Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
30 | * conf->bm_write is the number of the last batch successfully written. | |
31 | * conf->bm_flush is the number of the last batch that was closed to | |
32 | * new additions. | |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
35 | * the number of the batch it will be in. This is bm_flush+1. | |
36 | * When we are ready to do a write, if that batch hasn't been written yet, | |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
91c00924 | 48 | #include <linux/async_tx.h> |
bff61975 | 49 | #include <linux/seq_file.h> |
43b2e5d8 | 50 | #include "md.h" |
bff61975 | 51 | #include "raid5.h" |
ef740c37 CH |
52 | #include "raid6.h" |
53 | #include "bitmap.h" | |
72626685 | 54 | |
1da177e4 LT |
55 | /* |
56 | * Stripe cache | |
57 | */ | |
58 | ||
59 | #define NR_STRIPES 256 | |
60 | #define STRIPE_SIZE PAGE_SIZE | |
61 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
62 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
63 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 64 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 65 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
66 | #define HASH_MASK (NR_HASH - 1) |
67 | ||
fccddba0 | 68 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
69 | |
70 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
71 | * order without overlap. There may be several bio's per stripe+device, and | |
72 | * a bio could span several devices. | |
73 | * When walking this list for a particular stripe+device, we must never proceed | |
74 | * beyond a bio that extends past this device, as the next bio might no longer | |
75 | * be valid. | |
76 | * This macro is used to determine the 'next' bio in the list, given the sector | |
77 | * of the current stripe+device | |
78 | */ | |
79 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
80 | /* | |
81 | * The following can be used to debug the driver | |
82 | */ | |
1da177e4 LT |
83 | #define RAID5_PARANOIA 1 |
84 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
85 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
86 | #else | |
87 | # define CHECK_DEVLOCK() | |
88 | #endif | |
89 | ||
45b4233c | 90 | #ifdef DEBUG |
1da177e4 LT |
91 | #define inline |
92 | #define __inline__ | |
93 | #endif | |
94 | ||
6be9d494 BS |
95 | #define printk_rl(args...) ((void) (printk_ratelimit() && printk(args))) |
96 | ||
16a53ecc N |
97 | #if !RAID6_USE_EMPTY_ZERO_PAGE |
98 | /* In .bss so it's zeroed */ | |
99 | const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256))); | |
100 | #endif | |
101 | ||
960e739d | 102 | /* |
5b99c2ff JA |
103 | * We maintain a biased count of active stripes in the bottom 16 bits of |
104 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d JA |
105 | */ |
106 | static inline int raid5_bi_phys_segments(struct bio *bio) | |
107 | { | |
5b99c2ff | 108 | return bio->bi_phys_segments & 0xffff; |
960e739d JA |
109 | } |
110 | ||
111 | static inline int raid5_bi_hw_segments(struct bio *bio) | |
112 | { | |
5b99c2ff | 113 | return (bio->bi_phys_segments >> 16) & 0xffff; |
960e739d JA |
114 | } |
115 | ||
116 | static inline int raid5_dec_bi_phys_segments(struct bio *bio) | |
117 | { | |
118 | --bio->bi_phys_segments; | |
119 | return raid5_bi_phys_segments(bio); | |
120 | } | |
121 | ||
122 | static inline int raid5_dec_bi_hw_segments(struct bio *bio) | |
123 | { | |
124 | unsigned short val = raid5_bi_hw_segments(bio); | |
125 | ||
126 | --val; | |
5b99c2ff | 127 | bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio); |
960e739d JA |
128 | return val; |
129 | } | |
130 | ||
131 | static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt) | |
132 | { | |
5b99c2ff | 133 | bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16); |
960e739d JA |
134 | } |
135 | ||
d0dabf7e N |
136 | /* Find first data disk in a raid6 stripe */ |
137 | static inline int raid6_d0(struct stripe_head *sh) | |
138 | { | |
67cc2b81 N |
139 | if (sh->ddf_layout) |
140 | /* ddf always start from first device */ | |
141 | return 0; | |
142 | /* md starts just after Q block */ | |
d0dabf7e N |
143 | if (sh->qd_idx == sh->disks - 1) |
144 | return 0; | |
145 | else | |
146 | return sh->qd_idx + 1; | |
147 | } | |
16a53ecc N |
148 | static inline int raid6_next_disk(int disk, int raid_disks) |
149 | { | |
150 | disk++; | |
151 | return (disk < raid_disks) ? disk : 0; | |
152 | } | |
a4456856 | 153 | |
d0dabf7e N |
154 | /* When walking through the disks in a raid5, starting at raid6_d0, |
155 | * We need to map each disk to a 'slot', where the data disks are slot | |
156 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
157 | * is raid_disks-1. This help does that mapping. | |
158 | */ | |
67cc2b81 N |
159 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
160 | int *count, int syndrome_disks) | |
d0dabf7e N |
161 | { |
162 | int slot; | |
67cc2b81 | 163 | |
d0dabf7e | 164 | if (idx == sh->pd_idx) |
67cc2b81 | 165 | return syndrome_disks; |
d0dabf7e | 166 | if (idx == sh->qd_idx) |
67cc2b81 | 167 | return syndrome_disks + 1; |
d0dabf7e N |
168 | slot = (*count)++; |
169 | return slot; | |
170 | } | |
171 | ||
a4456856 DW |
172 | static void return_io(struct bio *return_bi) |
173 | { | |
174 | struct bio *bi = return_bi; | |
175 | while (bi) { | |
a4456856 DW |
176 | |
177 | return_bi = bi->bi_next; | |
178 | bi->bi_next = NULL; | |
179 | bi->bi_size = 0; | |
0e13fe23 | 180 | bio_endio(bi, 0); |
a4456856 DW |
181 | bi = return_bi; |
182 | } | |
183 | } | |
184 | ||
1da177e4 LT |
185 | static void print_raid5_conf (raid5_conf_t *conf); |
186 | ||
600aa109 DW |
187 | static int stripe_operations_active(struct stripe_head *sh) |
188 | { | |
189 | return sh->check_state || sh->reconstruct_state || | |
190 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
191 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
192 | } | |
193 | ||
858119e1 | 194 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
195 | { |
196 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
197 | BUG_ON(!list_empty(&sh->lru)); |
198 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 199 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 200 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 201 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
202 | blk_plug_device(conf->mddev->queue); |
203 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 204 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 205 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
206 | blk_plug_device(conf->mddev->queue); |
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 | ||
278 | static void shrink_buffers(struct stripe_head *sh, int num) | |
279 | { | |
280 | struct page *p; | |
281 | int i; | |
282 | ||
283 | for (i=0; i<num ; i++) { | |
284 | p = sh->dev[i].page; | |
285 | if (!p) | |
286 | continue; | |
287 | sh->dev[i].page = NULL; | |
2d1f3b5d | 288 | put_page(p); |
1da177e4 LT |
289 | } |
290 | } | |
291 | ||
292 | static int grow_buffers(struct stripe_head *sh, int num) | |
293 | { | |
294 | int i; | |
295 | ||
296 | for (i=0; i<num; i++) { | |
297 | struct page *page; | |
298 | ||
299 | if (!(page = alloc_page(GFP_KERNEL))) { | |
300 | return 1; | |
301 | } | |
302 | sh->dev[i].page = page; | |
303 | } | |
304 | return 0; | |
305 | } | |
306 | ||
d710e138 | 307 | static void raid5_build_block(struct stripe_head *sh, int i); |
911d4ee8 N |
308 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
309 | struct stripe_head *sh); | |
1da177e4 | 310 | |
b5663ba4 | 311 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 LT |
312 | { |
313 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 314 | int i; |
1da177e4 | 315 | |
78bafebd ES |
316 | BUG_ON(atomic_read(&sh->count) != 0); |
317 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 318 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 319 | |
1da177e4 | 320 | CHECK_DEVLOCK(); |
45b4233c | 321 | pr_debug("init_stripe called, stripe %llu\n", |
1da177e4 LT |
322 | (unsigned long long)sh->sector); |
323 | ||
324 | remove_hash(sh); | |
16a53ecc | 325 | |
b5663ba4 | 326 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 327 | sh->sector = sector; |
911d4ee8 | 328 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
329 | sh->state = 0; |
330 | ||
7ecaa1e6 N |
331 | |
332 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
333 | struct r5dev *dev = &sh->dev[i]; |
334 | ||
d84e0f10 | 335 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 336 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 337 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 338 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 339 | dev->read, dev->towrite, dev->written, |
1da177e4 LT |
340 | test_bit(R5_LOCKED, &dev->flags)); |
341 | BUG(); | |
342 | } | |
343 | dev->flags = 0; | |
344 | raid5_build_block(sh, i); | |
345 | } | |
346 | insert_hash(conf, sh); | |
347 | } | |
348 | ||
7ecaa1e6 | 349 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, int disks) |
1da177e4 LT |
350 | { |
351 | struct stripe_head *sh; | |
fccddba0 | 352 | struct hlist_node *hn; |
1da177e4 LT |
353 | |
354 | CHECK_DEVLOCK(); | |
45b4233c | 355 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
fccddba0 | 356 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
7ecaa1e6 | 357 | if (sh->sector == sector && sh->disks == disks) |
1da177e4 | 358 | return sh; |
45b4233c | 359 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
360 | return NULL; |
361 | } | |
362 | ||
363 | static void unplug_slaves(mddev_t *mddev); | |
165125e1 | 364 | static void raid5_unplug_device(struct request_queue *q); |
1da177e4 | 365 | |
b5663ba4 N |
366 | static struct stripe_head * |
367 | get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
368 | int previous, int noblock) | |
1da177e4 LT |
369 | { |
370 | struct stripe_head *sh; | |
b5663ba4 | 371 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 372 | |
45b4233c | 373 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 LT |
374 | |
375 | spin_lock_irq(&conf->device_lock); | |
376 | ||
377 | do { | |
72626685 N |
378 | wait_event_lock_irq(conf->wait_for_stripe, |
379 | conf->quiesce == 0, | |
380 | conf->device_lock, /* nothing */); | |
7ecaa1e6 | 381 | sh = __find_stripe(conf, sector, disks); |
1da177e4 LT |
382 | if (!sh) { |
383 | if (!conf->inactive_blocked) | |
384 | sh = get_free_stripe(conf); | |
385 | if (noblock && sh == NULL) | |
386 | break; | |
387 | if (!sh) { | |
388 | conf->inactive_blocked = 1; | |
389 | wait_event_lock_irq(conf->wait_for_stripe, | |
390 | !list_empty(&conf->inactive_list) && | |
5036805b N |
391 | (atomic_read(&conf->active_stripes) |
392 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
393 | || !conf->inactive_blocked), |
394 | conf->device_lock, | |
f4370781 | 395 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
396 | ); |
397 | conf->inactive_blocked = 0; | |
398 | } else | |
b5663ba4 | 399 | init_stripe(sh, sector, previous); |
1da177e4 LT |
400 | } else { |
401 | if (atomic_read(&sh->count)) { | |
78bafebd | 402 | BUG_ON(!list_empty(&sh->lru)); |
1da177e4 LT |
403 | } else { |
404 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
405 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
406 | if (list_empty(&sh->lru) && |
407 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
408 | BUG(); |
409 | list_del_init(&sh->lru); | |
1da177e4 LT |
410 | } |
411 | } | |
412 | } while (sh == NULL); | |
413 | ||
414 | if (sh) | |
415 | atomic_inc(&sh->count); | |
416 | ||
417 | spin_unlock_irq(&conf->device_lock); | |
418 | return sh; | |
419 | } | |
420 | ||
6712ecf8 N |
421 | static void |
422 | raid5_end_read_request(struct bio *bi, int error); | |
423 | static void | |
424 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 425 | |
c4e5ac0a | 426 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 DW |
427 | { |
428 | raid5_conf_t *conf = sh->raid_conf; | |
429 | int i, disks = sh->disks; | |
430 | ||
431 | might_sleep(); | |
432 | ||
433 | for (i = disks; i--; ) { | |
434 | int rw; | |
435 | struct bio *bi; | |
436 | mdk_rdev_t *rdev; | |
437 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
438 | rw = WRITE; | |
439 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
440 | rw = READ; | |
441 | else | |
442 | continue; | |
443 | ||
444 | bi = &sh->dev[i].req; | |
445 | ||
446 | bi->bi_rw = rw; | |
447 | if (rw == WRITE) | |
448 | bi->bi_end_io = raid5_end_write_request; | |
449 | else | |
450 | bi->bi_end_io = raid5_end_read_request; | |
451 | ||
452 | rcu_read_lock(); | |
453 | rdev = rcu_dereference(conf->disks[i].rdev); | |
454 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
455 | rdev = NULL; | |
456 | if (rdev) | |
457 | atomic_inc(&rdev->nr_pending); | |
458 | rcu_read_unlock(); | |
459 | ||
460 | if (rdev) { | |
c4e5ac0a | 461 | if (s->syncing || s->expanding || s->expanded) |
91c00924 DW |
462 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
463 | ||
2b7497f0 DW |
464 | set_bit(STRIPE_IO_STARTED, &sh->state); |
465 | ||
91c00924 DW |
466 | bi->bi_bdev = rdev->bdev; |
467 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", | |
e46b272b | 468 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
469 | bi->bi_rw, i); |
470 | atomic_inc(&sh->count); | |
471 | bi->bi_sector = sh->sector + rdev->data_offset; | |
472 | bi->bi_flags = 1 << BIO_UPTODATE; | |
473 | bi->bi_vcnt = 1; | |
474 | bi->bi_max_vecs = 1; | |
475 | bi->bi_idx = 0; | |
476 | bi->bi_io_vec = &sh->dev[i].vec; | |
477 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
478 | bi->bi_io_vec[0].bv_offset = 0; | |
479 | bi->bi_size = STRIPE_SIZE; | |
480 | bi->bi_next = NULL; | |
481 | if (rw == WRITE && | |
482 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
483 | atomic_add(STRIPE_SECTORS, | |
484 | &rdev->corrected_errors); | |
485 | generic_make_request(bi); | |
486 | } else { | |
487 | if (rw == WRITE) | |
488 | set_bit(STRIPE_DEGRADED, &sh->state); | |
489 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
490 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
491 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
492 | set_bit(STRIPE_HANDLE, &sh->state); | |
493 | } | |
494 | } | |
495 | } | |
496 | ||
497 | static struct dma_async_tx_descriptor * | |
498 | async_copy_data(int frombio, struct bio *bio, struct page *page, | |
499 | sector_t sector, struct dma_async_tx_descriptor *tx) | |
500 | { | |
501 | struct bio_vec *bvl; | |
502 | struct page *bio_page; | |
503 | int i; | |
504 | int page_offset; | |
505 | ||
506 | if (bio->bi_sector >= sector) | |
507 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
508 | else | |
509 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
510 | bio_for_each_segment(bvl, bio, i) { | |
511 | int len = bio_iovec_idx(bio, i)->bv_len; | |
512 | int clen; | |
513 | int b_offset = 0; | |
514 | ||
515 | if (page_offset < 0) { | |
516 | b_offset = -page_offset; | |
517 | page_offset += b_offset; | |
518 | len -= b_offset; | |
519 | } | |
520 | ||
521 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
522 | clen = STRIPE_SIZE - page_offset; | |
523 | else | |
524 | clen = len; | |
525 | ||
526 | if (clen > 0) { | |
527 | b_offset += bio_iovec_idx(bio, i)->bv_offset; | |
528 | bio_page = bio_iovec_idx(bio, i)->bv_page; | |
529 | if (frombio) | |
530 | tx = async_memcpy(page, bio_page, page_offset, | |
531 | b_offset, clen, | |
eb0645a8 | 532 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
533 | tx, NULL, NULL); |
534 | else | |
535 | tx = async_memcpy(bio_page, page, b_offset, | |
536 | page_offset, clen, | |
eb0645a8 | 537 | ASYNC_TX_DEP_ACK, |
91c00924 DW |
538 | tx, NULL, NULL); |
539 | } | |
540 | if (clen < len) /* hit end of page */ | |
541 | break; | |
542 | page_offset += len; | |
543 | } | |
544 | ||
545 | return tx; | |
546 | } | |
547 | ||
548 | static void ops_complete_biofill(void *stripe_head_ref) | |
549 | { | |
550 | struct stripe_head *sh = stripe_head_ref; | |
551 | struct bio *return_bi = NULL; | |
552 | raid5_conf_t *conf = sh->raid_conf; | |
e4d84909 | 553 | int i; |
91c00924 | 554 | |
e46b272b | 555 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
556 | (unsigned long long)sh->sector); |
557 | ||
558 | /* clear completed biofills */ | |
83de75cc | 559 | spin_lock_irq(&conf->device_lock); |
91c00924 DW |
560 | for (i = sh->disks; i--; ) { |
561 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
562 | |
563 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
564 | /* and check if we need to reply to a read request, |
565 | * new R5_Wantfill requests are held off until | |
83de75cc | 566 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
567 | */ |
568 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 569 | struct bio *rbi, *rbi2; |
91c00924 | 570 | |
91c00924 DW |
571 | BUG_ON(!dev->read); |
572 | rbi = dev->read; | |
573 | dev->read = NULL; | |
574 | while (rbi && rbi->bi_sector < | |
575 | dev->sector + STRIPE_SECTORS) { | |
576 | rbi2 = r5_next_bio(rbi, dev->sector); | |
960e739d | 577 | if (!raid5_dec_bi_phys_segments(rbi)) { |
91c00924 DW |
578 | rbi->bi_next = return_bi; |
579 | return_bi = rbi; | |
580 | } | |
91c00924 DW |
581 | rbi = rbi2; |
582 | } | |
583 | } | |
584 | } | |
83de75cc DW |
585 | spin_unlock_irq(&conf->device_lock); |
586 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
91c00924 DW |
587 | |
588 | return_io(return_bi); | |
589 | ||
e4d84909 | 590 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
591 | release_stripe(sh); |
592 | } | |
593 | ||
594 | static void ops_run_biofill(struct stripe_head *sh) | |
595 | { | |
596 | struct dma_async_tx_descriptor *tx = NULL; | |
597 | raid5_conf_t *conf = sh->raid_conf; | |
598 | int i; | |
599 | ||
e46b272b | 600 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
601 | (unsigned long long)sh->sector); |
602 | ||
603 | for (i = sh->disks; i--; ) { | |
604 | struct r5dev *dev = &sh->dev[i]; | |
605 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
606 | struct bio *rbi; | |
607 | spin_lock_irq(&conf->device_lock); | |
608 | dev->read = rbi = dev->toread; | |
609 | dev->toread = NULL; | |
610 | spin_unlock_irq(&conf->device_lock); | |
611 | while (rbi && rbi->bi_sector < | |
612 | dev->sector + STRIPE_SECTORS) { | |
613 | tx = async_copy_data(0, rbi, dev->page, | |
614 | dev->sector, tx); | |
615 | rbi = r5_next_bio(rbi, dev->sector); | |
616 | } | |
617 | } | |
618 | } | |
619 | ||
620 | atomic_inc(&sh->count); | |
621 | async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
622 | ops_complete_biofill, sh); | |
623 | } | |
624 | ||
625 | static void ops_complete_compute5(void *stripe_head_ref) | |
626 | { | |
627 | struct stripe_head *sh = stripe_head_ref; | |
628 | int target = sh->ops.target; | |
629 | struct r5dev *tgt = &sh->dev[target]; | |
630 | ||
e46b272b | 631 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
632 | (unsigned long long)sh->sector); |
633 | ||
634 | set_bit(R5_UPTODATE, &tgt->flags); | |
635 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
636 | clear_bit(R5_Wantcompute, &tgt->flags); | |
ecc65c9b DW |
637 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
638 | if (sh->check_state == check_state_compute_run) | |
639 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
640 | set_bit(STRIPE_HANDLE, &sh->state); |
641 | release_stripe(sh); | |
642 | } | |
643 | ||
7b3a871e | 644 | static struct dma_async_tx_descriptor *ops_run_compute5(struct stripe_head *sh) |
91c00924 DW |
645 | { |
646 | /* kernel stack size limits the total number of disks */ | |
647 | int disks = sh->disks; | |
648 | struct page *xor_srcs[disks]; | |
649 | int target = sh->ops.target; | |
650 | struct r5dev *tgt = &sh->dev[target]; | |
651 | struct page *xor_dest = tgt->page; | |
652 | int count = 0; | |
653 | struct dma_async_tx_descriptor *tx; | |
654 | int i; | |
655 | ||
656 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 657 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
658 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
659 | ||
660 | for (i = disks; i--; ) | |
661 | if (i != target) | |
662 | xor_srcs[count++] = sh->dev[i].page; | |
663 | ||
664 | atomic_inc(&sh->count); | |
665 | ||
666 | if (unlikely(count == 1)) | |
667 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
668 | 0, NULL, ops_complete_compute5, sh); | |
669 | else | |
670 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
671 | ASYNC_TX_XOR_ZERO_DST, NULL, | |
672 | ops_complete_compute5, sh); | |
673 | ||
91c00924 DW |
674 | return tx; |
675 | } | |
676 | ||
677 | static void ops_complete_prexor(void *stripe_head_ref) | |
678 | { | |
679 | struct stripe_head *sh = stripe_head_ref; | |
680 | ||
e46b272b | 681 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 682 | (unsigned long long)sh->sector); |
91c00924 DW |
683 | } |
684 | ||
685 | static struct dma_async_tx_descriptor * | |
686 | ops_run_prexor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) | |
687 | { | |
688 | /* kernel stack size limits the total number of disks */ | |
689 | int disks = sh->disks; | |
690 | struct page *xor_srcs[disks]; | |
691 | int count = 0, pd_idx = sh->pd_idx, i; | |
692 | ||
693 | /* existing parity data subtracted */ | |
694 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
695 | ||
e46b272b | 696 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
697 | (unsigned long long)sh->sector); |
698 | ||
699 | for (i = disks; i--; ) { | |
700 | struct r5dev *dev = &sh->dev[i]; | |
701 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 702 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
703 | xor_srcs[count++] = dev->page; |
704 | } | |
705 | ||
706 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
707 | ASYNC_TX_DEP_ACK | ASYNC_TX_XOR_DROP_DST, tx, | |
708 | ops_complete_prexor, sh); | |
709 | ||
710 | return tx; | |
711 | } | |
712 | ||
713 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 714 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
715 | { |
716 | int disks = sh->disks; | |
d8ee0728 | 717 | int i; |
91c00924 | 718 | |
e46b272b | 719 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
720 | (unsigned long long)sh->sector); |
721 | ||
722 | for (i = disks; i--; ) { | |
723 | struct r5dev *dev = &sh->dev[i]; | |
724 | struct bio *chosen; | |
91c00924 | 725 | |
d8ee0728 | 726 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
727 | struct bio *wbi; |
728 | ||
729 | spin_lock(&sh->lock); | |
730 | chosen = dev->towrite; | |
731 | dev->towrite = NULL; | |
732 | BUG_ON(dev->written); | |
733 | wbi = dev->written = chosen; | |
734 | spin_unlock(&sh->lock); | |
735 | ||
736 | while (wbi && wbi->bi_sector < | |
737 | dev->sector + STRIPE_SECTORS) { | |
738 | tx = async_copy_data(1, wbi, dev->page, | |
739 | dev->sector, tx); | |
740 | wbi = r5_next_bio(wbi, dev->sector); | |
741 | } | |
742 | } | |
743 | } | |
744 | ||
745 | return tx; | |
746 | } | |
747 | ||
748 | static void ops_complete_postxor(void *stripe_head_ref) | |
91c00924 DW |
749 | { |
750 | struct stripe_head *sh = stripe_head_ref; | |
751 | int disks = sh->disks, i, pd_idx = sh->pd_idx; | |
752 | ||
e46b272b | 753 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
754 | (unsigned long long)sh->sector); |
755 | ||
756 | for (i = disks; i--; ) { | |
757 | struct r5dev *dev = &sh->dev[i]; | |
758 | if (dev->written || i == pd_idx) | |
759 | set_bit(R5_UPTODATE, &dev->flags); | |
760 | } | |
761 | ||
d8ee0728 DW |
762 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
763 | sh->reconstruct_state = reconstruct_state_drain_result; | |
764 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
765 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
766 | else { | |
767 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
