633 | async_trigger_callback(&submit); | |
91c00924 DW |
634 | } |
635 | ||
4e7d2c0a | 636 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 637 | { |
4e7d2c0a | 638 | struct r5dev *tgt; |
91c00924 | 639 | |
4e7d2c0a DW |
640 | if (target < 0) |
641 | return; | |
91c00924 | 642 | |
4e7d2c0a | 643 | tgt = &sh->dev[target]; |
91c00924 DW |
644 | set_bit(R5_UPTODATE, &tgt->flags); |
645 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
646 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
647 | } |
648 | ||
ac6b53b6 | 649 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
650 | { |
651 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 652 | |
e46b272b | 653 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
654 | (unsigned long long)sh->sector); |
655 | ||
ac6b53b6 | 656 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 657 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 658 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 659 | |
ecc65c9b DW |
660 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
661 | if (sh->check_state == check_state_compute_run) | |
662 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
663 | set_bit(STRIPE_HANDLE, &sh->state); |
664 | release_stripe(sh); | |
665 | } | |
666 | ||
d6f38f31 DW |
667 | /* return a pointer to the address conversion region of the scribble buffer */ |
668 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
669 | struct raid5_percpu *percpu) | |
670 | { | |
671 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
672 | } | |
673 | ||
674 | static struct dma_async_tx_descriptor * | |
675 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 676 | { |
91c00924 | 677 | int disks = sh->disks; |
d6f38f31 | 678 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
679 | int target = sh->ops.target; |
680 | struct r5dev *tgt = &sh->dev[target]; | |
681 | struct page *xor_dest = tgt->page; | |
682 | int count = 0; | |
683 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 684 | struct async_submit_ctl submit; |
91c00924 DW |
685 | int i; |
686 | ||
687 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 688 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
689 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
690 | ||
691 | for (i = disks; i--; ) | |
692 | if (i != target) | |
693 | xor_srcs[count++] = sh->dev[i].page; | |
694 | ||
695 | atomic_inc(&sh->count); | |
696 | ||
0403e382 | 697 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 698 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 699 | if (unlikely(count == 1)) |
a08abd8c | 700 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 701 | else |
a08abd8c | 702 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 703 | |
91c00924 DW |
704 | return tx; |
705 | } | |
706 | ||
ac6b53b6 DW |
707 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
708 | * @srcs - (struct page *) array of size sh->disks | |
709 | * @sh - stripe_head to parse | |
710 | * | |
711 | * Populates srcs in proper layout order for the stripe and returns the | |
712 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
713 | * destination buffer is recorded in srcs[count] and the Q destination | |
714 | * is recorded in srcs[count+1]]. | |
715 | */ | |
716 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
717 | { | |
718 | int disks = sh->disks; | |
719 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
720 | int d0_idx = raid6_d0(sh); | |
721 | int count; | |
722 | int i; | |
723 | ||
724 | for (i = 0; i < disks; i++) | |
5dd33c9a | 725 | srcs[i] = NULL; |
ac6b53b6 DW |
726 | |
727 | count = 0; | |
728 | i = d0_idx; | |
729 | do { | |
730 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
731 | ||
732 | srcs[slot] = sh->dev[i].page; | |
733 | i = raid6_next_disk(i, disks); | |
734 | } while (i != d0_idx); | |
ac6b53b6 | 735 | |
e4424fee | 736 | return syndrome_disks; |
ac6b53b6 DW |
737 | } |
738 | ||
739 | static struct dma_async_tx_descriptor * | |
740 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
741 | { | |
742 | int disks = sh->disks; | |
743 | struct page **blocks = percpu->scribble; | |
744 | int target; | |
745 | int qd_idx = sh->qd_idx; | |
746 | struct dma_async_tx_descriptor *tx; | |
747 | struct async_submit_ctl submit; | |
748 | struct r5dev *tgt; | |
749 | struct page *dest; | |
750 | int i; | |
751 | int count; | |
752 | ||
753 | if (sh->ops.target < 0) | |
754 | target = sh->ops.target2; | |
755 | else if (sh->ops.target2 < 0) | |
756 | target = sh->ops.target; | |
91c00924 | 757 | else |
ac6b53b6 DW |
758 | /* we should only have one valid target */ |
759 | BUG(); | |
760 | BUG_ON(target < 0); | |
761 | pr_debug("%s: stripe %llu block: %d\n", | |
762 | __func__, (unsigned long long)sh->sector, target); | |
763 | ||
764 | tgt = &sh->dev[target]; | |
765 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
766 | dest = tgt->page; | |
767 | ||
768 | atomic_inc(&sh->count); | |
769 | ||
770 | if (target == qd_idx) { | |
771 | count = set_syndrome_sources(blocks, sh); | |
772 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
773 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
774 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
775 | ops_complete_compute, sh, | |
ac6b53b6 DW |
776 | to_addr_conv(sh, percpu)); |
777 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
778 | } else { | |
779 | /* Compute any data- or p-drive using XOR */ | |
780 | count = 0; | |
781 | for (i = disks; i-- ; ) { | |
782 | if (i == target || i == qd_idx) | |
783 | continue; | |
784 | blocks[count++] = sh->dev[i].page; | |
785 | } | |
786 | ||
0403e382 DW |
787 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
788 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
789 | to_addr_conv(sh, percpu)); |
790 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
791 | } | |
91c00924 | 792 | |
91c00924 DW |
793 | return tx; |
794 | } | |
795 | ||
ac6b53b6 DW |
796 | static struct dma_async_tx_descriptor * |
797 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
798 | { | |
799 | int i, count, disks = sh->disks; | |
800 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
801 | int d0_idx = raid6_d0(sh); | |
802 | int faila = -1, failb = -1; | |
803 | int target = sh->ops.target; | |
804 | int target2 = sh->ops.target2; | |
805 | struct r5dev *tgt = &sh->dev[target]; | |
806 | struct r5dev *tgt2 = &sh->dev[target2]; | |
807 | struct dma_async_tx_descriptor *tx; | |
808 | struct page **blocks = percpu->scribble; | |
809 | struct async_submit_ctl submit; | |
810 | ||
811 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
812 | __func__, (unsigned long long)sh->sector, target, target2); | |
813 | BUG_ON(target < 0 || target2 < 0); | |
814 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
815 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
816 | ||
6c910a78 | 817 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
818 | * slot number conversion for 'faila' and 'failb' |
819 | */ | |
820 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 821 | blocks[i] = NULL; |
ac6b53b6 DW |
822 | count = 0; |
823 | i = d0_idx; | |
824 | do { | |
825 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
826 | ||
827 | blocks[slot] = sh->dev[i].page; | |
828 | ||
829 | if (i == target) | |
830 | faila = slot; | |
831 | if (i == target2) | |
832 | failb = slot; | |
833 | i = raid6_next_disk(i, disks); | |
834 | } while (i != d0_idx); | |
ac6b53b6 DW |
835 | |
836 | BUG_ON(faila == failb); | |
837 | if (failb < faila) | |
838 | swap(faila, failb); | |
839 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
840 | __func__, (unsigned long long)sh->sector, faila, failb); | |
841 | ||
842 | atomic_inc(&sh->count); | |
843 | ||
844 | if (failb == syndrome_disks+1) { | |
845 | /* Q disk is one of the missing disks */ | |
846 | if (faila == syndrome_disks) { | |
847 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
848 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
849 | ops_complete_compute, sh, | |
850 | to_addr_conv(sh, percpu)); | |
e4424fee | 851 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
852 | STRIPE_SIZE, &submit); |
853 | } else { | |
854 | struct page *dest; | |
855 | int data_target; | |
856 | int qd_idx = sh->qd_idx; | |
857 | ||
858 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
859 | if (target == qd_idx) | |
860 | data_target = target2; | |
861 | else | |
862 | data_target = target; | |
863 | ||
864 | count = 0; | |
865 | for (i = disks; i-- ; ) { | |
866 | if (i == data_target || i == qd_idx) | |
867 | continue; | |
868 | blocks[count++] = sh->dev[i].page; | |
869 | } | |
870 | dest = sh->dev[data_target].page; | |
0403e382 DW |
871 | init_async_submit(&submit, |
872 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
873 | NULL, NULL, NULL, | |
874 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
875 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
876 | &submit); | |
877 | ||
878 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
879 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
880 | ops_complete_compute, sh, | |
881 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
882 | return async_gen_syndrome(blocks, 0, count+2, |
883 | STRIPE_SIZE, &submit); | |
884 | } | |
ac6b53b6 | 885 | } else { |
6c910a78 DW |
886 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
887 | ops_complete_compute, sh, | |
888 | to_addr_conv(sh, percpu)); | |
889 | if (failb == syndrome_disks) { | |
890 | /* We're missing D+P. */ | |
891 | return async_raid6_datap_recov(syndrome_disks+2, | |
892 | STRIPE_SIZE, faila, | |
893 | blocks, &submit); | |
894 | } else { | |
895 | /* We're missing D+D. */ | |
896 | return async_raid6_2data_recov(syndrome_disks+2, | |
897 | STRIPE_SIZE, faila, failb, | |
898 | blocks, &submit); | |
899 | } | |
ac6b53b6 DW |
900 | } |
901 | } | |
902 | ||
903 | ||
91c00924 DW |
904 | static void ops_complete_prexor(void *stripe_head_ref) |
905 | { | |
906 | struct stripe_head *sh = stripe_head_ref; | |
907 | ||
e46b272b | 908 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 909 | (unsigned long long)sh->sector); |
91c00924 DW |
910 | } |
911 | ||
912 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
913 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
914 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 915 | { |
91c00924 | 916 | int disks = sh->disks; |
d6f38f31 | 917 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 918 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 919 | struct async_submit_ctl submit; |
91c00924 DW |
920 | |
921 | /* existing parity data subtracted */ | |
922 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
923 | ||
e46b272b | 924 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
925 | (unsigned long long)sh->sector); |
926 | ||
927 | for (i = disks; i--; ) { | |
928 | struct r5dev *dev = &sh->dev[i]; | |
929 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 930 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
931 | xor_srcs[count++] = dev->page; |
932 | } | |
933 | ||
0403e382 | 934 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 935 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 936 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
937 | |
938 | return tx; | |
939 | } | |
940 | ||
941 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 942 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
943 | { |
944 | int disks = sh->disks; | |
d8ee0728 | 945 | int i; |
91c00924 | 946 | |
e46b272b | 947 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
948 | (unsigned long long)sh->sector); |
949 | ||
950 | for (i = disks; i--; ) { | |
951 | struct r5dev *dev = &sh->dev[i]; | |
952 | struct bio *chosen; | |
91c00924 | 953 | |
d8ee0728 | 954 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
955 | struct bio *wbi; |
956 | ||
957 | spin_lock(&sh->lock); | |
958 | chosen = dev->towrite; | |
959 | dev->towrite = NULL; | |
960 | BUG_ON(dev->written); | |
961 | wbi = dev->written = chosen; | |
962 | spin_unlock(&sh->lock); | |
963 | ||
964 | while (wbi && wbi->bi_sector < | |
965 | dev->sector + STRIPE_SECTORS) { | |
966 | tx = async_copy_data(1, wbi, dev->page, | |
967 | dev->sector, tx); | |
968 | wbi = r5_next_bio(wbi, dev->sector); | |
969 | } | |
970 | } | |
971 | } | |
972 | ||
973 | return tx; | |
974 | } | |
975 | ||
ac6b53b6 | 976 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
977 | { |
978 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
979 | int disks = sh->disks; |
980 | int pd_idx = sh->pd_idx; | |
981 | int qd_idx = sh->qd_idx; | |
982 | int i; | |
91c00924 | 983 | |
e46b272b | 984 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
985 | (unsigned long long)sh->sector); |
986 | ||
987 | for (i = disks; i--; ) { | |
988 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 DW |
989 | |
990 | if (dev->written || i == pd_idx || i == qd_idx) | |
91c00924 DW |
991 | set_bit(R5_UPTODATE, &dev->flags); |
992 | } | |
993 | ||
d8ee0728 DW |
994 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
995 | sh->reconstruct_state = reconstruct_state_drain_result; | |
996 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
997 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
998 | else { | |
999 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1000 | sh->reconstruct_state = reconstruct_state_result; | |
1001 | } | |
91c00924 DW |
1002 | |
1003 | set_bit(STRIPE_HANDLE, &sh->state); | |
1004 | release_stripe(sh); | |
1005 | } | |
1006 | ||
1007 | static void | |
ac6b53b6 DW |
1008 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1009 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1010 | { |
91c00924 | 1011 | int disks = sh->disks; |
d6f38f31 | 1012 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1013 | struct async_submit_ctl submit; |
91c00924 DW |
1014 | int count = 0, pd_idx = sh->pd_idx, i; |
1015 | struct page *xor_dest; | |
d8ee0728 | 1016 | int prexor = 0; |
91c00924 | 1017 | unsigned long flags; |
91c00924 | 1018 | |
e46b272b | 1019 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1020 | (unsigned long long)sh->sector); |
1021 | ||
1022 | /* check if prexor is active which means only process blocks | |
1023 | * that are part of a read-modify-write (written) | |
1024 | */ | |
d8ee0728 DW |
1025 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1026 | prexor = 1; | |
91c00924 DW |
1027 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1028 | for (i = disks; i--; ) { | |
1029 | struct r5dev *dev = &sh->dev[i]; | |
1030 | if (dev->written) | |
1031 | xor_srcs[count++] = dev->page; | |
1032 | } | |
1033 | } else { | |
1034 | xor_dest = sh->dev[pd_idx].page; | |
1035 | for (i = disks; i--; ) { | |
1036 | struct r5dev *dev = &sh->dev[i]; | |
1037 | if (i != pd_idx) | |
1038 | xor_srcs[count++] = dev->page; | |
1039 | } | |
1040 | } | |
1041 | ||
91c00924 DW |
1042 | /* 1/ if we prexor'd then the dest is reused as a source |
1043 | * 2/ if we did not prexor then we are redoing the parity | |
1044 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1045 | * for the synchronous xor case | |
1046 | */ | |
88ba2aa5 | 1047 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1048 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1049 | ||
1050 | atomic_inc(&sh->count); | |
1051 | ||
ac6b53b6 | 1052 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1053 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1054 | if (unlikely(count == 1)) |
1055 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1056 | else | |
1057 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1058 | } |
1059 | ||
ac6b53b6 DW |
1060 | static void |
1061 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1062 | struct dma_async_tx_descriptor *tx) | |
1063 | { | |
1064 | struct async_submit_ctl submit; | |
1065 | struct page **blocks = percpu->scribble; | |
1066 | int count; | |
1067 | ||
1068 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1069 | ||
1070 | count = set_syndrome_sources(blocks, sh); | |
1071 | ||
1072 | atomic_inc(&sh->count); | |
1073 | ||
1074 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1075 | sh, to_addr_conv(sh, percpu)); | |
1076 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1077 | } |
1078 | ||
1079 | static void ops_complete_check(void *stripe_head_ref) | |
1080 | { | |
1081 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1082 | |
e46b272b | 1083 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1084 | (unsigned long long)sh->sector); |
1085 | ||
ecc65c9b | 1086 | sh->check_state = check_state_check_result; |
91c00924 DW |
1087 | set_bit(STRIPE_HANDLE, &sh->state); |
1088 | release_stripe(sh); | |
1089 | } | |
1090 | ||
ac6b53b6 | 1091 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1092 | { |
91c00924 | 1093 | int disks = sh->disks; |
ac6b53b6 DW |
1094 | int pd_idx = sh->pd_idx; |
1095 | int qd_idx = sh->qd_idx; | |
1096 | struct page *xor_dest; | |
d6f38f31 | 1097 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1098 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1099 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1100 | int count; |
1101 | int i; | |
91c00924 | 1102 | |
e46b272b | 1103 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1104 | (unsigned long long)sh->sector); |
1105 | ||
ac6b53b6 DW |
1106 | count = 0; |
1107 | xor_dest = sh->dev[pd_idx].page; | |
1108 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1109 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1110 | if (i == pd_idx || i == qd_idx) |
1111 | continue; | |
1112 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1113 | } |
1114 | ||
d6f38f31 DW |
1115 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1116 | to_addr_conv(sh, percpu)); | |
099f53cb | 1117 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1118 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1119 | |
91c00924 | 1120 | atomic_inc(&sh->count); |
a08abd8c DW |
1121 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1122 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1123 | } |
1124 | ||
ac6b53b6 DW |
1125 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1126 | { | |
1127 | struct page **srcs = percpu->scribble; | |
1128 | struct async_submit_ctl submit; | |
1129 | int count; | |
1130 | ||
1131 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1132 | (unsigned long long)sh->sector, checkp); | |
1133 | ||
1134 | count = set_syndrome_sources(srcs, sh); | |
1135 | if (!checkp) | |
1136 | srcs[count] = NULL; | |
91c00924 | 1137 | |
91c00924 | 1138 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1139 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1140 | sh, to_addr_conv(sh, percpu)); | |
1141 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1142 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1143 | } |
1144 | ||
417b8d4a | 1145 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1146 | { |
1147 | int overlap_clear = 0, i, disks = sh->disks; | |
1148 | struct dma_async_tx_descriptor *tx = NULL; | |
d6f38f31 | 1149 | raid5_conf_t *conf = sh->raid_conf; |
ac6b53b6 | 1150 | int level = conf->level; |
d6f38f31 DW |
1151 | struct raid5_percpu *percpu; |
1152 | unsigned long cpu; | |
91c00924 | 1153 | |
d6f38f31 DW |
1154 | cpu = get_cpu(); |
1155 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1156 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1157 | ops_run_biofill(sh); |
1158 | overlap_clear++; | |
1159 | } | |
1160 | ||
7b3a871e | 1161 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1162 | if (level < 6) |
1163 | tx = ops_run_compute5(sh, percpu); | |
1164 | else { | |
1165 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1166 | tx = ops_run_compute6_1(sh, percpu); | |
1167 | else | |
1168 | tx = ops_run_compute6_2(sh, percpu); | |
1169 | } | |
1170 | /* terminate the chain if reconstruct is not set to be run */ | |
1171 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1172 | async_tx_ack(tx); |
1173 | } | |
91c00924 | 1174 | |
600aa109 | 1175 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1176 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1177 | |
600aa109 | 1178 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1179 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1180 | overlap_clear++; |
1181 | } | |
1182 | ||
ac6b53b6 DW |
1183 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1184 | if (level < 6) | |
1185 | ops_run_reconstruct5(sh, percpu, tx); | |
1186 | else | |
1187 | ops_run_reconstruct6(sh, percpu, tx); | |
1188 | } | |
91c00924 | 1189 | |
ac6b53b6 DW |
1190 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1191 | if (sh->check_state == check_state_run) | |
1192 | ops_run_check_p(sh, percpu); | |
1193 | else if (sh->check_state == check_state_run_q) | |
1194 | ops_run_check_pq(sh, percpu, 0); | |
1195 | else if (sh->check_state == check_state_run_pq) | |
1196 | ops_run_check_pq(sh, percpu, 1); | |
1197 | else | |
1198 | BUG(); | |
1199 | } | |
91c00924 | 1200 | |
91c00924 DW |
1201 | if (overlap_clear) |
1202 | for (i = disks; i--; ) { | |
1203 | struct r5dev *dev = &sh->dev[i]; | |
1204 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1205 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1206 | } | |
d6f38f31 | 1207 | put_cpu(); |
91c00924 DW |
1208 | } |
1209 | ||
417b8d4a DW |
1210 | #ifdef CONFIG_MULTICORE_RAID456 |
1211 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1212 | { | |
1213 | struct stripe_head *sh = param; | |
1214 | unsigned long ops_request = sh->ops.request; | |
1215 | ||
1216 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1217 | wake_up(&sh->ops.wait_for_ops); | |
1218 | ||
1219 | __raid_run_ops(sh, ops_request); | |
1220 | release_stripe(sh); | |
1221 | } | |
1222 | ||
1223 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1224 | { | |
1225 | /* since handle_stripe can be called outside of raid5d context | |
1226 | * we need to ensure sh->ops.request is de-staged before another | |
1227 | * request arrives | |
1228 | */ | |
1229 | wait_event(sh->ops.wait_for_ops, | |
1230 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1231 | sh->ops.request = ops_request; | |
1232 | ||
1233 | atomic_inc(&sh->count); | |
1234 | async_schedule(async_run_ops, sh); | |
1235 | } | |
1236 | #else | |
1237 | #define raid_run_ops __raid_run_ops | |
1238 | #endif | |
1239 | ||
3f294f4f | 1240 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1241 | { |
1242 | struct stripe_head *sh; | |
5e5e3e78 | 1243 | int disks = max(conf->raid_disks, conf->previous_raid_disks); |
3f294f4f N |
1244 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
1245 | if (!sh) | |
1246 | return 0; | |
5e5e3e78 | 1247 | memset(sh, 0, sizeof(*sh) + (disks-1)*sizeof(struct r5dev)); |
3f294f4f N |
1248 | sh->raid_conf = conf; |
1249 | spin_lock_init(&sh->lock); | |
417b8d4a DW |
1250 | #ifdef CONFIG_MULTICORE_RAID456 |
1251 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1252 | #endif | |
3f294f4f | 1253 | |
5e5e3e78 N |
1254 | if (grow_buffers(sh, disks)) { |
1255 | shrink_buffers(sh, disks); | |
3f294f4f N |
1256 | kmem_cache_free(conf->slab_cache, sh); |
1257 | return 0; | |
1258 | } | |
1259 | /* we just created an active stripe so... */ | |
1260 | atomic_set(&sh->count, 1); | |
1261 | atomic_inc(&conf->active_stripes); | |
1262 | INIT_LIST_HEAD(&sh->lru); | |
1263 | release_stripe(sh); | |
1264 | return 1; | |
1265 | } | |
1266 | ||
1267 | static int grow_stripes(raid5_conf_t *conf, int num) | |
1268 | { | |
e18b890b | 1269 | struct kmem_cache *sc; |
5e5e3e78 | 1270 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1271 | |
245f46c2 N |
1272 | sprintf(conf->cache_name[0], |
1273 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1274 | sprintf(conf->cache_name[1], | |
1275 | "raid%d-%s-alt", conf->level, mdname(conf->mddev)); | |
ad01c9e3 N |
1276 | conf->active_name = 0; |
1277 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1278 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1279 | 0, 0, NULL); |
1da177e4 LT |
1280 | if (!sc) |
1281 | return 1; | |
1282 | conf->slab_cache = sc; | |
ad01c9e3 | 1283 | conf->pool_size = devs; |
16a53ecc | 1284 | while (num--) |
3f294f4f | 1285 | if (!grow_one_stripe(conf)) |
1da177e4 | 1286 | return 1; |
1da177e4 LT |
1287 | return 0; |
1288 | } | |
29269553 | 1289 | |
d6f38f31 DW |
1290 | /** |
1291 | * scribble_len - return the required size of the scribble region | |
1292 | * @num - total number of disks in the array | |
1293 | * | |
1294 | * The size must be enough to contain: | |
1295 | * 1/ a struct page pointer for each device in the array +2 | |
1296 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1297 | * (dma_map_page()) or page (page_address()) address. | |
1298 | * | |
1299 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1300 | * calculate over all devices (not just the data blocks), using zeros in place | |
1301 | * of the P and Q blocks. | |
1302 | */ | |
1303 | static size_t scribble_len(int num) | |
1304 | { | |
1305 | size_t len; | |
1306 | ||
1307 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1308 | ||
1309 | return len; | |
1310 | } | |
1311 | ||
ad01c9e3 N |
1312 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
1313 | { | |
1314 | /* Make all the stripes able to hold 'newsize' devices. | |
1315 | * New slots in each stripe get 'page' set to a new page. | |
1316 | * | |
1317 | * This happens in stages: | |
1318 | * 1/ create a new kmem_cache and allocate the required number of | |
1319 | * stripe_heads. | |
1320 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1321 | * to the new stripe_heads. This will have the side effect of | |
1322 | * freezing the array as once all stripe_heads have been collected, | |
1323 | * no IO will be possible. Old stripe heads are freed once their | |
1324 | * pages have been transferred over, and the old kmem_cache is | |
1325 | * freed when all stripes are done. | |
1326 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1327 | * we simple return a failre status - no need to clean anything up. | |
