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 | ||
cbe47ec5 | 1023 | spin_lock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1024 | chosen = dev->towrite; |
1025 | dev->towrite = NULL; | |
1026 | BUG_ON(dev->written); | |
1027 | wbi = dev->written = chosen; | |
cbe47ec5 | 1028 | spin_unlock_irq(&sh->raid_conf->device_lock); |
91c00924 DW |
1029 | |
1030 | while (wbi && wbi->bi_sector < | |
1031 | dev->sector + STRIPE_SECTORS) { | |
e9c7469b TH |
1032 | if (wbi->bi_rw & REQ_FUA) |
1033 | set_bit(R5_WantFUA, &dev->flags); | |
91c00924 DW |
1034 | tx = async_copy_data(1, wbi, dev->page, |
1035 | dev->sector, tx); | |
1036 | wbi = r5_next_bio(wbi, dev->sector); | |
1037 | } | |
1038 | } | |
1039 | } | |
1040 | ||
1041 | return tx; | |
1042 | } | |
1043 | ||
ac6b53b6 | 1044 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1045 | { |
1046 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1047 | int disks = sh->disks; |
1048 | int pd_idx = sh->pd_idx; | |
1049 | int qd_idx = sh->qd_idx; | |
1050 | int i; | |
e9c7469b | 1051 | bool fua = false; |
91c00924 | 1052 | |
e46b272b | 1053 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1054 | (unsigned long long)sh->sector); |
1055 | ||
e9c7469b TH |
1056 | for (i = disks; i--; ) |
1057 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); | |
1058 | ||
91c00924 DW |
1059 | for (i = disks; i--; ) { |
1060 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1061 | |
e9c7469b | 1062 | if (dev->written || i == pd_idx || i == qd_idx) { |
91c00924 | 1063 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1064 | if (fua) |
1065 | set_bit(R5_WantFUA, &dev->flags); | |
1066 | } | |
91c00924 DW |
1067 | } |
1068 | ||
d8ee0728 DW |
1069 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1070 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1071 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1072 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1073 | else { | |
1074 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1075 | sh->reconstruct_state = reconstruct_state_result; | |
1076 | } | |
91c00924 DW |
1077 | |
1078 | set_bit(STRIPE_HANDLE, &sh->state); | |
1079 | release_stripe(sh); | |
1080 | } | |
1081 | ||
1082 | static void | |
ac6b53b6 DW |
1083 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1084 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1085 | { |
91c00924 | 1086 | int disks = sh->disks; |
d6f38f31 | 1087 | struct page **xor_srcs = percpu->scribble; |
a08abd8c | 1088 | struct async_submit_ctl submit; |
91c00924 DW |
1089 | int count = 0, pd_idx = sh->pd_idx, i; |
1090 | struct page *xor_dest; | |
d8ee0728 | 1091 | int prexor = 0; |
91c00924 | 1092 | unsigned long flags; |
91c00924 | 1093 | |
e46b272b | 1094 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1095 | (unsigned long long)sh->sector); |
1096 | ||
1097 | /* check if prexor is active which means only process blocks | |
1098 | * that are part of a read-modify-write (written) | |
1099 | */ | |
d8ee0728 DW |
1100 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1101 | prexor = 1; | |
91c00924 DW |
1102 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1103 | for (i = disks; i--; ) { | |
1104 | struct r5dev *dev = &sh->dev[i]; | |
1105 | if (dev->written) | |
1106 | xor_srcs[count++] = dev->page; | |
1107 | } | |
1108 | } else { | |
1109 | xor_dest = sh->dev[pd_idx].page; | |
1110 | for (i = disks; i--; ) { | |
1111 | struct r5dev *dev = &sh->dev[i]; | |
1112 | if (i != pd_idx) | |
1113 | xor_srcs[count++] = dev->page; | |
1114 | } | |
1115 | } | |
1116 | ||
91c00924 DW |
1117 | /* 1/ if we prexor'd then the dest is reused as a source |
1118 | * 2/ if we did not prexor then we are redoing the parity | |
1119 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1120 | * for the synchronous xor case | |
1121 | */ | |
88ba2aa5 | 1122 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1123 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1124 | ||
1125 | atomic_inc(&sh->count); | |
1126 | ||
ac6b53b6 | 1127 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
d6f38f31 | 1128 | to_addr_conv(sh, percpu)); |
a08abd8c DW |
1129 | if (unlikely(count == 1)) |
1130 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1131 | else | |
1132 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1133 | } |
1134 | ||
ac6b53b6 DW |
1135 | static void |
1136 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1137 | struct dma_async_tx_descriptor *tx) | |
1138 | { | |
1139 | struct async_submit_ctl submit; | |
1140 | struct page **blocks = percpu->scribble; | |
1141 | int count; | |
1142 | ||
1143 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1144 | ||
1145 | count = set_syndrome_sources(blocks, sh); | |
1146 | ||
1147 | atomic_inc(&sh->count); | |
1148 | ||
1149 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
1150 | sh, to_addr_conv(sh, percpu)); | |
1151 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
91c00924 DW |
1152 | } |
1153 | ||
1154 | static void ops_complete_check(void *stripe_head_ref) | |
1155 | { | |
1156 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1157 | |
e46b272b | 1158 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1159 | (unsigned long long)sh->sector); |
1160 | ||
ecc65c9b | 1161 | sh->check_state = check_state_check_result; |
91c00924 DW |
1162 | set_bit(STRIPE_HANDLE, &sh->state); |
1163 | release_stripe(sh); | |
1164 | } | |
1165 | ||
ac6b53b6 | 1166 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1167 | { |
91c00924 | 1168 | int disks = sh->disks; |
ac6b53b6 DW |
1169 | int pd_idx = sh->pd_idx; |
1170 | int qd_idx = sh->qd_idx; | |
1171 | struct page *xor_dest; | |
d6f38f31 | 1172 | struct page **xor_srcs = percpu->scribble; |
91c00924 | 1173 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1174 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1175 | int count; |
1176 | int i; | |
91c00924 | 1177 | |
e46b272b | 1178 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1179 | (unsigned long long)sh->sector); |
1180 | ||
ac6b53b6 DW |
1181 | count = 0; |
1182 | xor_dest = sh->dev[pd_idx].page; | |
1183 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1184 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1185 | if (i == pd_idx || i == qd_idx) |
1186 | continue; | |
1187 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1188 | } |
1189 | ||
d6f38f31 DW |
1190 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
1191 | to_addr_conv(sh, percpu)); | |
099f53cb | 1192 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1193 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1194 | |
91c00924 | 1195 | atomic_inc(&sh->count); |
a08abd8c DW |
1196 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1197 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1198 | } |
1199 | ||
ac6b53b6 DW |
1200 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1201 | { | |
1202 | struct page **srcs = percpu->scribble; | |
1203 | struct async_submit_ctl submit; | |
1204 | int count; | |
1205 | ||
1206 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1207 | (unsigned long long)sh->sector, checkp); | |
1208 | ||
1209 | count = set_syndrome_sources(srcs, sh); | |
1210 | if (!checkp) | |
1211 | srcs[count] = NULL; | |
91c00924 | 1212 | |
91c00924 | 1213 | atomic_inc(&sh->count); |
ac6b53b6 DW |
1214 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
1215 | sh, to_addr_conv(sh, percpu)); | |
1216 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, | |
1217 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1218 | } |
1219 | ||
417b8d4a | 1220 | static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1221 | { |
1222 | int overlap_clear = 0, i, disks = sh->disks; | |
1223 | struct dma_async_tx_descriptor *tx = NULL; | |
d6f38f31 | 1224 | raid5_conf_t *conf = sh->raid_conf; |
ac6b53b6 | 1225 | int level = conf->level; |
d6f38f31 DW |
1226 | struct raid5_percpu *percpu; |
1227 | unsigned long cpu; | |
91c00924 | 1228 | |
d6f38f31 DW |
1229 | cpu = get_cpu(); |
1230 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1231 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1232 | ops_run_biofill(sh); |
1233 | overlap_clear++; | |
1234 | } | |
1235 | ||
7b3a871e | 1236 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1237 | if (level < 6) |
1238 | tx = ops_run_compute5(sh, percpu); | |
1239 | else { | |
1240 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1241 | tx = ops_run_compute6_1(sh, percpu); | |
1242 | else | |
1243 | tx = ops_run_compute6_2(sh, percpu); | |
1244 | } | |
1245 | /* terminate the chain if reconstruct is not set to be run */ | |
1246 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1247 | async_tx_ack(tx); |
1248 | } | |
91c00924 | 1249 | |
600aa109 | 1250 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1251 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1252 | |
600aa109 | 1253 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1254 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1255 | overlap_clear++; |
1256 | } | |
1257 | ||
ac6b53b6 DW |
1258 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1259 | if (level < 6) | |
1260 | ops_run_reconstruct5(sh, percpu, tx); | |
1261 | else | |
1262 | ops_run_reconstruct6(sh, percpu, tx); | |
1263 | } | |
91c00924 | 1264 | |
ac6b53b6 DW |
1265 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1266 | if (sh->check_state == check_state_run) | |
1267 | ops_run_check_p(sh, percpu); | |
1268 | else if (sh->check_state == check_state_run_q) | |
1269 | ops_run_check_pq(sh, percpu, 0); | |
1270 | else if (sh->check_state == check_state_run_pq) | |
1271 | ops_run_check_pq(sh, percpu, 1); | |
1272 | else | |
1273 | BUG(); | |
1274 | } | |
91c00924 | 1275 | |
91c00924 DW |
1276 | if (overlap_clear) |
1277 | for (i = disks; i--; ) { | |
1278 | struct r5dev *dev = &sh->dev[i]; | |
1279 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1280 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1281 | } | |
d6f38f31 | 1282 | put_cpu(); |
91c00924 DW |
1283 | } |
1284 | ||
417b8d4a DW |
1285 | #ifdef CONFIG_MULTICORE_RAID456 |
1286 | static void async_run_ops(void *param, async_cookie_t cookie) | |
1287 | { | |
1288 | struct stripe_head *sh = param; | |
1289 | unsigned long ops_request = sh->ops.request; | |
1290 | ||
1291 | clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state); | |
1292 | wake_up(&sh->ops.wait_for_ops); | |
1293 | ||
1294 | __raid_run_ops(sh, ops_request); | |
1295 | release_stripe(sh); | |
1296 | } | |
1297 | ||
1298 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) | |
1299 | { | |
1300 | /* since handle_stripe can be called outside of raid5d context | |
1301 | * we need to ensure sh->ops.request is de-staged before another | |
1302 | * request arrives | |
1303 | */ | |
1304 | wait_event(sh->ops.wait_for_ops, | |
1305 | !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state)); | |
1306 | sh->ops.request = ops_request; | |
1307 | ||
1308 | atomic_inc(&sh->count); | |
1309 | async_schedule(async_run_ops, sh); | |
1310 | } | |
1311 | #else | |
1312 | #define raid_run_ops __raid_run_ops | |
1313 | #endif | |
1314 | ||
3f294f4f | 1315 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1316 | { |
1317 | struct stripe_head *sh; | |
6ce32846 | 1318 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1319 | if (!sh) |
1320 | return 0; | |
6ce32846 | 1321 | |
3f294f4f | 1322 | sh->raid_conf = conf; |
417b8d4a DW |
1323 | #ifdef CONFIG_MULTICORE_RAID456 |
1324 | init_waitqueue_head(&sh->ops.wait_for_ops); | |
1325 | #endif | |
3f294f4f | 1326 | |
e4e11e38 N |
1327 | if (grow_buffers(sh)) { |
1328 | shrink_buffers(sh); | |
3f294f4f N |
1329 | kmem_cache_free(conf->slab_cache, sh); |
1330 | return 0; | |
1331 | } | |
1332 | /* we just created an active stripe so... */ | |
1333 | atomic_set(&sh->count, 1); | |
1334 | atomic_inc(&conf->active_stripes); | |
1335 | INIT_LIST_HEAD(&sh->lru); | |
1336 | release_stripe(sh); | |
1337 | return 1; | |
1338 | } | |
1339 | ||
1340 | static int grow_stripes(raid5_conf_t *conf, int num) | |
1341 | { | |
e18b890b | 1342 | struct kmem_cache *sc; |
5e5e3e78 | 1343 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 1344 | |
f4be6b43 N |
1345 | if (conf->mddev->gendisk) |
1346 | sprintf(conf->cache_name[0], | |
1347 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1348 | else | |
1349 | sprintf(conf->cache_name[0], | |
1350 | "raid%d-%p", conf->level, conf->mddev); | |
1351 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1352 | ||
ad01c9e3 N |
1353 | conf->active_name = 0; |
1354 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1355 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1356 | 0, 0, NULL); |
1da177e4 LT |
1357 | if (!sc) |
1358 | return 1; | |
1359 | conf->slab_cache = sc; | |
ad01c9e3 | 1360 | conf->pool_size = devs; |
16a53ecc | 1361 | while (num--) |
3f294f4f | 1362 | if (!grow_one_stripe(conf)) |
1da177e4 | 1363 | return 1; |
1da177e4 LT |
1364 | return 0; |
1365 | } | |
29269553 | 1366 | |
d6f38f31 DW |
1367 | /** |
1368 | * scribble_len - return the required size of the scribble region | |
1369 | * @num - total number of disks in the array | |
1370 | * | |
1371 | * The size must be enough to contain: | |
1372 | * 1/ a struct page pointer for each device in the array +2 | |
1373 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1374 | * (dma_map_page()) or page (page_address()) address. | |
1375 | * | |
1376 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1377 | * calculate over all devices (not just the data blocks), using zeros in place | |
1378 | * of the P and Q blocks. | |
1379 | */ | |
1380 | static size_t scribble_len(int num) | |
1381 | { | |
1382 | size_t len; | |
1383 | ||
1384 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
1385 | ||
1386 | return len; | |
1387 | } | |
1388 | ||
ad01c9e3 N |
1389 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
1390 | { | |
1391 | /* Make all the stripes able to hold 'newsize' devices. | |
1392 | * New slots in each stripe get 'page' set to a new page. | |
1393 | * | |
1394 | * This happens in stages: | |
1395 | * 1/ create a new kmem_cache and allocate the required number of | |
1396 | * stripe_heads. | |
1397 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
1398 | * to the new stripe_heads. This will have the side effect of | |
1399 | * freezing the array as once all stripe_heads have been collected, | |
1400 | * no IO will be possible. Old stripe heads are freed once their | |
1401 | * pages have been transferred over, and the old kmem_cache is | |
1402 | * freed when all stripes are done. | |
1403 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1404 | * we simple return a failre status - no need to clean anything up. | |
1405 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1406 | * If this fails, we don't bother trying the shrink the | |
1407 | * stripe_heads down again, we just leave them as they are. | |
1408 | * As each stripe_head is processed the new one is released into | |
1409 | * active service. | |
1410 | * | |
1411 | * Once step2 is started, we cannot afford to wait for a write, | |
1412 | * so we use GFP_NOIO allocations. | |
1413 | */ | |
1414 | struct stripe_head *osh, *nsh; | |
1415 | LIST_HEAD(newstripes); | |
1416 | struct disk_info *ndisks; | |
d6f38f31 | 1417 | unsigned long cpu; |
b5470dc5 | 1418 | int err; |
e18b890b | 1419 | struct kmem_cache *sc; |
ad01c9e3 N |
1420 | int i; |
1421 | ||
1422 | if (newsize <= conf->pool_size) | |
1423 | return 0; /* never bother to shrink */ | |
1424 | ||
b5470dc5 DW |
1425 | err = md_allow_write(conf->mddev); |
1426 | if (err) | |
1427 | return err; | |
2a2275d6 | 1428 | |
ad01c9e3 N |
1429 | /* Step 1 */ |
1430 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1431 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1432 | 0, 0, NULL); |
ad01c9e3 N |
1433 | if (!sc) |
1434 | return -ENOMEM; | |
1435 | ||
1436 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1437 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1438 | if (!nsh) |
1439 | break; | |
1440 | ||
ad01c9e3 | 1441 | nsh->raid_conf = conf; |
417b8d4a DW |
1442 | #ifdef CONFIG_MULTICORE_RAID456 |
1443 | init_waitqueue_head(&nsh->ops.wait_for_ops); | |
1444 | #endif | |
ad01c9e3 N |
1445 | |
1446 | list_add(&nsh->lru, &newstripes); | |
1447 | } | |
1448 | if (i) { | |
1449 | /* didn't get enough, give up */ | |
1450 | while (!list_empty(&newstripes)) { | |
1451 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1452 | list_del(&nsh->lru); | |
1453 | kmem_cache_free(sc, nsh); | |
1454 | } | |
1455 | kmem_cache_destroy(sc); | |
1456 | return -ENOMEM; | |
1457 | } | |
1458 | /* Step 2 - Must use GFP_NOIO now. | |
1459 | * OK, we have enough stripes, start collecting inactive | |
1460 | * stripes and copying them over | |
1461 | */ | |
1462 | list_for_each_entry(nsh, &newstripes, lru) { | |
1463 | spin_lock_irq(&conf->device_lock); | |
1464 | wait_event_lock_irq(conf->wait_for_stripe, | |
1465 | !list_empty(&conf->inactive_list), | |
1466 | conf->device_lock, | |
482c0834 | 1467 | ); |
ad01c9e3 N |
1468 | osh = get_free_stripe(conf); |
1469 | spin_unlock_irq(&conf->device_lock); | |
1470 | atomic_set(&nsh->count, 1); | |
1471 | for(i=0; i<conf->pool_size; i++) | |
1472 | nsh->dev[i].page = osh->dev[i].page; | |
1473 | for( ; i<newsize; i++) | |
1474 | nsh->dev[i].page = NULL; | |
1475 | kmem_cache_free(conf->slab_cache, osh); | |
1476 | } | |
1477 | kmem_cache_destroy(conf->slab_cache); | |
1478 | ||
1479 | /* Step 3. | |
1480 | * At this point, we are holding all the stripes so the array | |
1481 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1482 | * conf->disks and the scribble region |
ad01c9e3 N |
1483 | */ |
1484 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1485 | if (ndisks) { | |
1486 | for (i=0; i<conf->raid_disks; i++) | |
1487 | ndisks[i] = conf->disks[i]; | |
1488 | kfree(conf->disks); | |
1489 | conf->disks = ndisks; | |
1490 | } else | |
1491 | err = -ENOMEM; | |
1492 | ||
d6f38f31 DW |
1493 | get_online_cpus(); |
1494 | conf->scribble_len = scribble_len(newsize); | |
1495 | for_each_present_cpu(cpu) { | |
1496 | struct raid5_percpu *percpu; | |
1497 | void *scribble; | |
1498 | ||
1499 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
1500 | scribble = kmalloc(conf->scribble_len, GFP_NOIO); | |
1501 | ||
1502 | if (scribble) { | |
1503 | kfree(percpu->scribble); | |
1504 | percpu->scribble = scribble; | |
1505 | } else { | |
1506 | err = -ENOMEM; | |
1507 | break; | |
1508 | } | |
1509 | } | |
1510 | put_online_cpus(); | |
1511 | ||
ad01c9e3 N |
1512 | /* Step 4, return new stripes to service */ |
1513 | while(!list_empty(&newstripes)) { | |
1514 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1515 | list_del_init(&nsh->lru); | |
d6f38f31 | 1516 | |
ad01c9e3 N |
1517 | for (i=conf->raid_disks; i < newsize; i++) |
1518 | if (nsh->dev[i].page == NULL) { | |
1519 | struct page *p = alloc_page(GFP_NOIO); | |
1520 | nsh->dev[i].page = p; | |
1521 | if (!p) | |
1522 | err = -ENOMEM; | |
1523 | } | |
1524 | release_stripe(nsh); | |
1525 | } | |
1526 | /* critical section pass, GFP_NOIO no longer needed */ | |
1527 | ||
1528 | conf->slab_cache = sc; | |
1529 | conf->active_name = 1-conf->active_name; | |
1530 | conf->pool_size = newsize; | |
1531 | return err; | |
1532 | } | |
1da177e4 | 1533 | |
3f294f4f | 1534 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1535 | { |
1536 | struct stripe_head *sh; | |
1537 | ||
3f294f4f N |
1538 | spin_lock_irq(&conf->device_lock); |
1539 | sh = get_free_stripe(conf); | |
1540 | spin_unlock_irq(&conf->device_lock); | |
1541 | if (!sh) | |
1542 | return 0; | |
78bafebd | 1543 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1544 | shrink_buffers(sh); |
3f294f4f N |
1545 | kmem_cache_free(conf->slab_cache, sh); |
1546 | atomic_dec(&conf->active_stripes); | |
1547 | return 1; | |
1548 | } | |
1549 | ||
1550 | static void shrink_stripes(raid5_conf_t *conf) | |
1551 | { | |
1552 | while (drop_one_stripe(conf)) | |
1553 | ; | |
1554 | ||
29fc7e3e N |
1555 | if (conf->slab_cache) |
1556 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1557 | conf->slab_cache = NULL; |
1558 | } | |
1559 | ||
6712ecf8 | 1560 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1561 | { |
99c0fb5f | 1562 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1563 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1564 | int disks = sh->disks, i; |
1da177e4 | 1565 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1566 | char b[BDEVNAME_SIZE]; |
1567 | mdk_rdev_t *rdev; | |
1da177e4 | 1568 | |
1da177e4 LT |
1569 | |
1570 | for (i=0 ; i<disks; i++) | |
1571 | if (bi == &sh->dev[i].req) | |
1572 | break; | |
1573 | ||
45b4233c DW |
1574 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1575 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1576 | uptodate); |
1577 | if (i == disks) { | |
1578 | BUG(); | |
6712ecf8 | 1579 | return; |
1da177e4 LT |
1580 | } |
1581 | ||
1582 | if (uptodate) { | |
1da177e4 | 1583 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1584 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1585 | rdev = conf->disks[i].rdev; |
0c55e022 | 1586 | printk_rl(KERN_INFO "md/raid:%s: read error corrected" |
6be9d494 BS |
1587 | " (%lu sectors at %llu on %s)\n", |
1588 | mdname(conf->mddev), STRIPE_SECTORS, | |
1589 | (unsigned long long)(sh->sector | |
1590 | + rdev->data_offset), | |
1591 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1592 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1593 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1594 | } | |
ba22dcbf N |
1595 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1596 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1597 | } else { |
d6950432 | 1598 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1599 | int retry = 0; |
d6950432 N |
1600 | rdev = conf->disks[i].rdev; |
1601 | ||
1da177e4 | 1602 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1603 | atomic_inc(&rdev->read_errors); |
7b0bb536 | 1604 | if (conf->mddev->degraded >= conf->max_degraded) |
6be9d494 | 1605 | printk_rl(KERN_WARNING |
0c55e022 | 1606 | "md/raid:%s: read error not correctable " |
6be9d494 BS |
1607 | "(sector %llu on %s).\n", |
1608 | mdname(conf->mddev), | |
1609 | (unsigned long long)(sh->sector | |
1610 | + rdev->data_offset), | |
1611 | bdn); | |
ba22dcbf | 1612 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1613 | /* Oh, no!!! */ |
6be9d494 | 1614 | printk_rl(KERN_WARNING |
0c55e022 | 1615 | "md/raid:%s: read error NOT corrected!! " |
6be9d494 BS |
1616 | "(sector %llu on %s).\n", |
1617 | mdname(conf->mddev), | |
1618 | (unsigned long long)(sh->sector | |
1619 | + rdev->data_offset), | |
1620 | bdn); | |
d6950432 | 1621 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1622 | > conf->max_nr_stripes) |
14f8d26b | 1623 | printk(KERN_WARNING |
0c55e022 | 1624 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 1625 | mdname(conf->mddev), bdn); |
ba22dcbf N |
1626 | else |
1627 | retry = 1; | |
1628 | if (retry) | |
1629 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1630 | else { | |
4e5314b5 N |
1631 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1632 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1633 | md_error(conf->mddev, rdev); |
ba22dcbf | 1634 | } |
1da177e4 LT |
1635 | } |
1636 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1637 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1638 | set_bit(STRIPE_HANDLE, &sh->state); | |
1639 | release_stripe(sh); | |
1da177e4 LT |
1640 | } |
1641 | ||
d710e138 | 1642 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1643 | { |
99c0fb5f | 1644 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1645 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1646 | int disks = sh->disks, i; |
1da177e4 LT |
1647 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1648 | ||
1da177e4 LT |
1649 | for (i=0 ; i<disks; i++) |
1650 | if (bi == &sh->dev[i].req) | |
1651 | break; | |
1652 | ||
45b4233c | 1653 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1654 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1655 | uptodate); | |
1656 | if (i == disks) { | |
1657 | BUG(); | |
