blob: 47da0af6322be1bd7930f902c96c57875799a358 [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
NeilBrown16a53ec2006-06-26 00:27:38 -07005 * Copyright (C) 2002, 2003 H. Peter Anvin
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
NeilBrown16a53ec2006-06-26 00:27:38 -07007 * RAID-4/5/6 management functions.
8 * Thanks to Penguin Computing for making the RAID-6 development possible
9 * by donating a test server!
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
16 * You should have received a copy of the GNU General Public License
17 * (for example /usr/src/linux/COPYING); if not, write to the Free
18 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20
NeilBrownae3c20c2006-07-10 04:44:17 -070021/*
22 * BITMAP UNPLUGGING:
23 *
24 * The sequencing for updating the bitmap reliably is a little
25 * subtle (and I got it wrong the first time) so it deserves some
26 * explanation.
27 *
28 * We group bitmap updates into batches. Each batch has a number.
29 * We may write out several batches at once, but that isn't very important.
NeilBrown7c13edc2011-04-18 18:25:43 +100030 * conf->seq_write is the number of the last batch successfully written.
31 * conf->seq_flush is the number of the last batch that was closed to
NeilBrownae3c20c2006-07-10 04:44:17 -070032 * new additions.
33 * When we discover that we will need to write to any block in a stripe
34 * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq
NeilBrown7c13edc2011-04-18 18:25:43 +100035 * the number of the batch it will be in. This is seq_flush+1.
NeilBrownae3c20c2006-07-10 04:44:17 -070036 * When we are ready to do a write, if that batch hasn't been written yet,
37 * we plug the array and queue the stripe for later.
38 * When an unplug happens, we increment bm_flush, thus closing the current
39 * batch.
40 * When we notice that bm_flush > bm_write, we write out all pending updates
41 * to the bitmap, and advance bm_write to where bm_flush was.
42 * This may occasionally write a bit out twice, but is sure never to
43 * miss any bits.
44 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
NeilBrownbff61972009-03-31 14:33:13 +110046#include <linux/blkdev.h>
NeilBrownf6705572006-03-27 01:18:11 -080047#include <linux/kthread.h>
Dan Williamsf701d582009-03-31 15:09:39 +110048#include <linux/raid/pq.h>
Dan Williams91c00922007-01-02 13:52:30 -070049#include <linux/async_tx.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040050#include <linux/module.h>
Dan Williams07a3b412009-08-29 19:13:13 -070051#include <linux/async.h>
NeilBrownbff61972009-03-31 14:33:13 +110052#include <linux/seq_file.h>
Dan Williams36d1c642009-07-14 11:48:22 -070053#include <linux/cpu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090054#include <linux/slab.h>
Christian Dietrich8bda4702011-07-27 11:00:36 +100055#include <linux/ratelimit.h>
Shaohua Li851c30c2013-08-28 14:30:16 +080056#include <linux/nodemask.h>
NeilBrowna9add5d2012-10-31 11:59:09 +110057#include <trace/events/block.h>
58
NeilBrown43b2e5d2009-03-31 14:33:13 +110059#include "md.h"
NeilBrownbff61972009-03-31 14:33:13 +110060#include "raid5.h"
Trela Maciej54071b32010-03-08 16:02:42 +110061#include "raid0.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110062#include "bitmap.h"
NeilBrown72626682005-09-09 16:23:54 -070063
Shaohua Li851c30c2013-08-28 14:30:16 +080064#define cpu_to_group(cpu) cpu_to_node(cpu)
65#define ANY_GROUP NUMA_NO_NODE
66
67static struct workqueue_struct *raid5_wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/*
69 * Stripe cache
70 */
71
72#define NR_STRIPES 256
73#define STRIPE_SIZE PAGE_SIZE
74#define STRIPE_SHIFT (PAGE_SHIFT - 9)
75#define STRIPE_SECTORS (STRIPE_SIZE>>9)
76#define IO_THRESHOLD 1
Dan Williams8b3e6cd2008-04-28 02:15:53 -070077#define BYPASS_THRESHOLD 1
NeilBrownfccddba2006-01-06 00:20:33 -080078#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#define HASH_MASK (NR_HASH - 1)
Shaohua Libfc90cb2013-08-29 15:40:32 +080080#define MAX_STRIPE_BATCH 8
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
NeilBrownd1688a62011-10-11 16:49:52 +110082static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
NeilBrowndb298e12011-10-07 14:23:00 +110083{
84 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
85 return &conf->stripe_hashtbl[hash];
86}
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Shaohua Li566c09c2013-11-14 15:16:17 +110088static inline int stripe_hash_locks_hash(sector_t sect)
89{
90 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
91}
92
93static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
94{
95 spin_lock_irq(conf->hash_locks + hash);
96 spin_lock(&conf->device_lock);
97}
98
99static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
100{
101 spin_unlock(&conf->device_lock);
102 spin_unlock_irq(conf->hash_locks + hash);
103}
104
105static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
106{
107 int i;
108 local_irq_disable();
109 spin_lock(conf->hash_locks);
110 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
111 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
112 spin_lock(&conf->device_lock);
113}
114
115static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
116{
117 int i;
118 spin_unlock(&conf->device_lock);
119 for (i = NR_STRIPE_HASH_LOCKS; i; i--)
120 spin_unlock(conf->hash_locks + i - 1);
121 local_irq_enable();
122}
123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124/* bio's attached to a stripe+device for I/O are linked together in bi_sector
125 * order without overlap. There may be several bio's per stripe+device, and
126 * a bio could span several devices.
127 * When walking this list for a particular stripe+device, we must never proceed
128 * beyond a bio that extends past this device, as the next bio might no longer
129 * be valid.
NeilBrowndb298e12011-10-07 14:23:00 +1100130 * This function is used to determine the 'next' bio in the list, given the sector
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 * of the current stripe+device
132 */
NeilBrowndb298e12011-10-07 14:23:00 +1100133static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
134{
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800135 int sectors = bio_sectors(bio);
NeilBrowndb298e12011-10-07 14:23:00 +1100136 if (bio->bi_sector + sectors < sector + STRIPE_SECTORS)
137 return bio->bi_next;
138 else
139 return NULL;
140}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Jens Axboe960e7392008-08-15 10:41:18 +0200142/*
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200143 * We maintain a biased count of active stripes in the bottom 16 bits of
144 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
Jens Axboe960e7392008-08-15 10:41:18 +0200145 */
Shaohua Lie7836bd62012-07-19 16:01:31 +1000146static inline int raid5_bi_processed_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200147{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000148 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
149 return (atomic_read(segments) >> 16) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200150}
151
Shaohua Lie7836bd62012-07-19 16:01:31 +1000152static inline int raid5_dec_bi_active_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200153{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000154 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
155 return atomic_sub_return(1, segments) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200156}
157
Shaohua Lie7836bd62012-07-19 16:01:31 +1000158static inline void raid5_inc_bi_active_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200159{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000160 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
161 atomic_inc(segments);
Jens Axboe960e7392008-08-15 10:41:18 +0200162}
163
Shaohua Lie7836bd62012-07-19 16:01:31 +1000164static inline void raid5_set_bi_processed_stripes(struct bio *bio,
165 unsigned int cnt)
Jens Axboe960e7392008-08-15 10:41:18 +0200166{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000167 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
168 int old, new;
Jens Axboe960e7392008-08-15 10:41:18 +0200169
Shaohua Lie7836bd62012-07-19 16:01:31 +1000170 do {
171 old = atomic_read(segments);
172 new = (old & 0xffff) | (cnt << 16);
173 } while (atomic_cmpxchg(segments, old, new) != old);
Jens Axboe960e7392008-08-15 10:41:18 +0200174}
175
Shaohua Lie7836bd62012-07-19 16:01:31 +1000176static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt)
Jens Axboe960e7392008-08-15 10:41:18 +0200177{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000178 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
179 atomic_set(segments, cnt);
Jens Axboe960e7392008-08-15 10:41:18 +0200180}
181
NeilBrownd0dabf72009-03-31 14:39:38 +1100182/* Find first data disk in a raid6 stripe */
183static inline int raid6_d0(struct stripe_head *sh)
184{
NeilBrown67cc2b82009-03-31 14:39:38 +1100185 if (sh->ddf_layout)
186 /* ddf always start from first device */
187 return 0;
188 /* md starts just after Q block */
NeilBrownd0dabf72009-03-31 14:39:38 +1100189 if (sh->qd_idx == sh->disks - 1)
190 return 0;
191 else
192 return sh->qd_idx + 1;
193}
NeilBrown16a53ec2006-06-26 00:27:38 -0700194static inline int raid6_next_disk(int disk, int raid_disks)
195{
196 disk++;
197 return (disk < raid_disks) ? disk : 0;
198}
Dan Williamsa4456852007-07-09 11:56:43 -0700199
NeilBrownd0dabf72009-03-31 14:39:38 +1100200/* When walking through the disks in a raid5, starting at raid6_d0,
201 * We need to map each disk to a 'slot', where the data disks are slot
202 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
203 * is raid_disks-1. This help does that mapping.
