Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
30 | * conf->bm_write is the number of the last batch successfully written. | |
31 | * conf->bm_flush is the number of the last batch that was closed to | |
32 | * new additions. | |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
35 | * the number of the batch it will be in. This is bm_flush+1. | |
36 | * When we are ready to do a write, if that batch hasn't been written yet, | |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
91c00924 | 48 | #include <linux/async_tx.h> |
bff61975 | 49 | #include <linux/seq_file.h> |
43b2e5d8 | 50 | #include "md.h" |
bff61975 | 51 | #include "raid5.h" |
ef740c37 CH |
52 | #include "raid6.h" |
53 | #include "bitmap.h" | |
72626685 | 54 | |
1da177e4 LT |
55 | /* |
56 | * Stripe cache | |
57 | */ | |
58 | ||
59 | #define NR_STRIPES 256 | |
60 | #define STRIPE_SIZE PAGE_SIZE | |
61 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
62 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
63 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 64 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 65 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
66 | #define HASH_MASK (NR_HASH - 1) |
67 | ||
fccddba0 | 68 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
69 | |
70 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
71 | * order without overlap. There may be several bio's per stripe+device, and | |
72 | * a bio could span several devices. | |
73 | * When walking this list for a particular stripe+device, we must never proceed | |
74 | * beyond a bio that extends past this device, as the next bio might no longer | |
75 | * be valid. | |
76 | * This macro is used to determine the 'next' bio in the list, given the sector | |
77 | * of the current stripe+device | |
78 | */ | |
79 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
80 | /* | |
81 | * The following can be used to debug the driver | |
82 | */ | |
1da177e4 LT |
83 | #define RAID5_PARANOIA 1 |
84 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
85 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
86 | #else | |
87 | # define CHECK_DEVLOCK() | |
88 | #endif | |
89 | ||
45b4233c | 90 | #ifdef DEBUG |
1da177e4 LT |
91 | #define inline |
92 | #define __inline__ | |
93 | #endif | |
94 | ||
6be9d494 BS |
95 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
96 | ||
16a53ecc N |
97 | #if !RAID6_USE_EMPTY_ZERO_PAGE |
98 | /* In .bss so it's zeroed */ | |
99 | const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256))); | |
100 | #endif | |
101 | ||
960e739d | 102 | /* |
5b99c2ff JA |
103 | * We maintain a biased count of active stripes in the bottom 16 bits of |
104 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
105 | */ |
106 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
107 | { | |
5b99c2ff | 108 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
109 | } |
110 | ||
111 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
112 | { | |
5b99c2ff | 113 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
114 | } |
115 | ||
116 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
117 | { | |
118 | --bio->bi_phys_segments; | |
119 | return raid5_bi_phys_segments(bio); | |
120 | } | |
121 | ||
122 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
123 | { | |
124 | unsigned short val = raid5_bi_hw_segments(bio); | |
125 | ||
126 | --val; | |
5b99c2ff | 127 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
128 | return val; |
129 | } | |
130 | ||
131 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
132 | { | |
5b99c2ff | 133 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16); |
960e739d JA |
134 | } |
135 | ||
16a53ecc N |
136 | static inline int raid6_next_disk(int disk, int raid_disks) |
137 | { | |
138 | disk++; | |
139 | return (disk < raid_disks) ? disk : 0; | |
140 | } | |
a4456856 DW |
141 | |
142 | static void return_io(struct bio *return_bi) | |
143 | { | |
144 | struct bio *bi = return_bi; | |
145 | while (bi) { | |
a4456856 DW |
146 | |
147 | return_bi = bi->bi_next; | |
148 | bi->bi_next = NULL; | |
149 | bi->bi_size = 0; | |
0e13fe23 | 150 | bio_endio(bi, 0); |
a4456856 DW |
151 | bi = return_bi; |
152 | } | |
153 | } | |
154 | ||
1da177e4 LT |
155 | static void print_raid5_conf (raid5_conf_t *conf); |
156 | ||
600aa109 DW |
157 | static int stripe_operations_active(struct stripe_head *sh) |
158 | { | |
159 | return sh->check_state || sh->reconstruct_state || | |
160 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
161 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
162 | } | |
163 | ||
858119e1 | 164 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
165 | { |
166 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
167 | BUG_ON(!list_empty(&sh->lru)); |
168 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 169 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 170 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 171 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
172 | blk_plug_device(conf->mddev->queue); |
173 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 174 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 175 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
176 | blk_plug_device(conf->mddev->queue); |
177 | } else { | |
72626685 | 178 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 179 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 180 | } |
1da177e4 LT |
181 | md_wakeup_thread(conf->mddev->thread); |
182 | } else { | |
600aa109 | 183 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
184 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
185 | atomic_dec(&conf->preread_active_stripes); | |
186 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
187 | md_wakeup_thread(conf->mddev->thread); | |
188 | } | |
1da177e4 | 189 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
190 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
191 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 192 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
193 | if (conf->retry_read_aligned) |
194 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 195 | } |
1da177e4 LT |
196 | } |
197 | } | |
198 | } | |
199 | static void release_stripe(struct stripe_head *sh) | |
200 | { | |
201 | raid5_conf_t *conf = sh->raid_conf; | |
202 | unsigned long flags; | |
16a53ecc | 203 | |
1da177e4 LT |
204 | spin_lock_irqsave(&conf->device_lock, flags); |
205 | __release_stripe(conf, sh); | |
206 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
207 | } | |
208 | ||
fccddba0 | 209 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 210 | { |
45b4233c DW |
211 | pr_debug("remove_hash(), stripe %llu\n", |
212 | (unsigned long long)sh->sector); | |
1da177e4 | 213 | |
fccddba0 | 214 | hlist_del_init(&sh->hash); |
1da177e4 LT |
215 | } |
216 | ||
16a53ecc | 217 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 218 | { |
fccddba0 | 219 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 220 | |
45b4233c DW |
221 | pr_debug("insert_hash(), stripe %llu\n", |
222 | (unsigned long long)sh->sector); | |
1da177e4 LT |
223 | |
224 | CHECK_DEVLOCK(); | |
fccddba0 | 225 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
226 | } |
227 | ||
228 | ||
229 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
230 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
231 | { | |
232 | struct stripe_head *sh = NULL; | |
233 | struct list_head *first; | |
234 | ||
235 | CHECK_DEVLOCK(); | |
236 | if (list_empty(&conf->inactive_list)) | |
237 | goto out; | |
238 | first = conf->inactive_list.next; | |
239 | sh = list_entry(first, struct stripe_head, lru); | |
240 | list_del_init(first); | |
241 | remove_hash(sh); | |
242 | atomic_inc(&conf->active_stripes); | |
243 | out: | |
244 | return sh; | |
245 | } | |
246 | ||
247 | static void shrink_buffers(struct stripe_head *sh, int num) | |
248 | { | |
249 | struct page *p; | |
250 | int i; | |
251 | ||
252 | for (i=0; i<num ; i++) { | |
253 | p = sh->dev[i].page; | |
254 | if (!p) | |
255 | continue; | |
256 | sh->dev[i].page = NULL; | |
2d1f3b5d | 257 | put_page(p); |
1da177e4 LT |
258 | } |
259 | } | |
260 | ||
261 | static int grow_buffers(struct stripe_head *sh, int num) | |
262 | { | |
263 | int i; | |
264 | ||
265 | for (i=0; i<num; i++) { | |
266 | struct page *page; | |
267 | ||
268 | if (!(page = alloc_page(GFP_KERNEL))) { | |
269 | return 1; | |
270 | } | |
271 | sh->dev[i].page = page; | |
272 | } | |
273 | return 0; | |
274 | } | |
275 | ||
d710e138 | 276 | static void raid5_build_block(struct stripe_head *sh, int i); |
112bf897 | 277 | static int stripe_to_pdidx(sector_t stripe, raid5_conf_t *conf, int previous); |
1da177e4 | 278 | |
b5663ba4 | 279 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
280 | { |
281 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 282 | int i; |
1da177e4 | 283 | |
78bafebd ES |
284 | BUG_ON(atomic_read(&sh->count) != 0); |
285 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 286 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 287 | |
1da177e4 | 288 | CHECK_DEVLOCK(); |
45b4233c | 289 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
290 | (unsigned long long)sh->sector); |
291 | ||
292 | remove_hash(sh); | |
16a53ecc | 293 | |
b5663ba4 | 294 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 295 | sh->sector = sector; |
112bf897 | 296 | sh->pd_idx = stripe_to_pdidx(sector, conf, previous); |
1da177e4 LT |
297 | sh->state = 0; |
298 | ||
7ecaa1e6 N |
299 | |
300 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
301 | struct r5dev *dev = &sh->dev[i]; |
302 | ||
d84e0f10 | 303 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 304 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 305 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 306 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 307 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
308 | test_bit(R5_LOCKED, &dev->flags)); |
309 | BUG(); | |
310 | } | |
311 | dev->flags = 0; | |
312 | raid5_build_block(sh, i); | |
313 | } | |
314 | insert_hash(conf, sh); | |
315 | } | |
316 | ||
7ecaa1e6 | 317 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, int disks) |
1da177e4 LT |
318 | { |
319 | struct stripe_head *sh; | |
fccddba0 | 320 | struct hlist_node *hn; |
1da177e4 LT |
321 | |
322 | CHECK_DEVLOCK(); | |
45b4233c | 323 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 324 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
7ecaa1e6 | 325 | if (sh->sector == sector && sh->disks == disks) |
1da177e4 | 326 | return sh; |
45b4233c | 327 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
328 | return NULL; |
329 | } | |
330 | ||
331 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 332 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 333 | |
b5663ba4 N |
334 | static struct stripe_head * |
335 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
336 | int previous, int noblock) | |
1da177e4 LT |
337 | { |
338 | struct stripe_head *sh; | |
b5663ba4 | 339 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 340 | |
45b4233c | 341 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
342 | |
343 | spin_lock_irq(&conf->device_lock); | |
344 | ||
345 | do { | |
72626685 N |
346 | wait_event_lock_irq(conf->wait_for_stripe, |
347 | conf->quiesce == 0, | |
348 | conf->device_lock, /* nothing */); | |
7ecaa1e6 | 349 | sh = __find_stripe(conf, sector, disks); |
1da177e4 LT |
350 | if (!sh) { |
351 | if (!conf->inactive_blocked) | |
352 | sh = get_free_stripe(conf); | |
353 | if (noblock && sh == NULL) | |
354 | break; | |
355 | if (!sh) { | |
356 | conf->inactive_blocked = 1; | |
357 | wait_event_lock_irq(conf->wait_for_stripe, | |
358 | !list_empty(&conf->inactive_list) && | |
5036805b N |
359 | (atomic_read(&conf->active_stripes) |
360 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
361 | || !conf->inactive_blocked), |
362 | conf->device_lock, | |
f4370781 | 363 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
364 | ); |
365 | conf->inactive_blocked = 0; | |
366 | } else | |
b5663ba4 | 367 | init_stripe(sh, sector, previous); |
1da177e4 LT |
368 | } else { |
369 | if (atomic_read(&sh->count)) { | |
78bafebd | 370 | BUG_ON(!list_empty(&sh->lru)); |
1da177e4 LT |
371 | } else { |
372 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
373 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
374 | if (list_empty(&sh->lru) && |
375 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
376 | BUG(); |
377 | list_del_init(&sh->lru); | |
1da177e4 LT |
378 | } |
379 | } | |
380 | } while (sh == NULL); | |
381 | ||
382 | if (sh) | |
383 | atomic_inc(&sh->count); | |
384 | ||
385 | spin_unlock_irq(&conf->device_lock); | |
386 | return sh; | |
387 | } | |
388 | ||
6712ecf8 N |
389 | static void |
390 | raid5_end_read_request(struct bio *bi, int error); | |
391 | static void | |
392 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 393 | |
c4e5ac0a | 394 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
395 | { |
396 | raid5_conf_t *conf = sh->raid_conf; | |
397 | int i, disks = sh->disks; | |
398 | ||
399 | might_sleep(); | |
400 | ||
401 | for (i = disks; i--; ) { | |
402 | int rw; | |
403 | struct bio *bi; | |
404 | mdk_rdev_t *rdev; | |
405 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
406 | rw = WRITE; | |
407 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
408 | rw = READ; | |
409 | else | |
410 | continue; | |
411 | ||
412 | bi = &sh->dev[i].req; | |
413 | ||
414 | bi->bi_rw = rw; | |
415 | if (rw == WRITE) | |
416 | bi->bi_end_io = raid5_end_write_request; | |
417 | else | |
418 | bi->bi_end_io = raid5_end_read_request; | |
419 | ||
420 | rcu_read_lock(); | |
421 | rdev = rcu_dereference(conf->disks[i].rdev); | |
422 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
423 | rdev = NULL; | |
424 | if (rdev) | |
425 | atomic_inc(&rdev->nr_pending); | |
426 | rcu_read_unlock(); | |
427 | ||
428 | if (rdev) { | |
c4e5ac0a | 429 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
430 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
431 | ||
2b7497f0 DW |
432 | set_bit(STRIPE_IO_STARTED, &sh->state); |
433 | ||
91c00924 DW |
434 | bi->bi_bdev = rdev->bdev; |
435 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 436 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
437 | bi->bi_rw, i); |
438 | atomic_inc(&sh->count); | |
439 | bi->bi_sector = sh->sector + rdev->data_offset; | |
440 | bi->bi_flags = 1 << BIO_UPTODATE; | |
441 | bi->bi_vcnt = 1; | |
442 | bi->bi_max_vecs = 1; | |
443 | bi->bi_idx = 0; | |
444 | bi->bi_io_vec = &sh->dev[i].vec; | |
445 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
446 | bi->bi_io_vec[0].bv_offset = 0; | |
447 | bi->bi_size = STRIPE_SIZE; | |
448 | bi->bi_next = NULL; | |
449 | if (rw == WRITE && | |
450 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
451 | atomic_add(STRIPE_SECTORS, | |
452 | &rdev->corrected_errors); | |
453 | generic_make_request(bi); | |
454 | } else { | |
455 | if (rw == WRITE) | |
456 | set_bit(STRIPE_DEGRADED, &sh->state); | |
457 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
458 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
459 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
460 | set_bit(STRIPE_HANDLE, &sh->state); | |
461 | } | |
462 | } | |
463 | } | |
464 | ||
465 | static struct dma_async_tx_descriptor * | |
466 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
467 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
468 | { | |
469 | struct bio_vec *bvl; | |
470 | struct page *bio_page; | |
471 | int i; | |
472 | int page_offset; | |
473 | ||
474 | if (bio->bi_sector >= sector) | |
475 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
476 | else | |
477 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
478 | bio_for_each_segment(bvl, bio, i) { | |
479 | int len = bio_iovec_idx(bio, i)->bv_len; | |
480 | int clen; | |
481 | int b_offset = 0; | |
482 | ||
483 | if (page_offset < 0) { | |
484 | b_offset = -page_offset; | |
485 | page_offset += b_offset; | |
486 | len -= b_offset; | |
487 | } | |
488 | ||
489 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
490 | clen = STRIPE_SIZE - page_offset; | |
491 | else | |
492 | clen = len; | |
493 | ||
494 | if (clen > 0) { | |
495 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
496 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
497 | if (frombio) | |
498 | tx = async_memcpy(page, bio_page, page_offset, | |
499 | b_offset, clen, | |
eb0645a8 | 500 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
501 | tx, NULL, NULL); |
502 | else | |
503 | tx = async_memcpy(bio_page, page, b_offset, | |
504 | page_offset, clen, | |
eb0645a8 | 505 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
506 | tx, NULL, NULL); |
507 | } | |
508 | if (clen < len) /* hit end of page */ | |
509 | break; | |
510 | page_offset += len; | |
511 | } | |
512 | ||
513 | return tx; | |
514 | } | |
515 | ||
516 | static void ops_complete_biofill(void *stripe_head_ref) | |
517 | { | |
518 | struct stripe_head *sh = stripe_head_ref; | |
519 | struct bio *return_bi = NULL; | |
520 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 521 | int i; |
91c00924 | 522 | |
e46b272b | 523 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
524 | (unsigned long long)sh->sector); |
525 | ||
526 | /* clear completed biofills */ | |
83de75cc | 527 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
528 | for (i = sh->disks; i--; ) { |
529 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
530 | |
531 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
532 | /* and check if we need to reply to a read request, |
533 | * new R5_Wantfill requests are held off until | |
83de75cc | 534 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
535 | */ |
536 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 537 | struct bio *rbi, *rbi2; |
91c00924 | 538 | |
91c00924 DW |
539 | BUG_ON(!dev->read); |
540 | rbi = dev->read; | |
541 | dev->read = NULL; | |
542 | while (rbi && rbi->bi_sector < | |
543 | dev->sector + STRIPE_SECTORS) { | |
544 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 545 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
546 | rbi->bi_next = return_bi; |
547 | return_bi = rbi; | |
548 | } | |
91c00924 DW |
549 | rbi = rbi2; |
550 | } | |
551 | } | |
552 | } | |
83de75cc DW |
553 | spin_unlock_irq(&conf->device_lock); |
554 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
555 | |
556 | return_io(return_bi); | |
557 | ||
e4d84909 | 558 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
559 | release_stripe(sh); |
560 | } | |
561 | ||
562 | static void ops_run_biofill(struct stripe_head *sh) | |
563 | { | |
564 | struct dma_async_tx_descriptor *tx = NULL; | |
565 | raid5_conf_t *conf = sh->raid_conf; | |
566 | int i; | |
567 | ||
e46b272b | 568 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
569 | (unsigned long long)sh->sector); |
570 | ||
571 | for (i = sh->disks; i--; ) { | |
572 | struct r5dev *dev = &sh->dev[i]; | |
573 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
574 | struct bio *rbi; | |
575 | spin_lock_irq(&conf->device_lock); | |
576 | dev->read = rbi = dev->toread; | |
577 | dev->toread = NULL; | |
578 | spin_unlock_irq(&conf->device_lock); | |
579 | while (rbi && rbi->bi_sector < | |
580 | dev->sector + STRIPE_SECTORS) { | |
581 | tx = async_copy_data(0, rbi, dev->page, | |
582 | dev->sector, tx); | |
583 | rbi = r5_next_bio(rbi, dev->sector); | |
584 | } | |
585 | } | |
586 | } | |
587 | ||
588 | atomic_inc(&sh->count); | |
589 | async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
590 | ops_complete_biofill, sh); | |
591 | } | |
592 | ||
593 | static void ops_complete_compute5(void *stripe_head_ref) | |
594 | { | |
595 | struct stripe_head *sh = stripe_head_ref; | |
596 | int target = sh->ops.target; | |
597 | struct r5dev *tgt = &sh->dev[target]; | |
598 | ||
e46b272b | 599 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
600 | (unsigned long long)sh->sector); |
601 | ||
602 | set_bit(R5_UPTODATE, &tgt->flags); | |
603 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
604 | clear_bit(R5_Wantcompute, &tgt->flags); | |
ecc65c9b DW |
605 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
606 | if (sh->check_state == check_state_compute_run) | |
607 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
608 | set_bit(STRIPE_HANDLE, &sh->state); |
609 | release_stripe(sh); | |
610 | } | |
611 | ||
7b3a871e | 612 | static struct dma_async_tx_descriptor *ops_run_compute5(struct stripe_head *sh) |
91c00924 DW |
613 | { |
614 | /* kernel stack size limits the total number of disks */ | |
615 | int disks = sh->disks; | |
616 | struct page *xor_srcs[disks]; | |
617 | int target = sh->ops.target; | |
618 | struct r5dev *tgt = &sh->dev[target]; | |
619 | struct page *xor_dest = tgt->page; | |
620 | int count = 0; | |
621 | struct dma_async_tx_descriptor *tx; | |
622 | int i; | |
623 | ||
624 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 625 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
626 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
627 | ||
628 | for (i = disks; i--; ) | |
629 | if (i != target) | |
630 | xor_srcs[count++] = sh->dev[i].page; | |
631 | ||
632 | atomic_inc(&sh->count); | |
633 | ||
634 | if (unlikely(count == 1)) | |
635 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
636 | 0, NULL, ops_complete_compute5, sh); | |
637 | else | |
638 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
639 | ASYNC_TX_XOR_ZERO_DST, NULL, | |
640 | ops_complete_compute5, sh); | |
641 | ||
91c00924 DW |
642 | return tx; |
643 | } | |
644 | ||
645 | static void ops_complete_prexor(void *stripe_head_ref) | |
646 | { | |
647 | struct stripe_head *sh = stripe_head_ref; | |
648 | ||
e46b272b | 649 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 650 | (unsigned long long)sh->sector); |
91c00924 DW |
651 | } |
652 | ||
653 | static struct dma_async_tx_descriptor * | |
654 | ops_run_prexor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) | |
655 | { | |
656 | /* kernel stack size limits the total number of disks */ | |
657 | int disks = sh->disks; | |
658 | struct page *xor_srcs[disks]; | |
659 | int count = 0, pd_idx = sh->pd_idx, i; | |
660 | ||
661 | /* existing parity data subtracted */ | |
662 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
663 | ||
e46b272b | 664 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
665 | (unsigned long long)sh->sector); |
666 | ||
667 | for (i = disks; i--; ) { | |
668 | struct r5dev *dev = &sh->dev[i]; | |
669 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 670 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
671 | xor_srcs[count++] = dev->page; |
672 | } | |
673 | ||
674 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
675 | ASYNC_TX_DEP_ACK | ASYNC_TX_XOR_DROP_DST, tx, | |
