]>
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 | ||
42b9bebe N |
936 | sprintf(conf->cache_name[0], "raid5-%s", mdname(conf->mddev)); |
937 | sprintf(conf->cache_name[1], "raid5-%s-alt", mdname(conf->mddev)); | |
ad01c9e3 N |
938 | conf->active_name = 0; |
939 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 940 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 941 | 0, 0, NULL); |
1da177e4 LT |
942 | if (!sc) |
943 | return 1; | |
944 | conf->slab_cache = sc; | |
ad01c9e3 | 945 | conf->pool_size = devs; |
16a53ecc | 946 | while (num--) |
3f294f4f | 947 | if (!grow_one_stripe(conf)) |
1da177e4 | 948 | return 1; |
1da177e4 LT |
949 | return 0; |
950 | } | |
29269553 N |
951 | |
952 | #ifdef CONFIG_MD_RAID5_RESHAPE | |
ad01c9e3 N |
953 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
954 | { | |
955 | /* Make all the stripes able to hold 'newsize' devices. | |
956 | * New slots in each stripe get 'page' set to a new page. | |
957 | * | |
958 | * This happens in stages: | |
959 | * 1/ create a new kmem_cache and allocate the required number of | |
960 | * stripe_heads. | |
961 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
962 | * to the new stripe_heads. This will have the side effect of | |
963 | * freezing the array as once all stripe_heads have been collected, | |
964 | * no IO will be possible. Old stripe heads are freed once their | |
965 | * pages have been transferred over, and the old kmem_cache is | |
966 | * freed when all stripes are done. | |
967 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
968 | * we simple return a failre status - no need to clean anything up. | |
969 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
970 | * If this fails, we don't bother trying the shrink the | |
971 | * stripe_heads down again, we just leave them as they are. | |
972 | * As each stripe_head is processed the new one is released into | |
973 | * active service. | |
974 | * | |
975 | * Once step2 is started, we cannot afford to wait for a write, | |
976 | * so we use GFP_NOIO allocations. | |
977 | */ | |
978 | struct stripe_head *osh, *nsh; | |
979 | LIST_HEAD(newstripes); | |
980 | struct disk_info *ndisks; | |
b5470dc5 | 981 | int err; |
e18b890b | 982 | struct kmem_cache *sc; |
ad01c9e3 N |
983 | int i; |
984 | ||
985 | if (newsize <= conf->pool_size) | |
986 | return 0; /* never bother to shrink */ | |
987 | ||
b5470dc5 DW |
988 | err = md_allow_write(conf->mddev); |
989 | if (err) | |
990 | return err; | |
2a2275d6 | 991 | |
ad01c9e3 N |
992 | /* Step 1 */ |
993 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
994 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 995 | 0, 0, NULL); |
ad01c9e3 N |
996 | if (!sc) |
997 | return -ENOMEM; | |
998 | ||
999 | for (i = conf->max_nr_stripes; i; i--) { | |
1000 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
1001 | if (!nsh) | |
1002 | break; | |
1003 | ||
1004 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
1005 | ||
1006 | nsh->raid_conf = conf; | |
1007 | spin_lock_init(&nsh->lock); | |
1008 | ||
1009 | list_add(&nsh->lru, &newstripes); | |
1010 | } | |
1011 | if (i) { | |
1012 | /* didn't get enough, give up */ | |
1013 | while (!list_empty(&newstripes)) { | |
1014 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1015 | list_del(&nsh->lru); | |
1016 | kmem_cache_free(sc, nsh); | |
1017 | } | |
1018 | kmem_cache_destroy(sc); | |
1019 | return -ENOMEM; | |
1020 | } | |
1021 | /* Step 2 - Must use GFP_NOIO now. | |
1022 | * OK, we have enough stripes, start collecting inactive | |
1023 | * stripes and copying them over | |
1024 | */ | |
1025 | list_for_each_entry(nsh, &newstripes, lru) { | |
1026 | spin_lock_irq(&conf->device_lock); | |
1027 | wait_event_lock_irq(conf->wait_for_stripe, | |
1028 | !list_empty(&conf->inactive_list), | |
1029 | conf->device_lock, | |
b3b46be3 | 1030 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
1031 | ); |
1032 | osh = get_free_stripe(conf); | |
1033 | spin_unlock_irq(&conf->device_lock); | |
1034 | atomic_set(&nsh->count, 1); | |
1035 | for(i=0; i<conf->pool_size; i++) | |
1036 | nsh->dev[i].page = osh->dev[i].page; | |
1037 | for( ; i<newsize; i++) | |
1038 | nsh->dev[i].page = NULL; | |
1039 | kmem_cache_free(conf->slab_cache, osh); | |
1040 | } | |
1041 | kmem_cache_destroy(conf->slab_cache); | |
1042 | ||
1043 | /* Step 3. | |
1044 | * At this point, we are holding all the stripes so the array | |
1045 | * is completely stalled, so now is a good time to resize | |
1046 | * conf->disks. | |
1047 | */ | |
1048 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1049 | if (ndisks) { | |
1050 | for (i=0; i<conf->raid_disks; i++) | |
1051 | ndisks[i] = conf->disks[i]; | |
1052 | kfree(conf->disks); | |
1053 | conf->disks = ndisks; | |
1054 | } else | |
1055 | err = -ENOMEM; | |
1056 | ||
1057 | /* Step 4, return new stripes to service */ | |
1058 | while(!list_empty(&newstripes)) { | |
1059 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1060 | list_del_init(&nsh->lru); | |
1061 | for (i=conf->raid_disks; i < newsize; i++) | |
1062 | if (nsh->dev[i].page == NULL) { | |
1063 | struct page *p = alloc_page(GFP_NOIO); | |
1064 | nsh->dev[i].page = p; | |
1065 | if (!p) | |
1066 | err = -ENOMEM; | |
1067 | } | |
1068 | release_stripe(nsh); | |
1069 | } | |
1070 | /* critical section pass, GFP_NOIO no longer needed */ | |
1071 | ||
1072 | conf->slab_cache = sc; | |
1073 | conf->active_name = 1-conf->active_name; | |
1074 | conf->pool_size = newsize; | |
1075 | return err; | |
1076 | } | |
29269553 | 1077 | #endif |
1da177e4 | 1078 | |
3f294f4f | 1079 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
1080 | { |
1081 | struct stripe_head *sh; | |
1082 | ||
3f294f4f N |
1083 | spin_lock_irq(&conf->device_lock); |
1084 | sh = get_free_stripe(conf); | |
1085 | spin_unlock_irq(&conf->device_lock); | |
1086 | if (!sh) | |
1087 | return 0; | |
78bafebd | 1088 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 1089 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
1090 | kmem_cache_free(conf->slab_cache, sh); |
1091 | atomic_dec(&conf->active_stripes); | |
1092 | return 1; | |
1093 | } | |
1094 | ||
1095 | static void shrink_stripes(raid5_conf_t *conf) | |
1096 | { | |
1097 | while (drop_one_stripe(conf)) | |
1098 | ; | |
1099 | ||
29fc7e3e N |
1100 | if (conf->slab_cache) |
1101 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1102 | conf->slab_cache = NULL; |
1103 | } | |
1104 | ||
6712ecf8 | 1105 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1106 | { |
99c0fb5f | 1107 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1108 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1109 | int disks = sh->disks, i; |
1da177e4 | 1110 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
1111 | char b[BDEVNAME_SIZE]; |
1112 | mdk_rdev_t *rdev; | |
1da177e4 | 1113 | |
1da177e4 LT |
1114 | |
1115 | for (i=0 ; i<disks; i++) | |
1116 | if (bi == &sh->dev[i].req) | |
1117 | break; | |
1118 | ||
45b4233c DW |
1119 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
1120 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
1121 | uptodate); |
1122 | if (i == disks) { | |
1123 | BUG(); | |
6712ecf8 | 1124 | return; |
1da177e4 LT |
1125 | } |
1126 | ||
1127 | if (uptodate) { | |
1da177e4 | 1128 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 1129 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 | 1130 | rdev = conf->disks[i].rdev; |
6be9d494 BS |
1131 | printk_rl(KERN_INFO "raid5:%s: read error corrected" |
1132 | " (%lu sectors at %llu on %s)\n", | |
1133 | mdname(conf->mddev), STRIPE_SECTORS, | |
1134 | (unsigned long long)(sh->sector | |
1135 | + rdev->data_offset), | |
1136 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
1137 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1138 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
1139 | } | |
ba22dcbf N |
1140 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
1141 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 1142 | } else { |
d6950432 | 1143 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 1144 | int retry = 0; |
d6950432 N |
1145 | rdev = conf->disks[i].rdev; |
1146 | ||
1da177e4 | 1147 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 1148 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 1149 | if (conf->mddev->degraded) |
6be9d494 BS |
1150 | printk_rl(KERN_WARNING |
1151 | "raid5:%s: read error not correctable " | |
1152 | "(sector %llu on %s).\n", | |
1153 | mdname(conf->mddev), | |
1154 | (unsigned long long)(sh->sector | |
1155 | + rdev->data_offset), | |
1156 | bdn); | |
ba22dcbf | 1157 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 1158 | /* Oh, no!!! */ |
6be9d494 BS |
1159 | printk_rl(KERN_WARNING |
1160 | "raid5:%s: read error NOT corrected!! " | |
1161 | "(sector %llu on %s).\n", | |
1162 | mdname(conf->mddev), | |
1163 | (unsigned long long)(sh->sector | |
1164 | + rdev->data_offset), | |
1165 | bdn); | |
d6950432 | 1166 | else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 1167 | > conf->max_nr_stripes) |
14f8d26b | 1168 | printk(KERN_WARNING |
d6950432 N |
1169 | "raid5:%s: Too many read errors, failing device %s.\n", |
1170 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
1171 | else |
1172 | retry = 1; | |
1173 | if (retry) | |
1174 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
1175 | else { | |
4e5314b5 N |
1176 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
1177 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 1178 | md_error(conf->mddev, rdev); |
ba22dcbf | 1179 | } |
1da177e4 LT |
1180 | } |
1181 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
1182 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1183 | set_bit(STRIPE_HANDLE, &sh->state); | |
1184 | release_stripe(sh); | |
1da177e4 LT |
1185 | } |
1186 | ||
d710e138 | 1187 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 1188 | { |
99c0fb5f | 1189 | struct stripe_head *sh = bi->bi_private; |
1da177e4 | 1190 | raid5_conf_t *conf = sh->raid_conf; |
7ecaa1e6 | 1191 | int disks = sh->disks, i; |
1da177e4 LT |
1192 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
1193 | ||
1da177e4 LT |
1194 | for (i=0 ; i<disks; i++) |
1195 | if (bi == &sh->dev[i].req) | |
1196 | break; | |
1197 | ||
45b4233c | 1198 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
1199 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
1200 | uptodate); | |
1201 | if (i == disks) { | |
1202 | BUG(); | |
6712ecf8 | 1203 | return; |
1da177e4 LT |
1204 | } |
1205 | ||
1da177e4 LT |
1206 | if (!uptodate) |
1207 | md_error(conf->mddev, conf->disks[i].rdev); | |
1208 | ||
1209 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1210 | ||
1211 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1212 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 1213 | release_stripe(sh); |
1da177e4 LT |
1214 | } |
1215 | ||
1216 | ||
1217 | static sector_t compute_blocknr(struct stripe_head *sh, int i); | |
1218 | ||
d710e138 | 1219 | static void raid5_build_block(struct stripe_head *sh, int i) |
1da177e4 LT |
1220 | { |
1221 | struct r5dev *dev = &sh->dev[i]; | |
1222 | ||
1223 | bio_init(&dev->req); | |
1224 | dev->req.bi_io_vec = &dev->vec; | |
1225 | dev->req.bi_vcnt++; | |
1226 | dev->req.bi_max_vecs++; | |
1227 | dev->vec.bv_page = dev->page; | |
1228 | dev->vec.bv_len = STRIPE_SIZE; | |
1229 | dev->vec.bv_offset = 0; | |
1230 | ||
1231 | dev->req.bi_sector = sh->sector; | |
1232 | dev->req.bi_private = sh; | |
1233 | ||
1234 | dev->flags = 0; | |
16a53ecc | 1235 | dev->sector = compute_blocknr(sh, i); |
1da177e4 LT |
1236 | } |
1237 | ||
1238 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1239 | { | |
1240 | char b[BDEVNAME_SIZE]; | |
1241 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
45b4233c | 1242 | pr_debug("raid5: error called\n"); |
1da177e4 | 1243 | |
b2d444d7 | 1244 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 1245 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
1246 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1247 | unsigned long flags; | |
1248 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1249 | mddev->degraded++; |
c04be0aa | 1250 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1251 | /* |
1252 | * if recovery was running, make sure it aborts. | |
1253 | */ | |
dfc70645 | 1254 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1255 | } |
b2d444d7 | 1256 | set_bit(Faulty, &rdev->flags); |
d710e138 N |
1257 | printk(KERN_ALERT |
1258 | "raid5: Disk failure on %s, disabling device.\n" | |
1259 | "raid5: Operation continuing on %d devices.\n", | |
1260 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); | |
1da177e4 | 1261 | } |
16a53ecc | 1262 | } |
1da177e4 LT |
1263 | |
1264 | /* | |
1265 | * Input: a 'big' sector number, | |
1266 | * Output: index of the data and parity disk, and the sector # in them. | |
1267 | */ | |
112bf897 | 1268 | static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector, |
911d4ee8 N |
1269 | int previous, int *dd_idx, |
1270 | struct stripe_head *sh) | |
1da177e4 LT |
1271 | { |
1272 | long stripe; | |
1273 | unsigned long chunk_number; | |
1274 | unsigned int chunk_offset; | |
911d4ee8 | 1275 | int pd_idx, qd_idx; |
67cc2b81 | 1276 | int ddf_layout = 0; |
1da177e4 LT |
1277 | sector_t new_sector; |
1278 | int sectors_per_chunk = conf->chunk_size >> 9; | |
112bf897 N |
1279 | int raid_disks = previous ? conf->previous_raid_disks |
1280 | : conf->raid_disks; | |
1281 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1282 | |
1283 | /* First compute the information on this sector */ | |
1284 | ||
1285 | /* | |
1286 | * Compute the chunk number and the sector offset inside the chunk | |
1287 | */ | |
1288 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
1289 | chunk_number = r_sector; | |
1290 | BUG_ON(r_sector != chunk_number); | |
1291 | ||
1292 | /* | |
1293 | * Compute the stripe number | |
1294 | */ | |
1295 | stripe = chunk_number / data_disks; | |
1296 | ||
1297 | /* | |
1298 | * Compute the data disk and parity disk indexes inside the stripe | |
1299 | */ | |
1300 | *dd_idx = chunk_number % data_disks; | |
1301 | ||
1302 | /* | |
1303 | * Select the parity disk based on the user selected algorithm. | |
1304 | */ | |
911d4ee8 | 1305 | pd_idx = qd_idx = ~0; |
16a53ecc N |
1306 | switch(conf->level) { |
1307 | case 4: | |
911d4ee8 | 1308 | pd_idx = data_disks; |
16a53ecc N |
1309 | break; |
1310 | case 5: | |
1311 | switch (conf->algorithm) { | |
1da177e4 | 1312 | case ALGORITHM_LEFT_ASYMMETRIC: |
911d4ee8 N |
1313 | pd_idx = data_disks - stripe % raid_disks; |
1314 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1315 | (*dd_idx)++; |
1316 | break; | |
1317 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1318 | pd_idx = stripe % raid_disks; |
1319 | if (*dd_idx >= pd_idx) | |
1da177e4 LT |
1320 | (*dd_idx)++; |
1321 | break; | |
1322 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1323 | pd_idx = data_disks - stripe % raid_disks; |
1324 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 LT |
1325 | break; |
1326 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1327 | pd_idx = stripe % raid_disks; |
1328 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
1da177e4 | 1329 | break; |
99c0fb5f N |
1330 | case ALGORITHM_PARITY_0: |
1331 | pd_idx = 0; | |
1332 | (*dd_idx)++; | |
1333 | break; | |
1334 | case ALGORITHM_PARITY_N: | |
1335 | pd_idx = data_disks; | |
1336 | break; | |
1da177e4 | 1337 | default: |
14f8d26b | 1338 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 | 1339 | conf->algorithm); |
99c0fb5f | 1340 | BUG(); |
16a53ecc N |
1341 | } |
1342 | break; | |
1343 | case 6: | |
1344 | ||
16a53ecc N |
1345 | switch (conf->algorithm) { |
1346 | case ALGORITHM_LEFT_ASYMMETRIC: | |
911d4ee8 N |
1347 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1348 | qd_idx = pd_idx + 1; | |
1349 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1350 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1351 | qd_idx = 0; |
1352 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1353 | (*dd_idx) += 2; /* D D P Q D */ |
1354 | break; | |
1355 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
911d4ee8 N |
1356 | pd_idx = stripe % raid_disks; |
1357 | qd_idx = pd_idx + 1; | |
1358 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 1359 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
1360 | qd_idx = 0; |
1361 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
1362 | (*dd_idx) += 2; /* D D P Q D */ |
1363 | break; | |
1364 | case ALGORITHM_LEFT_SYMMETRIC: | |
911d4ee8 N |
1365 | pd_idx = raid_disks - 1 - (stripe % raid_disks); |
1366 | qd_idx = (pd_idx + 1) % raid_disks; | |
1367 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
1368 | break; |
1369 | case ALGORITHM_RIGHT_SYMMETRIC: | |
911d4ee8 N |
1370 | pd_idx = stripe % raid_disks; |
1371 | qd_idx = (pd_idx + 1) % raid_disks; | |
1372 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 1373 | break; |
99c0fb5f N |
1374 | |
1375 | case ALGORITHM_PARITY_0: | |
1376 | pd_idx = 0; | |
1377 | qd_idx = 1; | |
1378 | (*dd_idx) += 2; | |
1379 | break; | |
1380 | case ALGORITHM_PARITY_N: | |
1381 | pd_idx = data_disks; | |
1382 | qd_idx = data_disks + 1; | |
1383 | break; | |
1384 | ||
1385 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
1386 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
1387 | * of blocks for computing Q is different. | |
1388 | */ | |
1389 | pd_idx = stripe % raid_disks; | |
1390 | qd_idx = pd_idx + 1; | |
1391 | if (pd_idx == raid_disks-1) { | |
1392 | (*dd_idx)++; /* Q D D D P */ | |
1393 | qd_idx = 0; | |
1394 | } else if (*dd_idx >= pd_idx) | |
1395 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1396 | ddf_layout = 1; |
99c0fb5f N |
1397 | break; |
1398 | ||
1399 | case ALGORITHM_ROTATING_N_RESTART: | |
1400 | /* Same a left_asymmetric, by first stripe is | |
1401 | * D D D P Q rather than | |
1402 | * Q D D D P | |
1403 | */ | |
1404 | pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
1405 | qd_idx = pd_idx + 1; | |
1406 | if (pd_idx == raid_disks-1) { | |
1407 | (*dd_idx)++; /* Q D D D P */ | |
1408 | qd_idx = 0; | |
1409 | } else if (*dd_idx >= pd_idx) | |
1410 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 1411 | ddf_layout = 1; |
99c0fb5f N |
1412 | break; |
1413 | ||
1414 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1415 | /* Same as left_symmetric but Q is before P */ | |
1416 | pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
1417 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; | |
1418 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 1419 | ddf_layout = 1; |
99c0fb5f N |
1420 | break; |
1421 | ||
1422 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1423 | /* RAID5 left_asymmetric, with Q on last device */ | |
1424 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1425 | if (*dd_idx >= pd_idx) | |
1426 | (*dd_idx)++; | |
1427 | qd_idx = raid_disks - 1; | |
1428 | break; | |
1429 | ||
1430 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1431 | pd_idx = stripe % (raid_disks-1); | |
1432 | if (*dd_idx >= pd_idx) | |
1433 | (*dd_idx)++; | |
1434 | qd_idx = raid_disks - 1; | |
1435 | break; | |
1436 | ||
1437 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1438 | pd_idx = data_disks - stripe % (raid_disks-1); | |
1439 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1440 | qd_idx = raid_disks - 1; | |
1441 | break; | |
1442 | ||
1443 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1444 | pd_idx = stripe % (raid_disks-1); | |
1445 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); | |
1446 | qd_idx = raid_disks - 1; | |
1447 | break; | |
1448 | ||
1449 | case ALGORITHM_PARITY_0_6: | |
1450 | pd_idx = 0; | |
1451 | (*dd_idx)++; | |
1452 | qd_idx = raid_disks - 1; | |
1453 | break; | |
1454 | ||
1455 | ||
16a53ecc | 1456 | default: |
d710e138 N |
1457 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1458 | conf->algorithm); | |
99c0fb5f | 1459 | BUG(); |
16a53ecc N |
1460 | } |