6712ecf8 | 1658 | return; |
1da177e4 LT |
1659 | } |
1660 | ||
1da177e4 LT |
1661 | if (!uptodate) |
1662 | md_error(conf->mddev, conf->disks[i].rdev); | |
1663 | ||
1664 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1665 | ||
1666 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1667 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1668 | release_stripe(sh); |
1da177e4 LT |
1669 | } |
1670 | ||
1671 | ||
784052ec | 1672 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
1da177e4 | 1673 | |
784052ec | 1674 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1675 | { |
1676 | struct r5dev *dev = &sh->dev[i]; | |
1677 | ||
1678 | bio_init(&dev->req); | |
1679 | dev->req.bi_io_vec = &dev->vec; | |
1680 | dev->req.bi_vcnt++; | |
1681 | dev->req.bi_max_vecs++; | |
1682 | dev->vec.bv_page = dev->page; | |
1683 | dev->vec.bv_len = STRIPE_SIZE; | |
1684 | dev->vec.bv_offset = 0; | |
1685 | ||
1686 | dev->req.bi_sector = sh->sector; | |
1687 | dev->req.bi_private = sh; | |
1688 | ||
1689 | dev->flags = 0; | |
784052ec | 1690 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
1691 | } |
1692 | ||
1693 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1694 | { | |
1695 | char b[BDEVNAME_SIZE]; | |
7b92813c | 1696 | raid5_conf_t *conf = mddev->private; |
0c55e022 | 1697 | pr_debug("raid456: error called\n"); |
1da177e4 | 1698 | |
6f8d0c77 N |
1699 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1700 | unsigned long flags; | |
1701 | spin_lock_irqsave(&conf->device_lock, flags); | |
1702 | mddev->degraded++; | |
1703 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1704 | /* | |
1705 | * if recovery was running, make sure it aborts. | |
1706 | */ | |
1707 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
1da177e4 | 1708 | } |
6f8d0c77 N |
1709 | set_bit(Faulty, &rdev->flags); |
1710 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1711 | printk(KERN_ALERT | |
1712 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
1713 | "md/raid:%s: Operation continuing on %d devices.\n", | |
1714 | mdname(mddev), | |
1715 | bdevname(rdev->bdev, b), | |
1716 | mdname(mddev), | |
1717 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 1718 | } |
1da177e4 LT |
1719 | |
1720 | /* | |
1721 | * Input: a 'big' sector number, | |
1722 | * Output: index of the data and parity disk, and the sector # in them. | |
1723 | */ | |
112bf897 | 1724 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1725 | int previous, int *dd_idx, |
1726 | struct stripe_head *sh) | |
1da177e4 | 1727 | { |
6e3b96ed | 1728 | sector_t stripe, stripe2; |
35f2a591 | 1729 | sector_t chunk_number; |
1da177e4 | 1730 | unsigned int chunk_offset; |
911d4ee8 | 1731 | int pd_idx, qd_idx; |
67cc2b81 | 1732 | int ddf_layout = 0; |
1da177e4 | 1733 | sector_t new_sector; |
e183eaed N |
1734 | int algorithm = previous ? conf->prev_algo |
1735 | : conf->algorithm; | |
09c9e5fa AN |
1736 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1737 | : conf->chunk_sectors; | |
112bf897 N |
1738 | int raid_disks = previous ? conf->previous_raid_disks |
1739 | : conf->raid_disks; | |
1740 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1741 | |
1742 | /* First compute the information on this sector */ | |
1743 | ||
1744 | /* | |
1745 | * Compute the chunk number and the sector offset inside the chunk | |
1746 | */ | |
1747 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1748 | chunk_number = r_sector; | |
1da177e4 LT |
1749 | |
1750 | /* | |
1751 | * Compute the stripe number | |
1752 | */ | |
35f2a591 N |
1753 | stripe = chunk_number; |
1754 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 1755 | stripe2 = stripe; |
1da177e4 LT |
1756 | /* |
1757 | * Select the parity disk based on the user selected algorithm. | |
1758 | */ | |
911d4ee8 | 1759 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1760 | switch(conf->level) { |
1761 | case 4: | |
911d4ee8 | 1762 | pd_idx = data_disks; |
16a53ecc N |
1763 | break; |
1764 | case 5: | |
e183eaed | 1765 | switch (algorithm) { |
1da177e4 | 1766 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1767 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1768 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1769 | (*dd_idx)++; |
1770 | break; | |
1771 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1772 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1773 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
1774 | (*dd_idx)++; |
1775 | break; | |
1776 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1777 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 1778 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
1779 | break; |
1780 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1781 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 1782 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 1783 | break; |
99c0fb5f N |
1784 | case ALGORITHM_PARITY_0: |
1785 | pd_idx = 0; | |
1786 | (*dd_idx)++; | |
1787 | break; | |
1788 | case ALGORITHM_PARITY_N: | |
1789 | pd_idx = data_disks; | |
1790 | break; | |
1da177e4 | 1791 | default: |
99c0fb5f | 1792 | BUG(); |
16a53ecc N |
1793 | } |
1794 | break; | |
1795 | case 6: | |
1796 | ||
e183eaed | 1797 | switch (algorithm) { |
16a53ecc | 1798 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 1799 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1800 | qd_idx = pd_idx + 1; |
1801 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1802 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1803 | qd_idx = 0; |
1804 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1805 | (*dd_idx) += 2; /* D D P Q D */ |
1806 | break; | |
1807 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 1808 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1809 | qd_idx = pd_idx + 1; |
1810 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1811 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1812 | qd_idx = 0; |
1813 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1814 | (*dd_idx) += 2; /* D D P Q D */ |
1815 | break; | |
1816 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 1817 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
1818 | qd_idx = (pd_idx + 1) % raid_disks; |
1819 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1820 | break; |
1821 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 1822 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
1823 | qd_idx = (pd_idx + 1) % raid_disks; |
1824 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1825 | break; |
99c0fb5f N |
1826 | |
1827 | case ALGORITHM_PARITY_0: | |
1828 | pd_idx = 0; | |
1829 | qd_idx = 1; | |
1830 | (*dd_idx) += 2; | |
1831 | break; | |
1832 | case ALGORITHM_PARITY_N: | |
1833 | pd_idx = data_disks; | |
1834 | qd_idx = data_disks + 1; | |
1835 | break; | |
1836 | ||
1837 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1838 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1839 | * of blocks for computing Q is different. | |
1840 | */ | |
6e3b96ed | 1841 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
1842 | qd_idx = pd_idx + 1; |
1843 | if (pd_idx == raid_disks-1) { | |
1844 | (*dd_idx)++; /* Q D D D P */ | |
1845 | qd_idx = 0; | |
1846 | } else if (*dd_idx >= pd_idx) | |
1847 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1848 | ddf_layout = 1; |
99c0fb5f N |
1849 | break; |
1850 | ||
1851 | case ALGORITHM_ROTATING_N_RESTART: | |
1852 | /* Same a left_asymmetric, by first stripe is | |
1853 | * D D D P Q rather than | |
1854 | * Q D D D P | |
1855 | */ | |
6e3b96ed N |
1856 | stripe2 += 1; |
1857 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
1858 | qd_idx = pd_idx + 1; |
1859 | if (pd_idx == raid_disks-1) { | |
1860 | (*dd_idx)++; /* Q D D D P */ | |
1861 | qd_idx = 0; | |
1862 | } else if (*dd_idx >= pd_idx) | |
1863 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1864 | ddf_layout = 1; |
99c0fb5f N |
1865 | break; |
1866 | ||
1867 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1868 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 1869 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
1870 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
1871 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1872 | ddf_layout = 1; |
99c0fb5f N |
1873 | break; |
1874 | ||
1875 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1876 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 1877 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1878 | if (*dd_idx >= pd_idx) |
1879 | (*dd_idx)++; | |
1880 | qd_idx = raid_disks - 1; | |
1881 | break; | |
1882 | ||
1883 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 1884 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1885 | if (*dd_idx >= pd_idx) |
1886 | (*dd_idx)++; | |
1887 | qd_idx = raid_disks - 1; | |
1888 | break; | |
1889 | ||
1890 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 1891 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1892 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1893 | qd_idx = raid_disks - 1; | |
1894 | break; | |
1895 | ||
1896 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 1897 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
1898 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
1899 | qd_idx = raid_disks - 1; | |
1900 | break; | |
1901 | ||
1902 | case ALGORITHM_PARITY_0_6: | |
1903 | pd_idx = 0; | |
1904 | (*dd_idx)++; | |
1905 | qd_idx = raid_disks - 1; | |
1906 | break; | |
1907 | ||
16a53ecc | 1908 | default: |
99c0fb5f | 1909 | BUG(); |
16a53ecc N |
1910 | } |
1911 | break; | |
1da177e4 LT |
1912 | } |
1913 | ||
911d4ee8 N |
1914 | if (sh) { |
1915 | sh->pd_idx = pd_idx; | |
1916 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1917 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1918 | } |
1da177e4 LT |
1919 | /* |
1920 | * Finally, compute the new sector number | |
1921 | */ | |
1922 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1923 | return new_sector; | |
1924 | } | |
1925 | ||
1926 | ||
784052ec | 1927 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
1928 | { |
1929 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1930 | int raid_disks = sh->disks; |
1931 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 1932 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
1933 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
1934 | : conf->chunk_sectors; | |
e183eaed N |
1935 | int algorithm = previous ? conf->prev_algo |
1936 | : conf->algorithm; | |
1da177e4 LT |
1937 | sector_t stripe; |
1938 | int chunk_offset; | |
35f2a591 N |
1939 | sector_t chunk_number; |
1940 | int dummy1, dd_idx = i; | |
1da177e4 | 1941 | sector_t r_sector; |
911d4ee8 | 1942 | struct stripe_head sh2; |
1da177e4 | 1943 | |
16a53ecc | 1944 | |
1da177e4 LT |
1945 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1946 | stripe = new_sector; | |
1da177e4 | 1947 | |
16a53ecc N |
1948 | if (i == sh->pd_idx) |
1949 | return 0; | |
1950 | switch(conf->level) { | |
1951 | case 4: break; | |
1952 | case 5: | |
e183eaed | 1953 | switch (algorithm) { |
1da177e4 LT |
1954 | case ALGORITHM_LEFT_ASYMMETRIC: |
1955 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1956 | if (i > sh->pd_idx) | |
1957 | i--; | |
1958 | break; | |
1959 | case ALGORITHM_LEFT_SYMMETRIC: | |
1960 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1961 | if (i < sh->pd_idx) | |
1962 | i += raid_disks; | |
1963 | i -= (sh->pd_idx + 1); | |
1964 | break; | |
99c0fb5f N |
1965 | case ALGORITHM_PARITY_0: |
1966 | i -= 1; | |
1967 | break; | |
1968 | case ALGORITHM_PARITY_N: | |
1969 | break; | |
1da177e4 | 1970 | default: |
99c0fb5f | 1971 | BUG(); |
16a53ecc N |
1972 | } |
1973 | break; | |
1974 | case 6: | |
d0dabf7e | 1975 | if (i == sh->qd_idx) |
16a53ecc | 1976 | return 0; /* It is the Q disk */ |
e183eaed | 1977 | switch (algorithm) { |
16a53ecc N |
1978 | case ALGORITHM_LEFT_ASYMMETRIC: |
1979 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1980 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1981 | case ALGORITHM_ROTATING_N_RESTART: | |
1982 | if (sh->pd_idx == raid_disks-1) | |
1983 | i--; /* Q D D D P */ | |
16a53ecc N |
1984 | else if (i > sh->pd_idx) |
1985 | i -= 2; /* D D P Q D */ | |
1986 | break; | |
1987 | case ALGORITHM_LEFT_SYMMETRIC: | |
1988 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1989 | if (sh->pd_idx == raid_disks-1) | |
1990 | i--; /* Q D D D P */ | |
1991 | else { | |
1992 | /* D D P Q D */ | |
1993 | if (i < sh->pd_idx) | |
1994 | i += raid_disks; | |
1995 | i -= (sh->pd_idx + 2); | |
1996 | } | |
1997 | break; | |
99c0fb5f N |
1998 | case ALGORITHM_PARITY_0: |
1999 | i -= 2; | |
2000 | break; | |
2001 | case ALGORITHM_PARITY_N: | |
2002 | break; | |
2003 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2004 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2005 | if (sh->pd_idx == 0) |
2006 | i--; /* P D D D Q */ | |
e4424fee N |
2007 | else { |
2008 | /* D D Q P D */ | |
2009 | if (i < sh->pd_idx) | |
2010 | i += raid_disks; | |
2011 | i -= (sh->pd_idx + 1); | |
2012 | } | |
99c0fb5f N |
2013 | break; |
2014 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2015 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2016 | if (i > sh->pd_idx) | |
2017 | i--; | |
2018 | break; | |
2019 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2020 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2021 | if (i < sh->pd_idx) | |
2022 | i += data_disks + 1; | |
2023 | i -= (sh->pd_idx + 1); | |
2024 | break; | |
2025 | case ALGORITHM_PARITY_0_6: | |
2026 | i -= 1; | |
2027 | break; | |
16a53ecc | 2028 | default: |
99c0fb5f | 2029 | BUG(); |
16a53ecc N |
2030 | } |
2031 | break; | |
1da177e4 LT |
2032 | } |
2033 | ||
2034 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2035 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2036 | |
112bf897 | 2037 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2038 | previous, &dummy1, &sh2); |
911d4ee8 N |
2039 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2040 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2041 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2042 | mdname(conf->mddev)); | |
1da177e4 LT |
2043 | return 0; |
2044 | } | |
2045 | return r_sector; | |
2046 | } | |
2047 | ||
2048 | ||
600aa109 | 2049 | static void |
c0f7bddb | 2050 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2051 | int rcw, int expand) |
e33129d8 DW |
2052 | { |
2053 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
c0f7bddb YT |
2054 | raid5_conf_t *conf = sh->raid_conf; |
2055 | int level = conf->level; | |
e33129d8 DW |
2056 | |
2057 | if (rcw) { | |
2058 | /* if we are not expanding this is a proper write request, and | |
2059 | * there will be bios with new data to be drained into the | |
2060 | * stripe cache | |
2061 | */ | |
2062 | if (!expand) { | |
600aa109 DW |
2063 | sh->reconstruct_state = reconstruct_state_drain_run; |
2064 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2065 | } else | |
2066 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 2067 | |
ac6b53b6 | 2068 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2069 | |
2070 | for (i = disks; i--; ) { | |
2071 | struct r5dev *dev = &sh->dev[i]; | |
2072 | ||
2073 | if (dev->towrite) { | |
2074 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2075 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2076 | if (!expand) |
2077 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2078 | s->locked++; |
e33129d8 DW |
2079 | } |
2080 | } | |
c0f7bddb | 2081 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2082 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2083 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2084 | } else { |
c0f7bddb | 2085 | BUG_ON(level == 6); |
e33129d8 DW |
2086 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2087 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2088 | ||
d8ee0728 | 2089 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
2090 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
2091 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
ac6b53b6 | 2092 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); |
e33129d8 DW |
2093 | |
2094 | for (i = disks; i--; ) { | |
2095 | struct r5dev *dev = &sh->dev[i]; | |
2096 | if (i == pd_idx) | |
2097 | continue; | |
2098 | ||
e33129d8 DW |
2099 | if (dev->towrite && |
2100 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2101 | test_bit(R5_Wantcompute, &dev->flags))) { |
2102 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2103 | set_bit(R5_LOCKED, &dev->flags); |
2104 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2105 | s->locked++; |
e33129d8 DW |
2106 | } |
2107 | } | |
2108 | } | |
2109 | ||
c0f7bddb | 2110 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2111 | * are in flight |
2112 | */ | |
2113 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2114 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2115 | s->locked++; |
e33129d8 | 2116 | |
c0f7bddb YT |
2117 | if (level == 6) { |
2118 | int qd_idx = sh->qd_idx; | |
2119 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2120 | ||
2121 | set_bit(R5_LOCKED, &dev->flags); | |
2122 | clear_bit(R5_UPTODATE, &dev->flags); | |
2123 | s->locked++; | |
2124 | } | |
2125 | ||
600aa109 | 2126 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2127 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2128 | s->locked, s->ops_request); |
e33129d8 | 2129 | } |
16a53ecc | 2130 | |
1da177e4 LT |
2131 | /* |
2132 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2133 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2134 | * The bi_next chain must be in order. |
2135 | */ | |
2136 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
2137 | { | |
2138 | struct bio **bip; | |
2139 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 2140 | int firstwrite=0; |
1da177e4 | 2141 | |
cbe47ec5 | 2142 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
1da177e4 LT |
2143 | (unsigned long long)bi->bi_sector, |
2144 | (unsigned long long)sh->sector); | |
2145 | ||
2146 | ||
1da177e4 | 2147 | spin_lock_irq(&conf->device_lock); |
72626685 | 2148 | if (forwrite) { |
1da177e4 | 2149 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
2150 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
2151 | firstwrite = 1; | |
2152 | } else | |
1da177e4 LT |
2153 | bip = &sh->dev[dd_idx].toread; |
2154 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
2155 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
2156 | goto overlap; | |
2157 | bip = & (*bip)->bi_next; | |
2158 | } | |
2159 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
2160 | goto overlap; | |
2161 | ||
78bafebd | 2162 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2163 | if (*bip) |
2164 | bi->bi_next = *bip; | |
2165 | *bip = bi; | |
960e739d | 2166 | bi->bi_phys_segments++; |
72626685 | 2167 | |
1da177e4 LT |
2168 | if (forwrite) { |
2169 | /* check if page is covered */ | |
2170 | sector_t sector = sh->dev[dd_idx].sector; | |
2171 | for (bi=sh->dev[dd_idx].towrite; | |
2172 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
2173 | bi && bi->bi_sector <= sector; | |
2174 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
2175 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
2176 | sector = bi->bi_sector + (bi->bi_size>>9); | |
2177 | } | |
2178 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2179 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2180 | } | |
cbe47ec5 | 2181 | spin_unlock_irq(&conf->device_lock); |
cbe47ec5 N |
2182 | |
2183 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
2184 | (unsigned long long)(*bip)->bi_sector, | |
2185 | (unsigned long long)sh->sector, dd_idx); | |
2186 | ||
2187 | if (conf->mddev->bitmap && firstwrite) { | |
2188 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2189 | STRIPE_SECTORS, 0); | |
2190 | sh->bm_seq = conf->seq_flush+1; | |
2191 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2192 | } | |
1da177e4 LT |
2193 | return 1; |
2194 | ||
2195 | overlap: | |
2196 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
2197 | spin_unlock_irq(&conf->device_lock); | |
1da177e4 LT |
2198 | return 0; |
2199 | } | |
2200 | ||
29269553 N |
2201 | static void end_reshape(raid5_conf_t *conf); |
2202 | ||
911d4ee8 N |
2203 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
2204 | struct stripe_head *sh) | |
ccfcc3c1 | 2205 | { |
784052ec | 2206 | int sectors_per_chunk = |
09c9e5fa | 2207 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2208 | int dd_idx; |
2d2063ce | 2209 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2210 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2211 | |
112bf897 N |
2212 | raid5_compute_sector(conf, |
2213 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2214 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2215 | previous, |
911d4ee8 | 2216 | &dd_idx, sh); |
ccfcc3c1 N |
2217 | } |
2218 | ||
a4456856 | 2219 | static void |
1fe797e6 | 2220 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2221 | struct stripe_head_state *s, int disks, |
2222 | struct bio **return_bi) | |
2223 | { | |
2224 | int i; | |
2225 | for (i = disks; i--; ) { | |
2226 | struct bio *bi; | |
2227 | int bitmap_end = 0; | |
2228 | ||
2229 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2230 | mdk_rdev_t *rdev; | |
2231 | rcu_read_lock(); | |
2232 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2233 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2234 | /* multiple read failures in one stripe */ | |
2235 | md_error(conf->mddev, rdev); | |
2236 | rcu_read_unlock(); | |
2237 | } | |
2238 | spin_lock_irq(&conf->device_lock); | |
2239 | /* fail all writes first */ | |
2240 | bi = sh->dev[i].towrite; | |
2241 | sh->dev[i].towrite = NULL; | |
2242 | if (bi) { | |
2243 | s->to_write--; | |
2244 | bitmap_end = 1; | |
2245 | } | |
2246 | ||
2247 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2248 | wake_up(&conf->wait_for_overlap); | |
2249 | ||
2250 | while (bi && bi->bi_sector < | |
2251 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2252 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2253 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2254 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2255 | md_write_end(conf->mddev); |
2256 | bi->bi_next = *return_bi; | |
2257 | *return_bi = bi; | |
2258 | } | |
2259 | bi = nextbi; | |
2260 | } | |
2261 | /* and fail all 'written' */ | |
2262 | bi = sh->dev[i].written; | |
2263 | sh->dev[i].written = NULL; | |
2264 | if (bi) bitmap_end = 1; | |
2265 | while (bi && bi->bi_sector < | |
2266 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2267 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2268 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2269 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2270 | md_write_end(conf->mddev); |
2271 | bi->bi_next = *return_bi; | |
2272 | *return_bi = bi; | |
2273 | } | |
2274 | bi = bi2; | |
2275 | } | |
2276 | ||
b5e98d65 DW |
2277 | /* fail any reads if this device is non-operational and |
2278 | * the data has not reached the cache yet. | |
2279 | */ | |
2280 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2281 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2282 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2283 | bi = sh->dev[i].toread; |
2284 | sh->dev[i].toread = NULL; | |
2285 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2286 | wake_up(&conf->wait_for_overlap); | |
2287 | if (bi) s->to_read--; | |
2288 | while (bi && bi->bi_sector < | |
2289 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2290 | struct bio *nextbi = | |
2291 | r5_next_bio(bi, sh->dev[i].sector); | |
2292 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2293 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2294 | bi->bi_next = *return_bi; |
2295 | *return_bi = bi; | |
2296 | } | |
2297 | bi = nextbi; | |
2298 | } | |
2299 | } | |
2300 | spin_unlock_irq(&conf->device_lock); | |
2301 | if (bitmap_end) | |
2302 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2303 | STRIPE_SECTORS, 0, 0); | |
2304 | } | |
2305 | ||
8b3e6cdc DW |
2306 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2307 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2308 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2309 | } |
