blob: 089a32604305c339b9a1fb20019fec8b15352c9d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
5 *
6 * RAID-5 management functions.
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2, or (at your option)
11 * any later version.
12 *
13 * You should have received a copy of the GNU General Public License
14 * (for example /usr/src/linux/COPYING); if not, write to the Free
15 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
16 */
17
18
19#include <linux/config.h>
20#include <linux/module.h>
21#include <linux/slab.h>
22#include <linux/raid/raid5.h>
23#include <linux/highmem.h>
24#include <linux/bitops.h>
NeilBrownf6705572006-03-27 01:18:11 -080025#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/atomic.h>
27
NeilBrown72626682005-09-09 16:23:54 -070028#include <linux/raid/bitmap.h>
29
Linus Torvalds1da177e2005-04-16 15:20:36 -070030/*
31 * Stripe cache
32 */
33
34#define NR_STRIPES 256
35#define STRIPE_SIZE PAGE_SIZE
36#define STRIPE_SHIFT (PAGE_SHIFT - 9)
37#define STRIPE_SECTORS (STRIPE_SIZE>>9)
38#define IO_THRESHOLD 1
NeilBrownfccddba2006-01-06 00:20:33 -080039#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#define HASH_MASK (NR_HASH - 1)
41
NeilBrownfccddba2006-01-06 00:20:33 -080042#define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44/* bio's attached to a stripe+device for I/O are linked together in bi_sector
45 * order without overlap. There may be several bio's per stripe+device, and
46 * a bio could span several devices.
47 * When walking this list for a particular stripe+device, we must never proceed
48 * beyond a bio that extends past this device, as the next bio might no longer
49 * be valid.
50 * This macro is used to determine the 'next' bio in the list, given the sector
51 * of the current stripe+device
52 */
53#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
54/*
55 * The following can be used to debug the driver
56 */
57#define RAID5_DEBUG 0
58#define RAID5_PARANOIA 1
59#if RAID5_PARANOIA && defined(CONFIG_SMP)
60# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
61#else
62# define CHECK_DEVLOCK()
63#endif
64
65#define PRINTK(x...) ((void)(RAID5_DEBUG && printk(x)))
66#if RAID5_DEBUG
67#define inline
68#define __inline__
69#endif
70
71static void print_raid5_conf (raid5_conf_t *conf);
72
Arjan van de Ven858119e2006-01-14 13:20:43 -080073static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 if (atomic_dec_and_test(&sh->count)) {
76 if (!list_empty(&sh->lru))
77 BUG();
78 if (atomic_read(&conf->active_stripes)==0)
79 BUG();
80 if (test_bit(STRIPE_HANDLE, &sh->state)) {
81 if (test_bit(STRIPE_DELAYED, &sh->state))
82 list_add_tail(&sh->lru, &conf->delayed_list);
NeilBrown72626682005-09-09 16:23:54 -070083 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
84 conf->seq_write == sh->bm_seq)
85 list_add_tail(&sh->lru, &conf->bitmap_list);
86 else {
87 clear_bit(STRIPE_BIT_DELAY, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 list_add_tail(&sh->lru, &conf->handle_list);
NeilBrown72626682005-09-09 16:23:54 -070089 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 md_wakeup_thread(conf->mddev->thread);
91 } else {
92 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
93 atomic_dec(&conf->preread_active_stripes);
94 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
95 md_wakeup_thread(conf->mddev->thread);
96 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 atomic_dec(&conf->active_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -080098 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
99 list_add_tail(&sh->lru, &conf->inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 wake_up(&conf->wait_for_stripe);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 }
103 }
104}
105static void release_stripe(struct stripe_head *sh)
106{
107 raid5_conf_t *conf = sh->raid_conf;
108 unsigned long flags;
109
110 spin_lock_irqsave(&conf->device_lock, flags);
111 __release_stripe(conf, sh);
112 spin_unlock_irqrestore(&conf->device_lock, flags);
113}
114
NeilBrownfccddba2006-01-06 00:20:33 -0800115static inline void remove_hash(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
117 PRINTK("remove_hash(), stripe %llu\n", (unsigned long long)sh->sector);
118
NeilBrownfccddba2006-01-06 00:20:33 -0800119 hlist_del_init(&sh->hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120}
121
Arjan van de Ven858119e2006-01-14 13:20:43 -0800122static void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123{
NeilBrownfccddba2006-01-06 00:20:33 -0800124 struct hlist_head *hp = stripe_hash(conf, sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
126 PRINTK("insert_hash(), stripe %llu\n", (unsigned long long)sh->sector);
127
128 CHECK_DEVLOCK();
NeilBrownfccddba2006-01-06 00:20:33 -0800129 hlist_add_head(&sh->hash, hp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130}
131
132
133/* find an idle stripe, make sure it is unhashed, and return it. */
134static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
135{
136 struct stripe_head *sh = NULL;
137 struct list_head *first;
138
139 CHECK_DEVLOCK();
140 if (list_empty(&conf->inactive_list))
141 goto out;
142 first = conf->inactive_list.next;
143 sh = list_entry(first, struct stripe_head, lru);
144 list_del_init(first);
145 remove_hash(sh);
146 atomic_inc(&conf->active_stripes);
147out:
148 return sh;
149}
150
151static void shrink_buffers(struct stripe_head *sh, int num)
152{
153 struct page *p;
154 int i;
155
156 for (i=0; i<num ; i++) {
157 p = sh->dev[i].page;
158 if (!p)
159 continue;
160 sh->dev[i].page = NULL;
NeilBrown2d1f3b52006-01-06 00:20:31 -0800161 put_page(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 }
163}
164
165static int grow_buffers(struct stripe_head *sh, int num)
166{
167 int i;
168
169 for (i=0; i<num; i++) {
170 struct page *page;
171
172 if (!(page = alloc_page(GFP_KERNEL))) {
173 return 1;
174 }
175 sh->dev[i].page = page;
176 }
177 return 0;
178}
179
180static void raid5_build_block (struct stripe_head *sh, int i);
181
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800182static void init_stripe(struct stripe_head *sh, sector_t sector, int pd_idx, int disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
184 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800185 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
187 if (atomic_read(&sh->count) != 0)
188 BUG();
189 if (test_bit(STRIPE_HANDLE, &sh->state))
190 BUG();
191
192 CHECK_DEVLOCK();
193 PRINTK("init_stripe called, stripe %llu\n",
194 (unsigned long long)sh->sector);
195
196 remove_hash(sh);
197
198 sh->sector = sector;
199 sh->pd_idx = pd_idx;
200 sh->state = 0;
201
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800202 sh->disks = disks;
203
204 for (i = sh->disks; i--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 struct r5dev *dev = &sh->dev[i];
206
207 if (dev->toread || dev->towrite || dev->written ||
208 test_bit(R5_LOCKED, &dev->flags)) {
209 printk("sector=%llx i=%d %p %p %p %d\n",
210 (unsigned long long)sh->sector, i, dev->toread,
211 dev->towrite, dev->written,
212 test_bit(R5_LOCKED, &dev->flags));
213 BUG();
214 }
215 dev->flags = 0;
216 raid5_build_block(sh, i);
217 }
218 insert_hash(conf, sh);
219}
220
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800221static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, int disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
223 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -0800224 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
226 CHECK_DEVLOCK();
227 PRINTK("__find_stripe, sector %llu\n", (unsigned long long)sector);
NeilBrownfccddba2006-01-06 00:20:33 -0800228 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800229 if (sh->sector == sector && sh->disks == disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return sh;
231 PRINTK("__stripe %llu not in cache\n", (unsigned long long)sector);
232 return NULL;
233}
234
235static void unplug_slaves(mddev_t *mddev);
236static void raid5_unplug_device(request_queue_t *q);
237
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800238static struct stripe_head *get_active_stripe(raid5_conf_t *conf, sector_t sector, int disks,
239 int pd_idx, int noblock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
241 struct stripe_head *sh;
242
243 PRINTK("get_stripe, sector %llu\n", (unsigned long long)sector);
244
245 spin_lock_irq(&conf->device_lock);
246
247 do {
NeilBrown72626682005-09-09 16:23:54 -0700248 wait_event_lock_irq(conf->wait_for_stripe,
249 conf->quiesce == 0,
250 conf->device_lock, /* nothing */);
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800251 sh = __find_stripe(conf, sector, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (!sh) {
253 if (!conf->inactive_blocked)
254 sh = get_free_stripe(conf);
255 if (noblock && sh == NULL)
256 break;
257 if (!sh) {
258 conf->inactive_blocked = 1;
259 wait_event_lock_irq(conf->wait_for_stripe,
260 !list_empty(&conf->inactive_list) &&
NeilBrown50368052005-12-12 02:39:17 -0800261 (atomic_read(&conf->active_stripes)
262 < (conf->max_nr_stripes *3/4)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 || !conf->inactive_blocked),
264 conf->device_lock,
265 unplug_slaves(conf->mddev);
266 );
267 conf->inactive_blocked = 0;
268 } else
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800269 init_stripe(sh, sector, pd_idx, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 } else {
271 if (atomic_read(&sh->count)) {
272 if (!list_empty(&sh->lru))
273 BUG();
274 } else {
275 if (!test_bit(STRIPE_HANDLE, &sh->state))
276 atomic_inc(&conf->active_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800277 if (!list_empty(&sh->lru))
278 list_del_init(&sh->lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 }
280 }
281 } while (sh == NULL);
282
283 if (sh)
284 atomic_inc(&sh->count);
285
286 spin_unlock_irq(&conf->device_lock);
287 return sh;
288}
289
NeilBrown3f294f42005-11-08 21:39:25 -0800290static int grow_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
292 struct stripe_head *sh;
NeilBrown3f294f42005-11-08 21:39:25 -0800293 sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL);
294 if (!sh)
295 return 0;
296 memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev));
297 sh->raid_conf = conf;
298 spin_lock_init(&sh->lock);
299
300 if (grow_buffers(sh, conf->raid_disks)) {
301 shrink_buffers(sh, conf->raid_disks);
302 kmem_cache_free(conf->slab_cache, sh);
303 return 0;
304 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800305 sh->disks = conf->raid_disks;
NeilBrown3f294f42005-11-08 21:39:25 -0800306 /* we just created an active stripe so... */
307 atomic_set(&sh->count, 1);
308 atomic_inc(&conf->active_stripes);
309 INIT_LIST_HEAD(&sh->lru);
310 release_stripe(sh);
311 return 1;
312}
313
314static int grow_stripes(raid5_conf_t *conf, int num)
315{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 kmem_cache_t *sc;
317 int devs = conf->raid_disks;
318
NeilBrownad01c9e2006-03-27 01:18:07 -0800319 sprintf(conf->cache_name[0], "raid5/%s", mdname(conf->mddev));
320 sprintf(conf->cache_name[1], "raid5/%s-alt", mdname(conf->mddev));
321 conf->active_name = 0;
322 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
324 0, 0, NULL, NULL);
325 if (!sc)
326 return 1;
327 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -0800328 conf->pool_size = devs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 while (num--) {
NeilBrown3f294f42005-11-08 21:39:25 -0800330 if (!grow_one_stripe(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 }
333 return 0;
334}
NeilBrown29269552006-03-27 01:18:10 -0800335
336#ifdef CONFIG_MD_RAID5_RESHAPE
NeilBrownad01c9e2006-03-27 01:18:07 -0800337static int resize_stripes(raid5_conf_t *conf, int newsize)
338{
339 /* Make all the stripes able to hold 'newsize' devices.
340 * New slots in each stripe get 'page' set to a new page.
341 *
342 * This happens in stages:
343 * 1/ create a new kmem_cache and allocate the required number of
344 * stripe_heads.
345 * 2/ gather all the old stripe_heads and tranfer the pages across
346 * to the new stripe_heads. This will have the side effect of
347 * freezing the array as once all stripe_heads have been collected,
348 * no IO will be possible. Old stripe heads are freed once their
349 * pages have been transferred over, and the old kmem_cache is
350 * freed when all stripes are done.
351 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
352 * we simple return a failre status - no need to clean anything up.
353 * 4/ allocate new pages for the new slots in the new stripe_heads.
354 * If this fails, we don't bother trying the shrink the
355 * stripe_heads down again, we just leave them as they are.
356 * As each stripe_head is processed the new one is released into
357 * active service.
358 *
359 * Once step2 is started, we cannot afford to wait for a write,
360 * so we use GFP_NOIO allocations.
361 */
362 struct stripe_head *osh, *nsh;
363 LIST_HEAD(newstripes);
364 struct disk_info *ndisks;
365 int err = 0;
366 kmem_cache_t *sc;
367 int i;
368
369 if (newsize <= conf->pool_size)
370 return 0; /* never bother to shrink */
371
372 /* Step 1 */
373 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
374 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
375 0, 0, NULL, NULL);
376 if (!sc)
377 return -ENOMEM;
378
379 for (i = conf->max_nr_stripes; i; i--) {
380 nsh = kmem_cache_alloc(sc, GFP_KERNEL);
381 if (!nsh)
382 break;
383
384 memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev));
385
386 nsh->raid_conf = conf;
387 spin_lock_init(&nsh->lock);
388
389 list_add(&nsh->lru, &newstripes);
390 }
391 if (i) {
392 /* didn't get enough, give up */
393 while (!list_empty(&newstripes)) {
394 nsh = list_entry(newstripes.next, struct stripe_head, lru);
395 list_del(&nsh->lru);
396 kmem_cache_free(sc, nsh);
397 }
398 kmem_cache_destroy(sc);
399 return -ENOMEM;
400 }
401 /* Step 2 - Must use GFP_NOIO now.
