]>
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> |
43b2e5d8 | 54 | #include "md.h" |
bff61975 | 55 | #include "raid5.h" |
54071b38 | 56 | #include "raid0.h" |
ef740c37 | 57 | #include "bitmap.h" |
72626685 | 58 | |
1da177e4 LT |
59 | /* |
60 | * Stripe cache | |
61 | */ | |
62 | ||
63 | #define NR_STRIPES 256 | |
64 | #define STRIPE_SIZE PAGE_SIZE | |
65 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
66 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
67 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 68 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 69 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
70 | #define HASH_MASK (NR_HASH - 1) |
71 | ||
fccddba0 | 72 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
73 | |
74 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
75 | * order without overlap. There may be several bio's per stripe+device, and | |
76 | * a bio could span several devices. | |
77 | * When walking this list for a particular stripe+device, we must never proceed | |
78 | * beyond a bio that extends past this device, as the next bio might no longer | |
79 | * be valid. | |
80 | * This macro is used to determine the 'next' bio in the list, given the sector | |
81 | * of the current stripe+device | |
82 | */ | |
83 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
84 | /* | |
85 | * The following can be used to debug the driver | |
86 | */ | |
1da177e4 LT |
87 | #define RAID5_PARANOIA 1 |
88 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
89 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
90 | #else | |
91 | # define CHECK_DEVLOCK() | |
92 | #endif | |
93 | ||
45b4233c | 94 | #ifdef DEBUG |
1da177e4 LT |
95 | #define inline |
96 | #define __inline__ | |
97 | #endif | |
98 | ||
6be9d494 BS |
99 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
100 | ||
960e739d | 101 | /* |
5b99c2ff JA |
102 | * We maintain a biased count of active stripes in the bottom 16 bits of |
103 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
104 | */ |
105 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
106 | { | |
5b99c2ff | 107 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
108 | } |
109 | ||
110 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
111 | { | |
5b99c2ff | 112 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
113 | } |
114 | ||
115 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
116 | { | |
117 | --bio->bi_phys_segments; | |
118 | return raid5_bi_phys_segments(bio); | |
119 | } | |
120 | ||
121 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
122 | { | |
123 | unsigned short val = raid5_bi_hw_segments(bio); | |
124 | ||
125 | --val; | |
5b99c2ff | 126 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
127 | return val; |
128 | } | |
129 | ||
130 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
131 | { | |
9b2dc8b6 | 132 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) | (cnt << 16); |
960e739d JA |
133 | } |
134 | ||
d0dabf7e N |
135 | /* Find first data disk in a raid6 stripe */ |
136 | static inline int raid6_d0(struct stripe_head *sh) | |
137 | { | |
67cc2b81 N |
138 | if (sh->ddf_layout) |
139 | /* ddf always start from first device */ | |
140 | return 0; | |
141 | /* md starts just after Q block */ | |
d0dabf7e N |
142 | if (sh->qd_idx == sh->disks - 1) |
143 | return 0; | |
144 | else | |
145 | return sh->qd_idx + 1; | |
146 | } | |
16a53ecc N |
147 | static inline int raid6_next_disk(int disk, int raid_disks) |
148 | { | |
149 | disk++; | |
150 | return (disk < raid_disks) ? disk : 0; | |
151 | } | |
a4456856 | 152 | |
d0dabf7e N |
153 | /* When walking through the disks in a raid5, starting at raid6_d0, |
154 | * We need to map each disk to a 'slot', where the data disks are slot | |
155 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
156 | * is raid_disks-1. This help does that mapping. | |
157 | */ | |
67cc2b81 N |
158 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
159 | int *count, int syndrome_disks) | |
d0dabf7e | 160 | { |
6629542e | 161 | int slot = *count; |
67cc2b81 | 162 | |
e4424fee | 163 | if (sh->ddf_layout) |
6629542e | 164 | (*count)++; |
d0dabf7e | 165 | if (idx == sh->pd_idx) |
67cc2b81 | 166 | return syndrome_disks; |
d0dabf7e | 167 | if (idx == sh->qd_idx) |
67cc2b81 | 168 | return syndrome_disks + 1; |
e4424fee | 169 | if (!sh->ddf_layout) |
6629542e | 170 | (*count)++; |
d0dabf7e N |
171 | return slot; |
172 | } | |
173 | ||
a4456856 DW |
174 | static void return_io(struct bio *return_bi) |
175 | { | |
176 | struct bio *bi = return_bi; | |
177 | while (bi) { | |
a4456856 DW |
178 | |
179 | return_bi = bi->bi_next; | |
180 | bi->bi_next = NULL; | |
181 | bi->bi_size = 0; | |
0e13fe23 | 182 | bio_endio(bi, 0); |
a4456856 DW |
183 | bi = return_bi; |
184 | } | |
185 | } | |
186 | ||
1da177e4 LT |
187 | static void print_raid5_conf (raid5_conf_t *conf); |
188 | ||
600aa109 DW |
189 | static int stripe_operations_active(struct stripe_head *sh) |
190 | { | |
191 | return sh->check_state || sh->reconstruct_state || | |
192 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
193 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
194 | } | |
195 | ||
858119e1 | 196 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
197 | { |
198 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
199 | BUG_ON(!list_empty(&sh->lru)); |
200 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 201 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
482c0834 | 202 | if (test_bit(STRIPE_DELAYED, &sh->state)) |
1da177e4 | 203 | list_add_tail(&sh->lru, &conf->delayed_list); |
482c0834 N |
204 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
205 | sh->bm_seq - conf->seq_write > 0) | |
72626685 | 206 | list_add_tail(&sh->lru, &conf->bitmap_list); |
482c0834 | 207 | else { |
72626685 | 208 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 209 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 210 | } |
1da177e4 LT |
211 | md_wakeup_thread(conf->mddev->thread); |
212 | } else { | |
600aa109 | 213 | BUG_ON(stripe_operations_active(sh)); |
1da177e4 LT |
214 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { |
215 | atomic_dec(&conf->preread_active_stripes); | |
216 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
217 | md_wakeup_thread(conf->mddev->thread); | |
218 | } | |
1da177e4 | 219 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
220 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
221 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 222 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
223 | if (conf->retry_read_aligned) |
224 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 225 | } |
1da177e4 LT |
226 | } |
227 | } | |
228 | } | |
d0dabf7e | 229 | |
1da177e4 LT |
230 | static void release_stripe(struct stripe_head *sh) |
231 | { | |
232 | raid5_conf_t *conf = sh->raid_conf; | |
233 | unsigned long flags; | |
16a53ecc | 234 | |
1da177e4 LT |
235 | spin_lock_irqsave(&conf->device_lock, flags); |
236 | __release_stripe(conf, sh); | |
237 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
238 | } | |
239 | ||
fccddba0 | 240 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 241 | { |
45b4233c DW |
242 | pr_debug("remove_hash(), stripe %llu\n", |
243 | (unsigned long long)sh->sector); | |
1da177e4 | 244 | |
fccddba0 | 245 | hlist_del_init(&sh->hash); |
1da177e4 LT |
246 | } |
247 | ||
16a53ecc | 248 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 249 | { |
fccddba0 | 250 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 251 | |
45b4233c DW |
252 | pr_debug("insert_hash(), stripe %llu\n", |
253 | (unsigned long long)sh->sector); | |
1da177e4 LT |
254 | |
255 | CHECK_DEVLOCK(); | |
fccddba0 | 256 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
257 | } |
258 | ||
259 | ||
260 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
261 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
262 | { | |
263 | struct stripe_head *sh = NULL; | |
264 | struct list_head *first; | |
265 | ||
266 | CHECK_DEVLOCK(); | |
267 | if (list_empty(&conf->inactive_list)) | |
268 | goto out; | |
269 | first = conf->inactive_list.next; | |
270 | sh = list_entry(first, struct stripe_head, lru); | |
271 | list_del_init(first); | |
272 | remove_hash(sh); | |
273 | atomic_inc(&conf->active_stripes); | |
274 | out: | |
275 | return sh; | |
276 | } | |
277 | ||
e4e11e38 | 278 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
279 | { |
280 | struct page *p; | |
281 | int i; | |
e4e11e38 | 282 | int num = sh->raid_conf->pool_size; |
1da177e4 | 283 | |
e4e11e38 | 284 | for (i = 0; i < num ; i++) { |
1da177e4 LT |
285 | p = sh->dev[i].page; |
286 | if (!p) | |
287 | continue; | |
288 | sh->dev[i].page = NULL; | |
2d1f3b5d | 289 | put_page(p); |
1da177e4 LT |
290 | } |
291 | } | |
292 | ||
e4e11e38 | 293 | static int grow_buffers(struct stripe_head *sh) |
1da177e4 LT |
294 | { |
295 | int i; | |
e4e11e38 | 296 | int num = sh->raid_conf->pool_size; |
1da177e4 | 297 | |
e4e11e38 | 298 | for (i = 0; i < num; i++) { |
1da177e4 LT |
299 | struct page *page; |
300 | ||
301 | if (!(page = alloc_page(GFP_KERNEL))) { | |
302 | return 1; | |
303 | } | |
304 | sh->dev[i].page = page; | |
305 | } | |
306 | return 0; | |
307 | } | |
308 | ||
784052ec | 309 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
911d4ee8 N |
310 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
311 | struct stripe_head *sh); | |
1da177e4 | 312 | |
b5663ba4 | 313 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
314 | { |
315 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 316 | int i; |
1da177e4 | 317 | |
78bafebd ES |
318 | BUG_ON(atomic_read(&sh->count) != 0); |
319 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 320 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 321 | |
1da177e4 | 322 | CHECK_DEVLOCK(); |
45b4233c | 323 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
324 | (unsigned long long)sh->sector); |
325 | ||
326 | remove_hash(sh); | |
16a53ecc | 327 | |
86b42c71 | 328 | sh->generation = conf->generation - previous; |
b5663ba4 | 329 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 330 | sh->sector = sector; |
911d4ee8 | 331 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
332 | sh->state = 0; |
333 | ||
7ecaa1e6 N |
334 | |
335 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
336 | struct r5dev *dev = &sh->dev[i]; |
337 | ||
d84e0f10 | 338 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 339 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 340 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 341 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 342 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
343 | test_bit(R5_LOCKED, &dev->flags)); |
344 | BUG(); | |
345 | } | |
346 | dev->flags = 0; | |
784052ec | 347 | raid5_build_block(sh, i, previous); |
1da177e4 LT |
348 | } |
349 | insert_hash(conf, sh); | |
350 | } | |
351 | ||
86b42c71 N |
352 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, |
353 | short generation) | |
1da177e4 LT |
354 | { |
355 | struct stripe_head *sh; | |
fccddba0 | 356 | struct hlist_node *hn; |
1da177e4 LT |
357 | |
358 | CHECK_DEVLOCK(); | |
45b4233c | 359 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 360 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
86b42c71 | 361 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 362 | return sh; |
45b4233c | 363 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
364 | return NULL; |
365 | } | |
366 | ||
674806d6 N |
367 | /* |
368 | * Need to check if array has failed when deciding whether to: | |
369 | * - start an array | |
370 | * - remove non-faulty devices | |
371 | * - add a spare | |
372 | * - allow a reshape | |
373 | * This determination is simple when no reshape is happening. | |
374 | * However if there is a reshape, we need to carefully check | |
375 | * both the before and after sections. | |
376 | * This is because some failed devices may only affect one | |
377 | * of the two sections, and some non-in_sync devices may | |
378 | * be insync in the section most affected by failed devices. | |
379 | */ | |
380 | static int has_failed(raid5_conf_t *conf) | |
381 | { | |
382 | int degraded; | |
383 | int i; | |
384 | if (conf->mddev->reshape_position == MaxSector) | |
385 | return conf->mddev->degraded > conf->max_degraded; | |
386 | ||
387 | rcu_read_lock(); | |
388 | degraded = 0; | |
389 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
390 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
391 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
392 | degraded++; | |
393 | else if (test_bit(In_sync, &rdev->flags)) | |
394 | ; | |
395 | else | |
396 | /* not in-sync or faulty. | |
397 | * If the reshape increases the number of devices, | |
398 | * this is being recovered by the reshape, so | |
399 | * this 'previous' section is not in_sync. | |
400 | * If the number of devices is being reduced however, | |
401 | * the device can only be part of the array if | |
402 | * we are reverting a reshape, so this section will | |
403 | * be in-sync. | |
404 | */ | |
405 | if (conf->raid_disks >= conf->previous_raid_disks) | |
406 | degraded++; | |
407 | } | |
408 | rcu_read_unlock(); | |
409 | if (degraded > conf->max_degraded) | |
410 | return 1; | |
411 | rcu_read_lock(); | |
412 | degraded = 0; | |
413 | for (i = 0; i < conf->raid_disks; i++) { | |
414 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
415 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
416 | degraded++; | |
417 | else if (test_bit(In_sync, &rdev->flags)) | |
418 | ; | |
419 | else | |
420 | /* not in-sync or faulty. | |
421 | * If reshape increases the number of devices, this | |
422 | * section has already been recovered, else it | |
423 | * almost certainly hasn't. | |
424 | */ | |
425 | if (conf->raid_disks <= conf->previous_raid_disks) | |
426 | degraded++; | |
427 | } | |
428 | rcu_read_unlock(); | |
429 | if (degraded > conf->max_degraded) | |
430 | return 1; | |
431 | return 0; | |
432 | } | |
433 | ||
b5663ba4 N |
434 | static struct stripe_head * |
435 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
a8c906ca | 436 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
437 | { |
438 | struct stripe_head *sh; | |
439 | ||
45b4233c | 440 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
441 | |
442 | spin_lock_irq(&conf->device_lock); | |
443 | ||
444 | do { | |
72626685 | 445 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 446 | conf->quiesce == 0 || noquiesce, |
72626685 | 447 | conf->device_lock, /* nothing */); |
86b42c71 | 448 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
449 | if (!sh) { |
450 | if (!conf->inactive_blocked) | |
451 | sh = get_free_stripe(conf); | |
452 | if (noblock && sh == NULL) | |
453 | break; | |
454 | if (!sh) { | |
455 | conf->inactive_blocked = 1; | |
456 | wait_event_lock_irq(conf->wait_for_stripe, | |
457 | !list_empty(&conf->inactive_list) && | |
5036805b N |
458 | (atomic_read(&conf->active_stripes) |
459 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
460 | || !conf->inactive_blocked), |
461 | conf->device_lock, | |
7c13edc8 | 462 | ); |
1da177e4 LT |
463 | conf->inactive_blocked = 0; |
464 | } else | |
b5663ba4 | 465 | init_stripe(sh, sector, previous); |
1da177e4 LT |
466 | } else { |
467 | if (atomic_read(&sh->count)) { | |
ab69ae12 N |
468 | BUG_ON(!list_empty(&sh->lru) |
469 | && !test_bit(STRIPE_EXPANDING, &sh->state)); | |
1da177e4 LT |
470 | } else { |
471 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
472 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
473 | if (list_empty(&sh->lru) && |
474 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
475 | BUG(); |
476 | list_del_init(&sh->lru); | |
1da177e4 LT |
477 | } |
478 | } | |
479 | } while (sh == NULL); | |
480 | ||
481 | if (sh) | |
482 | atomic_inc(&sh->count); | |
483 | ||
484 | spin_unlock_irq(&conf->device_lock); | |
485 | return sh; | |
486 | } | |
487 | ||
6712ecf8 N |
488 | static void |
489 | raid5_end_read_request(struct bio *bi, int error); | |
490 | static void | |
491 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 492 | |
c4e5ac0a | 493 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
494 | { |
495 | raid5_conf_t *conf = sh->raid_conf; | |
496 | int i, disks = sh->disks; | |
497 | ||
498 | might_sleep(); | |
499 | ||
500 | for (i = disks; i--; ) { | |
501 | int rw; | |
502 | struct bio *bi; | |
503 | mdk_rdev_t *rdev; | |
e9c7469b TH |
504 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
505 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
506 | rw = WRITE_FUA; | |
507 | else | |
508 | rw = WRITE; | |
509 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
91c00924 DW |
510 | rw = READ; |
511 | else | |
512 | continue; | |
513 | ||
514 | bi = &sh->dev[i].req; | |
515 | ||
516 | bi->bi_rw = rw; | |
b062962e | 517 | if (rw & WRITE) |
91c00924 DW |
518 | bi->bi_end_io = raid5_end_write_request; |
519 | else | |
520 | bi->bi_end_io = raid5_end_read_request; | |
521 | ||
522 | rcu_read_lock(); | |
523 | rdev = rcu_dereference(conf->disks[i].rdev); | |
524 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
525 | rdev = NULL; | |
526 | if (rdev) | |
527 | atomic_inc(&rdev->nr_pending); | |
528 | rcu_read_unlock(); | |
529 | ||
530 | if (rdev) { | |
c4e5ac0a | 531 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
532 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
533 | ||
2b7497f0 DW |
534 | set_bit(STRIPE_IO_STARTED, &sh->state); |
535 | ||
91c00924 DW |
536 | bi->bi_bdev = rdev->bdev; |
537 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 538 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
539 | bi->bi_rw, i); |
540 | atomic_inc(&sh->count); | |
541 | bi->bi_sector = sh->sector + rdev->data_offset; | |
542 | bi->bi_flags = 1 << BIO_UPTODATE; | |
543 | bi->bi_vcnt = 1; | |
544 | bi->bi_max_vecs = 1; | |
545 | bi->bi_idx = 0; | |
546 | bi->bi_io_vec = &sh->dev[i].vec; | |
547 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
548 | bi->bi_io_vec[0].bv_offset = 0; | |
549 | bi->bi_size = STRIPE_SIZE; | |
550 | bi->bi_next = NULL; | |
b062962e | 551 | if ((rw & WRITE) && |
91c00924 DW |
552 | test_bit(R5_ReWrite, &sh->dev[i].flags)) |
553 | atomic_add(STRIPE_SECTORS, | |
554 | &rdev->corrected_errors); | |
555 | generic_make_request(bi); | |
556 | } else { | |
b062962e | 557 | if (rw & WRITE) |
91c00924 DW |
558 | set_bit(STRIPE_DEGRADED, &sh->state); |
559 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
560 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
561 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
562 | set_bit(STRIPE_HANDLE, &sh->state); | |
563 | } | |
564 | } | |
565 | } | |
566 | ||
567 | static struct dma_async_tx_descriptor * | |
568 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
569 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
570 | { | |
571 | struct bio_vec *bvl; | |
572 | struct page *bio_page; | |
573 | int i; | |
574 | int page_offset; | |
a08abd8c | 575 | struct async_submit_ctl submit; |
0403e382 | 576 | enum async_tx_flags flags = 0; |
91c00924 DW |
577 | |
578 | if (bio->bi_sector >= sector) | |
579 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
580 | else | |
581 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
a08abd8c | 582 | |
0403e382 DW |
583 | if (frombio) |
584 | flags |= ASYNC_TX_FENCE; | |
585 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
586 | ||
91c00924 | 587 | bio_for_each_segment(bvl, bio, i) { |
fcde9075 | 588 | int len = bvl->bv_len; |
91c00924 DW |
589 | int clen; |
590 | int b_offset = 0; | |
591 | ||
592 | if (page_offset < 0) { | |
593 | b_offset = -page_offset; | |
594 | page_offset += b_offset; | |
595 | len -= b_offset; | |
596 | } | |
597 | ||
598 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
599 | clen = STRIPE_SIZE - page_offset; | |
600 | else | |
601 | clen = len; | |
602 | ||
603 | if (clen > 0) { | |
fcde9075 NK |
604 | b_offset += bvl->bv_offset; |
605 | bio_page = bvl->bv_page; | |
91c00924 DW |
606 | if (frombio) |
607 | tx = async_memcpy(page, bio_page, page_offset, | |
a08abd8c | 608 | b_offset, clen, &submit); |
91c00924 DW |
609 | else |
610 | tx = async_memcpy(bio_page, page, b_offset, | |
a08abd8c | 611 | page_offset, clen, &submit); |
91c00924 | 612 | } |
a08abd8c DW |
613 | /* chain the operations */ |
614 | submit.depend_tx = tx; | |
615 | ||
91c00924 DW |
616 | if (clen < len) /* hit end of page */ |
617 | break; | |
618 | page_offset += len; | |
619 | } | |
620 | ||
621 | return tx; | |
622 | } | |
623 | ||
624 | static void ops_complete_biofill(void *stripe_head_ref) | |
625 | { | |
626 | struct stripe_head *sh = stripe_head_ref; | |
627 | struct bio *return_bi = NULL; | |
628 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 629 | int i; |
91c00924 | 630 | |
e46b272b | 631 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
632 | (unsigned long long)sh->sector); |
633 | ||
634 | /* clear completed biofills */ | |
83de75cc | 635 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
636 | for (i = sh->disks; i--; ) { |
637 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
638 | |
639 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
640 | /* and check if we need to reply to a read request, |
641 | * new R5_Wantfill requests are held off until | |
83de75cc | 642 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
643 | */ |
644 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 645 | struct bio *rbi, *rbi2; |
91c00924 | 646 | |
91c00924 DW |
647 | BUG_ON(!dev->read); |
648 | rbi = dev->read; | |
649 | dev->read = NULL; | |
650 | while (rbi && rbi->bi_sector < | |
651 | dev->sector + STRIPE_SECTORS) { | |
652 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 653 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
654 | rbi->bi_next = return_bi; |
655 | return_bi = rbi; | |
656 | } | |
91c00924 DW |
657 | rbi = rbi2; |
658 | } | |
659 | } | |
660 | } | |
83de75cc DW |
661 | spin_unlock_irq(&conf->device_lock); |
662 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
663 | |
664 | return_io(return_bi); | |
665 | ||
e4d84909 | 666 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
667 | release_stripe(sh); |
668 | } | |
669 | ||
670 | static void ops_run_biofill(struct stripe_head *sh) | |
671 | { | |
672 | struct dma_async_tx_descriptor *tx = NULL; | |
673 | raid5_conf_t *conf = sh->raid_conf; | |
a08abd8c | 674 | struct async_submit_ctl submit; |
91c00924 DW |
675 | int i; |
676 | ||
e46b272b | 677 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
678 | (unsigned long long)sh->sector); |
679 | ||
680 | for (i = sh->disks; i--; ) { | |
681 | struct r5dev *dev = &sh->dev[i]; | |
682 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
683 | struct bio *rbi; | |
684 | spin_lock_irq(&conf->device_lock); | |
685 | dev->read = rbi = dev->toread; | |
686 | dev->toread = NULL; | |
687 | spin_unlock_irq(&conf->device_lock); | |
688 | while (rbi && rbi->bi_sector < | |
689 | dev->sector + STRIPE_SECTORS) { | |
690 | tx = async_copy_data(0, rbi, dev->page, | |
691 | dev->sector, tx); | |
692 | rbi = r5_next_bio(rbi, dev->sector); | |
693 | } | |
694 | } | |
695 | } | |
696 | ||
697 | atomic_inc(&sh->count); | |
a08abd8c DW |
698 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
699 | async_trigger_callback(&submit); | |
91c00924 DW |
700 | } |
701 | ||
4e7d2c0a | 702 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 703 | { |
4e7d2c0a | 704 | struct r5dev *tgt; |
91c00924 | 705 | |
4e7d2c0a DW |
706 | if (target < 0) |
707 | return; | |
91c00924 | 708 | |
4e7d2c0a | 709 | tgt = &sh->dev[target]; |
91c00924 DW |
710 | set_bit(R5_UPTODATE, &tgt->flags); |
711 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
712 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
713 | } |
714 | ||
ac6b53b6 | 715 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
716 | { |
717 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 718 | |
e46b272b | 719 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
720 | (unsigned long long)sh->sector); |
721 | ||
ac6b53b6 | 722 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 723 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 724 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 725 | |
ecc65c9b DW |
726 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
727 | if (sh->check_state == check_state_compute_run) | |
728 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
729 | set_bit(STRIPE_HANDLE, &sh->state); |
730 | release_stripe(sh); | |
731 | } | |
732 | ||
d6f38f31 DW |
733 | /* return a pointer to the address conversion region of the scribble buffer */ |
734 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
735 | struct raid5_percpu *percpu) | |
736 | { | |
737 | return percpu->scribble + sizeof(struct page *) * (sh->disks + 2); | |
738 | } | |
739 | ||
740 | static struct dma_async_tx_descriptor * | |
741 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 742 | { |
91c00924 | 743 | int disks = sh->disks; |
d6f38f31 | 744 | struct page **xor_srcs = percpu->scribble; |
91c00924 DW |
745 | int target = sh->ops.target; |
746 | struct r5dev *tgt = &sh->dev[target]; | |
747 | struct page *xor_dest = tgt->page; | |
748 | int count = 0; | |
749 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 750 | struct async_submit_ctl submit; |
91c00924 DW |
751 | int i; |
752 | ||
753 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 754 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
755 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
756 | ||
757 | for (i = disks; i--; ) | |
758 | if (i != target) | |
759 | xor_srcs[count++] = sh->dev[i].page; | |
760 | ||
761 | atomic_inc(&sh->count); | |
762 | ||
0403e382 | 763 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
ac6b53b6 | 764 | ops_complete_compute, sh, to_addr_conv(sh, percpu)); |
91c00924 | 765 | if (unlikely(count == 1)) |
a08abd8c | 766 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 767 | else |
a08abd8c | 768 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 769 | |
91c00924 DW |
770 | return tx; |
771 | } | |
772 | ||
ac6b53b6 DW |
773 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
774 | * @srcs - (struct page *) array of size sh->disks | |
775 | * @sh - stripe_head to parse | |
776 | * | |
777 | * Populates srcs in proper layout order for the stripe and returns the | |
778 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
779 | * destination buffer is recorded in srcs[count] and the Q destination | |
780 | * is recorded in srcs[count+1]]. | |
781 | */ | |
782 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
783 | { | |
784 | int disks = sh->disks; | |
785 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
786 | int d0_idx = raid6_d0(sh); | |
787 | int count; | |
788 | int i; | |
789 | ||
790 | for (i = 0; i < disks; i++) | |
5dd33c9a | 791 | srcs[i] = NULL; |
ac6b53b6 DW |
792 | |
793 | count = 0; | |
794 | i = d0_idx; | |
795 | do { | |
796 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
797 | ||
798 | srcs[slot] = sh->dev[i].page; | |
799 | i = raid6_next_disk(i, disks); | |
800 | } while (i != d0_idx); | |
ac6b53b6 | 801 | |
e4424fee | 802 | return syndrome_disks; |
ac6b53b6 DW |
803 | } |
804 | ||
805 | static struct dma_async_tx_descriptor * | |
806 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
807 | { | |
808 | int disks = sh->disks; | |
809 | struct page **blocks = percpu->scribble; | |
810 | int target; | |
811 | int qd_idx = sh->qd_idx; | |
812 | struct dma_async_tx_descriptor *tx; | |
813 | struct async_submit_ctl submit; | |
814 | struct r5dev *tgt; | |
815 | struct page *dest; | |
816 | int i; | |
817 | int count; | |
818 | ||
819 | if (sh->ops.target < 0) | |
820 | target = sh->ops.target2; | |
821 | else if (sh->ops.target2 < 0) | |
822 | target = sh->ops.target; | |
91c00924 | 823 | else |
ac6b53b6 DW |
824 | /* we should only have one valid target */ |
825 | BUG(); | |
826 | BUG_ON(target < 0); | |
827 | pr_debug("%s: stripe %llu block: %d\n", | |
828 | __func__, (unsigned long long)sh->sector, target); | |
829 | ||
830 | tgt = &sh->dev[target]; | |
831 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
832 | dest = tgt->page; | |
833 | ||
834 | atomic_inc(&sh->count); | |
835 | ||
836 | if (target == qd_idx) { | |
837 | count = set_syndrome_sources(blocks, sh); | |
838 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
839 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
840 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
841 | ops_complete_compute, sh, | |
ac6b53b6 DW |
842 | to_addr_conv(sh, percpu)); |
843 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
844 | } else { | |
845 | /* Compute any data- or p-drive using XOR */ | |
846 | count = 0; | |
847 | for (i = disks; i-- ; ) { | |
848 | if (i == target || i == qd_idx) | |
849 | continue; | |
850 | blocks[count++] = sh->dev[i].page; | |
851 | } | |
852 | ||
0403e382 DW |
853 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