1328 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1329 | * If this fails, we don't bother trying the shrink the | |
1330 | * stripe_heads down again, we just leave them as they are. | |
1331 | * As each stripe_head is processed the new one is released into | |
1332 | * active service. | |
1333 | * | |
1334 | * Once step2 is started, we cannot afford to wait for a write, | |
1335 | * so we use GFP_NOIO allocations. | |
1336 | */ | |
1337 | struct stripe_head *osh, *nsh; | |
1338 | LIST_HEAD(newstripes); | |
1339 | struct disk_info *ndisks; | |
d6f38f31 | 1340 | unsigned long cpu; |
b5470dc5 | 1341 | int err; |
e18b890b | 1342 | struct kmem_cache *sc; |
ad01c9e3 N |
1343 | int i; |
1344 | ||
1345 | if (newsize <= conf->pool_size) | |
1346 | return 0; /* never bother to shrink */ | |
1347 | ||
b5470dc5 DW |
1348 | err = md_allow_write(conf->mddev); |
1349 | if (err) | |
1350 | return err; | |
2a2275d6 | 1351 | |
ad01c9e3 N |
1352 | /* Step 1 */ |
1353 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1354 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1355 | 0, 0, NULL); |
ad01c9e3 N |
1356 | if (!sc) |
1357 | return -ENOMEM; | |
1358 | ||
1359 | for (i = conf->max_nr_stripes; i; i--) { | |
1360 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
1361 | if (!nsh) | |
1362 | break; | |
1363 | ||
1364 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
1365 | ||
1366 | nsh->raid_conf = conf; | |
1367 | spin_lock_init(&nsh->lock); | |
417b8d4a DW |
1368 | #ifdef CONFIG_MULTICORE_RAID456 |
1369 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1370 | #endif | |
ad01c9e3 N |
1371 | |
1372 | list_add(&nsh->lru, &newstripes); | |
1373 | } | |
1374 | if (i) { | |
1375 | /* didn't get enough, give up */ | |
1376 | while (!list_empty(&newstripes)) { | |
1377 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1378 | list_del(&nsh->lru); | |
1379 | kmem_cache_free(sc, nsh); | |
1380 | } | |
1381 | kmem_cache_destroy(sc); | |
1382 | return -ENOMEM; | |
1383 | } | |
1384 | /* Step 2 - Must use GFP_NOIO now. | |
1385 | * OK, we have enough stripes, start collecting inactive | |
1386 | * stripes and copying them over | |
1387 | */ | |
1388 | list_for_each_entry(nsh, &newstripes, lru) { | |
1389 | spin_lock_irq(&conf->device_lock); | |
1390 | wait_event_lock_irq(conf->wait_for_stripe, | |
1391 | !list_empty(&conf->inactive_list), | |
1392 | conf->device_lock, | |
b3b46be3 | 1393 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1394 | ); |
1395 | osh = get_free_stripe(conf); | |
1396 | spin_unlock_irq(&conf->device_lock); | |
1397 | atomic_set(&nsh->count, 1); | |
1398 | for(i=0; i<conf->pool_size; i++) | |
1399 | nsh->dev[i].page = osh->dev[i].page; | |
1400 | for( ; i<newsize; i++) | |
1401 | nsh->dev[i].page = NULL; | |
1402 | kmem_cache_free(conf->slab_cache, osh); | |
1403 | } | |
1404 | kmem_cache_destroy(conf->slab_cache); | |
1405 | ||
1406 | /* Step 3. | |
1407 | * At this point, we are holding all the stripes so the array | |
1408 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1409 | * conf->disks and the scribble region |
ad01c9e3 N |
1410 | */ |
1411 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1412 | if (ndisks) { | |
1413 | for (i=0; i<conf->raid_disks; i++) | |
1414 | ndisks[i] = conf->disks[i]; | |
1415 | kfree(conf->disks); | |
1416 | conf->disks = ndisks; | |
1417 | } else | |
1418 | err = -ENOMEM; | |
1419 | ||
d6f38f31 DW |
1420 | get_online_cpus(); |
1421 | conf->scribble_len = scribble_len(newsize); | |
1422 | for_each_present_cpu(cpu) { | |
1423 | struct raid5_percpu *percpu; | |
1424 | void *scribble; | |
1425 | ||
1426 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1427 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1428 | ||
1429 | if (scribble) { | |
1430 | kfree(percpu->scribble); | |
1431 | percpu->scribble = scribble; | |
1432 | } else { | |
1433 | err = -ENOMEM; | |
1434 | break; | |
1435 | } | |
1436 | } | |
1437 | put_online_cpus(); | |
1438 | ||
ad01c9e3 N |
1439 | /* Step 4, return new stripes to service */ |
1440 | while(!list_empty(&newstripes)) { | |
1441 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1442 | list_del_init(&nsh->lru); | |
d6f38f31 | 1443 | |
ad01c9e3 N |
1444 | for (i=conf->raid_disks; i < newsize; i++) |
1445 | if (nsh->dev[i].page == NULL) { | |
1446 | struct page *p = alloc_page(GFP_NOIO); | |
1447 | nsh->dev[i].page = p; | |
1448 | if (!p) | |
1449 | err = -ENOMEM; | |
1450 | } | |
1451 | release_stripe(nsh); | |
1452 | } | |
1453 | /* critical section pass, GFP_NOIO no longer needed */ | |
1454 | ||
1455 | conf->slab_cache = sc; | |
1456 | conf->active_name = 1-conf->active_name; | |
1457 | conf->pool_size = newsize; | |
1458 | return err; | |
1459 | } | |
1da177e4 | 1460 | |
3f294f4f | 1461 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1462 | { |
1463 | struct stripe_head *sh; | |
1464 | ||
3f294f4f N |
1465 | spin_lock_irq(&conf->device_lock); |
1466 | sh = get_free_stripe(conf); | |
1467 | spin_unlock_irq(&conf->device_lock); | |
1468 | if (!sh) | |
1469 | return 0; | |
78bafebd | 1470 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1471 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1472 | kmem_cache_free(conf->slab_cache, sh); |
1473 | atomic_dec(&conf->active_stripes); | |
1474 | return 1; | |
1475 | } | |
1476 | ||
1477 | static void shrink_stripes(raid5_conf_t *conf) | |
1478 | { | |
1479 | while (drop_one_stripe(conf)) | |
1480 | ; | |
1481 | ||
29fc7e3e N |
1482 | if (conf->slab_cache) |
1483 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1484 | conf->slab_cache = NULL; |
1485 | } | |
1486 | ||
6712ecf8 | 1487 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1488 | { |
99c0fb5f | 1489 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1490 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1491 | int disks = sh->disks, i; |
1da177e4 | 1492 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1493 | char b[BDEVNAME_SIZE]; |
1494 | mdk_rdev_t *rdev; | |
1da177e4 | 1495 | |
1da177e4 LT |
1496 | |
1497 | for (i=0 ; i<disks; i++) | |
1498 | if (bi == &sh->dev[i].req) | |
1499 | break; | |
1500 | ||
45b4233c DW |
1501 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1502 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1503 | uptodate); |
1504 | if (i == disks) { | |
1505 | BUG(); | |
6712ecf8 | 1506 | return; |
1da177e4 LT |
1507 | } |
1508 | ||
1509 | if (uptodate) { | |
1da177e4 | 1510 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1511 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1512 | rdev = conf->disks[i].rdev; |
0c55e022 | 1513 | printk_rl(KERN_INFO "md/raid:%s: read error corrected" |
6be9d494 BS |
1514 | " (%lu sectors at %llu on %s)\n", |
1515 | mdname(conf->mddev), STRIPE_SECTORS, | |
1516 | (unsigned long long)(sh->sector | |
1517 | + rdev->data_offset), | |
1518 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1519 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1520 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1521 | } | |
ba22dcbf N |
1522 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1523 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1524 | } else { |
d6950432 | 1525 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1526 | int retry = 0; |
d6950432 N |
1527 | rdev = conf->disks[i].rdev; |
1528 | ||
1da177e4 | 1529 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1530 | atomic_inc(&rdev->read_errors); |
7b0bb536 | 1531 | if (conf->mddev->degraded >= conf->max_degraded) |
6be9d494 | 1532 | printk_rl(KERN_WARNING |
0c55e022 | 1533 | "md/raid:%s: read error not correctable " |
6be9d494 BS |
1534 | "(sector %llu on %s).\n", |
1535 | mdname(conf->mddev), | |
1536 | (unsigned long long)(sh->sector | |
1537 | + rdev->data_offset), | |
1538 | bdn); | |
ba22dcbf | 1539 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1540 | /* Oh, no!!! */ |
6be9d494 | 1541 | printk_rl(KERN_WARNING |
0c55e022 | 1542 | "md/raid:%s: read error NOT corrected!! " |
6be9d494 BS |
1543 | "(sector %llu on %s).\n", |
1544 | mdname(conf->mddev), | |
1545 | (unsigned long long)(sh->sector | |
1546 | + rdev->data_offset), | |
1547 | bdn); | |
d6950432 | 1548 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1549 | > conf->max_nr_stripes) |
14f8d26b | 1550 | printk(KERN_WARNING |
0c55e022 | 1551 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1552 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1553 | else |
1554 | retry = 1; | |
1555 | if (retry) | |
1556 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1557 | else { | |
4e5314b5 N |
1558 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1559 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1560 | md_error(conf->mddev, rdev); |
ba22dcbf | 1561 | } |
1da177e4 LT |
1562 | } |
1563 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1564 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1565 | set_bit(STRIPE_HANDLE, &sh->state); | |
1566 | release_stripe(sh); | |
1da177e4 LT |
1567 | } |
1568 | ||
d710e138 | 1569 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1570 | { |
99c0fb5f | 1571 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1572 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1573 | int disks = sh->disks, i; |
1da177e4 LT |
1574 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1575 | ||
1da177e4 LT |
1576 | for (i=0 ; i<disks; i++) |
1577 | if (bi == &sh->dev[i].req) | |
1578 | break; | |
1579 | ||
45b4233c | 1580 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1581 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1582 | uptodate); | |
1583 | if (i == disks) { | |
1584 | BUG(); | |
6712ecf8 | 1585 | return; |
1da177e4 LT |
1586 | } |
1587 | ||
1da177e4 LT |
1588 | if (!uptodate) |
1589 | md_error(conf->mddev, conf->disks[i].rdev); | |
1590 | ||
1591 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1592 | ||
1593 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1594 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1595 | release_stripe(sh); |
1da177e4 LT |
1596 | } |
1597 | ||
1598 | ||
784052ec | 1599 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1600 | |
784052ec | 1601 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1602 | { |
1603 | struct r5dev *dev = &sh->dev[i]; | |
1604 | ||
1605 | bio_init(&dev->req); | |
1606 | dev->req.bi_io_vec = &dev->vec; | |
1607 | dev->req.bi_vcnt++; | |
1608 | dev->req.bi_max_vecs++; | |
1609 | dev->vec.bv_page = dev->page; | |
1610 | dev->vec.bv_len = STRIPE_SIZE; | |
1611 | dev->vec.bv_offset = 0; | |
1612 | ||
1613 | dev->req.bi_sector = sh->sector; | |
1614 | dev->req.bi_private = sh; | |
1615 | ||
1616 | dev->flags = 0; | |
784052ec | 1617 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1618 | } |
1619 | ||
1620 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1621 | { | |
1622 | char b[BDEVNAME_SIZE]; | |
7b92813c | 1623 | raid5_conf_t *conf = mddev->private; |
0c55e022 | 1624 | pr_debug("raid456: error called\n"); |
1da177e4 | 1625 | |
b2d444d7 | 1626 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1627 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1628 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1629 | unsigned long flags; | |
1630 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1631 | mddev->degraded++; |
c04be0aa | 1632 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1633 | /* |
1634 | * if recovery was running, make sure it aborts. | |
1635 | */ | |
dfc70645 | 1636 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1637 | } |
b2d444d7 | 1638 | set_bit(Faulty, &rdev->flags); |
d710e138 | 1639 | printk(KERN_ALERT |
0c55e022 N |
1640 | "md/raid:%s: Disk failure on %s, disabling device.\n" |
1641 | KERN_ALERT | |
1642 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1643 | mdname(mddev), | |
1644 | bdevname(rdev->bdev, b), | |
1645 | mdname(mddev), | |
1646 | conf->raid_disks - mddev->degraded); | |
1da177e4 | 1647 | } |
16a53ecc | 1648 | } |
1da177e4 LT |
1649 | |
1650 | /* | |
1651 | * Input: a 'big' sector number, | |
1652 | * Output: index of the data and parity disk, and the sector # in them. | |
1653 | */ | |
112bf897 | 1654 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1655 | int previous, int *dd_idx, |
1656 | struct stripe_head *sh) | |
1da177e4 | 1657 | { |
6e3b96ed | 1658 | sector_t stripe, stripe2; |
35f2a591 | 1659 | sector_t chunk_number; |
1da177e4 | 1660 | unsigned int chunk_offset; |
911d4ee8 | 1661 | int pd_idx, qd_idx; |
67cc2b81 | 1662 | int ddf_layout = 0; |
1da177e4 | 1663 | sector_t new_sector; |
e183eaed N |
1664 | int algorithm = previous ? conf->prev_algo |
1665 | : conf->algorithm; | |
09c9e5fa AN |
1666 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1667 | : conf->chunk_sectors; | |
112bf897 N |
1668 | int raid_disks = previous ? conf->previous_raid_disks |
1669 | : conf->raid_disks; | |
1670 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1671 | |
1672 | /* First compute the information on this sector */ | |
1673 | ||
1674 | /* | |
1675 | * Compute the chunk number and the sector offset inside the chunk | |
1676 | */ | |
1677 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1678 | chunk_number = r_sector; | |
1da177e4 LT |
1679 | |
1680 | /* | |
1681 | * Compute the stripe number | |
1682 | */ | |
35f2a591 N |
1683 | stripe = chunk_number; |
1684 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1685 | stripe2 = stripe; |
1da177e4 LT |
1686 | /* |
1687 | * Select the parity disk based on the user selected algorithm. | |
1688 | */ | |
911d4ee8 | 1689 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1690 | switch(conf->level) { |
1691 | case 4: | |
911d4ee8 | 1692 | pd_idx = data_disks; |
16a53ecc N |
1693 | break; |
1694 | case 5: | |
e183eaed | 1695 | switch (algorithm) { |
1da177e4 | 1696 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1697 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1698 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1699 | (*dd_idx)++; |
1700 | break; | |
1701 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1702 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1703 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1704 | (*dd_idx)++; |
1705 | break; | |
1706 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1707 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1708 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1709 | break; |
1710 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1711 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1712 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 1713 | break; |
99c0fb5f N |
1714 | case ALGORITHM_PARITY_0: |
1715 | pd_idx = 0; | |
1716 | (*dd_idx)++; | |
1717 | break; | |
1718 | case ALGORITHM_PARITY_N: | |
1719 | pd_idx = data_disks; | |
1720 | break; | |
1da177e4 | 1721 | default: |
99c0fb5f | 1722 | BUG(); |
16a53ecc N |
1723 | } |
1724 | break; | |
1725 | case 6: | |
1726 | ||
e183eaed | 1727 | switch (algorithm) { |
16a53ecc | 1728 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1729 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1730 | qd_idx = pd_idx + 1; |
1731 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1732 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1733 | qd_idx = 0; |
1734 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1735 | (*dd_idx) += 2; /* D D P Q D */ |
1736 | break; | |
1737 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1738 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1739 | qd_idx = pd_idx + 1; |
1740 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1741 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1742 | qd_idx = 0; |
1743 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1744 | (*dd_idx) += 2; /* D D P Q D */ |
1745 | break; | |
1746 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1747 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1748 | qd_idx = (pd_idx + 1) % raid_disks; |
1749 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1750 | break; |
1751 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1752 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1753 | qd_idx = (pd_idx + 1) % raid_disks; |
1754 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1755 | break; |
99c0fb5f N |
1756 | |
1757 | case ALGORITHM_PARITY_0: | |
1758 | pd_idx = 0; | |
1759 | qd_idx = 1; | |
1760 | (*dd_idx) += 2; | |
1761 | break; | |
1762 | case ALGORITHM_PARITY_N: | |
1763 | pd_idx = data_disks; | |
1764 | qd_idx = data_disks + 1; | |
1765 | break; | |
1766 | ||
1767 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1768 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1769 | * of blocks for computing Q is different. | |
1770 | */ | |
6e3b96ed | 1771 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
1772 | qd_idx = pd_idx + 1; |
1773 | if (pd_idx == raid_disks-1) { | |
1774 | (*dd_idx)++; /* Q D D D P */ | |
1775 | qd_idx = 0; | |
1776 | } else if (*dd_idx >= pd_idx) | |
1777 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1778 | ddf_layout = 1; |
99c0fb5f N |
1779 | break; |
1780 | ||
1781 | case ALGORITHM_ROTATING_N_RESTART: | |
1782 | /* Same a left_asymmetric, by first stripe is | |
1783 | * D D D P Q rather than | |
1784 | * Q D D D P | |
1785 | */ | |
6e3b96ed N |
1786 | stripe2 += 1; |
1787 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
1788 | qd_idx = pd_idx + 1; |
1789 | if (pd_idx == raid_disks-1) { | |
1790 | (*dd_idx)++; /* Q D D D P */ | |
1791 | qd_idx = 0; | |
1792 | } else if (*dd_idx >= pd_idx) | |
1793 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1794 | ddf_layout = 1; |
99c0fb5f N |
1795 | break; |
1796 | ||
1797 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1798 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 1799 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
1800 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
1801 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1802 | ddf_layout = 1; |
99c0fb5f N |
1803 | break; |
1804 | ||
1805 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1806 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 1807 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1808 | if (*dd_idx >= pd_idx) |
1809 | (*dd_idx)++; | |
1810 | qd_idx = raid_disks - 1; | |
1811 | break; | |
1812 | ||
1813 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 1814 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1815 | if (*dd_idx >= pd_idx) |
1816 | (*dd_idx)++; | |
1817 | qd_idx = raid_disks - 1; | |
1818 | break; | |
1819 | ||
1820 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 1821 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1822 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1823 | qd_idx = raid_disks - 1; | |
1824 | break; | |
1825 | ||
1826 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 1827 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1828 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1829 | qd_idx = raid_disks - 1; | |
1830 | break; | |
1831 | ||
1832 | case ALGORITHM_PARITY_0_6: | |
1833 | pd_idx = 0; | |
1834 | (*dd_idx)++; | |
1835 | qd_idx = raid_disks - 1; | |
1836 | break; | |
1837 | ||
16a53ecc | 1838 | default: |
99c0fb5f | 1839 | BUG(); |
16a53ecc N |
1840 | } |
1841 | break; | |
1da177e4 LT |
1842 | } |
1843 | ||
911d4ee8 N |
1844 | if (sh) { |
1845 | sh->pd_idx = pd_idx; | |
1846 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1847 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1848 | } |
1da177e4 LT |
1849 | /* |
1850 | * Finally, compute the new sector number | |
1851 | */ | |
1852 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1853 | return new_sector; | |
1854 | } | |
1855 | ||
1856 | ||
784052ec | 1857 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1858 | { |
1859 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1860 | int raid_disks = sh->disks; |
1861 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1862 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
1863 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1864 | : conf->chunk_sectors; | |
e183eaed N |
1865 | int algorithm = previous ? conf->prev_algo |
1866 | : conf->algorithm; | |
1da177e4 LT |
1867 | sector_t stripe; |
1868 | int chunk_offset; | |
35f2a591 N |
1869 | sector_t chunk_number; |
1870 | int dummy1, dd_idx = i; | |
1da177e4 | 1871 | sector_t r_sector; |
911d4ee8 | 1872 | struct stripe_head sh2; |
1da177e4 | 1873 | |
16a53ecc | 1874 | |
1da177e4 LT |
1875 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1876 | stripe = new_sector; | |
1da177e4 | 1877 | |
16a53ecc N |
1878 | if (i == sh->pd_idx) |
1879 | return 0; | |
1880 | switch(conf->level) { | |
1881 | case 4: break; | |
1882 | case 5: | |
e183eaed | 1883 | switch (algorithm) { |
1da177e4 LT |
1884 | case ALGORITHM_LEFT_ASYMMETRIC: |
1885 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1886 | if (i > sh->pd_idx) | |
1887 | i--; | |
1888 | break; | |
1889 | case ALGORITHM_LEFT_SYMMETRIC: | |
1890 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1891 | if (i < sh->pd_idx) | |
1892 | i += raid_disks; | |
1893 | i -= (sh->pd_idx + 1); | |
1894 | break; | |
99c0fb5f N |
1895 | case ALGORITHM_PARITY_0: |
1896 | i -= 1; | |
1897 | break; | |
1898 | case ALGORITHM_PARITY_N: | |
1899 | break; | |
1da177e4 | 1900 | default: |
99c0fb5f | 1901 | BUG(); |
16a53ecc N |
1902 | } |
1903 | break; | |
1904 | case 6: | |
d0dabf7e | 1905 | if (i == sh->qd_idx) |
16a53ecc | 1906 | return 0; /* It is the Q disk */ |
e183eaed | 1907 | switch (algorithm) { |
16a53ecc N |
1908 | case ALGORITHM_LEFT_ASYMMETRIC: |
1909 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1910 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1911 | case ALGORITHM_ROTATING_N_RESTART: | |
1912 | if (sh->pd_idx == raid_disks-1) | |
1913 | i--; /* Q D D D P */ | |
16a53ecc N |
1914 | else if (i > sh->pd_idx) |
1915 | i -= 2; /* D D P Q D */ | |
1916 | break; | |
1917 | case ALGORITHM_LEFT_SYMMETRIC: | |
1918 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1919 | if (sh->pd_idx == raid_disks-1) | |
1920 | i--; /* Q D D D P */ | |
1921 | else { | |
1922 | /* D D P Q D */ | |
1923 | if (i < sh->pd_idx) | |
1924 | i += raid_disks; | |
1925 | i -= (sh->pd_idx + 2); | |
1926 | } | |
1927 | break; | |
99c0fb5f N |
1928 | case ALGORITHM_PARITY_0: |
1929 | i -= 2; | |
1930 | break; | |
1931 | case ALGORITHM_PARITY_N: | |
1932 | break; | |
1933 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 1934 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
1935 | if (sh->pd_idx == 0) |
1936 | i--; /* P D D D Q */ | |
e4424fee N |
1937 | else { |
1938 | /* D D Q P D */ | |
1939 | if (i < sh->pd_idx) | |
1940 | i += raid_disks; | |
1941 | i -= (sh->pd_idx + 1); | |
1942 | } | |
99c0fb5f N |
1943 | break; |
1944 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1945 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1946 | if (i > sh->pd_idx) | |
1947 | i--; | |
1948 | break; | |
1949 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1950 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1951 | if (i < sh->pd_idx) | |
1952 | i += data_disks + 1; | |
1953 | i -= (sh->pd_idx + 1); | |
1954 | break; | |
1955 | case ALGORITHM_PARITY_0_6: | |
1956 | i -= 1; | |
1957 | break; | |
16a53ecc | 1958 | default: |
99c0fb5f | 1959 | BUG(); |
16a53ecc N |
1960 | } |
1961 | break; | |
1da177e4 LT |
1962 | } |
1963 | ||
1964 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 1965 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 1966 | |
112bf897 | 1967 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 1968 | previous, &dummy1, &sh2); |
911d4ee8 N |
1969 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
1970 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
1971 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
1972 | mdname(conf->mddev)); | |
1da177e4 LT |
1973 | return 0; |
1974 | } | |
1975 | return r_sector; | |
1976 | } | |
1977 | ||
1978 | ||
600aa109 | 1979 | static void |
c0f7bddb | 1980 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 1981 | int rcw, int expand) |
e33129d8 DW |
1982 | { |
1983 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
c0f7bddb YT |
1984 | raid5_conf_t *conf = sh->raid_conf; |
1985 | int level = conf->level; | |
e33129d8 DW |
1986 | |
1987 | if (rcw) { | |
1988 | /* if we are not expanding this is a proper write request, and | |
1989 | * there will be bios with new data to be drained into the | |
1990 | * stripe cache | |
1991 | */ | |
1992 | if (!expand) { | |
600aa109 DW |
1993 | sh->reconstruct_state = reconstruct_state_drain_run; |
1994 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1995 | } else | |
1996 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 1997 | |
ac6b53b6 | 1998 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
1999 | |
2000 | for (i = disks; i--; ) { | |
2001 | struct r5dev *dev = &sh->dev[i]; | |
2002 | ||
2003 | if (dev->towrite) { | |
2004 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2005 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2006 | if (!expand) |
2007 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2008 | s->locked++; |
e33129d8 DW |
2009 | } |
2010 | } | |
c0f7bddb | 2011 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2012 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2013 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2014 | } else { |
c0f7bddb | 2015 | BUG_ON(level == 6); |
e33129d8 DW |
2016 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2017 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2018 | ||
d8ee0728 | 2019 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2020 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2021 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2022 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2023 | |
2024 | for (i = disks; i--; ) { | |
2025 | struct r5dev *dev = &sh->dev[i]; | |
2026 | if (i == pd_idx) | |
2027 | continue; | |