204 */
NeilBrown67cc2b82009-03-31 14:39:38 +1100205static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
206 int *count, int syndrome_disks)
NeilBrownd0dabf72009-03-31 14:39:38 +1100207{
Dan Williams66295422009-10-19 18:09:32 -0700208 int slot = *count;
NeilBrown67cc2b82009-03-31 14:39:38 +1100209
NeilBrowne4424fe2009-10-16 16:27:34 +1100210 if (sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700211 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100212 if (idx == sh->pd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100213 return syndrome_disks;
NeilBrownd0dabf72009-03-31 14:39:38 +1100214 if (idx == sh->qd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100215 return syndrome_disks + 1;
NeilBrowne4424fe2009-10-16 16:27:34 +1100216 if (!sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700217 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100218 return slot;
219}
220
Dan Williamsa4456852007-07-09 11:56:43 -0700221static void return_io(struct bio *return_bi)
222{
223 struct bio *bi = return_bi;
224 while (bi) {
Dan Williamsa4456852007-07-09 11:56:43 -0700225
226 return_bi = bi->bi_next;
227 bi->bi_next = NULL;
228 bi->bi_size = 0;
Linus Torvalds0a82a8d2013-04-18 09:00:26 -0700229 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
230 bi, 0);
Neil Brown0e13fe232008-06-28 08:31:20 +1000231 bio_endio(bi, 0);
Dan Williamsa4456852007-07-09 11:56:43 -0700232 bi = return_bi;
233 }
234}
235
NeilBrownd1688a62011-10-11 16:49:52 +1100236static void print_raid5_conf (struct r5conf *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Dan Williams600aa102008-06-28 08:32:05 +1000238static int stripe_operations_active(struct stripe_head *sh)
239{
240 return sh->check_state || sh->reconstruct_state ||
241 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
242 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
243}
244
Shaohua Li851c30c2013-08-28 14:30:16 +0800245static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
246{
247 struct r5conf *conf = sh->raid_conf;
248 struct r5worker_group *group;
Shaohua Libfc90cb2013-08-29 15:40:32 +0800249 int thread_cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +0800250 int i, cpu = sh->cpu;
251
252 if (!cpu_online(cpu)) {
253 cpu = cpumask_any(cpu_online_mask);
254 sh->cpu = cpu;
255 }
256
257 if (list_empty(&sh->lru)) {
258 struct r5worker_group *group;
259 group = conf->worker_groups + cpu_to_group(cpu);
260 list_add_tail(&sh->lru, &group->handle_list);
Shaohua Libfc90cb2013-08-29 15:40:32 +0800261 group->stripes_cnt++;
262 sh->group = group;
Shaohua Li851c30c2013-08-28 14:30:16 +0800263 }
264
265 if (conf->worker_cnt_per_group == 0) {
266 md_wakeup_thread(conf->mddev->thread);
267 return;
268 }
269
270 group = conf->worker_groups + cpu_to_group(sh->cpu);
271
Shaohua Libfc90cb2013-08-29 15:40:32 +0800272 group->workers[0].working = true;
273 /* at least one worker should run to avoid race */
274 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
275
276 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
277 /* wakeup more workers */
278 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
279 if (group->workers[i].working == false) {
280 group->workers[i].working = true;
281 queue_work_on(sh->cpu, raid5_wq,
282 &group->workers[i].work);
283 thread_cnt--;
284 }
285 }
Shaohua Li851c30c2013-08-28 14:30:16 +0800286}
287
Shaohua Li566c09c2013-11-14 15:16:17 +1100288static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
289 struct list_head *temp_inactive_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Shaohua Li4eb788d2012-07-19 16:01:31 +1000291 BUG_ON(!list_empty(&sh->lru));
292 BUG_ON(atomic_read(&conf->active_stripes)==0);
293 if (test_bit(STRIPE_HANDLE, &sh->state)) {
294 if (test_bit(STRIPE_DELAYED, &sh->state) &&
295 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
296 list_add_tail(&sh->lru, &conf->delayed_list);
297 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
298 sh->bm_seq - conf->seq_write > 0)
299 list_add_tail(&sh->lru, &conf->bitmap_list);
300 else {
301 clear_bit(STRIPE_DELAYED, &sh->state);
302 clear_bit(STRIPE_BIT_DELAY, &sh->state);
Shaohua Li851c30c2013-08-28 14:30:16 +0800303 if (conf->worker_cnt_per_group == 0) {
304 list_add_tail(&sh->lru, &conf->handle_list);
305 } else {
306 raid5_wakeup_stripe_thread(sh);
307 return;
308 }
Shaohua Li4eb788d2012-07-19 16:01:31 +1000309 }
310 md_wakeup_thread(conf->mddev->thread);
311 } else {
312 BUG_ON(stripe_operations_active(sh));
313 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
314 if (atomic_dec_return(&conf->preread_active_stripes)
315 < IO_THRESHOLD)
316 md_wakeup_thread(conf->mddev->thread);
317 atomic_dec(&conf->active_stripes);
Shaohua Li566c09c2013-11-14 15:16:17 +1100318 if (!test_bit(STRIPE_EXPANDING, &sh->state))
319 list_add_tail(&sh->lru, temp_inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 }
321}
NeilBrownd0dabf72009-03-31 14:39:38 +1100322
Shaohua Li566c09c2013-11-14 15:16:17 +1100323static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
324 struct list_head *temp_inactive_list)
Shaohua Li4eb788d2012-07-19 16:01:31 +1000325{
326 if (atomic_dec_and_test(&sh->count))
Shaohua Li566c09c2013-11-14 15:16:17 +1100327 do_release_stripe(conf, sh, temp_inactive_list);
328}
329
330/*
331 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
332 *
333 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
334 * given time. Adding stripes only takes device lock, while deleting stripes
335 * only takes hash lock.
336 */
337static void release_inactive_stripe_list(struct r5conf *conf,
338 struct list_head *temp_inactive_list,
339 int hash)
340{
341 int size;
342 bool do_wakeup = false;
343 unsigned long flags;
344
345 if (hash == NR_STRIPE_HASH_LOCKS) {
346 size = NR_STRIPE_HASH_LOCKS;
347 hash = NR_STRIPE_HASH_LOCKS - 1;
348 } else
349 size = 1;
350 while (size) {
351 struct list_head *list = &temp_inactive_list[size - 1];
352
353 /*
354 * We don't hold any lock here yet, get_active_stripe() might
355 * remove stripes from the list
356 */
357 if (!list_empty_careful(list)) {
358 spin_lock_irqsave(conf->hash_locks + hash, flags);
Shaohua Li4bda5562013-11-14 15:16:17 +1100359 if (list_empty(conf->inactive_list + hash) &&
360 !list_empty(list))
361 atomic_dec(&conf->empty_inactive_list_nr);
Shaohua Li566c09c2013-11-14 15:16:17 +1100362 list_splice_tail_init(list, conf->inactive_list + hash);
363 do_wakeup = true;
364 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
365 }
366 size--;
367 hash--;
368 }
369
370 if (do_wakeup) {
371 wake_up(&conf->wait_for_stripe);
372 if (conf->retry_read_aligned)
373 md_wakeup_thread(conf->mddev->thread);
374 }
Shaohua Li4eb788d2012-07-19 16:01:31 +1000375}
376
Shaohua Li773ca822013-08-27 17:50:39 +0800377/* should hold conf->device_lock already */
Shaohua Li566c09c2013-11-14 15:16:17 +1100378static int release_stripe_list(struct r5conf *conf,
379 struct list_head *temp_inactive_list)
Shaohua Li773ca822013-08-27 17:50:39 +0800380{
381 struct stripe_head *sh;
382 int count = 0;
383 struct llist_node *head;
384
385 head = llist_del_all(&conf->released_stripes);
Shaohua Lid265d9d2013-08-28 14:29:05 +0800386 head = llist_reverse_order(head);
Shaohua Li773ca822013-08-27 17:50:39 +0800387 while (head) {
Shaohua Li566c09c2013-11-14 15:16:17 +1100388 int hash;
389
Shaohua Li773ca822013-08-27 17:50:39 +0800390 sh = llist_entry(head, struct stripe_head, release_list);
391 head = llist_next(head);
392 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
393 smp_mb();
394 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
395 /*
396 * Don't worry the bit is set here, because if the bit is set
397 * again, the count is always > 1. This is true for
398 * STRIPE_ON_UNPLUG_LIST bit too.