1461 | break; | |
1da177e4 LT |
1462 | } |
1463 | ||
911d4ee8 N |
1464 | if (sh) { |
1465 | sh->pd_idx = pd_idx; | |
1466 | sh->qd_idx = qd_idx; | |
67cc2b81 | 1467 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 1468 | } |
1da177e4 LT |
1469 | /* |
1470 | * Finally, compute the new sector number | |
1471 | */ | |
1472 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
1473 | return new_sector; | |
1474 | } | |
1475 | ||
1476 | ||
1477 | static sector_t compute_blocknr(struct stripe_head *sh, int i) | |
1478 | { | |
1479 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
1480 | int raid_disks = sh->disks; |
1481 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
1482 | sector_t new_sector = sh->sector, check; |
1483 | int sectors_per_chunk = conf->chunk_size >> 9; | |
1484 | sector_t stripe; | |
1485 | int chunk_offset; | |
911d4ee8 | 1486 | int chunk_number, dummy1, dd_idx = i; |
1da177e4 | 1487 | sector_t r_sector; |
911d4ee8 | 1488 | struct stripe_head sh2; |
1da177e4 | 1489 | |
16a53ecc | 1490 | |
1da177e4 LT |
1491 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
1492 | stripe = new_sector; | |
1493 | BUG_ON(new_sector != stripe); | |
1494 | ||
16a53ecc N |
1495 | if (i == sh->pd_idx) |
1496 | return 0; | |
1497 | switch(conf->level) { | |
1498 | case 4: break; | |
1499 | case 5: | |
1500 | switch (conf->algorithm) { | |
1da177e4 LT |
1501 | case ALGORITHM_LEFT_ASYMMETRIC: |
1502 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
1503 | if (i > sh->pd_idx) | |
1504 | i--; | |
1505 | break; | |
1506 | case ALGORITHM_LEFT_SYMMETRIC: | |
1507 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1508 | if (i < sh->pd_idx) | |
1509 | i += raid_disks; | |
1510 | i -= (sh->pd_idx + 1); | |
1511 | break; | |
99c0fb5f N |
1512 | case ALGORITHM_PARITY_0: |
1513 | i -= 1; | |
1514 | break; | |
1515 | case ALGORITHM_PARITY_N: | |
1516 | break; | |
1da177e4 | 1517 | default: |
14f8d26b | 1518 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
16a53ecc | 1519 | conf->algorithm); |
99c0fb5f | 1520 | BUG(); |
16a53ecc N |
1521 | } |
1522 | break; | |
1523 | case 6: | |
d0dabf7e | 1524 | if (i == sh->qd_idx) |
16a53ecc N |
1525 | return 0; /* It is the Q disk */ |
1526 | switch (conf->algorithm) { | |
1527 | case ALGORITHM_LEFT_ASYMMETRIC: | |
1528 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
1529 | case ALGORITHM_ROTATING_ZERO_RESTART: |
1530 | case ALGORITHM_ROTATING_N_RESTART: | |
1531 | if (sh->pd_idx == raid_disks-1) | |
1532 | i--; /* Q D D D P */ | |
16a53ecc N |
1533 | else if (i > sh->pd_idx) |
1534 | i -= 2; /* D D P Q D */ | |
1535 | break; | |
1536 | case ALGORITHM_LEFT_SYMMETRIC: | |
1537 | case ALGORITHM_RIGHT_SYMMETRIC: | |
1538 | if (sh->pd_idx == raid_disks-1) | |
1539 | i--; /* Q D D D P */ | |
1540 | else { | |
1541 | /* D D P Q D */ | |
1542 | if (i < sh->pd_idx) | |
1543 | i += raid_disks; | |
1544 | i -= (sh->pd_idx + 2); | |
1545 | } | |
1546 | break; | |
99c0fb5f N |
1547 | case ALGORITHM_PARITY_0: |
1548 | i -= 2; | |
1549 | break; | |
1550 | case ALGORITHM_PARITY_N: | |
1551 | break; | |
1552 | case ALGORITHM_ROTATING_N_CONTINUE: | |
1553 | if (sh->pd_idx == 0) | |
1554 | i--; /* P D D D Q */ | |
1555 | else if (i > sh->pd_idx) | |
1556 | i -= 2; /* D D Q P D */ | |
1557 | break; | |
1558 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
1559 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
1560 | if (i > sh->pd_idx) | |
1561 | i--; | |
1562 | break; | |
1563 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
1564 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
1565 | if (i < sh->pd_idx) | |
1566 | i += data_disks + 1; | |
1567 | i -= (sh->pd_idx + 1); | |
1568 | break; | |
1569 | case ALGORITHM_PARITY_0_6: | |
1570 | i -= 1; | |
1571 | break; | |
16a53ecc | 1572 | default: |
d710e138 N |
1573 | printk(KERN_CRIT "raid6: unsupported algorithm %d\n", |
1574 | conf->algorithm); | |
99c0fb5f | 1575 | BUG(); |
16a53ecc N |
1576 | } |
1577 | break; | |
1da177e4 LT |
1578 | } |
1579 | ||
1580 | chunk_number = stripe * data_disks + i; | |
1581 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
1582 | ||
112bf897 N |
1583 | check = raid5_compute_sector(conf, r_sector, |
1584 | (raid_disks != conf->raid_disks), | |
911d4ee8 N |
1585 | &dummy1, &sh2); |
1586 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx | |
1587 | || sh2.qd_idx != sh->qd_idx) { | |
14f8d26b | 1588 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
1589 | return 0; |
1590 | } | |
1591 | return r_sector; | |
1592 | } | |
1593 | ||
1594 | ||
1595 | ||
1596 | /* | |
16a53ecc N |
1597 | * Copy data between a page in the stripe cache, and one or more bion |
1598 | * The page could align with the middle of the bio, or there could be | |
1599 | * several bion, each with several bio_vecs, which cover part of the page | |
1600 | * Multiple bion are linked together on bi_next. There may be extras | |
1601 | * at the end of this list. We ignore them. | |
1da177e4 LT |
1602 | */ |
1603 | static void copy_data(int frombio, struct bio *bio, | |
1604 | struct page *page, | |
1605 | sector_t sector) | |
1606 | { | |
1607 | char *pa = page_address(page); | |
1608 | struct bio_vec *bvl; | |
1609 | int i; | |
1610 | int page_offset; | |
1611 | ||
1612 | if (bio->bi_sector >= sector) | |
1613 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
1614 | else | |
1615 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
1616 | bio_for_each_segment(bvl, bio, i) { | |
1617 | int len = bio_iovec_idx(bio,i)->bv_len; | |
1618 | int clen; | |
1619 | int b_offset = 0; | |
1620 | ||
1621 | if (page_offset < 0) { | |
1622 | b_offset = -page_offset; | |
1623 | page_offset += b_offset; | |
1624 | len -= b_offset; | |
1625 | } | |
1626 | ||
1627 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1628 | clen = STRIPE_SIZE - page_offset; | |
1629 | else clen = len; | |
16a53ecc | 1630 | |
1da177e4 LT |
1631 | if (clen > 0) { |
1632 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
1633 | if (frombio) | |
1634 | memcpy(pa+page_offset, ba+b_offset, clen); | |
1635 | else | |
1636 | memcpy(ba+b_offset, pa+page_offset, clen); | |
1637 | __bio_kunmap_atomic(ba, KM_USER0); | |
1638 | } | |
1639 | if (clen < len) /* hit end of page */ | |
1640 | break; | |
1641 | page_offset += len; | |
1642 | } | |
1643 | } | |
1644 | ||
9bc89cd8 DW |
1645 | #define check_xor() do { \ |
1646 | if (count == MAX_XOR_BLOCKS) { \ | |
1647 | xor_blocks(count, STRIPE_SIZE, dest, ptr);\ | |
1648 | count = 0; \ | |
1649 | } \ | |
1da177e4 LT |
1650 | } while(0) |
1651 | ||
16a53ecc N |
1652 | static void compute_parity6(struct stripe_head *sh, int method) |
1653 | { | |
bff61975 | 1654 | raid5_conf_t *conf = sh->raid_conf; |
d0dabf7e | 1655 | int i, pd_idx, qd_idx, d0_idx, disks = sh->disks, count; |
67cc2b81 | 1656 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); |
16a53ecc N |
1657 | struct bio *chosen; |
1658 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
67cc2b81 | 1659 | void *ptrs[syndrome_disks+2]; |
16a53ecc | 1660 | |
d0dabf7e N |
1661 | pd_idx = sh->pd_idx; |
1662 | qd_idx = sh->qd_idx; | |
1663 | d0_idx = raid6_d0(sh); | |
16a53ecc | 1664 | |
45b4233c | 1665 | pr_debug("compute_parity, stripe %llu, method %d\n", |
16a53ecc N |
1666 | (unsigned long long)sh->sector, method); |
1667 | ||
1668 | switch(method) { | |
1669 | case READ_MODIFY_WRITE: | |
1670 | BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ | |
1671 | case RECONSTRUCT_WRITE: | |
1672 | for (i= disks; i-- ;) | |
1673 | if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { | |
1674 | chosen = sh->dev[i].towrite; | |
1675 | sh->dev[i].towrite = NULL; | |
1676 | ||
1677 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1678 | wake_up(&conf->wait_for_overlap); | |
1679 | ||
52e5f9d1 | 1680 | BUG_ON(sh->dev[i].written); |
16a53ecc N |
1681 | sh->dev[i].written = chosen; |
1682 | } | |
1683 | break; | |
1684 | case CHECK_PARITY: | |
1685 | BUG(); /* Not implemented yet */ | |
1686 | } | |
1687 | ||
1688 | for (i = disks; i--;) | |
1689 | if (sh->dev[i].written) { | |
1690 | sector_t sector = sh->dev[i].sector; | |
1691 | struct bio *wbi = sh->dev[i].written; | |
1692 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
1693 | copy_data(1, wbi, sh->dev[i].page, sector); | |
1694 | wbi = r5_next_bio(wbi, sector); | |
1695 | } | |
1696 | ||
1697 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
1698 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
1699 | } | |
1700 | ||
d0dabf7e | 1701 | /* Note that unlike RAID-5, the ordering of the disks matters greatly.*/ |
67cc2b81 N |
1702 | |
1703 | for (i = 0; i < disks; i++) | |
1704 | ptrs[i] = (void *)raid6_empty_zero_page; | |
1705 | ||
d0dabf7e N |
1706 | count = 0; |
1707 | i = d0_idx; | |
1708 | do { | |
67cc2b81 N |
1709 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); |
1710 | ||
d0dabf7e | 1711 | ptrs[slot] = page_address(sh->dev[i].page); |
67cc2b81 | 1712 | if (slot < syndrome_disks && |
d0dabf7e N |
1713 | !test_bit(R5_UPTODATE, &sh->dev[i].flags)) { |
1714 | printk(KERN_ERR "block %d/%d not uptodate " | |
1715 | "on parity calc\n", i, count); | |
1716 | BUG(); | |
1717 | } | |
67cc2b81 | 1718 | |
d0dabf7e N |
1719 | i = raid6_next_disk(i, disks); |
1720 | } while (i != d0_idx); | |
67cc2b81 | 1721 | BUG_ON(count != syndrome_disks); |
16a53ecc | 1722 | |
67cc2b81 | 1723 | raid6_call.gen_syndrome(syndrome_disks+2, STRIPE_SIZE, ptrs); |
16a53ecc N |
1724 | |
1725 | switch(method) { | |
1726 | case RECONSTRUCT_WRITE: | |
1727 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1728 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1729 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1730 | set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); | |
1731 | break; | |
1732 | case UPDATE_PARITY: | |
1733 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1734 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1735 | break; | |
1736 | } | |
1737 | } | |
1738 | ||
1739 | ||
1740 | /* Compute one missing block */ | |
1741 | static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero) | |
1742 | { | |
f416885e | 1743 | int i, count, disks = sh->disks; |
9bc89cd8 | 1744 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
d0dabf7e | 1745 | int qd_idx = sh->qd_idx; |
16a53ecc | 1746 | |
45b4233c | 1747 | pr_debug("compute_block_1, stripe %llu, idx %d\n", |
16a53ecc N |
1748 | (unsigned long long)sh->sector, dd_idx); |
1749 | ||
1750 | if ( dd_idx == qd_idx ) { | |
1751 | /* We're actually computing the Q drive */ | |
1752 | compute_parity6(sh, UPDATE_PARITY); | |
1753 | } else { | |
9bc89cd8 DW |
1754 | dest = page_address(sh->dev[dd_idx].page); |
1755 | if (!nozero) memset(dest, 0, STRIPE_SIZE); | |
1756 | count = 0; | |
16a53ecc N |
1757 | for (i = disks ; i--; ) { |
1758 | if (i == dd_idx || i == qd_idx) | |
1759 | continue; | |
1760 | p = page_address(sh->dev[i].page); | |
1761 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1762 | ptr[count++] = p; | |
1763 | else | |
1764 | printk("compute_block() %d, stripe %llu, %d" | |
1765 | " not present\n", dd_idx, | |
1766 | (unsigned long long)sh->sector, i); | |
1767 | ||
1768 | check_xor(); | |
1769 | } | |
9bc89cd8 DW |
1770 | if (count) |
1771 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
16a53ecc N |
1772 | if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
1773 | else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
1774 | } | |
1775 | } | |
1776 | ||
1777 | /* Compute two missing blocks */ | |
1778 | static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) | |
1779 | { | |
f416885e | 1780 | int i, count, disks = sh->disks; |
67cc2b81 | 1781 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; |
d0dabf7e N |
1782 | int d0_idx = raid6_d0(sh); |
1783 | int faila = -1, failb = -1; | |
1784 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
67cc2b81 | 1785 | void *ptrs[syndrome_disks+2]; |
16a53ecc | 1786 | |
67cc2b81 N |
1787 | for (i = 0; i < disks ; i++) |
1788 | ptrs[i] = (void *)raid6_empty_zero_page; | |
d0dabf7e N |
1789 | count = 0; |
1790 | i = d0_idx; | |
1791 | do { | |
67cc2b81 N |
1792 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); |
1793 | ||
d0dabf7e | 1794 | ptrs[slot] = page_address(sh->dev[i].page); |
67cc2b81 | 1795 | |
d0dabf7e N |
1796 | if (i == dd_idx1) |
1797 | faila = slot; | |
1798 | if (i == dd_idx2) | |
1799 | failb = slot; | |
1800 | i = raid6_next_disk(i, disks); | |
1801 | } while (i != d0_idx); | |
67cc2b81 | 1802 | BUG_ON(count != syndrome_disks); |
16a53ecc N |
1803 | |
1804 | BUG_ON(faila == failb); | |
1805 | if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } | |
1806 | ||
45b4233c | 1807 | pr_debug("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", |
d0dabf7e N |
1808 | (unsigned long long)sh->sector, dd_idx1, dd_idx2, |
1809 | faila, failb); | |
16a53ecc | 1810 | |
67cc2b81 | 1811 | if (failb == syndrome_disks+1) { |
16a53ecc | 1812 | /* Q disk is one of the missing disks */ |
67cc2b81 | 1813 | if (faila == syndrome_disks) { |
16a53ecc N |
1814 | /* Missing P+Q, just recompute */ |
1815 | compute_parity6(sh, UPDATE_PARITY); | |
1816 | return; | |
1817 | } else { | |
1818 | /* We're missing D+Q; recompute D from P */ | |
d0dabf7e N |
1819 | compute_block_1(sh, ((dd_idx1 == sh->qd_idx) ? |
1820 | dd_idx2 : dd_idx1), | |
1821 | 0); | |
16a53ecc N |
1822 | compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */ |
1823 | return; | |
1824 | } | |
1825 | } | |
1826 | ||
d0dabf7e | 1827 | /* We're missing D+P or D+D; */ |
67cc2b81 | 1828 | if (failb == syndrome_disks) { |
d0dabf7e | 1829 | /* We're missing D+P. */ |
67cc2b81 | 1830 | raid6_datap_recov(syndrome_disks+2, STRIPE_SIZE, faila, ptrs); |
d0dabf7e N |
1831 | } else { |
1832 | /* We're missing D+D. */ | |
67cc2b81 N |
1833 | raid6_2data_recov(syndrome_disks+2, STRIPE_SIZE, faila, failb, |
1834 | ptrs); | |
16a53ecc | 1835 | } |
d0dabf7e N |
1836 | |
1837 | /* Both the above update both missing blocks */ | |
1838 | set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); | |
1839 | set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); | |
16a53ecc N |
1840 | } |
1841 | ||
600aa109 | 1842 | static void |
1fe797e6 | 1843 | schedule_reconstruction5(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 1844 | int rcw, int expand) |
e33129d8 DW |
1845 | { |
1846 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
e33129d8 DW |
1847 | |
1848 | if (rcw) { | |
1849 | /* if we are not expanding this is a proper write request, and | |
1850 | * there will be bios with new data to be drained into the | |
1851 | * stripe cache | |
1852 | */ | |
1853 | if (!expand) { | |
600aa109 DW |
1854 | sh->reconstruct_state = reconstruct_state_drain_run; |
1855 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1856 | } else | |
1857 | sh->reconstruct_state = reconstruct_state_run; | |
16a53ecc | 1858 | |
600aa109 | 1859 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); |
e33129d8 DW |
1860 | |
1861 | for (i = disks; i--; ) { | |
1862 | struct r5dev *dev = &sh->dev[i]; | |
1863 | ||
1864 | if (dev->towrite) { | |
1865 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 1866 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
1867 | if (!expand) |
1868 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1869 | s->locked++; |
e33129d8 DW |
1870 | } |
1871 | } | |
600aa109 | 1872 | if (s->locked + 1 == disks) |
8b3e6cdc DW |
1873 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
1874 | atomic_inc(&sh->raid_conf->pending_full_writes); | |
e33129d8 DW |
1875 | } else { |
1876 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
1877 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
1878 | ||
d8ee0728 | 1879 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; |
600aa109 DW |
1880 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); |
1881 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
1882 | set_bit(STRIPE_OP_POSTXOR, &s->ops_request); | |
e33129d8 DW |
1883 | |
1884 | for (i = disks; i--; ) { | |
1885 | struct r5dev *dev = &sh->dev[i]; | |
1886 | if (i == pd_idx) | |
1887 | continue; | |
1888 | ||
e33129d8 DW |
1889 | if (dev->towrite && |
1890 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
1891 | test_bit(R5_Wantcompute, &dev->flags))) { |
1892 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
1893 | set_bit(R5_LOCKED, &dev->flags); |
1894 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 1895 | s->locked++; |
e33129d8 DW |
1896 | } |
1897 | } | |
1898 | } | |
1899 | ||
1900 | /* keep the parity disk locked while asynchronous operations | |
1901 | * are in flight | |
1902 | */ | |
1903 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1904 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 1905 | s->locked++; |
e33129d8 | 1906 | |
600aa109 | 1907 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 1908 | __func__, (unsigned long long)sh->sector, |
600aa109 | 1909 | s->locked, s->ops_request); |
e33129d8 | 1910 | } |
16a53ecc | 1911 | |
1da177e4 LT |
1912 | /* |
1913 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 1914 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
1915 | * The bi_next chain must be in order. |
1916 | */ | |
1917 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
1918 | { | |
1919 | struct bio **bip; | |
1920 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 1921 | int firstwrite=0; |
1da177e4 | 1922 | |
45b4233c | 1923 | pr_debug("adding bh b#%llu to stripe s#%llu\n", |
1da177e4 LT |
1924 | (unsigned long long)bi->bi_sector, |
1925 | (unsigned long long)sh->sector); | |
1926 | ||
1927 | ||
1928 | spin_lock(&sh->lock); | |
1929 | spin_lock_irq(&conf->device_lock); | |
72626685 | 1930 | if (forwrite) { |
1da177e4 | 1931 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
1932 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
1933 | firstwrite = 1; | |
1934 | } else | |
1da177e4 LT |
1935 | bip = &sh->dev[dd_idx].toread; |
1936 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
1937 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
1938 | goto overlap; | |
1939 | bip = & (*bip)->bi_next; | |
1940 | } | |
1941 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
1942 | goto overlap; | |
1943 | ||
78bafebd | 1944 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
1945 | if (*bip) |
1946 | bi->bi_next = *bip; | |
1947 | *bip = bi; | |
960e739d | 1948 | bi->bi_phys_segments++; |
1da177e4 LT |
1949 | spin_unlock_irq(&conf->device_lock); |
1950 | spin_unlock(&sh->lock); | |
1951 | ||
45b4233c | 1952 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", |
1da177e4 LT |
1953 | (unsigned long long)bi->bi_sector, |
1954 | (unsigned long long)sh->sector, dd_idx); | |
1955 | ||
72626685 | 1956 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
1957 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
1958 | STRIPE_SECTORS, 0); | |
ae3c20cc | 1959 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
1960 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
1961 | } | |
1962 | ||
1da177e4 LT |
1963 | if (forwrite) { |
1964 | /* check if page is covered */ | |
1965 | sector_t sector = sh->dev[dd_idx].sector; | |
1966 | for (bi=sh->dev[dd_idx].towrite; | |
1967 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
1968 | bi && bi->bi_sector <= sector; | |
1969 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
1970 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
1971 | sector = bi->bi_sector + (bi->bi_size>>9); | |
1972 | } | |
1973 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
1974 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
1975 | } | |
1976 | return 1; | |
1977 | ||
1978 | overlap: | |
1979 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
1980 | spin_unlock_irq(&conf->device_lock); | |
1981 | spin_unlock(&sh->lock); | |
1982 | return 0; | |
1983 | } | |
1984 | ||
29269553 N |
1985 | static void end_reshape(raid5_conf_t *conf); |
1986 | ||
16a53ecc N |
1987 | static int page_is_zero(struct page *p) |