2028 | ||
e33129d8 DW |
2029 | if (dev->towrite && |
2030 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2031 | test_bit(R5_Wantcompute, &dev->flags))) { |
2032 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2033 | set_bit(R5_LOCKED, &dev->flags); |
2034 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2035 | s->locked++; |
e33129d8 DW |
2036 | } |
2037 | } | |
2038 | } | |
2039 | ||
c0f7bddb | 2040 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2041 | * are in flight |
2042 | */ | |
2043 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2044 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2045 | s->locked++; |
e33129d8 | 2046 | |
c0f7bddb YT |
2047 | if (level == 6) { |
2048 | int qd_idx = sh->qd_idx; | |
2049 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2050 | ||
2051 | set_bit(R5_LOCKED, &dev->flags); | |
2052 | clear_bit(R5_UPTODATE, &dev->flags); | |
2053 | s->locked++; | |
2054 | } | |
2055 | ||
600aa109 | 2056 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2057 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2058 | s->locked, s->ops_request); |
e33129d8 | 2059 | } |
16a53ecc | 2060 | |
1da177e4 LT |
2061 | /* |
2062 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2063 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2064 | * The bi_next chain must be in order. |
2065 | */ | |
2066 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2067 | { | |
2068 | struct bio **bip; | |
2069 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 2070 | int firstwrite=0; |
1da177e4 | 2071 | |
45b4233c | 2072 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2073 | (unsigned long long)bi->bi_sector, |
2074 | (unsigned long long)sh->sector); | |
2075 | ||
2076 | ||
2077 | spin_lock(&sh->lock); | |
2078 | spin_lock_irq(&conf->device_lock); | |
72626685 | 2079 | if (forwrite) { |
1da177e4 | 2080 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2081 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2082 | firstwrite = 1; | |
2083 | } else | |
1da177e4 LT |
2084 | bip = &sh->dev[dd_idx].toread; |
2085 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2086 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2087 | goto overlap; | |
2088 | bip = & (*bip)->bi_next; | |
2089 | } | |
2090 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2091 | goto overlap; | |
2092 | ||
78bafebd | 2093 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2094 | if (*bip) |
2095 | bi->bi_next = *bip; | |
2096 | *bip = bi; | |
960e739d | 2097 | bi->bi_phys_segments++; |
1da177e4 LT |
2098 | spin_unlock_irq(&conf->device_lock); |
2099 | spin_unlock(&sh->lock); | |
2100 | ||
45b4233c | 2101 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
2102 | (unsigned long long)bi->bi_sector, |
2103 | (unsigned long long)sh->sector, dd_idx); | |
2104 | ||
72626685 | 2105 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
2106 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
2107 | STRIPE_SECTORS, 0); | |
ae3c20cc | 2108 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
2109 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
2110 | } | |
2111 | ||
1da177e4 LT |
2112 | if (forwrite) { |
2113 | /* check if page is covered */ | |
2114 | sector_t sector = sh->dev[dd_idx].sector; | |
2115 | for (bi=sh->dev[dd_idx].towrite; | |
2116 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2117 | bi && bi->bi_sector <= sector; | |
2118 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2119 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2120 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2121 | } | |
2122 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2123 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2124 | } | |
2125 | return 1; | |
2126 | ||
2127 | overlap: | |
2128 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2129 | spin_unlock_irq(&conf->device_lock); | |
2130 | spin_unlock(&sh->lock); | |
2131 | return 0; | |
2132 | } | |
2133 | ||
29269553 N |
2134 | static void end_reshape(raid5_conf_t *conf); |
2135 | ||
911d4ee8 N |
2136 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
2137 | struct stripe_head *sh) | |
ccfcc3c1 | 2138 | { |
784052ec | 2139 | int sectors_per_chunk = |
09c9e5fa | 2140 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2141 | int dd_idx; |
2d2063ce | 2142 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2143 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2144 | |
112bf897 N |
2145 | raid5_compute_sector(conf, |
2146 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2147 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2148 | previous, |
911d4ee8 | 2149 | &dd_idx, sh); |
ccfcc3c1 N |
2150 | } |
2151 | ||
a4456856 | 2152 | static void |
1fe797e6 | 2153 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2154 | struct stripe_head_state *s, int disks, |
2155 | struct bio **return_bi) | |
2156 | { | |
2157 | int i; | |
2158 | for (i = disks; i--; ) { | |
2159 | struct bio *bi; | |
2160 | int bitmap_end = 0; | |
2161 | ||
2162 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2163 | mdk_rdev_t *rdev; | |
2164 | rcu_read_lock(); | |
2165 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2166 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2167 | /* multiple read failures in one stripe */ | |
2168 | md_error(conf->mddev, rdev); | |
2169 | rcu_read_unlock(); | |
2170 | } | |
2171 | spin_lock_irq(&conf->device_lock); | |
2172 | /* fail all writes first */ | |
2173 | bi = sh->dev[i].towrite; | |
2174 | sh->dev[i].towrite = NULL; | |
2175 | if (bi) { | |
2176 | s->to_write--; | |
2177 | bitmap_end = 1; | |
2178 | } | |
2179 | ||
2180 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2181 | wake_up(&conf->wait_for_overlap); | |
2182 | ||
2183 | while (bi && bi->bi_sector < | |
2184 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2185 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2186 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2187 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2188 | md_write_end(conf->mddev); |
2189 | bi->bi_next = *return_bi; | |
2190 | *return_bi = bi; | |
2191 | } | |
2192 | bi = nextbi; | |
2193 | } | |
2194 | /* and fail all 'written' */ | |
2195 | bi = sh->dev[i].written; | |
2196 | sh->dev[i].written = NULL; | |
2197 | if (bi) bitmap_end = 1; | |
2198 | while (bi && bi->bi_sector < | |
2199 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2200 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2201 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2202 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2203 | md_write_end(conf->mddev); |
2204 | bi->bi_next = *return_bi; | |
2205 | *return_bi = bi; | |
2206 | } | |
2207 | bi = bi2; | |
2208 | } | |
2209 | ||
b5e98d65 DW |
2210 | /* fail any reads if this device is non-operational and |
2211 | * the data has not reached the cache yet. | |
2212 | */ | |
2213 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2214 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2215 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2216 | bi = sh->dev[i].toread; |
2217 | sh->dev[i].toread = NULL; | |
2218 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2219 | wake_up(&conf->wait_for_overlap); | |
2220 | if (bi) s->to_read--; | |
2221 | while (bi && bi->bi_sector < | |
2222 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2223 | struct bio *nextbi = | |
2224 | r5_next_bio(bi, sh->dev[i].sector); | |
2225 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2226 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2227 | bi->bi_next = *return_bi; |
2228 | *return_bi = bi; | |
2229 | } | |
2230 | bi = nextbi; | |
2231 | } | |
2232 | } | |
2233 | spin_unlock_irq(&conf->device_lock); | |
2234 | if (bitmap_end) | |
2235 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2236 | STRIPE_SECTORS, 0, 0); | |
2237 | } | |
2238 | ||
8b3e6cdc DW |
2239 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2240 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2241 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2242 | } |
2243 | ||
1fe797e6 DW |
2244 | /* fetch_block5 - checks the given member device to see if its data needs |
2245 | * to be read or computed to satisfy a request. | |
2246 | * | |
2247 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2248 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2249 | */ |
1fe797e6 DW |
2250 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2251 | int disk_idx, int disks) | |
f38e1219 DW |
2252 | { |
2253 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2254 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2255 | ||
f38e1219 DW |
2256 | /* is the data in this block needed, and can we get it? */ |
2257 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2258 | !test_bit(R5_UPTODATE, &dev->flags) && |
2259 | (dev->toread || | |
2260 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2261 | s->syncing || s->expanding || | |
2262 | (s->failed && | |
2263 | (failed_dev->toread || | |
2264 | (failed_dev->towrite && | |
2265 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2266 | /* We would like to get this block, possibly by computing it, |
2267 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2268 | */ |
2269 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2270 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2271 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2272 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2273 | set_bit(R5_Wantcompute, &dev->flags); |
2274 | sh->ops.target = disk_idx; | |
ac6b53b6 | 2275 | sh->ops.target2 = -1; |
f38e1219 | 2276 | s->req_compute = 1; |
f38e1219 | 2277 | /* Careful: from this point on 'uptodate' is in the eye |
ac6b53b6 | 2278 | * of raid_run_ops which services 'compute' operations |
f38e1219 DW |
2279 | * before writes. R5_Wantcompute flags a block that will |
2280 | * be R5_UPTODATE by the time it is needed for a | |
2281 | * subsequent operation. | |
2282 | */ | |
2283 | s->uptodate++; | |
1fe797e6 | 2284 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2285 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2286 | set_bit(R5_LOCKED, &dev->flags); |
2287 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2288 | s->locked++; |
2289 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2290 | s->syncing); | |
2291 | } | |
2292 | } | |
2293 | ||
1fe797e6 | 2294 | return 0; |
f38e1219 DW |
2295 | } |
2296 | ||
1fe797e6 DW |
2297 | /** |
2298 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2299 | */ | |
2300 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2301 | struct stripe_head_state *s, int disks) |
2302 | { | |
2303 | int i; | |
f38e1219 | 2304 | |
f38e1219 DW |
2305 | /* look for blocks to read/compute, skip this if a compute |
2306 | * is already in flight, or if the stripe contents are in the | |
2307 | * midst of changing due to a write | |
2308 | */ | |
976ea8d4 | 2309 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2310 | !sh->reconstruct_state) |
f38e1219 | 2311 | for (i = disks; i--; ) |
1fe797e6 | 2312 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2313 | break; |
a4456856 DW |
2314 | set_bit(STRIPE_HANDLE, &sh->state); |
2315 | } | |
2316 | ||
5599becc YT |
2317 | /* fetch_block6 - checks the given member device to see if its data needs |
2318 | * to be read or computed to satisfy a request. | |
2319 | * | |
2320 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2321 | * 0 to tell the loop in handle_stripe_fill6 to continue | |
2322 | */ | |
2323 | static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s, | |
2324 | struct r6_state *r6s, int disk_idx, int disks) | |
a4456856 | 2325 | { |
5599becc YT |
2326 | struct r5dev *dev = &sh->dev[disk_idx]; |
2327 | struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]], | |
2328 | &sh->dev[r6s->failed_num[1]] }; | |
2329 | ||
2330 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2331 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2332 | (dev->toread || | |
2333 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2334 | s->syncing || s->expanding || | |
2335 | (s->failed >= 1 && | |
2336 | (fdev[0]->toread || s->to_write)) || | |
2337 | (s->failed >= 2 && | |
2338 | (fdev[1]->toread || s->to_write)))) { | |
2339 | /* we would like to get this block, possibly by computing it, | |
2340 | * otherwise read it if the backing disk is insync | |
2341 | */ | |
2342 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2343 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2344 | if ((s->uptodate == disks - 1) && | |
2345 | (s->failed && (disk_idx == r6s->failed_num[0] || | |
2346 | disk_idx == r6s->failed_num[1]))) { | |
2347 | /* have disk failed, and we're requested to fetch it; | |
2348 | * do compute it | |
a4456856 | 2349 | */ |
5599becc YT |
2350 | pr_debug("Computing stripe %llu block %d\n", |
2351 | (unsigned long long)sh->sector, disk_idx); | |
2352 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2353 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2354 | set_bit(R5_Wantcompute, &dev->flags); | |
2355 | sh->ops.target = disk_idx; | |
2356 | sh->ops.target2 = -1; /* no 2nd target */ | |
2357 | s->req_compute = 1; | |
2358 | s->uptodate++; | |
2359 | return 1; | |
2360 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2361 | /* Computing 2-failure is *very* expensive; only | |
2362 | * do it if failed >= 2 | |
2363 | */ | |
2364 | int other; | |
2365 | for (other = disks; other--; ) { | |
2366 | if (other == disk_idx) | |
2367 | continue; | |
2368 | if (!test_bit(R5_UPTODATE, | |
2369 | &sh->dev[other].flags)) | |
2370 | break; | |
a4456856 | 2371 | } |
5599becc YT |
2372 | BUG_ON(other < 0); |
2373 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2374 | (unsigned long long)sh->sector, | |
2375 | disk_idx, other); | |
2376 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2377 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2378 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2379 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2380 | sh->ops.target = disk_idx; | |
2381 | sh->ops.target2 = other; | |
2382 | s->uptodate += 2; | |
2383 | s->req_compute = 1; | |
2384 | return 1; | |
2385 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2386 | set_bit(R5_LOCKED, &dev->flags); | |
2387 | set_bit(R5_Wantread, &dev->flags); | |
2388 | s->locked++; | |
2389 | pr_debug("Reading block %d (sync=%d)\n", | |
2390 | disk_idx, s->syncing); | |
a4456856 DW |
2391 | } |
2392 | } | |
5599becc YT |
2393 | |
2394 | return 0; | |
2395 | } | |
2396 | ||
2397 | /** | |
2398 | * handle_stripe_fill6 - read or compute data to satisfy pending requests. | |
2399 | */ | |
2400 | static void handle_stripe_fill6(struct stripe_head *sh, | |
2401 | struct stripe_head_state *s, struct r6_state *r6s, | |
2402 | int disks) | |
2403 | { | |
2404 | int i; | |
2405 | ||
2406 | /* look for blocks to read/compute, skip this if a compute | |
2407 | * is already in flight, or if the stripe contents are in the | |
2408 | * midst of changing due to a write | |
2409 | */ | |
2410 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2411 | !sh->reconstruct_state) | |
2412 | for (i = disks; i--; ) | |
2413 | if (fetch_block6(sh, s, r6s, i, disks)) | |
2414 | break; | |
a4456856 DW |
2415 | set_bit(STRIPE_HANDLE, &sh->state); |
2416 | } | |
2417 | ||
2418 | ||
1fe797e6 | 2419 | /* handle_stripe_clean_event |
a4456856 DW |
2420 | * any written block on an uptodate or failed drive can be returned. |
2421 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2422 | * never LOCKED, so we don't need to test 'failed' directly. | |
2423 | */ | |
1fe797e6 | 2424 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2425 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2426 | { | |
2427 | int i; | |
2428 | struct r5dev *dev; | |
2429 | ||
2430 | for (i = disks; i--; ) | |
2431 | if (sh->dev[i].written) { | |
2432 | dev = &sh->dev[i]; | |
2433 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2434 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2435 | /* We can return any write requests */ | |
2436 | struct bio *wbi, *wbi2; | |
2437 | int bitmap_end = 0; | |
45b4233c | 2438 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2439 | spin_lock_irq(&conf->device_lock); |
2440 | wbi = dev->written; | |
2441 | dev->written = NULL; | |
2442 | while (wbi && wbi->bi_sector < | |
2443 | dev->sector + STRIPE_SECTORS) { | |
2444 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2445 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2446 | md_write_end(conf->mddev); |
2447 | wbi->bi_next = *return_bi; | |
2448 | *return_bi = wbi; | |
2449 | } | |
2450 | wbi = wbi2; | |
2451 | } | |
2452 | if (dev->towrite == NULL) | |
2453 | bitmap_end = 1; | |
2454 | spin_unlock_irq(&conf->device_lock); | |
2455 | if (bitmap_end) | |
2456 | bitmap_endwrite(conf->mddev->bitmap, | |
2457 | sh->sector, | |
2458 | STRIPE_SECTORS, | |
2459 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2460 | 0); | |
2461 | } | |
2462 | } | |
8b3e6cdc DW |
2463 | |
2464 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2465 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2466 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2467 | } |
2468 | ||
1fe797e6 | 2469 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2470 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2471 | { | |
2472 | int rmw = 0, rcw = 0, i; | |
2473 | for (i = disks; i--; ) { | |
2474 | /* would I have to read this buffer for read_modify_write */ | |
2475 | struct r5dev *dev = &sh->dev[i]; | |
2476 | if ((dev->towrite || i == sh->pd_idx) && | |
2477 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2478 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2479 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2480 | if (test_bit(R5_Insync, &dev->flags)) |
2481 | rmw++; | |
2482 | else | |
2483 | rmw += 2*disks; /* cannot read it */ | |
2484 | } | |
2485 | /* Would I have to read this buffer for reconstruct_write */ | |
2486 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2487 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2488 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2489 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2490 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2491 | else |
2492 | rcw += 2*disks; | |
2493 | } | |
2494 | } | |
45b4233c | 2495 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2496 | (unsigned long long)sh->sector, rmw, rcw); |
2497 | set_bit(STRIPE_HANDLE, &sh->state); | |
2498 | if (rmw < rcw && rmw > 0) | |
2499 | /* prefer read-modify-write, but need to get some data */ | |
2500 | for (i = disks; i--; ) { | |
2501 | struct r5dev *dev = &sh->dev[i]; | |
2502 | if ((dev->towrite || i == sh->pd_idx) && | |
2503 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2504 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2505 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2506 | test_bit(R5_Insync, &dev->flags)) { |
2507 | if ( | |
2508 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2509 | pr_debug("Read_old block " |
a4456856 DW |
2510 | "%d for r-m-w\n", i); |
2511 | set_bit(R5_LOCKED, &dev->flags); | |
2512 | set_bit(R5_Wantread, &dev->flags); | |
2513 | s->locked++; | |
2514 | } else { | |
2515 | set_bit(STRIPE_DELAYED, &sh->state); | |
2516 | set_bit(STRIPE_HANDLE, &sh->state); | |
2517 | } | |
2518 | } | |
2519 | } | |
2520 | if (rcw <= rmw && rcw > 0) | |
2521 | /* want reconstruct write, but need to get some data */ | |
2522 | for (i = disks; i--; ) { | |
2523 | struct r5dev *dev = &sh->dev[i]; | |
2524 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2525 | i != sh->pd_idx && | |
2526 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2527 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2528 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2529 | test_bit(R5_Insync, &dev->flags)) { |
2530 | if ( | |
2531 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2532 | pr_debug("Read_old block " |
a4456856 DW |
2533 | "%d for Reconstruct\n", i); |
2534 | set_bit(R5_LOCKED, &dev->flags); | |
2535 | set_bit(R5_Wantread, &dev->flags); | |
2536 | s->locked++; | |
2537 | } else { | |
2538 | set_bit(STRIPE_DELAYED, &sh->state); | |
2539 | set_bit(STRIPE_HANDLE, &sh->state); | |
2540 | } | |
2541 | } | |
2542 | } | |
2543 | /* now if nothing is locked, and if we have enough data, | |
2544 | * we can start a write request | |
2545 | */ | |
f38e1219 DW |
2546 | /* since handle_stripe can be called at any time we need to handle the |
2547 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2548 | * subsequent call wants to start a write request. raid_run_ops only |
2549 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2550 | * simultaneously. If this is not the case then new writes need to be |
2551 | * held off until the compute completes. | |
2552 | */ | |
976ea8d4 DW |
2553 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2554 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2555 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2556 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2557 | } |
2558 | ||
1fe797e6 | 2559 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2560 | struct stripe_head *sh, struct stripe_head_state *s, |
2561 | struct r6_state *r6s, int disks) | |
2562 | { | |
a9b39a74 | 2563 | int rcw = 0, pd_idx = sh->pd_idx, i; |
34e04e87 | 2564 | int qd_idx = sh->qd_idx; |
a9b39a74 YT |
2565 | |
2566 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2567 | for (i = disks; i--; ) { |
2568 | struct r5dev *dev = &sh->dev[i]; | |
a9b39a74 YT |
2569 | /* check if we haven't enough data */ |
2570 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2571 | i != pd_idx && i != qd_idx && | |
2572 | !test_bit(R5_LOCKED, &dev->flags) && | |
2573 | !(test_bit(R5_UPTODATE, &dev->flags) || | |
2574 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2575 | rcw++; | |
2576 | if (!test_bit(R5_Insync, &dev->flags)) | |
2577 | continue; /* it's a failed drive */ | |
2578 | ||
2579 | if ( | |
2580 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2581 | pr_debug("Read_old stripe %llu " | |
2582 | "block %d for Reconstruct\n", | |
2583 | (unsigned long long)sh->sector, i); | |
2584 | set_bit(R5_LOCKED, &dev->flags); | |
2585 | set_bit(R5_Wantread, &dev->flags); | |
2586 | s->locked++; | |
2587 | } else { | |
2588 | pr_debug("Request delayed stripe %llu " | |
2589 | "block %d for Reconstruct\n", | |
2590 | (unsigned long long)sh->sector, i); | |
2591 | set_bit(STRIPE_DELAYED, &sh->state); | |
2592 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2593 | } |
2594 | } | |
2595 | } | |
a4456856 DW |
2596 | /* now if nothing is locked, and if we have enough data, we can start a |
2597 | * write request | |
2598 | */ | |
a9b39a74 YT |
2599 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2600 | s->locked == 0 && rcw == 0 && | |
a4456856 | 2601 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { |
a9b39a74 | 2602 | schedule_reconstruction(sh, s, 1, 0); |
a4456856 DW |
2603 | } |
2604 | } | |
2605 | ||
2606 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2607 | struct stripe_head_state *s, int disks) | |
2608 | { | |
ecc65c9b | 2609 | struct r5dev *dev = NULL; |
bd2ab670 | 2610 | |
a4456856 | 2611 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2612 | |
ecc65c9b DW |
2613 | switch (sh->check_state) { |
2614 | case check_state_idle: | |
2615 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2616 | if (s->failed == 0) { |
bd2ab670 | 2617 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2618 | sh->check_state = check_state_run; |
2619 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2620 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2621 | s->uptodate--; |
ecc65c9b | 2622 | break; |
bd2ab670 | 2623 | } |
ecc65c9b DW |
2624 | dev = &sh->dev[s->failed_num]; |
2625 | /* fall through */ | |
2626 | case check_state_compute_result: | |
2627 | sh->check_state = check_state_idle; | |
2628 | if (!dev) | |
2629 | dev = &sh->dev[sh->pd_idx]; | |
2630 | ||
2631 | /* check that a write has not made the stripe insync */ | |
2632 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2633 | break; | |
c8894419 | 2634 | |
a4456856 | 2635 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2636 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2637 | BUG_ON(s->uptodate != disks); | |
2638 | ||
2639 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2640 | s->locked++; |
a4456856 | 2641 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2642 | |
a4456856 | 2643 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2644 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2645 | break; |
2646 | case check_state_run: | |
2647 | break; /* we will be called again upon completion */ | |
2648 | case check_state_check_result: | |
2649 | sh->check_state = check_state_idle; | |
2650 | ||
2651 | /* if a failure occurred during the check operation, leave | |
2652 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2653 | */ | |
2654 | if (s->failed) | |
2655 | break; | |
2656 | ||
2657 | /* handle a successful check operation, if parity is correct | |
2658 | * we are done. Otherwise update the mismatch count and repair | |
2659 | * parity if !MD_RECOVERY_CHECK | |
2660 | */ | |
ad283ea4 | 2661 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2662 | /* parity is correct (on disc, |
2663 | * not in buffer any more) | |
2664 | */ | |
2665 | set_bit(STRIPE_INSYNC, &sh->state); | |
2666 | else { | |
2667 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2668 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2669 | /* don't try to repair!! */ | |
2670 | set_bit(STRIPE_INSYNC, &sh->state); | |
2671 | else { | |
2672 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2673 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2674 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2675 | set_bit(R5_Wantcompute, | |
