]>
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. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
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 | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
36 | * When we are ready to do a write, if that batch hasn't been written yet, |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
07a3b417 | 50 | #include <linux/async.h> |
bff61975 | 51 | #include <linux/seq_file.h> |
36d1c647 | 52 | #include <linux/cpu.h> |
5a0e3ad6 | 53 | #include <linux/slab.h> |
8bda470e | 54 | #include <linux/ratelimit.h> |
43b2e5d8 | 55 | #include "md.h" |
bff61975 | 56 | #include "raid5.h" |
54071b38 | 57 | #include "raid0.h" |
ef740c37 | 58 | #include "bitmap.h" |
72626685 | 59 | |
1da177e4 LT |
60 | /* |
61 | * Stripe cache | |
62 | */ | |
63 | ||
64 | #define NR_STRIPES 256 | |
65 | #define STRIPE_SIZE PAGE_SIZE | |
66 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
67 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
68 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 69 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 70 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
71 | #define HASH_MASK (NR_HASH - 1) |
72 | ||
fccddba0 | 73 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
74 | |
75 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
76 | * order without overlap. There may be several bio's per stripe+device, and | |
77 | * a bio could span several devices. | |
78 | * When walking this list for a particular stripe+device, we must never proceed | |
79 | * beyond a bio that extends past this device, as the next bio might no longer | |
80 | * be valid. | |
81 | * This macro is used to determine the 'next' bio in the list, given the sector | |
82 | * of the current stripe+device | |
83 | */ | |
84 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
85 | /* | |
86 | * The following can be used to debug the driver | |
87 | */ | |
1da177e4 LT |
88 | #define RAID5_PARANOIA 1 |
89 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
90 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
91 | #else | |
92 | # define CHECK_DEVLOCK() | |
93 | #endif | |
94 | ||
45b4233c | 95 | #ifdef DEBUG |
1da177e4 LT |
96 | #define inline |
97 | #define __inline__ | |
98 | #endif | |
99 | ||
960e739d | 100 | /* |
5b99c2ff JA |
101 | * We maintain a biased count of active stripes in the bottom 16 bits of |
102 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
103 | */ |
104 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
105 | { | |
5b99c2ff | 106 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
107 | } |
108 | ||
109 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
110 | { | |
5b99c2ff | 111 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
112 | } |
113 | ||
114 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
115 | { | |
116 | --bio->bi_phys_segments; | |
117 | return raid5_bi_phys_segments(bio); | |
118 | } | |
119 | ||
120 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
121 | { | |
122 | unsigned short val = raid5_bi_hw_segments(bio); | |
123 | ||
124 | --val; | |
5b99c2ff | 125 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
126 | return val; |
127 | } | |
128 | ||
129 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
130 | { | |
9b2dc8b6 | 131 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16); |
960e739d JA |
132 | } |
133 | ||
d0dabf7e N |
134 | /* Find first data disk in a raid6 stripe */ |
135 | static inline int raid6_d0(struct stripe_head *sh) | |
136 | { | |
67cc2b81 N |
137 | if (sh->ddf_layout) |
138 | /* ddf always start from first device */ | |
139 | return 0; | |
140 | /* md starts just after Q block */ | |
d0dabf7e N |
141 | if (sh->qd_idx == sh->disks - 1) |
142 | return 0; | |
143 | else | |
144 | return sh->qd_idx + 1; | |
145 | } | |
16a53ecc N |
146 | static inline int raid6_next_disk(int disk, int raid_disks) |
147 | { | |
148 | disk++; | |
149 | return (disk < raid_disks) ? disk : 0; | |
150 | } | |
a4456856 | 151 | |
d0dabf7e N |
152 | /* When walking through the disks in a raid5, starting at raid6_d0, |
153 | * We need to map each disk to a 'slot', where the data disks are slot | |
154 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
155 | * is raid_disks-1. This help does that mapping. | |
156 | */ | |
67cc2b81 N |
157 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
158 | int *count, int syndrome_disks) | |
d0dabf7e | 159 | { |
6629542e | 160 | int slot = *count; |
67cc2b81 | 161 | |
e4424fee | 162 | if (sh->ddf_layout) |
6629542e | 163 | (*count)++; |
d0dabf7e | 164 | if (idx == sh->pd_idx) |
67cc2b81 | 165 | return syndrome_disks; |
d0dabf7e | 166 | if (idx == sh->qd_idx) |
67cc2b81 | 167 | return syndrome_disks + 1; |
e4424fee | 168 | if (!sh->ddf_layout) |
6629542e | 169 | (*count)++; |
d0dabf7e N |
170 | return slot; |
171 | } | |
172 | ||
a4456856 DW |
173 | static void return_io(struct bio *return_bi) |
174 | { | |
175 | struct bio *bi = return_bi; | |
176 | while (bi) { | |
a4456856 DW |
177 | |
178 | return_bi = bi->bi_next; | |
179 | bi->bi_next = NULL; | |
180 | bi->bi_size = 0; | |
0e13fe23 | 181 | bio_endio(bi, 0); |
a4456856 DW |
182 | bi = return_bi; |
183 | } | |
184 | } | |
185 | ||
1da177e4 LT |
186 | static void print_raid5_conf (raid5_conf_t *conf); |
187 | ||
600aa109 DW |
188 | static int stripe_operations_active(struct stripe_head *sh) |
189 | { | |
190 | return sh->check_state || sh->reconstruct_state || | |
191 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
192 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
193 | } | |
194 | ||
858119e1 | 195 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
196 | { |
197 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
198 | BUG_ON(!list_empty(&sh->lru)); |
199 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 200 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
482c0834 | 201 | if (test_bit(STRIPE_DELAYED, &sh->state)) |
1da177e4 | 202 | list_add_tail(&sh->lru, &conf->delayed_list); |
482c0834 N |
203 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
204 | sh->bm_seq - conf->seq_write > 0) | |
72626685 | 205 | list_add_tail(&sh->lru, &conf->bitmap_list); |
482c0834 | 206 | else { |
72626685 | 207 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 208 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 209 | } |
1da177e4 LT |
210 | md_wakeup_thread(conf->mddev->thread); |
211 | } else { | |
600aa109 | 212 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
213 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
214 | atomic_dec(&conf->preread_active_stripes); | |
215 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
216 | md_wakeup_thread(conf->mddev->thread); | |
217 | } | |
1da177e4 | 218 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
219 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
220 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 221 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
222 | if (conf->retry_read_aligned) |
223 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 224 | } |
1da177e4 LT |
225 | } |
226 | } | |
227 | } | |
d0dabf7e | 228 | |
1da177e4 LT |
229 | static void release_stripe(struct stripe_head *sh) |
230 | { | |
231 | raid5_conf_t *conf = sh->raid_conf; | |
232 | unsigned long flags; | |
16a53ecc | 233 | |
1da177e4 LT |
234 | spin_lock_irqsave(&conf->device_lock, flags); |
235 | __release_stripe(conf, sh); | |
236 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
237 | } | |
238 | ||
fccddba0 | 239 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 240 | { |
45b4233c DW |
241 | pr_debug("remove_hash(), stripe %llu\n", |
242 | (unsigned long long)sh->sector); | |
1da177e4 | 243 | |
fccddba0 | 244 | hlist_del_init(&sh->hash); |
1da177e4 LT |
245 | } |
246 | ||
16a53ecc | 247 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 248 | { |
fccddba0 | 249 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 250 | |
45b4233c DW |
251 | pr_debug("insert_hash(), stripe %llu\n", |
252 | (unsigned long long)sh->sector); | |
1da177e4 LT |
253 | |
254 | CHECK_DEVLOCK(); | |
fccddba0 | 255 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
256 | } |
257 | ||
258 | ||
259 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
260 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
261 | { | |
262 | struct stripe_head *sh = NULL; | |
263 | struct list_head *first; | |
264 | ||
265 | CHECK_DEVLOCK(); | |
266 | if (list_empty(&conf->inactive_list)) | |
267 | goto out; | |
268 | first = conf->inactive_list.next; | |
269 | sh = list_entry(first, struct stripe_head, lru); | |
270 | list_del_init(first); | |
271 | remove_hash(sh); | |
272 | atomic_inc(&conf->active_stripes); | |
273 | out: | |
274 | return sh; | |
275 | } | |
276 | ||
e4e11e38 | 277 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
278 | { |
279 | struct page *p; | |
280 | int i; | |
e4e11e38 | 281 | int num = sh->raid_conf->pool_size; |
1da177e4 | 282 | |
e4e11e38 | 283 | for (i = 0; i < num ; i++) { |
1da177e4 LT |
284 | p = sh->dev[i].page; |
285 | if (!p) | |
286 | continue; | |
287 | sh->dev[i].page = NULL; | |
2d1f3b5d | 288 | put_page(p); |
1da177e4 LT |
289 | } |
290 | } | |
291 | ||
e4e11e38 | 292 | static int grow_buffers(struct stripe_head *sh) |
1da177e4 LT |
293 | { |
294 | int i; | |
e4e11e38 | 295 | int num = sh->raid_conf->pool_size; |
1da177e4 | 296 | |
e4e11e38 | 297 | for (i = 0; i < num; i++) { |
1da177e4 LT |
298 | struct page *page; |
299 | ||
300 | if (!(page = alloc_page(GFP_KERNEL))) { | |
301 | return 1; | |
302 | } | |
303 | sh->dev[i].page = page; | |
304 | } | |
305 | return 0; | |
306 | } | |
307 | ||
784052ec | 308 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
911d4ee8 N |
309 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
310 | struct stripe_head *sh); | |
1da177e4 | 311 | |
b5663ba4 | 312 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
313 | { |
314 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 315 | int i; |
1da177e4 | 316 | |
78bafebd ES |
317 | BUG_ON(atomic_read(&sh->count) != 0); |
318 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 319 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 320 | |
1da177e4 | 321 | CHECK_DEVLOCK(); |
45b4233c | 322 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
323 | (unsigned long long)sh->sector); |
324 | ||
325 | remove_hash(sh); | |
16a53ecc | 326 | |
86b42c71 | 327 | sh->generation = conf->generation - previous; |
b5663ba4 | 328 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 329 | sh->sector = sector; |
911d4ee8 | 330 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
331 | sh->state = 0; |
332 | ||
7ecaa1e6 N |
333 | |
334 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
335 | struct r5dev *dev = &sh->dev[i]; |
336 | ||
d84e0f10 | 337 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 338 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 339 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 340 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 341 | dev->read, dev->towrite, dev->written, |
1da177e4 | 342 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 343 | WARN_ON(1); |
1da177e4 LT |
344 | } |
345 | dev->flags = 0; | |
784052ec | 346 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
347 | } |
348 | insert_hash(conf, sh); | |
349 | } | |
350 | ||
86b42c71 N |
351 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, |
352 | short generation) | |
1da177e4 LT |
353 | { |
354 | struct stripe_head *sh; | |
fccddba0 | 355 | struct hlist_node *hn; |
1da177e4 LT |
356 | |
357 | CHECK_DEVLOCK(); | |
45b4233c | 358 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 359 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 360 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 361 | return sh; |
45b4233c | 362 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
363 | return NULL; |
364 | } | |
365 | ||
674806d6 N |
366 | /* |
367 | * Need to check if array has failed when deciding whether to: | |
368 | * - start an array | |
369 | * - remove non-faulty devices | |
370 | * - add a spare | |
371 | * - allow a reshape | |
372 | * This determination is simple when no reshape is happening. | |
373 | * However if there is a reshape, we need to carefully check | |
374 | * both the before and after sections. | |
375 | * This is because some failed devices may only affect one | |
376 | * of the two sections, and some non-in_sync devices may | |
377 | * be insync in the section most affected by failed devices. | |
378 | */ | |
379 | static int has_failed(raid5_conf_t *conf) | |
380 | { | |
381 | int degraded; | |
382 | int i; | |
383 | if (conf->mddev->reshape_position == MaxSector) | |
384 | return conf->mddev->degraded > conf->max_degraded; | |
385 | ||
386 | rcu_read_lock(); | |
387 | degraded = 0; | |
388 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
389 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
390 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
391 | degraded++; | |
392 | else if (test_bit(In_sync, &rdev->flags)) | |
393 | ; | |
394 | else | |
395 | /* not in-sync or faulty. | |
396 | * If the reshape increases the number of devices, | |
397 | * this is being recovered by the reshape, so | |
398 | * this 'previous' section is not in_sync. | |
399 | * If the number of devices is being reduced however, | |
400 | * the device can only be part of the array if | |
401 | * we are reverting a reshape, so this section will | |
402 | * be in-sync. | |
403 | */ | |
404 | if (conf->raid_disks >= conf->previous_raid_disks) | |
405 | degraded++; | |
406 | } | |
407 | rcu_read_unlock(); | |
408 | if (degraded > conf->max_degraded) | |
409 | return 1; | |
410 | rcu_read_lock(); | |
411 | degraded = 0; | |
412 | for (i = 0; i < conf->raid_disks; i++) { | |
413 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
414 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
415 | degraded++; | |
416 | else if (test_bit(In_sync, &rdev->flags)) | |
417 | ; | |
418 | else | |
419 | /* not in-sync or faulty. | |
420 | * If reshape increases the number of devices, this | |
421 | * section has already been recovered, else it | |
422 | * almost certainly hasn't. | |
423 | */ | |
424 | if (conf->raid_disks <= conf->previous_raid_disks) | |
425 | degraded++; | |
426 | } | |
427 | rcu_read_unlock(); | |
428 | if (degraded > conf->max_degraded) | |
429 | return 1; | |
430 | return 0; | |
431 | } | |
432 | ||
b5663ba4 N |
433 | static struct stripe_head * |
434 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
a8c906ca | 435 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
436 | { |
437 | struct stripe_head *sh; | |
438 | ||
45b4233c | 439 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
440 | |
441 | spin_lock_irq(&conf->device_lock); | |
442 | ||
443 | do { | |
72626685 | 444 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 445 | conf->quiesce == 0 || noquiesce, |
72626685 | 446 | conf->device_lock, /* nothing */); |
86b42c71 | 447 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
448 | if (!sh) { |
449 | if (!conf->inactive_blocked) | |
450 | sh = get_free_stripe(conf); | |
451 | if (noblock && sh == NULL) | |
452 | break; | |
453 | if (!sh) { | |
454 | conf->inactive_blocked = 1; | |
455 | wait_event_lock_irq(conf->wait_for_stripe, | |
456 | !list_empty(&conf->inactive_list) && | |
5036805b N |
457 | (atomic_read(&conf->active_stripes) |
458 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
459 | || !conf->inactive_blocked), |
460 | conf->device_lock, | |
7c13edc8 | 461 | ); |
1da177e4 LT |
462 | conf->inactive_blocked = 0; |
463 | } else | |
b5663ba4 | 464 | init_stripe(sh, sector, previous); |
1da177e4 LT |
465 | } else { |
466 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
467 | BUG_ON(!list_empty(&sh->lru) |
468 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
469 | } else { |
470 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
471 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
472 | if (list_empty(&sh->lru) && |
473 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
474 | BUG(); |
475 | list_del_init(&sh->lru); | |
1da177e4 LT |
476 | } |
477 | } | |
478 | } while (sh == NULL); | |
479 | ||
480 | if (sh) | |
481 | atomic_inc(&sh->count); | |
482 | ||
483 | spin_unlock_irq(&conf->device_lock); | |
484 | return sh; | |
485 | } | |
486 | ||
6712ecf8 N |
487 | static void |
488 | raid5_end_read_request(struct bio *bi, int error); | |
489 | static void | |
490 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 491 | |
c4e5ac0a | 492 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
493 | { |
494 | raid5_conf_t *conf = sh->raid_conf; | |
495 | int i, disks = sh->disks; | |
496 | ||
497 | might_sleep(); | |
498 | ||
499 | for (i = disks; i--; ) { | |
500 | int rw; | |
501 | struct bio *bi; | |
502 | mdk_rdev_t *rdev; | |
e9c7469b TH |
503 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
504 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
505 | rw = WRITE_FUA; | |
506 | else | |
507 | rw = WRITE; | |
508 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
91c00924 DW |
509 | rw = READ; |
510 | else | |
511 | continue; | |
512 | ||
513 | bi = &sh->dev[i].req; | |
514 | ||
515 | bi->bi_rw = rw; | |
b062962e | 516 | if (rw & WRITE) |
91c00924 DW |
517 | bi->bi_end_io = raid5_end_write_request; |
518 | else | |
519 | bi->bi_end_io = raid5_end_read_request; | |
520 | ||
521 | rcu_read_lock(); | |
522 | rdev = rcu_dereference(conf->disks[i].rdev); | |
523 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
524 | rdev = NULL; | |
525 | if (rdev) | |
526 | atomic_inc(&rdev->nr_pending); | |
527 | rcu_read_unlock(); | |
528 | ||
529 | if (rdev) { | |
c4e5ac0a | 530 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
531 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
532 | ||
2b7497f0 DW |
533 | set_bit(STRIPE_IO_STARTED, &sh->state); |
534 | ||
91c00924 DW |
535 | bi->bi_bdev = rdev->bdev; |
536 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 537 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
538 | bi->bi_rw, i); |
539 | atomic_inc(&sh->count); | |
540 | bi->bi_sector = sh->sector + rdev->data_offset; | |
541 | bi->bi_flags = 1 << BIO_UPTODATE; | |
542 | bi->bi_vcnt = 1; | |
543 | bi->bi_max_vecs = 1; | |
544 | bi->bi_idx = 0; | |
545 | bi->bi_io_vec = &sh->dev[i].vec; | |
546 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
547 | bi->bi_io_vec[0].bv_offset = 0; | |
548 | bi->bi_size = STRIPE_SIZE; | |
549 | bi->bi_next = NULL; | |
91c00924 DW |
550 | generic_make_request(bi); |
551 | } else { | |
b062962e | 552 | if (rw & WRITE) |
91c00924 DW |
553 | set_bit(STRIPE_DEGRADED, &sh->state); |
554 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
555 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
556 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
557 | set_bit(STRIPE_HANDLE, &sh->state); | |
558 | } | |
559 | } | |
560 | } | |
561 | ||
562 | static struct dma_async_tx_descriptor * | |
563 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
564 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
565 | { | |
566 | struct bio_vec *bvl; | |
567 | struct page *bio_page; | |
568 | int i; | |
569 | int page_offset; | |
a08abd8c | 570 | struct async_submit_ctl submit; |
0403e382 | 571 | enum async_tx_flags flags = 0; |
91c00924 DW |
572 | |
573 | if (bio->bi_sector >= sector) | |
574 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
575 | else | |
576 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 577 | |
0403e382 DW |
578 | if (frombio) |
579 | flags |= ASYNC_TX_FENCE; | |
580 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
581 | ||
91c00924 | 582 | bio_for_each_segment(bvl, bio, i) { |
fcde9075 | 583 | int len = bvl->bv_len; |
91c00924 DW |
584 | int clen; |
585 | int b_offset = 0; | |
586 | ||
587 | if (page_offset < 0) { | |
588 | b_offset = -page_offset; | |
589 | page_offset += b_offset; | |
590 | len -= b_offset; | |
591 | } | |
592 | ||
593 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
594 | clen = STRIPE_SIZE - page_offset; | |
595 | else | |
596 | clen = len; | |
597 | ||
598 | if (clen > 0) { | |
fcde9075 NK |
599 | b_offset += bvl->bv_offset; |
600 | bio_page = bvl->bv_page; | |
91c00924 DW |
601 | if (frombio) |
602 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 603 | b_offset, clen, &submit); |
91c00924 DW |
604 | else |
605 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 606 | page_offset, clen, &submit); |
91c00924 | 607 | } |
a08abd8c DW |
608 | /* chain the operations */ |
609 | submit.depend_tx = tx; | |
610 | ||
91c00924 DW |
611 | if (clen < len) /* hit end of page */ |
612 | break; | |
613 | page_offset += len; | |
614 | } | |
615 | ||
616 | return tx; | |
617 | } | |
618 | ||
619 | static void ops_complete_biofill(void *stripe_head_ref) | |
620 | { | |
621 | struct stripe_head *sh = stripe_head_ref; | |
622 | struct bio *return_bi = NULL; | |
623 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 624 | int i; |
91c00924 | 625 | |
e46b272b | 626 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
627 | (unsigned long long)sh->sector); |
628 | ||
629 | /* clear completed biofills */ | |
83de75cc | 630 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
631 | for (i = sh->disks; i--; ) { |
632 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
633 | |
634 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
635 | /* and check if we need to reply to a read request, |
636 | * new R5_Wantfill requests are held off until | |
83de75cc | 637 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
638 | */ |
639 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 640 | struct bio *rbi, *rbi2; |
91c00924 | 641 | |
91c00924 DW |
642 | BUG_ON(!dev->read); |
643 | rbi = dev->read; | |
644 | dev->read = NULL; | |
645 | while (rbi && rbi->bi_sector < | |
646 | dev->sector + STRIPE_SECTORS) { | |
647 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 648 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
649 | rbi->bi_next = return_bi; |
650 | return_bi = rbi; | |
651 | } | |
91c00924 DW |
652 | rbi = rbi2; |
653 | } | |
654 | } | |
655 | } | |
83de75cc DW |
656 | spin_unlock_irq(&conf->device_lock); |
657 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
658 | |
659 | return_io(return_bi); | |
660 | ||
e4d84909 | 661 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
662 | release_stripe(sh); |
663 | } | |
664 | ||
665 | static void ops_run_biofill(struct stripe_head *sh) | |
666 | { | |
667 | struct dma_async_tx_descriptor *tx = NULL; | |
668 | raid5_conf_t *conf = sh->raid_conf; | |
a08abd8c | 669 | struct async_submit_ctl submit; |
91c00924 DW |
670 | int i; |
671 | ||
e46b272b | 672 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
673 | (unsigned long long)sh->sector); |
674 | ||
675 | for (i = sh->disks; i--; ) { | |
676 | struct r5dev *dev = &sh->dev[i]; | |
677 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
678 | struct bio *rbi; | |
679 | spin_lock_irq(&conf->device_lock); | |
680 | dev->read = rbi = dev->toread; | |
681 | dev->toread = NULL; | |
682 | spin_unlock_irq(&conf->device_lock); | |
683 | while (rbi && rbi->bi_sector < | |
684 | dev->sector + STRIPE_SECTORS) { | |
685 | tx = async_copy_data(0, rbi, dev->page, | |
686 | dev->sector, tx); | |
687 | rbi = r5_next_bio(rbi, dev->sector); | |
688 | } | |
689 | } | |
690 | } | |
691 | ||
692 | atomic_inc(&sh->count); | |
a08abd8c DW |
693 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
694 | async_trigger_callback(&submit); | |
91c00924 DW |
695 | } |
696 | ||
4e7d2c0a | 697 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 698 | { |
4e7d2c0a | 699 | struct r5dev *tgt; |
91c00924 | 700 | |
4e7d2c0a DW |
701 | if (target < 0) |
702 | return; | |
91c00924 | 703 | |
4e7d2c0a | 704 | tgt = &sh->dev[target]; |
91c00924 DW |
705 | set_bit(R5_UPTODATE, &tgt->flags); |
706 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
707 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
708 | } |
709 | ||
ac6b53b6 | 710 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
711 | { |
712 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 713 | |
e46b272b | 714 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
715 | (unsigned long long)sh->sector); |
716 | ||
ac6b53b6 | 717 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 718 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 719 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 720 | |
ecc65c9b DW |
721 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
722 | if (sh->check_state == check_state_compute_run) | |
723 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
724 | set_bit(STRIPE_HANDLE, &sh->state); |
725 | release_stripe(sh); | |
726 | } | |
727 | ||
d6f38f31 DW |
728 | /* return a pointer to the address conversion region of the scribble buffer */ |
729 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
730 | struct raid5_percpu *percpu) | |
731 | { | |
732 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
733 | } | |
734 | ||
735 | static struct dma_async_tx_descriptor * | |
736 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 737 | { |
91c00924 | 738 | int disks = sh->disks; |
d6f38f31 | 739 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
740 | int target = sh->ops.target; |
741 | struct r5dev *tgt = &sh->dev[target]; | |
742 | struct page *xor_dest = tgt->page; | |
743 | int count = 0; | |
744 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 745 | struct async_submit_ctl submit; |
91c00924 DW |
746 | int i; |
747 | ||
748 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 749 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
750 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
751 | ||
752 | for (i = disks; i--; ) | |
753 | if (i != target) | |
754 | xor_srcs[count++] = sh->dev[i].page; | |
755 | ||
756 | atomic_inc(&sh->count); | |
757 | ||
0403e382 | 758 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 759 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 760 | if (unlikely(count == 1)) |
a08abd8c | 761 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 762 | else |
a08abd8c | 763 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 764 | |
91c00924 DW |
765 | return tx; |
766 | } | |
767 | ||
ac6b53b6 DW |
768 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
769 | * @srcs - (struct page *) array of size sh->disks | |
770 | * @sh - stripe_head to parse | |
771 | * | |
772 | * Populates srcs in proper layout order for the stripe and returns the | |
773 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
774 | * destination buffer is recorded in srcs[count] and the Q destination | |
775 | * is recorded in srcs[count+1]]. | |
776 | */ | |
777 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
778 | { | |
779 | int disks = sh->disks; | |
780 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
781 | int d0_idx = raid6_d0(sh); | |
782 | int count; | |
783 | int i; | |
784 | ||
785 | for (i = 0; i < disks; i++) | |