854 | NULL, ops_complete_compute, sh, | |
ac6b53b6 DW |
855 | to_addr_conv(sh, percpu)); |
856 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); | |
857 | } | |
91c00924 | 858 | |
91c00924 DW |
859 | return tx; |
860 | } | |
861 | ||
ac6b53b6 DW |
862 | static struct dma_async_tx_descriptor * |
863 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
864 | { | |
865 | int i, count, disks = sh->disks; | |
866 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
867 | int d0_idx = raid6_d0(sh); | |
868 | int faila = -1, failb = -1; | |
869 | int target = sh->ops.target; | |
870 | int target2 = sh->ops.target2; | |
871 | struct r5dev *tgt = &sh->dev[target]; | |
872 | struct r5dev *tgt2 = &sh->dev[target2]; | |
873 | struct dma_async_tx_descriptor *tx; | |
874 | struct page **blocks = percpu->scribble; | |
875 | struct async_submit_ctl submit; | |
876 | ||
877 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
878 | __func__, (unsigned long long)sh->sector, target, target2); | |
879 | BUG_ON(target < 0 || target2 < 0); | |
880 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
881 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
882 | ||
6c910a78 | 883 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
884 | * slot number conversion for 'faila' and 'failb' |
885 | */ | |
886 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 887 | blocks[i] = NULL; |
ac6b53b6 DW |
888 | count = 0; |
889 | i = d0_idx; | |
890 | do { | |
891 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
892 | ||
893 | blocks[slot] = sh->dev[i].page; | |
894 | ||
895 | if (i == target) | |
896 | faila = slot; | |
897 | if (i == target2) | |
898 | failb = slot; | |
899 | i = raid6_next_disk(i, disks); | |
900 | } while (i != d0_idx); | |
ac6b53b6 DW |
901 | |
902 | BUG_ON(faila == failb); | |
903 | if (failb < faila) | |
904 | swap(faila, failb); | |
905 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
906 | __func__, (unsigned long long)sh->sector, faila, failb); | |
907 | ||
908 | atomic_inc(&sh->count); | |
909 | ||
910 | if (failb == syndrome_disks+1) { | |
911 | /* Q disk is one of the missing disks */ | |
912 | if (faila == syndrome_disks) { | |
913 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
914 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
915 | ops_complete_compute, sh, | |
916 | to_addr_conv(sh, percpu)); | |
e4424fee | 917 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
918 | STRIPE_SIZE, &submit); |
919 | } else { | |
920 | struct page *dest; | |
921 | int data_target; | |
922 | int qd_idx = sh->qd_idx; | |
923 | ||
924 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
925 | if (target == qd_idx) | |
926 | data_target = target2; | |
927 | else | |
928 | data_target = target; | |
929 | ||
930 | count = 0; | |
931 | for (i = disks; i-- ; ) { | |
932 | if (i == data_target || i == qd_idx) | |
933 | continue; | |
934 | blocks[count++] = sh->dev[i].page; | |
935 | } | |
936 | dest = sh->dev[data_target].page; | |
0403e382 DW |
937 | init_async_submit(&submit, |
938 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
939 | NULL, NULL, NULL, | |
940 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
941 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
942 | &submit); | |
943 | ||
944 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
945 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
946 | ops_complete_compute, sh, | |
947 | to_addr_conv(sh, percpu)); | |
ac6b53b6 DW |
948 | return async_gen_syndrome(blocks, 0, count+2, |
949 | STRIPE_SIZE, &submit); | |
950 | } | |
ac6b53b6 | 951 | } else { |
6c910a78 DW |
952 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
953 | ops_complete_compute, sh, | |
954 | to_addr_conv(sh, percpu)); | |
955 | if (failb == syndrome_disks) { | |
956 | /* We're missing D+P. */ | |
957 | return async_raid6_datap_recov(syndrome_disks+2, | |
958 | STRIPE_SIZE, faila, | |
959 | blocks, &submit); | |
960 | } else { | |
961 | /* We're missing D+D. */ | |
962 | return async_raid6_2data_recov(syndrome_disks+2, | |
963 | STRIPE_SIZE, faila, failb, | |
964 | blocks, &submit); | |
965 | } | |
ac6b53b6 DW |
966 | } |
967 | } | |
968 | ||
969 | ||
91c00924 DW |
970 | static void ops_complete_prexor(void *stripe_head_ref) |
971 | { | |
972 | struct stripe_head *sh = stripe_head_ref; | |
973 | ||
e46b272b | 974 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 975 | (unsigned long long)sh->sector); |
91c00924 DW |
976 | } |
977 | ||
978 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
979 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
980 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 981 | { |
91c00924 | 982 | int disks = sh->disks; |
d6f38f31 | 983 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 984 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 985 | struct async_submit_ctl submit; |
91c00924 DW |
986 | |
987 | /* existing parity data subtracted */ | |
988 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
989 | ||
e46b272b | 990 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
991 | (unsigned long long)sh->sector); |
992 | ||
993 | for (i = disks; i--; ) { | |
994 | struct r5dev *dev = &sh->dev[i]; | |
995 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 996 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
997 | xor_srcs[count++] = dev->page; |
998 | } | |
999 | ||
0403e382 | 1000 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
d6f38f31 | 1001 | ops_complete_prexor, sh, to_addr_conv(sh, percpu)); |
a08abd8c | 1002 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1003 | |
1004 | return tx; | |
1005 | } | |
1006 | ||
1007 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1008 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1009 | { |
1010 | int disks = sh->disks; | |
d8ee0728 | 1011 | int i; |
91c00924 | 1012 | |
e46b272b | 1013 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1014 | (unsigned long long)sh->sector); |
1015 | ||
1016 | for (i = disks; i--; ) { | |
1017 | struct r5dev *dev = &sh->dev[i]; | |
1018 | struct bio *chosen; | |
91c00924 | 1019 | |
d8ee0728 | 1020 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1021 | struct bio *wbi; |
1022 | ||
1023 | spin_lock(&sh->lock); | |
cbe47ec5 | 1024 | spin_lock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1025 | chosen = dev->towrite; |
1026 | dev->towrite = NULL; | |
1027 | BUG_ON(dev->written); | |
1028 | wbi = dev->written = chosen; | |
cbe47ec5 | 1029 | spin_unlock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1030 | spin_unlock(&sh->lock); |
1031 | ||
1032 | while (wbi && wbi->bi_sector < | |
1033 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1034 | if (wbi->bi_rw & REQ_FUA) |
1035 | set_bit(R5_WantFUA, &dev->flags); | |
91c00924 DW |
1036 | tx = async_copy_data(1, wbi, dev->page, |
1037 | dev->sector, tx); | |
1038 | wbi = r5_next_bio(wbi, dev->sector); | |
1039 | } | |
1040 | } | |
1041 | } | |
1042 | ||
1043 | return tx; | |
1044 | } | |
1045 | ||
ac6b53b6 | 1046 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1047 | { |
1048 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1049 | int disks = sh->disks; |
1050 | int pd_idx = sh->pd_idx; | |
1051 | int qd_idx = sh->qd_idx; | |
1052 | int i; | |
e9c7469b | 1053 | bool fua = false; |
91c00924 | 1054 | |
e46b272b | 1055 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1056 | (unsigned long long)sh->sector); |
1057 | ||
e9c7469b TH |
1058 | for (i = disks; i--; ) |
1059 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); | |
1060 | ||
91c00924 DW |
1061 | for (i = disks; i--; ) { |
1062 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1063 | |
e9c7469b | 1064 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1065 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1066 | if (fua) |
1067 | set_bit(R5_WantFUA, &dev->flags); | |
1068 | } | |
91c00924 DW |
1069 | } |
1070 | ||
d8ee0728 DW |
1071 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1072 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1073 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1074 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1075 | else { | |
1076 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1077 | sh->reconstruct_state = reconstruct_state_result; | |
1078 | } | |
91c00924 DW |
1079 | |
1080 | set_bit(STRIPE_HANDLE, &sh->state); | |
1081 | release_stripe(sh); | |
1082 | } | |
1083 | ||
1084 | static void | |
ac6b53b6 DW |
1085 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1086 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1087 | { |
91c00924 | 1088 | int disks = sh->disks; |
d6f38f31 | 1089 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1090 | struct async_submit_ctl submit; |
91c00924 DW |
1091 | int count = 0, pd_idx = sh->pd_idx, i; |
1092 | struct page *xor_dest; | |
d8ee0728 | 1093 | int prexor = 0; |
91c00924 | 1094 | unsigned long flags; |
91c00924 | 1095 | |
e46b272b | 1096 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1097 | (unsigned long long)sh->sector); |
1098 | ||
1099 | /* check if prexor is active which means only process blocks | |
1100 | * that are part of a read-modify-write (written) | |
1101 | */ | |
d8ee0728 DW |
1102 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1103 | prexor = 1; | |
91c00924 DW |
1104 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1105 | for (i = disks; i--; ) { | |
1106 | struct r5dev *dev = &sh->dev[i]; | |
1107 | if (dev->written) | |
1108 | xor_srcs[count++] = dev->page; | |
1109 | } | |
1110 | } else { | |
1111 | xor_dest = sh->dev[pd_idx].page; | |
1112 | for (i = disks; i--; ) { | |
1113 | struct r5dev *dev = &sh->dev[i]; | |
1114 | if (i != pd_idx) | |
1115 | xor_srcs[count++] = dev->page; | |
1116 | } | |
1117 | } | |
1118 | ||
91c00924 DW |
1119 | /* 1/ if we prexor'd then the dest is reused as a source |
1120 | * 2/ if we did not prexor then we are redoing the parity | |
1121 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1122 | * for the synchronous xor case | |
1123 | */ | |
88ba2aa5 | 1124 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1125 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1126 | ||
1127 | atomic_inc(&sh->count); | |
1128 | ||
ac6b53b6 | 1129 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1130 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1131 | if (unlikely(count == 1)) |
1132 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1133 | else | |
1134 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1135 | } |
1136 | ||
ac6b53b6 DW |
1137 | static void |
1138 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1139 | struct dma_async_tx_descriptor *tx) | |
1140 | { | |
1141 | struct async_submit_ctl submit; | |
1142 | struct page **blocks = percpu->scribble; | |
1143 | int count; | |
1144 | ||
1145 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1146 | ||
1147 | count = set_syndrome_sources(blocks, sh); | |
1148 | ||
1149 | atomic_inc(&sh->count); | |
1150 | ||
1151 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1152 | sh, to_addr_conv(sh, percpu)); | |
1153 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1154 | } |
1155 | ||
1156 | static void ops_complete_check(void *stripe_head_ref) | |
1157 | { | |
1158 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1159 | |
e46b272b | 1160 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1161 | (unsigned long long)sh->sector); |
1162 | ||
ecc65c9b | 1163 | sh->check_state = check_state_check_result; |
91c00924 DW |
1164 | set_bit(STRIPE_HANDLE, &sh->state); |
1165 | release_stripe(sh); | |
1166 | } | |
1167 | ||
ac6b53b6 | 1168 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1169 | { |
91c00924 | 1170 | int disks = sh->disks; |
ac6b53b6 DW |
1171 | int pd_idx = sh->pd_idx; |
1172 | int qd_idx = sh->qd_idx; | |
1173 | struct page *xor_dest; | |
d6f38f31 | 1174 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1175 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1176 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1177 | int count; |
1178 | int i; | |
91c00924 | 1179 | |
e46b272b | 1180 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1181 | (unsigned long long)sh->sector); |
1182 | ||
ac6b53b6 DW |
1183 | count = 0; |
1184 | xor_dest = sh->dev[pd_idx].page; | |
1185 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1186 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1187 | if (i == pd_idx || i == qd_idx) |
1188 | continue; | |
1189 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1190 | } |
1191 | ||
d6f38f31 DW |
1192 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1193 | to_addr_conv(sh, percpu)); | |
099f53cb | 1194 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1195 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1196 | |
91c00924 | 1197 | atomic_inc(&sh->count); |
a08abd8c DW |
1198 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1199 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1200 | } |
1201 | ||
ac6b53b6 DW |
1202 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1203 | { | |
1204 | struct page **srcs = percpu->scribble; | |
1205 | struct async_submit_ctl submit; | |
1206 | int count; | |
1207 | ||
1208 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1209 | (unsigned long long)sh->sector, checkp); | |
1210 | ||
1211 | count = set_syndrome_sources(srcs, sh); | |
1212 | if (!checkp) | |
1213 | srcs[count] = NULL; | |
91c00924 | 1214 | |
91c00924 | 1215 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1216 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1217 | sh, to_addr_conv(sh, percpu)); | |
1218 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1219 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1220 | } |
1221 | ||
417b8d4a | 1222 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1223 | { |
1224 | int overlap_clear = 0, i, disks = sh->disks; | |
1225 | struct dma_async_tx_descriptor *tx = NULL; | |
d6f38f31 | 1226 | raid5_conf_t *conf = sh->raid_conf; |
ac6b53b6 | 1227 | int level = conf->level; |
d6f38f31 DW |
1228 | struct raid5_percpu *percpu; |
1229 | unsigned long cpu; | |
91c00924 | 1230 | |
d6f38f31 DW |
1231 | cpu = get_cpu(); |
1232 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1233 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1234 | ops_run_biofill(sh); |
1235 | overlap_clear++; | |
1236 | } | |
1237 | ||
7b3a871e | 1238 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1239 | if (level < 6) |
1240 | tx = ops_run_compute5(sh, percpu); | |
1241 | else { | |
1242 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1243 | tx = ops_run_compute6_1(sh, percpu); | |
1244 | else | |
1245 | tx = ops_run_compute6_2(sh, percpu); | |
1246 | } | |
1247 | /* terminate the chain if reconstruct is not set to be run */ | |
1248 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1249 | async_tx_ack(tx); |
1250 | } | |
91c00924 | 1251 | |
600aa109 | 1252 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1253 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1254 | |
600aa109 | 1255 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1256 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1257 | overlap_clear++; |
1258 | } | |
1259 | ||
ac6b53b6 DW |
1260 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1261 | if (level < 6) | |
1262 | ops_run_reconstruct5(sh, percpu, tx); | |
1263 | else | |
1264 | ops_run_reconstruct6(sh, percpu, tx); | |
1265 | } | |
91c00924 | 1266 | |
ac6b53b6 DW |
1267 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1268 | if (sh->check_state == check_state_run) | |
1269 | ops_run_check_p(sh, percpu); | |
1270 | else if (sh->check_state == check_state_run_q) | |
1271 | ops_run_check_pq(sh, percpu, 0); | |
1272 | else if (sh->check_state == check_state_run_pq) | |
1273 | ops_run_check_pq(sh, percpu, 1); | |
1274 | else | |
1275 | BUG(); | |
1276 | } | |
91c00924 | 1277 | |
91c00924 DW |
1278 | if (overlap_clear) |
1279 | for (i = disks; i--; ) { | |
1280 | struct r5dev *dev = &sh->dev[i]; | |
1281 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1282 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1283 | } | |
d6f38f31 | 1284 | put_cpu(); |
91c00924 DW |
1285 | } |
1286 | ||
417b8d4a DW |
1287 | #ifdef CONFIG_MULTICORE_RAID456 |
1288 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1289 | { | |
1290 | struct stripe_head *sh = param; | |
1291 | unsigned long ops_request = sh->ops.request; | |
1292 | ||
1293 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1294 | wake_up(&sh->ops.wait_for_ops); | |
1295 | ||
1296 | __raid_run_ops(sh, ops_request); | |
1297 | release_stripe(sh); | |
1298 | } | |
1299 | ||
1300 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1301 | { | |
1302 | /* since handle_stripe can be called outside of raid5d context | |
1303 | * we need to ensure sh->ops.request is de-staged before another | |
1304 | * request arrives | |
1305 | */ | |
1306 | wait_event(sh->ops.wait_for_ops, | |
1307 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1308 | sh->ops.request = ops_request; | |
1309 | ||
1310 | atomic_inc(&sh->count); | |
1311 | async_schedule(async_run_ops, sh); | |
1312 | } | |
1313 | #else | |
1314 | #define raid_run_ops __raid_run_ops | |
1315 | #endif | |
1316 | ||
3f294f4f | 1317 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1318 | { |
1319 | struct stripe_head *sh; | |
6ce32846 | 1320 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1321 | if (!sh) |
1322 | return 0; | |
6ce32846 | 1323 | |
3f294f4f N |
1324 | sh->raid_conf = conf; |
1325 | spin_lock_init(&sh->lock); | |
417b8d4a DW |
1326 | #ifdef CONFIG_MULTICORE_RAID456 |
1327 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1328 | #endif | |
3f294f4f | 1329 | |
e4e11e38 N |
1330 | if (grow_buffers(sh)) { |
1331 | shrink_buffers(sh); | |
3f294f4f N |
1332 | kmem_cache_free(conf->slab_cache, sh); |
1333 | return 0; | |
1334 | } | |
1335 | /* we just created an active stripe so... */ | |
1336 | atomic_set(&sh->count, 1); | |
1337 | atomic_inc(&conf->active_stripes); | |
1338 | INIT_LIST_HEAD(&sh->lru); | |
1339 | release_stripe(sh); | |
1340 | return 1; | |
1341 | } | |
1342 | ||
1343 | static int grow_stripes(raid5_conf_t *conf, int num) | |
1344 | { | |
e18b890b | 1345 | struct kmem_cache *sc; |
5e5e3e78 | 1346 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1347 | |
f4be6b43 N |
1348 | if (conf->mddev->gendisk) |
1349 | sprintf(conf->cache_name[0], | |
1350 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1351 | else | |
1352 | sprintf(conf->cache_name[0], | |
1353 | "raid%d-%p", conf->level, conf->mddev); | |
1354 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1355 | ||
ad01c9e3 N |
1356 | conf->active_name = 0; |
1357 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1358 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1359 | 0, 0, NULL); |
1da177e4 LT |
1360 | if (!sc) |
1361 | return 1; | |
1362 | conf->slab_cache = sc; | |
ad01c9e3 | 1363 | conf->pool_size = devs; |
16a53ecc | 1364 | while (num--) |
3f294f4f | 1365 | if (!grow_one_stripe(conf)) |
1da177e4 | 1366 | return 1; |
1da177e4 LT |
1367 | return 0; |
1368 | } | |
29269553 | 1369 | |
d6f38f31 DW |
1370 | /** |
1371 | * scribble_len - return the required size of the scribble region | |
1372 | * @num - total number of disks in the array | |
1373 | * | |
1374 | * The size must be enough to contain: | |
1375 | * 1/ a struct page pointer for each device in the array +2 | |
1376 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1377 | * (dma_map_page()) or page (page_address()) address. | |
1378 | * | |
1379 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1380 | * calculate over all devices (not just the data blocks), using zeros in place | |
1381 | * of the P and Q blocks. | |
1382 | */ | |
1383 | static size_t scribble_len(int num) | |
1384 | { | |
1385 | size_t len; | |
1386 | ||
1387 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1388 | ||
1389 | return len; | |
1390 | } | |
1391 | ||
ad01c9e3 N |
1392 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
1393 | { | |
1394 | /* Make all the stripes able to hold 'newsize' devices. | |
1395 | * New slots in each stripe get 'page' set to a new page. | |
1396 | * | |
1397 | * This happens in stages: | |
1398 | * 1/ create a new kmem_cache and allocate the required number of | |
1399 | * stripe_heads. | |
1400 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1401 | * to the new stripe_heads. This will have the side effect of | |
1402 | * freezing the array as once all stripe_heads have been collected, | |
1403 | * no IO will be possible. Old stripe heads are freed once their | |
1404 | * pages have been transferred over, and the old kmem_cache is | |
1405 | * freed when all stripes are done. | |
1406 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1407 | * we simple return a failre status - no need to clean anything up. | |
1408 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1409 | * If this fails, we don't bother trying the shrink the | |
1410 | * stripe_heads down again, we just leave them as they are. | |
1411 | * As each stripe_head is processed the new one is released into | |
1412 | * active service. | |
1413 | * | |
1414 | * Once step2 is started, we cannot afford to wait for a write, | |
1415 | * so we use GFP_NOIO allocations. | |
1416 | */ | |
1417 | struct stripe_head *osh, *nsh; | |
1418 | LIST_HEAD(newstripes); | |
1419 | struct disk_info *ndisks; | |
d6f38f31 | 1420 | unsigned long cpu; |
b5470dc5 | 1421 | int err; |
e18b890b | 1422 | struct kmem_cache *sc; |
ad01c9e3 N |
1423 | int i; |
1424 | ||
1425 | if (newsize <= conf->pool_size) | |
1426 | return 0; /* never bother to shrink */ | |
1427 | ||
b5470dc5 DW |
1428 | err = md_allow_write(conf->mddev); |
1429 | if (err) | |
1430 | return err; | |
2a2275d6 | 1431 | |
ad01c9e3 N |
1432 | /* Step 1 */ |
1433 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1434 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1435 | 0, 0, NULL); |
ad01c9e3 N |
1436 | if (!sc) |
1437 | return -ENOMEM; | |
1438 | ||
1439 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1440 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1441 | if (!nsh) |
1442 | break; | |
1443 | ||
ad01c9e3 N |
1444 | nsh->raid_conf = conf; |
1445 | spin_lock_init(&nsh->lock); | |
417b8d4a DW |
1446 | #ifdef CONFIG_MULTICORE_RAID456 |
1447 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1448 | #endif | |
ad01c9e3 N |
1449 | |
1450 | list_add(&nsh->lru, &newstripes); | |
1451 | } | |
1452 | if (i) { | |
1453 | /* didn't get enough, give up */ | |
1454 | while (!list_empty(&newstripes)) { | |
1455 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1456 | list_del(&nsh->lru); | |
1457 | kmem_cache_free(sc, nsh); | |
1458 | } | |
1459 | kmem_cache_destroy(sc); | |
1460 | return -ENOMEM; | |
1461 | } | |
1462 | /* Step 2 - Must use GFP_NOIO now. | |
1463 | * OK, we have enough stripes, start collecting inactive | |
1464 | * stripes and copying them over | |
1465 | */ | |
1466 | list_for_each_entry(nsh, &newstripes, lru) { | |
1467 | spin_lock_irq(&conf->device_lock); | |
1468 | wait_event_lock_irq(conf->wait_for_stripe, | |
1469 | !list_empty(&conf->inactive_list), | |
1470 | conf->device_lock, | |
482c0834 | 1471 | ); |
ad01c9e3 N |
1472 | osh = get_free_stripe(conf); |
1473 | spin_unlock_irq(&conf->device_lock); | |
1474 | atomic_set(&nsh->count, 1); | |
1475 | for(i=0; i<conf->pool_size; i++) | |
1476 | nsh->dev[i].page = osh->dev[i].page; | |
1477 | for( ; i<newsize; i++) | |
1478 | nsh->dev[i].page = NULL; | |
1479 | kmem_cache_free(conf->slab_cache, osh); | |
1480 | } | |
1481 | kmem_cache_destroy(conf->slab_cache); | |
1482 | ||
1483 | /* Step 3. | |
1484 | * At this point, we are holding all the stripes so the array | |
1485 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1486 | * conf->disks and the scribble region |
ad01c9e3 N |
1487 | */ |
1488 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1489 | if (ndisks) { | |
1490 | for (i=0; i<conf->raid_disks; i++) | |
1491 | ndisks[i] = conf->disks[i]; | |
1492 | kfree(conf->disks); | |
1493 | conf->disks = ndisks; | |
1494 | } else | |
1495 | err = -ENOMEM; | |
1496 | ||
d6f38f31 DW |
1497 | get_online_cpus(); |
1498 | conf->scribble_len = scribble_len(newsize); | |
1499 | for_each_present_cpu(cpu) { | |
1500 | struct raid5_percpu *percpu; | |
1501 | void *scribble; | |
1502 | ||
1503 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1504 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1505 | ||
1506 | if (scribble) { | |
1507 | kfree(percpu->scribble); | |
1508 | percpu->scribble = scribble; | |
1509 | } else { | |
1510 | err = -ENOMEM; | |
1511 | break; | |
1512 | } | |
1513 | } | |
1514 | put_online_cpus(); | |
1515 | ||
ad01c9e3 N |
1516 | /* Step 4, return new stripes to service */ |
1517 | while(!list_empty(&newstripes)) { | |
1518 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1519 | list_del_init(&nsh->lru); | |
d6f38f31 | 1520 | |
ad01c9e3 N |
1521 | for (i=conf->raid_disks; i < newsize; i++) |
1522 | if (nsh->dev[i].page == NULL) { | |
1523 | struct page *p = alloc_page(GFP_NOIO); | |
1524 | nsh->dev[i].page = p; | |
1525 | if (!p) | |
1526 | err = -ENOMEM; | |
1527 | } | |
1528 | release_stripe(nsh); | |
1529 | } | |
1530 | /* critical section pass, GFP_NOIO no longer needed */ | |
1531 | ||
1532 | conf->slab_cache = sc; | |
1533 | conf->active_name = 1-conf->active_name; | |
1534 | conf->pool_size = newsize; | |
1535 | return err; | |
1536 | } | |
1da177e4 | 1537 | |
3f294f4f | 1538 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1539 | { |
1540 | struct stripe_head *sh; | |
1541 | ||
3f294f4f N |
1542 | spin_lock_irq(&conf->device_lock); |
1543 | sh = get_free_stripe(conf); | |
1544 | spin_unlock_irq(&conf->device_lock); | |
1545 | if (!sh) | |
1546 | return 0; | |
78bafebd | 1547 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1548 | shrink_buffers(sh); |
3f294f4f N |
1549 | kmem_cache_free(conf->slab_cache, sh); |
1550 | atomic_dec(&conf->active_stripes); | |
1551 | return 1; | |
1552 | } | |
1553 | ||
1554 | static void shrink_stripes(raid5_conf_t *conf) | |
1555 | { | |
1556 | while (drop_one_stripe(conf)) | |
1557 | ; | |
1558 | ||
29fc7e3e N |
1559 | if (conf->slab_cache) |
1560 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1561 | conf->slab_cache = NULL; |
1562 | } | |
1563 | ||
6712ecf8 | 1564 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1565 | { |
99c0fb5f | 1566 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1567 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1568 | int disks = sh->disks, i; |
1da177e4 | 1569 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1570 | char b[BDEVNAME_SIZE]; |
1571 | mdk_rdev_t *rdev; | |
1da177e4 | 1572 | |
1da177e4 LT |
1573 | |
1574 | for (i=0 ; i<disks; i++) | |
1575 | if (bi == &sh->dev[i].req) | |
1576 | break; | |
1577 | ||
45b4233c DW |
1578 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1579 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1580 | uptodate); |
1581 | if (i == disks) { | |
1582 | BUG(); | |
6712ecf8 | 1583 | return; |
1da177e4 LT |
1584 | } |
1585 | ||
1586 | if (uptodate) { | |
1da177e4 | 1587 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1588 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1589 | rdev = conf->disks[i].rdev; |
0c55e022 | 1590 | printk_rl(KERN_INFO "md/raid:%s: read error corrected" |
6be9d494 BS |
1591 | " (%lu sectors at %llu on %s)\n", |
1592 | mdname(conf->mddev), STRIPE_SECTORS, | |
1593 | (unsigned long long)(sh->sector | |
1594 | + rdev->data_offset), | |
1595 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1596 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1597 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1598 | } | |
ba22dcbf N |
1599 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1600 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1601 | } else { |
d6950432 | 1602 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1603 | int retry = 0; |
d6950432 N |
1604 | rdev = conf->disks[i].rdev; |
1605 | ||
1da177e4 | 1606 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1607 | atomic_inc(&rdev->read_errors); |
7b0bb536 | 1608 | if (conf->mddev->degraded >= conf->max_degraded) |
6be9d494 | 1609 | printk_rl(KERN_WARNING |
0c55e022 | 1610 | "md/raid:%s: read error not correctable " |
6be9d494 BS |
1611 | "(sector %llu on %s).\n", |
1612 | mdname(conf->mddev), | |
1613 | (unsigned long long)(sh->sector | |
1614 | + rdev->data_offset), | |
1615 | bdn); | |
ba22dcbf | 1616 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1617 | /* Oh, no!!! */ |
6be9d494 | 1618 | printk_rl(KERN_WARNING |
0c55e022 | 1619 | "md/raid:%s: read error NOT corrected!! " |
6be9d494 BS |
1620 | "(sector %llu on %s).\n", |
1621 | mdname(conf->mddev), | |
1622 | (unsigned long long)(sh->sector | |
1623 | + rdev->data_offset), | |
1624 | bdn); | |
d6950432 | 1625 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1626 | > conf->max_nr_stripes) |