676 | ops_complete_prexor, sh); | |
677 | ||
678 | return tx; | |
679 | } | |
680 | ||
681 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 682 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
683 | { |
684 | int disks = sh->disks; | |
d8ee0728 | 685 | int i; |
91c00924 | 686 | |
e46b272b | 687 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
688 | (unsigned long long)sh->sector); |
689 | ||
690 | for (i = disks; i--; ) { | |
691 | struct r5dev *dev = &sh->dev[i]; | |
692 | struct bio *chosen; | |
91c00924 | 693 | |
d8ee0728 | 694 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
695 | struct bio *wbi; |
696 | ||
697 | spin_lock(&sh->lock); | |
698 | chosen = dev->towrite; | |
699 | dev->towrite = NULL; | |
700 | BUG_ON(dev->written); | |
701 | wbi = dev->written = chosen; | |
702 | spin_unlock(&sh->lock); | |
703 | ||
704 | while (wbi && wbi->bi_sector < | |
705 | dev->sector + STRIPE_SECTORS) { | |
706 | tx = async_copy_data(1, wbi, dev->page, | |
707 | dev->sector, tx); | |
708 | wbi = r5_next_bio(wbi, dev->sector); | |
709 | } | |
710 | } | |
711 | } | |
712 | ||
713 | return tx; | |
714 | } | |
715 | ||
716 | static void ops_complete_postxor(void *stripe_head_ref) | |
91c00924 DW |
717 | { |
718 | struct stripe_head *sh = stripe_head_ref; | |
719 | int disks = sh->disks, i, pd_idx = sh->pd_idx; | |
720 | ||
e46b272b | 721 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
722 | (unsigned long long)sh->sector); |
723 | ||
724 | for (i = disks; i--; ) { | |
725 | struct r5dev *dev = &sh->dev[i]; | |
726 | if (dev->written || i == pd_idx) | |
727 | set_bit(R5_UPTODATE, &dev->flags); | |
728 | } | |
729 | ||
d8ee0728 DW |
730 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
731 | sh->reconstruct_state = reconstruct_state_drain_result; | |
732 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
733 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
734 | else { | |
735 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
736 | sh->reconstruct_state = reconstruct_state_result; | |
737 | } | |
91c00924 DW |
738 | |
739 | set_bit(STRIPE_HANDLE, &sh->state); | |
740 | release_stripe(sh); | |
741 | } | |
742 | ||
743 | static void | |
d8ee0728 | 744 | ops_run_postxor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
745 | { |
746 | /* kernel stack size limits the total number of disks */ | |
747 | int disks = sh->disks; | |
748 | struct page *xor_srcs[disks]; | |
749 | ||
750 | int count = 0, pd_idx = sh->pd_idx, i; | |
751 | struct page *xor_dest; | |
d8ee0728 | 752 | int prexor = 0; |
91c00924 | 753 | unsigned long flags; |
91c00924 | 754 | |
e46b272b | 755 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
756 | (unsigned long long)sh->sector); |
757 | ||
758 | /* check if prexor is active which means only process blocks | |
759 | * that are part of a read-modify-write (written) | |
760 | */ | |
d8ee0728 DW |
761 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
762 | prexor = 1; | |
91c00924 DW |
763 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
764 | for (i = disks; i--; ) { | |
765 | struct r5dev *dev = &sh->dev[i]; | |
766 | if (dev->written) | |
767 | xor_srcs[count++] = dev->page; | |
768 | } | |
769 | } else { | |
770 | xor_dest = sh->dev[pd_idx].page; | |
771 | for (i = disks; i--; ) { | |
772 | struct r5dev *dev = &sh->dev[i]; | |
773 | if (i != pd_idx) | |
774 | xor_srcs[count++] = dev->page; | |
775 | } | |
776 | } | |
777 | ||
91c00924 DW |
778 | /* 1/ if we prexor'd then the dest is reused as a source |
779 | * 2/ if we did not prexor then we are redoing the parity | |
780 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
781 | * for the synchronous xor case | |
782 | */ | |
783 | flags = ASYNC_TX_DEP_ACK | ASYNC_TX_ACK | | |
784 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
785 | ||
786 | atomic_inc(&sh->count); | |
787 | ||
788 | if (unlikely(count == 1)) { | |
789 | flags &= ~(ASYNC_TX_XOR_DROP_DST | ASYNC_TX_XOR_ZERO_DST); | |
790 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
d8ee0728 | 791 | flags, tx, ops_complete_postxor, sh); |
91c00924 DW |
792 | } else |
793 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
d8ee0728 | 794 | flags, tx, ops_complete_postxor, sh); |
91c00924 DW |
795 | } |
796 | ||
797 | static void ops_complete_check(void *stripe_head_ref) | |
798 | { | |
799 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 800 | |
e46b272b | 801 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
802 | (unsigned long long)sh->sector); |
803 | ||
ecc65c9b | 804 | sh->check_state = check_state_check_result; |
91c00924 DW |
805 | set_bit(STRIPE_HANDLE, &sh->state); |
806 | release_stripe(sh); | |
807 | } | |
808 | ||
809 | static void ops_run_check(struct stripe_head *sh) | |
810 | { | |
811 | /* kernel stack size limits the total number of disks */ | |
812 | int disks = sh->disks; | |
813 | struct page *xor_srcs[disks]; | |
814 | struct dma_async_tx_descriptor *tx; | |
815 | ||
816 | int count = 0, pd_idx = sh->pd_idx, i; | |
817 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
818 | ||
e46b272b | 819 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
820 | (unsigned long long)sh->sector); |
821 | ||
822 | for (i = disks; i--; ) { | |
823 | struct r5dev *dev = &sh->dev[i]; | |
824 | if (i != pd_idx) | |
825 | xor_srcs[count++] = dev->page; | |
826 | } | |
827 | ||
828 | tx = async_xor_zero_sum(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
829 | &sh->ops.zero_sum_result, 0, NULL, NULL, NULL); | |
830 | ||
91c00924 DW |
831 | atomic_inc(&sh->count); |
832 | tx = async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
833 | ops_complete_check, sh); | |
834 | } | |
835 | ||
600aa109 | 836 | static void raid5_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
837 | { |
838 | int overlap_clear = 0, i, disks = sh->disks; | |
839 | struct dma_async_tx_descriptor *tx = NULL; | |
840 | ||
83de75cc | 841 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
842 | ops_run_biofill(sh); |
843 | overlap_clear++; | |
844 | } | |
845 | ||
7b3a871e DW |
846 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
847 | tx = ops_run_compute5(sh); | |
848 | /* terminate the chain if postxor is not set to be run */ | |
849 | if (tx && !test_bit(STRIPE_OP_POSTXOR, &ops_request)) | |
850 | async_tx_ack(tx); | |
851 | } | |
91c00924 | 852 | |
600aa109 | 853 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
91c00924 DW |
854 | tx = ops_run_prexor(sh, tx); |
855 | ||
600aa109 | 856 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 857 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
858 | overlap_clear++; |
859 | } | |
860 | ||
600aa109 | 861 | if (test_bit(STRIPE_OP_POSTXOR, &ops_request)) |
d8ee0728 | 862 | ops_run_postxor(sh, tx); |
91c00924 | 863 | |
ecc65c9b | 864 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) |
91c00924 DW |
865 | ops_run_check(sh); |
866 | ||
91c00924 DW |
867 | if (overlap_clear) |
868 | for (i = disks; i--; ) { | |
869 | struct r5dev *dev = &sh->dev[i]; | |
870 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
871 | wake_up(&sh->raid_conf->wait_for_overlap); | |
872 | } | |
873 | } | |
874 | ||
3f294f4f | 875 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
876 | { |
877 | struct stripe_head *sh; | |
3f294f4f N |
878 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
879 | if (!sh) | |
880 | return 0; | |
881 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
882 | sh->raid_conf = conf; | |
883 | spin_lock_init(&sh->lock); | |
884 | ||
885 | if (grow_buffers(sh, conf->raid_disks)) { | |
886 | shrink_buffers(sh, conf->raid_disks); | |
887 | kmem_cache_free(conf->slab_cache, sh); | |
888 | return 0; | |
889 | } | |
7ecaa1e6 | 890 | sh->disks = conf->raid_disks; |
3f294f4f N |
891 | /* we just created an active stripe so... */ |
892 | atomic_set(&sh->count, 1); | |
893 | atomic_inc(&conf->active_stripes); | |
894 | INIT_LIST_HEAD(&sh->lru); | |
895 | release_stripe(sh); | |
896 | return 1; | |
897 | } | |
898 | ||
899 | static int grow_stripes(raid5_conf_t *conf, int num) | |
900 | { | |
e18b890b | 901 | struct kmem_cache *sc; |
1da177e4 LT |
902 | int devs = conf->raid_disks; |
903 | ||
42b9bebe N |
904 | sprintf(conf->cache_name[0], "raid5-%s", mdname(conf->mddev)); |
905 | sprintf(conf->cache_name[1], "raid5-%s-alt", mdname(conf->mddev)); | |
ad01c9e3 N |
906 | conf->active_name = 0; |
907 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 908 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 909 | 0, 0, NULL); |
1da177e4 LT |
910 | if (!sc) |
911 | return 1; | |
912 | conf->slab_cache = sc; | |
ad01c9e3 | 913 | conf->pool_size = devs; |
16a53ecc | 914 | while (num--) |
3f294f4f | 915 | if (!grow_one_stripe(conf)) |
1da177e4 | 916 | return 1; |
1da177e4 LT |
917 | return 0; |
918 | } | |
29269553 N |
919 | |
920 | #ifdef CONFIG_MD_RAID5_RESHAPE | |
ad01c9e3 N |
921 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
922 | { | |
923 | /* Make all the stripes able to hold 'newsize' devices. | |
924 | * New slots in each stripe get 'page' set to a new page. | |
925 | * | |
926 | * This happens in stages: | |
927 | * 1/ create a new kmem_cache and allocate the required number of | |
928 | * stripe_heads. | |
929 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
930 | * to the new stripe_heads. This will have the side effect of | |
931 | * freezing the array as once all stripe_heads have been collected, | |
932 | * no IO will be possible. Old stripe heads are freed once their | |
933 | * pages have been transferred over, and the old kmem_cache is | |
934 | * freed when all stripes are done. | |
935 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
936 | * we simple return a failre status - no need to clean anything up. | |
937 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
938 | * If this fails, we don't bother trying the shrink the | |
939 | * stripe_heads down again, we just leave them as they are. | |
940 | * As each stripe_head is processed the new one is released into | |
941 | * active service. | |
942 | * | |
943 | * Once step2 is started, we cannot afford to wait for a write, | |
944 | * so we use GFP_NOIO allocations. | |
945 | */ | |
946 | struct stripe_head *osh, *nsh; | |
947 | LIST_HEAD(newstripes); | |
948 | struct disk_info *ndisks; | |
b5470dc5 | 949 | int err; |
e18b890b | 950 | struct kmem_cache *sc; |
ad01c9e3 N |
951 | int i; |
952 | ||
953 | if (newsize <= conf->pool_size) | |
954 | return 0; /* never bother to shrink */ | |
955 | ||
b5470dc5 DW |
956 | err = md_allow_write(conf->mddev); |
957 | if (err) | |
958 | return err; | |
2a2275d6 | 959 | |
ad01c9e3 N |
960 | /* Step 1 */ |
961 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
962 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 963 | 0, 0, NULL); |
ad01c9e3 N |
964 | if (!sc) |
965 | return -ENOMEM; | |
966 | ||
967 | for (i = conf->max_nr_stripes; i; i--) { | |
968 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
969 | if (!nsh) | |
970 | break; | |
971 | ||
972 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
973 | ||
974 | nsh->raid_conf = conf; | |
975 | spin_lock_init(&nsh->lock); | |
976 | ||
977 | list_add(&nsh->lru, &newstripes); | |
978 | } | |
979 | if (i) { | |
980 | /* didn't get enough, give up */ | |
981 | while (!list_empty(&newstripes)) { | |
982 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
983 | list_del(&nsh->lru); | |
984 | kmem_cache_free(sc, nsh); | |
985 | } | |
986 | kmem_cache_destroy(sc); | |
987 | return -ENOMEM; | |
988 | } | |
989 | /* Step 2 - Must use GFP_NOIO now. | |
990 | * OK, we have enough stripes, start collecting inactive | |
991 | * stripes and copying them over | |
992 | */ | |
993 | list_for_each_entry(nsh, &newstripes, lru) { | |
994 | spin_lock_irq(&conf->device_lock); | |
995 | wait_event_lock_irq(conf->wait_for_stripe, | |
996 | !list_empty(&conf->inactive_list), | |
997 | conf->device_lock, | |
b3b46be3 | 998 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
999 | ); |
1000 | osh = get_free_stripe(conf); | |
1001 | spin_unlock_irq(&conf->device_lock); | |
1002 | atomic_set(&nsh->count, 1); | |
1003 | for(i=0; i<conf->pool_size; i++) | |
1004 | nsh->dev[i].page = osh->dev[i].page; | |
1005 | for( ; i<newsize; i++) | |
1006 | nsh->dev[i].page = NULL; | |
1007 | kmem_cache_free(conf->slab_cache, osh); | |
1008 | } | |
1009 | kmem_cache_destroy(conf->slab_cache); | |
1010 | ||
1011 | /* Step 3. | |
1012 | * At this point, we are holding all the stripes so the array | |
1013 | * is completely stalled, so now is a good time to resize | |
1014 | * conf->disks. | |
1015 | */ | |
1016 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1017 | if (ndisks) { | |
1018 | for (i=0; i<conf->raid_disks; i++) | |
1019 | ndisks[i] = conf->disks[i]; | |
1020 | kfree(conf->disks); | |
1021 | conf->disks = ndisks; | |
1022 | } else | |
1023 | err = -ENOMEM; | |
1024 | ||
1025 | /* Step 4, return new stripes to service */ | |
1026 | while(!list_empty(&newstripes)) { | |
1027 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1028 | list_del_init(&nsh->lru); | |
1029 | for (i=conf->raid_disks; i < newsize; i++) | |
1030 | if (nsh->dev[i].page == NULL) { | |
1031 | struct page *p = alloc_page(GFP_NOIO); | |
1032 | nsh->dev[i].page = p; | |
1033 | if (!p) | |
1034 | err = -ENOMEM; | |
1035 | } | |
1036 | release_stripe(nsh); | |
1037 | } | |
1038 | /* critical section pass, GFP_NOIO no longer needed */ | |
1039 | ||
1040 | conf->slab_cache = sc; | |
1041 | conf->active_name = 1-conf->active_name; | |
1042 | conf->pool_size = newsize; | |
1043 | return err; | |
1044 | } | |
29269553 | 1045 | #endif |
1da177e4 | 1046 | |
3f294f4f | 1047 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1048 | { |
1049 | struct stripe_head *sh; | |
1050 | ||
3f294f4f N |
1051 | spin_lock_irq(&conf->device_lock); |
1052 | sh = get_free_stripe(conf); | |
1053 | spin_unlock_irq(&conf->device_lock); | |
1054 | if (!sh) | |
1055 | return 0; | |
78bafebd | 1056 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1057 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1058 | kmem_cache_free(conf->slab_cache, sh); |
1059 | atomic_dec(&conf->active_stripes); | |
1060 | return 1; | |
1061 | } | |
1062 | ||
1063 | static void shrink_stripes(raid5_conf_t *conf) | |
1064 | { | |
1065 | while (drop_one_stripe(conf)) | |
1066 | ; | |
1067 | ||
29fc7e3e N |
1068 | if (conf->slab_cache) |
1069 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1070 | conf->slab_cache = NULL; |
1071 | } | |
1072 | ||
6712ecf8 | 1073 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 LT |
1074 | { |
1075 | struct stripe_head *sh = bi->bi_private; | |
1076 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 1077 | int disks = sh->disks, i; |
1da177e4 | 1078 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1079 | char b[BDEVNAME_SIZE]; |
1080 | mdk_rdev_t *rdev; | |
1da177e4 | 1081 | |
1da177e4 LT |
1082 | |
1083 | for (i=0 ; i<disks; i++) | |
1084 | if (bi == &sh->dev[i].req) | |
1085 | break; | |
1086 | ||
45b4233c DW |
1087 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1088 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1089 | uptodate); |
1090 | if (i == disks) { | |
1091 | BUG(); | |
6712ecf8 | 1092 | return; |
1da177e4 LT |
1093 | } |
1094 | ||
1095 | if (uptodate) { | |
1da177e4 | 1096 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1097 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1098 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1099 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1100 | " (%lu sectors at %llu on %s)\n", | |
1101 | mdname(conf->mddev), STRIPE_SECTORS, | |
1102 | (unsigned long long)(sh->sector | |
1103 | + rdev->data_offset), | |
1104 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1105 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1106 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1107 | } | |
ba22dcbf N |
1108 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1109 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1110 | } else { |
d6950432 | 1111 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1112 | int retry = 0; |
d6950432 N |
1113 | rdev = conf->disks[i].rdev; |
1114 | ||
1da177e4 | 1115 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1116 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1117 | if (conf->mddev->degraded) |
6be9d494 BS |
1118 | printk_rl(KERN_WARNING |
1119 | "raid5:%s: read error not correctable " | |
1120 | "(sector %llu on %s).\n", | |
1121 | mdname(conf->mddev), | |
1122 | (unsigned long long)(sh->sector | |
1123 | + rdev->data_offset), | |
1124 | bdn); | |
ba22dcbf | 1125 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1126 | /* Oh, no!!! */ |
6be9d494 BS |
1127 | printk_rl(KERN_WARNING |
1128 | "raid5:%s: read error NOT corrected!! " | |
1129 | "(sector %llu on %s).\n", | |
1130 | mdname(conf->mddev), | |
1131 | (unsigned long long)(sh->sector | |
1132 | + rdev->data_offset), | |
1133 | bdn); | |
d6950432 | 1134 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1135 | > conf->max_nr_stripes) |
14f8d26b | 1136 | printk(KERN_WARNING |
d6950432 N |
1137 | "raid5:%s: Too many read errors, failing device %s.\n", |
1138 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1139 | else |
1140 | retry = 1; | |
1141 | if (retry) | |
1142 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1143 | else { | |
4e5314b5 N |
1144 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1145 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1146 | md_error(conf->mddev, rdev); |
ba22dcbf | 1147 | } |
1da177e4 LT |
1148 | } |
1149 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1150 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1151 | set_bit(STRIPE_HANDLE, &sh->state); | |
1152 | release_stripe(sh); | |
1da177e4 LT |
1153 | } |
1154 | ||
d710e138 | 1155 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 LT |
1156 | { |
1157 | struct stripe_head *sh = bi->bi_private; | |
1158 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 1159 | int disks = sh->disks, i; |
1da177e4 LT |
1160 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1161 | ||
1da177e4 LT |
1162 | for (i=0 ; i<disks; i++) |
1163 | if (bi == &sh->dev[i].req) | |
1164 | break; | |
1165 | ||
45b4233c | 1166 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1167 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1168 | uptodate); | |
1169 | if (i == disks) { | |
1170 | BUG(); | |
6712ecf8 | 1171 | return; |
1da177e4 LT |
1172 | } |
1173 | ||
1da177e4 LT |
1174 | if (!uptodate) |
1175 | md_error(conf->mddev, conf->disks[i].rdev); | |
1176 | ||
1177 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1178 | ||
1179 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1180 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1181 | release_stripe(sh); |
1da177e4 LT |
1182 | } |
1183 | ||
1184 | ||
1185 | static sector_t compute_blocknr(struct stripe_head *sh, int i); | |
1186 | ||
d710e138 | 1187 | static void raid5_build_block(struct stripe_head *sh, int i) |
1da177e4 LT |
1188 | { |
1189 | struct r5dev *dev = &sh->dev[i]; | |
1190 | ||
1191 | bio_init(&dev->req); | |
1192 | dev->req.bi_io_vec = &dev->vec; | |
1193 | dev->req.bi_vcnt++; | |
1194 | dev->req.bi_max_vecs++; | |
1195 | dev->vec.bv_page = dev->page; | |
1196 | dev->vec.bv_len = STRIPE_SIZE; | |
1197 | dev->vec.bv_offset = 0; | |
1198 | ||
1199 | dev->req.bi_sector = sh->sector; | |
1200 | dev->req.bi_private = sh; | |
1201 | ||
1202 | dev->flags = 0; | |
16a53ecc | 1203 | dev->sector = compute_blocknr(sh, i); |
1da177e4 LT |
1204 | } |
1205 | ||
1206 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1207 | { | |
1208 | char b[BDEVNAME_SIZE]; | |
1209 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1210 | pr_debug("raid5: error called\n"); |
1da177e4 | 1211 | |
b2d444d7 | 1212 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1213 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1214 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1215 | unsigned long flags; | |
1216 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1217 | mddev->degraded++; |
c04be0aa | 1218 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1219 | /* |
1220 | * if recovery was running, make sure it aborts. | |
1221 | */ | |
dfc70645 | 1222 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1223 | } |
b2d444d7 | 1224 | set_bit(Faulty, &rdev->flags); |
d710e138 N |
1225 | printk(KERN_ALERT |
1226 | "raid5: Disk failure on %s, disabling device.\n" | |
1227 | "raid5: Operation continuing on %d devices.\n", | |
1228 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); | |
1da177e4 | 1229 | } |
16a53ecc | 1230 | } |
1da177e4 LT |
1231 | |
1232 | /* | |
1233 | * Input: a 'big' sector number, | |
1234 | * Output: index of the data and parity disk, and the sector # in them. | |
1235 | */ | |
112bf897 N |
1236 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
1237 | int previous, | |
1238 | int *dd_idx, int *pd_idx) | |
1da177e4 LT |
1239 | { |
1240 | long stripe; | |
1241 | unsigned long chunk_number; | |
1242 | unsigned int chunk_offset; | |
1243 | sector_t new_sector; | |
1244 | int sectors_per_chunk = conf->chunk_size >> 9; | |
112bf897 N |
1245 | int raid_disks = previous ? conf->previous_raid_disks |
1246 | : conf->raid_disks; | |
1247 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1248 | |
1249 | /* First compute the information on this sector */ | |
1250 | ||
1251 | /* | |
1252 | * Compute the chunk number and the sector offset inside the chunk | |
1253 | */ | |
1254 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1255 | chunk_number = r_sector; | |
1256 | BUG_ON(r_sector != chunk_number); | |
1257 | ||
1258 | /* | |
1259 | * Compute the stripe number | |
1260 | */ | |
1261 | stripe = chunk_number / data_disks; | |
1262 | ||
1263 | /* | |
1264 | * Compute the data disk and parity disk indexes inside the stripe | |
1265 | */ | |
1266 | *dd_idx = chunk_number % data_disks; | |
1267 | ||
1268 | /* | |
1269 | * Select the parity disk based on the user selected algorithm. | |
1270 | */ | |
16a53ecc N |
1271 | switch(conf->level) { |
1272 | case 4: | |
1da177e4 | 1273 | *pd_idx = data_disks; |
16a53ecc N |
1274 | break; |
1275 | case 5: | |
1276 | switch (conf->algorithm) { | |
1da177e4 LT |
1277 | case ALGORITHM_LEFT_ASYMMETRIC: |
1278 | *pd_idx = data_disks - stripe % raid_disks; | |
1279 | if (*dd_idx >= *pd_idx) | |
1280 | (*dd_idx)++; | |
1281 | break; | |
1282 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1283 | *pd_idx = stripe % raid_disks; | |
1284 | if (*dd_idx >= *pd_idx) | |
1285 | (*dd_idx)++; | |
1286 | break; | |
1287 | case ALGORITHM_LEFT_SYMMETRIC: | |
1288 | *pd_idx = data_disks - stripe % raid_disks; | |
1289 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
1290 | break; | |