2310 | ||
1fe797e6 DW |
2311 | /* fetch_block5 - checks the given member device to see if its data needs |
2312 | * to be read or computed to satisfy a request. | |
2313 | * | |
2314 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2315 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2316 | */ |
1fe797e6 DW |
2317 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2318 | int disk_idx, int disks) | |
f38e1219 DW |
2319 | { |
2320 | struct r5dev *dev = &sh->dev[disk_idx]; | |
f2b3b44d | 2321 | struct r5dev *failed_dev = &sh->dev[s->failed_num[0]]; |
f38e1219 | 2322 | |
f38e1219 DW |
2323 | /* is the data in this block needed, and can we get it? */ |
2324 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2325 | !test_bit(R5_UPTODATE, &dev->flags) && |
2326 | (dev->toread || | |
2327 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2328 | s->syncing || s->expanding || | |
5d35e09c N |
2329 | (s->failed && failed_dev->toread) || |
2330 | (s->failed && failed_dev->towrite && | |
1fe797e6 | 2331 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { |
976ea8d4 DW |
2332 | /* We would like to get this block, possibly by computing it, |
2333 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2334 | */ |
2335 | if ((s->uptodate == disks - 1) && | |
f2b3b44d | 2336 | (s->failed && disk_idx == s->failed_num[0])) { |
976ea8d4 DW |
2337 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2338 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2339 | set_bit(R5_Wantcompute, &dev->flags); |
2340 | sh->ops.target = disk_idx; | |
ac6b53b6 | 2341 | sh->ops.target2 = -1; |
f38e1219 | 2342 | s->req_compute = 1; |
f38e1219 | 2343 | /* Careful: from this point on 'uptodate' is in the eye |
ac6b53b6 | 2344 | * of raid_run_ops which services 'compute' operations |
f38e1219 DW |
2345 | * before writes. R5_Wantcompute flags a block that will |
2346 | * be R5_UPTODATE by the time it is needed for a | |
2347 | * subsequent operation. | |
2348 | */ | |
2349 | s->uptodate++; | |
1fe797e6 | 2350 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2351 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2352 | set_bit(R5_LOCKED, &dev->flags); |
2353 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2354 | s->locked++; |
2355 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2356 | s->syncing); | |
2357 | } | |
2358 | } | |
2359 | ||
1fe797e6 | 2360 | return 0; |
f38e1219 DW |
2361 | } |
2362 | ||
1fe797e6 DW |
2363 | /** |
2364 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2365 | */ | |
2366 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2367 | struct stripe_head_state *s, int disks) |
2368 | { | |
2369 | int i; | |
f38e1219 | 2370 | |
f38e1219 DW |
2371 | /* look for blocks to read/compute, skip this if a compute |
2372 | * is already in flight, or if the stripe contents are in the | |
2373 | * midst of changing due to a write | |
2374 | */ | |
976ea8d4 | 2375 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2376 | !sh->reconstruct_state) |
f38e1219 | 2377 | for (i = disks; i--; ) |
1fe797e6 | 2378 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2379 | break; |
a4456856 DW |
2380 | set_bit(STRIPE_HANDLE, &sh->state); |
2381 | } | |
2382 | ||
5599becc YT |
2383 | /* fetch_block6 - checks the given member device to see if its data needs |
2384 | * to be read or computed to satisfy a request. | |
2385 | * | |
2386 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2387 | * 0 to tell the loop in handle_stripe_fill6 to continue | |
2388 | */ | |
2389 | static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s, | |
f2b3b44d | 2390 | int disk_idx, int disks) |
a4456856 | 2391 | { |
5599becc | 2392 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2393 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2394 | &sh->dev[s->failed_num[1]] }; | |
5599becc YT |
2395 | |
2396 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2397 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2398 | (dev->toread || | |
2399 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2400 | s->syncing || s->expanding || | |
5d35e09c N |
2401 | (s->failed >= 1 && fdev[0]->toread) || |
2402 | (s->failed >= 2 && fdev[1]->toread) || | |
2403 | (s->failed && s->to_write)) { | |
5599becc YT |
2404 | /* we would like to get this block, possibly by computing it, |
2405 | * otherwise read it if the backing disk is insync | |
2406 | */ | |
2407 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
2408 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
2409 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
2410 | (s->failed && (disk_idx == s->failed_num[0] || |
2411 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
2412 | /* have disk failed, and we're requested to fetch it; |
2413 | * do compute it | |
a4456856 | 2414 | */ |
5599becc YT |
2415 | pr_debug("Computing stripe %llu block %d\n", |
2416 | (unsigned long long)sh->sector, disk_idx); | |
2417 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2418 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2419 | set_bit(R5_Wantcompute, &dev->flags); | |
2420 | sh->ops.target = disk_idx; | |
2421 | sh->ops.target2 = -1; /* no 2nd target */ | |
2422 | s->req_compute = 1; | |
2423 | s->uptodate++; | |
2424 | return 1; | |
2425 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
2426 | /* Computing 2-failure is *very* expensive; only | |
2427 | * do it if failed >= 2 | |
2428 | */ | |
2429 | int other; | |
2430 | for (other = disks; other--; ) { | |
2431 | if (other == disk_idx) | |
2432 | continue; | |
2433 | if (!test_bit(R5_UPTODATE, | |
2434 | &sh->dev[other].flags)) | |
2435 | break; | |
a4456856 | 2436 | } |
5599becc YT |
2437 | BUG_ON(other < 0); |
2438 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
2439 | (unsigned long long)sh->sector, | |
2440 | disk_idx, other); | |
2441 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2442 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2443 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
2444 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
2445 | sh->ops.target = disk_idx; | |
2446 | sh->ops.target2 = other; | |
2447 | s->uptodate += 2; | |
2448 | s->req_compute = 1; | |
2449 | return 1; | |
2450 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2451 | set_bit(R5_LOCKED, &dev->flags); | |
2452 | set_bit(R5_Wantread, &dev->flags); | |
2453 | s->locked++; | |
2454 | pr_debug("Reading block %d (sync=%d)\n", | |
2455 | disk_idx, s->syncing); | |
a4456856 DW |
2456 | } |
2457 | } | |
5599becc YT |
2458 | |
2459 | return 0; | |
2460 | } | |
2461 | ||
2462 | /** | |
2463 | * handle_stripe_fill6 - read or compute data to satisfy pending requests. | |
2464 | */ | |
2465 | static void handle_stripe_fill6(struct stripe_head *sh, | |
f2b3b44d | 2466 | struct stripe_head_state *s, |
5599becc YT |
2467 | int disks) |
2468 | { | |
2469 | int i; | |
2470 | ||
2471 | /* look for blocks to read/compute, skip this if a compute | |
2472 | * is already in flight, or if the stripe contents are in the | |
2473 | * midst of changing due to a write | |
2474 | */ | |
2475 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
2476 | !sh->reconstruct_state) | |
2477 | for (i = disks; i--; ) | |
f2b3b44d | 2478 | if (fetch_block6(sh, s, i, disks)) |
5599becc | 2479 | break; |
a4456856 DW |
2480 | set_bit(STRIPE_HANDLE, &sh->state); |
2481 | } | |
2482 | ||
2483 | ||
1fe797e6 | 2484 | /* handle_stripe_clean_event |
a4456856 DW |
2485 | * any written block on an uptodate or failed drive can be returned. |
2486 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2487 | * never LOCKED, so we don't need to test 'failed' directly. | |
2488 | */ | |
1fe797e6 | 2489 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2490 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2491 | { | |
2492 | int i; | |
2493 | struct r5dev *dev; | |
2494 | ||
2495 | for (i = disks; i--; ) | |
2496 | if (sh->dev[i].written) { | |
2497 | dev = &sh->dev[i]; | |
2498 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2499 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2500 | /* We can return any write requests */ | |
2501 | struct bio *wbi, *wbi2; | |
2502 | int bitmap_end = 0; | |
45b4233c | 2503 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2504 | spin_lock_irq(&conf->device_lock); |
2505 | wbi = dev->written; | |
2506 | dev->written = NULL; | |
2507 | while (wbi && wbi->bi_sector < | |
2508 | dev->sector + STRIPE_SECTORS) { | |
2509 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2510 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2511 | md_write_end(conf->mddev); |
2512 | wbi->bi_next = *return_bi; | |
2513 | *return_bi = wbi; | |
2514 | } | |
2515 | wbi = wbi2; | |
2516 | } | |
2517 | if (dev->towrite == NULL) | |
2518 | bitmap_end = 1; | |
2519 | spin_unlock_irq(&conf->device_lock); | |
2520 | if (bitmap_end) | |
2521 | bitmap_endwrite(conf->mddev->bitmap, | |
2522 | sh->sector, | |
2523 | STRIPE_SECTORS, | |
2524 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2525 | 0); | |
2526 | } | |
2527 | } | |
8b3e6cdc DW |
2528 | |
2529 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2530 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2531 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2532 | } |
2533 | ||
1fe797e6 | 2534 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2535 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2536 | { | |
2537 | int rmw = 0, rcw = 0, i; | |
2538 | for (i = disks; i--; ) { | |
2539 | /* would I have to read this buffer for read_modify_write */ | |
2540 | struct r5dev *dev = &sh->dev[i]; | |
2541 | if ((dev->towrite || i == sh->pd_idx) && | |
2542 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2543 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2544 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2545 | if (test_bit(R5_Insync, &dev->flags)) |
2546 | rmw++; | |
2547 | else | |
2548 | rmw += 2*disks; /* cannot read it */ | |
2549 | } | |
2550 | /* Would I have to read this buffer for reconstruct_write */ | |
2551 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2552 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2553 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2554 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2555 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2556 | else |
2557 | rcw += 2*disks; | |
2558 | } | |
2559 | } | |
45b4233c | 2560 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2561 | (unsigned long long)sh->sector, rmw, rcw); |
2562 | set_bit(STRIPE_HANDLE, &sh->state); | |
2563 | if (rmw < rcw && rmw > 0) | |
2564 | /* prefer read-modify-write, but need to get some data */ | |
2565 | for (i = disks; i--; ) { | |
2566 | struct r5dev *dev = &sh->dev[i]; | |
2567 | if ((dev->towrite || i == sh->pd_idx) && | |
2568 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2569 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2570 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2571 | test_bit(R5_Insync, &dev->flags)) { |
2572 | if ( | |
2573 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2574 | pr_debug("Read_old block " |
a4456856 DW |
2575 | "%d for r-m-w\n", i); |
2576 | set_bit(R5_LOCKED, &dev->flags); | |
2577 | set_bit(R5_Wantread, &dev->flags); | |
2578 | s->locked++; | |
2579 | } else { | |
2580 | set_bit(STRIPE_DELAYED, &sh->state); | |
2581 | set_bit(STRIPE_HANDLE, &sh->state); | |
2582 | } | |
2583 | } | |
2584 | } | |
2585 | if (rcw <= rmw && rcw > 0) | |
2586 | /* want reconstruct write, but need to get some data */ | |
2587 | for (i = disks; i--; ) { | |
2588 | struct r5dev *dev = &sh->dev[i]; | |
2589 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2590 | i != sh->pd_idx && | |
2591 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2592 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2593 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2594 | test_bit(R5_Insync, &dev->flags)) { |
2595 | if ( | |
2596 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2597 | pr_debug("Read_old block " |
a4456856 DW |
2598 | "%d for Reconstruct\n", i); |
2599 | set_bit(R5_LOCKED, &dev->flags); | |
2600 | set_bit(R5_Wantread, &dev->flags); | |
2601 | s->locked++; | |
2602 | } else { | |
2603 | set_bit(STRIPE_DELAYED, &sh->state); | |
2604 | set_bit(STRIPE_HANDLE, &sh->state); | |
2605 | } | |
2606 | } | |
2607 | } | |
2608 | /* now if nothing is locked, and if we have enough data, | |
2609 | * we can start a write request | |
2610 | */ | |
f38e1219 DW |
2611 | /* since handle_stripe can be called at any time we need to handle the |
2612 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
2613 | * subsequent call wants to start a write request. raid_run_ops only |
2614 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
2615 | * simultaneously. If this is not the case then new writes need to be |
2616 | * held off until the compute completes. | |
2617 | */ | |
976ea8d4 DW |
2618 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2619 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2620 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 2621 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
2622 | } |
2623 | ||
1fe797e6 | 2624 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 | 2625 | struct stripe_head *sh, struct stripe_head_state *s, |
f2b3b44d | 2626 | int disks) |
a4456856 | 2627 | { |
a9b39a74 | 2628 | int rcw = 0, pd_idx = sh->pd_idx, i; |
34e04e87 | 2629 | int qd_idx = sh->qd_idx; |
a9b39a74 YT |
2630 | |
2631 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2632 | for (i = disks; i--; ) { |
2633 | struct r5dev *dev = &sh->dev[i]; | |
a9b39a74 YT |
2634 | /* check if we haven't enough data */ |
2635 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2636 | i != pd_idx && i != qd_idx && | |
2637 | !test_bit(R5_LOCKED, &dev->flags) && | |
2638 | !(test_bit(R5_UPTODATE, &dev->flags) || | |
2639 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2640 | rcw++; | |
2641 | if (!test_bit(R5_Insync, &dev->flags)) | |
2642 | continue; /* it's a failed drive */ | |
2643 | ||
2644 | if ( | |
2645 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2646 | pr_debug("Read_old stripe %llu " | |
2647 | "block %d for Reconstruct\n", | |
2648 | (unsigned long long)sh->sector, i); | |
2649 | set_bit(R5_LOCKED, &dev->flags); | |
2650 | set_bit(R5_Wantread, &dev->flags); | |
2651 | s->locked++; | |
2652 | } else { | |
2653 | pr_debug("Request delayed stripe %llu " | |
2654 | "block %d for Reconstruct\n", | |
2655 | (unsigned long long)sh->sector, i); | |
2656 | set_bit(STRIPE_DELAYED, &sh->state); | |
2657 | set_bit(STRIPE_HANDLE, &sh->state); | |
a4456856 DW |
2658 | } |
2659 | } | |
2660 | } | |
a4456856 DW |
2661 | /* now if nothing is locked, and if we have enough data, we can start a |
2662 | * write request | |
2663 | */ | |
a9b39a74 YT |
2664 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2665 | s->locked == 0 && rcw == 0 && | |
a4456856 | 2666 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { |
a9b39a74 | 2667 | schedule_reconstruction(sh, s, 1, 0); |
a4456856 DW |
2668 | } |
2669 | } | |
2670 | ||
2671 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2672 | struct stripe_head_state *s, int disks) | |
2673 | { | |
ecc65c9b | 2674 | struct r5dev *dev = NULL; |
bd2ab670 | 2675 | |
a4456856 | 2676 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2677 | |
ecc65c9b DW |
2678 | switch (sh->check_state) { |
2679 | case check_state_idle: | |
2680 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2681 | if (s->failed == 0) { |
bd2ab670 | 2682 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2683 | sh->check_state = check_state_run; |
2684 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2685 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2686 | s->uptodate--; |
ecc65c9b | 2687 | break; |
bd2ab670 | 2688 | } |
f2b3b44d | 2689 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
2690 | /* fall through */ |
2691 | case check_state_compute_result: | |
2692 | sh->check_state = check_state_idle; | |
2693 | if (!dev) | |
2694 | dev = &sh->dev[sh->pd_idx]; | |
2695 | ||
2696 | /* check that a write has not made the stripe insync */ | |
2697 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2698 | break; | |
c8894419 | 2699 | |
a4456856 | 2700 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2701 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2702 | BUG_ON(s->uptodate != disks); | |
2703 | ||
2704 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2705 | s->locked++; |
a4456856 | 2706 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2707 | |
a4456856 | 2708 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2709 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2710 | break; |
2711 | case check_state_run: | |
2712 | break; /* we will be called again upon completion */ | |
2713 | case check_state_check_result: | |
2714 | sh->check_state = check_state_idle; | |
2715 | ||
2716 | /* if a failure occurred during the check operation, leave | |
2717 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2718 | */ | |
2719 | if (s->failed) | |
2720 | break; | |
2721 | ||
2722 | /* handle a successful check operation, if parity is correct | |
2723 | * we are done. Otherwise update the mismatch count and repair | |
2724 | * parity if !MD_RECOVERY_CHECK | |
2725 | */ | |
ad283ea4 | 2726 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
2727 | /* parity is correct (on disc, |
2728 | * not in buffer any more) | |
2729 | */ | |
2730 | set_bit(STRIPE_INSYNC, &sh->state); | |
2731 | else { | |
2732 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2733 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2734 | /* don't try to repair!! */ | |
2735 | set_bit(STRIPE_INSYNC, &sh->state); | |
2736 | else { | |
2737 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2738 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2739 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2740 | set_bit(R5_Wantcompute, | |
2741 | &sh->dev[sh->pd_idx].flags); | |
2742 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 2743 | sh->ops.target2 = -1; |
ecc65c9b DW |
2744 | s->uptodate++; |
2745 | } | |
2746 | } | |
2747 | break; | |
2748 | case check_state_compute_run: | |
2749 | break; | |
2750 | default: | |
2751 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2752 | __func__, sh->check_state, | |
2753 | (unsigned long long) sh->sector); | |
2754 | BUG(); | |
a4456856 DW |
2755 | } |
2756 | } | |
2757 | ||
2758 | ||
2759 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
36d1c647 | 2760 | struct stripe_head_state *s, |
f2b3b44d | 2761 | int disks) |
a4456856 | 2762 | { |
a4456856 | 2763 | int pd_idx = sh->pd_idx; |
34e04e87 | 2764 | int qd_idx = sh->qd_idx; |
d82dfee0 | 2765 | struct r5dev *dev; |
a4456856 DW |
2766 | |
2767 | set_bit(STRIPE_HANDLE, &sh->state); | |
2768 | ||
2769 | BUG_ON(s->failed > 2); | |
d82dfee0 | 2770 | |
a4456856 DW |
2771 | /* Want to check and possibly repair P and Q. |
2772 | * However there could be one 'failed' device, in which | |
2773 | * case we can only check one of them, possibly using the | |
2774 | * other to generate missing data | |
2775 | */ | |
2776 | ||
d82dfee0 DW |
2777 | switch (sh->check_state) { |
2778 | case check_state_idle: | |
2779 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 2780 | if (s->failed == s->q_failed) { |
d82dfee0 | 2781 | /* The only possible failed device holds Q, so it |
a4456856 DW |
2782 | * makes sense to check P (If anything else were failed, |
2783 | * we would have used P to recreate it). | |
2784 | */ | |
d82dfee0 | 2785 | sh->check_state = check_state_run; |
a4456856 | 2786 | } |
f2b3b44d | 2787 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 2788 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
2789 | * anything, so it makes sense to check it |
2790 | */ | |
d82dfee0 DW |
2791 | if (sh->check_state == check_state_run) |
2792 | sh->check_state = check_state_run_pq; | |
2793 | else | |
2794 | sh->check_state = check_state_run_q; | |
a4456856 | 2795 | } |
a4456856 | 2796 | |
d82dfee0 DW |
2797 | /* discard potentially stale zero_sum_result */ |
2798 | sh->ops.zero_sum_result = 0; | |
a4456856 | 2799 | |
d82dfee0 DW |
2800 | if (sh->check_state == check_state_run) { |
2801 | /* async_xor_zero_sum destroys the contents of P */ | |
2802 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
2803 | s->uptodate--; | |
a4456856 | 2804 | } |
d82dfee0 DW |
2805 | if (sh->check_state >= check_state_run && |
2806 | sh->check_state <= check_state_run_pq) { | |
2807 | /* async_syndrome_zero_sum preserves P and Q, so | |
2808 | * no need to mark them !uptodate here | |
2809 | */ | |
2810 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
2811 | break; | |
a4456856 DW |
2812 | } |
2813 | ||
d82dfee0 DW |
2814 | /* we have 2-disk failure */ |
2815 | BUG_ON(s->failed != 2); | |
2816 | /* fall through */ | |
2817 | case check_state_compute_result: | |
2818 | sh->check_state = check_state_idle; | |
a4456856 | 2819 | |
d82dfee0 DW |
2820 | /* check that a write has not made the stripe insync */ |
2821 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2822 | break; | |
a4456856 DW |
2823 | |
2824 | /* now write out any block on a failed drive, | |
d82dfee0 | 2825 | * or P or Q if they were recomputed |
a4456856 | 2826 | */ |
d82dfee0 | 2827 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 2828 | if (s->failed == 2) { |
f2b3b44d | 2829 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
2830 | s->locked++; |
2831 | set_bit(R5_LOCKED, &dev->flags); | |
2832 | set_bit(R5_Wantwrite, &dev->flags); | |
2833 | } | |
2834 | if (s->failed >= 1) { | |
f2b3b44d | 2835 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
2836 | s->locked++; |
2837 | set_bit(R5_LOCKED, &dev->flags); | |
2838 | set_bit(R5_Wantwrite, &dev->flags); | |
2839 | } | |
d82dfee0 | 2840 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
2841 | dev = &sh->dev[pd_idx]; |
2842 | s->locked++; | |
2843 | set_bit(R5_LOCKED, &dev->flags); | |
2844 | set_bit(R5_Wantwrite, &dev->flags); | |
2845 | } | |
d82dfee0 | 2846 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
2847 | dev = &sh->dev[qd_idx]; |
2848 | s->locked++; | |
2849 | set_bit(R5_LOCKED, &dev->flags); | |
2850 | set_bit(R5_Wantwrite, &dev->flags); | |
2851 | } | |
2852 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2853 | ||
2854 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
2855 | break; |
2856 | case check_state_run: | |
2857 | case check_state_run_q: | |
2858 | case check_state_run_pq: | |
2859 | break; /* we will be called again upon completion */ | |
2860 | case check_state_check_result: | |
2861 | sh->check_state = check_state_idle; | |
2862 | ||
2863 | /* handle a successful check operation, if parity is correct | |
2864 | * we are done. Otherwise update the mismatch count and repair | |
2865 | * parity if !MD_RECOVERY_CHECK | |
2866 | */ | |
2867 | if (sh->ops.zero_sum_result == 0) { | |
2868 | /* both parities are correct */ | |
2869 | if (!s->failed) | |
2870 | set_bit(STRIPE_INSYNC, &sh->state); | |
2871 | else { | |
2872 | /* in contrast to the raid5 case we can validate | |
2873 | * parity, but still have a failure to write | |
2874 | * back | |
2875 | */ | |
2876 | sh->check_state = check_state_compute_result; | |
2877 | /* Returning at this point means that we may go | |
2878 | * off and bring p and/or q uptodate again so | |
2879 | * we make sure to check zero_sum_result again | |
2880 | * to verify if p or q need writeback | |
2881 | */ | |
2882 | } | |
2883 | } else { | |
2884 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2885 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2886 | /* don't try to repair!! */ | |
2887 | set_bit(STRIPE_INSYNC, &sh->state); | |
2888 | else { | |
2889 | int *target = &sh->ops.target; | |
2890 | ||
2891 | sh->ops.target = -1; | |
2892 | sh->ops.target2 = -1; | |
2893 | sh->check_state = check_state_compute_run; | |
2894 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
2895 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
2896 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
2897 | set_bit(R5_Wantcompute, | |