14f8d26b | 1627 | printk(KERN_WARNING |
0c55e022 | 1628 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1629 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1630 | else |
1631 | retry = 1; | |
1632 | if (retry) | |
1633 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1634 | else { | |
4e5314b5 N |
1635 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1636 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1637 | md_error(conf->mddev, rdev); |
ba22dcbf | 1638 | } |
1da177e4 LT |
1639 | } |
1640 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1641 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1642 | set_bit(STRIPE_HANDLE, &sh->state); | |
1643 | release_stripe(sh); | |
1da177e4 LT |
1644 | } |
1645 | ||
d710e138 | 1646 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1647 | { |
99c0fb5f | 1648 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1649 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1650 | int disks = sh->disks, i; |
1da177e4 LT |
1651 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1652 | ||
1da177e4 LT |
1653 | for (i=0 ; i<disks; i++) |
1654 | if (bi == &sh->dev[i].req) | |
1655 | break; | |
1656 | ||
45b4233c | 1657 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1658 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1659 | uptodate); | |
1660 | if (i == disks) { | |
1661 | BUG(); | |
6712ecf8 | 1662 | return; |
1da177e4 LT |
1663 | } |
1664 | ||
1da177e4 LT |
1665 | if (!uptodate) |
1666 | md_error(conf->mddev, conf->disks[i].rdev); | |
1667 | ||
1668 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1669 | ||
1670 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1671 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1672 | release_stripe(sh); |
1da177e4 LT |
1673 | } |
1674 | ||
1675 | ||
784052ec | 1676 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1677 | |
784052ec | 1678 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1679 | { |
1680 | struct r5dev *dev = &sh->dev[i]; | |
1681 | ||
1682 | bio_init(&dev->req); | |
1683 | dev->req.bi_io_vec = &dev->vec; | |
1684 | dev->req.bi_vcnt++; | |
1685 | dev->req.bi_max_vecs++; | |
1686 | dev->vec.bv_page = dev->page; | |
1687 | dev->vec.bv_len = STRIPE_SIZE; | |
1688 | dev->vec.bv_offset = 0; | |
1689 | ||
1690 | dev->req.bi_sector = sh->sector; | |
1691 | dev->req.bi_private = sh; | |
1692 | ||
1693 | dev->flags = 0; | |
784052ec | 1694 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1695 | } |
1696 | ||
1697 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1698 | { | |
1699 | char b[BDEVNAME_SIZE]; | |
7b92813c | 1700 | raid5_conf_t *conf = mddev->private; |
0c55e022 | 1701 | pr_debug("raid456: error called\n"); |
1da177e4 | 1702 | |
6f8d0c77 N |
1703 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1704 | unsigned long flags; | |
1705 | spin_lock_irqsave(&conf->device_lock, flags); | |
1706 | mddev->degraded++; | |
1707 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1708 | /* | |
1709 | * if recovery was running, make sure it aborts. | |
1710 | */ | |
1711 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1da177e4 | 1712 | } |
6f8d0c77 N |
1713 | set_bit(Faulty, &rdev->flags); |
1714 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1715 | printk(KERN_ALERT | |
1716 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1717 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1718 | mdname(mddev), | |
1719 | bdevname(rdev->bdev, b), | |
1720 | mdname(mddev), | |
1721 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1722 | } |
1da177e4 LT |
1723 | |
1724 | /* | |
1725 | * Input: a 'big' sector number, | |
1726 | * Output: index of the data and parity disk, and the sector # in them. | |
1727 | */ | |
112bf897 | 1728 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1729 | int previous, int *dd_idx, |
1730 | struct stripe_head *sh) | |
1da177e4 | 1731 | { |
6e3b96ed | 1732 | sector_t stripe, stripe2; |
35f2a591 | 1733 | sector_t chunk_number; |
1da177e4 | 1734 | unsigned int chunk_offset; |
911d4ee8 | 1735 | int pd_idx, qd_idx; |
67cc2b81 | 1736 | int ddf_layout = 0; |
1da177e4 | 1737 | sector_t new_sector; |
e183eaed N |
1738 | int algorithm = previous ? conf->prev_algo |
1739 | : conf->algorithm; | |
09c9e5fa AN |
1740 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1741 | : conf->chunk_sectors; | |
112bf897 N |
1742 | int raid_disks = previous ? conf->previous_raid_disks |
1743 | : conf->raid_disks; | |
1744 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1745 | |
1746 | /* First compute the information on this sector */ | |
1747 | ||
1748 | /* | |
1749 | * Compute the chunk number and the sector offset inside the chunk | |
1750 | */ | |
1751 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1752 | chunk_number = r_sector; | |
1da177e4 LT |
1753 | |
1754 | /* | |
1755 | * Compute the stripe number | |
1756 | */ | |
35f2a591 N |
1757 | stripe = chunk_number; |
1758 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1759 | stripe2 = stripe; |
1da177e4 LT |
1760 | /* |
1761 | * Select the parity disk based on the user selected algorithm. | |
1762 | */ | |
911d4ee8 | 1763 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1764 | switch(conf->level) { |
1765 | case 4: | |
911d4ee8 | 1766 | pd_idx = data_disks; |
16a53ecc N |
1767 | break; |
1768 | case 5: | |
e183eaed | 1769 | switch (algorithm) { |
1da177e4 | 1770 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1771 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1772 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1773 | (*dd_idx)++; |
1774 | break; | |
1775 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1776 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1777 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1778 | (*dd_idx)++; |
1779 | break; | |
1780 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1781 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1782 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1783 | break; |
1784 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1785 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1786 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 1787 | break; |
99c0fb5f N |
1788 | case ALGORITHM_PARITY_0: |
1789 | pd_idx = 0; | |
1790 | (*dd_idx)++; | |
1791 | break; | |
1792 | case ALGORITHM_PARITY_N: | |
1793 | pd_idx = data_disks; | |
1794 | break; | |
1da177e4 | 1795 | default: |
99c0fb5f | 1796 | BUG(); |
16a53ecc N |
1797 | } |
1798 | break; | |
1799 | case 6: | |
1800 | ||
e183eaed | 1801 | switch (algorithm) { |
16a53ecc | 1802 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1803 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1804 | qd_idx = pd_idx + 1; |
1805 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1806 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1807 | qd_idx = 0; |
1808 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1809 | (*dd_idx) += 2; /* D D P Q D */ |
1810 | break; | |
1811 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1812 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1813 | qd_idx = pd_idx + 1; |
1814 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1815 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1816 | qd_idx = 0; |
1817 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1818 | (*dd_idx) += 2; /* D D P Q D */ |
1819 | break; | |
1820 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1821 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1822 | qd_idx = (pd_idx + 1) % raid_disks; |
1823 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1824 | break; |
1825 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1826 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1827 | qd_idx = (pd_idx + 1) % raid_disks; |
1828 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1829 | break; |
99c0fb5f N |
1830 | |
1831 | case ALGORITHM_PARITY_0: | |
1832 | pd_idx = 0; | |
1833 | qd_idx = 1; | |
1834 | (*dd_idx) += 2; | |
1835 | break; | |
1836 | case ALGORITHM_PARITY_N: | |
1837 | pd_idx = data_disks; | |
1838 | qd_idx = data_disks + 1; | |
1839 | break; | |
1840 | ||
1841 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1842 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1843 | * of blocks for computing Q is different. | |
1844 | */ | |
6e3b96ed | 1845 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
1846 | qd_idx = pd_idx + 1; |
1847 | if (pd_idx == raid_disks-1) { | |
1848 | (*dd_idx)++; /* Q D D D P */ | |
1849 | qd_idx = 0; | |
1850 | } else if (*dd_idx >= pd_idx) | |
1851 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1852 | ddf_layout = 1; |
99c0fb5f N |
1853 | break; |
1854 | ||
1855 | case ALGORITHM_ROTATING_N_RESTART: | |
1856 | /* Same a left_asymmetric, by first stripe is | |
1857 | * D D D P Q rather than | |
1858 | * Q D D D P | |
1859 | */ | |
6e3b96ed N |
1860 | stripe2 += 1; |
1861 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
1862 | qd_idx = pd_idx + 1; |
1863 | if (pd_idx == raid_disks-1) { | |
1864 | (*dd_idx)++; /* Q D D D P */ | |
1865 | qd_idx = 0; | |
1866 | } else if (*dd_idx >= pd_idx) | |
1867 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1868 | ddf_layout = 1; |
99c0fb5f N |
1869 | break; |
1870 | ||
1871 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1872 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 1873 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
1874 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
1875 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1876 | ddf_layout = 1; |
99c0fb5f N |
1877 | break; |
1878 | ||
1879 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1880 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 1881 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1882 | if (*dd_idx >= pd_idx) |
1883 | (*dd_idx)++; | |
1884 | qd_idx = raid_disks - 1; | |
1885 | break; | |
1886 | ||
1887 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 1888 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1889 | if (*dd_idx >= pd_idx) |
1890 | (*dd_idx)++; | |
1891 | qd_idx = raid_disks - 1; | |
1892 | break; | |
1893 | ||
1894 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 1895 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1896 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1897 | qd_idx = raid_disks - 1; | |
1898 | break; | |
1899 | ||
1900 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 1901 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1902 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1903 | qd_idx = raid_disks - 1; | |
1904 | break; | |
1905 | ||
1906 | case ALGORITHM_PARITY_0_6: | |
1907 | pd_idx = 0; | |
1908 | (*dd_idx)++; | |
1909 | qd_idx = raid_disks - 1; | |
1910 | break; | |
1911 | ||
16a53ecc | 1912 | default: |
99c0fb5f | 1913 | BUG(); |
16a53ecc N |
1914 | } |
1915 | break; | |
1da177e4 LT |
1916 | } |
1917 | ||
911d4ee8 N |
1918 | if (sh) { |
1919 | sh->pd_idx = pd_idx; | |
1920 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1921 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1922 | } |
1da177e4 LT |
1923 | /* |
1924 | * Finally, compute the new sector number | |
1925 | */ | |
1926 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1927 | return new_sector; | |
1928 | } | |
1929 | ||
1930 | ||
784052ec | 1931 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1932 | { |
1933 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1934 | int raid_disks = sh->disks; |
1935 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1936 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
1937 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1938 | : conf->chunk_sectors; | |
e183eaed N |
1939 | int algorithm = previous ? conf->prev_algo |
1940 | : conf->algorithm; | |
1da177e4 LT |
1941 | sector_t stripe; |
1942 | int chunk_offset; | |
35f2a591 N |
1943 | sector_t chunk_number; |
1944 | int dummy1, dd_idx = i; | |
1da177e4 | 1945 | sector_t r_sector; |
911d4ee8 | 1946 | struct stripe_head sh2; |
1da177e4 | 1947 | |
16a53ecc | 1948 | |
1da177e4 LT |
1949 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1950 | stripe = new_sector; | |
1da177e4 | 1951 | |
16a53ecc N |
1952 | if (i == sh->pd_idx) |
1953 | return 0; | |
1954 | switch(conf->level) { | |
1955 | case 4: break; | |
1956 | case 5: | |
e183eaed | 1957 | switch (algorithm) { |
1da177e4 LT |
1958 | case ALGORITHM_LEFT_ASYMMETRIC: |
1959 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1960 | if (i > sh->pd_idx) | |
1961 | i--; | |
1962 | break; | |
1963 | case ALGORITHM_LEFT_SYMMETRIC: | |
1964 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1965 | if (i < sh->pd_idx) | |
1966 | i += raid_disks; | |
1967 | i -= (sh->pd_idx + 1); | |
1968 | break; | |
99c0fb5f N |
1969 | case ALGORITHM_PARITY_0: |
1970 | i -= 1; | |
1971 | break; | |
1972 | case ALGORITHM_PARITY_N: | |
1973 | break; | |
1da177e4 | 1974 | default: |
99c0fb5f | 1975 | BUG(); |
16a53ecc N |
1976 | } |
1977 | break; | |
1978 | case 6: | |
d0dabf7e | 1979 | if (i == sh->qd_idx) |
16a53ecc | 1980 | return 0; /* It is the Q disk */ |
e183eaed | 1981 | switch (algorithm) { |
16a53ecc N |
1982 | case ALGORITHM_LEFT_ASYMMETRIC: |
1983 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1984 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1985 | case ALGORITHM_ROTATING_N_RESTART: | |
1986 | if (sh->pd_idx == raid_disks-1) | |
1987 | i--; /* Q D D D P */ | |
16a53ecc N |
1988 | else if (i > sh->pd_idx) |
1989 | i -= 2; /* D D P Q D */ | |
1990 | break; | |
1991 | case ALGORITHM_LEFT_SYMMETRIC: | |
1992 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1993 | if (sh->pd_idx == raid_disks-1) | |
1994 | i--; /* Q D D D P */ | |
1995 | else { | |
1996 | /* D D P Q D */ | |
1997 | if (i < sh->pd_idx) | |
1998 | i += raid_disks; | |
1999 | i -= (sh->pd_idx + 2); | |
2000 | } | |
2001 | break; | |
99c0fb5f N |
2002 | case ALGORITHM_PARITY_0: |
2003 | i -= 2; | |
2004 | break; | |
2005 | case ALGORITHM_PARITY_N: | |
2006 | break; | |
2007 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2008 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2009 | if (sh->pd_idx == 0) |
2010 | i--; /* P D D D Q */ | |
e4424fee N |
2011 | else { |
2012 | /* D D Q P D */ | |
2013 | if (i < sh->pd_idx) | |
2014 | i += raid_disks; | |
2015 | i -= (sh->pd_idx + 1); | |
2016 | } | |
99c0fb5f N |
2017 | break; |
2018 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2019 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2020 | if (i > sh->pd_idx) | |
2021 | i--; | |
2022 | break; | |
2023 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2024 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2025 | if (i < sh->pd_idx) | |
2026 | i += data_disks + 1; | |
2027 | i -= (sh->pd_idx + 1); | |
2028 | break; | |
2029 | case ALGORITHM_PARITY_0_6: | |
2030 | i -= 1; | |
2031 | break; | |
16a53ecc | 2032 | default: |
99c0fb5f | 2033 | BUG(); |
16a53ecc N |
2034 | } |
2035 | break; | |
1da177e4 LT |
2036 | } |
2037 | ||
2038 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2039 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2040 | |
112bf897 | 2041 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2042 | previous, &dummy1, &sh2); |
911d4ee8 N |
2043 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2044 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2045 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2046 | mdname(conf->mddev)); | |
1da177e4 LT |
2047 | return 0; |
2048 | } | |
2049 | return r_sector; | |
2050 | } | |
2051 | ||
2052 | ||
600aa109 | 2053 | static void |
c0f7bddb | 2054 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2055 | int rcw, int expand) |
e33129d8 DW |
2056 | { |
2057 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
c0f7bddb YT |
2058 | raid5_conf_t *conf = sh->raid_conf; |
2059 | int level = conf->level; | |
e33129d8 DW |
2060 | |
2061 | if (rcw) { | |
2062 | /* if we are not expanding this is a proper write request, and | |
2063 | * there will be bios with new data to be drained into the | |
2064 | * stripe cache | |
2065 | */ | |
2066 | if (!expand) { | |
600aa109 DW |
2067 | sh->reconstruct_state = reconstruct_state_drain_run; |
2068 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2069 | } else | |
2070 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2071 | |
ac6b53b6 | 2072 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2073 | |
2074 | for (i = disks; i--; ) { | |
2075 | struct r5dev *dev = &sh->dev[i]; | |
2076 | ||
2077 | if (dev->towrite) { | |
2078 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2079 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2080 | if (!expand) |
2081 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2082 | s->locked++; |
e33129d8 DW |
2083 | } |
2084 | } | |
c0f7bddb | 2085 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2086 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2087 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2088 | } else { |
c0f7bddb | 2089 | BUG_ON(level == 6); |
e33129d8 DW |
2090 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2091 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2092 | ||
d8ee0728 | 2093 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2094 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2095 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2096 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2097 | |
2098 | for (i = disks; i--; ) { | |
2099 | struct r5dev *dev = &sh->dev[i]; | |
2100 | if (i == pd_idx) | |
2101 | continue; | |
2102 | ||
e33129d8 DW |
2103 | if (dev->towrite && |
2104 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2105 | test_bit(R5_Wantcompute, &dev->flags))) { |
2106 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2107 | set_bit(R5_LOCKED, &dev->flags); |
2108 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2109 | s->locked++; |
e33129d8 DW |
2110 | } |
2111 | } | |
2112 | } | |
2113 | ||
c0f7bddb | 2114 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2115 | * are in flight |
2116 | */ | |
2117 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2118 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2119 | s->locked++; |
e33129d8 | 2120 | |
c0f7bddb YT |
2121 | if (level == 6) { |
2122 | int qd_idx = sh->qd_idx; | |
2123 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2124 | ||
2125 | set_bit(R5_LOCKED, &dev->flags); | |
2126 | clear_bit(R5_UPTODATE, &dev->flags); | |
2127 | s->locked++; | |
2128 | } | |
2129 | ||
600aa109 | 2130 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2131 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2132 | s->locked, s->ops_request); |
e33129d8 | 2133 | } |
16a53ecc | 2134 | |
1da177e4 LT |
2135 | /* |
2136 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2137 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2138 | * The bi_next chain must be in order. |
2139 | */ | |
2140 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2141 | { | |
2142 | struct bio **bip; | |
2143 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 2144 | int firstwrite=0; |
1da177e4 | 2145 | |
cbe47ec5 | 2146 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2147 | (unsigned long long)bi->bi_sector, |
2148 | (unsigned long long)sh->sector); | |
2149 | ||
2150 | ||
2151 | spin_lock(&sh->lock); | |
2152 | spin_lock_irq(&conf->device_lock); | |
72626685 | 2153 | if (forwrite) { |
1da177e4 | 2154 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2155 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2156 | firstwrite = 1; | |
2157 | } else | |
1da177e4 LT |
2158 | bip = &sh->dev[dd_idx].toread; |
2159 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2160 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2161 | goto overlap; | |
2162 | bip = & (*bip)->bi_next; | |
2163 | } | |
2164 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2165 | goto overlap; | |
2166 | ||
78bafebd | 2167 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2168 | if (*bip) |
2169 | bi->bi_next = *bip; | |
2170 | *bip = bi; | |
960e739d | 2171 | bi->bi_phys_segments++; |
72626685 | 2172 | |
1da177e4 LT |
2173 | if (forwrite) { |
2174 | /* check if page is covered */ | |
2175 | sector_t sector = sh->dev[dd_idx].sector; | |
2176 | for (bi=sh->dev[dd_idx].towrite; | |
2177 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2178 | bi && bi->bi_sector <= sector; | |
2179 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2180 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2181 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2182 | } | |
2183 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2184 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2185 | } | |
cbe47ec5 N |
2186 | spin_unlock_irq(&conf->device_lock); |
2187 | spin_unlock(&sh->lock); | |
2188 | ||
2189 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2190 | (unsigned long long)(*bip)->bi_sector, | |
2191 | (unsigned long long)sh->sector, dd_idx); | |
2192 | ||
2193 | if (conf->mddev->bitmap && firstwrite) { | |
2194 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2195 | STRIPE_SECTORS, 0); | |
2196 | sh->bm_seq = conf->seq_flush+1; | |
2197 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2198 | } | |
1da177e4 LT |
2199 | return 1; |
2200 | ||
2201 | overlap: | |
2202 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2203 | spin_unlock_irq(&conf->device_lock); | |
2204 | spin_unlock(&sh->lock); | |
2205 | return 0; | |
2206 | } | |
2207 | ||
29269553 N |
2208 | static void end_reshape(raid5_conf_t *conf); |
2209 | ||
911d4ee8 N |
2210 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
2211 | struct stripe_head *sh) | |
ccfcc3c1 | 2212 | { |
784052ec | 2213 | int sectors_per_chunk = |
09c9e5fa | 2214 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2215 | int dd_idx; |
2d2063ce | 2216 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2217 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2218 | |
112bf897 N |
2219 | raid5_compute_sector(conf, |
2220 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2221 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2222 | previous, |
911d4ee8 | 2223 | &dd_idx, sh); |
ccfcc3c1 N |
2224 | } |
2225 | ||
a4456856 | 2226 | static void |
1fe797e6 | 2227 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2228 | struct stripe_head_state *s, int disks, |
2229 | struct bio **return_bi) | |
2230 | { | |
2231 | int i; | |
2232 | for (i = disks; i--; ) { | |
2233 | struct bio *bi; | |
2234 | int bitmap_end = 0; | |
2235 | ||
2236 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2237 | mdk_rdev_t *rdev; | |
2238 | rcu_read_lock(); | |
2239 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2240 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2241 | /* multiple read failures in one stripe */ | |
2242 | md_error(conf->mddev, rdev); | |
2243 | rcu_read_unlock(); | |
2244 | } | |
2245 | spin_lock_irq(&conf->device_lock); | |
2246 | /* fail all writes first */ | |
2247 | bi = sh->dev[i].towrite; | |
2248 | sh->dev[i].towrite = NULL; | |
2249 | if (bi) { | |
2250 | s->to_write--; | |
2251 | bitmap_end = 1; | |
2252 | } | |
2253 | ||
2254 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2255 | wake_up(&conf->wait_for_overlap); | |
2256 | ||
2257 | while (bi && bi->bi_sector < | |
2258 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2259 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2260 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2261 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2262 | md_write_end(conf->mddev); |
2263 | bi->bi_next = *return_bi; | |
2264 | *return_bi = bi; | |
2265 | } | |
2266 | bi = nextbi; | |
2267 | } | |
2268 | /* and fail all 'written' */ | |
2269 | bi = sh->dev[i].written; | |
2270 | sh->dev[i].written = NULL; | |
2271 | if (bi) bitmap_end = 1; | |
2272 | while (bi && bi->bi_sector < | |
2273 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2274 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2275 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2276 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2277 | md_write_end(conf->mddev); |
2278 | bi->bi_next = *return_bi; | |
2279 | *return_bi = bi; | |
2280 | } | |
2281 | bi = bi2; | |
2282 | } | |
2283 | ||
b5e98d65 DW |
2284 | /* fail any reads if this device is non-operational and |
2285 | * the data has not reached the cache yet. | |
2286 | */ | |
2287 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2288 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2289 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2290 | bi = sh->dev[i].toread; |
2291 | sh->dev[i].toread = NULL; | |
2292 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2293 | wake_up(&conf->wait_for_overlap); | |
2294 | if (bi) s->to_read--; | |
2295 | while (bi && bi->bi_sector < | |
2296 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2297 | struct bio *nextbi = | |
2298 | r5_next_bio(bi, sh->dev[i].sector); | |
2299 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2300 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2301 | bi->bi_next = *return_bi; |
2302 | *return_bi = bi; | |
2303 | } | |
2304 | bi = nextbi; | |
2305 | } | |
2306 | } | |
2307 | spin_unlock_irq(&conf->device_lock); | |
2308 | if (bitmap_end) | |
2309 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2310 | STRIPE_SECTORS, 0, 0); | |
2311 | } | |
2312 | ||
8b3e6cdc DW |
2313 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2314 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2315 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2316 | } |
2317 | ||
1fe797e6 DW |
2318 | /* fetch_block5 - checks the given member device to see if its data needs |
2319 | * to be read or computed to satisfy a request. | |
2320 | * | |
2321 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2322 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2323 | */ |
1fe797e6 DW |
2324 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2325 | int disk_idx, int disks) | |
f38e1219 DW |
2326 | { |
2327 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2328 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2329 | ||
f38e1219 DW |
2330 | /* is the data in this block needed, and can we get it? */ |
2331 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2332 | !test_bit(R5_UPTODATE, &dev->flags) && |
2333 | (dev->toread || | |
2334 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2335 | s->syncing || s->expanding || | |
2336 | (s->failed && | |
2337 | (failed_dev->toread || | |
2338 | (failed_dev->towrite && | |
2339 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2340 | /* We would like to get this block, possibly by computing it, |
2341 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2342 | */ |
2343 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2344 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2345 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2346 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2347 | set_bit(R5_Wantcompute, &dev->flags); |
2348 | sh->ops.target = disk_idx; | |
ac6b53b6 | 2349 | sh->ops.target2 = -1; |
f38e1219 | 2350 | s->req_compute = 1; |
f38e1219 | 2351 | /* Careful: from this point on 'uptodate' is in the eye |
ac6b53b6 | 2352 | * of raid_run_ops which services 'compute' operations |
f38e1219 DW |
2353 | * before writes. R5_Wantcompute flags a block that will |
2354 | * be R5_UPTODATE by the time it is needed for a | |
2355 | * subsequent operation. | |
2356 | */ | |
2357 | s->uptodate++; | |
1fe797e6 | 2358 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2359 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2360 | set_bit(R5_LOCKED, &dev->flags); |
2361 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2362 | s->locked++; |
2363 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2364 | s->syncing); | |
2365 | } | |
2366 | } | |
2367 | ||
1fe797e6 | 2368 | return 0; |
f38e1219 DW |
2369 | } |
2370 | ||
1fe797e6 DW |
2371 | /** |
2372 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2373 | */ | |
2374 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2375 | struct stripe_head_state *s, int disks) |
2376 | { | |
2377 | int i; | |
f38e1219 | 2378 | |
f38e1219 DW |
2379 | /* look for blocks to read/compute, skip this if a compute |
2380 | * is already in flight, or if the stripe contents are in the | |
2381 | * midst of changing due to a write | |
2382 | */ | |
976ea8d4 | 2383 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2384 | !sh->reconstruct_state) |
f38e1219 | 2385 | for (i = disks; i--; ) |
1fe797e6 | 2386 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2387 | break; |
a4456856 DW |
2388 | set_bit(STRIPE_HANDLE, &sh->state); |
2389 | } | |
2390 | ||
5599becc YT |
2391 | /* fetch_block6 - checks the given member device to see if its data needs |
2392 | * to be read or computed to satisfy a request. | |
2393 | * | |
2394 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2395 | * 0 to tell the loop in handle_stripe_fill6 to continue | |
2396 | */ | |