1291 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1292 | *pd_idx = stripe % raid_disks; | |
1293 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
1294 | break; | |
1295 | default: | |
14f8d26b | 1296 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 | 1297 | conf->algorithm); |
16a53ecc N |
1298 | } |
1299 | break; | |
1300 | case 6: | |
1301 | ||
1302 | /**** FIX THIS ****/ | |
1303 | switch (conf->algorithm) { | |
1304 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1305 | *pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1306 | if (*pd_idx == raid_disks-1) | |
1307 | (*dd_idx)++; /* Q D D D P */ | |
1308 | else if (*dd_idx >= *pd_idx) | |
1309 | (*dd_idx) += 2; /* D D P Q D */ | |
1310 | break; | |
1311 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1312 | *pd_idx = stripe % raid_disks; | |
1313 | if (*pd_idx == raid_disks-1) | |
1314 | (*dd_idx)++; /* Q D D D P */ | |
1315 | else if (*dd_idx >= *pd_idx) | |
1316 | (*dd_idx) += 2; /* D D P Q D */ | |
1317 | break; | |
1318 | case ALGORITHM_LEFT_SYMMETRIC: | |
1319 | *pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1320 | *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; | |
1321 | break; | |
1322 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1323 | *pd_idx = stripe % raid_disks; | |
1324 | *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; | |
1325 | break; | |
1326 | default: | |
d710e138 N |
1327 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1328 | conf->algorithm); | |
16a53ecc N |
1329 | } |
1330 | break; | |
1da177e4 LT |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * Finally, compute the new sector number | |
1335 | */ | |
1336 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1337 | return new_sector; | |
1338 | } | |
1339 | ||
1340 | ||
1341 | static sector_t compute_blocknr(struct stripe_head *sh, int i) | |
1342 | { | |
1343 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1344 | int raid_disks = sh->disks; |
1345 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1346 | sector_t new_sector = sh->sector, check; |
1347 | int sectors_per_chunk = conf->chunk_size >> 9; | |
1348 | sector_t stripe; | |
1349 | int chunk_offset; | |
1350 | int chunk_number, dummy1, dummy2, dd_idx = i; | |
1351 | sector_t r_sector; | |
1352 | ||
16a53ecc | 1353 | |
1da177e4 LT |
1354 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1355 | stripe = new_sector; | |
1356 | BUG_ON(new_sector != stripe); | |
1357 | ||
16a53ecc N |
1358 | if (i == sh->pd_idx) |
1359 | return 0; | |
1360 | switch(conf->level) { | |
1361 | case 4: break; | |
1362 | case 5: | |
1363 | switch (conf->algorithm) { | |
1da177e4 LT |
1364 | case ALGORITHM_LEFT_ASYMMETRIC: |
1365 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1366 | if (i > sh->pd_idx) | |
1367 | i--; | |
1368 | break; | |
1369 | case ALGORITHM_LEFT_SYMMETRIC: | |
1370 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1371 | if (i < sh->pd_idx) | |
1372 | i += raid_disks; | |
1373 | i -= (sh->pd_idx + 1); | |
1374 | break; | |
1375 | default: | |
14f8d26b | 1376 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
16a53ecc N |
1377 | conf->algorithm); |
1378 | } | |
1379 | break; | |
1380 | case 6: | |
16a53ecc N |
1381 | if (i == raid6_next_disk(sh->pd_idx, raid_disks)) |
1382 | return 0; /* It is the Q disk */ | |
1383 | switch (conf->algorithm) { | |
1384 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1385 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1386 | if (sh->pd_idx == raid_disks-1) | |
1387 | i--; /* Q D D D P */ | |
1388 | else if (i > sh->pd_idx) | |
1389 | i -= 2; /* D D P Q D */ | |
1390 | break; | |
1391 | case ALGORITHM_LEFT_SYMMETRIC: | |
1392 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1393 | if (sh->pd_idx == raid_disks-1) | |
1394 | i--; /* Q D D D P */ | |
1395 | else { | |
1396 | /* D D P Q D */ | |
1397 | if (i < sh->pd_idx) | |
1398 | i += raid_disks; | |
1399 | i -= (sh->pd_idx + 2); | |
1400 | } | |
1401 | break; | |
1402 | default: | |
d710e138 N |
1403 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1404 | conf->algorithm); | |
16a53ecc N |
1405 | } |
1406 | break; | |
1da177e4 LT |
1407 | } |
1408 | ||
1409 | chunk_number = stripe * data_disks + i; | |
1410 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1411 | ||
112bf897 N |
1412 | check = raid5_compute_sector(conf, r_sector, |
1413 | (raid_disks != conf->raid_disks), | |
1414 | &dummy1, &dummy2); | |
1da177e4 | 1415 | if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) { |
14f8d26b | 1416 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1417 | return 0; |
1418 | } | |
1419 | return r_sector; | |
1420 | } | |
1421 | ||
1422 | ||
1423 | ||
1424 | /* | |
16a53ecc N |
1425 | * Copy data between a page in the stripe cache, and one or more bion |
1426 | * The page could align with the middle of the bio, or there could be | |
1427 | * several bion, each with several bio_vecs, which cover part of the page | |
1428 | * Multiple bion are linked together on bi_next. There may be extras | |
1429 | * at the end of this list. We ignore them. | |
1da177e4 LT |
1430 | */ |
1431 | static void copy_data(int frombio, struct bio *bio, | |
1432 | struct page *page, | |
1433 | sector_t sector) | |
1434 | { | |
1435 | char *pa = page_address(page); | |
1436 | struct bio_vec *bvl; | |
1437 | int i; | |
1438 | int page_offset; | |
1439 | ||
1440 | if (bio->bi_sector >= sector) | |
1441 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
1442 | else | |
1443 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
1444 | bio_for_each_segment(bvl, bio, i) { | |
1445 | int len = bio_iovec_idx(bio,i)->bv_len; | |
1446 | int clen; | |
1447 | int b_offset = 0; | |
1448 | ||
1449 | if (page_offset < 0) { | |
1450 | b_offset = -page_offset; | |
1451 | page_offset += b_offset; | |
1452 | len -= b_offset; | |
1453 | } | |
1454 | ||
1455 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1456 | clen = STRIPE_SIZE - page_offset; | |
1457 | else clen = len; | |
16a53ecc | 1458 | |
1da177e4 LT |
1459 | if (clen > 0) { |
1460 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
1461 | if (frombio) | |
1462 | memcpy(pa+page_offset, ba+b_offset, clen); | |
1463 | else | |
1464 | memcpy(ba+b_offset, pa+page_offset, clen); | |
1465 | __bio_kunmap_atomic(ba, KM_USER0); | |
1466 | } | |
1467 | if (clen < len) /* hit end of page */ | |
1468 | break; | |
1469 | page_offset += len; | |
1470 | } | |
1471 | } | |
1472 | ||
9bc89cd8 DW |
1473 | #define check_xor() do { \ |
1474 | if (count == MAX_XOR_BLOCKS) { \ | |
1475 | xor_blocks(count, STRIPE_SIZE, dest, ptr);\ | |
1476 | count = 0; \ | |
1477 | } \ | |
1da177e4 LT |
1478 | } while(0) |
1479 | ||
16a53ecc N |
1480 | static void compute_parity6(struct stripe_head *sh, int method) |
1481 | { | |
bff61975 | 1482 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 1483 | int i, pd_idx = sh->pd_idx, qd_idx, d0_idx, disks = sh->disks, count; |
16a53ecc N |
1484 | struct bio *chosen; |
1485 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1486 | void *ptrs[disks]; | |
1487 | ||
1488 | qd_idx = raid6_next_disk(pd_idx, disks); | |
1489 | d0_idx = raid6_next_disk(qd_idx, disks); | |
1490 | ||
45b4233c | 1491 | pr_debug("compute_parity, stripe %llu, method %d\n", |
16a53ecc N |
1492 | (unsigned long long)sh->sector, method); |
1493 | ||
1494 | switch(method) { | |
1495 | case READ_MODIFY_WRITE: | |
1496 | BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ | |
1497 | case RECONSTRUCT_WRITE: | |
1498 | for (i= disks; i-- ;) | |
1499 | if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { | |
1500 | chosen = sh->dev[i].towrite; | |
1501 | sh->dev[i].towrite = NULL; | |
1502 | ||
1503 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1504 | wake_up(&conf->wait_for_overlap); | |
1505 | ||
52e5f9d1 | 1506 | BUG_ON(sh->dev[i].written); |
16a53ecc N |
1507 | sh->dev[i].written = chosen; |
1508 | } | |
1509 | break; | |
1510 | case CHECK_PARITY: | |
1511 | BUG(); /* Not implemented yet */ | |
1512 | } | |
1513 | ||
1514 | for (i = disks; i--;) | |
1515 | if (sh->dev[i].written) { | |
1516 | sector_t sector = sh->dev[i].sector; | |
1517 | struct bio *wbi = sh->dev[i].written; | |
1518 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
1519 | copy_data(1, wbi, sh->dev[i].page, sector); | |
1520 | wbi = r5_next_bio(wbi, sector); | |
1521 | } | |
1522 | ||
1523 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
1524 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
1525 | } | |
1526 | ||
1527 | // switch(method) { | |
1528 | // case RECONSTRUCT_WRITE: | |
1529 | // case CHECK_PARITY: | |
1530 | // case UPDATE_PARITY: | |
1531 | /* Note that unlike RAID-5, the ordering of the disks matters greatly. */ | |
1532 | /* FIX: Is this ordering of drives even remotely optimal? */ | |
1533 | count = 0; | |
1534 | i = d0_idx; | |
1535 | do { | |
1536 | ptrs[count++] = page_address(sh->dev[i].page); | |
1537 | if (count <= disks-2 && !test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1538 | printk("block %d/%d not uptodate on parity calc\n", i,count); | |
1539 | i = raid6_next_disk(i, disks); | |
1540 | } while ( i != d0_idx ); | |
1541 | // break; | |
1542 | // } | |
1543 | ||
1544 | raid6_call.gen_syndrome(disks, STRIPE_SIZE, ptrs); | |
1545 | ||
1546 | switch(method) { | |
1547 | case RECONSTRUCT_WRITE: | |
1548 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1549 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1550 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1551 | set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); | |
1552 | break; | |
1553 | case UPDATE_PARITY: | |
1554 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1555 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1556 | break; | |
1557 | } | |
1558 | } | |
1559 | ||
1560 | ||
1561 | /* Compute one missing block */ | |
1562 | static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero) | |
1563 | { | |
f416885e | 1564 | int i, count, disks = sh->disks; |
9bc89cd8 | 1565 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
16a53ecc N |
1566 | int pd_idx = sh->pd_idx; |
1567 | int qd_idx = raid6_next_disk(pd_idx, disks); | |
1568 | ||
45b4233c | 1569 | pr_debug("compute_block_1, stripe %llu, idx %d\n", |
16a53ecc N |
1570 | (unsigned long long)sh->sector, dd_idx); |
1571 | ||
1572 | if ( dd_idx == qd_idx ) { | |
1573 | /* We're actually computing the Q drive */ | |
1574 | compute_parity6(sh, UPDATE_PARITY); | |
1575 | } else { | |
9bc89cd8 DW |
1576 | dest = page_address(sh->dev[dd_idx].page); |
1577 | if (!nozero) memset(dest, 0, STRIPE_SIZE); | |
1578 | count = 0; | |
16a53ecc N |
1579 | for (i = disks ; i--; ) { |
1580 | if (i == dd_idx || i == qd_idx) | |
1581 | continue; | |
1582 | p = page_address(sh->dev[i].page); | |
1583 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1584 | ptr[count++] = p; | |
1585 | else | |
1586 | printk("compute_block() %d, stripe %llu, %d" | |
1587 | " not present\n", dd_idx, | |
1588 | (unsigned long long)sh->sector, i); | |
1589 | ||
1590 | check_xor(); | |
1591 | } | |
9bc89cd8 DW |
1592 | if (count) |
1593 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
16a53ecc N |
1594 | if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
1595 | else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
1596 | } | |
1597 | } | |
1598 | ||
1599 | /* Compute two missing blocks */ | |
1600 | static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) | |
1601 | { | |
f416885e | 1602 | int i, count, disks = sh->disks; |
16a53ecc N |
1603 | int pd_idx = sh->pd_idx; |
1604 | int qd_idx = raid6_next_disk(pd_idx, disks); | |
1605 | int d0_idx = raid6_next_disk(qd_idx, disks); | |
1606 | int faila, failb; | |
1607 | ||
1608 | /* faila and failb are disk numbers relative to d0_idx */ | |
1609 | /* pd_idx become disks-2 and qd_idx become disks-1 */ | |
1610 | faila = (dd_idx1 < d0_idx) ? dd_idx1+(disks-d0_idx) : dd_idx1-d0_idx; | |
1611 | failb = (dd_idx2 < d0_idx) ? dd_idx2+(disks-d0_idx) : dd_idx2-d0_idx; | |
1612 | ||
1613 | BUG_ON(faila == failb); | |
1614 | if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } | |
1615 | ||
45b4233c | 1616 | pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", |
16a53ecc N |
1617 | (unsigned long long)sh->sector, dd_idx1, dd_idx2, faila, failb); |
1618 | ||
1619 | if ( failb == disks-1 ) { | |
1620 | /* Q disk is one of the missing disks */ | |
1621 | if ( faila == disks-2 ) { | |
1622 | /* Missing P+Q, just recompute */ | |
1623 | compute_parity6(sh, UPDATE_PARITY); | |
1624 | return; | |
1625 | } else { | |
1626 | /* We're missing D+Q; recompute D from P */ | |
1627 | compute_block_1(sh, (dd_idx1 == qd_idx) ? dd_idx2 : dd_idx1, 0); | |
1628 | compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */ | |
1629 | return; | |
1630 | } | |
1631 | } | |
1632 | ||
1633 | /* We're missing D+P or D+D; build pointer table */ | |
1634 | { | |
1635 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1636 | void *ptrs[disks]; | |
1637 | ||
1638 | count = 0; | |
1639 | i = d0_idx; | |
1640 | do { | |
1641 | ptrs[count++] = page_address(sh->dev[i].page); | |
1642 | i = raid6_next_disk(i, disks); | |
1643 | if (i != dd_idx1 && i != dd_idx2 && | |
1644 | !test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1645 | printk("compute_2 with missing block %d/%d\n", count, i); | |
1646 | } while ( i != d0_idx ); | |
1647 | ||
1648 | if ( failb == disks-2 ) { | |
1649 | /* We're missing D+P. */ | |
1650 | raid6_datap_recov(disks, STRIPE_SIZE, faila, ptrs); | |
1651 | } else { | |
1652 | /* We're missing D+D. */ | |
1653 | raid6_2data_recov(disks, STRIPE_SIZE, faila, failb, ptrs); | |
1654 | } | |
1655 | ||
1656 | /* Both the above update both missing blocks */ | |
1657 | set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); | |
1658 | set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); | |
1659 | } | |
1660 | } | |
1661 | ||
600aa109 | 1662 | static void |
1fe797e6 | 1663 | schedule_reconstruction5(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 1664 | int rcw, int expand) |
e33129d8 DW |
1665 | { |
1666 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
e33129d8 DW |
1667 | |
1668 | if (rcw) { | |
1669 | /* if we are not expanding this is a proper write request, and | |
1670 | * there will be bios with new data to be drained into the | |
1671 | * stripe cache | |
1672 | */ | |
1673 | if (!expand) { | |
600aa109 DW |
1674 | sh->reconstruct_state = reconstruct_state_drain_run; |
1675 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1676 | } else | |
1677 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 1678 | |
600aa109 | 1679 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); |
e33129d8 DW |
1680 | |
1681 | for (i = disks; i--; ) { | |
1682 | struct r5dev *dev = &sh->dev[i]; | |
1683 | ||
1684 | if (dev->towrite) { | |
1685 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 1686 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
1687 | if (!expand) |
1688 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1689 | s->locked++; |
e33129d8 DW |
1690 | } |
1691 | } | |
600aa109 | 1692 | if (s->locked + 1 == disks) |
8b3e6cdc DW |
1693 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
1694 | atomic_inc(&sh->raid_conf->pending_full_writes); | |
e33129d8 DW |
1695 | } else { |
1696 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
1697 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
1698 | ||
d8ee0728 | 1699 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
1700 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
1701 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1702 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); | |
e33129d8 DW |
1703 | |
1704 | for (i = disks; i--; ) { | |
1705 | struct r5dev *dev = &sh->dev[i]; | |
1706 | if (i == pd_idx) | |
1707 | continue; | |
1708 | ||
e33129d8 DW |
1709 | if (dev->towrite && |
1710 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
1711 | test_bit(R5_Wantcompute, &dev->flags))) { |
1712 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
1713 | set_bit(R5_LOCKED, &dev->flags); |
1714 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1715 | s->locked++; |
e33129d8 DW |
1716 | } |
1717 | } | |
1718 | } | |
1719 | ||
1720 | /* keep the parity disk locked while asynchronous operations | |
1721 | * are in flight | |
1722 | */ | |
1723 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1724 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 1725 | s->locked++; |
e33129d8 | 1726 | |
600aa109 | 1727 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 1728 | __func__, (unsigned long long)sh->sector, |
600aa109 | 1729 | s->locked, s->ops_request); |
e33129d8 | 1730 | } |
16a53ecc | 1731 | |
1da177e4 LT |
1732 | /* |
1733 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 1734 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
1735 | * The bi_next chain must be in order. |
1736 | */ | |
1737 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
1738 | { | |
1739 | struct bio **bip; | |
1740 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 1741 | int firstwrite=0; |
1da177e4 | 1742 | |
45b4233c | 1743 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
1744 | (unsigned long long)bi->bi_sector, |
1745 | (unsigned long long)sh->sector); | |
1746 | ||
1747 | ||
1748 | spin_lock(&sh->lock); | |
1749 | spin_lock_irq(&conf->device_lock); | |
72626685 | 1750 | if (forwrite) { |
1da177e4 | 1751 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
1752 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
1753 | firstwrite = 1; | |
1754 | } else | |
1da177e4 LT |
1755 | bip = &sh->dev[dd_idx].toread; |
1756 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
1757 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
1758 | goto overlap; | |
1759 | bip = & (*bip)->bi_next; | |
1760 | } | |
1761 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
1762 | goto overlap; | |
1763 | ||
78bafebd | 1764 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
1765 | if (*bip) |
1766 | bi->bi_next = *bip; | |
1767 | *bip = bi; | |
960e739d | 1768 | bi->bi_phys_segments++; |
1da177e4 LT |
1769 | spin_unlock_irq(&conf->device_lock); |
1770 | spin_unlock(&sh->lock); | |
1771 | ||
45b4233c | 1772 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
1773 | (unsigned long long)bi->bi_sector, |
1774 | (unsigned long long)sh->sector, dd_idx); | |
1775 | ||
72626685 | 1776 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
1777 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
1778 | STRIPE_SECTORS, 0); | |
ae3c20cc | 1779 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
1780 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
1781 | } | |
1782 | ||
1da177e4 LT |
1783 | if (forwrite) { |
1784 | /* check if page is covered */ | |
1785 | sector_t sector = sh->dev[dd_idx].sector; | |
1786 | for (bi=sh->dev[dd_idx].towrite; | |
1787 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
1788 | bi && bi->bi_sector <= sector; | |
1789 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
1790 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
1791 | sector = bi->bi_sector + (bi->bi_size>>9); | |
1792 | } | |
1793 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
1794 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
1795 | } | |
1796 | return 1; | |
1797 | ||
1798 | overlap: | |
1799 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
1800 | spin_unlock_irq(&conf->device_lock); | |
1801 | spin_unlock(&sh->lock); | |
1802 | return 0; | |
1803 | } | |
1804 | ||
29269553 N |
1805 | static void end_reshape(raid5_conf_t *conf); |
1806 | ||
16a53ecc N |
1807 | static int page_is_zero(struct page *p) |
1808 | { | |
1809 | char *a = page_address(p); | |
1810 | return ((*(u32*)a) == 0 && | |
1811 | memcmp(a, a+4, STRIPE_SIZE-4)==0); | |
1812 | } | |
1813 | ||
112bf897 | 1814 | static int stripe_to_pdidx(sector_t stripe, raid5_conf_t *conf, int previous) |
ccfcc3c1 N |
1815 | { |
1816 | int sectors_per_chunk = conf->chunk_size >> 9; | |
ccfcc3c1 | 1817 | int pd_idx, dd_idx; |
2d2063ce | 1818 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 1819 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 1820 | |
112bf897 N |
1821 | raid5_compute_sector(conf, |
1822 | stripe * (disks - conf->max_degraded) | |
b875e531 | 1823 | *sectors_per_chunk + chunk_offset, |
112bf897 N |
1824 | previous, |
1825 | &dd_idx, &pd_idx); | |
ccfcc3c1 N |
1826 | return pd_idx; |
1827 | } | |
1828 | ||
a4456856 | 1829 | static void |
1fe797e6 | 1830 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
1831 | struct stripe_head_state *s, int disks, |
1832 | struct bio **return_bi) | |
1833 | { | |
1834 | int i; | |
1835 | for (i = disks; i--; ) { | |
1836 | struct bio *bi; | |
1837 | int bitmap_end = 0; | |
1838 | ||
1839 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
1840 | mdk_rdev_t *rdev; | |
1841 | rcu_read_lock(); | |
1842 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1843 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
1844 | /* multiple read failures in one stripe */ | |
1845 | md_error(conf->mddev, rdev); | |
1846 | rcu_read_unlock(); | |
1847 | } | |
1848 | spin_lock_irq(&conf->device_lock); | |
1849 | /* fail all writes first */ | |
1850 | bi = sh->dev[i].towrite; | |
1851 | sh->dev[i].towrite = NULL; | |
1852 | if (bi) { | |
1853 | s->to_write--; | |
1854 | bitmap_end = 1; | |
1855 | } | |
1856 | ||
1857 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1858 | wake_up(&conf->wait_for_overlap); | |
1859 | ||
1860 | while (bi && bi->bi_sector < | |
1861 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1862 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
1863 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 1864 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
1865 | md_write_end(conf->mddev); |
1866 | bi->bi_next = *return_bi; | |
1867 | *return_bi = bi; | |
1868 | } | |
1869 | bi = nextbi; | |
1870 | } | |
1871 | /* and fail all 'written' */ | |
1872 | bi = sh->dev[i].written; | |
1873 | sh->dev[i].written = NULL; | |
1874 | if (bi) bitmap_end = 1; | |
1875 | while (bi && bi->bi_sector < | |
1876 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1877 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
1878 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 1879 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
1880 | md_write_end(conf->mddev); |
1881 | bi->bi_next = *return_bi; | |
1882 | *return_bi = bi; | |
1883 | } | |
1884 | bi = bi2; | |
1885 | } | |
1886 | ||
b5e98d65 DW |
1887 | /* fail any reads if this device is non-operational and |
1888 | * the data has not reached the cache yet. | |
1889 | */ | |
1890 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
1891 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
1892 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