5dd33c9a | 786 | srcs[i] = NULL; |
ac6b53b6 DW |
787 | |
788 | count = 0; | |
789 | i = d0_idx; | |
790 | do { | |
791 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
792 | ||
793 | srcs[slot] = sh->dev[i].page; | |
794 | i = raid6_next_disk(i, disks); | |
795 | } while (i != d0_idx); | |
ac6b53b6 | 796 | |
e4424fee | 797 | return syndrome_disks; |
ac6b53b6 DW |
798 | } |
799 | ||
800 | static struct dma_async_tx_descriptor * | |
801 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
802 | { | |
803 | int disks = sh->disks; | |
804 | struct page **blocks = percpu->scribble; | |
805 | int target; | |
806 | int qd_idx = sh->qd_idx; | |
807 | struct dma_async_tx_descriptor *tx; | |
808 | struct async_submit_ctl submit; | |
809 | struct r5dev *tgt; | |
810 | struct page *dest; | |
811 | int i; | |
812 | int count; | |
813 | ||
814 | if (sh->ops.target < 0) | |
815 | target = sh->ops.target2; | |
816 | else if (sh->ops.target2 < 0) | |
817 | target = sh->ops.target; | |
91c00924 | 818 | else |
ac6b53b6 DW |
819 | /* we should only have one valid target */ |
820 | BUG(); | |
821 | BUG_ON(target < 0); | |
822 | pr_debug("%s: stripe %llu block: %d\n", | |
823 | __func__, (unsigned long long)sh->sector, target); | |
824 | ||
825 | tgt = &sh->dev[target]; | |
826 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
827 | dest = tgt->page; | |
828 | ||
829 | atomic_inc(&sh->count); | |
830 | ||
831 | if (target == qd_idx) { | |
832 | count = set_syndrome_sources(blocks, sh); | |
833 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
834 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
835 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
836 | ops_complete_compute, sh, | |
ac6b53b6 DW |
837 | to_addr_conv(sh, percpu)); |
838 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
839 | } else { | |
840 | /* Compute any data- or p-drive using XOR */ | |
841 | count = 0; | |
842 | for (i = disks; i-- ; ) { | |
843 | if (i == target || i == qd_idx) | |
844 | continue; | |
845 | blocks[count++] = sh->dev[i].page; | |
846 | } | |
847 | ||
0403e382 DW |
848 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
849 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
850 | to_addr_conv(sh, percpu)); |
851 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
852 | } | |
91c00924 | 853 | |
91c00924 DW |
854 | return tx; |
855 | } | |
856 | ||
ac6b53b6 DW |
857 | static struct dma_async_tx_descriptor * |
858 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
859 | { | |
860 | int i, count, disks = sh->disks; | |
861 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
862 | int d0_idx = raid6_d0(sh); | |
863 | int faila = -1, failb = -1; | |
864 | int target = sh->ops.target; | |
865 | int target2 = sh->ops.target2; | |
866 | struct r5dev *tgt = &sh->dev[target]; | |
867 | struct r5dev *tgt2 = &sh->dev[target2]; | |
868 | struct dma_async_tx_descriptor *tx; | |
869 | struct page **blocks = percpu->scribble; | |
870 | struct async_submit_ctl submit; | |
871 | ||
872 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
873 | __func__, (unsigned long long)sh->sector, target, target2); | |
874 | BUG_ON(target < 0 || target2 < 0); | |
875 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
876 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
877 | ||
6c910a78 | 878 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
879 | * slot number conversion for 'faila' and 'failb' |
880 | */ | |
881 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 882 | blocks[i] = NULL; |
ac6b53b6 DW |
883 | count = 0; |
884 | i = d0_idx; | |
885 | do { | |
886 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
887 | ||
888 | blocks[slot] = sh->dev[i].page; | |
889 | ||
890 | if (i == target) | |
891 | faila = slot; | |
892 | if (i == target2) | |
893 | failb = slot; | |
894 | i = raid6_next_disk(i, disks); | |
895 | } while (i != d0_idx); | |
ac6b53b6 DW |
896 | |
897 | BUG_ON(faila == failb); | |
898 | if (failb < faila) | |
899 | swap(faila, failb); | |
900 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
901 | __func__, (unsigned long long)sh->sector, faila, failb); | |
902 | ||
903 | atomic_inc(&sh->count); | |
904 | ||
905 | if (failb == syndrome_disks+1) { | |
906 | /* Q disk is one of the missing disks */ | |
907 | if (faila == syndrome_disks) { | |
908 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
909 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
910 | ops_complete_compute, sh, | |
911 | to_addr_conv(sh, percpu)); | |
e4424fee | 912 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
913 | STRIPE_SIZE, &submit); |
914 | } else { | |
915 | struct page *dest; | |
916 | int data_target; | |
917 | int qd_idx = sh->qd_idx; | |
918 | ||
919 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
920 | if (target == qd_idx) | |
921 | data_target = target2; | |
922 | else | |
923 | data_target = target; | |
924 | ||
925 | count = 0; | |
926 | for (i = disks; i-- ; ) { | |
927 | if (i == data_target || i == qd_idx) | |
928 | continue; | |
929 | blocks[count++] = sh->dev[i].page; | |
930 | } | |
931 | dest = sh->dev[data_target].page; | |
0403e382 DW |
932 | init_async_submit(&submit, |
933 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
934 | NULL, NULL, NULL, | |
935 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
936 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
937 | &submit); | |
938 | ||
939 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
940 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
941 | ops_complete_compute, sh, | |
942 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
943 | return async_gen_syndrome(blocks, 0, count+2, |
944 | STRIPE_SIZE, &submit); | |
945 | } | |
ac6b53b6 | 946 | } else { |
6c910a78 DW |
947 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
948 | ops_complete_compute, sh, | |
949 | to_addr_conv(sh, percpu)); | |
950 | if (failb == syndrome_disks) { | |
951 | /* We're missing D+P. */ | |
952 | return async_raid6_datap_recov(syndrome_disks+2, | |
953 | STRIPE_SIZE, faila, | |
954 | blocks, &submit); | |
955 | } else { | |
956 | /* We're missing D+D. */ | |
957 | return async_raid6_2data_recov(syndrome_disks+2, | |
958 | STRIPE_SIZE, faila, failb, | |
959 | blocks, &submit); | |
960 | } | |
ac6b53b6 DW |
961 | } |
962 | } | |
963 | ||
964 | ||
91c00924 DW |
965 | static void ops_complete_prexor(void *stripe_head_ref) |
966 | { | |
967 | struct stripe_head *sh = stripe_head_ref; | |
968 | ||
e46b272b | 969 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 970 | (unsigned long long)sh->sector); |
91c00924 DW |
971 | } |
972 | ||
973 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
974 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
975 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 976 | { |
91c00924 | 977 | int disks = sh->disks; |
d6f38f31 | 978 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 979 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 980 | struct async_submit_ctl submit; |
91c00924 DW |
981 | |
982 | /* existing parity data subtracted */ | |
983 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
984 | ||
e46b272b | 985 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
986 | (unsigned long long)sh->sector); |
987 | ||
988 | for (i = disks; i--; ) { | |
989 | struct r5dev *dev = &sh->dev[i]; | |
990 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 991 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
992 | xor_srcs[count++] = dev->page; |
993 | } | |
994 | ||
0403e382 | 995 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 996 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 997 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
998 | |
999 | return tx; | |
1000 | } | |
1001 | ||
1002 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1003 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1004 | { |
1005 | int disks = sh->disks; | |
d8ee0728 | 1006 | int i; |
91c00924 | 1007 | |
e46b272b | 1008 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1009 | (unsigned long long)sh->sector); |
1010 | ||
1011 | for (i = disks; i--; ) { | |
1012 | struct r5dev *dev = &sh->dev[i]; | |
1013 | struct bio *chosen; | |
91c00924 | 1014 | |
d8ee0728 | 1015 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1016 | struct bio *wbi; |
1017 | ||
cbe47ec5 | 1018 | spin_lock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1019 | chosen = dev->towrite; |
1020 | dev->towrite = NULL; | |
1021 | BUG_ON(dev->written); | |
1022 | wbi = dev->written = chosen; | |
cbe47ec5 | 1023 | spin_unlock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1024 | |
1025 | while (wbi && wbi->bi_sector < | |
1026 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1027 | if (wbi->bi_rw & REQ_FUA) |
1028 | set_bit(R5_WantFUA, &dev->flags); | |
91c00924 DW |
1029 | tx = async_copy_data(1, wbi, dev->page, |
1030 | dev->sector, tx); | |
1031 | wbi = r5_next_bio(wbi, dev->sector); | |
1032 | } | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | return tx; | |
1037 | } | |
1038 | ||
ac6b53b6 | 1039 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1040 | { |
1041 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1042 | int disks = sh->disks; |
1043 | int pd_idx = sh->pd_idx; | |
1044 | int qd_idx = sh->qd_idx; | |
1045 | int i; | |
e9c7469b | 1046 | bool fua = false; |
91c00924 | 1047 | |
e46b272b | 1048 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1049 | (unsigned long long)sh->sector); |
1050 | ||
e9c7469b TH |
1051 | for (i = disks; i--; ) |
1052 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); | |
1053 | ||
91c00924 DW |
1054 | for (i = disks; i--; ) { |
1055 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1056 | |
e9c7469b | 1057 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1058 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1059 | if (fua) |
1060 | set_bit(R5_WantFUA, &dev->flags); | |
1061 | } | |
91c00924 DW |
1062 | } |
1063 | ||
d8ee0728 DW |
1064 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1065 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1066 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1067 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1068 | else { | |
1069 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1070 | sh->reconstruct_state = reconstruct_state_result; | |
1071 | } | |
91c00924 DW |
1072 | |
1073 | set_bit(STRIPE_HANDLE, &sh->state); | |
1074 | release_stripe(sh); | |
1075 | } | |
1076 | ||
1077 | static void | |
ac6b53b6 DW |
1078 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1079 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1080 | { |
91c00924 | 1081 | int disks = sh->disks; |
d6f38f31 | 1082 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1083 | struct async_submit_ctl submit; |
91c00924 DW |
1084 | int count = 0, pd_idx = sh->pd_idx, i; |
1085 | struct page *xor_dest; | |
d8ee0728 | 1086 | int prexor = 0; |
91c00924 | 1087 | unsigned long flags; |
91c00924 | 1088 | |
e46b272b | 1089 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1090 | (unsigned long long)sh->sector); |
1091 | ||
1092 | /* check if prexor is active which means only process blocks | |
1093 | * that are part of a read-modify-write (written) | |
1094 | */ | |
d8ee0728 DW |
1095 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1096 | prexor = 1; | |
91c00924 DW |
1097 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1098 | for (i = disks; i--; ) { | |
1099 | struct r5dev *dev = &sh->dev[i]; | |
1100 | if (dev->written) | |
1101 | xor_srcs[count++] = dev->page; | |
1102 | } | |
1103 | } else { | |
1104 | xor_dest = sh->dev[pd_idx].page; | |
1105 | for (i = disks; i--; ) { | |
1106 | struct r5dev *dev = &sh->dev[i]; | |
1107 | if (i != pd_idx) | |
1108 | xor_srcs[count++] = dev->page; | |
1109 | } | |
1110 | } | |
1111 | ||
91c00924 DW |
1112 | /* 1/ if we prexor'd then the dest is reused as a source |
1113 | * 2/ if we did not prexor then we are redoing the parity | |
1114 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1115 | * for the synchronous xor case | |
1116 | */ | |
88ba2aa5 | 1117 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1118 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1119 | ||
1120 | atomic_inc(&sh->count); | |
1121 | ||
ac6b53b6 | 1122 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1123 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1124 | if (unlikely(count == 1)) |
1125 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1126 | else | |
1127 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1128 | } |
1129 | ||
ac6b53b6 DW |
1130 | static void |
1131 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1132 | struct dma_async_tx_descriptor *tx) | |
1133 | { | |
1134 | struct async_submit_ctl submit; | |
1135 | struct page **blocks = percpu->scribble; | |
1136 | int count; | |
1137 | ||
1138 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1139 | ||
1140 | count = set_syndrome_sources(blocks, sh); | |
1141 | ||
1142 | atomic_inc(&sh->count); | |
1143 | ||
1144 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1145 | sh, to_addr_conv(sh, percpu)); | |
1146 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1147 | } |
1148 | ||
1149 | static void ops_complete_check(void *stripe_head_ref) | |
1150 | { | |
1151 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1152 | |
e46b272b | 1153 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1154 | (unsigned long long)sh->sector); |
1155 | ||
ecc65c9b | 1156 | sh->check_state = check_state_check_result; |
91c00924 DW |
1157 | set_bit(STRIPE_HANDLE, &sh->state); |
1158 | release_stripe(sh); | |
1159 | } | |
1160 | ||
ac6b53b6 | 1161 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1162 | { |
91c00924 | 1163 | int disks = sh->disks; |
ac6b53b6 DW |
1164 | int pd_idx = sh->pd_idx; |
1165 | int qd_idx = sh->qd_idx; | |
1166 | struct page *xor_dest; | |
d6f38f31 | 1167 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1168 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1169 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1170 | int count; |
1171 | int i; | |
91c00924 | 1172 | |
e46b272b | 1173 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1174 | (unsigned long long)sh->sector); |
1175 | ||
ac6b53b6 DW |
1176 | count = 0; |
1177 | xor_dest = sh->dev[pd_idx].page; | |
1178 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1179 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1180 | if (i == pd_idx || i == qd_idx) |
1181 | continue; | |
1182 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1183 | } |
1184 | ||
d6f38f31 DW |
1185 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1186 | to_addr_conv(sh, percpu)); | |
099f53cb | 1187 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1188 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1189 | |
91c00924 | 1190 | atomic_inc(&sh->count); |
a08abd8c DW |
1191 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1192 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1193 | } |
1194 | ||
ac6b53b6 DW |
1195 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1196 | { | |
1197 | struct page **srcs = percpu->scribble; | |
1198 | struct async_submit_ctl submit; | |
1199 | int count; | |
1200 | ||
1201 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1202 | (unsigned long long)sh->sector, checkp); | |
1203 | ||
1204 | count = set_syndrome_sources(srcs, sh); | |
1205 | if (!checkp) | |
1206 | srcs[count] = NULL; | |
91c00924 | 1207 | |
91c00924 | 1208 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1209 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1210 | sh, to_addr_conv(sh, percpu)); | |
1211 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1212 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1213 | } |
1214 | ||
417b8d4a | 1215 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1216 | { |
1217 | int overlap_clear = 0, i, disks = sh->disks; | |
1218 | struct dma_async_tx_descriptor *tx = NULL; | |
d6f38f31 | 1219 | raid5_conf_t *conf = sh->raid_conf; |
ac6b53b6 | 1220 | int level = conf->level; |
d6f38f31 DW |
1221 | struct raid5_percpu *percpu; |
1222 | unsigned long cpu; | |
91c00924 | 1223 | |
d6f38f31 DW |
1224 | cpu = get_cpu(); |
1225 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1226 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1227 | ops_run_biofill(sh); |
1228 | overlap_clear++; | |
1229 | } | |
1230 | ||
7b3a871e | 1231 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1232 | if (level < 6) |
1233 | tx = ops_run_compute5(sh, percpu); | |
1234 | else { | |
1235 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1236 | tx = ops_run_compute6_1(sh, percpu); | |
1237 | else | |
1238 | tx = ops_run_compute6_2(sh, percpu); | |
1239 | } | |
1240 | /* terminate the chain if reconstruct is not set to be run */ | |
1241 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1242 | async_tx_ack(tx); |
1243 | } | |
91c00924 | 1244 | |
600aa109 | 1245 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1246 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1247 | |
600aa109 | 1248 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1249 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1250 | overlap_clear++; |
1251 | } | |
1252 | ||
ac6b53b6 DW |
1253 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1254 | if (level < 6) | |
1255 | ops_run_reconstruct5(sh, percpu, tx); | |
1256 | else | |
1257 | ops_run_reconstruct6(sh, percpu, tx); | |
1258 | } | |
91c00924 | 1259 | |
ac6b53b6 DW |
1260 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1261 | if (sh->check_state == check_state_run) | |
1262 | ops_run_check_p(sh, percpu); | |
1263 | else if (sh->check_state == check_state_run_q) | |
1264 | ops_run_check_pq(sh, percpu, 0); | |
1265 | else if (sh->check_state == check_state_run_pq) | |
1266 | ops_run_check_pq(sh, percpu, 1); | |
1267 | else | |
1268 | BUG(); | |
1269 | } | |
91c00924 | 1270 | |
91c00924 DW |
1271 | if (overlap_clear) |
1272 | for (i = disks; i--; ) { | |
1273 | struct r5dev *dev = &sh->dev[i]; | |
1274 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1275 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1276 | } | |
d6f38f31 | 1277 | put_cpu(); |
91c00924 DW |
1278 | } |
1279 | ||
417b8d4a DW |
1280 | #ifdef CONFIG_MULTICORE_RAID456 |
1281 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1282 | { | |
1283 | struct stripe_head *sh = param; | |
1284 | unsigned long ops_request = sh->ops.request; | |
1285 | ||
1286 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1287 | wake_up(&sh->ops.wait_for_ops); | |
1288 | ||
1289 | __raid_run_ops(sh, ops_request); | |
1290 | release_stripe(sh); | |
1291 | } | |
1292 | ||
1293 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1294 | { | |
1295 | /* since handle_stripe can be called outside of raid5d context | |
1296 | * we need to ensure sh->ops.request is de-staged before another | |
1297 | * request arrives | |
1298 | */ | |
1299 | wait_event(sh->ops.wait_for_ops, | |
1300 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1301 | sh->ops.request = ops_request; | |
1302 | ||
1303 | atomic_inc(&sh->count); | |
1304 | async_schedule(async_run_ops, sh); | |
1305 | } | |
1306 | #else | |
1307 | #define raid_run_ops __raid_run_ops | |
1308 | #endif | |
1309 | ||
3f294f4f | 1310 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1311 | { |
1312 | struct stripe_head *sh; | |
6ce32846 | 1313 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1314 | if (!sh) |
1315 | return 0; | |
6ce32846 | 1316 | |
3f294f4f | 1317 | sh->raid_conf = conf; |
417b8d4a DW |
1318 | #ifdef CONFIG_MULTICORE_RAID456 |
1319 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1320 | #endif | |
3f294f4f | 1321 | |
e4e11e38 N |
1322 | if (grow_buffers(sh)) { |
1323 | shrink_buffers(sh); | |
3f294f4f N |
1324 | kmem_cache_free(conf->slab_cache, sh); |
1325 | return 0; | |
1326 | } | |
1327 | /* we just created an active stripe so... */ | |
1328 | atomic_set(&sh->count, 1); | |
1329 | atomic_inc(&conf->active_stripes); | |
1330 | INIT_LIST_HEAD(&sh->lru); | |
1331 | release_stripe(sh); | |
1332 | return 1; | |
1333 | } | |
1334 | ||
1335 | static int grow_stripes(raid5_conf_t *conf, int num) | |
1336 | { | |
e18b890b | 1337 | struct kmem_cache *sc; |
5e5e3e78 | 1338 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1339 | |
f4be6b43 N |
1340 | if (conf->mddev->gendisk) |
1341 | sprintf(conf->cache_name[0], | |
1342 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1343 | else | |
1344 | sprintf(conf->cache_name[0], | |
1345 | "raid%d-%p", conf->level, conf->mddev); | |
1346 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1347 | ||
ad01c9e3 N |
1348 | conf->active_name = 0; |
1349 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1350 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1351 | 0, 0, NULL); |
1da177e4 LT |
1352 | if (!sc) |
1353 | return 1; | |
1354 | conf->slab_cache = sc; | |
ad01c9e3 | 1355 | conf->pool_size = devs; |
16a53ecc | 1356 | while (num--) |
3f294f4f | 1357 | if (!grow_one_stripe(conf)) |
1da177e4 | 1358 | return 1; |
1da177e4 LT |
1359 | return 0; |
1360 | } | |
29269553 | 1361 | |
d6f38f31 DW |
1362 | /** |
1363 | * scribble_len - return the required size of the scribble region | |
1364 | * @num - total number of disks in the array | |
1365 | * | |
1366 | * The size must be enough to contain: | |
1367 | * 1/ a struct page pointer for each device in the array +2 | |
1368 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1369 | * (dma_map_page()) or page (page_address()) address. | |
1370 | * | |
1371 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1372 | * calculate over all devices (not just the data blocks), using zeros in place | |
1373 | * of the P and Q blocks. | |
1374 | */ | |
1375 | static size_t scribble_len(int num) | |
1376 | { | |
1377 | size_t len; | |
1378 | ||
1379 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1380 | ||
1381 | return len; | |
1382 | } | |
1383 | ||
ad01c9e3 N |
1384 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
1385 | { | |
1386 | /* Make all the stripes able to hold 'newsize' devices. | |
1387 | * New slots in each stripe get 'page' set to a new page. | |
1388 | * | |
1389 | * This happens in stages: | |
1390 | * 1/ create a new kmem_cache and allocate the required number of | |
1391 | * stripe_heads. | |
1392 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1393 | * to the new stripe_heads. This will have the side effect of | |
1394 | * freezing the array as once all stripe_heads have been collected, | |
1395 | * no IO will be possible. Old stripe heads are freed once their | |
1396 | * pages have been transferred over, and the old kmem_cache is | |
1397 | * freed when all stripes are done. | |
1398 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1399 | * we simple return a failre status - no need to clean anything up. | |
1400 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1401 | * If this fails, we don't bother trying the shrink the | |
1402 | * stripe_heads down again, we just leave them as they are. | |
1403 | * As each stripe_head is processed the new one is released into | |
1404 | * active service. | |
1405 | * | |
1406 | * Once step2 is started, we cannot afford to wait for a write, | |
1407 | * so we use GFP_NOIO allocations. | |
1408 | */ | |
1409 | struct stripe_head *osh, *nsh; | |
1410 | LIST_HEAD(newstripes); | |
1411 | struct disk_info *ndisks; | |
d6f38f31 | 1412 | unsigned long cpu; |
b5470dc5 | 1413 | int err; |
e18b890b | 1414 | struct kmem_cache *sc; |
ad01c9e3 N |
1415 | int i; |
1416 | ||
1417 | if (newsize <= conf->pool_size) | |
1418 | return 0; /* never bother to shrink */ | |
1419 | ||
b5470dc5 DW |
1420 | err = md_allow_write(conf->mddev); |
1421 | if (err) | |
1422 | return err; | |
2a2275d6 | 1423 | |
ad01c9e3 N |
1424 | /* Step 1 */ |
1425 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1426 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1427 | 0, 0, NULL); |
ad01c9e3 N |
1428 | if (!sc) |
1429 | return -ENOMEM; | |
1430 | ||
1431 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1432 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1433 | if (!nsh) |
1434 | break; | |
1435 | ||
ad01c9e3 | 1436 | nsh->raid_conf = conf; |
417b8d4a DW |
1437 | #ifdef CONFIG_MULTICORE_RAID456 |
1438 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1439 | #endif | |
ad01c9e3 N |
1440 | |
1441 | list_add(&nsh->lru, &newstripes); | |
1442 | } | |
1443 | if (i) { | |
1444 | /* didn't get enough, give up */ | |
1445 | while (!list_empty(&newstripes)) { | |
1446 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1447 | list_del(&nsh->lru); | |
1448 | kmem_cache_free(sc, nsh); | |
1449 | } | |
1450 | kmem_cache_destroy(sc); | |
1451 | return -ENOMEM; | |
1452 | } | |
1453 | /* Step 2 - Must use GFP_NOIO now. | |
1454 | * OK, we have enough stripes, start collecting inactive | |
1455 | * stripes and copying them over | |
1456 | */ | |
1457 | list_for_each_entry(nsh, &newstripes, lru) { | |
1458 | spin_lock_irq(&conf->device_lock); | |
1459 | wait_event_lock_irq(conf->wait_for_stripe, | |
1460 | !list_empty(&conf->inactive_list), | |
1461 | conf->device_lock, | |
482c0834 | 1462 | ); |
ad01c9e3 N |
1463 | osh = get_free_stripe(conf); |
1464 | spin_unlock_irq(&conf->device_lock); | |
1465 | atomic_set(&nsh->count, 1); | |
1466 | for(i=0; i<conf->pool_size; i++) | |
1467 | nsh->dev[i].page = osh->dev[i].page; | |
1468 | for( ; i<newsize; i++) | |
1469 | nsh->dev[i].page = NULL; | |
1470 | kmem_cache_free(conf->slab_cache, osh); | |
1471 | } | |
1472 | kmem_cache_destroy(conf->slab_cache); | |
1473 | ||
1474 | /* Step 3. | |
1475 | * At this point, we are holding all the stripes so the array | |
1476 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1477 | * conf->disks and the scribble region |
ad01c9e3 N |
1478 | */ |
1479 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1480 | if (ndisks) { | |
1481 | for (i=0; i<conf->raid_disks; i++) | |
1482 | ndisks[i] = conf->disks[i]; | |
1483 | kfree(conf->disks); | |
1484 | conf->disks = ndisks; | |
1485 | } else | |
1486 | err = -ENOMEM; | |
1487 | ||
d6f38f31 DW |
1488 | get_online_cpus(); |
1489 | conf->scribble_len = scribble_len(newsize); | |
1490 | for_each_present_cpu(cpu) { | |
1491 | struct raid5_percpu *percpu; | |
1492 | void *scribble; | |
1493 | ||
1494 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1495 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1496 | ||
1497 | if (scribble) { | |
1498 | kfree(percpu->scribble); | |
1499 | percpu->scribble = scribble; | |
1500 | } else { | |
1501 | err = -ENOMEM; | |
1502 | break; | |
1503 | } | |
1504 | } | |
1505 | put_online_cpus(); | |
1506 | ||
ad01c9e3 N |
1507 | /* Step 4, return new stripes to service */ |
1508 | while(!list_empty(&newstripes)) { | |
1509 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1510 | list_del_init(&nsh->lru); | |
d6f38f31 | 1511 | |
ad01c9e3 N |
1512 | for (i=conf->raid_disks; i < newsize; i++) |