2397 | static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s, | |
2398 | struct r6_state *r6s, int disk_idx, int disks) | |
a4456856 | 2399 | { |
5599becc YT |
2400 | struct r5dev *dev = &sh->dev[disk_idx]; |
2401 | struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]], | |
2402 | &sh->dev[r6s->failed_num[1]] }; | |
2403 | ||
2404 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2405 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2406 | (dev->toread || | |
2407 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2408 | s->syncing || s->expanding || | |
2409 | (s->failed >= 1 && | |
2410 | (fdev[0]->toread || s->to_write)) || | |
2411 | (s->failed >= 2 && | |
2412 | (fdev[1]->toread || s->to_write)))) { | |
2413 | /* we would like to get this block, possibly by computing it, | |
2414 | * otherwise read it if the backing disk is insync | |
2415 | */ | |
2416 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2417 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2418 | if ((s->uptodate == disks - 1) && | |
2419 | (s->failed && (disk_idx == r6s->failed_num[0] || | |
2420 | disk_idx == r6s->failed_num[1]))) { | |
2421 | /* have disk failed, and we're requested to fetch it; | |
2422 | * do compute it | |
a4456856 | 2423 | */ |
5599becc YT |
2424 | pr_debug("Computing stripe %llu block %d\n", |
2425 | (unsigned long long)sh->sector, disk_idx); | |
2426 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2427 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2428 | set_bit(R5_Wantcompute, &dev->flags); | |
2429 | sh->ops.target = disk_idx; | |
2430 | sh->ops.target2 = -1; /* no 2nd target */ | |
2431 | s->req_compute = 1; | |
2432 | s->uptodate++; | |
2433 | return 1; | |
2434 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2435 | /* Computing 2-failure is *very* expensive; only | |
2436 | * do it if failed >= 2 | |
2437 | */ | |
2438 | int other; | |
2439 | for (other = disks; other--; ) { | |
2440 | if (other == disk_idx) | |
2441 | continue; | |
2442 | if (!test_bit(R5_UPTODATE, | |
2443 | &sh->dev[other].flags)) | |
2444 | break; | |
a4456856 | 2445 | } |
5599becc YT |
2446 | BUG_ON(other < 0); |
2447 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2448 | (unsigned long long)sh->sector, | |
2449 | disk_idx, other); | |
2450 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2451 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2452 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2453 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2454 | sh->ops.target = disk_idx; | |
2455 | sh->ops.target2 = other; | |
2456 | s->uptodate += 2; | |
2457 | s->req_compute = 1; | |
2458 | return 1; | |
2459 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2460 | set_bit(R5_LOCKED, &dev->flags); | |
2461 | set_bit(R5_Wantread, &dev->flags); | |
2462 | s->locked++; | |
2463 | pr_debug("Reading block %d (sync=%d)\n", | |
2464 | disk_idx, s->syncing); | |
a4456856 DW |
2465 | } |
2466 | } | |
5599becc YT |
2467 | |
2468 | return 0; | |
2469 | } | |
2470 | ||
2471 | /** | |
2472 | * handle_stripe_fill6 - read or compute data to satisfy pending requests. | |
2473 | */ | |
2474 | static void handle_stripe_fill6(struct stripe_head *sh, | |
2475 | struct stripe_head_state *s, struct r6_state *r6s, | |
2476 | int disks) | |
2477 | { | |
2478 | int i; | |
2479 | ||
2480 | /* look for blocks to read/compute, skip this if a compute | |
2481 | * is already in flight, or if the stripe contents are in the | |
2482 | * midst of changing due to a write | |
2483 | */ | |
2484 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2485 | !sh->reconstruct_state) | |
2486 | for (i = disks; i--; ) | |
2487 | if (fetch_block6(sh, s, r6s, i, disks)) | |
2488 | break; | |
a4456856 DW |
2489 | set_bit(STRIPE_HANDLE, &sh->state); |
2490 | } | |
2491 | ||
2492 | ||
1fe797e6 | 2493 | /* handle_stripe_clean_event |
a4456856 DW |
2494 | * any written block on an uptodate or failed drive can be returned. |
2495 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2496 | * never LOCKED, so we don't need to test 'failed' directly. | |
2497 | */ | |
1fe797e6 | 2498 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2499 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2500 | { | |
2501 | int i; | |
2502 | struct r5dev *dev; | |
2503 | ||
2504 | for (i = disks; i--; ) | |
2505 | if (sh->dev[i].written) { | |
2506 | dev = &sh->dev[i]; | |
2507 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2508 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2509 | /* We can return any write requests */ | |
2510 | struct bio *wbi, *wbi2; | |
2511 | int bitmap_end = 0; | |
45b4233c | 2512 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2513 | spin_lock_irq(&conf->device_lock); |
2514 | wbi = dev->written; | |
2515 | dev->written = NULL; | |
2516 | while (wbi && wbi->bi_sector < | |
2517 | dev->sector + STRIPE_SECTORS) { | |
2518 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2519 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2520 | md_write_end(conf->mddev); |
2521 | wbi->bi_next = *return_bi; | |
2522 | *return_bi = wbi; | |
2523 | } | |
2524 | wbi = wbi2; | |
2525 | } | |
2526 | if (dev->towrite == NULL) | |
2527 | bitmap_end = 1; | |
2528 | spin_unlock_irq(&conf->device_lock); | |
2529 | if (bitmap_end) | |
2530 | bitmap_endwrite(conf->mddev->bitmap, | |
2531 | sh->sector, | |
2532 | STRIPE_SECTORS, | |
2533 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2534 | 0); | |
2535 | } | |
2536 | } | |
8b3e6cdc DW |
2537 | |
2538 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2539 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2540 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2541 | } |
2542 | ||
1fe797e6 | 2543 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2544 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2545 | { | |
2546 | int rmw = 0, rcw = 0, i; | |
2547 | for (i = disks; i--; ) { | |
2548 | /* would I have to read this buffer for read_modify_write */ | |
2549 | struct r5dev *dev = &sh->dev[i]; | |
2550 | if ((dev->towrite || i == sh->pd_idx) && | |
2551 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2552 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2553 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2554 | if (test_bit(R5_Insync, &dev->flags)) |
2555 | rmw++; | |
2556 | else | |
2557 | rmw += 2*disks; /* cannot read it */ | |
2558 | } | |
2559 | /* Would I have to read this buffer for reconstruct_write */ | |
2560 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2561 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2562 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2563 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2564 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2565 | else |
2566 | rcw += 2*disks; | |
2567 | } | |
2568 | } | |
45b4233c | 2569 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2570 | (unsigned long long)sh->sector, rmw, rcw); |
2571 | set_bit(STRIPE_HANDLE, &sh->state); | |
2572 | if (rmw < rcw && rmw > 0) | |
2573 | /* prefer read-modify-write, but need to get some data */ | |
2574 | for (i = disks; i--; ) { | |
2575 | struct r5dev *dev = &sh->dev[i]; | |
2576 | if ((dev->towrite || i == sh->pd_idx) && | |
2577 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2578 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2579 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2580 | test_bit(R5_Insync, &dev->flags)) { |
2581 | if ( | |
2582 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2583 | pr_debug("Read_old block " |
a4456856 DW |
2584 | "%d for r-m-w\n", i); |
2585 | set_bit(R5_LOCKED, &dev->flags); | |
2586 | set_bit(R5_Wantread, &dev->flags); | |
2587 | s->locked++; | |
2588 | } else { | |
2589 | set_bit(STRIPE_DELAYED, &sh->state); | |
2590 | set_bit(STRIPE_HANDLE, &sh->state); | |
2591 | } | |
2592 | } | |
2593 | } | |
2594 | if (rcw <= rmw && rcw > 0) | |
2595 | /* want reconstruct write, but need to get some data */ | |
2596 | for (i = disks; i--; ) { | |
2597 | struct r5dev *dev = &sh->dev[i]; | |
2598 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2599 | i != sh->pd_idx && | |
2600 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2601 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2602 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2603 | test_bit(R5_Insync, &dev->flags)) { |
2604 | if ( | |
2605 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2606 | pr_debug("Read_old block " |
a4456856 DW |
2607 | "%d for Reconstruct\n", i); |
2608 | set_bit(R5_LOCKED, &dev->flags); | |
2609 | set_bit(R5_Wantread, &dev->flags); | |
2610 | s->locked++; | |
2611 | } else { | |
2612 | set_bit(STRIPE_DELAYED, &sh->state); | |
2613 | set_bit(STRIPE_HANDLE, &sh->state); | |
2614 | } | |
2615 | } | |
2616 | } | |
2617 | /* now if nothing is locked, and if we have enough data, | |
2618 | * we can start a write request | |
2619 | */ | |
f38e1219 DW |
2620 | /* since handle_stripe can be called at any time we need to handle the |
2621 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2622 | * subsequent call wants to start a write request. raid_run_ops only |
2623 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2624 | * simultaneously. If this is not the case then new writes need to be |
2625 | * held off until the compute completes. | |
2626 | */ | |
976ea8d4 DW |
2627 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2628 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2629 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2630 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2631 | } |
2632 | ||
1fe797e6 | 2633 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2634 | struct stripe_head *sh, struct stripe_head_state *s, |
2635 | struct r6_state *r6s, int disks) | |
2636 | { | |
a9b39a74 | 2637 | int rcw = 0, pd_idx = sh->pd_idx, i; |
34e04e87 | 2638 | int qd_idx = sh->qd_idx; |
a9b39a74 YT |
2639 | |
2640 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2641 | for (i = disks; i--; ) { |
2642 | struct r5dev *dev = &sh->dev[i]; | |
a9b39a74 YT |
2643 | /* check if we haven't enough data */ |
2644 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2645 | i != pd_idx && i != qd_idx && | |
2646 | !test_bit(R5_LOCKED, &dev->flags) && | |
2647 | !(test_bit(R5_UPTODATE, &dev->flags) || | |
2648 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2649 | rcw++; | |
2650 | if (!test_bit(R5_Insync, &dev->flags)) | |
2651 | continue; /* it's a failed drive */ | |
2652 | ||
2653 | if ( | |
2654 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2655 | pr_debug("Read_old stripe %llu " | |
2656 | "block %d for Reconstruct\n", | |
2657 | (unsigned long long)sh->sector, i); | |
2658 | set_bit(R5_LOCKED, &dev->flags); | |
2659 | set_bit(R5_Wantread, &dev->flags); | |
2660 | s->locked++; | |
2661 | } else { | |
2662 | pr_debug("Request delayed stripe %llu " | |
2663 | "block %d for Reconstruct\n", | |
2664 | (unsigned long long)sh->sector, i); | |
2665 | set_bit(STRIPE_DELAYED, &sh->state); | |
2666 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2667 | } |
2668 | } | |
2669 | } | |
a4456856 DW |
2670 | /* now if nothing is locked, and if we have enough data, we can start a |
2671 | * write request | |
2672 | */ | |
a9b39a74 YT |
2673 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2674 | s->locked == 0 && rcw == 0 && | |
a4456856 | 2675 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { |
a9b39a74 | 2676 | schedule_reconstruction(sh, s, 1, 0); |
a4456856 DW |
2677 | } |
2678 | } | |
2679 | ||
2680 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2681 | struct stripe_head_state *s, int disks) | |
2682 | { | |
ecc65c9b | 2683 | struct r5dev *dev = NULL; |
bd2ab670 | 2684 | |
a4456856 | 2685 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2686 | |
ecc65c9b DW |
2687 | switch (sh->check_state) { |
2688 | case check_state_idle: | |
2689 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2690 | if (s->failed == 0) { |
bd2ab670 | 2691 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2692 | sh->check_state = check_state_run; |
2693 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2694 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2695 | s->uptodate--; |
ecc65c9b | 2696 | break; |
bd2ab670 | 2697 | } |
ecc65c9b DW |
2698 | dev = &sh->dev[s->failed_num]; |
2699 | /* fall through */ | |
2700 | case check_state_compute_result: | |
2701 | sh->check_state = check_state_idle; | |
2702 | if (!dev) | |
2703 | dev = &sh->dev[sh->pd_idx]; | |
2704 | ||
2705 | /* check that a write has not made the stripe insync */ | |
2706 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2707 | break; | |
c8894419 | 2708 | |
a4456856 | 2709 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2710 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2711 | BUG_ON(s->uptodate != disks); | |
2712 | ||
2713 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2714 | s->locked++; |
a4456856 | 2715 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2716 | |
a4456856 | 2717 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2718 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2719 | break; |
2720 | case check_state_run: | |
2721 | break; /* we will be called again upon completion */ | |
2722 | case check_state_check_result: | |
2723 | sh->check_state = check_state_idle; | |
2724 | ||
2725 | /* if a failure occurred during the check operation, leave | |
2726 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2727 | */ | |
2728 | if (s->failed) | |
2729 | break; | |
2730 | ||
2731 | /* handle a successful check operation, if parity is correct | |
2732 | * we are done. Otherwise update the mismatch count and repair | |
2733 | * parity if !MD_RECOVERY_CHECK | |
2734 | */ | |
ad283ea4 | 2735 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2736 | /* parity is correct (on disc, |
2737 | * not in buffer any more) | |
2738 | */ | |
2739 | set_bit(STRIPE_INSYNC, &sh->state); | |
2740 | else { | |
2741 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2742 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2743 | /* don't try to repair!! */ | |
2744 | set_bit(STRIPE_INSYNC, &sh->state); | |
2745 | else { | |
2746 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2747 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2748 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2749 | set_bit(R5_Wantcompute, | |
2750 | &sh->dev[sh->pd_idx].flags); | |
2751 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2752 | sh->ops.target2 = -1; |
ecc65c9b DW |
2753 | s->uptodate++; |
2754 | } | |
2755 | } | |
2756 | break; | |
2757 | case check_state_compute_run: | |
2758 | break; | |
2759 | default: | |
2760 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2761 | __func__, sh->check_state, | |
2762 | (unsigned long long) sh->sector); | |
2763 | BUG(); | |
a4456856 DW |
2764 | } |
2765 | } | |
2766 | ||
2767 | ||
2768 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
36d1c647 DW |
2769 | struct stripe_head_state *s, |
2770 | struct r6_state *r6s, int disks) | |
a4456856 | 2771 | { |
a4456856 | 2772 | int pd_idx = sh->pd_idx; |
34e04e87 | 2773 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2774 | struct r5dev *dev; |
a4456856 DW |
2775 | |
2776 | set_bit(STRIPE_HANDLE, &sh->state); | |
2777 | ||
2778 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2779 | |
a4456856 DW |
2780 | /* Want to check and possibly repair P and Q. |
2781 | * However there could be one 'failed' device, in which | |
2782 | * case we can only check one of them, possibly using the | |
2783 | * other to generate missing data | |
2784 | */ | |
2785 | ||
d82dfee0 DW |
2786 | switch (sh->check_state) { |
2787 | case check_state_idle: | |
2788 | /* start a new check operation if there are < 2 failures */ | |
a4456856 | 2789 | if (s->failed == r6s->q_failed) { |
d82dfee0 | 2790 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2791 | * makes sense to check P (If anything else were failed, |
2792 | * we would have used P to recreate it). | |
2793 | */ | |
d82dfee0 | 2794 | sh->check_state = check_state_run; |
a4456856 DW |
2795 | } |
2796 | if (!r6s->q_failed && s->failed < 2) { | |
d82dfee0 | 2797 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2798 | * anything, so it makes sense to check it |
2799 | */ | |
d82dfee0 DW |
2800 | if (sh->check_state == check_state_run) |
2801 | sh->check_state = check_state_run_pq; | |
2802 | else | |
2803 | sh->check_state = check_state_run_q; | |
a4456856 | 2804 | } |
a4456856 | 2805 | |
d82dfee0 DW |
2806 | /* discard potentially stale zero_sum_result */ |
2807 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2808 | |
d82dfee0 DW |
2809 | if (sh->check_state == check_state_run) { |
2810 | /* async_xor_zero_sum destroys the contents of P */ | |
2811 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2812 | s->uptodate--; | |
a4456856 | 2813 | } |
d82dfee0 DW |
2814 | if (sh->check_state >= check_state_run && |
2815 | sh->check_state <= check_state_run_pq) { | |
2816 | /* async_syndrome_zero_sum preserves P and Q, so | |
2817 | * no need to mark them !uptodate here | |
2818 | */ | |
2819 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2820 | break; | |
a4456856 DW |
2821 | } |
2822 | ||
d82dfee0 DW |
2823 | /* we have 2-disk failure */ |
2824 | BUG_ON(s->failed != 2); | |
2825 | /* fall through */ | |
2826 | case check_state_compute_result: | |
2827 | sh->check_state = check_state_idle; | |
a4456856 | 2828 | |
d82dfee0 DW |
2829 | /* check that a write has not made the stripe insync */ |
2830 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2831 | break; | |
a4456856 DW |
2832 | |
2833 | /* now write out any block on a failed drive, | |
d82dfee0 | 2834 | * or P or Q if they were recomputed |
a4456856 | 2835 | */ |
d82dfee0 | 2836 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 DW |
2837 | if (s->failed == 2) { |
2838 | dev = &sh->dev[r6s->failed_num[1]]; | |
2839 | s->locked++; | |
2840 | set_bit(R5_LOCKED, &dev->flags); | |
2841 | set_bit(R5_Wantwrite, &dev->flags); | |
2842 | } | |
2843 | if (s->failed >= 1) { | |
2844 | dev = &sh->dev[r6s->failed_num[0]]; | |
2845 | s->locked++; | |
2846 | set_bit(R5_LOCKED, &dev->flags); | |
2847 | set_bit(R5_Wantwrite, &dev->flags); | |
2848 | } | |
d82dfee0 | 2849 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2850 | dev = &sh->dev[pd_idx]; |
2851 | s->locked++; | |
2852 | set_bit(R5_LOCKED, &dev->flags); | |
2853 | set_bit(R5_Wantwrite, &dev->flags); | |
2854 | } | |
d82dfee0 | 2855 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2856 | dev = &sh->dev[qd_idx]; |
2857 | s->locked++; | |
2858 | set_bit(R5_LOCKED, &dev->flags); | |
2859 | set_bit(R5_Wantwrite, &dev->flags); | |
2860 | } | |
2861 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2862 | ||
2863 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2864 | break; |
2865 | case check_state_run: | |
2866 | case check_state_run_q: | |
2867 | case check_state_run_pq: | |
2868 | break; /* we will be called again upon completion */ | |
2869 | case check_state_check_result: | |
2870 | sh->check_state = check_state_idle; | |
2871 | ||
2872 | /* handle a successful check operation, if parity is correct | |
2873 | * we are done. Otherwise update the mismatch count and repair | |
2874 | * parity if !MD_RECOVERY_CHECK | |
2875 | */ | |
2876 | if (sh->ops.zero_sum_result == 0) { | |
2877 | /* both parities are correct */ | |
2878 | if (!s->failed) | |
2879 | set_bit(STRIPE_INSYNC, &sh->state); | |
2880 | else { | |
2881 | /* in contrast to the raid5 case we can validate | |
2882 | * parity, but still have a failure to write | |
2883 | * back | |
2884 | */ | |
2885 | sh->check_state = check_state_compute_result; | |
2886 | /* Returning at this point means that we may go | |
2887 | * off and bring p and/or q uptodate again so | |
2888 | * we make sure to check zero_sum_result again | |
2889 | * to verify if p or q need writeback | |
2890 | */ | |
2891 | } | |
2892 | } else { | |
2893 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2894 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2895 | /* don't try to repair!! */ | |
2896 | set_bit(STRIPE_INSYNC, &sh->state); | |
2897 | else { | |
2898 | int *target = &sh->ops.target; | |
2899 | ||
2900 | sh->ops.target = -1; | |
2901 | sh->ops.target2 = -1; | |
2902 | sh->check_state = check_state_compute_run; | |
2903 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2904 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2905 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2906 | set_bit(R5_Wantcompute, | |
2907 | &sh->dev[pd_idx].flags); | |
2908 | *target = pd_idx; | |
2909 | target = &sh->ops.target2; | |
2910 | s->uptodate++; | |
2911 | } | |
2912 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2913 | set_bit(R5_Wantcompute, | |
2914 | &sh->dev[qd_idx].flags); | |
2915 | *target = qd_idx; | |
2916 | s->uptodate++; | |
2917 | } | |
2918 | } | |
2919 | } | |
2920 | break; | |
2921 | case check_state_compute_run: | |
2922 | break; | |
2923 | default: | |
2924 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2925 | __func__, sh->check_state, | |
2926 | (unsigned long long) sh->sector); | |
2927 | BUG(); | |
a4456856 DW |
2928 | } |
2929 | } | |
2930 | ||
2931 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2932 | struct r6_state *r6s) | |
2933 | { | |
2934 | int i; | |
2935 | ||
2936 | /* We have read all the blocks in this stripe and now we need to | |
2937 | * copy some of them into a target stripe for expand. | |
2938 | */ | |
f0a50d37 | 2939 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2940 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2941 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 2942 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 2943 | int dd_idx, j; |
a4456856 | 2944 | struct stripe_head *sh2; |
a08abd8c | 2945 | struct async_submit_ctl submit; |
a4456856 | 2946 | |
784052ec | 2947 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
2948 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2949 | &dd_idx, NULL); | |
a8c906ca | 2950 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
2951 | if (sh2 == NULL) |
2952 | /* so far only the early blocks of this stripe | |
2953 | * have been requested. When later blocks | |
2954 | * get requested, we will try again | |
2955 | */ | |
2956 | continue; | |
2957 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2958 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2959 | /* must have already done this block */ | |
2960 | release_stripe(sh2); | |
2961 | continue; | |
2962 | } | |
f0a50d37 DW |
2963 | |
2964 | /* place all the copies on one channel */ | |
a08abd8c | 2965 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 2966 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 2967 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 2968 | &submit); |
f0a50d37 | 2969 | |
a4456856 DW |
2970 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2971 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2972 | for (j = 0; j < conf->raid_disks; j++) | |
2973 | if (j != sh2->pd_idx && | |
d0dabf7e | 2974 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
2975 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2976 | break; | |
2977 | if (j == conf->raid_disks) { | |
2978 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2979 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2980 | } | |
2981 | release_stripe(sh2); | |
f0a50d37 | 2982 | |
a4456856 | 2983 | } |
a2e08551 N |
2984 | /* done submitting copies, wait for them to complete */ |
2985 | if (tx) { | |
2986 | async_tx_ack(tx); | |
2987 | dma_wait_for_async_tx(tx); | |
2988 | } | |
a4456856 | 2989 | } |
1da177e4 | 2990 | |
6bfe0b49 | 2991 | |
1da177e4 LT |
2992 | /* |
2993 | * handle_stripe - do things to a stripe. | |
2994 | * | |
2995 | * We lock the stripe and then examine the state of various bits | |
2996 | * to see what needs to be done. | |
2997 | * Possible results: | |
2998 | * return some read request which now have data | |
2999 | * return some write requests which are safely on disc | |
3000 | * schedule a read on some buffers | |
3001 | * schedule a write of some buffers | |
3002 | * return confirmation of parity correctness | |
3003 | * | |
1da177e4 LT |
3004 | * buffers are taken off read_list or write_list, and bh_cache buffers |
3005 | * get BH_Lock set before the stripe lock is released. | |
3006 | * | |
3007 | */ | |
a4456856 | 3008 | |
1442577b | 3009 | static void handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
3010 | { |
3011 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
3012 | int disks = sh->disks, i; |
3013 | struct bio *return_bi = NULL; | |
3014 | struct stripe_head_state s; | |
1da177e4 | 3015 | struct r5dev *dev; |
6bfe0b49 | 3016 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 3017 | int prexor; |
729a1866 | 3018 | int dec_preread_active = 0; |
1da177e4 | 3019 | |
a4456856 | 3020 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
3021 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
3022 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
3023 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
3024 | sh->reconstruct_state); | |
1da177e4 LT |
3025 | |
3026 | spin_lock(&sh->lock); | |
83206d66 N |
3027 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { |
3028 | set_bit(STRIPE_SYNCING, &sh->state); | |
3029 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3030 | } | |
1da177e4 LT |
3031 | clear_bit(STRIPE_HANDLE, &sh->state); |
3032 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3033 | ||
a4456856 DW |
3034 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3035 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3036 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 3037 | |
83de75cc | 3038 | /* Now to look around and see what can be done */ |
9910f16a | 3039 | rcu_read_lock(); |
1da177e4 LT |
3040 | for (i=disks; i--; ) { |
3041 | mdk_rdev_t *rdev; | |
a9f326eb N |
3042 | |
3043 | dev = &sh->dev[i]; | |
1da177e4 | 3044 | |
b5e98d65 DW |
3045 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
3046 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
3047 | dev->towrite, dev->written); | |
3048 | ||
3049 | /* maybe we can request a biofill operation | |
3050 | * | |
3051 | * new wantfill requests are only permitted while | |
83de75cc | 3052 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
3053 | */ |
3054 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 3055 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 3056 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
3057 | |
3058 | /* now count some things */ | |
a4456856 DW |
3059 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3060 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 3061 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 3062 | |
b5e98d65 DW |
3063 | if (test_bit(R5_Wantfill, &dev->flags)) |
3064 | s.to_fill++; | |
3065 | else if (dev->toread) | |
a4456856 | 3066 | s.to_read++; |
1da177e4 | 3067 | if (dev->towrite) { |
a4456856 | 3068 | s.to_write++; |
1da177e4 | 3069 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3070 | s.non_overwrite++; |
1da177e4 | 3071 | } |
a4456856 DW |
3072 | if (dev->written) |
3073 | s.written++; | |
9910f16a | 3074 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3075 | if (blocked_rdev == NULL && |
3076 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3077 | blocked_rdev = rdev; |
3078 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3079 | } |
415e72d0 N |
3080 | clear_bit(R5_Insync, &dev->flags); |
3081 | if (!rdev) | |
3082 | /* Not in-sync */; | |
3083 | else if (test_bit(In_sync, &rdev->flags)) | |
3084 | set_bit(R5_Insync, &dev->flags); | |
3085 | else { | |
3086 | /* could be in-sync depending on recovery/reshape status */ | |
3087 | if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) | |
3088 | set_bit(R5_Insync, &dev->flags); | |
3089 | } | |
3090 | if (!test_bit(R5_Insync, &dev->flags)) { | |
14f8d26b | 3091 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
3092 | clear_bit(R5_ReadError, &dev->flags); |
3093 | clear_bit(R5_ReWrite, &dev->flags); | |
3094 | } | |
415e72d0 N |
3095 | if (test_bit(R5_ReadError, &dev->flags)) |
3096 | clear_bit(R5_Insync, &dev->flags); | |
3097 | if (!test_bit(R5_Insync, &dev->flags)) { | |
a4456856 DW |
3098 | s.failed++; |
3099 | s.failed_num = i; | |
415e72d0 | 3100 | } |
1da177e4 | 3101 | } |
9910f16a | 3102 | rcu_read_unlock(); |
b5e98d65 | 3103 | |
6bfe0b49 | 3104 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
3105 | if (s.syncing || s.expanding || s.expanded || |
3106 | s.to_write || s.written) { | |
3107 | set_bit(STRIPE_HANDLE, &sh->state); | |
3108 | goto unlock; | |
3109 | } | |
3110 | /* There is nothing for the blocked_rdev to block */ | |
3111 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3112 | blocked_rdev = NULL; | |
6bfe0b49 DW |
3113 | } |
3114 | ||
83de75cc DW |