1893 | bi = sh->dev[i].toread; |
1894 | sh->dev[i].toread = NULL; | |
1895 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1896 | wake_up(&conf->wait_for_overlap); | |
1897 | if (bi) s->to_read--; | |
1898 | while (bi && bi->bi_sector < | |
1899 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1900 | struct bio *nextbi = | |
1901 | r5_next_bio(bi, sh->dev[i].sector); | |
1902 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 1903 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
1904 | bi->bi_next = *return_bi; |
1905 | *return_bi = bi; | |
1906 | } | |
1907 | bi = nextbi; | |
1908 | } | |
1909 | } | |
1910 | spin_unlock_irq(&conf->device_lock); | |
1911 | if (bitmap_end) | |
1912 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
1913 | STRIPE_SECTORS, 0, 0); | |
1914 | } | |
1915 | ||
8b3e6cdc DW |
1916 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
1917 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
1918 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
1919 | } |
1920 | ||
1fe797e6 DW |
1921 | /* fetch_block5 - checks the given member device to see if its data needs |
1922 | * to be read or computed to satisfy a request. | |
1923 | * | |
1924 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
1925 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 1926 | */ |
1fe797e6 DW |
1927 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
1928 | int disk_idx, int disks) | |
f38e1219 DW |
1929 | { |
1930 | struct r5dev *dev = &sh->dev[disk_idx]; | |
1931 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
1932 | ||
f38e1219 DW |
1933 | /* is the data in this block needed, and can we get it? */ |
1934 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
1935 | !test_bit(R5_UPTODATE, &dev->flags) && |
1936 | (dev->toread || | |
1937 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
1938 | s->syncing || s->expanding || | |
1939 | (s->failed && | |
1940 | (failed_dev->toread || | |
1941 | (failed_dev->towrite && | |
1942 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
1943 | /* We would like to get this block, possibly by computing it, |
1944 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
1945 | */ |
1946 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 1947 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
1948 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1949 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
1950 | set_bit(R5_Wantcompute, &dev->flags); |
1951 | sh->ops.target = disk_idx; | |
1952 | s->req_compute = 1; | |
f38e1219 DW |
1953 | /* Careful: from this point on 'uptodate' is in the eye |
1954 | * of raid5_run_ops which services 'compute' operations | |
1955 | * before writes. R5_Wantcompute flags a block that will | |
1956 | * be R5_UPTODATE by the time it is needed for a | |
1957 | * subsequent operation. | |
1958 | */ | |
1959 | s->uptodate++; | |
1fe797e6 | 1960 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 1961 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
1962 | set_bit(R5_LOCKED, &dev->flags); |
1963 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
1964 | s->locked++; |
1965 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
1966 | s->syncing); | |
1967 | } | |
1968 | } | |
1969 | ||
1fe797e6 | 1970 | return 0; |
f38e1219 DW |
1971 | } |
1972 | ||
1fe797e6 DW |
1973 | /** |
1974 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
1975 | */ | |
1976 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
1977 | struct stripe_head_state *s, int disks) |
1978 | { | |
1979 | int i; | |
f38e1219 | 1980 | |
f38e1219 DW |
1981 | /* look for blocks to read/compute, skip this if a compute |
1982 | * is already in flight, or if the stripe contents are in the | |
1983 | * midst of changing due to a write | |
1984 | */ | |
976ea8d4 | 1985 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 1986 | !sh->reconstruct_state) |
f38e1219 | 1987 | for (i = disks; i--; ) |
1fe797e6 | 1988 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 1989 | break; |
a4456856 DW |
1990 | set_bit(STRIPE_HANDLE, &sh->state); |
1991 | } | |
1992 | ||
1fe797e6 | 1993 | static void handle_stripe_fill6(struct stripe_head *sh, |
a4456856 DW |
1994 | struct stripe_head_state *s, struct r6_state *r6s, |
1995 | int disks) | |
1996 | { | |
1997 | int i; | |
1998 | for (i = disks; i--; ) { | |
1999 | struct r5dev *dev = &sh->dev[i]; | |
2000 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2001 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2002 | (dev->toread || (dev->towrite && | |
2003 | !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2004 | s->syncing || s->expanding || | |
2005 | (s->failed >= 1 && | |
2006 | (sh->dev[r6s->failed_num[0]].toread || | |
2007 | s->to_write)) || | |
2008 | (s->failed >= 2 && | |
2009 | (sh->dev[r6s->failed_num[1]].toread || | |
2010 | s->to_write)))) { | |
2011 | /* we would like to get this block, possibly | |
2012 | * by computing it, but we might not be able to | |
2013 | */ | |
c337869d DW |
2014 | if ((s->uptodate == disks - 1) && |
2015 | (s->failed && (i == r6s->failed_num[0] || | |
2016 | i == r6s->failed_num[1]))) { | |
45b4233c | 2017 | pr_debug("Computing stripe %llu block %d\n", |
a4456856 DW |
2018 | (unsigned long long)sh->sector, i); |
2019 | compute_block_1(sh, i, 0); | |
2020 | s->uptodate++; | |
2021 | } else if ( s->uptodate == disks-2 && s->failed >= 2 ) { | |
2022 | /* Computing 2-failure is *very* expensive; only | |
2023 | * do it if failed >= 2 | |
2024 | */ | |
2025 | int other; | |
2026 | for (other = disks; other--; ) { | |
2027 | if (other == i) | |
2028 | continue; | |
2029 | if (!test_bit(R5_UPTODATE, | |
2030 | &sh->dev[other].flags)) | |
2031 | break; | |
2032 | } | |
2033 | BUG_ON(other < 0); | |
45b4233c | 2034 | pr_debug("Computing stripe %llu blocks %d,%d\n", |
a4456856 DW |
2035 | (unsigned long long)sh->sector, |
2036 | i, other); | |
2037 | compute_block_2(sh, i, other); | |
2038 | s->uptodate += 2; | |
2039 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2040 | set_bit(R5_LOCKED, &dev->flags); | |
2041 | set_bit(R5_Wantread, &dev->flags); | |
2042 | s->locked++; | |
45b4233c | 2043 | pr_debug("Reading block %d (sync=%d)\n", |
a4456856 DW |
2044 | i, s->syncing); |
2045 | } | |
2046 | } | |
2047 | } | |
2048 | set_bit(STRIPE_HANDLE, &sh->state); | |
2049 | } | |
2050 | ||
2051 | ||
1fe797e6 | 2052 | /* handle_stripe_clean_event |
a4456856 DW |
2053 | * any written block on an uptodate or failed drive can be returned. |
2054 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2055 | * never LOCKED, so we don't need to test 'failed' directly. | |
2056 | */ | |
1fe797e6 | 2057 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2058 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2059 | { | |
2060 | int i; | |
2061 | struct r5dev *dev; | |
2062 | ||
2063 | for (i = disks; i--; ) | |
2064 | if (sh->dev[i].written) { | |
2065 | dev = &sh->dev[i]; | |
2066 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2067 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2068 | /* We can return any write requests */ | |
2069 | struct bio *wbi, *wbi2; | |
2070 | int bitmap_end = 0; | |
45b4233c | 2071 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2072 | spin_lock_irq(&conf->device_lock); |
2073 | wbi = dev->written; | |
2074 | dev->written = NULL; | |
2075 | while (wbi && wbi->bi_sector < | |
2076 | dev->sector + STRIPE_SECTORS) { | |
2077 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2078 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2079 | md_write_end(conf->mddev); |
2080 | wbi->bi_next = *return_bi; | |
2081 | *return_bi = wbi; | |
2082 | } | |
2083 | wbi = wbi2; | |
2084 | } | |
2085 | if (dev->towrite == NULL) | |
2086 | bitmap_end = 1; | |
2087 | spin_unlock_irq(&conf->device_lock); | |
2088 | if (bitmap_end) | |
2089 | bitmap_endwrite(conf->mddev->bitmap, | |
2090 | sh->sector, | |
2091 | STRIPE_SECTORS, | |
2092 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2093 | 0); | |
2094 | } | |
2095 | } | |
8b3e6cdc DW |
2096 | |
2097 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2098 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2099 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2100 | } |
2101 | ||
1fe797e6 | 2102 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2103 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2104 | { | |
2105 | int rmw = 0, rcw = 0, i; | |
2106 | for (i = disks; i--; ) { | |
2107 | /* would I have to read this buffer for read_modify_write */ | |
2108 | struct r5dev *dev = &sh->dev[i]; | |
2109 | if ((dev->towrite || i == sh->pd_idx) && | |
2110 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2111 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2112 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2113 | if (test_bit(R5_Insync, &dev->flags)) |
2114 | rmw++; | |
2115 | else | |
2116 | rmw += 2*disks; /* cannot read it */ | |
2117 | } | |
2118 | /* Would I have to read this buffer for reconstruct_write */ | |
2119 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2120 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2121 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2122 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2123 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2124 | else |
2125 | rcw += 2*disks; | |
2126 | } | |
2127 | } | |
45b4233c | 2128 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2129 | (unsigned long long)sh->sector, rmw, rcw); |
2130 | set_bit(STRIPE_HANDLE, &sh->state); | |
2131 | if (rmw < rcw && rmw > 0) | |
2132 | /* prefer read-modify-write, but need to get some data */ | |
2133 | for (i = disks; i--; ) { | |
2134 | struct r5dev *dev = &sh->dev[i]; | |
2135 | if ((dev->towrite || i == sh->pd_idx) && | |
2136 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2137 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2138 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2139 | test_bit(R5_Insync, &dev->flags)) { |
2140 | if ( | |
2141 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2142 | pr_debug("Read_old block " |
a4456856 DW |
2143 | "%d for r-m-w\n", i); |
2144 | set_bit(R5_LOCKED, &dev->flags); | |
2145 | set_bit(R5_Wantread, &dev->flags); | |
2146 | s->locked++; | |
2147 | } else { | |
2148 | set_bit(STRIPE_DELAYED, &sh->state); | |
2149 | set_bit(STRIPE_HANDLE, &sh->state); | |
2150 | } | |
2151 | } | |
2152 | } | |
2153 | if (rcw <= rmw && rcw > 0) | |
2154 | /* want reconstruct write, but need to get some data */ | |
2155 | for (i = disks; i--; ) { | |
2156 | struct r5dev *dev = &sh->dev[i]; | |
2157 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2158 | i != sh->pd_idx && | |
2159 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2160 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2161 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2162 | test_bit(R5_Insync, &dev->flags)) { |
2163 | if ( | |
2164 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2165 | pr_debug("Read_old block " |
a4456856 DW |
2166 | "%d for Reconstruct\n", i); |
2167 | set_bit(R5_LOCKED, &dev->flags); | |
2168 | set_bit(R5_Wantread, &dev->flags); | |
2169 | s->locked++; | |
2170 | } else { | |
2171 | set_bit(STRIPE_DELAYED, &sh->state); | |
2172 | set_bit(STRIPE_HANDLE, &sh->state); | |
2173 | } | |
2174 | } | |
2175 | } | |
2176 | /* now if nothing is locked, and if we have enough data, | |
2177 | * we can start a write request | |
2178 | */ | |
f38e1219 DW |
2179 | /* since handle_stripe can be called at any time we need to handle the |
2180 | * case where a compute block operation has been submitted and then a | |
2181 | * subsequent call wants to start a write request. raid5_run_ops only | |
2182 | * handles the case where compute block and postxor are requested | |
2183 | * simultaneously. If this is not the case then new writes need to be | |
2184 | * held off until the compute completes. | |
2185 | */ | |
976ea8d4 DW |
2186 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2187 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2188 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
1fe797e6 | 2189 | schedule_reconstruction5(sh, s, rcw == 0, 0); |
a4456856 DW |
2190 | } |
2191 | ||
1fe797e6 | 2192 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2193 | struct stripe_head *sh, struct stripe_head_state *s, |
2194 | struct r6_state *r6s, int disks) | |
2195 | { | |
2196 | int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i; | |
2197 | int qd_idx = r6s->qd_idx; | |
2198 | for (i = disks; i--; ) { | |
2199 | struct r5dev *dev = &sh->dev[i]; | |
2200 | /* Would I have to read this buffer for reconstruct_write */ | |
2201 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2202 | && i != pd_idx && i != qd_idx | |
2203 | && (!test_bit(R5_LOCKED, &dev->flags) | |
2204 | ) && | |
2205 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
2206 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
2207 | else { | |
45b4233c | 2208 | pr_debug("raid6: must_compute: " |
a4456856 DW |
2209 | "disk %d flags=%#lx\n", i, dev->flags); |
2210 | must_compute++; | |
2211 | } | |
2212 | } | |
2213 | } | |
45b4233c | 2214 | pr_debug("for sector %llu, rcw=%d, must_compute=%d\n", |
a4456856 DW |
2215 | (unsigned long long)sh->sector, rcw, must_compute); |
2216 | set_bit(STRIPE_HANDLE, &sh->state); | |
2217 | ||
2218 | if (rcw > 0) | |
2219 | /* want reconstruct write, but need to get some data */ | |
2220 | for (i = disks; i--; ) { | |
2221 | struct r5dev *dev = &sh->dev[i]; | |
2222 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2223 | && !(s->failed == 0 && (i == pd_idx || i == qd_idx)) | |
2224 | && !test_bit(R5_LOCKED, &dev->flags) && | |
2225 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2226 | test_bit(R5_Insync, &dev->flags)) { | |
2227 | if ( | |
2228 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2229 | pr_debug("Read_old stripe %llu " |
a4456856 DW |
2230 | "block %d for Reconstruct\n", |
2231 | (unsigned long long)sh->sector, i); | |
2232 | set_bit(R5_LOCKED, &dev->flags); | |
2233 | set_bit(R5_Wantread, &dev->flags); | |
2234 | s->locked++; | |
2235 | } else { | |
45b4233c | 2236 | pr_debug("Request delayed stripe %llu " |
a4456856 DW |
2237 | "block %d for Reconstruct\n", |
2238 | (unsigned long long)sh->sector, i); | |
2239 | set_bit(STRIPE_DELAYED, &sh->state); | |
2240 | set_bit(STRIPE_HANDLE, &sh->state); | |
2241 | } | |
2242 | } | |
2243 | } | |
2244 | /* now if nothing is locked, and if we have enough data, we can start a | |
2245 | * write request | |
2246 | */ | |
2247 | if (s->locked == 0 && rcw == 0 && | |
2248 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
2249 | if (must_compute > 0) { | |
2250 | /* We have failed blocks and need to compute them */ | |
2251 | switch (s->failed) { | |
2252 | case 0: | |
2253 | BUG(); | |
2254 | case 1: | |
2255 | compute_block_1(sh, r6s->failed_num[0], 0); | |
2256 | break; | |
2257 | case 2: | |
2258 | compute_block_2(sh, r6s->failed_num[0], | |
2259 | r6s->failed_num[1]); | |
2260 | break; | |
2261 | default: /* This request should have been failed? */ | |
2262 | BUG(); | |
2263 | } | |
2264 | } | |
2265 | ||
45b4233c | 2266 | pr_debug("Computing parity for stripe %llu\n", |
a4456856 DW |
2267 | (unsigned long long)sh->sector); |
2268 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
2269 | /* now every locked buffer is ready to be written */ | |
2270 | for (i = disks; i--; ) | |
2271 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
45b4233c | 2272 | pr_debug("Writing stripe %llu block %d\n", |
a4456856 DW |
2273 | (unsigned long long)sh->sector, i); |
2274 | s->locked++; | |
2275 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
2276 | } | |
8b3e6cdc DW |
2277 | if (s->locked == disks) |
2278 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2279 | atomic_inc(&conf->pending_full_writes); | |
a4456856 DW |
2280 | /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */ |
2281 | set_bit(STRIPE_INSYNC, &sh->state); | |
2282 | ||
2283 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2284 | atomic_dec(&conf->preread_active_stripes); | |
2285 | if (atomic_read(&conf->preread_active_stripes) < | |
2286 | IO_THRESHOLD) | |
2287 | md_wakeup_thread(conf->mddev->thread); | |
2288 | } | |
2289 | } | |
2290 | } | |
2291 | ||
2292 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2293 | struct stripe_head_state *s, int disks) | |
2294 | { | |
ecc65c9b | 2295 | struct r5dev *dev = NULL; |
bd2ab670 | 2296 | |
a4456856 | 2297 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2298 | |
ecc65c9b DW |
2299 | switch (sh->check_state) { |
2300 | case check_state_idle: | |
2301 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2302 | if (s->failed == 0) { |
bd2ab670 | 2303 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2304 | sh->check_state = check_state_run; |
2305 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2306 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2307 | s->uptodate--; |
ecc65c9b | 2308 | break; |
bd2ab670 | 2309 | } |
ecc65c9b DW |
2310 | dev = &sh->dev[s->failed_num]; |
2311 | /* fall through */ | |
2312 | case check_state_compute_result: | |
2313 | sh->check_state = check_state_idle; | |
2314 | if (!dev) | |
2315 | dev = &sh->dev[sh->pd_idx]; | |
2316 | ||
2317 | /* check that a write has not made the stripe insync */ | |
2318 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2319 | break; | |
c8894419 | 2320 | |
a4456856 | 2321 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2322 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2323 | BUG_ON(s->uptodate != disks); | |
2324 | ||
2325 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2326 | s->locked++; |
a4456856 | 2327 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2328 | |
a4456856 | 2329 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2330 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2331 | break; |
2332 | case check_state_run: | |
2333 | break; /* we will be called again upon completion */ | |
2334 | case check_state_check_result: | |
2335 | sh->check_state = check_state_idle; | |
2336 | ||
2337 | /* if a failure occurred during the check operation, leave | |
2338 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2339 | */ | |
2340 | if (s->failed) | |
2341 | break; | |
2342 | ||
2343 | /* handle a successful check operation, if parity is correct | |
2344 | * we are done. Otherwise update the mismatch count and repair | |
2345 | * parity if !MD_RECOVERY_CHECK | |
2346 | */ | |
2347 | if (sh->ops.zero_sum_result == 0) | |
2348 | /* parity is correct (on disc, | |
2349 | * not in buffer any more) | |
2350 | */ | |
2351 | set_bit(STRIPE_INSYNC, &sh->state); | |
2352 | else { | |
2353 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2354 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2355 | /* don't try to repair!! */ | |
2356 | set_bit(STRIPE_INSYNC, &sh->state); | |
2357 | else { | |
2358 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2359 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2360 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2361 | set_bit(R5_Wantcompute, | |
2362 | &sh->dev[sh->pd_idx].flags); | |
2363 | sh->ops.target = sh->pd_idx; | |
2364 | s->uptodate++; | |
2365 | } | |
2366 | } | |
2367 | break; | |
2368 | case check_state_compute_run: | |
2369 | break; | |
2370 | default: | |
2371 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2372 | __func__, sh->check_state, | |
2373 | (unsigned long long) sh->sector); | |
2374 | BUG(); | |
a4456856 DW |
2375 | } |
2376 | } | |
2377 | ||
2378 | ||
2379 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
2380 | struct stripe_head_state *s, | |
2381 | struct r6_state *r6s, struct page *tmp_page, | |
2382 | int disks) | |
2383 | { | |
2384 | int update_p = 0, update_q = 0; | |
2385 | struct r5dev *dev; | |
2386 | int pd_idx = sh->pd_idx; | |
2387 | int qd_idx = r6s->qd_idx; | |
2388 | ||
2389 | set_bit(STRIPE_HANDLE, &sh->state); | |
2390 | ||
2391 | BUG_ON(s->failed > 2); | |
2392 | BUG_ON(s->uptodate < disks); | |
2393 | /* Want to check and possibly repair P and Q. | |
2394 | * However there could be one 'failed' device, in which | |
2395 | * case we can only check one of them, possibly using the | |
2396 | * other to generate missing data | |
2397 | */ | |
2398 | ||
2399 | /* If !tmp_page, we cannot do the calculations, | |
2400 | * but as we have set STRIPE_HANDLE, we will soon be called | |
2401 | * by stripe_handle with a tmp_page - just wait until then. | |
2402 | */ | |
2403 | if (tmp_page) { | |
2404 | if (s->failed == r6s->q_failed) { | |
2405 | /* The only possible failed device holds 'Q', so it | |
2406 | * makes sense to check P (If anything else were failed, | |
2407 | * we would have used P to recreate it). | |
2408 | */ | |
2409 | compute_block_1(sh, pd_idx, 1); | |
2410 | if (!page_is_zero(sh->dev[pd_idx].page)) { | |
2411 | compute_block_1(sh, pd_idx, 0); | |
2412 | update_p = 1; | |
2413 | } | |
2414 | } | |
2415 | if (!r6s->q_failed && s->failed < 2) { | |
2416 | /* q is not failed, and we didn't use it to generate | |
2417 | * anything, so it makes sense to check it | |
2418 | */ | |
2419 | memcpy(page_address(tmp_page), | |
2420 | page_address(sh->dev[qd_idx].page), | |
2421 | STRIPE_SIZE); | |
2422 | compute_parity6(sh, UPDATE_PARITY); | |
2423 | if (memcmp(page_address(tmp_page), | |
2424 | page_address(sh->dev[qd_idx].page), | |
2425 | STRIPE_SIZE) != 0) { | |
2426 | clear_bit(STRIPE_INSYNC, &sh->state); | |
2427 | update_q = 1; | |
2428 | } | |
2429 | } | |
2430 | if (update_p || update_q) { | |
2431 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2432 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2433 | /* don't try to repair!! */ | |
2434 | update_p = update_q = 0; | |
2435 | } | |
2436 | ||
2437 | /* now write out any block on a failed drive, | |
2438 | * or P or Q if they need it | |
2439 | */ | |
2440 | ||
2441 | if (s->failed == 2) { | |
2442 | dev = &sh->dev[r6s->failed_num[1]]; | |
2443 | s->locked++; | |
2444 | set_bit(R5_LOCKED, &dev->flags); | |
2445 | set_bit(R5_Wantwrite, &dev->flags); | |
2446 | } | |
2447 | if (s->failed >= 1) { | |
2448 | dev = &sh->dev[r6s->failed_num[0]]; | |
2449 | s->locked++; | |
2450 | set_bit(R5_LOCKED, &dev->flags); | |
2451 | set_bit(R5_Wantwrite, &dev->flags); | |
2452 | } | |
2453 | ||
2454 | if (update_p) { | |
2455 | dev = &sh->dev[pd_idx]; | |
2456 | s->locked++; | |
2457 | set_bit(R5_LOCKED, &dev->flags); | |
2458 | set_bit(R5_Wantwrite, &dev->flags); | |
2459 | } | |
2460 | if (update_q) { | |
2461 | dev = &sh->dev[qd_idx]; | |
2462 | s->locked++; | |
2463 | set_bit(R5_LOCKED, &dev->flags); | |
2464 | set_bit(R5_Wantwrite, &dev->flags); | |
2465 | } | |
2466 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2467 | ||