2898 | &sh->dev[pd_idx].flags); | |
2899 | *target = pd_idx; | |
2900 | target = &sh->ops.target2; | |
2901 | s->uptodate++; | |
2902 | } | |
2903 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
2904 | set_bit(R5_Wantcompute, | |
2905 | &sh->dev[qd_idx].flags); | |
2906 | *target = qd_idx; | |
2907 | s->uptodate++; | |
2908 | } | |
2909 | } | |
2910 | } | |
2911 | break; | |
2912 | case check_state_compute_run: | |
2913 | break; | |
2914 | default: | |
2915 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2916 | __func__, sh->check_state, | |
2917 | (unsigned long long) sh->sector); | |
2918 | BUG(); | |
a4456856 DW |
2919 | } |
2920 | } | |
2921 | ||
86c374ba | 2922 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh) |
a4456856 DW |
2923 | { |
2924 | int i; | |
2925 | ||
2926 | /* We have read all the blocks in this stripe and now we need to | |
2927 | * copy some of them into a target stripe for expand. | |
2928 | */ | |
f0a50d37 | 2929 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2930 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2931 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 2932 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 2933 | int dd_idx, j; |
a4456856 | 2934 | struct stripe_head *sh2; |
a08abd8c | 2935 | struct async_submit_ctl submit; |
a4456856 | 2936 | |
784052ec | 2937 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
2938 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2939 | &dd_idx, NULL); | |
a8c906ca | 2940 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
2941 | if (sh2 == NULL) |
2942 | /* so far only the early blocks of this stripe | |
2943 | * have been requested. When later blocks | |
2944 | * get requested, we will try again | |
2945 | */ | |
2946 | continue; | |
2947 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2948 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2949 | /* must have already done this block */ | |
2950 | release_stripe(sh2); | |
2951 | continue; | |
2952 | } | |
f0a50d37 DW |
2953 | |
2954 | /* place all the copies on one channel */ | |
a08abd8c | 2955 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 2956 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 2957 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 2958 | &submit); |
f0a50d37 | 2959 | |
a4456856 DW |
2960 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2961 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2962 | for (j = 0; j < conf->raid_disks; j++) | |
2963 | if (j != sh2->pd_idx && | |
86c374ba | 2964 | j != sh2->qd_idx && |
a4456856 DW |
2965 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2966 | break; | |
2967 | if (j == conf->raid_disks) { | |
2968 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2969 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2970 | } | |
2971 | release_stripe(sh2); | |
f0a50d37 | 2972 | |
a4456856 | 2973 | } |
a2e08551 N |
2974 | /* done submitting copies, wait for them to complete */ |
2975 | if (tx) { | |
2976 | async_tx_ack(tx); | |
2977 | dma_wait_for_async_tx(tx); | |
2978 | } | |
a4456856 | 2979 | } |
1da177e4 | 2980 | |
6bfe0b49 | 2981 | |
1da177e4 LT |
2982 | /* |
2983 | * handle_stripe - do things to a stripe. | |
2984 | * | |
2985 | * We lock the stripe and then examine the state of various bits | |
2986 | * to see what needs to be done. | |
2987 | * Possible results: | |
2988 | * return some read request which now have data | |
2989 | * return some write requests which are safely on disc | |
2990 | * schedule a read on some buffers | |
2991 | * schedule a write of some buffers | |
2992 | * return confirmation of parity correctness | |
2993 | * | |
1da177e4 LT |
2994 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2995 | * get BH_Lock set before the stripe lock is released. | |
2996 | * | |
2997 | */ | |
a4456856 | 2998 | |
3687c061 | 2999 | static int handle_stripe5(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 LT |
3000 | { |
3001 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 | 3002 | int disks = sh->disks, i; |
1da177e4 | 3003 | struct r5dev *dev; |
e0a115e5 | 3004 | int prexor; |
1da177e4 | 3005 | |
83de75cc | 3006 | /* Now to look around and see what can be done */ |
9910f16a | 3007 | rcu_read_lock(); |
c4c1663b | 3008 | spin_lock_irq(&conf->device_lock); |
1da177e4 LT |
3009 | for (i=disks; i--; ) { |
3010 | mdk_rdev_t *rdev; | |
a9f326eb N |
3011 | |
3012 | dev = &sh->dev[i]; | |
1da177e4 | 3013 | |
b5e98d65 DW |
3014 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
3015 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
3016 | dev->towrite, dev->written); | |
3017 | ||
3018 | /* maybe we can request a biofill operation | |
3019 | * | |
3020 | * new wantfill requests are only permitted while | |
83de75cc | 3021 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
3022 | */ |
3023 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 3024 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 3025 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
3026 | |
3027 | /* now count some things */ | |
cc94015a N |
3028 | if (test_bit(R5_LOCKED, &dev->flags)) |
3029 | s->locked++; | |
3030 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3031 | s->uptodate++; | |
3032 | if (test_bit(R5_Wantcompute, &dev->flags)) | |
3033 | s->compute++; | |
1da177e4 | 3034 | |
b5e98d65 | 3035 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 3036 | s->to_fill++; |
b5e98d65 | 3037 | else if (dev->toread) |
cc94015a | 3038 | s->to_read++; |
1da177e4 | 3039 | if (dev->towrite) { |
cc94015a | 3040 | s->to_write++; |
1da177e4 | 3041 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3042 | s->non_overwrite++; |
1da177e4 | 3043 | } |
a4456856 | 3044 | if (dev->written) |
cc94015a | 3045 | s->written++; |
9910f16a | 3046 | rdev = rcu_dereference(conf->disks[i].rdev); |
cc94015a | 3047 | if (s->blocked_rdev == NULL && |
ac4090d2 | 3048 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
cc94015a | 3049 | s->blocked_rdev = rdev; |
6bfe0b49 | 3050 | atomic_inc(&rdev->nr_pending); |
6bfe0b49 | 3051 | } |
415e72d0 N |
3052 | clear_bit(R5_Insync, &dev->flags); |
3053 | if (!rdev) | |
3054 | /* Not in-sync */; | |
3055 | else if (test_bit(In_sync, &rdev->flags)) | |
3056 | set_bit(R5_Insync, &dev->flags); | |
3057 | else { | |
3058 | /* could be in-sync depending on recovery/reshape status */ | |
3059 | if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) | |
3060 | set_bit(R5_Insync, &dev->flags); | |
3061 | } | |
3062 | if (!test_bit(R5_Insync, &dev->flags)) { | |
14f8d26b | 3063 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
3064 | clear_bit(R5_ReadError, &dev->flags); |
3065 | clear_bit(R5_ReWrite, &dev->flags); | |
3066 | } | |
415e72d0 N |
3067 | if (test_bit(R5_ReadError, &dev->flags)) |
3068 | clear_bit(R5_Insync, &dev->flags); | |
3069 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3070 | s->failed++; |
3071 | s->failed_num[0] = i; | |
415e72d0 | 3072 | } |
1da177e4 | 3073 | } |
c4c1663b | 3074 | spin_unlock_irq(&conf->device_lock); |
9910f16a | 3075 | rcu_read_unlock(); |
b5e98d65 | 3076 | |
cc94015a N |
3077 | if (unlikely(s->blocked_rdev)) { |
3078 | if (s->syncing || s->expanding || s->expanded || | |
3079 | s->to_write || s->written) { | |
ac4090d2 | 3080 | set_bit(STRIPE_HANDLE, &sh->state); |
3687c061 | 3081 | return 1; |
ac4090d2 N |
3082 | } |
3083 | /* There is nothing for the blocked_rdev to block */ | |
cc94015a N |
3084 | rdev_dec_pending(s->blocked_rdev, conf->mddev); |
3085 | s->blocked_rdev = NULL; | |
6bfe0b49 DW |
3086 | } |
3087 | ||
cc94015a N |
3088 | if (s->to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3089 | set_bit(STRIPE_OP_BIOFILL, &s->ops_request); | |
83de75cc DW |
3090 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); |
3091 | } | |
b5e98d65 | 3092 | |
45b4233c | 3093 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 3094 | " to_write=%d failed=%d failed_num=%d\n", |
cc94015a N |
3095 | s->locked, s->uptodate, s->to_read, s->to_write, |
3096 | s->failed, s->failed_num[0]); | |
1da177e4 LT |
3097 | /* check if the array has lost two devices and, if so, some requests might |
3098 | * need to be failed | |
3099 | */ | |
cc94015a N |
3100 | if (s->failed > 1 && s->to_read+s->to_write+s->written) |
3101 | handle_failed_stripe(conf, sh, s, disks, &s->return_bi); | |
3102 | if (s->failed > 1 && s->syncing) { | |
1da177e4 LT |
3103 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3104 | clear_bit(STRIPE_SYNCING, &sh->state); | |
cc94015a | 3105 | s->syncing = 0; |
1da177e4 LT |
3106 | } |
3107 | ||
3108 | /* might be able to return some write requests if the parity block | |
3109 | * is safe, or on a failed drive | |
3110 | */ | |
3111 | dev = &sh->dev[sh->pd_idx]; | |
cc94015a N |
3112 | if (s->written && |
3113 | ((test_bit(R5_Insync, &dev->flags) && | |
3114 | !test_bit(R5_LOCKED, &dev->flags) && | |
3115 | test_bit(R5_UPTODATE, &dev->flags)) || | |
3116 | (s->failed == 1 && s->failed_num[0] == sh->pd_idx))) | |
3117 | handle_stripe_clean_event(conf, sh, disks, &s->return_bi); | |
1da177e4 LT |
3118 | |
3119 | /* Now we might consider reading some blocks, either to check/generate | |
3120 | * parity, or to satisfy requests | |
3121 | * or to load a block that is being partially written. | |
3122 | */ | |
cc94015a N |
3123 | if (s->to_read || s->non_overwrite || |
3124 | (s->syncing && (s->uptodate + s->compute < disks)) || s->expanding) | |
3125 | handle_stripe_fill5(sh, s, disks); | |
1da177e4 | 3126 | |
e33129d8 DW |
3127 | /* Now we check to see if any write operations have recently |
3128 | * completed | |
3129 | */ | |
e0a115e5 | 3130 | prexor = 0; |
d8ee0728 | 3131 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 3132 | prexor = 1; |
d8ee0728 DW |
3133 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
3134 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 3135 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
3136 | |
3137 | /* All the 'written' buffers and the parity block are ready to | |
3138 | * be written back to disk | |
3139 | */ | |
3140 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3141 | for (i = disks; i--; ) { | |
3142 | dev = &sh->dev[i]; | |
3143 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3144 | (i == sh->pd_idx || dev->written)) { | |
3145 | pr_debug("Writing block %d\n", i); | |
3146 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
3147 | if (prexor) |
3148 | continue; | |
e33129d8 | 3149 | if (!test_bit(R5_Insync, &dev->flags) || |
cc94015a | 3150 | (i == sh->pd_idx && s->failed == 0)) |
e33129d8 DW |
3151 | set_bit(STRIPE_INSYNC, &sh->state); |
3152 | } | |
3153 | } | |
729a1866 | 3154 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
cc94015a | 3155 | s->dec_preread_active = 1; |
e33129d8 DW |
3156 | } |
3157 | ||
3158 | /* Now to consider new write requests and what else, if anything | |
3159 | * should be read. We do not handle new writes when: | |
3160 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3161 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3162 | * block. | |
3163 | */ | |
cc94015a N |
3164 | if (s->to_write && !sh->reconstruct_state && !sh->check_state) |
3165 | handle_stripe_dirtying5(conf, sh, s, disks); | |
1da177e4 LT |
3166 | |
3167 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
3168 | * Any reads will already have been scheduled, so we just see if enough |
3169 | * data is available. The parity check is held off while parity | |
3170 | * dependent operations are in flight. | |
1da177e4 | 3171 | */ |
ecc65c9b | 3172 | if (sh->check_state || |
cc94015a | 3173 | (s->syncing && s->locked == 0 && |
976ea8d4 | 3174 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 3175 | !test_bit(STRIPE_INSYNC, &sh->state))) |
cc94015a | 3176 | handle_parity_checks5(conf, sh, s, disks); |
3687c061 | 3177 | return 0; |
1da177e4 LT |
3178 | } |
3179 | ||
3687c061 | 3180 | static int handle_stripe6(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3181 | { |
bff61975 | 3182 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 3183 | int disks = sh->disks; |
34e04e87 | 3184 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx; |
16a53ecc | 3185 | struct r5dev *dev, *pdev, *qdev; |
1da177e4 | 3186 | |
16a53ecc | 3187 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3188 | |
3189 | rcu_read_lock(); | |
c4c1663b | 3190 | spin_lock_irq(&conf->device_lock); |
16a53ecc N |
3191 | for (i=disks; i--; ) { |
3192 | mdk_rdev_t *rdev; | |
3193 | dev = &sh->dev[i]; | |
1da177e4 | 3194 | |
45b4233c | 3195 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc | 3196 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
6c0069c0 YT |
3197 | /* maybe we can reply to a read |
3198 | * | |
3199 | * new wantfill requests are only permitted while | |
3200 | * ops_complete_biofill is guaranteed to be inactive | |
3201 | */ | |
3202 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3203 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3204 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3205 | |
16a53ecc | 3206 | /* now count some things */ |
cc94015a N |
3207 | if (test_bit(R5_LOCKED, &dev->flags)) |
3208 | s->locked++; | |
3209 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3210 | s->uptodate++; | |
2d6e4ecc | 3211 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
3212 | s->compute++; |
3213 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 3214 | } |
1da177e4 | 3215 | |
6c0069c0 | 3216 | if (test_bit(R5_Wantfill, &dev->flags)) { |
cc94015a | 3217 | s->to_fill++; |
6c0069c0 | 3218 | } else if (dev->toread) |
cc94015a | 3219 | s->to_read++; |
16a53ecc | 3220 | if (dev->towrite) { |
cc94015a | 3221 | s->to_write++; |
16a53ecc | 3222 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3223 | s->non_overwrite++; |
16a53ecc | 3224 | } |
a4456856 | 3225 | if (dev->written) |
cc94015a | 3226 | s->written++; |
16a53ecc | 3227 | rdev = rcu_dereference(conf->disks[i].rdev); |
cc94015a | 3228 | if (s->blocked_rdev == NULL && |
ac4090d2 | 3229 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
cc94015a | 3230 | s->blocked_rdev = rdev; |
6bfe0b49 | 3231 | atomic_inc(&rdev->nr_pending); |
6bfe0b49 | 3232 | } |
415e72d0 N |
3233 | clear_bit(R5_Insync, &dev->flags); |
3234 | if (!rdev) | |
3235 | /* Not in-sync */; | |
3236 | else if (test_bit(In_sync, &rdev->flags)) | |
3237 | set_bit(R5_Insync, &dev->flags); | |
3238 | else { | |
3239 | /* in sync if before recovery_offset */ | |
3240 | if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) | |
3241 | set_bit(R5_Insync, &dev->flags); | |
3242 | } | |
3243 | if (!test_bit(R5_Insync, &dev->flags)) { | |
16a53ecc N |
3244 | /* The ReadError flag will just be confusing now */ |
3245 | clear_bit(R5_ReadError, &dev->flags); | |
3246 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3247 | } |
415e72d0 N |
3248 | if (test_bit(R5_ReadError, &dev->flags)) |
3249 | clear_bit(R5_Insync, &dev->flags); | |
3250 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3251 | if (s->failed < 2) |
3252 | s->failed_num[s->failed] = i; | |
3253 | s->failed++; | |
415e72d0 | 3254 | } |
1da177e4 | 3255 | } |
c4c1663b | 3256 | spin_unlock_irq(&conf->device_lock); |
1da177e4 | 3257 | rcu_read_unlock(); |
6bfe0b49 | 3258 | |
cc94015a N |
3259 | if (unlikely(s->blocked_rdev)) { |
3260 | if (s->syncing || s->expanding || s->expanded || | |
3261 | s->to_write || s->written) { | |
ac4090d2 | 3262 | set_bit(STRIPE_HANDLE, &sh->state); |
3687c061 | 3263 | return 1; |
ac4090d2 N |
3264 | } |
3265 | /* There is nothing for the blocked_rdev to block */ | |
cc94015a N |
3266 | rdev_dec_pending(s->blocked_rdev, conf->mddev); |
3267 | s->blocked_rdev = NULL; | |
6bfe0b49 | 3268 | } |
ac4090d2 | 3269 | |
cc94015a N |
3270 | if (s->to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
3271 | set_bit(STRIPE_OP_BIOFILL, &s->ops_request); | |
6c0069c0 YT |
3272 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); |
3273 | } | |
3274 | ||
45b4233c | 3275 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3276 | " to_write=%d failed=%d failed_num=%d,%d\n", |
cc94015a N |
3277 | s->locked, s->uptodate, s->to_read, s->to_write, s->failed, |
3278 | s->failed_num[0], s->failed_num[1]); | |
a4456856 DW |
3279 | /* check if the array has lost >2 devices and, if so, some requests |
3280 | * might need to be failed | |
16a53ecc | 3281 | */ |
cc94015a N |
3282 | if (s->failed > 2 && s->to_read+s->to_write+s->written) |
3283 | handle_failed_stripe(conf, sh, s, disks, &s->return_bi); | |
3284 | if (s->failed > 2 && s->syncing) { | |
16a53ecc N |
3285 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3286 | clear_bit(STRIPE_SYNCING, &sh->state); | |
cc94015a | 3287 | s->syncing = 0; |
16a53ecc N |
3288 | } |
3289 | ||
3290 | /* | |
3291 | * might be able to return some write requests if the parity blocks | |
3292 | * are safe, or on a failed drive | |
3293 | */ | |
3294 | pdev = &sh->dev[pd_idx]; | |
cc94015a N |
3295 | s->p_failed = (s->failed >= 1 && s->failed_num[0] == pd_idx) |
3296 | || (s->failed >= 2 && s->failed_num[1] == pd_idx); | |
34e04e87 | 3297 | qdev = &sh->dev[qd_idx]; |
cc94015a N |
3298 | s->q_failed = (s->failed >= 1 && s->failed_num[0] == qd_idx) |
3299 | || (s->failed >= 2 && s->failed_num[1] == qd_idx); | |
a4456856 | 3300 | |
cc94015a N |
3301 | if (s->written && |
3302 | (s->p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3303 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 | 3304 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
cc94015a | 3305 | (s->q_failed || ((test_bit(R5_Insync, &qdev->flags) |
16a53ecc | 3306 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3307 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
cc94015a | 3308 | handle_stripe_clean_event(conf, sh, disks, &s->return_bi); |
16a53ecc N |
3309 | |
3310 | /* Now we might consider reading some blocks, either to check/generate | |
3311 | * parity, or to satisfy requests | |
3312 | * or to load a block that is being partially written. | |
3313 | */ | |
cc94015a N |
3314 | if (s->to_read || s->non_overwrite || (s->to_write && s->failed) || |
3315 | (s->syncing && (s->uptodate + s->compute < disks)) || s->expanding) | |
3316 | handle_stripe_fill6(sh, s, disks); | |
16a53ecc | 3317 | |
6c0069c0 YT |
3318 | /* Now we check to see if any write operations have recently |
3319 | * completed | |
3320 | */ | |
3321 | if (sh->reconstruct_state == reconstruct_state_drain_result) { | |
6c0069c0 YT |
3322 | |
3323 | sh->reconstruct_state = reconstruct_state_idle; | |
3324 | /* All the 'written' buffers and the parity blocks are ready to | |
3325 | * be written back to disk | |
3326 | */ | |
3327 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
3328 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags)); | |
3329 | for (i = disks; i--; ) { | |
3330 | dev = &sh->dev[i]; | |
3331 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3332 | (i == sh->pd_idx || i == qd_idx || | |
3333 | dev->written)) { | |
3334 | pr_debug("Writing block %d\n", i); | |
3335 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); | |
3336 | set_bit(R5_Wantwrite, &dev->flags); | |
3337 | if (!test_bit(R5_Insync, &dev->flags) || | |
3338 | ((i == sh->pd_idx || i == qd_idx) && | |
cc94015a | 3339 | s->failed == 0)) |
6c0069c0 YT |
3340 | set_bit(STRIPE_INSYNC, &sh->state); |
3341 | } | |
3342 | } | |
729a1866 | 3343 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
cc94015a | 3344 | s->dec_preread_active = 1; |
6c0069c0 YT |
3345 | } |
3346 | ||
a9b39a74 YT |
3347 | /* Now to consider new write requests and what else, if anything |
3348 | * should be read. We do not handle new writes when: | |
3349 | * 1/ A 'write' operation (copy+gen_syndrome) is already in flight. | |
3350 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3351 | * block. | |
3352 | */ | |
cc94015a N |
3353 | if (s->to_write && !sh->reconstruct_state && !sh->check_state) |
3354 | handle_stripe_dirtying6(conf, sh, s, disks); | |
16a53ecc N |
3355 | |
3356 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 | 3357 | * Any reads will already have been scheduled, so we just see if enough |
6c0069c0 YT |
3358 | * data is available. The parity check is held off while parity |
3359 | * dependent operations are in flight. | |
16a53ecc | 3360 | */ |
6c0069c0 | 3361 | if (sh->check_state || |
cc94015a | 3362 | (s->syncing && s->locked == 0 && |
6c0069c0 YT |
3363 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
3364 | !test_bit(STRIPE_INSYNC, &sh->state))) | |
cc94015a | 3365 | handle_parity_checks6(conf, sh, s, disks); |
3687c061 | 3366 | return 0; |
cc94015a N |
3367 | } |
3368 | ||
3369 | static void handle_stripe(struct stripe_head *sh) | |
3370 | { | |
3371 | struct stripe_head_state s; | |
3687c061 N |
3372 | int done; |
3373 | int i; | |
cc94015a N |
3374 | raid5_conf_t *conf = sh->raid_conf; |
3375 | ||
3376 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3377 | if (test_and_set_bit(STRIPE_ACTIVE, &sh->state)) { | |
3378 | /* already being handled, ensure it gets handled | |
3379 | * again when current action finishes */ | |
3380 | set_bit(STRIPE_HANDLE, &sh->state); | |
3381 | return; | |
3382 | } | |
3383 | ||
3384 | if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3385 | set_bit(STRIPE_SYNCING, &sh->state); | |
3386 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3387 | } | |
3388 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3389 | ||
3390 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3391 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3392 | (unsigned long long)sh->sector, sh->state, | |
3393 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3394 | sh->check_state, sh->reconstruct_state); | |
3395 | memset(&s, 0, sizeof(s)); | |
3396 | ||
3397 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); | |
3398 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3399 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
3400 | ||
3401 | if (conf->level == 6) | |
3687c061 | 3402 | done = handle_stripe6(sh, &s); |
cc94015a | 3403 | else |
3687c061 N |
3404 | done = handle_stripe5(sh, &s); |
3405 | ||
3406 | if (done) | |
3407 | goto finish; | |
c5a31000 N |
3408 | |
3409 | ||
3410 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { | |
3411 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3412 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3413 | } | |
3414 | ||
3415 | /* If the failed drives are just a ReadError, then we might need | |
3416 | * to progress the repair/check process | |
3417 | */ | |
3418 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
3419 | for (i = 0; i < s.failed; i++) { | |
3420 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
3421 | if (test_bit(R5_ReadError, &dev->flags) | |
3422 | && !test_bit(R5_LOCKED, &dev->flags) | |
3423 | && test_bit(R5_UPTODATE, &dev->flags) | |
3424 | ) { | |
3425 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3426 | set_bit(R5_Wantwrite, &dev->flags); | |
3427 | set_bit(R5_ReWrite, &dev->flags); | |
3428 | set_bit(R5_LOCKED, &dev->flags); | |
3429 | s.locked++; | |
3430 | } else { | |
3431 | /* let's read it back */ | |
3432 | set_bit(R5_Wantread, &dev->flags); | |
3433 | set_bit(R5_LOCKED, &dev->flags); | |
3434 | s.locked++; | |
3435 | } | |
3436 | } | |
3437 | } | |
3438 | ||
3439 | ||
3687c061 N |
3440 | /* Finish reconstruct operations initiated by the expansion process */ |
3441 | if (sh->reconstruct_state == reconstruct_state_result) { | |
3442 | struct stripe_head *sh_src | |
3443 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
3444 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
3445 | /* sh cannot be written until sh_src has been read. | |
3446 | * so arrange for sh to be delayed a little | |
3447 | */ | |
3448 | set_bit(STRIPE_DELAYED, &sh->state); | |
3449 | set_bit(STRIPE_HANDLE, &sh->state); | |
3450 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
3451 | &sh_src->state)) | |
3452 | atomic_inc(&conf->preread_active_stripes); | |
3453 | release_stripe(sh_src); | |
3454 | goto finish; | |
3455 | } | |
3456 | if (sh_src) | |
3457 | release_stripe(sh_src); | |
3458 | ||
3459 | sh->reconstruct_state = reconstruct_state_idle; | |