1513 | if (nsh->dev[i].page == NULL) { | |
1514 | struct page *p = alloc_page(GFP_NOIO); | |
1515 | nsh->dev[i].page = p; | |
1516 | if (!p) | |
1517 | err = -ENOMEM; | |
1518 | } | |
1519 | release_stripe(nsh); | |
1520 | } | |
1521 | /* critical section pass, GFP_NOIO no longer needed */ | |
1522 | ||
1523 | conf->slab_cache = sc; | |
1524 | conf->active_name = 1-conf->active_name; | |
1525 | conf->pool_size = newsize; | |
1526 | return err; | |
1527 | } | |
1da177e4 | 1528 | |
3f294f4f | 1529 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1530 | { |
1531 | struct stripe_head *sh; | |
1532 | ||
3f294f4f N |
1533 | spin_lock_irq(&conf->device_lock); |
1534 | sh = get_free_stripe(conf); | |
1535 | spin_unlock_irq(&conf->device_lock); | |
1536 | if (!sh) | |
1537 | return 0; | |
78bafebd | 1538 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1539 | shrink_buffers(sh); |
3f294f4f N |
1540 | kmem_cache_free(conf->slab_cache, sh); |
1541 | atomic_dec(&conf->active_stripes); | |
1542 | return 1; | |
1543 | } | |
1544 | ||
1545 | static void shrink_stripes(raid5_conf_t *conf) | |
1546 | { | |
1547 | while (drop_one_stripe(conf)) | |
1548 | ; | |
1549 | ||
29fc7e3e N |
1550 | if (conf->slab_cache) |
1551 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1552 | conf->slab_cache = NULL; |
1553 | } | |
1554 | ||
6712ecf8 | 1555 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1556 | { |
99c0fb5f | 1557 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1558 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1559 | int disks = sh->disks, i; |
1da177e4 | 1560 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1561 | char b[BDEVNAME_SIZE]; |
1562 | mdk_rdev_t *rdev; | |
1da177e4 | 1563 | |
1da177e4 LT |
1564 | |
1565 | for (i=0 ; i<disks; i++) | |
1566 | if (bi == &sh->dev[i].req) | |
1567 | break; | |
1568 | ||
45b4233c DW |
1569 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1570 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1571 | uptodate); |
1572 | if (i == disks) { | |
1573 | BUG(); | |
6712ecf8 | 1574 | return; |
1da177e4 LT |
1575 | } |
1576 | ||
1577 | if (uptodate) { | |
1da177e4 | 1578 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1579 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1580 | rdev = conf->disks[i].rdev; |
8bda470e CD |
1581 | printk_ratelimited( |
1582 | KERN_INFO | |
1583 | "md/raid:%s: read error corrected" | |
1584 | " (%lu sectors at %llu on %s)\n", | |
1585 | mdname(conf->mddev), STRIPE_SECTORS, | |
1586 | (unsigned long long)(sh->sector | |
1587 | + rdev->data_offset), | |
1588 | bdevname(rdev->bdev, b)); | |
ddd5115f | 1589 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
1590 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1591 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1592 | } | |
ba22dcbf N |
1593 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1594 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1595 | } else { |
d6950432 | 1596 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1597 | int retry = 0; |
d6950432 N |
1598 | rdev = conf->disks[i].rdev; |
1599 | ||
1da177e4 | 1600 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1601 | atomic_inc(&rdev->read_errors); |
7b0bb536 | 1602 | if (conf->mddev->degraded >= conf->max_degraded) |
8bda470e CD |
1603 | printk_ratelimited( |
1604 | KERN_WARNING | |
1605 | "md/raid:%s: read error not correctable " | |
1606 | "(sector %llu on %s).\n", | |
1607 | mdname(conf->mddev), | |
1608 | (unsigned long long)(sh->sector | |
1609 | + rdev->data_offset), | |
1610 | bdn); | |
ba22dcbf | 1611 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1612 | /* Oh, no!!! */ |
8bda470e CD |
1613 | printk_ratelimited( |
1614 | KERN_WARNING | |
1615 | "md/raid:%s: read error NOT corrected!! " | |
1616 | "(sector %llu on %s).\n", | |
1617 | mdname(conf->mddev), | |
1618 | (unsigned long long)(sh->sector | |
1619 | + rdev->data_offset), | |
1620 | bdn); | |
d6950432 | 1621 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1622 | > conf->max_nr_stripes) |
14f8d26b | 1623 | printk(KERN_WARNING |
0c55e022 | 1624 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1625 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1626 | else |
1627 | retry = 1; | |
1628 | if (retry) | |
1629 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1630 | else { | |
4e5314b5 N |
1631 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1632 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1633 | md_error(conf->mddev, rdev); |
ba22dcbf | 1634 | } |
1da177e4 LT |
1635 | } |
1636 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1637 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1638 | set_bit(STRIPE_HANDLE, &sh->state); | |
1639 | release_stripe(sh); | |
1da177e4 LT |
1640 | } |
1641 | ||
d710e138 | 1642 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1643 | { |
99c0fb5f | 1644 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1645 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1646 | int disks = sh->disks, i; |
1da177e4 LT |
1647 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1648 | ||
1da177e4 LT |
1649 | for (i=0 ; i<disks; i++) |
1650 | if (bi == &sh->dev[i].req) | |
1651 | break; | |
1652 | ||
45b4233c | 1653 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1654 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1655 | uptodate); | |
1656 | if (i == disks) { | |
1657 | BUG(); | |
6712ecf8 | 1658 | return; |
1da177e4 LT |
1659 | } |
1660 | ||
1da177e4 LT |
1661 | if (!uptodate) |
1662 | md_error(conf->mddev, conf->disks[i].rdev); | |
1663 | ||
1664 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1665 | ||
1666 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1667 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1668 | release_stripe(sh); |
1da177e4 LT |
1669 | } |
1670 | ||
1671 | ||
784052ec | 1672 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1673 | |
784052ec | 1674 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1675 | { |
1676 | struct r5dev *dev = &sh->dev[i]; | |
1677 | ||
1678 | bio_init(&dev->req); | |
1679 | dev->req.bi_io_vec = &dev->vec; | |
1680 | dev->req.bi_vcnt++; | |
1681 | dev->req.bi_max_vecs++; | |
1682 | dev->vec.bv_page = dev->page; | |
1683 | dev->vec.bv_len = STRIPE_SIZE; | |
1684 | dev->vec.bv_offset = 0; | |
1685 | ||
1686 | dev->req.bi_sector = sh->sector; | |
1687 | dev->req.bi_private = sh; | |
1688 | ||
1689 | dev->flags = 0; | |
784052ec | 1690 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1691 | } |
1692 | ||
1693 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1694 | { | |
1695 | char b[BDEVNAME_SIZE]; | |
7b92813c | 1696 | raid5_conf_t *conf = mddev->private; |
0c55e022 | 1697 | pr_debug("raid456: error called\n"); |
1da177e4 | 1698 | |
6f8d0c77 N |
1699 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1700 | unsigned long flags; | |
1701 | spin_lock_irqsave(&conf->device_lock, flags); | |
1702 | mddev->degraded++; | |
1703 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1704 | /* | |
1705 | * if recovery was running, make sure it aborts. | |
1706 | */ | |
1707 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1da177e4 | 1708 | } |
de393cde | 1709 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
1710 | set_bit(Faulty, &rdev->flags); |
1711 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1712 | printk(KERN_ALERT | |
1713 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1714 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1715 | mdname(mddev), | |
1716 | bdevname(rdev->bdev, b), | |
1717 | mdname(mddev), | |
1718 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1719 | } |
1da177e4 LT |
1720 | |
1721 | /* | |
1722 | * Input: a 'big' sector number, | |
1723 | * Output: index of the data and parity disk, and the sector # in them. | |
1724 | */ | |
112bf897 | 1725 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1726 | int previous, int *dd_idx, |
1727 | struct stripe_head *sh) | |
1da177e4 | 1728 | { |
6e3b96ed | 1729 | sector_t stripe, stripe2; |
35f2a591 | 1730 | sector_t chunk_number; |
1da177e4 | 1731 | unsigned int chunk_offset; |
911d4ee8 | 1732 | int pd_idx, qd_idx; |
67cc2b81 | 1733 | int ddf_layout = 0; |
1da177e4 | 1734 | sector_t new_sector; |
e183eaed N |
1735 | int algorithm = previous ? conf->prev_algo |
1736 | : conf->algorithm; | |
09c9e5fa AN |
1737 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1738 | : conf->chunk_sectors; | |
112bf897 N |
1739 | int raid_disks = previous ? conf->previous_raid_disks |
1740 | : conf->raid_disks; | |
1741 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1742 | |
1743 | /* First compute the information on this sector */ | |
1744 | ||
1745 | /* | |
1746 | * Compute the chunk number and the sector offset inside the chunk | |
1747 | */ | |
1748 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1749 | chunk_number = r_sector; | |
1da177e4 LT |
1750 | |
1751 | /* | |
1752 | * Compute the stripe number | |
1753 | */ | |
35f2a591 N |
1754 | stripe = chunk_number; |
1755 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1756 | stripe2 = stripe; |
1da177e4 LT |
1757 | /* |
1758 | * Select the parity disk based on the user selected algorithm. | |
1759 | */ | |
84789554 | 1760 | pd_idx = qd_idx = -1; |
16a53ecc N |
1761 | switch(conf->level) { |
1762 | case 4: | |
911d4ee8 | 1763 | pd_idx = data_disks; |
16a53ecc N |
1764 | break; |
1765 | case 5: | |
e183eaed | 1766 | switch (algorithm) { |
1da177e4 | 1767 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1768 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1769 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1770 | (*dd_idx)++; |
1771 | break; | |
1772 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1773 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1774 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1775 | (*dd_idx)++; |
1776 | break; | |
1777 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1778 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1779 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1780 | break; |
1781 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1782 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1783 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 1784 | break; |
99c0fb5f N |
1785 | case ALGORITHM_PARITY_0: |
1786 | pd_idx = 0; | |
1787 | (*dd_idx)++; | |
1788 | break; | |
1789 | case ALGORITHM_PARITY_N: | |
1790 | pd_idx = data_disks; | |
1791 | break; | |
1da177e4 | 1792 | default: |
99c0fb5f | 1793 | BUG(); |
16a53ecc N |
1794 | } |
1795 | break; | |
1796 | case 6: | |
1797 | ||
e183eaed | 1798 | switch (algorithm) { |
16a53ecc | 1799 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1800 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1801 | qd_idx = pd_idx + 1; |
1802 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1803 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1804 | qd_idx = 0; |
1805 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1806 | (*dd_idx) += 2; /* D D P Q D */ |
1807 | break; | |
1808 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1809 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1810 | qd_idx = pd_idx + 1; |
1811 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1812 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1813 | qd_idx = 0; |
1814 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1815 | (*dd_idx) += 2; /* D D P Q D */ |
1816 | break; | |
1817 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1818 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1819 | qd_idx = (pd_idx + 1) % raid_disks; |
1820 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1821 | break; |
1822 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1823 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1824 | qd_idx = (pd_idx + 1) % raid_disks; |
1825 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1826 | break; |
99c0fb5f N |
1827 | |
1828 | case ALGORITHM_PARITY_0: | |
1829 | pd_idx = 0; | |
1830 | qd_idx = 1; | |
1831 | (*dd_idx) += 2; | |
1832 | break; | |
1833 | case ALGORITHM_PARITY_N: | |
1834 | pd_idx = data_disks; | |
1835 | qd_idx = data_disks + 1; | |
1836 | break; | |
1837 | ||
1838 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1839 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1840 | * of blocks for computing Q is different. | |
1841 | */ | |
6e3b96ed | 1842 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
1843 | qd_idx = pd_idx + 1; |
1844 | if (pd_idx == raid_disks-1) { | |
1845 | (*dd_idx)++; /* Q D D D P */ | |
1846 | qd_idx = 0; | |
1847 | } else if (*dd_idx >= pd_idx) | |
1848 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1849 | ddf_layout = 1; |
99c0fb5f N |
1850 | break; |
1851 | ||
1852 | case ALGORITHM_ROTATING_N_RESTART: | |
1853 | /* Same a left_asymmetric, by first stripe is | |
1854 | * D D D P Q rather than | |
1855 | * Q D D D P | |
1856 | */ | |
6e3b96ed N |
1857 | stripe2 += 1; |
1858 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
1859 | qd_idx = pd_idx + 1; |
1860 | if (pd_idx == raid_disks-1) { | |
1861 | (*dd_idx)++; /* Q D D D P */ | |
1862 | qd_idx = 0; | |
1863 | } else if (*dd_idx >= pd_idx) | |
1864 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1865 | ddf_layout = 1; |
99c0fb5f N |
1866 | break; |
1867 | ||
1868 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1869 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 1870 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
1871 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
1872 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1873 | ddf_layout = 1; |
99c0fb5f N |
1874 | break; |
1875 | ||
1876 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1877 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 1878 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1879 | if (*dd_idx >= pd_idx) |
1880 | (*dd_idx)++; | |
1881 | qd_idx = raid_disks - 1; | |
1882 | break; | |
1883 | ||
1884 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 1885 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1886 | if (*dd_idx >= pd_idx) |
1887 | (*dd_idx)++; | |
1888 | qd_idx = raid_disks - 1; | |
1889 | break; | |
1890 | ||
1891 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 1892 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1893 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1894 | qd_idx = raid_disks - 1; | |
1895 | break; | |
1896 | ||
1897 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 1898 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1899 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1900 | qd_idx = raid_disks - 1; | |
1901 | break; | |
1902 | ||
1903 | case ALGORITHM_PARITY_0_6: | |
1904 | pd_idx = 0; | |
1905 | (*dd_idx)++; | |
1906 | qd_idx = raid_disks - 1; | |
1907 | break; | |
1908 | ||
16a53ecc | 1909 | default: |
99c0fb5f | 1910 | BUG(); |
16a53ecc N |
1911 | } |
1912 | break; | |
1da177e4 LT |
1913 | } |
1914 | ||
911d4ee8 N |
1915 | if (sh) { |
1916 | sh->pd_idx = pd_idx; | |
1917 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1918 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1919 | } |
1da177e4 LT |
1920 | /* |
1921 | * Finally, compute the new sector number | |
1922 | */ | |
1923 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1924 | return new_sector; | |
1925 | } | |
1926 | ||
1927 | ||
784052ec | 1928 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1929 | { |
1930 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1931 | int raid_disks = sh->disks; |
1932 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1933 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
1934 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1935 | : conf->chunk_sectors; | |
e183eaed N |
1936 | int algorithm = previous ? conf->prev_algo |
1937 | : conf->algorithm; | |
1da177e4 LT |
1938 | sector_t stripe; |
1939 | int chunk_offset; | |
35f2a591 N |
1940 | sector_t chunk_number; |
1941 | int dummy1, dd_idx = i; | |
1da177e4 | 1942 | sector_t r_sector; |
911d4ee8 | 1943 | struct stripe_head sh2; |
1da177e4 | 1944 | |
16a53ecc | 1945 | |
1da177e4 LT |
1946 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1947 | stripe = new_sector; | |
1da177e4 | 1948 | |
16a53ecc N |
1949 | if (i == sh->pd_idx) |
1950 | return 0; | |
1951 | switch(conf->level) { | |
1952 | case 4: break; | |
1953 | case 5: | |
e183eaed | 1954 | switch (algorithm) { |
1da177e4 LT |
1955 | case ALGORITHM_LEFT_ASYMMETRIC: |
1956 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1957 | if (i > sh->pd_idx) | |
1958 | i--; | |
1959 | break; | |
1960 | case ALGORITHM_LEFT_SYMMETRIC: | |
1961 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1962 | if (i < sh->pd_idx) | |
1963 | i += raid_disks; | |
1964 | i -= (sh->pd_idx + 1); | |
1965 | break; | |
99c0fb5f N |
1966 | case ALGORITHM_PARITY_0: |
1967 | i -= 1; | |
1968 | break; | |
1969 | case ALGORITHM_PARITY_N: | |
1970 | break; | |
1da177e4 | 1971 | default: |
99c0fb5f | 1972 | BUG(); |
16a53ecc N |
1973 | } |
1974 | break; | |
1975 | case 6: | |
d0dabf7e | 1976 | if (i == sh->qd_idx) |
16a53ecc | 1977 | return 0; /* It is the Q disk */ |
e183eaed | 1978 | switch (algorithm) { |
16a53ecc N |
1979 | case ALGORITHM_LEFT_ASYMMETRIC: |
1980 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1981 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1982 | case ALGORITHM_ROTATING_N_RESTART: | |
1983 | if (sh->pd_idx == raid_disks-1) | |
1984 | i--; /* Q D D D P */ | |
16a53ecc N |
1985 | else if (i > sh->pd_idx) |
1986 | i -= 2; /* D D P Q D */ | |
1987 | break; | |
1988 | case ALGORITHM_LEFT_SYMMETRIC: | |
1989 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1990 | if (sh->pd_idx == raid_disks-1) | |
1991 | i--; /* Q D D D P */ | |
1992 | else { | |
1993 | /* D D P Q D */ | |
1994 | if (i < sh->pd_idx) | |
1995 | i += raid_disks; | |
1996 | i -= (sh->pd_idx + 2); | |
1997 | } | |
1998 | break; | |
99c0fb5f N |
1999 | case ALGORITHM_PARITY_0: |
2000 | i -= 2; | |
2001 | break; | |
2002 | case ALGORITHM_PARITY_N: | |
2003 | break; | |
2004 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2005 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2006 | if (sh->pd_idx == 0) |
2007 | i--; /* P D D D Q */ | |
e4424fee N |
2008 | else { |
2009 | /* D D Q P D */ | |
2010 | if (i < sh->pd_idx) | |
2011 | i += raid_disks; | |
2012 | i -= (sh->pd_idx + 1); | |
2013 | } | |
99c0fb5f N |
2014 | break; |
2015 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2016 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2017 | if (i > sh->pd_idx) | |
2018 | i--; | |
2019 | break; | |
2020 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2021 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2022 | if (i < sh->pd_idx) | |
2023 | i += data_disks + 1; | |
2024 | i -= (sh->pd_idx + 1); | |
2025 | break; | |
2026 | case ALGORITHM_PARITY_0_6: | |
2027 | i -= 1; | |
2028 | break; | |
16a53ecc | 2029 | default: |
99c0fb5f | 2030 | BUG(); |
16a53ecc N |
2031 | } |
2032 | break; | |
1da177e4 LT |
2033 | } |
2034 | ||
2035 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2036 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2037 | |
112bf897 | 2038 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2039 | previous, &dummy1, &sh2); |
911d4ee8 N |
2040 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2041 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2042 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2043 | mdname(conf->mddev)); | |
1da177e4 LT |
2044 | return 0; |
2045 | } | |
2046 | return r_sector; | |
2047 | } | |
2048 | ||
2049 | ||
600aa109 | 2050 | static void |
c0f7bddb | 2051 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2052 | int rcw, int expand) |
e33129d8 DW |
2053 | { |
2054 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
c0f7bddb YT |
2055 | raid5_conf_t *conf = sh->raid_conf; |
2056 | int level = conf->level; | |
e33129d8 DW |
2057 | |
2058 | if (rcw) { | |
2059 | /* if we are not expanding this is a proper write request, and | |
2060 | * there will be bios with new data to be drained into the | |
2061 | * stripe cache | |
2062 | */ | |
2063 | if (!expand) { | |
600aa109 DW |
2064 | sh->reconstruct_state = reconstruct_state_drain_run; |
2065 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2066 | } else | |
2067 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2068 | |
ac6b53b6 | 2069 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2070 | |
2071 | for (i = disks; i--; ) { | |
2072 | struct r5dev *dev = &sh->dev[i]; | |
2073 | ||
2074 | if (dev->towrite) { | |
2075 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2076 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2077 | if (!expand) |
2078 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2079 | s->locked++; |
e33129d8 DW |
2080 | } |
2081 | } | |
c0f7bddb | 2082 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2083 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2084 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2085 | } else { |
c0f7bddb | 2086 | BUG_ON(level == 6); |
e33129d8 DW |
2087 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2088 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2089 | ||
d8ee0728 | 2090 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2091 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2092 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2093 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2094 | |
2095 | for (i = disks; i--; ) { | |
2096 | struct r5dev *dev = &sh->dev[i]; | |
2097 | if (i == pd_idx) | |
2098 | continue; | |
2099 | ||
e33129d8 DW |
2100 | if (dev->towrite && |
2101 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2102 | test_bit(R5_Wantcompute, &dev->flags))) { |
2103 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2104 | set_bit(R5_LOCKED, &dev->flags); |
2105 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2106 | s->locked++; |
e33129d8 DW |
2107 | } |
2108 | } | |
2109 | } | |
2110 | ||
c0f7bddb | 2111 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2112 | * are in flight |
2113 | */ | |
2114 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2115 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2116 | s->locked++; |
e33129d8 | 2117 | |
c0f7bddb YT |
2118 | if (level == 6) { |
2119 | int qd_idx = sh->qd_idx; | |
2120 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2121 | ||
2122 | set_bit(R5_LOCKED, &dev->flags); | |
2123 | clear_bit(R5_UPTODATE, &dev->flags); | |
2124 | s->locked++; | |
2125 | } | |
2126 | ||
600aa109 | 2127 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2128 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2129 | s->locked, s->ops_request); |
e33129d8 | 2130 | } |
16a53ecc | 2131 | |
1da177e4 LT |
2132 | /* |
2133 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2134 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2135 | * The bi_next chain must be in order. |
2136 | */ | |
2137 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2138 | { | |
2139 | struct bio **bip; | |
2140 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 2141 | int firstwrite=0; |
1da177e4 | 2142 | |
cbe47ec5 | 2143 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2144 | (unsigned long long)bi->bi_sector, |
2145 | (unsigned long long)sh->sector); | |
2146 | ||
2147 | ||
1da177e4 | 2148 | spin_lock_irq(&conf->device_lock); |
72626685 | 2149 | if (forwrite) { |
1da177e4 | 2150 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2151 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2152 | firstwrite = 1; | |
2153 | } else | |
1da177e4 LT |
2154 | bip = &sh->dev[dd_idx].toread; |
2155 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2156 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2157 | goto overlap; | |
2158 | bip = & (*bip)->bi_next; | |
2159 | } | |
2160 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2161 | goto overlap; | |
2162 | ||
78bafebd | 2163 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2164 | if (*bip) |
2165 | bi->bi_next = *bip; | |
2166 | *bip = bi; | |
960e739d | 2167 | bi->bi_phys_segments++; |
72626685 | 2168 | |
1da177e4 LT |
2169 | if (forwrite) { |
2170 | /* check if page is covered */ | |
2171 | sector_t sector = sh->dev[dd_idx].sector; | |
2172 | for (bi=sh->dev[dd_idx].towrite; | |
2173 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2174 | bi && bi->bi_sector <= sector; | |
2175 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2176 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2177 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2178 | } | |
2179 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2180 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2181 | } | |
cbe47ec5 | 2182 | spin_unlock_irq(&conf->device_lock); |
cbe47ec5 N |
2183 | |
2184 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2185 | (unsigned long long)(*bip)->bi_sector, | |
2186 | (unsigned long long)sh->sector, dd_idx); | |
2187 | ||
2188 | if (conf->mddev->bitmap && firstwrite) { | |
2189 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2190 | STRIPE_SECTORS, 0); | |
2191 | sh->bm_seq = conf->seq_flush+1; | |
2192 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2193 | } | |
1da177e4 LT |
2194 | return 1; |
2195 | ||
2196 | overlap: | |
2197 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2198 | spin_unlock_irq(&conf->device_lock); | |
1da177e4 LT |
2199 | return 0; |
2200 | } | |
2201 | ||
29269553 N |
2202 | static void end_reshape(raid5_conf_t *conf); |
2203 | ||
911d4ee8 N |
2204 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
2205 | struct stripe_head *sh) | |
ccfcc3c1 | 2206 | { |
784052ec | 2207 | int sectors_per_chunk = |
09c9e5fa | 2208 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2209 | int dd_idx; |
2d2063ce | 2210 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2211 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2212 | |
112bf897 N |
2213 | raid5_compute_sector(conf, |
2214 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2215 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2216 | previous, |
911d4ee8 | 2217 | &dd_idx, sh); |
ccfcc3c1 N |
2218 | } |
2219 | ||
a4456856 | 2220 | static void |
1fe797e6 | 2221 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2222 | struct stripe_head_state *s, int disks, |
2223 | struct bio **return_bi) | |
2224 | { | |
2225 | int i; | |
2226 | for (i = disks; i--; ) { | |
2227 | struct bio *bi; | |
2228 | int bitmap_end = 0; | |
2229 | ||
2230 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2231 | mdk_rdev_t *rdev; | |
2232 | rcu_read_lock(); | |
2233 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2234 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2235 | /* multiple read failures in one stripe */ | |
2236 | md_error(conf->mddev, rdev); | |
2237 | rcu_read_unlock(); | |
2238 | } | |
2239 | spin_lock_irq(&conf->device_lock); | |
2240 | /* fail all writes first */ | |
2241 | bi = sh->dev[i].towrite; | |
2242 | sh->dev[i].towrite = NULL; | |
2243 | if (bi) { | |
2244 | s->to_write--; | |