2468 | set_bit(STRIPE_INSYNC, &sh->state); | |
2469 | } | |
2470 | } | |
2471 | ||
2472 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2473 | struct r6_state *r6s) | |
2474 | { | |
2475 | int i; | |
2476 | ||
2477 | /* We have read all the blocks in this stripe and now we need to | |
2478 | * copy some of them into a target stripe for expand. | |
2479 | */ | |
f0a50d37 | 2480 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2481 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2482 | for (i = 0; i < sh->disks; i++) | |
a2e08551 | 2483 | if (i != sh->pd_idx && (!r6s || i != r6s->qd_idx)) { |
a4456856 DW |
2484 | int dd_idx, pd_idx, j; |
2485 | struct stripe_head *sh2; | |
2486 | ||
2487 | sector_t bn = compute_blocknr(sh, i); | |
112bf897 N |
2488 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2489 | &dd_idx, &pd_idx); | |
b5663ba4 | 2490 | sh2 = get_active_stripe(conf, s, 0, 1); |
a4456856 DW |
2491 | if (sh2 == NULL) |
2492 | /* so far only the early blocks of this stripe | |
2493 | * have been requested. When later blocks | |
2494 | * get requested, we will try again | |
2495 | */ | |
2496 | continue; | |
2497 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2498 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2499 | /* must have already done this block */ | |
2500 | release_stripe(sh2); | |
2501 | continue; | |
2502 | } | |
f0a50d37 DW |
2503 | |
2504 | /* place all the copies on one channel */ | |
2505 | tx = async_memcpy(sh2->dev[dd_idx].page, | |
2506 | sh->dev[i].page, 0, 0, STRIPE_SIZE, | |
2507 | ASYNC_TX_DEP_ACK, tx, NULL, NULL); | |
2508 | ||
a4456856 DW |
2509 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2510 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2511 | for (j = 0; j < conf->raid_disks; j++) | |
2512 | if (j != sh2->pd_idx && | |
a2e08551 N |
2513 | (!r6s || j != raid6_next_disk(sh2->pd_idx, |
2514 | sh2->disks)) && | |
a4456856 DW |
2515 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2516 | break; | |
2517 | if (j == conf->raid_disks) { | |
2518 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2519 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2520 | } | |
2521 | release_stripe(sh2); | |
f0a50d37 | 2522 | |
a4456856 | 2523 | } |
a2e08551 N |
2524 | /* done submitting copies, wait for them to complete */ |
2525 | if (tx) { | |
2526 | async_tx_ack(tx); | |
2527 | dma_wait_for_async_tx(tx); | |
2528 | } | |
a4456856 | 2529 | } |
1da177e4 | 2530 | |
6bfe0b49 | 2531 | |
1da177e4 LT |
2532 | /* |
2533 | * handle_stripe - do things to a stripe. | |
2534 | * | |
2535 | * We lock the stripe and then examine the state of various bits | |
2536 | * to see what needs to be done. | |
2537 | * Possible results: | |
2538 | * return some read request which now have data | |
2539 | * return some write requests which are safely on disc | |
2540 | * schedule a read on some buffers | |
2541 | * schedule a write of some buffers | |
2542 | * return confirmation of parity correctness | |
2543 | * | |
1da177e4 LT |
2544 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2545 | * get BH_Lock set before the stripe lock is released. | |
2546 | * | |
2547 | */ | |
a4456856 | 2548 | |
df10cfbc | 2549 | static bool handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2550 | { |
2551 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2552 | int disks = sh->disks, i; |
2553 | struct bio *return_bi = NULL; | |
2554 | struct stripe_head_state s; | |
1da177e4 | 2555 | struct r5dev *dev; |
6bfe0b49 | 2556 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2557 | int prexor; |
1da177e4 | 2558 | |
a4456856 | 2559 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
2560 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
2561 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
2562 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
2563 | sh->reconstruct_state); | |
1da177e4 LT |
2564 | |
2565 | spin_lock(&sh->lock); | |
2566 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2567 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2568 | ||
a4456856 DW |
2569 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2570 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2571 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 2572 | |
83de75cc | 2573 | /* Now to look around and see what can be done */ |
9910f16a | 2574 | rcu_read_lock(); |
1da177e4 LT |
2575 | for (i=disks; i--; ) { |
2576 | mdk_rdev_t *rdev; | |
a4456856 | 2577 | struct r5dev *dev = &sh->dev[i]; |
1da177e4 | 2578 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2579 | |
b5e98d65 DW |
2580 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2581 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2582 | dev->towrite, dev->written); | |
2583 | ||
2584 | /* maybe we can request a biofill operation | |
2585 | * | |
2586 | * new wantfill requests are only permitted while | |
83de75cc | 2587 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
2588 | */ |
2589 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 2590 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 2591 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
2592 | |
2593 | /* now count some things */ | |
a4456856 DW |
2594 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2595 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2596 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2597 | |
b5e98d65 DW |
2598 | if (test_bit(R5_Wantfill, &dev->flags)) |
2599 | s.to_fill++; | |
2600 | else if (dev->toread) | |
a4456856 | 2601 | s.to_read++; |
1da177e4 | 2602 | if (dev->towrite) { |
a4456856 | 2603 | s.to_write++; |
1da177e4 | 2604 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2605 | s.non_overwrite++; |
1da177e4 | 2606 | } |
a4456856 DW |
2607 | if (dev->written) |
2608 | s.written++; | |
9910f16a | 2609 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
2610 | if (blocked_rdev == NULL && |
2611 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
2612 | blocked_rdev = rdev; |
2613 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 2614 | } |
b2d444d7 | 2615 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 2616 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
2617 | clear_bit(R5_ReadError, &dev->flags); |
2618 | clear_bit(R5_ReWrite, &dev->flags); | |
2619 | } | |
b2d444d7 | 2620 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 2621 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
2622 | s.failed++; |
2623 | s.failed_num = i; | |
1da177e4 LT |
2624 | } else |
2625 | set_bit(R5_Insync, &dev->flags); | |
2626 | } | |
9910f16a | 2627 | rcu_read_unlock(); |
b5e98d65 | 2628 | |
6bfe0b49 | 2629 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
2630 | if (s.syncing || s.expanding || s.expanded || |
2631 | s.to_write || s.written) { | |
2632 | set_bit(STRIPE_HANDLE, &sh->state); | |
2633 | goto unlock; | |
2634 | } | |
2635 | /* There is nothing for the blocked_rdev to block */ | |
2636 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
2637 | blocked_rdev = NULL; | |
6bfe0b49 DW |
2638 | } |
2639 | ||
83de75cc DW |
2640 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
2641 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
2642 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
2643 | } | |
b5e98d65 | 2644 | |
45b4233c | 2645 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 2646 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
2647 | s.locked, s.uptodate, s.to_read, s.to_write, |
2648 | s.failed, s.failed_num); | |
1da177e4 LT |
2649 | /* check if the array has lost two devices and, if so, some requests might |
2650 | * need to be failed | |
2651 | */ | |
a4456856 | 2652 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 2653 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 2654 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
2655 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2656 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2657 | s.syncing = 0; |
1da177e4 LT |
2658 | } |
2659 | ||
2660 | /* might be able to return some write requests if the parity block | |
2661 | * is safe, or on a failed drive | |
2662 | */ | |
2663 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
2664 | if ( s.written && |
2665 | ((test_bit(R5_Insync, &dev->flags) && | |
2666 | !test_bit(R5_LOCKED, &dev->flags) && | |
2667 | test_bit(R5_UPTODATE, &dev->flags)) || | |
2668 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 2669 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
2670 | |
2671 | /* Now we might consider reading some blocks, either to check/generate | |
2672 | * parity, or to satisfy requests | |
2673 | * or to load a block that is being partially written. | |
2674 | */ | |
a4456856 | 2675 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 2676 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 2677 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 2678 | |
e33129d8 DW |
2679 | /* Now we check to see if any write operations have recently |
2680 | * completed | |
2681 | */ | |
e0a115e5 | 2682 | prexor = 0; |
d8ee0728 | 2683 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 2684 | prexor = 1; |
d8ee0728 DW |
2685 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
2686 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 2687 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
2688 | |
2689 | /* All the 'written' buffers and the parity block are ready to | |
2690 | * be written back to disk | |
2691 | */ | |
2692 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
2693 | for (i = disks; i--; ) { | |
2694 | dev = &sh->dev[i]; | |
2695 | if (test_bit(R5_LOCKED, &dev->flags) && | |
2696 | (i == sh->pd_idx || dev->written)) { | |
2697 | pr_debug("Writing block %d\n", i); | |
2698 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
2699 | if (prexor) |
2700 | continue; | |
e33129d8 DW |
2701 | if (!test_bit(R5_Insync, &dev->flags) || |
2702 | (i == sh->pd_idx && s.failed == 0)) | |
2703 | set_bit(STRIPE_INSYNC, &sh->state); | |
2704 | } | |
2705 | } | |
2706 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2707 | atomic_dec(&conf->preread_active_stripes); | |
2708 | if (atomic_read(&conf->preread_active_stripes) < | |
2709 | IO_THRESHOLD) | |
2710 | md_wakeup_thread(conf->mddev->thread); | |
2711 | } | |
2712 | } | |
2713 | ||
2714 | /* Now to consider new write requests and what else, if anything | |
2715 | * should be read. We do not handle new writes when: | |
2716 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
2717 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
2718 | * block. | |
2719 | */ | |
600aa109 | 2720 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 2721 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
2722 | |
2723 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
2724 | * Any reads will already have been scheduled, so we just see if enough |
2725 | * data is available. The parity check is held off while parity | |
2726 | * dependent operations are in flight. | |
1da177e4 | 2727 | */ |
ecc65c9b DW |
2728 | if (sh->check_state || |
2729 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 2730 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 2731 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 2732 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 2733 | |
a4456856 | 2734 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
2735 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2736 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2737 | } | |
4e5314b5 N |
2738 | |
2739 | /* If the failed drive is just a ReadError, then we might need to progress | |
2740 | * the repair/check process | |
2741 | */ | |
a4456856 DW |
2742 | if (s.failed == 1 && !conf->mddev->ro && |
2743 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
2744 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
2745 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 2746 | ) { |
a4456856 | 2747 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
2748 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
2749 | set_bit(R5_Wantwrite, &dev->flags); | |
2750 | set_bit(R5_ReWrite, &dev->flags); | |
2751 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2752 | s.locked++; |
4e5314b5 N |
2753 | } else { |
2754 | /* let's read it back */ | |
2755 | set_bit(R5_Wantread, &dev->flags); | |
2756 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2757 | s.locked++; |
4e5314b5 N |
2758 | } |
2759 | } | |
2760 | ||
600aa109 DW |
2761 | /* Finish reconstruct operations initiated by the expansion process */ |
2762 | if (sh->reconstruct_state == reconstruct_state_result) { | |
2763 | sh->reconstruct_state = reconstruct_state_idle; | |
f0a50d37 | 2764 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 2765 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 2766 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 2767 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 2768 | s.locked++; |
23397883 | 2769 | } |
f0a50d37 DW |
2770 | } |
2771 | ||
2772 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 2773 | !sh->reconstruct_state) { |
f0a50d37 DW |
2774 | /* Need to write out all blocks after computing parity */ |
2775 | sh->disks = conf->raid_disks; | |
112bf897 | 2776 | sh->pd_idx = stripe_to_pdidx(sh->sector, conf, 0); |
1fe797e6 | 2777 | schedule_reconstruction5(sh, &s, 1, 1); |
600aa109 | 2778 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 2779 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 2780 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
2781 | wake_up(&conf->wait_for_overlap); |
2782 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
2783 | } | |
2784 | ||
0f94e87c | 2785 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 2786 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 2787 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 2788 | |
6bfe0b49 | 2789 | unlock: |
1da177e4 LT |
2790 | spin_unlock(&sh->lock); |
2791 | ||
6bfe0b49 DW |
2792 | /* wait for this device to become unblocked */ |
2793 | if (unlikely(blocked_rdev)) | |
2794 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
2795 | ||
600aa109 DW |
2796 | if (s.ops_request) |
2797 | raid5_run_ops(sh, s.ops_request); | |
d84e0f10 | 2798 | |
c4e5ac0a | 2799 | ops_run_io(sh, &s); |
1da177e4 | 2800 | |
a4456856 | 2801 | return_io(return_bi); |
df10cfbc DW |
2802 | |
2803 | return blocked_rdev == NULL; | |
1da177e4 LT |
2804 | } |
2805 | ||
df10cfbc | 2806 | static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page) |
1da177e4 | 2807 | { |
bff61975 | 2808 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 2809 | int disks = sh->disks; |
a4456856 DW |
2810 | struct bio *return_bi = NULL; |
2811 | int i, pd_idx = sh->pd_idx; | |
2812 | struct stripe_head_state s; | |
2813 | struct r6_state r6s; | |
16a53ecc | 2814 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 2815 | mdk_rdev_t *blocked_rdev = NULL; |
1da177e4 | 2816 | |
a4456856 | 2817 | r6s.qd_idx = raid6_next_disk(pd_idx, disks); |
45b4233c | 2818 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
a4456856 DW |
2819 | "pd_idx=%d, qd_idx=%d\n", |
2820 | (unsigned long long)sh->sector, sh->state, | |
2821 | atomic_read(&sh->count), pd_idx, r6s.qd_idx); | |
2822 | memset(&s, 0, sizeof(s)); | |
72626685 | 2823 | |
16a53ecc N |
2824 | spin_lock(&sh->lock); |
2825 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2826 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2827 | ||
a4456856 DW |
2828 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2829 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2830 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 2831 | /* Now to look around and see what can be done */ |
1da177e4 LT |
2832 | |
2833 | rcu_read_lock(); | |
16a53ecc N |
2834 | for (i=disks; i--; ) { |
2835 | mdk_rdev_t *rdev; | |
2836 | dev = &sh->dev[i]; | |
2837 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 2838 | |
45b4233c | 2839 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc N |
2840 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
2841 | /* maybe we can reply to a read */ | |
2842 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
2843 | struct bio *rbi, *rbi2; | |
45b4233c | 2844 | pr_debug("Return read for disc %d\n", i); |
16a53ecc N |
2845 | spin_lock_irq(&conf->device_lock); |
2846 | rbi = dev->toread; | |
2847 | dev->toread = NULL; | |
2848 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2849 | wake_up(&conf->wait_for_overlap); | |
2850 | spin_unlock_irq(&conf->device_lock); | |
2851 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
2852 | copy_data(0, rbi, dev->page, dev->sector); | |
2853 | rbi2 = r5_next_bio(rbi, dev->sector); | |
2854 | spin_lock_irq(&conf->device_lock); | |
960e739d | 2855 | if (!raid5_dec_bi_phys_segments(rbi)) { |
16a53ecc N |
2856 | rbi->bi_next = return_bi; |
2857 | return_bi = rbi; | |
2858 | } | |
2859 | spin_unlock_irq(&conf->device_lock); | |
2860 | rbi = rbi2; | |
2861 | } | |
2862 | } | |
1da177e4 | 2863 | |
16a53ecc | 2864 | /* now count some things */ |
a4456856 DW |
2865 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2866 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 2867 | |
16a53ecc | 2868 | |
a4456856 DW |
2869 | if (dev->toread) |
2870 | s.to_read++; | |
16a53ecc | 2871 | if (dev->towrite) { |
a4456856 | 2872 | s.to_write++; |
16a53ecc | 2873 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2874 | s.non_overwrite++; |
16a53ecc | 2875 | } |
a4456856 DW |
2876 | if (dev->written) |
2877 | s.written++; | |
16a53ecc | 2878 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
2879 | if (blocked_rdev == NULL && |
2880 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
2881 | blocked_rdev = rdev; |
2882 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 2883 | } |
16a53ecc N |
2884 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
2885 | /* The ReadError flag will just be confusing now */ | |
2886 | clear_bit(R5_ReadError, &dev->flags); | |
2887 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 2888 | } |
16a53ecc N |
2889 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
2890 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
2891 | if (s.failed < 2) |
2892 | r6s.failed_num[s.failed] = i; | |
2893 | s.failed++; | |
16a53ecc N |
2894 | } else |
2895 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
2896 | } |
2897 | rcu_read_unlock(); | |
6bfe0b49 DW |
2898 | |
2899 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
2900 | if (s.syncing || s.expanding || s.expanded || |
2901 | s.to_write || s.written) { | |
2902 | set_bit(STRIPE_HANDLE, &sh->state); | |
2903 | goto unlock; | |
2904 | } | |
2905 | /* There is nothing for the blocked_rdev to block */ | |
2906 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
2907 | blocked_rdev = NULL; | |
6bfe0b49 | 2908 | } |
ac4090d2 | 2909 | |
45b4233c | 2910 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 2911 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
2912 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
2913 | r6s.failed_num[0], r6s.failed_num[1]); | |
2914 | /* check if the array has lost >2 devices and, if so, some requests | |
2915 | * might need to be failed | |
16a53ecc | 2916 | */ |
a4456856 | 2917 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 2918 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 2919 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
2920 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2921 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2922 | s.syncing = 0; |
16a53ecc N |
2923 | } |
2924 | ||
2925 | /* | |
2926 | * might be able to return some write requests if the parity blocks | |
2927 | * are safe, or on a failed drive | |
2928 | */ | |
2929 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
2930 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
2931 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
2932 | qdev = &sh->dev[r6s.qd_idx]; | |
2933 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == r6s.qd_idx) | |
2934 | || (s.failed >= 2 && r6s.failed_num[1] == r6s.qd_idx); | |
2935 | ||
2936 | if ( s.written && | |
2937 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 2938 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
2939 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
2940 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 2941 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 2942 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 2943 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
2944 | |
2945 | /* Now we might consider reading some blocks, either to check/generate | |
2946 | * parity, or to satisfy requests | |
2947 | * or to load a block that is being partially written. | |
2948 | */ | |
a4456856 DW |
2949 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
2950 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
1fe797e6 | 2951 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc N |
2952 | |
2953 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 | 2954 | if (s.to_write) |
1fe797e6 | 2955 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
2956 | |
2957 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 DW |
2958 | * Any reads will already have been scheduled, so we just see if enough |
2959 | * data is available | |
16a53ecc | 2960 | */ |
a4456856 DW |
2961 | if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state)) |
2962 | handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks); | |
16a53ecc | 2963 | |
a4456856 | 2964 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
2965 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2966 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2967 | } | |
2968 | ||
2969 | /* If the failed drives are just a ReadError, then we might need | |
2970 | * to progress the repair/check process | |
2971 | */ | |
a4456856 DW |
2972 | if (s.failed <= 2 && !conf->mddev->ro) |
2973 | for (i = 0; i < s.failed; i++) { | |
2974 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
2975 | if (test_bit(R5_ReadError, &dev->flags) |
2976 | && !test_bit(R5_LOCKED, &dev->flags) | |
2977 | && test_bit(R5_UPTODATE, &dev->flags) | |
2978 | ) { | |
2979 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
2980 | set_bit(R5_Wantwrite, &dev->flags); | |
2981 | set_bit(R5_ReWrite, &dev->flags); | |
2982 | set_bit(R5_LOCKED, &dev->flags); | |
2983 | } else { | |
2984 | /* let's read it back */ | |
2985 | set_bit(R5_Wantread, &dev->flags); | |
2986 | set_bit(R5_LOCKED, &dev->flags); | |
2987 | } | |
2988 | } | |
2989 | } | |
f416885e | 2990 | |
a4456856 | 2991 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
f416885e N |
2992 | /* Need to write out all blocks after computing P&Q */ |
2993 | sh->disks = conf->raid_disks; | |
112bf897 | 2994 | sh->pd_idx = stripe_to_pdidx(sh->sector, conf, 0); |
f416885e N |