2676 | &sh->dev[sh->pd_idx].flags); | |
2677 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2678 | sh->ops.target2 = -1; |
ecc65c9b DW |
2679 | s->uptodate++; |
2680 | } | |
2681 | } | |
2682 | break; | |
2683 | case check_state_compute_run: | |
2684 | break; | |
2685 | default: | |
2686 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2687 | __func__, sh->check_state, | |
2688 | (unsigned long long) sh->sector); | |
2689 | BUG(); | |
a4456856 DW |
2690 | } |
2691 | } | |
2692 | ||
2693 | ||
2694 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
36d1c647 DW |
2695 | struct stripe_head_state *s, |
2696 | struct r6_state *r6s, int disks) | |
a4456856 | 2697 | { |
a4456856 | 2698 | int pd_idx = sh->pd_idx; |
34e04e87 | 2699 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2700 | struct r5dev *dev; |
a4456856 DW |
2701 | |
2702 | set_bit(STRIPE_HANDLE, &sh->state); | |
2703 | ||
2704 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2705 | |
a4456856 DW |
2706 | /* Want to check and possibly repair P and Q. |
2707 | * However there could be one 'failed' device, in which | |
2708 | * case we can only check one of them, possibly using the | |
2709 | * other to generate missing data | |
2710 | */ | |
2711 | ||
d82dfee0 DW |
2712 | switch (sh->check_state) { |
2713 | case check_state_idle: | |
2714 | /* start a new check operation if there are < 2 failures */ | |
a4456856 | 2715 | if (s->failed == r6s->q_failed) { |
d82dfee0 | 2716 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2717 | * makes sense to check P (If anything else were failed, |
2718 | * we would have used P to recreate it). | |
2719 | */ | |
d82dfee0 | 2720 | sh->check_state = check_state_run; |
a4456856 DW |
2721 | } |
2722 | if (!r6s->q_failed && s->failed < 2) { | |
d82dfee0 | 2723 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2724 | * anything, so it makes sense to check it |
2725 | */ | |
d82dfee0 DW |
2726 | if (sh->check_state == check_state_run) |
2727 | sh->check_state = check_state_run_pq; | |
2728 | else | |
2729 | sh->check_state = check_state_run_q; | |
a4456856 | 2730 | } |
a4456856 | 2731 | |
d82dfee0 DW |
2732 | /* discard potentially stale zero_sum_result */ |
2733 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2734 | |
d82dfee0 DW |
2735 | if (sh->check_state == check_state_run) { |
2736 | /* async_xor_zero_sum destroys the contents of P */ | |
2737 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2738 | s->uptodate--; | |
a4456856 | 2739 | } |
d82dfee0 DW |
2740 | if (sh->check_state >= check_state_run && |
2741 | sh->check_state <= check_state_run_pq) { | |
2742 | /* async_syndrome_zero_sum preserves P and Q, so | |
2743 | * no need to mark them !uptodate here | |
2744 | */ | |
2745 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2746 | break; | |
a4456856 DW |
2747 | } |
2748 | ||
d82dfee0 DW |
2749 | /* we have 2-disk failure */ |
2750 | BUG_ON(s->failed != 2); | |
2751 | /* fall through */ | |
2752 | case check_state_compute_result: | |
2753 | sh->check_state = check_state_idle; | |
a4456856 | 2754 | |
d82dfee0 DW |
2755 | /* check that a write has not made the stripe insync */ |
2756 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2757 | break; | |
a4456856 DW |
2758 | |
2759 | /* now write out any block on a failed drive, | |
d82dfee0 | 2760 | * or P or Q if they were recomputed |
a4456856 | 2761 | */ |
d82dfee0 | 2762 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 DW |
2763 | if (s->failed == 2) { |
2764 | dev = &sh->dev[r6s->failed_num[1]]; | |
2765 | s->locked++; | |
2766 | set_bit(R5_LOCKED, &dev->flags); | |
2767 | set_bit(R5_Wantwrite, &dev->flags); | |
2768 | } | |
2769 | if (s->failed >= 1) { | |
2770 | dev = &sh->dev[r6s->failed_num[0]]; | |
2771 | s->locked++; | |
2772 | set_bit(R5_LOCKED, &dev->flags); | |
2773 | set_bit(R5_Wantwrite, &dev->flags); | |
2774 | } | |
d82dfee0 | 2775 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2776 | dev = &sh->dev[pd_idx]; |
2777 | s->locked++; | |
2778 | set_bit(R5_LOCKED, &dev->flags); | |
2779 | set_bit(R5_Wantwrite, &dev->flags); | |
2780 | } | |
d82dfee0 | 2781 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2782 | dev = &sh->dev[qd_idx]; |
2783 | s->locked++; | |
2784 | set_bit(R5_LOCKED, &dev->flags); | |
2785 | set_bit(R5_Wantwrite, &dev->flags); | |
2786 | } | |
2787 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2788 | ||
2789 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2790 | break; |
2791 | case check_state_run: | |
2792 | case check_state_run_q: | |
2793 | case check_state_run_pq: | |
2794 | break; /* we will be called again upon completion */ | |
2795 | case check_state_check_result: | |
2796 | sh->check_state = check_state_idle; | |
2797 | ||
2798 | /* handle a successful check operation, if parity is correct | |
2799 | * we are done. Otherwise update the mismatch count and repair | |
2800 | * parity if !MD_RECOVERY_CHECK | |
2801 | */ | |
2802 | if (sh->ops.zero_sum_result == 0) { | |
2803 | /* both parities are correct */ | |
2804 | if (!s->failed) | |
2805 | set_bit(STRIPE_INSYNC, &sh->state); | |
2806 | else { | |
2807 | /* in contrast to the raid5 case we can validate | |
2808 | * parity, but still have a failure to write | |
2809 | * back | |
2810 | */ | |
2811 | sh->check_state = check_state_compute_result; | |
2812 | /* Returning at this point means that we may go | |
2813 | * off and bring p and/or q uptodate again so | |
2814 | * we make sure to check zero_sum_result again | |
2815 | * to verify if p or q need writeback | |
2816 | */ | |
2817 | } | |
2818 | } else { | |
2819 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2820 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2821 | /* don't try to repair!! */ | |
2822 | set_bit(STRIPE_INSYNC, &sh->state); | |
2823 | else { | |
2824 | int *target = &sh->ops.target; | |
2825 | ||
2826 | sh->ops.target = -1; | |
2827 | sh->ops.target2 = -1; | |
2828 | sh->check_state = check_state_compute_run; | |
2829 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2830 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2831 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2832 | set_bit(R5_Wantcompute, | |
2833 | &sh->dev[pd_idx].flags); | |
2834 | *target = pd_idx; | |
2835 | target = &sh->ops.target2; | |
2836 | s->uptodate++; | |
2837 | } | |
2838 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2839 | set_bit(R5_Wantcompute, | |
2840 | &sh->dev[qd_idx].flags); | |
2841 | *target = qd_idx; | |
2842 | s->uptodate++; | |
2843 | } | |
2844 | } | |
2845 | } | |
2846 | break; | |
2847 | case check_state_compute_run: | |
2848 | break; | |
2849 | default: | |
2850 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2851 | __func__, sh->check_state, | |
2852 | (unsigned long long) sh->sector); | |
2853 | BUG(); | |
a4456856 DW |
2854 | } |
2855 | } | |
2856 | ||
2857 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2858 | struct r6_state *r6s) | |
2859 | { | |
2860 | int i; | |
2861 | ||
2862 | /* We have read all the blocks in this stripe and now we need to | |
2863 | * copy some of them into a target stripe for expand. | |
2864 | */ | |
f0a50d37 | 2865 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2866 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2867 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 2868 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 2869 | int dd_idx, j; |
a4456856 | 2870 | struct stripe_head *sh2; |
a08abd8c | 2871 | struct async_submit_ctl submit; |
a4456856 | 2872 | |
784052ec | 2873 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
2874 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2875 | &dd_idx, NULL); | |
a8c906ca | 2876 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
2877 | if (sh2 == NULL) |
2878 | /* so far only the early blocks of this stripe | |
2879 | * have been requested. When later blocks | |
2880 | * get requested, we will try again | |
2881 | */ | |
2882 | continue; | |
2883 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2884 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2885 | /* must have already done this block */ | |
2886 | release_stripe(sh2); | |
2887 | continue; | |
2888 | } | |
f0a50d37 DW |
2889 | |
2890 | /* place all the copies on one channel */ | |
a08abd8c | 2891 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 2892 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 2893 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 2894 | &submit); |
f0a50d37 | 2895 | |
a4456856 DW |
2896 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2897 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2898 | for (j = 0; j < conf->raid_disks; j++) | |
2899 | if (j != sh2->pd_idx && | |
d0dabf7e | 2900 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
2901 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2902 | break; | |
2903 | if (j == conf->raid_disks) { | |
2904 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2905 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2906 | } | |
2907 | release_stripe(sh2); | |
f0a50d37 | 2908 | |
a4456856 | 2909 | } |
a2e08551 N |
2910 | /* done submitting copies, wait for them to complete */ |
2911 | if (tx) { | |
2912 | async_tx_ack(tx); | |
2913 | dma_wait_for_async_tx(tx); | |
2914 | } | |
a4456856 | 2915 | } |
1da177e4 | 2916 | |
6bfe0b49 | 2917 | |
1da177e4 LT |
2918 | /* |
2919 | * handle_stripe - do things to a stripe. | |
2920 | * | |
2921 | * We lock the stripe and then examine the state of various bits | |
2922 | * to see what needs to be done. | |
2923 | * Possible results: | |
2924 | * return some read request which now have data | |
2925 | * return some write requests which are safely on disc | |
2926 | * schedule a read on some buffers | |
2927 | * schedule a write of some buffers | |
2928 | * return confirmation of parity correctness | |
2929 | * | |
1da177e4 LT |
2930 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2931 | * get BH_Lock set before the stripe lock is released. | |
2932 | * | |
2933 | */ | |
a4456856 | 2934 | |
1442577b | 2935 | static void handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2936 | { |
2937 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2938 | int disks = sh->disks, i; |
2939 | struct bio *return_bi = NULL; | |
2940 | struct stripe_head_state s; | |
1da177e4 | 2941 | struct r5dev *dev; |
6bfe0b49 | 2942 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2943 | int prexor; |
729a1866 | 2944 | int dec_preread_active = 0; |
1da177e4 | 2945 | |
a4456856 | 2946 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
2947 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
2948 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
2949 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
2950 | sh->reconstruct_state); | |
1da177e4 LT |
2951 | |
2952 | spin_lock(&sh->lock); | |
2953 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2954 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2955 | ||
a4456856 DW |
2956 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2957 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2958 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 2959 | |
83de75cc | 2960 | /* Now to look around and see what can be done */ |
9910f16a | 2961 | rcu_read_lock(); |
1da177e4 LT |
2962 | for (i=disks; i--; ) { |
2963 | mdk_rdev_t *rdev; | |
a9f326eb N |
2964 | |
2965 | dev = &sh->dev[i]; | |
1da177e4 | 2966 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2967 | |
b5e98d65 DW |
2968 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2969 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2970 | dev->towrite, dev->written); | |
2971 | ||
2972 | /* maybe we can request a biofill operation | |
2973 | * | |
2974 | * new wantfill requests are only permitted while | |
83de75cc | 2975 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
2976 | */ |
2977 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 2978 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 2979 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
2980 | |
2981 | /* now count some things */ | |
a4456856 DW |
2982 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2983 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2984 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2985 | |
b5e98d65 DW |
2986 | if (test_bit(R5_Wantfill, &dev->flags)) |
2987 | s.to_fill++; | |
2988 | else if (dev->toread) | |
a4456856 | 2989 | s.to_read++; |
1da177e4 | 2990 | if (dev->towrite) { |
a4456856 | 2991 | s.to_write++; |
1da177e4 | 2992 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2993 | s.non_overwrite++; |
1da177e4 | 2994 | } |
a4456856 DW |
2995 | if (dev->written) |
2996 | s.written++; | |
9910f16a | 2997 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
2998 | if (blocked_rdev == NULL && |
2999 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3000 | blocked_rdev = rdev; |
3001 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3002 | } |
b2d444d7 | 3003 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 3004 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
3005 | clear_bit(R5_ReadError, &dev->flags); |
3006 | clear_bit(R5_ReWrite, &dev->flags); | |
3007 | } | |
b2d444d7 | 3008 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 3009 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
3010 | s.failed++; |
3011 | s.failed_num = i; | |
1da177e4 LT |
3012 | } else |
3013 | set_bit(R5_Insync, &dev->flags); | |
3014 | } | |
9910f16a | 3015 | rcu_read_unlock(); |
b5e98d65 | 3016 | |
6bfe0b49 | 3017 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
3018 | if (s.syncing || s.expanding || s.expanded || |
3019 | s.to_write || s.written) { | |
3020 | set_bit(STRIPE_HANDLE, &sh->state); | |
3021 | goto unlock; | |
3022 | } | |
3023 | /* There is nothing for the blocked_rdev to block */ | |
3024 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3025 | blocked_rdev = NULL; | |
6bfe0b49 DW |
3026 | } |
3027 | ||
83de75cc DW |
3028 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3029 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3030 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3031 | } | |
b5e98d65 | 3032 | |
45b4233c | 3033 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 3034 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
3035 | s.locked, s.uptodate, s.to_read, s.to_write, |
3036 | s.failed, s.failed_num); | |
1da177e4 LT |
3037 | /* check if the array has lost two devices and, if so, some requests might |
3038 | * need to be failed | |
3039 | */ | |
a4456856 | 3040 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3041 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3042 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
3043 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3044 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3045 | s.syncing = 0; |
1da177e4 LT |
3046 | } |
3047 | ||
3048 | /* might be able to return some write requests if the parity block | |
3049 | * is safe, or on a failed drive | |
3050 | */ | |
3051 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
3052 | if ( s.written && |
3053 | ((test_bit(R5_Insync, &dev->flags) && | |
3054 | !test_bit(R5_LOCKED, &dev->flags) && | |
3055 | test_bit(R5_UPTODATE, &dev->flags)) || | |
3056 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 3057 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
3058 | |
3059 | /* Now we might consider reading some blocks, either to check/generate | |
3060 | * parity, or to satisfy requests | |
3061 | * or to load a block that is being partially written. | |
3062 | */ | |
a4456856 | 3063 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 3064 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3065 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 3066 | |
e33129d8 DW |
3067 | /* Now we check to see if any write operations have recently |
3068 | * completed | |
3069 | */ | |
e0a115e5 | 3070 | prexor = 0; |
d8ee0728 | 3071 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 3072 | prexor = 1; |
d8ee0728 DW |
3073 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
3074 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 3075 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
3076 | |
3077 | /* All the 'written' buffers and the parity block are ready to | |
3078 | * be written back to disk | |
3079 | */ | |
3080 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3081 | for (i = disks; i--; ) { | |
3082 | dev = &sh->dev[i]; | |
3083 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3084 | (i == sh->pd_idx || dev->written)) { | |
3085 | pr_debug("Writing block %d\n", i); | |
3086 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
3087 | if (prexor) |
3088 | continue; | |
e33129d8 DW |
3089 | if (!test_bit(R5_Insync, &dev->flags) || |
3090 | (i == sh->pd_idx && s.failed == 0)) | |
3091 | set_bit(STRIPE_INSYNC, &sh->state); | |
3092 | } | |
3093 | } | |
729a1866 N |
3094 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3095 | dec_preread_active = 1; | |
e33129d8 DW |
3096 | } |
3097 | ||
3098 | /* Now to consider new write requests and what else, if anything | |
3099 | * should be read. We do not handle new writes when: | |
3100 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3101 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3102 | * block. | |
3103 | */ | |
600aa109 | 3104 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 3105 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
3106 | |
3107 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
3108 | * Any reads will already have been scheduled, so we just see if enough |
3109 | * data is available. The parity check is held off while parity | |
3110 | * dependent operations are in flight. | |
1da177e4 | 3111 | */ |
ecc65c9b DW |
3112 | if (sh->check_state || |
3113 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 3114 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 3115 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 3116 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 3117 | |
a4456856 | 3118 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
3119 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3120 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3121 | } | |
4e5314b5 N |
3122 | |
3123 | /* If the failed drive is just a ReadError, then we might need to progress | |
3124 | * the repair/check process | |
3125 | */ | |
a4456856 DW |
3126 | if (s.failed == 1 && !conf->mddev->ro && |
3127 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
3128 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
3129 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 3130 | ) { |
a4456856 | 3131 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
3132 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
3133 | set_bit(R5_Wantwrite, &dev->flags); | |
3134 | set_bit(R5_ReWrite, &dev->flags); | |
3135 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3136 | s.locked++; |
4e5314b5 N |
3137 | } else { |
3138 | /* let's read it back */ | |
3139 | set_bit(R5_Wantread, &dev->flags); | |
3140 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3141 | s.locked++; |
4e5314b5 N |
3142 | } |
3143 | } | |
3144 | ||
600aa109 DW |
3145 | /* Finish reconstruct operations initiated by the expansion process */ |
3146 | if (sh->reconstruct_state == reconstruct_state_result) { | |
ab69ae12 | 3147 | struct stripe_head *sh2 |
a8c906ca | 3148 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3149 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3150 | /* sh cannot be written until sh2 has been read. | |
3151 | * so arrange for sh to be delayed a little | |
3152 | */ | |
3153 | set_bit(STRIPE_DELAYED, &sh->state); | |
3154 | set_bit(STRIPE_HANDLE, &sh->state); | |
3155 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3156 | &sh2->state)) | |
3157 | atomic_inc(&conf->preread_active_stripes); | |
3158 | release_stripe(sh2); | |
3159 | goto unlock; | |
3160 | } | |
3161 | if (sh2) | |
3162 | release_stripe(sh2); | |
3163 | ||
600aa109 | 3164 | sh->reconstruct_state = reconstruct_state_idle; |
f0a50d37 | 3165 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 3166 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 3167 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 3168 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 3169 | s.locked++; |
23397883 | 3170 | } |
f0a50d37 DW |
3171 | } |
3172 | ||
3173 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 3174 | !sh->reconstruct_state) { |
f0a50d37 DW |
3175 | /* Need to write out all blocks after computing parity */ |
3176 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3177 | stripe_set_idx(sh->sector, conf, 0, sh); |
c0f7bddb | 3178 | schedule_reconstruction(sh, &s, 1, 1); |
600aa109 | 3179 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 3180 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 3181 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
3182 | wake_up(&conf->wait_for_overlap); |
3183 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3184 | } | |
3185 | ||
0f94e87c | 3186 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3187 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3188 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 3189 | |
6bfe0b49 | 3190 | unlock: |
1da177e4 LT |
3191 | spin_unlock(&sh->lock); |
3192 | ||
6bfe0b49 DW |
3193 | /* wait for this device to become unblocked */ |
3194 | if (unlikely(blocked_rdev)) | |
3195 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3196 | ||
600aa109 | 3197 | if (s.ops_request) |
ac6b53b6 | 3198 | raid_run_ops(sh, s.ops_request); |
d84e0f10 | 3199 | |
c4e5ac0a | 3200 | ops_run_io(sh, &s); |
1da177e4 | 3201 | |
729a1866 N |
3202 | if (dec_preread_active) { |
3203 | /* We delay this until after ops_run_io so that if make_request | |
3204 | * is waiting on a barrier, it won't continue until the writes | |
3205 | * have actually been submitted. | |
3206 | */ | |
3207 | atomic_dec(&conf->preread_active_stripes); | |
3208 | if (atomic_read(&conf->preread_active_stripes) < | |
3209 | IO_THRESHOLD) | |
3210 | md_wakeup_thread(conf->mddev->thread); | |
3211 | } | |
a4456856 | 3212 | return_io(return_bi); |
1da177e4 LT |
3213 | } |
3214 | ||
1442577b | 3215 | static void handle_stripe6(struct stripe_head *sh) |
1da177e4 | 3216 | { |
bff61975 | 3217 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 3218 | int disks = sh->disks; |
a4456856 | 3219 | struct bio *return_bi = NULL; |
34e04e87 | 3220 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx; |
a4456856 DW |
3221 | struct stripe_head_state s; |
3222 | struct r6_state r6s; | |
16a53ecc | 3223 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 3224 | mdk_rdev_t *blocked_rdev = NULL; |
729a1866 | 3225 | int dec_preread_active = 0; |
1da177e4 | 3226 | |
45b4233c | 3227 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
6c0069c0 | 3228 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", |
a4456856 | 3229 | (unsigned long long)sh->sector, sh->state, |
6c0069c0 YT |
3230 | atomic_read(&sh->count), pd_idx, qd_idx, |
3231 | sh->check_state, sh->reconstruct_state); | |
a4456856 | 3232 | memset(&s, 0, sizeof(s)); |
72626685 | 3233 | |
16a53ecc N |
3234 | spin_lock(&sh->lock); |
3235 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3236 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3237 | ||
a4456856 DW |
3238 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3239 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3240 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 3241 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3242 | |
3243 | rcu_read_lock(); | |
16a53ecc N |
3244 | for (i=disks; i--; ) { |
3245 | mdk_rdev_t *rdev; | |
3246 | dev = &sh->dev[i]; | |
3247 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 3248 | |
45b4233c | 3249 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc | 3250 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
6c0069c0 YT |
3251 | /* maybe we can reply to a read |
3252 | * | |
3253 | * new wantfill requests are only permitted while | |
3254 | * ops_complete_biofill is guaranteed to be inactive | |
3255 | */ | |
3256 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3257 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3258 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3259 | |
16a53ecc | 3260 | /* now count some things */ |
a4456856 DW |
3261 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3262 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
2d6e4ecc DW |
3263 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
3264 | s.compute++; | |
3265 | BUG_ON(s.compute > 2); | |
3266 | } | |
1da177e4 | 3267 | |
6c0069c0 YT |
3268 | if (test_bit(R5_Wantfill, &dev->flags)) { |
3269 | s.to_fill++; | |
3270 | } else if (dev->toread) | |
a4456856 | 3271 | s.to_read++; |
16a53ecc | 3272 | if (dev->towrite) { |
a4456856 | 3273 | s.to_write++; |
16a53ecc | 3274 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3275 | s.non_overwrite++; |
16a53ecc | 3276 | } |
a4456856 DW |
3277 | if (dev->written) |
3278 | s.written++; | |
16a53ecc | 3279 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3280 | if (blocked_rdev == NULL && |
3281 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3282 | blocked_rdev = rdev; |
3283 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3284 | } |
16a53ecc N |
3285 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3286 | /* The ReadError flag will just be confusing now */ | |
3287 | clear_bit(R5_ReadError, &dev->flags); | |
3288 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3289 | } |
16a53ecc N |
3290 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3291 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3292 | if (s.failed < 2) |
3293 | r6s.failed_num[s.failed] = i; | |
3294 | s.failed++; | |
16a53ecc N |
3295 | } else |
3296 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3297 | } |
3298 | rcu_read_unlock(); | |