399 */
Shaohua Li566c09c2013-11-14 15:16:17 +1100400 hash = sh->hash_lock_index;
401 __release_stripe(conf, sh, &temp_inactive_list[hash]);
Shaohua Li773ca822013-08-27 17:50:39 +0800402 count++;
403 }
404
405 return count;
406}
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408static void release_stripe(struct stripe_head *sh)
409{
NeilBrownd1688a62011-10-11 16:49:52 +1100410 struct r5conf *conf = sh->raid_conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 unsigned long flags;
Shaohua Li566c09c2013-11-14 15:16:17 +1100412 struct list_head list;
413 int hash;
Shaohua Li773ca822013-08-27 17:50:39 +0800414 bool wakeup;
NeilBrown16a53ec2006-06-26 00:27:38 -0700415
majianpengad4068d2013-11-14 15:16:15 +1100416 if (unlikely(!conf->mddev->thread) ||
417 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
Shaohua Li773ca822013-08-27 17:50:39 +0800418 goto slow_path;
419 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
420 if (wakeup)
421 md_wakeup_thread(conf->mddev->thread);
422 return;
423slow_path:
Shaohua Li4eb788d2012-07-19 16:01:31 +1000424 local_irq_save(flags);
Shaohua Li773ca822013-08-27 17:50:39 +0800425 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
Shaohua Li4eb788d2012-07-19 16:01:31 +1000426 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
Shaohua Li566c09c2013-11-14 15:16:17 +1100427 INIT_LIST_HEAD(&list);
428 hash = sh->hash_lock_index;
429 do_release_stripe(conf, sh, &list);
Shaohua Li4eb788d2012-07-19 16:01:31 +1000430 spin_unlock(&conf->device_lock);
Shaohua Li566c09c2013-11-14 15:16:17 +1100431 release_inactive_stripe_list(conf, &list, hash);
Shaohua Li4eb788d2012-07-19 16:01:31 +1000432 }
433 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
NeilBrownfccddba2006-01-06 00:20:33 -0800436static inline void remove_hash(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
Dan Williams45b42332007-07-09 11:56:43 -0700438 pr_debug("remove_hash(), stripe %llu\n",
439 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
NeilBrownfccddba2006-01-06 00:20:33 -0800441 hlist_del_init(&sh->hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443
NeilBrownd1688a62011-10-11 16:49:52 +1100444static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
NeilBrownfccddba2006-01-06 00:20:33 -0800446 struct hlist_head *hp = stripe_hash(conf, sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Dan Williams45b42332007-07-09 11:56:43 -0700448 pr_debug("insert_hash(), stripe %llu\n",
449 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
NeilBrownfccddba2006-01-06 00:20:33 -0800451 hlist_add_head(&sh->hash, hp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
454
455/* find an idle stripe, make sure it is unhashed, and return it. */
Shaohua Li566c09c2013-11-14 15:16:17 +1100456static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 struct stripe_head *sh = NULL;
459 struct list_head *first;
460
Shaohua Li566c09c2013-11-14 15:16:17 +1100461 if (list_empty(conf->inactive_list + hash))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 goto out;
Shaohua Li566c09c2013-11-14 15:16:17 +1100463 first = (conf->inactive_list + hash)->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 sh = list_entry(first, struct stripe_head, lru);
465 list_del_init(first);
466 remove_hash(sh);
467 atomic_inc(&conf->active_stripes);
Shaohua Li566c09c2013-11-14 15:16:17 +1100468 BUG_ON(hash != sh->hash_lock_index);
Shaohua Li4bda5562013-11-14 15:16:17 +1100469 if (list_empty(conf->inactive_list + hash))
470 atomic_inc(&conf->empty_inactive_list_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471out:
472 return sh;
473}
474
NeilBrowne4e11e32010-06-16 16:45:16 +1000475static void shrink_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 struct page *p;
478 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000479 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
NeilBrowne4e11e32010-06-16 16:45:16 +1000481 for (i = 0; i < num ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 p = sh->dev[i].page;
483 if (!p)
484 continue;
485 sh->dev[i].page = NULL;
NeilBrown2d1f3b52006-01-06 00:20:31 -0800486 put_page(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 }
488}
489
NeilBrowne4e11e32010-06-16 16:45:16 +1000490static int grow_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
492 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000493 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
NeilBrowne4e11e32010-06-16 16:45:16 +1000495 for (i = 0; i < num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 struct page *page;
497
498 if (!(page = alloc_page(GFP_KERNEL))) {
499 return 1;
500 }
501 sh->dev[i].page = page;
502 }
503 return 0;
504}
505
NeilBrown784052e2009-03-31 15:19:07 +1100506static void raid5_build_block(struct stripe_head *sh, int i, int previous);
NeilBrownd1688a62011-10-11 16:49:52 +1100507static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
NeilBrown911d4ee2009-03-31 14:39:38 +1100508 struct stripe_head *sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
NeilBrownb5663ba2009-03-31 14:39:38 +1100510static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
NeilBrownd1688a62011-10-11 16:49:52 +1100512 struct r5conf *conf = sh->raid_conf;
Shaohua Li566c09c2013-11-14 15:16:17 +1100513 int i, seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200515 BUG_ON(atomic_read(&sh->count) != 0);
516 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
Dan Williams600aa102008-06-28 08:32:05 +1000517 BUG_ON(stripe_operations_active(sh));
Dan Williamsd84e0f12007-01-02 13:52:30 -0700518
Dan Williams45b42332007-07-09 11:56:43 -0700519 pr_debug("init_stripe called, stripe %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 (unsigned long long)sh->sector);
521
522 remove_hash(sh);
Shaohua Li566c09c2013-11-14 15:16:17 +1100523retry:
524 seq = read_seqcount_begin(&conf->gen_lock);
NeilBrown86b42c72009-03-31 15:19:03 +1100525 sh->generation = conf->generation - previous;
NeilBrownb5663ba2009-03-31 14:39:38 +1100526 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 sh->sector = sector;
NeilBrown911d4ee2009-03-31 14:39:38 +1100528 stripe_set_idx(sector, conf, previous, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 sh->state = 0;
530
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800531
532 for (i = sh->disks; i--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 struct r5dev *dev = &sh->dev[i];
534
Dan Williamsd84e0f12007-01-02 13:52:30 -0700535 if (dev->toread || dev->read || dev->towrite || dev->written ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 test_bit(R5_LOCKED, &dev->flags)) {
Dan Williamsd84e0f12007-01-02 13:52:30 -0700537 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 (unsigned long long)sh->sector, i, dev->toread,
Dan Williamsd84e0f12007-01-02 13:52:30 -0700539 dev->read, dev->towrite, dev->written,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 test_bit(R5_LOCKED, &dev->flags));
NeilBrown8cfa7b02011-07-27 11:00:36 +1000541 WARN_ON(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +1100544 raid5_build_block(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 }
Shaohua Li566c09c2013-11-14 15:16:17 +1100546 if (read_seqcount_retry(&conf->gen_lock, seq))
547 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 insert_hash(conf, sh);
Shaohua Li851c30c2013-08-28 14:30:16 +0800549 sh->cpu = smp_processor_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
NeilBrownd1688a62011-10-11 16:49:52 +1100552static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
NeilBrown86b42c72009-03-31 15:19:03 +1100553 short generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 struct stripe_head *sh;
556
Dan Williams45b42332007-07-09 11:56:43 -0700557 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800558 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
NeilBrown86b42c72009-03-31 15:19:03 +1100559 if (sh->sector == sector && sh->generation == generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 return sh;
Dan Williams45b42332007-07-09 11:56:43 -0700561 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 return NULL;
563}
564
NeilBrown674806d2010-06-16 17:17:53 +1000565/*
566 * Need to check if array has failed when deciding whether to:
567 * - start an array
568 * - remove non-faulty devices
569 * - add a spare
570 * - allow a reshape
571 * This determination is simple when no reshape is happening.
572 * However if there is a reshape, we need to carefully check
573 * both the before and after sections.
574 * This is because some failed devices may only affect one
575 * of the two sections, and some non-in_sync devices may
576 * be insync in the section most affected by failed devices.
577 */
NeilBrown908f4fb2011-12-23 10:17:50 +1100578static int calc_degraded(struct r5conf *conf)
NeilBrown674806d2010-06-16 17:17:53 +1000579{
NeilBrown908f4fb2011-12-23 10:17:50 +1100580 int degraded, degraded2;
NeilBrown674806d2010-06-16 17:17:53 +1000581 int i;
NeilBrown674806d2010-06-16 17:17:53 +1000582
583 rcu_read_lock();
584 degraded = 0;
585 for (i = 0; i < conf->previous_raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100586 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowne5c86472012-09-19 12:52:30 +1000587 if (rdev && test_bit(Faulty, &rdev->flags))
588 rdev = rcu_dereference(conf->disks[i].replacement);
NeilBrown674806d2010-06-16 17:17:53 +1000589 if (!rdev || test_bit(Faulty, &rdev->flags))
590 degraded++;
591 else if (test_bit(In_sync, &rdev->flags))
592 ;
593 else
594 /* not in-sync or faulty.