1988 | { | |
1989 | char *a = page_address(p); | |
1990 | return ((*(u32*)a) == 0 && | |
1991 | memcmp(a, a+4, STRIPE_SIZE-4)==0); | |
1992 | } | |
1993 | ||
911d4ee8 N |
1994 | static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous, |
1995 | struct stripe_head *sh) | |
ccfcc3c1 N |
1996 | { |
1997 | int sectors_per_chunk = conf->chunk_size >> 9; | |
911d4ee8 | 1998 | int dd_idx; |
2d2063ce | 1999 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2000 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2001 | |
112bf897 N |
2002 | raid5_compute_sector(conf, |
2003 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2004 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2005 | previous, |
911d4ee8 | 2006 | &dd_idx, sh); |
ccfcc3c1 N |
2007 | } |
2008 | ||
a4456856 | 2009 | static void |
1fe797e6 | 2010 | handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh, |
a4456856 DW |
2011 | struct stripe_head_state *s, int disks, |
2012 | struct bio **return_bi) | |
2013 | { | |
2014 | int i; | |
2015 | for (i = disks; i--; ) { | |
2016 | struct bio *bi; | |
2017 | int bitmap_end = 0; | |
2018 | ||
2019 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
2020 | mdk_rdev_t *rdev; | |
2021 | rcu_read_lock(); | |
2022 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2023 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
2024 | /* multiple read failures in one stripe */ | |
2025 | md_error(conf->mddev, rdev); | |
2026 | rcu_read_unlock(); | |
2027 | } | |
2028 | spin_lock_irq(&conf->device_lock); | |
2029 | /* fail all writes first */ | |
2030 | bi = sh->dev[i].towrite; | |
2031 | sh->dev[i].towrite = NULL; | |
2032 | if (bi) { | |
2033 | s->to_write--; | |
2034 | bitmap_end = 1; | |
2035 | } | |
2036 | ||
2037 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2038 | wake_up(&conf->wait_for_overlap); | |
2039 | ||
2040 | while (bi && bi->bi_sector < | |
2041 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2042 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2043 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2044 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2045 | md_write_end(conf->mddev); |
2046 | bi->bi_next = *return_bi; | |
2047 | *return_bi = bi; | |
2048 | } | |
2049 | bi = nextbi; | |
2050 | } | |
2051 | /* and fail all 'written' */ | |
2052 | bi = sh->dev[i].written; | |
2053 | sh->dev[i].written = NULL; | |
2054 | if (bi) bitmap_end = 1; | |
2055 | while (bi && bi->bi_sector < | |
2056 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2057 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2058 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2059 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2060 | md_write_end(conf->mddev); |
2061 | bi->bi_next = *return_bi; | |
2062 | *return_bi = bi; | |
2063 | } | |
2064 | bi = bi2; | |
2065 | } | |
2066 | ||
b5e98d65 DW |
2067 | /* fail any reads if this device is non-operational and |
2068 | * the data has not reached the cache yet. | |
2069 | */ | |
2070 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2071 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2072 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
a4456856 DW |
2073 | bi = sh->dev[i].toread; |
2074 | sh->dev[i].toread = NULL; | |
2075 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2076 | wake_up(&conf->wait_for_overlap); | |
2077 | if (bi) s->to_read--; | |
2078 | while (bi && bi->bi_sector < | |
2079 | sh->dev[i].sector + STRIPE_SECTORS) { | |
2080 | struct bio *nextbi = | |
2081 | r5_next_bio(bi, sh->dev[i].sector); | |
2082 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
960e739d | 2083 | if (!raid5_dec_bi_phys_segments(bi)) { |
a4456856 DW |
2084 | bi->bi_next = *return_bi; |
2085 | *return_bi = bi; | |
2086 | } | |
2087 | bi = nextbi; | |
2088 | } | |
2089 | } | |
2090 | spin_unlock_irq(&conf->device_lock); | |
2091 | if (bitmap_end) | |
2092 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2093 | STRIPE_SECTORS, 0, 0); | |
2094 | } | |
2095 | ||
8b3e6cdc DW |
2096 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2097 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2098 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2099 | } |
2100 | ||
1fe797e6 DW |
2101 | /* fetch_block5 - checks the given member device to see if its data needs |
2102 | * to be read or computed to satisfy a request. | |
2103 | * | |
2104 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
2105 | * 0 to tell the loop in handle_stripe_fill5 to continue | |
f38e1219 | 2106 | */ |
1fe797e6 DW |
2107 | static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s, |
2108 | int disk_idx, int disks) | |
f38e1219 DW |
2109 | { |
2110 | struct r5dev *dev = &sh->dev[disk_idx]; | |
2111 | struct r5dev *failed_dev = &sh->dev[s->failed_num]; | |
2112 | ||
f38e1219 DW |
2113 | /* is the data in this block needed, and can we get it? */ |
2114 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1fe797e6 DW |
2115 | !test_bit(R5_UPTODATE, &dev->flags) && |
2116 | (dev->toread || | |
2117 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2118 | s->syncing || s->expanding || | |
2119 | (s->failed && | |
2120 | (failed_dev->toread || | |
2121 | (failed_dev->towrite && | |
2122 | !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) { | |
976ea8d4 DW |
2123 | /* We would like to get this block, possibly by computing it, |
2124 | * otherwise read it if the backing disk is insync | |
f38e1219 DW |
2125 | */ |
2126 | if ((s->uptodate == disks - 1) && | |
ecc65c9b | 2127 | (s->failed && disk_idx == s->failed_num)) { |
976ea8d4 DW |
2128 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
2129 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
f38e1219 DW |
2130 | set_bit(R5_Wantcompute, &dev->flags); |
2131 | sh->ops.target = disk_idx; | |
2132 | s->req_compute = 1; | |
f38e1219 DW |
2133 | /* Careful: from this point on 'uptodate' is in the eye |
2134 | * of raid5_run_ops which services 'compute' operations | |
2135 | * before writes. R5_Wantcompute flags a block that will | |
2136 | * be R5_UPTODATE by the time it is needed for a | |
2137 | * subsequent operation. | |
2138 | */ | |
2139 | s->uptodate++; | |
1fe797e6 | 2140 | return 1; /* uptodate + compute == disks */ |
7a1fc53c | 2141 | } else if (test_bit(R5_Insync, &dev->flags)) { |
f38e1219 DW |
2142 | set_bit(R5_LOCKED, &dev->flags); |
2143 | set_bit(R5_Wantread, &dev->flags); | |
f38e1219 DW |
2144 | s->locked++; |
2145 | pr_debug("Reading block %d (sync=%d)\n", disk_idx, | |
2146 | s->syncing); | |
2147 | } | |
2148 | } | |
2149 | ||
1fe797e6 | 2150 | return 0; |
f38e1219 DW |
2151 | } |
2152 | ||
1fe797e6 DW |
2153 | /** |
2154 | * handle_stripe_fill5 - read or compute data to satisfy pending requests. | |
2155 | */ | |
2156 | static void handle_stripe_fill5(struct stripe_head *sh, | |
a4456856 DW |
2157 | struct stripe_head_state *s, int disks) |
2158 | { | |
2159 | int i; | |
f38e1219 | 2160 | |
f38e1219 DW |
2161 | /* look for blocks to read/compute, skip this if a compute |
2162 | * is already in flight, or if the stripe contents are in the | |
2163 | * midst of changing due to a write | |
2164 | */ | |
976ea8d4 | 2165 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && |
1fe797e6 | 2166 | !sh->reconstruct_state) |
f38e1219 | 2167 | for (i = disks; i--; ) |
1fe797e6 | 2168 | if (fetch_block5(sh, s, i, disks)) |
f38e1219 | 2169 | break; |
a4456856 DW |
2170 | set_bit(STRIPE_HANDLE, &sh->state); |
2171 | } | |
2172 | ||
1fe797e6 | 2173 | static void handle_stripe_fill6(struct stripe_head *sh, |
a4456856 DW |
2174 | struct stripe_head_state *s, struct r6_state *r6s, |
2175 | int disks) | |
2176 | { | |
2177 | int i; | |
2178 | for (i = disks; i--; ) { | |
2179 | struct r5dev *dev = &sh->dev[i]; | |
2180 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2181 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2182 | (dev->toread || (dev->towrite && | |
2183 | !test_bit(R5_OVERWRITE, &dev->flags)) || | |
2184 | s->syncing || s->expanding || | |
2185 | (s->failed >= 1 && | |
2186 | (sh->dev[r6s->failed_num[0]].toread || | |
2187 | s->to_write)) || | |
2188 | (s->failed >= 2 && | |
2189 | (sh->dev[r6s->failed_num[1]].toread || | |
2190 | s->to_write)))) { | |
2191 | /* we would like to get this block, possibly | |
2192 | * by computing it, but we might not be able to | |
2193 | */ | |
c337869d DW |
2194 | if ((s->uptodate == disks - 1) && |
2195 | (s->failed && (i == r6s->failed_num[0] || | |
2196 | i == r6s->failed_num[1]))) { | |
45b4233c | 2197 | pr_debug("Computing stripe %llu block %d\n", |
a4456856 DW |
2198 | (unsigned long long)sh->sector, i); |
2199 | compute_block_1(sh, i, 0); | |
2200 | s->uptodate++; | |
2201 | } else if ( s->uptodate == disks-2 && s->failed >= 2 ) { | |
2202 | /* Computing 2-failure is *very* expensive; only | |
2203 | * do it if failed >= 2 | |
2204 | */ | |
2205 | int other; | |
2206 | for (other = disks; other--; ) { | |
2207 | if (other == i) | |
2208 | continue; | |
2209 | if (!test_bit(R5_UPTODATE, | |
2210 | &sh->dev[other].flags)) | |
2211 | break; | |
2212 | } | |
2213 | BUG_ON(other < 0); | |
45b4233c | 2214 | pr_debug("Computing stripe %llu blocks %d,%d\n", |
a4456856 DW |
2215 | (unsigned long long)sh->sector, |
2216 | i, other); | |
2217 | compute_block_2(sh, i, other); | |
2218 | s->uptodate += 2; | |
2219 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
2220 | set_bit(R5_LOCKED, &dev->flags); | |
2221 | set_bit(R5_Wantread, &dev->flags); | |
2222 | s->locked++; | |
45b4233c | 2223 | pr_debug("Reading block %d (sync=%d)\n", |
a4456856 DW |
2224 | i, s->syncing); |
2225 | } | |
2226 | } | |
2227 | } | |
2228 | set_bit(STRIPE_HANDLE, &sh->state); | |
2229 | } | |
2230 | ||
2231 | ||
1fe797e6 | 2232 | /* handle_stripe_clean_event |
a4456856 DW |
2233 | * any written block on an uptodate or failed drive can be returned. |
2234 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
2235 | * never LOCKED, so we don't need to test 'failed' directly. | |
2236 | */ | |
1fe797e6 | 2237 | static void handle_stripe_clean_event(raid5_conf_t *conf, |
a4456856 DW |
2238 | struct stripe_head *sh, int disks, struct bio **return_bi) |
2239 | { | |
2240 | int i; | |
2241 | struct r5dev *dev; | |
2242 | ||
2243 | for (i = disks; i--; ) | |
2244 | if (sh->dev[i].written) { | |
2245 | dev = &sh->dev[i]; | |
2246 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
2247 | test_bit(R5_UPTODATE, &dev->flags)) { | |
2248 | /* We can return any write requests */ | |
2249 | struct bio *wbi, *wbi2; | |
2250 | int bitmap_end = 0; | |
45b4233c | 2251 | pr_debug("Return write for disc %d\n", i); |
a4456856 DW |
2252 | spin_lock_irq(&conf->device_lock); |
2253 | wbi = dev->written; | |
2254 | dev->written = NULL; | |
2255 | while (wbi && wbi->bi_sector < | |
2256 | dev->sector + STRIPE_SECTORS) { | |
2257 | wbi2 = r5_next_bio(wbi, dev->sector); | |
960e739d | 2258 | if (!raid5_dec_bi_phys_segments(wbi)) { |
a4456856 DW |
2259 | md_write_end(conf->mddev); |
2260 | wbi->bi_next = *return_bi; | |
2261 | *return_bi = wbi; | |
2262 | } | |
2263 | wbi = wbi2; | |
2264 | } | |
2265 | if (dev->towrite == NULL) | |
2266 | bitmap_end = 1; | |
2267 | spin_unlock_irq(&conf->device_lock); | |
2268 | if (bitmap_end) | |
2269 | bitmap_endwrite(conf->mddev->bitmap, | |
2270 | sh->sector, | |
2271 | STRIPE_SECTORS, | |
2272 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
2273 | 0); | |
2274 | } | |
2275 | } | |
8b3e6cdc DW |
2276 | |
2277 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2278 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2279 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2280 | } |
2281 | ||
1fe797e6 | 2282 | static void handle_stripe_dirtying5(raid5_conf_t *conf, |
a4456856 DW |
2283 | struct stripe_head *sh, struct stripe_head_state *s, int disks) |
2284 | { | |
2285 | int rmw = 0, rcw = 0, i; | |
2286 | for (i = disks; i--; ) { | |
2287 | /* would I have to read this buffer for read_modify_write */ | |
2288 | struct r5dev *dev = &sh->dev[i]; | |
2289 | if ((dev->towrite || i == sh->pd_idx) && | |
2290 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2291 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2292 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
2293 | if (test_bit(R5_Insync, &dev->flags)) |
2294 | rmw++; | |
2295 | else | |
2296 | rmw += 2*disks; /* cannot read it */ | |
2297 | } | |
2298 | /* Would I have to read this buffer for reconstruct_write */ | |
2299 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
2300 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2301 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2302 | test_bit(R5_Wantcompute, &dev->flags))) { | |
2303 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
a4456856 DW |
2304 | else |
2305 | rcw += 2*disks; | |
2306 | } | |
2307 | } | |
45b4233c | 2308 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
2309 | (unsigned long long)sh->sector, rmw, rcw); |
2310 | set_bit(STRIPE_HANDLE, &sh->state); | |
2311 | if (rmw < rcw && rmw > 0) | |
2312 | /* prefer read-modify-write, but need to get some data */ | |
2313 | for (i = disks; i--; ) { | |
2314 | struct r5dev *dev = &sh->dev[i]; | |
2315 | if ((dev->towrite || i == sh->pd_idx) && | |
2316 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2317 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2318 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2319 | test_bit(R5_Insync, &dev->flags)) { |
2320 | if ( | |
2321 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2322 | pr_debug("Read_old block " |
a4456856 DW |
2323 | "%d for r-m-w\n", i); |
2324 | set_bit(R5_LOCKED, &dev->flags); | |
2325 | set_bit(R5_Wantread, &dev->flags); | |
2326 | s->locked++; | |
2327 | } else { | |
2328 | set_bit(STRIPE_DELAYED, &sh->state); | |
2329 | set_bit(STRIPE_HANDLE, &sh->state); | |
2330 | } | |
2331 | } | |
2332 | } | |
2333 | if (rcw <= rmw && rcw > 0) | |
2334 | /* want reconstruct write, but need to get some data */ | |
2335 | for (i = disks; i--; ) { | |
2336 | struct r5dev *dev = &sh->dev[i]; | |
2337 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2338 | i != sh->pd_idx && | |
2339 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
2340 | !(test_bit(R5_UPTODATE, &dev->flags) || |
2341 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 DW |
2342 | test_bit(R5_Insync, &dev->flags)) { |
2343 | if ( | |
2344 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2345 | pr_debug("Read_old block " |
a4456856 DW |
2346 | "%d for Reconstruct\n", i); |
2347 | set_bit(R5_LOCKED, &dev->flags); | |
2348 | set_bit(R5_Wantread, &dev->flags); | |
2349 | s->locked++; | |
2350 | } else { | |
2351 | set_bit(STRIPE_DELAYED, &sh->state); | |
2352 | set_bit(STRIPE_HANDLE, &sh->state); | |
2353 | } | |
2354 | } | |
2355 | } | |
2356 | /* now if nothing is locked, and if we have enough data, | |
2357 | * we can start a write request | |
2358 | */ | |
f38e1219 DW |
2359 | /* since handle_stripe can be called at any time we need to handle the |
2360 | * case where a compute block operation has been submitted and then a | |
2361 | * subsequent call wants to start a write request. raid5_run_ops only | |
2362 | * handles the case where compute block and postxor are requested | |
2363 | * simultaneously. If this is not the case then new writes need to be | |
2364 | * held off until the compute completes. | |
2365 | */ | |
976ea8d4 DW |
2366 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
2367 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
2368 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
1fe797e6 | 2369 | schedule_reconstruction5(sh, s, rcw == 0, 0); |
a4456856 DW |
2370 | } |
2371 | ||
1fe797e6 | 2372 | static void handle_stripe_dirtying6(raid5_conf_t *conf, |
a4456856 DW |
2373 | struct stripe_head *sh, struct stripe_head_state *s, |
2374 | struct r6_state *r6s, int disks) | |
2375 | { | |
2376 | int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i; | |
2377 | int qd_idx = r6s->qd_idx; | |
2378 | for (i = disks; i--; ) { | |
2379 | struct r5dev *dev = &sh->dev[i]; | |
2380 | /* Would I have to read this buffer for reconstruct_write */ | |
2381 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2382 | && i != pd_idx && i != qd_idx | |
2383 | && (!test_bit(R5_LOCKED, &dev->flags) | |
2384 | ) && | |
2385 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
2386 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
2387 | else { | |
45b4233c | 2388 | pr_debug("raid6: must_compute: " |
a4456856 DW |
2389 | "disk %d flags=%#lx\n", i, dev->flags); |
2390 | must_compute++; | |
2391 | } | |
2392 | } | |
2393 | } | |
45b4233c | 2394 | pr_debug("for sector %llu, rcw=%d, must_compute=%d\n", |
a4456856 DW |
2395 | (unsigned long long)sh->sector, rcw, must_compute); |
2396 | set_bit(STRIPE_HANDLE, &sh->state); | |
2397 | ||
2398 | if (rcw > 0) | |
2399 | /* want reconstruct write, but need to get some data */ | |
2400 | for (i = disks; i--; ) { | |
2401 | struct r5dev *dev = &sh->dev[i]; | |
2402 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
2403 | && !(s->failed == 0 && (i == pd_idx || i == qd_idx)) | |
2404 | && !test_bit(R5_LOCKED, &dev->flags) && | |
2405 | !test_bit(R5_UPTODATE, &dev->flags) && | |
2406 | test_bit(R5_Insync, &dev->flags)) { | |
2407 | if ( | |
2408 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
45b4233c | 2409 | pr_debug("Read_old stripe %llu " |
a4456856 DW |
2410 | "block %d for Reconstruct\n", |
2411 | (unsigned long long)sh->sector, i); | |
2412 | set_bit(R5_LOCKED, &dev->flags); | |
2413 | set_bit(R5_Wantread, &dev->flags); | |
2414 | s->locked++; | |
2415 | } else { | |
45b4233c | 2416 | pr_debug("Request delayed stripe %llu " |
a4456856 DW |
2417 | "block %d for Reconstruct\n", |
2418 | (unsigned long long)sh->sector, i); | |
2419 | set_bit(STRIPE_DELAYED, &sh->state); | |
2420 | set_bit(STRIPE_HANDLE, &sh->state); | |
2421 | } | |
2422 | } | |
2423 | } | |
2424 | /* now if nothing is locked, and if we have enough data, we can start a | |
2425 | * write request | |
2426 | */ | |
2427 | if (s->locked == 0 && rcw == 0 && | |
2428 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
2429 | if (must_compute > 0) { | |
2430 | /* We have failed blocks and need to compute them */ | |
2431 | switch (s->failed) { | |
2432 | case 0: | |
2433 | BUG(); | |
2434 | case 1: | |
2435 | compute_block_1(sh, r6s->failed_num[0], 0); | |
2436 | break; | |
2437 | case 2: | |
2438 | compute_block_2(sh, r6s->failed_num[0], | |
2439 | r6s->failed_num[1]); | |
2440 | break; | |
2441 | default: /* This request should have been failed? */ | |
2442 | BUG(); | |
2443 | } | |
2444 | } | |
2445 | ||
45b4233c | 2446 | pr_debug("Computing parity for stripe %llu\n", |
a4456856 DW |
2447 | (unsigned long long)sh->sector); |
2448 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
2449 | /* now every locked buffer is ready to be written */ | |
2450 | for (i = disks; i--; ) | |
2451 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
45b4233c | 2452 | pr_debug("Writing stripe %llu block %d\n", |
a4456856 DW |
2453 | (unsigned long long)sh->sector, i); |
2454 | s->locked++; | |
2455 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
2456 | } | |
8b3e6cdc DW |
2457 | if (s->locked == disks) |
2458 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) | |
2459 | atomic_inc(&conf->pending_full_writes); | |
a4456856 DW |
2460 | /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */ |
2461 | set_bit(STRIPE_INSYNC, &sh->state); | |
2462 | ||
2463 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2464 | atomic_dec(&conf->preread_active_stripes); | |
2465 | if (atomic_read(&conf->preread_active_stripes) < | |
2466 | IO_THRESHOLD) | |
2467 | md_wakeup_thread(conf->mddev->thread); | |
2468 | } | |
2469 | } | |
2470 | } | |
2471 | ||
2472 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
2473 | struct stripe_head_state *s, int disks) | |
2474 | { | |
ecc65c9b | 2475 | struct r5dev *dev = NULL; |