402 * OK, we have enough stripes, start collecting inactive
403 * stripes and copying them over
404 */
405 list_for_each_entry(nsh, &newstripes, lru) {
406 spin_lock_irq(&conf->device_lock);
407 wait_event_lock_irq(conf->wait_for_stripe,
408 !list_empty(&conf->inactive_list),
409 conf->device_lock,
410 unplug_slaves(conf->mddev);
411 );
412 osh = get_free_stripe(conf);
413 spin_unlock_irq(&conf->device_lock);
414 atomic_set(&nsh->count, 1);
415 for(i=0; i<conf->pool_size; i++)
416 nsh->dev[i].page = osh->dev[i].page;
417 for( ; i<newsize; i++)
418 nsh->dev[i].page = NULL;
419 kmem_cache_free(conf->slab_cache, osh);
420 }
421 kmem_cache_destroy(conf->slab_cache);
422
423 /* Step 3.
424 * At this point, we are holding all the stripes so the array
425 * is completely stalled, so now is a good time to resize
426 * conf->disks.
427 */
428 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
429 if (ndisks) {
430 for (i=0; i<conf->raid_disks; i++)
431 ndisks[i] = conf->disks[i];
432 kfree(conf->disks);
433 conf->disks = ndisks;
434 } else
435 err = -ENOMEM;
436
437 /* Step 4, return new stripes to service */
438 while(!list_empty(&newstripes)) {
439 nsh = list_entry(newstripes.next, struct stripe_head, lru);
440 list_del_init(&nsh->lru);
441 for (i=conf->raid_disks; i < newsize; i++)
442 if (nsh->dev[i].page == NULL) {
443 struct page *p = alloc_page(GFP_NOIO);
444 nsh->dev[i].page = p;
445 if (!p)
446 err = -ENOMEM;
447 }
448 release_stripe(nsh);
449 }
450 /* critical section pass, GFP_NOIO no longer needed */
451
452 conf->slab_cache = sc;
453 conf->active_name = 1-conf->active_name;
454 conf->pool_size = newsize;
455 return err;
456}
NeilBrown29269552006-03-27 01:18:10 -0800457#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
NeilBrown3f294f42005-11-08 21:39:25 -0800459static int drop_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
461 struct stripe_head *sh;
462
NeilBrown3f294f42005-11-08 21:39:25 -0800463 spin_lock_irq(&conf->device_lock);
464 sh = get_free_stripe(conf);
465 spin_unlock_irq(&conf->device_lock);
466 if (!sh)
467 return 0;
468 if (atomic_read(&sh->count))
469 BUG();
NeilBrownad01c9e2006-03-27 01:18:07 -0800470 shrink_buffers(sh, conf->pool_size);
NeilBrown3f294f42005-11-08 21:39:25 -0800471 kmem_cache_free(conf->slab_cache, sh);
472 atomic_dec(&conf->active_stripes);
473 return 1;
474}
475
476static void shrink_stripes(raid5_conf_t *conf)
477{
478 while (drop_one_stripe(conf))
479 ;
480
NeilBrown29fc7e32006-02-03 03:03:41 -0800481 if (conf->slab_cache)
482 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 conf->slab_cache = NULL;
484}
485
NeilBrown4e5314b2005-11-08 21:39:22 -0800486static int raid5_end_read_request(struct bio * bi, unsigned int bytes_done,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 int error)
488{
489 struct stripe_head *sh = bi->bi_private;
490 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800491 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
493
494 if (bi->bi_size)
495 return 1;
496
497 for (i=0 ; i<disks; i++)
498 if (bi == &sh->dev[i].req)
499 break;
500
501 PRINTK("end_read_request %llu/%d, count: %d, uptodate %d.\n",
502 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
503 uptodate);
504 if (i == disks) {
505 BUG();
506 return 0;
507 }
508
509 if (uptodate) {
510#if 0
511 struct bio *bio;
512 unsigned long flags;
513 spin_lock_irqsave(&conf->device_lock, flags);
514 /* we can return a buffer if we bypassed the cache or
515 * if the top buffer is not in highmem. If there are
516 * multiple buffers, leave the extra work to
517 * handle_stripe
518 */
519 buffer = sh->bh_read[i];
520 if (buffer &&
521 (!PageHighMem(buffer->b_page)
522 || buffer->b_page == bh->b_page )
523 ) {
524 sh->bh_read[i] = buffer->b_reqnext;
525 buffer->b_reqnext = NULL;
526 } else
527 buffer = NULL;
528 spin_unlock_irqrestore(&conf->device_lock, flags);
529 if (sh->bh_page[i]==bh->b_page)
530 set_buffer_uptodate(bh);
531 if (buffer) {
532 if (buffer->b_page != bh->b_page)
533 memcpy(buffer->b_data, bh->b_data, bh->b_size);
534 buffer->b_end_io(buffer, 1);
535 }
536#else
537 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -0800538#endif
539 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown14f8d262006-01-06 00:20:14 -0800540 printk(KERN_INFO "raid5: read error corrected!!\n");
NeilBrown4e5314b2005-11-08 21:39:22 -0800541 clear_bit(R5_ReadError, &sh->dev[i].flags);
542 clear_bit(R5_ReWrite, &sh->dev[i].flags);
543 }
NeilBrownba22dcb2005-11-08 21:39:31 -0800544 if (atomic_read(&conf->disks[i].rdev->read_errors))
545 atomic_set(&conf->disks[i].rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 } else {
NeilBrownba22dcb2005-11-08 21:39:31 -0800547 int retry = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownba22dcb2005-11-08 21:39:31 -0800549 atomic_inc(&conf->disks[i].rdev->read_errors);
550 if (conf->mddev->degraded)
NeilBrown14f8d262006-01-06 00:20:14 -0800551 printk(KERN_WARNING "raid5: read error not correctable.\n");
NeilBrownba22dcb2005-11-08 21:39:31 -0800552 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
NeilBrown4e5314b2005-11-08 21:39:22 -0800553 /* Oh, no!!! */
NeilBrown14f8d262006-01-06 00:20:14 -0800554 printk(KERN_WARNING "raid5: read error NOT corrected!!\n");
NeilBrownba22dcb2005-11-08 21:39:31 -0800555 else if (atomic_read(&conf->disks[i].rdev->read_errors)
556 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -0800557 printk(KERN_WARNING
558 "raid5: Too many read errors, failing device.\n");
NeilBrownba22dcb2005-11-08 21:39:31 -0800559 else
560 retry = 1;
561 if (retry)
562 set_bit(R5_ReadError, &sh->dev[i].flags);
563 else {
NeilBrown4e5314b2005-11-08 21:39:22 -0800564 clear_bit(R5_ReadError, &sh->dev[i].flags);
565 clear_bit(R5_ReWrite, &sh->dev[i].flags);
566 md_error(conf->mddev, conf->disks[i].rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -0800567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
569 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
570#if 0
571 /* must restore b_page before unlocking buffer... */
572 if (sh->bh_page[i] != bh->b_page) {
573 bh->b_page = sh->bh_page[i];
574 bh->b_data = page_address(bh->b_page);
575 clear_buffer_uptodate(bh);
576 }
577#endif
578 clear_bit(R5_LOCKED, &sh->dev[i].flags);
579 set_bit(STRIPE_HANDLE, &sh->state);
580 release_stripe(sh);
581 return 0;
582}
583
584static int raid5_end_write_request (struct bio *bi, unsigned int bytes_done,
585 int error)
586{
587 struct stripe_head *sh = bi->bi_private;
588 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800589 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 unsigned long flags;
591 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
592
593 if (bi->bi_size)
594 return 1;
595
596 for (i=0 ; i<disks; i++)
597 if (bi == &sh->dev[i].req)
598 break;
599
600 PRINTK("end_write_request %llu/%d, count %d, uptodate: %d.\n",
601 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
602 uptodate);
603 if (i == disks) {
604 BUG();
605 return 0;
606 }
607
608 spin_lock_irqsave(&conf->device_lock, flags);
609 if (!uptodate)
610 md_error(conf->mddev, conf->disks[i].rdev);
611
612 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
613
614 clear_bit(R5_LOCKED, &sh->dev[i].flags);
615 set_bit(STRIPE_HANDLE, &sh->state);
616 __release_stripe(conf, sh);
617 spin_unlock_irqrestore(&conf->device_lock, flags);
618 return 0;
619}
620
621
622static sector_t compute_blocknr(struct stripe_head *sh, int i);
623
624static void raid5_build_block (struct stripe_head *sh, int i)
625{
626 struct r5dev *dev = &sh->dev[i];
627
628 bio_init(&dev->req);
629 dev->req.bi_io_vec = &dev->vec;
630 dev->req.bi_vcnt++;
631 dev->req.bi_max_vecs++;
632 dev->vec.bv_page = dev->page;
633 dev->vec.bv_len = STRIPE_SIZE;
634 dev->vec.bv_offset = 0;
635
636 dev->req.bi_sector = sh->sector;
637 dev->req.bi_private = sh;
638
639 dev->flags = 0;
640 if (i != sh->pd_idx)
641 dev->sector = compute_blocknr(sh, i);
642}
643
644static void error(mddev_t *mddev, mdk_rdev_t *rdev)
645{
646 char b[BDEVNAME_SIZE];
647 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
648 PRINTK("raid5: error called\n");
649
NeilBrownb2d444d2005-11-08 21:39:31 -0800650 if (!test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 mddev->sb_dirty = 1;
NeilBrownb2d444d2005-11-08 21:39:31 -0800652 if (test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 conf->working_disks--;
654 mddev->degraded++;
655 conf->failed_disks++;
NeilBrownb2d444d2005-11-08 21:39:31 -0800656 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 /*
658 * if recovery was running, make sure it aborts.
659 */
660 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
661 }
NeilBrownb2d444d2005-11-08 21:39:31 -0800662 set_bit(Faulty, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 printk (KERN_ALERT
664 "raid5: Disk failure on %s, disabling device."
665 " Operation continuing on %d devices\n",
666 bdevname(rdev->bdev,b), conf->working_disks);
667 }
668}
669
670/*
671 * Input: a 'big' sector number,
672 * Output: index of the data and parity disk, and the sector # in them.
673 */
674static sector_t raid5_compute_sector(sector_t r_sector, unsigned int raid_disks,
675 unsigned int data_disks, unsigned int * dd_idx,
676 unsigned int * pd_idx, raid5_conf_t *conf)
677{
678 long stripe;
679 unsigned long chunk_number;
680 unsigned int chunk_offset;
681 sector_t new_sector;
682 int sectors_per_chunk = conf->chunk_size >> 9;
683
684 /* First compute the information on this sector */
685
686 /*
687 * Compute the chunk number and the sector offset inside the chunk
688 */
689 chunk_offset = sector_div(r_sector, sectors_per_chunk);
690 chunk_number = r_sector;
691 BUG_ON(r_sector != chunk_number);
692
693 /*
694 * Compute the stripe number
695 */
696 stripe = chunk_number / data_disks;
697
698 /*
699 * Compute the data disk and parity disk indexes inside the stripe
700 */
701 *dd_idx = chunk_number % data_disks;
702
703 /*
704 * Select the parity disk based on the user selected algorithm.
705 */
706 if (conf->level == 4)
707 *pd_idx = data_disks;
708 else switch (conf->algorithm) {
709 case ALGORITHM_LEFT_ASYMMETRIC:
710 *pd_idx = data_disks - stripe % raid_disks;
711 if (*dd_idx >= *pd_idx)
712 (*dd_idx)++;
713 break;
714 case ALGORITHM_RIGHT_ASYMMETRIC:
715 *pd_idx = stripe % raid_disks;
716 if (*dd_idx >= *pd_idx)
717 (*dd_idx)++;
718 break;
719 case ALGORITHM_LEFT_SYMMETRIC:
720 *pd_idx = data_disks - stripe % raid_disks;
721 *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks;
722 break;
723 case ALGORITHM_RIGHT_SYMMETRIC:
724 *pd_idx = stripe % raid_disks;
725 *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks;
726 break;
727 default:
NeilBrown14f8d262006-01-06 00:20:14 -0800728 printk(KERN_ERR "raid5: unsupported algorithm %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 conf->algorithm);
730 }
731
732 /*
733 * Finally, compute the new sector number
734 */
735 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
736 return new_sector;
737}
738
739
740static sector_t compute_blocknr(struct stripe_head *sh, int i)
741{
742 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800743 int raid_disks = sh->disks, data_disks = raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 sector_t new_sector = sh->sector, check;
745 int sectors_per_chunk = conf->chunk_size >> 9;
746 sector_t stripe;
747 int chunk_offset;
748 int chunk_number, dummy1, dummy2, dd_idx = i;
749 sector_t r_sector;
750
751 chunk_offset = sector_div(new_sector, sectors_per_chunk);
752 stripe = new_sector;
753 BUG_ON(new_sector != stripe);
754
755
756 switch (conf->algorithm) {
757 case ALGORITHM_LEFT_ASYMMETRIC:
758 case ALGORITHM_RIGHT_ASYMMETRIC:
759 if (i > sh->pd_idx)
760 i--;
761 break;
762 case ALGORITHM_LEFT_SYMMETRIC:
763 case ALGORITHM_RIGHT_SYMMETRIC:
764 if (i < sh->pd_idx)
765 i += raid_disks;
766 i -= (sh->pd_idx + 1);
767 break;
768 default:
NeilBrown14f8d262006-01-06 00:20:14 -0800769 printk(KERN_ERR "raid5: unsupported algorithm %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 conf->algorithm);
771 }
772
773 chunk_number = stripe * data_disks + i;
774 r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset;
775
776 check = raid5_compute_sector (r_sector, raid_disks, data_disks, &dummy1, &dummy2, conf);
777 if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) {
NeilBrown14f8d262006-01-06 00:20:14 -0800778 printk(KERN_ERR "compute_blocknr: map not correct\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return 0;
780 }
781 return r_sector;
782}
783
784
785
786/*
787 * Copy data between a page in the stripe cache, and a bio.