3460 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
3461 | for (i = conf->raid_disks; i--; ) { | |
3462 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
3463 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3464 | s.locked++; | |
3465 | } | |
3466 | } | |
f416885e | 3467 | |
3687c061 N |
3468 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
3469 | !sh->reconstruct_state) { | |
3470 | /* Need to write out all blocks after computing parity */ | |
3471 | sh->disks = conf->raid_disks; | |
3472 | stripe_set_idx(sh->sector, conf, 0, sh); | |
3473 | schedule_reconstruction(sh, &s, 1, 1); | |
3474 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
3475 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
3476 | atomic_dec(&conf->reshape_stripes); | |
3477 | wake_up(&conf->wait_for_overlap); | |
3478 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3479 | } | |
3480 | ||
3481 | if (s.expanding && s.locked == 0 && | |
3482 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
3483 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 3484 | |
3687c061 | 3485 | finish: |
6bfe0b49 | 3486 | /* wait for this device to become unblocked */ |
c5709ef6 N |
3487 | if (unlikely(s.blocked_rdev)) |
3488 | md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev); | |
6bfe0b49 | 3489 | |
6c0069c0 YT |
3490 | if (s.ops_request) |
3491 | raid_run_ops(sh, s.ops_request); | |
3492 | ||
f0e43bcd | 3493 | ops_run_io(sh, &s); |
16a53ecc | 3494 | |
729a1866 | 3495 | |
c5709ef6 | 3496 | if (s.dec_preread_active) { |
729a1866 | 3497 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 3498 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
3499 | * have actually been submitted. |
3500 | */ | |
3501 | atomic_dec(&conf->preread_active_stripes); | |
3502 | if (atomic_read(&conf->preread_active_stripes) < | |
3503 | IO_THRESHOLD) | |
3504 | md_wakeup_thread(conf->mddev->thread); | |
3505 | } | |
3506 | ||
c5709ef6 | 3507 | return_io(s.return_bi); |
16a53ecc | 3508 | |
c4c1663b | 3509 | clear_bit(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
3510 | } |
3511 | ||
16a53ecc N |
3512 | static void raid5_activate_delayed(raid5_conf_t *conf) |
3513 | { | |
3514 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3515 | while (!list_empty(&conf->delayed_list)) { | |
3516 | struct list_head *l = conf->delayed_list.next; | |
3517 | struct stripe_head *sh; | |
3518 | sh = list_entry(l, struct stripe_head, lru); | |
3519 | list_del_init(l); | |
3520 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3521 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3522 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3523 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3524 | } |
482c0834 | 3525 | } |
16a53ecc N |
3526 | } |
3527 | ||
3528 | static void activate_bit_delay(raid5_conf_t *conf) | |
3529 | { | |
3530 | /* device_lock is held */ | |
3531 | struct list_head head; | |
3532 | list_add(&head, &conf->bitmap_list); | |
3533 | list_del_init(&conf->bitmap_list); | |
3534 | while (!list_empty(&head)) { | |
3535 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3536 | list_del_init(&sh->lru); | |
3537 | atomic_inc(&sh->count); | |
3538 | __release_stripe(conf, sh); | |
3539 | } | |
3540 | } | |
3541 | ||
11d8a6e3 | 3542 | int md_raid5_congested(mddev_t *mddev, int bits) |
f022b2fd | 3543 | { |
070ec55d | 3544 | raid5_conf_t *conf = mddev->private; |
f022b2fd N |
3545 | |
3546 | /* No difference between reads and writes. Just check | |
3547 | * how busy the stripe_cache is | |
3548 | */ | |
3fa841d7 | 3549 | |
f022b2fd N |
3550 | if (conf->inactive_blocked) |
3551 | return 1; | |
3552 | if (conf->quiesce) | |
3553 | return 1; | |
3554 | if (list_empty_careful(&conf->inactive_list)) | |
3555 | return 1; | |
3556 | ||
3557 | return 0; | |
3558 | } | |
11d8a6e3 N |
3559 | EXPORT_SYMBOL_GPL(md_raid5_congested); |
3560 | ||
3561 | static int raid5_congested(void *data, int bits) | |
3562 | { | |
3563 | mddev_t *mddev = data; | |
3564 | ||
3565 | return mddev_congested(mddev, bits) || | |
3566 | md_raid5_congested(mddev, bits); | |
3567 | } | |
f022b2fd | 3568 | |
23032a0e RBJ |
3569 | /* We want read requests to align with chunks where possible, |
3570 | * but write requests don't need to. | |
3571 | */ | |
cc371e66 AK |
3572 | static int raid5_mergeable_bvec(struct request_queue *q, |
3573 | struct bvec_merge_data *bvm, | |
3574 | struct bio_vec *biovec) | |
23032a0e RBJ |
3575 | { |
3576 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3577 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 3578 | int max; |
9d8f0363 | 3579 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 3580 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3581 | |
cc371e66 | 3582 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3583 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3584 | ||
664e7c41 AN |
3585 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3586 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
3587 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
3588 | if (max < 0) max = 0; | |
3589 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3590 | return biovec->bv_len; | |
3591 | else | |
3592 | return max; | |
3593 | } | |
3594 | ||
f679623f RBJ |
3595 | |
3596 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3597 | { | |
3598 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
9d8f0363 | 3599 | unsigned int chunk_sectors = mddev->chunk_sectors; |
f679623f RBJ |
3600 | unsigned int bio_sectors = bio->bi_size >> 9; |
3601 | ||
664e7c41 AN |
3602 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
3603 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
3604 | return chunk_sectors >= |
3605 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3606 | } | |
3607 | ||
46031f9a RBJ |
3608 | /* |
3609 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3610 | * later sampled by raid5d. | |
3611 | */ | |
3612 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3613 | { | |
3614 | unsigned long flags; | |
3615 | ||
3616 | spin_lock_irqsave(&conf->device_lock, flags); | |
3617 | ||
3618 | bi->bi_next = conf->retry_read_aligned_list; | |
3619 | conf->retry_read_aligned_list = bi; | |
3620 | ||
3621 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3622 | md_wakeup_thread(conf->mddev->thread); | |
3623 | } | |
3624 | ||
3625 | ||
3626 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3627 | { | |
3628 | struct bio *bi; | |
3629 | ||
3630 | bi = conf->retry_read_aligned; | |
3631 | if (bi) { | |
3632 | conf->retry_read_aligned = NULL; | |
3633 | return bi; | |
3634 | } | |
3635 | bi = conf->retry_read_aligned_list; | |
3636 | if(bi) { | |
387bb173 | 3637 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3638 | bi->bi_next = NULL; |
960e739d JA |
3639 | /* |
3640 | * this sets the active strip count to 1 and the processed | |
3641 | * strip count to zero (upper 8 bits) | |
3642 | */ | |
46031f9a | 3643 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3644 | } |
3645 | ||
3646 | return bi; | |
3647 | } | |
3648 | ||
3649 | ||
f679623f RBJ |
3650 | /* |
3651 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3652 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3653 | * first). | |
3654 | * If the read failed.. | |
3655 | */ | |
6712ecf8 | 3656 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3657 | { |
3658 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3659 | mddev_t *mddev; |
3660 | raid5_conf_t *conf; | |
3661 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3662 | mdk_rdev_t *rdev; | |
3663 | ||
f679623f | 3664 | bio_put(bi); |
46031f9a | 3665 | |
46031f9a RBJ |
3666 | rdev = (void*)raid_bi->bi_next; |
3667 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
3668 | mddev = rdev->mddev; |
3669 | conf = mddev->private; | |
46031f9a RBJ |
3670 | |
3671 | rdev_dec_pending(rdev, conf->mddev); | |
3672 | ||
3673 | if (!error && uptodate) { | |
6712ecf8 | 3674 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3675 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3676 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3677 | return; |
46031f9a RBJ |
3678 | } |
3679 | ||
3680 | ||
45b4233c | 3681 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3682 | |
3683 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3684 | } |
3685 | ||
387bb173 NB |
3686 | static int bio_fits_rdev(struct bio *bi) |
3687 | { | |
165125e1 | 3688 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 3689 | |
ae03bf63 | 3690 | if ((bi->bi_size>>9) > queue_max_sectors(q)) |
387bb173 NB |
3691 | return 0; |
3692 | blk_recount_segments(q, bi); | |
8a78362c | 3693 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
3694 | return 0; |
3695 | ||
3696 | if (q->merge_bvec_fn) | |
3697 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3698 | * just just give up | |
3699 | */ | |
3700 | return 0; | |
3701 | ||
3702 | return 1; | |
3703 | } | |
3704 | ||
3705 | ||
21a52c6d | 3706 | static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio) |
f679623f | 3707 | { |
070ec55d | 3708 | raid5_conf_t *conf = mddev->private; |
8553fe7e | 3709 | int dd_idx; |
f679623f RBJ |
3710 | struct bio* align_bi; |
3711 | mdk_rdev_t *rdev; | |
3712 | ||
3713 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3714 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3715 | return 0; |
3716 | } | |
3717 | /* | |
a167f663 | 3718 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 3719 | */ |
a167f663 | 3720 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
3721 | if (!align_bi) |
3722 | return 0; | |
3723 | /* | |
3724 | * set bi_end_io to a new function, and set bi_private to the | |
3725 | * original bio. | |
3726 | */ | |
3727 | align_bi->bi_end_io = raid5_align_endio; | |
3728 | align_bi->bi_private = raid_bio; | |
3729 | /* | |
3730 | * compute position | |
3731 | */ | |
112bf897 N |
3732 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3733 | 0, | |
911d4ee8 | 3734 | &dd_idx, NULL); |
f679623f RBJ |
3735 | |
3736 | rcu_read_lock(); | |
3737 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3738 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3739 | atomic_inc(&rdev->nr_pending); |
3740 | rcu_read_unlock(); | |
46031f9a RBJ |
3741 | raid_bio->bi_next = (void*)rdev; |
3742 | align_bi->bi_bdev = rdev->bdev; | |
3743 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3744 | align_bi->bi_sector += rdev->data_offset; | |
3745 | ||
387bb173 NB |
3746 | if (!bio_fits_rdev(align_bi)) { |
3747 | /* too big in some way */ | |
3748 | bio_put(align_bi); | |
3749 | rdev_dec_pending(rdev, mddev); | |
3750 | return 0; | |
3751 | } | |
3752 | ||
46031f9a RBJ |
3753 | spin_lock_irq(&conf->device_lock); |
3754 | wait_event_lock_irq(conf->wait_for_stripe, | |
3755 | conf->quiesce == 0, | |
3756 | conf->device_lock, /* nothing */); | |
3757 | atomic_inc(&conf->active_aligned_reads); | |
3758 | spin_unlock_irq(&conf->device_lock); | |
3759 | ||
f679623f RBJ |
3760 | generic_make_request(align_bi); |
3761 | return 1; | |
3762 | } else { | |
3763 | rcu_read_unlock(); | |
46031f9a | 3764 | bio_put(align_bi); |
f679623f RBJ |
3765 | return 0; |
3766 | } | |
3767 | } | |
3768 | ||
8b3e6cdc DW |
3769 | /* __get_priority_stripe - get the next stripe to process |
3770 | * | |
3771 | * Full stripe writes are allowed to pass preread active stripes up until | |
3772 | * the bypass_threshold is exceeded. In general the bypass_count | |
3773 | * increments when the handle_list is handled before the hold_list; however, it | |
3774 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3775 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3776 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3777 | * handle_list. | |
3778 | */ | |
3779 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3780 | { | |
3781 | struct stripe_head *sh; | |
3782 | ||
3783 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3784 | __func__, | |
3785 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3786 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3787 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3788 | ||
3789 | if (!list_empty(&conf->handle_list)) { | |
3790 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3791 | ||
3792 | if (list_empty(&conf->hold_list)) | |
3793 | conf->bypass_count = 0; | |
3794 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3795 | if (conf->hold_list.next == conf->last_hold) | |
3796 | conf->bypass_count++; | |
3797 | else { | |
3798 | conf->last_hold = conf->hold_list.next; | |
3799 | conf->bypass_count -= conf->bypass_threshold; | |
3800 | if (conf->bypass_count < 0) | |
3801 | conf->bypass_count = 0; | |
3802 | } | |
3803 | } | |
3804 | } else if (!list_empty(&conf->hold_list) && | |
3805 | ((conf->bypass_threshold && | |
3806 | conf->bypass_count > conf->bypass_threshold) || | |
3807 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3808 | sh = list_entry(conf->hold_list.next, | |
3809 | typeof(*sh), lru); | |
3810 | conf->bypass_count -= conf->bypass_threshold; | |
3811 | if (conf->bypass_count < 0) | |
3812 | conf->bypass_count = 0; | |
3813 | } else | |
3814 | return NULL; | |
3815 | ||
3816 | list_del_init(&sh->lru); | |
3817 | atomic_inc(&sh->count); | |
3818 | BUG_ON(atomic_read(&sh->count) != 1); | |
3819 | return sh; | |
3820 | } | |
f679623f | 3821 | |
21a52c6d | 3822 | static int make_request(mddev_t *mddev, struct bio * bi) |
1da177e4 | 3823 | { |
070ec55d | 3824 | raid5_conf_t *conf = mddev->private; |
911d4ee8 | 3825 | int dd_idx; |
1da177e4 LT |
3826 | sector_t new_sector; |
3827 | sector_t logical_sector, last_sector; | |
3828 | struct stripe_head *sh; | |
a362357b | 3829 | const int rw = bio_data_dir(bi); |
49077326 | 3830 | int remaining; |
7c13edc8 | 3831 | int plugged; |
1da177e4 | 3832 | |
e9c7469b TH |
3833 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
3834 | md_flush_request(mddev, bi); | |
e5dcdd80 N |
3835 | return 0; |
3836 | } | |
3837 | ||
3d310eb7 | 3838 | md_write_start(mddev, bi); |
06d91a5f | 3839 | |
802ba064 | 3840 | if (rw == READ && |
52488615 | 3841 | mddev->reshape_position == MaxSector && |
21a52c6d | 3842 | chunk_aligned_read(mddev,bi)) |
99c0fb5f | 3843 | return 0; |
52488615 | 3844 | |
1da177e4 LT |
3845 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3846 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3847 | bi->bi_next = NULL; | |
3848 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3849 | |
7c13edc8 | 3850 | plugged = mddev_check_plugged(mddev); |
1da177e4 LT |
3851 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3852 | DEFINE_WAIT(w); | |
16a53ecc | 3853 | int disks, data_disks; |
b5663ba4 | 3854 | int previous; |
b578d55f | 3855 | |
7ecaa1e6 | 3856 | retry: |
b5663ba4 | 3857 | previous = 0; |
b0f9ec04 | 3858 | disks = conf->raid_disks; |
b578d55f | 3859 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
b0f9ec04 | 3860 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 3861 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
3862 | * 64bit on a 32bit platform, and so it might be |
3863 | * possible to see a half-updated value | |
aeb878b0 | 3864 | * Of course reshape_progress could change after |
df8e7f76 N |
3865 | * the lock is dropped, so once we get a reference |
3866 | * to the stripe that we think it is, we will have | |
3867 | * to check again. | |
3868 | */ | |
7ecaa1e6 | 3869 | spin_lock_irq(&conf->device_lock); |
fef9c61f N |
3870 | if (mddev->delta_disks < 0 |
3871 | ? logical_sector < conf->reshape_progress | |
3872 | : logical_sector >= conf->reshape_progress) { | |
7ecaa1e6 | 3873 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3874 | previous = 1; |
3875 | } else { | |
fef9c61f N |
3876 | if (mddev->delta_disks < 0 |
3877 | ? logical_sector < conf->reshape_safe | |
3878 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
3879 | spin_unlock_irq(&conf->device_lock); |
3880 | schedule(); | |
3881 | goto retry; | |
3882 | } | |
3883 | } | |
7ecaa1e6 N |
3884 | spin_unlock_irq(&conf->device_lock); |
3885 | } | |
16a53ecc N |
3886 | data_disks = disks - conf->max_degraded; |
3887 | ||
112bf897 N |
3888 | new_sector = raid5_compute_sector(conf, logical_sector, |
3889 | previous, | |
911d4ee8 | 3890 | &dd_idx, NULL); |
0c55e022 | 3891 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3892 | (unsigned long long)new_sector, |
3893 | (unsigned long long)logical_sector); | |
3894 | ||
b5663ba4 | 3895 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 3896 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 3897 | if (sh) { |
b0f9ec04 | 3898 | if (unlikely(previous)) { |
7ecaa1e6 | 3899 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
3900 | * stripe, so we must do the range check again. |
3901 | * Expansion could still move past after this | |
3902 | * test, but as we are holding a reference to | |
3903 | * 'sh', we know that if that happens, | |
3904 | * STRIPE_EXPANDING will get set and the expansion | |
3905 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3906 | */ |
3907 | int must_retry = 0; | |
3908 | spin_lock_irq(&conf->device_lock); | |
b0f9ec04 N |
3909 | if (mddev->delta_disks < 0 |
3910 | ? logical_sector >= conf->reshape_progress | |
3911 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
3912 | /* mismatch, need to try again */ |
3913 | must_retry = 1; | |
3914 | spin_unlock_irq(&conf->device_lock); | |
3915 | if (must_retry) { | |
3916 | release_stripe(sh); | |
7a3ab908 | 3917 | schedule(); |
7ecaa1e6 N |
3918 | goto retry; |
3919 | } | |
3920 | } | |
e62e58a5 | 3921 | |
ffd96e35 | 3922 | if (rw == WRITE && |
a5c308d4 | 3923 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
3924 | logical_sector < mddev->suspend_hi) { |
3925 | release_stripe(sh); | |
e62e58a5 N |
3926 | /* As the suspend_* range is controlled by |
3927 | * userspace, we want an interruptible | |
3928 | * wait. | |
3929 | */ | |
3930 | flush_signals(current); | |
3931 | prepare_to_wait(&conf->wait_for_overlap, | |
3932 | &w, TASK_INTERRUPTIBLE); | |
3933 | if (logical_sector >= mddev->suspend_lo && | |
3934 | logical_sector < mddev->suspend_hi) | |
3935 | schedule(); | |
e464eafd N |
3936 | goto retry; |
3937 | } | |
7ecaa1e6 N |
3938 | |
3939 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
ffd96e35 | 3940 | !add_stripe_bio(sh, bi, dd_idx, rw)) { |
7ecaa1e6 N |
3941 | /* Stripe is busy expanding or |
3942 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3943 | * and wait a while |
3944 | */ | |
482c0834 | 3945 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
3946 | release_stripe(sh); |
3947 | schedule(); | |
3948 | goto retry; | |
3949 | } | |
3950 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3951 | set_bit(STRIPE_HANDLE, &sh->state); |
3952 | clear_bit(STRIPE_DELAYED, &sh->state); | |
e9c7469b | 3953 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
3954 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
3955 | atomic_inc(&conf->preread_active_stripes); | |
1da177e4 | 3956 | release_stripe(sh); |
1da177e4 LT |
3957 | } else { |
3958 | /* cannot get stripe for read-ahead, just give-up */ | |
3959 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3960 | finish_wait(&conf->wait_for_overlap, &w); | |
3961 | break; | |
3962 | } | |
3963 | ||
3964 | } | |
7c13edc8 N |
3965 | if (!plugged) |
3966 | md_wakeup_thread(mddev->thread); | |
3967 | ||
1da177e4 | 3968 | spin_lock_irq(&conf->device_lock); |
960e739d | 3969 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3970 | spin_unlock_irq(&conf->device_lock); |
3971 | if (remaining == 0) { | |
1da177e4 | 3972 | |
16a53ecc | 3973 | if ( rw == WRITE ) |
1da177e4 | 3974 | md_write_end(mddev); |
6712ecf8 | 3975 | |
0e13fe23 | 3976 | bio_endio(bi, 0); |
1da177e4 | 3977 | } |
729a1866 | 3978 | |
1da177e4 LT |
3979 | return 0; |
3980 | } | |
3981 | ||
b522adcd DW |
3982 | static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks); |
3983 | ||
52c03291 | 3984 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3985 | { |
52c03291 N |
3986 | /* reshaping is quite different to recovery/resync so it is |
3987 | * handled quite separately ... here. | |
3988 | * | |
3989 | * On each call to sync_request, we gather one chunk worth of | |
3990 | * destination stripes and flag them as expanding. | |
3991 | * Then we find all the source stripes and request reads. | |
3992 | * As the reads complete, handle_stripe will copy the data | |
3993 | * into the destination stripe and release that stripe. | |
3994 | */ | |
7b92813c | 3995 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 3996 | struct stripe_head *sh; |
ccfcc3c1 | 3997 | sector_t first_sector, last_sector; |
f416885e N |
3998 | int raid_disks = conf->previous_raid_disks; |
3999 | int data_disks = raid_disks - conf->max_degraded; | |
4000 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4001 | int i; |
4002 | int dd_idx; | |
c8f517c4 | 4003 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4004 | sector_t stripe_addr; |
7a661381 | 4005 | int reshape_sectors; |
ab69ae12 | 4006 | struct list_head stripes; |
52c03291 | 4007 | |
fef9c61f N |
4008 | if (sector_nr == 0) { |
4009 | /* If restarting in the middle, skip the initial sectors */ | |
4010 | if (mddev->delta_disks < 0 && | |
4011 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { | |
4012 | sector_nr = raid5_size(mddev, 0, 0) | |
4013 | - conf->reshape_progress; | |
a639755c | 4014 | } else if (mddev->delta_disks >= 0 && |
fef9c61f N |
4015 | conf->reshape_progress > 0) |
4016 | sector_nr = conf->reshape_progress; | |
f416885e | 4017 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4018 | if (sector_nr) { |
8dee7211 N |
4019 | mddev->curr_resync_completed = sector_nr; |
4020 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4021 | *skipped = 1; |
4022 | return sector_nr; | |
4023 | } | |
52c03291 N |
4024 | } |
4025 | ||
7a661381 N |
4026 | /* We need to process a full chunk at a time. |
4027 | * If old and new chunk sizes differ, we need to process the | |
4028 | * largest of these | |
4029 | */ | |
664e7c41 AN |
4030 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4031 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4032 | else |
9d8f0363 | 4033 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4034 | |
52c03291 N |
4035 | /* we update the metadata when there is more than 3Meg |
4036 | * in the block range (that is rather arbitrary, should | |
4037 | * probably be time based) or when the data about to be | |
4038 | * copied would over-write the source of the data at | |
4039 | * the front of the range. | |
fef9c61f N |
4040 | * i.e. one new_stripe along from reshape_progress new_maps |
4041 | * to after where reshape_safe old_maps to | |
52c03291 | 4042 | */ |
fef9c61f | 4043 | writepos = conf->reshape_progress; |
f416885e | 4044 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4045 | readpos = conf->reshape_progress; |
4046 | sector_div(readpos, data_disks); | |
fef9c61f | 4047 | safepos = conf->reshape_safe; |
f416885e | 4048 | sector_div(safepos, data_disks); |
fef9c61f | 4049 | if (mddev->delta_disks < 0) { |
ed37d83e | 4050 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4051 | readpos += reshape_sectors; |
7a661381 | 4052 | safepos += reshape_sectors; |
fef9c61f | 4053 | } else { |
7a661381 | 4054 | writepos += reshape_sectors; |
ed37d83e N |
4055 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4056 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4057 | } |
52c03291 | 4058 | |
c8f517c4 N |
4059 | /* 'writepos' is the most advanced device address we might write. |
4060 | * 'readpos' is the least advanced device address we might read. | |
4061 | * 'safepos' is the least address recorded in the metadata as having | |
4062 | * been reshaped. | |
4063 | * If 'readpos' is behind 'writepos', then there is no way that we can | |
4064 | * ensure safety in the face of a crash - that must be done by userspace | |
4065 | * making a backup of the data. So in that case there is no particular | |
4066 | * rush to update metadata. | |
4067 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4068 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4069 | * we can be safe in the event of a crash. | |
4070 | * So we insist on updating metadata if safepos is behind writepos and | |