6bfe0b49 DW |
3299 | |
3300 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3301 | if (s.syncing || s.expanding || s.expanded || |
3302 | s.to_write || s.written) { | |
3303 | set_bit(STRIPE_HANDLE, &sh->state); | |
3304 | goto unlock; | |
3305 | } | |
3306 | /* There is nothing for the blocked_rdev to block */ | |
3307 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3308 | blocked_rdev = NULL; | |
6bfe0b49 | 3309 | } |
ac4090d2 | 3310 | |
6c0069c0 YT |
3311 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3312 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3313 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3314 | } | |
3315 | ||
45b4233c | 3316 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3317 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3318 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3319 | r6s.failed_num[0], r6s.failed_num[1]); | |
3320 | /* check if the array has lost >2 devices and, if so, some requests | |
3321 | * might need to be failed | |
16a53ecc | 3322 | */ |
a4456856 | 3323 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3324 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3325 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3326 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3327 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3328 | s.syncing = 0; |
16a53ecc N |
3329 | } |
3330 | ||
3331 | /* | |
3332 | * might be able to return some write requests if the parity blocks | |
3333 | * are safe, or on a failed drive | |
3334 | */ | |
3335 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3336 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3337 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
34e04e87 N |
3338 | qdev = &sh->dev[qd_idx]; |
3339 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx) | |
3340 | || (s.failed >= 2 && r6s.failed_num[1] == qd_idx); | |
a4456856 DW |
3341 | |
3342 | if ( s.written && | |
3343 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3344 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3345 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3346 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3347 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3348 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3349 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3350 | |
3351 | /* Now we might consider reading some blocks, either to check/generate | |
3352 | * parity, or to satisfy requests | |
3353 | * or to load a block that is being partially written. | |
3354 | */ | |
a4456856 | 3355 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
6c0069c0 | 3356 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3357 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc | 3358 | |
6c0069c0 YT |
3359 | /* Now we check to see if any write operations have recently |
3360 | * completed | |
3361 | */ | |
3362 | if (sh->reconstruct_state == reconstruct_state_drain_result) { | |
6c0069c0 YT |
3363 | |
3364 | sh->reconstruct_state = reconstruct_state_idle; | |
3365 | /* All the 'written' buffers and the parity blocks are ready to | |
3366 | * be written back to disk | |
3367 | */ | |
3368 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3369 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags)); | |
3370 | for (i = disks; i--; ) { | |
3371 | dev = &sh->dev[i]; | |
3372 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3373 | (i == sh->pd_idx || i == qd_idx || | |
3374 | dev->written)) { | |
3375 | pr_debug("Writing block %d\n", i); | |
3376 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); | |
3377 | set_bit(R5_Wantwrite, &dev->flags); | |
3378 | if (!test_bit(R5_Insync, &dev->flags) || | |
3379 | ((i == sh->pd_idx || i == qd_idx) && | |
3380 | s.failed == 0)) | |
3381 | set_bit(STRIPE_INSYNC, &sh->state); | |
3382 | } | |
3383 | } | |
729a1866 N |
3384 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3385 | dec_preread_active = 1; | |
6c0069c0 YT |
3386 | } |
3387 | ||
a9b39a74 YT |
3388 | /* Now to consider new write requests and what else, if anything |
3389 | * should be read. We do not handle new writes when: | |
3390 | * 1/ A 'write' operation (copy+gen_syndrome) is already in flight. | |
3391 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3392 | * block. | |
3393 | */ | |
3394 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
1fe797e6 | 3395 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3396 | |
3397 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 | 3398 | * Any reads will already have been scheduled, so we just see if enough |
6c0069c0 YT |
3399 | * data is available. The parity check is held off while parity |
3400 | * dependent operations are in flight. | |
16a53ecc | 3401 | */ |
6c0069c0 YT |
3402 | if (sh->check_state || |
3403 | (s.syncing && s.locked == 0 && | |
3404 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3405 | !test_bit(STRIPE_INSYNC, &sh->state))) | |
36d1c647 | 3406 | handle_parity_checks6(conf, sh, &s, &r6s, disks); |
16a53ecc | 3407 | |
a4456856 | 3408 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3409 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3410 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3411 | } | |
3412 | ||
3413 | /* If the failed drives are just a ReadError, then we might need | |
3414 | * to progress the repair/check process | |
3415 | */ | |
a4456856 DW |
3416 | if (s.failed <= 2 && !conf->mddev->ro) |
3417 | for (i = 0; i < s.failed; i++) { | |
3418 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3419 | if (test_bit(R5_ReadError, &dev->flags) |
3420 | && !test_bit(R5_LOCKED, &dev->flags) | |
3421 | && test_bit(R5_UPTODATE, &dev->flags) | |
3422 | ) { | |
3423 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3424 | set_bit(R5_Wantwrite, &dev->flags); | |
3425 | set_bit(R5_ReWrite, &dev->flags); | |
3426 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3427 | s.locked++; |
16a53ecc N |
3428 | } else { |
3429 | /* let's read it back */ | |
3430 | set_bit(R5_Wantread, &dev->flags); | |
3431 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3432 | s.locked++; |
16a53ecc N |
3433 | } |
3434 | } | |
3435 | } | |
f416885e | 3436 | |
6c0069c0 YT |
3437 | /* Finish reconstruct operations initiated by the expansion process */ |
3438 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3439 | sh->reconstruct_state = reconstruct_state_idle; | |
3440 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3441 | for (i = conf->raid_disks; i--; ) { | |
3442 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3443 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3444 | s.locked++; | |
3445 | } | |
3446 | } | |
3447 | ||
3448 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
3449 | !sh->reconstruct_state) { | |
ab69ae12 | 3450 | struct stripe_head *sh2 |
a8c906ca | 3451 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3452 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3453 | /* sh cannot be written until sh2 has been read. | |
3454 | * so arrange for sh to be delayed a little | |
3455 | */ | |
3456 | set_bit(STRIPE_DELAYED, &sh->state); | |
3457 | set_bit(STRIPE_HANDLE, &sh->state); | |
3458 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3459 | &sh2->state)) | |
3460 | atomic_inc(&conf->preread_active_stripes); | |
3461 | release_stripe(sh2); | |
3462 | goto unlock; | |
3463 | } | |
3464 | if (sh2) | |
3465 | release_stripe(sh2); | |
3466 | ||
f416885e N |
3467 | /* Need to write out all blocks after computing P&Q */ |
3468 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3469 | stripe_set_idx(sh->sector, conf, 0, sh); |
6c0069c0 YT |
3470 | schedule_reconstruction(sh, &s, 1, 1); |
3471 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
f416885e N |
3472 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3473 | atomic_dec(&conf->reshape_stripes); | |
3474 | wake_up(&conf->wait_for_overlap); | |
3475 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3476 | } | |
3477 | ||
0f94e87c | 3478 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3479 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3480 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3481 | |
6bfe0b49 | 3482 | unlock: |
16a53ecc N |
3483 | spin_unlock(&sh->lock); |
3484 | ||
6bfe0b49 DW |
3485 | /* wait for this device to become unblocked */ |
3486 | if (unlikely(blocked_rdev)) | |
3487 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3488 | ||
6c0069c0 YT |
3489 | if (s.ops_request) |
3490 | raid_run_ops(sh, s.ops_request); | |
3491 | ||
f0e43bcd | 3492 | ops_run_io(sh, &s); |
16a53ecc | 3493 | |
729a1866 N |
3494 | |
3495 | if (dec_preread_active) { | |
3496 | /* We delay this until after ops_run_io so that if make_request | |
3497 | * is waiting on a barrier, it won't continue until the writes | |
3498 | * have actually been submitted. | |
3499 | */ | |
3500 | atomic_dec(&conf->preread_active_stripes); | |
3501 | if (atomic_read(&conf->preread_active_stripes) < | |
3502 | IO_THRESHOLD) | |
3503 | md_wakeup_thread(conf->mddev->thread); | |
3504 | } | |
3505 | ||
f0e43bcd | 3506 | return_io(return_bi); |
16a53ecc N |
3507 | } |
3508 | ||
1442577b | 3509 | static void handle_stripe(struct stripe_head *sh) |
16a53ecc N |
3510 | { |
3511 | if (sh->raid_conf->level == 6) | |
1442577b | 3512 | handle_stripe6(sh); |
16a53ecc | 3513 | else |
1442577b | 3514 | handle_stripe5(sh); |
16a53ecc N |
3515 | } |
3516 | ||
16a53ecc N |
3517 | static void raid5_activate_delayed(raid5_conf_t *conf) |
3518 | { | |
3519 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3520 | while (!list_empty(&conf->delayed_list)) { | |
3521 | struct list_head *l = conf->delayed_list.next; | |
3522 | struct stripe_head *sh; | |
3523 | sh = list_entry(l, struct stripe_head, lru); | |
3524 | list_del_init(l); | |
3525 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3526 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3527 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3528 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3529 | } |
6ed3003c N |
3530 | } else |
3531 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3532 | } |
3533 | ||
3534 | static void activate_bit_delay(raid5_conf_t *conf) | |
3535 | { | |
3536 | /* device_lock is held */ | |
3537 | struct list_head head; | |
3538 | list_add(&head, &conf->bitmap_list); | |
3539 | list_del_init(&conf->bitmap_list); | |
3540 | while (!list_empty(&head)) { | |
3541 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3542 | list_del_init(&sh->lru); | |
3543 | atomic_inc(&sh->count); | |
3544 | __release_stripe(conf, sh); | |
3545 | } | |
3546 | } | |
3547 | ||
3548 | static void unplug_slaves(mddev_t *mddev) | |
3549 | { | |
070ec55d | 3550 | raid5_conf_t *conf = mddev->private; |
16a53ecc | 3551 | int i; |
5e5e3e78 | 3552 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
16a53ecc N |
3553 | |
3554 | rcu_read_lock(); | |
5e5e3e78 | 3555 | for (i = 0; i < devs; i++) { |
16a53ecc N |
3556 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); |
3557 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3558 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3559 | |
3560 | atomic_inc(&rdev->nr_pending); | |
3561 | rcu_read_unlock(); | |
3562 | ||
2ad8b1ef | 3563 | blk_unplug(r_queue); |
16a53ecc N |
3564 | |
3565 | rdev_dec_pending(rdev, mddev); | |
3566 | rcu_read_lock(); | |
3567 | } | |
3568 | } | |
3569 | rcu_read_unlock(); | |
3570 | } | |
3571 | ||
165125e1 | 3572 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3573 | { |
3574 | mddev_t *mddev = q->queuedata; | |
070ec55d | 3575 | raid5_conf_t *conf = mddev->private; |
16a53ecc N |
3576 | unsigned long flags; |
3577 | ||
3578 | spin_lock_irqsave(&conf->device_lock, flags); | |
3579 | ||
3580 | if (blk_remove_plug(q)) { | |
3581 | conf->seq_flush++; | |
3582 | raid5_activate_delayed(conf); | |
72626685 | 3583 | } |
1da177e4 LT |
3584 | md_wakeup_thread(mddev->thread); |
3585 | ||
3586 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3587 | ||
3588 | unplug_slaves(mddev); | |
3589 | } | |
3590 | ||
f022b2fd N |
3591 | static int raid5_congested(void *data, int bits) |
3592 | { | |
3593 | mddev_t *mddev = data; | |
070ec55d | 3594 | raid5_conf_t *conf = mddev->private; |
f022b2fd N |
3595 | |
3596 | /* No difference between reads and writes. Just check | |
3597 | * how busy the stripe_cache is | |
3598 | */ | |
3fa841d7 N |
3599 | |
3600 | if (mddev_congested(mddev, bits)) | |
3601 | return 1; | |
f022b2fd N |
3602 | if (conf->inactive_blocked) |
3603 | return 1; | |
3604 | if (conf->quiesce) | |
3605 | return 1; | |
3606 | if (list_empty_careful(&conf->inactive_list)) | |
3607 | return 1; | |
3608 | ||
3609 | return 0; | |
3610 | } | |
3611 | ||
23032a0e RBJ |
3612 | /* We want read requests to align with chunks where possible, |
3613 | * but write requests don't need to. | |
3614 | */ | |
cc371e66 AK |
3615 | static int raid5_mergeable_bvec(struct request_queue *q, |
3616 | struct bvec_merge_data *bvm, | |
3617 | struct bio_vec *biovec) | |
23032a0e RBJ |
3618 | { |
3619 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3620 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3621 | int max; |
9d8f0363 | 3622 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3623 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3624 | |
cc371e66 | 3625 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3626 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3627 | ||
664e7c41 AN |
3628 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3629 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3630 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3631 | if (max < 0) max = 0; | |
3632 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3633 | return biovec->bv_len; | |
3634 | else | |
3635 | return max; | |
3636 | } | |
3637 | ||
f679623f RBJ |
3638 | |
3639 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3640 | { | |
3641 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3642 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3643 | unsigned int bio_sectors = bio->bi_size >> 9; |
3644 | ||
664e7c41 AN |
3645 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3646 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3647 | return chunk_sectors >= |
3648 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3649 | } | |
3650 | ||
46031f9a RBJ |
3651 | /* |
3652 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3653 | * later sampled by raid5d. | |
3654 | */ | |
3655 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3656 | { | |
3657 | unsigned long flags; | |
3658 | ||
3659 | spin_lock_irqsave(&conf->device_lock, flags); | |
3660 | ||
3661 | bi->bi_next = conf->retry_read_aligned_list; | |
3662 | conf->retry_read_aligned_list = bi; | |
3663 | ||
3664 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3665 | md_wakeup_thread(conf->mddev->thread); | |
3666 | } | |
3667 | ||
3668 | ||
3669 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3670 | { | |
3671 | struct bio *bi; | |
3672 | ||
3673 | bi = conf->retry_read_aligned; | |
3674 | if (bi) { | |
3675 | conf->retry_read_aligned = NULL; | |
3676 | return bi; | |
3677 | } | |
3678 | bi = conf->retry_read_aligned_list; | |
3679 | if(bi) { | |
387bb173 | 3680 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3681 | bi->bi_next = NULL; |
960e739d JA |
3682 | /* |
3683 | * this sets the active strip count to 1 and the processed | |
3684 | * strip count to zero (upper 8 bits) | |
3685 | */ | |
46031f9a | 3686 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3687 | } |
3688 | ||
3689 | return bi; | |
3690 | } | |
3691 | ||
3692 | ||
f679623f RBJ |
3693 | /* |
3694 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3695 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3696 | * first). | |
3697 | * If the read failed.. | |
3698 | */ | |
6712ecf8 | 3699 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3700 | { |
3701 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3702 | mddev_t *mddev; |
3703 | raid5_conf_t *conf; | |
3704 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3705 | mdk_rdev_t *rdev; | |
3706 | ||
f679623f | 3707 | bio_put(bi); |
46031f9a | 3708 | |
46031f9a RBJ |
3709 | rdev = (void*)raid_bi->bi_next; |
3710 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3711 | mddev = rdev->mddev; |
3712 | conf = mddev->private; | |
46031f9a RBJ |
3713 | |
3714 | rdev_dec_pending(rdev, conf->mddev); | |
3715 | ||
3716 | if (!error && uptodate) { | |
6712ecf8 | 3717 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3718 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3719 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3720 | return; |
46031f9a RBJ |
3721 | } |
3722 | ||
3723 | ||
45b4233c | 3724 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3725 | |
3726 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3727 | } |
3728 | ||
387bb173 NB |
3729 | static int bio_fits_rdev(struct bio *bi) |
3730 | { | |
165125e1 | 3731 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3732 | |
ae03bf63 | 3733 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3734 | return 0; |
3735 | blk_recount_segments(q, bi); | |
8a78362c | 3736 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3737 | return 0; |
3738 | ||
3739 | if (q->merge_bvec_fn) | |
3740 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3741 | * just just give up | |
3742 | */ | |
3743 | return 0; | |
3744 | ||
3745 | return 1; | |
3746 | } | |
3747 | ||
3748 | ||
21a52c6d | 3749 | static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio) |
f679623f | 3750 | { |
070ec55d | 3751 | raid5_conf_t *conf = mddev->private; |
8553fe7e | 3752 | int dd_idx; |
f679623f RBJ |
3753 | struct bio* align_bi; |
3754 | mdk_rdev_t *rdev; | |
3755 | ||
3756 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3757 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3758 | return 0; |
3759 | } | |
3760 | /* | |
99c0fb5f | 3761 | * use bio_clone to make a copy of the bio |
f679623f RBJ |
3762 | */ |
3763 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3764 | if (!align_bi) | |
3765 | return 0; | |
3766 | /* | |
3767 | * set bi_end_io to a new function, and set bi_private to the | |
3768 | * original bio. | |
3769 | */ | |
3770 | align_bi->bi_end_io = raid5_align_endio; | |
3771 | align_bi->bi_private = raid_bio; | |
3772 | /* | |
3773 | * compute position | |
3774 | */ | |
112bf897 N |
3775 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3776 | 0, | |
911d4ee8 | 3777 | &dd_idx, NULL); |
f679623f RBJ |
3778 | |
3779 | rcu_read_lock(); | |
3780 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3781 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3782 | atomic_inc(&rdev->nr_pending); |
3783 | rcu_read_unlock(); | |
46031f9a RBJ |
3784 | raid_bio->bi_next = (void*)rdev; |
3785 | align_bi->bi_bdev = rdev->bdev; | |
3786 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3787 | align_bi->bi_sector += rdev->data_offset; | |
3788 | ||
387bb173 NB |
3789 | if (!bio_fits_rdev(align_bi)) { |
3790 | /* too big in some way */ | |
3791 | bio_put(align_bi); | |
3792 | rdev_dec_pending(rdev, mddev); | |
3793 | return 0; | |
3794 | } | |
3795 | ||
46031f9a RBJ |
3796 | spin_lock_irq(&conf->device_lock); |
3797 | wait_event_lock_irq(conf->wait_for_stripe, | |
3798 | conf->quiesce == 0, | |
3799 | conf->device_lock, /* nothing */); | |
3800 | atomic_inc(&conf->active_aligned_reads); | |
3801 | spin_unlock_irq(&conf->device_lock); | |
3802 | ||
f679623f RBJ |
3803 | generic_make_request(align_bi); |
3804 | return 1; | |
3805 | } else { | |
3806 | rcu_read_unlock(); | |
46031f9a | 3807 | bio_put(align_bi); |
f679623f RBJ |
3808 | return 0; |
3809 | } | |
3810 | } | |
3811 | ||
8b3e6cdc DW |
3812 | /* __get_priority_stripe - get the next stripe to process |
3813 | * | |
3814 | * Full stripe writes are allowed to pass preread active stripes up until | |
3815 | * the bypass_threshold is exceeded. In general the bypass_count | |
3816 | * increments when the handle_list is handled before the hold_list; however, it | |
3817 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3818 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3819 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3820 | * handle_list. | |
3821 | */ | |
3822 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3823 | { | |
3824 | struct stripe_head *sh; | |
3825 | ||
3826 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3827 | __func__, | |
3828 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3829 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3830 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3831 | ||
3832 | if (!list_empty(&conf->handle_list)) { | |
3833 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3834 | ||
3835 | if (list_empty(&conf->hold_list)) | |
3836 | conf->bypass_count = 0; | |
3837 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3838 | if (conf->hold_list.next == conf->last_hold) | |
3839 | conf->bypass_count++; | |
3840 | else { | |
3841 | conf->last_hold = conf->hold_list.next; | |
3842 | conf->bypass_count -= conf->bypass_threshold; | |
3843 | if (conf->bypass_count < 0) | |
3844 | conf->bypass_count = 0; | |
3845 | } | |
3846 | } | |
3847 | } else if (!list_empty(&conf->hold_list) && | |
3848 | ((conf->bypass_threshold && | |
3849 | conf->bypass_count > conf->bypass_threshold) || | |
3850 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3851 | sh = list_entry(conf->hold_list.next, | |
3852 | typeof(*sh), lru); | |
3853 | conf->bypass_count -= conf->bypass_threshold; | |
3854 | if (conf->bypass_count < 0) | |
3855 | conf->bypass_count = 0; | |
3856 | } else | |
3857 | return NULL; | |
3858 | ||
3859 | list_del_init(&sh->lru); | |
3860 | atomic_inc(&sh->count); | |
3861 | BUG_ON(atomic_read(&sh->count) != 1); | |
3862 | return sh; | |
3863 | } | |
f679623f | 3864 | |
21a52c6d | 3865 | static int make_request(mddev_t *mddev, struct bio * bi) |
1da177e4 | 3866 | { |
070ec55d | 3867 | raid5_conf_t *conf = mddev->private; |
911d4ee8 | 3868 | int dd_idx; |
1da177e4 LT |
3869 | sector_t new_sector; |
3870 | sector_t logical_sector, last_sector; | |
3871 | struct stripe_head *sh; | |
a362357b | 3872 | const int rw = bio_data_dir(bi); |
49077326 | 3873 | int remaining; |
1da177e4 | 3874 | |
1f98a13f | 3875 | if (unlikely(bio_rw_flagged(bi, BIO_RW_BARRIER))) { |
a2826aa9 N |
3876 | /* Drain all pending writes. We only really need |
3877 | * to ensure they have been submitted, but this is | |
3878 | * easier. | |
3879 | */ | |
3880 | mddev->pers->quiesce(mddev, 1); | |
3881 | mddev->pers->quiesce(mddev, 0); | |
3882 | md_barrier_request(mddev, bi); | |
e5dcdd80 N |
3883 | return 0; |
3884 | } | |
3885 | ||
3d310eb7 | 3886 | md_write_start(mddev, bi); |
06d91a5f | 3887 | |
802ba064 | 3888 | if (rw == READ && |
52488615 | 3889 | mddev->reshape_position == MaxSector && |
21a52c6d | 3890 | chunk_aligned_read(mddev,bi)) |
99c0fb5f | 3891 | return 0; |
52488615 | 3892 | |
1da177e4 LT |
3893 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3894 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3895 | bi->bi_next = NULL; | |
3896 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3897 | |
1da177e4 LT |
3898 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3899 | DEFINE_WAIT(w); | |
16a53ecc | 3900 | int disks, data_disks; |
b5663ba4 | 3901 | int previous; |
b578d55f | 3902 | |
7ecaa1e6 | 3903 | retry: |
b5663ba4 | 3904 | previous = 0; |
b0f9ec04 | 3905 | disks = conf->raid_disks; |
b578d55f | 3906 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3907 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3908 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3909 | * 64bit on a 32bit platform, and so it might be |
3910 | * possible to see a half-updated value | |
fef9c61f | 3911 | * Ofcourse reshape_progress could change after |
df8e7f76 N |
3912 | * the lock is dropped, so once we get a reference |
3913 | * to the stripe that we think it is, we will have | |
3914 | * to check again. | |
3915 | */ | |
7ecaa1e6 | 3916 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3917 | if (mddev->delta_disks < 0 |
3918 | ? logical_sector < conf->reshape_progress | |
3919 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3920 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3921 | previous = 1; |
3922 | } else { | |
fef9c61f N |
3923 | if (mddev->delta_disks < 0 |
3924 | ? logical_sector < conf->reshape_safe | |
3925 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3926 | spin_unlock_irq(&conf->device_lock); |
3927 | schedule(); | |
3928 | goto retry; | |
3929 | } | |
3930 | } | |
7ecaa1e6 N |
3931 | spin_unlock_irq(&conf->device_lock); |
3932 | } | |
16a53ecc N |
3933 | data_disks = disks - conf->max_degraded; |
3934 | ||
112bf897 N |
3935 | new_sector = raid5_compute_sector(conf, logical_sector, |
3936 | previous, | |
911d4ee8 | 3937 | &dd_idx, NULL); |
0c55e022 | 3938 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3939 | (unsigned long long)new_sector, |
3940 | (unsigned long long)logical_sector); | |
3941 | ||
b5663ba4 | 3942 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3943 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3944 | if (sh) { |
b0f9ec04 | 3945 | if (unlikely(previous)) { |
7ecaa1e6 | 3946 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3947 | * stripe, so we must do the range check again. |
3948 | * Expansion could still move past after this | |
3949 | * test, but as we are holding a reference to | |
3950 | * 'sh', we know that if that happens, | |
3951 | * STRIPE_EXPANDING will get set and the expansion | |
3952 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3953 | */ |
3954 | int must_retry = 0; | |
3955 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
3956 | if (mddev->delta_disks < 0 |
3957 | ? logical_sector >= conf->reshape_progress | |
3958 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
3959 | /* mismatch, need to try again */ |