768 | sh->reconstruct_state = reconstruct_state_result; | |
769 | } | |
91c00924 DW |
770 | |
771 | set_bit(STRIPE_HANDLE, &sh->state); | |
772 | release_stripe(sh); | |
773 | } | |
774 | ||
775 | static void | |
d8ee0728 | 776 | ops_run_postxor(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
777 | { |
778 | /* kernel stack size limits the total number of disks */ | |
779 | int disks = sh->disks; | |
780 | struct page *xor_srcs[disks]; | |
781 | ||
782 | int count = 0, pd_idx = sh->pd_idx, i; | |
783 | struct page *xor_dest; | |
d8ee0728 | 784 | int prexor = 0; |
91c00924 | 785 | unsigned long flags; |
91c00924 | 786 | |
e46b272b | 787 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
788 | (unsigned long long)sh->sector); |
789 | ||
790 | /* check if prexor is active which means only process blocks | |
791 | * that are part of a read-modify-write (written) | |
792 | */ | |
d8ee0728 DW |
793 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
794 | prexor = 1; | |
91c00924 DW |
795 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
796 | for (i = disks; i--; ) { | |
797 | struct r5dev *dev = &sh->dev[i]; | |
798 | if (dev->written) | |
799 | xor_srcs[count++] = dev->page; | |
800 | } | |
801 | } else { | |
802 | xor_dest = sh->dev[pd_idx].page; | |
803 | for (i = disks; i--; ) { | |
804 | struct r5dev *dev = &sh->dev[i]; | |
805 | if (i != pd_idx) | |
806 | xor_srcs[count++] = dev->page; | |
807 | } | |
808 | } | |
809 | ||
91c00924 DW |
810 | /* 1/ if we prexor'd then the dest is reused as a source |
811 | * 2/ if we did not prexor then we are redoing the parity | |
812 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
813 | * for the synchronous xor case | |
814 | */ | |
815 | flags = ASYNC_TX_DEP_ACK | ASYNC_TX_ACK | | |
816 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
817 | ||
818 | atomic_inc(&sh->count); | |
819 | ||
820 | if (unlikely(count == 1)) { | |
821 | flags &= ~(ASYNC_TX_XOR_DROP_DST | ASYNC_TX_XOR_ZERO_DST); | |
822 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, | |
d8ee0728 | 823 | flags, tx, ops_complete_postxor, sh); |
91c00924 DW |
824 | } else |
825 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
d8ee0728 | 826 | flags, tx, ops_complete_postxor, sh); |
91c00924 DW |
827 | } |
828 | ||
829 | static void ops_complete_check(void *stripe_head_ref) | |
830 | { | |
831 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 832 | |
e46b272b | 833 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
834 | (unsigned long long)sh->sector); |
835 | ||
ecc65c9b | 836 | sh->check_state = check_state_check_result; |
91c00924 DW |
837 | set_bit(STRIPE_HANDLE, &sh->state); |
838 | release_stripe(sh); | |
839 | } | |
840 | ||
841 | static void ops_run_check(struct stripe_head *sh) | |
842 | { | |
843 | /* kernel stack size limits the total number of disks */ | |
844 | int disks = sh->disks; | |
845 | struct page *xor_srcs[disks]; | |
846 | struct dma_async_tx_descriptor *tx; | |
847 | ||
848 | int count = 0, pd_idx = sh->pd_idx, i; | |
849 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
850 | ||
e46b272b | 851 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
852 | (unsigned long long)sh->sector); |
853 | ||
854 | for (i = disks; i--; ) { | |
855 | struct r5dev *dev = &sh->dev[i]; | |
856 | if (i != pd_idx) | |
857 | xor_srcs[count++] = dev->page; | |
858 | } | |
859 | ||
860 | tx = async_xor_zero_sum(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, | |
861 | &sh->ops.zero_sum_result, 0, NULL, NULL, NULL); | |
862 | ||
91c00924 DW |
863 | atomic_inc(&sh->count); |
864 | tx = async_trigger_callback(ASYNC_TX_DEP_ACK | ASYNC_TX_ACK, tx, | |
865 | ops_complete_check, sh); | |
866 | } | |
867 | ||
600aa109 | 868 | static void raid5_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
869 | { |
870 | int overlap_clear = 0, i, disks = sh->disks; | |
871 | struct dma_async_tx_descriptor *tx = NULL; | |
872 | ||
83de75cc | 873 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
874 | ops_run_biofill(sh); |
875 | overlap_clear++; | |
876 | } | |
877 | ||
7b3a871e DW |
878 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
879 | tx = ops_run_compute5(sh); | |
880 | /* terminate the chain if postxor is not set to be run */ | |
881 | if (tx && !test_bit(STRIPE_OP_POSTXOR, &ops_request)) | |
882 | async_tx_ack(tx); | |
883 | } | |
91c00924 | 884 | |
600aa109 | 885 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
91c00924 DW |
886 | tx = ops_run_prexor(sh, tx); |
887 | ||
600aa109 | 888 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 889 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
890 | overlap_clear++; |
891 | } | |
892 | ||
600aa109 | 893 | if (test_bit(STRIPE_OP_POSTXOR, &ops_request)) |
d8ee0728 | 894 | ops_run_postxor(sh, tx); |
91c00924 | 895 | |
ecc65c9b | 896 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) |
91c00924 DW |
897 | ops_run_check(sh); |
898 | ||
91c00924 DW |
899 | if (overlap_clear) |
900 | for (i = disks; i--; ) { | |
901 | struct r5dev *dev = &sh->dev[i]; | |
902 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
903 | wake_up(&sh->raid_conf->wait_for_overlap); | |
904 | } | |
905 | } | |
906 | ||
3f294f4f | 907 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
908 | { |
909 | struct stripe_head *sh; | |
3f294f4f N |
910 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
911 | if (!sh) | |
912 | return 0; | |
913 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
914 | sh->raid_conf = conf; | |
915 | spin_lock_init(&sh->lock); | |
916 | ||
917 | if (grow_buffers(sh, conf->raid_disks)) { | |
918 | shrink_buffers(sh, conf->raid_disks); | |
919 | kmem_cache_free(conf->slab_cache, sh); | |
920 | return 0; | |
921 | } | |
7ecaa1e6 | 922 | sh->disks = conf->raid_disks; |
3f294f4f N |
923 | /* we just created an active stripe so... */ |
924 | atomic_set(&sh->count, 1); | |
925 | atomic_inc(&conf->active_stripes); | |
926 | INIT_LIST_HEAD(&sh->lru); | |
927 | release_stripe(sh); | |
928 | return 1; | |
929 | } | |
930 | ||
931 | static int grow_stripes(raid5_conf_t *conf, int num) | |
932 | { | |
e18b890b | 933 | struct kmem_cache *sc; |
1da177e4 LT |
934 | int devs = conf->raid_disks; |
935 | ||
245f46c2 N |
936 | sprintf(conf->cache_name[0], |
937 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
938 | sprintf(conf->cache_name[1], | |
939 | "raid%d-%s-alt", conf->level, mdname(conf->mddev)); | |
ad01c9e3 N |
940 | conf->active_name = 0; |
941 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 942 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 943 | 0, 0, NULL); |
1da177e4 LT |
944 | if (!sc) |
945 | return 1; | |
946 | conf->slab_cache = sc; | |
ad01c9e3 | 947 | conf->pool_size = devs; |
16a53ecc | 948 | while (num--) |
3f294f4f | 949 | if (!grow_one_stripe(conf)) |
1da177e4 | 950 | return 1; |
1da177e4 LT |
951 | return 0; |
952 | } | |
29269553 N |
953 | |
954 | #ifdef CONFIG_MD_RAID5_RESHAPE | |
ad01c9e3 N |
955 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
956 | { | |
957 | /* Make all the stripes able to hold 'newsize' devices. | |
958 | * New slots in each stripe get 'page' set to a new page. | |
959 | * | |
960 | * This happens in stages: | |
961 | * 1/ create a new kmem_cache and allocate the required number of | |
962 | * stripe_heads. | |
963 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
964 | * to the new stripe_heads. This will have the side effect of | |
965 | * freezing the array as once all stripe_heads have been collected, | |
966 | * no IO will be possible. Old stripe heads are freed once their | |
967 | * pages have been transferred over, and the old kmem_cache is | |
968 | * freed when all stripes are done. | |
969 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
970 | * we simple return a failre status - no need to clean anything up. | |
971 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
972 | * If this fails, we don't bother trying the shrink the | |
973 | * stripe_heads down again, we just leave them as they are. | |
974 | * As each stripe_head is processed the new one is released into | |
975 | * active service. | |
976 | * | |
977 | * Once step2 is started, we cannot afford to wait for a write, | |
978 | * so we use GFP_NOIO allocations. | |
979 | */ | |
980 | struct stripe_head *osh, *nsh; | |
981 | LIST_HEAD(newstripes); | |
982 | struct disk_info *ndisks; | |
b5470dc5 | 983 | int err; |
e18b890b | 984 | struct kmem_cache *sc; |
ad01c9e3 N |
985 | int i; |
986 | ||
987 | if (newsize <= conf->pool_size) | |
988 | return 0; /* never bother to shrink */ | |
989 | ||
b5470dc5 DW |
990 | err = md_allow_write(conf->mddev); |
991 | if (err) | |
992 | return err; | |
2a2275d6 | 993 | |
ad01c9e3 N |
994 | /* Step 1 */ |
995 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
996 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 997 | 0, 0, NULL); |
ad01c9e3 N |
998 | if (!sc) |
999 | return -ENOMEM; | |
1000 | ||
1001 | for (i = conf->max_nr_stripes; i; i--) { | |
1002 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
1003 | if (!nsh) | |
1004 | break; | |
1005 | ||
1006 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
1007 | ||
1008 | nsh->raid_conf = conf; | |
1009 | spin_lock_init(&nsh->lock); | |
1010 | ||
1011 | list_add(&nsh->lru, &newstripes); | |
1012 | } | |
1013 | if (i) { | |
1014 | /* didn't get enough, give up */ | |
1015 | while (!list_empty(&newstripes)) { | |
1016 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1017 | list_del(&nsh->lru); | |
1018 | kmem_cache_free(sc, nsh); | |
1019 | } | |
1020 | kmem_cache_destroy(sc); | |
1021 | return -ENOMEM; | |
1022 | } | |
1023 | /* Step 2 - Must use GFP_NOIO now. | |
1024 | * OK, we have enough stripes, start collecting inactive | |
1025 | * stripes and copying them over | |
1026 | */ | |
1027 | list_for_each_entry(nsh, &newstripes, lru) { | |
1028 | spin_lock_irq(&conf->device_lock); | |
1029 | wait_event_lock_irq(conf->wait_for_stripe, | |
1030 | !list_empty(&conf->inactive_list), | |
1031 | conf->device_lock, | |
b3b46be3 | 1032 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1033 | ); |
1034 | osh = get_free_stripe(conf); | |
1035 | spin_unlock_irq(&conf->device_lock); | |
1036 | atomic_set(&nsh->count, 1); | |
1037 | for(i=0; i<conf->pool_size; i++) | |
1038 | nsh->dev[i].page = osh->dev[i].page; | |
1039 | for( ; i<newsize; i++) | |
1040 | nsh->dev[i].page = NULL; | |
1041 | kmem_cache_free(conf->slab_cache, osh); | |
1042 | } | |
1043 | kmem_cache_destroy(conf->slab_cache); | |
1044 | ||
1045 | /* Step 3. | |
1046 | * At this point, we are holding all the stripes so the array | |
1047 | * is completely stalled, so now is a good time to resize | |
1048 | * conf->disks. | |
1049 | */ | |
1050 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1051 | if (ndisks) { | |
1052 | for (i=0; i<conf->raid_disks; i++) | |
1053 | ndisks[i] = conf->disks[i]; | |
1054 | kfree(conf->disks); | |
1055 | conf->disks = ndisks; | |
1056 | } else | |
1057 | err = -ENOMEM; | |
1058 | ||
1059 | /* Step 4, return new stripes to service */ | |
1060 | while(!list_empty(&newstripes)) { | |
1061 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1062 | list_del_init(&nsh->lru); | |
1063 | for (i=conf->raid_disks; i < newsize; i++) | |
1064 | if (nsh->dev[i].page == NULL) { | |
1065 | struct page *p = alloc_page(GFP_NOIO); | |
1066 | nsh->dev[i].page = p; | |
1067 | if (!p) | |
1068 | err = -ENOMEM; | |
1069 | } | |
1070 | release_stripe(nsh); | |
1071 | } | |
1072 | /* critical section pass, GFP_NOIO no longer needed */ | |
1073 | ||
1074 | conf->slab_cache = sc; | |
1075 | conf->active_name = 1-conf->active_name; | |
1076 | conf->pool_size = newsize; | |
1077 | return err; | |
1078 | } | |
29269553 | 1079 | #endif |
1da177e4 | 1080 | |
3f294f4f | 1081 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1082 | { |
1083 | struct stripe_head *sh; | |
1084 | ||
3f294f4f N |
1085 | spin_lock_irq(&conf->device_lock); |
1086 | sh = get_free_stripe(conf); | |
1087 | spin_unlock_irq(&conf->device_lock); | |
1088 | if (!sh) | |
1089 | return 0; | |
78bafebd | 1090 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1091 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1092 | kmem_cache_free(conf->slab_cache, sh); |
1093 | atomic_dec(&conf->active_stripes); | |
1094 | return 1; | |
1095 | } | |
1096 | ||
1097 | static void shrink_stripes(raid5_conf_t *conf) | |
1098 | { | |
1099 | while (drop_one_stripe(conf)) | |
1100 | ; | |
1101 | ||
29fc7e3e N |
1102 | if (conf->slab_cache) |
1103 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1104 | conf->slab_cache = NULL; |
1105 | } | |
1106 | ||
6712ecf8 | 1107 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1108 | { |
99c0fb5f | 1109 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1110 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1111 | int disks = sh->disks, i; |
1da177e4 | 1112 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1113 | char b[BDEVNAME_SIZE]; |
1114 | mdk_rdev_t *rdev; | |
1da177e4 | 1115 | |
1da177e4 LT |
1116 | |
1117 | for (i=0 ; i<disks; i++) | |
1118 | if (bi == &sh->dev[i].req) | |
1119 | break; | |
1120 | ||
45b4233c DW |
1121 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1122 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1123 | uptodate); |
1124 | if (i == disks) { | |
1125 | BUG(); | |
6712ecf8 | 1126 | return; |
1da177e4 LT |
1127 | } |
1128 | ||
1129 | if (uptodate) { | |
1da177e4 | 1130 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1131 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1132 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1133 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1134 | " (%lu sectors at %llu on %s)\n", | |
1135 | mdname(conf->mddev), STRIPE_SECTORS, | |
1136 | (unsigned long long)(sh->sector | |
1137 | + rdev->data_offset), | |
1138 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1139 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1140 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1141 | } | |
ba22dcbf N |
1142 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1143 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1144 | } else { |
d6950432 | 1145 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1146 | int retry = 0; |
d6950432 N |
1147 | rdev = conf->disks[i].rdev; |
1148 | ||
1da177e4 | 1149 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1150 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1151 | if (conf->mddev->degraded) |
6be9d494 BS |
1152 | printk_rl(KERN_WARNING |
1153 | "raid5:%s: read error not correctable " | |
1154 | "(sector %llu on %s).\n", | |
1155 | mdname(conf->mddev), | |
1156 | (unsigned long long)(sh->sector | |
1157 | + rdev->data_offset), | |
1158 | bdn); | |
ba22dcbf | 1159 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1160 | /* Oh, no!!! */ |
6be9d494 BS |
1161 | printk_rl(KERN_WARNING |
1162 | "raid5:%s: read error NOT corrected!! " | |
1163 | "(sector %llu on %s).\n", | |
1164 | mdname(conf->mddev), | |
1165 | (unsigned long long)(sh->sector | |
1166 | + rdev->data_offset), | |
1167 | bdn); | |
d6950432 | 1168 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1169 | > conf->max_nr_stripes) |
14f8d26b | 1170 | printk(KERN_WARNING |
d6950432 N |
1171 | "raid5:%s: Too many read errors, failing device %s.\n", |
1172 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1173 | else |
1174 | retry = 1; | |
1175 | if (retry) | |
1176 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1177 | else { | |
4e5314b5 N |
1178 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1179 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1180 | md_error(conf->mddev, rdev); |
ba22dcbf | 1181 | } |
1da177e4 LT |
1182 | } |
1183 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1184 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1185 | set_bit(STRIPE_HANDLE, &sh->state); | |
1186 | release_stripe(sh); | |
1da177e4 LT |
1187 | } |
1188 | ||
d710e138 | 1189 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1190 | { |
99c0fb5f | 1191 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1192 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1193 | int disks = sh->disks, i; |
1da177e4 LT |
1194 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1195 | ||
1da177e4 LT |
1196 | for (i=0 ; i<disks; i++) |
1197 | if (bi == &sh->dev[i].req) | |
1198 | break; | |
1199 | ||
45b4233c | 1200 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1201 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1202 | uptodate); | |
1203 | if (i == disks) { | |
1204 | BUG(); | |
6712ecf8 | 1205 | return; |
1da177e4 LT |
1206 | } |
1207 | ||
1da177e4 LT |
1208 | if (!uptodate) |
1209 | md_error(conf->mddev, conf->disks[i].rdev); | |
1210 | ||
1211 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1212 | ||
1213 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1214 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1215 | release_stripe(sh); |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | ||
1219 | static sector_t compute_blocknr(struct stripe_head *sh, int i); | |
1220 | ||
d710e138 | 1221 | static void raid5_build_block(struct stripe_head *sh, int i) |
1da177e4 LT |
1222 | { |
1223 | struct r5dev *dev = &sh->dev[i]; | |
1224 | ||
1225 | bio_init(&dev->req); | |
1226 | dev->req.bi_io_vec = &dev->vec; | |
1227 | dev->req.bi_vcnt++; | |
1228 | dev->req.bi_max_vecs++; | |
1229 | dev->vec.bv_page = dev->page; | |
1230 | dev->vec.bv_len = STRIPE_SIZE; | |
1231 | dev->vec.bv_offset = 0; | |
1232 | ||
1233 | dev->req.bi_sector = sh->sector; | |
1234 | dev->req.bi_private = sh; | |
1235 | ||
1236 | dev->flags = 0; | |
16a53ecc | 1237 | dev->sector = compute_blocknr(sh, i); |
1da177e4 LT |
1238 | } |
1239 | ||
1240 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1241 | { | |
1242 | char b[BDEVNAME_SIZE]; | |
1243 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1244 | pr_debug("raid5: error called\n"); |
1da177e4 | 1245 | |
b2d444d7 | 1246 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1247 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1248 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1249 | unsigned long flags; | |
1250 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1251 | mddev->degraded++; |
c04be0aa | 1252 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1253 | /* |
1254 | * if recovery was running, make sure it aborts. | |
1255 | */ | |
dfc70645 | 1256 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1257 | } |
b2d444d7 | 1258 | set_bit(Faulty, &rdev->flags); |
d710e138 N |
1259 | printk(KERN_ALERT |
1260 | "raid5: Disk failure on %s, disabling device.\n" | |
1261 | "raid5: Operation continuing on %d devices.\n", | |
1262 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); | |
1da177e4 | 1263 | } |
16a53ecc | 1264 | } |
1da177e4 LT |
1265 | |
1266 | /* | |
1267 | * Input: a 'big' sector number, | |
1268 | * Output: index of the data and parity disk, and the sector # in them. | |
1269 | */ | |
112bf897 | 1270 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1271 | int previous, int *dd_idx, |
1272 | struct stripe_head *sh) | |
1da177e4 LT |
1273 | { |
1274 | long stripe; | |
1275 | unsigned long chunk_number; | |
1276 | unsigned int chunk_offset; | |
911d4ee8 | 1277 | int pd_idx, qd_idx; |
67cc2b81 | 1278 | int ddf_layout = 0; |
1da177e4 LT |
1279 | sector_t new_sector; |
1280 | int sectors_per_chunk = conf->chunk_size >> 9; | |
112bf897 N |
1281 | int raid_disks = previous ? conf->previous_raid_disks |
1282 | : conf->raid_disks; | |
1283 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1284 | |
1285 | /* First compute the information on this sector */ | |
1286 | ||
1287 | /* | |
1288 | * Compute the chunk number and the sector offset inside the chunk | |
1289 | */ | |
1290 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1291 | chunk_number = r_sector; | |
1292 | BUG_ON(r_sector != chunk_number); | |
1293 | ||
1294 | /* | |
1295 | * Compute the stripe number | |
1296 | */ | |
1297 | stripe = chunk_number / data_disks; | |
1298 | ||
1299 | /* | |
1300 | * Compute the data disk and parity disk indexes inside the stripe | |
1301 | */ | |
1302 | *dd_idx = chunk_number % data_disks; | |
1303 | ||
1304 | /* | |
1305 | * Select the parity disk based on the user selected algorithm. | |
1306 | */ | |
911d4ee8 | 1307 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1308 | switch(conf->level) { |
1309 | case 4: | |
911d4ee8 | 1310 | pd_idx = data_disks; |
16a53ecc N |
1311 | break; |
1312 | case 5: | |
1313 | switch (conf->algorithm) { | |
1da177e4 | 1314 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1315 | pd_idx = data_disks - stripe % raid_disks; |
1316 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1317 | (*dd_idx)++; |
1318 | break; | |
1319 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1320 | pd_idx = stripe % raid_disks; |
1321 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1322 | (*dd_idx)++; |
1323 | break; | |
1324 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1325 | pd_idx = data_disks - stripe % raid_disks; |
1326 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 LT |
1327 | break; |
1328 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1329 | pd_idx = stripe % raid_disks; |
1330 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 | 1331 | break; |
99c0fb5f N |
1332 | case ALGORITHM_PARITY_0: |
1333 | pd_idx = 0; | |
1334 | (*dd_idx)++; | |
1335 | break; | |
1336 | case ALGORITHM_PARITY_N: | |
1337 | pd_idx = data_disks; | |
1338 | break; | |
1da177e4 | 1339 | default: |
14f8d26b | 1340 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 | 1341 | conf->algorithm); |
99c0fb5f | 1342 | BUG(); |
16a53ecc N |
1343 | } |
1344 | break; | |
1345 | case 6: | |
1346 | ||
16a53ecc N |
1347 | switch (conf->algorithm) { |
1348 | case ALGORITHM_LEFT_ASYMMETRIC: | |
911d4ee8 N |
1349 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1350 | qd_idx = pd_idx + 1; | |
1351 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1352 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1353 | qd_idx = 0; |
1354 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1355 | (*dd_idx) += 2; /* D D P Q D */ |
1356 | break; | |
1357 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1358 | pd_idx = stripe % raid_disks; |
1359 | qd_idx = pd_idx + 1; | |
1360 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1361 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1362 | qd_idx = 0; |
1363 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1364 | (*dd_idx) += 2; /* D D P Q D */ |
1365 | break; | |
1366 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1367 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1368 | qd_idx = (pd_idx + 1) % raid_disks; | |
1369 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1370 | break; |
1371 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1372 | pd_idx = stripe % raid_disks; |
1373 | qd_idx = (pd_idx + 1) % raid_disks; | |
1374 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1375 | break; |
99c0fb5f N |
1376 | |
1377 | case ALGORITHM_PARITY_0: | |
1378 | pd_idx = 0; | |
1379 | qd_idx = 1; | |
1380 | (*dd_idx) += 2; | |
1381 | break; | |
1382 | case ALGORITHM_PARITY_N: | |
1383 | pd_idx = data_disks; | |
1384 | qd_idx = data_disks + 1; | |
1385 | break; | |
1386 | ||
1387 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1388 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1389 | * of blocks for computing Q is different. | |
1390 | */ | |
1391 | pd_idx = stripe % raid_disks; | |
1392 | qd_idx = pd_idx + 1; | |
1393 | if (pd_idx == raid_disks-1) { | |
1394 | (*dd_idx)++; /* Q D D D P */ | |
1395 | qd_idx = 0; | |