4071 | * readpos is beyond writepos. | |
4072 | * In any case, update the metadata every 10 seconds. | |
4073 | * Maybe that number should be configurable, but I'm not sure it is | |
4074 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4075 | */ | |
fef9c61f | 4076 | if ((mddev->delta_disks < 0 |
c8f517c4 N |
4077 | ? (safepos > writepos && readpos < writepos) |
4078 | : (safepos < writepos && readpos > writepos)) || | |
4079 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4080 | /* Cannot proceed until we've updated the superblock... */ |
4081 | wait_event(conf->wait_for_overlap, | |
4082 | atomic_read(&conf->reshape_stripes)==0); | |
fef9c61f | 4083 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4084 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4085 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4086 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4087 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4088 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
4089 | kthread_should_stop()); |
4090 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4091 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4092 | spin_unlock_irq(&conf->device_lock); |
4093 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4094 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4095 | } |
4096 | ||
ec32a2bd N |
4097 | if (mddev->delta_disks < 0) { |
4098 | BUG_ON(conf->reshape_progress == 0); | |
4099 | stripe_addr = writepos; | |
4100 | BUG_ON((mddev->dev_sectors & | |
7a661381 N |
4101 | ~((sector_t)reshape_sectors - 1)) |
4102 | - reshape_sectors - stripe_addr | |
ec32a2bd N |
4103 | != sector_nr); |
4104 | } else { | |
7a661381 | 4105 | BUG_ON(writepos != sector_nr + reshape_sectors); |
ec32a2bd N |
4106 | stripe_addr = sector_nr; |
4107 | } | |
ab69ae12 | 4108 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4109 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4110 | int j; |
a9f326eb | 4111 | int skipped_disk = 0; |
a8c906ca | 4112 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4113 | set_bit(STRIPE_EXPANDING, &sh->state); |
4114 | atomic_inc(&conf->reshape_stripes); | |
4115 | /* If any of this stripe is beyond the end of the old | |
4116 | * array, then we need to zero those blocks | |
4117 | */ | |
4118 | for (j=sh->disks; j--;) { | |
4119 | sector_t s; | |
4120 | if (j == sh->pd_idx) | |
4121 | continue; | |
f416885e | 4122 | if (conf->level == 6 && |
d0dabf7e | 4123 | j == sh->qd_idx) |
f416885e | 4124 | continue; |
784052ec | 4125 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4126 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 4127 | skipped_disk = 1; |
52c03291 N |
4128 | continue; |
4129 | } | |
4130 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
4131 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
4132 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
4133 | } | |
a9f326eb | 4134 | if (!skipped_disk) { |
52c03291 N |
4135 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
4136 | set_bit(STRIPE_HANDLE, &sh->state); | |
4137 | } | |
ab69ae12 | 4138 | list_add(&sh->lru, &stripes); |
52c03291 N |
4139 | } |
4140 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4141 | if (mddev->delta_disks < 0) |
7a661381 | 4142 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 4143 | else |
7a661381 | 4144 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
4145 | spin_unlock_irq(&conf->device_lock); |
4146 | /* Ok, those stripe are ready. We can start scheduling | |
4147 | * reads on the source stripes. | |
4148 | * The source stripes are determined by mapping the first and last | |
4149 | * block on the destination stripes. | |
4150 | */ | |
52c03291 | 4151 | first_sector = |
ec32a2bd | 4152 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 4153 | 1, &dd_idx, NULL); |
52c03291 | 4154 | last_sector = |
0e6e0271 | 4155 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 4156 | * new_data_disks - 1), |
911d4ee8 | 4157 | 1, &dd_idx, NULL); |
58c0fed4 AN |
4158 | if (last_sector >= mddev->dev_sectors) |
4159 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 4160 | while (first_sector <= last_sector) { |
a8c906ca | 4161 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
4162 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4163 | set_bit(STRIPE_HANDLE, &sh->state); | |
4164 | release_stripe(sh); | |
4165 | first_sector += STRIPE_SECTORS; | |
4166 | } | |
ab69ae12 N |
4167 | /* Now that the sources are clearly marked, we can release |
4168 | * the destination stripes | |
4169 | */ | |
4170 | while (!list_empty(&stripes)) { | |
4171 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
4172 | list_del_init(&sh->lru); | |
4173 | release_stripe(sh); | |
4174 | } | |
c6207277 N |
4175 | /* If this takes us to the resync_max point where we have to pause, |
4176 | * then we need to write out the superblock. | |
4177 | */ | |
7a661381 | 4178 | sector_nr += reshape_sectors; |
c03f6a19 N |
4179 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
4180 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
4181 | /* Cannot proceed until we've updated the superblock... */ |
4182 | wait_event(conf->wait_for_overlap, | |
4183 | atomic_read(&conf->reshape_stripes) == 0); | |
fef9c61f | 4184 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4185 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4186 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
4187 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
4188 | md_wakeup_thread(mddev->thread); | |
4189 | wait_event(mddev->sb_wait, | |
4190 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
4191 | || kthread_should_stop()); | |
4192 | spin_lock_irq(&conf->device_lock); | |
fef9c61f | 4193 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
4194 | spin_unlock_irq(&conf->device_lock); |
4195 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4196 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 4197 | } |
7a661381 | 4198 | return reshape_sectors; |
52c03291 N |
4199 | } |
4200 | ||
4201 | /* FIXME go_faster isn't used */ | |
4202 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
4203 | { | |
7b92813c | 4204 | raid5_conf_t *conf = mddev->private; |
52c03291 | 4205 | struct stripe_head *sh; |
58c0fed4 | 4206 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 4207 | sector_t sync_blocks; |
16a53ecc N |
4208 | int still_degraded = 0; |
4209 | int i; | |
1da177e4 | 4210 | |
72626685 | 4211 | if (sector_nr >= max_sector) { |
1da177e4 | 4212 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 4213 | |
29269553 N |
4214 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
4215 | end_reshape(conf); | |
4216 | return 0; | |
4217 | } | |
72626685 N |
4218 | |
4219 | if (mddev->curr_resync < max_sector) /* aborted */ | |
4220 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
4221 | &sync_blocks, 1); | |
16a53ecc | 4222 | else /* completed sync */ |
72626685 N |
4223 | conf->fullsync = 0; |
4224 | bitmap_close_sync(mddev->bitmap); | |
4225 | ||
1da177e4 LT |
4226 | return 0; |
4227 | } | |
ccfcc3c1 | 4228 | |
64bd660b N |
4229 | /* Allow raid5_quiesce to complete */ |
4230 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
4231 | ||
52c03291 N |
4232 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
4233 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 4234 | |
c6207277 N |
4235 | /* No need to check resync_max as we never do more than one |
4236 | * stripe, and as resync_max will always be on a chunk boundary, | |
4237 | * if the check in md_do_sync didn't fire, there is no chance | |
4238 | * of overstepping resync_max here | |
4239 | */ | |
4240 | ||
16a53ecc | 4241 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
4242 | * to resync, then assert that we are finished, because there is |
4243 | * nothing we can do. | |
4244 | */ | |
3285edf1 | 4245 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 4246 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 4247 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 4248 | *skipped = 1; |
1da177e4 LT |
4249 | return rv; |
4250 | } | |
72626685 | 4251 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 4252 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
4253 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
4254 | /* we can skip this block, and probably more */ | |
4255 | sync_blocks /= STRIPE_SECTORS; | |
4256 | *skipped = 1; | |
4257 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
4258 | } | |
1da177e4 | 4259 | |
b47490c9 N |
4260 | |
4261 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
4262 | ||
a8c906ca | 4263 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 4264 | if (sh == NULL) { |
a8c906ca | 4265 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 4266 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 4267 | * is trying to get access |
1da177e4 | 4268 | */ |
66c006a5 | 4269 | schedule_timeout_uninterruptible(1); |
1da177e4 | 4270 | } |
16a53ecc N |
4271 | /* Need to check if array will still be degraded after recovery/resync |
4272 | * We don't need to check the 'failed' flag as when that gets set, | |
4273 | * recovery aborts. | |
4274 | */ | |
f001a70c | 4275 | for (i = 0; i < conf->raid_disks; i++) |
16a53ecc N |
4276 | if (conf->disks[i].rdev == NULL) |
4277 | still_degraded = 1; | |
4278 | ||
4279 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
4280 | ||
83206d66 | 4281 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
1da177e4 | 4282 | |
1442577b | 4283 | handle_stripe(sh); |
1da177e4 LT |
4284 | release_stripe(sh); |
4285 | ||
4286 | return STRIPE_SECTORS; | |
4287 | } | |
4288 | ||
46031f9a RBJ |
4289 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
4290 | { | |
4291 | /* We may not be able to submit a whole bio at once as there | |
4292 | * may not be enough stripe_heads available. | |
4293 | * We cannot pre-allocate enough stripe_heads as we may need | |
4294 | * more than exist in the cache (if we allow ever large chunks). | |
4295 | * So we do one stripe head at a time and record in | |
4296 | * ->bi_hw_segments how many have been done. | |
4297 | * | |
4298 | * We *know* that this entire raid_bio is in one chunk, so | |
4299 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
4300 | */ | |
4301 | struct stripe_head *sh; | |
911d4ee8 | 4302 | int dd_idx; |
46031f9a RBJ |
4303 | sector_t sector, logical_sector, last_sector; |
4304 | int scnt = 0; | |
4305 | int remaining; | |
4306 | int handled = 0; | |
4307 | ||
4308 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 4309 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 4310 | 0, &dd_idx, NULL); |
46031f9a RBJ |
4311 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
4312 | ||
4313 | for (; logical_sector < last_sector; | |
387bb173 NB |
4314 | logical_sector += STRIPE_SECTORS, |
4315 | sector += STRIPE_SECTORS, | |
4316 | scnt++) { | |
46031f9a | 4317 | |
960e739d | 4318 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
4319 | /* already done this stripe */ |
4320 | continue; | |
4321 | ||
a8c906ca | 4322 | sh = get_active_stripe(conf, sector, 0, 1, 0); |
46031f9a RBJ |
4323 | |
4324 | if (!sh) { | |
4325 | /* failed to get a stripe - must wait */ | |
960e739d | 4326 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
4327 | conf->retry_read_aligned = raid_bio; |
4328 | return handled; | |
4329 | } | |
4330 | ||
4331 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
4332 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
4333 | release_stripe(sh); | |
960e739d | 4334 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
4335 | conf->retry_read_aligned = raid_bio; |
4336 | return handled; | |
4337 | } | |
4338 | ||
36d1c647 | 4339 | handle_stripe(sh); |
46031f9a RBJ |
4340 | release_stripe(sh); |
4341 | handled++; | |
4342 | } | |
4343 | spin_lock_irq(&conf->device_lock); | |
960e739d | 4344 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 4345 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
4346 | if (remaining == 0) |
4347 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
4348 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4349 | wake_up(&conf->wait_for_stripe); | |
4350 | return handled; | |
4351 | } | |
4352 | ||
46031f9a | 4353 | |
1da177e4 LT |
4354 | /* |
4355 | * This is our raid5 kernel thread. | |
4356 | * | |
4357 | * We scan the hash table for stripes which can be handled now. | |
4358 | * During the scan, completed stripes are saved for us by the interrupt | |
4359 | * handler, so that they will not have to wait for our next wakeup. | |
4360 | */ | |
6ed3003c | 4361 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4362 | { |
4363 | struct stripe_head *sh; | |
070ec55d | 4364 | raid5_conf_t *conf = mddev->private; |
1da177e4 | 4365 | int handled; |
e1dfa0a2 | 4366 | struct blk_plug plug; |
1da177e4 | 4367 | |
45b4233c | 4368 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4369 | |
4370 | md_check_recovery(mddev); | |
1da177e4 | 4371 | |
e1dfa0a2 | 4372 | blk_start_plug(&plug); |
1da177e4 LT |
4373 | handled = 0; |
4374 | spin_lock_irq(&conf->device_lock); | |
4375 | while (1) { | |
46031f9a | 4376 | struct bio *bio; |
1da177e4 | 4377 | |
7c13edc8 N |
4378 | if (atomic_read(&mddev->plug_cnt) == 0 && |
4379 | !list_empty(&conf->bitmap_list)) { | |
4380 | /* Now is a good time to flush some bitmap updates */ | |
4381 | conf->seq_flush++; | |
700e432d | 4382 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4383 | bitmap_unplug(mddev->bitmap); |
700e432d | 4384 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 4385 | conf->seq_write = conf->seq_flush; |
72626685 N |
4386 | activate_bit_delay(conf); |
4387 | } | |
7c13edc8 N |
4388 | if (atomic_read(&mddev->plug_cnt) == 0) |
4389 | raid5_activate_delayed(conf); | |
72626685 | 4390 | |
46031f9a RBJ |
4391 | while ((bio = remove_bio_from_retry(conf))) { |
4392 | int ok; | |
4393 | spin_unlock_irq(&conf->device_lock); | |
4394 | ok = retry_aligned_read(conf, bio); | |
4395 | spin_lock_irq(&conf->device_lock); | |
4396 | if (!ok) | |
4397 | break; | |
4398 | handled++; | |
4399 | } | |
4400 | ||
8b3e6cdc DW |
4401 | sh = __get_priority_stripe(conf); |
4402 | ||
c9f21aaf | 4403 | if (!sh) |
1da177e4 | 4404 | break; |
1da177e4 LT |
4405 | spin_unlock_irq(&conf->device_lock); |
4406 | ||
4407 | handled++; | |
417b8d4a DW |
4408 | handle_stripe(sh); |
4409 | release_stripe(sh); | |
4410 | cond_resched(); | |
1da177e4 LT |
4411 | |
4412 | spin_lock_irq(&conf->device_lock); | |
4413 | } | |
45b4233c | 4414 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4415 | |
4416 | spin_unlock_irq(&conf->device_lock); | |
4417 | ||
c9f21aaf | 4418 | async_tx_issue_pending_all(); |
e1dfa0a2 | 4419 | blk_finish_plug(&plug); |
1da177e4 | 4420 | |
45b4233c | 4421 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4422 | } |
4423 | ||
3f294f4f | 4424 | static ssize_t |
007583c9 | 4425 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4426 | { |
070ec55d | 4427 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4428 | if (conf) |
4429 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4430 | else | |
4431 | return 0; | |
3f294f4f N |
4432 | } |
4433 | ||
c41d4ac4 N |
4434 | int |
4435 | raid5_set_cache_size(mddev_t *mddev, int size) | |
3f294f4f | 4436 | { |
070ec55d | 4437 | raid5_conf_t *conf = mddev->private; |
b5470dc5 DW |
4438 | int err; |
4439 | ||
c41d4ac4 | 4440 | if (size <= 16 || size > 32768) |
3f294f4f | 4441 | return -EINVAL; |
c41d4ac4 | 4442 | while (size < conf->max_nr_stripes) { |
3f294f4f N |
4443 | if (drop_one_stripe(conf)) |
4444 | conf->max_nr_stripes--; | |
4445 | else | |
4446 | break; | |
4447 | } | |
b5470dc5 DW |
4448 | err = md_allow_write(mddev); |
4449 | if (err) | |
4450 | return err; | |
c41d4ac4 | 4451 | while (size > conf->max_nr_stripes) { |
3f294f4f N |
4452 | if (grow_one_stripe(conf)) |
4453 | conf->max_nr_stripes++; | |
4454 | else break; | |
4455 | } | |
c41d4ac4 N |
4456 | return 0; |
4457 | } | |
4458 | EXPORT_SYMBOL(raid5_set_cache_size); | |
4459 | ||
4460 | static ssize_t | |
4461 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) | |
4462 | { | |
4463 | raid5_conf_t *conf = mddev->private; | |
4464 | unsigned long new; | |
4465 | int err; | |
4466 | ||
4467 | if (len >= PAGE_SIZE) | |
4468 | return -EINVAL; | |
4469 | if (!conf) | |
4470 | return -ENODEV; | |
4471 | ||
4472 | if (strict_strtoul(page, 10, &new)) | |
4473 | return -EINVAL; | |
4474 | err = raid5_set_cache_size(mddev, new); | |
4475 | if (err) | |
4476 | return err; | |
3f294f4f N |
4477 | return len; |
4478 | } | |
007583c9 | 4479 | |
96de1e66 N |
4480 | static struct md_sysfs_entry |
4481 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4482 | raid5_show_stripe_cache_size, | |
4483 | raid5_store_stripe_cache_size); | |
3f294f4f | 4484 | |
8b3e6cdc DW |
4485 | static ssize_t |
4486 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4487 | { | |
070ec55d | 4488 | raid5_conf_t *conf = mddev->private; |
8b3e6cdc DW |
4489 | if (conf) |
4490 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4491 | else | |
4492 | return 0; | |
4493 | } | |
4494 | ||
4495 | static ssize_t | |
4496 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4497 | { | |
070ec55d | 4498 | raid5_conf_t *conf = mddev->private; |
4ef197d8 | 4499 | unsigned long new; |
8b3e6cdc DW |
4500 | if (len >= PAGE_SIZE) |
4501 | return -EINVAL; | |
4502 | if (!conf) | |
4503 | return -ENODEV; | |
4504 | ||
4ef197d8 | 4505 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4506 | return -EINVAL; |
4ef197d8 | 4507 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4508 | return -EINVAL; |
4509 | conf->bypass_threshold = new; | |
4510 | return len; | |
4511 | } | |
4512 | ||
4513 | static struct md_sysfs_entry | |
4514 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4515 | S_IRUGO | S_IWUSR, | |
4516 | raid5_show_preread_threshold, | |
4517 | raid5_store_preread_threshold); | |
4518 | ||
3f294f4f | 4519 | static ssize_t |
96de1e66 | 4520 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4521 | { |
070ec55d | 4522 | raid5_conf_t *conf = mddev->private; |
96de1e66 N |
4523 | if (conf) |
4524 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4525 | else | |
4526 | return 0; | |
3f294f4f N |
4527 | } |
4528 | ||
96de1e66 N |
4529 | static struct md_sysfs_entry |
4530 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4531 | |
007583c9 | 4532 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4533 | &raid5_stripecache_size.attr, |
4534 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4535 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4536 | NULL, |
4537 | }; | |
007583c9 N |
4538 | static struct attribute_group raid5_attrs_group = { |
4539 | .name = NULL, | |
4540 | .attrs = raid5_attrs, | |
3f294f4f N |
4541 | }; |
4542 | ||
80c3a6ce DW |
4543 | static sector_t |
4544 | raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
4545 | { | |
070ec55d | 4546 | raid5_conf_t *conf = mddev->private; |
80c3a6ce DW |
4547 | |
4548 | if (!sectors) | |
4549 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 4550 | if (!raid_disks) |
7ec05478 | 4551 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 4552 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 4553 | |
9d8f0363 | 4554 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 4555 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
4556 | return sectors * (raid_disks - conf->max_degraded); |
4557 | } | |
4558 | ||
36d1c647 DW |
4559 | static void raid5_free_percpu(raid5_conf_t *conf) |
4560 | { | |
4561 | struct raid5_percpu *percpu; | |
4562 | unsigned long cpu; | |
4563 | ||
4564 | if (!conf->percpu) | |
4565 | return; | |
4566 | ||
4567 | get_online_cpus(); | |
4568 | for_each_possible_cpu(cpu) { | |
4569 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
4570 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4571 | kfree(percpu->scribble); |
36d1c647 DW |
4572 | } |
4573 | #ifdef CONFIG_HOTPLUG_CPU | |
4574 | unregister_cpu_notifier(&conf->cpu_notify); | |
4575 | #endif | |
4576 | put_online_cpus(); | |
4577 | ||
4578 | free_percpu(conf->percpu); | |
4579 | } | |
4580 | ||
95fc17aa DW |
4581 | static void free_conf(raid5_conf_t *conf) |
4582 | { | |
4583 | shrink_stripes(conf); | |
36d1c647 | 4584 | raid5_free_percpu(conf); |
95fc17aa DW |
4585 | kfree(conf->disks); |
4586 | kfree(conf->stripe_hashtbl); | |
4587 | kfree(conf); | |
4588 | } | |
4589 | ||
36d1c647 DW |
4590 | #ifdef CONFIG_HOTPLUG_CPU |
4591 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
4592 | void *hcpu) | |
4593 | { | |
4594 | raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify); | |
4595 | long cpu = (long)hcpu; | |
4596 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
4597 | ||
4598 | switch (action) { | |
4599 | case CPU_UP_PREPARE: | |
4600 | case CPU_UP_PREPARE_FROZEN: | |
d6f38f31 | 4601 | if (conf->level == 6 && !percpu->spare_page) |
36d1c647 | 4602 | percpu->spare_page = alloc_page(GFP_KERNEL); |
d6f38f31 DW |
4603 | if (!percpu->scribble) |
4604 | percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL); | |
4605 | ||
4606 | if (!percpu->scribble || | |
4607 | (conf->level == 6 && !percpu->spare_page)) { | |
4608 | safe_put_page(percpu->spare_page); | |
4609 | kfree(percpu->scribble); | |
36d1c647 DW |
4610 | pr_err("%s: failed memory allocation for cpu%ld\n", |
4611 | __func__, cpu); | |
55af6bb5 | 4612 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
4613 | } |
4614 | break; | |
4615 | case CPU_DEAD: | |
4616 | case CPU_DEAD_FROZEN: | |
4617 | safe_put_page(percpu->spare_page); | |
d6f38f31 | 4618 | kfree(percpu->scribble); |
36d1c647 | 4619 | percpu->spare_page = NULL; |
d6f38f31 | 4620 | percpu->scribble = NULL; |
36d1c647 DW |
4621 | break; |
4622 | default: | |
4623 | break; | |
4624 | } | |
4625 | return NOTIFY_OK; | |
4626 | } | |
4627 | #endif | |
4628 | ||
4629 | static int raid5_alloc_percpu(raid5_conf_t *conf) | |
4630 | { | |
4631 | unsigned long cpu; | |
4632 | struct page *spare_page; | |
a29d8b8e | 4633 | struct raid5_percpu __percpu *allcpus; |
d6f38f31 | 4634 | void *scribble; |
36d1c647 DW |
4635 | int err; |
4636 | ||
36d1c647 DW |
4637 | allcpus = alloc_percpu(struct raid5_percpu); |
4638 | if (!allcpus) | |
4639 | return -ENOMEM; | |
4640 | conf->percpu = allcpus; | |
4641 | ||
4642 | get_online_cpus(); | |
4643 | err = 0; | |
4644 | for_each_present_cpu(cpu) { | |
d6f38f31 DW |
4645 | if (conf->level == 6) { |
4646 | spare_page = alloc_page(GFP_KERNEL); | |
4647 | if (!spare_page) { | |
4648 | err = -ENOMEM; | |
4649 | break; | |
4650 | } | |
4651 | per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page; | |
4652 | } | |
5e5e3e78 | 4653 | scribble = kmalloc(conf->scribble_len, GFP_KERNEL); |
d6f38f31 | 4654 | if (!scribble) { |
36d1c647 DW |
4655 | err = -ENOMEM; |
4656 | break; | |
4657 | } | |
d6f38f31 | 4658 | per_cpu_ptr(conf->percpu, cpu)->scribble = scribble; |
36d1c647 DW |
4659 | } |
4660 | #ifdef CONFIG_HOTPLUG_CPU | |
4661 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
4662 | conf->cpu_notify.priority = 0; | |
4663 | if (err == 0) | |
4664 | err = register_cpu_notifier(&conf->cpu_notify); | |
4665 | #endif | |
4666 | put_online_cpus(); | |
4667 | ||
4668 | return err; | |
4669 | } | |
4670 | ||
91adb564 | 4671 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4672 | { |
4673 | raid5_conf_t *conf; | |
5e5e3e78 | 4674 | int raid_disk, memory, max_disks; |
1da177e4 LT |
4675 | mdk_rdev_t *rdev; |
4676 | struct disk_info *disk; | |
1da177e4 | 4677 | |
91adb564 N |
4678 | if (mddev->new_level != 5 |
4679 | && mddev->new_level != 4 | |
4680 | && mddev->new_level != 6) { | |
0c55e022 | 4681 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4682 | mdname(mddev), mddev->new_level); |
4683 | return ERR_PTR(-EIO); | |
1da177e4 | 4684 | } |
91adb564 N |
4685 | if ((mddev->new_level == 5 |
4686 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4687 | (mddev->new_level == 6 | |
4688 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 4689 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
4690 | mdname(mddev), mddev->new_layout); |
4691 | return ERR_PTR(-EIO); | |
99c0fb5f | 4692 | } |