595 * If the reshape increases the number of devices,
596 * this is being recovered by the reshape, so
597 * this 'previous' section is not in_sync.
598 * If the number of devices is being reduced however,
599 * the device can only be part of the array if
600 * we are reverting a reshape, so this section will
601 * be in-sync.
602 */
603 if (conf->raid_disks >= conf->previous_raid_disks)
604 degraded++;
605 }
606 rcu_read_unlock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100607 if (conf->raid_disks == conf->previous_raid_disks)
608 return degraded;
NeilBrown674806d2010-06-16 17:17:53 +1000609 rcu_read_lock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100610 degraded2 = 0;
NeilBrown674806d2010-06-16 17:17:53 +1000611 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100612 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowne5c86472012-09-19 12:52:30 +1000613 if (rdev && test_bit(Faulty, &rdev->flags))
614 rdev = rcu_dereference(conf->disks[i].replacement);
NeilBrown674806d2010-06-16 17:17:53 +1000615 if (!rdev || test_bit(Faulty, &rdev->flags))
NeilBrown908f4fb2011-12-23 10:17:50 +1100616 degraded2++;
NeilBrown674806d2010-06-16 17:17:53 +1000617 else if (test_bit(In_sync, &rdev->flags))
618 ;
619 else
620 /* not in-sync or faulty.
621 * If reshape increases the number of devices, this
622 * section has already been recovered, else it
623 * almost certainly hasn't.
624 */
625 if (conf->raid_disks <= conf->previous_raid_disks)
NeilBrown908f4fb2011-12-23 10:17:50 +1100626 degraded2++;
NeilBrown674806d2010-06-16 17:17:53 +1000627 }
628 rcu_read_unlock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100629 if (degraded2 > degraded)
630 return degraded2;
631 return degraded;
632}
633
634static int has_failed(struct r5conf *conf)
635{
636 int degraded;
637
638 if (conf->mddev->reshape_position == MaxSector)
639 return conf->mddev->degraded > conf->max_degraded;
640
641 degraded = calc_degraded(conf);
NeilBrown674806d2010-06-16 17:17:53 +1000642 if (degraded > conf->max_degraded)
643 return 1;
644 return 0;
645}
646
NeilBrownb5663ba2009-03-31 14:39:38 +1100647static struct stripe_head *
NeilBrownd1688a62011-10-11 16:49:52 +1100648get_active_stripe(struct r5conf *conf, sector_t sector,
NeilBrowna8c906c2009-06-09 14:39:59 +1000649 int previous, int noblock, int noquiesce)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 struct stripe_head *sh;
Shaohua Li566c09c2013-11-14 15:16:17 +1100652 int hash = stripe_hash_locks_hash(sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Dan Williams45b42332007-07-09 11:56:43 -0700654 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Shaohua Li566c09c2013-11-14 15:16:17 +1100656 spin_lock_irq(conf->hash_locks + hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658 do {
NeilBrown72626682005-09-09 16:23:54 -0700659 wait_event_lock_irq(conf->wait_for_stripe,
NeilBrowna8c906c2009-06-09 14:39:59 +1000660 conf->quiesce == 0 || noquiesce,
Shaohua Li566c09c2013-11-14 15:16:17 +1100661 *(conf->hash_locks + hash));
NeilBrown86b42c72009-03-31 15:19:03 +1100662 sh = __find_stripe(conf, sector, conf->generation - previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 if (!sh) {
664 if (!conf->inactive_blocked)
Shaohua Li566c09c2013-11-14 15:16:17 +1100665 sh = get_free_stripe(conf, hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 if (noblock && sh == NULL)
667 break;
668 if (!sh) {
669 conf->inactive_blocked = 1;
Shaohua Li566c09c2013-11-14 15:16:17 +1100670 wait_event_lock_irq(
671 conf->wait_for_stripe,
672 !list_empty(conf->inactive_list + hash) &&
673 (atomic_read(&conf->active_stripes)
674 < (conf->max_nr_stripes * 3 / 4)
675 || !conf->inactive_blocked),
676 *(conf->hash_locks + hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 conf->inactive_blocked = 0;
678 } else
NeilBrownb5663ba2009-03-31 14:39:38 +1100679 init_stripe(sh, sector, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 } else {
681 if (atomic_read(&sh->count)) {
NeilBrownab69ae12009-03-31 15:26:47 +1100682 BUG_ON(!list_empty(&sh->lru)
Shaohua Li8811b592012-08-02 08:33:00 +1000683 && !test_bit(STRIPE_EXPANDING, &sh->state)
Shaohua Li773ca822013-08-27 17:50:39 +0800684 && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)
685 && !test_bit(STRIPE_ON_RELEASE_LIST, &sh->state));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 } else {
Shaohua Li566c09c2013-11-14 15:16:17 +1100687 spin_lock(&conf->device_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (!test_bit(STRIPE_HANDLE, &sh->state))
689 atomic_inc(&conf->active_stripes);
NeilBrownff4e8d92006-07-10 04:44:16 -0700690 if (list_empty(&sh->lru) &&
Shaohua Li566c09c2013-11-14 15:16:17 +1100691 !test_bit(STRIPE_ON_RELEASE_LIST, &sh->state) &&
NeilBrownff4e8d92006-07-10 04:44:16 -0700692 !test_bit(STRIPE_EXPANDING, &sh->state))
NeilBrown16a53ec2006-06-26 00:27:38 -0700693 BUG();
694 list_del_init(&sh->lru);
Shaohua Libfc90cb2013-08-29 15:40:32 +0800695 if (sh->group) {
696 sh->group->stripes_cnt--;
697 sh->group = NULL;
698 }
Shaohua Li566c09c2013-11-14 15:16:17 +1100699 spin_unlock(&conf->device_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701 }
702 } while (sh == NULL);
703
704 if (sh)
705 atomic_inc(&sh->count);
706
Shaohua Li566c09c2013-11-14 15:16:17 +1100707 spin_unlock_irq(conf->hash_locks + hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return sh;
709}
710
NeilBrown05616be2012-05-21 09:27:00 +1000711/* Determine if 'data_offset' or 'new_data_offset' should be used
712 * in this stripe_head.
713 */
714static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
715{
716 sector_t progress = conf->reshape_progress;
717 /* Need a memory barrier to make sure we see the value
718 * of conf->generation, or ->data_offset that was set before
719 * reshape_progress was updated.
720 */
721 smp_rmb();
722 if (progress == MaxSector)
723 return 0;
724 if (sh->generation == conf->generation - 1)
725 return 0;
726 /* We are in a reshape, and this is a new-generation stripe,
727 * so use new_data_offset.