1396 | } else if (*dd_idx >= pd_idx) | |
1397 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1398 | ddf_layout = 1; |
99c0fb5f N |
1399 | break; |
1400 | ||
1401 | case ALGORITHM_ROTATING_N_RESTART: | |
1402 | /* Same a left_asymmetric, by first stripe is | |
1403 | * D D D P Q rather than | |
1404 | * Q D D D P | |
1405 | */ | |
1406 | pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
1407 | qd_idx = pd_idx + 1; | |
1408 | if (pd_idx == raid_disks-1) { | |
1409 | (*dd_idx)++; /* Q D D D P */ | |
1410 | qd_idx = 0; | |
1411 | } else if (*dd_idx >= pd_idx) | |
1412 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1413 | ddf_layout = 1; |
99c0fb5f N |
1414 | break; |
1415 | ||
1416 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1417 | /* Same as left_symmetric but Q is before P */ | |
1418 | pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1419 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; | |
1420 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1421 | ddf_layout = 1; |
99c0fb5f N |
1422 | break; |
1423 | ||
1424 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1425 | /* RAID5 left_asymmetric, with Q on last device */ | |
1426 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1427 | if (*dd_idx >= pd_idx) | |
1428 | (*dd_idx)++; | |
1429 | qd_idx = raid_disks - 1; | |
1430 | break; | |
1431 | ||
1432 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1433 | pd_idx = stripe % (raid_disks-1); | |
1434 | if (*dd_idx >= pd_idx) | |
1435 | (*dd_idx)++; | |
1436 | qd_idx = raid_disks - 1; | |
1437 | break; | |
1438 | ||
1439 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1440 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1441 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1442 | qd_idx = raid_disks - 1; | |
1443 | break; | |
1444 | ||
1445 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1446 | pd_idx = stripe % (raid_disks-1); | |
1447 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1448 | qd_idx = raid_disks - 1; | |
1449 | break; | |
1450 | ||
1451 | case ALGORITHM_PARITY_0_6: | |
1452 | pd_idx = 0; | |
1453 | (*dd_idx)++; | |
1454 | qd_idx = raid_disks - 1; | |
1455 | break; | |
1456 | ||
1457 | ||
16a53ecc | 1458 | default: |
d710e138 N |
1459 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1460 | conf->algorithm); | |
99c0fb5f | 1461 | BUG(); |
16a53ecc N |
1462 | } |
1463 | break; | |
1da177e4 LT |
1464 | } |
1465 | ||
911d4ee8 N |
1466 | if (sh) { |
1467 | sh->pd_idx = pd_idx; | |
1468 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1469 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1470 | } |
1da177e4 LT |
1471 | /* |
1472 | * Finally, compute the new sector number | |
1473 | */ | |
1474 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1475 | return new_sector; | |
1476 | } | |
1477 | ||
1478 | ||
1479 | static sector_t compute_blocknr(struct stripe_head *sh, int i) | |
1480 | { | |
1481 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1482 | int raid_disks = sh->disks; |
1483 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1484 | sector_t new_sector = sh->sector, check; |
1485 | int sectors_per_chunk = conf->chunk_size >> 9; | |
1486 | sector_t stripe; | |
1487 | int chunk_offset; | |
911d4ee8 | 1488 | int chunk_number, dummy1, dd_idx = i; |
1da177e4 | 1489 | sector_t r_sector; |
911d4ee8 | 1490 | struct stripe_head sh2; |
1da177e4 | 1491 | |
16a53ecc | 1492 | |
1da177e4 LT |
1493 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1494 | stripe = new_sector; | |
1495 | BUG_ON(new_sector != stripe); | |
1496 | ||
16a53ecc N |
1497 | if (i == sh->pd_idx) |
1498 | return 0; | |
1499 | switch(conf->level) { | |
1500 | case 4: break; | |
1501 | case 5: | |
1502 | switch (conf->algorithm) { | |
1da177e4 LT |
1503 | case ALGORITHM_LEFT_ASYMMETRIC: |
1504 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1505 | if (i > sh->pd_idx) | |
1506 | i--; | |
1507 | break; | |
1508 | case ALGORITHM_LEFT_SYMMETRIC: | |
1509 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1510 | if (i < sh->pd_idx) | |
1511 | i += raid_disks; | |
1512 | i -= (sh->pd_idx + 1); | |
1513 | break; | |
99c0fb5f N |
1514 | case ALGORITHM_PARITY_0: |
1515 | i -= 1; | |
1516 | break; | |
1517 | case ALGORITHM_PARITY_N: | |
1518 | break; | |
1da177e4 | 1519 | default: |
14f8d26b | 1520 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
16a53ecc | 1521 | conf->algorithm); |
99c0fb5f | 1522 | BUG(); |
16a53ecc N |
1523 | } |
1524 | break; | |
1525 | case 6: | |
d0dabf7e | 1526 | if (i == sh->qd_idx) |
16a53ecc N |
1527 | return 0; /* It is the Q disk */ |
1528 | switch (conf->algorithm) { | |
1529 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1530 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1531 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1532 | case ALGORITHM_ROTATING_N_RESTART: | |
1533 | if (sh->pd_idx == raid_disks-1) | |
1534 | i--; /* Q D D D P */ | |
16a53ecc N |
1535 | else if (i > sh->pd_idx) |
1536 | i -= 2; /* D D P Q D */ | |
1537 | break; | |
1538 | case ALGORITHM_LEFT_SYMMETRIC: | |
1539 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1540 | if (sh->pd_idx == raid_disks-1) | |
1541 | i--; /* Q D D D P */ | |
1542 | else { | |
1543 | /* D D P Q D */ | |
1544 | if (i < sh->pd_idx) | |
1545 | i += raid_disks; | |
1546 | i -= (sh->pd_idx + 2); | |
1547 | } | |
1548 | break; | |
99c0fb5f N |
1549 | case ALGORITHM_PARITY_0: |
1550 | i -= 2; | |
1551 | break; | |
1552 | case ALGORITHM_PARITY_N: | |
1553 | break; | |
1554 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1555 | if (sh->pd_idx == 0) | |
1556 | i--; /* P D D D Q */ | |
1557 | else if (i > sh->pd_idx) | |
1558 | i -= 2; /* D D Q P D */ | |
1559 | break; | |
1560 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1561 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1562 | if (i > sh->pd_idx) | |
1563 | i--; | |
1564 | break; | |
1565 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1566 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1567 | if (i < sh->pd_idx) | |
1568 | i += data_disks + 1; | |
1569 | i -= (sh->pd_idx + 1); | |
1570 | break; | |
1571 | case ALGORITHM_PARITY_0_6: | |
1572 | i -= 1; | |
1573 | break; | |
16a53ecc | 1574 | default: |
d710e138 N |
1575 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1576 | conf->algorithm); | |
99c0fb5f | 1577 | BUG(); |
16a53ecc N |
1578 | } |
1579 | break; | |
1da177e4 LT |
1580 | } |
1581 | ||
1582 | chunk_number = stripe * data_disks + i; | |
1583 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1584 | ||
112bf897 N |
1585 | check = raid5_compute_sector(conf, r_sector, |
1586 | (raid_disks != conf->raid_disks), | |
911d4ee8 N |
1587 | &dummy1, &sh2); |
1588 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx | |
1589 | || sh2.qd_idx != sh->qd_idx) { | |
14f8d26b | 1590 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1591 | return 0; |
1592 | } | |
1593 | return r_sector; | |
1594 | } | |
1595 | ||
1596 | ||
1597 | ||
1598 | /* | |
16a53ecc N |
1599 | * Copy data between a page in the stripe cache, and one or more bion |
1600 | * The page could align with the middle of the bio, or there could be | |
1601 | * several bion, each with several bio_vecs, which cover part of the page | |
1602 | * Multiple bion are linked together on bi_next. There may be extras | |
1603 | * at the end of this list. We ignore them. | |
1da177e4 LT |
1604 | */ |
1605 | static void copy_data(int frombio, struct bio *bio, | |
1606 | struct page *page, | |
1607 | sector_t sector) | |
1608 | { | |
1609 | char *pa = page_address(page); | |
1610 | struct bio_vec *bvl; | |
1611 | int i; | |
1612 | int page_offset; | |
1613 | ||
1614 | if (bio->bi_sector >= sector) | |
1615 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
1616 | else | |
1617 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
1618 | bio_for_each_segment(bvl, bio, i) { | |
1619 | int len = bio_iovec_idx(bio,i)->bv_len; | |
1620 | int clen; | |
1621 | int b_offset = 0; | |
1622 | ||
1623 | if (page_offset < 0) { | |
1624 | b_offset = -page_offset; | |
1625 | page_offset += b_offset; | |
1626 | len -= b_offset; | |
1627 | } | |
1628 | ||
1629 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1630 | clen = STRIPE_SIZE - page_offset; | |
1631 | else clen = len; | |
16a53ecc | 1632 | |
1da177e4 LT |
1633 | if (clen > 0) { |
1634 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
1635 | if (frombio) | |
1636 | memcpy(pa+page_offset, ba+b_offset, clen); | |
1637 | else | |
1638 | memcpy(ba+b_offset, pa+page_offset, clen); | |
1639 | __bio_kunmap_atomic(ba, KM_USER0); | |
1640 | } | |
1641 | if (clen < len) /* hit end of page */ | |
1642 | break; | |
1643 | page_offset += len; | |
1644 | } | |
1645 | } | |
1646 | ||
9bc89cd8 DW |
1647 | #define check_xor() do { \ |
1648 | if (count == MAX_XOR_BLOCKS) { \ | |
1649 | xor_blocks(count, STRIPE_SIZE, dest, ptr);\ | |
1650 | count = 0; \ | |
1651 | } \ | |
1da177e4 LT |
1652 | } while(0) |
1653 | ||
16a53ecc N |
1654 | static void compute_parity6(struct stripe_head *sh, int method) |
1655 | { | |
bff61975 | 1656 | raid5_conf_t *conf = sh->raid_conf; |
d0dabf7e | 1657 | int i, pd_idx, qd_idx, d0_idx, disks = sh->disks, count; |
67cc2b81 | 1658 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); |
16a53ecc N |
1659 | struct bio *chosen; |
1660 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
67cc2b81 | 1661 | void *ptrs[syndrome_disks+2]; |
16a53ecc | 1662 | |
d0dabf7e N |
1663 | pd_idx = sh->pd_idx; |
1664 | qd_idx = sh->qd_idx; | |
1665 | d0_idx = raid6_d0(sh); | |
16a53ecc | 1666 | |
45b4233c | 1667 | pr_debug("compute_parity, stripe %llu, method %d\n", |
16a53ecc N |
1668 | (unsigned long long)sh->sector, method); |
1669 | ||
1670 | switch(method) { | |
1671 | case READ_MODIFY_WRITE: | |
1672 | BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ | |
1673 | case RECONSTRUCT_WRITE: | |
1674 | for (i= disks; i-- ;) | |
1675 | if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { | |
1676 | chosen = sh->dev[i].towrite; | |
1677 | sh->dev[i].towrite = NULL; | |
1678 | ||
1679 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1680 | wake_up(&conf->wait_for_overlap); | |
1681 | ||
52e5f9d1 | 1682 | BUG_ON(sh->dev[i].written); |
16a53ecc N |
1683 | sh->dev[i].written = chosen; |
1684 | } | |
1685 | break; | |
1686 | case CHECK_PARITY: | |
1687 | BUG(); /* Not implemented yet */ | |
1688 | } | |
1689 | ||
1690 | for (i = disks; i--;) | |
1691 | if (sh->dev[i].written) { | |
1692 | sector_t sector = sh->dev[i].sector; | |
1693 | struct bio *wbi = sh->dev[i].written; | |
1694 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
1695 | copy_data(1, wbi, sh->dev[i].page, sector); | |
1696 | wbi = r5_next_bio(wbi, sector); | |
1697 | } | |
1698 | ||
1699 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
1700 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
1701 | } | |
1702 | ||
d0dabf7e | 1703 | /* Note that unlike RAID-5, the ordering of the disks matters greatly.*/ |
67cc2b81 N |
1704 | |
1705 | for (i = 0; i < disks; i++) | |
1706 | ptrs[i] = (void *)raid6_empty_zero_page; | |
1707 | ||
d0dabf7e N |
1708 | count = 0; |
1709 | i = d0_idx; | |
1710 | do { | |
67cc2b81 N |
1711 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); |
1712 | ||
d0dabf7e | 1713 | ptrs[slot] = page_address(sh->dev[i].page); |
67cc2b81 | 1714 | if (slot < syndrome_disks && |
d0dabf7e N |
1715 | !test_bit(R5_UPTODATE, &sh->dev[i].flags)) { |
1716 | printk(KERN_ERR "block %d/%d not uptodate " | |
1717 | "on parity calc\n", i, count); | |
1718 | BUG(); | |
1719 | } | |
67cc2b81 | 1720 | |
d0dabf7e N |
1721 | i = raid6_next_disk(i, disks); |
1722 | } while (i != d0_idx); | |
67cc2b81 | 1723 | BUG_ON(count != syndrome_disks); |
16a53ecc | 1724 | |
67cc2b81 | 1725 | raid6_call.gen_syndrome(syndrome_disks+2, STRIPE_SIZE, ptrs); |
16a53ecc N |
1726 | |
1727 | switch(method) { | |
1728 | case RECONSTRUCT_WRITE: | |
1729 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1730 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1731 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1732 | set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); | |
1733 | break; | |
1734 | case UPDATE_PARITY: | |
1735 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1736 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1737 | break; | |
1738 | } | |
1739 | } | |
1740 | ||
1741 | ||
1742 | /* Compute one missing block */ | |
1743 | static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero) | |
1744 | { | |
f416885e | 1745 | int i, count, disks = sh->disks; |
9bc89cd8 | 1746 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
d0dabf7e | 1747 | int qd_idx = sh->qd_idx; |
16a53ecc | 1748 | |
45b4233c | 1749 | pr_debug("compute_block_1, stripe %llu, idx %d\n", |
16a53ecc N |
1750 | (unsigned long long)sh->sector, dd_idx); |
1751 | ||
1752 | if ( dd_idx == qd_idx ) { | |
1753 | /* We're actually computing the Q drive */ | |
1754 | compute_parity6(sh, UPDATE_PARITY); | |
1755 | } else { | |
9bc89cd8 DW |
1756 | dest = page_address(sh->dev[dd_idx].page); |
1757 | if (!nozero) memset(dest, 0, STRIPE_SIZE); | |
1758 | count = 0; | |
16a53ecc N |
1759 | for (i = disks ; i--; ) { |
1760 | if (i == dd_idx || i == qd_idx) | |
1761 | continue; | |
1762 | p = page_address(sh->dev[i].page); | |
1763 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1764 | ptr[count++] = p; | |
1765 | else | |
1766 | printk("compute_block() %d, stripe %llu, %d" | |
1767 | " not present\n", dd_idx, | |
1768 | (unsigned long long)sh->sector, i); | |
1769 | ||
1770 | check_xor(); | |
1771 | } | |
9bc89cd8 DW |
1772 | if (count) |
1773 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
16a53ecc N |
1774 | if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
1775 | else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
1776 | } | |
1777 | } | |
1778 | ||
1779 | /* Compute two missing blocks */ | |
1780 | static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) | |
1781 | { | |
f416885e | 1782 | int i, count, disks = sh->disks; |
67cc2b81 | 1783 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; |
d0dabf7e N |
1784 | int d0_idx = raid6_d0(sh); |
1785 | int faila = -1, failb = -1; | |
1786 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
67cc2b81 | 1787 | void *ptrs[syndrome_disks+2]; |
16a53ecc | 1788 | |
67cc2b81 N |
1789 | for (i = 0; i < disks ; i++) |
1790 | ptrs[i] = (void *)raid6_empty_zero_page; | |
d0dabf7e N |
1791 | count = 0; |
1792 | i = d0_idx; | |
1793 | do { | |
67cc2b81 N |
1794 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); |
1795 | ||
d0dabf7e | 1796 | ptrs[slot] = page_address(sh->dev[i].page); |
67cc2b81 | 1797 | |
d0dabf7e N |
1798 | if (i == dd_idx1) |
1799 | faila = slot; | |
1800 | if (i == dd_idx2) | |
1801 | failb = slot; | |
1802 | i = raid6_next_disk(i, disks); | |
1803 | } while (i != d0_idx); | |
67cc2b81 | 1804 | BUG_ON(count != syndrome_disks); |
16a53ecc N |
1805 | |
1806 | BUG_ON(faila == failb); | |
1807 | if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } | |
1808 | ||
45b4233c | 1809 | pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", |
d0dabf7e N |
1810 | (unsigned long long)sh->sector, dd_idx1, dd_idx2, |
1811 | faila, failb); | |
16a53ecc | 1812 | |
67cc2b81 | 1813 | if (failb == syndrome_disks+1) { |
16a53ecc | 1814 | /* Q disk is one of the missing disks */ |
67cc2b81 | 1815 | if (faila == syndrome_disks) { |
16a53ecc N |
1816 | /* Missing P+Q, just recompute */ |
1817 | compute_parity6(sh, UPDATE_PARITY); | |
1818 | return; | |
1819 | } else { | |
1820 | /* We're missing D+Q; recompute D from P */ | |
d0dabf7e N |
1821 | compute_block_1(sh, ((dd_idx1 == sh->qd_idx) ? |
1822 | dd_idx2 : dd_idx1), | |
1823 | 0); | |
16a53ecc N |
1824 | compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */ |
1825 | return; | |
1826 | } | |
1827 | } | |
1828 | ||
d0dabf7e | 1829 | /* We're missing D+P or D+D; */ |
67cc2b81 | 1830 | if (failb == syndrome_disks) { |
d0dabf7e | 1831 | /* We're missing D+P. */ |
67cc2b81 | 1832 | raid6_datap_recov(syndrome_disks+2, STRIPE_SIZE, faila, ptrs); |
d0dabf7e N |
1833 | } else { |
1834 | /* We're missing D+D. */ | |
67cc2b81 N |
1835 | raid6_2data_recov(syndrome_disks+2, STRIPE_SIZE, faila, failb, |
1836 | ptrs); | |
16a53ecc | 1837 | } |
d0dabf7e N |
1838 | |
1839 | /* Both the above update both missing blocks */ | |
1840 | set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); | |
1841 | set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); | |
16a53ecc N |
1842 | } |
1843 | ||
600aa109 | 1844 | static void |
1fe797e6 | 1845 | schedule_reconstruction5(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 1846 | int rcw, int expand) |
e33129d8 DW |
1847 | { |
1848 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
e33129d8 DW |
1849 | |
1850 | if (rcw) { | |
1851 | /* if we are not expanding this is a proper write request, and | |
1852 | * there will be bios with new data to be drained into the | |
1853 | * stripe cache | |
1854 | */ | |
1855 | if (!expand) { | |
600aa109 DW |
1856 | sh->reconstruct_state = reconstruct_state_drain_run; |
1857 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1858 | } else | |
1859 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 1860 | |
600aa109 | 1861 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); |
e33129d8 DW |
1862 | |
1863 | for (i = disks; i--; ) { | |
1864 | struct r5dev *dev = &sh->dev[i]; | |
1865 | ||
1866 | if (dev->towrite) { | |
1867 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 1868 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
1869 | if (!expand) |
1870 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1871 | s->locked++; |
e33129d8 DW |
1872 | } |
1873 | } | |
600aa109 | 1874 | if (s->locked + 1 == disks) |
8b3e6cdc DW |
1875 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
1876 | atomic_inc(&sh->raid_conf->pending_full_writes); | |
e33129d8 DW |
1877 | } else { |
1878 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
1879 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
1880 | ||
d8ee0728 | 1881 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
1882 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
1883 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1884 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); | |
e33129d8 DW |
1885 | |
1886 | for (i = disks; i--; ) { | |
1887 | struct r5dev *dev = &sh->dev[i]; | |
1888 | if (i == pd_idx) | |
1889 | continue; | |
1890 | ||
e33129d8 DW |
1891 | if (dev->towrite && |
1892 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
1893 | test_bit(R5_Wantcompute, &dev->flags))) { |
1894 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
1895 | set_bit(R5_LOCKED, &dev->flags); |
1896 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1897 | s->locked++; |
e33129d8 DW |
1898 | } |
1899 | } | |
1900 | } | |
1901 | ||
1902 | /* keep the parity disk locked while asynchronous operations | |
1903 | * are in flight | |
1904 | */ | |
1905 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1906 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 1907 | s->locked++; |
e33129d8 | 1908 | |
600aa109 | 1909 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 1910 | __func__, (unsigned long long)sh->sector, |
600aa109 | 1911 | s->locked, s->ops_request); |
e33129d8 | 1912 | } |
16a53ecc | 1913 | |
1da177e4 LT |
1914 | /* |
1915 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 1916 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
1917 | * The bi_next chain must be in order. |
1918 | */ | |
1919 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
1920 | { | |
1921 | struct bio **bip; | |
1922 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 1923 | int firstwrite=0; |
1da177e4 | 1924 | |
45b4233c | 1925 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
1926 | (unsigned long long)bi->bi_sector, |
1927 | (unsigned long long)sh->sector); | |
1928 | ||
1929 | ||
1930 | spin_lock(&sh->lock); | |
1931 | spin_lock_irq(&conf->device_lock); | |
72626685 | 1932 | if (forwrite) { |
1da177e4 | 1933 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
1934 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
1935 | firstwrite = 1; | |
1936 | } else | |
1da177e4 LT |
1937 | bip = &sh->dev[dd_idx].toread; |
1938 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
1939 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
1940 | goto overlap; | |
1941 | bip = & (*bip)->bi_next; | |
1942 | } | |
1943 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
1944 | goto overlap; | |
1945 | ||
78bafebd | 1946 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
1947 | if (*bip) |
1948 | bi->bi_next = *bip; | |
1949 | *bip = bi; | |
960e739d | 1950 | bi->bi_phys_segments++; |
1da177e4 LT |
1951 | spin_unlock_irq(&conf->device_lock); |
1952 | spin_unlock(&sh->lock); | |
1953 | ||
45b4233c | 1954 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
1955 | (unsigned long long)bi->bi_sector, |
1956 | (unsigned long long)sh->sector, dd_idx); | |
1957 | ||
72626685 | 1958 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
1959 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
1960 | STRIPE_SECTORS, 0); | |
ae3c20cc | 1961 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
1962 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
1963 | } | |
1964 | ||
1da177e4 LT |
1965 | if (forwrite) { |
1966 | /* check if page is covered */ | |
1967 | sector_t sector = sh->dev[dd_idx].sector; | |
1968 | for (bi=sh->dev[dd_idx].towrite; | |
1969 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
1970 | bi && bi->bi_sector <= sector; | |
1971 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
1972 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
1973 | sector = bi->bi_sector + (bi->bi_size>>9); | |
1974 | } | |
1975 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
1976 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
1977 | } | |
1978 | return 1; | |
1979 | ||
1980 | overlap: | |
1981 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
1982 | spin_unlock_irq(&conf->device_lock); | |
1983 | spin_unlock(&sh->lock); | |
1984 | return 0; | |
1985 | } | |
1986 | ||
29269553 N |
1987 | static void end_reshape(raid5_conf_t *conf); |
1988 | ||
16a53ecc N |
1989 | static int page_is_zero(struct page *p) |
1990 | { | |
1991 | char *a = page_address(p); | |
1992 | return ((*(u32*)a) == 0 && | |
1993 | memcmp(a, a+4, STRIPE_SIZE-4)==0); | |
1994 | } | |
1995 | ||
911d4ee8 N |
1996 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
1997 | struct stripe_head *sh) | |
ccfcc3c1 N |
1998 | { |
1999 | int sectors_per_chunk = conf->chunk_size >> 9; | |
911d4ee8 | 2000 | int dd_idx; |
2d2063ce | 2001 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2002 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2003 | |
112bf897 N |
2004 | raid5_compute_sector(conf, |
2005 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2006 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2007 | previous, |
911d4ee8 | 2008 | &dd_idx, sh); |
ccfcc3c1 N |
2009 | } |
2010 | ||
a4456856 | 2011 | static void |
1fe797e6 | 2012 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2013 | struct stripe_head_state *s, int disks, |
2014 | struct bio **return_bi) | |
2015 | { | |
2016 | int i; | |
2017 | for (i = disks; i--; ) { | |
2018 | struct bio *bi; | |
2019 | int bitmap_end = 0; | |
2020 | ||
2021 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2022 | mdk_rdev_t *rdev; | |
2023 | rcu_read_lock(); | |
2024 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2025 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2026 | /* multiple read failures in one stripe */ | |