788 * There are no alignment or size guarantees between the page or the
789 * bio except that there is some overlap.
790 * All iovecs in the bio must be considered.
791 */
792static void copy_data(int frombio, struct bio *bio,
793 struct page *page,
794 sector_t sector)
795{
796 char *pa = page_address(page);
797 struct bio_vec *bvl;
798 int i;
799 int page_offset;
800
801 if (bio->bi_sector >= sector)
802 page_offset = (signed)(bio->bi_sector - sector) * 512;
803 else
804 page_offset = (signed)(sector - bio->bi_sector) * -512;
805 bio_for_each_segment(bvl, bio, i) {
806 int len = bio_iovec_idx(bio,i)->bv_len;
807 int clen;
808 int b_offset = 0;
809
810 if (page_offset < 0) {
811 b_offset = -page_offset;
812 page_offset += b_offset;
813 len -= b_offset;
814 }
815
816 if (len > 0 && page_offset + len > STRIPE_SIZE)
817 clen = STRIPE_SIZE - page_offset;
818 else clen = len;
819
820 if (clen > 0) {
821 char *ba = __bio_kmap_atomic(bio, i, KM_USER0);
822 if (frombio)
823 memcpy(pa+page_offset, ba+b_offset, clen);
824 else
825 memcpy(ba+b_offset, pa+page_offset, clen);
826 __bio_kunmap_atomic(ba, KM_USER0);
827 }
828 if (clen < len) /* hit end of page */
829 break;
830 page_offset += len;
831 }
832}
833
834#define check_xor() do { \
835 if (count == MAX_XOR_BLOCKS) { \
836 xor_block(count, STRIPE_SIZE, ptr); \
837 count = 1; \
838 } \
839 } while(0)
840
841
842static void compute_block(struct stripe_head *sh, int dd_idx)
843{
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800844 int i, count, disks = sh->disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 void *ptr[MAX_XOR_BLOCKS], *p;
846
847 PRINTK("compute_block, stripe %llu, idx %d\n",
848 (unsigned long long)sh->sector, dd_idx);
849
850 ptr[0] = page_address(sh->dev[dd_idx].page);
851 memset(ptr[0], 0, STRIPE_SIZE);
852 count = 1;
853 for (i = disks ; i--; ) {
854 if (i == dd_idx)
855 continue;
856 p = page_address(sh->dev[i].page);
857 if (test_bit(R5_UPTODATE, &sh->dev[i].flags))
858 ptr[count++] = p;
859 else
NeilBrown14f8d262006-01-06 00:20:14 -0800860 printk(KERN_ERR "compute_block() %d, stripe %llu, %d"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 " not present\n", dd_idx,
862 (unsigned long long)sh->sector, i);
863
864 check_xor();
865 }
866 if (count != 1)
867 xor_block(count, STRIPE_SIZE, ptr);
868 set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags);
869}
870
871static void compute_parity(struct stripe_head *sh, int method)
872{
873 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800874 int i, pd_idx = sh->pd_idx, disks = sh->disks, count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 void *ptr[MAX_XOR_BLOCKS];
876 struct bio *chosen;
877
878 PRINTK("compute_parity, stripe %llu, method %d\n",
879 (unsigned long long)sh->sector, method);
880
881 count = 1;
882 ptr[0] = page_address(sh->dev[pd_idx].page);
883 switch(method) {
884 case READ_MODIFY_WRITE:
885 if (!test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags))
886 BUG();
887 for (i=disks ; i-- ;) {
888 if (i==pd_idx)
889 continue;
890 if (sh->dev[i].towrite &&
891 test_bit(R5_UPTODATE, &sh->dev[i].flags)) {
892 ptr[count++] = page_address(sh->dev[i].page);
893 chosen = sh->dev[i].towrite;
894 sh->dev[i].towrite = NULL;
895
896 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
897 wake_up(&conf->wait_for_overlap);
898
899 if (sh->dev[i].written) BUG();
900 sh->dev[i].written = chosen;
901 check_xor();
902 }
903 }
904 break;
905 case RECONSTRUCT_WRITE:
906 memset(ptr[0], 0, STRIPE_SIZE);
907 for (i= disks; i-- ;)
908 if (i!=pd_idx && sh->dev[i].towrite) {
909 chosen = sh->dev[i].towrite;
910 sh->dev[i].towrite = NULL;
911
912 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
913 wake_up(&conf->wait_for_overlap);
914
915 if (sh->dev[i].written) BUG();
916 sh->dev[i].written = chosen;
917 }
918 break;
919 case CHECK_PARITY:
920 break;
921 }
922 if (count>1) {
923 xor_block(count, STRIPE_SIZE, ptr);
924 count = 1;
925 }
926
927 for (i = disks; i--;)
928 if (sh->dev[i].written) {
929 sector_t sector = sh->dev[i].sector;
930 struct bio *wbi = sh->dev[i].written;
931 while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) {
932 copy_data(1, wbi, sh->dev[i].page, sector);
933 wbi = r5_next_bio(wbi, sector);
934 }
935
936 set_bit(R5_LOCKED, &sh->dev[i].flags);
937 set_bit(R5_UPTODATE, &sh->dev[i].flags);
938 }
939
940 switch(method) {
941 case RECONSTRUCT_WRITE:
942 case CHECK_PARITY:
943 for (i=disks; i--;)
944 if (i != pd_idx) {
945 ptr[count++] = page_address(sh->dev[i].page);
946 check_xor();
947 }
948 break;
949 case READ_MODIFY_WRITE:
950 for (i = disks; i--;)
951 if (sh->dev[i].written) {
952 ptr[count++] = page_address(sh->dev[i].page);
953 check_xor();
954 }
955 }
956 if (count != 1)
957 xor_block(count, STRIPE_SIZE, ptr);
958
959 if (method != CHECK_PARITY) {
960 set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
961 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
962 } else
963 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
964}
965
966/*
967 * Each stripe/dev can have one or more bion attached.
968 * toread/towrite point to the first in a chain.
969 * The bi_next chain must be in order.
970 */
971static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
972{
973 struct bio **bip;
974 raid5_conf_t *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -0700975 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977 PRINTK("adding bh b#%llu to stripe s#%llu\n",
978 (unsigned long long)bi->bi_sector,
979 (unsigned long long)sh->sector);
980
981
982 spin_lock(&sh->lock);
983 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -0700984 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 bip = &sh->dev[dd_idx].towrite;
NeilBrown72626682005-09-09 16:23:54 -0700986 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
987 firstwrite = 1;
988 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 bip = &sh->dev[dd_idx].toread;
990 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
991 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
992 goto overlap;
993 bip = & (*bip)->bi_next;
994 }
995 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
996 goto overlap;
997
998 if (*bip && bi->bi_next && (*bip) != bi->bi_next)
999 BUG();
1000 if (*bip)
1001 bi->bi_next = *bip;
1002 *bip = bi;
1003 bi->bi_phys_segments ++;
1004 spin_unlock_irq(&conf->device_lock);
1005 spin_unlock(&sh->lock);
1006
1007 PRINTK("added bi b#%llu to stripe s#%llu, disk %d.\n",
1008 (unsigned long long)bi->bi_sector,
1009 (unsigned long long)sh->sector, dd_idx);
1010
NeilBrown72626682005-09-09 16:23:54 -07001011 if (conf->mddev->bitmap && firstwrite) {
1012 sh->bm_seq = conf->seq_write;
1013 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
1014 STRIPE_SECTORS, 0);
1015 set_bit(STRIPE_BIT_DELAY, &sh->state);
1016 }
1017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (forwrite) {
1019 /* check if page is covered */
1020 sector_t sector = sh->dev[dd_idx].sector;
1021 for (bi=sh->dev[dd_idx].towrite;
1022 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
1023 bi && bi->bi_sector <= sector;
1024 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
1025 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
1026 sector = bi->bi_sector + (bi->bi_size>>9);
1027 }
1028 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
1029 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
1030 }
1031 return 1;
1032
1033 overlap:
1034 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
1035 spin_unlock_irq(&conf->device_lock);
1036 spin_unlock(&sh->lock);
1037 return 0;
1038}
1039
NeilBrown29269552006-03-27 01:18:10 -08001040static void end_reshape(raid5_conf_t *conf);
1041
NeilBrownccfcc3c2006-03-27 01:18:09 -08001042static int stripe_to_pdidx(sector_t stripe, raid5_conf_t *conf, int disks)
1043{
1044 int sectors_per_chunk = conf->chunk_size >> 9;
1045 sector_t x = stripe;
1046 int pd_idx, dd_idx;
1047 int chunk_offset = sector_div(x, sectors_per_chunk);
1048 stripe = x;
1049 raid5_compute_sector(stripe*(disks-1)*sectors_per_chunk
1050 + chunk_offset, disks, disks-1, &dd_idx, &pd_idx, conf);
1051 return pd_idx;
1052}
1053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055/*
1056 * handle_stripe - do things to a stripe.
1057 *
1058 * We lock the stripe and then examine the state of various bits
1059 * to see what needs to be done.
1060 * Possible results:
1061 * return some read request which now have data
1062 * return some write requests which are safely on disc
1063 * schedule a read on some buffers
1064 * schedule a write of some buffers
1065 * return confirmation of parity correctness
1066 *
1067 * Parity calculations are done inside the stripe lock
1068 * buffers are taken off read_list or write_list, and bh_cache buffers
1069 * get BH_Lock set before the stripe lock is released.