bd2ab670 | 2476 | |
a4456856 | 2477 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 2478 | |
ecc65c9b DW |
2479 | switch (sh->check_state) { |
2480 | case check_state_idle: | |
2481 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 2482 | if (s->failed == 0) { |
bd2ab670 | 2483 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
2484 | sh->check_state = check_state_run; |
2485 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 2486 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 2487 | s->uptodate--; |
ecc65c9b | 2488 | break; |
bd2ab670 | 2489 | } |
ecc65c9b DW |
2490 | dev = &sh->dev[s->failed_num]; |
2491 | /* fall through */ | |
2492 | case check_state_compute_result: | |
2493 | sh->check_state = check_state_idle; | |
2494 | if (!dev) | |
2495 | dev = &sh->dev[sh->pd_idx]; | |
2496 | ||
2497 | /* check that a write has not made the stripe insync */ | |
2498 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
2499 | break; | |
c8894419 | 2500 | |
a4456856 | 2501 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
2502 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
2503 | BUG_ON(s->uptodate != disks); | |
2504 | ||
2505 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 2506 | s->locked++; |
a4456856 | 2507 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 2508 | |
a4456856 | 2509 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 2510 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
2511 | break; |
2512 | case check_state_run: | |
2513 | break; /* we will be called again upon completion */ | |
2514 | case check_state_check_result: | |
2515 | sh->check_state = check_state_idle; | |
2516 | ||
2517 | /* if a failure occurred during the check operation, leave | |
2518 | * STRIPE_INSYNC not set and let the stripe be handled again | |
2519 | */ | |
2520 | if (s->failed) | |
2521 | break; | |
2522 | ||
2523 | /* handle a successful check operation, if parity is correct | |
2524 | * we are done. Otherwise update the mismatch count and repair | |
2525 | * parity if !MD_RECOVERY_CHECK | |
2526 | */ | |
2527 | if (sh->ops.zero_sum_result == 0) | |
2528 | /* parity is correct (on disc, | |
2529 | * not in buffer any more) | |
2530 | */ | |
2531 | set_bit(STRIPE_INSYNC, &sh->state); | |
2532 | else { | |
2533 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2534 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2535 | /* don't try to repair!! */ | |
2536 | set_bit(STRIPE_INSYNC, &sh->state); | |
2537 | else { | |
2538 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 2539 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
2540 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
2541 | set_bit(R5_Wantcompute, | |
2542 | &sh->dev[sh->pd_idx].flags); | |
2543 | sh->ops.target = sh->pd_idx; | |
2544 | s->uptodate++; | |
2545 | } | |
2546 | } | |
2547 | break; | |
2548 | case check_state_compute_run: | |
2549 | break; | |
2550 | default: | |
2551 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
2552 | __func__, sh->check_state, | |
2553 | (unsigned long long) sh->sector); | |
2554 | BUG(); | |
a4456856 DW |
2555 | } |
2556 | } | |
2557 | ||
2558 | ||
2559 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
2560 | struct stripe_head_state *s, | |
2561 | struct r6_state *r6s, struct page *tmp_page, | |
2562 | int disks) | |
2563 | { | |
2564 | int update_p = 0, update_q = 0; | |
2565 | struct r5dev *dev; | |
2566 | int pd_idx = sh->pd_idx; | |
2567 | int qd_idx = r6s->qd_idx; | |
2568 | ||
2569 | set_bit(STRIPE_HANDLE, &sh->state); | |
2570 | ||
2571 | BUG_ON(s->failed > 2); | |
2572 | BUG_ON(s->uptodate < disks); | |
2573 | /* Want to check and possibly repair P and Q. | |
2574 | * However there could be one 'failed' device, in which | |
2575 | * case we can only check one of them, possibly using the | |
2576 | * other to generate missing data | |
2577 | */ | |
2578 | ||
2579 | /* If !tmp_page, we cannot do the calculations, | |
2580 | * but as we have set STRIPE_HANDLE, we will soon be called | |
2581 | * by stripe_handle with a tmp_page - just wait until then. | |
2582 | */ | |
2583 | if (tmp_page) { | |
2584 | if (s->failed == r6s->q_failed) { | |
2585 | /* The only possible failed device holds 'Q', so it | |
2586 | * makes sense to check P (If anything else were failed, | |
2587 | * we would have used P to recreate it). | |
2588 | */ | |
2589 | compute_block_1(sh, pd_idx, 1); | |
2590 | if (!page_is_zero(sh->dev[pd_idx].page)) { | |
2591 | compute_block_1(sh, pd_idx, 0); | |
2592 | update_p = 1; | |
2593 | } | |
2594 | } | |
2595 | if (!r6s->q_failed && s->failed < 2) { | |
2596 | /* q is not failed, and we didn't use it to generate | |
2597 | * anything, so it makes sense to check it | |
2598 | */ | |
2599 | memcpy(page_address(tmp_page), | |
2600 | page_address(sh->dev[qd_idx].page), | |
2601 | STRIPE_SIZE); | |
2602 | compute_parity6(sh, UPDATE_PARITY); | |
2603 | if (memcmp(page_address(tmp_page), | |
2604 | page_address(sh->dev[qd_idx].page), | |
2605 | STRIPE_SIZE) != 0) { | |
2606 | clear_bit(STRIPE_INSYNC, &sh->state); | |
2607 | update_q = 1; | |
2608 | } | |
2609 | } | |
2610 | if (update_p || update_q) { | |
2611 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
2612 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
2613 | /* don't try to repair!! */ | |
2614 | update_p = update_q = 0; | |
2615 | } | |
2616 | ||
2617 | /* now write out any block on a failed drive, | |
2618 | * or P or Q if they need it | |
2619 | */ | |
2620 | ||
2621 | if (s->failed == 2) { | |
2622 | dev = &sh->dev[r6s->failed_num[1]]; | |
2623 | s->locked++; | |
2624 | set_bit(R5_LOCKED, &dev->flags); | |
2625 | set_bit(R5_Wantwrite, &dev->flags); | |
2626 | } | |
2627 | if (s->failed >= 1) { | |
2628 | dev = &sh->dev[r6s->failed_num[0]]; | |
2629 | s->locked++; | |
2630 | set_bit(R5_LOCKED, &dev->flags); | |
2631 | set_bit(R5_Wantwrite, &dev->flags); | |
2632 | } | |
2633 | ||
2634 | if (update_p) { | |
2635 | dev = &sh->dev[pd_idx]; | |
2636 | s->locked++; | |
2637 | set_bit(R5_LOCKED, &dev->flags); | |
2638 | set_bit(R5_Wantwrite, &dev->flags); | |
2639 | } | |
2640 | if (update_q) { | |
2641 | dev = &sh->dev[qd_idx]; | |
2642 | s->locked++; | |
2643 | set_bit(R5_LOCKED, &dev->flags); | |
2644 | set_bit(R5_Wantwrite, &dev->flags); | |
2645 | } | |
2646 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
2647 | ||
2648 | set_bit(STRIPE_INSYNC, &sh->state); | |
2649 | } | |
2650 | } | |
2651 | ||
2652 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
2653 | struct r6_state *r6s) | |
2654 | { | |
2655 | int i; | |
2656 | ||
2657 | /* We have read all the blocks in this stripe and now we need to | |
2658 | * copy some of them into a target stripe for expand. | |
2659 | */ | |
f0a50d37 | 2660 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
2661 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
2662 | for (i = 0; i < sh->disks; i++) | |
a2e08551 | 2663 | if (i != sh->pd_idx && (!r6s || i != r6s->qd_idx)) { |
911d4ee8 | 2664 | int dd_idx, j; |
a4456856 DW |
2665 | struct stripe_head *sh2; |
2666 | ||
2667 | sector_t bn = compute_blocknr(sh, i); | |
911d4ee8 N |
2668 | sector_t s = raid5_compute_sector(conf, bn, 0, |
2669 | &dd_idx, NULL); | |
b5663ba4 | 2670 | sh2 = get_active_stripe(conf, s, 0, 1); |
a4456856 DW |
2671 | if (sh2 == NULL) |
2672 | /* so far only the early blocks of this stripe | |
2673 | * have been requested. When later blocks | |
2674 | * get requested, we will try again | |
2675 | */ | |
2676 | continue; | |
2677 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
2678 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
2679 | /* must have already done this block */ | |
2680 | release_stripe(sh2); | |
2681 | continue; | |
2682 | } | |
f0a50d37 DW |
2683 | |
2684 | /* place all the copies on one channel */ | |
2685 | tx = async_memcpy(sh2->dev[dd_idx].page, | |
2686 | sh->dev[i].page, 0, 0, STRIPE_SIZE, | |
2687 | ASYNC_TX_DEP_ACK, tx, NULL, NULL); | |
2688 | ||
a4456856 DW |
2689 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
2690 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
2691 | for (j = 0; j < conf->raid_disks; j++) | |
2692 | if (j != sh2->pd_idx && | |
d0dabf7e | 2693 | (!r6s || j != sh2->qd_idx) && |
a4456856 DW |
2694 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
2695 | break; | |
2696 | if (j == conf->raid_disks) { | |
2697 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
2698 | set_bit(STRIPE_HANDLE, &sh2->state); | |
2699 | } | |
2700 | release_stripe(sh2); | |
f0a50d37 | 2701 | |
a4456856 | 2702 | } |
a2e08551 N |
2703 | /* done submitting copies, wait for them to complete */ |
2704 | if (tx) { | |
2705 | async_tx_ack(tx); | |
2706 | dma_wait_for_async_tx(tx); | |
2707 | } | |
a4456856 | 2708 | } |
1da177e4 | 2709 | |
6bfe0b49 | 2710 | |
1da177e4 LT |
2711 | /* |
2712 | * handle_stripe - do things to a stripe. | |
2713 | * | |
2714 | * We lock the stripe and then examine the state of various bits | |
2715 | * to see what needs to be done. | |
2716 | * Possible results: | |
2717 | * return some read request which now have data | |
2718 | * return some write requests which are safely on disc | |
2719 | * schedule a read on some buffers | |
2720 | * schedule a write of some buffers | |
2721 | * return confirmation of parity correctness | |
2722 | * | |
1da177e4 LT |
2723 | * buffers are taken off read_list or write_list, and bh_cache buffers |
2724 | * get BH_Lock set before the stripe lock is released. | |
2725 | * | |
2726 | */ | |
a4456856 | 2727 | |
df10cfbc | 2728 | static bool handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
2729 | { |
2730 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
2731 | int disks = sh->disks, i; |
2732 | struct bio *return_bi = NULL; | |
2733 | struct stripe_head_state s; | |
1da177e4 | 2734 | struct r5dev *dev; |
6bfe0b49 | 2735 | mdk_rdev_t *blocked_rdev = NULL; |
e0a115e5 | 2736 | int prexor; |
1da177e4 | 2737 | |
a4456856 | 2738 | memset(&s, 0, sizeof(s)); |
600aa109 DW |
2739 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d " |
2740 | "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state, | |
2741 | atomic_read(&sh->count), sh->pd_idx, sh->check_state, | |
2742 | sh->reconstruct_state); | |
1da177e4 LT |
2743 | |
2744 | spin_lock(&sh->lock); | |
2745 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2746 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2747 | ||
a4456856 DW |
2748 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2749 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2750 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
def6ae26 | 2751 | |
83de75cc | 2752 | /* Now to look around and see what can be done */ |
9910f16a | 2753 | rcu_read_lock(); |
1da177e4 LT |
2754 | for (i=disks; i--; ) { |
2755 | mdk_rdev_t *rdev; | |
a4456856 | 2756 | struct r5dev *dev = &sh->dev[i]; |
1da177e4 | 2757 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 | 2758 | |
b5e98d65 DW |
2759 | pr_debug("check %d: state 0x%lx toread %p read %p write %p " |
2760 | "written %p\n", i, dev->flags, dev->toread, dev->read, | |
2761 | dev->towrite, dev->written); | |
2762 | ||
2763 | /* maybe we can request a biofill operation | |
2764 | * | |
2765 | * new wantfill requests are only permitted while | |
83de75cc | 2766 | * ops_complete_biofill is guaranteed to be inactive |
b5e98d65 DW |
2767 | */ |
2768 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
83de75cc | 2769 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) |
b5e98d65 | 2770 | set_bit(R5_Wantfill, &dev->flags); |
1da177e4 LT |
2771 | |
2772 | /* now count some things */ | |
a4456856 DW |
2773 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2774 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
f38e1219 | 2775 | if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++; |
1da177e4 | 2776 | |
b5e98d65 DW |
2777 | if (test_bit(R5_Wantfill, &dev->flags)) |
2778 | s.to_fill++; | |
2779 | else if (dev->toread) | |
a4456856 | 2780 | s.to_read++; |
1da177e4 | 2781 | if (dev->towrite) { |
a4456856 | 2782 | s.to_write++; |
1da177e4 | 2783 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2784 | s.non_overwrite++; |
1da177e4 | 2785 | } |
a4456856 DW |
2786 | if (dev->written) |
2787 | s.written++; | |
9910f16a | 2788 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
2789 | if (blocked_rdev == NULL && |
2790 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
2791 | blocked_rdev = rdev; |
2792 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 2793 | } |
b2d444d7 | 2794 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 2795 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
2796 | clear_bit(R5_ReadError, &dev->flags); |
2797 | clear_bit(R5_ReWrite, &dev->flags); | |
2798 | } | |
b2d444d7 | 2799 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 2800 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
2801 | s.failed++; |
2802 | s.failed_num = i; | |
1da177e4 LT |
2803 | } else |
2804 | set_bit(R5_Insync, &dev->flags); | |
2805 | } | |
9910f16a | 2806 | rcu_read_unlock(); |
b5e98d65 | 2807 | |
6bfe0b49 | 2808 | if (unlikely(blocked_rdev)) { |
ac4090d2 N |
2809 | if (s.syncing || s.expanding || s.expanded || |
2810 | s.to_write || s.written) { | |
2811 | set_bit(STRIPE_HANDLE, &sh->state); | |
2812 | goto unlock; | |
2813 | } | |
2814 | /* There is nothing for the blocked_rdev to block */ | |
2815 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
2816 | blocked_rdev = NULL; | |
6bfe0b49 DW |
2817 | } |
2818 | ||
83de75cc DW |
2819 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { |
2820 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
2821 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
2822 | } | |
b5e98d65 | 2823 | |
45b4233c | 2824 | pr_debug("locked=%d uptodate=%d to_read=%d" |
1da177e4 | 2825 | " to_write=%d failed=%d failed_num=%d\n", |
a4456856 DW |
2826 | s.locked, s.uptodate, s.to_read, s.to_write, |
2827 | s.failed, s.failed_num); | |
1da177e4 LT |
2828 | /* check if the array has lost two devices and, if so, some requests might |
2829 | * need to be failed | |
2830 | */ | |
a4456856 | 2831 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
1fe797e6 | 2832 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 2833 | if (s.failed > 1 && s.syncing) { |
1da177e4 LT |
2834 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2835 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2836 | s.syncing = 0; |
1da177e4 LT |
2837 | } |
2838 | ||
2839 | /* might be able to return some write requests if the parity block | |
2840 | * is safe, or on a failed drive | |
2841 | */ | |
2842 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
2843 | if ( s.written && |
2844 | ((test_bit(R5_Insync, &dev->flags) && | |
2845 | !test_bit(R5_LOCKED, &dev->flags) && | |
2846 | test_bit(R5_UPTODATE, &dev->flags)) || | |
2847 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
1fe797e6 | 2848 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
1da177e4 LT |
2849 | |
2850 | /* Now we might consider reading some blocks, either to check/generate | |
2851 | * parity, or to satisfy requests | |
2852 | * or to load a block that is being partially written. | |
2853 | */ | |
a4456856 | 2854 | if (s.to_read || s.non_overwrite || |
976ea8d4 | 2855 | (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding) |
1fe797e6 | 2856 | handle_stripe_fill5(sh, &s, disks); |
1da177e4 | 2857 | |
e33129d8 DW |
2858 | /* Now we check to see if any write operations have recently |
2859 | * completed | |
2860 | */ | |
e0a115e5 | 2861 | prexor = 0; |
d8ee0728 | 2862 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) |
e0a115e5 | 2863 | prexor = 1; |
d8ee0728 DW |
2864 | if (sh->reconstruct_state == reconstruct_state_drain_result || |
2865 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
600aa109 | 2866 | sh->reconstruct_state = reconstruct_state_idle; |
e33129d8 DW |
2867 | |
2868 | /* All the 'written' buffers and the parity block are ready to | |
2869 | * be written back to disk | |
2870 | */ | |
2871 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags)); | |
2872 | for (i = disks; i--; ) { | |
2873 | dev = &sh->dev[i]; | |
2874 | if (test_bit(R5_LOCKED, &dev->flags) && | |
2875 | (i == sh->pd_idx || dev->written)) { | |
2876 | pr_debug("Writing block %d\n", i); | |
2877 | set_bit(R5_Wantwrite, &dev->flags); | |
e0a115e5 DW |
2878 | if (prexor) |
2879 | continue; | |
e33129d8 DW |
2880 | if (!test_bit(R5_Insync, &dev->flags) || |
2881 | (i == sh->pd_idx && s.failed == 0)) | |
2882 | set_bit(STRIPE_INSYNC, &sh->state); | |
2883 | } | |
2884 | } | |
2885 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
2886 | atomic_dec(&conf->preread_active_stripes); | |
2887 | if (atomic_read(&conf->preread_active_stripes) < | |
2888 | IO_THRESHOLD) | |
2889 | md_wakeup_thread(conf->mddev->thread); | |
2890 | } | |
2891 | } | |
2892 | ||
2893 | /* Now to consider new write requests and what else, if anything | |
2894 | * should be read. We do not handle new writes when: | |
2895 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
2896 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
2897 | * block. | |
2898 | */ | |
600aa109 | 2899 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) |
1fe797e6 | 2900 | handle_stripe_dirtying5(conf, sh, &s, disks); |
1da177e4 LT |
2901 | |
2902 | /* maybe we need to check and possibly fix the parity for this stripe | |
e89f8962 DW |
2903 | * Any reads will already have been scheduled, so we just see if enough |
2904 | * data is available. The parity check is held off while parity | |
2905 | * dependent operations are in flight. | |
1da177e4 | 2906 | */ |
ecc65c9b DW |
2907 | if (sh->check_state || |
2908 | (s.syncing && s.locked == 0 && | |
976ea8d4 | 2909 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
ecc65c9b | 2910 | !test_bit(STRIPE_INSYNC, &sh->state))) |
a4456856 | 2911 | handle_parity_checks5(conf, sh, &s, disks); |
e89f8962 | 2912 | |
a4456856 | 2913 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
2914 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2915 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2916 | } | |
4e5314b5 N |
2917 | |
2918 | /* If the failed drive is just a ReadError, then we might need to progress | |
2919 | * the repair/check process | |
2920 | */ | |
a4456856 DW |
2921 | if (s.failed == 1 && !conf->mddev->ro && |
2922 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
2923 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
2924 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 2925 | ) { |
a4456856 | 2926 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
2927 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
2928 | set_bit(R5_Wantwrite, &dev->flags); | |
2929 | set_bit(R5_ReWrite, &dev->flags); | |
2930 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2931 | s.locked++; |
4e5314b5 N |
2932 | } else { |
2933 | /* let's read it back */ | |
2934 | set_bit(R5_Wantread, &dev->flags); | |
2935 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2936 | s.locked++; |
4e5314b5 N |
2937 | } |
2938 | } | |
2939 | ||
600aa109 DW |
2940 | /* Finish reconstruct operations initiated by the expansion process */ |
2941 | if (sh->reconstruct_state == reconstruct_state_result) { | |
2942 | sh->reconstruct_state = reconstruct_state_idle; | |
f0a50d37 | 2943 | clear_bit(STRIPE_EXPANDING, &sh->state); |
23397883 | 2944 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 2945 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
23397883 | 2946 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
efe31143 | 2947 | s.locked++; |
23397883 | 2948 | } |
f0a50d37 DW |
2949 | } |
2950 | ||
2951 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && | |
600aa109 | 2952 | !sh->reconstruct_state) { |
f0a50d37 DW |
2953 | /* Need to write out all blocks after computing parity */ |
2954 | sh->disks = conf->raid_disks; | |
911d4ee8 | 2955 | stripe_set_idx(sh->sector, conf, 0, sh); |