3115 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3116 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3117 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3118 | } | |
b5e98d65 | 3119 | |
45b4233c | 3120 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 3121 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
3122 | s.locked, s.uptodate, s.to_read, s.to_write, |
3123 | s.failed, s.failed_num); | |
1da177e4 LT |
3124 | /* check if the array has lost two devices and, if so, some requests might |
3125 | * need to be failed | |
3126 | */ | |
a4456856 | 3127 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3128 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3129 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
3130 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3131 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3132 | s.syncing = 0; |
1da177e4 LT |
3133 | } |
3134 | ||
3135 | /* might be able to return some write requests if the parity block | |
3136 | * is safe, or on a failed drive | |
3137 | */ | |
3138 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
3139 | if ( s.written && |
3140 | ((test_bit(R5_Insync, &dev->flags) && | |
3141 | !test_bit(R5_LOCKED, &dev->flags) && | |
3142 | test_bit(R5_UPTODATE, &dev->flags)) || | |
3143 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 3144 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
3145 | |
3146 | /* Now we might consider reading some blocks, either to check/generate | |
3147 | * parity, or to satisfy requests | |
3148 | * or to load a block that is being partially written. | |
3149 | */ | |
a4456856 | 3150 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 3151 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3152 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 3153 | |
e33129d8 DW |
3154 | /* Now we check to see if any write operations have recently |
3155 | * completed | |
3156 | */ | |
e0a115e5 | 3157 | prexor = 0; |
d8ee0728 | 3158 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 3159 | prexor = 1; |
d8ee0728 DW |
3160 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
3161 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 3162 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
3163 | |
3164 | /* All the 'written' buffers and the parity block are ready to | |
3165 | * be written back to disk | |
3166 | */ | |
3167 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3168 | for (i = disks; i--; ) { | |
3169 | dev = &sh->dev[i]; | |
3170 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3171 | (i == sh->pd_idx || dev->written)) { | |
3172 | pr_debug("Writing block %d\n", i); | |
3173 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
3174 | if (prexor) |
3175 | continue; | |
e33129d8 DW |
3176 | if (!test_bit(R5_Insync, &dev->flags) || |
3177 | (i == sh->pd_idx && s.failed == 0)) | |
3178 | set_bit(STRIPE_INSYNC, &sh->state); | |
3179 | } | |
3180 | } | |
729a1866 N |
3181 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3182 | dec_preread_active = 1; | |
e33129d8 DW |
3183 | } |
3184 | ||
3185 | /* Now to consider new write requests and what else, if anything | |
3186 | * should be read. We do not handle new writes when: | |
3187 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3188 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3189 | * block. | |
3190 | */ | |
600aa109 | 3191 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 3192 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
3193 | |
3194 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
3195 | * Any reads will already have been scheduled, so we just see if enough |
3196 | * data is available. The parity check is held off while parity | |
3197 | * dependent operations are in flight. | |
1da177e4 | 3198 | */ |
ecc65c9b DW |
3199 | if (sh->check_state || |
3200 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 3201 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 3202 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 3203 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 3204 | |
a4456856 | 3205 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
3206 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3207 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3208 | } | |
4e5314b5 N |
3209 | |
3210 | /* If the failed drive is just a ReadError, then we might need to progress | |
3211 | * the repair/check process | |
3212 | */ | |
a4456856 DW |
3213 | if (s.failed == 1 && !conf->mddev->ro && |
3214 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
3215 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
3216 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 3217 | ) { |
a4456856 | 3218 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
3219 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
3220 | set_bit(R5_Wantwrite, &dev->flags); | |
3221 | set_bit(R5_ReWrite, &dev->flags); | |
3222 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3223 | s.locked++; |
4e5314b5 N |
3224 | } else { |
3225 | /* let's read it back */ | |
3226 | set_bit(R5_Wantread, &dev->flags); | |
3227 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 3228 | s.locked++; |
4e5314b5 N |
3229 | } |
3230 | } | |
3231 | ||
600aa109 DW |
3232 | /* Finish reconstruct operations initiated by the expansion process */ |
3233 | if (sh->reconstruct_state == reconstruct_state_result) { | |
ab69ae12 | 3234 | struct stripe_head *sh2 |
a8c906ca | 3235 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3236 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3237 | /* sh cannot be written until sh2 has been read. | |
3238 | * so arrange for sh to be delayed a little | |
3239 | */ | |
3240 | set_bit(STRIPE_DELAYED, &sh->state); | |
3241 | set_bit(STRIPE_HANDLE, &sh->state); | |
3242 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3243 | &sh2->state)) | |
3244 | atomic_inc(&conf->preread_active_stripes); | |
3245 | release_stripe(sh2); | |
3246 | goto unlock; | |
3247 | } | |
3248 | if (sh2) | |
3249 | release_stripe(sh2); | |
3250 | ||
600aa109 | 3251 | sh->reconstruct_state = reconstruct_state_idle; |
f0a50d37 | 3252 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 3253 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 3254 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 3255 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 3256 | s.locked++; |
23397883 | 3257 | } |
f0a50d37 DW |
3258 | } |
3259 | ||
3260 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 3261 | !sh->reconstruct_state) { |
f0a50d37 DW |
3262 | /* Need to write out all blocks after computing parity */ |
3263 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3264 | stripe_set_idx(sh->sector, conf, 0, sh); |
c0f7bddb | 3265 | schedule_reconstruction(sh, &s, 1, 1); |
600aa109 | 3266 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 3267 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 3268 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
3269 | wake_up(&conf->wait_for_overlap); |
3270 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3271 | } | |
3272 | ||
0f94e87c | 3273 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3274 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3275 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 3276 | |
6bfe0b49 | 3277 | unlock: |
1da177e4 LT |
3278 | spin_unlock(&sh->lock); |
3279 | ||
6bfe0b49 DW |
3280 | /* wait for this device to become unblocked */ |
3281 | if (unlikely(blocked_rdev)) | |
3282 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3283 | ||
600aa109 | 3284 | if (s.ops_request) |
ac6b53b6 | 3285 | raid_run_ops(sh, s.ops_request); |
d84e0f10 | 3286 | |
c4e5ac0a | 3287 | ops_run_io(sh, &s); |
1da177e4 | 3288 | |
729a1866 N |
3289 | if (dec_preread_active) { |
3290 | /* We delay this until after ops_run_io so that if make_request | |
e9c7469b | 3291 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3292 | * have actually been submitted. |
3293 | */ | |
3294 | atomic_dec(&conf->preread_active_stripes); | |
3295 | if (atomic_read(&conf->preread_active_stripes) < | |
3296 | IO_THRESHOLD) | |
3297 | md_wakeup_thread(conf->mddev->thread); | |
3298 | } | |
a4456856 | 3299 | return_io(return_bi); |
1da177e4 LT |
3300 | } |
3301 | ||
1442577b | 3302 | static void handle_stripe6(struct stripe_head *sh) |
1da177e4 | 3303 | { |
bff61975 | 3304 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 3305 | int disks = sh->disks; |
a4456856 | 3306 | struct bio *return_bi = NULL; |
34e04e87 | 3307 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx; |
a4456856 DW |
3308 | struct stripe_head_state s; |
3309 | struct r6_state r6s; | |
16a53ecc | 3310 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 3311 | mdk_rdev_t *blocked_rdev = NULL; |
729a1866 | 3312 | int dec_preread_active = 0; |
1da177e4 | 3313 | |
45b4233c | 3314 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
6c0069c0 | 3315 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", |
a4456856 | 3316 | (unsigned long long)sh->sector, sh->state, |
6c0069c0 YT |
3317 | atomic_read(&sh->count), pd_idx, qd_idx, |
3318 | sh->check_state, sh->reconstruct_state); | |
a4456856 | 3319 | memset(&s, 0, sizeof(s)); |
72626685 | 3320 | |
16a53ecc | 3321 | spin_lock(&sh->lock); |
83206d66 N |
3322 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { |
3323 | set_bit(STRIPE_SYNCING, &sh->state); | |
3324 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3325 | } | |
16a53ecc N |
3326 | clear_bit(STRIPE_HANDLE, &sh->state); |
3327 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3328 | ||
a4456856 DW |
3329 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3330 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3331 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 3332 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3333 | |
3334 | rcu_read_lock(); | |
16a53ecc N |
3335 | for (i=disks; i--; ) { |
3336 | mdk_rdev_t *rdev; | |
3337 | dev = &sh->dev[i]; | |
1da177e4 | 3338 | |
45b4233c | 3339 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc | 3340 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
6c0069c0 YT |
3341 | /* maybe we can reply to a read |
3342 | * | |
3343 | * new wantfill requests are only permitted while | |
3344 | * ops_complete_biofill is guaranteed to be inactive | |
3345 | */ | |
3346 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3347 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3348 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3349 | |
16a53ecc | 3350 | /* now count some things */ |
a4456856 DW |
3351 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3352 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
2d6e4ecc DW |
3353 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
3354 | s.compute++; | |
3355 | BUG_ON(s.compute > 2); | |
3356 | } | |
1da177e4 | 3357 | |
6c0069c0 YT |
3358 | if (test_bit(R5_Wantfill, &dev->flags)) { |
3359 | s.to_fill++; | |
3360 | } else if (dev->toread) | |
a4456856 | 3361 | s.to_read++; |
16a53ecc | 3362 | if (dev->towrite) { |
a4456856 | 3363 | s.to_write++; |
16a53ecc | 3364 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3365 | s.non_overwrite++; |
16a53ecc | 3366 | } |
a4456856 DW |
3367 | if (dev->written) |
3368 | s.written++; | |
16a53ecc | 3369 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3370 | if (blocked_rdev == NULL && |
3371 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3372 | blocked_rdev = rdev; |
3373 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3374 | } |
415e72d0 N |
3375 | clear_bit(R5_Insync, &dev->flags); |
3376 | if (!rdev) | |
3377 | /* Not in-sync */; | |
3378 | else if (test_bit(In_sync, &rdev->flags)) | |
3379 | set_bit(R5_Insync, &dev->flags); | |
3380 | else { | |
3381 | /* in sync if before recovery_offset */ | |
3382 | if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) | |
3383 | set_bit(R5_Insync, &dev->flags); | |
3384 | } | |
3385 | if (!test_bit(R5_Insync, &dev->flags)) { | |
16a53ecc N |
3386 | /* The ReadError flag will just be confusing now */ |
3387 | clear_bit(R5_ReadError, &dev->flags); | |
3388 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3389 | } |
415e72d0 N |
3390 | if (test_bit(R5_ReadError, &dev->flags)) |
3391 | clear_bit(R5_Insync, &dev->flags); | |
3392 | if (!test_bit(R5_Insync, &dev->flags)) { | |
a4456856 DW |
3393 | if (s.failed < 2) |
3394 | r6s.failed_num[s.failed] = i; | |
3395 | s.failed++; | |
415e72d0 | 3396 | } |
1da177e4 LT |
3397 | } |
3398 | rcu_read_unlock(); | |
6bfe0b49 DW |
3399 | |
3400 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3401 | if (s.syncing || s.expanding || s.expanded || |
3402 | s.to_write || s.written) { | |
3403 | set_bit(STRIPE_HANDLE, &sh->state); | |
3404 | goto unlock; | |
3405 | } | |
3406 | /* There is nothing for the blocked_rdev to block */ | |
3407 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3408 | blocked_rdev = NULL; | |
6bfe0b49 | 3409 | } |
ac4090d2 | 3410 | |
6c0069c0 YT |
3411 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3412 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3413 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3414 | } | |
3415 | ||
45b4233c | 3416 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3417 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3418 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3419 | r6s.failed_num[0], r6s.failed_num[1]); | |
3420 | /* check if the array has lost >2 devices and, if so, some requests | |
3421 | * might need to be failed | |
16a53ecc | 3422 | */ |
a4456856 | 3423 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3424 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3425 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3426 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3427 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3428 | s.syncing = 0; |
16a53ecc N |
3429 | } |
3430 | ||
3431 | /* | |
3432 | * might be able to return some write requests if the parity blocks | |
3433 | * are safe, or on a failed drive | |
3434 | */ | |
3435 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3436 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3437 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
34e04e87 N |
3438 | qdev = &sh->dev[qd_idx]; |
3439 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx) | |
3440 | || (s.failed >= 2 && r6s.failed_num[1] == qd_idx); | |
a4456856 DW |
3441 | |
3442 | if ( s.written && | |
3443 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3444 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3445 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3446 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3447 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3448 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3449 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3450 | |
3451 | /* Now we might consider reading some blocks, either to check/generate | |
3452 | * parity, or to satisfy requests | |
3453 | * or to load a block that is being partially written. | |
3454 | */ | |
a4456856 | 3455 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
6c0069c0 | 3456 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 3457 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc | 3458 | |
6c0069c0 YT |
3459 | /* Now we check to see if any write operations have recently |
3460 | * completed | |
3461 | */ | |
3462 | if (sh->reconstruct_state == reconstruct_state_drain_result) { | |
6c0069c0 YT |
3463 | |
3464 | sh->reconstruct_state = reconstruct_state_idle; | |
3465 | /* All the 'written' buffers and the parity blocks are ready to | |
3466 | * be written back to disk | |
3467 | */ | |
3468 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3469 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags)); | |
3470 | for (i = disks; i--; ) { | |
3471 | dev = &sh->dev[i]; | |
3472 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3473 | (i == sh->pd_idx || i == qd_idx || | |
3474 | dev->written)) { | |
3475 | pr_debug("Writing block %d\n", i); | |
3476 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); | |
3477 | set_bit(R5_Wantwrite, &dev->flags); | |
3478 | if (!test_bit(R5_Insync, &dev->flags) || | |
3479 | ((i == sh->pd_idx || i == qd_idx) && | |
3480 | s.failed == 0)) | |
3481 | set_bit(STRIPE_INSYNC, &sh->state); | |
3482 | } | |
3483 | } | |
729a1866 N |
3484 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3485 | dec_preread_active = 1; | |
6c0069c0 YT |
3486 | } |
3487 | ||
a9b39a74 YT |
3488 | /* Now to consider new write requests and what else, if anything |
3489 | * should be read. We do not handle new writes when: | |
3490 | * 1/ A 'write' operation (copy+gen_syndrome) is already in flight. | |
3491 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3492 | * block. | |
3493 | */ | |
3494 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
1fe797e6 | 3495 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3496 | |
3497 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 | 3498 | * Any reads will already have been scheduled, so we just see if enough |
6c0069c0 YT |
3499 | * data is available. The parity check is held off while parity |
3500 | * dependent operations are in flight. | |
16a53ecc | 3501 | */ |
6c0069c0 YT |
3502 | if (sh->check_state || |
3503 | (s.syncing && s.locked == 0 && | |
3504 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3505 | !test_bit(STRIPE_INSYNC, &sh->state))) | |
36d1c647 | 3506 | handle_parity_checks6(conf, sh, &s, &r6s, disks); |
16a53ecc | 3507 | |
a4456856 | 3508 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3509 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3510 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3511 | } | |
3512 | ||
3513 | /* If the failed drives are just a ReadError, then we might need | |
3514 | * to progress the repair/check process | |
3515 | */ | |
a4456856 DW |
3516 | if (s.failed <= 2 && !conf->mddev->ro) |
3517 | for (i = 0; i < s.failed; i++) { | |
3518 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3519 | if (test_bit(R5_ReadError, &dev->flags) |
3520 | && !test_bit(R5_LOCKED, &dev->flags) | |
3521 | && test_bit(R5_UPTODATE, &dev->flags) | |
3522 | ) { | |
3523 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3524 | set_bit(R5_Wantwrite, &dev->flags); | |
3525 | set_bit(R5_ReWrite, &dev->flags); | |
3526 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3527 | s.locked++; |
16a53ecc N |
3528 | } else { |
3529 | /* let's read it back */ | |
3530 | set_bit(R5_Wantread, &dev->flags); | |
3531 | set_bit(R5_LOCKED, &dev->flags); | |
6c0069c0 | 3532 | s.locked++; |
16a53ecc N |
3533 | } |
3534 | } | |
3535 | } | |
f416885e | 3536 | |
6c0069c0 YT |
3537 | /* Finish reconstruct operations initiated by the expansion process */ |
3538 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3539 | sh->reconstruct_state = reconstruct_state_idle; | |
3540 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3541 | for (i = conf->raid_disks; i--; ) { | |
3542 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3543 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3544 | s.locked++; | |
3545 | } | |
3546 | } | |
3547 | ||
3548 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
3549 | !sh->reconstruct_state) { | |
ab69ae12 | 3550 | struct stripe_head *sh2 |
a8c906ca | 3551 | = get_active_stripe(conf, sh->sector, 1, 1, 1); |
ab69ae12 N |
3552 | if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) { |
3553 | /* sh cannot be written until sh2 has been read. | |
3554 | * so arrange for sh to be delayed a little | |
3555 | */ | |
3556 | set_bit(STRIPE_DELAYED, &sh->state); | |
3557 | set_bit(STRIPE_HANDLE, &sh->state); | |
3558 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3559 | &sh2->state)) | |
3560 | atomic_inc(&conf->preread_active_stripes); | |
3561 | release_stripe(sh2); | |
3562 | goto unlock; | |
3563 | } | |
3564 | if (sh2) | |
3565 | release_stripe(sh2); | |
3566 | ||
f416885e N |
3567 | /* Need to write out all blocks after computing P&Q */ |
3568 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3569 | stripe_set_idx(sh->sector, conf, 0, sh); |
6c0069c0 YT |
3570 | schedule_reconstruction(sh, &s, 1, 1); |
3571 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
f416885e N |
3572 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3573 | atomic_dec(&conf->reshape_stripes); | |
3574 | wake_up(&conf->wait_for_overlap); | |
3575 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3576 | } | |
3577 | ||
0f94e87c | 3578 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3579 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3580 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3581 | |
6bfe0b49 | 3582 | unlock: |
16a53ecc N |
3583 | spin_unlock(&sh->lock); |
3584 | ||
6bfe0b49 DW |
3585 | /* wait for this device to become unblocked */ |
3586 | if (unlikely(blocked_rdev)) | |
3587 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3588 | ||
6c0069c0 YT |
3589 | if (s.ops_request) |
3590 | raid_run_ops(sh, s.ops_request); | |
3591 | ||
f0e43bcd | 3592 | ops_run_io(sh, &s); |
16a53ecc | 3593 | |
729a1866 N |
3594 | |
3595 | if (dec_preread_active) { | |
3596 | /* We delay this until after ops_run_io so that if make_request | |
e9c7469b | 3597 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3598 | * have actually been submitted. |
3599 | */ | |
3600 | atomic_dec(&conf->preread_active_stripes); | |
3601 | if (atomic_read(&conf->preread_active_stripes) < | |
3602 | IO_THRESHOLD) | |
3603 | md_wakeup_thread(conf->mddev->thread); | |
3604 | } | |
3605 | ||
f0e43bcd | 3606 | return_io(return_bi); |
16a53ecc N |
3607 | } |
3608 | ||
1442577b | 3609 | static void handle_stripe(struct stripe_head *sh) |
16a53ecc N |
3610 | { |
3611 | if (sh->raid_conf->level == 6) | |
1442577b | 3612 | handle_stripe6(sh); |
16a53ecc | 3613 | else |
1442577b | 3614 | handle_stripe5(sh); |
16a53ecc N |
3615 | } |
3616 | ||
16a53ecc N |
3617 | static void raid5_activate_delayed(raid5_conf_t *conf) |
3618 | { | |
3619 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3620 | while (!list_empty(&conf->delayed_list)) { | |
3621 | struct list_head *l = conf->delayed_list.next; | |
3622 | struct stripe_head *sh; | |
3623 | sh = list_entry(l, struct stripe_head, lru); | |
3624 | list_del_init(l); | |
3625 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3626 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3627 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3628 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3629 | } |
482c0834 | 3630 | } |
16a53ecc N |
3631 | } |
3632 | ||
3633 | static void activate_bit_delay(raid5_conf_t *conf) | |
3634 | { | |
3635 | /* device_lock is held */ | |
3636 | struct list_head head; | |
3637 | list_add(&head, &conf->bitmap_list); | |
3638 | list_del_init(&conf->bitmap_list); | |
3639 | while (!list_empty(&head)) { | |
3640 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3641 | list_del_init(&sh->lru); | |
3642 | atomic_inc(&sh->count); | |
3643 | __release_stripe(conf, sh); | |
3644 | } | |
3645 | } | |
3646 | ||
11d8a6e3 | 3647 | int md_raid5_congested(mddev_t *mddev, int bits) |
f022b2fd | 3648 | { |
070ec55d | 3649 | raid5_conf_t *conf = mddev->private; |
f022b2fd N |
3650 | |
3651 | /* No difference between reads and writes. Just check | |
3652 | * how busy the stripe_cache is | |
3653 | */ | |
3fa841d7 | 3654 | |
f022b2fd N |
3655 | if (conf->inactive_blocked) |
3656 | return 1; | |
3657 | if (conf->quiesce) | |
3658 | return 1; | |
3659 | if (list_empty_careful(&conf->inactive_list)) | |
3660 | return 1; | |
3661 | ||
3662 | return 0; | |
3663 | } | |
11d8a6e3 N |
3664 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3665 | ||
3666 | static int raid5_congested(void *data, int bits) | |
3667 | { | |
3668 | mddev_t *mddev = data; | |
3669 | ||
3670 | return mddev_congested(mddev, bits) || | |
3671 | md_raid5_congested(mddev, bits); | |
3672 | } | |
f022b2fd | 3673 | |
23032a0e RBJ |
3674 | /* We want read requests to align with chunks where possible, |
3675 | * but write requests don't need to. | |
3676 | */ | |
cc371e66 AK |
3677 | static int raid5_mergeable_bvec(struct request_queue *q, |
3678 | struct bvec_merge_data *bvm, | |
3679 | struct bio_vec *biovec) | |
23032a0e RBJ |
3680 | { |
3681 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3682 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3683 | int max; |
9d8f0363 | 3684 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3685 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3686 | |
cc371e66 | 3687 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3688 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3689 | ||
664e7c41 AN |
3690 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3691 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3692 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3693 | if (max < 0) max = 0; | |
3694 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3695 | return biovec->bv_len; | |
3696 | else | |
3697 | return max; | |
3698 | } | |
3699 | ||
f679623f RBJ |
3700 | |
3701 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3702 | { | |
3703 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3704 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3705 | unsigned int bio_sectors = bio->bi_size >> 9; |
3706 | ||
664e7c41 AN |
3707 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3708 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3709 | return chunk_sectors >= |
3710 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3711 | } | |
3712 | ||
46031f9a RBJ |
3713 | /* |
3714 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3715 | * later sampled by raid5d. | |
3716 | */ | |
3717 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3718 | { | |
3719 | unsigned long flags; | |
3720 | ||
3721 | spin_lock_irqsave(&conf->device_lock, flags); | |
3722 | ||
3723 | bi->bi_next = conf->retry_read_aligned_list; | |
3724 | conf->retry_read_aligned_list = bi; | |
3725 | ||
3726 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3727 | md_wakeup_thread(conf->mddev->thread); | |
3728 | } | |
3729 | ||
3730 | ||
3731 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3732 | { | |
3733 | struct bio *bi; | |
3734 | ||
3735 | bi = conf->retry_read_aligned; | |
3736 | if (bi) { | |
3737 | conf->retry_read_aligned = NULL; | |
3738 | return bi; | |
3739 | } | |
3740 | bi = conf->retry_read_aligned_list; | |
3741 | if(bi) { | |
387bb173 | 3742 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3743 | bi->bi_next = NULL; |
960e739d JA |
3744 | /* |
3745 | * this sets the active strip count to 1 and the processed | |
3746 | * strip count to zero (upper 8 bits) | |
3747 | */ | |
46031f9a | 3748 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3749 | } |
3750 | ||
3751 | return bi; | |
3752 | } | |
3753 | ||
3754 | ||
f679623f RBJ |
3755 | /* |
3756 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3757 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3758 | * first). | |
3759 | * If the read failed.. | |
3760 | */ | |
6712ecf8 | 3761 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3762 | { |
3763 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3764 | mddev_t *mddev; |
3765 | raid5_conf_t *conf; | |
3766 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3767 | mdk_rdev_t *rdev; | |
3768 | ||
f679623f | 3769 | bio_put(bi); |
46031f9a | 3770 | |
46031f9a RBJ |
3771 | rdev = (void*)raid_bi->bi_next; |
3772 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3773 | mddev = rdev->mddev; |
3774 | conf = mddev->private; | |
46031f9a RBJ |
3775 | |
3776 | rdev_dec_pending(rdev, conf->mddev); | |
3777 | ||
3778 | if (!error && uptodate) { | |
6712ecf8 | 3779 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3780 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3781 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3782 | return; |
46031f9a RBJ |
3783 | } |
3784 | ||
3785 | ||
45b4233c | 3786 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3787 | |
3788 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3789 | } |
3790 | ||
387bb173 NB |
3791 | static int bio_fits_rdev(struct bio *bi) |
3792 | { | |
165125e1 | 3793 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3794 | |
ae03bf63 | 3795 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3796 | return 0; |
3797 | blk_recount_segments(q, bi); | |
8a78362c | 3798 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3799 | return 0; |
3800 | ||
3801 | if (q->merge_bvec_fn) | |
3802 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3803 | * just just give up | |
3804 | */ | |
3805 | return 0; | |
3806 | ||
3807 | return 1; | |
3808 | } | |
3809 | ||
3810 | ||
21a52c6d | 3811 | static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio) |
f679623f | 3812 | { |
070ec55d | 3813 | raid5_conf_t *conf = mddev->private; |
8553fe7e | 3814 | int dd_idx; |
f679623f RBJ |
3815 | struct bio* align_bi; |
3816 | mdk_rdev_t *rdev; | |
3817 | ||