2245 | bitmap_end = 1; | |
2246 | } | |
2247 | ||
2248 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2249 | wake_up(&conf->wait_for_overlap); | |
2250 | ||
2251 | while (bi && bi->bi_sector < | |
2252 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2253 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2254 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2255 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2256 | md_write_end(conf->mddev); |
2257 | bi->bi_next = *return_bi; | |
2258 | *return_bi = bi; | |
2259 | } | |
2260 | bi = nextbi; | |
2261 | } | |
2262 | /* and fail all 'written' */ | |
2263 | bi = sh->dev[i].written; | |
2264 | sh->dev[i].written = NULL; | |
2265 | if (bi) bitmap_end = 1; | |
2266 | while (bi && bi->bi_sector < | |
2267 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2268 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2269 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2270 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2271 | md_write_end(conf->mddev); |
2272 | bi->bi_next = *return_bi; | |
2273 | *return_bi = bi; | |
2274 | } | |
2275 | bi = bi2; | |
2276 | } | |
2277 | ||
b5e98d65 DW |
2278 | /* fail any reads if this device is non-operational and |
2279 | * the data has not reached the cache yet. | |
2280 | */ | |
2281 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2282 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2283 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2284 | bi = sh->dev[i].toread; |
2285 | sh->dev[i].toread = NULL; | |
2286 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2287 | wake_up(&conf->wait_for_overlap); | |
2288 | if (bi) s->to_read--; | |
2289 | while (bi && bi->bi_sector < | |
2290 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2291 | struct bio *nextbi = | |
2292 | r5_next_bio(bi, sh->dev[i].sector); | |
2293 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2294 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2295 | bi->bi_next = *return_bi; |
2296 | *return_bi = bi; | |
2297 | } | |
2298 | bi = nextbi; | |
2299 | } | |
2300 | } | |
2301 | spin_unlock_irq(&conf->device_lock); | |
2302 | if (bitmap_end) | |
2303 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2304 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
2305 | /* If we were in the middle of a write the parity block might |
2306 | * still be locked - so just clear all R5_LOCKED flags | |
2307 | */ | |
2308 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
2309 | } |
2310 | ||
8b3e6cdc DW |
2311 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2312 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2313 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2314 | } |
2315 | ||
93b3dbce | 2316 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
2317 | * to be read or computed to satisfy a request. |
2318 | * | |
2319 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 2320 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 2321 | */ |
93b3dbce N |
2322 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
2323 | int disk_idx, int disks) | |
a4456856 | 2324 | { |
5599becc | 2325 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2326 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2327 | &sh->dev[s->failed_num[1]] }; | |
5599becc | 2328 | |
93b3dbce | 2329 | /* is the data in this block needed, and can we get it? */ |
5599becc YT |
2330 | if (!test_bit(R5_LOCKED, &dev->flags) && |
2331 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2332 | (dev->toread || | |
2333 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2334 | s->syncing || s->expanding || | |
5d35e09c N |
2335 | (s->failed >= 1 && fdev[0]->toread) || |
2336 | (s->failed >= 2 && fdev[1]->toread) || | |
93b3dbce N |
2337 | (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite && |
2338 | !test_bit(R5_OVERWRITE, &fdev[0]->flags)) || | |
2339 | (sh->raid_conf->level == 6 && s->failed && s->to_write))) { | |
5599becc YT |
2340 | /* we would like to get this block, possibly by computing it, |
2341 | * otherwise read it if the backing disk is insync | |
2342 | */ | |
2343 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2344 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2345 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
2346 | (s->failed && (disk_idx == s->failed_num[0] || |
2347 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
2348 | /* have disk failed, and we're requested to fetch it; |
2349 | * do compute it | |
a4456856 | 2350 | */ |
5599becc YT |
2351 | pr_debug("Computing stripe %llu block %d\n", |
2352 | (unsigned long long)sh->sector, disk_idx); | |
2353 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2354 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2355 | set_bit(R5_Wantcompute, &dev->flags); | |
2356 | sh->ops.target = disk_idx; | |
2357 | sh->ops.target2 = -1; /* no 2nd target */ | |
2358 | s->req_compute = 1; | |
93b3dbce N |
2359 | /* Careful: from this point on 'uptodate' is in the eye |
2360 | * of raid_run_ops which services 'compute' operations | |
2361 | * before writes. R5_Wantcompute flags a block that will | |
2362 | * be R5_UPTODATE by the time it is needed for a | |
2363 | * subsequent operation. | |
2364 | */ | |
5599becc YT |
2365 | s->uptodate++; |
2366 | return 1; | |
2367 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2368 | /* Computing 2-failure is *very* expensive; only | |
2369 | * do it if failed >= 2 | |
2370 | */ | |
2371 | int other; | |
2372 | for (other = disks; other--; ) { | |
2373 | if (other == disk_idx) | |
2374 | continue; | |
2375 | if (!test_bit(R5_UPTODATE, | |
2376 | &sh->dev[other].flags)) | |
2377 | break; | |
a4456856 | 2378 | } |
5599becc YT |
2379 | BUG_ON(other < 0); |
2380 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2381 | (unsigned long long)sh->sector, | |
2382 | disk_idx, other); | |
2383 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2384 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2385 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2386 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2387 | sh->ops.target = disk_idx; | |
2388 | sh->ops.target2 = other; | |
2389 | s->uptodate += 2; | |
2390 | s->req_compute = 1; | |
2391 | return 1; | |
2392 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2393 | set_bit(R5_LOCKED, &dev->flags); | |
2394 | set_bit(R5_Wantread, &dev->flags); | |
2395 | s->locked++; | |
2396 | pr_debug("Reading block %d (sync=%d)\n", | |
2397 | disk_idx, s->syncing); | |
a4456856 DW |
2398 | } |
2399 | } | |
5599becc YT |
2400 | |
2401 | return 0; | |
2402 | } | |
2403 | ||
2404 | /** | |
93b3dbce | 2405 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 2406 | */ |
93b3dbce N |
2407 | static void handle_stripe_fill(struct stripe_head *sh, |
2408 | struct stripe_head_state *s, | |
2409 | int disks) | |
5599becc YT |
2410 | { |
2411 | int i; | |
2412 | ||
2413 | /* look for blocks to read/compute, skip this if a compute | |
2414 | * is already in flight, or if the stripe contents are in the | |
2415 | * midst of changing due to a write | |
2416 | */ | |
2417 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2418 | !sh->reconstruct_state) | |
2419 | for (i = disks; i--; ) | |
93b3dbce | 2420 | if (fetch_block(sh, s, i, disks)) |
5599becc | 2421 | break; |
a4456856 DW |
2422 | set_bit(STRIPE_HANDLE, &sh->state); |
2423 | } | |
2424 | ||
2425 | ||
1fe797e6 | 2426 | /* handle_stripe_clean_event |
a4456856 DW |
2427 | * any written block on an uptodate or failed drive can be returned. |
2428 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2429 | * never LOCKED, so we don't need to test 'failed' directly. | |
2430 | */ | |
1fe797e6 | 2431 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2432 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2433 | { | |
2434 | int i; | |
2435 | struct r5dev *dev; | |
2436 | ||
2437 | for (i = disks; i--; ) | |
2438 | if (sh->dev[i].written) { | |
2439 | dev = &sh->dev[i]; | |
2440 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2441 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2442 | /* We can return any write requests */ | |
2443 | struct bio *wbi, *wbi2; | |
2444 | int bitmap_end = 0; | |
45b4233c | 2445 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2446 | spin_lock_irq(&conf->device_lock); |
2447 | wbi = dev->written; | |
2448 | dev->written = NULL; | |
2449 | while (wbi && wbi->bi_sector < | |
2450 | dev->sector + STRIPE_SECTORS) { | |
2451 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2452 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2453 | md_write_end(conf->mddev); |
2454 | wbi->bi_next = *return_bi; | |
2455 | *return_bi = wbi; | |
2456 | } | |
2457 | wbi = wbi2; | |
2458 | } | |
2459 | if (dev->towrite == NULL) | |
2460 | bitmap_end = 1; | |
2461 | spin_unlock_irq(&conf->device_lock); | |
2462 | if (bitmap_end) | |
2463 | bitmap_endwrite(conf->mddev->bitmap, | |
2464 | sh->sector, | |
2465 | STRIPE_SECTORS, | |
2466 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2467 | 0); | |
2468 | } | |
2469 | } | |
8b3e6cdc DW |
2470 | |
2471 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2472 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2473 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2474 | } |
2475 | ||
c8ac1803 N |
2476 | static void handle_stripe_dirtying(raid5_conf_t *conf, |
2477 | struct stripe_head *sh, | |
2478 | struct stripe_head_state *s, | |
2479 | int disks) | |
a4456856 DW |
2480 | { |
2481 | int rmw = 0, rcw = 0, i; | |
c8ac1803 N |
2482 | if (conf->max_degraded == 2) { |
2483 | /* RAID6 requires 'rcw' in current implementation | |
2484 | * Calculate the real rcw later - for now fake it | |
2485 | * look like rcw is cheaper | |
2486 | */ | |
2487 | rcw = 1; rmw = 2; | |
2488 | } else for (i = disks; i--; ) { | |
a4456856 DW |
2489 | /* would I have to read this buffer for read_modify_write */ |
2490 | struct r5dev *dev = &sh->dev[i]; | |
2491 | if ((dev->towrite || i == sh->pd_idx) && | |
2492 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2493 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2494 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2495 | if (test_bit(R5_Insync, &dev->flags)) |
2496 | rmw++; | |
2497 | else | |
2498 | rmw += 2*disks; /* cannot read it */ | |
2499 | } | |
2500 | /* Would I have to read this buffer for reconstruct_write */ | |
2501 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2502 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2503 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2504 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2505 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2506 | else |
2507 | rcw += 2*disks; | |
2508 | } | |
2509 | } | |
45b4233c | 2510 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2511 | (unsigned long long)sh->sector, rmw, rcw); |
2512 | set_bit(STRIPE_HANDLE, &sh->state); | |
2513 | if (rmw < rcw && rmw > 0) | |
2514 | /* prefer read-modify-write, but need to get some data */ | |
2515 | for (i = disks; i--; ) { | |
2516 | struct r5dev *dev = &sh->dev[i]; | |
2517 | if ((dev->towrite || i == sh->pd_idx) && | |
2518 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2519 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2520 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2521 | test_bit(R5_Insync, &dev->flags)) { |
2522 | if ( | |
2523 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2524 | pr_debug("Read_old block " |
a4456856 DW |
2525 | "%d for r-m-w\n", i); |
2526 | set_bit(R5_LOCKED, &dev->flags); | |
2527 | set_bit(R5_Wantread, &dev->flags); | |
2528 | s->locked++; | |
2529 | } else { | |
2530 | set_bit(STRIPE_DELAYED, &sh->state); | |
2531 | set_bit(STRIPE_HANDLE, &sh->state); | |
2532 | } | |
2533 | } | |
2534 | } | |
c8ac1803 | 2535 | if (rcw <= rmw && rcw > 0) { |
a4456856 | 2536 | /* want reconstruct write, but need to get some data */ |
c8ac1803 | 2537 | rcw = 0; |
a4456856 DW |
2538 | for (i = disks; i--; ) { |
2539 | struct r5dev *dev = &sh->dev[i]; | |
2540 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 2541 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 2542 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 2543 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
2544 | test_bit(R5_Wantcompute, &dev->flags))) { |
2545 | rcw++; | |
2546 | if (!test_bit(R5_Insync, &dev->flags)) | |
2547 | continue; /* it's a failed drive */ | |
a4456856 DW |
2548 | if ( |
2549 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2550 | pr_debug("Read_old block " |
a4456856 DW |
2551 | "%d for Reconstruct\n", i); |
2552 | set_bit(R5_LOCKED, &dev->flags); | |
2553 | set_bit(R5_Wantread, &dev->flags); | |
2554 | s->locked++; | |
2555 | } else { | |
2556 | set_bit(STRIPE_DELAYED, &sh->state); | |
2557 | set_bit(STRIPE_HANDLE, &sh->state); | |
2558 | } | |
2559 | } | |
2560 | } | |
c8ac1803 | 2561 | } |
a4456856 DW |
2562 | /* now if nothing is locked, and if we have enough data, |
2563 | * we can start a write request | |
2564 | */ | |
f38e1219 DW |
2565 | /* since handle_stripe can be called at any time we need to handle the |
2566 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2567 | * subsequent call wants to start a write request. raid_run_ops only |
2568 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2569 | * simultaneously. If this is not the case then new writes need to be |
2570 | * held off until the compute completes. | |
2571 | */ | |
976ea8d4 DW |
2572 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2573 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2574 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2575 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2576 | } |
2577 | ||
a4456856 DW |
2578 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, |
2579 | struct stripe_head_state *s, int disks) | |
2580 | { | |
ecc65c9b | 2581 | struct r5dev *dev = NULL; |
bd2ab670 | 2582 | |
a4456856 | 2583 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2584 | |
ecc65c9b DW |
2585 | switch (sh->check_state) { |
2586 | case check_state_idle: | |
2587 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2588 | if (s->failed == 0) { |
bd2ab670 | 2589 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2590 | sh->check_state = check_state_run; |
2591 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2592 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2593 | s->uptodate--; |
ecc65c9b | 2594 | break; |
bd2ab670 | 2595 | } |
f2b3b44d | 2596 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
2597 | /* fall through */ |
2598 | case check_state_compute_result: | |
2599 | sh->check_state = check_state_idle; | |
2600 | if (!dev) | |
2601 | dev = &sh->dev[sh->pd_idx]; | |
2602 | ||
2603 | /* check that a write has not made the stripe insync */ | |
2604 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2605 | break; | |
c8894419 | 2606 | |
a4456856 | 2607 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2608 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2609 | BUG_ON(s->uptodate != disks); | |
2610 | ||
2611 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2612 | s->locked++; |
a4456856 | 2613 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2614 | |
a4456856 | 2615 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2616 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2617 | break; |
2618 | case check_state_run: | |
2619 | break; /* we will be called again upon completion */ | |
2620 | case check_state_check_result: | |
2621 | sh->check_state = check_state_idle; | |
2622 | ||
2623 | /* if a failure occurred during the check operation, leave | |
2624 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2625 | */ | |
2626 | if (s->failed) | |
2627 | break; | |
2628 | ||
2629 | /* handle a successful check operation, if parity is correct | |
2630 | * we are done. Otherwise update the mismatch count and repair | |
2631 | * parity if !MD_RECOVERY_CHECK | |
2632 | */ | |
ad283ea4 | 2633 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2634 | /* parity is correct (on disc, |
2635 | * not in buffer any more) | |
2636 | */ | |
2637 | set_bit(STRIPE_INSYNC, &sh->state); | |
2638 | else { | |
2639 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2640 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2641 | /* don't try to repair!! */ | |
2642 | set_bit(STRIPE_INSYNC, &sh->state); | |
2643 | else { | |
2644 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2645 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2646 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2647 | set_bit(R5_Wantcompute, | |
2648 | &sh->dev[sh->pd_idx].flags); | |
2649 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2650 | sh->ops.target2 = -1; |
ecc65c9b DW |
2651 | s->uptodate++; |
2652 | } | |
2653 | } | |
2654 | break; | |
2655 | case check_state_compute_run: | |
2656 | break; | |
2657 | default: | |
2658 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2659 | __func__, sh->check_state, | |
2660 | (unsigned long long) sh->sector); | |
2661 | BUG(); | |
a4456856 DW |
2662 | } |
2663 | } | |
2664 | ||
2665 | ||
2666 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
36d1c647 | 2667 | struct stripe_head_state *s, |
f2b3b44d | 2668 | int disks) |
a4456856 | 2669 | { |
a4456856 | 2670 | int pd_idx = sh->pd_idx; |
34e04e87 | 2671 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2672 | struct r5dev *dev; |
a4456856 DW |
2673 | |
2674 | set_bit(STRIPE_HANDLE, &sh->state); | |
2675 | ||
2676 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2677 | |
a4456856 DW |
2678 | /* Want to check and possibly repair P and Q. |
2679 | * However there could be one 'failed' device, in which | |
2680 | * case we can only check one of them, possibly using the | |
2681 | * other to generate missing data | |
2682 | */ | |
2683 | ||
d82dfee0 DW |
2684 | switch (sh->check_state) { |
2685 | case check_state_idle: | |
2686 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 2687 | if (s->failed == s->q_failed) { |
d82dfee0 | 2688 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2689 | * makes sense to check P (If anything else were failed, |
2690 | * we would have used P to recreate it). | |
2691 | */ | |
d82dfee0 | 2692 | sh->check_state = check_state_run; |
a4456856 | 2693 | } |
f2b3b44d | 2694 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 2695 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2696 | * anything, so it makes sense to check it |
2697 | */ | |
d82dfee0 DW |
2698 | if (sh->check_state == check_state_run) |
2699 | sh->check_state = check_state_run_pq; | |
2700 | else | |
2701 | sh->check_state = check_state_run_q; | |
a4456856 | 2702 | } |
a4456856 | 2703 | |
d82dfee0 DW |
2704 | /* discard potentially stale zero_sum_result */ |
2705 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2706 | |
d82dfee0 DW |
2707 | if (sh->check_state == check_state_run) { |
2708 | /* async_xor_zero_sum destroys the contents of P */ | |
2709 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2710 | s->uptodate--; | |
a4456856 | 2711 | } |
d82dfee0 DW |
2712 | if (sh->check_state >= check_state_run && |
2713 | sh->check_state <= check_state_run_pq) { | |
2714 | /* async_syndrome_zero_sum preserves P and Q, so | |
2715 | * no need to mark them !uptodate here | |
2716 | */ | |
2717 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2718 | break; | |
a4456856 DW |
2719 | } |
2720 | ||
d82dfee0 DW |
2721 | /* we have 2-disk failure */ |
2722 | BUG_ON(s->failed != 2); | |
2723 | /* fall through */ | |
2724 | case check_state_compute_result: | |
2725 | sh->check_state = check_state_idle; | |
a4456856 | 2726 | |
d82dfee0 DW |
2727 | /* check that a write has not made the stripe insync */ |
2728 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2729 | break; | |
a4456856 DW |
2730 | |
2731 | /* now write out any block on a failed drive, | |
d82dfee0 | 2732 | * or P or Q if they were recomputed |
a4456856 | 2733 | */ |
d82dfee0 | 2734 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 2735 | if (s->failed == 2) { |
f2b3b44d | 2736 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
2737 | s->locked++; |
2738 | set_bit(R5_LOCKED, &dev->flags); | |
2739 | set_bit(R5_Wantwrite, &dev->flags); | |
2740 | } | |
2741 | if (s->failed >= 1) { | |
f2b3b44d | 2742 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
2743 | s->locked++; |
2744 | set_bit(R5_LOCKED, &dev->flags); | |
2745 | set_bit(R5_Wantwrite, &dev->flags); | |
2746 | } | |
d82dfee0 | 2747 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2748 | dev = &sh->dev[pd_idx]; |
2749 | s->locked++; | |
2750 | set_bit(R5_LOCKED, &dev->flags); | |
2751 | set_bit(R5_Wantwrite, &dev->flags); | |
2752 | } | |
d82dfee0 | 2753 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2754 | dev = &sh->dev[qd_idx]; |
2755 | s->locked++; | |
2756 | set_bit(R5_LOCKED, &dev->flags); | |
2757 | set_bit(R5_Wantwrite, &dev->flags); | |
2758 | } | |
2759 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2760 | ||
2761 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2762 | break; |
2763 | case check_state_run: | |
2764 | case check_state_run_q: | |
2765 | case check_state_run_pq: | |
2766 | break; /* we will be called again upon completion */ | |
2767 | case check_state_check_result: | |
2768 | sh->check_state = check_state_idle; | |
2769 | ||
2770 | /* handle a successful check operation, if parity is correct | |
2771 | * we are done. Otherwise update the mismatch count and repair | |
2772 | * parity if !MD_RECOVERY_CHECK | |
2773 | */ | |
2774 | if (sh->ops.zero_sum_result == 0) { | |
2775 | /* both parities are correct */ | |
2776 | if (!s->failed) | |
2777 | set_bit(STRIPE_INSYNC, &sh->state); | |
2778 | else { | |
2779 | /* in contrast to the raid5 case we can validate | |
2780 | * parity, but still have a failure to write | |
2781 | * back | |
2782 | */ | |
2783 | sh->check_state = check_state_compute_result; | |
2784 | /* Returning at this point means that we may go | |
2785 | * off and bring p and/or q uptodate again so | |
2786 | * we make sure to check zero_sum_result again | |
2787 | * to verify if p or q need writeback | |
2788 | */ | |
2789 | } | |
2790 | } else { | |
2791 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2792 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2793 | /* don't try to repair!! */ | |
2794 | set_bit(STRIPE_INSYNC, &sh->state); | |
2795 | else { | |
2796 | int *target = &sh->ops.target; | |
2797 | ||
2798 | sh->ops.target = -1; | |
2799 | sh->ops.target2 = -1; | |
2800 | sh->check_state = check_state_compute_run; | |
2801 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2802 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2803 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2804 | set_bit(R5_Wantcompute, | |
2805 | &sh->dev[pd_idx].flags); | |
2806 | *target = pd_idx; | |
2807 | target = &sh->ops.target2; | |
2808 | s->uptodate++; | |
2809 | } | |
2810 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2811 | set_bit(R5_Wantcompute, | |
2812 | &sh->dev[qd_idx].flags); | |
2813 | *target = qd_idx; | |
2814 | s->uptodate++; | |
2815 | } | |
2816 | } | |
2817 | } | |
2818 | break; | |
2819 | case check_state_compute_run: | |
2820 | break; | |
2821 | default: | |
2822 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2823 | __func__, sh->check_state, | |
2824 | (unsigned long long) sh->sector); | |
2825 | BUG(); | |
a4456856 DW |
2826 | } |
2827 | } | |
2828 | ||
86c374ba | 2829 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh) |
a4456856 DW |
2830 | { |
2831 | int i; | |
2832 | ||
2833 | /* We have read all the blocks in this stripe and now we need to | |
2834 | * copy some of them into a target stripe for expand. | |
2835 | */ | |
f0a50d37 | 2836 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2837 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2838 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 2839 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 2840 | int dd_idx, j; |
a4456856 | 2841 | struct stripe_head *sh2; |
a08abd8c | 2842 | struct async_submit_ctl submit; |
a4456856 | 2843 | |
784052ec | 2844 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
2845 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2846 | &dd_idx, NULL); | |
a8c906ca | 2847 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
2848 | if (sh2 == NULL) |
2849 | /* so far only the early blocks of this stripe | |
2850 | * have been requested. When later blocks | |
2851 | * get requested, we will try again | |
2852 | */ | |
2853 | continue; | |
2854 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2855 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2856 | /* must have already done this block */ | |
2857 | release_stripe(sh2); | |
2858 | continue; | |
2859 | } | |
f0a50d37 DW |
2860 | |
2861 | /* place all the copies on one channel */ | |
a08abd8c | 2862 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 2863 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 2864 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 2865 | &submit); |
f0a50d37 | 2866 | |
a4456856 DW |
2867 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2868 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2869 | for (j = 0; j < conf->raid_disks; j++) | |
2870 | if (j != sh2->pd_idx && | |
86c374ba | 2871 | j != sh2->qd_idx && |
a4456856 DW |
2872 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2873 | break; | |
2874 | if (j == conf->raid_disks) { | |
2875 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2876 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2877 | } | |
2878 | release_stripe(sh2); | |
f0a50d37 | 2879 | |
a4456856 | 2880 | } |
a2e08551 N |
2881 | /* done submitting copies, wait for them to complete */ |
2882 | if (tx) { | |
2883 | async_tx_ack(tx); | |
2884 | dma_wait_for_async_tx(tx); | |
2885 | } | |
a4456856 | 2886 | } |
1da177e4 | 2887 | |
6bfe0b49 | 2888 | |
1da177e4 LT |
2889 | /* |
2890 | * handle_stripe - do things to a stripe. | |
2891 | * | |
2892 | * We lock the stripe and then examine the state of various bits | |
2893 | * to see what needs to be done. | |
2894 | * Possible results: | |
2895 | * return some read request which now have data | |
2896 | * return some write requests which are safely on disc | |
2897 | * schedule a read on some buffers | |
2898 | * schedule a write of some buffers | |
2899 | * return confirmation of parity correctness | |
2900 | * | |
1da177e4 LT |
2901 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2902 | * get BH_Lock set before the stripe lock is released. | |
2903 | * | |
2904 | */ | |
a4456856 | 2905 | |
acfe726b | 2906 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 2907 | { |
bff61975 | 2908 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 2909 | int disks = sh->disks; |
474af965 N |
2910 | struct r5dev *dev; |
2911 | int i; | |
1da177e4 | 2912 | |
acfe726b N |
2913 | memset(s, 0, sizeof(*s)); |
2914 | ||
2915 | s->syncing = test_bit(STRIPE_SYNCING, &sh->state); | |