2995 | compute_parity6(sh, RECONSTRUCT_WRITE); |
2996 | for (i = conf->raid_disks ; i-- ; ) { | |
2997 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 2998 | s.locked++; |
f416885e N |
2999 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
3000 | } | |
3001 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 3002 | } else if (s.expanded) { |
f416885e N |
3003 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3004 | atomic_dec(&conf->reshape_stripes); | |
3005 | wake_up(&conf->wait_for_overlap); | |
3006 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3007 | } | |
3008 | ||
0f94e87c | 3009 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3010 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3011 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3012 | |
6bfe0b49 | 3013 | unlock: |
16a53ecc N |
3014 | spin_unlock(&sh->lock); |
3015 | ||
6bfe0b49 DW |
3016 | /* wait for this device to become unblocked */ |
3017 | if (unlikely(blocked_rdev)) | |
3018 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3019 | ||
f0e43bcd | 3020 | ops_run_io(sh, &s); |
16a53ecc | 3021 | |
f0e43bcd | 3022 | return_io(return_bi); |
df10cfbc DW |
3023 | |
3024 | return blocked_rdev == NULL; | |
16a53ecc N |
3025 | } |
3026 | ||
df10cfbc DW |
3027 | /* returns true if the stripe was handled */ |
3028 | static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page) | |
16a53ecc N |
3029 | { |
3030 | if (sh->raid_conf->level == 6) | |
df10cfbc | 3031 | return handle_stripe6(sh, tmp_page); |
16a53ecc | 3032 | else |
df10cfbc | 3033 | return handle_stripe5(sh); |
16a53ecc N |
3034 | } |
3035 | ||
3036 | ||
3037 | ||
3038 | static void raid5_activate_delayed(raid5_conf_t *conf) | |
3039 | { | |
3040 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3041 | while (!list_empty(&conf->delayed_list)) { | |
3042 | struct list_head *l = conf->delayed_list.next; | |
3043 | struct stripe_head *sh; | |
3044 | sh = list_entry(l, struct stripe_head, lru); | |
3045 | list_del_init(l); | |
3046 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3047 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3048 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3049 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3050 | } |
6ed3003c N |
3051 | } else |
3052 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3053 | } |
3054 | ||
3055 | static void activate_bit_delay(raid5_conf_t *conf) | |
3056 | { | |
3057 | /* device_lock is held */ | |
3058 | struct list_head head; | |
3059 | list_add(&head, &conf->bitmap_list); | |
3060 | list_del_init(&conf->bitmap_list); | |
3061 | while (!list_empty(&head)) { | |
3062 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3063 | list_del_init(&sh->lru); | |
3064 | atomic_inc(&sh->count); | |
3065 | __release_stripe(conf, sh); | |
3066 | } | |
3067 | } | |
3068 | ||
3069 | static void unplug_slaves(mddev_t *mddev) | |
3070 | { | |
3071 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3072 | int i; | |
3073 | ||
3074 | rcu_read_lock(); | |
3075 | for (i=0; i<mddev->raid_disks; i++) { | |
3076 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
3077 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3078 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3079 | |
3080 | atomic_inc(&rdev->nr_pending); | |
3081 | rcu_read_unlock(); | |
3082 | ||
2ad8b1ef | 3083 | blk_unplug(r_queue); |
16a53ecc N |
3084 | |
3085 | rdev_dec_pending(rdev, mddev); | |
3086 | rcu_read_lock(); | |
3087 | } | |
3088 | } | |
3089 | rcu_read_unlock(); | |
3090 | } | |
3091 | ||
165125e1 | 3092 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3093 | { |
3094 | mddev_t *mddev = q->queuedata; | |
3095 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3096 | unsigned long flags; | |
3097 | ||
3098 | spin_lock_irqsave(&conf->device_lock, flags); | |
3099 | ||
3100 | if (blk_remove_plug(q)) { | |
3101 | conf->seq_flush++; | |
3102 | raid5_activate_delayed(conf); | |
72626685 | 3103 | } |
1da177e4 LT |
3104 | md_wakeup_thread(mddev->thread); |
3105 | ||
3106 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3107 | ||
3108 | unplug_slaves(mddev); | |
3109 | } | |
3110 | ||
f022b2fd N |
3111 | static int raid5_congested(void *data, int bits) |
3112 | { | |
3113 | mddev_t *mddev = data; | |
3114 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3115 | ||
3116 | /* No difference between reads and writes. Just check | |
3117 | * how busy the stripe_cache is | |
3118 | */ | |
3119 | if (conf->inactive_blocked) | |
3120 | return 1; | |
3121 | if (conf->quiesce) | |
3122 | return 1; | |
3123 | if (list_empty_careful(&conf->inactive_list)) | |
3124 | return 1; | |
3125 | ||
3126 | return 0; | |
3127 | } | |
3128 | ||
23032a0e RBJ |
3129 | /* We want read requests to align with chunks where possible, |
3130 | * but write requests don't need to. | |
3131 | */ | |
cc371e66 AK |
3132 | static int raid5_mergeable_bvec(struct request_queue *q, |
3133 | struct bvec_merge_data *bvm, | |
3134 | struct bio_vec *biovec) | |
23032a0e RBJ |
3135 | { |
3136 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3137 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e RBJ |
3138 | int max; |
3139 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
cc371e66 | 3140 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3141 | |
cc371e66 | 3142 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3143 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3144 | ||
3145 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
3146 | if (max < 0) max = 0; | |
3147 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3148 | return biovec->bv_len; | |
3149 | else | |
3150 | return max; | |
3151 | } | |
3152 | ||
f679623f RBJ |
3153 | |
3154 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3155 | { | |
3156 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
3157 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
3158 | unsigned int bio_sectors = bio->bi_size >> 9; | |
3159 | ||
3160 | return chunk_sectors >= | |
3161 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3162 | } | |
3163 | ||
46031f9a RBJ |
3164 | /* |
3165 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3166 | * later sampled by raid5d. | |
3167 | */ | |
3168 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3169 | { | |
3170 | unsigned long flags; | |
3171 | ||
3172 | spin_lock_irqsave(&conf->device_lock, flags); | |
3173 | ||
3174 | bi->bi_next = conf->retry_read_aligned_list; | |
3175 | conf->retry_read_aligned_list = bi; | |
3176 | ||
3177 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3178 | md_wakeup_thread(conf->mddev->thread); | |
3179 | } | |
3180 | ||
3181 | ||
3182 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3183 | { | |
3184 | struct bio *bi; | |
3185 | ||
3186 | bi = conf->retry_read_aligned; | |
3187 | if (bi) { | |
3188 | conf->retry_read_aligned = NULL; | |
3189 | return bi; | |
3190 | } | |
3191 | bi = conf->retry_read_aligned_list; | |
3192 | if(bi) { | |
387bb173 | 3193 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3194 | bi->bi_next = NULL; |
960e739d JA |
3195 | /* |
3196 | * this sets the active strip count to 1 and the processed | |
3197 | * strip count to zero (upper 8 bits) | |
3198 | */ | |
46031f9a | 3199 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3200 | } |
3201 | ||
3202 | return bi; | |
3203 | } | |
3204 | ||
3205 | ||
f679623f RBJ |
3206 | /* |
3207 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3208 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3209 | * first). | |
3210 | * If the read failed.. | |
3211 | */ | |
6712ecf8 | 3212 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3213 | { |
3214 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3215 | mddev_t *mddev; |
3216 | raid5_conf_t *conf; | |
3217 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3218 | mdk_rdev_t *rdev; | |
3219 | ||
f679623f | 3220 | bio_put(bi); |
46031f9a RBJ |
3221 | |
3222 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
3223 | conf = mddev_to_conf(mddev); | |
3224 | rdev = (void*)raid_bi->bi_next; | |
3225 | raid_bi->bi_next = NULL; | |
3226 | ||
3227 | rdev_dec_pending(rdev, conf->mddev); | |
3228 | ||
3229 | if (!error && uptodate) { | |
6712ecf8 | 3230 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3231 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3232 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3233 | return; |
46031f9a RBJ |
3234 | } |
3235 | ||
3236 | ||
45b4233c | 3237 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3238 | |
3239 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3240 | } |
3241 | ||
387bb173 NB |
3242 | static int bio_fits_rdev(struct bio *bi) |
3243 | { | |
165125e1 | 3244 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 NB |
3245 | |
3246 | if ((bi->bi_size>>9) > q->max_sectors) | |
3247 | return 0; | |
3248 | blk_recount_segments(q, bi); | |
960e739d | 3249 | if (bi->bi_phys_segments > q->max_phys_segments) |
387bb173 NB |
3250 | return 0; |
3251 | ||
3252 | if (q->merge_bvec_fn) | |
3253 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3254 | * just just give up | |
3255 | */ | |
3256 | return 0; | |
3257 | ||
3258 | return 1; | |
3259 | } | |
3260 | ||
3261 | ||
165125e1 | 3262 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3263 | { |
3264 | mddev_t *mddev = q->queuedata; | |
3265 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
f679623f RBJ |
3266 | unsigned int dd_idx, pd_idx; |
3267 | struct bio* align_bi; | |
3268 | mdk_rdev_t *rdev; | |
3269 | ||
3270 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3271 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3272 | return 0; |
3273 | } | |
3274 | /* | |
3275 | * use bio_clone to make a copy of the bio | |
3276 | */ | |
3277 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3278 | if (!align_bi) | |
3279 | return 0; | |
3280 | /* | |
3281 | * set bi_end_io to a new function, and set bi_private to the | |
3282 | * original bio. | |
3283 | */ | |
3284 | align_bi->bi_end_io = raid5_align_endio; | |
3285 | align_bi->bi_private = raid_bio; | |
3286 | /* | |
3287 | * compute position | |
3288 | */ | |
112bf897 N |
3289 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3290 | 0, | |
3291 | &dd_idx, &pd_idx); | |
f679623f RBJ |
3292 | |
3293 | rcu_read_lock(); | |
3294 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3295 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3296 | atomic_inc(&rdev->nr_pending); |
3297 | rcu_read_unlock(); | |
46031f9a RBJ |
3298 | raid_bio->bi_next = (void*)rdev; |
3299 | align_bi->bi_bdev = rdev->bdev; | |
3300 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3301 | align_bi->bi_sector += rdev->data_offset; | |
3302 | ||
387bb173 NB |
3303 | if (!bio_fits_rdev(align_bi)) { |
3304 | /* too big in some way */ | |
3305 | bio_put(align_bi); | |
3306 | rdev_dec_pending(rdev, mddev); | |
3307 | return 0; | |
3308 | } | |
3309 | ||
46031f9a RBJ |
3310 | spin_lock_irq(&conf->device_lock); |
3311 | wait_event_lock_irq(conf->wait_for_stripe, | |
3312 | conf->quiesce == 0, | |
3313 | conf->device_lock, /* nothing */); | |
3314 | atomic_inc(&conf->active_aligned_reads); | |
3315 | spin_unlock_irq(&conf->device_lock); | |
3316 | ||
f679623f RBJ |
3317 | generic_make_request(align_bi); |
3318 | return 1; | |
3319 | } else { | |
3320 | rcu_read_unlock(); | |
46031f9a | 3321 | bio_put(align_bi); |
f679623f RBJ |
3322 | return 0; |
3323 | } | |
3324 | } | |
3325 | ||
8b3e6cdc DW |
3326 | /* __get_priority_stripe - get the next stripe to process |
3327 | * | |
3328 | * Full stripe writes are allowed to pass preread active stripes up until | |
3329 | * the bypass_threshold is exceeded. In general the bypass_count | |
3330 | * increments when the handle_list is handled before the hold_list; however, it | |
3331 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3332 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3333 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3334 | * handle_list. | |
3335 | */ | |
3336 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3337 | { | |
3338 | struct stripe_head *sh; | |
3339 | ||
3340 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3341 | __func__, | |
3342 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3343 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3344 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3345 | ||
3346 | if (!list_empty(&conf->handle_list)) { | |
3347 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3348 | ||
3349 | if (list_empty(&conf->hold_list)) | |
3350 | conf->bypass_count = 0; | |
3351 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3352 | if (conf->hold_list.next == conf->last_hold) | |
3353 | conf->bypass_count++; | |
3354 | else { | |
3355 | conf->last_hold = conf->hold_list.next; | |
3356 | conf->bypass_count -= conf->bypass_threshold; | |
3357 | if (conf->bypass_count < 0) | |
3358 | conf->bypass_count = 0; | |
3359 | } | |
3360 | } | |
3361 | } else if (!list_empty(&conf->hold_list) && | |
3362 | ((conf->bypass_threshold && | |
3363 | conf->bypass_count > conf->bypass_threshold) || | |
3364 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3365 | sh = list_entry(conf->hold_list.next, | |
3366 | typeof(*sh), lru); | |
3367 | conf->bypass_count -= conf->bypass_threshold; | |
3368 | if (conf->bypass_count < 0) | |
3369 | conf->bypass_count = 0; | |
3370 | } else | |
3371 | return NULL; | |
3372 | ||
3373 | list_del_init(&sh->lru); | |
3374 | atomic_inc(&sh->count); | |
3375 | BUG_ON(atomic_read(&sh->count) != 1); | |
3376 | return sh; | |
3377 | } | |
f679623f | 3378 | |
165125e1 | 3379 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3380 | { |
3381 | mddev_t *mddev = q->queuedata; | |
3382 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1da177e4 LT |
3383 | unsigned int dd_idx, pd_idx; |
3384 | sector_t new_sector; | |
3385 | sector_t logical_sector, last_sector; | |
3386 | struct stripe_head *sh; | |
a362357b | 3387 | const int rw = bio_data_dir(bi); |
c9959059 | 3388 | int cpu, remaining; |
1da177e4 | 3389 | |
e5dcdd80 | 3390 | if (unlikely(bio_barrier(bi))) { |
6712ecf8 | 3391 | bio_endio(bi, -EOPNOTSUPP); |
e5dcdd80 N |
3392 | return 0; |
3393 | } | |
3394 | ||
3d310eb7 | 3395 | md_write_start(mddev, bi); |
06d91a5f | 3396 | |
074a7aca TH |
3397 | cpu = part_stat_lock(); |
3398 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
3399 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
3400 | bio_sectors(bi)); | |
3401 | part_stat_unlock(); | |
1da177e4 | 3402 | |
802ba064 | 3403 | if (rw == READ && |
52488615 RBJ |
3404 | mddev->reshape_position == MaxSector && |
3405 | chunk_aligned_read(q,bi)) | |
3406 | return 0; | |
3407 | ||
1da177e4 LT |
3408 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3409 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3410 | bi->bi_next = NULL; | |
3411 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3412 | |
1da177e4 LT |
3413 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3414 | DEFINE_WAIT(w); | |
16a53ecc | 3415 | int disks, data_disks; |
b5663ba4 | 3416 | int previous; |
b578d55f | 3417 | |
7ecaa1e6 | 3418 | retry: |
b5663ba4 | 3419 | previous = 0; |
b578d55f | 3420 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
7ecaa1e6 N |
3421 | if (likely(conf->expand_progress == MaxSector)) |
3422 | disks = conf->raid_disks; | |
3423 | else { | |
df8e7f76 N |
3424 | /* spinlock is needed as expand_progress may be |
3425 | * 64bit on a 32bit platform, and so it might be | |
3426 | * possible to see a half-updated value | |
3427 | * Ofcourse expand_progress could change after | |
3428 | * the lock is dropped, so once we get a reference | |
3429 | * to the stripe that we think it is, we will have | |
3430 | * to check again. | |
3431 | */ | |
7ecaa1e6 N |
3432 | spin_lock_irq(&conf->device_lock); |
3433 | disks = conf->raid_disks; | |
b5663ba4 | 3434 | if (logical_sector >= conf->expand_progress) { |
7ecaa1e6 | 3435 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3436 | previous = 1; |
3437 | } else { | |
b578d55f N |
3438 | if (logical_sector >= conf->expand_lo) { |
3439 | spin_unlock_irq(&conf->device_lock); | |
3440 | schedule(); | |
3441 | goto retry; | |
3442 | } | |
3443 | } | |
7ecaa1e6 N |
3444 | spin_unlock_irq(&conf->device_lock); |
3445 | } | |
16a53ecc N |
3446 | data_disks = disks - conf->max_degraded; |
3447 | ||
112bf897 N |
3448 | new_sector = raid5_compute_sector(conf, logical_sector, |
3449 | previous, | |
3450 | &dd_idx, &pd_idx); | |
45b4233c | 3451 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3452 | (unsigned long long)new_sector, |
3453 | (unsigned long long)logical_sector); | |
3454 | ||
b5663ba4 N |
3455 | sh = get_active_stripe(conf, new_sector, previous, |
3456 | (bi->bi_rw&RWA_MASK)); | |
1da177e4 | 3457 | if (sh) { |
7ecaa1e6 N |
3458 | if (unlikely(conf->expand_progress != MaxSector)) { |
3459 | /* expansion might have moved on while waiting for a | |
df8e7f76 N |
3460 | * stripe, so we must do the range check again. |
3461 | * Expansion could still move past after this | |
3462 | * test, but as we are holding a reference to | |
3463 | * 'sh', we know that if that happens, | |
3464 | * STRIPE_EXPANDING will get set and the expansion | |
3465 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3466 | */ |
3467 | int must_retry = 0; | |
3468 | spin_lock_irq(&conf->device_lock); | |
3469 | if (logical_sector < conf->expand_progress && | |
3470 | disks == conf->previous_raid_disks) | |
3471 | /* mismatch, need to try again */ | |
3472 | must_retry = 1; | |
3473 | spin_unlock_irq(&conf->device_lock); | |
3474 | if (must_retry) { | |
3475 | release_stripe(sh); | |
3476 | goto retry; | |
3477 | } | |
3478 | } | |
e464eafd N |
3479 | /* FIXME what if we get a false positive because these |
3480 | * are being updated. | |
3481 | */ | |
3482 | if (logical_sector >= mddev->suspend_lo && | |
3483 | logical_sector < mddev->suspend_hi) { | |
3484 | release_stripe(sh); | |
3485 | schedule(); | |
3486 | goto retry; | |
3487 | } | |
7ecaa1e6 N |
3488 | |
3489 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
3490 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
3491 | /* Stripe is busy expanding or | |
3492 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3493 | * and wait a while |
3494 | */ | |
3495 | raid5_unplug_device(mddev->queue); | |
3496 | release_stripe(sh); | |
3497 | schedule(); | |
3498 | goto retry; | |
3499 | } | |
3500 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3501 | set_bit(STRIPE_HANDLE, &sh->state); |
3502 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1da177e4 | 3503 | release_stripe(sh); |
1da177e4 LT |
3504 | } else { |
3505 | /* cannot get stripe for read-ahead, just give-up */ | |
3506 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3507 | finish_wait(&conf->wait_for_overlap, &w); | |
3508 | break; | |
3509 | } | |
3510 | ||
3511 | } | |
3512 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3513 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3514 | spin_unlock_irq(&conf->device_lock); |
3515 | if (remaining == 0) { | |
1da177e4 | 3516 | |
16a53ecc | 3517 | if ( rw == WRITE ) |
1da177e4 | 3518 | md_write_end(mddev); |
6712ecf8 | 3519 | |
0e13fe23 | 3520 | bio_endio(bi, 0); |
1da177e4 | 3521 | } |
1da177e4 LT |
3522 | return 0; |
3523 | } | |
3524 | ||
52c03291 | 3525 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3526 | { |
52c03291 N |
3527 | /* reshaping is quite different to recovery/resync so it is |
3528 | * handled quite separately ... here. | |
3529 | * | |
3530 | * On each call to sync_request, we gather one chunk worth of | |
3531 | * destination stripes and flag them as expanding. | |
3532 | * Then we find all the source stripes and request reads. | |
3533 | * As the reads complete, handle_stripe will copy the data | |
3534 | * into the destination stripe and release that stripe. | |
3535 | */ | |
1da177e4 LT |
3536 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
3537 | struct stripe_head *sh; | |
ccfcc3c1 N |
3538 | int pd_idx; |
3539 | sector_t first_sector, last_sector; | |
f416885e N |
3540 | int raid_disks = conf->previous_raid_disks; |
3541 | int data_disks = raid_disks - conf->max_degraded; | |
3542 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3543 | int i; |
3544 | int dd_idx; | |
3545 | sector_t writepos, safepos, gap; | |
3546 | ||
3547 | if (sector_nr == 0 && | |
3548 | conf->expand_progress != 0) { | |
3549 | /* restarting in the middle, skip the initial sectors */ | |
3550 | sector_nr = conf->expand_progress; | |
f416885e | 3551 | sector_div(sector_nr, new_data_disks); |
52c03291 N |
3552 | *skipped = 1; |
3553 | return sector_nr; | |
3554 | } | |
3555 | ||
3556 | /* we update the metadata when there is more than 3Meg | |
3557 | * in the block range (that is rather arbitrary, should | |
3558 | * probably be time based) or when the data about to be | |
3559 | * copied would over-write the source of the data at | |
3560 | * the front of the range. | |
3561 | * i.e. one new_stripe forward from expand_progress new_maps | |
3562 | * to after where expand_lo old_maps to | |
3563 | */ | |
3564 | writepos = conf->expand_progress + | |
f416885e N |
3565 | conf->chunk_size/512*(new_data_disks); |
3566 | sector_div(writepos, new_data_disks); | |
52c03291 | 3567 | safepos = conf->expand_lo; |
f416885e | 3568 | sector_div(safepos, data_disks); |
52c03291 N |
3569 | gap = conf->expand_progress - conf->expand_lo; |
3570 | ||
3571 | if (writepos >= safepos || | |
f416885e | 3572 | gap > (new_data_disks)*3000*2 /*3Meg*/) { |
52c03291 N |
3573 | /* Cannot proceed until we've updated the superblock... */ |
3574 | wait_event(conf->wait_for_overlap, | |
3575 | atomic_read(&conf->reshape_stripes)==0); | |
3576 | mddev->reshape_position = conf->expand_progress; | |
850b2b42 | 3577 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 3578 | md_wakeup_thread(mddev->thread); |
850b2b42 | 3579 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
3580 | kthread_should_stop()); |
3581 | spin_lock_irq(&conf->device_lock); | |
3582 | conf->expand_lo = mddev->reshape_position; | |
3583 | spin_unlock_irq(&conf->device_lock); | |
3584 | wake_up(&conf->wait_for_overlap); | |
3585 | } | |
3586 | ||
3587 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { | |
3588 | int j; | |
3589 | int skipped = 0; | |
b5663ba4 | 3590 | sh = get_active_stripe(conf, sector_nr+i, 0, 0); |
52c03291 N |
3591 | set_bit(STRIPE_EXPANDING, &sh->state); |
3592 | atomic_inc(&conf->reshape_stripes); | |
3593 | /* If any of this stripe is beyond the end of the old | |
3594 | * array, then we need to zero those blocks | |
3595 | */ | |
3596 | for (j=sh->disks; j--;) { | |
3597 | sector_t s; | |
3598 | if (j == sh->pd_idx) | |
3599 | continue; | |
f416885e N |
3600 | if (conf->level == 6 && |