1070 *
1071 */
1072
1073static void handle_stripe(struct stripe_head *sh)
1074{
1075 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001076 int disks = sh->disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 struct bio *return_bi= NULL;
1078 struct bio *bi;
1079 int i;
NeilBrownccfcc3c2006-03-27 01:18:09 -08001080 int syncing, expanding, expanded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 int locked=0, uptodate=0, to_read=0, to_write=0, failed=0, written=0;
1082 int non_overwrite = 0;
1083 int failed_num=0;
1084 struct r5dev *dev;
1085
1086 PRINTK("handling stripe %llu, cnt=%d, pd_idx=%d\n",
1087 (unsigned long long)sh->sector, atomic_read(&sh->count),
1088 sh->pd_idx);
1089
1090 spin_lock(&sh->lock);
1091 clear_bit(STRIPE_HANDLE, &sh->state);
1092 clear_bit(STRIPE_DELAYED, &sh->state);
1093
1094 syncing = test_bit(STRIPE_SYNCING, &sh->state);
NeilBrownccfcc3c2006-03-27 01:18:09 -08001095 expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
1096 expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 /* Now to look around and see what can be done */
1098
NeilBrown9910f162006-01-06 00:20:24 -08001099 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 for (i=disks; i--; ) {
1101 mdk_rdev_t *rdev;
1102 dev = &sh->dev[i];
1103 clear_bit(R5_Insync, &dev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
1105 PRINTK("check %d: state 0x%lx read %p write %p written %p\n",
1106 i, dev->flags, dev->toread, dev->towrite, dev->written);
1107 /* maybe we can reply to a read */
1108 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) {
1109 struct bio *rbi, *rbi2;
1110 PRINTK("Return read for disc %d\n", i);
1111 spin_lock_irq(&conf->device_lock);
1112 rbi = dev->toread;
1113 dev->toread = NULL;
1114 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1115 wake_up(&conf->wait_for_overlap);
1116 spin_unlock_irq(&conf->device_lock);
1117 while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) {
1118 copy_data(0, rbi, dev->page, dev->sector);
1119 rbi2 = r5_next_bio(rbi, dev->sector);
1120 spin_lock_irq(&conf->device_lock);
1121 if (--rbi->bi_phys_segments == 0) {
1122 rbi->bi_next = return_bi;
1123 return_bi = rbi;
1124 }
1125 spin_unlock_irq(&conf->device_lock);
1126 rbi = rbi2;
1127 }
1128 }
1129
1130 /* now count some things */
1131 if (test_bit(R5_LOCKED, &dev->flags)) locked++;
1132 if (test_bit(R5_UPTODATE, &dev->flags)) uptodate++;
1133
1134
1135 if (dev->toread) to_read++;
1136 if (dev->towrite) {
1137 to_write++;
1138 if (!test_bit(R5_OVERWRITE, &dev->flags))
1139 non_overwrite++;
1140 }
1141 if (dev->written) written++;
NeilBrown9910f162006-01-06 00:20:24 -08001142 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -08001143 if (!rdev || !test_bit(In_sync, &rdev->flags)) {
NeilBrown14f8d262006-01-06 00:20:14 -08001144 /* The ReadError flag will just be confusing now */
NeilBrown4e5314b2005-11-08 21:39:22 -08001145 clear_bit(R5_ReadError, &dev->flags);
1146 clear_bit(R5_ReWrite, &dev->flags);
1147 }
NeilBrownb2d444d2005-11-08 21:39:31 -08001148 if (!rdev || !test_bit(In_sync, &rdev->flags)
NeilBrown4e5314b2005-11-08 21:39:22 -08001149 || test_bit(R5_ReadError, &dev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 failed++;
1151 failed_num = i;
1152 } else
1153 set_bit(R5_Insync, &dev->flags);
1154 }
NeilBrown9910f162006-01-06 00:20:24 -08001155 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 PRINTK("locked=%d uptodate=%d to_read=%d"
1157 " to_write=%d failed=%d failed_num=%d\n",
1158 locked, uptodate, to_read, to_write, failed, failed_num);
1159 /* check if the array has lost two devices and, if so, some requests might
1160 * need to be failed
1161 */
1162 if (failed > 1 && to_read+to_write+written) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 for (i=disks; i--; ) {
NeilBrown72626682005-09-09 16:23:54 -07001164 int bitmap_end = 0;
NeilBrown4e5314b2005-11-08 21:39:22 -08001165
1166 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown9910f162006-01-06 00:20:24 -08001167 mdk_rdev_t *rdev;
1168 rcu_read_lock();
1169 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -08001170 if (rdev && test_bit(In_sync, &rdev->flags))
NeilBrown4e5314b2005-11-08 21:39:22 -08001171 /* multiple read failures in one stripe */
1172 md_error(conf->mddev, rdev);
NeilBrown9910f162006-01-06 00:20:24 -08001173 rcu_read_unlock();
NeilBrown4e5314b2005-11-08 21:39:22 -08001174 }
1175
NeilBrown72626682005-09-09 16:23:54 -07001176 spin_lock_irq(&conf->device_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 /* fail all writes first */
1178 bi = sh->dev[i].towrite;
1179 sh->dev[i].towrite = NULL;
NeilBrown72626682005-09-09 16:23:54 -07001180 if (bi) { to_write--; bitmap_end = 1; }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
1183 wake_up(&conf->wait_for_overlap);
1184
1185 while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){
1186 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
1187 clear_bit(BIO_UPTODATE, &bi->bi_flags);
1188 if (--bi->bi_phys_segments == 0) {
1189 md_write_end(conf->mddev);
1190 bi->bi_next = return_bi;
1191 return_bi = bi;
1192 }
1193 bi = nextbi;
1194 }
1195 /* and fail all 'written' */
1196 bi = sh->dev[i].written;
1197 sh->dev[i].written = NULL;
NeilBrown72626682005-09-09 16:23:54 -07001198 if (bi) bitmap_end = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS) {
1200 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
1201 clear_bit(BIO_UPTODATE, &bi->bi_flags);
1202 if (--bi->bi_phys_segments == 0) {
1203 md_write_end(conf->mddev);
1204 bi->bi_next = return_bi;
1205 return_bi = bi;
1206 }
1207 bi = bi2;
1208 }
1209
1210 /* fail any reads if this device is non-operational */
NeilBrown4e5314b2005-11-08 21:39:22 -08001211 if (!test_bit(R5_Insync, &sh->dev[i].flags) ||
1212 test_bit(R5_ReadError, &sh->dev[i].flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 bi = sh->dev[i].toread;
1214 sh->dev[i].toread = NULL;
1215 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
1216 wake_up(&conf->wait_for_overlap);
1217 if (bi) to_read--;
1218 while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){
1219 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
1220 clear_bit(BIO_UPTODATE, &bi->bi_flags);
1221 if (--bi->bi_phys_segments == 0) {
1222 bi->bi_next = return_bi;
1223 return_bi = bi;
1224 }
1225 bi = nextbi;
1226 }
1227 }
NeilBrown72626682005-09-09 16:23:54 -07001228 spin_unlock_irq(&conf->device_lock);
1229 if (bitmap_end)
1230 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
1231 STRIPE_SECTORS, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 }
1234 if (failed > 1 && syncing) {
1235 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
1236 clear_bit(STRIPE_SYNCING, &sh->state);
1237 syncing = 0;
1238 }
1239
1240 /* might be able to return some write requests if the parity block
1241 * is safe, or on a failed drive
1242 */
1243 dev = &sh->dev[sh->pd_idx];
1244 if ( written &&
1245 ( (test_bit(R5_Insync, &dev->flags) && !test_bit(R5_LOCKED, &dev->flags) &&
1246 test_bit(R5_UPTODATE, &dev->flags))
1247 || (failed == 1 && failed_num == sh->pd_idx))
1248 ) {
1249 /* any written block on an uptodate or failed drive can be returned.
1250 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
1251 * never LOCKED, so we don't need to test 'failed' directly.
1252 */
1253 for (i=disks; i--; )
1254 if (sh->dev[i].written) {
1255 dev = &sh->dev[i];
1256 if (!test_bit(R5_LOCKED, &dev->flags) &&
1257 test_bit(R5_UPTODATE, &dev->flags) ) {
1258 /* We can return any write requests */
1259 struct bio *wbi, *wbi2;
NeilBrown72626682005-09-09 16:23:54 -07001260 int bitmap_end = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 PRINTK("Return write for disc %d\n", i);
1262 spin_lock_irq(&conf->device_lock);
1263 wbi = dev->written;
1264 dev->written = NULL;
1265 while (wbi && wbi->bi_sector < dev->sector + STRIPE_SECTORS) {
1266 wbi2 = r5_next_bio(wbi, dev->sector);
1267 if (--wbi->bi_phys_segments == 0) {
1268 md_write_end(conf->mddev);
1269 wbi->bi_next = return_bi;
1270 return_bi = wbi;
1271 }
1272 wbi = wbi2;
1273 }
NeilBrown72626682005-09-09 16:23:54 -07001274 if (dev->towrite == NULL)
1275 bitmap_end = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07001277 if (bitmap_end)
1278 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
1279 STRIPE_SECTORS,
1280 !test_bit(STRIPE_DEGRADED, &sh->state), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 }
1282 }
1283 }
1284
1285 /* Now we might consider reading some blocks, either to check/generate
1286 * parity, or to satisfy requests
1287 * or to load a block that is being partially written.
1288 */
NeilBrownccfcc3c2006-03-27 01:18:09 -08001289 if (to_read || non_overwrite || (syncing && (uptodate < disks)) || expanding) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 for (i=disks; i--;) {
1291 dev = &sh->dev[i];
1292 if (!test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) &&
1293 (dev->toread ||
1294 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
1295 syncing ||
NeilBrownccfcc3c2006-03-27 01:18:09 -08001296 expanding ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 (failed && (sh->dev[failed_num].toread ||
1298 (sh->dev[failed_num].towrite && !test_bit(R5_OVERWRITE, &sh->dev[failed_num].flags))))
1299 )
1300 ) {
1301 /* we would like to get this block, possibly
1302 * by computing it, but we might not be able to
1303 */
1304 if (uptodate == disks-1) {
1305 PRINTK("Computing block %d\n", i);
1306 compute_block(sh, i);
1307 uptodate++;
1308 } else if (test_bit(R5_Insync, &dev->flags)) {
1309 set_bit(R5_LOCKED, &dev->flags);
1310 set_bit(R5_Wantread, &dev->flags);
1311#if 0
1312 /* if I am just reading this block and we don't have
1313 a failed drive, or any pending writes then sidestep the cache */
1314 if (sh->bh_read[i] && !sh->bh_read[i]->b_reqnext &&
1315 ! syncing && !failed && !to_write) {
1316 sh->bh_cache[i]->b_page = sh->bh_read[i]->b_page;
1317 sh->bh_cache[i]->b_data = sh->bh_read[i]->b_data;
1318 }
1319#endif
1320 locked++;
1321 PRINTK("Reading block %d (sync=%d)\n",
1322 i, syncing);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 }
1324 }
1325 }
1326 set_bit(STRIPE_HANDLE, &sh->state);
1327 }
1328
1329 /* now to consider writing and what else, if anything should be read */
1330 if (to_write) {
1331 int rmw=0, rcw=0;
1332 for (i=disks ; i--;) {
1333 /* would I have to read this buffer for read_modify_write */
1334 dev = &sh->dev[i];
1335 if ((dev->towrite || i == sh->pd_idx) &&
1336 (!test_bit(R5_LOCKED, &dev->flags)
1337#if 0
1338|| sh->bh_page[i]!=bh->b_page
1339#endif
1340 ) &&
1341 !test_bit(R5_UPTODATE, &dev->flags)) {
1342 if (test_bit(R5_Insync, &dev->flags)
1343/* && !(!mddev->insync && i == sh->pd_idx) */
1344 )
1345 rmw++;
1346 else rmw += 2*disks; /* cannot read it */
1347 }
1348 /* Would I have to read this buffer for reconstruct_write */
1349 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
1350 (!test_bit(R5_LOCKED, &dev->flags)
1351#if 0
1352|| sh->bh_page[i] != bh->b_page
1353#endif
1354 ) &&
1355 !test_bit(R5_UPTODATE, &dev->flags)) {
1356 if (test_bit(R5_Insync, &dev->flags)) rcw++;
1357 else rcw += 2*disks;
1358 }
1359 }
1360 PRINTK("for sector %llu, rmw=%d rcw=%d\n",
1361 (unsigned long long)sh->sector, rmw, rcw);
1362 set_bit(STRIPE_HANDLE, &sh->state);
1363 if (rmw < rcw && rmw > 0)
1364 /* prefer read-modify-write, but need to get some data */
1365 for (i=disks; i--;) {
1366 dev = &sh->dev[i];
1367 if ((dev->towrite || i == sh->pd_idx) &&
1368 !test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) &&
1369 test_bit(R5_Insync, &dev->flags)) {
1370 if (test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
1371 {
1372 PRINTK("Read_old block %d for r-m-w\n", i);
1373 set_bit(R5_LOCKED, &dev->flags);
1374 set_bit(R5_Wantread, &dev->flags);
1375 locked++;
1376 } else {
1377 set_bit(STRIPE_DELAYED, &sh->state);
1378 set_bit(STRIPE_HANDLE, &sh->state);
1379 }
1380 }
1381 }
1382 if (rcw <= rmw && rcw > 0)
1383 /* want reconstruct write, but need to get some data */
1384 for (i=disks; i--;) {
1385 dev = &sh->dev[i];
1386 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
1387 !test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) &&
1388 test_bit(R5_Insync, &dev->flags)) {
1389 if (test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
1390 {
1391 PRINTK("Read_old block %d for Reconstruct\n", i);
1392 set_bit(R5_LOCKED, &dev->flags);
1393 set_bit(R5_Wantread, &dev->flags);
1394 locked++;
1395 } else {
1396 set_bit(STRIPE_DELAYED, &sh->state);
1397 set_bit(STRIPE_HANDLE, &sh->state);
1398 }
1399 }
1400 }
1401 /* now if nothing is locked, and if we have enough data, we can start a write request */
NeilBrown72626682005-09-09 16:23:54 -07001402 if (locked == 0 && (rcw == 0 ||rmw == 0) &&
1403 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 PRINTK("Computing parity...\n");
1405 compute_parity(sh, rcw==0 ? RECONSTRUCT_WRITE : READ_MODIFY_WRITE);
1406 /* now every locked buffer is ready to be written */
1407 for (i=disks; i--;)
1408 if (test_bit(R5_LOCKED, &sh->dev[i].flags)) {
1409 PRINTK("Writing block %d\n", i);
1410 locked++;
1411 set_bit(R5_Wantwrite, &sh->dev[i].flags);
1412 if (!test_bit(R5_Insync, &sh->dev[i].flags)
1413 || (i==sh->pd_idx && failed == 0))
1414 set_bit(STRIPE_INSYNC, &sh->state);
1415 }
1416 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
1417 atomic_dec(&conf->preread_active_stripes);
1418 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
1419 md_wakeup_thread(conf->mddev->thread);
1420 }
1421 }
1422 }
1423
1424 /* maybe we need to check and possibly fix the parity for this stripe
1425 * Any reads will already have been scheduled, so we just see if enough data
1426 * is available
1427 */
1428 if (syncing && locked == 0 &&
NeilBrown14f8d262006-01-06 00:20:14 -08001429 !test_bit(STRIPE_INSYNC, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 set_bit(STRIPE_HANDLE, &sh->state);
1431 if (failed == 0) {
1432 char *pagea;
1433 if (uptodate != disks)
1434 BUG();
1435 compute_parity(sh, CHECK_PARITY);
1436 uptodate--;
1437 pagea = page_address(sh->dev[sh->pd_idx].page);
1438 if ((*(u32*)pagea) == 0 &&
1439 !memcmp(pagea, pagea+4, STRIPE_SIZE-4)) {
1440 /* parity is correct (on disc, not in buffer any more) */
1441 set_bit(STRIPE_INSYNC, &sh->state);
NeilBrown9d888832005-11-08 21:39:26 -08001442 } else {
1443 conf->mddev->resync_mismatches += STRIPE_SECTORS;
1444 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
1445 /* don't try to repair!! */
1446 set_bit(STRIPE_INSYNC, &sh->state);
NeilBrown14f8d262006-01-06 00:20:14 -08001447 else {
1448 compute_block(sh, sh->pd_idx);
1449 uptodate++;
1450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 }
1452 }
1453 if (!test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrown14f8d262006-01-06 00:20:14 -08001454 /* either failed parity check, or recovery is happening */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (failed==0)
1456 failed_num = sh->pd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 dev = &sh->dev[failed_num];
NeilBrown14f8d262006-01-06 00:20:14 -08001458 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
1459 BUG_ON(uptodate != disks);
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 set_bit(R5_LOCKED, &dev->flags);
1462 set_bit(R5_Wantwrite, &dev->flags);
NeilBrown72626682005-09-09 16:23:54 -07001463 clear_bit(STRIPE_DEGRADED, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 locked++;
1465 set_bit(STRIPE_INSYNC, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 }
1467 }
1468 if (syncing && locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
1469 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
1470 clear_bit(STRIPE_SYNCING, &sh->state);
1471 }
NeilBrown4e5314b2005-11-08 21:39:22 -08001472
1473 /* If the failed drive is just a ReadError, then we might need to progress
1474 * the repair/check process
1475 */
NeilBrownba22dcb2005-11-08 21:39:31 -08001476 if (failed == 1 && ! conf->mddev->ro &&
1477 test_bit(R5_ReadError, &sh->dev[failed_num].flags)
NeilBrown4e5314b2005-11-08 21:39:22 -08001478 && !test_bit(R5_LOCKED, &sh->dev[failed_num].flags)
1479 && test_bit(R5_UPTODATE, &sh->dev[failed_num].flags)
1480 ) {
1481 dev = &sh->dev[failed_num];
1482 if (!test_bit(R5_ReWrite, &dev->flags)) {
1483 set_bit(R5_Wantwrite, &dev->flags);
1484 set_bit(R5_ReWrite, &dev->flags);
1485 set_bit(R5_LOCKED, &dev->flags);
NeilBrownccfcc3c2006-03-27 01:18:09 -08001486 locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08001487 } else {
1488 /* let's read it back */
1489 set_bit(R5_Wantread, &dev->flags);
1490 set_bit(R5_LOCKED, &dev->flags);
NeilBrownccfcc3c2006-03-27 01:18:09 -08001491 locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08001492 }
1493 }
1494
NeilBrownccfcc3c2006-03-27 01:18:09 -08001495 if (expanded && test_bit(STRIPE_EXPANDING, &sh->state)) {
1496 /* Need to write out all blocks after computing parity */
1497 sh->disks = conf->raid_disks;
1498 sh->pd_idx = stripe_to_pdidx(sh->sector, conf, conf->raid_disks);
1499 compute_parity(sh, RECONSTRUCT_WRITE);
1500 for (i= conf->raid_disks; i--;) {
1501 set_bit(R5_LOCKED, &sh->dev[i].flags);
1502 locked++;
1503 set_bit(R5_Wantwrite, &sh->dev[i].flags);
1504 }
1505 clear_bit(STRIPE_EXPANDING, &sh->state);
1506 } else if (expanded) {
1507 clear_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrownf6705572006-03-27 01:18:11 -08001508 atomic_dec(&conf->reshape_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -08001509 wake_up(&conf->wait_for_overlap);
1510 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
1511 }
1512
1513 if (expanding && locked == 0) {
1514 /* We have read all the blocks in this stripe and now we need to
1515 * copy some of them into a target stripe for expand.