2027 | md_error(conf->mddev, rdev); | |
2028 | rcu_read_unlock(); | |
2029 | } | |
2030 | spin_lock_irq(&conf->device_lock); | |
2031 | /* fail all writes first */ | |
2032 | bi = sh->dev[i].towrite; | |
2033 | sh->dev[i].towrite = NULL; | |
2034 | if (bi) { | |
2035 | s->to_write--; | |
2036 | bitmap_end = 1; | |
2037 | } | |
2038 | ||
2039 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2040 | wake_up(&conf->wait_for_overlap); | |
2041 | ||
2042 | while (bi && bi->bi_sector < | |
2043 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2044 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2045 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2046 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2047 | md_write_end(conf->mddev); |
2048 | bi->bi_next = *return_bi; | |
2049 | *return_bi = bi; | |
2050 | } | |
2051 | bi = nextbi; | |
2052 | } | |
2053 | /* and fail all 'written' */ | |
2054 | bi = sh->dev[i].written; | |
2055 | sh->dev[i].written = NULL; | |
2056 | if (bi) bitmap_end = 1; | |
2057 | while (bi && bi->bi_sector < | |
2058 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2059 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2060 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2061 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2062 | md_write_end(conf->mddev); |
2063 | bi->bi_next = *return_bi; | |
2064 | *return_bi = bi; | |
2065 | } | |
2066 | bi = bi2; | |
2067 | } | |
2068 | ||
b5e98d65 DW |
2069 | /* fail any reads if this device is non-operational and |
2070 | * the data has not reached the cache yet. | |
2071 | */ | |
2072 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2073 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2074 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2075 | bi = sh->dev[i].toread; |
2076 | sh->dev[i].toread = NULL; | |
2077 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2078 | wake_up(&conf->wait_for_overlap); | |
2079 | if (bi) s->to_read--; | |
2080 | while (bi && bi->bi_sector < | |
2081 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2082 | struct bio *nextbi = | |
2083 | r5_next_bio(bi, sh->dev[i].sector); | |
2084 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2085 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2086 | bi->bi_next = *return_bi; |
2087 | *return_bi = bi; | |
2088 | } | |
2089 | bi = nextbi; | |
2090 | } | |
2091 | } | |
2092 | spin_unlock_irq(&conf->device_lock); | |
2093 | if (bitmap_end) | |
2094 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2095 | STRIPE_SECTORS, 0, 0); | |
2096 | } | |
2097 | ||
8b3e6cdc DW |
2098 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2099 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2100 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2101 | } |
2102 | ||
1fe797e6 DW |
2103 | /* fetch_block5 - checks the given member device to see if its data needs |
2104 | * to be read or computed to satisfy a request. | |
2105 | * | |
2106 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2107 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2108 | */ |
1fe797e6 DW |
2109 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2110 | int disk_idx, int disks) | |
f38e1219 DW |
2111 | { |
2112 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2113 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2114 | ||
f38e1219 DW |
2115 | /* is the data in this block needed, and can we get it? */ |
2116 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2117 | !test_bit(R5_UPTODATE, &dev->flags) && |
2118 | (dev->toread || | |
2119 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2120 | s->syncing || s->expanding || | |
2121 | (s->failed && | |
2122 | (failed_dev->toread || | |
2123 | (failed_dev->towrite && | |
2124 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2125 | /* We would like to get this block, possibly by computing it, |
2126 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2127 | */ |
2128 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2129 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2130 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2131 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2132 | set_bit(R5_Wantcompute, &dev->flags); |
2133 | sh->ops.target = disk_idx; | |
2134 | s->req_compute = 1; | |
f38e1219 DW |
2135 | /* Careful: from this point on 'uptodate' is in the eye |
2136 | * of raid5_run_ops which services 'compute' operations | |
2137 | * before writes. R5_Wantcompute flags a block that will | |
2138 | * be R5_UPTODATE by the time it is needed for a | |
2139 | * subsequent operation. | |
2140 | */ | |
2141 | s->uptodate++; | |
1fe797e6 | 2142 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2143 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2144 | set_bit(R5_LOCKED, &dev->flags); |
2145 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2146 | s->locked++; |
2147 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2148 | s->syncing); | |
2149 | } | |
2150 | } | |
2151 | ||
1fe797e6 | 2152 | return 0; |
f38e1219 DW |
2153 | } |
2154 | ||
1fe797e6 DW |
2155 | /** |
2156 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2157 | */ | |
2158 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2159 | struct stripe_head_state *s, int disks) |
2160 | { | |
2161 | int i; | |
f38e1219 | 2162 | |
f38e1219 DW |
2163 | /* look for blocks to read/compute, skip this if a compute |
2164 | * is already in flight, or if the stripe contents are in the | |
2165 | * midst of changing due to a write | |
2166 | */ | |
976ea8d4 | 2167 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2168 | !sh->reconstruct_state) |
f38e1219 | 2169 | for (i = disks; i--; ) |
1fe797e6 | 2170 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2171 | break; |
a4456856 DW |
2172 | set_bit(STRIPE_HANDLE, &sh->state); |
2173 | } | |
2174 | ||
1fe797e6 | 2175 | static void handle_stripe_fill6(struct stripe_head *sh, |
a4456856 DW |
2176 | struct stripe_head_state *s, struct r6_state *r6s, |
2177 | int disks) | |
2178 | { | |
2179 | int i; | |
2180 | for (i = disks; i--; ) { | |
2181 | struct r5dev *dev = &sh->dev[i]; | |
2182 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2183 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2184 | (dev->toread || (dev->towrite && | |
2185 | !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2186 | s->syncing || s->expanding || | |
2187 | (s->failed >= 1 && | |
2188 | (sh->dev[r6s->failed_num[0]].toread || | |
2189 | s->to_write)) || | |
2190 | (s->failed >= 2 && | |
2191 | (sh->dev[r6s->failed_num[1]].toread || | |
2192 | s->to_write)))) { | |
2193 | /* we would like to get this block, possibly | |
2194 | * by computing it, but we might not be able to | |
2195 | */ | |
c337869d DW |
2196 | if ((s->uptodate == disks - 1) && |
2197 | (s->failed && (i == r6s->failed_num[0] || | |
2198 | i == r6s->failed_num[1]))) { | |
45b4233c | 2199 | pr_debug("Computing stripe %llu block %d\n", |
a4456856 DW |
2200 | (unsigned long long)sh->sector, i); |
2201 | compute_block_1(sh, i, 0); | |
2202 | s->uptodate++; | |
2203 | } else if ( s->uptodate == disks-2 && s->failed >= 2 ) { | |
2204 | /* Computing 2-failure is *very* expensive; only | |
2205 | * do it if failed >= 2 | |
2206 | */ | |
2207 | int other; | |
2208 | for (other = disks; other--; ) { | |
2209 | if (other == i) | |
2210 | continue; | |
2211 | if (!test_bit(R5_UPTODATE, | |
2212 | &sh->dev[other].flags)) | |
2213 | break; | |
2214 | } | |
2215 | BUG_ON(other < 0); | |
45b4233c | 2216 | pr_debug("Computing stripe %llu blocks %d,%d\n", |
a4456856 DW |
2217 | (unsigned long long)sh->sector, |
2218 | i, other); | |
2219 | compute_block_2(sh, i, other); | |
2220 | s->uptodate += 2; | |
2221 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2222 | set_bit(R5_LOCKED, &dev->flags); | |
2223 | set_bit(R5_Wantread, &dev->flags); | |
2224 | s->locked++; | |
45b4233c | 2225 | pr_debug("Reading block %d (sync=%d)\n", |
a4456856 DW |
2226 | i, s->syncing); |
2227 | } | |
2228 | } | |
2229 | } | |
2230 | set_bit(STRIPE_HANDLE, &sh->state); | |
2231 | } | |
2232 | ||
2233 | ||
1fe797e6 | 2234 | /* handle_stripe_clean_event |
a4456856 DW |
2235 | * any written block on an uptodate or failed drive can be returned. |
2236 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2237 | * never LOCKED, so we don't need to test 'failed' directly. | |
2238 | */ | |
1fe797e6 | 2239 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2240 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2241 | { | |
2242 | int i; | |
2243 | struct r5dev *dev; | |
2244 | ||
2245 | for (i = disks; i--; ) | |
2246 | if (sh->dev[i].written) { | |
2247 | dev = &sh->dev[i]; | |
2248 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2249 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2250 | /* We can return any write requests */ | |
2251 | struct bio *wbi, *wbi2; | |
2252 | int bitmap_end = 0; | |
45b4233c | 2253 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2254 | spin_lock_irq(&conf->device_lock); |
2255 | wbi = dev->written; | |
2256 | dev->written = NULL; | |
2257 | while (wbi && wbi->bi_sector < | |
2258 | dev->sector + STRIPE_SECTORS) { | |
2259 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2260 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2261 | md_write_end(conf->mddev); |
2262 | wbi->bi_next = *return_bi; | |
2263 | *return_bi = wbi; | |
2264 | } | |
2265 | wbi = wbi2; | |
2266 | } | |
2267 | if (dev->towrite == NULL) | |
2268 | bitmap_end = 1; | |
2269 | spin_unlock_irq(&conf->device_lock); | |
2270 | if (bitmap_end) | |
2271 | bitmap_endwrite(conf->mddev->bitmap, | |
2272 | sh->sector, | |
2273 | STRIPE_SECTORS, | |
2274 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2275 | 0); | |
2276 | } | |
2277 | } | |
8b3e6cdc DW |
2278 | |
2279 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2280 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2281 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2282 | } |
2283 | ||
1fe797e6 | 2284 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2285 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2286 | { | |
2287 | int rmw = 0, rcw = 0, i; | |
2288 | for (i = disks; i--; ) { | |
2289 | /* would I have to read this buffer for read_modify_write */ | |
2290 | struct r5dev *dev = &sh->dev[i]; | |
2291 | if ((dev->towrite || i == sh->pd_idx) && | |
2292 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2293 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2294 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2295 | if (test_bit(R5_Insync, &dev->flags)) |
2296 | rmw++; | |
2297 | else | |
2298 | rmw += 2*disks; /* cannot read it */ | |
2299 | } | |
2300 | /* Would I have to read this buffer for reconstruct_write */ | |
2301 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2302 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2303 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2304 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2305 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2306 | else |
2307 | rcw += 2*disks; | |
2308 | } | |
2309 | } | |
45b4233c | 2310 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2311 | (unsigned long long)sh->sector, rmw, rcw); |
2312 | set_bit(STRIPE_HANDLE, &sh->state); | |
2313 | if (rmw < rcw && rmw > 0) | |
2314 | /* prefer read-modify-write, but need to get some data */ | |
2315 | for (i = disks; i--; ) { | |
2316 | struct r5dev *dev = &sh->dev[i]; | |
2317 | if ((dev->towrite || i == sh->pd_idx) && | |
2318 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2319 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2320 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2321 | test_bit(R5_Insync, &dev->flags)) { |
2322 | if ( | |
2323 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2324 | pr_debug("Read_old block " |
a4456856 DW |
2325 | "%d for r-m-w\n", i); |
2326 | set_bit(R5_LOCKED, &dev->flags); | |
2327 | set_bit(R5_Wantread, &dev->flags); | |
2328 | s->locked++; | |
2329 | } else { | |
2330 | set_bit(STRIPE_DELAYED, &sh->state); | |
2331 | set_bit(STRIPE_HANDLE, &sh->state); | |
2332 | } | |
2333 | } | |
2334 | } | |
2335 | if (rcw <= rmw && rcw > 0) | |
2336 | /* want reconstruct write, but need to get some data */ | |
2337 | for (i = disks; i--; ) { | |
2338 | struct r5dev *dev = &sh->dev[i]; | |
2339 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2340 | i != sh->pd_idx && | |
2341 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2342 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2343 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2344 | test_bit(R5_Insync, &dev->flags)) { |
2345 | if ( | |
2346 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2347 | pr_debug("Read_old block " |
a4456856 DW |
2348 | "%d for Reconstruct\n", i); |
2349 | set_bit(R5_LOCKED, &dev->flags); | |
2350 | set_bit(R5_Wantread, &dev->flags); | |
2351 | s->locked++; | |
2352 | } else { | |
2353 | set_bit(STRIPE_DELAYED, &sh->state); | |
2354 | set_bit(STRIPE_HANDLE, &sh->state); | |
2355 | } | |
2356 | } | |
2357 | } | |
2358 | /* now if nothing is locked, and if we have enough data, | |
2359 | * we can start a write request | |
2360 | */ | |
f38e1219 DW |
2361 | /* since handle_stripe can be called at any time we need to handle the |
2362 | * case where a compute block operation has been submitted and then a | |
2363 | * subsequent call wants to start a write request. raid5_run_ops only | |
2364 | * handles the case where compute block and postxor are requested | |
2365 | * simultaneously. If this is not the case then new writes need to be | |
2366 | * held off until the compute completes. | |
2367 | */ | |
976ea8d4 DW |
2368 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2369 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2370 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
1fe797e6 | 2371 | schedule_reconstruction5(sh, s, rcw == 0, 0); |
a4456856 DW |
2372 | } |
2373 | ||
1fe797e6 | 2374 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2375 | struct stripe_head *sh, struct stripe_head_state *s, |
2376 | struct r6_state *r6s, int disks) | |
2377 | { | |
2378 | int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i; | |
2379 | int qd_idx = r6s->qd_idx; | |
2380 | for (i = disks; i--; ) { | |
2381 | struct r5dev *dev = &sh->dev[i]; | |
2382 | /* Would I have to read this buffer for reconstruct_write */ | |
2383 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2384 | && i != pd_idx && i != qd_idx | |
2385 | && (!test_bit(R5_LOCKED, &dev->flags) | |
2386 | ) && | |
2387 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
2388 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
2389 | else { | |
45b4233c | 2390 | pr_debug("raid6: must_compute: " |
a4456856 DW |
2391 | "disk %d flags=%#lx\n", i, dev->flags); |
2392 | must_compute++; | |
2393 | } | |
2394 | } | |
2395 | } | |
45b4233c | 2396 | pr_debug("for sector %llu, rcw=%d, must_compute=%d\n", |
a4456856 DW |
2397 | (unsigned long long)sh->sector, rcw, must_compute); |
2398 | set_bit(STRIPE_HANDLE, &sh->state); | |
2399 | ||
2400 | if (rcw > 0) | |
2401 | /* want reconstruct write, but need to get some data */ | |
2402 | for (i = disks; i--; ) { | |
2403 | struct r5dev *dev = &sh->dev[i]; | |
2404 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2405 | && !(s->failed == 0 && (i == pd_idx || i == qd_idx)) | |
2406 | && !test_bit(R5_LOCKED, &dev->flags) && | |
2407 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2408 | test_bit(R5_Insync, &dev->flags)) { | |
2409 | if ( | |
2410 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2411 | pr_debug("Read_old stripe %llu " |
a4456856 DW |
2412 | "block %d for Reconstruct\n", |
2413 | (unsigned long long)sh->sector, i); | |
2414 | set_bit(R5_LOCKED, &dev->flags); | |
2415 | set_bit(R5_Wantread, &dev->flags); | |
2416 | s->locked++; | |
2417 | } else { | |
45b4233c | 2418 | pr_debug("Request delayed stripe %llu " |
a4456856 DW |
2419 | "block %d for Reconstruct\n", |
2420 | (unsigned long long)sh->sector, i); | |
2421 | set_bit(STRIPE_DELAYED, &sh->state); | |
2422 | set_bit(STRIPE_HANDLE, &sh->state); | |
2423 | } | |
2424 | } | |
2425 | } | |
2426 | /* now if nothing is locked, and if we have enough data, we can start a | |
2427 | * write request | |
2428 | */ | |
2429 | if (s->locked == 0 && rcw == 0 && | |
2430 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
2431 | if (must_compute > 0) { | |
2432 | /* We have failed blocks and need to compute them */ | |
2433 | switch (s->failed) { | |
2434 | case 0: | |
2435 | BUG(); | |
2436 | case 1: | |
2437 | compute_block_1(sh, r6s->failed_num[0], 0); | |
2438 | break; | |
2439 | case 2: | |
2440 | compute_block_2(sh, r6s->failed_num[0], | |
2441 | r6s->failed_num[1]); | |
2442 | break; | |
2443 | default: /* This request should have been failed? */ | |
2444 | BUG(); | |
2445 | } | |
2446 | } | |
2447 | ||
45b4233c | 2448 | pr_debug("Computing parity for stripe %llu\n", |
a4456856 DW |
2449 | (unsigned long long)sh->sector); |
2450 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
2451 | /* now every locked buffer is ready to be written */ | |
2452 | for (i = disks; i--; ) | |
2453 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
45b4233c | 2454 | pr_debug("Writing stripe %llu block %d\n", |
a4456856 DW |
2455 | (unsigned long long)sh->sector, i); |
2456 | s->locked++; | |
2457 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
2458 | } | |
8b3e6cdc DW |
2459 | if (s->locked == disks) |
2460 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2461 | atomic_inc(&conf->pending_full_writes); | |
a4456856 DW |
2462 | /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */ |
2463 | set_bit(STRIPE_INSYNC, &sh->state); | |
2464 | ||
2465 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2466 | atomic_dec(&conf->preread_active_stripes); | |
2467 | if (atomic_read(&conf->preread_active_stripes) < | |
2468 | IO_THRESHOLD) | |
2469 | md_wakeup_thread(conf->mddev->thread); | |
2470 | } | |
2471 | } | |
2472 | } | |
2473 | ||
2474 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2475 | struct stripe_head_state *s, int disks) | |
2476 | { | |
ecc65c9b | 2477 | struct r5dev *dev = NULL; |
bd2ab670 | 2478 | |
a4456856 | 2479 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2480 | |
ecc65c9b DW |
2481 | switch (sh->check_state) { |
2482 | case check_state_idle: | |
2483 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2484 | if (s->failed == 0) { |
bd2ab670 | 2485 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2486 | sh->check_state = check_state_run; |
2487 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2488 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2489 | s->uptodate--; |
ecc65c9b | 2490 | break; |
bd2ab670 | 2491 | } |
ecc65c9b DW |
2492 | dev = &sh->dev[s->failed_num]; |
2493 | /* fall through */ | |
2494 | case check_state_compute_result: | |
2495 | sh->check_state = check_state_idle; | |
2496 | if (!dev) | |
2497 | dev = &sh->dev[sh->pd_idx]; | |
2498 | ||
2499 | /* check that a write has not made the stripe insync */ | |
2500 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2501 | break; | |
c8894419 | 2502 | |
a4456856 | 2503 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2504 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2505 | BUG_ON(s->uptodate != disks); | |
2506 | ||
2507 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2508 | s->locked++; |
a4456856 | 2509 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2510 | |
a4456856 | 2511 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2512 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2513 | break; |
2514 | case check_state_run: | |
2515 | break; /* we will be called again upon completion */ | |
2516 | case check_state_check_result: | |
2517 | sh->check_state = check_state_idle; | |
2518 | ||
2519 | /* if a failure occurred during the check operation, leave | |
2520 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2521 | */ | |
2522 | if (s->failed) | |
2523 | break; | |
2524 | ||
2525 | /* handle a successful check operation, if parity is correct | |
2526 | * we are done. Otherwise update the mismatch count and repair | |
2527 | * parity if !MD_RECOVERY_CHECK | |
2528 | */ | |
2529 | if (sh->ops.zero_sum_result == 0) | |
2530 | /* parity is correct (on disc, | |
2531 | * not in buffer any more) | |
2532 | */ | |
2533 | set_bit(STRIPE_INSYNC, &sh->state); | |
2534 | else { | |
2535 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2536 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2537 | /* don't try to repair!! */ | |
2538 | set_bit(STRIPE_INSYNC, &sh->state); | |
2539 | else { | |
2540 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2541 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2542 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2543 | set_bit(R5_Wantcompute, | |
2544 | &sh->dev[sh->pd_idx].flags); | |
2545 | sh->ops.target = sh->pd_idx; | |
2546 | s->uptodate++; | |
2547 | } | |
2548 | } | |
2549 | break; | |
2550 | case check_state_compute_run: | |
2551 | break; | |
2552 | default: | |
2553 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2554 | __func__, sh->check_state, | |
2555 | (unsigned long long) sh->sector); | |
2556 | BUG(); | |
a4456856 DW |
2557 | } |
2558 | } | |
2559 | ||
2560 | ||
2561 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
2562 | struct stripe_head_state *s, | |
2563 | struct r6_state *r6s, struct page *tmp_page, | |
2564 | int disks) | |
2565 | { | |
2566 | int update_p = 0, update_q = 0; | |
2567 | struct r5dev *dev; | |
2568 | int pd_idx = sh->pd_idx; | |
2569 | int qd_idx = r6s->qd_idx; | |
2570 | ||
2571 | set_bit(STRIPE_HANDLE, &sh->state); | |
2572 | ||
2573 | BUG_ON(s->failed > 2); | |
2574 | BUG_ON(s->uptodate < disks); | |
2575 | /* Want to check and possibly repair P and Q. | |
2576 | * However there could be one 'failed' device, in which | |
2577 | * case we can only check one of them, possibly using the | |
2578 | * other to generate missing data | |
2579 | */ | |
2580 | ||
2581 | /* If !tmp_page, we cannot do the calculations, | |
2582 | * but as we have set STRIPE_HANDLE, we will soon be called | |
2583 | * by stripe_handle with a tmp_page - just wait until then. | |
2584 | */ | |
2585 | if (tmp_page) { | |
2586 | if (s->failed == r6s->q_failed) { | |
2587 | /* The only possible failed device holds 'Q', so it | |
2588 | * makes sense to check P (If anything else were failed, | |
2589 | * we would have used P to recreate it). | |
2590 | */ | |
2591 | compute_block_1(sh, pd_idx, 1); | |
2592 | if (!page_is_zero(sh->dev[pd_idx].page)) { | |
2593 | compute_block_1(sh, pd_idx, 0); | |
2594 | update_p = 1; | |
2595 | } | |
2596 | } | |
2597 | if (!r6s->q_failed && s->failed < 2) { | |
2598 | /* q is not failed, and we didn't use it to generate | |
2599 | * anything, so it makes sense to check it | |
2600 | */ | |
2601 | memcpy(page_address(tmp_page), | |
2602 | page_address(sh->dev[qd_idx].page), | |
2603 | STRIPE_SIZE); | |
2604 | compute_parity6(sh, UPDATE_PARITY); | |
2605 | if (memcmp(page_address(tmp_page), | |
2606 | page_address(sh->dev[qd_idx].page), | |
2607 | STRIPE_SIZE) != 0) { | |
2608 | clear_bit(STRIPE_INSYNC, &sh->state); | |
2609 | update_q = 1; | |
2610 | } | |
2611 | } | |
2612 | if (update_p || update_q) { | |
2613 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2614 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2615 | /* don't try to repair!! */ | |
2616 | update_p = update_q = 0; | |
2617 | } | |
2618 | ||
2619 | /* now write out any block on a failed drive, | |
2620 | * or P or Q if they need it | |
2621 | */ | |
2622 | ||
2623 | if (s->failed == 2) { | |