2916 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2917 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
2918 | s->failed_num[0] = -1; | |
2919 | s->failed_num[1] = -1; | |
1da177e4 | 2920 | |
acfe726b | 2921 | /* Now to look around and see what can be done */ |
1da177e4 | 2922 | rcu_read_lock(); |
c4c1663b | 2923 | spin_lock_irq(&conf->device_lock); |
16a53ecc N |
2924 | for (i=disks; i--; ) { |
2925 | mdk_rdev_t *rdev; | |
acfe726b | 2926 | |
16a53ecc | 2927 | dev = &sh->dev[i]; |
1da177e4 | 2928 | |
45b4233c | 2929 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc | 2930 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
6c0069c0 YT |
2931 | /* maybe we can reply to a read |
2932 | * | |
2933 | * new wantfill requests are only permitted while | |
2934 | * ops_complete_biofill is guaranteed to be inactive | |
2935 | */ | |
2936 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
2937 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
2938 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 2939 | |
16a53ecc | 2940 | /* now count some things */ |
cc94015a N |
2941 | if (test_bit(R5_LOCKED, &dev->flags)) |
2942 | s->locked++; | |
2943 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
2944 | s->uptodate++; | |
2d6e4ecc | 2945 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
2946 | s->compute++; |
2947 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 2948 | } |
1da177e4 | 2949 | |
acfe726b | 2950 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 2951 | s->to_fill++; |
acfe726b | 2952 | else if (dev->toread) |
cc94015a | 2953 | s->to_read++; |
16a53ecc | 2954 | if (dev->towrite) { |
cc94015a | 2955 | s->to_write++; |
16a53ecc | 2956 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 2957 | s->non_overwrite++; |
16a53ecc | 2958 | } |
a4456856 | 2959 | if (dev->written) |
cc94015a | 2960 | s->written++; |
16a53ecc | 2961 | rdev = rcu_dereference(conf->disks[i].rdev); |
cc94015a | 2962 | if (s->blocked_rdev == NULL && |
ac4090d2 | 2963 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
cc94015a | 2964 | s->blocked_rdev = rdev; |
6bfe0b49 | 2965 | atomic_inc(&rdev->nr_pending); |
6bfe0b49 | 2966 | } |
415e72d0 N |
2967 | clear_bit(R5_Insync, &dev->flags); |
2968 | if (!rdev) | |
2969 | /* Not in-sync */; | |
2970 | else if (test_bit(In_sync, &rdev->flags)) | |
2971 | set_bit(R5_Insync, &dev->flags); | |
2972 | else { | |
2973 | /* in sync if before recovery_offset */ | |
2974 | if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) | |
2975 | set_bit(R5_Insync, &dev->flags); | |
2976 | } | |
2977 | if (!test_bit(R5_Insync, &dev->flags)) { | |
16a53ecc N |
2978 | /* The ReadError flag will just be confusing now */ |
2979 | clear_bit(R5_ReadError, &dev->flags); | |
2980 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 2981 | } |
415e72d0 N |
2982 | if (test_bit(R5_ReadError, &dev->flags)) |
2983 | clear_bit(R5_Insync, &dev->flags); | |
2984 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
2985 | if (s->failed < 2) |
2986 | s->failed_num[s->failed] = i; | |
2987 | s->failed++; | |
415e72d0 | 2988 | } |
1da177e4 | 2989 | } |
c4c1663b | 2990 | spin_unlock_irq(&conf->device_lock); |
1da177e4 | 2991 | rcu_read_unlock(); |
cc94015a N |
2992 | } |
2993 | ||
2994 | static void handle_stripe(struct stripe_head *sh) | |
2995 | { | |
2996 | struct stripe_head_state s; | |
474af965 | 2997 | raid5_conf_t *conf = sh->raid_conf; |
3687c061 | 2998 | int i; |
84789554 N |
2999 | int prexor; |
3000 | int disks = sh->disks; | |
474af965 | 3001 | struct r5dev *pdev, *qdev; |
cc94015a N |
3002 | |
3003 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3004 | if (test_and_set_bit(STRIPE_ACTIVE, &sh->state)) { | |
3005 | /* already being handled, ensure it gets handled | |
3006 | * again when current action finishes */ | |
3007 | set_bit(STRIPE_HANDLE, &sh->state); | |
3008 | return; | |
3009 | } | |
3010 | ||
3011 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3012 | set_bit(STRIPE_SYNCING, &sh->state); | |
3013 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3014 | } | |
3015 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3016 | ||
3017 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3018 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3019 | (unsigned long long)sh->sector, sh->state, | |
3020 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3021 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 3022 | |
acfe726b | 3023 | analyse_stripe(sh, &s); |
c5a31000 | 3024 | |
474af965 N |
3025 | if (unlikely(s.blocked_rdev)) { |
3026 | if (s.syncing || s.expanding || s.expanded || | |
3027 | s.to_write || s.written) { | |
3028 | set_bit(STRIPE_HANDLE, &sh->state); | |
3029 | goto finish; | |
3030 | } | |
3031 | /* There is nothing for the blocked_rdev to block */ | |
3032 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
3033 | s.blocked_rdev = NULL; | |
3034 | } | |
3035 | ||
3036 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
3037 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3038 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3039 | } | |
3040 | ||
3041 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
3042 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
3043 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
3044 | s.failed_num[0], s.failed_num[1]); | |
3045 | /* check if the array has lost more than max_degraded devices and, | |
3046 | * if so, some requests might need to be failed. | |
3047 | */ | |
3048 | if (s.failed > conf->max_degraded && s.to_read+s.to_write+s.written) | |
3049 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
3050 | if (s.failed > conf->max_degraded && s.syncing) { | |
3051 | md_done_sync(conf->mddev, STRIPE_SECTORS, 0); | |
3052 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3053 | s.syncing = 0; | |
3054 | } | |
3055 | ||
3056 | /* | |
3057 | * might be able to return some write requests if the parity blocks | |
3058 | * are safe, or on a failed drive | |
3059 | */ | |
3060 | pdev = &sh->dev[sh->pd_idx]; | |
3061 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
3062 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
3063 | qdev = &sh->dev[sh->qd_idx]; | |
3064 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
3065 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
3066 | || conf->level < 6; | |
3067 | ||
3068 | if (s.written && | |
3069 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
3070 | && !test_bit(R5_LOCKED, &pdev->flags) | |
3071 | && test_bit(R5_UPTODATE, &pdev->flags)))) && | |
3072 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
3073 | && !test_bit(R5_LOCKED, &qdev->flags) | |
3074 | && test_bit(R5_UPTODATE, &qdev->flags))))) | |
3075 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
3076 | ||
3077 | /* Now we might consider reading some blocks, either to check/generate | |
3078 | * parity, or to satisfy requests | |
3079 | * or to load a block that is being partially written. | |
3080 | */ | |
3081 | if (s.to_read || s.non_overwrite | |
3082 | || (conf->level == 6 && s.to_write && s.failed) | |
3083 | || (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) | |
3084 | handle_stripe_fill(sh, &s, disks); | |
3085 | ||
84789554 N |
3086 | /* Now we check to see if any write operations have recently |
3087 | * completed | |
3088 | */ | |
3089 | prexor = 0; | |
3090 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
3091 | prexor = 1; | |
3092 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
3093 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
3094 | sh->reconstruct_state = reconstruct_state_idle; | |
3095 | ||
3096 | /* All the 'written' buffers and the parity block are ready to | |
3097 | * be written back to disk | |
3098 | */ | |
3099 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3100 | BUG_ON(sh->qd_idx >= 0 && | |
3101 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags)); | |
3102 | for (i = disks; i--; ) { | |
3103 | struct r5dev *dev = &sh->dev[i]; | |
3104 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3105 | (i == sh->pd_idx || i == sh->qd_idx || | |
3106 | dev->written)) { | |
3107 | pr_debug("Writing block %d\n", i); | |
3108 | set_bit(R5_Wantwrite, &dev->flags); | |
3109 | if (prexor) | |
3110 | continue; | |
3111 | if (!test_bit(R5_Insync, &dev->flags) || | |
3112 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
3113 | s.failed == 0)) | |
3114 | set_bit(STRIPE_INSYNC, &sh->state); | |
3115 | } | |
3116 | } | |
3117 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3118 | s.dec_preread_active = 1; | |
3119 | } | |
3120 | ||
3121 | /* Now to consider new write requests and what else, if anything | |
3122 | * should be read. We do not handle new writes when: | |
3123 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3124 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3125 | * block. | |
3126 | */ | |
3127 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
3128 | handle_stripe_dirtying(conf, sh, &s, disks); | |
3129 | ||
3130 | /* maybe we need to check and possibly fix the parity for this stripe | |
3131 | * Any reads will already have been scheduled, so we just see if enough | |
3132 | * data is available. The parity check is held off while parity | |
3133 | * dependent operations are in flight. | |
3134 | */ | |
3135 | if (sh->check_state || | |
3136 | (s.syncing && s.locked == 0 && | |
3137 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3138 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
3139 | if (conf->level == 6) | |
3140 | handle_parity_checks6(conf, sh, &s, disks); | |
3141 | else | |
3142 | handle_parity_checks5(conf, sh, &s, disks); | |
3143 | } | |
c5a31000 N |
3144 | |
3145 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { | |
3146 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3147 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3148 | } | |
3149 | ||
3150 | /* If the failed drives are just a ReadError, then we might need | |
3151 | * to progress the repair/check process | |
3152 | */ | |
3153 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
3154 | for (i = 0; i < s.failed; i++) { | |
3155 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
3156 | if (test_bit(R5_ReadError, &dev->flags) | |
3157 | && !test_bit(R5_LOCKED, &dev->flags) | |
3158 | && test_bit(R5_UPTODATE, &dev->flags) | |
3159 | ) { | |
3160 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3161 | set_bit(R5_Wantwrite, &dev->flags); | |
3162 | set_bit(R5_ReWrite, &dev->flags); | |
3163 | set_bit(R5_LOCKED, &dev->flags); | |
3164 | s.locked++; | |
3165 | } else { | |
3166 | /* let's read it back */ | |
3167 | set_bit(R5_Wantread, &dev->flags); | |
3168 | set_bit(R5_LOCKED, &dev->flags); | |
3169 | s.locked++; | |
3170 | } | |
3171 | } | |
3172 | } | |
3173 | ||
3174 | ||
3687c061 N |
3175 | /* Finish reconstruct operations initiated by the expansion process */ |
3176 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3177 | struct stripe_head *sh_src | |
3178 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
3179 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
3180 | /* sh cannot be written until sh_src has been read. | |
3181 | * so arrange for sh to be delayed a little | |
3182 | */ | |
3183 | set_bit(STRIPE_DELAYED, &sh->state); | |
3184 | set_bit(STRIPE_HANDLE, &sh->state); | |
3185 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3186 | &sh_src->state)) | |
3187 | atomic_inc(&conf->preread_active_stripes); | |
3188 | release_stripe(sh_src); | |
3189 | goto finish; | |
3190 | } | |
3191 | if (sh_src) | |
3192 | release_stripe(sh_src); | |
3193 | ||
3194 | sh->reconstruct_state = reconstruct_state_idle; | |
3195 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3196 | for (i = conf->raid_disks; i--; ) { | |
3197 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3198 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3199 | s.locked++; | |
3200 | } | |
3201 | } | |
f416885e | 3202 | |
3687c061 N |
3203 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
3204 | !sh->reconstruct_state) { | |
3205 | /* Need to write out all blocks after computing parity */ | |
3206 | sh->disks = conf->raid_disks; | |
3207 | stripe_set_idx(sh->sector, conf, 0, sh); | |
3208 | schedule_reconstruction(sh, &s, 1, 1); | |
3209 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
3210 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
3211 | atomic_dec(&conf->reshape_stripes); | |
3212 | wake_up(&conf->wait_for_overlap); | |
3213 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3214 | } | |
3215 | ||
3216 | if (s.expanding && s.locked == 0 && | |
3217 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
3218 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 3219 | |
3687c061 | 3220 | finish: |
6bfe0b49 | 3221 | /* wait for this device to become unblocked */ |
c5709ef6 N |
3222 | if (unlikely(s.blocked_rdev)) |
3223 | md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev); | |
6bfe0b49 | 3224 | |
6c0069c0 YT |
3225 | if (s.ops_request) |
3226 | raid_run_ops(sh, s.ops_request); | |
3227 | ||
f0e43bcd | 3228 | ops_run_io(sh, &s); |
16a53ecc | 3229 | |
729a1866 | 3230 | |
c5709ef6 | 3231 | if (s.dec_preread_active) { |
729a1866 | 3232 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 3233 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3234 | * have actually been submitted. |
3235 | */ | |
3236 | atomic_dec(&conf->preread_active_stripes); | |
3237 | if (atomic_read(&conf->preread_active_stripes) < | |
3238 | IO_THRESHOLD) | |
3239 | md_wakeup_thread(conf->mddev->thread); | |
3240 | } | |
3241 | ||
c5709ef6 | 3242 | return_io(s.return_bi); |
16a53ecc | 3243 | |
c4c1663b | 3244 | clear_bit(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
3245 | } |
3246 | ||
16a53ecc N |
3247 | static void raid5_activate_delayed(raid5_conf_t *conf) |
3248 | { | |
3249 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3250 | while (!list_empty(&conf->delayed_list)) { | |
3251 | struct list_head *l = conf->delayed_list.next; | |
3252 | struct stripe_head *sh; | |
3253 | sh = list_entry(l, struct stripe_head, lru); | |
3254 | list_del_init(l); | |
3255 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3256 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3257 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3258 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3259 | } |
482c0834 | 3260 | } |
16a53ecc N |
3261 | } |
3262 | ||
3263 | static void activate_bit_delay(raid5_conf_t *conf) | |
3264 | { | |
3265 | /* device_lock is held */ | |
3266 | struct list_head head; | |
3267 | list_add(&head, &conf->bitmap_list); | |
3268 | list_del_init(&conf->bitmap_list); | |
3269 | while (!list_empty(&head)) { | |
3270 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3271 | list_del_init(&sh->lru); | |
3272 | atomic_inc(&sh->count); | |
3273 | __release_stripe(conf, sh); | |
3274 | } | |
3275 | } | |
3276 | ||
11d8a6e3 | 3277 | int md_raid5_congested(mddev_t *mddev, int bits) |
f022b2fd | 3278 | { |
070ec55d | 3279 | raid5_conf_t *conf = mddev->private; |
f022b2fd N |
3280 | |
3281 | /* No difference between reads and writes. Just check | |
3282 | * how busy the stripe_cache is | |
3283 | */ | |
3fa841d7 | 3284 | |
f022b2fd N |
3285 | if (conf->inactive_blocked) |
3286 | return 1; | |
3287 | if (conf->quiesce) | |
3288 | return 1; | |
3289 | if (list_empty_careful(&conf->inactive_list)) | |
3290 | return 1; | |
3291 | ||
3292 | return 0; | |
3293 | } | |
11d8a6e3 N |
3294 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3295 | ||
3296 | static int raid5_congested(void *data, int bits) | |
3297 | { | |
3298 | mddev_t *mddev = data; | |
3299 | ||
3300 | return mddev_congested(mddev, bits) || | |
3301 | md_raid5_congested(mddev, bits); | |
3302 | } | |
f022b2fd | 3303 | |
23032a0e RBJ |
3304 | /* We want read requests to align with chunks where possible, |
3305 | * but write requests don't need to. | |
3306 | */ | |
cc371e66 AK |
3307 | static int raid5_mergeable_bvec(struct request_queue *q, |
3308 | struct bvec_merge_data *bvm, | |
3309 | struct bio_vec *biovec) | |
23032a0e RBJ |
3310 | { |
3311 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3312 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3313 | int max; |
9d8f0363 | 3314 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3315 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3316 | |
cc371e66 | 3317 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3318 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3319 | ||
664e7c41 AN |
3320 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3321 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3322 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3323 | if (max < 0) max = 0; | |
3324 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3325 | return biovec->bv_len; | |
3326 | else | |
3327 | return max; | |
3328 | } | |
3329 | ||
f679623f RBJ |
3330 | |
3331 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3332 | { | |
3333 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3334 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3335 | unsigned int bio_sectors = bio->bi_size >> 9; |
3336 | ||
664e7c41 AN |
3337 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3338 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3339 | return chunk_sectors >= |
3340 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3341 | } | |
3342 | ||
46031f9a RBJ |
3343 | /* |
3344 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3345 | * later sampled by raid5d. | |
3346 | */ | |
3347 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3348 | { | |
3349 | unsigned long flags; | |
3350 | ||
3351 | spin_lock_irqsave(&conf->device_lock, flags); | |
3352 | ||
3353 | bi->bi_next = conf->retry_read_aligned_list; | |
3354 | conf->retry_read_aligned_list = bi; | |
3355 | ||
3356 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3357 | md_wakeup_thread(conf->mddev->thread); | |
3358 | } | |
3359 | ||
3360 | ||
3361 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3362 | { | |
3363 | struct bio *bi; | |
3364 | ||
3365 | bi = conf->retry_read_aligned; | |
3366 | if (bi) { | |
3367 | conf->retry_read_aligned = NULL; | |
3368 | return bi; | |
3369 | } | |
3370 | bi = conf->retry_read_aligned_list; | |
3371 | if(bi) { | |
387bb173 | 3372 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3373 | bi->bi_next = NULL; |
960e739d JA |
3374 | /* |
3375 | * this sets the active strip count to 1 and the processed | |
3376 | * strip count to zero (upper 8 bits) | |
3377 | */ | |
46031f9a | 3378 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3379 | } |
3380 | ||
3381 | return bi; | |
3382 | } | |
3383 | ||
3384 | ||
f679623f RBJ |
3385 | /* |
3386 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3387 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3388 | * first). | |
3389 | * If the read failed.. | |
3390 | */ | |
6712ecf8 | 3391 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3392 | { |
3393 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3394 | mddev_t *mddev; |
3395 | raid5_conf_t *conf; | |
3396 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3397 | mdk_rdev_t *rdev; | |
3398 | ||
f679623f | 3399 | bio_put(bi); |
46031f9a | 3400 | |
46031f9a RBJ |
3401 | rdev = (void*)raid_bi->bi_next; |
3402 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3403 | mddev = rdev->mddev; |
3404 | conf = mddev->private; | |
46031f9a RBJ |
3405 | |
3406 | rdev_dec_pending(rdev, conf->mddev); | |
3407 | ||
3408 | if (!error && uptodate) { | |
6712ecf8 | 3409 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3410 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3411 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3412 | return; |
46031f9a RBJ |
3413 | } |
3414 | ||
3415 | ||
45b4233c | 3416 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3417 | |
3418 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3419 | } |
3420 | ||
387bb173 NB |
3421 | static int bio_fits_rdev(struct bio *bi) |
3422 | { | |
165125e1 | 3423 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3424 | |
ae03bf63 | 3425 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3426 | return 0; |
3427 | blk_recount_segments(q, bi); | |
8a78362c | 3428 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3429 | return 0; |
3430 | ||
3431 | if (q->merge_bvec_fn) | |
3432 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3433 | * just just give up | |
3434 | */ | |
3435 | return 0; | |
3436 | ||
3437 | return 1; | |
3438 | } | |
3439 | ||
3440 | ||
21a52c6d | 3441 | static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio) |
f679623f | 3442 | { |
070ec55d | 3443 | raid5_conf_t *conf = mddev->private; |
8553fe7e | 3444 | int dd_idx; |
f679623f RBJ |
3445 | struct bio* align_bi; |
3446 | mdk_rdev_t *rdev; | |
3447 | ||
3448 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3449 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3450 | return 0; |
3451 | } | |
3452 | /* | |
a167f663 | 3453 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3454 | */ |
a167f663 | 3455 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3456 | if (!align_bi) |
3457 | return 0; | |
3458 | /* | |
3459 | * set bi_end_io to a new function, and set bi_private to the | |
3460 | * original bio. | |
3461 | */ | |
3462 | align_bi->bi_end_io = raid5_align_endio; | |
3463 | align_bi->bi_private = raid_bio; | |
3464 | /* | |
3465 | * compute position | |
3466 | */ | |
112bf897 N |
3467 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3468 | 0, | |
911d4ee8 | 3469 | &dd_idx, NULL); |
f679623f RBJ |
3470 | |
3471 | rcu_read_lock(); | |
3472 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3473 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3474 | atomic_inc(&rdev->nr_pending); |
3475 | rcu_read_unlock(); | |
46031f9a RBJ |
3476 | raid_bio->bi_next = (void*)rdev; |
3477 | align_bi->bi_bdev = rdev->bdev; | |
3478 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3479 | align_bi->bi_sector += rdev->data_offset; | |
3480 | ||
387bb173 NB |
3481 | if (!bio_fits_rdev(align_bi)) { |
3482 | /* too big in some way */ | |
3483 | bio_put(align_bi); | |
3484 | rdev_dec_pending(rdev, mddev); | |
3485 | return 0; | |
3486 | } | |
3487 | ||
46031f9a RBJ |
3488 | spin_lock_irq(&conf->device_lock); |
3489 | wait_event_lock_irq(conf->wait_for_stripe, | |
3490 | conf->quiesce == 0, | |
3491 | conf->device_lock, /* nothing */); | |
3492 | atomic_inc(&conf->active_aligned_reads); | |
3493 | spin_unlock_irq(&conf->device_lock); | |
3494 | ||
f679623f RBJ |
3495 | generic_make_request(align_bi); |
3496 | return 1; | |
3497 | } else { | |
3498 | rcu_read_unlock(); | |
46031f9a | 3499 | bio_put(align_bi); |
f679623f RBJ |
3500 | return 0; |
3501 | } | |
3502 | } | |
3503 | ||
8b3e6cdc DW |
3504 | /* __get_priority_stripe - get the next stripe to process |
3505 | * | |
3506 | * Full stripe writes are allowed to pass preread active stripes up until | |
3507 | * the bypass_threshold is exceeded. In general the bypass_count | |
3508 | * increments when the handle_list is handled before the hold_list; however, it | |
3509 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3510 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3511 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3512 | * handle_list. | |
3513 | */ | |
3514 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3515 | { | |
3516 | struct stripe_head *sh; | |
3517 | ||
3518 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3519 | __func__, | |
3520 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3521 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3522 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3523 | ||
3524 | if (!list_empty(&conf->handle_list)) { | |
3525 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3526 | ||
3527 | if (list_empty(&conf->hold_list)) | |
3528 | conf->bypass_count = 0; | |
3529 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3530 | if (conf->hold_list.next == conf->last_hold) | |
3531 | conf->bypass_count++; | |
3532 | else { | |
3533 | conf->last_hold = conf->hold_list.next; | |
3534 | conf->bypass_count -= conf->bypass_threshold; | |
3535 | if (conf->bypass_count < 0) | |
3536 | conf->bypass_count = 0; | |
3537 | } | |
3538 | } | |
3539 | } else if (!list_empty(&conf->hold_list) && | |
3540 | ((conf->bypass_threshold && | |
3541 | conf->bypass_count > conf->bypass_threshold) || | |
3542 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3543 | sh = list_entry(conf->hold_list.next, | |
3544 | typeof(*sh), lru); | |
3545 | conf->bypass_count -= conf->bypass_threshold; | |
3546 | if (conf->bypass_count < 0) | |
3547 | conf->bypass_count = 0; | |
3548 | } else | |
3549 | return NULL; | |
3550 | ||
3551 | list_del_init(&sh->lru); | |
3552 | atomic_inc(&sh->count); | |
3553 | BUG_ON(atomic_read(&sh->count) != 1); | |
3554 | return sh; | |
3555 | } | |
f679623f | 3556 | |
21a52c6d | 3557 | static int make_request(mddev_t *mddev, struct bio * bi) |
1da177e4 | 3558 | { |
070ec55d | 3559 | raid5_conf_t *conf = mddev->private; |
911d4ee8 | 3560 | int dd_idx; |
1da177e4 LT |
3561 | sector_t new_sector; |
3562 | sector_t logical_sector, last_sector; | |
3563 | struct stripe_head *sh; | |
a362357b | 3564 | const int rw = bio_data_dir(bi); |
49077326 | 3565 | int remaining; |
7c13edc8 | 3566 | int plugged; |
1da177e4 | 3567 | |
e9c7469b TH |
3568 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
3569 | md_flush_request(mddev, bi); | |
e5dcdd80 N |
3570 | return 0; |
3571 | } | |
3572 | ||
3d310eb7 | 3573 | md_write_start(mddev, bi); |
06d91a5f | 3574 | |
802ba064 | 3575 | if (rw == READ && |
52488615 | 3576 | mddev->reshape_position == MaxSector && |
21a52c6d | 3577 | chunk_aligned_read(mddev,bi)) |
99c0fb5f | 3578 | return 0; |
52488615 | 3579 | |
1da177e4 LT |
3580 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3581 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3582 | bi->bi_next = NULL; | |
3583 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3584 | |
7c13edc8 | 3585 | plugged = mddev_check_plugged(mddev); |
1da177e4 LT |
3586 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3587 | DEFINE_WAIT(w); | |
16a53ecc | 3588 | int disks, data_disks; |
b5663ba4 | 3589 | int previous; |
b578d55f | 3590 | |
7ecaa1e6 | 3591 | retry: |
b5663ba4 | 3592 | previous = 0; |
b0f9ec04 | 3593 | disks = conf->raid_disks; |
b578d55f | 3594 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3595 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3596 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3597 | * 64bit on a 32bit platform, and so it might be |
3598 | * possible to see a half-updated value | |
aeb878b0 | 3599 | * Of course reshape_progress could change after |
df8e7f76 N |
3600 | * the lock is dropped, so once we get a reference |
3601 | * to the stripe that we think it is, we will have | |
3602 | * to check again. | |
3603 | */ | |
7ecaa1e6 | 3604 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3605 | if (mddev->delta_disks < 0 |