3601 | j == raid6_next_disk(sh->pd_idx, sh->disks)) | |
3602 | continue; | |
52c03291 | 3603 | s = compute_blocknr(sh, j); |
f233ea5c | 3604 | if (s < mddev->array_sectors) { |
52c03291 N |
3605 | skipped = 1; |
3606 | continue; | |
3607 | } | |
3608 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
3609 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
3610 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
3611 | } | |
3612 | if (!skipped) { | |
3613 | set_bit(STRIPE_EXPAND_READY, &sh->state); | |
3614 | set_bit(STRIPE_HANDLE, &sh->state); | |
3615 | } | |
3616 | release_stripe(sh); | |
3617 | } | |
3618 | spin_lock_irq(&conf->device_lock); | |
6d3baf2e | 3619 | conf->expand_progress = (sector_nr + i) * new_data_disks; |
52c03291 N |
3620 | spin_unlock_irq(&conf->device_lock); |
3621 | /* Ok, those stripe are ready. We can start scheduling | |
3622 | * reads on the source stripes. | |
3623 | * The source stripes are determined by mapping the first and last | |
3624 | * block on the destination stripes. | |
3625 | */ | |
52c03291 | 3626 | first_sector = |
112bf897 N |
3627 | raid5_compute_sector(conf, sector_nr*(new_data_disks), |
3628 | 1, &dd_idx, &pd_idx); | |
52c03291 | 3629 | last_sector = |
112bf897 N |
3630 | raid5_compute_sector(conf, ((sector_nr+conf->chunk_size/512) |
3631 | *(new_data_disks) - 1), | |
3632 | 1, &dd_idx, &pd_idx); | |
58c0fed4 AN |
3633 | if (last_sector >= mddev->dev_sectors) |
3634 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 3635 | while (first_sector <= last_sector) { |
b5663ba4 | 3636 | sh = get_active_stripe(conf, first_sector, 1, 0); |
52c03291 N |
3637 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3638 | set_bit(STRIPE_HANDLE, &sh->state); | |
3639 | release_stripe(sh); | |
3640 | first_sector += STRIPE_SECTORS; | |
3641 | } | |
c6207277 N |
3642 | /* If this takes us to the resync_max point where we have to pause, |
3643 | * then we need to write out the superblock. | |
3644 | */ | |
3645 | sector_nr += conf->chunk_size>>9; | |
3646 | if (sector_nr >= mddev->resync_max) { | |
3647 | /* Cannot proceed until we've updated the superblock... */ | |
3648 | wait_event(conf->wait_for_overlap, | |
3649 | atomic_read(&conf->reshape_stripes) == 0); | |
3650 | mddev->reshape_position = conf->expand_progress; | |
3651 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3652 | md_wakeup_thread(mddev->thread); | |
3653 | wait_event(mddev->sb_wait, | |
3654 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
3655 | || kthread_should_stop()); | |
3656 | spin_lock_irq(&conf->device_lock); | |
3657 | conf->expand_lo = mddev->reshape_position; | |
3658 | spin_unlock_irq(&conf->device_lock); | |
3659 | wake_up(&conf->wait_for_overlap); | |
3660 | } | |
52c03291 N |
3661 | return conf->chunk_size>>9; |
3662 | } | |
3663 | ||
3664 | /* FIXME go_faster isn't used */ | |
3665 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
3666 | { | |
3667 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3668 | struct stripe_head *sh; | |
58c0fed4 | 3669 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 3670 | int sync_blocks; |
16a53ecc N |
3671 | int still_degraded = 0; |
3672 | int i; | |
1da177e4 | 3673 | |
72626685 | 3674 | if (sector_nr >= max_sector) { |
1da177e4 LT |
3675 | /* just being told to finish up .. nothing much to do */ |
3676 | unplug_slaves(mddev); | |
29269553 N |
3677 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3678 | end_reshape(conf); | |
3679 | return 0; | |
3680 | } | |
72626685 N |
3681 | |
3682 | if (mddev->curr_resync < max_sector) /* aborted */ | |
3683 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3684 | &sync_blocks, 1); | |
16a53ecc | 3685 | else /* completed sync */ |
72626685 N |
3686 | conf->fullsync = 0; |
3687 | bitmap_close_sync(mddev->bitmap); | |
3688 | ||
1da177e4 LT |
3689 | return 0; |
3690 | } | |
ccfcc3c1 | 3691 | |
52c03291 N |
3692 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3693 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 3694 | |
c6207277 N |
3695 | /* No need to check resync_max as we never do more than one |
3696 | * stripe, and as resync_max will always be on a chunk boundary, | |
3697 | * if the check in md_do_sync didn't fire, there is no chance | |
3698 | * of overstepping resync_max here | |
3699 | */ | |
3700 | ||
16a53ecc | 3701 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
3702 | * to resync, then assert that we are finished, because there is |
3703 | * nothing we can do. | |
3704 | */ | |
3285edf1 | 3705 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 3706 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 3707 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 3708 | *skipped = 1; |
1da177e4 LT |
3709 | return rv; |
3710 | } | |
72626685 | 3711 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 3712 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
3713 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
3714 | /* we can skip this block, and probably more */ | |
3715 | sync_blocks /= STRIPE_SECTORS; | |
3716 | *skipped = 1; | |
3717 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
3718 | } | |
1da177e4 | 3719 | |
b47490c9 N |
3720 | |
3721 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
3722 | ||
b5663ba4 | 3723 | sh = get_active_stripe(conf, sector_nr, 0, 1); |
1da177e4 | 3724 | if (sh == NULL) { |
b5663ba4 | 3725 | sh = get_active_stripe(conf, sector_nr, 0, 0); |
1da177e4 | 3726 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 3727 | * is trying to get access |
1da177e4 | 3728 | */ |
66c006a5 | 3729 | schedule_timeout_uninterruptible(1); |
1da177e4 | 3730 | } |
16a53ecc N |
3731 | /* Need to check if array will still be degraded after recovery/resync |
3732 | * We don't need to check the 'failed' flag as when that gets set, | |
3733 | * recovery aborts. | |
3734 | */ | |
3735 | for (i=0; i<mddev->raid_disks; i++) | |
3736 | if (conf->disks[i].rdev == NULL) | |
3737 | still_degraded = 1; | |
3738 | ||
3739 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
3740 | ||
3741 | spin_lock(&sh->lock); | |
1da177e4 LT |
3742 | set_bit(STRIPE_SYNCING, &sh->state); |
3743 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3744 | spin_unlock(&sh->lock); | |
3745 | ||
df10cfbc DW |
3746 | /* wait for any blocked device to be handled */ |
3747 | while(unlikely(!handle_stripe(sh, NULL))) | |
3748 | ; | |
1da177e4 LT |
3749 | release_stripe(sh); |
3750 | ||
3751 | return STRIPE_SECTORS; | |
3752 | } | |
3753 | ||
46031f9a RBJ |
3754 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
3755 | { | |
3756 | /* We may not be able to submit a whole bio at once as there | |
3757 | * may not be enough stripe_heads available. | |
3758 | * We cannot pre-allocate enough stripe_heads as we may need | |
3759 | * more than exist in the cache (if we allow ever large chunks). | |
3760 | * So we do one stripe head at a time and record in | |
3761 | * ->bi_hw_segments how many have been done. | |
3762 | * | |
3763 | * We *know* that this entire raid_bio is in one chunk, so | |
3764 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
3765 | */ | |
3766 | struct stripe_head *sh; | |
3767 | int dd_idx, pd_idx; | |
3768 | sector_t sector, logical_sector, last_sector; | |
3769 | int scnt = 0; | |
3770 | int remaining; | |
3771 | int handled = 0; | |
3772 | ||
3773 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 N |
3774 | sector = raid5_compute_sector(conf, logical_sector, |
3775 | 0, &dd_idx, &pd_idx); | |
46031f9a RBJ |
3776 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
3777 | ||
3778 | for (; logical_sector < last_sector; | |
387bb173 NB |
3779 | logical_sector += STRIPE_SECTORS, |
3780 | sector += STRIPE_SECTORS, | |
3781 | scnt++) { | |
46031f9a | 3782 | |
960e739d | 3783 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
3784 | /* already done this stripe */ |
3785 | continue; | |
3786 | ||
b5663ba4 | 3787 | sh = get_active_stripe(conf, sector, 0, 1); |
46031f9a RBJ |
3788 | |
3789 | if (!sh) { | |
3790 | /* failed to get a stripe - must wait */ | |
960e739d | 3791 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
3792 | conf->retry_read_aligned = raid_bio; |
3793 | return handled; | |
3794 | } | |
3795 | ||
3796 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
3797 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
3798 | release_stripe(sh); | |
960e739d | 3799 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
3800 | conf->retry_read_aligned = raid_bio; |
3801 | return handled; | |
3802 | } | |
3803 | ||
46031f9a RBJ |
3804 | handle_stripe(sh, NULL); |
3805 | release_stripe(sh); | |
3806 | handled++; | |
3807 | } | |
3808 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3809 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 3810 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
3811 | if (remaining == 0) |
3812 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
3813 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3814 | wake_up(&conf->wait_for_stripe); | |
3815 | return handled; | |
3816 | } | |
3817 | ||
3818 | ||
3819 | ||
1da177e4 LT |
3820 | /* |
3821 | * This is our raid5 kernel thread. | |
3822 | * | |
3823 | * We scan the hash table for stripes which can be handled now. | |
3824 | * During the scan, completed stripes are saved for us by the interrupt | |
3825 | * handler, so that they will not have to wait for our next wakeup. | |
3826 | */ | |
6ed3003c | 3827 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
3828 | { |
3829 | struct stripe_head *sh; | |
3830 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3831 | int handled; | |
3832 | ||
45b4233c | 3833 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
3834 | |
3835 | md_check_recovery(mddev); | |
1da177e4 LT |
3836 | |
3837 | handled = 0; | |
3838 | spin_lock_irq(&conf->device_lock); | |
3839 | while (1) { | |
46031f9a | 3840 | struct bio *bio; |
1da177e4 | 3841 | |
ae3c20cc | 3842 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 3843 | int seq = conf->seq_flush; |
700e432d | 3844 | spin_unlock_irq(&conf->device_lock); |
72626685 | 3845 | bitmap_unplug(mddev->bitmap); |
700e432d | 3846 | spin_lock_irq(&conf->device_lock); |
72626685 N |
3847 | conf->seq_write = seq; |
3848 | activate_bit_delay(conf); | |
3849 | } | |
3850 | ||
46031f9a RBJ |
3851 | while ((bio = remove_bio_from_retry(conf))) { |
3852 | int ok; | |
3853 | spin_unlock_irq(&conf->device_lock); | |
3854 | ok = retry_aligned_read(conf, bio); | |
3855 | spin_lock_irq(&conf->device_lock); | |
3856 | if (!ok) | |
3857 | break; | |
3858 | handled++; | |
3859 | } | |
3860 | ||
8b3e6cdc DW |
3861 | sh = __get_priority_stripe(conf); |
3862 | ||
c9f21aaf | 3863 | if (!sh) |
1da177e4 | 3864 | break; |
1da177e4 LT |
3865 | spin_unlock_irq(&conf->device_lock); |
3866 | ||
3867 | handled++; | |
16a53ecc | 3868 | handle_stripe(sh, conf->spare_page); |
1da177e4 LT |
3869 | release_stripe(sh); |
3870 | ||
3871 | spin_lock_irq(&conf->device_lock); | |
3872 | } | |
45b4233c | 3873 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
3874 | |
3875 | spin_unlock_irq(&conf->device_lock); | |
3876 | ||
c9f21aaf | 3877 | async_tx_issue_pending_all(); |
1da177e4 LT |
3878 | unplug_slaves(mddev); |
3879 | ||
45b4233c | 3880 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
3881 | } |
3882 | ||
3f294f4f | 3883 | static ssize_t |
007583c9 | 3884 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 3885 | { |
007583c9 | 3886 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
3887 | if (conf) |
3888 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
3889 | else | |
3890 | return 0; | |
3f294f4f N |
3891 | } |
3892 | ||
3893 | static ssize_t | |
007583c9 | 3894 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 3895 | { |
007583c9 | 3896 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4ef197d8 | 3897 | unsigned long new; |
b5470dc5 DW |
3898 | int err; |
3899 | ||
3f294f4f N |
3900 | if (len >= PAGE_SIZE) |
3901 | return -EINVAL; | |
96de1e66 N |
3902 | if (!conf) |
3903 | return -ENODEV; | |
3f294f4f | 3904 | |
4ef197d8 | 3905 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
3906 | return -EINVAL; |
3907 | if (new <= 16 || new > 32768) | |
3908 | return -EINVAL; | |
3909 | while (new < conf->max_nr_stripes) { | |
3910 | if (drop_one_stripe(conf)) | |
3911 | conf->max_nr_stripes--; | |
3912 | else | |
3913 | break; | |
3914 | } | |
b5470dc5 DW |
3915 | err = md_allow_write(mddev); |
3916 | if (err) | |
3917 | return err; | |
3f294f4f N |
3918 | while (new > conf->max_nr_stripes) { |
3919 | if (grow_one_stripe(conf)) | |
3920 | conf->max_nr_stripes++; | |
3921 | else break; | |
3922 | } | |
3923 | return len; | |
3924 | } | |
007583c9 | 3925 | |
96de1e66 N |
3926 | static struct md_sysfs_entry |
3927 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
3928 | raid5_show_stripe_cache_size, | |
3929 | raid5_store_stripe_cache_size); | |
3f294f4f | 3930 | |
8b3e6cdc DW |
3931 | static ssize_t |
3932 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
3933 | { | |
3934 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3935 | if (conf) | |
3936 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
3937 | else | |
3938 | return 0; | |
3939 | } | |
3940 | ||
3941 | static ssize_t | |
3942 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
3943 | { | |
3944 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4ef197d8 | 3945 | unsigned long new; |
8b3e6cdc DW |
3946 | if (len >= PAGE_SIZE) |
3947 | return -EINVAL; | |
3948 | if (!conf) | |
3949 | return -ENODEV; | |
3950 | ||
4ef197d8 | 3951 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 3952 | return -EINVAL; |
4ef197d8 | 3953 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
3954 | return -EINVAL; |
3955 | conf->bypass_threshold = new; | |
3956 | return len; | |
3957 | } | |
3958 | ||
3959 | static struct md_sysfs_entry | |
3960 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
3961 | S_IRUGO | S_IWUSR, | |
3962 | raid5_show_preread_threshold, | |
3963 | raid5_store_preread_threshold); | |
3964 | ||
3f294f4f | 3965 | static ssize_t |
96de1e66 | 3966 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 3967 | { |
007583c9 | 3968 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
3969 | if (conf) |
3970 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
3971 | else | |
3972 | return 0; | |
3f294f4f N |
3973 | } |
3974 | ||
96de1e66 N |
3975 | static struct md_sysfs_entry |
3976 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 3977 | |
007583c9 | 3978 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
3979 | &raid5_stripecache_size.attr, |
3980 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 3981 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
3982 | NULL, |
3983 | }; | |
007583c9 N |
3984 | static struct attribute_group raid5_attrs_group = { |
3985 | .name = NULL, | |
3986 | .attrs = raid5_attrs, | |
3f294f4f N |
3987 | }; |
3988 | ||
72626685 | 3989 | static int run(mddev_t *mddev) |
1da177e4 LT |
3990 | { |
3991 | raid5_conf_t *conf; | |
3992 | int raid_disk, memory; | |
3993 | mdk_rdev_t *rdev; | |
3994 | struct disk_info *disk; | |
02c2de8c | 3995 | int working_disks = 0; |
1da177e4 | 3996 | |
16a53ecc N |
3997 | if (mddev->level != 5 && mddev->level != 4 && mddev->level != 6) { |
3998 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", | |
14f8d26b | 3999 | mdname(mddev), mddev->level); |
1da177e4 LT |
4000 | return -EIO; |
4001 | } | |
4002 | ||
4bbf3771 N |
4003 | if (mddev->chunk_size < PAGE_SIZE) { |
4004 | printk(KERN_ERR "md/raid5: chunk_size must be at least " | |
4005 | "PAGE_SIZE but %d < %ld\n", | |
4006 | mddev->chunk_size, PAGE_SIZE); | |
4007 | return -EINVAL; | |
4008 | } | |
4009 | ||
f6705578 N |
4010 | if (mddev->reshape_position != MaxSector) { |
4011 | /* Check that we can continue the reshape. | |
4012 | * Currently only disks can change, it must | |
4013 | * increase, and we must be past the point where | |
4014 | * a stripe over-writes itself | |
4015 | */ | |
4016 | sector_t here_new, here_old; | |
4017 | int old_disks; | |
f416885e | 4018 | int max_degraded = (mddev->level == 5 ? 1 : 2); |
f6705578 N |
4019 | |
4020 | if (mddev->new_level != mddev->level || | |
4021 | mddev->new_layout != mddev->layout || | |
4022 | mddev->new_chunk != mddev->chunk_size) { | |
f416885e N |
4023 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4024 | "required - aborting.\n", | |
f6705578 N |
4025 | mdname(mddev)); |
4026 | return -EINVAL; | |
4027 | } | |
4028 | if (mddev->delta_disks <= 0) { | |
f416885e N |
4029 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
4030 | "(reduce disks) required - aborting.\n", | |
f6705578 N |
4031 | mdname(mddev)); |
4032 | return -EINVAL; | |
4033 | } | |
4034 | old_disks = mddev->raid_disks - mddev->delta_disks; | |
4035 | /* reshape_position must be on a new-stripe boundary, and one | |
f416885e N |
4036 | * further up in new geometry must map after here in old |
4037 | * geometry. | |
f6705578 N |
4038 | */ |
4039 | here_new = mddev->reshape_position; | |
f416885e N |
4040 | if (sector_div(here_new, (mddev->chunk_size>>9)* |
4041 | (mddev->raid_disks - max_degraded))) { | |
4042 | printk(KERN_ERR "raid5: reshape_position not " | |
4043 | "on a stripe boundary\n"); | |
f6705578 N |
4044 | return -EINVAL; |
4045 | } | |
4046 | /* here_new is the stripe we will write to */ | |
4047 | here_old = mddev->reshape_position; | |
f416885e N |
4048 | sector_div(here_old, (mddev->chunk_size>>9)* |
4049 | (old_disks-max_degraded)); | |
4050 | /* here_old is the first stripe that we might need to read | |
4051 | * from */ | |
f6705578 N |
4052 | if (here_new >= here_old) { |
4053 | /* Reading from the same stripe as writing to - bad */ | |
f416885e N |
4054 | printk(KERN_ERR "raid5: reshape_position too early for " |
4055 | "auto-recovery - aborting.\n"); | |
f6705578 N |
4056 | return -EINVAL; |
4057 | } | |
4058 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4059 | /* OK, we should be able to continue; */ | |
4060 | } | |
4061 | ||
4062 | ||
b55e6bfc | 4063 | mddev->private = kzalloc(sizeof (raid5_conf_t), GFP_KERNEL); |
1da177e4 LT |
4064 | if ((conf = mddev->private) == NULL) |
4065 | goto abort; | |
f6705578 N |
4066 | if (mddev->reshape_position == MaxSector) { |
4067 | conf->previous_raid_disks = conf->raid_disks = mddev->raid_disks; | |
4068 | } else { | |
4069 | conf->raid_disks = mddev->raid_disks; | |
4070 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; | |
4071 | } | |
4072 | ||
4073 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
4074 | GFP_KERNEL); |
4075 | if (!conf->disks) | |
4076 | goto abort; | |
9ffae0cf | 4077 | |
1da177e4 LT |
4078 | conf->mddev = mddev; |
4079 | ||
fccddba0 | 4080 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4081 | goto abort; |
1da177e4 | 4082 | |
16a53ecc N |
4083 | if (mddev->level == 6) { |
4084 | conf->spare_page = alloc_page(GFP_KERNEL); | |
4085 | if (!conf->spare_page) | |
4086 | goto abort; | |
4087 | } | |
1da177e4 | 4088 | spin_lock_init(&conf->device_lock); |
e7e72bf6 | 4089 | mddev->queue->queue_lock = &conf->device_lock; |
1da177e4 LT |
4090 | init_waitqueue_head(&conf->wait_for_stripe); |
4091 | init_waitqueue_head(&conf->wait_for_overlap); | |
4092 | INIT_LIST_HEAD(&conf->handle_list); | |
8b3e6cdc | 4093 | INIT_LIST_HEAD(&conf->hold_list); |
1da177e4 | 4094 | INIT_LIST_HEAD(&conf->delayed_list); |
72626685 | 4095 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
4096 | INIT_LIST_HEAD(&conf->inactive_list); |
4097 | atomic_set(&conf->active_stripes, 0); | |
4098 | atomic_set(&conf->preread_active_stripes, 0); | |
46031f9a | 4099 | atomic_set(&conf->active_aligned_reads, 0); |
8b3e6cdc | 4100 | conf->bypass_threshold = BYPASS_THRESHOLD; |
1da177e4 | 4101 | |
45b4233c | 4102 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4103 | |
159ec1fc | 4104 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4105 | raid_disk = rdev->raid_disk; |
f6705578 | 4106 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
4107 | || raid_disk < 0) |
4108 | continue; | |
4109 | disk = conf->disks + raid_disk; | |
4110 | ||
4111 | disk->rdev = rdev; | |
4112 | ||
b2d444d7 | 4113 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4114 | char b[BDEVNAME_SIZE]; |
4115 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4116 | " disk %d\n", bdevname(rdev->bdev,b), | |
4117 | raid_disk); | |
02c2de8c | 4118 | working_disks++; |
8c2e870a NB |
4119 | } else |
4120 | /* Cannot rely on bitmap to complete recovery */ | |
4121 | conf->fullsync = 1; | |
1da177e4 LT |
4122 | } |
4123 | ||
1da177e4 | 4124 | /* |
16a53ecc | 4125 | * 0 for a fully functional array, 1 or 2 for a degraded array. |
1da177e4 | 4126 | */ |
02c2de8c | 4127 | mddev->degraded = conf->raid_disks - working_disks; |
1da177e4 LT |
4128 | conf->mddev = mddev; |
4129 | conf->chunk_size = mddev->chunk_size; | |
4130 | conf->level = mddev->level; | |
16a53ecc N |
4131 | if (conf->level == 6) |
4132 | conf->max_degraded = 2; | |
4133 | else | |
4134 | conf->max_degraded = 1; | |
1da177e4 LT |
4135 | conf->algorithm = mddev->layout; |
4136 | conf->max_nr_stripes = NR_STRIPES; | |
f6705578 | 4137 | conf->expand_progress = mddev->reshape_position; |
1da177e4 LT |
4138 | |
4139 | /* device size must be a multiple of chunk size */ | |
58c0fed4 AN |
4140 | mddev->dev_sectors &= ~(mddev->chunk_size / 512 - 1); |
4141 | mddev->resync_max_sectors = mddev->dev_sectors; | |
1da177e4 | 4142 | |
16a53ecc N |
4143 | if (conf->level == 6 && conf->raid_disks < 4) { |
4144 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4145 | mdname(mddev), conf->raid_disks); | |
4146 | goto abort; | |
4147 | } | |
1da177e4 LT |
4148 | if (!conf->chunk_size || conf->chunk_size % 4) { |
4149 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
4150 | conf->chunk_size, mdname(mddev)); | |
4151 | goto abort; | |
4152 | } | |
4153 | if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) { | |
4154 | printk(KERN_ERR | |
4155 | "raid5: unsupported parity algorithm %d for %s\n", | |
4156 | conf->algorithm, mdname(mddev)); | |
4157 | goto abort; | |
4158 | } | |
16a53ecc | 4159 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
4160 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
4161 | " (%d/%d failed)\n", | |
02c2de8c | 4162 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4163 | goto abort; |
4164 | } | |
4165 | ||
16a53ecc | 4166 | if (mddev->degraded > 0 && |
1da177e4 | 4167 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4168 | if (mddev->ok_start_degraded) |
4169 | printk(KERN_WARNING | |
4170 | "raid5: starting dirty degraded array: %s" | |
4171 | "- data corruption possible.\n", | |
4172 | mdname(mddev)); | |
4173 | else { | |
4174 | printk(KERN_ERR | |
4175 | "raid5: cannot start dirty degraded array for %s\n", | |
4176 | mdname(mddev)); | |
4177 | goto abort; | |