91adb564 | 4693 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 4694 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
4695 | mdname(mddev), mddev->raid_disks); |
4696 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4697 | } |
4698 | ||
664e7c41 AN |
4699 | if (!mddev->new_chunk_sectors || |
4700 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
4701 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
4702 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
4703 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 4704 | return ERR_PTR(-EINVAL); |
f6705578 N |
4705 | } |
4706 | ||
91adb564 N |
4707 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4708 | if (conf == NULL) | |
1da177e4 | 4709 | goto abort; |
f5efd45a DW |
4710 | spin_lock_init(&conf->device_lock); |
4711 | init_waitqueue_head(&conf->wait_for_stripe); | |
4712 | init_waitqueue_head(&conf->wait_for_overlap); | |
4713 | INIT_LIST_HEAD(&conf->handle_list); | |
4714 | INIT_LIST_HEAD(&conf->hold_list); | |
4715 | INIT_LIST_HEAD(&conf->delayed_list); | |
4716 | INIT_LIST_HEAD(&conf->bitmap_list); | |
4717 | INIT_LIST_HEAD(&conf->inactive_list); | |
4718 | atomic_set(&conf->active_stripes, 0); | |
4719 | atomic_set(&conf->preread_active_stripes, 0); | |
4720 | atomic_set(&conf->active_aligned_reads, 0); | |
4721 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
91adb564 N |
4722 | |
4723 | conf->raid_disks = mddev->raid_disks; | |
4724 | if (mddev->reshape_position == MaxSector) | |
4725 | conf->previous_raid_disks = mddev->raid_disks; | |
4726 | else | |
f6705578 | 4727 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 N |
4728 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
4729 | conf->scribble_len = scribble_len(max_disks); | |
f6705578 | 4730 | |
5e5e3e78 | 4731 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
4732 | GFP_KERNEL); |
4733 | if (!conf->disks) | |
4734 | goto abort; | |
9ffae0cf | 4735 | |
1da177e4 LT |
4736 | conf->mddev = mddev; |
4737 | ||
fccddba0 | 4738 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4739 | goto abort; |
1da177e4 | 4740 | |
36d1c647 DW |
4741 | conf->level = mddev->new_level; |
4742 | if (raid5_alloc_percpu(conf) != 0) | |
4743 | goto abort; | |
4744 | ||
0c55e022 | 4745 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4746 | |
159ec1fc | 4747 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4748 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 4749 | if (raid_disk >= max_disks |
1da177e4 LT |
4750 | || raid_disk < 0) |
4751 | continue; | |
4752 | disk = conf->disks + raid_disk; | |
4753 | ||
4754 | disk->rdev = rdev; | |
4755 | ||
b2d444d7 | 4756 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 4757 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
4758 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
4759 | " disk %d\n", | |
4760 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 4761 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
4762 | /* Cannot rely on bitmap to complete recovery */ |
4763 | conf->fullsync = 1; | |
1da177e4 LT |
4764 | } |
4765 | ||
09c9e5fa | 4766 | conf->chunk_sectors = mddev->new_chunk_sectors; |
91adb564 | 4767 | conf->level = mddev->new_level; |
16a53ecc N |
4768 | if (conf->level == 6) |
4769 | conf->max_degraded = 2; | |
4770 | else | |
4771 | conf->max_degraded = 1; | |
91adb564 | 4772 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4773 | conf->max_nr_stripes = NR_STRIPES; |
fef9c61f | 4774 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 4775 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 4776 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
4777 | conf->prev_algo = mddev->layout; |
4778 | } | |
1da177e4 | 4779 | |
91adb564 | 4780 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 4781 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
91adb564 N |
4782 | if (grow_stripes(conf, conf->max_nr_stripes)) { |
4783 | printk(KERN_ERR | |
0c55e022 N |
4784 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
4785 | mdname(mddev), memory); | |
91adb564 N |
4786 | goto abort; |
4787 | } else | |
0c55e022 N |
4788 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
4789 | mdname(mddev), memory); | |
1da177e4 | 4790 | |
0da3c619 | 4791 | conf->thread = md_register_thread(raid5d, mddev, NULL); |
91adb564 N |
4792 | if (!conf->thread) { |
4793 | printk(KERN_ERR | |
0c55e022 | 4794 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 4795 | mdname(mddev)); |
16a53ecc N |
4796 | goto abort; |
4797 | } | |
91adb564 N |
4798 | |
4799 | return conf; | |
4800 | ||
4801 | abort: | |
4802 | if (conf) { | |
95fc17aa | 4803 | free_conf(conf); |
91adb564 N |
4804 | return ERR_PTR(-EIO); |
4805 | } else | |
4806 | return ERR_PTR(-ENOMEM); | |
4807 | } | |
4808 | ||
c148ffdc N |
4809 | |
4810 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) | |
4811 | { | |
4812 | switch (algo) { | |
4813 | case ALGORITHM_PARITY_0: | |
4814 | if (raid_disk < max_degraded) | |
4815 | return 1; | |
4816 | break; | |
4817 | case ALGORITHM_PARITY_N: | |
4818 | if (raid_disk >= raid_disks - max_degraded) | |
4819 | return 1; | |
4820 | break; | |
4821 | case ALGORITHM_PARITY_0_6: | |
4822 | if (raid_disk == 0 || | |
4823 | raid_disk == raid_disks - 1) | |
4824 | return 1; | |
4825 | break; | |
4826 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
4827 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
4828 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
4829 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
4830 | if (raid_disk == raid_disks - 1) | |
4831 | return 1; | |
4832 | } | |
4833 | return 0; | |
4834 | } | |
4835 | ||
91adb564 N |
4836 | static int run(mddev_t *mddev) |
4837 | { | |
4838 | raid5_conf_t *conf; | |
9f7c2220 | 4839 | int working_disks = 0; |
c148ffdc | 4840 | int dirty_parity_disks = 0; |
91adb564 | 4841 | mdk_rdev_t *rdev; |
c148ffdc | 4842 | sector_t reshape_offset = 0; |
91adb564 | 4843 | |
8c6ac868 | 4844 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 4845 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
4846 | " -- starting background reconstruction\n", |
4847 | mdname(mddev)); | |
91adb564 N |
4848 | if (mddev->reshape_position != MaxSector) { |
4849 | /* Check that we can continue the reshape. | |
4850 | * Currently only disks can change, it must | |
4851 | * increase, and we must be past the point where | |
4852 | * a stripe over-writes itself | |
4853 | */ | |
4854 | sector_t here_new, here_old; | |
4855 | int old_disks; | |
18b00334 | 4856 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 4857 | |
88ce4930 | 4858 | if (mddev->new_level != mddev->level) { |
0c55e022 | 4859 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
4860 | "required - aborting.\n", |
4861 | mdname(mddev)); | |
4862 | return -EINVAL; | |
4863 | } | |
91adb564 N |
4864 | old_disks = mddev->raid_disks - mddev->delta_disks; |
4865 | /* reshape_position must be on a new-stripe boundary, and one | |
4866 | * further up in new geometry must map after here in old | |
4867 | * geometry. | |
4868 | */ | |
4869 | here_new = mddev->reshape_position; | |
664e7c41 | 4870 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 4871 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
4872 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
4873 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
4874 | return -EINVAL; |
4875 | } | |
c148ffdc | 4876 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
4877 | /* here_new is the stripe we will write to */ |
4878 | here_old = mddev->reshape_position; | |
9d8f0363 | 4879 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
4880 | (old_disks-max_degraded)); |
4881 | /* here_old is the first stripe that we might need to read | |
4882 | * from */ | |
67ac6011 N |
4883 | if (mddev->delta_disks == 0) { |
4884 | /* We cannot be sure it is safe to start an in-place | |
4885 | * reshape. It is only safe if user-space if monitoring | |
4886 | * and taking constant backups. | |
4887 | * mdadm always starts a situation like this in | |
4888 | * readonly mode so it can take control before | |
4889 | * allowing any writes. So just check for that. | |
4890 | */ | |
4891 | if ((here_new * mddev->new_chunk_sectors != | |
4892 | here_old * mddev->chunk_sectors) || | |
4893 | mddev->ro == 0) { | |
0c55e022 N |
4894 | printk(KERN_ERR "md/raid:%s: in-place reshape must be started" |
4895 | " in read-only mode - aborting\n", | |
4896 | mdname(mddev)); | |
67ac6011 N |
4897 | return -EINVAL; |
4898 | } | |
4899 | } else if (mddev->delta_disks < 0 | |
4900 | ? (here_new * mddev->new_chunk_sectors <= | |
4901 | here_old * mddev->chunk_sectors) | |
4902 | : (here_new * mddev->new_chunk_sectors >= | |
4903 | here_old * mddev->chunk_sectors)) { | |
91adb564 | 4904 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
4905 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
4906 | "auto-recovery - aborting.\n", | |
4907 | mdname(mddev)); | |
91adb564 N |
4908 | return -EINVAL; |
4909 | } | |
0c55e022 N |
4910 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
4911 | mdname(mddev)); | |
91adb564 N |
4912 | /* OK, we should be able to continue; */ |
4913 | } else { | |
4914 | BUG_ON(mddev->level != mddev->new_level); | |
4915 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 4916 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 4917 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 4918 | } |
91adb564 | 4919 | |
245f46c2 N |
4920 | if (mddev->private == NULL) |
4921 | conf = setup_conf(mddev); | |
4922 | else | |
4923 | conf = mddev->private; | |
4924 | ||
91adb564 N |
4925 | if (IS_ERR(conf)) |
4926 | return PTR_ERR(conf); | |
4927 | ||
4928 | mddev->thread = conf->thread; | |
4929 | conf->thread = NULL; | |
4930 | mddev->private = conf; | |
4931 | ||
4932 | /* | |
4933 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4934 | */ | |
c148ffdc N |
4935 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
4936 | if (rdev->raid_disk < 0) | |
4937 | continue; | |
2f115882 | 4938 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 4939 | working_disks++; |
2f115882 N |
4940 | continue; |
4941 | } | |
c148ffdc N |
4942 | /* This disc is not fully in-sync. However if it |
4943 | * just stored parity (beyond the recovery_offset), | |
4944 | * when we don't need to be concerned about the | |
4945 | * array being dirty. | |
4946 | * When reshape goes 'backwards', we never have | |
4947 | * partially completed devices, so we only need | |
4948 | * to worry about reshape going forwards. | |
4949 | */ | |
4950 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
4951 | if (mddev->major_version == 0 && | |
4952 | mddev->minor_version > 90) | |
4953 | rdev->recovery_offset = reshape_offset; | |
4954 | ||
c148ffdc N |
4955 | if (rdev->recovery_offset < reshape_offset) { |
4956 | /* We need to check old and new layout */ | |
4957 | if (!only_parity(rdev->raid_disk, | |
4958 | conf->algorithm, | |
4959 | conf->raid_disks, | |
4960 | conf->max_degraded)) | |
4961 | continue; | |
4962 | } | |
4963 | if (!only_parity(rdev->raid_disk, | |
4964 | conf->prev_algo, | |
4965 | conf->previous_raid_disks, | |
4966 | conf->max_degraded)) | |
4967 | continue; | |
4968 | dirty_parity_disks++; | |
4969 | } | |
91adb564 | 4970 | |
5e5e3e78 N |
4971 | mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks) |
4972 | - working_disks); | |
91adb564 | 4973 | |
674806d6 | 4974 | if (has_failed(conf)) { |
0c55e022 | 4975 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 4976 | " (%d/%d failed)\n", |
02c2de8c | 4977 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4978 | goto abort; |
4979 | } | |
4980 | ||
91adb564 | 4981 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 4982 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
4983 | mddev->resync_max_sectors = mddev->dev_sectors; |
4984 | ||
c148ffdc | 4985 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 4986 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4987 | if (mddev->ok_start_degraded) |
4988 | printk(KERN_WARNING | |
0c55e022 N |
4989 | "md/raid:%s: starting dirty degraded array" |
4990 | " - data corruption possible.\n", | |
6ff8d8ec N |
4991 | mdname(mddev)); |
4992 | else { | |
4993 | printk(KERN_ERR | |
0c55e022 | 4994 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
4995 | mdname(mddev)); |
4996 | goto abort; | |
4997 | } | |
1da177e4 LT |
4998 | } |
4999 | ||
1da177e4 | 5000 | if (mddev->degraded == 0) |
0c55e022 N |
5001 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
5002 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
5003 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
5004 | mddev->new_layout); | |
1da177e4 | 5005 | else |
0c55e022 N |
5006 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
5007 | " out of %d devices, algorithm %d\n", | |
5008 | mdname(mddev), conf->level, | |
5009 | mddev->raid_disks - mddev->degraded, | |
5010 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
5011 | |
5012 | print_raid5_conf(conf); | |
5013 | ||
fef9c61f | 5014 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 5015 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
5016 | atomic_set(&conf->reshape_stripes, 0); |
5017 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
5018 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5019 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5020 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5021 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5022 | "reshape"); |
f6705578 N |
5023 | } |
5024 | ||
1da177e4 LT |
5025 | |
5026 | /* Ok, everything is just fine now */ | |
a64c876f N |
5027 | if (mddev->to_remove == &raid5_attrs_group) |
5028 | mddev->to_remove = NULL; | |
00bcb4ac N |
5029 | else if (mddev->kobj.sd && |
5030 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 5031 | printk(KERN_WARNING |
4a5add49 | 5032 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 5033 | mdname(mddev)); |
4a5add49 | 5034 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 5035 | |
4a5add49 | 5036 | if (mddev->queue) { |
9f7c2220 | 5037 | int chunk_size; |
4a5add49 N |
5038 | /* read-ahead size must cover two whole stripes, which |
5039 | * is 2 * (datadisks) * chunksize where 'n' is the | |
5040 | * number of raid devices | |
5041 | */ | |
5042 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
5043 | int stripe = data_disks * | |
5044 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
5045 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
5046 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 5047 | |
4a5add49 | 5048 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
f022b2fd | 5049 | |
11d8a6e3 N |
5050 | mddev->queue->backing_dev_info.congested_data = mddev; |
5051 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; | |
7a5febe9 | 5052 | |
9f7c2220 N |
5053 | chunk_size = mddev->chunk_sectors << 9; |
5054 | blk_queue_io_min(mddev->queue, chunk_size); | |
5055 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5056 | (conf->raid_disks - conf->max_degraded)); | |
8f6c2e4b | 5057 | |
9f7c2220 N |
5058 | list_for_each_entry(rdev, &mddev->disks, same_set) |
5059 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
5060 | rdev->data_offset << 9); | |
5061 | } | |
23032a0e | 5062 | |
1da177e4 LT |
5063 | return 0; |
5064 | abort: | |
e0cf8f04 | 5065 | md_unregister_thread(mddev->thread); |
91adb564 | 5066 | mddev->thread = NULL; |
1da177e4 LT |
5067 | if (conf) { |
5068 | print_raid5_conf(conf); | |
95fc17aa | 5069 | free_conf(conf); |
1da177e4 LT |
5070 | } |
5071 | mddev->private = NULL; | |
0c55e022 | 5072 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
5073 | return -EIO; |
5074 | } | |
5075 | ||
3f294f4f | 5076 | static int stop(mddev_t *mddev) |
1da177e4 | 5077 | { |
7b92813c | 5078 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
5079 | |
5080 | md_unregister_thread(mddev->thread); | |
5081 | mddev->thread = NULL; | |
11d8a6e3 N |
5082 | if (mddev->queue) |
5083 | mddev->queue->backing_dev_info.congested_fn = NULL; | |
95fc17aa | 5084 | free_conf(conf); |
a64c876f N |
5085 | mddev->private = NULL; |
5086 | mddev->to_remove = &raid5_attrs_group; | |
1da177e4 LT |
5087 | return 0; |
5088 | } | |
5089 | ||
45b4233c | 5090 | #ifdef DEBUG |
d710e138 | 5091 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
5092 | { |
5093 | int i; | |
5094 | ||
16a53ecc N |
5095 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
5096 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
5097 | seq_printf(seq, "sh %llu, count %d.\n", | |
5098 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
5099 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 5100 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
5101 | seq_printf(seq, "(cache%d: %p %ld) ", |
5102 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 5103 | } |
16a53ecc | 5104 | seq_printf(seq, "\n"); |
1da177e4 LT |
5105 | } |
5106 | ||
d710e138 | 5107 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
5108 | { |
5109 | struct stripe_head *sh; | |
fccddba0 | 5110 | struct hlist_node *hn; |
1da177e4 LT |
5111 | int i; |
5112 | ||
5113 | spin_lock_irq(&conf->device_lock); | |
5114 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 5115 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
5116 | if (sh->raid_conf != conf) |
5117 | continue; | |
16a53ecc | 5118 | print_sh(seq, sh); |
1da177e4 LT |
5119 | } |
5120 | } | |
5121 | spin_unlock_irq(&conf->device_lock); | |
5122 | } | |
5123 | #endif | |
5124 | ||
d710e138 | 5125 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 | 5126 | { |
7b92813c | 5127 | raid5_conf_t *conf = mddev->private; |
1da177e4 LT |
5128 | int i; |
5129 | ||
9d8f0363 AN |
5130 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
5131 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 5132 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
5133 | for (i = 0; i < conf->raid_disks; i++) |
5134 | seq_printf (seq, "%s", | |
5135 | conf->disks[i].rdev && | |
b2d444d7 | 5136 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 5137 | seq_printf (seq, "]"); |
45b4233c | 5138 | #ifdef DEBUG |
16a53ecc N |
5139 | seq_printf (seq, "\n"); |
5140 | printall(seq, conf); | |
1da177e4 LT |
5141 | #endif |
5142 | } | |
5143 | ||
5144 | static void print_raid5_conf (raid5_conf_t *conf) | |
5145 | { | |
5146 | int i; | |
5147 | struct disk_info *tmp; | |
5148 | ||
0c55e022 | 5149 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
5150 | if (!conf) { |
5151 | printk("(conf==NULL)\n"); | |
5152 | return; | |
5153 | } | |
0c55e022 N |
5154 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
5155 | conf->raid_disks, | |
5156 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
5157 | |
5158 | for (i = 0; i < conf->raid_disks; i++) { | |
5159 | char b[BDEVNAME_SIZE]; | |
5160 | tmp = conf->disks + i; | |
5161 | if (tmp->rdev) | |
0c55e022 N |
5162 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
5163 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
5164 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
5165 | } |
5166 | } | |
5167 | ||
5168 | static int raid5_spare_active(mddev_t *mddev) | |
5169 | { | |
5170 | int i; | |
5171 | raid5_conf_t *conf = mddev->private; | |
5172 | struct disk_info *tmp; | |
6b965620 N |
5173 | int count = 0; |
5174 | unsigned long flags; | |
1da177e4 LT |
5175 | |
5176 | for (i = 0; i < conf->raid_disks; i++) { | |
5177 | tmp = conf->disks + i; | |
5178 | if (tmp->rdev | |
70fffd0b | 5179 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 5180 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 5181 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 5182 | count++; |
43c73ca4 | 5183 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
5184 | } |
5185 | } | |
6b965620 N |
5186 | spin_lock_irqsave(&conf->device_lock, flags); |
5187 | mddev->degraded -= count; | |
5188 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 5189 | print_raid5_conf(conf); |
6b965620 | 5190 | return count; |
1da177e4 LT |
5191 | } |
5192 | ||
5193 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
5194 | { | |
5195 | raid5_conf_t *conf = mddev->private; | |
5196 | int err = 0; | |
5197 | mdk_rdev_t *rdev; | |
5198 | struct disk_info *p = conf->disks + number; | |
5199 | ||
5200 | print_raid5_conf(conf); | |
5201 | rdev = p->rdev; | |
5202 | if (rdev) { | |
ec32a2bd N |
5203 | if (number >= conf->raid_disks && |
5204 | conf->reshape_progress == MaxSector) | |
5205 | clear_bit(In_sync, &rdev->flags); | |
5206 | ||
b2d444d7 | 5207 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
5208 | atomic_read(&rdev->nr_pending)) { |
5209 | err = -EBUSY; | |
5210 | goto abort; | |
5211 | } | |
dfc70645 N |
5212 | /* Only remove non-faulty devices if recovery |
5213 | * isn't possible. | |
5214 | */ | |
5215 | if (!test_bit(Faulty, &rdev->flags) && | |
674806d6 | 5216 | !has_failed(conf) && |
ec32a2bd | 5217 | number < conf->raid_disks) { |
dfc70645 N |
5218 | err = -EBUSY; |
5219 | goto abort; | |
5220 | } | |
1da177e4 | 5221 | p->rdev = NULL; |
fbd568a3 | 5222 | synchronize_rcu(); |
1da177e4 LT |
5223 | if (atomic_read(&rdev->nr_pending)) { |
5224 | /* lost the race, try later */ | |
5225 | err = -EBUSY; | |
5226 | p->rdev = rdev; | |
5227 | } | |
5228 | } | |
5229 | abort: | |
5230 | ||
5231 | print_raid5_conf(conf); | |
5232 | return err; | |
5233 | } | |
5234 | ||
5235 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
5236 | { | |
5237 | raid5_conf_t *conf = mddev->private; | |
199050ea | 5238 | int err = -EEXIST; |
1da177e4 LT |
5239 | int disk; |
5240 | struct disk_info *p; | |
6c2fce2e NB |
5241 | int first = 0; |
5242 | int last = conf->raid_disks - 1; | |
1da177e4 | 5243 | |
674806d6 | 5244 | if (has_failed(conf)) |
1da177e4 | 5245 | /* no point adding a device */ |
199050ea | 5246 | return -EINVAL; |
1da177e4 | 5247 | |
6c2fce2e NB |
5248 | if (rdev->raid_disk >= 0) |
5249 | first = last = rdev->raid_disk; | |
1da177e4 LT |
5250 | |
5251 | /* | |
16a53ecc N |
5252 | * find the disk ... but prefer rdev->saved_raid_disk |
5253 | * if possible. | |
1da177e4 | 5254 | */ |
16a53ecc | 5255 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 5256 | rdev->saved_raid_disk >= first && |
16a53ecc N |
5257 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5258 | disk = rdev->saved_raid_disk; | |
5259 | else | |
6c2fce2e NB |
5260 | disk = first; |
5261 | for ( ; disk <= last ; disk++) | |
1da177e4 | 5262 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 5263 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 5264 | rdev->raid_disk = disk; |
199050ea | 5265 | err = 0; |
72626685 N |
5266 | if (rdev->saved_raid_disk != disk) |
5267 | conf->fullsync = 1; | |
d6065f7b | 5268 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
5269 | break; |
5270 | } | |
5271 | print_raid5_conf(conf); | |
199050ea | 5272 | return err; |
1da177e4 LT |
5273 | } |
5274 | ||
5275 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
5276 | { | |
5277 | /* no resync is happening, and there is enough space | |
5278 | * on all devices, so we can resize. | |
5279 | * We need to make sure resync covers any new space. | |
5280 | * If the array is shrinking we should possibly wait until | |
5281 | * any io in the removed space completes, but it hardly seems | |
5282 | * worth it. | |
5283 | */ | |
9d8f0363 | 5284 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
1f403624 DW |
5285 | md_set_array_sectors(mddev, raid5_size(mddev, sectors, |
5286 | mddev->raid_disks)); | |
b522adcd DW |
5287 | if (mddev->array_sectors > |
5288 | raid5_size(mddev, sectors, mddev->raid_disks)) | |
5289 | return -EINVAL; | |
f233ea5c | 5290 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 5291 | revalidate_disk(mddev->gendisk); |
b098636c N |
5292 | if (sectors > mddev->dev_sectors && |
5293 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 5294 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
5295 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
5296 | } | |
58c0fed4 | 5297 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 5298 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
5299 | return 0; |
5300 | } | |
5301 | ||
01ee22b4 N |
5302 | static int check_stripe_cache(mddev_t *mddev) |
5303 | { | |