2624 | dev = &sh->dev[r6s->failed_num[1]]; | |
2625 | s->locked++; | |
2626 | set_bit(R5_LOCKED, &dev->flags); | |
2627 | set_bit(R5_Wantwrite, &dev->flags); | |
2628 | } | |
2629 | if (s->failed >= 1) { | |
2630 | dev = &sh->dev[r6s->failed_num[0]]; | |
2631 | s->locked++; | |
2632 | set_bit(R5_LOCKED, &dev->flags); | |
2633 | set_bit(R5_Wantwrite, &dev->flags); | |
2634 | } | |
2635 | ||
2636 | if (update_p) { | |
2637 | dev = &sh->dev[pd_idx]; | |
2638 | s->locked++; | |
2639 | set_bit(R5_LOCKED, &dev->flags); | |
2640 | set_bit(R5_Wantwrite, &dev->flags); | |
2641 | } | |
2642 | if (update_q) { | |
2643 | dev = &sh->dev[qd_idx]; | |
2644 | s->locked++; | |
2645 | set_bit(R5_LOCKED, &dev->flags); | |
2646 | set_bit(R5_Wantwrite, &dev->flags); | |
2647 | } | |
2648 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2649 | ||
2650 | set_bit(STRIPE_INSYNC, &sh->state); | |
2651 | } | |
2652 | } | |
2653 | ||
2654 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2655 | struct r6_state *r6s) | |
2656 | { | |
2657 | int i; | |
2658 | ||
2659 | /* We have read all the blocks in this stripe and now we need to | |
2660 | * copy some of them into a target stripe for expand. | |
2661 | */ | |
f0a50d37 | 2662 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2663 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2664 | for (i = 0; i < sh->disks; i++) | |
a2e08551 | 2665 | if (i != sh->pd_idx && (!r6s || i != r6s->qd_idx)) { |
911d4ee8 | 2666 | int dd_idx, j; |
a4456856 DW |
2667 | struct stripe_head *sh2; |
2668 | ||
2669 | sector_t bn = compute_blocknr(sh, i); | |
911d4ee8 N |
2670 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2671 | &dd_idx, NULL); | |
b5663ba4 | 2672 | sh2 = get_active_stripe(conf, s, 0, 1); |
a4456856 DW |
2673 | if (sh2 == NULL) |
2674 | /* so far only the early blocks of this stripe | |
2675 | * have been requested. When later blocks | |
2676 | * get requested, we will try again | |
2677 | */ | |
2678 | continue; | |
2679 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2680 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2681 | /* must have already done this block */ | |
2682 | release_stripe(sh2); | |
2683 | continue; | |
2684 | } | |
f0a50d37 DW |
2685 | |
2686 | /* place all the copies on one channel */ | |
2687 | tx = async_memcpy(sh2->dev[dd_idx].page, | |
2688 | sh->dev[i].page, 0, 0, STRIPE_SIZE, | |
2689 | ASYNC_TX_DEP_ACK, tx, NULL, NULL); | |
2690 | ||
a4456856 DW |
2691 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2692 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2693 | for (j = 0; j < conf->raid_disks; j++) | |
2694 | if (j != sh2->pd_idx && | |
d0dabf7e | 2695 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
2696 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2697 | break; | |
2698 | if (j == conf->raid_disks) { | |
2699 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2700 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2701 | } | |
2702 | release_stripe(sh2); | |
f0a50d37 | 2703 | |
a4456856 | 2704 | } |
a2e08551 N |
2705 | /* done submitting copies, wait for them to complete */ |
2706 | if (tx) { | |
2707 | async_tx_ack(tx); | |
2708 | dma_wait_for_async_tx(tx); | |
2709 | } | |
a4456856 | 2710 | } |
1da177e4 | 2711 | |
6bfe0b49 | 2712 | |
1da177e4 LT |
2713 | /* |
2714 | * handle_stripe - do things to a stripe. | |
2715 | * | |
2716 | * We lock the stripe and then examine the state of various bits | |
2717 | * to see what needs to be done. | |
2718 | * Possible results: | |
2719 | * return some read request which now have data | |
2720 | * return some write requests which are safely on disc | |
2721 | * schedule a read on some buffers | |
2722 | * schedule a write of some buffers | |
2723 | * return confirmation of parity correctness | |
2724 | * | |
1da177e4 LT |
2725 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2726 | * get BH_Lock set before the stripe lock is released. | |
2727 | * | |
2728 | */ | |
a4456856 | 2729 | |
df10cfbc | 2730 | static bool handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2731 | { |
2732 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2733 | int disks = sh->disks, i; |
2734 | struct bio *return_bi = NULL; | |
2735 | struct stripe_head_state s; | |
1da177e4 | 2736 | struct r5dev *dev; |
6bfe0b49 | 2737 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2738 | int prexor; |
1da177e4 | 2739 | |
a4456856 | 2740 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
2741 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
2742 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
2743 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
2744 | sh->reconstruct_state); | |
1da177e4 LT |
2745 | |
2746 | spin_lock(&sh->lock); | |
2747 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2748 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2749 | ||
a4456856 DW |
2750 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2751 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2752 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 2753 | |
83de75cc | 2754 | /* Now to look around and see what can be done */ |
9910f16a | 2755 | rcu_read_lock(); |
1da177e4 LT |
2756 | for (i=disks; i--; ) { |
2757 | mdk_rdev_t *rdev; | |
a4456856 | 2758 | struct r5dev *dev = &sh->dev[i]; |
1da177e4 | 2759 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2760 | |
b5e98d65 DW |
2761 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2762 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2763 | dev->towrite, dev->written); | |
2764 | ||
2765 | /* maybe we can request a biofill operation | |
2766 | * | |
2767 | * new wantfill requests are only permitted while | |
83de75cc | 2768 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
2769 | */ |
2770 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 2771 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 2772 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
2773 | |
2774 | /* now count some things */ | |
a4456856 DW |
2775 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2776 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2777 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2778 | |
b5e98d65 DW |
2779 | if (test_bit(R5_Wantfill, &dev->flags)) |
2780 | s.to_fill++; | |
2781 | else if (dev->toread) | |
a4456856 | 2782 | s.to_read++; |
1da177e4 | 2783 | if (dev->towrite) { |
a4456856 | 2784 | s.to_write++; |
1da177e4 | 2785 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2786 | s.non_overwrite++; |
1da177e4 | 2787 | } |
a4456856 DW |
2788 | if (dev->written) |
2789 | s.written++; | |
9910f16a | 2790 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
2791 | if (blocked_rdev == NULL && |
2792 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
2793 | blocked_rdev = rdev; |
2794 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 2795 | } |
b2d444d7 | 2796 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 2797 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
2798 | clear_bit(R5_ReadError, &dev->flags); |
2799 | clear_bit(R5_ReWrite, &dev->flags); | |
2800 | } | |
b2d444d7 | 2801 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 2802 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
2803 | s.failed++; |
2804 | s.failed_num = i; | |
1da177e4 LT |
2805 | } else |
2806 | set_bit(R5_Insync, &dev->flags); | |
2807 | } | |
9910f16a | 2808 | rcu_read_unlock(); |
b5e98d65 | 2809 | |
6bfe0b49 | 2810 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
2811 | if (s.syncing || s.expanding || s.expanded || |
2812 | s.to_write || s.written) { | |
2813 | set_bit(STRIPE_HANDLE, &sh->state); | |
2814 | goto unlock; | |
2815 | } | |
2816 | /* There is nothing for the blocked_rdev to block */ | |
2817 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
2818 | blocked_rdev = NULL; | |
6bfe0b49 DW |
2819 | } |
2820 | ||
83de75cc DW |
2821 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
2822 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
2823 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
2824 | } | |
b5e98d65 | 2825 | |
45b4233c | 2826 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 2827 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
2828 | s.locked, s.uptodate, s.to_read, s.to_write, |
2829 | s.failed, s.failed_num); | |
1da177e4 LT |
2830 | /* check if the array has lost two devices and, if so, some requests might |
2831 | * need to be failed | |
2832 | */ | |
a4456856 | 2833 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 2834 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 2835 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
2836 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2837 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2838 | s.syncing = 0; |
1da177e4 LT |
2839 | } |
2840 | ||
2841 | /* might be able to return some write requests if the parity block | |
2842 | * is safe, or on a failed drive | |
2843 | */ | |
2844 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
2845 | if ( s.written && |
2846 | ((test_bit(R5_Insync, &dev->flags) && | |
2847 | !test_bit(R5_LOCKED, &dev->flags) && | |
2848 | test_bit(R5_UPTODATE, &dev->flags)) || | |
2849 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 2850 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
2851 | |
2852 | /* Now we might consider reading some blocks, either to check/generate | |
2853 | * parity, or to satisfy requests | |
2854 | * or to load a block that is being partially written. | |
2855 | */ | |
a4456856 | 2856 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 2857 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 2858 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 2859 | |
e33129d8 DW |
2860 | /* Now we check to see if any write operations have recently |
2861 | * completed | |
2862 | */ | |
e0a115e5 | 2863 | prexor = 0; |
d8ee0728 | 2864 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 2865 | prexor = 1; |
d8ee0728 DW |
2866 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
2867 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 2868 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
2869 | |
2870 | /* All the 'written' buffers and the parity block are ready to | |
2871 | * be written back to disk | |
2872 | */ | |
2873 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
2874 | for (i = disks; i--; ) { | |
2875 | dev = &sh->dev[i]; | |
2876 | if (test_bit(R5_LOCKED, &dev->flags) && | |
2877 | (i == sh->pd_idx || dev->written)) { | |
2878 | pr_debug("Writing block %d\n", i); | |
2879 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
2880 | if (prexor) |
2881 | continue; | |
e33129d8 DW |
2882 | if (!test_bit(R5_Insync, &dev->flags) || |
2883 | (i == sh->pd_idx && s.failed == 0)) | |
2884 | set_bit(STRIPE_INSYNC, &sh->state); | |
2885 | } | |
2886 | } | |
2887 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2888 | atomic_dec(&conf->preread_active_stripes); | |
2889 | if (atomic_read(&conf->preread_active_stripes) < | |
2890 | IO_THRESHOLD) | |
2891 | md_wakeup_thread(conf->mddev->thread); | |
2892 | } | |
2893 | } | |
2894 | ||
2895 | /* Now to consider new write requests and what else, if anything | |
2896 | * should be read. We do not handle new writes when: | |
2897 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
2898 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
2899 | * block. | |
2900 | */ | |
600aa109 | 2901 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 2902 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
2903 | |
2904 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
2905 | * Any reads will already have been scheduled, so we just see if enough |
2906 | * data is available. The parity check is held off while parity | |
2907 | * dependent operations are in flight. | |
1da177e4 | 2908 | */ |
ecc65c9b DW |
2909 | if (sh->check_state || |
2910 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 2911 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 2912 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 2913 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 2914 | |
a4456856 | 2915 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
2916 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2917 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2918 | } | |
4e5314b5 N |
2919 | |
2920 | /* If the failed drive is just a ReadError, then we might need to progress | |
2921 | * the repair/check process | |
2922 | */ | |
a4456856 DW |
2923 | if (s.failed == 1 && !conf->mddev->ro && |
2924 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
2925 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
2926 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 2927 | ) { |
a4456856 | 2928 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
2929 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
2930 | set_bit(R5_Wantwrite, &dev->flags); | |
2931 | set_bit(R5_ReWrite, &dev->flags); | |
2932 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2933 | s.locked++; |
4e5314b5 N |
2934 | } else { |
2935 | /* let's read it back */ | |
2936 | set_bit(R5_Wantread, &dev->flags); | |
2937 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2938 | s.locked++; |
4e5314b5 N |
2939 | } |
2940 | } | |
2941 | ||
600aa109 DW |
2942 | /* Finish reconstruct operations initiated by the expansion process */ |
2943 | if (sh->reconstruct_state == reconstruct_state_result) { | |
2944 | sh->reconstruct_state = reconstruct_state_idle; | |
f0a50d37 | 2945 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 2946 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 2947 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 2948 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 2949 | s.locked++; |
23397883 | 2950 | } |
f0a50d37 DW |
2951 | } |
2952 | ||
2953 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 2954 | !sh->reconstruct_state) { |
f0a50d37 DW |
2955 | /* Need to write out all blocks after computing parity */ |
2956 | sh->disks = conf->raid_disks; | |
911d4ee8 | 2957 | stripe_set_idx(sh->sector, conf, 0, sh); |
1fe797e6 | 2958 | schedule_reconstruction5(sh, &s, 1, 1); |
600aa109 | 2959 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 2960 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 2961 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
2962 | wake_up(&conf->wait_for_overlap); |
2963 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
2964 | } | |
2965 | ||
0f94e87c | 2966 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 2967 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 2968 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 2969 | |
6bfe0b49 | 2970 | unlock: |
1da177e4 LT |
2971 | spin_unlock(&sh->lock); |
2972 | ||
6bfe0b49 DW |
2973 | /* wait for this device to become unblocked */ |
2974 | if (unlikely(blocked_rdev)) | |
2975 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
2976 | ||
600aa109 DW |
2977 | if (s.ops_request) |
2978 | raid5_run_ops(sh, s.ops_request); | |
d84e0f10 | 2979 | |
c4e5ac0a | 2980 | ops_run_io(sh, &s); |
1da177e4 | 2981 | |
a4456856 | 2982 | return_io(return_bi); |
df10cfbc DW |
2983 | |
2984 | return blocked_rdev == NULL; | |
1da177e4 LT |
2985 | } |
2986 | ||
df10cfbc | 2987 | static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page) |
1da177e4 | 2988 | { |
bff61975 | 2989 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 2990 | int disks = sh->disks; |
a4456856 DW |
2991 | struct bio *return_bi = NULL; |
2992 | int i, pd_idx = sh->pd_idx; | |
2993 | struct stripe_head_state s; | |
2994 | struct r6_state r6s; | |
16a53ecc | 2995 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 2996 | mdk_rdev_t *blocked_rdev = NULL; |
1da177e4 | 2997 | |
d0dabf7e | 2998 | r6s.qd_idx = sh->qd_idx; |
45b4233c | 2999 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
a4456856 DW |
3000 | "pd_idx=%d, qd_idx=%d\n", |
3001 | (unsigned long long)sh->sector, sh->state, | |
3002 | atomic_read(&sh->count), pd_idx, r6s.qd_idx); | |
3003 | memset(&s, 0, sizeof(s)); | |
72626685 | 3004 | |
16a53ecc N |
3005 | spin_lock(&sh->lock); |
3006 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3007 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3008 | ||
a4456856 DW |
3009 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3010 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3011 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 3012 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3013 | |
3014 | rcu_read_lock(); | |
16a53ecc N |
3015 | for (i=disks; i--; ) { |
3016 | mdk_rdev_t *rdev; | |
3017 | dev = &sh->dev[i]; | |
3018 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 3019 | |
45b4233c | 3020 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc N |
3021 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
3022 | /* maybe we can reply to a read */ | |
3023 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
3024 | struct bio *rbi, *rbi2; | |
45b4233c | 3025 | pr_debug("Return read for disc %d\n", i); |
16a53ecc N |
3026 | spin_lock_irq(&conf->device_lock); |
3027 | rbi = dev->toread; | |
3028 | dev->toread = NULL; | |
3029 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
3030 | wake_up(&conf->wait_for_overlap); | |
3031 | spin_unlock_irq(&conf->device_lock); | |
3032 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
3033 | copy_data(0, rbi, dev->page, dev->sector); | |
3034 | rbi2 = r5_next_bio(rbi, dev->sector); | |
3035 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3036 | if (!raid5_dec_bi_phys_segments(rbi)) { |
16a53ecc N |
3037 | rbi->bi_next = return_bi; |
3038 | return_bi = rbi; | |
3039 | } | |
3040 | spin_unlock_irq(&conf->device_lock); | |
3041 | rbi = rbi2; | |
3042 | } | |
3043 | } | |
1da177e4 | 3044 | |
16a53ecc | 3045 | /* now count some things */ |
a4456856 DW |
3046 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3047 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 3048 | |
16a53ecc | 3049 | |
a4456856 DW |
3050 | if (dev->toread) |
3051 | s.to_read++; | |
16a53ecc | 3052 | if (dev->towrite) { |
a4456856 | 3053 | s.to_write++; |
16a53ecc | 3054 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3055 | s.non_overwrite++; |
16a53ecc | 3056 | } |
a4456856 DW |
3057 | if (dev->written) |
3058 | s.written++; | |
16a53ecc | 3059 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3060 | if (blocked_rdev == NULL && |
3061 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3062 | blocked_rdev = rdev; |
3063 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3064 | } |
16a53ecc N |
3065 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3066 | /* The ReadError flag will just be confusing now */ | |
3067 | clear_bit(R5_ReadError, &dev->flags); | |
3068 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3069 | } |
16a53ecc N |
3070 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3071 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3072 | if (s.failed < 2) |
3073 | r6s.failed_num[s.failed] = i; | |
3074 | s.failed++; | |
16a53ecc N |
3075 | } else |
3076 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3077 | } |
3078 | rcu_read_unlock(); | |
6bfe0b49 DW |
3079 | |
3080 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3081 | if (s.syncing || s.expanding || s.expanded || |
3082 | s.to_write || s.written) { | |
3083 | set_bit(STRIPE_HANDLE, &sh->state); | |
3084 | goto unlock; | |
3085 | } | |
3086 | /* There is nothing for the blocked_rdev to block */ | |
3087 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3088 | blocked_rdev = NULL; | |
6bfe0b49 | 3089 | } |
ac4090d2 | 3090 | |
45b4233c | 3091 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3092 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3093 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3094 | r6s.failed_num[0], r6s.failed_num[1]); | |
3095 | /* check if the array has lost >2 devices and, if so, some requests | |
3096 | * might need to be failed | |
16a53ecc | 3097 | */ |
a4456856 | 3098 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3099 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3100 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3101 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3102 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3103 | s.syncing = 0; |
16a53ecc N |
3104 | } |
3105 | ||
3106 | /* | |
3107 | * might be able to return some write requests if the parity blocks | |
3108 | * are safe, or on a failed drive | |
3109 | */ | |
3110 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3111 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3112 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
3113 | qdev = &sh->dev[r6s.qd_idx]; | |
3114 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == r6s.qd_idx) | |
3115 | || (s.failed >= 2 && r6s.failed_num[1] == r6s.qd_idx); | |
3116 | ||
3117 | if ( s.written && | |
3118 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3119 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3120 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3121 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3122 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3123 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3124 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3125 | |
3126 | /* Now we might consider reading some blocks, either to check/generate | |
3127 | * parity, or to satisfy requests | |
3128 | * or to load a block that is being partially written. | |
3129 | */ | |
a4456856 DW |
3130 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
3131 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
1fe797e6 | 3132 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc N |
3133 | |
3134 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 | 3135 | if (s.to_write) |
1fe797e6 | 3136 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3137 | |
3138 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 DW |
3139 | * Any reads will already have been scheduled, so we just see if enough |
3140 | * data is available | |
16a53ecc | 3141 | */ |
a4456856 DW |
3142 | if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state)) |
3143 | handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks); | |
16a53ecc | 3144 | |
a4456856 | 3145 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3146 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3147 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3148 | } | |
3149 | ||
3150 | /* If the failed drives are just a ReadError, then we might need | |
3151 | * to progress the repair/check process | |
3152 | */ | |
a4456856 DW |
3153 | if (s.failed <= 2 && !conf->mddev->ro) |
3154 | for (i = 0; i < s.failed; i++) { | |
3155 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3156 | if (test_bit(R5_ReadError, &dev->flags) |
3157 | && !test_bit(R5_LOCKED, &dev->flags) | |
3158 | && test_bit(R5_UPTODATE, &dev->flags) | |
3159 | ) { | |
3160 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3161 | set_bit(R5_Wantwrite, &dev->flags); | |
3162 | set_bit(R5_ReWrite, &dev->flags); | |
3163 | set_bit(R5_LOCKED, &dev->flags); | |
3164 | } else { | |
3165 | /* let's read it back */ | |
3166 | set_bit(R5_Wantread, &dev->flags); | |
3167 | set_bit(R5_LOCKED, &dev->flags); | |
3168 | } | |
3169 | } | |
3170 | } | |
f416885e | 3171 | |
a4456856 | 3172 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
f416885e N |
3173 | /* Need to write out all blocks after computing P&Q */ |
3174 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3175 | stripe_set_idx(sh->sector, conf, 0, sh); |
f416885e N |
3176 | compute_parity6(sh, RECONSTRUCT_WRITE); |
3177 | for (i = conf->raid_disks ; i-- ; ) { | |
3178 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3179 | s.locked++; |