1fe797e6 | 2956 | schedule_reconstruction5(sh, &s, 1, 1); |
600aa109 | 2957 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { |
ccfcc3c1 | 2958 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 2959 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
2960 | wake_up(&conf->wait_for_overlap); |
2961 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
2962 | } | |
2963 | ||
0f94e87c | 2964 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 2965 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 2966 | handle_stripe_expansion(conf, sh, NULL); |
ccfcc3c1 | 2967 | |
6bfe0b49 | 2968 | unlock: |
1da177e4 LT |
2969 | spin_unlock(&sh->lock); |
2970 | ||
6bfe0b49 DW |
2971 | /* wait for this device to become unblocked */ |
2972 | if (unlikely(blocked_rdev)) | |
2973 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
2974 | ||
600aa109 DW |
2975 | if (s.ops_request) |
2976 | raid5_run_ops(sh, s.ops_request); | |
d84e0f10 | 2977 | |
c4e5ac0a | 2978 | ops_run_io(sh, &s); |
1da177e4 | 2979 | |
a4456856 | 2980 | return_io(return_bi); |
df10cfbc DW |
2981 | |
2982 | return blocked_rdev == NULL; | |
1da177e4 LT |
2983 | } |
2984 | ||
df10cfbc | 2985 | static bool handle_stripe6(struct stripe_head *sh, struct page *tmp_page) |
1da177e4 | 2986 | { |
bff61975 | 2987 | raid5_conf_t *conf = sh->raid_conf; |
f416885e | 2988 | int disks = sh->disks; |
a4456856 DW |
2989 | struct bio *return_bi = NULL; |
2990 | int i, pd_idx = sh->pd_idx; | |
2991 | struct stripe_head_state s; | |
2992 | struct r6_state r6s; | |
16a53ecc | 2993 | struct r5dev *dev, *pdev, *qdev; |
6bfe0b49 | 2994 | mdk_rdev_t *blocked_rdev = NULL; |
1da177e4 | 2995 | |
d0dabf7e | 2996 | r6s.qd_idx = sh->qd_idx; |
45b4233c | 2997 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " |
a4456856 DW |
2998 | "pd_idx=%d, qd_idx=%d\n", |
2999 | (unsigned long long)sh->sector, sh->state, | |
3000 | atomic_read(&sh->count), pd_idx, r6s.qd_idx); | |
3001 | memset(&s, 0, sizeof(s)); | |
72626685 | 3002 | |
16a53ecc N |
3003 | spin_lock(&sh->lock); |
3004 | clear_bit(STRIPE_HANDLE, &sh->state); | |
3005 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3006 | ||
a4456856 DW |
3007 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
3008 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3009 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 3010 | /* Now to look around and see what can be done */ |
1da177e4 LT |
3011 | |
3012 | rcu_read_lock(); | |
16a53ecc N |
3013 | for (i=disks; i--; ) { |
3014 | mdk_rdev_t *rdev; | |
3015 | dev = &sh->dev[i]; | |
3016 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 3017 | |
45b4233c | 3018 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
16a53ecc N |
3019 | i, dev->flags, dev->toread, dev->towrite, dev->written); |
3020 | /* maybe we can reply to a read */ | |
3021 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
3022 | struct bio *rbi, *rbi2; | |
45b4233c | 3023 | pr_debug("Return read for disc %d\n", i); |
16a53ecc N |
3024 | spin_lock_irq(&conf->device_lock); |
3025 | rbi = dev->toread; | |
3026 | dev->toread = NULL; | |
3027 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
3028 | wake_up(&conf->wait_for_overlap); | |
3029 | spin_unlock_irq(&conf->device_lock); | |
3030 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
3031 | copy_data(0, rbi, dev->page, dev->sector); | |
3032 | rbi2 = r5_next_bio(rbi, dev->sector); | |
3033 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3034 | if (!raid5_dec_bi_phys_segments(rbi)) { |
16a53ecc N |
3035 | rbi->bi_next = return_bi; |
3036 | return_bi = rbi; | |
3037 | } | |
3038 | spin_unlock_irq(&conf->device_lock); | |
3039 | rbi = rbi2; | |
3040 | } | |
3041 | } | |
1da177e4 | 3042 | |
16a53ecc | 3043 | /* now count some things */ |
a4456856 DW |
3044 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
3045 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 3046 | |
16a53ecc | 3047 | |
a4456856 DW |
3048 | if (dev->toread) |
3049 | s.to_read++; | |
16a53ecc | 3050 | if (dev->towrite) { |
a4456856 | 3051 | s.to_write++; |
16a53ecc | 3052 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 3053 | s.non_overwrite++; |
16a53ecc | 3054 | } |
a4456856 DW |
3055 | if (dev->written) |
3056 | s.written++; | |
16a53ecc | 3057 | rdev = rcu_dereference(conf->disks[i].rdev); |
ac4090d2 N |
3058 | if (blocked_rdev == NULL && |
3059 | rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
6bfe0b49 DW |
3060 | blocked_rdev = rdev; |
3061 | atomic_inc(&rdev->nr_pending); | |
6bfe0b49 | 3062 | } |
16a53ecc N |
3063 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
3064 | /* The ReadError flag will just be confusing now */ | |
3065 | clear_bit(R5_ReadError, &dev->flags); | |
3066 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3067 | } |
16a53ecc N |
3068 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
3069 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
3070 | if (s.failed < 2) |
3071 | r6s.failed_num[s.failed] = i; | |
3072 | s.failed++; | |
16a53ecc N |
3073 | } else |
3074 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
3075 | } |
3076 | rcu_read_unlock(); | |
6bfe0b49 DW |
3077 | |
3078 | if (unlikely(blocked_rdev)) { | |
ac4090d2 N |
3079 | if (s.syncing || s.expanding || s.expanded || |
3080 | s.to_write || s.written) { | |
3081 | set_bit(STRIPE_HANDLE, &sh->state); | |
3082 | goto unlock; | |
3083 | } | |
3084 | /* There is nothing for the blocked_rdev to block */ | |
3085 | rdev_dec_pending(blocked_rdev, conf->mddev); | |
3086 | blocked_rdev = NULL; | |
6bfe0b49 | 3087 | } |
ac4090d2 | 3088 | |
45b4233c | 3089 | pr_debug("locked=%d uptodate=%d to_read=%d" |
16a53ecc | 3090 | " to_write=%d failed=%d failed_num=%d,%d\n", |
a4456856 DW |
3091 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
3092 | r6s.failed_num[0], r6s.failed_num[1]); | |
3093 | /* check if the array has lost >2 devices and, if so, some requests | |
3094 | * might need to be failed | |
16a53ecc | 3095 | */ |
a4456856 | 3096 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
1fe797e6 | 3097 | handle_failed_stripe(conf, sh, &s, disks, &return_bi); |
a4456856 | 3098 | if (s.failed > 2 && s.syncing) { |
16a53ecc N |
3099 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
3100 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 3101 | s.syncing = 0; |
16a53ecc N |
3102 | } |
3103 | ||
3104 | /* | |
3105 | * might be able to return some write requests if the parity blocks | |
3106 | * are safe, or on a failed drive | |
3107 | */ | |
3108 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
3109 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
3110 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
3111 | qdev = &sh->dev[r6s.qd_idx]; | |
3112 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == r6s.qd_idx) | |
3113 | || (s.failed >= 2 && r6s.failed_num[1] == r6s.qd_idx); | |
3114 | ||
3115 | if ( s.written && | |
3116 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 3117 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
3118 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
3119 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 3120 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 | 3121 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
1fe797e6 | 3122 | handle_stripe_clean_event(conf, sh, disks, &return_bi); |
16a53ecc N |
3123 | |
3124 | /* Now we might consider reading some blocks, either to check/generate | |
3125 | * parity, or to satisfy requests | |
3126 | * or to load a block that is being partially written. | |
3127 | */ | |
a4456856 DW |
3128 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
3129 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
1fe797e6 | 3130 | handle_stripe_fill6(sh, &s, &r6s, disks); |
16a53ecc N |
3131 | |
3132 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 | 3133 | if (s.to_write) |
1fe797e6 | 3134 | handle_stripe_dirtying6(conf, sh, &s, &r6s, disks); |
16a53ecc N |
3135 | |
3136 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 DW |
3137 | * Any reads will already have been scheduled, so we just see if enough |
3138 | * data is available | |
16a53ecc | 3139 | */ |
a4456856 DW |
3140 | if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state)) |
3141 | handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks); | |
16a53ecc | 3142 | |
a4456856 | 3143 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
3144 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
3145 | clear_bit(STRIPE_SYNCING, &sh->state); | |
3146 | } | |
3147 | ||
3148 | /* If the failed drives are just a ReadError, then we might need | |
3149 | * to progress the repair/check process | |
3150 | */ | |
a4456856 DW |
3151 | if (s.failed <= 2 && !conf->mddev->ro) |
3152 | for (i = 0; i < s.failed; i++) { | |
3153 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
3154 | if (test_bit(R5_ReadError, &dev->flags) |
3155 | && !test_bit(R5_LOCKED, &dev->flags) | |
3156 | && test_bit(R5_UPTODATE, &dev->flags) | |
3157 | ) { | |
3158 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
3159 | set_bit(R5_Wantwrite, &dev->flags); | |
3160 | set_bit(R5_ReWrite, &dev->flags); | |
3161 | set_bit(R5_LOCKED, &dev->flags); | |
3162 | } else { | |
3163 | /* let's read it back */ | |
3164 | set_bit(R5_Wantread, &dev->flags); | |
3165 | set_bit(R5_LOCKED, &dev->flags); | |
3166 | } | |
3167 | } | |
3168 | } | |
f416885e | 3169 | |
a4456856 | 3170 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
f416885e N |
3171 | /* Need to write out all blocks after computing P&Q */ |
3172 | sh->disks = conf->raid_disks; | |
911d4ee8 | 3173 | stripe_set_idx(sh->sector, conf, 0, sh); |
f416885e N |
3174 | compute_parity6(sh, RECONSTRUCT_WRITE); |
3175 | for (i = conf->raid_disks ; i-- ; ) { | |
3176 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3177 | s.locked++; |
f416885e N |
3178 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
3179 | } | |
3180 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 3181 | } else if (s.expanded) { |
f416885e N |
3182 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
3183 | atomic_dec(&conf->reshape_stripes); | |
3184 | wake_up(&conf->wait_for_overlap); | |
3185 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
3186 | } | |
3187 | ||
0f94e87c | 3188 | if (s.expanding && s.locked == 0 && |
976ea8d4 | 3189 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) |
a4456856 | 3190 | handle_stripe_expansion(conf, sh, &r6s); |
f416885e | 3191 | |
6bfe0b49 | 3192 | unlock: |
16a53ecc N |
3193 | spin_unlock(&sh->lock); |
3194 | ||
6bfe0b49 DW |
3195 | /* wait for this device to become unblocked */ |
3196 | if (unlikely(blocked_rdev)) | |
3197 | md_wait_for_blocked_rdev(blocked_rdev, conf->mddev); | |
3198 | ||
f0e43bcd | 3199 | ops_run_io(sh, &s); |
16a53ecc | 3200 | |
f0e43bcd | 3201 | return_io(return_bi); |
df10cfbc DW |
3202 | |
3203 | return blocked_rdev == NULL; | |
16a53ecc N |
3204 | } |
3205 | ||
df10cfbc DW |
3206 | /* returns true if the stripe was handled */ |
3207 | static bool handle_stripe(struct stripe_head *sh, struct page *tmp_page) | |
16a53ecc N |
3208 | { |
3209 | if (sh->raid_conf->level == 6) | |
df10cfbc | 3210 | return handle_stripe6(sh, tmp_page); |
16a53ecc | 3211 | else |
df10cfbc | 3212 | return handle_stripe5(sh); |
16a53ecc N |
3213 | } |
3214 | ||
3215 | ||
3216 | ||
3217 | static void raid5_activate_delayed(raid5_conf_t *conf) | |
3218 | { | |
3219 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
3220 | while (!list_empty(&conf->delayed_list)) { | |
3221 | struct list_head *l = conf->delayed_list.next; | |
3222 | struct stripe_head *sh; | |
3223 | sh = list_entry(l, struct stripe_head, lru); | |
3224 | list_del_init(l); | |
3225 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3226 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3227 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 3228 | list_add_tail(&sh->lru, &conf->hold_list); |
16a53ecc | 3229 | } |
6ed3003c N |
3230 | } else |
3231 | blk_plug_device(conf->mddev->queue); | |
16a53ecc N |
3232 | } |
3233 | ||
3234 | static void activate_bit_delay(raid5_conf_t *conf) | |
3235 | { | |
3236 | /* device_lock is held */ | |
3237 | struct list_head head; | |
3238 | list_add(&head, &conf->bitmap_list); | |
3239 | list_del_init(&conf->bitmap_list); | |
3240 | while (!list_empty(&head)) { | |
3241 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
3242 | list_del_init(&sh->lru); | |
3243 | atomic_inc(&sh->count); | |
3244 | __release_stripe(conf, sh); | |
3245 | } | |
3246 | } | |
3247 | ||
3248 | static void unplug_slaves(mddev_t *mddev) | |
3249 | { | |
3250 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3251 | int i; | |
3252 | ||
3253 | rcu_read_lock(); | |
3254 | for (i=0; i<mddev->raid_disks; i++) { | |
3255 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
3256 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
165125e1 | 3257 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
16a53ecc N |
3258 | |
3259 | atomic_inc(&rdev->nr_pending); | |
3260 | rcu_read_unlock(); | |
3261 | ||
2ad8b1ef | 3262 | blk_unplug(r_queue); |
16a53ecc N |
3263 | |
3264 | rdev_dec_pending(rdev, mddev); | |
3265 | rcu_read_lock(); | |
3266 | } | |
3267 | } | |
3268 | rcu_read_unlock(); | |
3269 | } | |
3270 | ||
165125e1 | 3271 | static void raid5_unplug_device(struct request_queue *q) |
16a53ecc N |
3272 | { |
3273 | mddev_t *mddev = q->queuedata; | |
3274 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3275 | unsigned long flags; | |
3276 | ||
3277 | spin_lock_irqsave(&conf->device_lock, flags); | |
3278 | ||
3279 | if (blk_remove_plug(q)) { | |
3280 | conf->seq_flush++; | |
3281 | raid5_activate_delayed(conf); | |
72626685 | 3282 | } |
1da177e4 LT |
3283 | md_wakeup_thread(mddev->thread); |
3284 | ||
3285 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3286 | ||
3287 | unplug_slaves(mddev); | |
3288 | } | |
3289 | ||
f022b2fd N |
3290 | static int raid5_congested(void *data, int bits) |
3291 | { | |
3292 | mddev_t *mddev = data; | |
3293 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3294 | ||
3295 | /* No difference between reads and writes. Just check | |
3296 | * how busy the stripe_cache is | |
3297 | */ | |
3298 | if (conf->inactive_blocked) | |
3299 | return 1; | |
3300 | if (conf->quiesce) | |
3301 | return 1; | |
3302 | if (list_empty_careful(&conf->inactive_list)) | |
3303 | return 1; | |
3304 | ||
3305 | return 0; | |
3306 | } | |
3307 | ||
23032a0e RBJ |
3308 | /* We want read requests to align with chunks where possible, |
3309 | * but write requests don't need to. | |
3310 | */ | |
cc371e66 AK |
3311 | static int raid5_mergeable_bvec(struct request_queue *q, |
3312 | struct bvec_merge_data *bvm, | |
3313 | struct bio_vec *biovec) | |
23032a0e RBJ |
3314 | { |
3315 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 3316 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e RBJ |
3317 | int max; |
3318 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
cc371e66 | 3319 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 3320 | |
cc371e66 | 3321 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
3322 | return biovec->bv_len; /* always allow writes to be mergeable */ |
3323 | ||
3324 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
3325 | if (max < 0) max = 0; | |
3326 | if (max <= biovec->bv_len && bio_sectors == 0) | |
3327 | return biovec->bv_len; | |
3328 | else | |
3329 | return max; | |
3330 | } | |
3331 | ||
f679623f RBJ |
3332 | |
3333 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
3334 | { | |
3335 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
3336 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
3337 | unsigned int bio_sectors = bio->bi_size >> 9; | |
3338 | ||
3339 | return chunk_sectors >= | |
3340 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
3341 | } | |
3342 | ||
46031f9a RBJ |
3343 | /* |
3344 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
3345 | * later sampled by raid5d. | |
3346 | */ | |
3347 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
3348 | { | |
3349 | unsigned long flags; | |
3350 | ||
3351 | spin_lock_irqsave(&conf->device_lock, flags); | |
3352 | ||
3353 | bi->bi_next = conf->retry_read_aligned_list; | |
3354 | conf->retry_read_aligned_list = bi; | |
3355 | ||
3356 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3357 | md_wakeup_thread(conf->mddev->thread); | |
3358 | } | |
3359 | ||
3360 | ||
3361 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
3362 | { | |
3363 | struct bio *bi; | |
3364 | ||
3365 | bi = conf->retry_read_aligned; | |
3366 | if (bi) { | |
3367 | conf->retry_read_aligned = NULL; | |
3368 | return bi; | |
3369 | } | |
3370 | bi = conf->retry_read_aligned_list; | |
3371 | if(bi) { | |
387bb173 | 3372 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 3373 | bi->bi_next = NULL; |
960e739d JA |
3374 | /* |
3375 | * this sets the active strip count to 1 and the processed | |
3376 | * strip count to zero (upper 8 bits) | |
3377 | */ | |
46031f9a | 3378 | bi->bi_phys_segments = 1; /* biased count of active stripes */ |
46031f9a RBJ |
3379 | } |
3380 | ||
3381 | return bi; | |
3382 | } | |
3383 | ||
3384 | ||
f679623f RBJ |
3385 | /* |
3386 | * The "raid5_align_endio" should check if the read succeeded and if it | |
3387 | * did, call bio_endio on the original bio (having bio_put the new bio | |
3388 | * first). | |
3389 | * If the read failed.. | |
3390 | */ | |
6712ecf8 | 3391 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
3392 | { |
3393 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
3394 | mddev_t *mddev; |
3395 | raid5_conf_t *conf; | |
3396 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
3397 | mdk_rdev_t *rdev; | |
3398 | ||
f679623f | 3399 | bio_put(bi); |
46031f9a RBJ |
3400 | |
3401 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
3402 | conf = mddev_to_conf(mddev); | |
3403 | rdev = (void*)raid_bi->bi_next; | |
3404 | raid_bi->bi_next = NULL; | |
3405 | ||
3406 | rdev_dec_pending(rdev, conf->mddev); | |
3407 | ||
3408 | if (!error && uptodate) { | |
6712ecf8 | 3409 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
3410 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3411 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 3412 | return; |
46031f9a RBJ |
3413 | } |
3414 | ||
3415 | ||
45b4233c | 3416 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
3417 | |
3418 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
3419 | } |
3420 | ||
387bb173 NB |
3421 | static int bio_fits_rdev(struct bio *bi) |
3422 | { | |
165125e1 | 3423 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 NB |
3424 | |
3425 | if ((bi->bi_size>>9) > q->max_sectors) | |
3426 | return 0; | |
3427 | blk_recount_segments(q, bi); | |
960e739d | 3428 | if (bi->bi_phys_segments > q->max_phys_segments) |
387bb173 NB |
3429 | return 0; |
3430 | ||
3431 | if (q->merge_bvec_fn) | |
3432 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
3433 | * just just give up | |
3434 | */ | |
3435 | return 0; | |
3436 | ||
3437 | return 1; | |
3438 | } | |
3439 | ||
3440 | ||
165125e1 | 3441 | static int chunk_aligned_read(struct request_queue *q, struct bio * raid_bio) |
f679623f RBJ |
3442 | { |
3443 | mddev_t *mddev = q->queuedata; | |
3444 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3445 | unsigned int dd_idx; |
f679623f RBJ |
3446 | struct bio* align_bi; |
3447 | mdk_rdev_t *rdev; | |
3448 | ||
3449 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 3450 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
3451 | return 0; |
3452 | } | |
3453 | /* | |
99c0fb5f | 3454 | * use bio_clone to make a copy of the bio |
f679623f RBJ |
3455 | */ |
3456 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
3457 | if (!align_bi) | |
3458 | return 0; | |
3459 | /* | |
3460 | * set bi_end_io to a new function, and set bi_private to the | |
3461 | * original bio. | |
3462 | */ | |
3463 | align_bi->bi_end_io = raid5_align_endio; | |