3606 | ? logical_sector < conf->reshape_progress | |
3607 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3608 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3609 | previous = 1; |
3610 | } else { | |
fef9c61f N |
3611 | if (mddev->delta_disks < 0 |
3612 | ? logical_sector < conf->reshape_safe | |
3613 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3614 | spin_unlock_irq(&conf->device_lock); |
3615 | schedule(); | |
3616 | goto retry; | |
3617 | } | |
3618 | } | |
7ecaa1e6 N |
3619 | spin_unlock_irq(&conf->device_lock); |
3620 | } | |
16a53ecc N |
3621 | data_disks = disks - conf->max_degraded; |
3622 | ||
112bf897 N |
3623 | new_sector = raid5_compute_sector(conf, logical_sector, |
3624 | previous, | |
911d4ee8 | 3625 | &dd_idx, NULL); |
0c55e022 | 3626 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3627 | (unsigned long long)new_sector, |
3628 | (unsigned long long)logical_sector); | |
3629 | ||
b5663ba4 | 3630 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3631 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3632 | if (sh) { |
b0f9ec04 | 3633 | if (unlikely(previous)) { |
7ecaa1e6 | 3634 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3635 | * stripe, so we must do the range check again. |
3636 | * Expansion could still move past after this | |
3637 | * test, but as we are holding a reference to | |
3638 | * 'sh', we know that if that happens, | |
3639 | * STRIPE_EXPANDING will get set and the expansion | |
3640 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3641 | */ |
3642 | int must_retry = 0; | |
3643 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
3644 | if (mddev->delta_disks < 0 |
3645 | ? logical_sector >= conf->reshape_progress | |
3646 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
3647 | /* mismatch, need to try again */ |
3648 | must_retry = 1; | |
3649 | spin_unlock_irq(&conf->device_lock); | |
3650 | if (must_retry) { | |
3651 | release_stripe(sh); | |
7a3ab908 | 3652 | schedule(); |
7ecaa1e6 N |
3653 | goto retry; |
3654 | } | |
3655 | } | |
e62e58a5 | 3656 | |
ffd96e35 | 3657 | if (rw == WRITE && |
a5c308d4 | 3658 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
3659 | logical_sector < mddev->suspend_hi) { |
3660 | release_stripe(sh); | |
e62e58a5 N |
3661 | /* As the suspend_* range is controlled by |
3662 | * userspace, we want an interruptible | |
3663 | * wait. | |
3664 | */ | |
3665 | flush_signals(current); | |
3666 | prepare_to_wait(&conf->wait_for_overlap, | |
3667 | &w, TASK_INTERRUPTIBLE); | |
3668 | if (logical_sector >= mddev->suspend_lo && | |
3669 | logical_sector < mddev->suspend_hi) | |
3670 | schedule(); | |
e464eafd N |
3671 | goto retry; |
3672 | } | |
7ecaa1e6 N |
3673 | |
3674 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 3675 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
3676 | /* Stripe is busy expanding or |
3677 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3678 | * and wait a while |
3679 | */ | |
482c0834 | 3680 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
3681 | release_stripe(sh); |
3682 | schedule(); | |
3683 | goto retry; | |
3684 | } | |
3685 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3686 | set_bit(STRIPE_HANDLE, &sh->state); |
3687 | clear_bit(STRIPE_DELAYED, &sh->state); | |
e9c7469b | 3688 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
3689 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3690 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 3691 | release_stripe(sh); |
1da177e4 LT |
3692 | } else { |
3693 | /* cannot get stripe for read-ahead, just give-up */ | |
3694 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3695 | finish_wait(&conf->wait_for_overlap, &w); | |
3696 | break; | |
3697 | } | |
3698 | ||
3699 | } | |
7c13edc8 N |
3700 | if (!plugged) |
3701 | md_wakeup_thread(mddev->thread); | |
3702 | ||
1da177e4 | 3703 | spin_lock_irq(&conf->device_lock); |
960e739d | 3704 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3705 | spin_unlock_irq(&conf->device_lock); |
3706 | if (remaining == 0) { | |
1da177e4 | 3707 | |
16a53ecc | 3708 | if ( rw == WRITE ) |
1da177e4 | 3709 | md_write_end(mddev); |
6712ecf8 | 3710 | |
0e13fe23 | 3711 | bio_endio(bi, 0); |
1da177e4 | 3712 | } |
729a1866 | 3713 | |
1da177e4 LT |
3714 | return 0; |
3715 | } | |
3716 | ||
b522adcd DW |
3717 | static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks); |
3718 | ||
52c03291 | 3719 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3720 | { |
52c03291 N |
3721 | /* reshaping is quite different to recovery/resync so it is |
3722 | * handled quite separately ... here. | |
3723 | * | |
3724 | * On each call to sync_request, we gather one chunk worth of | |
3725 | * destination stripes and flag them as expanding. | |
3726 | * Then we find all the source stripes and request reads. | |
3727 | * As the reads complete, handle_stripe will copy the data | |
3728 | * into the destination stripe and release that stripe. | |
3729 | */ | |
7b92813c | 3730 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 3731 | struct stripe_head *sh; |
ccfcc3c1 | 3732 | sector_t first_sector, last_sector; |
f416885e N |
3733 | int raid_disks = conf->previous_raid_disks; |
3734 | int data_disks = raid_disks - conf->max_degraded; | |
3735 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3736 | int i; |
3737 | int dd_idx; | |
c8f517c4 | 3738 | sector_t writepos, readpos, safepos; |
ec32a2bd | 3739 | sector_t stripe_addr; |
7a661381 | 3740 | int reshape_sectors; |
ab69ae12 | 3741 | struct list_head stripes; |
52c03291 | 3742 | |
fef9c61f N |
3743 | if (sector_nr == 0) { |
3744 | /* If restarting in the middle, skip the initial sectors */ | |
3745 | if (mddev->delta_disks < 0 && | |
3746 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
3747 | sector_nr = raid5_size(mddev, 0, 0) | |
3748 | - conf->reshape_progress; | |
a639755c | 3749 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
3750 | conf->reshape_progress > 0) |
3751 | sector_nr = conf->reshape_progress; | |
f416885e | 3752 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 3753 | if (sector_nr) { |
8dee7211 N |
3754 | mddev->curr_resync_completed = sector_nr; |
3755 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
3756 | *skipped = 1; |
3757 | return sector_nr; | |
3758 | } | |
52c03291 N |
3759 | } |
3760 | ||
7a661381 N |
3761 | /* We need to process a full chunk at a time. |
3762 | * If old and new chunk sizes differ, we need to process the | |
3763 | * largest of these | |
3764 | */ | |
664e7c41 AN |
3765 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
3766 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 3767 | else |
9d8f0363 | 3768 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 3769 | |
52c03291 N |
3770 | /* we update the metadata when there is more than 3Meg |
3771 | * in the block range (that is rather arbitrary, should | |
3772 | * probably be time based) or when the data about to be | |
3773 | * copied would over-write the source of the data at | |
3774 | * the front of the range. | |
fef9c61f N |
3775 | * i.e. one new_stripe along from reshape_progress new_maps |
3776 | * to after where reshape_safe old_maps to | |
52c03291 | 3777 | */ |
fef9c61f | 3778 | writepos = conf->reshape_progress; |
f416885e | 3779 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
3780 | readpos = conf->reshape_progress; |
3781 | sector_div(readpos, data_disks); | |
fef9c61f | 3782 | safepos = conf->reshape_safe; |
f416885e | 3783 | sector_div(safepos, data_disks); |
fef9c61f | 3784 | if (mddev->delta_disks < 0) { |
ed37d83e | 3785 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 3786 | readpos += reshape_sectors; |
7a661381 | 3787 | safepos += reshape_sectors; |
fef9c61f | 3788 | } else { |
7a661381 | 3789 | writepos += reshape_sectors; |
ed37d83e N |
3790 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
3791 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 3792 | } |
52c03291 | 3793 | |
c8f517c4 N |
3794 | /* 'writepos' is the most advanced device address we might write. |
3795 | * 'readpos' is the least advanced device address we might read. | |
3796 | * 'safepos' is the least address recorded in the metadata as having | |
3797 | * been reshaped. | |
3798 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
3799 | * ensure safety in the face of a crash - that must be done by userspace | |
3800 | * making a backup of the data. So in that case there is no particular | |
3801 | * rush to update metadata. | |
3802 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
3803 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
3804 | * we can be safe in the event of a crash. | |
3805 | * So we insist on updating metadata if safepos is behind writepos and | |
3806 | * readpos is beyond writepos. | |
3807 | * In any case, update the metadata every 10 seconds. | |
3808 | * Maybe that number should be configurable, but I'm not sure it is | |
3809 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
3810 | */ | |
fef9c61f | 3811 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
3812 | ? (safepos > writepos && readpos < writepos) |
3813 | : (safepos < writepos && readpos > writepos)) || | |
3814 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
3815 | /* Cannot proceed until we've updated the superblock... */ |
3816 | wait_event(conf->wait_for_overlap, | |
3817 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 3818 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 3819 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 3820 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 3821 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 3822 | md_wakeup_thread(mddev->thread); |
850b2b42 | 3823 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
3824 | kthread_should_stop()); |
3825 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 3826 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
3827 | spin_unlock_irq(&conf->device_lock); |
3828 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 3829 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
3830 | } |
3831 | ||
ec32a2bd N |
3832 | if (mddev->delta_disks < 0) { |
3833 | BUG_ON(conf->reshape_progress == 0); | |
3834 | stripe_addr = writepos; | |
3835 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
3836 | ~((sector_t)reshape_sectors - 1)) |
3837 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
3838 | != sector_nr); |
3839 | } else { | |
7a661381 | 3840 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
3841 | stripe_addr = sector_nr; |
3842 | } | |
ab69ae12 | 3843 | INIT_LIST_HEAD(&stripes); |
7a661381 | 3844 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 3845 | int j; |
a9f326eb | 3846 | int skipped_disk = 0; |
a8c906ca | 3847 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
3848 | set_bit(STRIPE_EXPANDING, &sh->state); |
3849 | atomic_inc(&conf->reshape_stripes); | |
3850 | /* If any of this stripe is beyond the end of the old | |
3851 | * array, then we need to zero those blocks | |
3852 | */ | |
3853 | for (j=sh->disks; j--;) { | |
3854 | sector_t s; | |
3855 | if (j == sh->pd_idx) | |
3856 | continue; | |
f416885e | 3857 | if (conf->level == 6 && |
d0dabf7e | 3858 | j == sh->qd_idx) |
f416885e | 3859 | continue; |
784052ec | 3860 | s = compute_blocknr(sh, j, 0); |
b522adcd | 3861 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 3862 | skipped_disk = 1; |
52c03291 N |
3863 | continue; |
3864 | } | |
3865 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
3866 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
3867 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
3868 | } | |
a9f326eb | 3869 | if (!skipped_disk) { |
52c03291 N |
3870 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
3871 | set_bit(STRIPE_HANDLE, &sh->state); | |
3872 | } | |
ab69ae12 | 3873 | list_add(&sh->lru, &stripes); |
52c03291 N |
3874 | } |
3875 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 3876 | if (mddev->delta_disks < 0) |
7a661381 | 3877 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 3878 | else |
7a661381 | 3879 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
3880 | spin_unlock_irq(&conf->device_lock); |
3881 | /* Ok, those stripe are ready. We can start scheduling | |
3882 | * reads on the source stripes. | |
3883 | * The source stripes are determined by mapping the first and last | |
3884 | * block on the destination stripes. | |
3885 | */ | |
52c03291 | 3886 | first_sector = |
ec32a2bd | 3887 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 3888 | 1, &dd_idx, NULL); |
52c03291 | 3889 | last_sector = |
0e6e0271 | 3890 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 3891 | * new_data_disks - 1), |
911d4ee8 | 3892 | 1, &dd_idx, NULL); |
58c0fed4 AN |
3893 | if (last_sector >= mddev->dev_sectors) |
3894 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 3895 | while (first_sector <= last_sector) { |
a8c906ca | 3896 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
3897 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3898 | set_bit(STRIPE_HANDLE, &sh->state); | |
3899 | release_stripe(sh); | |
3900 | first_sector += STRIPE_SECTORS; | |
3901 | } | |
ab69ae12 N |
3902 | /* Now that the sources are clearly marked, we can release |
3903 | * the destination stripes | |
3904 | */ | |
3905 | while (!list_empty(&stripes)) { | |
3906 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
3907 | list_del_init(&sh->lru); | |
3908 | release_stripe(sh); | |
3909 | } | |
c6207277 N |
3910 | /* If this takes us to the resync_max point where we have to pause, |
3911 | * then we need to write out the superblock. | |
3912 | */ | |
7a661381 | 3913 | sector_nr += reshape_sectors; |
c03f6a19 N |
3914 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
3915 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
3916 | /* Cannot proceed until we've updated the superblock... */ |
3917 | wait_event(conf->wait_for_overlap, | |
3918 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 3919 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 3920 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 3921 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
3922 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
3923 | md_wakeup_thread(mddev->thread); | |
3924 | wait_event(mddev->sb_wait, | |
3925 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
3926 | || kthread_should_stop()); | |
3927 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 3928 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
3929 | spin_unlock_irq(&conf->device_lock); |
3930 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 3931 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 3932 | } |
7a661381 | 3933 | return reshape_sectors; |
52c03291 N |
3934 | } |
3935 | ||
3936 | /* FIXME go_faster isn't used */ | |
3937 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
3938 | { | |
7b92813c | 3939 | raid5_conf_t *conf = mddev->private; |
52c03291 | 3940 | struct stripe_head *sh; |
58c0fed4 | 3941 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 3942 | sector_t sync_blocks; |
16a53ecc N |
3943 | int still_degraded = 0; |
3944 | int i; | |
1da177e4 | 3945 | |
72626685 | 3946 | if (sector_nr >= max_sector) { |
1da177e4 | 3947 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 3948 | |
29269553 N |
3949 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3950 | end_reshape(conf); | |
3951 | return 0; | |
3952 | } | |
72626685 N |
3953 | |
3954 | if (mddev->curr_resync < max_sector) /* aborted */ | |
3955 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3956 | &sync_blocks, 1); | |
16a53ecc | 3957 | else /* completed sync */ |
72626685 N |
3958 | conf->fullsync = 0; |
3959 | bitmap_close_sync(mddev->bitmap); | |
3960 | ||
1da177e4 LT |
3961 | return 0; |
3962 | } | |
ccfcc3c1 | 3963 | |
64bd660b N |
3964 | /* Allow raid5_quiesce to complete */ |
3965 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
3966 | ||
52c03291 N |
3967 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3968 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 3969 | |
c6207277 N |
3970 | /* No need to check resync_max as we never do more than one |
3971 | * stripe, and as resync_max will always be on a chunk boundary, | |
3972 | * if the check in md_do_sync didn't fire, there is no chance | |
3973 | * of overstepping resync_max here | |
3974 | */ | |
3975 | ||
16a53ecc | 3976 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
3977 | * to resync, then assert that we are finished, because there is |
3978 | * nothing we can do. | |
3979 | */ | |
3285edf1 | 3980 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 3981 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 3982 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 3983 | *skipped = 1; |
1da177e4 LT |
3984 | return rv; |
3985 | } | |
72626685 | 3986 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 3987 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
3988 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
3989 | /* we can skip this block, and probably more */ | |
3990 | sync_blocks /= STRIPE_SECTORS; | |
3991 | *skipped = 1; | |
3992 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
3993 | } | |
1da177e4 | 3994 | |
b47490c9 N |
3995 | |
3996 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
3997 | ||
a8c906ca | 3998 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 3999 | if (sh == NULL) { |
a8c906ca | 4000 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4001 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4002 | * is trying to get access |
1da177e4 | 4003 | */ |
66c006a5 | 4004 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4005 | } |
16a53ecc N |
4006 | /* Need to check if array will still be degraded after recovery/resync |
4007 | * We don't need to check the 'failed' flag as when that gets set, | |
4008 | * recovery aborts. | |
4009 | */ | |
f001a70c | 4010 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4011 | if (conf->disks[i].rdev == NULL) |
4012 | still_degraded = 1; | |
4013 | ||
4014 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4015 | ||
83206d66 | 4016 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4017 | |
1442577b | 4018 | handle_stripe(sh); |
1da177e4 LT |
4019 | release_stripe(sh); |
4020 | ||
4021 | return STRIPE_SECTORS; | |
4022 | } | |
4023 | ||
46031f9a RBJ |
4024 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
4025 | { | |
4026 | /* We may not be able to submit a whole bio at once as there | |
4027 | * may not be enough stripe_heads available. | |
4028 | * We cannot pre-allocate enough stripe_heads as we may need | |
4029 | * more than exist in the cache (if we allow ever large chunks). | |
4030 | * So we do one stripe head at a time and record in | |
4031 | * ->bi_hw_segments how many have been done. | |
4032 | * | |
4033 | * We *know* that this entire raid_bio is in one chunk, so | |
4034 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4035 | */ | |
4036 | struct stripe_head *sh; | |
911d4ee8 | 4037 | int dd_idx; |
46031f9a RBJ |
4038 | sector_t sector, logical_sector, last_sector; |
4039 | int scnt = 0; | |
4040 | int remaining; | |
4041 | int handled = 0; | |
4042 | ||
4043 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4044 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4045 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4046 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4047 | ||
4048 | for (; logical_sector < last_sector; | |
387bb173 NB |
4049 | logical_sector += STRIPE_SECTORS, |
4050 | sector += STRIPE_SECTORS, | |
4051 | scnt++) { | |
46031f9a | 4052 | |
960e739d | 4053 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4054 | /* already done this stripe */ |
4055 | continue; | |
4056 | ||
a8c906ca | 4057 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4058 | |
4059 | if (!sh) { | |
4060 | /* failed to get a stripe - must wait */ | |
960e739d | 4061 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4062 | conf->retry_read_aligned = raid_bio; |
4063 | return handled; | |
4064 | } | |
4065 | ||
4066 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4067 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4068 | release_stripe(sh); | |
960e739d | 4069 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4070 | conf->retry_read_aligned = raid_bio; |
4071 | return handled; | |
4072 | } | |
4073 | ||
36d1c647 | 4074 | handle_stripe(sh); |
46031f9a RBJ |
4075 | release_stripe(sh); |
4076 | handled++; | |
4077 | } | |
4078 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4079 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4080 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4081 | if (remaining == 0) |
4082 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4083 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4084 | wake_up(&conf->wait_for_stripe); | |
4085 | return handled; | |
4086 | } | |
4087 | ||
46031f9a | 4088 | |
1da177e4 LT |
4089 | /* |
4090 | * This is our raid5 kernel thread. | |
4091 | * | |
4092 | * We scan the hash table for stripes which can be handled now. | |
4093 | * During the scan, completed stripes are saved for us by the interrupt | |
4094 | * handler, so that they will not have to wait for our next wakeup. | |
4095 | */ | |
6ed3003c | 4096 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4097 | { |
4098 | struct stripe_head *sh; | |
070ec55d | 4099 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 4100 | int handled; |
e1dfa0a2 | 4101 | struct blk_plug plug; |
1da177e4 | 4102 | |
45b4233c | 4103 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4104 | |
4105 | md_check_recovery(mddev); | |
1da177e4 | 4106 | |
e1dfa0a2 | 4107 | blk_start_plug(&plug); |
1da177e4 LT |
4108 | handled = 0; |
4109 | spin_lock_irq(&conf->device_lock); | |
4110 | while (1) { | |
46031f9a | 4111 | struct bio *bio; |
1da177e4 | 4112 | |
7c13edc8 N |
4113 | if (atomic_read(&mddev->plug_cnt) == 0 && |
4114 | !list_empty(&conf->bitmap_list)) { | |
4115 | /* Now is a good time to flush some bitmap updates */ | |
4116 | conf->seq_flush++; | |
700e432d | 4117 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4118 | bitmap_unplug(mddev->bitmap); |
700e432d | 4119 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4120 | conf->seq_write = conf->seq_flush; |
72626685 N |
4121 | activate_bit_delay(conf); |
4122 | } | |
7c13edc8 N |
4123 | if (atomic_read(&mddev->plug_cnt) == 0) |
4124 | raid5_activate_delayed(conf); | |
72626685 | 4125 | |
46031f9a RBJ |
4126 | while ((bio = remove_bio_from_retry(conf))) { |
4127 | int ok; | |
4128 | spin_unlock_irq(&conf->device_lock); | |
4129 | ok = retry_aligned_read(conf, bio); | |
4130 | spin_lock_irq(&conf->device_lock); | |
4131 | if (!ok) | |
4132 | break; | |
4133 | handled++; | |
4134 | } | |
4135 | ||
8b3e6cdc DW |
4136 | sh = __get_priority_stripe(conf); |
4137 | ||
c9f21aaf | 4138 | if (!sh) |
1da177e4 | 4139 | break; |
1da177e4 LT |
4140 | spin_unlock_irq(&conf->device_lock); |
4141 | ||
4142 | handled++; | |
417b8d4a DW |
4143 | handle_stripe(sh); |
4144 | release_stripe(sh); | |
4145 | cond_resched(); | |
1da177e4 | 4146 | |
de393cde N |
4147 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
4148 | md_check_recovery(mddev); | |
4149 | ||
1da177e4 LT |
4150 | spin_lock_irq(&conf->device_lock); |
4151 | } | |
45b4233c | 4152 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4153 | |
4154 | spin_unlock_irq(&conf->device_lock); | |
4155 | ||
c9f21aaf | 4156 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4157 | blk_finish_plug(&plug); |
1da177e4 | 4158 | |
45b4233c | 4159 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4160 | } |
4161 | ||
3f294f4f | 4162 | static ssize_t |
007583c9 | 4163 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4164 | { |
070ec55d | 4165 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4166 | if (conf) |
4167 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4168 | else | |
4169 | return 0; | |
3f294f4f N |
4170 | } |
4171 | ||
c41d4ac4 N |
4172 | int |
4173 | raid5_set_cache_size(mddev_t *mddev, int size) | |
3f294f4f | 4174 | { |
070ec55d | 4175 | raid5_conf_t *conf = mddev->private; |
b5470dc5 DW |
4176 | int err; |
4177 | ||
c41d4ac4 | 4178 | if (size <= 16 || size > 32768) |
3f294f4f | 4179 | return -EINVAL; |
c41d4ac4 | 4180 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4181 | if (drop_one_stripe(conf)) |
4182 | conf->max_nr_stripes--; | |
4183 | else | |
4184 | break; | |
4185 | } | |
b5470dc5 DW |
4186 | err = md_allow_write(mddev); |
4187 | if (err) | |
4188 | return err; | |
c41d4ac4 | 4189 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4190 | if (grow_one_stripe(conf)) |
4191 | conf->max_nr_stripes++; | |
4192 | else break; | |
4193 | } | |
c41d4ac4 N |
4194 | return 0; |
4195 | } | |
4196 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4197 | ||
4198 | static ssize_t | |
4199 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) | |
4200 | { | |
4201 | raid5_conf_t *conf = mddev->private; | |
4202 | unsigned long new; | |
4203 | int err; | |
4204 | ||
4205 | if (len >= PAGE_SIZE) | |
4206 | return -EINVAL; | |
4207 | if (!conf) | |
4208 | return -ENODEV; | |
4209 | ||
4210 | if (strict_strtoul(page, 10, &new)) | |
4211 | return -EINVAL; | |
4212 | err = raid5_set_cache_size(mddev, new); | |
4213 | if (err) | |
4214 | return err; | |
3f294f4f N |
4215 | return len; |
4216 | } | |
007583c9 | 4217 | |
96de1e66 N |
4218 | static struct md_sysfs_entry |
4219 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4220 | raid5_show_stripe_cache_size, | |
4221 | raid5_store_stripe_cache_size); | |
3f294f4f | 4222 | |
8b3e6cdc DW |
4223 | static ssize_t |
4224 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4225 | { | |
070ec55d | 4226 | raid5_conf_t *conf = mddev->private; |
8b3e6cdc DW |
4227 | if (conf) |
4228 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4229 | else | |
4230 | return 0; | |
4231 | } | |
4232 | ||
4233 | static ssize_t | |
4234 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4235 | { | |
070ec55d | 4236 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4237 | unsigned long new; |
8b3e6cdc DW |
4238 | if (len >= PAGE_SIZE) |
4239 | return -EINVAL; | |
4240 | if (!conf) | |
4241 | return -ENODEV; | |
4242 | ||
4ef197d8 | 4243 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4244 | return -EINVAL; |
4ef197d8 | 4245 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4246 | return -EINVAL; |
4247 | conf->bypass_threshold = new; | |
4248 | return len; | |
4249 | } | |
4250 | ||
4251 | static struct md_sysfs_entry | |
4252 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4253 | S_IRUGO | S_IWUSR, | |
4254 | raid5_show_preread_threshold, | |
4255 | raid5_store_preread_threshold); | |
4256 | ||
3f294f4f | 4257 | static ssize_t |
96de1e66 | 4258 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4259 | { |
070ec55d | 4260 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4261 | if (conf) |
4262 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4263 | else | |
4264 | return 0; | |