1516 */
1517 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
1518 for (i=0; i< sh->disks; i++)
1519 if (i != sh->pd_idx) {
1520 int dd_idx, pd_idx, j;
1521 struct stripe_head *sh2;
1522
1523 sector_t bn = compute_blocknr(sh, i);
1524 sector_t s = raid5_compute_sector(bn, conf->raid_disks,
1525 conf->raid_disks-1,
1526 &dd_idx, &pd_idx, conf);
1527 sh2 = get_active_stripe(conf, s, conf->raid_disks, pd_idx, 1);
1528 if (sh2 == NULL)
1529 /* so far only the early blocks of this stripe
1530 * have been requested. When later blocks
1531 * get requested, we will try again
1532 */
1533 continue;
1534 if(!test_bit(STRIPE_EXPANDING, &sh2->state) ||
1535 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
1536 /* must have already done this block */
1537 release_stripe(sh2);
1538 continue;
1539 }
1540 memcpy(page_address(sh2->dev[dd_idx].page),
1541 page_address(sh->dev[i].page),
1542 STRIPE_SIZE);
1543 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
1544 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
1545 for (j=0; j<conf->raid_disks; j++)
1546 if (j != sh2->pd_idx &&
1547 !test_bit(R5_Expanded, &sh2->dev[j].flags))
1548 break;
1549 if (j == conf->raid_disks) {
1550 set_bit(STRIPE_EXPAND_READY, &sh2->state);
1551 set_bit(STRIPE_HANDLE, &sh2->state);
1552 }
1553 release_stripe(sh2);
1554 }
1555 }
1556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 spin_unlock(&sh->lock);
1558
1559 while ((bi=return_bi)) {
1560 int bytes = bi->bi_size;
1561
1562 return_bi = bi->bi_next;
1563 bi->bi_next = NULL;
1564 bi->bi_size = 0;
1565 bi->bi_end_io(bi, bytes, 0);
1566 }
1567 for (i=disks; i-- ;) {
1568 int rw;
1569 struct bio *bi;
1570 mdk_rdev_t *rdev;
1571 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags))
1572 rw = 1;
1573 else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
1574 rw = 0;
1575 else
1576 continue;
1577
1578 bi = &sh->dev[i].req;
1579
1580 bi->bi_rw = rw;
1581 if (rw)
1582 bi->bi_end_io = raid5_end_write_request;
1583 else
1584 bi->bi_end_io = raid5_end_read_request;
1585
1586 rcu_read_lock();
Suzanne Woodd6065f72005-11-08 21:39:27 -08001587 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -08001588 if (rdev && test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 rdev = NULL;
1590 if (rdev)
1591 atomic_inc(&rdev->nr_pending);
1592 rcu_read_unlock();
1593
1594 if (rdev) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08001595 if (syncing || expanding || expanded)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
1597
1598 bi->bi_bdev = rdev->bdev;
1599 PRINTK("for %llu schedule op %ld on disc %d\n",
1600 (unsigned long long)sh->sector, bi->bi_rw, i);
1601 atomic_inc(&sh->count);
1602 bi->bi_sector = sh->sector + rdev->data_offset;
1603 bi->bi_flags = 1 << BIO_UPTODATE;
1604 bi->bi_vcnt = 1;
1605 bi->bi_max_vecs = 1;
1606 bi->bi_idx = 0;
1607 bi->bi_io_vec = &sh->dev[i].vec;
1608 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1609 bi->bi_io_vec[0].bv_offset = 0;
1610 bi->bi_size = STRIPE_SIZE;
1611 bi->bi_next = NULL;
NeilBrown4dbcdc72006-01-06 00:20:52 -08001612 if (rw == WRITE &&
1613 test_bit(R5_ReWrite, &sh->dev[i].flags))
1614 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 generic_make_request(bi);
1616 } else {
NeilBrown72626682005-09-09 16:23:54 -07001617 if (rw == 1)
1618 set_bit(STRIPE_DEGRADED, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 PRINTK("skip op %ld on disc %d for sector %llu\n",
1620 bi->bi_rw, i, (unsigned long long)sh->sector);
1621 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1622 set_bit(STRIPE_HANDLE, &sh->state);
1623 }
1624 }
1625}
1626
Arjan van de Ven858119e2006-01-14 13:20:43 -08001627static void raid5_activate_delayed(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628{
1629 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
1630 while (!list_empty(&conf->delayed_list)) {
1631 struct list_head *l = conf->delayed_list.next;
1632 struct stripe_head *sh;
1633 sh = list_entry(l, struct stripe_head, lru);
1634 list_del_init(l);
1635 clear_bit(STRIPE_DELAYED, &sh->state);
1636 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
1637 atomic_inc(&conf->preread_active_stripes);
1638 list_add_tail(&sh->lru, &conf->handle_list);
1639 }
1640 }
1641}
1642
Arjan van de Ven858119e2006-01-14 13:20:43 -08001643static void activate_bit_delay(raid5_conf_t *conf)
NeilBrown72626682005-09-09 16:23:54 -07001644{
1645 /* device_lock is held */
1646 struct list_head head;
1647 list_add(&head, &conf->bitmap_list);
1648 list_del_init(&conf->bitmap_list);
1649 while (!list_empty(&head)) {
1650 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
1651 list_del_init(&sh->lru);
1652 atomic_inc(&sh->count);
1653 __release_stripe(conf, sh);
1654 }
1655}
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657static void unplug_slaves(mddev_t *mddev)
1658{
1659 raid5_conf_t *conf = mddev_to_conf(mddev);
1660 int i;
1661
1662 rcu_read_lock();
1663 for (i=0; i<mddev->raid_disks; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -08001664 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -08001665 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
1667
1668 atomic_inc(&rdev->nr_pending);
1669 rcu_read_unlock();
1670
1671 if (r_queue->unplug_fn)
1672 r_queue->unplug_fn(r_queue);
1673
1674 rdev_dec_pending(rdev, mddev);
1675 rcu_read_lock();
1676 }
1677 }
1678 rcu_read_unlock();
1679}
1680
1681static void raid5_unplug_device(request_queue_t *q)
1682{
1683 mddev_t *mddev = q->queuedata;
1684 raid5_conf_t *conf = mddev_to_conf(mddev);
1685 unsigned long flags;
1686
1687 spin_lock_irqsave(&conf->device_lock, flags);
1688
NeilBrown72626682005-09-09 16:23:54 -07001689 if (blk_remove_plug(q)) {
1690 conf->seq_flush++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07001692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 md_wakeup_thread(mddev->thread);
1694
1695 spin_unlock_irqrestore(&conf->device_lock, flags);
1696
1697 unplug_slaves(mddev);
1698}
1699
1700static int raid5_issue_flush(request_queue_t *q, struct gendisk *disk,
1701 sector_t *error_sector)
1702{
1703 mddev_t *mddev = q->queuedata;
1704 raid5_conf_t *conf = mddev_to_conf(mddev);
1705 int i, ret = 0;
1706
1707 rcu_read_lock();
1708 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -08001709 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -08001710 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 struct block_device *bdev = rdev->bdev;
1712 request_queue_t *r_queue = bdev_get_queue(bdev);
1713
1714 if (!r_queue->issue_flush_fn)
1715 ret = -EOPNOTSUPP;
1716 else {
1717 atomic_inc(&rdev->nr_pending);
1718 rcu_read_unlock();
1719 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
1720 error_sector);
1721 rdev_dec_pending(rdev, mddev);
1722 rcu_read_lock();
1723 }
1724 }
1725 }
1726 rcu_read_unlock();
1727 return ret;
1728}
1729
1730static inline void raid5_plug_device(raid5_conf_t *conf)
1731{
1732 spin_lock_irq(&conf->device_lock);
1733 blk_plug_device(conf->mddev->queue);
1734 spin_unlock_irq(&conf->device_lock);
1735}
1736
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001737static int make_request(request_queue_t *q, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738{
1739 mddev_t *mddev = q->queuedata;
1740 raid5_conf_t *conf = mddev_to_conf(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 unsigned int dd_idx, pd_idx;
1742 sector_t new_sector;
1743 sector_t logical_sector, last_sector;
1744 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01001745 const int rw = bio_data_dir(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
NeilBrowne5dcdd82005-09-09 16:23:41 -07001747 if (unlikely(bio_barrier(bi))) {
1748 bio_endio(bi, bi->bi_size, -EOPNOTSUPP);
1749 return 0;
1750 }
1751
NeilBrown3d310eb2005-06-21 17:17:26 -07001752 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07001753
Jens Axboea3623572005-11-01 09:26:16 +01001754 disk_stat_inc(mddev->gendisk, ios[rw]);
1755 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bi));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756
1757 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
1758 last_sector = bi->bi_sector + (bi->bi_size>>9);
1759 bi->bi_next = NULL;
1760 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07001761
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
1763 DEFINE_WAIT(w);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001764 int disks;
NeilBrownb578d552006-03-27 01:18:12 -08001765
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001766 retry:
NeilBrownb578d552006-03-27 01:18:12 -08001767 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001768 if (likely(conf->expand_progress == MaxSector))
1769 disks = conf->raid_disks;
1770 else {
1771 spin_lock_irq(&conf->device_lock);
1772 disks = conf->raid_disks;
1773 if (logical_sector >= conf->expand_progress)
1774 disks = conf->previous_raid_disks;
NeilBrownb578d552006-03-27 01:18:12 -08001775 else {
1776 if (logical_sector >= conf->expand_lo) {
1777 spin_unlock_irq(&conf->device_lock);
1778 schedule();
1779 goto retry;
1780 }
1781 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001782 spin_unlock_irq(&conf->device_lock);
1783 }
1784 new_sector = raid5_compute_sector(logical_sector, disks, disks - 1,
1785 &dd_idx, &pd_idx, conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 PRINTK("raid5: make_request, sector %llu logical %llu\n",
1787 (unsigned long long)new_sector,
1788 (unsigned long long)logical_sector);
1789
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001790 sh = get_active_stripe(conf, new_sector, disks, pd_idx, (bi->bi_rw&RWA_MASK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 if (sh) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001792 if (unlikely(conf->expand_progress != MaxSector)) {
1793 /* expansion might have moved on while waiting for a
1794 * stripe, so we much do the range check again.