3464 | align_bi->bi_private = raid_bio; | |
3465 | /* | |
3466 | * compute position | |
3467 | */ | |
112bf897 N |
3468 | align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector, |
3469 | 0, | |
911d4ee8 | 3470 | &dd_idx, NULL); |
f679623f RBJ |
3471 | |
3472 | rcu_read_lock(); | |
3473 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
3474 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
3475 | atomic_inc(&rdev->nr_pending); |
3476 | rcu_read_unlock(); | |
46031f9a RBJ |
3477 | raid_bio->bi_next = (void*)rdev; |
3478 | align_bi->bi_bdev = rdev->bdev; | |
3479 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3480 | align_bi->bi_sector += rdev->data_offset; | |
3481 | ||
387bb173 NB |
3482 | if (!bio_fits_rdev(align_bi)) { |
3483 | /* too big in some way */ | |
3484 | bio_put(align_bi); | |
3485 | rdev_dec_pending(rdev, mddev); | |
3486 | return 0; | |
3487 | } | |
3488 | ||
46031f9a RBJ |
3489 | spin_lock_irq(&conf->device_lock); |
3490 | wait_event_lock_irq(conf->wait_for_stripe, | |
3491 | conf->quiesce == 0, | |
3492 | conf->device_lock, /* nothing */); | |
3493 | atomic_inc(&conf->active_aligned_reads); | |
3494 | spin_unlock_irq(&conf->device_lock); | |
3495 | ||
f679623f RBJ |
3496 | generic_make_request(align_bi); |
3497 | return 1; | |
3498 | } else { | |
3499 | rcu_read_unlock(); | |
46031f9a | 3500 | bio_put(align_bi); |
f679623f RBJ |
3501 | return 0; |
3502 | } | |
3503 | } | |
3504 | ||
8b3e6cdc DW |
3505 | /* __get_priority_stripe - get the next stripe to process |
3506 | * | |
3507 | * Full stripe writes are allowed to pass preread active stripes up until | |
3508 | * the bypass_threshold is exceeded. In general the bypass_count | |
3509 | * increments when the handle_list is handled before the hold_list; however, it | |
3510 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
3511 | * stripe with in flight i/o. The bypass_count will be reset when the | |
3512 | * head of the hold_list has changed, i.e. the head was promoted to the | |
3513 | * handle_list. | |
3514 | */ | |
3515 | static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf) | |
3516 | { | |
3517 | struct stripe_head *sh; | |
3518 | ||
3519 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
3520 | __func__, | |
3521 | list_empty(&conf->handle_list) ? "empty" : "busy", | |
3522 | list_empty(&conf->hold_list) ? "empty" : "busy", | |
3523 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
3524 | ||
3525 | if (!list_empty(&conf->handle_list)) { | |
3526 | sh = list_entry(conf->handle_list.next, typeof(*sh), lru); | |
3527 | ||
3528 | if (list_empty(&conf->hold_list)) | |
3529 | conf->bypass_count = 0; | |
3530 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
3531 | if (conf->hold_list.next == conf->last_hold) | |
3532 | conf->bypass_count++; | |
3533 | else { | |
3534 | conf->last_hold = conf->hold_list.next; | |
3535 | conf->bypass_count -= conf->bypass_threshold; | |
3536 | if (conf->bypass_count < 0) | |
3537 | conf->bypass_count = 0; | |
3538 | } | |
3539 | } | |
3540 | } else if (!list_empty(&conf->hold_list) && | |
3541 | ((conf->bypass_threshold && | |
3542 | conf->bypass_count > conf->bypass_threshold) || | |
3543 | atomic_read(&conf->pending_full_writes) == 0)) { | |
3544 | sh = list_entry(conf->hold_list.next, | |
3545 | typeof(*sh), lru); | |
3546 | conf->bypass_count -= conf->bypass_threshold; | |
3547 | if (conf->bypass_count < 0) | |
3548 | conf->bypass_count = 0; | |
3549 | } else | |
3550 | return NULL; | |
3551 | ||
3552 | list_del_init(&sh->lru); | |
3553 | atomic_inc(&sh->count); | |
3554 | BUG_ON(atomic_read(&sh->count) != 1); | |
3555 | return sh; | |
3556 | } | |
f679623f | 3557 | |
165125e1 | 3558 | static int make_request(struct request_queue *q, struct bio * bi) |
1da177e4 LT |
3559 | { |
3560 | mddev_t *mddev = q->queuedata; | |
3561 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
911d4ee8 | 3562 | int dd_idx; |
1da177e4 LT |
3563 | sector_t new_sector; |
3564 | sector_t logical_sector, last_sector; | |
3565 | struct stripe_head *sh; | |
a362357b | 3566 | const int rw = bio_data_dir(bi); |
c9959059 | 3567 | int cpu, remaining; |
1da177e4 | 3568 | |
e5dcdd80 | 3569 | if (unlikely(bio_barrier(bi))) { |
6712ecf8 | 3570 | bio_endio(bi, -EOPNOTSUPP); |
e5dcdd80 N |
3571 | return 0; |
3572 | } | |
3573 | ||
3d310eb7 | 3574 | md_write_start(mddev, bi); |
06d91a5f | 3575 | |
074a7aca TH |
3576 | cpu = part_stat_lock(); |
3577 | part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]); | |
3578 | part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], | |
3579 | bio_sectors(bi)); | |
3580 | part_stat_unlock(); | |
1da177e4 | 3581 | |
802ba064 | 3582 | if (rw == READ && |
52488615 RBJ |
3583 | mddev->reshape_position == MaxSector && |
3584 | chunk_aligned_read(q,bi)) | |
99c0fb5f | 3585 | return 0; |
52488615 | 3586 | |
1da177e4 LT |
3587 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
3588 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
3589 | bi->bi_next = NULL; | |
3590 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 3591 | |
1da177e4 LT |
3592 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
3593 | DEFINE_WAIT(w); | |
16a53ecc | 3594 | int disks, data_disks; |
b5663ba4 | 3595 | int previous; |
b578d55f | 3596 | |
7ecaa1e6 | 3597 | retry: |
b5663ba4 | 3598 | previous = 0; |
b578d55f | 3599 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
7ecaa1e6 N |
3600 | if (likely(conf->expand_progress == MaxSector)) |
3601 | disks = conf->raid_disks; | |
3602 | else { | |
df8e7f76 N |
3603 | /* spinlock is needed as expand_progress may be |
3604 | * 64bit on a 32bit platform, and so it might be | |
3605 | * possible to see a half-updated value | |
3606 | * Ofcourse expand_progress could change after | |
3607 | * the lock is dropped, so once we get a reference | |
3608 | * to the stripe that we think it is, we will have | |
3609 | * to check again. | |
3610 | */ | |
7ecaa1e6 N |
3611 | spin_lock_irq(&conf->device_lock); |
3612 | disks = conf->raid_disks; | |
b5663ba4 | 3613 | if (logical_sector >= conf->expand_progress) { |
7ecaa1e6 | 3614 | disks = conf->previous_raid_disks; |
b5663ba4 N |
3615 | previous = 1; |
3616 | } else { | |
b578d55f N |
3617 | if (logical_sector >= conf->expand_lo) { |
3618 | spin_unlock_irq(&conf->device_lock); | |
3619 | schedule(); | |
3620 | goto retry; | |
3621 | } | |
3622 | } | |
7ecaa1e6 N |
3623 | spin_unlock_irq(&conf->device_lock); |
3624 | } | |
16a53ecc N |
3625 | data_disks = disks - conf->max_degraded; |
3626 | ||
112bf897 N |
3627 | new_sector = raid5_compute_sector(conf, logical_sector, |
3628 | previous, | |
911d4ee8 | 3629 | &dd_idx, NULL); |
45b4233c | 3630 | pr_debug("raid5: make_request, sector %llu logical %llu\n", |
1da177e4 LT |
3631 | (unsigned long long)new_sector, |
3632 | (unsigned long long)logical_sector); | |
3633 | ||
b5663ba4 N |
3634 | sh = get_active_stripe(conf, new_sector, previous, |
3635 | (bi->bi_rw&RWA_MASK)); | |
1da177e4 | 3636 | if (sh) { |
7ecaa1e6 N |
3637 | if (unlikely(conf->expand_progress != MaxSector)) { |
3638 | /* expansion might have moved on while waiting for a | |
df8e7f76 N |
3639 | * stripe, so we must do the range check again. |
3640 | * Expansion could still move past after this | |
3641 | * test, but as we are holding a reference to | |
3642 | * 'sh', we know that if that happens, | |
3643 | * STRIPE_EXPANDING will get set and the expansion | |
3644 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
3645 | */ |
3646 | int must_retry = 0; | |
3647 | spin_lock_irq(&conf->device_lock); | |
3648 | if (logical_sector < conf->expand_progress && | |
3649 | disks == conf->previous_raid_disks) | |
3650 | /* mismatch, need to try again */ | |
3651 | must_retry = 1; | |
3652 | spin_unlock_irq(&conf->device_lock); | |
3653 | if (must_retry) { | |
3654 | release_stripe(sh); | |
3655 | goto retry; | |
3656 | } | |
3657 | } | |
e464eafd N |
3658 | /* FIXME what if we get a false positive because these |
3659 | * are being updated. | |
3660 | */ | |
3661 | if (logical_sector >= mddev->suspend_lo && | |
3662 | logical_sector < mddev->suspend_hi) { | |
3663 | release_stripe(sh); | |
3664 | schedule(); | |
3665 | goto retry; | |
3666 | } | |
7ecaa1e6 N |
3667 | |
3668 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
3669 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
3670 | /* Stripe is busy expanding or | |
3671 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
3672 | * and wait a while |
3673 | */ | |
3674 | raid5_unplug_device(mddev->queue); | |
3675 | release_stripe(sh); | |
3676 | schedule(); | |
3677 | goto retry; | |
3678 | } | |
3679 | finish_wait(&conf->wait_for_overlap, &w); | |
6ed3003c N |
3680 | set_bit(STRIPE_HANDLE, &sh->state); |
3681 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1da177e4 | 3682 | release_stripe(sh); |
1da177e4 LT |
3683 | } else { |
3684 | /* cannot get stripe for read-ahead, just give-up */ | |
3685 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
3686 | finish_wait(&conf->wait_for_overlap, &w); | |
3687 | break; | |
3688 | } | |
3689 | ||
3690 | } | |
3691 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3692 | remaining = raid5_dec_bi_phys_segments(bi); |
f6344757 N |
3693 | spin_unlock_irq(&conf->device_lock); |
3694 | if (remaining == 0) { | |
1da177e4 | 3695 | |
16a53ecc | 3696 | if ( rw == WRITE ) |
1da177e4 | 3697 | md_write_end(mddev); |
6712ecf8 | 3698 | |
0e13fe23 | 3699 | bio_endio(bi, 0); |
1da177e4 | 3700 | } |
1da177e4 LT |
3701 | return 0; |
3702 | } | |
3703 | ||
52c03291 | 3704 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 3705 | { |
52c03291 N |
3706 | /* reshaping is quite different to recovery/resync so it is |
3707 | * handled quite separately ... here. | |
3708 | * | |
3709 | * On each call to sync_request, we gather one chunk worth of | |
3710 | * destination stripes and flag them as expanding. | |
3711 | * Then we find all the source stripes and request reads. | |
3712 | * As the reads complete, handle_stripe will copy the data | |
3713 | * into the destination stripe and release that stripe. | |
3714 | */ | |
1da177e4 LT |
3715 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
3716 | struct stripe_head *sh; | |
ccfcc3c1 | 3717 | sector_t first_sector, last_sector; |
f416885e N |
3718 | int raid_disks = conf->previous_raid_disks; |
3719 | int data_disks = raid_disks - conf->max_degraded; | |
3720 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
3721 | int i; |
3722 | int dd_idx; | |
3723 | sector_t writepos, safepos, gap; | |
3724 | ||
3725 | if (sector_nr == 0 && | |
3726 | conf->expand_progress != 0) { | |
3727 | /* restarting in the middle, skip the initial sectors */ | |
3728 | sector_nr = conf->expand_progress; | |
f416885e | 3729 | sector_div(sector_nr, new_data_disks); |
52c03291 N |
3730 | *skipped = 1; |
3731 | return sector_nr; | |
3732 | } | |
3733 | ||
3734 | /* we update the metadata when there is more than 3Meg | |
3735 | * in the block range (that is rather arbitrary, should | |
3736 | * probably be time based) or when the data about to be | |
3737 | * copied would over-write the source of the data at | |
3738 | * the front of the range. | |
3739 | * i.e. one new_stripe forward from expand_progress new_maps | |
3740 | * to after where expand_lo old_maps to | |
3741 | */ | |
3742 | writepos = conf->expand_progress + | |
f416885e N |
3743 | conf->chunk_size/512*(new_data_disks); |
3744 | sector_div(writepos, new_data_disks); | |
52c03291 | 3745 | safepos = conf->expand_lo; |
f416885e | 3746 | sector_div(safepos, data_disks); |
52c03291 N |
3747 | gap = conf->expand_progress - conf->expand_lo; |
3748 | ||
3749 | if (writepos >= safepos || | |
f416885e | 3750 | gap > (new_data_disks)*3000*2 /*3Meg*/) { |
52c03291 N |
3751 | /* Cannot proceed until we've updated the superblock... */ |
3752 | wait_event(conf->wait_for_overlap, | |
3753 | atomic_read(&conf->reshape_stripes)==0); | |
3754 | mddev->reshape_position = conf->expand_progress; | |
850b2b42 | 3755 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 3756 | md_wakeup_thread(mddev->thread); |
850b2b42 | 3757 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
3758 | kthread_should_stop()); |
3759 | spin_lock_irq(&conf->device_lock); | |
3760 | conf->expand_lo = mddev->reshape_position; | |
3761 | spin_unlock_irq(&conf->device_lock); | |
3762 | wake_up(&conf->wait_for_overlap); | |
3763 | } | |
3764 | ||
3765 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { | |
3766 | int j; | |
3767 | int skipped = 0; | |
b5663ba4 | 3768 | sh = get_active_stripe(conf, sector_nr+i, 0, 0); |
52c03291 N |
3769 | set_bit(STRIPE_EXPANDING, &sh->state); |
3770 | atomic_inc(&conf->reshape_stripes); | |
3771 | /* If any of this stripe is beyond the end of the old | |
3772 | * array, then we need to zero those blocks | |
3773 | */ | |
3774 | for (j=sh->disks; j--;) { | |
3775 | sector_t s; | |
3776 | if (j == sh->pd_idx) | |
3777 | continue; | |
f416885e | 3778 | if (conf->level == 6 && |
d0dabf7e | 3779 | j == sh->qd_idx) |
f416885e | 3780 | continue; |
52c03291 | 3781 | s = compute_blocknr(sh, j); |
f233ea5c | 3782 | if (s < mddev->array_sectors) { |
52c03291 N |
3783 | skipped = 1; |
3784 | continue; | |
3785 | } | |
3786 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
3787 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
3788 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
3789 | } | |
3790 | if (!skipped) { | |
3791 | set_bit(STRIPE_EXPAND_READY, &sh->state); | |
3792 | set_bit(STRIPE_HANDLE, &sh->state); | |
3793 | } | |
3794 | release_stripe(sh); | |
3795 | } | |
3796 | spin_lock_irq(&conf->device_lock); | |
6d3baf2e | 3797 | conf->expand_progress = (sector_nr + i) * new_data_disks; |
52c03291 N |
3798 | spin_unlock_irq(&conf->device_lock); |
3799 | /* Ok, those stripe are ready. We can start scheduling | |
3800 | * reads on the source stripes. | |
3801 | * The source stripes are determined by mapping the first and last | |
3802 | * block on the destination stripes. | |
3803 | */ | |
52c03291 | 3804 | first_sector = |
112bf897 | 3805 | raid5_compute_sector(conf, sector_nr*(new_data_disks), |
911d4ee8 | 3806 | 1, &dd_idx, NULL); |
52c03291 | 3807 | last_sector = |
112bf897 N |
3808 | raid5_compute_sector(conf, ((sector_nr+conf->chunk_size/512) |
3809 | *(new_data_disks) - 1), | |
911d4ee8 | 3810 | 1, &dd_idx, NULL); |
58c0fed4 AN |
3811 | if (last_sector >= mddev->dev_sectors) |
3812 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 3813 | while (first_sector <= last_sector) { |
b5663ba4 | 3814 | sh = get_active_stripe(conf, first_sector, 1, 0); |
52c03291 N |
3815 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3816 | set_bit(STRIPE_HANDLE, &sh->state); | |
3817 | release_stripe(sh); | |
3818 | first_sector += STRIPE_SECTORS; | |
3819 | } | |
c6207277 N |
3820 | /* If this takes us to the resync_max point where we have to pause, |
3821 | * then we need to write out the superblock. | |
3822 | */ | |
3823 | sector_nr += conf->chunk_size>>9; | |
3824 | if (sector_nr >= mddev->resync_max) { | |
3825 | /* Cannot proceed until we've updated the superblock... */ | |
3826 | wait_event(conf->wait_for_overlap, | |
3827 | atomic_read(&conf->reshape_stripes) == 0); | |
3828 | mddev->reshape_position = conf->expand_progress; | |
3829 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
3830 | md_wakeup_thread(mddev->thread); | |
3831 | wait_event(mddev->sb_wait, | |
3832 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
3833 | || kthread_should_stop()); | |
3834 | spin_lock_irq(&conf->device_lock); | |
3835 | conf->expand_lo = mddev->reshape_position; | |
3836 | spin_unlock_irq(&conf->device_lock); | |
3837 | wake_up(&conf->wait_for_overlap); | |
3838 | } | |
52c03291 N |
3839 | return conf->chunk_size>>9; |
3840 | } | |
3841 | ||
3842 | /* FIXME go_faster isn't used */ | |
3843 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
3844 | { | |
3845 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3846 | struct stripe_head *sh; | |
58c0fed4 | 3847 | sector_t max_sector = mddev->dev_sectors; |
72626685 | 3848 | int sync_blocks; |
16a53ecc N |
3849 | int still_degraded = 0; |
3850 | int i; | |
1da177e4 | 3851 | |
72626685 | 3852 | if (sector_nr >= max_sector) { |
1da177e4 LT |
3853 | /* just being told to finish up .. nothing much to do */ |
3854 | unplug_slaves(mddev); | |
29269553 N |
3855 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3856 | end_reshape(conf); | |
3857 | return 0; | |
3858 | } | |
72626685 N |
3859 | |
3860 | if (mddev->curr_resync < max_sector) /* aborted */ | |
3861 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3862 | &sync_blocks, 1); | |
16a53ecc | 3863 | else /* completed sync */ |
72626685 N |
3864 | conf->fullsync = 0; |
3865 | bitmap_close_sync(mddev->bitmap); | |
3866 | ||
1da177e4 LT |
3867 | return 0; |
3868 | } | |
ccfcc3c1 | 3869 | |
52c03291 N |
3870 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3871 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 3872 | |
c6207277 N |
3873 | /* No need to check resync_max as we never do more than one |
3874 | * stripe, and as resync_max will always be on a chunk boundary, | |
3875 | * if the check in md_do_sync didn't fire, there is no chance | |
3876 | * of overstepping resync_max here | |
3877 | */ | |
3878 | ||
16a53ecc | 3879 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
3880 | * to resync, then assert that we are finished, because there is |
3881 | * nothing we can do. | |
3882 | */ | |
3285edf1 | 3883 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 3884 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 3885 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 3886 | *skipped = 1; |
1da177e4 LT |
3887 | return rv; |
3888 | } | |
72626685 | 3889 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 3890 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
3891 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
3892 | /* we can skip this block, and probably more */ | |
3893 | sync_blocks /= STRIPE_SECTORS; | |
3894 | *skipped = 1; | |
3895 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
3896 | } | |
1da177e4 | 3897 | |
b47490c9 N |
3898 | |
3899 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); | |
3900 | ||
b5663ba4 | 3901 | sh = get_active_stripe(conf, sector_nr, 0, 1); |
1da177e4 | 3902 | if (sh == NULL) { |
b5663ba4 | 3903 | sh = get_active_stripe(conf, sector_nr, 0, 0); |
1da177e4 | 3904 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 3905 | * is trying to get access |
1da177e4 | 3906 | */ |
66c006a5 | 3907 | schedule_timeout_uninterruptible(1); |
1da177e4 | 3908 | } |
16a53ecc N |
3909 | /* Need to check if array will still be degraded after recovery/resync |
3910 | * We don't need to check the 'failed' flag as when that gets set, | |
3911 | * recovery aborts. | |
3912 | */ | |
3913 | for (i=0; i<mddev->raid_disks; i++) | |
3914 | if (conf->disks[i].rdev == NULL) | |
3915 | still_degraded = 1; | |
3916 | ||
3917 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
3918 | ||
3919 | spin_lock(&sh->lock); | |
1da177e4 LT |
3920 | set_bit(STRIPE_SYNCING, &sh->state); |
3921 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3922 | spin_unlock(&sh->lock); | |
3923 | ||
df10cfbc DW |
3924 | /* wait for any blocked device to be handled */ |
3925 | while(unlikely(!handle_stripe(sh, NULL))) | |
3926 | ; | |
1da177e4 LT |
3927 | release_stripe(sh); |
3928 | ||
3929 | return STRIPE_SECTORS; | |
3930 | } | |
3931 | ||
46031f9a RBJ |
3932 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
3933 | { | |
3934 | /* We may not be able to submit a whole bio at once as there | |
3935 | * may not be enough stripe_heads available. | |
3936 | * We cannot pre-allocate enough stripe_heads as we may need | |
3937 | * more than exist in the cache (if we allow ever large chunks). | |
3938 | * So we do one stripe head at a time and record in | |
3939 | * ->bi_hw_segments how many have been done. | |
3940 | * | |