3960 | must_retry = 1; | |
3961 | spin_unlock_irq(&conf->device_lock); | |
3962 | if (must_retry) { | |
3963 | release_stripe(sh); | |
7a3ab908 | 3964 | schedule(); |
7ecaa1e6 N |
3965 | goto retry; |
3966 | } | |
3967 | } | |
e62e58a5 | 3968 | |
a5c308d4 N |
3969 | if (bio_data_dir(bi) == WRITE && |
3970 | logical_sector >= mddev->suspend_lo && | |
e464eafd N |
3971 | logical_sector < mddev->suspend_hi) { |
3972 | release_stripe(sh); | |
e62e58a5 N |
3973 | /* As the suspend_* range is controlled by |
3974 | * userspace, we want an interruptible | |
3975 | * wait. | |
3976 | */ | |
3977 | flush_signals(current); | |
3978 | prepare_to_wait(&conf->wait_for_overlap, | |
3979 | &w, TASK_INTERRUPTIBLE); | |
3980 | if (logical_sector >= mddev->suspend_lo && | |
3981 | logical_sector < mddev->suspend_hi) | |
3982 | schedule(); | |
e464eafd N |
3983 | goto retry; |
3984 | } | |
7ecaa1e6 N |
3985 | |
3986 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
3987 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
3988 | /* Stripe is busy expanding or | |
3989 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3990 | * and wait a while |
3991 | */ | |
3992 | raid5_unplug_device(mddev->queue); | |
3993 | release_stripe(sh); | |
3994 | schedule(); | |
3995 | goto retry; | |
3996 | } | |
3997 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3998 | set_bit(STRIPE_HANDLE, &sh->state); |
3999 | clear_bit(STRIPE_DELAYED, &sh->state); | |
729a1866 N |
4000 | if (mddev->barrier && |
4001 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4002 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 4003 | release_stripe(sh); |
1da177e4 LT |
4004 | } else { |
4005 | /* cannot get stripe for read-ahead, just give-up */ | |
4006 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4007 | finish_wait(&conf->wait_for_overlap, &w); | |
4008 | break; | |
4009 | } | |
4010 | ||
4011 | } | |
4012 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4013 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
4014 | spin_unlock_irq(&conf->device_lock); |
4015 | if (remaining == 0) { | |
1da177e4 | 4016 | |
16a53ecc | 4017 | if ( rw == WRITE ) |
1da177e4 | 4018 | md_write_end(mddev); |
6712ecf8 | 4019 | |
0e13fe23 | 4020 | bio_endio(bi, 0); |
1da177e4 | 4021 | } |
729a1866 N |
4022 | |
4023 | if (mddev->barrier) { | |
4024 | /* We need to wait for the stripes to all be handled. | |
4025 | * So: wait for preread_active_stripes to drop to 0. | |
4026 | */ | |
4027 | wait_event(mddev->thread->wqueue, | |
4028 | atomic_read(&conf->preread_active_stripes) == 0); | |
4029 | } | |
1da177e4 LT |
4030 | return 0; |
4031 | } | |
4032 | ||
b522adcd DW |
4033 | static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks); |
4034 | ||
52c03291 | 4035 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4036 | { |
52c03291 N |
4037 | /* reshaping is quite different to recovery/resync so it is |
4038 | * handled quite separately ... here. | |
4039 | * | |
4040 | * On each call to sync_request, we gather one chunk worth of | |
4041 | * destination stripes and flag them as expanding. | |
4042 | * Then we find all the source stripes and request reads. | |
4043 | * As the reads complete, handle_stripe will copy the data | |
4044 | * into the destination stripe and release that stripe. | |
4045 | */ | |
7b92813c | 4046 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 4047 | struct stripe_head *sh; |
ccfcc3c1 | 4048 | sector_t first_sector, last_sector; |
f416885e N |
4049 | int raid_disks = conf->previous_raid_disks; |
4050 | int data_disks = raid_disks - conf->max_degraded; | |
4051 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4052 | int i; |
4053 | int dd_idx; | |
c8f517c4 | 4054 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4055 | sector_t stripe_addr; |
7a661381 | 4056 | int reshape_sectors; |
ab69ae12 | 4057 | struct list_head stripes; |
52c03291 | 4058 | |
fef9c61f N |
4059 | if (sector_nr == 0) { |
4060 | /* If restarting in the middle, skip the initial sectors */ | |
4061 | if (mddev->delta_disks < 0 && | |
4062 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4063 | sector_nr = raid5_size(mddev, 0, 0) | |
4064 | - conf->reshape_progress; | |
a639755c | 4065 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
4066 | conf->reshape_progress > 0) |
4067 | sector_nr = conf->reshape_progress; | |
f416885e | 4068 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4069 | if (sector_nr) { |
8dee7211 N |
4070 | mddev->curr_resync_completed = sector_nr; |
4071 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4072 | *skipped = 1; |
4073 | return sector_nr; | |
4074 | } | |
52c03291 N |
4075 | } |
4076 | ||
7a661381 N |
4077 | /* We need to process a full chunk at a time. |
4078 | * If old and new chunk sizes differ, we need to process the | |
4079 | * largest of these | |
4080 | */ | |
664e7c41 AN |
4081 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4082 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4083 | else |
9d8f0363 | 4084 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4085 | |
52c03291 N |
4086 | /* we update the metadata when there is more than 3Meg |
4087 | * in the block range (that is rather arbitrary, should | |
4088 | * probably be time based) or when the data about to be | |
4089 | * copied would over-write the source of the data at | |
4090 | * the front of the range. | |
fef9c61f N |
4091 | * i.e. one new_stripe along from reshape_progress new_maps |
4092 | * to after where reshape_safe old_maps to | |
52c03291 | 4093 | */ |
fef9c61f | 4094 | writepos = conf->reshape_progress; |
f416885e | 4095 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4096 | readpos = conf->reshape_progress; |
4097 | sector_div(readpos, data_disks); | |
fef9c61f | 4098 | safepos = conf->reshape_safe; |
f416885e | 4099 | sector_div(safepos, data_disks); |
fef9c61f | 4100 | if (mddev->delta_disks < 0) { |
ed37d83e | 4101 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4102 | readpos += reshape_sectors; |
7a661381 | 4103 | safepos += reshape_sectors; |
fef9c61f | 4104 | } else { |
7a661381 | 4105 | writepos += reshape_sectors; |
ed37d83e N |
4106 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4107 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4108 | } |
52c03291 | 4109 | |
c8f517c4 N |
4110 | /* 'writepos' is the most advanced device address we might write. |
4111 | * 'readpos' is the least advanced device address we might read. | |
4112 | * 'safepos' is the least address recorded in the metadata as having | |
4113 | * been reshaped. | |
4114 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4115 | * ensure safety in the face of a crash - that must be done by userspace | |
4116 | * making a backup of the data. So in that case there is no particular | |
4117 | * rush to update metadata. | |
4118 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4119 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4120 | * we can be safe in the event of a crash. | |
4121 | * So we insist on updating metadata if safepos is behind writepos and | |
4122 | * readpos is beyond writepos. | |
4123 | * In any case, update the metadata every 10 seconds. | |
4124 | * Maybe that number should be configurable, but I'm not sure it is | |
4125 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4126 | */ | |
fef9c61f | 4127 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4128 | ? (safepos > writepos && readpos < writepos) |
4129 | : (safepos < writepos && readpos > writepos)) || | |
4130 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4131 | /* Cannot proceed until we've updated the superblock... */ |
4132 | wait_event(conf->wait_for_overlap, | |
4133 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4134 | mddev->reshape_position = conf->reshape_progress; |
acb180b0 | 4135 | mddev->curr_resync_completed = mddev->curr_resync; |
c8f517c4 | 4136 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4137 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4138 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4139 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4140 | kthread_should_stop()); |
4141 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4142 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4143 | spin_unlock_irq(&conf->device_lock); |
4144 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4145 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4146 | } |
4147 | ||
ec32a2bd N |
4148 | if (mddev->delta_disks < 0) { |
4149 | BUG_ON(conf->reshape_progress == 0); | |
4150 | stripe_addr = writepos; | |
4151 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4152 | ~((sector_t)reshape_sectors - 1)) |
4153 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4154 | != sector_nr); |
4155 | } else { | |
7a661381 | 4156 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4157 | stripe_addr = sector_nr; |
4158 | } | |
ab69ae12 | 4159 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4160 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4161 | int j; |
a9f326eb | 4162 | int skipped_disk = 0; |
a8c906ca | 4163 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4164 | set_bit(STRIPE_EXPANDING, &sh->state); |
4165 | atomic_inc(&conf->reshape_stripes); | |
4166 | /* If any of this stripe is beyond the end of the old | |
4167 | * array, then we need to zero those blocks | |
4168 | */ | |
4169 | for (j=sh->disks; j--;) { | |
4170 | sector_t s; | |
4171 | if (j == sh->pd_idx) | |
4172 | continue; | |
f416885e | 4173 | if (conf->level == 6 && |
d0dabf7e | 4174 | j == sh->qd_idx) |
f416885e | 4175 | continue; |
784052ec | 4176 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4177 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4178 | skipped_disk = 1; |
52c03291 N |
4179 | continue; |
4180 | } | |
4181 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4182 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4183 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4184 | } | |
a9f326eb | 4185 | if (!skipped_disk) { |
52c03291 N |
4186 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4187 | set_bit(STRIPE_HANDLE, &sh->state); | |
4188 | } | |
ab69ae12 | 4189 | list_add(&sh->lru, &stripes); |
52c03291 N |
4190 | } |
4191 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4192 | if (mddev->delta_disks < 0) |
7a661381 | 4193 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4194 | else |
7a661381 | 4195 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4196 | spin_unlock_irq(&conf->device_lock); |
4197 | /* Ok, those stripe are ready. We can start scheduling | |
4198 | * reads on the source stripes. | |
4199 | * The source stripes are determined by mapping the first and last | |
4200 | * block on the destination stripes. | |
4201 | */ | |
52c03291 | 4202 | first_sector = |
ec32a2bd | 4203 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4204 | 1, &dd_idx, NULL); |
52c03291 | 4205 | last_sector = |
0e6e0271 | 4206 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4207 | * new_data_disks - 1), |
911d4ee8 | 4208 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4209 | if (last_sector >= mddev->dev_sectors) |
4210 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4211 | while (first_sector <= last_sector) { |
a8c906ca | 4212 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4213 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4214 | set_bit(STRIPE_HANDLE, &sh->state); | |
4215 | release_stripe(sh); | |
4216 | first_sector += STRIPE_SECTORS; | |
4217 | } | |
ab69ae12 N |
4218 | /* Now that the sources are clearly marked, we can release |
4219 | * the destination stripes | |
4220 | */ | |
4221 | while (!list_empty(&stripes)) { | |
4222 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4223 | list_del_init(&sh->lru); | |
4224 | release_stripe(sh); | |
4225 | } | |
c6207277 N |
4226 | /* If this takes us to the resync_max point where we have to pause, |
4227 | * then we need to write out the superblock. | |
4228 | */ | |
7a661381 | 4229 | sector_nr += reshape_sectors; |
c03f6a19 N |
4230 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4231 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4232 | /* Cannot proceed until we've updated the superblock... */ |
4233 | wait_event(conf->wait_for_overlap, | |
4234 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4235 | mddev->reshape_position = conf->reshape_progress; |
48606a9f | 4236 | mddev->curr_resync_completed = mddev->curr_resync + reshape_sectors; |
c8f517c4 | 4237 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4238 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4239 | md_wakeup_thread(mddev->thread); | |
4240 | wait_event(mddev->sb_wait, | |
4241 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4242 | || kthread_should_stop()); | |
4243 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4244 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4245 | spin_unlock_irq(&conf->device_lock); |
4246 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4247 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4248 | } |
7a661381 | 4249 | return reshape_sectors; |
52c03291 N |
4250 | } |
4251 | ||
4252 | /* FIXME go_faster isn't used */ | |
4253 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
4254 | { | |
7b92813c | 4255 | raid5_conf_t *conf = mddev->private; |
52c03291 | 4256 | struct stripe_head *sh; |
58c0fed4 | 4257 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 4258 | int sync_blocks; |
16a53ecc N |
4259 | int still_degraded = 0; |
4260 | int i; | |
1da177e4 | 4261 | |
72626685 | 4262 | if (sector_nr >= max_sector) { |
1da177e4 LT |
4263 | /* just being told to finish up .. nothing much to do */ |
4264 | unplug_slaves(mddev); | |
cea9c228 | 4265 | |
29269553 N |
4266 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4267 | end_reshape(conf); | |
4268 | return 0; | |
4269 | } | |
72626685 N |
4270 | |
4271 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4272 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4273 | &sync_blocks, 1); | |
16a53ecc | 4274 | else /* completed sync */ |
72626685 N |
4275 | conf->fullsync = 0; |
4276 | bitmap_close_sync(mddev->bitmap); | |
4277 | ||
1da177e4 LT |
4278 | return 0; |
4279 | } | |
ccfcc3c1 | 4280 | |
64bd660b N |
4281 | /* Allow raid5_quiesce to complete */ |
4282 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4283 | ||
52c03291 N |
4284 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4285 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4286 | |
c6207277 N |
4287 | /* No need to check resync_max as we never do more than one |
4288 | * stripe, and as resync_max will always be on a chunk boundary, | |
4289 | * if the check in md_do_sync didn't fire, there is no chance | |
4290 | * of overstepping resync_max here | |
4291 | */ | |
4292 | ||
16a53ecc | 4293 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4294 | * to resync, then assert that we are finished, because there is |
4295 | * nothing we can do. | |
4296 | */ | |
3285edf1 | 4297 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4298 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4299 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4300 | *skipped = 1; |
1da177e4 LT |
4301 | return rv; |
4302 | } | |
72626685 | 4303 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4304 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4305 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4306 | /* we can skip this block, and probably more */ | |
4307 | sync_blocks /= STRIPE_SECTORS; | |
4308 | *skipped = 1; | |
4309 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4310 | } | |
1da177e4 | 4311 | |
b47490c9 N |
4312 | |
4313 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
4314 | ||
a8c906ca | 4315 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4316 | if (sh == NULL) { |
a8c906ca | 4317 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4318 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4319 | * is trying to get access |
1da177e4 | 4320 | */ |
66c006a5 | 4321 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4322 | } |
16a53ecc N |
4323 | /* Need to check if array will still be degraded after recovery/resync |
4324 | * We don't need to check the 'failed' flag as when that gets set, | |
4325 | * recovery aborts. | |
4326 | */ | |
f001a70c | 4327 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4328 | if (conf->disks[i].rdev == NULL) |
4329 | still_degraded = 1; | |
4330 | ||
4331 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4332 | ||
4333 | spin_lock(&sh->lock); | |
1da177e4 LT |
4334 | set_bit(STRIPE_SYNCING, &sh->state); |
4335 | clear_bit(STRIPE_INSYNC, &sh->state); | |
4336 | spin_unlock(&sh->lock); | |
4337 | ||
1442577b | 4338 | handle_stripe(sh); |
1da177e4 LT |
4339 | release_stripe(sh); |
4340 | ||
4341 | return STRIPE_SECTORS; | |
4342 | } | |
4343 | ||
46031f9a RBJ |
4344 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
4345 | { | |
4346 | /* We may not be able to submit a whole bio at once as there | |
4347 | * may not be enough stripe_heads available. | |
4348 | * We cannot pre-allocate enough stripe_heads as we may need | |
4349 | * more than exist in the cache (if we allow ever large chunks). | |
4350 | * So we do one stripe head at a time and record in | |
4351 | * ->bi_hw_segments how many have been done. | |
4352 | * | |
4353 | * We *know* that this entire raid_bio is in one chunk, so | |
4354 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4355 | */ | |
4356 | struct stripe_head *sh; | |
911d4ee8 | 4357 | int dd_idx; |
46031f9a RBJ |
4358 | sector_t sector, logical_sector, last_sector; |
4359 | int scnt = 0; | |
4360 | int remaining; | |
4361 | int handled = 0; | |
4362 | ||
4363 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4364 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4365 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4366 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4367 | ||
4368 | for (; logical_sector < last_sector; | |
387bb173 NB |
4369 | logical_sector += STRIPE_SECTORS, |
4370 | sector += STRIPE_SECTORS, | |
4371 | scnt++) { | |
46031f9a | 4372 | |
960e739d | 4373 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4374 | /* already done this stripe */ |
4375 | continue; | |
4376 | ||
a8c906ca | 4377 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4378 | |
4379 | if (!sh) { | |
4380 | /* failed to get a stripe - must wait */ | |
960e739d | 4381 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4382 | conf->retry_read_aligned = raid_bio; |
4383 | return handled; | |
4384 | } | |
4385 | ||
4386 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4387 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4388 | release_stripe(sh); | |
960e739d | 4389 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4390 | conf->retry_read_aligned = raid_bio; |
4391 | return handled; | |
4392 | } | |
4393 | ||
36d1c647 | 4394 | handle_stripe(sh); |
46031f9a RBJ |
4395 | release_stripe(sh); |
4396 | handled++; | |
4397 | } | |
4398 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4399 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4400 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4401 | if (remaining == 0) |
4402 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4403 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4404 | wake_up(&conf->wait_for_stripe); | |
4405 | return handled; | |
4406 | } | |
4407 | ||
46031f9a | 4408 | |
1da177e4 LT |
4409 | /* |
4410 | * This is our raid5 kernel thread. | |
4411 | * | |
4412 | * We scan the hash table for stripes which can be handled now. | |
4413 | * During the scan, completed stripes are saved for us by the interrupt | |
4414 | * handler, so that they will not have to wait for our next wakeup. | |
4415 | */ | |
6ed3003c | 4416 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4417 | { |
4418 | struct stripe_head *sh; | |
070ec55d | 4419 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
4420 | int handled; |
4421 | ||
45b4233c | 4422 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4423 | |
4424 | md_check_recovery(mddev); | |
1da177e4 LT |
4425 | |
4426 | handled = 0; | |
4427 | spin_lock_irq(&conf->device_lock); | |
4428 | while (1) { | |
46031f9a | 4429 | struct bio *bio; |
1da177e4 | 4430 | |
ae3c20cc | 4431 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 4432 | int seq = conf->seq_flush; |
700e432d | 4433 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4434 | bitmap_unplug(mddev->bitmap); |
700e432d | 4435 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4436 | conf->seq_write = seq; |
4437 | activate_bit_delay(conf); | |
4438 | } | |
4439 | ||
46031f9a RBJ |
4440 | while ((bio = remove_bio_from_retry(conf))) { |
4441 | int ok; | |
4442 | spin_unlock_irq(&conf->device_lock); | |
4443 | ok = retry_aligned_read(conf, bio); | |
4444 | spin_lock_irq(&conf->device_lock); | |
4445 | if (!ok) | |
4446 | break; | |
4447 | handled++; | |
4448 | } | |
4449 | ||
8b3e6cdc DW |
4450 | sh = __get_priority_stripe(conf); |
4451 | ||
c9f21aaf | 4452 | if (!sh) |
1da177e4 | 4453 | break; |
1da177e4 LT |
4454 | spin_unlock_irq(&conf->device_lock); |
4455 | ||
4456 | handled++; | |
417b8d4a DW |
4457 | handle_stripe(sh); |
4458 | release_stripe(sh); | |
4459 | cond_resched(); | |
1da177e4 LT |
4460 | |
4461 | spin_lock_irq(&conf->device_lock); | |
4462 | } | |
45b4233c | 4463 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4464 | |
4465 | spin_unlock_irq(&conf->device_lock); | |
4466 | ||
c9f21aaf | 4467 | async_tx_issue_pending_all(); |
1da177e4 LT |
4468 | unplug_slaves(mddev); |
4469 | ||
45b4233c | 4470 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4471 | } |
4472 | ||
3f294f4f | 4473 | static ssize_t |
007583c9 | 4474 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4475 | { |
070ec55d | 4476 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4477 | if (conf) |
4478 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4479 | else | |
4480 | return 0; | |
3f294f4f N |
4481 | } |
4482 | ||
4483 | static ssize_t | |
007583c9 | 4484 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4485 | { |
070ec55d | 4486 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4487 | unsigned long new; |
b5470dc5 DW |
4488 | int err; |
4489 | ||
3f294f4f N |
4490 | if (len >= PAGE_SIZE) |
4491 | return -EINVAL; | |
96de1e66 N |
4492 | if (!conf) |
4493 | return -ENODEV; | |
3f294f4f | 4494 | |
4ef197d8 | 4495 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4496 | return -EINVAL; |
4497 | if (new <= 16 || new > 32768) | |
4498 | return -EINVAL; | |
4499 | while (new < conf->max_nr_stripes) { | |
4500 | if (drop_one_stripe(conf)) | |
4501 | conf->max_nr_stripes--; | |
4502 | else | |
4503 | break; | |
4504 | } | |
b5470dc5 DW |
4505 | err = md_allow_write(mddev); |
4506 | if (err) | |
4507 | return err; | |
3f294f4f N |
4508 | while (new > conf->max_nr_stripes) { |
4509 | if (grow_one_stripe(conf)) | |
4510 | conf->max_nr_stripes++; | |
4511 | else break; | |
4512 | } | |
4513 | return len; | |
4514 | } | |
007583c9 | 4515 | |
96de1e66 N |
4516 | static struct md_sysfs_entry |
4517 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4518 | raid5_show_stripe_cache_size, | |
4519 | raid5_store_stripe_cache_size); | |
3f294f4f | 4520 | |
8b3e6cdc DW |
4521 | static ssize_t |
4522 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4523 | { | |
070ec55d | 4524 | raid5_conf_t *conf = mddev->private; |
8b3e6cdc DW |
4525 | if (conf) |
4526 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4527 | else | |
4528 | return 0; | |
4529 | } | |
4530 | ||
4531 | static ssize_t | |
4532 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4533 | { | |
070ec55d | 4534 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4535 | unsigned long new; |
8b3e6cdc DW |
4536 | if (len >= PAGE_SIZE) |
4537 | return -EINVAL; | |
4538 | if (!conf) | |
4539 | return -ENODEV; | |
4540 | ||
4ef197d8 | 4541 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4542 | return -EINVAL; |
4ef197d8 | 4543 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4544 | return -EINVAL; |
4545 | conf->bypass_threshold = new; | |
4546 | return len; | |
4547 | } | |
4548 | ||
4549 | static struct md_sysfs_entry | |
4550 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4551 | S_IRUGO | S_IWUSR, | |
4552 | raid5_show_preread_threshold, | |
4553 | raid5_store_preread_threshold); | |
4554 | ||
3f294f4f | 4555 | static ssize_t |
96de1e66 | 4556 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4557 | { |
070ec55d | 4558 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4559 | if (conf) |
4560 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4561 | else | |
4562 | return 0; | |
3f294f4f N |
4563 | } |
4564 | ||
96de1e66 N |
4565 | static struct md_sysfs_entry |
4566 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4567 | |
007583c9 | 4568 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4569 | &raid5_stripecache_size.attr, |
4570 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4571 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4572 | NULL, |
4573 | }; | |
007583c9 N |
4574 | static struct attribute_group raid5_attrs_group = { |
4575 | .name = NULL, | |
4576 | .attrs = raid5_attrs, | |
3f294f4f N |
4577 | }; |
4578 | ||
80c3a6ce DW |
4579 | static sector_t |
4580 | raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
4581 | { | |
070ec55d | 4582 | raid5_conf_t *conf = mddev->private; |
80c3a6ce DW |
4583 | |
4584 | if (!sectors) | |
4585 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4586 | if (!raid_disks) |
7ec05478 | 4587 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4588 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4589 | |
9d8f0363 | 4590 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4591 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4592 | return sectors * (raid_disks - conf->max_degraded); |
4593 | } | |
4594 | ||