f416885e N |
3180 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
3181 | } | |
3182 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 3183 | } else if (s.expanded) { |
f416885e N |
3184 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3185 | atomic_dec(&conf->reshape_stripes); | |
3186 | wake_up(&conf->wait_for_overlap); | |
3187 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3188 | } | |
3189 | ||
0f94e87c | 3190 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3191 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3192 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3193 | |
6bfe0b49 | 3194 | unlock: |
16a53ecc N |
3195 | spin_unlock(&sh->lock); |
3196 | ||
6bfe0b49 DW |
3197 | /* wait for this device to become unblocked */ |
3198 | if (unlikely(blocked_rdev)) | |
3199 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3200 | ||
f0e43bcd | 3201 | ops_run_io(sh, &s); |
16a53ecc | 3202 | |
f0e43bcd | 3203 | return_io(return_bi); |
df10cfbc DW |
3204 | |
3205 | return blocked_rdev == NULL; | |
16a53ecc N |
3206 | } |
3207 | ||
df10cfbc DW |
3208 | /* returns true if the stripe was handled */ |
3209 | static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page) | |
16a53ecc N |
3210 | { |
3211 | if (sh->raid_conf->level == 6) | |
df10cfbc | 3212 | return handle_stripe6(sh, tmp_page); |
16a53ecc | 3213 | else |
df10cfbc | 3214 | return handle_stripe5(sh); |
16a53ecc N |
3215 | } |
3216 | ||
3217 | ||
3218 | ||
3219 | static void raid5_activate_delayed(raid5_conf_t *conf) | |
3220 | { | |
3221 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3222 | while (!list_empty(&conf->delayed_list)) { | |
3223 | struct list_head *l = conf->delayed_list.next; | |
3224 | struct stripe_head *sh; | |
3225 | sh = list_entry(l, struct stripe_head, lru); | |
3226 | list_del_init(l); | |
3227 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3228 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3229 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3230 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3231 | } |
6ed3003c N |
3232 | } else |
3233 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3234 | } |
3235 | ||
3236 | static void activate_bit_delay(raid5_conf_t *conf) | |
3237 | { | |
3238 | /* device_lock is held */ | |
3239 | struct list_head head; | |
3240 | list_add(&head, &conf->bitmap_list); | |
3241 | list_del_init(&conf->bitmap_list); | |
3242 | while (!list_empty(&head)) { | |
3243 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3244 | list_del_init(&sh->lru); | |
3245 | atomic_inc(&sh->count); | |
3246 | __release_stripe(conf, sh); | |
3247 | } | |
3248 | } | |
3249 | ||
3250 | static void unplug_slaves(mddev_t *mddev) | |
3251 | { | |
3252 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3253 | int i; | |
3254 | ||
3255 | rcu_read_lock(); | |
3256 | for (i=0; i<mddev->raid_disks; i++) { | |
3257 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
3258 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3259 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3260 | |
3261 | atomic_inc(&rdev->nr_pending); | |
3262 | rcu_read_unlock(); | |
3263 | ||
2ad8b1ef | 3264 | blk_unplug(r_queue); |
16a53ecc N |
3265 | |
3266 | rdev_dec_pending(rdev, mddev); | |
3267 | rcu_read_lock(); | |
3268 | } | |
3269 | } | |
3270 | rcu_read_unlock(); | |
3271 | } | |
3272 | ||
165125e1 | 3273 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3274 | { |
3275 | mddev_t *mddev = q->queuedata; | |
3276 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3277 | unsigned long flags; | |
3278 | ||
3279 | spin_lock_irqsave(&conf->device_lock, flags); | |
3280 | ||
3281 | if (blk_remove_plug(q)) { | |
3282 | conf->seq_flush++; | |
3283 | raid5_activate_delayed(conf); | |
72626685 | 3284 | } |
1da177e4 LT |
3285 | md_wakeup_thread(mddev->thread); |
3286 | ||
3287 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3288 | ||
3289 | unplug_slaves(mddev); | |
3290 | } | |
3291 | ||
f022b2fd N |
3292 | static int raid5_congested(void *data, int bits) |
3293 | { | |
3294 | mddev_t *mddev = data; | |
3295 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3296 | ||
3297 | /* No difference between reads and writes. Just check | |
3298 | * how busy the stripe_cache is | |
3299 | */ | |
3300 | if (conf->inactive_blocked) | |
3301 | return 1; | |
3302 | if (conf->quiesce) | |
3303 | return 1; | |
3304 | if (list_empty_careful(&conf->inactive_list)) | |
3305 | return 1; | |
3306 | ||
3307 | return 0; | |
3308 | } | |
3309 | ||
23032a0e RBJ |
3310 | /* We want read requests to align with chunks where possible, |
3311 | * but write requests don't need to. | |
3312 | */ | |
cc371e66 AK |
3313 | static int raid5_mergeable_bvec(struct request_queue *q, |
3314 | struct bvec_merge_data *bvm, | |
3315 | struct bio_vec *biovec) | |
23032a0e RBJ |
3316 | { |
3317 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3318 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e RBJ |
3319 | int max; |
3320 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
cc371e66 | 3321 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3322 | |
cc371e66 | 3323 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3324 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3325 | ||
3326 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
3327 | if (max < 0) max = 0; | |
3328 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3329 | return biovec->bv_len; | |
3330 | else | |
3331 | return max; | |
3332 | } | |
3333 | ||
f679623f RBJ |
3334 | |
3335 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3336 | { | |
3337 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
3338 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
3339 | unsigned int bio_sectors = bio->bi_size >> 9; | |
3340 | ||
3341 | return chunk_sectors >= | |
3342 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3343 | } | |
3344 | ||
46031f9a RBJ |
3345 | /* |
3346 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3347 | * later sampled by raid5d. | |
3348 | */ | |
3349 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3350 | { | |
3351 | unsigned long flags; | |
3352 | ||
3353 | spin_lock_irqsave(&conf->device_lock, flags); | |
3354 | ||
3355 | bi->bi_next = conf->retry_read_aligned_list; | |
3356 | conf->retry_read_aligned_list = bi; | |
3357 | ||
3358 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3359 | md_wakeup_thread(conf->mddev->thread); | |
3360 | } | |
3361 | ||
3362 | ||
3363 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3364 | { | |
3365 | struct bio *bi; | |
3366 | ||
3367 | bi = conf->retry_read_aligned; | |
3368 | if (bi) { | |
3369 | conf->retry_read_aligned = NULL; | |
3370 | return bi; | |
3371 | } | |
3372 | bi = conf->retry_read_aligned_list; | |
3373 | if(bi) { | |
387bb173 | 3374 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3375 | bi->bi_next = NULL; |
960e739d JA |
3376 | /* |
3377 | * this sets the active strip count to 1 and the processed | |
3378 | * strip count to zero (upper 8 bits) | |
3379 | */ | |
46031f9a | 3380 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3381 | } |
3382 | ||
3383 | return bi; | |
3384 | } | |
3385 | ||
3386 | ||
f679623f RBJ |
3387 | /* |
3388 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3389 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3390 | * first). | |
3391 | * If the read failed.. | |
3392 | */ | |
6712ecf8 | 3393 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3394 | { |
3395 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3396 | mddev_t *mddev; |
3397 | raid5_conf_t *conf; | |
3398 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3399 | mdk_rdev_t *rdev; | |
3400 | ||
f679623f | 3401 | bio_put(bi); |
46031f9a RBJ |
3402 | |
3403 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
3404 | conf = mddev_to_conf(mddev); | |
3405 | rdev = (void*)raid_bi->bi_next; | |
3406 | raid_bi->bi_next = NULL; | |
3407 | ||
3408 | rdev_dec_pending(rdev, conf->mddev); | |
3409 | ||
3410 | if (!error && uptodate) { | |
6712ecf8 | 3411 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3412 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3413 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3414 | return; |
46031f9a RBJ |
3415 | } |
3416 | ||
3417 | ||
45b4233c | 3418 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3419 | |
3420 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3421 | } |
3422 | ||
387bb173 NB |
3423 | static int bio_fits_rdev(struct bio *bi) |
3424 | { | |
165125e1 | 3425 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 NB |
3426 | |
3427 | if ((bi->bi_size>>9) > q->max_sectors) | |
3428 | return 0; | |
3429 | blk_recount_segments(q, bi); | |
960e739d | 3430 | if (bi->bi_phys_segments > q->max_phys_segments) |
387bb173 NB |
3431 | return 0; |
3432 | ||
3433 | if (q->merge_bvec_fn) | |
3434 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3435 | * just just give up | |
3436 | */ | |
3437 | return 0; | |
3438 | ||
3439 | return 1; | |
3440 | } | |
3441 | ||
3442 | ||
165125e1 | 3443 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3444 | { |
3445 | mddev_t *mddev = q->queuedata; | |
3446 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3447 | unsigned int dd_idx; |
f679623f RBJ |
3448 | struct bio* align_bi; |
3449 | mdk_rdev_t *rdev; | |
3450 | ||
3451 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3452 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3453 | return 0; |
3454 | } | |
3455 | /* | |
99c0fb5f | 3456 | * use bio_clone to make a copy of the bio |
f679623f RBJ |
3457 | */ |
3458 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3459 | if (!align_bi) | |
3460 | return 0; | |
3461 | /* | |
3462 | * set bi_end_io to a new function, and set bi_private to the | |
3463 | * original bio. | |
3464 | */ | |
3465 | align_bi->bi_end_io = raid5_align_endio; | |
3466 | align_bi->bi_private = raid_bio; | |
3467 | /* | |
3468 | * compute position | |
3469 | */ | |
112bf897 N |
3470 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3471 | 0, | |
911d4ee8 | 3472 | &dd_idx, NULL); |
f679623f RBJ |
3473 | |
3474 | rcu_read_lock(); | |
3475 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3476 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3477 | atomic_inc(&rdev->nr_pending); |
3478 | rcu_read_unlock(); | |
46031f9a RBJ |
3479 | raid_bio->bi_next = (void*)rdev; |
3480 | align_bi->bi_bdev = rdev->bdev; | |
3481 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3482 | align_bi->bi_sector += rdev->data_offset; | |
3483 | ||
387bb173 NB |
3484 | if (!bio_fits_rdev(align_bi)) { |
3485 | /* too big in some way */ | |
3486 | bio_put(align_bi); | |
3487 | rdev_dec_pending(rdev, mddev); | |
3488 | return 0; | |
3489 | } | |
3490 | ||
46031f9a RBJ |
3491 | spin_lock_irq(&conf->device_lock); |
3492 | wait_event_lock_irq(conf->wait_for_stripe, | |
3493 | conf->quiesce == 0, | |
3494 | conf->device_lock, /* nothing */); | |
3495 | atomic_inc(&conf->active_aligned_reads); | |
3496 | spin_unlock_irq(&conf->device_lock); | |
3497 | ||
f679623f RBJ |
3498 | generic_make_request(align_bi); |
3499 | return 1; | |
3500 | } else { | |
3501 | rcu_read_unlock(); | |
46031f9a | 3502 | bio_put(align_bi); |
f679623f RBJ |
3503 | return 0; |
3504 | } | |
3505 | } | |
3506 | ||
8b3e6cdc DW |
3507 | /* __get_priority_stripe - get the next stripe to process |
3508 | * | |
3509 | * Full stripe writes are allowed to pass preread active stripes up until | |
3510 | * the bypass_threshold is exceeded. In general the bypass_count | |
3511 | * increments when the handle_list is handled before the hold_list; however, it | |
3512 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3513 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3514 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3515 | * handle_list. | |
3516 | */ | |
3517 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3518 | { | |
3519 | struct stripe_head *sh; | |
3520 | ||
3521 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3522 | __func__, | |
3523 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3524 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3525 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3526 | ||
3527 | if (!list_empty(&conf->handle_list)) { | |
3528 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3529 | ||
3530 | if (list_empty(&conf->hold_list)) | |
3531 | conf->bypass_count = 0; | |
3532 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3533 | if (conf->hold_list.next == conf->last_hold) | |
3534 | conf->bypass_count++; | |
3535 | else { | |
3536 | conf->last_hold = conf->hold_list.next; | |
3537 | conf->bypass_count -= conf->bypass_threshold; | |
3538 | if (conf->bypass_count < 0) | |
3539 | conf->bypass_count = 0; | |
3540 | } | |
3541 | } | |
3542 | } else if (!list_empty(&conf->hold_list) && | |
3543 | ((conf->bypass_threshold && | |
3544 | conf->bypass_count > conf->bypass_threshold) || | |
3545 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3546 | sh = list_entry(conf->hold_list.next, | |
3547 | typeof(*sh), lru); | |
3548 | conf->bypass_count -= conf->bypass_threshold; | |
3549 | if (conf->bypass_count < 0) | |
3550 | conf->bypass_count = 0; | |
3551 | } else | |
3552 | return NULL; | |
3553 | ||
3554 | list_del_init(&sh->lru); | |
3555 | atomic_inc(&sh->count); | |
3556 | BUG_ON(atomic_read(&sh->count) != 1); | |
3557 | return sh; | |
3558 | } | |
f679623f | 3559 | |
165125e1 | 3560 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3561 | { |
3562 | mddev_t *mddev = q->queuedata; | |
3563 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3564 | int dd_idx; |
1da177e4 LT |
3565 | sector_t new_sector; |
3566 | sector_t logical_sector, last_sector; | |
3567 | struct stripe_head *sh; | |
a362357b | 3568 | const int rw = bio_data_dir(bi); |
c9959059 | 3569 | int cpu, remaining; |
1da177e4 | 3570 | |
e5dcdd80 | 3571 | if (unlikely(bio_barrier(bi))) { |
6712ecf8 | 3572 | bio_endio(bi, -EOPNOTSUPP); |
e5dcdd80 N |
3573 | return 0; |
3574 | } | |
3575 | ||
3d310eb7 | 3576 | md_write_start(mddev, bi); |
06d91a5f | 3577 | |
074a7aca TH |
3578 | cpu = part_stat_lock(); |
3579 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
3580 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
3581 | bio_sectors(bi)); | |
3582 | part_stat_unlock(); | |
1da177e4 | 3583 | |
802ba064 | 3584 | if (rw == READ && |
52488615 RBJ |
3585 | mddev->reshape_position == MaxSector && |
3586 | chunk_aligned_read(q,bi)) | |
99c0fb5f | 3587 | return 0; |
52488615 | 3588 | |
1da177e4 LT |
3589 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3590 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3591 | bi->bi_next = NULL; | |
3592 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3593 | |
1da177e4 LT |
3594 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3595 | DEFINE_WAIT(w); | |
16a53ecc | 3596 | int disks, data_disks; |
b5663ba4 | 3597 | int previous; |
b578d55f | 3598 | |
7ecaa1e6 | 3599 | retry: |
b5663ba4 | 3600 | previous = 0; |
b578d55f | 3601 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
7ecaa1e6 N |
3602 | if (likely(conf->expand_progress == MaxSector)) |
3603 | disks = conf->raid_disks; | |
3604 | else { | |
df8e7f76 N |
3605 | /* spinlock is needed as expand_progress may be |
3606 | * 64bit on a 32bit platform, and so it might be | |
3607 | * possible to see a half-updated value | |
3608 | * Ofcourse expand_progress could change after | |
3609 | * the lock is dropped, so once we get a reference | |
3610 | * to the stripe that we think it is, we will have | |
3611 | * to check again. | |
3612 | */ | |
7ecaa1e6 N |
3613 | spin_lock_irq(&conf->device_lock); |
3614 | disks = conf->raid_disks; | |
b5663ba4 | 3615 | if (logical_sector >= conf->expand_progress) { |
7ecaa1e6 | 3616 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3617 | previous = 1; |
3618 | } else { | |
b578d55f N |
3619 | if (logical_sector >= conf->expand_lo) { |
3620 | spin_unlock_irq(&conf->device_lock); | |
3621 | schedule(); | |
3622 | goto retry; | |
3623 | } | |
3624 | } | |
7ecaa1e6 N |
3625 | spin_unlock_irq(&conf->device_lock); |
3626 | } | |
16a53ecc N |
3627 | data_disks = disks - conf->max_degraded; |
3628 | ||
112bf897 N |
3629 | new_sector = raid5_compute_sector(conf, logical_sector, |
3630 | previous, | |
911d4ee8 | 3631 | &dd_idx, NULL); |
45b4233c | 3632 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3633 | (unsigned long long)new_sector, |
3634 | (unsigned long long)logical_sector); | |
3635 | ||
b5663ba4 N |
3636 | sh = get_active_stripe(conf, new_sector, previous, |
3637 | (bi->bi_rw&RWA_MASK)); | |
1da177e4 | 3638 | if (sh) { |
7ecaa1e6 N |
3639 | if (unlikely(conf->expand_progress != MaxSector)) { |
3640 | /* expansion might have moved on while waiting for a | |
df8e7f76 N |
3641 | * stripe, so we must do the range check again. |
3642 | * Expansion could still move past after this | |
3643 | * test, but as we are holding a reference to | |
3644 | * 'sh', we know that if that happens, | |
3645 | * STRIPE_EXPANDING will get set and the expansion | |
3646 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3647 | */ |
3648 | int must_retry = 0; | |
3649 | spin_lock_irq(&conf->device_lock); | |
3650 | if (logical_sector < conf->expand_progress && | |
3651 | disks == conf->previous_raid_disks) | |
3652 | /* mismatch, need to try again */ | |
3653 | must_retry = 1; | |
3654 | spin_unlock_irq(&conf->device_lock); | |
3655 | if (must_retry) { | |
3656 | release_stripe(sh); | |
3657 | goto retry; | |
3658 | } | |
3659 | } | |
e464eafd N |
3660 | /* FIXME what if we get a false positive because these |
3661 | * are being updated. | |
3662 | */ | |
3663 | if (logical_sector >= mddev->suspend_lo && | |
3664 | logical_sector < mddev->suspend_hi) { | |
3665 | release_stripe(sh); | |
3666 | schedule(); | |
3667 | goto retry; | |
3668 | } | |
7ecaa1e6 N |
3669 | |
3670 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
3671 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
3672 | /* Stripe is busy expanding or | |
3673 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3674 | * and wait a while |
3675 | */ | |
3676 | raid5_unplug_device(mddev->queue); | |
3677 | release_stripe(sh); | |
3678 | schedule(); | |
3679 | goto retry; | |
3680 | } | |
3681 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3682 | set_bit(STRIPE_HANDLE, &sh->state); |
3683 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1da177e4 | 3684 | release_stripe(sh); |
1da177e4 LT |
3685 | } else { |
3686 | /* cannot get stripe for read-ahead, just give-up */ | |
3687 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3688 | finish_wait(&conf->wait_for_overlap, &w); | |
3689 | break; | |
3690 | } | |
3691 | ||
3692 | } | |
3693 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3694 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3695 | spin_unlock_irq(&conf->device_lock); |
3696 | if (remaining == 0) { | |
1da177e4 | 3697 | |
16a53ecc | 3698 | if ( rw == WRITE ) |
1da177e4 | 3699 | md_write_end(mddev); |
6712ecf8 | 3700 | |
0e13fe23 | 3701 | bio_endio(bi, 0); |
1da177e4 | 3702 | } |
1da177e4 LT |
3703 | return 0; |
3704 | } | |
3705 | ||
52c03291 | 3706 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3707 | { |
52c03291 N |
3708 | /* reshaping is quite different to recovery/resync so it is |
3709 | * handled quite separately ... here. | |
3710 | * | |
3711 | * On each call to sync_request, we gather one chunk worth of | |
3712 | * destination stripes and flag them as expanding. | |
3713 | * Then we find all the source stripes and request reads. | |
3714 | * As the reads complete, handle_stripe will copy the data | |
3715 | * into the destination stripe and release that stripe. | |
3716 | */ | |
1da177e4 LT |
3717 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
3718 | struct stripe_head *sh; | |
ccfcc3c1 | 3719 | sector_t first_sector, last_sector; |
f416885e N |
3720 | int raid_disks = conf->previous_raid_disks; |
3721 | int data_disks = raid_disks - conf->max_degraded; | |
3722 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3723 | int i; |
3724 | int dd_idx; | |
3725 | sector_t writepos, safepos, gap; | |
3726 | ||
3727 | if (sector_nr == 0 && | |
3728 | conf->expand_progress != 0) { | |
3729 | /* restarting in the middle, skip the initial sectors */ | |
3730 | sector_nr = conf->expand_progress; | |
f416885e | 3731 | sector_div(sector_nr, new_data_disks); |
52c03291 N |
3732 | *skipped = 1; |
3733 | return sector_nr; | |
3734 | } | |
3735 | ||
3736 | /* we update the metadata when there is more than 3Meg | |
3737 | * in the block range (that is rather arbitrary, should | |
3738 | * probably be time based) or when the data about to be | |
3739 | * copied would over-write the source of the data at | |
3740 | * the front of the range. | |
3741 | * i.e. one new_stripe forward from expand_progress new_maps | |
3742 | * to after where expand_lo old_maps to | |
3743 | */ | |
3744 | writepos = conf->expand_progress + | |
f416885e N |
3745 | conf->chunk_size/512*(new_data_disks); |
3746 | sector_div(writepos, new_data_disks); | |
52c03291 | 3747 | safepos = conf->expand_lo; |
f416885e | 3748 | sector_div(safepos, data_disks); |
52c03291 N |
3749 | gap = conf->expand_progress - conf->expand_lo; |
3750 | ||
3751 | if (writepos >= safepos || | |
f416885e | 3752 | gap > (new_data_disks)*3000*2 /*3Meg*/) { |
52c03291 N |
3753 | /* Cannot proceed until we've updated the superblock... */ |
3754 | wait_event(conf->wait_for_overlap, | |
3755 | atomic_read(&conf->reshape_stripes)==0); | |
3756 | mddev->reshape_position = conf->expand_progress; | |
850b2b42 | 3757 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 3758 | md_wakeup_thread(mddev->thread); |
850b2b42 | 3759 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
3760 | kthread_should_stop()); |
3761 | spin_lock_irq(&conf->device_lock); | |
3762 | conf->expand_lo = mddev->reshape_position; | |
3763 | spin_unlock_irq(&conf->device_lock); | |
3764 | wake_up(&conf->wait_for_overlap); | |
3765 | } | |
3766 | ||
3767 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { | |
3768 | int j; | |
3769 | int skipped = 0; | |
b5663ba4 | 3770 | sh = get_active_stripe(conf, sector_nr+i, 0, 0); |
52c03291 N |
3771 | set_bit(STRIPE_EXPANDING, &sh->state); |
3772 | atomic_inc(&conf->reshape_stripes); | |
3773 | /* If any of this stripe is beyond the end of the old | |
3774 | * array, then we need to zero those blocks | |
3775 | */ | |
3776 | for (j=sh->disks; j--;) { | |
3777 | sector_t s; | |
3778 | if (j == sh->pd_idx) | |
3779 | continue; | |
f416885e | 3780 | if (conf->level == 6 && |
d0dabf7e | 3781 | j == sh->qd_idx) |
f416885e | 3782 | continue; |
52c03291 | 3783 | s = compute_blocknr(sh, j); |
f233ea5c | 3784 | if (s < mddev->array_sectors) { |
52c03291 N |
3785 | skipped = 1; |
3786 | continue; | |
3787 | } | |
3788 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
3789 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
3790 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
3791 | } | |
3792 | if (!skipped) { | |
3793 | set_bit(STRIPE_EXPAND_READY, &sh->state); | |
3794 | set_bit(STRIPE_HANDLE, &sh->state); | |
3795 | } | |
3796 | release_stripe(sh); | |
3797 | } | |
3798 | spin_lock_irq(&conf->device_lock); | |
6d3baf2e | 3799 | conf->expand_progress = (sector_nr + i) * new_data_disks; |
52c03291 N |
3800 | spin_unlock_irq(&conf->device_lock); |
3801 | /* Ok, those stripe are ready. We can start scheduling | |
3802 | * reads on the source stripes. | |
3803 | * The source stripes are determined by mapping the first and last | |
3804 | * block on the destination stripes. | |
3805 | */ | |