3941 | * We *know* that this entire raid_bio is in one chunk, so | |
3942 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
3943 | */ | |
3944 | struct stripe_head *sh; | |
911d4ee8 | 3945 | int dd_idx; |
46031f9a RBJ |
3946 | sector_t sector, logical_sector, last_sector; |
3947 | int scnt = 0; | |
3948 | int remaining; | |
3949 | int handled = 0; | |
3950 | ||
3951 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 3952 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 3953 | 0, &dd_idx, NULL); |
46031f9a RBJ |
3954 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); |
3955 | ||
3956 | for (; logical_sector < last_sector; | |
387bb173 NB |
3957 | logical_sector += STRIPE_SECTORS, |
3958 | sector += STRIPE_SECTORS, | |
3959 | scnt++) { | |
46031f9a | 3960 | |
960e739d | 3961 | if (scnt < raid5_bi_hw_segments(raid_bio)) |
46031f9a RBJ |
3962 | /* already done this stripe */ |
3963 | continue; | |
3964 | ||
b5663ba4 | 3965 | sh = get_active_stripe(conf, sector, 0, 1); |
46031f9a RBJ |
3966 | |
3967 | if (!sh) { | |
3968 | /* failed to get a stripe - must wait */ | |
960e739d | 3969 | raid5_set_bi_hw_segments(raid_bio, scnt); |
46031f9a RBJ |
3970 | conf->retry_read_aligned = raid_bio; |
3971 | return handled; | |
3972 | } | |
3973 | ||
3974 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
3975 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
3976 | release_stripe(sh); | |
960e739d | 3977 | raid5_set_bi_hw_segments(raid_bio, scnt); |
387bb173 NB |
3978 | conf->retry_read_aligned = raid_bio; |
3979 | return handled; | |
3980 | } | |
3981 | ||
46031f9a RBJ |
3982 | handle_stripe(sh, NULL); |
3983 | release_stripe(sh); | |
3984 | handled++; | |
3985 | } | |
3986 | spin_lock_irq(&conf->device_lock); | |
960e739d | 3987 | remaining = raid5_dec_bi_phys_segments(raid_bio); |
46031f9a | 3988 | spin_unlock_irq(&conf->device_lock); |
0e13fe23 NB |
3989 | if (remaining == 0) |
3990 | bio_endio(raid_bio, 0); | |
46031f9a RBJ |
3991 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
3992 | wake_up(&conf->wait_for_stripe); | |
3993 | return handled; | |
3994 | } | |
3995 | ||
3996 | ||
3997 | ||
1da177e4 LT |
3998 | /* |
3999 | * This is our raid5 kernel thread. | |
4000 | * | |
4001 | * We scan the hash table for stripes which can be handled now. | |
4002 | * During the scan, completed stripes are saved for us by the interrupt | |
4003 | * handler, so that they will not have to wait for our next wakeup. | |
4004 | */ | |
6ed3003c | 4005 | static void raid5d(mddev_t *mddev) |
1da177e4 LT |
4006 | { |
4007 | struct stripe_head *sh; | |
4008 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4009 | int handled; | |
4010 | ||
45b4233c | 4011 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
4012 | |
4013 | md_check_recovery(mddev); | |
1da177e4 LT |
4014 | |
4015 | handled = 0; | |
4016 | spin_lock_irq(&conf->device_lock); | |
4017 | while (1) { | |
46031f9a | 4018 | struct bio *bio; |
1da177e4 | 4019 | |
ae3c20cc | 4020 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 4021 | int seq = conf->seq_flush; |
700e432d | 4022 | spin_unlock_irq(&conf->device_lock); |
72626685 | 4023 | bitmap_unplug(mddev->bitmap); |
700e432d | 4024 | spin_lock_irq(&conf->device_lock); |
72626685 N |
4025 | conf->seq_write = seq; |
4026 | activate_bit_delay(conf); | |
4027 | } | |
4028 | ||
46031f9a RBJ |
4029 | while ((bio = remove_bio_from_retry(conf))) { |
4030 | int ok; | |
4031 | spin_unlock_irq(&conf->device_lock); | |
4032 | ok = retry_aligned_read(conf, bio); | |
4033 | spin_lock_irq(&conf->device_lock); | |
4034 | if (!ok) | |
4035 | break; | |
4036 | handled++; | |
4037 | } | |
4038 | ||
8b3e6cdc DW |
4039 | sh = __get_priority_stripe(conf); |
4040 | ||
c9f21aaf | 4041 | if (!sh) |
1da177e4 | 4042 | break; |
1da177e4 LT |
4043 | spin_unlock_irq(&conf->device_lock); |
4044 | ||
4045 | handled++; | |
16a53ecc | 4046 | handle_stripe(sh, conf->spare_page); |
1da177e4 LT |
4047 | release_stripe(sh); |
4048 | ||
4049 | spin_lock_irq(&conf->device_lock); | |
4050 | } | |
45b4233c | 4051 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
4052 | |
4053 | spin_unlock_irq(&conf->device_lock); | |
4054 | ||
c9f21aaf | 4055 | async_tx_issue_pending_all(); |
1da177e4 LT |
4056 | unplug_slaves(mddev); |
4057 | ||
45b4233c | 4058 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
4059 | } |
4060 | ||
3f294f4f | 4061 | static ssize_t |
007583c9 | 4062 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 4063 | { |
007583c9 | 4064 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4065 | if (conf) |
4066 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
4067 | else | |
4068 | return 0; | |
3f294f4f N |
4069 | } |
4070 | ||
4071 | static ssize_t | |
007583c9 | 4072 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 4073 | { |
007583c9 | 4074 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4ef197d8 | 4075 | unsigned long new; |
b5470dc5 DW |
4076 | int err; |
4077 | ||
3f294f4f N |
4078 | if (len >= PAGE_SIZE) |
4079 | return -EINVAL; | |
96de1e66 N |
4080 | if (!conf) |
4081 | return -ENODEV; | |
3f294f4f | 4082 | |
4ef197d8 | 4083 | if (strict_strtoul(page, 10, &new)) |
3f294f4f N |
4084 | return -EINVAL; |
4085 | if (new <= 16 || new > 32768) | |
4086 | return -EINVAL; | |
4087 | while (new < conf->max_nr_stripes) { | |
4088 | if (drop_one_stripe(conf)) | |
4089 | conf->max_nr_stripes--; | |
4090 | else | |
4091 | break; | |
4092 | } | |
b5470dc5 DW |
4093 | err = md_allow_write(mddev); |
4094 | if (err) | |
4095 | return err; | |
3f294f4f N |
4096 | while (new > conf->max_nr_stripes) { |
4097 | if (grow_one_stripe(conf)) | |
4098 | conf->max_nr_stripes++; | |
4099 | else break; | |
4100 | } | |
4101 | return len; | |
4102 | } | |
007583c9 | 4103 | |
96de1e66 N |
4104 | static struct md_sysfs_entry |
4105 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
4106 | raid5_show_stripe_cache_size, | |
4107 | raid5_store_stripe_cache_size); | |
3f294f4f | 4108 | |
8b3e6cdc DW |
4109 | static ssize_t |
4110 | raid5_show_preread_threshold(mddev_t *mddev, char *page) | |
4111 | { | |
4112 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4113 | if (conf) | |
4114 | return sprintf(page, "%d\n", conf->bypass_threshold); | |
4115 | else | |
4116 | return 0; | |
4117 | } | |
4118 | ||
4119 | static ssize_t | |
4120 | raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len) | |
4121 | { | |
4122 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4ef197d8 | 4123 | unsigned long new; |
8b3e6cdc DW |
4124 | if (len >= PAGE_SIZE) |
4125 | return -EINVAL; | |
4126 | if (!conf) | |
4127 | return -ENODEV; | |
4128 | ||
4ef197d8 | 4129 | if (strict_strtoul(page, 10, &new)) |
8b3e6cdc | 4130 | return -EINVAL; |
4ef197d8 | 4131 | if (new > conf->max_nr_stripes) |
8b3e6cdc DW |
4132 | return -EINVAL; |
4133 | conf->bypass_threshold = new; | |
4134 | return len; | |
4135 | } | |
4136 | ||
4137 | static struct md_sysfs_entry | |
4138 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
4139 | S_IRUGO | S_IWUSR, | |
4140 | raid5_show_preread_threshold, | |
4141 | raid5_store_preread_threshold); | |
4142 | ||
3f294f4f | 4143 | static ssize_t |
96de1e66 | 4144 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 4145 | { |
007583c9 | 4146 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
4147 | if (conf) |
4148 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
4149 | else | |
4150 | return 0; | |
3f294f4f N |
4151 | } |
4152 | ||
96de1e66 N |
4153 | static struct md_sysfs_entry |
4154 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 4155 | |
007583c9 | 4156 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
4157 | &raid5_stripecache_size.attr, |
4158 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 4159 | &raid5_preread_bypass_threshold.attr, |
3f294f4f N |
4160 | NULL, |
4161 | }; | |
007583c9 N |
4162 | static struct attribute_group raid5_attrs_group = { |
4163 | .name = NULL, | |
4164 | .attrs = raid5_attrs, | |
3f294f4f N |
4165 | }; |
4166 | ||
91adb564 | 4167 | static raid5_conf_t *setup_conf(mddev_t *mddev) |
1da177e4 LT |
4168 | { |
4169 | raid5_conf_t *conf; | |
4170 | int raid_disk, memory; | |
4171 | mdk_rdev_t *rdev; | |
4172 | struct disk_info *disk; | |
1da177e4 | 4173 | |
91adb564 N |
4174 | if (mddev->new_level != 5 |
4175 | && mddev->new_level != 4 | |
4176 | && mddev->new_level != 6) { | |
16a53ecc | 4177 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
4178 | mdname(mddev), mddev->new_level); |
4179 | return ERR_PTR(-EIO); | |
1da177e4 | 4180 | } |
91adb564 N |
4181 | if ((mddev->new_level == 5 |
4182 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
4183 | (mddev->new_level == 6 | |
4184 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
99c0fb5f | 4185 | printk(KERN_ERR "raid5: %s: layout %d not supported\n", |
91adb564 N |
4186 | mdname(mddev), mddev->new_layout); |
4187 | return ERR_PTR(-EIO); | |
99c0fb5f | 4188 | } |
91adb564 N |
4189 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
4190 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
4191 | mdname(mddev), mddev->raid_disks); | |
4192 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
4193 | } |
4194 | ||
91adb564 N |
4195 | if (!mddev->new_chunk || mddev->new_chunk % PAGE_SIZE) { |
4196 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
4197 | mddev->new_chunk, mdname(mddev)); | |
4198 | return ERR_PTR(-EINVAL); | |
f6705578 N |
4199 | } |
4200 | ||
91adb564 N |
4201 | conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL); |
4202 | if (conf == NULL) | |
1da177e4 | 4203 | goto abort; |
91adb564 N |
4204 | |
4205 | conf->raid_disks = mddev->raid_disks; | |
4206 | if (mddev->reshape_position == MaxSector) | |
4207 | conf->previous_raid_disks = mddev->raid_disks; | |
4208 | else | |
f6705578 | 4209 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
f6705578 N |
4210 | |
4211 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
4212 | GFP_KERNEL); |
4213 | if (!conf->disks) | |
4214 | goto abort; | |
9ffae0cf | 4215 | |
1da177e4 LT |
4216 | conf->mddev = mddev; |
4217 | ||
fccddba0 | 4218 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 4219 | goto abort; |
1da177e4 | 4220 | |
91adb564 | 4221 | if (mddev->new_level == 6) { |
16a53ecc N |
4222 | conf->spare_page = alloc_page(GFP_KERNEL); |
4223 | if (!conf->spare_page) | |
4224 | goto abort; | |
4225 | } | |
1da177e4 LT |
4226 | spin_lock_init(&conf->device_lock); |
4227 | init_waitqueue_head(&conf->wait_for_stripe); | |
4228 | init_waitqueue_head(&conf->wait_for_overlap); | |
4229 | INIT_LIST_HEAD(&conf->handle_list); | |
8b3e6cdc | 4230 | INIT_LIST_HEAD(&conf->hold_list); |
1da177e4 | 4231 | INIT_LIST_HEAD(&conf->delayed_list); |
72626685 | 4232 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
4233 | INIT_LIST_HEAD(&conf->inactive_list); |
4234 | atomic_set(&conf->active_stripes, 0); | |
4235 | atomic_set(&conf->preread_active_stripes, 0); | |
46031f9a | 4236 | atomic_set(&conf->active_aligned_reads, 0); |
8b3e6cdc | 4237 | conf->bypass_threshold = BYPASS_THRESHOLD; |
1da177e4 | 4238 | |
45b4233c | 4239 | pr_debug("raid5: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 4240 | |
159ec1fc | 4241 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 4242 | raid_disk = rdev->raid_disk; |
f6705578 | 4243 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
4244 | || raid_disk < 0) |
4245 | continue; | |
4246 | disk = conf->disks + raid_disk; | |
4247 | ||
4248 | disk->rdev = rdev; | |
4249 | ||
b2d444d7 | 4250 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
4251 | char b[BDEVNAME_SIZE]; |
4252 | printk(KERN_INFO "raid5: device %s operational as raid" | |
4253 | " disk %d\n", bdevname(rdev->bdev,b), | |
4254 | raid_disk); | |
8c2e870a NB |
4255 | } else |
4256 | /* Cannot rely on bitmap to complete recovery */ | |
4257 | conf->fullsync = 1; | |
1da177e4 LT |
4258 | } |
4259 | ||
91adb564 N |
4260 | conf->chunk_size = mddev->new_chunk; |
4261 | conf->level = mddev->new_level; | |
16a53ecc N |
4262 | if (conf->level == 6) |
4263 | conf->max_degraded = 2; | |
4264 | else | |
4265 | conf->max_degraded = 1; | |
91adb564 | 4266 | conf->algorithm = mddev->new_layout; |
1da177e4 | 4267 | conf->max_nr_stripes = NR_STRIPES; |
f6705578 | 4268 | conf->expand_progress = mddev->reshape_position; |
1da177e4 | 4269 | |
91adb564 N |
4270 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
4271 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; | |
4272 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
4273 | printk(KERN_ERR | |
4274 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
4275 | goto abort; | |
4276 | } else | |
4277 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
4278 | memory, mdname(mddev)); | |
1da177e4 | 4279 | |
91adb564 N |
4280 | conf->thread = md_register_thread(raid5d, mddev, "%s_raid5"); |
4281 | if (!conf->thread) { | |
4282 | printk(KERN_ERR | |
4283 | "raid5: couldn't allocate thread for %s\n", | |
4284 | mdname(mddev)); | |
16a53ecc N |
4285 | goto abort; |
4286 | } | |
91adb564 N |
4287 | |
4288 | return conf; | |
4289 | ||
4290 | abort: | |
4291 | if (conf) { | |
4292 | shrink_stripes(conf); | |
4293 | safe_put_page(conf->spare_page); | |
4294 | kfree(conf->disks); | |
4295 | kfree(conf->stripe_hashtbl); | |
4296 | kfree(conf); | |
4297 | return ERR_PTR(-EIO); | |
4298 | } else | |
4299 | return ERR_PTR(-ENOMEM); | |
4300 | } | |
4301 | ||
4302 | static int run(mddev_t *mddev) | |
4303 | { | |
4304 | raid5_conf_t *conf; | |
4305 | int working_disks = 0; | |
4306 | mdk_rdev_t *rdev; | |
4307 | ||
4308 | if (mddev->reshape_position != MaxSector) { | |
4309 | /* Check that we can continue the reshape. | |
4310 | * Currently only disks can change, it must | |
4311 | * increase, and we must be past the point where | |
4312 | * a stripe over-writes itself | |
4313 | */ | |
4314 | sector_t here_new, here_old; | |
4315 | int old_disks; | |
4316 | int max_degraded = (mddev->level == 5 ? 1 : 2); | |
4317 | ||
4318 | if (mddev->new_level != mddev->level || | |
4319 | mddev->new_layout != mddev->layout || | |
4320 | mddev->new_chunk != mddev->chunk_size) { | |
4321 | printk(KERN_ERR "raid5: %s: unsupported reshape " | |
4322 | "required - aborting.\n", | |
4323 | mdname(mddev)); | |
4324 | return -EINVAL; | |
4325 | } | |
4326 | if (mddev->delta_disks <= 0) { | |
4327 | printk(KERN_ERR "raid5: %s: unsupported reshape " | |
4328 | "(reduce disks) required - aborting.\n", | |
4329 | mdname(mddev)); | |
4330 | return -EINVAL; | |
4331 | } | |
4332 | old_disks = mddev->raid_disks - mddev->delta_disks; | |
4333 | /* reshape_position must be on a new-stripe boundary, and one | |
4334 | * further up in new geometry must map after here in old | |
4335 | * geometry. | |
4336 | */ | |
4337 | here_new = mddev->reshape_position; | |
4338 | if (sector_div(here_new, (mddev->chunk_size>>9)* | |
4339 | (mddev->raid_disks - max_degraded))) { | |
4340 | printk(KERN_ERR "raid5: reshape_position not " | |
4341 | "on a stripe boundary\n"); | |
4342 | return -EINVAL; | |
4343 | } | |
4344 | /* here_new is the stripe we will write to */ | |
4345 | here_old = mddev->reshape_position; | |
4346 | sector_div(here_old, (mddev->chunk_size>>9)* | |
4347 | (old_disks-max_degraded)); | |
4348 | /* here_old is the first stripe that we might need to read | |
4349 | * from */ | |
4350 | if (here_new >= here_old) { | |
4351 | /* Reading from the same stripe as writing to - bad */ | |
4352 | printk(KERN_ERR "raid5: reshape_position too early for " | |
4353 | "auto-recovery - aborting.\n"); | |
4354 | return -EINVAL; | |
4355 | } | |
4356 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
4357 | /* OK, we should be able to continue; */ | |
4358 | } else { | |
4359 | BUG_ON(mddev->level != mddev->new_level); | |
4360 | BUG_ON(mddev->layout != mddev->new_layout); | |
4361 | BUG_ON(mddev->chunk_size != mddev->new_chunk); | |
4362 | BUG_ON(mddev->delta_disks != 0); | |
1da177e4 | 4363 | } |
91adb564 N |
4364 | conf = setup_conf(mddev); |
4365 | ||
4366 | if (conf == NULL) | |
4367 | return -EIO; | |
4368 | if (IS_ERR(conf)) | |
4369 | return PTR_ERR(conf); | |
4370 | ||
4371 | mddev->thread = conf->thread; | |
4372 | conf->thread = NULL; | |
4373 | mddev->private = conf; | |
4374 | ||
4375 | /* | |
4376 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
4377 | */ | |
4378 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
4379 | if (rdev->raid_disk >= 0 && | |
4380 | test_bit(In_sync, &rdev->flags)) | |
4381 | working_disks++; | |
4382 | ||
4383 | mddev->degraded = conf->raid_disks - working_disks; | |
4384 | ||
16a53ecc | 4385 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
4386 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
4387 | " (%d/%d failed)\n", | |
02c2de8c | 4388 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
4389 | goto abort; |
4390 | } | |
4391 | ||
91adb564 N |
4392 | /* device size must be a multiple of chunk size */ |
4393 | mddev->dev_sectors &= ~(mddev->chunk_size / 512 - 1); | |
4394 | mddev->resync_max_sectors = mddev->dev_sectors; | |
4395 | ||
16a53ecc | 4396 | if (mddev->degraded > 0 && |
1da177e4 | 4397 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
4398 | if (mddev->ok_start_degraded) |
4399 | printk(KERN_WARNING | |
4400 | "raid5: starting dirty degraded array: %s" | |
4401 | "- data corruption possible.\n", | |
4402 | mdname(mddev)); | |
4403 | else { | |
4404 | printk(KERN_ERR | |
4405 | "raid5: cannot start dirty degraded array for %s\n", | |
4406 | mdname(mddev)); | |
4407 | goto abort; | |
4408 | } | |
1da177e4 LT |
4409 | } |
4410 | ||
1da177e4 LT |
4411 | if (mddev->degraded == 0) |
4412 | printk("raid5: raid level %d set %s active with %d out of %d" | |
4413 | " devices, algorithm %d\n", conf->level, mdname(mddev), | |
4414 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
4415 | conf->algorithm); | |
4416 | else | |
4417 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
4418 | " out of %d devices, algorithm %d\n", conf->level, | |
4419 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
4420 | mddev->raid_disks, conf->algorithm); | |
4421 | ||
4422 | print_raid5_conf(conf); | |
4423 | ||
f6705578 N |
4424 | if (conf->expand_progress != MaxSector) { |
4425 | printk("...ok start reshape thread\n"); | |
b578d55f | 4426 | conf->expand_lo = conf->expand_progress; |
f6705578 N |
4427 | atomic_set(&conf->reshape_stripes, 0); |
4428 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4429 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4430 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4431 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4432 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4433 | "%s_reshape"); | |
f6705578 N |
4434 | } |
4435 | ||
1da177e4 | 4436 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 4437 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
4438 | */ |
4439 | { | |
16a53ecc N |
4440 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
4441 | int stripe = data_disks * | |
8932c2e0 | 4442 | (mddev->chunk_size / PAGE_SIZE); |
1da177e4 LT |
4443 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
4444 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4445 | } | |
4446 | ||
4447 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
4448 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
4449 | printk(KERN_WARNING | |
4450 | "raid5: failed to create sysfs attributes for %s\n", | |
4451 | mdname(mddev)); | |
7a5febe9 | 4452 | |
91adb564 N |
4453 | mddev->queue->queue_lock = &conf->device_lock; |
4454 | ||
7a5febe9 | 4455 | mddev->queue->unplug_fn = raid5_unplug_device; |
f022b2fd | 4456 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 4457 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 4458 | |
58c0fed4 AN |
4459 | mddev->array_sectors = mddev->dev_sectors * |
4460 | (conf->previous_raid_disks - conf->max_degraded); | |
7a5febe9 | 4461 | |
23032a0e RBJ |
4462 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
4463 | ||
1da177e4 LT |