728 */
729 return 1;
730}
731
NeilBrown6712ecf2007-09-27 12:47:43 +0200732static void
733raid5_end_read_request(struct bio *bi, int error);
734static void
735raid5_end_write_request(struct bio *bi, int error);
Dan Williams91c00922007-01-02 13:52:30 -0700736
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000737static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
Dan Williams91c00922007-01-02 13:52:30 -0700738{
NeilBrownd1688a62011-10-11 16:49:52 +1100739 struct r5conf *conf = sh->raid_conf;
Dan Williams91c00922007-01-02 13:52:30 -0700740 int i, disks = sh->disks;
741
742 might_sleep();
743
744 for (i = disks; i--; ) {
745 int rw;
NeilBrown9a3e1102011-12-23 10:17:53 +1100746 int replace_only = 0;
NeilBrown977df362011-12-23 10:17:53 +1100747 struct bio *bi, *rbi;
748 struct md_rdev *rdev, *rrdev = NULL;
Tejun Heoe9c74692010-09-03 11:56:18 +0200749 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
750 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
751 rw = WRITE_FUA;
752 else
753 rw = WRITE;
Shaohua Li9e4447682012-10-11 13:49:49 +1100754 if (test_bit(R5_Discard, &sh->dev[i].flags))
Shaohua Li620125f2012-10-11 13:49:05 +1100755 rw |= REQ_DISCARD;
Tejun Heoe9c74692010-09-03 11:56:18 +0200756 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
Dan Williams91c00922007-01-02 13:52:30 -0700757 rw = READ;
NeilBrown9a3e1102011-12-23 10:17:53 +1100758 else if (test_and_clear_bit(R5_WantReplace,
759 &sh->dev[i].flags)) {
760 rw = WRITE;
761 replace_only = 1;
762 } else
Dan Williams91c00922007-01-02 13:52:30 -0700763 continue;
Shaohua Libc0934f2012-05-22 13:55:05 +1000764 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
765 rw |= REQ_SYNC;
Dan Williams91c00922007-01-02 13:52:30 -0700766
767 bi = &sh->dev[i].req;
NeilBrown977df362011-12-23 10:17:53 +1100768 rbi = &sh->dev[i].rreq; /* For writing to replacement */
Dan Williams91c00922007-01-02 13:52:30 -0700769
Dan Williams91c00922007-01-02 13:52:30 -0700770 rcu_read_lock();
NeilBrown9a3e1102011-12-23 10:17:53 +1100771 rrdev = rcu_dereference(conf->disks[i].replacement);
NeilBrowndd054fc2011-12-23 10:17:53 +1100772 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
773 rdev = rcu_dereference(conf->disks[i].rdev);
774 if (!rdev) {
775 rdev = rrdev;
776 rrdev = NULL;
777 }
NeilBrown9a3e1102011-12-23 10:17:53 +1100778 if (rw & WRITE) {
779 if (replace_only)
780 rdev = NULL;
NeilBrowndd054fc2011-12-23 10:17:53 +1100781 if (rdev == rrdev)
782 /* We raced and saw duplicates */
783 rrdev = NULL;
NeilBrown9a3e1102011-12-23 10:17:53 +1100784 } else {
NeilBrowndd054fc2011-12-23 10:17:53 +1100785 if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev)
NeilBrown9a3e1102011-12-23 10:17:53 +1100786 rdev = rrdev;
787 rrdev = NULL;
788 }
NeilBrown977df362011-12-23 10:17:53 +1100789
Dan Williams91c00922007-01-02 13:52:30 -0700790 if (rdev && test_bit(Faulty, &rdev->flags))
791 rdev = NULL;
792 if (rdev)
793 atomic_inc(&rdev->nr_pending);
NeilBrown977df362011-12-23 10:17:53 +1100794 if (rrdev && test_bit(Faulty, &rrdev->flags))
795 rrdev = NULL;
796 if (rrdev)
797 atomic_inc(&rrdev->nr_pending);
Dan Williams91c00922007-01-02 13:52:30 -0700798 rcu_read_unlock();
799
NeilBrown73e92e52011-07-28 11:39:22 +1000800 /* We have already checked bad blocks for reads. Now
NeilBrown977df362011-12-23 10:17:53 +1100801 * need to check for writes. We never accept write errors
802 * on the replacement, so we don't to check rrdev.
NeilBrown73e92e52011-07-28 11:39:22 +1000803 */
804 while ((rw & WRITE) && rdev &&
805 test_bit(WriteErrorSeen, &rdev->flags)) {
806 sector_t first_bad;
807 int bad_sectors;
808 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
809 &first_bad, &bad_sectors);
810 if (!bad)
811 break;
812
813 if (bad < 0) {
814 set_bit(BlockedBadBlocks, &rdev->flags);
815 if (!conf->mddev->external &&
816 conf->mddev->flags) {
817 /* It is very unlikely, but we might
818 * still need to write out the
819 * bad block log - better give it
820 * a chance*/
821 md_check_recovery(conf->mddev);
822 }
majianpeng18507532012-07-03 12:11:54 +1000823 /*
824 * Because md_wait_for_blocked_rdev
825 * will dec nr_pending, we must
826 * increment it first.
827 */
828 atomic_inc(&rdev->nr_pending);
NeilBrown73e92e52011-07-28 11:39:22 +1000829 md_wait_for_blocked_rdev(rdev, conf->mddev);
830 } else {
831 /* Acknowledged bad block - skip the write */
832 rdev_dec_pending(rdev, conf->mddev);
833 rdev = NULL;
834 }
835 }
836
Dan Williams91c00922007-01-02 13:52:30 -0700837 if (rdev) {
NeilBrown9a3e1102011-12-23 10:17:53 +1100838 if (s->syncing || s->expanding || s->expanded
839 || s->replacing)
Dan Williams91c00922007-01-02 13:52:30 -0700840 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
841
Dan Williams2b7497f2008-06-28 08:31:52 +1000842 set_bit(STRIPE_IO_STARTED, &sh->state);
843
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700844 bio_reset(bi);
Dan Williams91c00922007-01-02 13:52:30 -0700845 bi->bi_bdev = rdev->bdev;
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700846 bi->bi_rw = rw;
847 bi->bi_end_io = (rw & WRITE)
848 ? raid5_end_write_request
849 : raid5_end_read_request;
850 bi->bi_private = sh;
851
Dan Williams91c00922007-01-02 13:52:30 -0700852 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700853 __func__, (unsigned long long)sh->sector,
Dan Williams91c00922007-01-02 13:52:30 -0700854 bi->bi_rw, i);
855 atomic_inc(&sh->count);
NeilBrown05616be2012-05-21 09:27:00 +1000856 if (use_new_offset(conf, sh))
857 bi->bi_sector = (sh->sector
858 + rdev->new_data_offset);
859 else
860 bi->bi_sector = (sh->sector
861 + rdev->data_offset);
majianpeng3f9e7c12012-07-31 10:04:21 +1000862 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
majianpenge59aa232013-11-14 15:16:19 +1100863 bi->bi_rw |= REQ_NOMERGE;
majianpeng3f9e7c12012-07-31 10:04:21 +1000864
Kent Overstreet4997b722013-05-30 08:44:39 +0200865 bi->bi_vcnt = 1;
Dan Williams91c00922007-01-02 13:52:30 -0700866 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
867 bi->bi_io_vec[0].bv_offset = 0;
868 bi->bi_size = STRIPE_SIZE;
Shaohua Li37c61ff2013-10-19 14:50:28 +0800869 /*
870 * If this is discard request, set bi_vcnt 0. We don't
871 * want to confuse SCSI because SCSI will replace payload
872 */
873 if (rw & REQ_DISCARD)
874 bi->bi_vcnt = 0;
NeilBrown977df362011-12-23 10:17:53 +1100875 if (rrdev)
876 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
Jonathan Brassowe3620a32013-03-07 16:22:01 -0600877
878 if (conf->mddev->gendisk)
879 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
880 bi, disk_devt(conf->mddev->gendisk),
881 sh->dev[i].sector);
Dan Williams91c00922007-01-02 13:52:30 -0700882 generic_make_request(bi);
NeilBrown977df362011-12-23 10:17:53 +1100883 }
884 if (rrdev) {
NeilBrown9a3e1102011-12-23 10:17:53 +1100885 if (s->syncing || s->expanding || s->expanded
886 || s->replacing)
NeilBrown977df362011-12-23 10:17:53 +1100887 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
888
889 set_bit(STRIPE_IO_STARTED, &sh->state);
890
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700891 bio_reset(rbi);
NeilBrown977df362011-12-23 10:17:53 +1100892 rbi->bi_bdev = rrdev->bdev;
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700893 rbi->bi_rw = rw;
894 BUG_ON(!(rw & WRITE));
895 rbi->bi_end_io = raid5_end_write_request;
896 rbi->bi_private = sh;
897
NeilBrown977df362011-12-23 10:17:53 +1100898 pr_debug("%s: for %llu schedule op %ld on "
899 "replacement disc %d\n",
900 __func__, (unsigned long long)sh->sector,
901 rbi->bi_rw, i);
902 atomic_inc(&sh->count);
NeilBrown05616be2012-05-21 09:27:00 +1000903 if (use_new_offset(conf, sh))
904 rbi->bi_sector = (sh->sector
905 + rrdev->new_data_offset);
906 else
907 rbi->bi_sector = (sh->sector
908 + rrdev->data_offset);
Kent Overstreet4997b722013-05-30 08:44:39 +0200909 rbi->bi_vcnt = 1;
NeilBrown977df362011-12-23 10:17:53 +1100910 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
911 rbi->bi_io_vec[0].bv_offset = 0;
912 rbi->bi_size = STRIPE_SIZE;
Shaohua Li37c61ff2013-10-19 14:50:28 +0800913 /*
914 * If this is discard request, set bi_vcnt 0. We don't
915 * want to confuse SCSI because SCSI will replace payload
916 */
917 if (rw & REQ_DISCARD)
918 rbi->bi_vcnt = 0;
Jonathan Brassowe3620a32013-03-07 16:22:01 -0600919 if (conf->mddev->gendisk)
920 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
921 rbi, disk_devt(conf->mddev->gendisk),
922 sh->dev[i].sector);
NeilBrown977df362011-12-23 10:17:53 +1100923 generic_make_request(rbi);
924 }
925 if (!rdev && !rrdev) {
Namhyung Kimb0629622011-06-14 14:20:19 +1000926 if (rw & WRITE)
Dan Williams91c00922007-01-02 13:52:30 -0700927 set_bit(STRIPE_DEGRADED, &sh->state);
928 pr_debug("skip op %ld on disc %d for sector %llu\n",
929 bi->bi_rw, i, (unsigned long long)sh->sector);
930 clear_bit(R5_LOCKED, &sh->dev[i].flags);
931 set_bit(STRIPE_HANDLE, &sh->state);