36d1c647 DW |
4595 | static void raid5_free_percpu(raid5_conf_t *conf) |
4596 | { | |
4597 | struct raid5_percpu *percpu; | |
4598 | unsigned long cpu; | |
4599 | ||
4600 | if (!conf->percpu) | |
4601 | return; | |
4602 | ||
4603 | get_online_cpus(); | |
4604 | for_each_possible_cpu(cpu) { | |
4605 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4606 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4607 | kfree(percpu->scribble); |
36d1c647 DW |
4608 | } |
4609 | #ifdef CONFIG_HOTPLUG_CPU | |
4610 | unregister_cpu_notifier(&conf->cpu_notify); | |
4611 | #endif | |
4612 | put_online_cpus(); | |
4613 | ||
4614 | free_percpu(conf->percpu); | |
4615 | } | |
4616 | ||
95fc17aa DW |
4617 | static void free_conf(raid5_conf_t *conf) |
4618 | { | |
4619 | shrink_stripes(conf); | |
36d1c647 | 4620 | raid5_free_percpu(conf); |
95fc17aa DW |
4621 | kfree(conf->disks); |
4622 | kfree(conf->stripe_hashtbl); | |
4623 | kfree(conf); | |
4624 | } | |
4625 | ||
36d1c647 DW |
4626 | #ifdef CONFIG_HOTPLUG_CPU |
4627 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4628 | void *hcpu) | |
4629 | { | |
4630 | raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify); | |
4631 | long cpu = (long)hcpu; | |
4632 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4633 | ||
4634 | switch (action) { | |
4635 | case CPU_UP_PREPARE: | |
4636 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4637 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4638 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4639 | if (!percpu->scribble) |
4640 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4641 | ||
4642 | if (!percpu->scribble || | |
4643 | (conf->level == 6 && !percpu->spare_page)) { | |
4644 | safe_put_page(percpu->spare_page); | |
4645 | kfree(percpu->scribble); | |
36d1c647 DW |
4646 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4647 | __func__, cpu); | |
55af6bb5 | 4648 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4649 | } |
4650 | break; | |
4651 | case CPU_DEAD: | |
4652 | case CPU_DEAD_FROZEN: | |
4653 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4654 | kfree(percpu->scribble); |
36d1c647 | 4655 | percpu->spare_page = NULL; |
d6f38f31 | 4656 | percpu->scribble = NULL; |
36d1c647 DW |
4657 | break; |
4658 | default: | |
4659 | break; | |
4660 | } | |
4661 | return NOTIFY_OK; | |
4662 | } | |
4663 | #endif | |
4664 | ||
4665 | static int raid5_alloc_percpu(raid5_conf_t *conf) | |
4666 | { | |
4667 | unsigned long cpu; | |
4668 | struct page *spare_page; | |
a29d8b8e | 4669 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4670 | void *scribble; |
36d1c647 DW |
4671 | int err; |
4672 | ||
36d1c647 DW |
4673 | allcpus = alloc_percpu(struct raid5_percpu); |
4674 | if (!allcpus) | |
4675 | return -ENOMEM; | |
4676 | conf->percpu = allcpus; | |
4677 | ||
4678 | get_online_cpus(); | |
4679 | err = 0; | |
4680 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4681 | if (conf->level == 6) { |
4682 | spare_page = alloc_page(GFP_KERNEL); | |
4683 | if (!spare_page) { | |
4684 | err = -ENOMEM; | |
4685 | break; | |
4686 | } | |
4687 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4688 | } | |
5e5e3e78 | 4689 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4690 | if (!scribble) { |
36d1c647 DW |
4691 | err = -ENOMEM; |
4692 | break; | |
4693 | } | |
d6f38f31 | 4694 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4695 | } |
4696 | #ifdef CONFIG_HOTPLUG_CPU | |
4697 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4698 | conf->cpu_notify.priority = 0; | |
4699 | if (err == 0) | |
4700 | err = register_cpu_notifier(&conf->cpu_notify); | |
4701 | #endif | |
4702 | put_online_cpus(); | |
4703 | ||
4704 | return err; | |
4705 | } | |
4706 | ||
91adb564 | 4707 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4708 | { |
4709 | raid5_conf_t *conf; | |
5e5e3e78 | 4710 | int raid_disk, memory, max_disks; |
1da177e4 LT |
4711 | mdk_rdev_t *rdev; |
4712 | struct disk_info *disk; | |
1da177e4 | 4713 | |
91adb564 N |
4714 | if (mddev->new_level != 5 |
4715 | && mddev->new_level != 4 | |
4716 | && mddev->new_level != 6) { | |
0c55e022 | 4717 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4718 | mdname(mddev), mddev->new_level); |
4719 | return ERR_PTR(-EIO); | |
1da177e4 | 4720 | } |
91adb564 N |
4721 | if ((mddev->new_level == 5 |
4722 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4723 | (mddev->new_level == 6 | |
4724 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4725 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4726 | mdname(mddev), mddev->new_layout); |
4727 | return ERR_PTR(-EIO); | |
99c0fb5f | 4728 | } |
91adb564 | 4729 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4730 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4731 | mdname(mddev), mddev->raid_disks); |
4732 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4733 | } |
4734 | ||
664e7c41 AN |
4735 | if (!mddev->new_chunk_sectors || |
4736 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4737 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4738 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4739 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4740 | return ERR_PTR(-EINVAL); |
f6705578 N |
4741 | } |
4742 | ||
91adb564 N |
4743 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4744 | if (conf == NULL) | |
1da177e4 | 4745 | goto abort; |
f5efd45a DW |
4746 | spin_lock_init(&conf->device_lock); |
4747 | init_waitqueue_head(&conf->wait_for_stripe); | |
4748 | init_waitqueue_head(&conf->wait_for_overlap); | |
4749 | INIT_LIST_HEAD(&conf->handle_list); | |
4750 | INIT_LIST_HEAD(&conf->hold_list); | |
4751 | INIT_LIST_HEAD(&conf->delayed_list); | |
4752 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4753 | INIT_LIST_HEAD(&conf->inactive_list); | |
4754 | atomic_set(&conf->active_stripes, 0); | |
4755 | atomic_set(&conf->preread_active_stripes, 0); | |
4756 | atomic_set(&conf->active_aligned_reads, 0); | |
4757 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
91adb564 N |
4758 | |
4759 | conf->raid_disks = mddev->raid_disks; | |
4760 | if (mddev->reshape_position == MaxSector) | |
4761 | conf->previous_raid_disks = mddev->raid_disks; | |
4762 | else | |
f6705578 | 4763 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4764 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4765 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4766 | |
5e5e3e78 | 4767 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4768 | GFP_KERNEL); |
4769 | if (!conf->disks) | |
4770 | goto abort; | |
9ffae0cf | 4771 | |
1da177e4 LT |
4772 | conf->mddev = mddev; |
4773 | ||
fccddba0 | 4774 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4775 | goto abort; |
1da177e4 | 4776 | |
36d1c647 DW |
4777 | conf->level = mddev->new_level; |
4778 | if (raid5_alloc_percpu(conf) != 0) | |
4779 | goto abort; | |
4780 | ||
0c55e022 | 4781 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4782 | |
159ec1fc | 4783 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4784 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4785 | if (raid_disk >= max_disks |
1da177e4 LT |
4786 | || raid_disk < 0) |
4787 | continue; | |
4788 | disk = conf->disks + raid_disk; | |
4789 | ||
4790 | disk->rdev = rdev; | |
4791 | ||
b2d444d7 | 4792 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 4793 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
4794 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
4795 | " disk %d\n", | |
4796 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
8c2e870a NB |
4797 | } else |
4798 | /* Cannot rely on bitmap to complete recovery */ | |
4799 | conf->fullsync = 1; | |
1da177e4 LT |
4800 | } |
4801 | ||
09c9e5fa | 4802 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4803 | conf->level = mddev->new_level; |
16a53ecc N |
4804 | if (conf->level == 6) |
4805 | conf->max_degraded = 2; | |
4806 | else | |
4807 | conf->max_degraded = 1; | |
91adb564 | 4808 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4809 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4810 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4811 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4812 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4813 | conf->prev_algo = mddev->layout; |
4814 | } | |
1da177e4 | 4815 | |
91adb564 | 4816 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4817 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4818 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4819 | printk(KERN_ERR | |
0c55e022 N |
4820 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
4821 | mdname(mddev), memory); | |
91adb564 N |
4822 | goto abort; |
4823 | } else | |
0c55e022 N |
4824 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
4825 | mdname(mddev), memory); | |
1da177e4 | 4826 | |
0da3c619 | 4827 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4828 | if (!conf->thread) { |
4829 | printk(KERN_ERR | |
0c55e022 | 4830 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 4831 | mdname(mddev)); |
16a53ecc N |
4832 | goto abort; |
4833 | } | |
91adb564 N |
4834 | |
4835 | return conf; | |
4836 | ||
4837 | abort: | |
4838 | if (conf) { | |
95fc17aa | 4839 | free_conf(conf); |
91adb564 N |
4840 | return ERR_PTR(-EIO); |
4841 | } else | |
4842 | return ERR_PTR(-ENOMEM); | |
4843 | } | |
4844 | ||
c148ffdc N |
4845 | |
4846 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4847 | { | |
4848 | switch (algo) { | |
4849 | case ALGORITHM_PARITY_0: | |
4850 | if (raid_disk < max_degraded) | |
4851 | return 1; | |
4852 | break; | |
4853 | case ALGORITHM_PARITY_N: | |
4854 | if (raid_disk >= raid_disks - max_degraded) | |
4855 | return 1; | |
4856 | break; | |
4857 | case ALGORITHM_PARITY_0_6: | |
4858 | if (raid_disk == 0 || | |
4859 | raid_disk == raid_disks - 1) | |
4860 | return 1; | |
4861 | break; | |
4862 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4863 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4864 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4865 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4866 | if (raid_disk == raid_disks - 1) | |
4867 | return 1; | |
4868 | } | |
4869 | return 0; | |
4870 | } | |
4871 | ||
91adb564 N |
4872 | static int run(mddev_t *mddev) |
4873 | { | |
4874 | raid5_conf_t *conf; | |
8f6c2e4b | 4875 | int working_disks = 0, chunk_size; |
c148ffdc | 4876 | int dirty_parity_disks = 0; |
91adb564 | 4877 | mdk_rdev_t *rdev; |
c148ffdc | 4878 | sector_t reshape_offset = 0; |
91adb564 | 4879 | |
8c6ac868 | 4880 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 4881 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
4882 | " -- starting background reconstruction\n", |
4883 | mdname(mddev)); | |
91adb564 N |
4884 | if (mddev->reshape_position != MaxSector) { |
4885 | /* Check that we can continue the reshape. | |
4886 | * Currently only disks can change, it must | |
4887 | * increase, and we must be past the point where | |
4888 | * a stripe over-writes itself | |
4889 | */ | |
4890 | sector_t here_new, here_old; | |
4891 | int old_disks; | |
18b00334 | 4892 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4893 | |
88ce4930 | 4894 | if (mddev->new_level != mddev->level) { |
0c55e022 | 4895 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
4896 | "required - aborting.\n", |
4897 | mdname(mddev)); | |
4898 | return -EINVAL; | |
4899 | } | |
91adb564 N |
4900 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4901 | /* reshape_position must be on a new-stripe boundary, and one | |
4902 | * further up in new geometry must map after here in old | |
4903 | * geometry. | |
4904 | */ | |
4905 | here_new = mddev->reshape_position; | |
664e7c41 | 4906 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 4907 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
4908 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
4909 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
4910 | return -EINVAL; |
4911 | } | |
c148ffdc | 4912 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4913 | /* here_new is the stripe we will write to */ |
4914 | here_old = mddev->reshape_position; | |
9d8f0363 | 4915 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4916 | (old_disks-max_degraded)); |
4917 | /* here_old is the first stripe that we might need to read | |
4918 | * from */ | |
67ac6011 N |
4919 | if (mddev->delta_disks == 0) { |
4920 | /* We cannot be sure it is safe to start an in-place | |
4921 | * reshape. It is only safe if user-space if monitoring | |
4922 | * and taking constant backups. | |
4923 | * mdadm always starts a situation like this in | |
4924 | * readonly mode so it can take control before | |
4925 | * allowing any writes. So just check for that. | |
4926 | */ | |
4927 | if ((here_new * mddev->new_chunk_sectors != | |
4928 | here_old * mddev->chunk_sectors) || | |
4929 | mddev->ro == 0) { | |
0c55e022 N |
4930 | printk(KERN_ERR "md/raid:%s: in-place reshape must be started" |
4931 | " in read-only mode - aborting\n", | |
4932 | mdname(mddev)); | |
67ac6011 N |
4933 | return -EINVAL; |
4934 | } | |
4935 | } else if (mddev->delta_disks < 0 | |
4936 | ? (here_new * mddev->new_chunk_sectors <= | |
4937 | here_old * mddev->chunk_sectors) | |
4938 | : (here_new * mddev->new_chunk_sectors >= | |
4939 | here_old * mddev->chunk_sectors)) { | |
91adb564 | 4940 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
4941 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
4942 | "auto-recovery - aborting.\n", | |
4943 | mdname(mddev)); | |
91adb564 N |
4944 | return -EINVAL; |
4945 | } | |
0c55e022 N |
4946 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
4947 | mdname(mddev)); | |
91adb564 N |
4948 | /* OK, we should be able to continue; */ |
4949 | } else { | |
4950 | BUG_ON(mddev->level != mddev->new_level); | |
4951 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 4952 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 4953 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 4954 | } |
91adb564 | 4955 | |
245f46c2 N |
4956 | if (mddev->private == NULL) |
4957 | conf = setup_conf(mddev); | |
4958 | else | |
4959 | conf = mddev->private; | |
4960 | ||
91adb564 N |
4961 | if (IS_ERR(conf)) |
4962 | return PTR_ERR(conf); | |
4963 | ||
4964 | mddev->thread = conf->thread; | |
4965 | conf->thread = NULL; | |
4966 | mddev->private = conf; | |
4967 | ||
4968 | /* | |
4969 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4970 | */ | |
c148ffdc N |
4971 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
4972 | if (rdev->raid_disk < 0) | |
4973 | continue; | |
4974 | if (test_bit(In_sync, &rdev->flags)) | |
91adb564 | 4975 | working_disks++; |
c148ffdc N |
4976 | /* This disc is not fully in-sync. However if it |
4977 | * just stored parity (beyond the recovery_offset), | |
4978 | * when we don't need to be concerned about the | |
4979 | * array being dirty. | |
4980 | * When reshape goes 'backwards', we never have | |
4981 | * partially completed devices, so we only need | |
4982 | * to worry about reshape going forwards. | |
4983 | */ | |
4984 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
4985 | if (mddev->major_version == 0 && | |
4986 | mddev->minor_version > 90) | |
4987 | rdev->recovery_offset = reshape_offset; | |
4988 | ||
c148ffdc N |
4989 | if (rdev->recovery_offset < reshape_offset) { |
4990 | /* We need to check old and new layout */ | |
4991 | if (!only_parity(rdev->raid_disk, | |
4992 | conf->algorithm, | |
4993 | conf->raid_disks, | |
4994 | conf->max_degraded)) | |
4995 | continue; | |
4996 | } | |
4997 | if (!only_parity(rdev->raid_disk, | |
4998 | conf->prev_algo, | |
4999 | conf->previous_raid_disks, | |
5000 | conf->max_degraded)) | |
5001 | continue; | |
5002 | dirty_parity_disks++; | |
5003 | } | |
91adb564 | 5004 | |
5e5e3e78 N |
5005 | mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks) |
5006 | - working_disks); | |
91adb564 | 5007 | |
16a53ecc | 5008 | if (mddev->degraded > conf->max_degraded) { |
0c55e022 | 5009 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 5010 | " (%d/%d failed)\n", |
02c2de8c | 5011 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
5012 | goto abort; |
5013 | } | |
5014 | ||
91adb564 | 5015 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 5016 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
5017 | mddev->resync_max_sectors = mddev->dev_sectors; |
5018 | ||
c148ffdc | 5019 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 5020 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
5021 | if (mddev->ok_start_degraded) |
5022 | printk(KERN_WARNING | |
0c55e022 N |
5023 | "md/raid:%s: starting dirty degraded array" |
5024 | " - data corruption possible.\n", | |
6ff8d8ec N |
5025 | mdname(mddev)); |
5026 | else { | |
5027 | printk(KERN_ERR | |
0c55e022 | 5028 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
5029 | mdname(mddev)); |
5030 | goto abort; | |
5031 | } | |
1da177e4 LT |
5032 | } |
5033 | ||
1da177e4 | 5034 | if (mddev->degraded == 0) |
0c55e022 N |
5035 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
5036 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
5037 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
5038 | mddev->new_layout); | |
1da177e4 | 5039 | else |
0c55e022 N |
5040 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
5041 | " out of %d devices, algorithm %d\n", | |
5042 | mdname(mddev), conf->level, | |
5043 | mddev->raid_disks - mddev->degraded, | |
5044 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
5045 | |
5046 | print_raid5_conf(conf); | |
5047 | ||
fef9c61f | 5048 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 5049 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5050 | atomic_set(&conf->reshape_stripes, 0); |
5051 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5052 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5053 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5054 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5055 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5056 | "reshape"); |
f6705578 N |
5057 | } |
5058 | ||
1da177e4 | 5059 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 5060 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
5061 | */ |
5062 | { | |
16a53ecc N |
5063 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
5064 | int stripe = data_disks * | |
9d8f0363 | 5065 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
1da177e4 LT |
5066 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5067 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5068 | } | |
5069 | ||
5070 | /* Ok, everything is just fine now */ | |
a64c876f N |
5071 | if (mddev->to_remove == &raid5_attrs_group) |
5072 | mddev->to_remove = NULL; | |
5073 | else if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 5074 | printk(KERN_WARNING |
0c55e022 | 5075 | "md/raid:%s: failed to create sysfs attributes.\n", |
5e55e2f5 | 5076 | mdname(mddev)); |
7a5febe9 | 5077 | |
91adb564 N |
5078 | mddev->queue->queue_lock = &conf->device_lock; |
5079 | ||
7a5febe9 | 5080 | mddev->queue->unplug_fn = raid5_unplug_device; |
f022b2fd | 5081 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 5082 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 5083 | |
1f403624 | 5084 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5085 | |
23032a0e | 5086 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
8f6c2e4b MP |
5087 | chunk_size = mddev->chunk_sectors << 9; |
5088 | blk_queue_io_min(mddev->queue, chunk_size); | |
5089 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5090 | (conf->raid_disks - conf->max_degraded)); | |
5091 | ||
5092 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5093 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
5094 | rdev->data_offset << 9); | |
23032a0e | 5095 | |
1da177e4 LT |
5096 | return 0; |
5097 | abort: | |
e0cf8f04 | 5098 | md_unregister_thread(mddev->thread); |
91adb564 | 5099 | mddev->thread = NULL; |
1da177e4 LT |
5100 | if (conf) { |
5101 | print_raid5_conf(conf); | |
95fc17aa | 5102 | free_conf(conf); |
1da177e4 LT |
5103 | } |
5104 | mddev->private = NULL; | |
0c55e022 | 5105 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
5106 | return -EIO; |
5107 | } | |
5108 | ||
3f294f4f | 5109 | static int stop(mddev_t *mddev) |
1da177e4 | 5110 | { |
7b92813c | 5111 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
5112 | |
5113 | md_unregister_thread(mddev->thread); | |
5114 | mddev->thread = NULL; | |
041ae52e | 5115 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 5116 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
95fc17aa | 5117 | free_conf(conf); |
a64c876f N |
5118 | mddev->private = NULL; |
5119 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
5120 | return 0; |
5121 | } | |
5122 | ||
45b4233c | 5123 | #ifdef DEBUG |
d710e138 | 5124 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
5125 | { |
5126 | int i; | |
5127 | ||
16a53ecc N |
5128 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
5129 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
5130 | seq_printf(seq, "sh %llu, count %d.\n", | |
5131 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
5132 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 5133 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
5134 | seq_printf(seq, "(cache%d: %p %ld) ", |
5135 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 5136 | } |
16a53ecc | 5137 | seq_printf(seq, "\n"); |
1da177e4 LT |
5138 | } |
5139 | ||
d710e138 | 5140 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
5141 | { |
5142 | struct stripe_head *sh; | |
fccddba0 | 5143 | struct hlist_node *hn; |
1da177e4 LT |
5144 | int i; |
5145 | ||
5146 | spin_lock_irq(&conf->device_lock); | |
5147 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 5148 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
5149 | if (sh->raid_conf != conf) |
5150 | continue; | |
16a53ecc | 5151 | print_sh(seq, sh); |
1da177e4 LT |
5152 | } |
5153 | } | |
5154 | spin_unlock_irq(&conf->device_lock); | |
5155 | } | |
5156 | #endif | |
5157 | ||
d710e138 | 5158 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 | 5159 | { |
7b92813c | 5160 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
5161 | int i; |
5162 | ||
9d8f0363 AN |
5163 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5164 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5165 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5166 | for (i = 0; i < conf->raid_disks; i++) |
5167 | seq_printf (seq, "%s", | |
5168 | conf->disks[i].rdev && | |
b2d444d7 | 5169 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5170 | seq_printf (seq, "]"); |
45b4233c | 5171 | #ifdef DEBUG |
16a53ecc N |
5172 | seq_printf (seq, "\n"); |
5173 | printall(seq, conf); | |
1da177e4 LT |
5174 | #endif |
5175 | } | |
5176 | ||
5177 | static void print_raid5_conf (raid5_conf_t *conf) | |
5178 | { | |
5179 | int i; | |
5180 | struct disk_info *tmp; | |
5181 | ||
0c55e022 | 5182 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
5183 | if (!conf) { |
5184 | printk("(conf==NULL)\n"); | |
5185 | return; | |
5186 | } | |
0c55e022 N |
5187 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
5188 | conf->raid_disks, | |
5189 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5190 | |
5191 | for (i = 0; i < conf->raid_disks; i++) { | |
5192 | char b[BDEVNAME_SIZE]; | |
5193 | tmp = conf->disks + i; | |
5194 | if (tmp->rdev) | |
0c55e022 N |
5195 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
5196 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
5197 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
5198 | } |
5199 | } | |
5200 | ||
5201 | static int raid5_spare_active(mddev_t *mddev) | |
5202 | { | |
5203 | int i; | |
5204 | raid5_conf_t *conf = mddev->private; | |
5205 | struct disk_info *tmp; | |
5206 | ||
5207 | for (i = 0; i < conf->raid_disks; i++) { | |
5208 | tmp = conf->disks + i; | |
5209 | if (tmp->rdev | |
b2d444d7 | 5210 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
5211 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
5212 | unsigned long flags; | |
5213 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 5214 | mddev->degraded--; |
c04be0aa | 5215 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
5216 | } |
5217 | } | |
5218 | print_raid5_conf(conf); | |
5219 | return 0; | |
5220 | } | |
5221 | ||
5222 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
5223 | { | |
5224 | raid5_conf_t *conf = mddev->private; | |
5225 | int err = 0; | |
5226 | mdk_rdev_t *rdev; | |
5227 | struct disk_info *p = conf->disks + number; | |
5228 | ||
5229 | print_raid5_conf(conf); | |
5230 | rdev = p->rdev; | |
5231 | if (rdev) { | |
ec32a2bd N |
5232 | if (number >= conf->raid_disks && |
5233 | conf->reshape_progress == MaxSector) | |
5234 | clear_bit(In_sync, &rdev->flags); | |
5235 | ||
b2d444d7 | 5236 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
5237 | atomic_read(&rdev->nr_pending)) { |
5238 | err = -EBUSY; | |
5239 | goto abort; | |
5240 | } | |
dfc70645 N |
5241 | /* Only remove non-faulty devices if recovery |
5242 | * isn't possible. | |
5243 | */ | |
5244 | if (!test_bit(Faulty, &rdev->flags) && | |
ec32a2bd N |
5245 | mddev->degraded <= conf->max_degraded && |
5246 | number < conf->raid_disks) { | |
dfc70645 N |
5247 | err = -EBUSY; |
5248 | goto abort; | |
5249 | } | |
1da177e4 | 5250 | p->rdev = NULL; |
fbd568a3 | 5251 | synchronize_rcu(); |
1da177e4 LT |
5252 | if (atomic_read(&rdev->nr_pending)) { |
5253 | /* lost the race, try later */ | |
5254 | err = -EBUSY; | |
5255 | p->rdev = rdev; | |
5256 | } | |
5257 | } | |
5258 | abort: | |
5259 | ||
5260 | print_raid5_conf(conf); | |
5261 | return err; | |
5262 | } | |
5263 | ||
5264 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
5265 | { | |
5266 | raid5_conf_t *conf = mddev->private; | |
199050ea | 5267 | int err = -EEXIST; |
1da177e4 LT |
5268 | int disk; |
5269 | struct disk_info *p; | |
6c2fce2e NB |
5270 | int first = 0; |
5271 | int last = conf->raid_disks - 1; | |
1da177e4 | 5272 | |