3818 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3819 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3820 | return 0; |
3821 | } | |
3822 | /* | |
a167f663 | 3823 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3824 | */ |
a167f663 | 3825 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3826 | if (!align_bi) |
3827 | return 0; | |
3828 | /* | |
3829 | * set bi_end_io to a new function, and set bi_private to the | |
3830 | * original bio. | |
3831 | */ | |
3832 | align_bi->bi_end_io = raid5_align_endio; | |
3833 | align_bi->bi_private = raid_bio; | |
3834 | /* | |
3835 | * compute position | |
3836 | */ | |
112bf897 N |
3837 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3838 | 0, | |
911d4ee8 | 3839 | &dd_idx, NULL); |
f679623f RBJ |
3840 | |
3841 | rcu_read_lock(); | |
3842 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3843 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3844 | atomic_inc(&rdev->nr_pending); |
3845 | rcu_read_unlock(); | |
46031f9a RBJ |
3846 | raid_bio->bi_next = (void*)rdev; |
3847 | align_bi->bi_bdev = rdev->bdev; | |
3848 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3849 | align_bi->bi_sector += rdev->data_offset; | |
3850 | ||
387bb173 NB |
3851 | if (!bio_fits_rdev(align_bi)) { |
3852 | /* too big in some way */ | |
3853 | bio_put(align_bi); | |
3854 | rdev_dec_pending(rdev, mddev); | |
3855 | return 0; | |
3856 | } | |
3857 | ||
46031f9a RBJ |
3858 | spin_lock_irq(&conf->device_lock); |
3859 | wait_event_lock_irq(conf->wait_for_stripe, | |
3860 | conf->quiesce == 0, | |
3861 | conf->device_lock, /* nothing */); | |
3862 | atomic_inc(&conf->active_aligned_reads); | |
3863 | spin_unlock_irq(&conf->device_lock); | |
3864 | ||
f679623f RBJ |
3865 | generic_make_request(align_bi); |
3866 | return 1; | |
3867 | } else { | |
3868 | rcu_read_unlock(); | |
46031f9a | 3869 | bio_put(align_bi); |
f679623f RBJ |
3870 | return 0; |
3871 | } | |
3872 | } | |
3873 | ||
8b3e6cdc DW |
3874 | /* __get_priority_stripe - get the next stripe to process |
3875 | * | |
3876 | * Full stripe writes are allowed to pass preread active stripes up until | |
3877 | * the bypass_threshold is exceeded. In general the bypass_count | |
3878 | * increments when the handle_list is handled before the hold_list; however, it | |
3879 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3880 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3881 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3882 | * handle_list. | |
3883 | */ | |
3884 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3885 | { | |
3886 | struct stripe_head *sh; | |
3887 | ||
3888 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3889 | __func__, | |
3890 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3891 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3892 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3893 | ||
3894 | if (!list_empty(&conf->handle_list)) { | |
3895 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3896 | ||
3897 | if (list_empty(&conf->hold_list)) | |
3898 | conf->bypass_count = 0; | |
3899 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3900 | if (conf->hold_list.next == conf->last_hold) | |
3901 | conf->bypass_count++; | |
3902 | else { | |
3903 | conf->last_hold = conf->hold_list.next; | |
3904 | conf->bypass_count -= conf->bypass_threshold; | |
3905 | if (conf->bypass_count < 0) | |
3906 | conf->bypass_count = 0; | |
3907 | } | |
3908 | } | |
3909 | } else if (!list_empty(&conf->hold_list) && | |
3910 | ((conf->bypass_threshold && | |
3911 | conf->bypass_count > conf->bypass_threshold) || | |
3912 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3913 | sh = list_entry(conf->hold_list.next, | |
3914 | typeof(*sh), lru); | |
3915 | conf->bypass_count -= conf->bypass_threshold; | |
3916 | if (conf->bypass_count < 0) | |
3917 | conf->bypass_count = 0; | |
3918 | } else | |
3919 | return NULL; | |
3920 | ||
3921 | list_del_init(&sh->lru); | |
3922 | atomic_inc(&sh->count); | |
3923 | BUG_ON(atomic_read(&sh->count) != 1); | |
3924 | return sh; | |
3925 | } | |
f679623f | 3926 | |
21a52c6d | 3927 | static int make_request(mddev_t *mddev, struct bio * bi) |
1da177e4 | 3928 | { |
070ec55d | 3929 | raid5_conf_t *conf = mddev->private; |
911d4ee8 | 3930 | int dd_idx; |
1da177e4 LT |
3931 | sector_t new_sector; |
3932 | sector_t logical_sector, last_sector; | |
3933 | struct stripe_head *sh; | |
a362357b | 3934 | const int rw = bio_data_dir(bi); |
49077326 | 3935 | int remaining; |
7c13edc8 | 3936 | int plugged; |
1da177e4 | 3937 | |
e9c7469b TH |
3938 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
3939 | md_flush_request(mddev, bi); | |
e5dcdd80 N |
3940 | return 0; |
3941 | } | |
3942 | ||
3d310eb7 | 3943 | md_write_start(mddev, bi); |
06d91a5f | 3944 | |
802ba064 | 3945 | if (rw == READ && |
52488615 | 3946 | mddev->reshape_position == MaxSector && |
21a52c6d | 3947 | chunk_aligned_read(mddev,bi)) |
99c0fb5f | 3948 | return 0; |
52488615 | 3949 | |
1da177e4 LT |
3950 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3951 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3952 | bi->bi_next = NULL; | |
3953 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3954 | |
7c13edc8 | 3955 | plugged = mddev_check_plugged(mddev); |
1da177e4 LT |
3956 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3957 | DEFINE_WAIT(w); | |
16a53ecc | 3958 | int disks, data_disks; |
b5663ba4 | 3959 | int previous; |
b578d55f | 3960 | |
7ecaa1e6 | 3961 | retry: |
b5663ba4 | 3962 | previous = 0; |
b0f9ec04 | 3963 | disks = conf->raid_disks; |
b578d55f | 3964 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3965 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3966 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3967 | * 64bit on a 32bit platform, and so it might be |
3968 | * possible to see a half-updated value | |
aeb878b0 | 3969 | * Of course reshape_progress could change after |
df8e7f76 N |
3970 | * the lock is dropped, so once we get a reference |
3971 | * to the stripe that we think it is, we will have | |
3972 | * to check again. | |
3973 | */ | |
7ecaa1e6 | 3974 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3975 | if (mddev->delta_disks < 0 |
3976 | ? logical_sector < conf->reshape_progress | |
3977 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3978 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3979 | previous = 1; |
3980 | } else { | |
fef9c61f N |
3981 | if (mddev->delta_disks < 0 |
3982 | ? logical_sector < conf->reshape_safe | |
3983 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3984 | spin_unlock_irq(&conf->device_lock); |
3985 | schedule(); | |
3986 | goto retry; | |
3987 | } | |
3988 | } | |
7ecaa1e6 N |
3989 | spin_unlock_irq(&conf->device_lock); |
3990 | } | |
16a53ecc N |
3991 | data_disks = disks - conf->max_degraded; |
3992 | ||
112bf897 N |
3993 | new_sector = raid5_compute_sector(conf, logical_sector, |
3994 | previous, | |
911d4ee8 | 3995 | &dd_idx, NULL); |
0c55e022 | 3996 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3997 | (unsigned long long)new_sector, |
3998 | (unsigned long long)logical_sector); | |
3999 | ||
b5663ba4 | 4000 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 4001 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 4002 | if (sh) { |
b0f9ec04 | 4003 | if (unlikely(previous)) { |
7ecaa1e6 | 4004 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
4005 | * stripe, so we must do the range check again. |
4006 | * Expansion could still move past after this | |
4007 | * test, but as we are holding a reference to | |
4008 | * 'sh', we know that if that happens, | |
4009 | * STRIPE_EXPANDING will get set and the expansion | |
4010 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
4011 | */ |
4012 | int must_retry = 0; | |
4013 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
4014 | if (mddev->delta_disks < 0 |
4015 | ? logical_sector >= conf->reshape_progress | |
4016 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
4017 | /* mismatch, need to try again */ |
4018 | must_retry = 1; | |
4019 | spin_unlock_irq(&conf->device_lock); | |
4020 | if (must_retry) { | |
4021 | release_stripe(sh); | |
7a3ab908 | 4022 | schedule(); |
7ecaa1e6 N |
4023 | goto retry; |
4024 | } | |
4025 | } | |
e62e58a5 | 4026 | |
ffd96e35 | 4027 | if (rw == WRITE && |
a5c308d4 | 4028 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
4029 | logical_sector < mddev->suspend_hi) { |
4030 | release_stripe(sh); | |
e62e58a5 N |
4031 | /* As the suspend_* range is controlled by |
4032 | * userspace, we want an interruptible | |
4033 | * wait. | |
4034 | */ | |
4035 | flush_signals(current); | |
4036 | prepare_to_wait(&conf->wait_for_overlap, | |
4037 | &w, TASK_INTERRUPTIBLE); | |
4038 | if (logical_sector >= mddev->suspend_lo && | |
4039 | logical_sector < mddev->suspend_hi) | |
4040 | schedule(); | |
e464eafd N |
4041 | goto retry; |
4042 | } | |
7ecaa1e6 N |
4043 | |
4044 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 4045 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
4046 | /* Stripe is busy expanding or |
4047 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4048 | * and wait a while |
4049 | */ | |
482c0834 | 4050 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4051 | release_stripe(sh); |
4052 | schedule(); | |
4053 | goto retry; | |
4054 | } | |
4055 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
4056 | set_bit(STRIPE_HANDLE, &sh->state); |
4057 | clear_bit(STRIPE_DELAYED, &sh->state); | |
e9c7469b | 4058 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
4059 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4060 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 4061 | release_stripe(sh); |
1da177e4 LT |
4062 | } else { |
4063 | /* cannot get stripe for read-ahead, just give-up */ | |
4064 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
4065 | finish_wait(&conf->wait_for_overlap, &w); | |
4066 | break; | |
4067 | } | |
4068 | ||
4069 | } | |
7c13edc8 N |
4070 | if (!plugged) |
4071 | md_wakeup_thread(mddev->thread); | |
4072 | ||
1da177e4 | 4073 | spin_lock_irq(&conf->device_lock); |
960e739d | 4074 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
4075 | spin_unlock_irq(&conf->device_lock); |
4076 | if (remaining == 0) { | |
1da177e4 | 4077 | |
16a53ecc | 4078 | if ( rw == WRITE ) |
1da177e4 | 4079 | md_write_end(mddev); |
6712ecf8 | 4080 | |
0e13fe23 | 4081 | bio_endio(bi, 0); |
1da177e4 | 4082 | } |
729a1866 | 4083 | |
1da177e4 LT |
4084 | return 0; |
4085 | } | |
4086 | ||
b522adcd DW |
4087 | static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks); |
4088 | ||
52c03291 | 4089 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4090 | { |
52c03291 N |
4091 | /* reshaping is quite different to recovery/resync so it is |
4092 | * handled quite separately ... here. | |
4093 | * | |
4094 | * On each call to sync_request, we gather one chunk worth of | |
4095 | * destination stripes and flag them as expanding. | |
4096 | * Then we find all the source stripes and request reads. | |
4097 | * As the reads complete, handle_stripe will copy the data | |
4098 | * into the destination stripe and release that stripe. | |
4099 | */ | |
7b92813c | 4100 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 4101 | struct stripe_head *sh; |
ccfcc3c1 | 4102 | sector_t first_sector, last_sector; |
f416885e N |
4103 | int raid_disks = conf->previous_raid_disks; |
4104 | int data_disks = raid_disks - conf->max_degraded; | |
4105 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4106 | int i; |
4107 | int dd_idx; | |
c8f517c4 | 4108 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4109 | sector_t stripe_addr; |
7a661381 | 4110 | int reshape_sectors; |
ab69ae12 | 4111 | struct list_head stripes; |
52c03291 | 4112 | |
fef9c61f N |
4113 | if (sector_nr == 0) { |
4114 | /* If restarting in the middle, skip the initial sectors */ | |
4115 | if (mddev->delta_disks < 0 && | |
4116 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4117 | sector_nr = raid5_size(mddev, 0, 0) | |
4118 | - conf->reshape_progress; | |
a639755c | 4119 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
4120 | conf->reshape_progress > 0) |
4121 | sector_nr = conf->reshape_progress; | |
f416885e | 4122 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4123 | if (sector_nr) { |
8dee7211 N |
4124 | mddev->curr_resync_completed = sector_nr; |
4125 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4126 | *skipped = 1; |
4127 | return sector_nr; | |
4128 | } | |
52c03291 N |
4129 | } |
4130 | ||
7a661381 N |
4131 | /* We need to process a full chunk at a time. |
4132 | * If old and new chunk sizes differ, we need to process the | |
4133 | * largest of these | |
4134 | */ | |
664e7c41 AN |
4135 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4136 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4137 | else |
9d8f0363 | 4138 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4139 | |
52c03291 N |
4140 | /* we update the metadata when there is more than 3Meg |
4141 | * in the block range (that is rather arbitrary, should | |
4142 | * probably be time based) or when the data about to be | |
4143 | * copied would over-write the source of the data at | |
4144 | * the front of the range. | |
fef9c61f N |
4145 | * i.e. one new_stripe along from reshape_progress new_maps |
4146 | * to after where reshape_safe old_maps to | |
52c03291 | 4147 | */ |
fef9c61f | 4148 | writepos = conf->reshape_progress; |
f416885e | 4149 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4150 | readpos = conf->reshape_progress; |
4151 | sector_div(readpos, data_disks); | |
fef9c61f | 4152 | safepos = conf->reshape_safe; |
f416885e | 4153 | sector_div(safepos, data_disks); |
fef9c61f | 4154 | if (mddev->delta_disks < 0) { |
ed37d83e | 4155 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4156 | readpos += reshape_sectors; |
7a661381 | 4157 | safepos += reshape_sectors; |
fef9c61f | 4158 | } else { |
7a661381 | 4159 | writepos += reshape_sectors; |
ed37d83e N |
4160 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4161 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4162 | } |
52c03291 | 4163 | |
c8f517c4 N |
4164 | /* 'writepos' is the most advanced device address we might write. |
4165 | * 'readpos' is the least advanced device address we might read. | |
4166 | * 'safepos' is the least address recorded in the metadata as having | |
4167 | * been reshaped. | |
4168 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4169 | * ensure safety in the face of a crash - that must be done by userspace | |
4170 | * making a backup of the data. So in that case there is no particular | |
4171 | * rush to update metadata. | |
4172 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4173 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4174 | * we can be safe in the event of a crash. | |
4175 | * So we insist on updating metadata if safepos is behind writepos and | |
4176 | * readpos is beyond writepos. | |
4177 | * In any case, update the metadata every 10 seconds. | |
4178 | * Maybe that number should be configurable, but I'm not sure it is | |
4179 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4180 | */ | |
fef9c61f | 4181 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4182 | ? (safepos > writepos && readpos < writepos) |
4183 | : (safepos < writepos && readpos > writepos)) || | |
4184 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4185 | /* Cannot proceed until we've updated the superblock... */ |
4186 | wait_event(conf->wait_for_overlap, | |
4187 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4188 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4189 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4190 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4191 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4192 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4193 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4194 | kthread_should_stop()); |
4195 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4196 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4197 | spin_unlock_irq(&conf->device_lock); |
4198 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4199 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4200 | } |
4201 | ||
ec32a2bd N |
4202 | if (mddev->delta_disks < 0) { |
4203 | BUG_ON(conf->reshape_progress == 0); | |
4204 | stripe_addr = writepos; | |
4205 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4206 | ~((sector_t)reshape_sectors - 1)) |
4207 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4208 | != sector_nr); |
4209 | } else { | |
7a661381 | 4210 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4211 | stripe_addr = sector_nr; |
4212 | } | |
ab69ae12 | 4213 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4214 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4215 | int j; |
a9f326eb | 4216 | int skipped_disk = 0; |
a8c906ca | 4217 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4218 | set_bit(STRIPE_EXPANDING, &sh->state); |
4219 | atomic_inc(&conf->reshape_stripes); | |
4220 | /* If any of this stripe is beyond the end of the old | |
4221 | * array, then we need to zero those blocks | |
4222 | */ | |
4223 | for (j=sh->disks; j--;) { | |
4224 | sector_t s; | |
4225 | if (j == sh->pd_idx) | |
4226 | continue; | |
f416885e | 4227 | if (conf->level == 6 && |
d0dabf7e | 4228 | j == sh->qd_idx) |
f416885e | 4229 | continue; |
784052ec | 4230 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4231 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4232 | skipped_disk = 1; |
52c03291 N |
4233 | continue; |
4234 | } | |
4235 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4236 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4237 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4238 | } | |
a9f326eb | 4239 | if (!skipped_disk) { |
52c03291 N |
4240 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4241 | set_bit(STRIPE_HANDLE, &sh->state); | |
4242 | } | |
ab69ae12 | 4243 | list_add(&sh->lru, &stripes); |
52c03291 N |
4244 | } |
4245 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4246 | if (mddev->delta_disks < 0) |
7a661381 | 4247 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4248 | else |
7a661381 | 4249 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4250 | spin_unlock_irq(&conf->device_lock); |
4251 | /* Ok, those stripe are ready. We can start scheduling | |
4252 | * reads on the source stripes. | |
4253 | * The source stripes are determined by mapping the first and last | |
4254 | * block on the destination stripes. | |
4255 | */ | |
52c03291 | 4256 | first_sector = |
ec32a2bd | 4257 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4258 | 1, &dd_idx, NULL); |
52c03291 | 4259 | last_sector = |
0e6e0271 | 4260 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4261 | * new_data_disks - 1), |
911d4ee8 | 4262 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4263 | if (last_sector >= mddev->dev_sectors) |
4264 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4265 | while (first_sector <= last_sector) { |
a8c906ca | 4266 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4267 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4268 | set_bit(STRIPE_HANDLE, &sh->state); | |
4269 | release_stripe(sh); | |
4270 | first_sector += STRIPE_SECTORS; | |
4271 | } | |
ab69ae12 N |
4272 | /* Now that the sources are clearly marked, we can release |
4273 | * the destination stripes | |
4274 | */ | |
4275 | while (!list_empty(&stripes)) { | |
4276 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4277 | list_del_init(&sh->lru); | |
4278 | release_stripe(sh); | |
4279 | } | |
c6207277 N |
4280 | /* If this takes us to the resync_max point where we have to pause, |
4281 | * then we need to write out the superblock. | |
4282 | */ | |
7a661381 | 4283 | sector_nr += reshape_sectors; |
c03f6a19 N |
4284 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4285 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4286 | /* Cannot proceed until we've updated the superblock... */ |
4287 | wait_event(conf->wait_for_overlap, | |
4288 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4289 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4290 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4291 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4292 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4293 | md_wakeup_thread(mddev->thread); | |
4294 | wait_event(mddev->sb_wait, | |
4295 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4296 | || kthread_should_stop()); | |
4297 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4298 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4299 | spin_unlock_irq(&conf->device_lock); |
4300 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4301 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4302 | } |
7a661381 | 4303 | return reshape_sectors; |
52c03291 N |
4304 | } |
4305 | ||
4306 | /* FIXME go_faster isn't used */ | |
4307 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
4308 | { | |
7b92813c | 4309 | raid5_conf_t *conf = mddev->private; |
52c03291 | 4310 | struct stripe_head *sh; |
58c0fed4 | 4311 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 4312 | sector_t sync_blocks; |
16a53ecc N |
4313 | int still_degraded = 0; |
4314 | int i; | |
1da177e4 | 4315 | |
72626685 | 4316 | if (sector_nr >= max_sector) { |
1da177e4 | 4317 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 4318 | |
29269553 N |
4319 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4320 | end_reshape(conf); | |
4321 | return 0; | |
4322 | } | |
72626685 N |
4323 | |
4324 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4325 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4326 | &sync_blocks, 1); | |
16a53ecc | 4327 | else /* completed sync */ |
72626685 N |
4328 | conf->fullsync = 0; |
4329 | bitmap_close_sync(mddev->bitmap); | |
4330 | ||
1da177e4 LT |
4331 | return 0; |
4332 | } | |
ccfcc3c1 | 4333 | |
64bd660b N |
4334 | /* Allow raid5_quiesce to complete */ |
4335 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4336 | ||
52c03291 N |
4337 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4338 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4339 | |
c6207277 N |
4340 | /* No need to check resync_max as we never do more than one |
4341 | * stripe, and as resync_max will always be on a chunk boundary, | |
4342 | * if the check in md_do_sync didn't fire, there is no chance | |
4343 | * of overstepping resync_max here | |
4344 | */ | |
4345 | ||
16a53ecc | 4346 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4347 | * to resync, then assert that we are finished, because there is |
4348 | * nothing we can do. | |
4349 | */ | |
3285edf1 | 4350 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4351 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4352 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4353 | *skipped = 1; |
1da177e4 LT |
4354 | return rv; |
4355 | } | |
72626685 | 4356 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4357 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4358 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4359 | /* we can skip this block, and probably more */ | |
4360 | sync_blocks /= STRIPE_SECTORS; | |
4361 | *skipped = 1; | |
4362 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4363 | } | |
1da177e4 | 4364 | |
b47490c9 N |
4365 | |
4366 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
4367 | ||
a8c906ca | 4368 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4369 | if (sh == NULL) { |
a8c906ca | 4370 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4371 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4372 | * is trying to get access |
1da177e4 | 4373 | */ |
66c006a5 | 4374 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4375 | } |
16a53ecc N |
4376 | /* Need to check if array will still be degraded after recovery/resync |
4377 | * We don't need to check the 'failed' flag as when that gets set, | |
4378 | * recovery aborts. | |
4379 | */ | |
f001a70c | 4380 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4381 | if (conf->disks[i].rdev == NULL) |
4382 | still_degraded = 1; | |
4383 | ||
4384 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4385 | ||
83206d66 | 4386 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4387 | |
1442577b | 4388 | handle_stripe(sh); |
1da177e4 LT |
4389 | release_stripe(sh); |
4390 | ||
4391 | return STRIPE_SECTORS; | |
4392 | } | |
4393 | ||
46031f9a RBJ |
4394 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
4395 | { | |
4396 | /* We may not be able to submit a whole bio at once as there | |
4397 | * may not be enough stripe_heads available. | |
4398 | * We cannot pre-allocate enough stripe_heads as we may need | |
4399 | * more than exist in the cache (if we allow ever large chunks). | |
4400 | * So we do one stripe head at a time and record in | |
4401 | * ->bi_hw_segments how many have been done. | |
4402 | * | |
4403 | * We *know* that this entire raid_bio is in one chunk, so | |
4404 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4405 | */ | |
4406 | struct stripe_head *sh; | |
911d4ee8 | 4407 | int dd_idx; |
46031f9a RBJ |
4408 | sector_t sector, logical_sector, last_sector; |
4409 | int scnt = 0; | |
4410 | int remaining; | |
4411 | int handled = 0; | |
4412 | ||
4413 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4414 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4415 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4416 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4417 | ||
4418 | for (; logical_sector < last_sector; | |
387bb173 NB |
4419 | logical_sector += STRIPE_SECTORS, |
4420 | sector += STRIPE_SECTORS, | |
4421 | scnt++) { | |
46031f9a | 4422 | |
960e739d | 4423 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4424 | /* already done this stripe */ |
4425 | continue; | |
4426 | ||
a8c906ca | 4427 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4428 | |
4429 | if (!sh) { | |
4430 | /* failed to get a stripe - must wait */ | |
960e739d | 4431 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4432 | conf->retry_read_aligned = raid_bio; |
4433 | return handled; | |
4434 | } | |
4435 | ||
4436 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4437 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4438 | release_stripe(sh); | |
960e739d | 4439 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4440 | conf->retry_read_aligned = raid_bio; |
4441 | return handled; | |
4442 | } | |
4443 | ||
36d1c647 | 4444 | handle_stripe(sh); |
46031f9a RBJ |
4445 | release_stripe(sh); |
4446 | handled++; | |
4447 | } | |
4448 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4449 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4450 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4451 | if (remaining == 0) |
4452 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4453 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4454 | wake_up(&conf->wait_for_stripe); | |
4455 | return handled; | |
4456 | } | |
4457 | ||
46031f9a | 4458 | |
1da177e4 LT |
4459 | /* |
4460 | * This is our raid5 kernel thread. | |
4461 | * | |
4462 | * We scan the hash table for stripes which can be handled now. | |
4463 | * During the scan, completed stripes are saved for us by the interrupt | |
4464 | * handler, so that they will not have to wait for our next wakeup. | |
4465 | */ | |
6ed3003c | 4466 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4467 | { |
4468 | struct stripe_head *sh; | |
070ec55d | 4469 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 4470 | int handled; |
e1dfa0a2 | 4471 | struct blk_plug plug; |
1da177e4 | 4472 | |
45b4233c | 4473 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4474 | |
4475 | md_check_recovery(mddev); | |
1da177e4 | 4476 | |
e1dfa0a2 | 4477 | blk_start_plug(&plug); |
1da177e4 LT |
4478 | handled = 0; |
4479 | spin_lock_irq(&conf->device_lock); | |
4480 | while (1) { | |
46031f9a | 4481 | struct bio *bio; |
1da177e4 | 4482 | |
7c13edc8 N |
4483 | if (atomic_read(&mddev->plug_cnt) == 0 && |
4484 | !list_empty(&conf->bitmap_list)) { | |
4485 | /* Now is a good time to flush some bitmap updates */ | |
4486 | conf->seq_flush++; | |
700e432d | 4487 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4488 | bitmap_unplug(mddev->bitmap); |
700e432d | 4489 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4490 | conf->seq_write = conf->seq_flush; |
72626685 N |
4491 | activate_bit_delay(conf); |
4492 | } | |
7c13edc8 N |
4493 | if (atomic_read(&mddev->plug_cnt) == 0) |
4494 | raid5_activate_delayed(conf); | |
72626685 | 4495 | |
46031f9a RBJ |
4496 | while ((bio = remove_bio_from_retry(conf))) { |
4497 | int ok; | |
4498 | spin_unlock_irq(&conf->device_lock); | |
4499 | ok = retry_aligned_read(conf, bio); | |
4500 | spin_lock_irq(&conf->device_lock); | |
4501 | if (!ok) | |
4502 | break; | |
4503 | handled++; | |
4504 | } | |
4505 | ||
8b3e6cdc DW |
4506 | sh = __get_priority_stripe(conf); |
4507 | ||
c9f21aaf | 4508 | if (!sh) |