932 }
933 }
934}
935
936static struct dma_async_tx_descriptor *
937async_copy_data(int frombio, struct bio *bio, struct page *page,
938 sector_t sector, struct dma_async_tx_descriptor *tx)
939{
940 struct bio_vec *bvl;
941 struct page *bio_page;
942 int i;
943 int page_offset;
Dan Williamsa08abd82009-06-03 11:43:59 -0700944 struct async_submit_ctl submit;
Dan Williams0403e382009-09-08 17:42:50 -0700945 enum async_tx_flags flags = 0;
Dan Williams91c00922007-01-02 13:52:30 -0700946
947 if (bio->bi_sector >= sector)
948 page_offset = (signed)(bio->bi_sector - sector) * 512;
949 else
950 page_offset = (signed)(sector - bio->bi_sector) * -512;
Dan Williamsa08abd82009-06-03 11:43:59 -0700951
Dan Williams0403e382009-09-08 17:42:50 -0700952 if (frombio)
953 flags |= ASYNC_TX_FENCE;
954 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
955
Dan Williams91c00922007-01-02 13:52:30 -0700956 bio_for_each_segment(bvl, bio, i) {
Namhyung Kimfcde9072011-06-14 14:23:57 +1000957 int len = bvl->bv_len;
Dan Williams91c00922007-01-02 13:52:30 -0700958 int clen;
959 int b_offset = 0;
960
961 if (page_offset < 0) {
962 b_offset = -page_offset;
963 page_offset += b_offset;
964 len -= b_offset;
965 }
966
967 if (len > 0 && page_offset + len > STRIPE_SIZE)
968 clen = STRIPE_SIZE - page_offset;
969 else
970 clen = len;
971
972 if (clen > 0) {
Namhyung Kimfcde9072011-06-14 14:23:57 +1000973 b_offset += bvl->bv_offset;
974 bio_page = bvl->bv_page;
Dan Williams91c00922007-01-02 13:52:30 -0700975 if (frombio)
976 tx = async_memcpy(page, bio_page, page_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700977 b_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700978 else
979 tx = async_memcpy(bio_page, page, b_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700980 page_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700981 }
Dan Williamsa08abd82009-06-03 11:43:59 -0700982 /* chain the operations */
983 submit.depend_tx = tx;
984
Dan Williams91c00922007-01-02 13:52:30 -0700985 if (clen < len) /* hit end of page */
986 break;
987 page_offset += len;
988 }
989
990 return tx;
991}
992
993static void ops_complete_biofill(void *stripe_head_ref)
994{
995 struct stripe_head *sh = stripe_head_ref;
996 struct bio *return_bi = NULL;
Dan Williamse4d84902007-09-24 10:06:13 -0700997 int i;
Dan Williams91c00922007-01-02 13:52:30 -0700998
Harvey Harrisone46b2722008-04-28 02:15:50 -0700999 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001000 (unsigned long long)sh->sector);
1001
1002 /* clear completed biofills */
1003 for (i = sh->disks; i--; ) {
1004 struct r5dev *dev = &sh->dev[i];
Dan Williams91c00922007-01-02 13:52:30 -07001005
1006 /* acknowledge completion of a biofill operation */
Dan Williamse4d84902007-09-24 10:06:13 -07001007 /* and check if we need to reply to a read request,
1008 * new R5_Wantfill requests are held off until
Dan Williams83de75c2008-06-28 08:31:58 +10001009 * !STRIPE_BIOFILL_RUN
Dan Williamse4d84902007-09-24 10:06:13 -07001010 */
1011 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001012 struct bio *rbi, *rbi2;
Dan Williams91c00922007-01-02 13:52:30 -07001013
Dan Williams91c00922007-01-02 13:52:30 -07001014 BUG_ON(!dev->read);
1015 rbi = dev->read;
1016 dev->read = NULL;
1017 while (rbi && rbi->bi_sector <
1018 dev->sector + STRIPE_SECTORS) {
1019 rbi2 = r5_next_bio(rbi, dev->sector);
Shaohua Lie7836bd62012-07-19 16:01:31 +10001020 if (!raid5_dec_bi_active_stripes(rbi)) {
Dan Williams91c00922007-01-02 13:52:30 -07001021 rbi->bi_next = return_bi;
1022 return_bi = rbi;
1023 }
Dan Williams91c00922007-01-02 13:52:30 -07001024 rbi = rbi2;
1025 }
1026 }
1027 }
Dan Williams83de75c2008-06-28 08:31:58 +10001028 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -07001029
1030 return_io(return_bi);
1031
Dan Williamse4d84902007-09-24 10:06:13 -07001032 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -07001033 release_stripe(sh);
1034}
1035
1036static void ops_run_biofill(struct stripe_head *sh)
1037{
1038 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa08abd82009-06-03 11:43:59 -07001039 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001040 int i;
1041
Harvey Harrisone46b2722008-04-28 02:15:50 -07001042 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001043 (unsigned long long)sh->sector);
1044
1045 for (i = sh->disks; i--; ) {
1046 struct r5dev *dev = &sh->dev[i];
1047 if (test_bit(R5_Wantfill, &dev->flags)) {
1048 struct bio *rbi;
Shaohua Lib17459c2012-07-19 16:01:31 +10001049 spin_lock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001050 dev->read = rbi = dev->toread;
1051 dev->toread = NULL;
Shaohua Lib17459c2012-07-19 16:01:31 +10001052 spin_unlock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001053 while (rbi && rbi->bi_sector <
1054 dev->sector + STRIPE_SECTORS) {
1055 tx = async_copy_data(0, rbi, dev->page,
1056 dev->sector, tx);
1057 rbi = r5_next_bio(rbi, dev->sector);
1058 }
1059 }
1060 }
1061
1062 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001063 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1064 async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001065}
1066
Dan Williams4e7d2c02009-08-29 19:13:11 -07001067static void mark_target_uptodate(struct stripe_head *sh, int target)
1068{
1069 struct r5dev *tgt;
1070
1071 if (target < 0)
1072 return;
1073
1074 tgt = &sh->dev[target];
1075 set_bit(R5_UPTODATE, &tgt->flags);
1076 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1077 clear_bit(R5_Wantcompute, &tgt->flags);
1078}
1079
Dan Williamsac6b53b2009-07-14 13:40:19 -07001080static void ops_complete_compute(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001081{
1082 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001083
Harvey Harrisone46b2722008-04-28 02:15:50 -07001084 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001085 (unsigned long long)sh->sector);
1086
Dan Williamsac6b53b2009-07-14 13:40:19 -07001087 /* mark the computed target(s) as uptodate */
Dan Williams4e7d2c02009-08-29 19:13:11 -07001088 mark_target_uptodate(sh, sh->ops.target);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001089 mark_target_uptodate(sh, sh->ops.target2);
Dan Williams4e7d2c02009-08-29 19:13:11 -07001090
Dan Williamsecc65c92008-06-28 08:31:57 +10001091 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1092 if (sh->check_state == check_state_compute_run)
1093 sh->check_state = check_state_compute_result;
Dan Williams91c00922007-01-02 13:52:30 -07001094 set_bit(STRIPE_HANDLE, &sh->state);
1095 release_stripe(sh);
1096}
1097
Dan Williamsd6f38f32009-07-14 11:50:52 -07001098/* return a pointer to the address conversion region of the scribble buffer */
1099static addr_conv_t *to_addr_conv(struct stripe_head *sh,
1100 struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001101{
Dan Williamsd6f38f32009-07-14 11:50:52 -07001102 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
1103}
1104
1105static struct dma_async_tx_descriptor *
1106ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
1107{
Dan Williams91c00922007-01-02 13:52:30 -07001108 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001109 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001110 int target = sh->ops.target;
1111 struct r5dev *tgt = &sh->dev[target];
1112 struct page *xor_dest = tgt->page;
1113 int count = 0;
1114 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001115 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001116 int i;
1117
1118 pr_debug("%s: stripe %llu block: %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07001119 __func__, (unsigned long long)sh->sector, target);
Dan Williams91c00922007-01-02 13:52:30 -07001120 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1121
1122 for (i = disks; i--; )
1123 if (i != target)
1124 xor_srcs[count++] = sh->dev[i].page;
1125
1126 atomic_inc(&sh->count);
1127
Dan Williams0403e382009-09-08 17:42:50 -07001128 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001129 ops_complete_compute, sh, to_addr_conv(sh, percpu));
Dan Williams91c00922007-01-02 13:52:30 -07001130 if (unlikely(count == 1))
Dan Williamsa08abd82009-06-03 11:43:59 -07001131 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001132 else
Dan Williamsa08abd82009-06-03 11:43:59 -07001133 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001134
Dan Williams91c00922007-01-02 13:52:30 -07001135 return tx;
1136}
1137
Dan Williamsac6b53b2009-07-14 13:40:19 -07001138/* set_syndrome_sources - populate source buffers for gen_syndrome
1139 * @srcs - (struct page *) array of size sh->disks
1140 * @sh - stripe_head to parse
1141 *
1142 * Populates srcs in proper layout order for the stripe and returns the
1143 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1144 * destination buffer is recorded in srcs[count] and the Q destination
1145 * is recorded in srcs[count+1]].