5304 | /* Can only proceed if there are plenty of stripe_heads. | |
5305 | * We need a minimum of one full stripe,, and for sensible progress | |
5306 | * it is best to have about 4 times that. | |
5307 | * If we require 4 times, then the default 256 4K stripe_heads will | |
5308 | * allow for chunk sizes up to 256K, which is probably OK. | |
5309 | * If the chunk size is greater, user-space should request more | |
5310 | * stripe_heads first. | |
5311 | */ | |
5312 | raid5_conf_t *conf = mddev->private; | |
5313 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5314 | > conf->max_nr_stripes || | |
5315 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
5316 | > conf->max_nr_stripes) { | |
0c55e022 N |
5317 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
5318 | mdname(mddev), | |
01ee22b4 N |
5319 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
5320 | / STRIPE_SIZE)*4); | |
5321 | return 0; | |
5322 | } | |
5323 | return 1; | |
5324 | } | |
5325 | ||
50ac168a | 5326 | static int check_reshape(mddev_t *mddev) |
29269553 | 5327 | { |
070ec55d | 5328 | raid5_conf_t *conf = mddev->private; |
29269553 | 5329 | |
88ce4930 N |
5330 | if (mddev->delta_disks == 0 && |
5331 | mddev->new_layout == mddev->layout && | |
664e7c41 | 5332 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 5333 | return 0; /* nothing to do */ |
dba034ee N |
5334 | if (mddev->bitmap) |
5335 | /* Cannot grow a bitmap yet */ | |
5336 | return -EBUSY; | |
674806d6 | 5337 | if (has_failed(conf)) |
ec32a2bd N |
5338 | return -EINVAL; |
5339 | if (mddev->delta_disks < 0) { | |
5340 | /* We might be able to shrink, but the devices must | |
5341 | * be made bigger first. | |
5342 | * For raid6, 4 is the minimum size. | |
5343 | * Otherwise 2 is the minimum | |
5344 | */ | |
5345 | int min = 2; | |
5346 | if (mddev->level == 6) | |
5347 | min = 4; | |
5348 | if (mddev->raid_disks + mddev->delta_disks < min) | |
5349 | return -EINVAL; | |
5350 | } | |
29269553 | 5351 | |
01ee22b4 | 5352 | if (!check_stripe_cache(mddev)) |
29269553 | 5353 | return -ENOSPC; |
29269553 | 5354 | |
ec32a2bd | 5355 | return resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
63c70c4f N |
5356 | } |
5357 | ||
5358 | static int raid5_start_reshape(mddev_t *mddev) | |
5359 | { | |
070ec55d | 5360 | raid5_conf_t *conf = mddev->private; |
63c70c4f | 5361 | mdk_rdev_t *rdev; |
63c70c4f | 5362 | int spares = 0; |
c04be0aa | 5363 | unsigned long flags; |
63c70c4f | 5364 | |
f416885e | 5365 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
5366 | return -EBUSY; |
5367 | ||
01ee22b4 N |
5368 | if (!check_stripe_cache(mddev)) |
5369 | return -ENOSPC; | |
5370 | ||
159ec1fc | 5371 | list_for_each_entry(rdev, &mddev->disks, same_set) |
469518a3 N |
5372 | if (!test_bit(In_sync, &rdev->flags) |
5373 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 5374 | spares++; |
63c70c4f | 5375 | |
f416885e | 5376 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
5377 | /* Not enough devices even to make a degraded array |
5378 | * of that size | |
5379 | */ | |
5380 | return -EINVAL; | |
5381 | ||
ec32a2bd N |
5382 | /* Refuse to reduce size of the array. Any reductions in |
5383 | * array size must be through explicit setting of array_size | |
5384 | * attribute. | |
5385 | */ | |
5386 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
5387 | < mddev->array_sectors) { | |
0c55e022 | 5388 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
5389 | "before number of disks\n", mdname(mddev)); |
5390 | return -EINVAL; | |
5391 | } | |
5392 | ||
f6705578 | 5393 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
5394 | spin_lock_irq(&conf->device_lock); |
5395 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 5396 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
5397 | conf->prev_chunk_sectors = conf->chunk_sectors; |
5398 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
5399 | conf->prev_algo = conf->algorithm; |
5400 | conf->algorithm = mddev->new_layout; | |
fef9c61f N |
5401 | if (mddev->delta_disks < 0) |
5402 | conf->reshape_progress = raid5_size(mddev, 0, 0); | |
5403 | else | |
5404 | conf->reshape_progress = 0; | |
5405 | conf->reshape_safe = conf->reshape_progress; | |
86b42c71 | 5406 | conf->generation++; |
29269553 N |
5407 | spin_unlock_irq(&conf->device_lock); |
5408 | ||
5409 | /* Add some new drives, as many as will fit. | |
5410 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
5411 | * Don't add devices if we are reducing the number of |
5412 | * devices in the array. This is because it is not possible | |
5413 | * to correctly record the "partially reconstructed" state of | |
5414 | * such devices during the reshape and confusion could result. | |
29269553 | 5415 | */ |
87a8dec9 N |
5416 | if (mddev->delta_disks >= 0) { |
5417 | int added_devices = 0; | |
5418 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
5419 | if (rdev->raid_disk < 0 && | |
5420 | !test_bit(Faulty, &rdev->flags)) { | |
5421 | if (raid5_add_disk(mddev, rdev) == 0) { | |
5422 | char nm[20]; | |
5423 | if (rdev->raid_disk | |
5424 | >= conf->previous_raid_disks) { | |
5425 | set_bit(In_sync, &rdev->flags); | |
5426 | added_devices++; | |
5427 | } else | |
5428 | rdev->recovery_offset = 0; | |
5429 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5430 | if (sysfs_create_link(&mddev->kobj, | |
5431 | &rdev->kobj, nm)) | |
5432 | /* Failure here is OK */; | |
50da0840 | 5433 | } |
87a8dec9 N |
5434 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
5435 | && !test_bit(Faulty, &rdev->flags)) { | |
5436 | /* This is a spare that was manually added */ | |
5437 | set_bit(In_sync, &rdev->flags); | |
5438 | added_devices++; | |
5439 | } | |
29269553 | 5440 | |
87a8dec9 N |
5441 | /* When a reshape changes the number of devices, |
5442 | * ->degraded is measured against the larger of the | |
5443 | * pre and post number of devices. | |
5444 | */ | |
ec32a2bd | 5445 | spin_lock_irqsave(&conf->device_lock, flags); |
9eb07c25 | 5446 | mddev->degraded += (conf->raid_disks - conf->previous_raid_disks) |
ec32a2bd N |
5447 | - added_devices; |
5448 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5449 | } | |
63c70c4f | 5450 | mddev->raid_disks = conf->raid_disks; |
e516402c | 5451 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 5452 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 5453 | |
29269553 N |
5454 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
5455 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
5456 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
5457 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
5458 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 5459 | "reshape"); |
29269553 N |
5460 | if (!mddev->sync_thread) { |
5461 | mddev->recovery = 0; | |
5462 | spin_lock_irq(&conf->device_lock); | |
5463 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
fef9c61f | 5464 | conf->reshape_progress = MaxSector; |
29269553 N |
5465 | spin_unlock_irq(&conf->device_lock); |
5466 | return -EAGAIN; | |
5467 | } | |
c8f517c4 | 5468 | conf->reshape_checkpoint = jiffies; |
29269553 N |
5469 | md_wakeup_thread(mddev->sync_thread); |
5470 | md_new_event(mddev); | |
5471 | return 0; | |
5472 | } | |
29269553 | 5473 | |
ec32a2bd N |
5474 | /* This is called from the reshape thread and should make any |
5475 | * changes needed in 'conf' | |
5476 | */ | |
29269553 N |
5477 | static void end_reshape(raid5_conf_t *conf) |
5478 | { | |
29269553 | 5479 | |
f6705578 | 5480 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 5481 | |
f6705578 | 5482 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 5483 | conf->previous_raid_disks = conf->raid_disks; |
fef9c61f | 5484 | conf->reshape_progress = MaxSector; |
f6705578 | 5485 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 5486 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
5487 | |
5488 | /* read-ahead size must cover two whole stripes, which is | |
5489 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
5490 | */ | |
4a5add49 | 5491 | if (conf->mddev->queue) { |
cea9c228 | 5492 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 5493 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 5494 | / PAGE_SIZE); |
16a53ecc N |
5495 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
5496 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
5497 | } | |
29269553 | 5498 | } |
29269553 N |
5499 | } |
5500 | ||
ec32a2bd N |
5501 | /* This is called from the raid5d thread with mddev_lock held. |
5502 | * It makes config changes to the device. | |
5503 | */ | |
cea9c228 N |
5504 | static void raid5_finish_reshape(mddev_t *mddev) |
5505 | { | |
070ec55d | 5506 | raid5_conf_t *conf = mddev->private; |
cea9c228 N |
5507 | |
5508 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
5509 | ||
ec32a2bd N |
5510 | if (mddev->delta_disks > 0) { |
5511 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
5512 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 5513 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
5514 | } else { |
5515 | int d; | |
ec32a2bd N |
5516 | mddev->degraded = conf->raid_disks; |
5517 | for (d = 0; d < conf->raid_disks ; d++) | |
5518 | if (conf->disks[d].rdev && | |
5519 | test_bit(In_sync, | |
5520 | &conf->disks[d].rdev->flags)) | |
5521 | mddev->degraded--; | |
5522 | for (d = conf->raid_disks ; | |
5523 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 N |
5524 | d++) { |
5525 | mdk_rdev_t *rdev = conf->disks[d].rdev; | |
5526 | if (rdev && raid5_remove_disk(mddev, d) == 0) { | |
5527 | char nm[20]; | |
5528 | sprintf(nm, "rd%d", rdev->raid_disk); | |
5529 | sysfs_remove_link(&mddev->kobj, nm); | |
5530 | rdev->raid_disk = -1; | |
5531 | } | |
5532 | } | |
cea9c228 | 5533 | } |
88ce4930 | 5534 | mddev->layout = conf->algorithm; |
09c9e5fa | 5535 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
5536 | mddev->reshape_position = MaxSector; |
5537 | mddev->delta_disks = 0; | |
cea9c228 N |
5538 | } |
5539 | } | |
5540 | ||
72626685 N |
5541 | static void raid5_quiesce(mddev_t *mddev, int state) |
5542 | { | |
070ec55d | 5543 | raid5_conf_t *conf = mddev->private; |
72626685 N |
5544 | |
5545 | switch(state) { | |
e464eafd N |
5546 | case 2: /* resume for a suspend */ |
5547 | wake_up(&conf->wait_for_overlap); | |
5548 | break; | |
5549 | ||
72626685 N |
5550 | case 1: /* stop all writes */ |
5551 | spin_lock_irq(&conf->device_lock); | |
64bd660b N |
5552 | /* '2' tells resync/reshape to pause so that all |
5553 | * active stripes can drain | |
5554 | */ | |
5555 | conf->quiesce = 2; | |
72626685 | 5556 | wait_event_lock_irq(conf->wait_for_stripe, |
46031f9a RBJ |
5557 | atomic_read(&conf->active_stripes) == 0 && |
5558 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 | 5559 | conf->device_lock, /* nothing */); |
64bd660b | 5560 | conf->quiesce = 1; |
72626685 | 5561 | spin_unlock_irq(&conf->device_lock); |
64bd660b N |
5562 | /* allow reshape to continue */ |
5563 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
5564 | break; |
5565 | ||
5566 | case 0: /* re-enable writes */ | |
5567 | spin_lock_irq(&conf->device_lock); | |
5568 | conf->quiesce = 0; | |
5569 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 5570 | wake_up(&conf->wait_for_overlap); |
72626685 N |
5571 | spin_unlock_irq(&conf->device_lock); |
5572 | break; | |
5573 | } | |
72626685 | 5574 | } |
b15c2e57 | 5575 | |
d562b0c4 | 5576 | |
f1b29bca | 5577 | static void *raid45_takeover_raid0(mddev_t *mddev, int level) |
54071b38 | 5578 | { |
f1b29bca | 5579 | struct raid0_private_data *raid0_priv = mddev->private; |
d76c8420 | 5580 | sector_t sectors; |
54071b38 | 5581 | |
f1b29bca DW |
5582 | /* for raid0 takeover only one zone is supported */ |
5583 | if (raid0_priv->nr_strip_zones > 1) { | |
0c55e022 N |
5584 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
5585 | mdname(mddev)); | |
f1b29bca DW |
5586 | return ERR_PTR(-EINVAL); |
5587 | } | |
5588 | ||
3b71bd93 N |
5589 | sectors = raid0_priv->strip_zone[0].zone_end; |
5590 | sector_div(sectors, raid0_priv->strip_zone[0].nb_dev); | |
5591 | mddev->dev_sectors = sectors; | |
f1b29bca | 5592 | mddev->new_level = level; |
54071b38 TM |
5593 | mddev->new_layout = ALGORITHM_PARITY_N; |
5594 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
5595 | mddev->raid_disks += 1; | |
5596 | mddev->delta_disks = 1; | |
5597 | /* make sure it will be not marked as dirty */ | |
5598 | mddev->recovery_cp = MaxSector; | |
5599 | ||
5600 | return setup_conf(mddev); | |
5601 | } | |
5602 | ||
5603 | ||
d562b0c4 N |
5604 | static void *raid5_takeover_raid1(mddev_t *mddev) |
5605 | { | |
5606 | int chunksect; | |
5607 | ||
5608 | if (mddev->raid_disks != 2 || | |
5609 | mddev->degraded > 1) | |
5610 | return ERR_PTR(-EINVAL); | |
5611 | ||
5612 | /* Should check if there are write-behind devices? */ | |
5613 | ||
5614 | chunksect = 64*2; /* 64K by default */ | |
5615 | ||
5616 | /* The array must be an exact multiple of chunksize */ | |
5617 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
5618 | chunksect >>= 1; | |
5619 | ||
5620 | if ((chunksect<<9) < STRIPE_SIZE) | |
5621 | /* array size does not allow a suitable chunk size */ | |
5622 | return ERR_PTR(-EINVAL); | |
5623 | ||
5624 | mddev->new_level = 5; | |
5625 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 5626 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
5627 | |
5628 | return setup_conf(mddev); | |
5629 | } | |
5630 | ||
fc9739c6 N |
5631 | static void *raid5_takeover_raid6(mddev_t *mddev) |
5632 | { | |
5633 | int new_layout; | |
5634 | ||
5635 | switch (mddev->layout) { | |
5636 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
5637 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
5638 | break; | |
5639 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
5640 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
5641 | break; | |
5642 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
5643 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
5644 | break; | |
5645 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
5646 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
5647 | break; | |
5648 | case ALGORITHM_PARITY_0_6: | |
5649 | new_layout = ALGORITHM_PARITY_0; | |
5650 | break; | |
5651 | case ALGORITHM_PARITY_N: | |
5652 | new_layout = ALGORITHM_PARITY_N; | |
5653 | break; | |
5654 | default: | |
5655 | return ERR_PTR(-EINVAL); | |
5656 | } | |
5657 | mddev->new_level = 5; | |
5658 | mddev->new_layout = new_layout; | |
5659 | mddev->delta_disks = -1; | |
5660 | mddev->raid_disks -= 1; | |
5661 | return setup_conf(mddev); | |
5662 | } | |
5663 | ||
d562b0c4 | 5664 | |
50ac168a | 5665 | static int raid5_check_reshape(mddev_t *mddev) |
b3546035 | 5666 | { |
88ce4930 N |
5667 | /* For a 2-drive array, the layout and chunk size can be changed |
5668 | * immediately as not restriping is needed. | |
5669 | * For larger arrays we record the new value - after validation | |
5670 | * to be used by a reshape pass. | |
b3546035 | 5671 | */ |
070ec55d | 5672 | raid5_conf_t *conf = mddev->private; |
597a711b | 5673 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 5674 | |
597a711b | 5675 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
5676 | return -EINVAL; |
5677 | if (new_chunk > 0) { | |
0ba459d2 | 5678 | if (!is_power_of_2(new_chunk)) |
b3546035 | 5679 | return -EINVAL; |
597a711b | 5680 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 5681 | return -EINVAL; |
597a711b | 5682 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
5683 | /* not factor of array size */ |
5684 | return -EINVAL; | |
5685 | } | |
5686 | ||
5687 | /* They look valid */ | |
5688 | ||
88ce4930 | 5689 | if (mddev->raid_disks == 2) { |
597a711b N |
5690 | /* can make the change immediately */ |
5691 | if (mddev->new_layout >= 0) { | |
5692 | conf->algorithm = mddev->new_layout; | |
5693 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
5694 | } |
5695 | if (new_chunk > 0) { | |
597a711b N |
5696 | conf->chunk_sectors = new_chunk ; |
5697 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
5698 | } |
5699 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
5700 | md_wakeup_thread(mddev->thread); | |
b3546035 | 5701 | } |
50ac168a | 5702 | return check_reshape(mddev); |
88ce4930 N |
5703 | } |
5704 | ||
50ac168a | 5705 | static int raid6_check_reshape(mddev_t *mddev) |
88ce4930 | 5706 | { |
597a711b | 5707 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 5708 | |
597a711b | 5709 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 5710 | return -EINVAL; |
b3546035 | 5711 | if (new_chunk > 0) { |
0ba459d2 | 5712 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 5713 | return -EINVAL; |
597a711b | 5714 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 5715 | return -EINVAL; |
597a711b | 5716 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
5717 | /* not factor of array size */ |
5718 | return -EINVAL; | |
b3546035 | 5719 | } |
88ce4930 N |
5720 | |
5721 | /* They look valid */ | |
50ac168a | 5722 | return check_reshape(mddev); |
b3546035 N |
5723 | } |
5724 | ||
d562b0c4 N |
5725 | static void *raid5_takeover(mddev_t *mddev) |
5726 | { | |
5727 | /* raid5 can take over: | |
f1b29bca | 5728 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
5729 | * raid1 - if there are two drives. We need to know the chunk size |
5730 | * raid4 - trivial - just use a raid4 layout. | |
5731 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 5732 | */ |
f1b29bca DW |
5733 | if (mddev->level == 0) |
5734 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
5735 | if (mddev->level == 1) |
5736 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
5737 | if (mddev->level == 4) { |
5738 | mddev->new_layout = ALGORITHM_PARITY_N; | |
5739 | mddev->new_level = 5; | |
5740 | return setup_conf(mddev); | |
5741 | } | |
fc9739c6 N |
5742 | if (mddev->level == 6) |
5743 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
5744 | |
5745 | return ERR_PTR(-EINVAL); | |
5746 | } | |
5747 | ||
a78d38a1 N |
5748 | static void *raid4_takeover(mddev_t *mddev) |
5749 | { | |
f1b29bca DW |
5750 | /* raid4 can take over: |
5751 | * raid0 - if there is only one strip zone | |
5752 | * raid5 - if layout is right | |
a78d38a1 | 5753 | */ |
f1b29bca DW |
5754 | if (mddev->level == 0) |
5755 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
5756 | if (mddev->level == 5 && |
5757 | mddev->layout == ALGORITHM_PARITY_N) { | |
5758 | mddev->new_layout = 0; | |
5759 | mddev->new_level = 4; | |
5760 | return setup_conf(mddev); | |
5761 | } | |
5762 | return ERR_PTR(-EINVAL); | |
5763 | } | |
d562b0c4 | 5764 | |
245f46c2 N |
5765 | static struct mdk_personality raid5_personality; |
5766 | ||
5767 | static void *raid6_takeover(mddev_t *mddev) | |
5768 | { | |
5769 | /* Currently can only take over a raid5. We map the | |
5770 | * personality to an equivalent raid6 personality | |
5771 | * with the Q block at the end. | |
5772 | */ | |
5773 | int new_layout; | |
5774 | ||
5775 | if (mddev->pers != &raid5_personality) | |
5776 | return ERR_PTR(-EINVAL); | |
5777 | if (mddev->degraded > 1) | |
5778 | return ERR_PTR(-EINVAL); | |
5779 | if (mddev->raid_disks > 253) | |
5780 | return ERR_PTR(-EINVAL); | |
5781 | if (mddev->raid_disks < 3) | |
5782 | return ERR_PTR(-EINVAL); | |
5783 | ||
5784 | switch (mddev->layout) { | |
5785 | case ALGORITHM_LEFT_ASYMMETRIC: | |
5786 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
5787 | break; | |
5788 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
5789 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
5790 | break; | |
5791 | case ALGORITHM_LEFT_SYMMETRIC: | |
5792 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
5793 | break; | |
5794 | case ALGORITHM_RIGHT_SYMMETRIC: | |
5795 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
5796 | break; | |
5797 | case ALGORITHM_PARITY_0: | |
5798 | new_layout = ALGORITHM_PARITY_0_6; | |
5799 | break; | |
5800 | case ALGORITHM_PARITY_N: | |
5801 | new_layout = ALGORITHM_PARITY_N; | |
5802 | break; | |
5803 | default: | |
5804 | return ERR_PTR(-EINVAL); | |
5805 | } | |
5806 | mddev->new_level = 6; | |
5807 | mddev->new_layout = new_layout; | |
5808 | mddev->delta_disks = 1; | |
5809 | mddev->raid_disks += 1; | |
5810 | return setup_conf(mddev); | |
5811 | } | |
5812 | ||
5813 | ||
16a53ecc N |
5814 | static struct mdk_personality raid6_personality = |
5815 | { | |
5816 | .name = "raid6", | |
5817 | .level = 6, | |
5818 | .owner = THIS_MODULE, | |
5819 | .make_request = make_request, | |
5820 | .run = run, | |
5821 | .stop = stop, | |
5822 | .status = status, | |
5823 | .error_handler = error, | |
5824 | .hot_add_disk = raid5_add_disk, | |
5825 | .hot_remove_disk= raid5_remove_disk, | |
5826 | .spare_active = raid5_spare_active, | |
5827 | .sync_request = sync_request, | |
5828 | .resize = raid5_resize, | |
80c3a6ce | 5829 | .size = raid5_size, |
50ac168a | 5830 | .check_reshape = raid6_check_reshape, |
f416885e | 5831 | .start_reshape = raid5_start_reshape, |
cea9c228 | 5832 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 5833 | .quiesce = raid5_quiesce, |
245f46c2 | 5834 | .takeover = raid6_takeover, |
16a53ecc | 5835 | }; |
2604b703 | 5836 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
5837 | { |
5838 | .name = "raid5", | |
2604b703 | 5839 | .level = 5, |
1da177e4 LT |
5840 | .owner = THIS_MODULE, |
5841 | .make_request = make_request, | |
5842 | .run = run, | |
5843 | .stop = stop, | |
5844 | .status = status, | |
5845 | .error_handler = error, | |
5846 | .hot_add_disk = raid5_add_disk, | |
5847 | .hot_remove_disk= raid5_remove_disk, | |
5848 | .spare_active = raid5_spare_active, | |
5849 | .sync_request = sync_request, | |
5850 | .resize = raid5_resize, | |
80c3a6ce | 5851 | .size = raid5_size, |
63c70c4f N |
5852 | .check_reshape = raid5_check_reshape, |
5853 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5854 | .finish_reshape = raid5_finish_reshape, |
72626685 | 5855 | .quiesce = raid5_quiesce, |
d562b0c4 | 5856 | .takeover = raid5_takeover, |
1da177e4 LT |
5857 | }; |
5858 | ||
2604b703 | 5859 | static struct mdk_personality raid4_personality = |
1da177e4 | 5860 | { |
2604b703 N |
5861 | .name = "raid4", |
5862 | .level = 4, | |
5863 | .owner = THIS_MODULE, | |
5864 | .make_request = make_request, | |
5865 | .run = run, | |
5866 | .stop = stop, | |
5867 | .status = status, | |
5868 | .error_handler = error, | |
5869 | .hot_add_disk = raid5_add_disk, | |
5870 | .hot_remove_disk= raid5_remove_disk, | |
5871 | .spare_active = raid5_spare_active, | |
5872 | .sync_request = sync_request, | |
5873 | .resize = raid5_resize, | |
80c3a6ce | 5874 | .size = raid5_size, |
3d37890b N |
5875 | .check_reshape = raid5_check_reshape, |
5876 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 5877 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 5878 | .quiesce = raid5_quiesce, |
a78d38a1 | 5879 | .takeover = raid4_takeover, |
2604b703 N |
5880 | }; |
5881 | ||
5882 | static int __init raid5_init(void) | |
5883 | { | |
16a53ecc | 5884 | register_md_personality(&raid6_personality); |
2604b703 N |
5885 | register_md_personality(&raid5_personality); |
5886 | register_md_personality(&raid4_personality); | |
5887 | return 0; | |
1da177e4 LT |
5888 | } |
5889 | ||
2604b703 | 5890 | static void raid5_exit(void) |
1da177e4 | 5891 | { |
16a53ecc | 5892 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5893 | unregister_md_personality(&raid5_personality); |
5894 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5895 | } |
5896 | ||
5897 | module_init(raid5_init); | |
5898 | module_exit(raid5_exit); | |
5899 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5900 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 5901 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
5902 | MODULE_ALIAS("md-raid5"); |
5903 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5904 | MODULE_ALIAS("md-level-5"); |
5905 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5906 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5907 | MODULE_ALIAS("md-raid6"); | |
5908 | MODULE_ALIAS("md-level-6"); | |
5909 | ||
5910 | /* This used to be two separate modules, they were: */ | |
5911 | MODULE_ALIAS("raid5"); | |
5912 | MODULE_ALIAS("raid6"); |