52c03291 | 3806 | first_sector = |
112bf897 | 3807 | raid5_compute_sector(conf, sector_nr*(new_data_disks), |
911d4ee8 | 3808 | 1, &dd_idx, NULL); |
52c03291 | 3809 | last_sector = |
112bf897 N |
3810 | raid5_compute_sector(conf, ((sector_nr+conf->chunk_size/512) |
3811 | *(new_data_disks) - 1), | |
911d4ee8 | 3812 | 1, &dd_idx, NULL); |
58c0fed4 AN |
3813 | if (last_sector >= mddev->dev_sectors) |
3814 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 3815 | while (first_sector <= last_sector) { |
b5663ba4 | 3816 | sh = get_active_stripe(conf, first_sector, 1, 0); |
52c03291 N |
3817 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3818 | set_bit(STRIPE_HANDLE, &sh->state); | |
3819 | release_stripe(sh); | |
3820 | first_sector += STRIPE_SECTORS; | |
3821 | } | |
c6207277 N |
3822 | /* If this takes us to the resync_max point where we have to pause, |
3823 | * then we need to write out the superblock. | |
3824 | */ | |
3825 | sector_nr += conf->chunk_size>>9; | |
3826 | if (sector_nr >= mddev->resync_max) { | |
3827 | /* Cannot proceed until we've updated the superblock... */ | |
3828 | wait_event(conf->wait_for_overlap, | |
3829 | atomic_read(&conf->reshape_stripes) == 0); | |
3830 | mddev->reshape_position = conf->expand_progress; | |
3831 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3832 | md_wakeup_thread(mddev->thread); | |
3833 | wait_event(mddev->sb_wait, | |
3834 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
3835 | || kthread_should_stop()); | |
3836 | spin_lock_irq(&conf->device_lock); | |
3837 | conf->expand_lo = mddev->reshape_position; | |
3838 | spin_unlock_irq(&conf->device_lock); | |
3839 | wake_up(&conf->wait_for_overlap); | |
3840 | } | |
52c03291 N |
3841 | return conf->chunk_size>>9; |
3842 | } | |
3843 | ||
3844 | /* FIXME go_faster isn't used */ | |
3845 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
3846 | { | |
3847 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3848 | struct stripe_head *sh; | |
58c0fed4 | 3849 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 3850 | int sync_blocks; |
16a53ecc N |
3851 | int still_degraded = 0; |
3852 | int i; | |
1da177e4 | 3853 | |
72626685 | 3854 | if (sector_nr >= max_sector) { |
1da177e4 LT |
3855 | /* just being told to finish up .. nothing much to do */ |
3856 | unplug_slaves(mddev); | |
29269553 N |
3857 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3858 | end_reshape(conf); | |
3859 | return 0; | |
3860 | } | |
72626685 N |
3861 | |
3862 | if (mddev->curr_resync < max_sector) /* aborted */ | |
3863 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3864 | &sync_blocks, 1); | |
16a53ecc | 3865 | else /* completed sync */ |
72626685 N |
3866 | conf->fullsync = 0; |
3867 | bitmap_close_sync(mddev->bitmap); | |
3868 | ||
1da177e4 LT |
3869 | return 0; |
3870 | } | |
ccfcc3c1 | 3871 | |
52c03291 N |
3872 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3873 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 3874 | |
c6207277 N |
3875 | /* No need to check resync_max as we never do more than one |
3876 | * stripe, and as resync_max will always be on a chunk boundary, | |
3877 | * if the check in md_do_sync didn't fire, there is no chance | |
3878 | * of overstepping resync_max here | |
3879 | */ | |
3880 | ||
16a53ecc | 3881 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
3882 | * to resync, then assert that we are finished, because there is |
3883 | * nothing we can do. | |
3884 | */ | |
3285edf1 | 3885 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 3886 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 3887 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 3888 | *skipped = 1; |
1da177e4 LT |
3889 | return rv; |
3890 | } | |
72626685 | 3891 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 3892 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
3893 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
3894 | /* we can skip this block, and probably more */ | |
3895 | sync_blocks /= STRIPE_SECTORS; | |
3896 | *skipped = 1; | |
3897 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
3898 | } | |
1da177e4 | 3899 | |
b47490c9 N |
3900 | |
3901 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
3902 | ||
b5663ba4 | 3903 | sh = get_active_stripe(conf, sector_nr, 0, 1); |
1da177e4 | 3904 | if (sh == NULL) { |
b5663ba4 | 3905 | sh = get_active_stripe(conf, sector_nr, 0, 0); |
1da177e4 | 3906 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 3907 | * is trying to get access |
1da177e4 | 3908 | */ |
66c006a5 | 3909 | schedule_timeout_uninterruptible(1); |
1da177e4 | 3910 | } |
16a53ecc N |
3911 | /* Need to check if array will still be degraded after recovery/resync |
3912 | * We don't need to check the 'failed' flag as when that gets set, | |
3913 | * recovery aborts. | |
3914 | */ | |
3915 | for (i=0; i<mddev->raid_disks; i++) | |
3916 | if (conf->disks[i].rdev == NULL) | |
3917 | still_degraded = 1; | |
3918 | ||
3919 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
3920 | ||
3921 | spin_lock(&sh->lock); | |
1da177e4 LT |
3922 | set_bit(STRIPE_SYNCING, &sh->state); |
3923 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3924 | spin_unlock(&sh->lock); | |
3925 | ||
df10cfbc DW |
3926 | /* wait for any blocked device to be handled */ |
3927 | while(unlikely(!handle_stripe(sh, NULL))) | |
3928 | ; | |
1da177e4 LT |
3929 | release_stripe(sh); |
3930 | ||
3931 | return STRIPE_SECTORS; | |
3932 | } | |
3933 | ||
46031f9a RBJ |
3934 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
3935 | { | |
3936 | /* We may not be able to submit a whole bio at once as there | |
3937 | * may not be enough stripe_heads available. | |
3938 | * We cannot pre-allocate enough stripe_heads as we may need | |
3939 | * more than exist in the cache (if we allow ever large chunks). | |
3940 | * So we do one stripe head at a time and record in | |
3941 | * ->bi_hw_segments how many have been done. | |
3942 | * | |
3943 | * We *know* that this entire raid_bio is in one chunk, so | |
3944 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
3945 | */ | |
3946 | struct stripe_head *sh; | |
911d4ee8 | 3947 | int dd_idx; |
46031f9a RBJ |
3948 | sector_t sector, logical_sector, last_sector; |
3949 | int scnt = 0; | |
3950 | int remaining; | |
3951 | int handled = 0; | |
3952 | ||
3953 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 3954 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 3955 | 0, &dd_idx, NULL); |
46031f9a RBJ |
3956 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
3957 | ||
3958 | for (; logical_sector < last_sector; | |
387bb173 NB |
3959 | logical_sector += STRIPE_SECTORS, |
3960 | sector += STRIPE_SECTORS, | |
3961 | scnt++) { | |
46031f9a | 3962 | |
960e739d | 3963 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
3964 | /* already done this stripe */ |
3965 | continue; | |
3966 | ||
b5663ba4 | 3967 | sh = get_active_stripe(conf, sector, 0, 1); |
46031f9a RBJ |
3968 | |
3969 | if (!sh) { | |
3970 | /* failed to get a stripe - must wait */ | |
960e739d | 3971 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
3972 | conf->retry_read_aligned = raid_bio; |
3973 | return handled; | |
3974 | } | |
3975 | ||
3976 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
3977 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
3978 | release_stripe(sh); | |
960e739d | 3979 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
3980 | conf->retry_read_aligned = raid_bio; |
3981 | return handled; | |
3982 | } | |
3983 | ||
46031f9a RBJ |
3984 | handle_stripe(sh, NULL); |
3985 | release_stripe(sh); | |
3986 | handled++; | |
3987 | } | |
3988 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3989 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 3990 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
3991 | if (remaining == 0) |
3992 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
3993 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3994 | wake_up(&conf->wait_for_stripe); | |
3995 | return handled; | |
3996 | } | |
3997 | ||
3998 | ||
3999 | ||
1da177e4 LT |
4000 | /* |
4001 | * This is our raid5 kernel thread. | |
4002 | * | |
4003 | * We scan the hash table for stripes which can be handled now. | |
4004 | * During the scan, completed stripes are saved for us by the interrupt | |
4005 | * handler, so that they will not have to wait for our next wakeup. | |
4006 | */ | |
6ed3003c | 4007 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4008 | { |
4009 | struct stripe_head *sh; | |
4010 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4011 | int handled; | |
4012 | ||
45b4233c | 4013 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4014 | |
4015 | md_check_recovery(mddev); | |
1da177e4 LT |
4016 | |
4017 | handled = 0; | |
4018 | spin_lock_irq(&conf->device_lock); | |
4019 | while (1) { | |
46031f9a | 4020 | struct bio *bio; |
1da177e4 | 4021 | |
ae3c20cc | 4022 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 4023 | int seq = conf->seq_flush; |
700e432d | 4024 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4025 | bitmap_unplug(mddev->bitmap); |
700e432d | 4026 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4027 | conf->seq_write = seq; |
4028 | activate_bit_delay(conf); | |
4029 | } | |
4030 | ||
46031f9a RBJ |
4031 | while ((bio = remove_bio_from_retry(conf))) { |
4032 | int ok; | |
4033 | spin_unlock_irq(&conf->device_lock); | |
4034 | ok = retry_aligned_read(conf, bio); | |
4035 | spin_lock_irq(&conf->device_lock); | |
4036 | if (!ok) | |
4037 | break; | |
4038 | handled++; | |
4039 | } | |
4040 | ||
8b3e6cdc DW |
4041 | sh = __get_priority_stripe(conf); |
4042 | ||
c9f21aaf | 4043 | if (!sh) |
1da177e4 | 4044 | break; |
1da177e4 LT |
4045 | spin_unlock_irq(&conf->device_lock); |
4046 | ||
4047 | handled++; | |
16a53ecc | 4048 | handle_stripe(sh, conf->spare_page); |
1da177e4 LT |
4049 | release_stripe(sh); |
4050 | ||
4051 | spin_lock_irq(&conf->device_lock); | |
4052 | } | |
45b4233c | 4053 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4054 | |
4055 | spin_unlock_irq(&conf->device_lock); | |
4056 | ||
c9f21aaf | 4057 | async_tx_issue_pending_all(); |
1da177e4 LT |
4058 | unplug_slaves(mddev); |
4059 | ||
45b4233c | 4060 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4061 | } |
4062 | ||
3f294f4f | 4063 | static ssize_t |
007583c9 | 4064 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4065 | { |
007583c9 | 4066 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4067 | if (conf) |
4068 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4069 | else | |
4070 | return 0; | |
3f294f4f N |
4071 | } |
4072 | ||
4073 | static ssize_t | |
007583c9 | 4074 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4075 | { |
007583c9 | 4076 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4ef197d8 | 4077 | unsigned long new; |
b5470dc5 DW |
4078 | int err; |
4079 | ||
3f294f4f N |
4080 | if (len >= PAGE_SIZE) |
4081 | return -EINVAL; | |
96de1e66 N |
4082 | if (!conf) |
4083 | return -ENODEV; | |
3f294f4f | 4084 | |
4ef197d8 | 4085 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4086 | return -EINVAL; |
4087 | if (new <= 16 || new > 32768) | |
4088 | return -EINVAL; | |
4089 | while (new < conf->max_nr_stripes) { | |
4090 | if (drop_one_stripe(conf)) | |
4091 | conf->max_nr_stripes--; | |
4092 | else | |
4093 | break; | |
4094 | } | |
b5470dc5 DW |
4095 | err = md_allow_write(mddev); |
4096 | if (err) | |
4097 | return err; | |
3f294f4f N |
4098 | while (new > conf->max_nr_stripes) { |
4099 | if (grow_one_stripe(conf)) | |
4100 | conf->max_nr_stripes++; | |
4101 | else break; | |
4102 | } | |
4103 | return len; | |
4104 | } | |
007583c9 | 4105 | |
96de1e66 N |
4106 | static struct md_sysfs_entry |
4107 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4108 | raid5_show_stripe_cache_size, | |
4109 | raid5_store_stripe_cache_size); | |
3f294f4f | 4110 | |
8b3e6cdc DW |
4111 | static ssize_t |
4112 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4113 | { | |
4114 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4115 | if (conf) | |
4116 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4117 | else | |
4118 | return 0; | |
4119 | } | |
4120 | ||
4121 | static ssize_t | |
4122 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4123 | { | |
4124 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4ef197d8 | 4125 | unsigned long new; |
8b3e6cdc DW |
4126 | if (len >= PAGE_SIZE) |
4127 | return -EINVAL; | |
4128 | if (!conf) | |
4129 | return -ENODEV; | |
4130 | ||
4ef197d8 | 4131 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4132 | return -EINVAL; |
4ef197d8 | 4133 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4134 | return -EINVAL; |
4135 | conf->bypass_threshold = new; | |
4136 | return len; | |
4137 | } | |
4138 | ||
4139 | static struct md_sysfs_entry | |
4140 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4141 | S_IRUGO | S_IWUSR, | |
4142 | raid5_show_preread_threshold, | |
4143 | raid5_store_preread_threshold); | |
4144 | ||
3f294f4f | 4145 | static ssize_t |
96de1e66 | 4146 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4147 | { |
007583c9 | 4148 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4149 | if (conf) |
4150 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4151 | else | |
4152 | return 0; | |
3f294f4f N |
4153 | } |
4154 | ||
96de1e66 N |
4155 | static struct md_sysfs_entry |
4156 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4157 | |
007583c9 | 4158 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4159 | &raid5_stripecache_size.attr, |
4160 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4161 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4162 | NULL, |
4163 | }; | |
007583c9 N |
4164 | static struct attribute_group raid5_attrs_group = { |
4165 | .name = NULL, | |
4166 | .attrs = raid5_attrs, | |
3f294f4f N |
4167 | }; |
4168 | ||
91adb564 | 4169 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4170 | { |
4171 | raid5_conf_t *conf; | |
4172 | int raid_disk, memory; | |
4173 | mdk_rdev_t *rdev; | |
4174 | struct disk_info *disk; | |
1da177e4 | 4175 | |
91adb564 N |
4176 | if (mddev->new_level != 5 |
4177 | && mddev->new_level != 4 | |
4178 | && mddev->new_level != 6) { | |
16a53ecc | 4179 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4180 | mdname(mddev), mddev->new_level); |
4181 | return ERR_PTR(-EIO); | |
1da177e4 | 4182 | } |
91adb564 N |
4183 | if ((mddev->new_level == 5 |
4184 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4185 | (mddev->new_level == 6 | |
4186 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
99c0fb5f | 4187 | printk(KERN_ERR "raid5: %s: layout %d not supported\n", |
91adb564 N |
4188 | mdname(mddev), mddev->new_layout); |
4189 | return ERR_PTR(-EIO); | |
99c0fb5f | 4190 | } |
91adb564 N |
4191 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
4192 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4193 | mdname(mddev), mddev->raid_disks); | |
4194 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4195 | } |
4196 | ||
91adb564 N |
4197 | if (!mddev->new_chunk || mddev->new_chunk % PAGE_SIZE) { |
4198 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
4199 | mddev->new_chunk, mdname(mddev)); | |
4200 | return ERR_PTR(-EINVAL); | |
f6705578 N |
4201 | } |
4202 | ||
91adb564 N |
4203 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4204 | if (conf == NULL) | |
1da177e4 | 4205 | goto abort; |
91adb564 N |
4206 | |
4207 | conf->raid_disks = mddev->raid_disks; | |
4208 | if (mddev->reshape_position == MaxSector) | |
4209 | conf->previous_raid_disks = mddev->raid_disks; | |
4210 | else | |
f6705578 | 4211 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
f6705578 N |
4212 | |
4213 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
4214 | GFP_KERNEL); |
4215 | if (!conf->disks) | |
4216 | goto abort; | |
9ffae0cf | 4217 | |
1da177e4 LT |
4218 | conf->mddev = mddev; |
4219 | ||
fccddba0 | 4220 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4221 | goto abort; |
1da177e4 | 4222 | |
91adb564 | 4223 | if (mddev->new_level == 6) { |
16a53ecc N |
4224 | conf->spare_page = alloc_page(GFP_KERNEL); |
4225 | if (!conf->spare_page) | |
4226 | goto abort; | |
4227 | } | |
1da177e4 LT |
4228 | spin_lock_init(&conf->device_lock); |
4229 | init_waitqueue_head(&conf->wait_for_stripe); | |
4230 | init_waitqueue_head(&conf->wait_for_overlap); | |
4231 | INIT_LIST_HEAD(&conf->handle_list); | |
8b3e6cdc | 4232 | INIT_LIST_HEAD(&conf->hold_list); |
1da177e4 | 4233 | INIT_LIST_HEAD(&conf->delayed_list); |
72626685 | 4234 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
4235 | INIT_LIST_HEAD(&conf->inactive_list); |
4236 | atomic_set(&conf->active_stripes, 0); | |
4237 | atomic_set(&conf->preread_active_stripes, 0); | |
46031f9a | 4238 | atomic_set(&conf->active_aligned_reads, 0); |
8b3e6cdc | 4239 | conf->bypass_threshold = BYPASS_THRESHOLD; |
1da177e4 | 4240 | |
45b4233c | 4241 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4242 | |
159ec1fc | 4243 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4244 | raid_disk = rdev->raid_disk; |
f6705578 | 4245 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
4246 | || raid_disk < 0) |
4247 | continue; | |
4248 | disk = conf->disks + raid_disk; | |
4249 | ||
4250 | disk->rdev = rdev; | |
4251 | ||
b2d444d7 | 4252 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4253 | char b[BDEVNAME_SIZE]; |
4254 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4255 | " disk %d\n", bdevname(rdev->bdev,b), | |
4256 | raid_disk); | |
8c2e870a NB |
4257 | } else |
4258 | /* Cannot rely on bitmap to complete recovery */ | |
4259 | conf->fullsync = 1; | |
1da177e4 LT |
4260 | } |
4261 | ||
91adb564 N |
4262 | conf->chunk_size = mddev->new_chunk; |
4263 | conf->level = mddev->new_level; | |
16a53ecc N |
4264 | if (conf->level == 6) |
4265 | conf->max_degraded = 2; | |
4266 | else | |
4267 | conf->max_degraded = 1; | |
91adb564 | 4268 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4269 | conf->max_nr_stripes = NR_STRIPES; |
f6705578 | 4270 | conf->expand_progress = mddev->reshape_position; |
1da177e4 | 4271 | |
91adb564 N |
4272 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
4273 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; | |
4274 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
4275 | printk(KERN_ERR | |
4276 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4277 | goto abort; | |
4278 | } else | |
4279 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4280 | memory, mdname(mddev)); | |
1da177e4 | 4281 | |
91adb564 N |
4282 | conf->thread = md_register_thread(raid5d, mddev, "%s_raid5"); |
4283 | if (!conf->thread) { | |
4284 | printk(KERN_ERR | |
4285 | "raid5: couldn't allocate thread for %s\n", | |
4286 | mdname(mddev)); | |
16a53ecc N |
4287 | goto abort; |
4288 | } | |
91adb564 N |
4289 | |
4290 | return conf; | |
4291 | ||
4292 | abort: | |
4293 | if (conf) { | |
4294 | shrink_stripes(conf); | |
4295 | safe_put_page(conf->spare_page); | |
4296 | kfree(conf->disks); | |
4297 | kfree(conf->stripe_hashtbl); | |
4298 | kfree(conf); | |
4299 | return ERR_PTR(-EIO); | |
4300 | } else | |
4301 | return ERR_PTR(-ENOMEM); | |
4302 | } | |
4303 | ||
4304 | static int run(mddev_t *mddev) | |
4305 | { | |
4306 | raid5_conf_t *conf; | |
4307 | int working_disks = 0; | |
4308 | mdk_rdev_t *rdev; | |
4309 | ||
4310 | if (mddev->reshape_position != MaxSector) { | |
4311 | /* Check that we can continue the reshape. | |
4312 | * Currently only disks can change, it must | |
4313 | * increase, and we must be past the point where | |
4314 | * a stripe over-writes itself | |
4315 | */ | |
4316 | sector_t here_new, here_old; | |
4317 | int old_disks; | |
4318 | int max_degraded = (mddev->level == 5 ? 1 : 2); | |
4319 | ||
4320 | if (mddev->new_level != mddev->level || | |
4321 | mddev->new_layout != mddev->layout || | |
4322 | mddev->new_chunk != mddev->chunk_size) { | |
4323 | printk(KERN_ERR "raid5: %s: unsupported reshape " | |
4324 | "required - aborting.\n", | |
4325 | mdname(mddev)); | |
4326 | return -EINVAL; | |
4327 | } | |
4328 | if (mddev->delta_disks <= 0) { | |
4329 | printk(KERN_ERR "raid5: %s: unsupported reshape " | |
4330 | "(reduce disks) required - aborting.\n", | |
4331 | mdname(mddev)); | |
4332 | return -EINVAL; | |
4333 | } | |
4334 | old_disks = mddev->raid_disks - mddev->delta_disks; | |
4335 | /* reshape_position must be on a new-stripe boundary, and one | |
4336 | * further up in new geometry must map after here in old | |
4337 | * geometry. | |
4338 | */ | |
4339 | here_new = mddev->reshape_position; | |
4340 | if (sector_div(here_new, (mddev->chunk_size>>9)* | |
4341 | (mddev->raid_disks - max_degraded))) { | |
4342 | printk(KERN_ERR "raid5: reshape_position not " | |
4343 | "on a stripe boundary\n"); | |
4344 | return -EINVAL; | |
4345 | } | |
4346 | /* here_new is the stripe we will write to */ | |
4347 | here_old = mddev->reshape_position; | |
4348 | sector_div(here_old, (mddev->chunk_size>>9)* | |
4349 | (old_disks-max_degraded)); | |
4350 | /* here_old is the first stripe that we might need to read | |
4351 | * from */ | |
4352 | if (here_new >= here_old) { | |
4353 | /* Reading from the same stripe as writing to - bad */ | |
4354 | printk(KERN_ERR "raid5: reshape_position too early for " | |
4355 | "auto-recovery - aborting.\n"); | |
4356 | return -EINVAL; | |
4357 | } | |
4358 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4359 | /* OK, we should be able to continue; */ | |
4360 | } else { | |
4361 | BUG_ON(mddev->level != mddev->new_level); | |
4362 | BUG_ON(mddev->layout != mddev->new_layout); | |
4363 | BUG_ON(mddev->chunk_size != mddev->new_chunk); | |
4364 | BUG_ON(mddev->delta_disks != 0); | |
1da177e4 | 4365 | } |
91adb564 | 4366 | |
245f46c2 N |
4367 | if (mddev->private == NULL) |
4368 | conf = setup_conf(mddev); | |
4369 | else | |
4370 | conf = mddev->private; | |
4371 | ||
91adb564 N |
4372 | if (IS_ERR(conf)) |
4373 | return PTR_ERR(conf); | |
4374 | ||
4375 | mddev->thread = conf->thread; | |
4376 | conf->thread = NULL; | |
4377 | mddev->private = conf; | |
4378 | ||
4379 | /* | |
4380 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4381 | */ | |
4382 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4383 | if (rdev->raid_disk >= 0 && | |
4384 | test_bit(In_sync, &rdev->flags)) | |
4385 | working_disks++; | |
4386 | ||
4387 | mddev->degraded = conf->raid_disks - working_disks; | |
4388 | ||
16a53ecc | 4389 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
4390 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
4391 | " (%d/%d failed)\n", | |
02c2de8c | 4392 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4393 | goto abort; |
4394 | } | |
4395 | ||
91adb564 N |
4396 | /* device size must be a multiple of chunk size */ |
4397 | mddev->dev_sectors &= ~(mddev->chunk_size / 512 - 1); | |
4398 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4399 | ||
16a53ecc | 4400 | if (mddev->degraded > 0 && |
1da177e4 | 4401 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4402 | if (mddev->ok_start_degraded) |
4403 | printk(KERN_WARNING | |
4404 | "raid5: starting dirty degraded array: %s" | |
4405 | "- data corruption possible.\n", | |
4406 | mdname(mddev)); | |
4407 | else { | |
4408 | printk(KERN_ERR | |
4409 | "raid5: cannot start dirty degraded array for %s\n", | |
4410 | mdname(mddev)); | |
4411 | goto abort; | |
4412 | } | |
1da177e4 LT |
4413 | } |
4414 | ||
1da177e4 LT |
4415 | if (mddev->degraded == 0) |
4416 | printk("raid5: raid level %d set %s active with %d out of %d" | |
4417 | " devices, algorithm %d\n", conf->level, mdname(mddev), | |
4418 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
4419 | conf->algorithm); | |
4420 | else | |