3f294f4f N |
4265 | } |
4266 | ||
96de1e66 N |
4267 | static struct md_sysfs_entry |
4268 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4269 | |
007583c9 | 4270 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4271 | &raid5_stripecache_size.attr, |
4272 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4273 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4274 | NULL, |
4275 | }; | |
007583c9 N |
4276 | static struct attribute_group raid5_attrs_group = { |
4277 | .name = NULL, | |
4278 | .attrs = raid5_attrs, | |
3f294f4f N |
4279 | }; |
4280 | ||
80c3a6ce DW |
4281 | static sector_t |
4282 | raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
4283 | { | |
070ec55d | 4284 | raid5_conf_t *conf = mddev->private; |
80c3a6ce DW |
4285 | |
4286 | if (!sectors) | |
4287 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4288 | if (!raid_disks) |
7ec05478 | 4289 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4290 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4291 | |
9d8f0363 | 4292 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4293 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4294 | return sectors * (raid_disks - conf->max_degraded); |
4295 | } | |
4296 | ||
36d1c647 DW |
4297 | static void raid5_free_percpu(raid5_conf_t *conf) |
4298 | { | |
4299 | struct raid5_percpu *percpu; | |
4300 | unsigned long cpu; | |
4301 | ||
4302 | if (!conf->percpu) | |
4303 | return; | |
4304 | ||
4305 | get_online_cpus(); | |
4306 | for_each_possible_cpu(cpu) { | |
4307 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4308 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4309 | kfree(percpu->scribble); |
36d1c647 DW |
4310 | } |
4311 | #ifdef CONFIG_HOTPLUG_CPU | |
4312 | unregister_cpu_notifier(&conf->cpu_notify); | |
4313 | #endif | |
4314 | put_online_cpus(); | |
4315 | ||
4316 | free_percpu(conf->percpu); | |
4317 | } | |
4318 | ||
95fc17aa DW |
4319 | static void free_conf(raid5_conf_t *conf) |
4320 | { | |
4321 | shrink_stripes(conf); | |
36d1c647 | 4322 | raid5_free_percpu(conf); |
95fc17aa DW |
4323 | kfree(conf->disks); |
4324 | kfree(conf->stripe_hashtbl); | |
4325 | kfree(conf); | |
4326 | } | |
4327 | ||
36d1c647 DW |
4328 | #ifdef CONFIG_HOTPLUG_CPU |
4329 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4330 | void *hcpu) | |
4331 | { | |
4332 | raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify); | |
4333 | long cpu = (long)hcpu; | |
4334 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4335 | ||
4336 | switch (action) { | |
4337 | case CPU_UP_PREPARE: | |
4338 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4339 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4340 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4341 | if (!percpu->scribble) |
4342 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4343 | ||
4344 | if (!percpu->scribble || | |
4345 | (conf->level == 6 && !percpu->spare_page)) { | |
4346 | safe_put_page(percpu->spare_page); | |
4347 | kfree(percpu->scribble); | |
36d1c647 DW |
4348 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4349 | __func__, cpu); | |
55af6bb5 | 4350 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4351 | } |
4352 | break; | |
4353 | case CPU_DEAD: | |
4354 | case CPU_DEAD_FROZEN: | |
4355 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4356 | kfree(percpu->scribble); |
36d1c647 | 4357 | percpu->spare_page = NULL; |
d6f38f31 | 4358 | percpu->scribble = NULL; |
36d1c647 DW |
4359 | break; |
4360 | default: | |
4361 | break; | |
4362 | } | |
4363 | return NOTIFY_OK; | |
4364 | } | |
4365 | #endif | |
4366 | ||
4367 | static int raid5_alloc_percpu(raid5_conf_t *conf) | |
4368 | { | |
4369 | unsigned long cpu; | |
4370 | struct page *spare_page; | |
a29d8b8e | 4371 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4372 | void *scribble; |
36d1c647 DW |
4373 | int err; |
4374 | ||
36d1c647 DW |
4375 | allcpus = alloc_percpu(struct raid5_percpu); |
4376 | if (!allcpus) | |
4377 | return -ENOMEM; | |
4378 | conf->percpu = allcpus; | |
4379 | ||
4380 | get_online_cpus(); | |
4381 | err = 0; | |
4382 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4383 | if (conf->level == 6) { |
4384 | spare_page = alloc_page(GFP_KERNEL); | |
4385 | if (!spare_page) { | |
4386 | err = -ENOMEM; | |
4387 | break; | |
4388 | } | |
4389 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4390 | } | |
5e5e3e78 | 4391 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4392 | if (!scribble) { |
36d1c647 DW |
4393 | err = -ENOMEM; |
4394 | break; | |
4395 | } | |
d6f38f31 | 4396 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4397 | } |
4398 | #ifdef CONFIG_HOTPLUG_CPU | |
4399 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4400 | conf->cpu_notify.priority = 0; | |
4401 | if (err == 0) | |
4402 | err = register_cpu_notifier(&conf->cpu_notify); | |
4403 | #endif | |
4404 | put_online_cpus(); | |
4405 | ||
4406 | return err; | |
4407 | } | |
4408 | ||
91adb564 | 4409 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4410 | { |
4411 | raid5_conf_t *conf; | |
5e5e3e78 | 4412 | int raid_disk, memory, max_disks; |
1da177e4 LT |
4413 | mdk_rdev_t *rdev; |
4414 | struct disk_info *disk; | |
1da177e4 | 4415 | |
91adb564 N |
4416 | if (mddev->new_level != 5 |
4417 | && mddev->new_level != 4 | |
4418 | && mddev->new_level != 6) { | |
0c55e022 | 4419 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4420 | mdname(mddev), mddev->new_level); |
4421 | return ERR_PTR(-EIO); | |
1da177e4 | 4422 | } |
91adb564 N |
4423 | if ((mddev->new_level == 5 |
4424 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4425 | (mddev->new_level == 6 | |
4426 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4427 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4428 | mdname(mddev), mddev->new_layout); |
4429 | return ERR_PTR(-EIO); | |
99c0fb5f | 4430 | } |
91adb564 | 4431 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4432 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4433 | mdname(mddev), mddev->raid_disks); |
4434 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4435 | } |
4436 | ||
664e7c41 AN |
4437 | if (!mddev->new_chunk_sectors || |
4438 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4439 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4440 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4441 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4442 | return ERR_PTR(-EINVAL); |
f6705578 N |
4443 | } |
4444 | ||
91adb564 N |
4445 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4446 | if (conf == NULL) | |
1da177e4 | 4447 | goto abort; |
f5efd45a DW |
4448 | spin_lock_init(&conf->device_lock); |
4449 | init_waitqueue_head(&conf->wait_for_stripe); | |
4450 | init_waitqueue_head(&conf->wait_for_overlap); | |
4451 | INIT_LIST_HEAD(&conf->handle_list); | |
4452 | INIT_LIST_HEAD(&conf->hold_list); | |
4453 | INIT_LIST_HEAD(&conf->delayed_list); | |
4454 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4455 | INIT_LIST_HEAD(&conf->inactive_list); | |
4456 | atomic_set(&conf->active_stripes, 0); | |
4457 | atomic_set(&conf->preread_active_stripes, 0); | |
4458 | atomic_set(&conf->active_aligned_reads, 0); | |
4459 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
91adb564 N |
4460 | |
4461 | conf->raid_disks = mddev->raid_disks; | |
4462 | if (mddev->reshape_position == MaxSector) | |
4463 | conf->previous_raid_disks = mddev->raid_disks; | |
4464 | else | |
f6705578 | 4465 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4466 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4467 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4468 | |
5e5e3e78 | 4469 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4470 | GFP_KERNEL); |
4471 | if (!conf->disks) | |
4472 | goto abort; | |
9ffae0cf | 4473 | |
1da177e4 LT |
4474 | conf->mddev = mddev; |
4475 | ||
fccddba0 | 4476 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4477 | goto abort; |
1da177e4 | 4478 | |
36d1c647 DW |
4479 | conf->level = mddev->new_level; |
4480 | if (raid5_alloc_percpu(conf) != 0) | |
4481 | goto abort; | |
4482 | ||
0c55e022 | 4483 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4484 | |
159ec1fc | 4485 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4486 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4487 | if (raid_disk >= max_disks |
1da177e4 LT |
4488 | || raid_disk < 0) |
4489 | continue; | |
4490 | disk = conf->disks + raid_disk; | |
4491 | ||
4492 | disk->rdev = rdev; | |
4493 | ||
b2d444d7 | 4494 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 4495 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
4496 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
4497 | " disk %d\n", | |
4498 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 4499 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
4500 | /* Cannot rely on bitmap to complete recovery */ |
4501 | conf->fullsync = 1; | |
1da177e4 LT |
4502 | } |
4503 | ||
09c9e5fa | 4504 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4505 | conf->level = mddev->new_level; |
16a53ecc N |
4506 | if (conf->level == 6) |
4507 | conf->max_degraded = 2; | |
4508 | else | |
4509 | conf->max_degraded = 1; | |
91adb564 | 4510 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4511 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4512 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4513 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4514 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4515 | conf->prev_algo = mddev->layout; |
4516 | } | |
1da177e4 | 4517 | |
91adb564 | 4518 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4519 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4520 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4521 | printk(KERN_ERR | |
0c55e022 N |
4522 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
4523 | mdname(mddev), memory); | |
91adb564 N |
4524 | goto abort; |
4525 | } else | |
0c55e022 N |
4526 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
4527 | mdname(mddev), memory); | |
1da177e4 | 4528 | |
0da3c619 | 4529 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4530 | if (!conf->thread) { |
4531 | printk(KERN_ERR | |
0c55e022 | 4532 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 4533 | mdname(mddev)); |
16a53ecc N |
4534 | goto abort; |
4535 | } | |
91adb564 N |
4536 | |
4537 | return conf; | |
4538 | ||
4539 | abort: | |
4540 | if (conf) { | |
95fc17aa | 4541 | free_conf(conf); |
91adb564 N |
4542 | return ERR_PTR(-EIO); |
4543 | } else | |
4544 | return ERR_PTR(-ENOMEM); | |
4545 | } | |
4546 | ||
c148ffdc N |
4547 | |
4548 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4549 | { | |
4550 | switch (algo) { | |
4551 | case ALGORITHM_PARITY_0: | |
4552 | if (raid_disk < max_degraded) | |
4553 | return 1; | |
4554 | break; | |
4555 | case ALGORITHM_PARITY_N: | |
4556 | if (raid_disk >= raid_disks - max_degraded) | |
4557 | return 1; | |
4558 | break; | |
4559 | case ALGORITHM_PARITY_0_6: | |
4560 | if (raid_disk == 0 || | |
4561 | raid_disk == raid_disks - 1) | |
4562 | return 1; | |
4563 | break; | |
4564 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4565 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4566 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4567 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4568 | if (raid_disk == raid_disks - 1) | |
4569 | return 1; | |
4570 | } | |
4571 | return 0; | |
4572 | } | |
4573 | ||
91adb564 N |
4574 | static int run(mddev_t *mddev) |
4575 | { | |
4576 | raid5_conf_t *conf; | |
9f7c2220 | 4577 | int working_disks = 0; |
c148ffdc | 4578 | int dirty_parity_disks = 0; |
91adb564 | 4579 | mdk_rdev_t *rdev; |
c148ffdc | 4580 | sector_t reshape_offset = 0; |
91adb564 | 4581 | |
8c6ac868 | 4582 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 4583 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
4584 | " -- starting background reconstruction\n", |
4585 | mdname(mddev)); | |
91adb564 N |
4586 | if (mddev->reshape_position != MaxSector) { |
4587 | /* Check that we can continue the reshape. | |
4588 | * Currently only disks can change, it must | |
4589 | * increase, and we must be past the point where | |
4590 | * a stripe over-writes itself | |
4591 | */ | |
4592 | sector_t here_new, here_old; | |
4593 | int old_disks; | |
18b00334 | 4594 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4595 | |
88ce4930 | 4596 | if (mddev->new_level != mddev->level) { |
0c55e022 | 4597 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
4598 | "required - aborting.\n", |
4599 | mdname(mddev)); | |
4600 | return -EINVAL; | |
4601 | } | |
91adb564 N |
4602 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4603 | /* reshape_position must be on a new-stripe boundary, and one | |
4604 | * further up in new geometry must map after here in old | |
4605 | * geometry. | |
4606 | */ | |
4607 | here_new = mddev->reshape_position; | |
664e7c41 | 4608 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 4609 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
4610 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
4611 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
4612 | return -EINVAL; |
4613 | } | |
c148ffdc | 4614 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4615 | /* here_new is the stripe we will write to */ |
4616 | here_old = mddev->reshape_position; | |
9d8f0363 | 4617 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4618 | (old_disks-max_degraded)); |
4619 | /* here_old is the first stripe that we might need to read | |
4620 | * from */ | |
67ac6011 N |
4621 | if (mddev->delta_disks == 0) { |
4622 | /* We cannot be sure it is safe to start an in-place | |
4623 | * reshape. It is only safe if user-space if monitoring | |
4624 | * and taking constant backups. | |
4625 | * mdadm always starts a situation like this in | |
4626 | * readonly mode so it can take control before | |
4627 | * allowing any writes. So just check for that. | |
4628 | */ | |
4629 | if ((here_new * mddev->new_chunk_sectors != | |
4630 | here_old * mddev->chunk_sectors) || | |
4631 | mddev->ro == 0) { | |
0c55e022 N |
4632 | printk(KERN_ERR "md/raid:%s: in-place reshape must be started" |
4633 | " in read-only mode - aborting\n", | |
4634 | mdname(mddev)); | |
67ac6011 N |
4635 | return -EINVAL; |
4636 | } | |
4637 | } else if (mddev->delta_disks < 0 | |
4638 | ? (here_new * mddev->new_chunk_sectors <= | |
4639 | here_old * mddev->chunk_sectors) | |
4640 | : (here_new * mddev->new_chunk_sectors >= | |
4641 | here_old * mddev->chunk_sectors)) { | |
91adb564 | 4642 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
4643 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
4644 | "auto-recovery - aborting.\n", | |
4645 | mdname(mddev)); | |
91adb564 N |
4646 | return -EINVAL; |
4647 | } | |
0c55e022 N |
4648 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
4649 | mdname(mddev)); | |
91adb564 N |
4650 | /* OK, we should be able to continue; */ |
4651 | } else { | |
4652 | BUG_ON(mddev->level != mddev->new_level); | |
4653 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 4654 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 4655 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 4656 | } |
91adb564 | 4657 | |
245f46c2 N |
4658 | if (mddev->private == NULL) |
4659 | conf = setup_conf(mddev); | |
4660 | else | |
4661 | conf = mddev->private; | |
4662 | ||
91adb564 N |
4663 | if (IS_ERR(conf)) |
4664 | return PTR_ERR(conf); | |
4665 | ||
4666 | mddev->thread = conf->thread; | |
4667 | conf->thread = NULL; | |
4668 | mddev->private = conf; | |
4669 | ||
4670 | /* | |
4671 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4672 | */ | |
c148ffdc | 4673 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
34b343cf N |
4674 | if (rdev->badblocks.count) { |
4675 | printk(KERN_ERR "md/raid5: cannot handle bad blocks yet\n"); | |
4676 | goto abort; | |
4677 | } | |
c148ffdc N |
4678 | if (rdev->raid_disk < 0) |
4679 | continue; | |
2f115882 | 4680 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 4681 | working_disks++; |
2f115882 N |
4682 | continue; |
4683 | } | |
c148ffdc N |
4684 | /* This disc is not fully in-sync. However if it |
4685 | * just stored parity (beyond the recovery_offset), | |
4686 | * when we don't need to be concerned about the | |
4687 | * array being dirty. | |
4688 | * When reshape goes 'backwards', we never have | |
4689 | * partially completed devices, so we only need | |
4690 | * to worry about reshape going forwards. | |
4691 | */ | |
4692 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
4693 | if (mddev->major_version == 0 && | |
4694 | mddev->minor_version > 90) | |
4695 | rdev->recovery_offset = reshape_offset; | |
4696 | ||
c148ffdc N |
4697 | if (rdev->recovery_offset < reshape_offset) { |
4698 | /* We need to check old and new layout */ | |
4699 | if (!only_parity(rdev->raid_disk, | |
4700 | conf->algorithm, | |
4701 | conf->raid_disks, | |
4702 | conf->max_degraded)) | |
4703 | continue; | |
4704 | } | |
4705 | if (!only_parity(rdev->raid_disk, | |
4706 | conf->prev_algo, | |
4707 | conf->previous_raid_disks, | |
4708 | conf->max_degraded)) | |
4709 | continue; | |
4710 | dirty_parity_disks++; | |
4711 | } | |
91adb564 | 4712 | |
5e5e3e78 N |
4713 | mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks) |
4714 | - working_disks); | |
91adb564 | 4715 | |
674806d6 | 4716 | if (has_failed(conf)) { |
0c55e022 | 4717 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 4718 | " (%d/%d failed)\n", |
02c2de8c | 4719 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4720 | goto abort; |
4721 | } | |
4722 | ||
91adb564 | 4723 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 4724 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
4725 | mddev->resync_max_sectors = mddev->dev_sectors; |
4726 | ||
c148ffdc | 4727 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 4728 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4729 | if (mddev->ok_start_degraded) |
4730 | printk(KERN_WARNING | |
0c55e022 N |
4731 | "md/raid:%s: starting dirty degraded array" |
4732 | " - data corruption possible.\n", | |
6ff8d8ec N |
4733 | mdname(mddev)); |
4734 | else { | |
4735 | printk(KERN_ERR | |
0c55e022 | 4736 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
4737 | mdname(mddev)); |
4738 | goto abort; | |
4739 | } | |
1da177e4 LT |
4740 | } |
4741 | ||
1da177e4 | 4742 | if (mddev->degraded == 0) |
0c55e022 N |
4743 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
4744 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
4745 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
4746 | mddev->new_layout); | |
1da177e4 | 4747 | else |
0c55e022 N |
4748 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
4749 | " out of %d devices, algorithm %d\n", | |
4750 | mdname(mddev), conf->level, | |
4751 | mddev->raid_disks - mddev->degraded, | |
4752 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
4753 | |
4754 | print_raid5_conf(conf); | |
4755 | ||
fef9c61f | 4756 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 4757 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
4758 | atomic_set(&conf->reshape_stripes, 0); |
4759 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4760 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4761 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4762 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4763 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 4764 | "reshape"); |
f6705578 N |
4765 | } |
4766 | ||
1da177e4 LT |
4767 | |
4768 | /* Ok, everything is just fine now */ | |
a64c876f N |
4769 | if (mddev->to_remove == &raid5_attrs_group) |
4770 | mddev->to_remove = NULL; | |
00bcb4ac N |
4771 | else if (mddev->kobj.sd && |
4772 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 4773 | printk(KERN_WARNING |
4a5add49 | 4774 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 4775 | mdname(mddev)); |
4a5add49 | 4776 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 4777 | |
4a5add49 | 4778 | if (mddev->queue) { |
9f7c2220 | 4779 | int chunk_size; |
4a5add49 N |
4780 | /* read-ahead size must cover two whole stripes, which |
4781 | * is 2 * (datadisks) * chunksize where 'n' is the | |
4782 | * number of raid devices | |
4783 | */ | |
4784 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
4785 | int stripe = data_disks * | |
4786 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
4787 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4788 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 4789 | |
4a5add49 | 4790 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 4791 | |
11d8a6e3 N |
4792 | mddev->queue->backing_dev_info.congested_data = mddev; |
4793 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 4794 | |
9f7c2220 N |
4795 | chunk_size = mddev->chunk_sectors << 9; |
4796 | blk_queue_io_min(mddev->queue, chunk_size); | |
4797 | blk_queue_io_opt(mddev->queue, chunk_size * | |
4798 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 4799 | |
9f7c2220 N |
4800 | list_for_each_entry(rdev, &mddev->disks, same_set) |
4801 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
4802 | rdev->data_offset << 9); | |
4803 | } | |
23032a0e | 4804 | |
1da177e4 LT |
4805 | return 0; |
4806 | abort: | |
e0cf8f04 | 4807 | md_unregister_thread(mddev->thread); |
91adb564 | 4808 | mddev->thread = NULL; |
1da177e4 LT |
4809 | if (conf) { |
4810 | print_raid5_conf(conf); | |
95fc17aa | 4811 | free_conf(conf); |
1da177e4 LT |
4812 | } |
4813 | mddev->private = NULL; | |
0c55e022 | 4814 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
4815 | return -EIO; |
4816 | } | |
4817 | ||
3f294f4f | 4818 | static int stop(mddev_t *mddev) |
1da177e4 | 4819 | { |
7b92813c | 4820 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
4821 | |
4822 | md_unregister_thread(mddev->thread); | |
4823 | mddev->thread = NULL; | |
11d8a6e3 N |
4824 | if (mddev->queue) |
4825 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 4826 | free_conf(conf); |
a64c876f N |
4827 | mddev->private = NULL; |
4828 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
4829 | return 0; |
4830 | } | |
4831 | ||
45b4233c | 4832 | #ifdef DEBUG |
d710e138 | 4833 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
4834 | { |
4835 | int i; | |
4836 | ||
16a53ecc N |
4837 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
4838 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
4839 | seq_printf(seq, "sh %llu, count %d.\n", | |
4840 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
4841 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 4842 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
4843 | seq_printf(seq, "(cache%d: %p %ld) ", |
4844 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 4845 | } |
16a53ecc | 4846 | seq_printf(seq, "\n"); |
1da177e4 LT |
4847 | } |
4848 | ||
d710e138 | 4849 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
4850 | { |
4851 | struct stripe_head *sh; | |
fccddba0 | 4852 | struct hlist_node *hn; |
1da177e4 LT |
4853 | int i; |
4854 | ||
4855 | spin_lock_irq(&conf->device_lock); | |
4856 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 4857 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
4858 | if (sh->raid_conf != conf) |
4859 | continue; | |
16a53ecc | 4860 | print_sh(seq, sh); |
1da177e4 LT |
4861 | } |
4862 | } | |
4863 | spin_unlock_irq(&conf->device_lock); | |
4864 | } | |
4865 | #endif | |
4866 | ||
d710e138 | 4867 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 | 4868 | { |
7b92813c | 4869 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
4870 | int i; |
4871 | ||
9d8f0363 AN |
4872 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
4873 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 4874 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
4875 | for (i = 0; i < conf->raid_disks; i++) |
4876 | seq_printf (seq, "%s", | |
4877 | conf->disks[i].rdev && | |
b2d444d7 | 4878 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 4879 | seq_printf (seq, "]"); |
45b4233c | 4880 | #ifdef DEBUG |
16a53ecc N |
4881 | seq_printf (seq, "\n"); |
4882 | printall(seq, conf); | |
1da177e4 LT |
4883 | #endif |
4884 | } | |
4885 | ||
4886 | static void print_raid5_conf (raid5_conf_t *conf) | |
4887 | { | |
4888 | int i; | |
4889 | struct disk_info *tmp; | |
4890 | ||
0c55e022 | 4891 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
4892 | if (!conf) { |
4893 | printk("(conf==NULL)\n"); | |
4894 | return; | |
4895 | } | |
0c55e022 N |
4896 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
4897 | conf->raid_disks, | |
4898 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
4899 | |
4900 | for (i = 0; i < conf->raid_disks; i++) { | |
4901 | char b[BDEVNAME_SIZE]; | |
4902 | tmp = conf->disks + i; | |
4903 | if (tmp->rdev) | |
0c55e022 N |
4904 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
4905 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
4906 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
4907 | } |
4908 | } | |
4909 | ||
4910 | static int raid5_spare_active(mddev_t *mddev) | |
4911 | { | |
4912 | int i; | |
4913 | raid5_conf_t *conf = mddev->private; | |
4914 | struct disk_info *tmp; | |
6b965620 N |
4915 | int count = 0; |
4916 | unsigned long flags; | |
1da177e4 LT |
4917 | |
4918 | for (i = 0; i < conf->raid_disks; i++) { | |
4919 | tmp = conf->disks + i; | |
4920 | if (tmp->rdev | |
70fffd0b | 4921 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 4922 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 4923 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 4924 | count++; |
43c73ca4 | 4925 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
4926 | } |
4927 | } | |
6b965620 N |
4928 | spin_lock_irqsave(&conf->device_lock, flags); |
4929 | mddev->degraded -= count; | |
4930 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 4931 | print_raid5_conf(conf); |
6b965620 | 4932 | return count; |
1da177e4 LT |
4933 | } |
4934 | ||
4935 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
4936 | { | |
4937 | raid5_conf_t *conf = mddev->private; | |
4938 | int err = 0; | |
4939 | mdk_rdev_t *rdev; | |
4940 | struct disk_info *p = conf->disks + number; | |
4941 | ||
4942 | print_raid5_conf(conf); | |
4943 | rdev = p->rdev; | |
4944 | if (rdev) { | |
ec32a2bd N |
4945 | if (number >= conf->raid_disks && |
4946 | conf->reshape_progress == MaxSector) | |
4947 | clear_bit(In_sync, &rdev->flags); | |
4948 | ||
b2d444d7 | 4949 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
4950 | atomic_read(&rdev->nr_pending)) { |
4951 | err = -EBUSY; | |
4952 | goto abort; | |
4953 | } | |
dfc70645 N |
4954 | /* Only remove non-faulty devices if recovery |
4955 | * isn't possible. | |
4956 | */ | |
4957 | if (!test_bit(Faulty, &rdev->flags) && | |
674806d6 | 4958 | !has_failed(conf) && |
ec32a2bd | 4959 | number < conf->raid_disks) { |
dfc70645 N |
4960 | err = -EBUSY; |
4961 | goto abort; | |
4962 | } | |
1da177e4 | 4963 | p->rdev = NULL; |
fbd568a3 | 4964 | synchronize_rcu(); |
1da177e4 LT |
4965 | if (atomic_read(&rdev->nr_pending)) { |
4966 | /* lost the race, try later */ | |