16a53ecc | 5273 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 5274 | /* no point adding a device */ |
199050ea | 5275 | return -EINVAL; |
1da177e4 | 5276 | |
6c2fce2e NB |
5277 | if (rdev->raid_disk >= 0) |
5278 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5279 | |
5280 | /* | |
16a53ecc N |
5281 | * find the disk ... but prefer rdev->saved_raid_disk |
5282 | * if possible. | |
1da177e4 | 5283 | */ |
16a53ecc | 5284 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5285 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5286 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5287 | disk = rdev->saved_raid_disk; | |
5288 | else | |
6c2fce2e NB |
5289 | disk = first; |
5290 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5291 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5292 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5293 | rdev->raid_disk = disk; |
199050ea | 5294 | err = 0; |
72626685 N |
5295 | if (rdev->saved_raid_disk != disk) |
5296 | conf->fullsync = 1; | |
d6065f7b | 5297 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5298 | break; |
5299 | } | |
5300 | print_raid5_conf(conf); | |
199050ea | 5301 | return err; |
1da177e4 LT |
5302 | } |
5303 | ||
5304 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
5305 | { | |
5306 | /* no resync is happening, and there is enough space | |
5307 | * on all devices, so we can resize. | |
5308 | * We need to make sure resync covers any new space. | |
5309 | * If the array is shrinking we should possibly wait until | |
5310 | * any io in the removed space completes, but it hardly seems | |
5311 | * worth it. | |
5312 | */ | |
9d8f0363 | 5313 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5314 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5315 | mddev->raid_disks)); | |
b522adcd DW |
5316 | if (mddev->array_sectors > |
5317 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5318 | return -EINVAL; | |
f233ea5c | 5319 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5320 | revalidate_disk(mddev->gendisk); |
58c0fed4 AN |
5321 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
5322 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
5323 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5324 | } | |
58c0fed4 | 5325 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5326 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5327 | return 0; |
5328 | } | |
5329 | ||
01ee22b4 N |
5330 | static int check_stripe_cache(mddev_t *mddev) |
5331 | { | |
5332 | /* Can only proceed if there are plenty of stripe_heads. | |
5333 | * We need a minimum of one full stripe,, and for sensible progress | |
5334 | * it is best to have about 4 times that. | |
5335 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5336 | * allow for chunk sizes up to 256K, which is probably OK. | |
5337 | * If the chunk size is greater, user-space should request more | |
5338 | * stripe_heads first. | |
5339 | */ | |
5340 | raid5_conf_t *conf = mddev->private; | |
5341 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5342 | > conf->max_nr_stripes || | |
5343 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5344 | > conf->max_nr_stripes) { | |
0c55e022 N |
5345 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5346 | mdname(mddev), | |
01ee22b4 N |
5347 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5348 | / STRIPE_SIZE)*4); | |
5349 | return 0; | |
5350 | } | |
5351 | return 1; | |
5352 | } | |
5353 | ||
50ac168a | 5354 | static int check_reshape(mddev_t *mddev) |
29269553 | 5355 | { |
070ec55d | 5356 | raid5_conf_t *conf = mddev->private; |
29269553 | 5357 | |
88ce4930 N |
5358 | if (mddev->delta_disks == 0 && |
5359 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5360 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5361 | return 0; /* nothing to do */ |
dba034ee N |
5362 | if (mddev->bitmap) |
5363 | /* Cannot grow a bitmap yet */ | |
5364 | return -EBUSY; | |
ec32a2bd N |
5365 | if (mddev->degraded > conf->max_degraded) |
5366 | return -EINVAL; | |
5367 | if (mddev->delta_disks < 0) { | |
5368 | /* We might be able to shrink, but the devices must | |
5369 | * be made bigger first. | |
5370 | * For raid6, 4 is the minimum size. | |
5371 | * Otherwise 2 is the minimum | |
5372 | */ | |
5373 | int min = 2; | |
5374 | if (mddev->level == 6) | |
5375 | min = 4; | |
5376 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5377 | return -EINVAL; | |
5378 | } | |
29269553 | 5379 | |
01ee22b4 | 5380 | if (!check_stripe_cache(mddev)) |
29269553 | 5381 | return -ENOSPC; |
29269553 | 5382 | |
ec32a2bd | 5383 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5384 | } |
5385 | ||
5386 | static int raid5_start_reshape(mddev_t *mddev) | |
5387 | { | |
070ec55d | 5388 | raid5_conf_t *conf = mddev->private; |
63c70c4f | 5389 | mdk_rdev_t *rdev; |
63c70c4f N |
5390 | int spares = 0; |
5391 | int added_devices = 0; | |
c04be0aa | 5392 | unsigned long flags; |
63c70c4f | 5393 | |
f416885e | 5394 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5395 | return -EBUSY; |
5396 | ||
01ee22b4 N |
5397 | if (!check_stripe_cache(mddev)) |
5398 | return -ENOSPC; | |
5399 | ||
159ec1fc | 5400 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5401 | if (rdev->raid_disk < 0 && |
5402 | !test_bit(Faulty, &rdev->flags)) | |
5403 | spares++; | |
63c70c4f | 5404 | |
f416885e | 5405 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5406 | /* Not enough devices even to make a degraded array |
5407 | * of that size | |
5408 | */ | |
5409 | return -EINVAL; | |
5410 | ||
ec32a2bd N |
5411 | /* Refuse to reduce size of the array. Any reductions in |
5412 | * array size must be through explicit setting of array_size | |
5413 | * attribute. | |
5414 | */ | |
5415 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5416 | < mddev->array_sectors) { | |
0c55e022 | 5417 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5418 | "before number of disks\n", mdname(mddev)); |
5419 | return -EINVAL; | |
5420 | } | |
5421 | ||
f6705578 | 5422 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5423 | spin_lock_irq(&conf->device_lock); |
5424 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5425 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5426 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5427 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5428 | conf->prev_algo = conf->algorithm; |
5429 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5430 | if (mddev->delta_disks < 0) |
5431 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5432 | else | |
5433 | conf->reshape_progress = 0; | |
5434 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5435 | conf->generation++; |
29269553 N |
5436 | spin_unlock_irq(&conf->device_lock); |
5437 | ||
5438 | /* Add some new drives, as many as will fit. | |
5439 | * We know there are enough to make the newly sized array work. | |
5440 | */ | |
159ec1fc | 5441 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
5442 | if (rdev->raid_disk < 0 && |
5443 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 5444 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 | 5445 | char nm[20]; |
9eb07c25 | 5446 | if (rdev->raid_disk >= conf->previous_raid_disks) { |
7ef90146 | 5447 | set_bit(In_sync, &rdev->flags); |
9eb07c25 N |
5448 | added_devices++; |
5449 | } else | |
7ef90146 | 5450 | rdev->recovery_offset = 0; |
29269553 | 5451 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
5452 | if (sysfs_create_link(&mddev->kobj, |
5453 | &rdev->kobj, nm)) | |
5454 | printk(KERN_WARNING | |
0c55e022 N |
5455 | "md/raid:%s: failed to create " |
5456 | " link %s\n", | |
5457 | mdname(mddev), nm); | |
29269553 N |
5458 | } else |
5459 | break; | |
5460 | } | |
5461 | ||
9eb07c25 N |
5462 | /* When a reshape changes the number of devices, ->degraded |
5463 | * is measured against the large of the pre and post number of | |
5464 | * devices.*/ | |
ec32a2bd N |
5465 | if (mddev->delta_disks > 0) { |
5466 | spin_lock_irqsave(&conf->device_lock, flags); | |
9eb07c25 | 5467 | mddev->degraded += (conf->raid_disks - conf->previous_raid_disks) |
ec32a2bd N |
5468 | - added_devices; |
5469 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5470 | } | |
63c70c4f | 5471 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5472 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5473 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5474 | |
29269553 N |
5475 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5476 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5477 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5478 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5479 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5480 | "reshape"); |
29269553 N |
5481 | if (!mddev->sync_thread) { |
5482 | mddev->recovery = 0; | |
5483 | spin_lock_irq(&conf->device_lock); | |
5484 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5485 | conf->reshape_progress = MaxSector; |
29269553 N |
5486 | spin_unlock_irq(&conf->device_lock); |
5487 | return -EAGAIN; | |
5488 | } | |
c8f517c4 | 5489 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5490 | md_wakeup_thread(mddev->sync_thread); |
5491 | md_new_event(mddev); | |
5492 | return 0; | |
5493 | } | |
29269553 | 5494 | |
ec32a2bd N |
5495 | /* This is called from the reshape thread and should make any |
5496 | * changes needed in 'conf' | |
5497 | */ | |
29269553 N |
5498 | static void end_reshape(raid5_conf_t *conf) |
5499 | { | |
29269553 | 5500 | |
f6705578 | 5501 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5502 | |
f6705578 | 5503 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5504 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5505 | conf->reshape_progress = MaxSector; |
f6705578 | 5506 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5507 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5508 | |
5509 | /* read-ahead size must cover two whole stripes, which is | |
5510 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5511 | */ | |
5512 | { | |
cea9c228 | 5513 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5514 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5515 | / PAGE_SIZE); |
16a53ecc N |
5516 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5517 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5518 | } | |
29269553 | 5519 | } |
29269553 N |
5520 | } |
5521 | ||
ec32a2bd N |
5522 | /* This is called from the raid5d thread with mddev_lock held. |
5523 | * It makes config changes to the device. | |
5524 | */ | |
cea9c228 N |
5525 | static void raid5_finish_reshape(mddev_t *mddev) |
5526 | { | |
070ec55d | 5527 | raid5_conf_t *conf = mddev->private; |
cea9c228 N |
5528 | |
5529 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5530 | ||
ec32a2bd N |
5531 | if (mddev->delta_disks > 0) { |
5532 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5533 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5534 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5535 | } else { |
5536 | int d; | |
ec32a2bd N |
5537 | mddev->degraded = conf->raid_disks; |
5538 | for (d = 0; d < conf->raid_disks ; d++) | |
5539 | if (conf->disks[d].rdev && | |
5540 | test_bit(In_sync, | |
5541 | &conf->disks[d].rdev->flags)) | |
5542 | mddev->degraded--; | |
5543 | for (d = conf->raid_disks ; | |
5544 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 N |
5545 | d++) { |
5546 | mdk_rdev_t *rdev = conf->disks[d].rdev; | |
5547 | if (rdev && raid5_remove_disk(mddev, d) == 0) { | |
5548 | char nm[20]; | |
5549 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5550 | sysfs_remove_link(&mddev->kobj, nm); | |
5551 | rdev->raid_disk = -1; | |
5552 | } | |
5553 | } | |
cea9c228 | 5554 | } |
88ce4930 | 5555 | mddev->layout = conf->algorithm; |
09c9e5fa | 5556 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5557 | mddev->reshape_position = MaxSector; |
5558 | mddev->delta_disks = 0; | |
cea9c228 N |
5559 | } |
5560 | } | |
5561 | ||
72626685 N |
5562 | static void raid5_quiesce(mddev_t *mddev, int state) |
5563 | { | |
070ec55d | 5564 | raid5_conf_t *conf = mddev->private; |
72626685 N |
5565 | |
5566 | switch(state) { | |
e464eafd N |
5567 | case 2: /* resume for a suspend */ |
5568 | wake_up(&conf->wait_for_overlap); | |
5569 | break; | |
5570 | ||
72626685 N |
5571 | case 1: /* stop all writes */ |
5572 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5573 | /* '2' tells resync/reshape to pause so that all |
5574 | * active stripes can drain | |
5575 | */ | |
5576 | conf->quiesce = 2; | |
72626685 | 5577 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5578 | atomic_read(&conf->active_stripes) == 0 && |
5579 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5580 | conf->device_lock, /* nothing */); |
64bd660b | 5581 | conf->quiesce = 1; |
72626685 | 5582 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5583 | /* allow reshape to continue */ |
5584 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5585 | break; |
5586 | ||
5587 | case 0: /* re-enable writes */ | |
5588 | spin_lock_irq(&conf->device_lock); | |
5589 | conf->quiesce = 0; | |
5590 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5591 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5592 | spin_unlock_irq(&conf->device_lock); |
5593 | break; | |
5594 | } | |
72626685 | 5595 | } |
b15c2e57 | 5596 | |
d562b0c4 | 5597 | |
f1b29bca | 5598 | static void *raid45_takeover_raid0(mddev_t *mddev, int level) |
54071b38 | 5599 | { |
f1b29bca | 5600 | struct raid0_private_data *raid0_priv = mddev->private; |
54071b38 | 5601 | |
f1b29bca DW |
5602 | /* for raid0 takeover only one zone is supported */ |
5603 | if (raid0_priv->nr_strip_zones > 1) { | |
0c55e022 N |
5604 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5605 | mdname(mddev)); | |
f1b29bca DW |
5606 | return ERR_PTR(-EINVAL); |
5607 | } | |
5608 | ||
5609 | mddev->new_level = level; | |
54071b38 TM |
5610 | mddev->new_layout = ALGORITHM_PARITY_N; |
5611 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5612 | mddev->raid_disks += 1; | |
5613 | mddev->delta_disks = 1; | |
5614 | /* make sure it will be not marked as dirty */ | |
5615 | mddev->recovery_cp = MaxSector; | |
5616 | ||
5617 | return setup_conf(mddev); | |
5618 | } | |
5619 | ||
5620 | ||
d562b0c4 N |
5621 | static void *raid5_takeover_raid1(mddev_t *mddev) |
5622 | { | |
5623 | int chunksect; | |
5624 | ||
5625 | if (mddev->raid_disks != 2 || | |
5626 | mddev->degraded > 1) | |
5627 | return ERR_PTR(-EINVAL); | |
5628 | ||
5629 | /* Should check if there are write-behind devices? */ | |
5630 | ||
5631 | chunksect = 64*2; /* 64K by default */ | |
5632 | ||
5633 | /* The array must be an exact multiple of chunksize */ | |
5634 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5635 | chunksect >>= 1; | |
5636 | ||
5637 | if ((chunksect<<9) < STRIPE_SIZE) | |
5638 | /* array size does not allow a suitable chunk size */ | |
5639 | return ERR_PTR(-EINVAL); | |
5640 | ||
5641 | mddev->new_level = 5; | |
5642 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5643 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5644 | |
5645 | return setup_conf(mddev); | |
5646 | } | |
5647 | ||
fc9739c6 N |
5648 | static void *raid5_takeover_raid6(mddev_t *mddev) |
5649 | { | |
5650 | int new_layout; | |
5651 | ||
5652 | switch (mddev->layout) { | |
5653 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5654 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5655 | break; | |
5656 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5657 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5658 | break; | |
5659 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5660 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5661 | break; | |
5662 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5663 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5664 | break; | |
5665 | case ALGORITHM_PARITY_0_6: | |
5666 | new_layout = ALGORITHM_PARITY_0; | |
5667 | break; | |
5668 | case ALGORITHM_PARITY_N: | |
5669 | new_layout = ALGORITHM_PARITY_N; | |
5670 | break; | |
5671 | default: | |
5672 | return ERR_PTR(-EINVAL); | |
5673 | } | |
5674 | mddev->new_level = 5; | |
5675 | mddev->new_layout = new_layout; | |
5676 | mddev->delta_disks = -1; | |
5677 | mddev->raid_disks -= 1; | |
5678 | return setup_conf(mddev); | |
5679 | } | |
5680 | ||
d562b0c4 | 5681 | |
50ac168a | 5682 | static int raid5_check_reshape(mddev_t *mddev) |
b3546035 | 5683 | { |
88ce4930 N |
5684 | /* For a 2-drive array, the layout and chunk size can be changed |
5685 | * immediately as not restriping is needed. | |
5686 | * For larger arrays we record the new value - after validation | |
5687 | * to be used by a reshape pass. | |
b3546035 | 5688 | */ |
070ec55d | 5689 | raid5_conf_t *conf = mddev->private; |
597a711b | 5690 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5691 | |
597a711b | 5692 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5693 | return -EINVAL; |
5694 | if (new_chunk > 0) { | |
0ba459d2 | 5695 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5696 | return -EINVAL; |
597a711b | 5697 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5698 | return -EINVAL; |
597a711b | 5699 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5700 | /* not factor of array size */ |
5701 | return -EINVAL; | |
5702 | } | |
5703 | ||
5704 | /* They look valid */ | |
5705 | ||
88ce4930 | 5706 | if (mddev->raid_disks == 2) { |
597a711b N |
5707 | /* can make the change immediately */ |
5708 | if (mddev->new_layout >= 0) { | |
5709 | conf->algorithm = mddev->new_layout; | |
5710 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5711 | } |
5712 | if (new_chunk > 0) { | |
597a711b N |
5713 | conf->chunk_sectors = new_chunk ; |
5714 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5715 | } |
5716 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5717 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5718 | } |
50ac168a | 5719 | return check_reshape(mddev); |
88ce4930 N |
5720 | } |
5721 | ||
50ac168a | 5722 | static int raid6_check_reshape(mddev_t *mddev) |
88ce4930 | 5723 | { |
597a711b | 5724 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5725 | |
597a711b | 5726 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5727 | return -EINVAL; |
b3546035 | 5728 | if (new_chunk > 0) { |
0ba459d2 | 5729 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5730 | return -EINVAL; |
597a711b | 5731 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5732 | return -EINVAL; |
597a711b | 5733 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5734 | /* not factor of array size */ |
5735 | return -EINVAL; | |
b3546035 | 5736 | } |
88ce4930 N |
5737 | |
5738 | /* They look valid */ | |
50ac168a | 5739 | return check_reshape(mddev); |
b3546035 N |
5740 | } |
5741 | ||
d562b0c4 N |
5742 | static void *raid5_takeover(mddev_t *mddev) |
5743 | { | |
5744 | /* raid5 can take over: | |
f1b29bca | 5745 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
5746 | * raid1 - if there are two drives. We need to know the chunk size |
5747 | * raid4 - trivial - just use a raid4 layout. | |
5748 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 5749 | */ |
f1b29bca DW |
5750 | if (mddev->level == 0) |
5751 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
5752 | if (mddev->level == 1) |
5753 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5754 | if (mddev->level == 4) { |
5755 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5756 | mddev->new_level = 5; | |
5757 | return setup_conf(mddev); | |
5758 | } | |
fc9739c6 N |
5759 | if (mddev->level == 6) |
5760 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5761 | |
5762 | return ERR_PTR(-EINVAL); | |
5763 | } | |
5764 | ||
a78d38a1 N |
5765 | static void *raid4_takeover(mddev_t *mddev) |
5766 | { | |
f1b29bca DW |
5767 | /* raid4 can take over: |
5768 | * raid0 - if there is only one strip zone | |
5769 | * raid5 - if layout is right | |
a78d38a1 | 5770 | */ |
f1b29bca DW |
5771 | if (mddev->level == 0) |
5772 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
5773 | if (mddev->level == 5 && |
5774 | mddev->layout == ALGORITHM_PARITY_N) { | |
5775 | mddev->new_layout = 0; | |
5776 | mddev->new_level = 4; | |
5777 | return setup_conf(mddev); | |
5778 | } | |
5779 | return ERR_PTR(-EINVAL); | |
5780 | } | |
d562b0c4 | 5781 | |
245f46c2 N |
5782 | static struct mdk_personality raid5_personality; |
5783 | ||
5784 | static void *raid6_takeover(mddev_t *mddev) | |
5785 | { | |
5786 | /* Currently can only take over a raid5. We map the | |
5787 | * personality to an equivalent raid6 personality | |
5788 | * with the Q block at the end. | |
5789 | */ | |
5790 | int new_layout; | |
5791 | ||
5792 | if (mddev->pers != &raid5_personality) | |
5793 | return ERR_PTR(-EINVAL); | |
5794 | if (mddev->degraded > 1) | |
5795 | return ERR_PTR(-EINVAL); | |
5796 | if (mddev->raid_disks > 253) | |
5797 | return ERR_PTR(-EINVAL); | |
5798 | if (mddev->raid_disks < 3) | |
5799 | return ERR_PTR(-EINVAL); | |
5800 | ||
5801 | switch (mddev->layout) { | |
5802 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5803 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5804 | break; | |
5805 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5806 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5807 | break; | |
5808 | case ALGORITHM_LEFT_SYMMETRIC: | |
5809 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5810 | break; | |
5811 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5812 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5813 | break; | |
5814 | case ALGORITHM_PARITY_0: | |
5815 | new_layout = ALGORITHM_PARITY_0_6; | |
5816 | break; | |
5817 | case ALGORITHM_PARITY_N: | |
5818 | new_layout = ALGORITHM_PARITY_N; | |
5819 | break; | |
5820 | default: | |
5821 | return ERR_PTR(-EINVAL); | |
5822 | } | |
5823 | mddev->new_level = 6; | |
5824 | mddev->new_layout = new_layout; | |
5825 | mddev->delta_disks = 1; | |
5826 | mddev->raid_disks += 1; | |
5827 | return setup_conf(mddev); | |
5828 | } | |
5829 | ||
5830 | ||
16a53ecc N |
5831 | static struct mdk_personality raid6_personality = |
5832 | { | |
5833 | .name = "raid6", | |
5834 | .level = 6, | |
5835 | .owner = THIS_MODULE, | |
5836 | .make_request = make_request, | |
5837 | .run = run, | |
5838 | .stop = stop, | |
5839 | .status = status, | |
5840 | .error_handler = error, | |
5841 | .hot_add_disk = raid5_add_disk, | |
5842 | .hot_remove_disk= raid5_remove_disk, | |
5843 | .spare_active = raid5_spare_active, | |
5844 | .sync_request = sync_request, | |
5845 | .resize = raid5_resize, | |
80c3a6ce | 5846 | .size = raid5_size, |
50ac168a | 5847 | .check_reshape = raid6_check_reshape, |
f416885e | 5848 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5849 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5850 | .quiesce = raid5_quiesce, |
245f46c2 | 5851 | .takeover = raid6_takeover, |
16a53ecc | 5852 | }; |
2604b703 | 5853 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
5854 | { |
5855 | .name = "raid5", | |
2604b703 | 5856 | .level = 5, |
1da177e4 LT |
5857 | .owner = THIS_MODULE, |
5858 | .make_request = make_request, | |
5859 | .run = run, | |
5860 | .stop = stop, | |
5861 | .status = status, | |
5862 | .error_handler = error, | |
5863 | .hot_add_disk = raid5_add_disk, | |
5864 | .hot_remove_disk= raid5_remove_disk, | |
5865 | .spare_active = raid5_spare_active, | |
5866 | .sync_request = sync_request, | |
5867 | .resize = raid5_resize, | |
80c3a6ce | 5868 | .size = raid5_size, |
63c70c4f N |
5869 | .check_reshape = raid5_check_reshape, |
5870 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5871 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5872 | .quiesce = raid5_quiesce, |
d562b0c4 | 5873 | .takeover = raid5_takeover, |
1da177e4 LT |
5874 | }; |
5875 | ||
2604b703 | 5876 | static struct mdk_personality raid4_personality = |
1da177e4 | 5877 | { |
2604b703 N |
5878 | .name = "raid4", |
5879 | .level = 4, | |
5880 | .owner = THIS_MODULE, | |
5881 | .make_request = make_request, | |
5882 | .run = run, | |
5883 | .stop = stop, | |
5884 | .status = status, | |
5885 | .error_handler = error, | |
5886 | .hot_add_disk = raid5_add_disk, | |
5887 | .hot_remove_disk= raid5_remove_disk, | |
5888 | .spare_active = raid5_spare_active, | |
5889 | .sync_request = sync_request, | |
5890 | .resize = raid5_resize, | |
80c3a6ce | 5891 | .size = raid5_size, |
3d37890b N |
5892 | .check_reshape = raid5_check_reshape, |
5893 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5894 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 5895 | .quiesce = raid5_quiesce, |
a78d38a1 | 5896 | .takeover = raid4_takeover, |
2604b703 N |
5897 | }; |
5898 | ||
5899 | static int __init raid5_init(void) | |
5900 | { | |
16a53ecc | 5901 | register_md_personality(&raid6_personality); |
2604b703 N |
5902 | register_md_personality(&raid5_personality); |
5903 | register_md_personality(&raid4_personality); | |
5904 | return 0; | |
1da177e4 LT |
5905 | } |
5906 | ||
2604b703 | 5907 | static void raid5_exit(void) |
1da177e4 | 5908 | { |
16a53ecc | 5909 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5910 | unregister_md_personality(&raid5_personality); |
5911 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5912 | } |
5913 | ||
5914 | module_init(raid5_init); | |
5915 | module_exit(raid5_exit); | |
5916 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5917 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 5918 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
5919 | MODULE_ALIAS("md-raid5"); |
5920 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5921 | MODULE_ALIAS("md-level-5"); |
5922 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5923 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5924 | MODULE_ALIAS("md-raid6"); | |
5925 | MODULE_ALIAS("md-level-6"); | |
5926 | ||
5927 | /* This used to be two separate modules, they were: */ | |
5928 | MODULE_ALIAS("raid5"); | |
5929 | MODULE_ALIAS("raid6"); |