1da177e4 | 4509 | break; |
1da177e4 LT |
4510 | spin_unlock_irq(&conf->device_lock); |
4511 | ||
4512 | handled++; | |
417b8d4a DW |
4513 | handle_stripe(sh); |
4514 | release_stripe(sh); | |
4515 | cond_resched(); | |
1da177e4 LT |
4516 | |
4517 | spin_lock_irq(&conf->device_lock); | |
4518 | } | |
45b4233c | 4519 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4520 | |
4521 | spin_unlock_irq(&conf->device_lock); | |
4522 | ||
c9f21aaf | 4523 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4524 | blk_finish_plug(&plug); |
1da177e4 | 4525 | |
45b4233c | 4526 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4527 | } |
4528 | ||
3f294f4f | 4529 | static ssize_t |
007583c9 | 4530 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4531 | { |
070ec55d | 4532 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4533 | if (conf) |
4534 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4535 | else | |
4536 | return 0; | |
3f294f4f N |
4537 | } |
4538 | ||
c41d4ac4 N |
4539 | int |
4540 | raid5_set_cache_size(mddev_t *mddev, int size) | |
3f294f4f | 4541 | { |
070ec55d | 4542 | raid5_conf_t *conf = mddev->private; |
b5470dc5 DW |
4543 | int err; |
4544 | ||
c41d4ac4 | 4545 | if (size <= 16 || size > 32768) |
3f294f4f | 4546 | return -EINVAL; |
c41d4ac4 | 4547 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4548 | if (drop_one_stripe(conf)) |
4549 | conf->max_nr_stripes--; | |
4550 | else | |
4551 | break; | |
4552 | } | |
b5470dc5 DW |
4553 | err = md_allow_write(mddev); |
4554 | if (err) | |
4555 | return err; | |
c41d4ac4 | 4556 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4557 | if (grow_one_stripe(conf)) |
4558 | conf->max_nr_stripes++; | |
4559 | else break; | |
4560 | } | |
c41d4ac4 N |
4561 | return 0; |
4562 | } | |
4563 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4564 | ||
4565 | static ssize_t | |
4566 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) | |
4567 | { | |
4568 | raid5_conf_t *conf = mddev->private; | |
4569 | unsigned long new; | |
4570 | int err; | |
4571 | ||
4572 | if (len >= PAGE_SIZE) | |
4573 | return -EINVAL; | |
4574 | if (!conf) | |
4575 | return -ENODEV; | |
4576 | ||
4577 | if (strict_strtoul(page, 10, &new)) | |
4578 | return -EINVAL; | |
4579 | err = raid5_set_cache_size(mddev, new); | |
4580 | if (err) | |
4581 | return err; | |
3f294f4f N |
4582 | return len; |
4583 | } | |
007583c9 | 4584 | |
96de1e66 N |
4585 | static struct md_sysfs_entry |
4586 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4587 | raid5_show_stripe_cache_size, | |
4588 | raid5_store_stripe_cache_size); | |
3f294f4f | 4589 | |
8b3e6cdc DW |
4590 | static ssize_t |
4591 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4592 | { | |
070ec55d | 4593 | raid5_conf_t *conf = mddev->private; |
8b3e6cdc DW |
4594 | if (conf) |
4595 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4596 | else | |
4597 | return 0; | |
4598 | } | |
4599 | ||
4600 | static ssize_t | |
4601 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4602 | { | |
070ec55d | 4603 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4604 | unsigned long new; |
8b3e6cdc DW |
4605 | if (len >= PAGE_SIZE) |
4606 | return -EINVAL; | |
4607 | if (!conf) | |
4608 | return -ENODEV; | |
4609 | ||
4ef197d8 | 4610 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4611 | return -EINVAL; |
4ef197d8 | 4612 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4613 | return -EINVAL; |
4614 | conf->bypass_threshold = new; | |
4615 | return len; | |
4616 | } | |
4617 | ||
4618 | static struct md_sysfs_entry | |
4619 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4620 | S_IRUGO | S_IWUSR, | |
4621 | raid5_show_preread_threshold, | |
4622 | raid5_store_preread_threshold); | |
4623 | ||
3f294f4f | 4624 | static ssize_t |
96de1e66 | 4625 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4626 | { |
070ec55d | 4627 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4628 | if (conf) |
4629 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4630 | else | |
4631 | return 0; | |
3f294f4f N |
4632 | } |
4633 | ||
96de1e66 N |
4634 | static struct md_sysfs_entry |
4635 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4636 | |
007583c9 | 4637 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4638 | &raid5_stripecache_size.attr, |
4639 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4640 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4641 | NULL, |
4642 | }; | |
007583c9 N |
4643 | static struct attribute_group raid5_attrs_group = { |
4644 | .name = NULL, | |
4645 | .attrs = raid5_attrs, | |
3f294f4f N |
4646 | }; |
4647 | ||
80c3a6ce DW |
4648 | static sector_t |
4649 | raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
4650 | { | |
070ec55d | 4651 | raid5_conf_t *conf = mddev->private; |
80c3a6ce DW |
4652 | |
4653 | if (!sectors) | |
4654 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4655 | if (!raid_disks) |
7ec05478 | 4656 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4657 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4658 | |
9d8f0363 | 4659 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4660 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4661 | return sectors * (raid_disks - conf->max_degraded); |
4662 | } | |
4663 | ||
36d1c647 DW |
4664 | static void raid5_free_percpu(raid5_conf_t *conf) |
4665 | { | |
4666 | struct raid5_percpu *percpu; | |
4667 | unsigned long cpu; | |
4668 | ||
4669 | if (!conf->percpu) | |
4670 | return; | |
4671 | ||
4672 | get_online_cpus(); | |
4673 | for_each_possible_cpu(cpu) { | |
4674 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4675 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4676 | kfree(percpu->scribble); |
36d1c647 DW |
4677 | } |
4678 | #ifdef CONFIG_HOTPLUG_CPU | |
4679 | unregister_cpu_notifier(&conf->cpu_notify); | |
4680 | #endif | |
4681 | put_online_cpus(); | |
4682 | ||
4683 | free_percpu(conf->percpu); | |
4684 | } | |
4685 | ||
95fc17aa DW |
4686 | static void free_conf(raid5_conf_t *conf) |
4687 | { | |
4688 | shrink_stripes(conf); | |
36d1c647 | 4689 | raid5_free_percpu(conf); |
95fc17aa DW |
4690 | kfree(conf->disks); |
4691 | kfree(conf->stripe_hashtbl); | |
4692 | kfree(conf); | |
4693 | } | |
4694 | ||
36d1c647 DW |
4695 | #ifdef CONFIG_HOTPLUG_CPU |
4696 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4697 | void *hcpu) | |
4698 | { | |
4699 | raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify); | |
4700 | long cpu = (long)hcpu; | |
4701 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4702 | ||
4703 | switch (action) { | |
4704 | case CPU_UP_PREPARE: | |
4705 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4706 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4707 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4708 | if (!percpu->scribble) |
4709 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4710 | ||
4711 | if (!percpu->scribble || | |
4712 | (conf->level == 6 && !percpu->spare_page)) { | |
4713 | safe_put_page(percpu->spare_page); | |
4714 | kfree(percpu->scribble); | |
36d1c647 DW |
4715 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4716 | __func__, cpu); | |
55af6bb5 | 4717 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4718 | } |
4719 | break; | |
4720 | case CPU_DEAD: | |
4721 | case CPU_DEAD_FROZEN: | |
4722 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4723 | kfree(percpu->scribble); |
36d1c647 | 4724 | percpu->spare_page = NULL; |
d6f38f31 | 4725 | percpu->scribble = NULL; |
36d1c647 DW |
4726 | break; |
4727 | default: | |
4728 | break; | |
4729 | } | |
4730 | return NOTIFY_OK; | |
4731 | } | |
4732 | #endif | |
4733 | ||
4734 | static int raid5_alloc_percpu(raid5_conf_t *conf) | |
4735 | { | |
4736 | unsigned long cpu; | |
4737 | struct page *spare_page; | |
a29d8b8e | 4738 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4739 | void *scribble; |
36d1c647 DW |
4740 | int err; |
4741 | ||
36d1c647 DW |
4742 | allcpus = alloc_percpu(struct raid5_percpu); |
4743 | if (!allcpus) | |
4744 | return -ENOMEM; | |
4745 | conf->percpu = allcpus; | |
4746 | ||
4747 | get_online_cpus(); | |
4748 | err = 0; | |
4749 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4750 | if (conf->level == 6) { |
4751 | spare_page = alloc_page(GFP_KERNEL); | |
4752 | if (!spare_page) { | |
4753 | err = -ENOMEM; | |
4754 | break; | |
4755 | } | |
4756 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4757 | } | |
5e5e3e78 | 4758 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4759 | if (!scribble) { |
36d1c647 DW |
4760 | err = -ENOMEM; |
4761 | break; | |
4762 | } | |
d6f38f31 | 4763 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4764 | } |
4765 | #ifdef CONFIG_HOTPLUG_CPU | |
4766 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4767 | conf->cpu_notify.priority = 0; | |
4768 | if (err == 0) | |
4769 | err = register_cpu_notifier(&conf->cpu_notify); | |
4770 | #endif | |
4771 | put_online_cpus(); | |
4772 | ||
4773 | return err; | |
4774 | } | |
4775 | ||
91adb564 | 4776 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4777 | { |
4778 | raid5_conf_t *conf; | |
5e5e3e78 | 4779 | int raid_disk, memory, max_disks; |
1da177e4 LT |
4780 | mdk_rdev_t *rdev; |
4781 | struct disk_info *disk; | |
1da177e4 | 4782 | |
91adb564 N |
4783 | if (mddev->new_level != 5 |
4784 | && mddev->new_level != 4 | |
4785 | && mddev->new_level != 6) { | |
0c55e022 | 4786 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4787 | mdname(mddev), mddev->new_level); |
4788 | return ERR_PTR(-EIO); | |
1da177e4 | 4789 | } |
91adb564 N |
4790 | if ((mddev->new_level == 5 |
4791 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4792 | (mddev->new_level == 6 | |
4793 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4794 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4795 | mdname(mddev), mddev->new_layout); |
4796 | return ERR_PTR(-EIO); | |
99c0fb5f | 4797 | } |
91adb564 | 4798 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4799 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4800 | mdname(mddev), mddev->raid_disks); |
4801 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4802 | } |
4803 | ||
664e7c41 AN |
4804 | if (!mddev->new_chunk_sectors || |
4805 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4806 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4807 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4808 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4809 | return ERR_PTR(-EINVAL); |
f6705578 N |
4810 | } |
4811 | ||
91adb564 N |
4812 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4813 | if (conf == NULL) | |
1da177e4 | 4814 | goto abort; |
f5efd45a DW |
4815 | spin_lock_init(&conf->device_lock); |
4816 | init_waitqueue_head(&conf->wait_for_stripe); | |
4817 | init_waitqueue_head(&conf->wait_for_overlap); | |
4818 | INIT_LIST_HEAD(&conf->handle_list); | |
4819 | INIT_LIST_HEAD(&conf->hold_list); | |
4820 | INIT_LIST_HEAD(&conf->delayed_list); | |
4821 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4822 | INIT_LIST_HEAD(&conf->inactive_list); | |
4823 | atomic_set(&conf->active_stripes, 0); | |
4824 | atomic_set(&conf->preread_active_stripes, 0); | |
4825 | atomic_set(&conf->active_aligned_reads, 0); | |
4826 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
91adb564 N |
4827 | |
4828 | conf->raid_disks = mddev->raid_disks; | |
4829 | if (mddev->reshape_position == MaxSector) | |
4830 | conf->previous_raid_disks = mddev->raid_disks; | |
4831 | else | |
f6705578 | 4832 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4833 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4834 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4835 | |
5e5e3e78 | 4836 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4837 | GFP_KERNEL); |
4838 | if (!conf->disks) | |
4839 | goto abort; | |
9ffae0cf | 4840 | |
1da177e4 LT |
4841 | conf->mddev = mddev; |
4842 | ||
fccddba0 | 4843 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4844 | goto abort; |
1da177e4 | 4845 | |
36d1c647 DW |
4846 | conf->level = mddev->new_level; |
4847 | if (raid5_alloc_percpu(conf) != 0) | |
4848 | goto abort; | |
4849 | ||
0c55e022 | 4850 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4851 | |
159ec1fc | 4852 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4853 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4854 | if (raid_disk >= max_disks |
1da177e4 LT |
4855 | || raid_disk < 0) |
4856 | continue; | |
4857 | disk = conf->disks + raid_disk; | |
4858 | ||
4859 | disk->rdev = rdev; | |
4860 | ||
b2d444d7 | 4861 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 4862 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
4863 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
4864 | " disk %d\n", | |
4865 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 4866 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
4867 | /* Cannot rely on bitmap to complete recovery */ |
4868 | conf->fullsync = 1; | |
1da177e4 LT |
4869 | } |
4870 | ||
09c9e5fa | 4871 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4872 | conf->level = mddev->new_level; |
16a53ecc N |
4873 | if (conf->level == 6) |
4874 | conf->max_degraded = 2; | |
4875 | else | |
4876 | conf->max_degraded = 1; | |
91adb564 | 4877 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4878 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4879 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4880 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4881 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4882 | conf->prev_algo = mddev->layout; |
4883 | } | |
1da177e4 | 4884 | |
91adb564 | 4885 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4886 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4887 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4888 | printk(KERN_ERR | |
0c55e022 N |
4889 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
4890 | mdname(mddev), memory); | |
91adb564 N |
4891 | goto abort; |
4892 | } else | |
0c55e022 N |
4893 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
4894 | mdname(mddev), memory); | |
1da177e4 | 4895 | |
0da3c619 | 4896 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4897 | if (!conf->thread) { |
4898 | printk(KERN_ERR | |
0c55e022 | 4899 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 4900 | mdname(mddev)); |
16a53ecc N |
4901 | goto abort; |
4902 | } | |
91adb564 N |
4903 | |
4904 | return conf; | |
4905 | ||
4906 | abort: | |
4907 | if (conf) { | |
95fc17aa | 4908 | free_conf(conf); |
91adb564 N |
4909 | return ERR_PTR(-EIO); |
4910 | } else | |
4911 | return ERR_PTR(-ENOMEM); | |
4912 | } | |
4913 | ||
c148ffdc N |
4914 | |
4915 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4916 | { | |
4917 | switch (algo) { | |
4918 | case ALGORITHM_PARITY_0: | |
4919 | if (raid_disk < max_degraded) | |
4920 | return 1; | |
4921 | break; | |
4922 | case ALGORITHM_PARITY_N: | |
4923 | if (raid_disk >= raid_disks - max_degraded) | |
4924 | return 1; | |
4925 | break; | |
4926 | case ALGORITHM_PARITY_0_6: | |
4927 | if (raid_disk == 0 || | |
4928 | raid_disk == raid_disks - 1) | |
4929 | return 1; | |
4930 | break; | |
4931 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4932 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4933 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4934 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4935 | if (raid_disk == raid_disks - 1) | |
4936 | return 1; | |
4937 | } | |
4938 | return 0; | |
4939 | } | |
4940 | ||
91adb564 N |
4941 | static int run(mddev_t *mddev) |
4942 | { | |
4943 | raid5_conf_t *conf; | |
9f7c2220 | 4944 | int working_disks = 0; |
c148ffdc | 4945 | int dirty_parity_disks = 0; |
91adb564 | 4946 | mdk_rdev_t *rdev; |
c148ffdc | 4947 | sector_t reshape_offset = 0; |
91adb564 | 4948 | |
8c6ac868 | 4949 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 4950 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
4951 | " -- starting background reconstruction\n", |
4952 | mdname(mddev)); | |
91adb564 N |
4953 | if (mddev->reshape_position != MaxSector) { |
4954 | /* Check that we can continue the reshape. | |
4955 | * Currently only disks can change, it must | |
4956 | * increase, and we must be past the point where | |
4957 | * a stripe over-writes itself | |
4958 | */ | |
4959 | sector_t here_new, here_old; | |
4960 | int old_disks; | |
18b00334 | 4961 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4962 | |
88ce4930 | 4963 | if (mddev->new_level != mddev->level) { |
0c55e022 | 4964 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
4965 | "required - aborting.\n", |
4966 | mdname(mddev)); | |
4967 | return -EINVAL; | |
4968 | } | |
91adb564 N |
4969 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4970 | /* reshape_position must be on a new-stripe boundary, and one | |
4971 | * further up in new geometry must map after here in old | |
4972 | * geometry. | |
4973 | */ | |
4974 | here_new = mddev->reshape_position; | |
664e7c41 | 4975 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 4976 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
4977 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
4978 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
4979 | return -EINVAL; |
4980 | } | |
c148ffdc | 4981 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4982 | /* here_new is the stripe we will write to */ |
4983 | here_old = mddev->reshape_position; | |
9d8f0363 | 4984 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4985 | (old_disks-max_degraded)); |
4986 | /* here_old is the first stripe that we might need to read | |
4987 | * from */ | |
67ac6011 N |
4988 | if (mddev->delta_disks == 0) { |
4989 | /* We cannot be sure it is safe to start an in-place | |
4990 | * reshape. It is only safe if user-space if monitoring | |
4991 | * and taking constant backups. | |
4992 | * mdadm always starts a situation like this in | |
4993 | * readonly mode so it can take control before | |
4994 | * allowing any writes. So just check for that. | |
4995 | */ | |
4996 | if ((here_new * mddev->new_chunk_sectors != | |
4997 | here_old * mddev->chunk_sectors) || | |
4998 | mddev->ro == 0) { | |
0c55e022 N |
4999 | printk(KERN_ERR "md/raid:%s: in-place reshape must be started" |
5000 | " in read-only mode - aborting\n", | |
5001 | mdname(mddev)); | |
67ac6011 N |
5002 | return -EINVAL; |
5003 | } | |
5004 | } else if (mddev->delta_disks < 0 | |
5005 | ? (here_new * mddev->new_chunk_sectors <= | |
5006 | here_old * mddev->chunk_sectors) | |
5007 | : (here_new * mddev->new_chunk_sectors >= | |
5008 | here_old * mddev->chunk_sectors)) { | |
91adb564 | 5009 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
5010 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
5011 | "auto-recovery - aborting.\n", | |
5012 | mdname(mddev)); | |
91adb564 N |
5013 | return -EINVAL; |
5014 | } | |
0c55e022 N |
5015 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
5016 | mdname(mddev)); | |
91adb564 N |
5017 | /* OK, we should be able to continue; */ |
5018 | } else { | |
5019 | BUG_ON(mddev->level != mddev->new_level); | |
5020 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 5021 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 5022 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 5023 | } |
91adb564 | 5024 | |
245f46c2 N |
5025 | if (mddev->private == NULL) |
5026 | conf = setup_conf(mddev); | |
5027 | else | |
5028 | conf = mddev->private; | |
5029 | ||
91adb564 N |
5030 | if (IS_ERR(conf)) |
5031 | return PTR_ERR(conf); | |
5032 | ||
5033 | mddev->thread = conf->thread; | |
5034 | conf->thread = NULL; | |
5035 | mddev->private = conf; | |
5036 | ||
5037 | /* | |
5038 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
5039 | */ | |
c148ffdc N |
5040 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
5041 | if (rdev->raid_disk < 0) | |
5042 | continue; | |
2f115882 | 5043 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 5044 | working_disks++; |
2f115882 N |
5045 | continue; |
5046 | } | |
c148ffdc N |
5047 | /* This disc is not fully in-sync. However if it |
5048 | * just stored parity (beyond the recovery_offset), | |
5049 | * when we don't need to be concerned about the | |
5050 | * array being dirty. | |
5051 | * When reshape goes 'backwards', we never have | |
5052 | * partially completed devices, so we only need | |
5053 | * to worry about reshape going forwards. | |
5054 | */ | |
5055 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
5056 | if (mddev->major_version == 0 && | |
5057 | mddev->minor_version > 90) | |
5058 | rdev->recovery_offset = reshape_offset; | |
5059 | ||
c148ffdc N |
5060 | if (rdev->recovery_offset < reshape_offset) { |
5061 | /* We need to check old and new layout */ | |
5062 | if (!only_parity(rdev->raid_disk, | |
5063 | conf->algorithm, | |
5064 | conf->raid_disks, | |
5065 | conf->max_degraded)) | |
5066 | continue; | |
5067 | } | |
5068 | if (!only_parity(rdev->raid_disk, | |
5069 | conf->prev_algo, | |
5070 | conf->previous_raid_disks, | |
5071 | conf->max_degraded)) | |
5072 | continue; | |
5073 | dirty_parity_disks++; | |
5074 | } | |
91adb564 | 5075 | |
5e5e3e78 N |
5076 | mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks) |
5077 | - working_disks); | |
91adb564 | 5078 | |
674806d6 | 5079 | if (has_failed(conf)) { |
0c55e022 | 5080 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 5081 | " (%d/%d failed)\n", |
02c2de8c | 5082 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
5083 | goto abort; |
5084 | } | |
5085 | ||
91adb564 | 5086 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 5087 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
5088 | mddev->resync_max_sectors = mddev->dev_sectors; |
5089 | ||
c148ffdc | 5090 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 5091 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
5092 | if (mddev->ok_start_degraded) |
5093 | printk(KERN_WARNING | |
0c55e022 N |
5094 | "md/raid:%s: starting dirty degraded array" |
5095 | " - data corruption possible.\n", | |
6ff8d8ec N |
5096 | mdname(mddev)); |
5097 | else { | |
5098 | printk(KERN_ERR | |
0c55e022 | 5099 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
5100 | mdname(mddev)); |
5101 | goto abort; | |
5102 | } | |
1da177e4 LT |
5103 | } |
5104 | ||
1da177e4 | 5105 | if (mddev->degraded == 0) |
0c55e022 N |
5106 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
5107 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
5108 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
5109 | mddev->new_layout); | |
1da177e4 | 5110 | else |
0c55e022 N |
5111 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
5112 | " out of %d devices, algorithm %d\n", | |
5113 | mdname(mddev), conf->level, | |
5114 | mddev->raid_disks - mddev->degraded, | |
5115 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
5116 | |
5117 | print_raid5_conf(conf); | |
5118 | ||
fef9c61f | 5119 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 5120 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5121 | atomic_set(&conf->reshape_stripes, 0); |
5122 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5123 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5124 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5125 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5126 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5127 | "reshape"); |
f6705578 N |
5128 | } |
5129 | ||
1da177e4 LT |
5130 | |
5131 | /* Ok, everything is just fine now */ | |
a64c876f N |
5132 | if (mddev->to_remove == &raid5_attrs_group) |
5133 | mddev->to_remove = NULL; | |
00bcb4ac N |
5134 | else if (mddev->kobj.sd && |
5135 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 5136 | printk(KERN_WARNING |
4a5add49 | 5137 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 5138 | mdname(mddev)); |
4a5add49 | 5139 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5140 | |
4a5add49 | 5141 | if (mddev->queue) { |
9f7c2220 | 5142 | int chunk_size; |
4a5add49 N |
5143 | /* read-ahead size must cover two whole stripes, which |
5144 | * is 2 * (datadisks) * chunksize where 'n' is the | |
5145 | * number of raid devices | |
5146 | */ | |
5147 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
5148 | int stripe = data_disks * | |
5149 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
5150 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
5151 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 5152 | |
4a5add49 | 5153 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 5154 | |
11d8a6e3 N |
5155 | mddev->queue->backing_dev_info.congested_data = mddev; |
5156 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 5157 | |
9f7c2220 N |
5158 | chunk_size = mddev->chunk_sectors << 9; |
5159 | blk_queue_io_min(mddev->queue, chunk_size); | |
5160 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5161 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 5162 | |
9f7c2220 N |
5163 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5164 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
5165 | rdev->data_offset << 9); | |
5166 | } | |
23032a0e | 5167 | |
1da177e4 LT |
5168 | return 0; |
5169 | abort: | |
e0cf8f04 | 5170 | md_unregister_thread(mddev->thread); |
91adb564 | 5171 | mddev->thread = NULL; |
1da177e4 LT |
5172 | if (conf) { |
5173 | print_raid5_conf(conf); | |
95fc17aa | 5174 | free_conf(conf); |
1da177e4 LT |
5175 | } |
5176 | mddev->private = NULL; | |
0c55e022 | 5177 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
5178 | return -EIO; |
5179 | } | |
5180 | ||
3f294f4f | 5181 | static int stop(mddev_t *mddev) |
1da177e4 | 5182 | { |
7b92813c | 5183 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
5184 | |
5185 | md_unregister_thread(mddev->thread); | |
5186 | mddev->thread = NULL; | |
11d8a6e3 N |
5187 | if (mddev->queue) |
5188 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 5189 | free_conf(conf); |
a64c876f N |
5190 | mddev->private = NULL; |
5191 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
5192 | return 0; |
5193 | } | |
5194 | ||
45b4233c | 5195 | #ifdef DEBUG |
d710e138 | 5196 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
5197 | { |
5198 | int i; | |
5199 | ||
16a53ecc N |
5200 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
5201 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
5202 | seq_printf(seq, "sh %llu, count %d.\n", | |
5203 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
5204 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 5205 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
5206 | seq_printf(seq, "(cache%d: %p %ld) ", |
5207 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 5208 | } |
16a53ecc | 5209 | seq_printf(seq, "\n"); |
1da177e4 LT |
5210 | } |
5211 | ||
d710e138 | 5212 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
5213 | { |
5214 | struct stripe_head *sh; | |
fccddba0 | 5215 | struct hlist_node *hn; |
1da177e4 LT |
5216 | int i; |
5217 | ||
5218 | spin_lock_irq(&conf->device_lock); | |
5219 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 5220 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
5221 | if (sh->raid_conf != conf) |
5222 | continue; | |
16a53ecc | 5223 | print_sh(seq, sh); |
1da177e4 LT |
5224 | } |
5225 | } | |
5226 | spin_unlock_irq(&conf->device_lock); | |
5227 | } | |
5228 | #endif | |
5229 | ||
d710e138 | 5230 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 | 5231 | { |
7b92813c | 5232 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
5233 | int i; |
5234 | ||
9d8f0363 AN |
5235 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5236 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5237 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5238 | for (i = 0; i < conf->raid_disks; i++) |
5239 | seq_printf (seq, "%s", | |
5240 | conf->disks[i].rdev && | |
b2d444d7 | 5241 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5242 | seq_printf (seq, "]"); |
45b4233c | 5243 | #ifdef DEBUG |
16a53ecc N |
5244 | seq_printf (seq, "\n"); |
5245 | printall(seq, conf); | |
1da177e4 LT |
5246 | #endif |
5247 | } | |
5248 | ||
5249 | static void print_raid5_conf (raid5_conf_t *conf) | |
5250 | { | |
5251 | int i; | |
5252 | struct disk_info *tmp; | |
5253 | ||
0c55e022 | 5254 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
5255 | if (!conf) { |
5256 | printk("(conf==NULL)\n"); | |
5257 | return; | |
5258 | } | |
0c55e022 N |
5259 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
5260 | conf->raid_disks, | |
5261 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5262 | |
5263 | for (i = 0; i < conf->raid_disks; i++) { | |
5264 | char b[BDEVNAME_SIZE]; | |
5265 | tmp = conf->disks + i; | |
5266 | if (tmp->rdev) | |
0c55e022 N |
5267 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
5268 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
5269 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
5270 | } |
5271 | } | |
5272 | ||
5273 | static int raid5_spare_active(mddev_t *mddev) | |
5274 | { | |
5275 | int i; | |
5276 | raid5_conf_t *conf = mddev->private; | |
5277 | struct disk_info *tmp; | |
6b965620 N |