1146 */
1147static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
1148{
1149 int disks = sh->disks;
1150 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1151 int d0_idx = raid6_d0(sh);
1152 int count;
1153 int i;
1154
1155 for (i = 0; i < disks; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +11001156 srcs[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001157
1158 count = 0;
1159 i = d0_idx;
1160 do {
1161 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1162
1163 srcs[slot] = sh->dev[i].page;
1164 i = raid6_next_disk(i, disks);
1165 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001166
NeilBrowne4424fe2009-10-16 16:27:34 +11001167 return syndrome_disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001168}
1169
1170static struct dma_async_tx_descriptor *
1171ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1172{
1173 int disks = sh->disks;
1174 struct page **blocks = percpu->scribble;
1175 int target;
1176 int qd_idx = sh->qd_idx;
1177 struct dma_async_tx_descriptor *tx;
1178 struct async_submit_ctl submit;
1179 struct r5dev *tgt;
1180 struct page *dest;
1181 int i;
1182 int count;
1183
1184 if (sh->ops.target < 0)
1185 target = sh->ops.target2;
1186 else if (sh->ops.target2 < 0)
1187 target = sh->ops.target;
1188 else
1189 /* we should only have one valid target */
1190 BUG();
1191 BUG_ON(target < 0);
1192 pr_debug("%s: stripe %llu block: %d\n",
1193 __func__, (unsigned long long)sh->sector, target);
1194
1195 tgt = &sh->dev[target];
1196 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1197 dest = tgt->page;
1198
1199 atomic_inc(&sh->count);
1200
1201 if (target == qd_idx) {
1202 count = set_syndrome_sources(blocks, sh);
1203 blocks[count] = NULL; /* regenerating p is not necessary */
1204 BUG_ON(blocks[count+1] != dest); /* q should already be set */
Dan Williams0403e382009-09-08 17:42:50 -07001205 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1206 ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001207 to_addr_conv(sh, percpu));
1208 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1209 } else {
1210 /* Compute any data- or p-drive using XOR */
1211 count = 0;
1212 for (i = disks; i-- ; ) {
1213 if (i == target || i == qd_idx)
1214 continue;
1215 blocks[count++] = sh->dev[i].page;
1216 }
1217
Dan Williams0403e382009-09-08 17:42:50 -07001218 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1219 NULL, ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001220 to_addr_conv(sh, percpu));
1221 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1222 }
1223
1224 return tx;
1225}
1226
1227static struct dma_async_tx_descriptor *
1228ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1229{
1230 int i, count, disks = sh->disks;
1231 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1232 int d0_idx = raid6_d0(sh);
1233 int faila = -1, failb = -1;
1234 int target = sh->ops.target;
1235 int target2 = sh->ops.target2;
1236 struct r5dev *tgt = &sh->dev[target];
1237 struct r5dev *tgt2 = &sh->dev[target2];
1238 struct dma_async_tx_descriptor *tx;
1239 struct page **blocks = percpu->scribble;
1240 struct async_submit_ctl submit;
1241
1242 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1243 __func__, (unsigned long long)sh->sector, target, target2);
1244 BUG_ON(target < 0 || target2 < 0);
1245 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1246 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1247
Dan Williams6c910a72009-09-16 12:24:54 -07001248 /* we need to open-code set_syndrome_sources to handle the
Dan Williamsac6b53b2009-07-14 13:40:19 -07001249 * slot number conversion for 'faila' and 'failb'
1250 */
1251 for (i = 0; i < disks ; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +11001252 blocks[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001253 count = 0;
1254 i = d0_idx;
1255 do {
1256 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1257
1258 blocks[slot] = sh->dev[i].page;
1259
1260 if (i == target)
1261 faila = slot;
1262 if (i == target2)
1263 failb = slot;
1264 i = raid6_next_disk(i, disks);
1265 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001266
1267 BUG_ON(faila == failb);
1268 if (failb < faila)
1269 swap(faila, failb);
1270 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1271 __func__, (unsigned long long)sh->sector, faila, failb);
1272
1273 atomic_inc(&sh->count);
1274
1275 if (failb == syndrome_disks+1) {
1276 /* Q disk is one of the missing disks */
1277 if (faila == syndrome_disks) {
1278 /* Missing P+Q, just recompute */
Dan Williams0403e382009-09-08 17:42:50 -07001279 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1280 ops_complete_compute, sh,
1281 to_addr_conv(sh, percpu));
NeilBrowne4424fe2009-10-16 16:27:34 +11001282 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001283 STRIPE_SIZE, &submit);
1284 } else {
1285 struct page *dest;
1286 int data_target;
1287 int qd_idx = sh->qd_idx;
1288
1289 /* Missing D+Q: recompute D from P, then recompute Q */
1290 if (target == qd_idx)
1291 data_target = target2;
1292 else
1293 data_target = target;
1294
1295 count = 0;
1296 for (i = disks; i-- ; ) {
1297 if (i == data_target || i == qd_idx)
1298 continue;
1299 blocks[count++] = sh->dev[i].page;
1300 }
1301 dest = sh->dev[data_target].page;
Dan Williams0403e382009-09-08 17:42:50 -07001302 init_async_submit(&submit,
1303 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1304 NULL, NULL, NULL,
1305 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001306 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1307 &submit);
1308
1309 count = set_syndrome_sources(blocks, sh);
Dan Williams0403e382009-09-08 17:42:50 -07001310 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1311 ops_complete_compute, sh,
1312 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001313 return async_gen_syndrome(blocks, 0, count+2,
1314 STRIPE_SIZE, &submit);
1315 }
Dan Williamsac6b53b2009-07-14 13:40:19 -07001316 } else {
Dan Williams6c910a72009-09-16 12:24:54 -07001317 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1318 ops_complete_compute, sh,
1319 to_addr_conv(sh, percpu));
1320 if (failb == syndrome_disks) {
1321 /* We're missing D+P. */
1322 return async_raid6_datap_recov(syndrome_disks+2,
1323 STRIPE_SIZE, faila,
1324 blocks, &submit);
1325 } else {
1326 /* We're missing D+D. */
1327 return async_raid6_2data_recov(syndrome_disks+2,
1328 STRIPE_SIZE, faila, failb,
1329 blocks, &submit);
1330 }
Dan Williamsac6b53b2009-07-14 13:40:19 -07001331 }
1332}
1333
1334
Dan Williams91c00922007-01-02 13:52:30 -07001335static void ops_complete_prexor(void *stripe_head_ref)
1336{
1337 struct stripe_head *sh = stripe_head_ref;
1338
Harvey Harrisone46b2722008-04-28 02:15:50 -07001339 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001340 (unsigned long long)sh->sector);
Dan Williams91c00922007-01-02 13:52:30 -07001341}
1342
1343static struct dma_async_tx_descriptor *
Dan Williamsd6f38f32009-07-14 11:50:52 -07001344ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
1345 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001346{
Dan Williams91c00922007-01-02 13:52:30 -07001347 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001348 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001349 int count = 0, pd_idx = sh->pd_idx, i;
Dan Williamsa08abd82009-06-03 11:43:59 -07001350 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001351
1352 /* existing parity data subtracted */
1353 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1354
Harvey Harrisone46b2722008-04-28 02:15:50 -07001355 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001356 (unsigned long long)sh->sector);
1357
1358 for (i = disks; i--; ) {
1359 struct r5dev *dev = &sh->dev[i];
1360 /* Only process blocks that are known to be uptodate */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001361 if (test_bit(R5_Wantdrain, &dev->flags))
Dan Williams91c00922007-01-02 13:52:30 -07001362 xor_srcs[count++] = dev->page;
1363 }
1364
Dan Williams0403e382009-09-08 17:42:50 -07001365 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001366 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001367 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001368
1369 return tx;
1370}
1371
1372static struct dma_async_tx_descriptor *
Dan Williamsd8ee0722008-06-28 08:32:06 +10001373ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001374{
1375 int disks = sh->disks;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001376 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001377
Harvey Harrisone46b2722008-04-28 02:15:50 -07001378 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001379 (unsigned long long)sh->sector);
1380
1381 for (i = disks; i--; ) {
1382 struct r5dev *dev = &sh->dev[i];
1383 struct bio *chosen;
Dan Williams91c00922007-01-02 13:52:30 -07001384
Dan Williamsd8ee0722008-06-28 08:32:06 +10001385 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001386 struct bio *wbi;
1387
Shaohua Lib17459c2012-07-19 16:01:31 +10001388 spin_lock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001389 chosen = dev->towrite;