4967 | err = -EBUSY; | |
4968 | p->rdev = rdev; | |
4969 | } | |
4970 | } | |
4971 | abort: | |
4972 | ||
4973 | print_raid5_conf(conf); | |
4974 | return err; | |
4975 | } | |
4976 | ||
4977 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
4978 | { | |
4979 | raid5_conf_t *conf = mddev->private; | |
199050ea | 4980 | int err = -EEXIST; |
1da177e4 LT |
4981 | int disk; |
4982 | struct disk_info *p; | |
6c2fce2e NB |
4983 | int first = 0; |
4984 | int last = conf->raid_disks - 1; | |
1da177e4 | 4985 | |
34b343cf N |
4986 | if (rdev->badblocks.count) |
4987 | return -EINVAL; | |
4988 | ||
674806d6 | 4989 | if (has_failed(conf)) |
1da177e4 | 4990 | /* no point adding a device */ |
199050ea | 4991 | return -EINVAL; |
1da177e4 | 4992 | |
6c2fce2e NB |
4993 | if (rdev->raid_disk >= 0) |
4994 | first = last = rdev->raid_disk; | |
1da177e4 LT |
4995 | |
4996 | /* | |
16a53ecc N |
4997 | * find the disk ... but prefer rdev->saved_raid_disk |
4998 | * if possible. | |
1da177e4 | 4999 | */ |
16a53ecc | 5000 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5001 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5002 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5003 | disk = rdev->saved_raid_disk; | |
5004 | else | |
6c2fce2e NB |
5005 | disk = first; |
5006 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5007 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5008 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5009 | rdev->raid_disk = disk; |
199050ea | 5010 | err = 0; |
72626685 N |
5011 | if (rdev->saved_raid_disk != disk) |
5012 | conf->fullsync = 1; | |
d6065f7b | 5013 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5014 | break; |
5015 | } | |
5016 | print_raid5_conf(conf); | |
199050ea | 5017 | return err; |
1da177e4 LT |
5018 | } |
5019 | ||
5020 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
5021 | { | |
5022 | /* no resync is happening, and there is enough space | |
5023 | * on all devices, so we can resize. | |
5024 | * We need to make sure resync covers any new space. | |
5025 | * If the array is shrinking we should possibly wait until | |
5026 | * any io in the removed space completes, but it hardly seems | |
5027 | * worth it. | |
5028 | */ | |
9d8f0363 | 5029 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5030 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5031 | mddev->raid_disks)); | |
b522adcd DW |
5032 | if (mddev->array_sectors > |
5033 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5034 | return -EINVAL; | |
f233ea5c | 5035 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5036 | revalidate_disk(mddev->gendisk); |
b098636c N |
5037 | if (sectors > mddev->dev_sectors && |
5038 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5039 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5040 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5041 | } | |
58c0fed4 | 5042 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5043 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5044 | return 0; |
5045 | } | |
5046 | ||
01ee22b4 N |
5047 | static int check_stripe_cache(mddev_t *mddev) |
5048 | { | |
5049 | /* Can only proceed if there are plenty of stripe_heads. | |
5050 | * We need a minimum of one full stripe,, and for sensible progress | |
5051 | * it is best to have about 4 times that. | |
5052 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5053 | * allow for chunk sizes up to 256K, which is probably OK. | |
5054 | * If the chunk size is greater, user-space should request more | |
5055 | * stripe_heads first. | |
5056 | */ | |
5057 | raid5_conf_t *conf = mddev->private; | |
5058 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5059 | > conf->max_nr_stripes || | |
5060 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5061 | > conf->max_nr_stripes) { | |
0c55e022 N |
5062 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5063 | mdname(mddev), | |
01ee22b4 N |
5064 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5065 | / STRIPE_SIZE)*4); | |
5066 | return 0; | |
5067 | } | |
5068 | return 1; | |
5069 | } | |
5070 | ||
50ac168a | 5071 | static int check_reshape(mddev_t *mddev) |
29269553 | 5072 | { |
070ec55d | 5073 | raid5_conf_t *conf = mddev->private; |
29269553 | 5074 | |
88ce4930 N |
5075 | if (mddev->delta_disks == 0 && |
5076 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5077 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5078 | return 0; /* nothing to do */ |
dba034ee N |
5079 | if (mddev->bitmap) |
5080 | /* Cannot grow a bitmap yet */ | |
5081 | return -EBUSY; | |
674806d6 | 5082 | if (has_failed(conf)) |
ec32a2bd N |
5083 | return -EINVAL; |
5084 | if (mddev->delta_disks < 0) { | |
5085 | /* We might be able to shrink, but the devices must | |
5086 | * be made bigger first. | |
5087 | * For raid6, 4 is the minimum size. | |
5088 | * Otherwise 2 is the minimum | |
5089 | */ | |
5090 | int min = 2; | |
5091 | if (mddev->level == 6) | |
5092 | min = 4; | |
5093 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5094 | return -EINVAL; | |
5095 | } | |
29269553 | 5096 | |
01ee22b4 | 5097 | if (!check_stripe_cache(mddev)) |
29269553 | 5098 | return -ENOSPC; |
29269553 | 5099 | |
ec32a2bd | 5100 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5101 | } |
5102 | ||
5103 | static int raid5_start_reshape(mddev_t *mddev) | |
5104 | { | |
070ec55d | 5105 | raid5_conf_t *conf = mddev->private; |
63c70c4f | 5106 | mdk_rdev_t *rdev; |
63c70c4f | 5107 | int spares = 0; |
c04be0aa | 5108 | unsigned long flags; |
63c70c4f | 5109 | |
f416885e | 5110 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5111 | return -EBUSY; |
5112 | ||
01ee22b4 N |
5113 | if (!check_stripe_cache(mddev)) |
5114 | return -ENOSPC; | |
5115 | ||
159ec1fc | 5116 | list_for_each_entry(rdev, &mddev->disks, same_set) |
469518a3 N |
5117 | if (!test_bit(In_sync, &rdev->flags) |
5118 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5119 | spares++; |
63c70c4f | 5120 | |
f416885e | 5121 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5122 | /* Not enough devices even to make a degraded array |
5123 | * of that size | |
5124 | */ | |
5125 | return -EINVAL; | |
5126 | ||
ec32a2bd N |
5127 | /* Refuse to reduce size of the array. Any reductions in |
5128 | * array size must be through explicit setting of array_size | |
5129 | * attribute. | |
5130 | */ | |
5131 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5132 | < mddev->array_sectors) { | |
0c55e022 | 5133 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5134 | "before number of disks\n", mdname(mddev)); |
5135 | return -EINVAL; | |
5136 | } | |
5137 | ||
f6705578 | 5138 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5139 | spin_lock_irq(&conf->device_lock); |
5140 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5141 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5142 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5143 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5144 | conf->prev_algo = conf->algorithm; |
5145 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5146 | if (mddev->delta_disks < 0) |
5147 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5148 | else | |
5149 | conf->reshape_progress = 0; | |
5150 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5151 | conf->generation++; |
29269553 N |
5152 | spin_unlock_irq(&conf->device_lock); |
5153 | ||
5154 | /* Add some new drives, as many as will fit. | |
5155 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5156 | * Don't add devices if we are reducing the number of |
5157 | * devices in the array. This is because it is not possible | |
5158 | * to correctly record the "partially reconstructed" state of | |
5159 | * such devices during the reshape and confusion could result. | |
29269553 | 5160 | */ |
87a8dec9 N |
5161 | if (mddev->delta_disks >= 0) { |
5162 | int added_devices = 0; | |
5163 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5164 | if (rdev->raid_disk < 0 && | |
5165 | !test_bit(Faulty, &rdev->flags)) { | |
5166 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 N |
5167 | if (rdev->raid_disk |
5168 | >= conf->previous_raid_disks) { | |
5169 | set_bit(In_sync, &rdev->flags); | |
5170 | added_devices++; | |
5171 | } else | |
5172 | rdev->recovery_offset = 0; | |
36fad858 NK |
5173 | |
5174 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 5175 | /* Failure here is OK */; |
50da0840 | 5176 | } |
87a8dec9 N |
5177 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5178 | && !test_bit(Faulty, &rdev->flags)) { | |
5179 | /* This is a spare that was manually added */ | |
5180 | set_bit(In_sync, &rdev->flags); | |
5181 | added_devices++; | |
5182 | } | |
29269553 | 5183 | |
87a8dec9 N |
5184 | /* When a reshape changes the number of devices, |
5185 | * ->degraded is measured against the larger of the | |
5186 | * pre and post number of devices. | |
5187 | */ | |
ec32a2bd | 5188 | spin_lock_irqsave(&conf->device_lock, flags); |
9eb07c25 | 5189 | mddev->degraded += (conf->raid_disks - conf->previous_raid_disks) |
ec32a2bd N |
5190 | - added_devices; |
5191 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5192 | } | |
63c70c4f | 5193 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5194 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5195 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5196 | |
29269553 N |
5197 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5198 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5199 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5200 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5201 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5202 | "reshape"); |
29269553 N |
5203 | if (!mddev->sync_thread) { |
5204 | mddev->recovery = 0; | |
5205 | spin_lock_irq(&conf->device_lock); | |
5206 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5207 | conf->reshape_progress = MaxSector; |
29269553 N |
5208 | spin_unlock_irq(&conf->device_lock); |
5209 | return -EAGAIN; | |
5210 | } | |
c8f517c4 | 5211 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5212 | md_wakeup_thread(mddev->sync_thread); |
5213 | md_new_event(mddev); | |
5214 | return 0; | |
5215 | } | |
29269553 | 5216 | |
ec32a2bd N |
5217 | /* This is called from the reshape thread and should make any |
5218 | * changes needed in 'conf' | |
5219 | */ | |
29269553 N |
5220 | static void end_reshape(raid5_conf_t *conf) |
5221 | { | |
29269553 | 5222 | |
f6705578 | 5223 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5224 | |
f6705578 | 5225 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5226 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5227 | conf->reshape_progress = MaxSector; |
f6705578 | 5228 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5229 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5230 | |
5231 | /* read-ahead size must cover two whole stripes, which is | |
5232 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5233 | */ | |
4a5add49 | 5234 | if (conf->mddev->queue) { |
cea9c228 | 5235 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5236 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5237 | / PAGE_SIZE); |
16a53ecc N |
5238 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5239 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5240 | } | |
29269553 | 5241 | } |
29269553 N |
5242 | } |
5243 | ||
ec32a2bd N |
5244 | /* This is called from the raid5d thread with mddev_lock held. |
5245 | * It makes config changes to the device. | |
5246 | */ | |
cea9c228 N |
5247 | static void raid5_finish_reshape(mddev_t *mddev) |
5248 | { | |
070ec55d | 5249 | raid5_conf_t *conf = mddev->private; |
cea9c228 N |
5250 | |
5251 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5252 | ||
ec32a2bd N |
5253 | if (mddev->delta_disks > 0) { |
5254 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5255 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5256 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5257 | } else { |
5258 | int d; | |
ec32a2bd N |
5259 | mddev->degraded = conf->raid_disks; |
5260 | for (d = 0; d < conf->raid_disks ; d++) | |
5261 | if (conf->disks[d].rdev && | |
5262 | test_bit(In_sync, | |
5263 | &conf->disks[d].rdev->flags)) | |
5264 | mddev->degraded--; | |
5265 | for (d = conf->raid_disks ; | |
5266 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 N |
5267 | d++) { |
5268 | mdk_rdev_t *rdev = conf->disks[d].rdev; | |
5269 | if (rdev && raid5_remove_disk(mddev, d) == 0) { | |
36fad858 | 5270 | sysfs_unlink_rdev(mddev, rdev); |
1a67dde0 N |
5271 | rdev->raid_disk = -1; |
5272 | } | |
5273 | } | |
cea9c228 | 5274 | } |
88ce4930 | 5275 | mddev->layout = conf->algorithm; |
09c9e5fa | 5276 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5277 | mddev->reshape_position = MaxSector; |
5278 | mddev->delta_disks = 0; | |
cea9c228 N |
5279 | } |
5280 | } | |
5281 | ||
72626685 N |
5282 | static void raid5_quiesce(mddev_t *mddev, int state) |
5283 | { | |
070ec55d | 5284 | raid5_conf_t *conf = mddev->private; |
72626685 N |
5285 | |
5286 | switch(state) { | |
e464eafd N |
5287 | case 2: /* resume for a suspend */ |
5288 | wake_up(&conf->wait_for_overlap); | |
5289 | break; | |
5290 | ||
72626685 N |
5291 | case 1: /* stop all writes */ |
5292 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5293 | /* '2' tells resync/reshape to pause so that all |
5294 | * active stripes can drain | |
5295 | */ | |
5296 | conf->quiesce = 2; | |
72626685 | 5297 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5298 | atomic_read(&conf->active_stripes) == 0 && |
5299 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5300 | conf->device_lock, /* nothing */); |
64bd660b | 5301 | conf->quiesce = 1; |
72626685 | 5302 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5303 | /* allow reshape to continue */ |
5304 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5305 | break; |
5306 | ||
5307 | case 0: /* re-enable writes */ | |
5308 | spin_lock_irq(&conf->device_lock); | |
5309 | conf->quiesce = 0; | |
5310 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5311 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5312 | spin_unlock_irq(&conf->device_lock); |
5313 | break; | |
5314 | } | |
72626685 | 5315 | } |
b15c2e57 | 5316 | |
d562b0c4 | 5317 | |
f1b29bca | 5318 | static void *raid45_takeover_raid0(mddev_t *mddev, int level) |
54071b38 | 5319 | { |
f1b29bca | 5320 | struct raid0_private_data *raid0_priv = mddev->private; |
d76c8420 | 5321 | sector_t sectors; |
54071b38 | 5322 | |
f1b29bca DW |
5323 | /* for raid0 takeover only one zone is supported */ |
5324 | if (raid0_priv->nr_strip_zones > 1) { | |
0c55e022 N |
5325 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5326 | mdname(mddev)); | |
f1b29bca DW |
5327 | return ERR_PTR(-EINVAL); |
5328 | } | |
5329 | ||
3b71bd93 N |
5330 | sectors = raid0_priv->strip_zone[0].zone_end; |
5331 | sector_div(sectors, raid0_priv->strip_zone[0].nb_dev); | |
5332 | mddev->dev_sectors = sectors; | |
f1b29bca | 5333 | mddev->new_level = level; |
54071b38 TM |
5334 | mddev->new_layout = ALGORITHM_PARITY_N; |
5335 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5336 | mddev->raid_disks += 1; | |
5337 | mddev->delta_disks = 1; | |
5338 | /* make sure it will be not marked as dirty */ | |
5339 | mddev->recovery_cp = MaxSector; | |
5340 | ||
5341 | return setup_conf(mddev); | |
5342 | } | |
5343 | ||
5344 | ||
d562b0c4 N |
5345 | static void *raid5_takeover_raid1(mddev_t *mddev) |
5346 | { | |
5347 | int chunksect; | |
5348 | ||
5349 | if (mddev->raid_disks != 2 || | |
5350 | mddev->degraded > 1) | |
5351 | return ERR_PTR(-EINVAL); | |
5352 | ||
5353 | /* Should check if there are write-behind devices? */ | |
5354 | ||
5355 | chunksect = 64*2; /* 64K by default */ | |
5356 | ||
5357 | /* The array must be an exact multiple of chunksize */ | |
5358 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5359 | chunksect >>= 1; | |
5360 | ||
5361 | if ((chunksect<<9) < STRIPE_SIZE) | |
5362 | /* array size does not allow a suitable chunk size */ | |
5363 | return ERR_PTR(-EINVAL); | |
5364 | ||
5365 | mddev->new_level = 5; | |
5366 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5367 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5368 | |
5369 | return setup_conf(mddev); | |
5370 | } | |
5371 | ||
fc9739c6 N |
5372 | static void *raid5_takeover_raid6(mddev_t *mddev) |
5373 | { | |
5374 | int new_layout; | |
5375 | ||
5376 | switch (mddev->layout) { | |
5377 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5378 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5379 | break; | |
5380 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5381 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5382 | break; | |
5383 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5384 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5385 | break; | |
5386 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5387 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5388 | break; | |
5389 | case ALGORITHM_PARITY_0_6: | |
5390 | new_layout = ALGORITHM_PARITY_0; | |
5391 | break; | |
5392 | case ALGORITHM_PARITY_N: | |
5393 | new_layout = ALGORITHM_PARITY_N; | |
5394 | break; | |
5395 | default: | |
5396 | return ERR_PTR(-EINVAL); | |
5397 | } | |
5398 | mddev->new_level = 5; | |
5399 | mddev->new_layout = new_layout; | |
5400 | mddev->delta_disks = -1; | |
5401 | mddev->raid_disks -= 1; | |
5402 | return setup_conf(mddev); | |
5403 | } | |
5404 | ||
d562b0c4 | 5405 | |
50ac168a | 5406 | static int raid5_check_reshape(mddev_t *mddev) |
b3546035 | 5407 | { |
88ce4930 N |
5408 | /* For a 2-drive array, the layout and chunk size can be changed |
5409 | * immediately as not restriping is needed. | |
5410 | * For larger arrays we record the new value - after validation | |
5411 | * to be used by a reshape pass. | |
b3546035 | 5412 | */ |
070ec55d | 5413 | raid5_conf_t *conf = mddev->private; |
597a711b | 5414 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5415 | |
597a711b | 5416 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5417 | return -EINVAL; |
5418 | if (new_chunk > 0) { | |
0ba459d2 | 5419 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5420 | return -EINVAL; |
597a711b | 5421 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5422 | return -EINVAL; |
597a711b | 5423 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5424 | /* not factor of array size */ |
5425 | return -EINVAL; | |
5426 | } | |
5427 | ||
5428 | /* They look valid */ | |
5429 | ||
88ce4930 | 5430 | if (mddev->raid_disks == 2) { |
597a711b N |
5431 | /* can make the change immediately */ |
5432 | if (mddev->new_layout >= 0) { | |
5433 | conf->algorithm = mddev->new_layout; | |
5434 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5435 | } |
5436 | if (new_chunk > 0) { | |
597a711b N |
5437 | conf->chunk_sectors = new_chunk ; |
5438 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5439 | } |
5440 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5441 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5442 | } |
50ac168a | 5443 | return check_reshape(mddev); |
88ce4930 N |
5444 | } |
5445 | ||
50ac168a | 5446 | static int raid6_check_reshape(mddev_t *mddev) |
88ce4930 | 5447 | { |
597a711b | 5448 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5449 | |
597a711b | 5450 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5451 | return -EINVAL; |
b3546035 | 5452 | if (new_chunk > 0) { |
0ba459d2 | 5453 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5454 | return -EINVAL; |
597a711b | 5455 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5456 | return -EINVAL; |
597a711b | 5457 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5458 | /* not factor of array size */ |
5459 | return -EINVAL; | |
b3546035 | 5460 | } |
88ce4930 N |
5461 | |
5462 | /* They look valid */ | |
50ac168a | 5463 | return check_reshape(mddev); |
b3546035 N |
5464 | } |
5465 | ||
d562b0c4 N |
5466 | static void *raid5_takeover(mddev_t *mddev) |
5467 | { | |
5468 | /* raid5 can take over: | |
f1b29bca | 5469 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
5470 | * raid1 - if there are two drives. We need to know the chunk size |
5471 | * raid4 - trivial - just use a raid4 layout. | |
5472 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 5473 | */ |
f1b29bca DW |
5474 | if (mddev->level == 0) |
5475 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
5476 | if (mddev->level == 1) |
5477 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5478 | if (mddev->level == 4) { |
5479 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5480 | mddev->new_level = 5; | |
5481 | return setup_conf(mddev); | |
5482 | } | |
fc9739c6 N |
5483 | if (mddev->level == 6) |
5484 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5485 | |
5486 | return ERR_PTR(-EINVAL); | |
5487 | } | |
5488 | ||
a78d38a1 N |
5489 | static void *raid4_takeover(mddev_t *mddev) |
5490 | { | |
f1b29bca DW |
5491 | /* raid4 can take over: |
5492 | * raid0 - if there is only one strip zone | |
5493 | * raid5 - if layout is right | |
a78d38a1 | 5494 | */ |
f1b29bca DW |
5495 | if (mddev->level == 0) |
5496 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
5497 | if (mddev->level == 5 && |
5498 | mddev->layout == ALGORITHM_PARITY_N) { | |
5499 | mddev->new_layout = 0; | |
5500 | mddev->new_level = 4; | |
5501 | return setup_conf(mddev); | |
5502 | } | |
5503 | return ERR_PTR(-EINVAL); | |
5504 | } | |
d562b0c4 | 5505 | |
245f46c2 N |
5506 | static struct mdk_personality raid5_personality; |
5507 | ||
5508 | static void *raid6_takeover(mddev_t *mddev) | |
5509 | { | |
5510 | /* Currently can only take over a raid5. We map the | |
5511 | * personality to an equivalent raid6 personality | |
5512 | * with the Q block at the end. | |
5513 | */ | |
5514 | int new_layout; | |
5515 | ||
5516 | if (mddev->pers != &raid5_personality) | |
5517 | return ERR_PTR(-EINVAL); | |
5518 | if (mddev->degraded > 1) | |
5519 | return ERR_PTR(-EINVAL); | |
5520 | if (mddev->raid_disks > 253) | |
5521 | return ERR_PTR(-EINVAL); | |
5522 | if (mddev->raid_disks < 3) | |
5523 | return ERR_PTR(-EINVAL); | |
5524 | ||
5525 | switch (mddev->layout) { | |
5526 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5527 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5528 | break; | |
5529 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5530 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5531 | break; | |
5532 | case ALGORITHM_LEFT_SYMMETRIC: | |
5533 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5534 | break; | |
5535 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5536 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5537 | break; | |
5538 | case ALGORITHM_PARITY_0: | |
5539 | new_layout = ALGORITHM_PARITY_0_6; | |
5540 | break; | |
5541 | case ALGORITHM_PARITY_N: | |
5542 | new_layout = ALGORITHM_PARITY_N; | |
5543 | break; | |
5544 | default: | |
5545 | return ERR_PTR(-EINVAL); | |
5546 | } | |
5547 | mddev->new_level = 6; | |
5548 | mddev->new_layout = new_layout; | |
5549 | mddev->delta_disks = 1; | |
5550 | mddev->raid_disks += 1; | |
5551 | return setup_conf(mddev); | |
5552 | } | |
5553 | ||
5554 | ||
16a53ecc N |
5555 | static struct mdk_personality raid6_personality = |
5556 | { | |
5557 | .name = "raid6", | |
5558 | .level = 6, | |
5559 | .owner = THIS_MODULE, | |
5560 | .make_request = make_request, | |
5561 | .run = run, | |
5562 | .stop = stop, | |
5563 | .status = status, | |
5564 | .error_handler = error, | |
5565 | .hot_add_disk = raid5_add_disk, | |
5566 | .hot_remove_disk= raid5_remove_disk, | |
5567 | .spare_active = raid5_spare_active, | |
5568 | .sync_request = sync_request, | |
5569 | .resize = raid5_resize, | |
80c3a6ce | 5570 | .size = raid5_size, |
50ac168a | 5571 | .check_reshape = raid6_check_reshape, |
f416885e | 5572 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5573 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5574 | .quiesce = raid5_quiesce, |
245f46c2 | 5575 | .takeover = raid6_takeover, |
16a53ecc | 5576 | }; |
2604b703 | 5577 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
5578 | { |
5579 | .name = "raid5", | |
2604b703 | 5580 | .level = 5, |
1da177e4 LT |
5581 | .owner = THIS_MODULE, |
5582 | .make_request = make_request, | |
5583 | .run = run, | |
5584 | .stop = stop, | |
5585 | .status = status, | |
5586 | .error_handler = error, | |
5587 | .hot_add_disk = raid5_add_disk, | |
5588 | .hot_remove_disk= raid5_remove_disk, | |
5589 | .spare_active = raid5_spare_active, | |
5590 | .sync_request = sync_request, | |
5591 | .resize = raid5_resize, | |
80c3a6ce | 5592 | .size = raid5_size, |
63c70c4f N |
5593 | .check_reshape = raid5_check_reshape, |
5594 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5595 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5596 | .quiesce = raid5_quiesce, |
d562b0c4 | 5597 | .takeover = raid5_takeover, |
1da177e4 LT |
5598 | }; |
5599 | ||
2604b703 | 5600 | static struct mdk_personality raid4_personality = |
1da177e4 | 5601 | { |
2604b703 N |
5602 | .name = "raid4", |
5603 | .level = 4, | |
5604 | .owner = THIS_MODULE, | |
5605 | .make_request = make_request, | |
5606 | .run = run, | |
5607 | .stop = stop, | |
5608 | .status = status, | |
5609 | .error_handler = error, | |
5610 | .hot_add_disk = raid5_add_disk, | |
5611 | .hot_remove_disk= raid5_remove_disk, | |
5612 | .spare_active = raid5_spare_active, | |
5613 | .sync_request = sync_request, | |
5614 | .resize = raid5_resize, | |
80c3a6ce | 5615 | .size = raid5_size, |
3d37890b N |
5616 | .check_reshape = raid5_check_reshape, |
5617 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5618 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 5619 | .quiesce = raid5_quiesce, |
a78d38a1 | 5620 | .takeover = raid4_takeover, |
2604b703 N |
5621 | }; |
5622 | ||
5623 | static int __init raid5_init(void) | |
5624 | { | |
16a53ecc | 5625 | register_md_personality(&raid6_personality); |
2604b703 N |
5626 | register_md_personality(&raid5_personality); |
5627 | register_md_personality(&raid4_personality); | |
5628 | return 0; | |
1da177e4 LT |
5629 | } |
5630 | ||
2604b703 | 5631 | static void raid5_exit(void) |
1da177e4 | 5632 | { |
16a53ecc | 5633 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5634 | unregister_md_personality(&raid5_personality); |
5635 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5636 | } |
5637 | ||
5638 | module_init(raid5_init); | |
5639 | module_exit(raid5_exit); | |
5640 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5641 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 5642 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
5643 | MODULE_ALIAS("md-raid5"); |
5644 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5645 | MODULE_ALIAS("md-level-5"); |
5646 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5647 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5648 | MODULE_ALIAS("md-raid6"); | |
5649 | MODULE_ALIAS("md-level-6"); | |
5650 | ||
5651 | /* This used to be two separate modules, they were: */ | |
5652 | MODULE_ALIAS("raid5"); | |
5653 | MODULE_ALIAS("raid6"); |