5278 | int count = 0; |
5279 | unsigned long flags; | |
1da177e4 LT |
5280 | |
5281 | for (i = 0; i < conf->raid_disks; i++) { | |
5282 | tmp = conf->disks + i; | |
5283 | if (tmp->rdev | |
70fffd0b | 5284 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 5285 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 5286 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 5287 | count++; |
43c73ca4 | 5288 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
5289 | } |
5290 | } | |
6b965620 N |
5291 | spin_lock_irqsave(&conf->device_lock, flags); |
5292 | mddev->degraded -= count; | |
5293 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 5294 | print_raid5_conf(conf); |
6b965620 | 5295 | return count; |
1da177e4 LT |
5296 | } |
5297 | ||
5298 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
5299 | { | |
5300 | raid5_conf_t *conf = mddev->private; | |
5301 | int err = 0; | |
5302 | mdk_rdev_t *rdev; | |
5303 | struct disk_info *p = conf->disks + number; | |
5304 | ||
5305 | print_raid5_conf(conf); | |
5306 | rdev = p->rdev; | |
5307 | if (rdev) { | |
ec32a2bd N |
5308 | if (number >= conf->raid_disks && |
5309 | conf->reshape_progress == MaxSector) | |
5310 | clear_bit(In_sync, &rdev->flags); | |
5311 | ||
b2d444d7 | 5312 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
5313 | atomic_read(&rdev->nr_pending)) { |
5314 | err = -EBUSY; | |
5315 | goto abort; | |
5316 | } | |
dfc70645 N |
5317 | /* Only remove non-faulty devices if recovery |
5318 | * isn't possible. | |
5319 | */ | |
5320 | if (!test_bit(Faulty, &rdev->flags) && | |
674806d6 | 5321 | !has_failed(conf) && |
ec32a2bd | 5322 | number < conf->raid_disks) { |
dfc70645 N |
5323 | err = -EBUSY; |
5324 | goto abort; | |
5325 | } | |
1da177e4 | 5326 | p->rdev = NULL; |
fbd568a3 | 5327 | synchronize_rcu(); |
1da177e4 LT |
5328 | if (atomic_read(&rdev->nr_pending)) { |
5329 | /* lost the race, try later */ | |
5330 | err = -EBUSY; | |
5331 | p->rdev = rdev; | |
5332 | } | |
5333 | } | |
5334 | abort: | |
5335 | ||
5336 | print_raid5_conf(conf); | |
5337 | return err; | |
5338 | } | |
5339 | ||
5340 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
5341 | { | |
5342 | raid5_conf_t *conf = mddev->private; | |
199050ea | 5343 | int err = -EEXIST; |
1da177e4 LT |
5344 | int disk; |
5345 | struct disk_info *p; | |
6c2fce2e NB |
5346 | int first = 0; |
5347 | int last = conf->raid_disks - 1; | |
1da177e4 | 5348 | |
674806d6 | 5349 | if (has_failed(conf)) |
1da177e4 | 5350 | /* no point adding a device */ |
199050ea | 5351 | return -EINVAL; |
1da177e4 | 5352 | |
6c2fce2e NB |
5353 | if (rdev->raid_disk >= 0) |
5354 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5355 | |
5356 | /* | |
16a53ecc N |
5357 | * find the disk ... but prefer rdev->saved_raid_disk |
5358 | * if possible. | |
1da177e4 | 5359 | */ |
16a53ecc | 5360 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5361 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5362 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5363 | disk = rdev->saved_raid_disk; | |
5364 | else | |
6c2fce2e NB |
5365 | disk = first; |
5366 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5367 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5368 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5369 | rdev->raid_disk = disk; |
199050ea | 5370 | err = 0; |
72626685 N |
5371 | if (rdev->saved_raid_disk != disk) |
5372 | conf->fullsync = 1; | |
d6065f7b | 5373 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5374 | break; |
5375 | } | |
5376 | print_raid5_conf(conf); | |
199050ea | 5377 | return err; |
1da177e4 LT |
5378 | } |
5379 | ||
5380 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
5381 | { | |
5382 | /* no resync is happening, and there is enough space | |
5383 | * on all devices, so we can resize. | |
5384 | * We need to make sure resync covers any new space. | |
5385 | * If the array is shrinking we should possibly wait until | |
5386 | * any io in the removed space completes, but it hardly seems | |
5387 | * worth it. | |
5388 | */ | |
9d8f0363 | 5389 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5390 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5391 | mddev->raid_disks)); | |
b522adcd DW |
5392 | if (mddev->array_sectors > |
5393 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5394 | return -EINVAL; | |
f233ea5c | 5395 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5396 | revalidate_disk(mddev->gendisk); |
b098636c N |
5397 | if (sectors > mddev->dev_sectors && |
5398 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5399 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5400 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5401 | } | |
58c0fed4 | 5402 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5403 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5404 | return 0; |
5405 | } | |
5406 | ||
01ee22b4 N |
5407 | static int check_stripe_cache(mddev_t *mddev) |
5408 | { | |
5409 | /* Can only proceed if there are plenty of stripe_heads. | |
5410 | * We need a minimum of one full stripe,, and for sensible progress | |
5411 | * it is best to have about 4 times that. | |
5412 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5413 | * allow for chunk sizes up to 256K, which is probably OK. | |
5414 | * If the chunk size is greater, user-space should request more | |
5415 | * stripe_heads first. | |
5416 | */ | |
5417 | raid5_conf_t *conf = mddev->private; | |
5418 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5419 | > conf->max_nr_stripes || | |
5420 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5421 | > conf->max_nr_stripes) { | |
0c55e022 N |
5422 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5423 | mdname(mddev), | |
01ee22b4 N |
5424 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5425 | / STRIPE_SIZE)*4); | |
5426 | return 0; | |
5427 | } | |
5428 | return 1; | |
5429 | } | |
5430 | ||
50ac168a | 5431 | static int check_reshape(mddev_t *mddev) |
29269553 | 5432 | { |
070ec55d | 5433 | raid5_conf_t *conf = mddev->private; |
29269553 | 5434 | |
88ce4930 N |
5435 | if (mddev->delta_disks == 0 && |
5436 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5437 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5438 | return 0; /* nothing to do */ |
dba034ee N |
5439 | if (mddev->bitmap) |
5440 | /* Cannot grow a bitmap yet */ | |
5441 | return -EBUSY; | |
674806d6 | 5442 | if (has_failed(conf)) |
ec32a2bd N |
5443 | return -EINVAL; |
5444 | if (mddev->delta_disks < 0) { | |
5445 | /* We might be able to shrink, but the devices must | |
5446 | * be made bigger first. | |
5447 | * For raid6, 4 is the minimum size. | |
5448 | * Otherwise 2 is the minimum | |
5449 | */ | |
5450 | int min = 2; | |
5451 | if (mddev->level == 6) | |
5452 | min = 4; | |
5453 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5454 | return -EINVAL; | |
5455 | } | |
29269553 | 5456 | |
01ee22b4 | 5457 | if (!check_stripe_cache(mddev)) |
29269553 | 5458 | return -ENOSPC; |
29269553 | 5459 | |
ec32a2bd | 5460 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5461 | } |
5462 | ||
5463 | static int raid5_start_reshape(mddev_t *mddev) | |
5464 | { | |
070ec55d | 5465 | raid5_conf_t *conf = mddev->private; |
63c70c4f | 5466 | mdk_rdev_t *rdev; |
63c70c4f | 5467 | int spares = 0; |
c04be0aa | 5468 | unsigned long flags; |
63c70c4f | 5469 | |
f416885e | 5470 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5471 | return -EBUSY; |
5472 | ||
01ee22b4 N |
5473 | if (!check_stripe_cache(mddev)) |
5474 | return -ENOSPC; | |
5475 | ||
159ec1fc | 5476 | list_for_each_entry(rdev, &mddev->disks, same_set) |
469518a3 N |
5477 | if (!test_bit(In_sync, &rdev->flags) |
5478 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5479 | spares++; |
63c70c4f | 5480 | |
f416885e | 5481 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5482 | /* Not enough devices even to make a degraded array |
5483 | * of that size | |
5484 | */ | |
5485 | return -EINVAL; | |
5486 | ||
ec32a2bd N |
5487 | /* Refuse to reduce size of the array. Any reductions in |
5488 | * array size must be through explicit setting of array_size | |
5489 | * attribute. | |
5490 | */ | |
5491 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5492 | < mddev->array_sectors) { | |
0c55e022 | 5493 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5494 | "before number of disks\n", mdname(mddev)); |
5495 | return -EINVAL; | |
5496 | } | |
5497 | ||
f6705578 | 5498 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5499 | spin_lock_irq(&conf->device_lock); |
5500 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5501 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5502 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5503 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5504 | conf->prev_algo = conf->algorithm; |
5505 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5506 | if (mddev->delta_disks < 0) |
5507 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5508 | else | |
5509 | conf->reshape_progress = 0; | |
5510 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5511 | conf->generation++; |
29269553 N |
5512 | spin_unlock_irq(&conf->device_lock); |
5513 | ||
5514 | /* Add some new drives, as many as will fit. | |
5515 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5516 | * Don't add devices if we are reducing the number of |
5517 | * devices in the array. This is because it is not possible | |
5518 | * to correctly record the "partially reconstructed" state of | |
5519 | * such devices during the reshape and confusion could result. | |
29269553 | 5520 | */ |
87a8dec9 N |
5521 | if (mddev->delta_disks >= 0) { |
5522 | int added_devices = 0; | |
5523 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5524 | if (rdev->raid_disk < 0 && | |
5525 | !test_bit(Faulty, &rdev->flags)) { | |
5526 | if (raid5_add_disk(mddev, rdev) == 0) { | |
5527 | char nm[20]; | |
5528 | if (rdev->raid_disk | |
5529 | >= conf->previous_raid_disks) { | |
5530 | set_bit(In_sync, &rdev->flags); | |
5531 | added_devices++; | |
5532 | } else | |
5533 | rdev->recovery_offset = 0; | |
5534 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5535 | if (sysfs_create_link(&mddev->kobj, | |
5536 | &rdev->kobj, nm)) | |
5537 | /* Failure here is OK */; | |
50da0840 | 5538 | } |
87a8dec9 N |
5539 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5540 | && !test_bit(Faulty, &rdev->flags)) { | |
5541 | /* This is a spare that was manually added */ | |
5542 | set_bit(In_sync, &rdev->flags); | |
5543 | added_devices++; | |
5544 | } | |
29269553 | 5545 | |
87a8dec9 N |
5546 | /* When a reshape changes the number of devices, |
5547 | * ->degraded is measured against the larger of the | |
5548 | * pre and post number of devices. | |
5549 | */ | |
ec32a2bd | 5550 | spin_lock_irqsave(&conf->device_lock, flags); |
9eb07c25 | 5551 | mddev->degraded += (conf->raid_disks - conf->previous_raid_disks) |
ec32a2bd N |
5552 | - added_devices; |
5553 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5554 | } | |
63c70c4f | 5555 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5556 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5557 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5558 | |
29269553 N |
5559 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5560 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5561 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5562 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5563 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5564 | "reshape"); |
29269553 N |
5565 | if (!mddev->sync_thread) { |
5566 | mddev->recovery = 0; | |
5567 | spin_lock_irq(&conf->device_lock); | |
5568 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5569 | conf->reshape_progress = MaxSector; |
29269553 N |
5570 | spin_unlock_irq(&conf->device_lock); |
5571 | return -EAGAIN; | |
5572 | } | |
c8f517c4 | 5573 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5574 | md_wakeup_thread(mddev->sync_thread); |
5575 | md_new_event(mddev); | |
5576 | return 0; | |
5577 | } | |
29269553 | 5578 | |
ec32a2bd N |
5579 | /* This is called from the reshape thread and should make any |
5580 | * changes needed in 'conf' | |
5581 | */ | |
29269553 N |
5582 | static void end_reshape(raid5_conf_t *conf) |
5583 | { | |
29269553 | 5584 | |
f6705578 | 5585 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5586 | |
f6705578 | 5587 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5588 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5589 | conf->reshape_progress = MaxSector; |
f6705578 | 5590 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5591 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5592 | |
5593 | /* read-ahead size must cover two whole stripes, which is | |
5594 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5595 | */ | |
4a5add49 | 5596 | if (conf->mddev->queue) { |
cea9c228 | 5597 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5598 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5599 | / PAGE_SIZE); |
16a53ecc N |
5600 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5601 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5602 | } | |
29269553 | 5603 | } |
29269553 N |
5604 | } |
5605 | ||
ec32a2bd N |
5606 | /* This is called from the raid5d thread with mddev_lock held. |
5607 | * It makes config changes to the device. | |
5608 | */ | |
cea9c228 N |
5609 | static void raid5_finish_reshape(mddev_t *mddev) |
5610 | { | |
070ec55d | 5611 | raid5_conf_t *conf = mddev->private; |
cea9c228 N |
5612 | |
5613 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5614 | ||
ec32a2bd N |
5615 | if (mddev->delta_disks > 0) { |
5616 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5617 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5618 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5619 | } else { |
5620 | int d; | |
ec32a2bd N |
5621 | mddev->degraded = conf->raid_disks; |
5622 | for (d = 0; d < conf->raid_disks ; d++) | |
5623 | if (conf->disks[d].rdev && | |
5624 | test_bit(In_sync, | |
5625 | &conf->disks[d].rdev->flags)) | |
5626 | mddev->degraded--; | |
5627 | for (d = conf->raid_disks ; | |
5628 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 N |
5629 | d++) { |
5630 | mdk_rdev_t *rdev = conf->disks[d].rdev; | |
5631 | if (rdev && raid5_remove_disk(mddev, d) == 0) { | |
5632 | char nm[20]; | |
5633 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5634 | sysfs_remove_link(&mddev->kobj, nm); | |
5635 | rdev->raid_disk = -1; | |
5636 | } | |
5637 | } | |
cea9c228 | 5638 | } |
88ce4930 | 5639 | mddev->layout = conf->algorithm; |
09c9e5fa | 5640 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5641 | mddev->reshape_position = MaxSector; |
5642 | mddev->delta_disks = 0; | |
cea9c228 N |
5643 | } |
5644 | } | |
5645 | ||
72626685 N |
5646 | static void raid5_quiesce(mddev_t *mddev, int state) |
5647 | { | |
070ec55d | 5648 | raid5_conf_t *conf = mddev->private; |
72626685 N |
5649 | |
5650 | switch(state) { | |
e464eafd N |
5651 | case 2: /* resume for a suspend */ |
5652 | wake_up(&conf->wait_for_overlap); | |
5653 | break; | |
5654 | ||
72626685 N |
5655 | case 1: /* stop all writes */ |
5656 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5657 | /* '2' tells resync/reshape to pause so that all |
5658 | * active stripes can drain | |
5659 | */ | |
5660 | conf->quiesce = 2; | |
72626685 | 5661 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5662 | atomic_read(&conf->active_stripes) == 0 && |
5663 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5664 | conf->device_lock, /* nothing */); |
64bd660b | 5665 | conf->quiesce = 1; |
72626685 | 5666 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5667 | /* allow reshape to continue */ |
5668 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5669 | break; |
5670 | ||
5671 | case 0: /* re-enable writes */ | |
5672 | spin_lock_irq(&conf->device_lock); | |
5673 | conf->quiesce = 0; | |
5674 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5675 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5676 | spin_unlock_irq(&conf->device_lock); |
5677 | break; | |
5678 | } | |
72626685 | 5679 | } |
b15c2e57 | 5680 | |
d562b0c4 | 5681 | |
f1b29bca | 5682 | static void *raid45_takeover_raid0(mddev_t *mddev, int level) |
54071b38 | 5683 | { |
f1b29bca | 5684 | struct raid0_private_data *raid0_priv = mddev->private; |
d76c8420 | 5685 | sector_t sectors; |
54071b38 | 5686 | |
f1b29bca DW |
5687 | /* for raid0 takeover only one zone is supported */ |
5688 | if (raid0_priv->nr_strip_zones > 1) { | |
0c55e022 N |
5689 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5690 | mdname(mddev)); | |
f1b29bca DW |
5691 | return ERR_PTR(-EINVAL); |
5692 | } | |
5693 | ||
3b71bd93 N |
5694 | sectors = raid0_priv->strip_zone[0].zone_end; |
5695 | sector_div(sectors, raid0_priv->strip_zone[0].nb_dev); | |
5696 | mddev->dev_sectors = sectors; | |
f1b29bca | 5697 | mddev->new_level = level; |
54071b38 TM |
5698 | mddev->new_layout = ALGORITHM_PARITY_N; |
5699 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5700 | mddev->raid_disks += 1; | |
5701 | mddev->delta_disks = 1; | |
5702 | /* make sure it will be not marked as dirty */ | |
5703 | mddev->recovery_cp = MaxSector; | |
5704 | ||
5705 | return setup_conf(mddev); | |
5706 | } | |
5707 | ||
5708 | ||
d562b0c4 N |
5709 | static void *raid5_takeover_raid1(mddev_t *mddev) |
5710 | { | |
5711 | int chunksect; | |
5712 | ||
5713 | if (mddev->raid_disks != 2 || | |
5714 | mddev->degraded > 1) | |
5715 | return ERR_PTR(-EINVAL); | |
5716 | ||
5717 | /* Should check if there are write-behind devices? */ | |
5718 | ||
5719 | chunksect = 64*2; /* 64K by default */ | |
5720 | ||
5721 | /* The array must be an exact multiple of chunksize */ | |
5722 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5723 | chunksect >>= 1; | |
5724 | ||
5725 | if ((chunksect<<9) < STRIPE_SIZE) | |
5726 | /* array size does not allow a suitable chunk size */ | |
5727 | return ERR_PTR(-EINVAL); | |
5728 | ||
5729 | mddev->new_level = 5; | |
5730 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5731 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5732 | |
5733 | return setup_conf(mddev); | |
5734 | } | |
5735 | ||
fc9739c6 N |
5736 | static void *raid5_takeover_raid6(mddev_t *mddev) |
5737 | { | |
5738 | int new_layout; | |
5739 | ||
5740 | switch (mddev->layout) { | |
5741 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5742 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5743 | break; | |
5744 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5745 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5746 | break; | |
5747 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5748 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5749 | break; | |
5750 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5751 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5752 | break; | |
5753 | case ALGORITHM_PARITY_0_6: | |
5754 | new_layout = ALGORITHM_PARITY_0; | |
5755 | break; | |
5756 | case ALGORITHM_PARITY_N: | |
5757 | new_layout = ALGORITHM_PARITY_N; | |
5758 | break; | |
5759 | default: | |
5760 | return ERR_PTR(-EINVAL); | |
5761 | } | |
5762 | mddev->new_level = 5; | |
5763 | mddev->new_layout = new_layout; | |
5764 | mddev->delta_disks = -1; | |
5765 | mddev->raid_disks -= 1; | |
5766 | return setup_conf(mddev); | |
5767 | } | |
5768 | ||
d562b0c4 | 5769 | |
50ac168a | 5770 | static int raid5_check_reshape(mddev_t *mddev) |
b3546035 | 5771 | { |
88ce4930 N |
5772 | /* For a 2-drive array, the layout and chunk size can be changed |
5773 | * immediately as not restriping is needed. | |
5774 | * For larger arrays we record the new value - after validation | |
5775 | * to be used by a reshape pass. | |
b3546035 | 5776 | */ |
070ec55d | 5777 | raid5_conf_t *conf = mddev->private; |
597a711b | 5778 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5779 | |
597a711b | 5780 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5781 | return -EINVAL; |
5782 | if (new_chunk > 0) { | |
0ba459d2 | 5783 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5784 | return -EINVAL; |
597a711b | 5785 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5786 | return -EINVAL; |
597a711b | 5787 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5788 | /* not factor of array size */ |
5789 | return -EINVAL; | |
5790 | } | |
5791 | ||
5792 | /* They look valid */ | |
5793 | ||
88ce4930 | 5794 | if (mddev->raid_disks == 2) { |
597a711b N |
5795 | /* can make the change immediately */ |
5796 | if (mddev->new_layout >= 0) { | |
5797 | conf->algorithm = mddev->new_layout; | |
5798 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5799 | } |
5800 | if (new_chunk > 0) { | |
597a711b N |
5801 | conf->chunk_sectors = new_chunk ; |
5802 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5803 | } |
5804 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5805 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5806 | } |
50ac168a | 5807 | return check_reshape(mddev); |
88ce4930 N |
5808 | } |
5809 | ||
50ac168a | 5810 | static int raid6_check_reshape(mddev_t *mddev) |
88ce4930 | 5811 | { |
597a711b | 5812 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5813 | |
597a711b | 5814 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5815 | return -EINVAL; |
b3546035 | 5816 | if (new_chunk > 0) { |
0ba459d2 | 5817 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5818 | return -EINVAL; |
597a711b | 5819 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5820 | return -EINVAL; |
597a711b | 5821 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5822 | /* not factor of array size */ |
5823 | return -EINVAL; | |
b3546035 | 5824 | } |
88ce4930 N |
5825 | |
5826 | /* They look valid */ | |
50ac168a | 5827 | return check_reshape(mddev); |
b3546035 N |
5828 | } |
5829 | ||
d562b0c4 N |
5830 | static void *raid5_takeover(mddev_t *mddev) |
5831 | { | |
5832 | /* raid5 can take over: | |
f1b29bca | 5833 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
5834 | * raid1 - if there are two drives. We need to know the chunk size |
5835 | * raid4 - trivial - just use a raid4 layout. | |
5836 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 5837 | */ |
f1b29bca DW |
5838 | if (mddev->level == 0) |
5839 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
5840 | if (mddev->level == 1) |
5841 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5842 | if (mddev->level == 4) { |
5843 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5844 | mddev->new_level = 5; | |
5845 | return setup_conf(mddev); | |
5846 | } | |
fc9739c6 N |
5847 | if (mddev->level == 6) |
5848 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5849 | |
5850 | return ERR_PTR(-EINVAL); | |
5851 | } | |
5852 | ||
a78d38a1 N |
5853 | static void *raid4_takeover(mddev_t *mddev) |
5854 | { | |
f1b29bca DW |
5855 | /* raid4 can take over: |
5856 | * raid0 - if there is only one strip zone | |
5857 | * raid5 - if layout is right | |
a78d38a1 | 5858 | */ |
f1b29bca DW |
5859 | if (mddev->level == 0) |
5860 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
5861 | if (mddev->level == 5 && |
5862 | mddev->layout == ALGORITHM_PARITY_N) { | |
5863 | mddev->new_layout = 0; | |
5864 | mddev->new_level = 4; | |
5865 | return setup_conf(mddev); | |
5866 | } | |
5867 | return ERR_PTR(-EINVAL); | |
5868 | } | |
d562b0c4 | 5869 | |
245f46c2 N |
5870 | static struct mdk_personality raid5_personality; |
5871 | ||
5872 | static void *raid6_takeover(mddev_t *mddev) | |
5873 | { | |
5874 | /* Currently can only take over a raid5. We map the | |
5875 | * personality to an equivalent raid6 personality | |
5876 | * with the Q block at the end. | |
5877 | */ | |
5878 | int new_layout; | |
5879 | ||
5880 | if (mddev->pers != &raid5_personality) | |
5881 | return ERR_PTR(-EINVAL); | |
5882 | if (mddev->degraded > 1) | |
5883 | return ERR_PTR(-EINVAL); | |
5884 | if (mddev->raid_disks > 253) | |
5885 | return ERR_PTR(-EINVAL); | |
5886 | if (mddev->raid_disks < 3) | |
5887 | return ERR_PTR(-EINVAL); | |
5888 | ||
5889 | switch (mddev->layout) { | |
5890 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5891 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5892 | break; | |
5893 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5894 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5895 | break; | |
5896 | case ALGORITHM_LEFT_SYMMETRIC: | |
5897 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5898 | break; | |
5899 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5900 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5901 | break; | |
5902 | case ALGORITHM_PARITY_0: | |
5903 | new_layout = ALGORITHM_PARITY_0_6; | |
5904 | break; | |
5905 | case ALGORITHM_PARITY_N: | |
5906 | new_layout = ALGORITHM_PARITY_N; | |
5907 | break; | |
5908 | default: | |
5909 | return ERR_PTR(-EINVAL); | |
5910 | } | |
5911 | mddev->new_level = 6; | |
5912 | mddev->new_layout = new_layout; | |
5913 | mddev->delta_disks = 1; | |
5914 | mddev->raid_disks += 1; | |
5915 | return setup_conf(mddev); | |
5916 | } | |
5917 | ||
5918 | ||
16a53ecc N |
5919 | static struct mdk_personality raid6_personality = |
5920 | { | |
5921 | .name = "raid6", | |
5922 | .level = 6, | |
5923 | .owner = THIS_MODULE, | |
5924 | .make_request = make_request, | |
5925 | .run = run, | |
5926 | .stop = stop, | |
5927 | .status = status, | |
5928 | .error_handler = error, | |
5929 | .hot_add_disk = raid5_add_disk, | |
5930 | .hot_remove_disk= raid5_remove_disk, | |
5931 | .spare_active = raid5_spare_active, | |
5932 | .sync_request = sync_request, | |
5933 | .resize = raid5_resize, | |
80c3a6ce | 5934 | .size = raid5_size, |
50ac168a | 5935 | .check_reshape = raid6_check_reshape, |
f416885e | 5936 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5937 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5938 | .quiesce = raid5_quiesce, |
245f46c2 | 5939 | .takeover = raid6_takeover, |
16a53ecc | 5940 | }; |
2604b703 | 5941 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
5942 | { |
5943 | .name = "raid5", | |
2604b703 | 5944 | .level = 5, |
1da177e4 LT |
5945 | .owner = THIS_MODULE, |
5946 | .make_request = make_request, | |
5947 | .run = run, | |
5948 | .stop = stop, | |
5949 | .status = status, | |
5950 | .error_handler = error, | |
5951 | .hot_add_disk = raid5_add_disk, | |
5952 | .hot_remove_disk= raid5_remove_disk, | |
5953 | .spare_active = raid5_spare_active, | |
5954 | .sync_request = sync_request, | |
5955 | .resize = raid5_resize, | |
80c3a6ce | 5956 | .size = raid5_size, |
63c70c4f N |
5957 | .check_reshape = raid5_check_reshape, |
5958 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5959 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5960 | .quiesce = raid5_quiesce, |
d562b0c4 | 5961 | .takeover = raid5_takeover, |
1da177e4 LT |
5962 | }; |
5963 | ||
2604b703 | 5964 | static struct mdk_personality raid4_personality = |
1da177e4 | 5965 | { |
2604b703 N |
5966 | .name = "raid4", |
5967 | .level = 4, | |
5968 | .owner = THIS_MODULE, | |
5969 | .make_request = make_request, | |
5970 | .run = run, | |
5971 | .stop = stop, | |
5972 | .status = status, | |
5973 | .error_handler = error, | |
5974 | .hot_add_disk = raid5_add_disk, | |
5975 | .hot_remove_disk= raid5_remove_disk, | |
5976 | .spare_active = raid5_spare_active, | |
5977 | .sync_request = sync_request, | |
5978 | .resize = raid5_resize, | |
80c3a6ce | 5979 | .size = raid5_size, |
3d37890b N |
5980 | .check_reshape = raid5_check_reshape, |
5981 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5982 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 5983 | .quiesce = raid5_quiesce, |
a78d38a1 | 5984 | .takeover = raid4_takeover, |
2604b703 N |
5985 | }; |
5986 | ||
5987 | static int __init raid5_init(void) | |
5988 | { | |
16a53ecc | 5989 | register_md_personality(&raid6_personality); |
2604b703 N |
5990 | register_md_personality(&raid5_personality); |
5991 | register_md_personality(&raid4_personality); | |
5992 | return 0; | |
1da177e4 LT |
5993 | } |
5994 | ||
2604b703 | 5995 | static void raid5_exit(void) |
1da177e4 | 5996 | { |
16a53ecc | 5997 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5998 | unregister_md_personality(&raid5_personality); |
5999 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
6000 | } |
6001 | ||
6002 | module_init(raid5_init); | |
6003 | module_exit(raid5_exit); | |
6004 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 6005 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 6006 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
6007 | MODULE_ALIAS("md-raid5"); |
6008 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
6009 | MODULE_ALIAS("md-level-5"); |
6010 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
6011 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
6012 | MODULE_ALIAS("md-raid6"); | |
6013 | MODULE_ALIAS("md-level-6"); | |
6014 | ||
6015 | /* This used to be two separate modules, they were: */ | |
6016 | MODULE_ALIAS("raid5"); | |
6017 | MODULE_ALIAS("raid6"); |