4178 | } | |
1da177e4 LT |
4179 | } |
4180 | ||
4181 | { | |
4182 | mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5"); | |
4183 | if (!mddev->thread) { | |
4184 | printk(KERN_ERR | |
4185 | "raid5: couldn't allocate thread for %s\n", | |
4186 | mdname(mddev)); | |
4187 | goto abort; | |
4188 | } | |
4189 | } | |
5036805b | 4190 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
1da177e4 LT |
4191 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4192 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
4193 | printk(KERN_ERR | |
4194 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4195 | shrink_stripes(conf); | |
4196 | md_unregister_thread(mddev->thread); | |
4197 | goto abort; | |
4198 | } else | |
4199 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4200 | memory, mdname(mddev)); | |
4201 | ||
4202 | if (mddev->degraded == 0) | |
4203 | printk("raid5: raid level %d set %s active with %d out of %d" | |
4204 | " devices, algorithm %d\n", conf->level, mdname(mddev), | |
4205 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
4206 | conf->algorithm); | |
4207 | else | |
4208 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
4209 | " out of %d devices, algorithm %d\n", conf->level, | |
4210 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
4211 | mddev->raid_disks, conf->algorithm); | |
4212 | ||
4213 | print_raid5_conf(conf); | |
4214 | ||
f6705578 N |
4215 | if (conf->expand_progress != MaxSector) { |
4216 | printk("...ok start reshape thread\n"); | |
b578d55f | 4217 | conf->expand_lo = conf->expand_progress; |
f6705578 N |
4218 | atomic_set(&conf->reshape_stripes, 0); |
4219 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4220 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4221 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4222 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4223 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4224 | "%s_reshape"); | |
f6705578 N |
4225 | } |
4226 | ||
1da177e4 | 4227 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 4228 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
4229 | */ |
4230 | { | |
16a53ecc N |
4231 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
4232 | int stripe = data_disks * | |
8932c2e0 | 4233 | (mddev->chunk_size / PAGE_SIZE); |
1da177e4 LT |
4234 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
4235 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4236 | } | |
4237 | ||
4238 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
4239 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
4240 | printk(KERN_WARNING | |
4241 | "raid5: failed to create sysfs attributes for %s\n", | |
4242 | mdname(mddev)); | |
7a5febe9 N |
4243 | |
4244 | mddev->queue->unplug_fn = raid5_unplug_device; | |
f022b2fd | 4245 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 4246 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 4247 | |
58c0fed4 AN |
4248 | mddev->array_sectors = mddev->dev_sectors * |
4249 | (conf->previous_raid_disks - conf->max_degraded); | |
7a5febe9 | 4250 | |
23032a0e RBJ |
4251 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
4252 | ||
1da177e4 LT |
4253 | return 0; |
4254 | abort: | |
4255 | if (conf) { | |
4256 | print_raid5_conf(conf); | |
16a53ecc | 4257 | safe_put_page(conf->spare_page); |
b55e6bfc | 4258 | kfree(conf->disks); |
fccddba0 | 4259 | kfree(conf->stripe_hashtbl); |
1da177e4 LT |
4260 | kfree(conf); |
4261 | } | |
4262 | mddev->private = NULL; | |
4263 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
4264 | return -EIO; | |
4265 | } | |
4266 | ||
4267 | ||
4268 | ||
3f294f4f | 4269 | static int stop(mddev_t *mddev) |
1da177e4 LT |
4270 | { |
4271 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4272 | ||
4273 | md_unregister_thread(mddev->thread); | |
4274 | mddev->thread = NULL; | |
4275 | shrink_stripes(conf); | |
fccddba0 | 4276 | kfree(conf->stripe_hashtbl); |
041ae52e | 4277 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4278 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 4279 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
b55e6bfc | 4280 | kfree(conf->disks); |
96de1e66 | 4281 | kfree(conf); |
1da177e4 LT |
4282 | mddev->private = NULL; |
4283 | return 0; | |
4284 | } | |
4285 | ||
45b4233c | 4286 | #ifdef DEBUG |
d710e138 | 4287 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
4288 | { |
4289 | int i; | |
4290 | ||
16a53ecc N |
4291 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
4292 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
4293 | seq_printf(seq, "sh %llu, count %d.\n", | |
4294 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
4295 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 4296 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
4297 | seq_printf(seq, "(cache%d: %p %ld) ", |
4298 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 4299 | } |
16a53ecc | 4300 | seq_printf(seq, "\n"); |
1da177e4 LT |
4301 | } |
4302 | ||
d710e138 | 4303 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
4304 | { |
4305 | struct stripe_head *sh; | |
fccddba0 | 4306 | struct hlist_node *hn; |
1da177e4 LT |
4307 | int i; |
4308 | ||
4309 | spin_lock_irq(&conf->device_lock); | |
4310 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 4311 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
4312 | if (sh->raid_conf != conf) |
4313 | continue; | |
16a53ecc | 4314 | print_sh(seq, sh); |
1da177e4 LT |
4315 | } |
4316 | } | |
4317 | spin_unlock_irq(&conf->device_lock); | |
4318 | } | |
4319 | #endif | |
4320 | ||
d710e138 | 4321 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
4322 | { |
4323 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4324 | int i; | |
4325 | ||
4326 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
02c2de8c | 4327 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
4328 | for (i = 0; i < conf->raid_disks; i++) |
4329 | seq_printf (seq, "%s", | |
4330 | conf->disks[i].rdev && | |
b2d444d7 | 4331 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 4332 | seq_printf (seq, "]"); |
45b4233c | 4333 | #ifdef DEBUG |
16a53ecc N |
4334 | seq_printf (seq, "\n"); |
4335 | printall(seq, conf); | |
1da177e4 LT |
4336 | #endif |
4337 | } | |
4338 | ||
4339 | static void print_raid5_conf (raid5_conf_t *conf) | |
4340 | { | |
4341 | int i; | |
4342 | struct disk_info *tmp; | |
4343 | ||
4344 | printk("RAID5 conf printout:\n"); | |
4345 | if (!conf) { | |
4346 | printk("(conf==NULL)\n"); | |
4347 | return; | |
4348 | } | |
02c2de8c N |
4349 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
4350 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
4351 | |
4352 | for (i = 0; i < conf->raid_disks; i++) { | |
4353 | char b[BDEVNAME_SIZE]; | |
4354 | tmp = conf->disks + i; | |
4355 | if (tmp->rdev) | |
4356 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 4357 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
4358 | bdevname(tmp->rdev->bdev,b)); |
4359 | } | |
4360 | } | |
4361 | ||
4362 | static int raid5_spare_active(mddev_t *mddev) | |
4363 | { | |
4364 | int i; | |
4365 | raid5_conf_t *conf = mddev->private; | |
4366 | struct disk_info *tmp; | |
4367 | ||
4368 | for (i = 0; i < conf->raid_disks; i++) { | |
4369 | tmp = conf->disks + i; | |
4370 | if (tmp->rdev | |
b2d444d7 | 4371 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
4372 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
4373 | unsigned long flags; | |
4374 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 4375 | mddev->degraded--; |
c04be0aa | 4376 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
4377 | } |
4378 | } | |
4379 | print_raid5_conf(conf); | |
4380 | return 0; | |
4381 | } | |
4382 | ||
4383 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
4384 | { | |
4385 | raid5_conf_t *conf = mddev->private; | |
4386 | int err = 0; | |
4387 | mdk_rdev_t *rdev; | |
4388 | struct disk_info *p = conf->disks + number; | |
4389 | ||
4390 | print_raid5_conf(conf); | |
4391 | rdev = p->rdev; | |
4392 | if (rdev) { | |
b2d444d7 | 4393 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
4394 | atomic_read(&rdev->nr_pending)) { |
4395 | err = -EBUSY; | |
4396 | goto abort; | |
4397 | } | |
dfc70645 N |
4398 | /* Only remove non-faulty devices if recovery |
4399 | * isn't possible. | |
4400 | */ | |
4401 | if (!test_bit(Faulty, &rdev->flags) && | |
4402 | mddev->degraded <= conf->max_degraded) { | |
4403 | err = -EBUSY; | |
4404 | goto abort; | |
4405 | } | |
1da177e4 | 4406 | p->rdev = NULL; |
fbd568a3 | 4407 | synchronize_rcu(); |
1da177e4 LT |
4408 | if (atomic_read(&rdev->nr_pending)) { |
4409 | /* lost the race, try later */ | |
4410 | err = -EBUSY; | |
4411 | p->rdev = rdev; | |
4412 | } | |
4413 | } | |
4414 | abort: | |
4415 | ||
4416 | print_raid5_conf(conf); | |
4417 | return err; | |
4418 | } | |
4419 | ||
4420 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
4421 | { | |
4422 | raid5_conf_t *conf = mddev->private; | |
199050ea | 4423 | int err = -EEXIST; |
1da177e4 LT |
4424 | int disk; |
4425 | struct disk_info *p; | |
6c2fce2e NB |
4426 | int first = 0; |
4427 | int last = conf->raid_disks - 1; | |
1da177e4 | 4428 | |
16a53ecc | 4429 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 4430 | /* no point adding a device */ |
199050ea | 4431 | return -EINVAL; |
1da177e4 | 4432 | |
6c2fce2e NB |
4433 | if (rdev->raid_disk >= 0) |
4434 | first = last = rdev->raid_disk; | |
1da177e4 LT |
4435 | |
4436 | /* | |
16a53ecc N |
4437 | * find the disk ... but prefer rdev->saved_raid_disk |
4438 | * if possible. | |
1da177e4 | 4439 | */ |
16a53ecc | 4440 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 4441 | rdev->saved_raid_disk >= first && |
16a53ecc N |
4442 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
4443 | disk = rdev->saved_raid_disk; | |
4444 | else | |
6c2fce2e NB |
4445 | disk = first; |
4446 | for ( ; disk <= last ; disk++) | |
1da177e4 | 4447 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 4448 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4449 | rdev->raid_disk = disk; |
199050ea | 4450 | err = 0; |
72626685 N |
4451 | if (rdev->saved_raid_disk != disk) |
4452 | conf->fullsync = 1; | |
d6065f7b | 4453 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
4454 | break; |
4455 | } | |
4456 | print_raid5_conf(conf); | |
199050ea | 4457 | return err; |
1da177e4 LT |
4458 | } |
4459 | ||
4460 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
4461 | { | |
4462 | /* no resync is happening, and there is enough space | |
4463 | * on all devices, so we can resize. | |
4464 | * We need to make sure resync covers any new space. | |
4465 | * If the array is shrinking we should possibly wait until | |
4466 | * any io in the removed space completes, but it hardly seems | |
4467 | * worth it. | |
4468 | */ | |
16a53ecc N |
4469 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4470 | ||
1da177e4 | 4471 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); |
f233ea5c AN |
4472 | mddev->array_sectors = sectors * (mddev->raid_disks |
4473 | - conf->max_degraded); | |
4474 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
44ce6294 | 4475 | mddev->changed = 1; |
58c0fed4 AN |
4476 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
4477 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
4478 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
4479 | } | |
58c0fed4 | 4480 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 4481 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
4482 | return 0; |
4483 | } | |
4484 | ||
29269553 | 4485 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f | 4486 | static int raid5_check_reshape(mddev_t *mddev) |
29269553 N |
4487 | { |
4488 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4489 | int err; | |
29269553 | 4490 | |
63c70c4f N |
4491 | if (mddev->delta_disks < 0 || |
4492 | mddev->new_level != mddev->level) | |
4493 | return -EINVAL; /* Cannot shrink array or change level yet */ | |
4494 | if (mddev->delta_disks == 0) | |
29269553 | 4495 | return 0; /* nothing to do */ |
dba034ee N |
4496 | if (mddev->bitmap) |
4497 | /* Cannot grow a bitmap yet */ | |
4498 | return -EBUSY; | |
29269553 N |
4499 | |
4500 | /* Can only proceed if there are plenty of stripe_heads. | |
4501 | * We need a minimum of one full stripe,, and for sensible progress | |
4502 | * it is best to have about 4 times that. | |
4503 | * If we require 4 times, then the default 256 4K stripe_heads will | |
4504 | * allow for chunk sizes up to 256K, which is probably OK. | |
4505 | * If the chunk size is greater, user-space should request more | |
4506 | * stripe_heads first. | |
4507 | */ | |
63c70c4f N |
4508 | if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes || |
4509 | (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) { | |
29269553 N |
4510 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", |
4511 | (mddev->chunk_size / STRIPE_SIZE)*4); | |
4512 | return -ENOSPC; | |
4513 | } | |
4514 | ||
63c70c4f N |
4515 | err = resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
4516 | if (err) | |
4517 | return err; | |
4518 | ||
b4c4c7b8 N |
4519 | if (mddev->degraded > conf->max_degraded) |
4520 | return -EINVAL; | |
63c70c4f N |
4521 | /* looks like we might be able to manage this */ |
4522 | return 0; | |
4523 | } | |
4524 | ||
4525 | static int raid5_start_reshape(mddev_t *mddev) | |
4526 | { | |
4527 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4528 | mdk_rdev_t *rdev; | |
63c70c4f N |
4529 | int spares = 0; |
4530 | int added_devices = 0; | |
c04be0aa | 4531 | unsigned long flags; |
63c70c4f | 4532 | |
f416885e | 4533 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
4534 | return -EBUSY; |
4535 | ||
159ec1fc | 4536 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4537 | if (rdev->raid_disk < 0 && |
4538 | !test_bit(Faulty, &rdev->flags)) | |
4539 | spares++; | |
63c70c4f | 4540 | |
f416885e | 4541 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
4542 | /* Not enough devices even to make a degraded array |
4543 | * of that size | |
4544 | */ | |
4545 | return -EINVAL; | |
4546 | ||
f6705578 | 4547 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
4548 | spin_lock_irq(&conf->device_lock); |
4549 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 4550 | conf->raid_disks += mddev->delta_disks; |
29269553 | 4551 | conf->expand_progress = 0; |
b578d55f | 4552 | conf->expand_lo = 0; |
29269553 N |
4553 | spin_unlock_irq(&conf->device_lock); |
4554 | ||
4555 | /* Add some new drives, as many as will fit. | |
4556 | * We know there are enough to make the newly sized array work. | |
4557 | */ | |
159ec1fc | 4558 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4559 | if (rdev->raid_disk < 0 && |
4560 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 4561 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 N |
4562 | char nm[20]; |
4563 | set_bit(In_sync, &rdev->flags); | |
29269553 | 4564 | added_devices++; |
5fd6c1dc | 4565 | rdev->recovery_offset = 0; |
29269553 | 4566 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
4567 | if (sysfs_create_link(&mddev->kobj, |
4568 | &rdev->kobj, nm)) | |
4569 | printk(KERN_WARNING | |
4570 | "raid5: failed to create " | |
4571 | " link %s for %s\n", | |
4572 | nm, mdname(mddev)); | |
29269553 N |
4573 | } else |
4574 | break; | |
4575 | } | |
4576 | ||
c04be0aa | 4577 | spin_lock_irqsave(&conf->device_lock, flags); |
63c70c4f | 4578 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) - added_devices; |
c04be0aa | 4579 | spin_unlock_irqrestore(&conf->device_lock, flags); |
63c70c4f | 4580 | mddev->raid_disks = conf->raid_disks; |
f6705578 | 4581 | mddev->reshape_position = 0; |
850b2b42 | 4582 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 4583 | |
29269553 N |
4584 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
4585 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4586 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4587 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4588 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4589 | "%s_reshape"); | |
4590 | if (!mddev->sync_thread) { | |
4591 | mddev->recovery = 0; | |
4592 | spin_lock_irq(&conf->device_lock); | |
4593 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
4594 | conf->expand_progress = MaxSector; | |
4595 | spin_unlock_irq(&conf->device_lock); | |
4596 | return -EAGAIN; | |
4597 | } | |
4598 | md_wakeup_thread(mddev->sync_thread); | |
4599 | md_new_event(mddev); | |
4600 | return 0; | |
4601 | } | |
4602 | #endif | |
4603 | ||
4604 | static void end_reshape(raid5_conf_t *conf) | |
4605 | { | |
4606 | struct block_device *bdev; | |
4607 | ||
f6705578 | 4608 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
58c0fed4 | 4609 | conf->mddev->array_sectors = conf->mddev->dev_sectors * |
f416885e | 4610 | (conf->raid_disks - conf->max_degraded); |
f233ea5c | 4611 | set_capacity(conf->mddev->gendisk, conf->mddev->array_sectors); |
44ce6294 | 4612 | conf->mddev->changed = 1; |
f6705578 N |
4613 | |
4614 | bdev = bdget_disk(conf->mddev->gendisk, 0); | |
4615 | if (bdev) { | |
4616 | mutex_lock(&bdev->bd_inode->i_mutex); | |
f233ea5c AN |
4617 | i_size_write(bdev->bd_inode, |
4618 | (loff_t)conf->mddev->array_sectors << 9); | |
f6705578 N |
4619 | mutex_unlock(&bdev->bd_inode->i_mutex); |
4620 | bdput(bdev); | |
4621 | } | |
4622 | spin_lock_irq(&conf->device_lock); | |
4623 | conf->expand_progress = MaxSector; | |
4624 | spin_unlock_irq(&conf->device_lock); | |
4625 | conf->mddev->reshape_position = MaxSector; | |
16a53ecc N |
4626 | |
4627 | /* read-ahead size must cover two whole stripes, which is | |
4628 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4629 | */ | |
4630 | { | |
4631 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
4632 | int stripe = data_disks * | |
4633 | (conf->mddev->chunk_size / PAGE_SIZE); | |
4634 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4635 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4636 | } | |
29269553 | 4637 | } |
29269553 N |
4638 | } |
4639 | ||
72626685 N |
4640 | static void raid5_quiesce(mddev_t *mddev, int state) |
4641 | { | |
4642 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4643 | ||
4644 | switch(state) { | |
e464eafd N |
4645 | case 2: /* resume for a suspend */ |
4646 | wake_up(&conf->wait_for_overlap); | |
4647 | break; | |
4648 | ||
72626685 N |
4649 | case 1: /* stop all writes */ |
4650 | spin_lock_irq(&conf->device_lock); | |
4651 | conf->quiesce = 1; | |
4652 | wait_event_lock_irq(conf->wait_for_stripe, | |
46031f9a RBJ |
4653 | atomic_read(&conf->active_stripes) == 0 && |
4654 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 N |
4655 | conf->device_lock, /* nothing */); |
4656 | spin_unlock_irq(&conf->device_lock); | |
4657 | break; | |
4658 | ||
4659 | case 0: /* re-enable writes */ | |
4660 | spin_lock_irq(&conf->device_lock); | |
4661 | conf->quiesce = 0; | |
4662 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 4663 | wake_up(&conf->wait_for_overlap); |
72626685 N |
4664 | spin_unlock_irq(&conf->device_lock); |
4665 | break; | |
4666 | } | |
72626685 | 4667 | } |
b15c2e57 | 4668 | |
16a53ecc N |
4669 | static struct mdk_personality raid6_personality = |
4670 | { | |
4671 | .name = "raid6", | |
4672 | .level = 6, | |
4673 | .owner = THIS_MODULE, | |
4674 | .make_request = make_request, | |
4675 | .run = run, | |
4676 | .stop = stop, | |
4677 | .status = status, | |
4678 | .error_handler = error, | |
4679 | .hot_add_disk = raid5_add_disk, | |
4680 | .hot_remove_disk= raid5_remove_disk, | |
4681 | .spare_active = raid5_spare_active, | |
4682 | .sync_request = sync_request, | |
4683 | .resize = raid5_resize, | |
f416885e N |
4684 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4685 | .check_reshape = raid5_check_reshape, | |
4686 | .start_reshape = raid5_start_reshape, | |
4687 | #endif | |
16a53ecc N |
4688 | .quiesce = raid5_quiesce, |
4689 | }; | |
2604b703 | 4690 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
4691 | { |
4692 | .name = "raid5", | |
2604b703 | 4693 | .level = 5, |
1da177e4 LT |
4694 | .owner = THIS_MODULE, |
4695 | .make_request = make_request, | |
4696 | .run = run, | |
4697 | .stop = stop, | |
4698 | .status = status, | |
4699 | .error_handler = error, | |
4700 | .hot_add_disk = raid5_add_disk, | |
4701 | .hot_remove_disk= raid5_remove_disk, | |
4702 | .spare_active = raid5_spare_active, | |
4703 | .sync_request = sync_request, | |
4704 | .resize = raid5_resize, | |
29269553 | 4705 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f N |
4706 | .check_reshape = raid5_check_reshape, |
4707 | .start_reshape = raid5_start_reshape, | |
29269553 | 4708 | #endif |
72626685 | 4709 | .quiesce = raid5_quiesce, |
1da177e4 LT |
4710 | }; |
4711 | ||
2604b703 | 4712 | static struct mdk_personality raid4_personality = |
1da177e4 | 4713 | { |
2604b703 N |
4714 | .name = "raid4", |
4715 | .level = 4, | |
4716 | .owner = THIS_MODULE, | |
4717 | .make_request = make_request, | |
4718 | .run = run, | |
4719 | .stop = stop, | |
4720 | .status = status, | |
4721 | .error_handler = error, | |
4722 | .hot_add_disk = raid5_add_disk, | |
4723 | .hot_remove_disk= raid5_remove_disk, | |
4724 | .spare_active = raid5_spare_active, | |
4725 | .sync_request = sync_request, | |
4726 | .resize = raid5_resize, | |
3d37890b N |
4727 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4728 | .check_reshape = raid5_check_reshape, | |
4729 | .start_reshape = raid5_start_reshape, | |
4730 | #endif | |
2604b703 N |
4731 | .quiesce = raid5_quiesce, |
4732 | }; | |
4733 | ||
4734 | static int __init raid5_init(void) | |
4735 | { | |
16a53ecc N |
4736 | int e; |
4737 | ||
4738 | e = raid6_select_algo(); | |
4739 | if ( e ) | |
4740 | return e; | |
4741 | register_md_personality(&raid6_personality); | |
2604b703 N |
4742 | register_md_personality(&raid5_personality); |
4743 | register_md_personality(&raid4_personality); | |
4744 | return 0; | |
1da177e4 LT |
4745 | } |
4746 | ||
2604b703 | 4747 | static void raid5_exit(void) |
1da177e4 | 4748 | { |
16a53ecc | 4749 | unregister_md_personality(&raid6_personality); |
2604b703 N |
4750 | unregister_md_personality(&raid5_personality); |
4751 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
4752 | } |
4753 | ||
4754 | module_init(raid5_init); | |
4755 | module_exit(raid5_exit); | |
4756 | MODULE_LICENSE("GPL"); | |
4757 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
4758 | MODULE_ALIAS("md-raid5"); |
4759 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
4760 | MODULE_ALIAS("md-level-5"); |
4761 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
4762 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
4763 | MODULE_ALIAS("md-raid6"); | |
4764 | MODULE_ALIAS("md-level-6"); | |
4765 | ||
4766 | /* This used to be two separate modules, they were: */ | |
4767 | MODULE_ALIAS("raid5"); | |
4768 | MODULE_ALIAS("raid6"); |