4464 | return 0; |
4465 | abort: | |
91adb564 N |
4466 | if (mddev->thread) |
4467 | md_unregister_thread(mddev->thread); | |
4468 | mddev->thread = NULL; | |
1da177e4 | 4469 | if (conf) { |
91adb564 | 4470 | shrink_stripes(conf); |
1da177e4 | 4471 | print_raid5_conf(conf); |
16a53ecc | 4472 | safe_put_page(conf->spare_page); |
b55e6bfc | 4473 | kfree(conf->disks); |
fccddba0 | 4474 | kfree(conf->stripe_hashtbl); |
1da177e4 LT |
4475 | kfree(conf); |
4476 | } | |
4477 | mddev->private = NULL; | |
4478 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
4479 | return -EIO; | |
4480 | } | |
4481 | ||
4482 | ||
4483 | ||
3f294f4f | 4484 | static int stop(mddev_t *mddev) |
1da177e4 LT |
4485 | { |
4486 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4487 | ||
4488 | md_unregister_thread(mddev->thread); | |
4489 | mddev->thread = NULL; | |
4490 | shrink_stripes(conf); | |
fccddba0 | 4491 | kfree(conf->stripe_hashtbl); |
041ae52e | 4492 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 4493 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 4494 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
b55e6bfc | 4495 | kfree(conf->disks); |
96de1e66 | 4496 | kfree(conf); |
1da177e4 LT |
4497 | mddev->private = NULL; |
4498 | return 0; | |
4499 | } | |
4500 | ||
45b4233c | 4501 | #ifdef DEBUG |
d710e138 | 4502 | static void print_sh(struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
4503 | { |
4504 | int i; | |
4505 | ||
16a53ecc N |
4506 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
4507 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
4508 | seq_printf(seq, "sh %llu, count %d.\n", | |
4509 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
4510 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 4511 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
4512 | seq_printf(seq, "(cache%d: %p %ld) ", |
4513 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 4514 | } |
16a53ecc | 4515 | seq_printf(seq, "\n"); |
1da177e4 LT |
4516 | } |
4517 | ||
d710e138 | 4518 | static void printall(struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
4519 | { |
4520 | struct stripe_head *sh; | |
fccddba0 | 4521 | struct hlist_node *hn; |
1da177e4 LT |
4522 | int i; |
4523 | ||
4524 | spin_lock_irq(&conf->device_lock); | |
4525 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 4526 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
4527 | if (sh->raid_conf != conf) |
4528 | continue; | |
16a53ecc | 4529 | print_sh(seq, sh); |
1da177e4 LT |
4530 | } |
4531 | } | |
4532 | spin_unlock_irq(&conf->device_lock); | |
4533 | } | |
4534 | #endif | |
4535 | ||
d710e138 | 4536 | static void status(struct seq_file *seq, mddev_t *mddev) |
1da177e4 LT |
4537 | { |
4538 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
4539 | int i; | |
4540 | ||
4541 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
02c2de8c | 4542 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
4543 | for (i = 0; i < conf->raid_disks; i++) |
4544 | seq_printf (seq, "%s", | |
4545 | conf->disks[i].rdev && | |
b2d444d7 | 4546 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 4547 | seq_printf (seq, "]"); |
45b4233c | 4548 | #ifdef DEBUG |
16a53ecc N |
4549 | seq_printf (seq, "\n"); |
4550 | printall(seq, conf); | |
1da177e4 LT |
4551 | #endif |
4552 | } | |
4553 | ||
4554 | static void print_raid5_conf (raid5_conf_t *conf) | |
4555 | { | |
4556 | int i; | |
4557 | struct disk_info *tmp; | |
4558 | ||
4559 | printk("RAID5 conf printout:\n"); | |
4560 | if (!conf) { | |
4561 | printk("(conf==NULL)\n"); | |
4562 | return; | |
4563 | } | |
02c2de8c N |
4564 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
4565 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
4566 | |
4567 | for (i = 0; i < conf->raid_disks; i++) { | |
4568 | char b[BDEVNAME_SIZE]; | |
4569 | tmp = conf->disks + i; | |
4570 | if (tmp->rdev) | |
4571 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 4572 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
4573 | bdevname(tmp->rdev->bdev,b)); |
4574 | } | |
4575 | } | |
4576 | ||
4577 | static int raid5_spare_active(mddev_t *mddev) | |
4578 | { | |
4579 | int i; | |
4580 | raid5_conf_t *conf = mddev->private; | |
4581 | struct disk_info *tmp; | |
4582 | ||
4583 | for (i = 0; i < conf->raid_disks; i++) { | |
4584 | tmp = conf->disks + i; | |
4585 | if (tmp->rdev | |
b2d444d7 | 4586 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
4587 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
4588 | unsigned long flags; | |
4589 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 4590 | mddev->degraded--; |
c04be0aa | 4591 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
4592 | } |
4593 | } | |
4594 | print_raid5_conf(conf); | |
4595 | return 0; | |
4596 | } | |
4597 | ||
4598 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
4599 | { | |
4600 | raid5_conf_t *conf = mddev->private; | |
4601 | int err = 0; | |
4602 | mdk_rdev_t *rdev; | |
4603 | struct disk_info *p = conf->disks + number; | |
4604 | ||
4605 | print_raid5_conf(conf); | |
4606 | rdev = p->rdev; | |
4607 | if (rdev) { | |
b2d444d7 | 4608 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
4609 | atomic_read(&rdev->nr_pending)) { |
4610 | err = -EBUSY; | |
4611 | goto abort; | |
4612 | } | |
dfc70645 N |
4613 | /* Only remove non-faulty devices if recovery |
4614 | * isn't possible. | |
4615 | */ | |
4616 | if (!test_bit(Faulty, &rdev->flags) && | |
4617 | mddev->degraded <= conf->max_degraded) { | |
4618 | err = -EBUSY; | |
4619 | goto abort; | |
4620 | } | |
1da177e4 | 4621 | p->rdev = NULL; |
fbd568a3 | 4622 | synchronize_rcu(); |
1da177e4 LT |
4623 | if (atomic_read(&rdev->nr_pending)) { |
4624 | /* lost the race, try later */ | |
4625 | err = -EBUSY; | |
4626 | p->rdev = rdev; | |
4627 | } | |
4628 | } | |
4629 | abort: | |
4630 | ||
4631 | print_raid5_conf(conf); | |
4632 | return err; | |
4633 | } | |
4634 | ||
4635 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
4636 | { | |
4637 | raid5_conf_t *conf = mddev->private; | |
199050ea | 4638 | int err = -EEXIST; |
1da177e4 LT |
4639 | int disk; |
4640 | struct disk_info *p; | |
6c2fce2e NB |
4641 | int first = 0; |
4642 | int last = conf->raid_disks - 1; | |
1da177e4 | 4643 | |
16a53ecc | 4644 | if (mddev->degraded > conf->max_degraded) |
1da177e4 | 4645 | /* no point adding a device */ |
199050ea | 4646 | return -EINVAL; |
1da177e4 | 4647 | |
6c2fce2e NB |
4648 | if (rdev->raid_disk >= 0) |
4649 | first = last = rdev->raid_disk; | |
1da177e4 LT |
4650 | |
4651 | /* | |
16a53ecc N |
4652 | * find the disk ... but prefer rdev->saved_raid_disk |
4653 | * if possible. | |
1da177e4 | 4654 | */ |
16a53ecc | 4655 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 4656 | rdev->saved_raid_disk >= first && |
16a53ecc N |
4657 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
4658 | disk = rdev->saved_raid_disk; | |
4659 | else | |
6c2fce2e NB |
4660 | disk = first; |
4661 | for ( ; disk <= last ; disk++) | |
1da177e4 | 4662 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 4663 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 4664 | rdev->raid_disk = disk; |
199050ea | 4665 | err = 0; |
72626685 N |
4666 | if (rdev->saved_raid_disk != disk) |
4667 | conf->fullsync = 1; | |
d6065f7b | 4668 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
4669 | break; |
4670 | } | |
4671 | print_raid5_conf(conf); | |
199050ea | 4672 | return err; |
1da177e4 LT |
4673 | } |
4674 | ||
4675 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
4676 | { | |
4677 | /* no resync is happening, and there is enough space | |
4678 | * on all devices, so we can resize. | |
4679 | * We need to make sure resync covers any new space. | |
4680 | * If the array is shrinking we should possibly wait until | |
4681 | * any io in the removed space completes, but it hardly seems | |
4682 | * worth it. | |
4683 | */ | |
16a53ecc N |
4684 | raid5_conf_t *conf = mddev_to_conf(mddev); |
4685 | ||
1da177e4 | 4686 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); |
f233ea5c AN |
4687 | mddev->array_sectors = sectors * (mddev->raid_disks |
4688 | - conf->max_degraded); | |
4689 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
44ce6294 | 4690 | mddev->changed = 1; |
58c0fed4 AN |
4691 | if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) { |
4692 | mddev->recovery_cp = mddev->dev_sectors; | |
1da177e4 LT |
4693 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
4694 | } | |
58c0fed4 | 4695 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 4696 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
4697 | return 0; |
4698 | } | |
4699 | ||
29269553 | 4700 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f | 4701 | static int raid5_check_reshape(mddev_t *mddev) |
29269553 N |
4702 | { |
4703 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4704 | int err; | |
29269553 | 4705 | |
63c70c4f N |
4706 | if (mddev->delta_disks < 0 || |
4707 | mddev->new_level != mddev->level) | |
4708 | return -EINVAL; /* Cannot shrink array or change level yet */ | |
4709 | if (mddev->delta_disks == 0) | |
29269553 | 4710 | return 0; /* nothing to do */ |
dba034ee N |
4711 | if (mddev->bitmap) |
4712 | /* Cannot grow a bitmap yet */ | |
4713 | return -EBUSY; | |
29269553 N |
4714 | |
4715 | /* Can only proceed if there are plenty of stripe_heads. | |
4716 | * We need a minimum of one full stripe,, and for sensible progress | |
4717 | * it is best to have about 4 times that. | |
4718 | * If we require 4 times, then the default 256 4K stripe_heads will | |
4719 | * allow for chunk sizes up to 256K, which is probably OK. | |
4720 | * If the chunk size is greater, user-space should request more | |
4721 | * stripe_heads first. | |
4722 | */ | |
63c70c4f N |
4723 | if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes || |
4724 | (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) { | |
29269553 N |
4725 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", |
4726 | (mddev->chunk_size / STRIPE_SIZE)*4); | |
4727 | return -ENOSPC; | |
4728 | } | |
4729 | ||
63c70c4f N |
4730 | err = resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
4731 | if (err) | |
4732 | return err; | |
4733 | ||
b4c4c7b8 N |
4734 | if (mddev->degraded > conf->max_degraded) |
4735 | return -EINVAL; | |
63c70c4f N |
4736 | /* looks like we might be able to manage this */ |
4737 | return 0; | |
4738 | } | |
4739 | ||
4740 | static int raid5_start_reshape(mddev_t *mddev) | |
4741 | { | |
4742 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4743 | mdk_rdev_t *rdev; | |
63c70c4f N |
4744 | int spares = 0; |
4745 | int added_devices = 0; | |
c04be0aa | 4746 | unsigned long flags; |
63c70c4f | 4747 | |
f416885e | 4748 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
4749 | return -EBUSY; |
4750 | ||
159ec1fc | 4751 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4752 | if (rdev->raid_disk < 0 && |
4753 | !test_bit(Faulty, &rdev->flags)) | |
4754 | spares++; | |
63c70c4f | 4755 | |
f416885e | 4756 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
4757 | /* Not enough devices even to make a degraded array |
4758 | * of that size | |
4759 | */ | |
4760 | return -EINVAL; | |
4761 | ||
f6705578 | 4762 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
4763 | spin_lock_irq(&conf->device_lock); |
4764 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 4765 | conf->raid_disks += mddev->delta_disks; |
29269553 | 4766 | conf->expand_progress = 0; |
b578d55f | 4767 | conf->expand_lo = 0; |
29269553 N |
4768 | spin_unlock_irq(&conf->device_lock); |
4769 | ||
4770 | /* Add some new drives, as many as will fit. | |
4771 | * We know there are enough to make the newly sized array work. | |
4772 | */ | |
159ec1fc | 4773 | list_for_each_entry(rdev, &mddev->disks, same_set) |
29269553 N |
4774 | if (rdev->raid_disk < 0 && |
4775 | !test_bit(Faulty, &rdev->flags)) { | |
199050ea | 4776 | if (raid5_add_disk(mddev, rdev) == 0) { |
29269553 N |
4777 | char nm[20]; |
4778 | set_bit(In_sync, &rdev->flags); | |
29269553 | 4779 | added_devices++; |
5fd6c1dc | 4780 | rdev->recovery_offset = 0; |
29269553 | 4781 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
4782 | if (sysfs_create_link(&mddev->kobj, |
4783 | &rdev->kobj, nm)) | |
4784 | printk(KERN_WARNING | |
4785 | "raid5: failed to create " | |
4786 | " link %s for %s\n", | |
4787 | nm, mdname(mddev)); | |
29269553 N |
4788 | } else |
4789 | break; | |
4790 | } | |
4791 | ||
c04be0aa | 4792 | spin_lock_irqsave(&conf->device_lock, flags); |
63c70c4f | 4793 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) - added_devices; |
c04be0aa | 4794 | spin_unlock_irqrestore(&conf->device_lock, flags); |
63c70c4f | 4795 | mddev->raid_disks = conf->raid_disks; |
f6705578 | 4796 | mddev->reshape_position = 0; |
850b2b42 | 4797 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 4798 | |
29269553 N |
4799 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
4800 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4801 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4802 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4803 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4804 | "%s_reshape"); | |
4805 | if (!mddev->sync_thread) { | |
4806 | mddev->recovery = 0; | |
4807 | spin_lock_irq(&conf->device_lock); | |
4808 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
4809 | conf->expand_progress = MaxSector; | |
4810 | spin_unlock_irq(&conf->device_lock); | |
4811 | return -EAGAIN; | |
4812 | } | |
4813 | md_wakeup_thread(mddev->sync_thread); | |
4814 | md_new_event(mddev); | |
4815 | return 0; | |
4816 | } | |
4817 | #endif | |
4818 | ||
4819 | static void end_reshape(raid5_conf_t *conf) | |
4820 | { | |
4821 | struct block_device *bdev; | |
4822 | ||
f6705578 | 4823 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
58c0fed4 | 4824 | conf->mddev->array_sectors = conf->mddev->dev_sectors * |
f416885e | 4825 | (conf->raid_disks - conf->max_degraded); |
f233ea5c | 4826 | set_capacity(conf->mddev->gendisk, conf->mddev->array_sectors); |
44ce6294 | 4827 | conf->mddev->changed = 1; |
f6705578 N |
4828 | |
4829 | bdev = bdget_disk(conf->mddev->gendisk, 0); | |
4830 | if (bdev) { | |
4831 | mutex_lock(&bdev->bd_inode->i_mutex); | |
f233ea5c AN |
4832 | i_size_write(bdev->bd_inode, |
4833 | (loff_t)conf->mddev->array_sectors << 9); | |
f6705578 N |
4834 | mutex_unlock(&bdev->bd_inode->i_mutex); |
4835 | bdput(bdev); | |
4836 | } | |
4837 | spin_lock_irq(&conf->device_lock); | |
4838 | conf->expand_progress = MaxSector; | |
4839 | spin_unlock_irq(&conf->device_lock); | |
4840 | conf->mddev->reshape_position = MaxSector; | |
16a53ecc N |
4841 | |
4842 | /* read-ahead size must cover two whole stripes, which is | |
4843 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4844 | */ | |
4845 | { | |
4846 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
4847 | int stripe = data_disks * | |
4848 | (conf->mddev->chunk_size / PAGE_SIZE); | |
4849 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4850 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4851 | } | |
29269553 | 4852 | } |
29269553 N |
4853 | } |
4854 | ||
72626685 N |
4855 | static void raid5_quiesce(mddev_t *mddev, int state) |
4856 | { | |
4857 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
4858 | ||
4859 | switch(state) { | |
e464eafd N |
4860 | case 2: /* resume for a suspend */ |
4861 | wake_up(&conf->wait_for_overlap); | |
4862 | break; | |
4863 | ||
72626685 N |
4864 | case 1: /* stop all writes */ |
4865 | spin_lock_irq(&conf->device_lock); | |
4866 | conf->quiesce = 1; | |
4867 | wait_event_lock_irq(conf->wait_for_stripe, | |
46031f9a RBJ |
4868 | atomic_read(&conf->active_stripes) == 0 && |
4869 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 N |
4870 | conf->device_lock, /* nothing */); |
4871 | spin_unlock_irq(&conf->device_lock); | |
4872 | break; | |
4873 | ||
4874 | case 0: /* re-enable writes */ | |
4875 | spin_lock_irq(&conf->device_lock); | |
4876 | conf->quiesce = 0; | |
4877 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 4878 | wake_up(&conf->wait_for_overlap); |
72626685 N |
4879 | spin_unlock_irq(&conf->device_lock); |
4880 | break; | |
4881 | } | |
72626685 | 4882 | } |
b15c2e57 | 4883 | |
16a53ecc N |
4884 | static struct mdk_personality raid6_personality = |
4885 | { | |
4886 | .name = "raid6", | |
4887 | .level = 6, | |
4888 | .owner = THIS_MODULE, | |
4889 | .make_request = make_request, | |
4890 | .run = run, | |
4891 | .stop = stop, | |
4892 | .status = status, | |
4893 | .error_handler = error, | |
4894 | .hot_add_disk = raid5_add_disk, | |
4895 | .hot_remove_disk= raid5_remove_disk, | |
4896 | .spare_active = raid5_spare_active, | |
4897 | .sync_request = sync_request, | |
4898 | .resize = raid5_resize, | |
f416885e N |
4899 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4900 | .check_reshape = raid5_check_reshape, | |
4901 | .start_reshape = raid5_start_reshape, | |
4902 | #endif | |
16a53ecc N |
4903 | .quiesce = raid5_quiesce, |
4904 | }; | |
2604b703 | 4905 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
4906 | { |
4907 | .name = "raid5", | |
2604b703 | 4908 | .level = 5, |
1da177e4 LT |
4909 | .owner = THIS_MODULE, |
4910 | .make_request = make_request, | |
4911 | .run = run, | |
4912 | .stop = stop, | |
4913 | .status = status, | |
4914 | .error_handler = error, | |
4915 | .hot_add_disk = raid5_add_disk, | |
4916 | .hot_remove_disk= raid5_remove_disk, | |
4917 | .spare_active = raid5_spare_active, | |
4918 | .sync_request = sync_request, | |
4919 | .resize = raid5_resize, | |
29269553 | 4920 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f N |
4921 | .check_reshape = raid5_check_reshape, |
4922 | .start_reshape = raid5_start_reshape, | |
29269553 | 4923 | #endif |
72626685 | 4924 | .quiesce = raid5_quiesce, |
1da177e4 LT |
4925 | }; |
4926 | ||
2604b703 | 4927 | static struct mdk_personality raid4_personality = |
1da177e4 | 4928 | { |
2604b703 N |
4929 | .name = "raid4", |
4930 | .level = 4, | |
4931 | .owner = THIS_MODULE, | |
4932 | .make_request = make_request, | |
4933 | .run = run, | |
4934 | .stop = stop, | |
4935 | .status = status, | |
4936 | .error_handler = error, | |
4937 | .hot_add_disk = raid5_add_disk, | |
4938 | .hot_remove_disk= raid5_remove_disk, | |
4939 | .spare_active = raid5_spare_active, | |
4940 | .sync_request = sync_request, | |
4941 | .resize = raid5_resize, | |
3d37890b N |
4942 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4943 | .check_reshape = raid5_check_reshape, | |
4944 | .start_reshape = raid5_start_reshape, | |
4945 | #endif | |
2604b703 N |
4946 | .quiesce = raid5_quiesce, |
4947 | }; | |
4948 | ||
4949 | static int __init raid5_init(void) | |
4950 | { | |
16a53ecc N |
4951 | int e; |
4952 | ||
4953 | e = raid6_select_algo(); | |
4954 | if ( e ) | |
4955 | return e; | |
4956 | register_md_personality(&raid6_personality); | |
2604b703 N |
4957 | register_md_personality(&raid5_personality); |
4958 | register_md_personality(&raid4_personality); | |
4959 | return 0; | |
1da177e4 LT |
4960 | } |
4961 | ||
2604b703 | 4962 | static void raid5_exit(void) |
1da177e4 | 4963 | { |
16a53ecc | 4964 | unregister_md_personality(&raid6_personality); |
2604b703 N |
4965 | unregister_md_personality(&raid5_personality); |
4966 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
4967 | } |
4968 | ||
4969 | module_init(raid5_init); | |
4970 | module_exit(raid5_exit); | |
4971 | MODULE_LICENSE("GPL"); | |
4972 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
4973 | MODULE_ALIAS("md-raid5"); |
4974 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
4975 | MODULE_ALIAS("md-level-5"); |
4976 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
4977 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
4978 | MODULE_ALIAS("md-raid6"); | |
4979 | MODULE_ALIAS("md-level-6"); | |
4980 | ||
4981 | /* This used to be two separate modules, they were: */ | |
4982 | MODULE_ALIAS("raid5"); | |
4983 | MODULE_ALIAS("raid6"); |