1795 */
1796 int must_retry = 0;
1797 spin_lock_irq(&conf->device_lock);
1798 if (logical_sector < conf->expand_progress &&
1799 disks == conf->previous_raid_disks)
1800 /* mismatch, need to try again */
1801 must_retry = 1;
1802 spin_unlock_irq(&conf->device_lock);
1803 if (must_retry) {
1804 release_stripe(sh);
1805 goto retry;
1806 }
1807 }
1808
1809 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
1810 !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) {
1811 /* Stripe is busy expanding or
1812 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 * and wait a while
1814 */
1815 raid5_unplug_device(mddev->queue);
1816 release_stripe(sh);
1817 schedule();
1818 goto retry;
1819 }
1820 finish_wait(&conf->wait_for_overlap, &w);
1821 raid5_plug_device(conf);
1822 handle_stripe(sh);
1823 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 } else {
1825 /* cannot get stripe for read-ahead, just give-up */
1826 clear_bit(BIO_UPTODATE, &bi->bi_flags);
1827 finish_wait(&conf->wait_for_overlap, &w);
1828 break;
1829 }
1830
1831 }
1832 spin_lock_irq(&conf->device_lock);
1833 if (--bi->bi_phys_segments == 0) {
1834 int bytes = bi->bi_size;
1835
1836 if ( bio_data_dir(bi) == WRITE )
1837 md_write_end(mddev);
1838 bi->bi_size = 0;
1839 bi->bi_end_io(bi, bytes, 0);
1840 }
1841 spin_unlock_irq(&conf->device_lock);
1842 return 0;
1843}
1844
1845/* FIXME go_faster isn't used */
NeilBrown57afd892005-06-21 17:17:13 -07001846static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847{
1848 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
1849 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08001850 int pd_idx;
1851 sector_t first_sector, last_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 int raid_disks = conf->raid_disks;
1853 int data_disks = raid_disks-1;
NeilBrown72626682005-09-09 16:23:54 -07001854 sector_t max_sector = mddev->size << 1;
1855 int sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
NeilBrown72626682005-09-09 16:23:54 -07001857 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 /* just being told to finish up .. nothing much to do */
1859 unplug_slaves(mddev);
NeilBrown29269552006-03-27 01:18:10 -08001860 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
1861 end_reshape(conf);
1862 return 0;
1863 }
NeilBrown72626682005-09-09 16:23:54 -07001864
1865 if (mddev->curr_resync < max_sector) /* aborted */
1866 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1867 &sync_blocks, 1);
1868 else /* compelted sync */
1869 conf->fullsync = 0;
1870 bitmap_close_sync(mddev->bitmap);
1871
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return 0;
1873 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08001874
1875 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
1876 /* reshaping is quite different to recovery/resync so it is
1877 * handled quite separately ... here.
1878 *
1879 * On each call to sync_request, we gather one chunk worth of
1880 * destination stripes and flag them as expanding.
1881 * Then we find all the source stripes and request reads.
1882 * As the reads complete, handle_stripe will copy the data
1883 * into the destination stripe and release that stripe.
1884 */
1885 int i;
1886 int dd_idx;
NeilBrownb578d552006-03-27 01:18:12 -08001887 sector_t writepos, safepos, gap;
NeilBrownf6705572006-03-27 01:18:11 -08001888
1889 if (sector_nr == 0 &&
1890 conf->expand_progress != 0) {
1891 /* restarting in the middle, skip the initial sectors */
1892 sector_nr = conf->expand_progress;
1893 sector_div(sector_nr, conf->raid_disks-1);
1894 *skipped = 1;
1895 return sector_nr;
1896 }
1897
NeilBrownb578d552006-03-27 01:18:12 -08001898 /* we update the metadata when there is more than 3Meg
1899 * in the block range (that is rather arbitrary, should
1900 * probably be time based) or when the data about to be
1901 * copied would over-write the source of the data at
1902 * the front of the range.
1903 * i.e. one new_stripe forward from expand_progress new_maps
1904 * to after where expand_lo old_maps to
1905 */
1906 writepos = conf->expand_progress +
1907 conf->chunk_size/512*(conf->raid_disks-1);
1908 sector_div(writepos, conf->raid_disks-1);
1909 safepos = conf->expand_lo;
1910 sector_div(safepos, conf->previous_raid_disks-1);
1911 gap = conf->expand_progress - conf->expand_lo;
NeilBrownf6705572006-03-27 01:18:11 -08001912
NeilBrownb578d552006-03-27 01:18:12 -08001913 if (writepos >= safepos ||
1914 gap > (conf->raid_disks-1)*3000*2 /*3Meg*/) {
1915 /* Cannot proceed until we've updated the superblock... */
1916 wait_event(conf->wait_for_overlap,
1917 atomic_read(&conf->reshape_stripes)==0);
1918 mddev->reshape_position = conf->expand_progress;
1919 mddev->sb_dirty = 1;
1920 md_wakeup_thread(mddev->thread);
1921 wait_event(mddev->sb_wait, mddev->sb_dirty == 0 ||
1922 kthread_should_stop());
1923 spin_lock_irq(&conf->device_lock);
1924 conf->expand_lo = mddev->reshape_position;
1925 spin_unlock_irq(&conf->device_lock);
1926 wake_up(&conf->wait_for_overlap);
1927 }
NeilBrownf6705572006-03-27 01:18:11 -08001928
NeilBrownccfcc3c2006-03-27 01:18:09 -08001929 for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) {
1930 int j;
1931 int skipped = 0;
1932 pd_idx = stripe_to_pdidx(sector_nr+i, conf, conf->raid_disks);
1933 sh = get_active_stripe(conf, sector_nr+i,
1934 conf->raid_disks, pd_idx, 0);
1935 set_bit(STRIPE_EXPANDING, &sh->state);
NeilBrownf6705572006-03-27 01:18:11 -08001936 atomic_inc(&conf->reshape_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -08001937 /* If any of this stripe is beyond the end of the old
1938 * array, then we need to zero those blocks
1939 */
1940 for (j=sh->disks; j--;) {
1941 sector_t s;
1942 if (j == sh->pd_idx)
1943 continue;
1944 s = compute_blocknr(sh, j);
1945 if (s < (mddev->array_size<<1)) {
1946 skipped = 1;
1947 continue;
1948 }
1949 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
1950 set_bit(R5_Expanded, &sh->dev[j].flags);
1951 set_bit(R5_UPTODATE, &sh->dev[j].flags);
1952 }
1953 if (!skipped) {
1954 set_bit(STRIPE_EXPAND_READY, &sh->state);
1955 set_bit(STRIPE_HANDLE, &sh->state);
1956 }
1957 release_stripe(sh);
1958 }
1959 spin_lock_irq(&conf->device_lock);
1960 conf->expand_progress = (sector_nr + i)*(conf->raid_disks-1);
1961 spin_unlock_irq(&conf->device_lock);
1962 /* Ok, those stripe are ready. We can start scheduling
1963 * reads on the source stripes.
1964 * The source stripes are determined by mapping the first and last
1965 * block on the destination stripes.
1966 */
1967 raid_disks = conf->previous_raid_disks;
1968 data_disks = raid_disks - 1;
1969 first_sector =
1970 raid5_compute_sector(sector_nr*(conf->raid_disks-1),
1971 raid_disks, data_disks,
1972 &dd_idx, &pd_idx, conf);
1973 last_sector =
1974 raid5_compute_sector((sector_nr+conf->chunk_size/512)
1975 *(conf->raid_disks-1) -1,
1976 raid_disks, data_disks,
1977 &dd_idx, &pd_idx, conf);
1978 if (last_sector >= (mddev->size<<1))
1979 last_sector = (mddev->size<<1)-1;
1980 while (first_sector <= last_sector) {
1981 pd_idx = stripe_to_pdidx(first_sector, conf, conf->previous_raid_disks);
1982 sh = get_active_stripe(conf, first_sector,
1983 conf->previous_raid_disks, pd_idx, 0);
1984 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
1985 set_bit(STRIPE_HANDLE, &sh->state);
1986 release_stripe(sh);
1987 first_sector += STRIPE_SECTORS;
1988 }
1989 return conf->chunk_size>>9;
1990 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 /* if there is 1 or more failed drives and we are trying
1992 * to resync, then assert that we are finished, because there is
1993 * nothing we can do.
1994 */
1995 if (mddev->degraded >= 1 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
NeilBrown57afd892005-06-21 17:17:13 -07001996 sector_t rv = (mddev->size << 1) - sector_nr;
1997 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 return rv;
1999 }
NeilBrown72626682005-09-09 16:23:54 -07002000 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrown3855ad92005-11-08 21:39:38 -08002001 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown72626682005-09-09 16:23:54 -07002002 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
2003 /* we can skip this block, and probably more */
2004 sync_blocks /= STRIPE_SECTORS;
2005 *skipped = 1;
2006 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
2007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
NeilBrownccfcc3c2006-03-27 01:18:09 -08002009 pd_idx = stripe_to_pdidx(sector_nr, conf, raid_disks);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08002010 sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 if (sh == NULL) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08002012 sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 /* make sure we don't swamp the stripe cache if someone else
2014 * is trying to get access
2015 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08002016 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 }
NeilBrown72626682005-09-09 16:23:54 -07002018 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 spin_lock(&sh->lock);
2020 set_bit(STRIPE_SYNCING, &sh->state);
2021 clear_bit(STRIPE_INSYNC, &sh->state);
2022 spin_unlock(&sh->lock);
2023
2024 handle_stripe(sh);
2025 release_stripe(sh);
2026
2027 return STRIPE_SECTORS;
2028}
2029
2030/*
2031 * This is our raid5 kernel thread.
2032 *
2033 * We scan the hash table for stripes which can be handled now.
2034 * During the scan, completed stripes are saved for us by the interrupt
2035 * handler, so that they will not have to wait for our next wakeup.
2036 */
2037static void raid5d (mddev_t *mddev)
2038{
2039 struct stripe_head *sh;
2040 raid5_conf_t *conf = mddev_to_conf(mddev);
2041 int handled;
2042
2043 PRINTK("+++ raid5d active\n");
2044
2045 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047 handled = 0;
2048 spin_lock_irq(&conf->device_lock);
2049 while (1) {
2050 struct list_head *first;
2051
NeilBrown72626682005-09-09 16:23:54 -07002052 if (conf->seq_flush - conf->seq_write > 0) {
2053 int seq = conf->seq_flush;
NeilBrown700e4322005-11-28 13:44:10 -08002054 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07002055 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08002056 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07002057 conf->seq_write = seq;
2058 activate_bit_delay(conf);
2059 }
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 if (list_empty(&conf->handle_list) &&
2062 atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD &&
2063 !blk_queue_plugged(mddev->queue) &&
2064 !list_empty(&conf->delayed_list))
2065 raid5_activate_delayed(conf);
2066
2067 if (list_empty(&conf->handle_list))
2068 break;
2069
2070 first = conf->handle_list.next;
2071 sh = list_entry(first, struct stripe_head, lru);
2072
2073 list_del_init(first);
2074 atomic_inc(&sh->count);
2075 if (atomic_read(&sh->count)!= 1)
2076 BUG();
2077 spin_unlock_irq(&conf->device_lock);
2078
2079 handled++;
2080 handle_stripe(sh);
2081 release_stripe(sh);
2082
2083 spin_lock_irq(&conf->device_lock);
2084 }
2085 PRINTK("%d stripes handled\n", handled);
2086
2087 spin_unlock_irq(&conf->device_lock);
2088
2089 unplug_slaves(mddev);
2090
2091 PRINTK("--- raid5d inactive\n");
2092}
2093
NeilBrown3f294f42005-11-08 21:39:25 -08002094static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08002095raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08002096{
NeilBrown007583c2005-11-08 21:39:30 -08002097 raid5_conf_t *conf = mddev_to_conf(mddev);
NeilBrown96de1e62005-11-08 21:39:39 -08002098 if (conf)
2099 return sprintf(page, "%d\n", conf->max_nr_stripes);
2100 else
2101 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08002102}
2103
2104static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08002105raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08002106{
NeilBrown007583c2005-11-08 21:39:30 -08002107 raid5_conf_t *conf = mddev_to_conf(mddev);
NeilBrown3f294f42005-11-08 21:39:25 -08002108 char *end;
2109 int new;
2110 if (len >= PAGE_SIZE)
2111 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08002112 if (!conf)
2113 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08002114
2115 new = simple_strtoul(page, &end, 10);
2116 if (!*page || (*end && *end != '\n') )
2117 return -EINVAL;
2118 if (new <= 16 || new > 32768)
2119 return -EINVAL;
2120 while (new < conf->max_nr_stripes) {
2121 if (drop_one_stripe(conf))
2122 conf->max_nr_stripes--;
2123 else
2124 break;
2125 }
2126 while (new > conf->max_nr_stripes) {
2127 if (grow_one_stripe(conf))
2128 conf->max_nr_stripes++;
2129 else break;
2130 }
2131 return len;
2132}
NeilBrown007583c2005-11-08 21:39:30 -08002133
NeilBrown96de1e62005-11-08 21:39:39 -08002134static struct md_sysfs_entry
2135raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
2136 raid5_show_stripe_cache_size,
2137 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08002138
2139static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08002140stripe_cache_active_show(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08002141{
NeilBrown007583c2005-11-08 21:39:30 -08002142 raid5_conf_t *conf = mddev_to_conf(mddev);
NeilBrown96de1e62005-11-08 21:39:39 -08002143 if (conf)
2144 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
2145 else
2146 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08002147}
2148
NeilBrown96de1e62005-11-08 21:39:39 -08002149static struct md_sysfs_entry
2150raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08002151
NeilBrown007583c2005-11-08 21:39:30 -08002152static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08002153 &raid5_stripecache_size.attr,
2154 &raid5_stripecache_active.attr,
2155 NULL,
2156};
NeilBrown007583c2005-11-08 21:39:30 -08002157static struct attribute_group raid5_attrs_group = {
2158 .name = NULL,
2159 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08002160};
2161
NeilBrown72626682005-09-09 16:23:54 -07002162static int run(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163{
2164 raid5_conf_t *conf;
2165 int raid_disk, memory;
2166 mdk_rdev_t *rdev;
2167 struct disk_info *disk;
2168 struct list_head *tmp;
2169
2170 if (mddev->level != 5 && mddev->level != 4) {
NeilBrown14f8d262006-01-06 00:20:14 -08002171 printk(KERN_ERR "raid5: %s: raid level not set to 4/5 (%d)\n",
2172 mdname(mddev), mddev->level);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 return -EIO;
2174 }
2175
NeilBrownf6705572006-03-27 01:18:11 -08002176 if (mddev->reshape_position != MaxSector) {
2177 /* Check that we can continue the reshape.