4421 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
4422 | " out of %d devices, algorithm %d\n", conf->level, | |
4423 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
4424 | mddev->raid_disks, conf->algorithm); | |
4425 | ||
4426 | print_raid5_conf(conf); | |
4427 | ||
f6705578 N |
4428 | if (conf->expand_progress != MaxSector) { |
4429 | printk("...ok start reshape thread\n"); | |
b578d55f | 4430 | conf->expand_lo = conf->expand_progress; |
f6705578 N |
4431 | atomic_set(&conf->reshape_stripes, 0); |
4432 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4433 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4434 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4435 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4436 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4437 | "%s_reshape"); | |
f6705578 N |
4438 | } |
4439 | ||
1da177e4 | 4440 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 4441 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
4442 | */ |
4443 | { | |
16a53ecc N |
4444 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
4445 | int stripe = data_disks * | |
8932c2e0 | 4446 | (mddev->chunk_size / PAGE_SIZE); |
1da177e4 LT |
4447 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
4448 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4449 | } | |
4450 | ||
4451 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
4452 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
4453 | printk(KERN_WARNING | |
4454 | "raid5: failed to create sysfs attributes for %s\n", | |
4455 | mdname(mddev)); | |
7a5febe9 | 4456 | |
91adb564 N |
4457 | mddev->queue->queue_lock = &conf->device_lock; |
4458 | ||
7a5febe9 | 4459 | mddev->queue->unplug_fn = raid5_unplug_device; |
f022b2fd | 4460 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 4461 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 4462 | |
58c0fed4 AN |
4463 | mddev->array_sectors = mddev->dev_sectors * |
4464 | (conf->previous_raid_disks - conf->max_degraded); | |
7a5febe9 | 4465 | |
23032a0e RBJ |
4466 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
4467 | ||
1da177e4 LT |
4468 | return 0; |
4469 | abort: | |
e0cf8f04 | 4470 | md_unregister_thread(mddev->thread); |
91adb564 | 4471 | mddev->thread = NULL; |
1da177e4 | 4472 | if (conf) { |
91adb564 | 4473 | shrink_stripes(conf); |
1da177e4 | 4474 | print_raid5_conf(conf); |
16a53ecc | 4475 | safe_put_page(conf->spare_page); |
b55e6bfc | 4476 | kfree(conf->disks); |
fccddba0 | 4477 | kfree(conf->stripe_hashtbl); |
1da177e4 LT |
4478 | kfree(conf); |
4479 | } | |
4480 | mddev->private = NULL; | |
4481 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
4482 | return -EIO; | |
4483 | } | |
4484 | ||
4485 | ||
4486 | ||
3f294f4f | 4487 | static int stop(mddev_t *mddev) |
1da177e4 LT |
4488 | { |
4489 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4490 | ||
4491 | md_unregister_thread(mddev->thread); | |
4492 | mddev->thread = NULL; | |
4493 | shrink_stripes(conf); | |
fccddba0 | 4494 | kfree(conf->stripe_hashtbl); |
041ae52e | 4495 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4496 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 4497 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
b55e6bfc | 4498 | kfree(conf->disks); |
96de1e66 | 4499 | kfree(conf); |
1da177e4 LT |
4500 | mddev->private = NULL; |
4501 | return 0; | |
4502 | } | |
4503 | ||
45b4233c | 4504 | #ifdef DEBUG |
d710e138 | 4505 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
4506 | { |
4507 | int i; | |
4508 | ||
16a53ecc N |
4509 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
4510 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
4511 | seq_printf(seq, "sh %llu, count %d.\n", | |
4512 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
4513 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 4514 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
4515 | seq_printf(seq, "(cache%d: %p %ld) ", |
4516 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 4517 | } |
16a53ecc | 4518 | seq_printf(seq, "\n"); |
1da177e4 LT |
4519 | } |
4520 | ||
d710e138 | 4521 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
4522 | { |
4523 | struct stripe_head *sh; | |
fccddba0 | 4524 | struct hlist_node *hn; |
1da177e4 LT |
4525 | int i; |
4526 | ||
4527 | spin_lock_irq(&conf->device_lock); | |
4528 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 4529 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
4530 | if (sh->raid_conf != conf) |
4531 | continue; | |
16a53ecc | 4532 | print_sh(seq, sh); |
1da177e4 LT |
4533 | } |
4534 | } | |
4535 | spin_unlock_irq(&conf->device_lock); | |
4536 | } | |
4537 | #endif | |
4538 | ||
d710e138 | 4539 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
4540 | { |
4541 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4542 | int i; | |
4543 | ||
4544 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
02c2de8c | 4545 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
4546 | for (i = 0; i < conf->raid_disks; i++) |
4547 | seq_printf (seq, "%s", | |
4548 | conf->disks[i].rdev && | |
b2d444d7 | 4549 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 4550 | seq_printf (seq, "]"); |
45b4233c | 4551 | #ifdef DEBUG |
16a53ecc N |
4552 | seq_printf (seq, "\n"); |
4553 | printall(seq, conf); | |
1da177e4 LT |
4554 | #endif |
4555 | } | |
4556 | ||
4557 | static void print_raid5_conf (raid5_conf_t *conf) | |
4558 | { | |
4559 | int i; | |
4560 | struct disk_info *tmp; | |
4561 | ||
4562 | printk("RAID5 conf printout:\n"); | |
4563 | if (!conf) { | |
4564 | printk("(conf==NULL)\n"); | |
4565 | return; | |
4566 | } | |
02c2de8c N |
4567 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
4568 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
4569 | |
4570 | for (i = 0; i < conf->raid_disks; i++) { | |
4571 | char b[BDEVNAME_SIZE]; | |
4572 | tmp = conf->disks + i; | |
4573 | if (tmp->rdev) | |
4574 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 4575 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
4576 | bdevname(tmp->rdev->bdev,b)); |
4577 | } | |
4578 | } | |
4579 | ||
4580 | static int raid5_spare_active(mddev_t *mddev) | |
4581 | { | |
4582 | int i; | |
4583 | raid5_conf_t *conf = mddev->private; | |
4584 | struct disk_info *tmp; | |
4585 | ||
4586 | for (i = 0; i < conf->raid_disks; i++) { | |
4587 | tmp = conf->disks + i; | |
4588 | if (tmp->rdev | |
b2d444d7 | 4589 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
4590 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
4591 | unsigned long flags; | |
4592 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 4593 | mddev->degraded--; |
c04be0aa | 4594 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
4595 | } |
4596 | } | |
4597 | print_raid5_conf(conf); | |
4598 | return 0; | |
4599 | } | |
4600 | ||
4601 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
4602 | { | |
4603 | raid5_conf_t *conf = mddev->private; | |
4604 | int err = 0; | |
4605 | mdk_rdev_t *rdev; | |
4606 | struct disk_info *p = conf->disks + number; | |
4607 | ||
4608 | print_raid5_conf(conf); | |
4609 | rdev = p->rdev; | |
4610 | if (rdev) { | |
b2d444d7 | 4611 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
4612 | atomic_read(&rdev->nr_pending)) { |
4613 | err = -EBUSY; | |
4614 | goto abort; | |
4615 | } | |
dfc70645 N |
4616 | /* Only remove non-faulty devices if recovery |
4617 | * isn't possible. | |
4618 | */ | |
4619 | if (!test_bit(Faulty, &rdev->flags) && | |
4620 | mddev->degraded <= conf->max_degraded) { | |
4621 | err = -EBUSY; | |
4622 | goto abort; | |
4623 | } | |
1da177e4 | 4624 | p->rdev = NULL; |
fbd568a3 | 4625 | synchronize_rcu(); |
1da177e4 LT |
4626 | if (atomic_read(&rdev->nr_pending)) { |
4627 | /* lost the race, try later */ | |
4628 | err = -EBUSY; | |
4629 | p->rdev = rdev; | |
4630 | } | |
4631 | } | |
4632 | abort: | |
4633 | ||
4634 | print_raid5_conf(conf); | |
4635 | return err; | |
4636 | } | |
4637 | ||
4638 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
4639 | { | |
4640 | raid5_conf_t *conf = mddev->private; | |
199050ea | 4641 | int err = -EEXIST; |
1da177e4 LT |
4642 | int disk; |
4643 | struct disk_info *p; | |
6c2fce2e NB |
4644 | int first = 0; |
4645 | int last = conf->raid_disks - 1; | |
1da177e4 | 4646 | |
16a53ecc | 4647 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 4648 | /* no point adding a device */ |
199050ea | 4649 | return -EINVAL; |
1da177e4 | 4650 | |
6c2fce2e NB |
4651 | if (rdev->raid_disk >= 0) |
4652 | first = last = rdev->raid_disk; | |
1da177e4 LT |
4653 | |
4654 | /* | |
16a53ecc N |
4655 | * find the disk ... but prefer rdev->saved_raid_disk |
4656 | * if possible. | |
1da177e4 | 4657 | */ |
16a53ecc | 4658 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 4659 | rdev->saved_raid_disk >= first && |
16a53ecc N |
4660 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
4661 | disk = rdev->saved_raid_disk; | |
4662 | else | |
6c2fce2e NB |
4663 | disk = first; |
4664 | for ( ; disk <= last ; disk++) | |
1da177e4 | 4665 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 4666 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4667 | rdev->raid_disk = disk; |
199050ea | 4668 | err = 0; |
72626685 N |
4669 | if (rdev->saved_raid_disk != disk) |
4670 | conf->fullsync = 1; | |
d6065f7b | 4671 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
4672 | break; |
4673 | } | |
4674 | print_raid5_conf(conf); | |
199050ea | 4675 | return err; |
1da177e4 LT |
4676 | } |
4677 | ||
4678 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
4679 | { | |
4680 | /* no resync is happening, and there is enough space | |
4681 | * on all devices, so we can resize. | |
4682 | * We need to make sure resync covers any new space. | |
4683 | * If the array is shrinking we should possibly wait until | |
4684 | * any io in the removed space completes, but it hardly seems | |
4685 | * worth it. | |
4686 | */ | |
16a53ecc N |
4687 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4688 | ||
1da177e4 | 4689 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); |
f233ea5c AN |
4690 | mddev->array_sectors = sectors * (mddev->raid_disks |
4691 | - conf->max_degraded); | |
4692 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
44ce6294 | 4693 | mddev->changed = 1; |
58c0fed4 AN |
4694 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
4695 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
4696 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
4697 | } | |
58c0fed4 | 4698 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 4699 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
4700 | return 0; |
4701 | } | |
4702 | ||
29269553 | 4703 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f | 4704 | static int raid5_check_reshape(mddev_t *mddev) |
29269553 N |
4705 | { |
4706 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4707 | int err; | |
29269553 | 4708 | |
63c70c4f N |
4709 | if (mddev->delta_disks < 0 || |
4710 | mddev->new_level != mddev->level) | |
4711 | return -EINVAL; /* Cannot shrink array or change level yet */ | |
4712 | if (mddev->delta_disks == 0) | |
29269553 | 4713 | return 0; /* nothing to do */ |
dba034ee N |
4714 | if (mddev->bitmap) |
4715 | /* Cannot grow a bitmap yet */ | |
4716 | return -EBUSY; | |
29269553 N |
4717 | |
4718 | /* Can only proceed if there are plenty of stripe_heads. | |
4719 | * We need a minimum of one full stripe,, and for sensible progress | |
4720 | * it is best to have about 4 times that. | |
4721 | * If we require 4 times, then the default 256 4K stripe_heads will | |
4722 | * allow for chunk sizes up to 256K, which is probably OK. | |
4723 | * If the chunk size is greater, user-space should request more | |
4724 | * stripe_heads first. | |
4725 | */ | |
63c70c4f N |
4726 | if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes || |
4727 | (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) { | |
29269553 N |
4728 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", |
4729 | (mddev->chunk_size / STRIPE_SIZE)*4); | |
4730 | return -ENOSPC; | |
4731 | } | |
4732 | ||
63c70c4f N |
4733 | err = resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
4734 | if (err) | |
4735 | return err; | |
4736 | ||
b4c4c7b8 N |
4737 | if (mddev->degraded > conf->max_degraded) |
4738 | return -EINVAL; | |
63c70c4f N |
4739 | /* looks like we might be able to manage this */ |
4740 | return 0; | |
4741 | } | |
4742 | ||
4743 | static int raid5_start_reshape(mddev_t *mddev) | |
4744 | { | |
4745 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4746 | mdk_rdev_t *rdev; | |
63c70c4f N |
4747 | int spares = 0; |
4748 | int added_devices = 0; | |
c04be0aa | 4749 | unsigned long flags; |
63c70c4f | 4750 | |
f416885e | 4751 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
4752 | return -EBUSY; |
4753 | ||
159ec1fc | 4754 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4755 | if (rdev->raid_disk < 0 && |
4756 | !test_bit(Faulty, &rdev->flags)) | |
4757 | spares++; | |
63c70c4f | 4758 | |
f416885e | 4759 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
4760 | /* Not enough devices even to make a degraded array |
4761 | * of that size | |
4762 | */ | |
4763 | return -EINVAL; | |
4764 | ||
f6705578 | 4765 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
4766 | spin_lock_irq(&conf->device_lock); |
4767 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 4768 | conf->raid_disks += mddev->delta_disks; |
29269553 | 4769 | conf->expand_progress = 0; |
b578d55f | 4770 | conf->expand_lo = 0; |
29269553 N |
4771 | spin_unlock_irq(&conf->device_lock); |
4772 | ||
4773 | /* Add some new drives, as many as will fit. | |
4774 | * We know there are enough to make the newly sized array work. | |
4775 | */ | |
159ec1fc | 4776 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4777 | if (rdev->raid_disk < 0 && |
4778 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 4779 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 N |
4780 | char nm[20]; |
4781 | set_bit(In_sync, &rdev->flags); | |
29269553 | 4782 | added_devices++; |
5fd6c1dc | 4783 | rdev->recovery_offset = 0; |
29269553 | 4784 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
4785 | if (sysfs_create_link(&mddev->kobj, |
4786 | &rdev->kobj, nm)) | |
4787 | printk(KERN_WARNING | |
4788 | "raid5: failed to create " | |
4789 | " link %s for %s\n", | |
4790 | nm, mdname(mddev)); | |
29269553 N |
4791 | } else |
4792 | break; | |
4793 | } | |
4794 | ||
c04be0aa | 4795 | spin_lock_irqsave(&conf->device_lock, flags); |
63c70c4f | 4796 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) - added_devices; |
c04be0aa | 4797 | spin_unlock_irqrestore(&conf->device_lock, flags); |
63c70c4f | 4798 | mddev->raid_disks = conf->raid_disks; |
f6705578 | 4799 | mddev->reshape_position = 0; |
850b2b42 | 4800 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 4801 | |
29269553 N |
4802 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
4803 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4804 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4805 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4806 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4807 | "%s_reshape"); | |
4808 | if (!mddev->sync_thread) { | |
4809 | mddev->recovery = 0; | |
4810 | spin_lock_irq(&conf->device_lock); | |
4811 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
4812 | conf->expand_progress = MaxSector; | |
4813 | spin_unlock_irq(&conf->device_lock); | |
4814 | return -EAGAIN; | |
4815 | } | |
4816 | md_wakeup_thread(mddev->sync_thread); | |
4817 | md_new_event(mddev); | |
4818 | return 0; | |
4819 | } | |
4820 | #endif | |
4821 | ||
4822 | static void end_reshape(raid5_conf_t *conf) | |
4823 | { | |
4824 | struct block_device *bdev; | |
4825 | ||
f6705578 | 4826 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
58c0fed4 | 4827 | conf->mddev->array_sectors = conf->mddev->dev_sectors * |
f416885e | 4828 | (conf->raid_disks - conf->max_degraded); |
f233ea5c | 4829 | set_capacity(conf->mddev->gendisk, conf->mddev->array_sectors); |
44ce6294 | 4830 | conf->mddev->changed = 1; |
f6705578 N |
4831 | |
4832 | bdev = bdget_disk(conf->mddev->gendisk, 0); | |
4833 | if (bdev) { | |
4834 | mutex_lock(&bdev->bd_inode->i_mutex); | |
f233ea5c AN |
4835 | i_size_write(bdev->bd_inode, |
4836 | (loff_t)conf->mddev->array_sectors << 9); | |
f6705578 N |
4837 | mutex_unlock(&bdev->bd_inode->i_mutex); |
4838 | bdput(bdev); | |
4839 | } | |
4840 | spin_lock_irq(&conf->device_lock); | |
4841 | conf->expand_progress = MaxSector; | |
4842 | spin_unlock_irq(&conf->device_lock); | |
4843 | conf->mddev->reshape_position = MaxSector; | |
16a53ecc N |
4844 | |
4845 | /* read-ahead size must cover two whole stripes, which is | |
4846 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4847 | */ | |
4848 | { | |
4849 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
4850 | int stripe = data_disks * | |
4851 | (conf->mddev->chunk_size / PAGE_SIZE); | |
4852 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4853 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4854 | } | |
29269553 | 4855 | } |
29269553 N |
4856 | } |
4857 | ||
72626685 N |
4858 | static void raid5_quiesce(mddev_t *mddev, int state) |
4859 | { | |
4860 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4861 | ||
4862 | switch(state) { | |
e464eafd N |
4863 | case 2: /* resume for a suspend */ |
4864 | wake_up(&conf->wait_for_overlap); | |
4865 | break; | |
4866 | ||
72626685 N |
4867 | case 1: /* stop all writes */ |
4868 | spin_lock_irq(&conf->device_lock); | |
4869 | conf->quiesce = 1; | |
4870 | wait_event_lock_irq(conf->wait_for_stripe, | |
46031f9a RBJ |
4871 | atomic_read(&conf->active_stripes) == 0 && |
4872 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 N |
4873 | conf->device_lock, /* nothing */); |
4874 | spin_unlock_irq(&conf->device_lock); | |
4875 | break; | |
4876 | ||
4877 | case 0: /* re-enable writes */ | |
4878 | spin_lock_irq(&conf->device_lock); | |
4879 | conf->quiesce = 0; | |
4880 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 4881 | wake_up(&conf->wait_for_overlap); |
72626685 N |
4882 | spin_unlock_irq(&conf->device_lock); |
4883 | break; | |
4884 | } | |
72626685 | 4885 | } |
b15c2e57 | 4886 | |
245f46c2 N |
4887 | static struct mdk_personality raid5_personality; |
4888 | ||
4889 | static void *raid6_takeover(mddev_t *mddev) | |
4890 | { | |
4891 | /* Currently can only take over a raid5. We map the | |
4892 | * personality to an equivalent raid6 personality | |
4893 | * with the Q block at the end. | |
4894 | */ | |
4895 | int new_layout; | |
4896 | ||
4897 | if (mddev->pers != &raid5_personality) | |
4898 | return ERR_PTR(-EINVAL); | |
4899 | if (mddev->degraded > 1) | |
4900 | return ERR_PTR(-EINVAL); | |
4901 | if (mddev->raid_disks > 253) | |
4902 | return ERR_PTR(-EINVAL); | |
4903 | if (mddev->raid_disks < 3) | |
4904 | return ERR_PTR(-EINVAL); | |
4905 | ||
4906 | switch (mddev->layout) { | |
4907 | case ALGORITHM_LEFT_ASYMMETRIC: | |
4908 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
4909 | break; | |
4910 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
4911 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
4912 | break; | |
4913 | case ALGORITHM_LEFT_SYMMETRIC: | |
4914 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
4915 | break; | |
4916 | case ALGORITHM_RIGHT_SYMMETRIC: | |
4917 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
4918 | break; | |
4919 | case ALGORITHM_PARITY_0: | |
4920 | new_layout = ALGORITHM_PARITY_0_6; | |
4921 | break; | |
4922 | case ALGORITHM_PARITY_N: | |
4923 | new_layout = ALGORITHM_PARITY_N; | |
4924 | break; | |
4925 | default: | |
4926 | return ERR_PTR(-EINVAL); | |
4927 | } | |
4928 | mddev->new_level = 6; | |
4929 | mddev->new_layout = new_layout; | |
4930 | mddev->delta_disks = 1; | |
4931 | mddev->raid_disks += 1; | |
4932 | return setup_conf(mddev); | |
4933 | } | |
4934 | ||
4935 | ||
16a53ecc N |
4936 | static struct mdk_personality raid6_personality = |
4937 | { | |
4938 | .name = "raid6", | |
4939 | .level = 6, | |
4940 | .owner = THIS_MODULE, | |
4941 | .make_request = make_request, | |
4942 | .run = run, | |
4943 | .stop = stop, | |
4944 | .status = status, | |
4945 | .error_handler = error, | |
4946 | .hot_add_disk = raid5_add_disk, | |
4947 | .hot_remove_disk= raid5_remove_disk, | |
4948 | .spare_active = raid5_spare_active, | |
4949 | .sync_request = sync_request, | |
4950 | .resize = raid5_resize, | |
f416885e N |
4951 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4952 | .check_reshape = raid5_check_reshape, | |
4953 | .start_reshape = raid5_start_reshape, | |
4954 | #endif | |
16a53ecc | 4955 | .quiesce = raid5_quiesce, |
245f46c2 | 4956 | .takeover = raid6_takeover, |
16a53ecc | 4957 | }; |
2604b703 | 4958 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
4959 | { |
4960 | .name = "raid5", | |
2604b703 | 4961 | .level = 5, |
1da177e4 LT |
4962 | .owner = THIS_MODULE, |
4963 | .make_request = make_request, | |
4964 | .run = run, | |
4965 | .stop = stop, | |
4966 | .status = status, | |
4967 | .error_handler = error, | |
4968 | .hot_add_disk = raid5_add_disk, | |
4969 | .hot_remove_disk= raid5_remove_disk, | |
4970 | .spare_active = raid5_spare_active, | |
4971 | .sync_request = sync_request, | |
4972 | .resize = raid5_resize, | |
29269553 | 4973 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f N |
4974 | .check_reshape = raid5_check_reshape, |
4975 | .start_reshape = raid5_start_reshape, | |
29269553 | 4976 | #endif |
72626685 | 4977 | .quiesce = raid5_quiesce, |
1da177e4 LT |
4978 | }; |
4979 | ||
2604b703 | 4980 | static struct mdk_personality raid4_personality = |
1da177e4 | 4981 | { |
2604b703 N |
4982 | .name = "raid4", |
4983 | .level = 4, | |
4984 | .owner = THIS_MODULE, | |
4985 | .make_request = make_request, | |
4986 | .run = run, | |
4987 | .stop = stop, | |
4988 | .status = status, | |
4989 | .error_handler = error, | |
4990 | .hot_add_disk = raid5_add_disk, | |
4991 | .hot_remove_disk= raid5_remove_disk, | |
4992 | .spare_active = raid5_spare_active, | |
4993 | .sync_request = sync_request, | |
4994 | .resize = raid5_resize, | |
3d37890b N |
4995 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4996 | .check_reshape = raid5_check_reshape, | |
4997 | .start_reshape = raid5_start_reshape, | |
4998 | #endif | |
2604b703 N |
4999 | .quiesce = raid5_quiesce, |
5000 | }; | |
5001 | ||
5002 | static int __init raid5_init(void) | |
5003 | { | |
16a53ecc N |
5004 | int e; |
5005 | ||
5006 | e = raid6_select_algo(); | |
5007 | if ( e ) | |
5008 | return e; | |
5009 | register_md_personality(&raid6_personality); | |
2604b703 N |
5010 | register_md_personality(&raid5_personality); |
5011 | register_md_personality(&raid4_personality); | |
5012 | return 0; | |
1da177e4 LT |
5013 | } |
5014 | ||
2604b703 | 5015 | static void raid5_exit(void) |
1da177e4 | 5016 | { |
16a53ecc | 5017 | unregister_md_personality(&raid6_personality); |
2604b703 N |
5018 | unregister_md_personality(&raid5_personality); |
5019 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
5020 | } |
5021 | ||
5022 | module_init(raid5_init); | |
5023 | module_exit(raid5_exit); | |
5024 | MODULE_LICENSE("GPL"); | |
5025 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
5026 | MODULE_ALIAS("md-raid5"); |
5027 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
5028 | MODULE_ALIAS("md-level-5"); |
5029 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
5030 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
5031 | MODULE_ALIAS("md-raid6"); | |
5032 | MODULE_ALIAS("md-level-6"); | |
5033 | ||
5034 | /* This used to be two separate modules, they were: */ | |
5035 | MODULE_ALIAS("raid5"); | |
5036 | MODULE_ALIAS("raid6"); |