2178 * Currently only disks can change, it must
2179 * increase, and we must be past the point where
2180 * a stripe over-writes itself
2181 */
2182 sector_t here_new, here_old;
2183 int old_disks;
2184
2185 if (mddev->new_level != mddev->level ||
2186 mddev->new_layout != mddev->layout ||
2187 mddev->new_chunk != mddev->chunk_size) {
2188 printk(KERN_ERR "raid5: %s: unsupported reshape required - aborting.\n",
2189 mdname(mddev));
2190 return -EINVAL;
2191 }
2192 if (mddev->delta_disks <= 0) {
2193 printk(KERN_ERR "raid5: %s: unsupported reshape (reduce disks) required - aborting.\n",
2194 mdname(mddev));
2195 return -EINVAL;
2196 }
2197 old_disks = mddev->raid_disks - mddev->delta_disks;
2198 /* reshape_position must be on a new-stripe boundary, and one
2199 * further up in new geometry must map after here in old geometry.
2200 */
2201 here_new = mddev->reshape_position;
2202 if (sector_div(here_new, (mddev->chunk_size>>9)*(mddev->raid_disks-1))) {
2203 printk(KERN_ERR "raid5: reshape_position not on a stripe boundary\n");
2204 return -EINVAL;
2205 }
2206 /* here_new is the stripe we will write to */
2207 here_old = mddev->reshape_position;
2208 sector_div(here_old, (mddev->chunk_size>>9)*(old_disks-1));
2209 /* here_old is the first stripe that we might need to read from */
2210 if (here_new >= here_old) {
2211 /* Reading from the same stripe as writing to - bad */
2212 printk(KERN_ERR "raid5: reshape_position too early for auto-recovery - aborting.\n");
2213 return -EINVAL;
2214 }
2215 printk(KERN_INFO "raid5: reshape will continue\n");
2216 /* OK, we should be able to continue; */
2217 }
2218
2219
NeilBrownb55e6bf2006-03-27 01:18:06 -08002220 mddev->private = kzalloc(sizeof (raid5_conf_t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 if ((conf = mddev->private) == NULL)
2222 goto abort;
NeilBrownf6705572006-03-27 01:18:11 -08002223 if (mddev->reshape_position == MaxSector) {
2224 conf->previous_raid_disks = conf->raid_disks = mddev->raid_disks;
2225 } else {
2226 conf->raid_disks = mddev->raid_disks;
2227 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
2228 }
2229
2230 conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info),
NeilBrownb55e6bf2006-03-27 01:18:06 -08002231 GFP_KERNEL);
2232 if (!conf->disks)
2233 goto abort;
NeilBrown9ffae0c2006-01-06 00:20:32 -08002234
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 conf->mddev = mddev;
2236
NeilBrownfccddba2006-01-06 00:20:33 -08002237 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
2240 spin_lock_init(&conf->device_lock);
2241 init_waitqueue_head(&conf->wait_for_stripe);
2242 init_waitqueue_head(&conf->wait_for_overlap);
2243 INIT_LIST_HEAD(&conf->handle_list);
2244 INIT_LIST_HEAD(&conf->delayed_list);
NeilBrown72626682005-09-09 16:23:54 -07002245 INIT_LIST_HEAD(&conf->bitmap_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 INIT_LIST_HEAD(&conf->inactive_list);
2247 atomic_set(&conf->active_stripes, 0);
2248 atomic_set(&conf->preread_active_stripes, 0);
2249
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 PRINTK("raid5: run(%s) called.\n", mdname(mddev));
2251
2252 ITERATE_RDEV(mddev,rdev,tmp) {
2253 raid_disk = rdev->raid_disk;
NeilBrownf6705572006-03-27 01:18:11 -08002254 if (raid_disk >= conf->raid_disks
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 || raid_disk < 0)
2256 continue;
2257 disk = conf->disks + raid_disk;
2258
2259 disk->rdev = rdev;
2260
NeilBrownb2d444d2005-11-08 21:39:31 -08002261 if (test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 char b[BDEVNAME_SIZE];
2263 printk(KERN_INFO "raid5: device %s operational as raid"
2264 " disk %d\n", bdevname(rdev->bdev,b),
2265 raid_disk);
2266 conf->working_disks++;
2267 }
2268 }
2269
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 /*
2271 * 0 for a fully functional array, 1 for a degraded array.
2272 */
2273 mddev->degraded = conf->failed_disks = conf->raid_disks - conf->working_disks;
2274 conf->mddev = mddev;
2275 conf->chunk_size = mddev->chunk_size;
2276 conf->level = mddev->level;
2277 conf->algorithm = mddev->layout;
2278 conf->max_nr_stripes = NR_STRIPES;
NeilBrownf6705572006-03-27 01:18:11 -08002279 conf->expand_progress = mddev->reshape_position;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
2281 /* device size must be a multiple of chunk size */
2282 mddev->size &= ~(mddev->chunk_size/1024 -1);
NeilBrownb1581562005-07-31 22:34:50 -07002283 mddev->resync_max_sectors = mddev->size << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
2285 if (!conf->chunk_size || conf->chunk_size % 4) {
2286 printk(KERN_ERR "raid5: invalid chunk size %d for %s\n",
2287 conf->chunk_size, mdname(mddev));
2288 goto abort;
2289 }
2290 if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) {
2291 printk(KERN_ERR
2292 "raid5: unsupported parity algorithm %d for %s\n",
2293 conf->algorithm, mdname(mddev));
2294 goto abort;
2295 }
2296 if (mddev->degraded > 1) {
2297 printk(KERN_ERR "raid5: not enough operational devices for %s"
2298 " (%d/%d failed)\n",
2299 mdname(mddev), conf->failed_disks, conf->raid_disks);
2300 goto abort;
2301 }
2302
2303 if (mddev->degraded == 1 &&
2304 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08002305 if (mddev->ok_start_degraded)
2306 printk(KERN_WARNING
2307 "raid5: starting dirty degraded array: %s"
2308 "- data corruption possible.\n",
2309 mdname(mddev));
2310 else {
2311 printk(KERN_ERR
2312 "raid5: cannot start dirty degraded array for %s\n",
2313 mdname(mddev));
2314 goto abort;
2315 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 }
2317
2318 {
2319 mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5");
2320 if (!mddev->thread) {
2321 printk(KERN_ERR
2322 "raid5: couldn't allocate thread for %s\n",
2323 mdname(mddev));
2324 goto abort;
2325 }
2326 }
NeilBrown50368052005-12-12 02:39:17 -08002327 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
2329 if (grow_stripes(conf, conf->max_nr_stripes)) {
2330 printk(KERN_ERR
2331 "raid5: couldn't allocate %dkB for buffers\n", memory);
2332 shrink_stripes(conf);
2333 md_unregister_thread(mddev->thread);
2334 goto abort;
2335 } else
2336 printk(KERN_INFO "raid5: allocated %dkB for %s\n",
2337 memory, mdname(mddev));
2338
2339 if (mddev->degraded == 0)
2340 printk("raid5: raid level %d set %s active with %d out of %d"
2341 " devices, algorithm %d\n", conf->level, mdname(mddev),
2342 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
2343 conf->algorithm);
2344 else
2345 printk(KERN_ALERT "raid5: raid level %d set %s active with %d"
2346 " out of %d devices, algorithm %d\n", conf->level,
2347 mdname(mddev), mddev->raid_disks - mddev->degraded,
2348 mddev->raid_disks, conf->algorithm);
2349
2350 print_raid5_conf(conf);
2351
NeilBrownf6705572006-03-27 01:18:11 -08002352 if (conf->expand_progress != MaxSector) {
2353 printk("...ok start reshape thread\n");
NeilBrownb578d552006-03-27 01:18:12 -08002354 conf->expand_lo = conf->expand_progress;
NeilBrownf6705572006-03-27 01:18:11 -08002355 atomic_set(&conf->reshape_stripes, 0);
2356 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2357 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2358 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
2359 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
2360 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
2361 "%s_reshape");
2362 /* FIXME if md_register_thread fails?? */
2363 md_wakeup_thread(mddev->sync_thread);
2364
2365 }
2366
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 /* read-ahead size must cover two whole stripes, which is
2368 * 2 * (n-1) * chunksize where 'n' is the number of raid devices
2369 */
2370 {
2371 int stripe = (mddev->raid_disks-1) * mddev->chunk_size
NeilBrown2d1f3b52006-01-06 00:20:31 -08002372 / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
2374 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
2375 }
2376
2377 /* Ok, everything is just fine now */
NeilBrown007583c2005-11-08 21:39:30 -08002378 sysfs_create_group(&mddev->kobj, &raid5_attrs_group);
NeilBrown7a5febe2005-05-16 21:53:16 -07002379
2380 mddev->queue->unplug_fn = raid5_unplug_device;
2381 mddev->queue->issue_flush_fn = raid5_issue_flush;
NeilBrownf6705572006-03-27 01:18:11 -08002382 mddev->array_size = mddev->size * (conf->previous_raid_disks - 1);
NeilBrown7a5febe2005-05-16 21:53:16 -07002383
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 return 0;
2385abort:
2386 if (conf) {
2387 print_raid5_conf(conf);
NeilBrownb55e6bf2006-03-27 01:18:06 -08002388 kfree(conf->disks);
NeilBrownfccddba2006-01-06 00:20:33 -08002389 kfree(conf->stripe_hashtbl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 kfree(conf);
2391 }
2392 mddev->private = NULL;
2393 printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev));
2394 return -EIO;
2395}
2396
2397
2398
NeilBrown3f294f42005-11-08 21:39:25 -08002399static int stop(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400{
2401 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
2402
2403 md_unregister_thread(mddev->thread);
2404 mddev->thread = NULL;
2405 shrink_stripes(conf);
NeilBrownfccddba2006-01-06 00:20:33 -08002406 kfree(conf->stripe_hashtbl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
NeilBrown007583c2005-11-08 21:39:30 -08002408 sysfs_remove_group(&mddev->kobj, &raid5_attrs_group);
NeilBrownb55e6bf2006-03-27 01:18:06 -08002409 kfree(conf->disks);
NeilBrown96de1e62005-11-08 21:39:39 -08002410 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 mddev->private = NULL;
2412 return 0;
2413}
2414
2415#if RAID5_DEBUG
2416static void print_sh (struct stripe_head *sh)
2417{
2418 int i;
2419
2420 printk("sh %llu, pd_idx %d, state %ld.\n",
2421 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
2422 printk("sh %llu, count %d.\n",
2423 (unsigned long long)sh->sector, atomic_read(&sh->count));
2424 printk("sh %llu, ", (unsigned long long)sh->sector);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08002425 for (i = 0; i < sh->disks; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 printk("(cache%d: %p %ld) ",
2427 i, sh->dev[i].page, sh->dev[i].flags);
2428 }
2429 printk("\n");
2430}
2431
2432static void printall (raid5_conf_t *conf)
2433{
2434 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -08002435 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 int i;
2437
2438 spin_lock_irq(&conf->device_lock);
2439 for (i = 0; i < NR_HASH; i++) {
NeilBrownfccddba2006-01-06 00:20:33 -08002440 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 if (sh->raid_conf != conf)
2442 continue;
2443 print_sh(sh);
2444 }
2445 }
2446 spin_unlock_irq(&conf->device_lock);
2447}
2448#endif
2449
2450static void status (struct seq_file *seq, mddev_t *mddev)
2451{
2452 raid5_conf_t *conf = (raid5_conf_t *) mddev->private;
2453 int i;
2454
2455 seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout);
2456 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->working_disks);
2457 for (i = 0; i < conf->raid_disks; i++)
2458 seq_printf (seq, "%s",
2459 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08002460 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 seq_printf (seq, "]");
2462#if RAID5_DEBUG
2463#define D(x) \
2464 seq_printf (seq, "<"#x":%d>", atomic_read(&conf->x))
2465 printall(conf);
2466#endif
2467}
2468
2469static void print_raid5_conf (raid5_conf_t *conf)
2470{
2471 int i;
2472 struct disk_info *tmp;
2473
2474 printk("RAID5 conf printout:\n");
2475 if (!conf) {
2476 printk("(conf==NULL)\n");
2477 return;
2478 }
2479 printk(" --- rd:%d wd:%d fd:%d\n", conf->raid_disks,
2480 conf->working_disks, conf->failed_disks);
2481
2482 for (i = 0; i < conf->raid_disks; i++) {
2483 char b[BDEVNAME_SIZE];
2484 tmp = conf->disks + i;
2485 if (tmp->rdev)
2486 printk(" disk %d, o:%d, dev:%s\n",
NeilBrownb2d444d2005-11-08 21:39:31 -08002487 i, !test_bit(Faulty, &tmp->rdev->flags),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 bdevname(tmp->rdev->bdev,b));
2489 }
2490}
2491
2492static int raid5_spare_active(mddev_t *mddev)
2493{
2494 int i;
2495 raid5_conf_t *conf = mddev->private;
2496 struct disk_info *tmp;
2497
2498 for (i = 0; i < conf->raid_disks; i++) {
2499 tmp = conf->disks + i;
2500 if (tmp->rdev
NeilBrownb2d444d2005-11-08 21:39:31 -08002501 && !test_bit(Faulty, &tmp->rdev->flags)
2502 && !test_bit(In_sync, &tmp->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 mddev->degraded--;
2504 conf->failed_disks--;
2505 conf->working_disks++;
NeilBrownb2d444d2005-11-08 21:39:31 -08002506 set_bit(In_sync, &tmp->rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 }
2508 }
2509 print_raid5_conf(conf);
2510 return 0;
2511}
2512
2513static int raid5_remove_disk(mddev_t *mddev, int number)
2514{
2515 raid5_conf_t *conf = mddev->private;
2516 int err = 0;
2517 mdk_rdev_t *rdev;
2518 struct disk_info *p = conf->disks + number;
2519
2520 print_raid5_conf(conf);
2521 rdev = p->rdev;
2522 if (rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08002523 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 atomic_read(&rdev->nr_pending)) {
2525 err = -EBUSY;
2526 goto abort;
2527 }
2528 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07002529 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (atomic_read(&rdev->nr_pending)) {
2531 /* lost the race, try later */
2532 err = -EBUSY;
2533 p->rdev = rdev;
2534 }
2535 }
2536abort:
2537
2538 print_raid5_conf(conf);
2539 return err;
2540}
2541
2542static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
2543{
2544 raid5_conf_t *conf = mddev->private;
2545 int found = 0;
2546 int disk;
2547 struct disk_info *p;
2548
2549 if (mddev->degraded > 1)
2550 /* no point adding a device */
2551 return 0;
2552
2553 /*
2554 * find the disk ...
2555 */
NeilBrownf6705572006-03-27 01:18:11 -08002556 for (disk=0; disk < conf->raid_disks; disk++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 if ((p=conf->disks + disk)->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08002558 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 rdev->raid_disk = disk;
2560 found = 1;
NeilBrown72626682005-09-09 16:23:54 -07002561 if (rdev->saved_raid_disk != disk)
2562 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08002563 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 break;
2565 }
2566 print_raid5_conf(conf);
2567 return found;
2568}
2569
2570static int raid5_resize(mddev_t *mddev, sector_t sectors)
2571{
2572 /* no resync is happening, and there is enough space
2573 * on all devices, so we can resize.
2574 * We need to make sure resync covers any new space.
2575 * If the array is shrinking we should possibly wait until
2576 * any io in the removed space completes, but it hardly seems
2577 * worth it.
2578 */
2579 sectors &= ~((sector_t)mddev->chunk_size/512 - 1);
2580 mddev->array_size = (sectors * (mddev->raid_disks-1))>>1;
2581 set_capacity(mddev->gendisk, mddev->array_size << 1);
2582 mddev->changed = 1;
2583 if (sectors/2 > mddev->size && mddev->recovery_cp == MaxSector) {
2584 mddev->recovery_cp = mddev->size << 1;
2585 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2586 }
2587 mddev->size = sectors /2;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002588 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 return 0;
2590}
2591
NeilBrown29269552006-03-27 01:18:10 -08002592#ifdef CONFIG_MD_RAID5_RESHAPE
2593static int raid5_reshape(mddev_t *mddev, int raid_disks)
2594{
2595 raid5_conf_t *conf = mddev_to_conf(mddev);
2596 int err;
2597 mdk_rdev_t *rdev;
2598 struct list_head *rtmp;
2599 int spares = 0;
2600 int added_devices = 0;
2601
2602 if (mddev->degraded ||
2603 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2604 return -EBUSY;
2605 if (conf->raid_disks > raid_disks)
2606 return -EINVAL; /* Cannot shrink array yet */
2607 if (conf->raid_disks == raid_disks)
2608 return 0; /* nothing to do */
2609
2610 /* Can only proceed if there are plenty of stripe_heads.
2611 * We need a minimum of one full stripe,, and for sensible progress
2612 * it is best to have about 4 times that.
2613 * If we require 4 times, then the default 256 4K stripe_heads will
2614 * allow for chunk sizes up to 256K, which is probably OK.
2615 * If the chunk size is greater, user-space should request more
2616 * stripe_heads first.
2617 */
2618 if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes) {
2619 printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n",
2620 (mddev->chunk_size / STRIPE_SIZE)*4);
2621 return -ENOSPC;
2622 }
2623
2624 ITERATE_RDEV(mddev, rdev, rtmp)
2625 if (rdev->raid_disk < 0 &&
2626 !test_bit(Faulty, &rdev->flags))
2627 spares++;
2628 if (conf->raid_disks + spares < raid_disks-1)
2629 /* Not enough devices even to make a degraded array
2630 * of that size
2631 */
2632 return -EINVAL;
2633
2634 err = resize_stripes(conf, raid_disks);
2635 if (err)
2636 return err;
2637
NeilBrownf6705572006-03-27 01:18:11 -08002638 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08002639 spin_lock_irq(&conf->device_lock);
2640 conf->previous_raid_disks = conf->raid_disks;
NeilBrownf6705572006-03-27 01:18:11 -08002641 conf->raid_disks = raid_disks;
NeilBrown29269552006-03-27 01:18:10 -08002642 conf->expand_progress = 0;
NeilBrownb578d552006-03-27 01:18:12 -08002643 conf->expand_lo = 0;
NeilBrown29269552006-03-27 01:18:10 -08002644 spin_unlock_irq(&conf->device_lock);
2645
2646 /* Add some new drives, as many as will fit.
2647 * We know there are enough to make the newly sized array work.
2648 */
2649 ITERATE_RDEV(mddev, rdev, rtmp)
2650 if (rdev->raid_disk < 0 &&
2651 !test_bit(Faulty, &rdev->flags)) {
2652 if (raid5_add_disk(mddev, rdev)) {
2653 char nm[20];
2654 set_bit(In_sync, &rdev->flags);
2655 conf->working_disks++;
2656 added_devices++;
2657 sprintf(nm, "rd%d", rdev->raid_disk);
2658 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2659 } else
2660 break;
2661 }
2662
2663 mddev->degraded = (raid_disks - conf->previous_raid_disks) - added_devices;
NeilBrownf6705572006-03-27 01:18:11 -08002664 mddev->new_chunk = mddev->chunk_size;
2665 mddev->new_layout = mddev->layout;
2666 mddev->new_level = mddev->level;
2667 mddev->raid_disks = raid_disks;
2668 mddev->delta_disks = raid_disks - conf->previous_raid_disks;
2669 mddev->reshape_position = 0;
2670 mddev->sb_dirty = 1;
2671
NeilBrown29269552006-03-27 01:18:10 -08002672 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2673 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2674 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
2675 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
2676 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
2677 "%s_reshape");
2678 if (!mddev->sync_thread) {
2679 mddev->recovery = 0;
2680 spin_lock_irq(&conf->device_lock);
2681 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownf6705572006-03-27 01:18:11 -08002682 mddev->delta_disks = 0;
NeilBrown29269552006-03-27 01:18:10 -08002683 conf->expand_progress = MaxSector;
2684 spin_unlock_irq(&conf->device_lock);
2685 return -EAGAIN;
2686 }
2687 md_wakeup_thread(mddev->sync_thread);
2688 md_new_event(mddev);
2689 return 0;
2690}
2691#endif
2692
2693static void end_reshape(raid5_conf_t *conf)
2694{
2695 struct block_device *bdev;
2696
NeilBrownf6705572006-03-27 01:18:11 -08002697 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
2698 conf->mddev->array_size = conf->mddev->size * (conf->raid_disks-1);
2699 set_capacity(conf->mddev->gendisk, conf->mddev->array_size << 1);
2700 conf->mddev->changed = 1;
NeilBrown29269552006-03-27 01:18:10 -08002701
NeilBrownf6705572006-03-27 01:18:11 -08002702 bdev = bdget_disk(conf->mddev->gendisk, 0);
2703 if (bdev) {
2704 mutex_lock(&bdev->bd_inode->i_mutex);
2705 i_size_write(bdev->bd_inode, conf->mddev->array_size << 10);
2706 mutex_unlock(&bdev->bd_inode->i_mutex);
2707 bdput(bdev);
2708 }
2709 spin_lock_irq(&conf->device_lock);
2710 conf->expand_progress = MaxSector;
2711 spin_unlock_irq(&conf->device_lock);
2712 conf->mddev->reshape_position = MaxSector;
NeilBrown29269552006-03-27 01:18:10 -08002713 }
NeilBrown29269552006-03-27 01:18:10 -08002714}
2715
NeilBrown72626682005-09-09 16:23:54 -07002716static void raid5_quiesce(mddev_t *mddev, int state)
2717{
2718 raid5_conf_t *conf = mddev_to_conf(mddev);
2719
2720 switch(state) {
2721 case 1: /* stop all writes */
2722 spin_lock_irq(&conf->device_lock);
2723 conf->quiesce = 1;
2724 wait_event_lock_irq(conf->wait_for_stripe,
2725 atomic_read(&conf->active_stripes) == 0,
2726 conf->device_lock, /* nothing */);
2727 spin_unlock_irq(&conf->device_lock);
2728 break;
2729
2730 case 0: /* re-enable writes */
2731 spin_lock_irq(&conf->device_lock);
2732 conf->quiesce = 0;
2733 wake_up(&conf->wait_for_stripe);
2734 spin_unlock_irq(&conf->device_lock);
2735 break;
2736 }
NeilBrown72626682005-09-09 16:23:54 -07002737}
NeilBrownb15c2e52006-01-06 00:20:16 -08002738
NeilBrown2604b702006-01-06 00:20:36 -08002739static struct mdk_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740{
2741 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08002742 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 .owner = THIS_MODULE,
2744 .make_request = make_request,
2745 .run = run,
2746 .stop = stop,
2747 .status = status,
2748 .error_handler = error,
2749 .hot_add_disk = raid5_add_disk,
2750 .hot_remove_disk= raid5_remove_disk,
2751 .spare_active = raid5_spare_active,
2752 .sync_request = sync_request,
2753 .resize = raid5_resize,
NeilBrown29269552006-03-27 01:18:10 -08002754#ifdef CONFIG_MD_RAID5_RESHAPE
2755 .reshape = raid5_reshape,
2756#endif
NeilBrown72626682005-09-09 16:23:54 -07002757 .quiesce = raid5_quiesce,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758};
2759
NeilBrown2604b702006-01-06 00:20:36 -08002760static struct mdk_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761{
NeilBrown2604b702006-01-06 00:20:36 -08002762 .name = "raid4",
2763 .level = 4,
2764 .owner = THIS_MODULE,
2765 .make_request = make_request,
2766 .run = run,
2767 .stop = stop,
2768 .status = status,
2769 .error_handler = error,
2770 .hot_add_disk = raid5_add_disk,
2771 .hot_remove_disk= raid5_remove_disk,
2772 .spare_active = raid5_spare_active,
2773 .sync_request = sync_request,
2774 .resize = raid5_resize,
2775 .quiesce = raid5_quiesce,
2776};
2777
2778static int __init raid5_init(void)
2779{
2780 register_md_personality(&raid5_personality);
2781 register_md_personality(&raid4_personality);
2782 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783}
2784
NeilBrown2604b702006-01-06 00:20:36 -08002785static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786{
NeilBrown2604b702006-01-06 00:20:36 -08002787 unregister_md_personality(&raid5_personality);
2788 unregister_md_personality(&raid4_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789}
2790
2791module_init(raid5_init);
2792module_exit(raid5_exit);
2793MODULE_LICENSE("GPL");
2794MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002795MODULE_ALIAS("md-raid5");
2796MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08002797MODULE_ALIAS("md-level-5");
2798MODULE_ALIAS("md-level-4");