1390 dev->towrite = NULL;
1391 BUG_ON(dev->written);
1392 wbi = dev->written = chosen;
Shaohua Lib17459c2012-07-19 16:01:31 +10001393 spin_unlock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001394
1395 while (wbi && wbi->bi_sector <
1396 dev->sector + STRIPE_SECTORS) {
Tejun Heoe9c74692010-09-03 11:56:18 +02001397 if (wbi->bi_rw & REQ_FUA)
1398 set_bit(R5_WantFUA, &dev->flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001399 if (wbi->bi_rw & REQ_SYNC)
1400 set_bit(R5_SyncIO, &dev->flags);
Shaohua Li9e4447682012-10-11 13:49:49 +11001401 if (wbi->bi_rw & REQ_DISCARD)
Shaohua Li620125f2012-10-11 13:49:05 +11001402 set_bit(R5_Discard, &dev->flags);
Shaohua Li9e4447682012-10-11 13:49:49 +11001403 else
Shaohua Li620125f2012-10-11 13:49:05 +11001404 tx = async_copy_data(1, wbi, dev->page,
1405 dev->sector, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001406 wbi = r5_next_bio(wbi, dev->sector);
1407 }
1408 }
1409 }
1410
1411 return tx;
1412}
1413
Dan Williamsac6b53b2009-07-14 13:40:19 -07001414static void ops_complete_reconstruct(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001415{
1416 struct stripe_head *sh = stripe_head_ref;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001417 int disks = sh->disks;
1418 int pd_idx = sh->pd_idx;
1419 int qd_idx = sh->qd_idx;
1420 int i;
Shaohua Li9e4447682012-10-11 13:49:49 +11001421 bool fua = false, sync = false, discard = false;
Dan Williams91c00922007-01-02 13:52:30 -07001422
Harvey Harrisone46b2722008-04-28 02:15:50 -07001423 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001424 (unsigned long long)sh->sector);
1425
Shaohua Libc0934f2012-05-22 13:55:05 +10001426 for (i = disks; i--; ) {
Tejun Heoe9c74692010-09-03 11:56:18 +02001427 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001428 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
Shaohua Li9e4447682012-10-11 13:49:49 +11001429 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001430 }
Tejun Heoe9c74692010-09-03 11:56:18 +02001431
Dan Williams91c00922007-01-02 13:52:30 -07001432 for (i = disks; i--; ) {
1433 struct r5dev *dev = &sh->dev[i];
Dan Williamsac6b53b2009-07-14 13:40:19 -07001434
Tejun Heoe9c74692010-09-03 11:56:18 +02001435 if (dev->written || i == pd_idx || i == qd_idx) {
Shaohua Li9e4447682012-10-11 13:49:49 +11001436 if (!discard)
1437 set_bit(R5_UPTODATE, &dev->flags);
Tejun Heoe9c74692010-09-03 11:56:18 +02001438 if (fua)
1439 set_bit(R5_WantFUA, &dev->flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001440 if (sync)
1441 set_bit(R5_SyncIO, &dev->flags);
Tejun Heoe9c74692010-09-03 11:56:18 +02001442 }
Dan Williams91c00922007-01-02 13:52:30 -07001443 }
1444
Dan Williamsd8ee0722008-06-28 08:32:06 +10001445 if (sh->reconstruct_state == reconstruct_state_drain_run)
1446 sh->reconstruct_state = reconstruct_state_drain_result;
1447 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1448 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1449 else {
1450 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1451 sh->reconstruct_state = reconstruct_state_result;
1452 }
Dan Williams91c00922007-01-02 13:52:30 -07001453
1454 set_bit(STRIPE_HANDLE, &sh->state);
1455 release_stripe(sh);
1456}
1457
1458static void
Dan Williamsac6b53b2009-07-14 13:40:19 -07001459ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1460 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001461{
Dan Williams91c00922007-01-02 13:52:30 -07001462 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001463 struct page **xor_srcs = percpu->scribble;
Dan Williamsa08abd82009-06-03 11:43:59 -07001464 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001465 int count = 0, pd_idx = sh->pd_idx, i;
1466 struct page *xor_dest;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001467 int prexor = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001468 unsigned long flags;
Dan Williams91c00922007-01-02 13:52:30 -07001469
Harvey Harrisone46b2722008-04-28 02:15:50 -07001470 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001471 (unsigned long long)sh->sector);
1472
Shaohua Li620125f2012-10-11 13:49:05 +11001473 for (i = 0; i < sh->disks; i++) {
1474 if (pd_idx == i)
1475 continue;
1476 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1477 break;
1478 }
1479 if (i >= sh->disks) {
1480 atomic_inc(&sh->count);
Shaohua Li620125f2012-10-11 13:49:05 +11001481 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1482 ops_complete_reconstruct(sh);
1483 return;
1484 }
Dan Williams91c00922007-01-02 13:52:30 -07001485 /* check if prexor is active which means only process blocks
1486 * that are part of a read-modify-write (written)
1487 */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001488 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1489 prexor = 1;
Dan Williams91c00922007-01-02 13:52:30 -07001490 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1491 for (i = disks; i--; ) {
1492 struct r5dev *dev = &sh->dev[i];
1493 if (dev->written)
1494 xor_srcs[count++] = dev->page;
1495 }
1496 } else {
1497 xor_dest = sh->dev[pd_idx].page;
1498 for (i = disks; i--; ) {
1499 struct r5dev *dev = &sh->dev[i];
1500 if (i != pd_idx)
1501 xor_srcs[count++] = dev->page;
1502 }
1503 }
1504
Dan Williams91c00922007-01-02 13:52:30 -07001505 /* 1/ if we prexor'd then the dest is reused as a source
1506 * 2/ if we did not prexor then we are redoing the parity
1507 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1508 * for the synchronous xor case
1509 */
Dan Williams88ba2aa2009-04-09 16:16:18 -07001510 flags = ASYNC_TX_ACK |
Dan Williams91c00922007-01-02 13:52:30 -07001511 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1512
1513 atomic_inc(&sh->count);
1514
Dan Williamsac6b53b2009-07-14 13:40:19 -07001515 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001516 to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001517 if (unlikely(count == 1))
1518 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1519 else
1520 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001521}
1522
Dan Williamsac6b53b2009-07-14 13:40:19 -07001523static void
1524ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1525 struct dma_async_tx_descriptor *tx)
1526{
1527 struct async_submit_ctl submit;
1528 struct page **blocks = percpu->scribble;
Shaohua Li620125f2012-10-11 13:49:05 +11001529 int count, i;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001530
1531 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1532
Shaohua Li620125f2012-10-11 13:49:05 +11001533 for (i = 0; i < sh->disks; i++) {
1534 if (sh->pd_idx == i || sh->qd_idx == i)
1535 continue;
1536 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1537 break;
1538 }
1539 if (i >= sh->disks) {
1540 atomic_inc(&sh->count);
Shaohua Li620125f2012-10-11 13:49:05 +11001541 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1542 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1543 ops_complete_reconstruct(sh);
1544 return;
1545 }
1546
Dan Williamsac6b53b2009-07-14 13:40:19 -07001547 count = set_syndrome_sources(blocks, sh);
1548
1549 atomic_inc(&sh->count);
1550
1551 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1552 sh, to_addr_conv(sh, percpu));
1553 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001554}
1555
1556static void ops_complete_check(void *stripe_head_ref)
1557{
1558 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001559
Harvey Harrisone46b2722008-04-28 02:15:50 -07001560 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001561 (unsigned long long)sh->sector);
1562
Dan Williamsecc65c92008-06-28 08:31:57 +10001563 sh->check_state = check_state_check_result;
Dan Williams91c00922007-01-02 13:52:30 -07001564 set_bit(STRIPE_HANDLE, &sh->state);
1565 release_stripe(sh);
1566}
1567
Dan Williamsac6b53b2009-07-14 13:40:19 -07001568static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001569{
Dan Williams91c00922007-01-02 13:52:30 -07001570 int disks = sh->disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001571 int pd_idx = sh->pd_idx;
1572 int qd_idx = sh->qd_idx;
1573 struct page *xor_dest;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001574 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001575 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001576 struct async_submit_ctl submit;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001577 int count;
1578 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001579
Harvey Harrisone46b2722008-04-28 02:15:50 -07001580 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001581 (unsigned long long)sh->sector);
1582
Dan Williamsac6b53b2009-07-14 13:40:19 -07001583 count = 0;
1584 xor_dest = sh->dev[pd_idx].page;
1585 xor_srcs[count++] = xor_dest;
Dan Williams91c00922007-01-02 13:52:30 -07001586 for (i = disks; i--; ) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001587 if (i == pd_idx || i == qd_idx)
1588 continue;
1589 xor_srcs[count++] = sh->dev[i].page;
Dan Williams91c00922007-01-02 13:52:30 -07001590 }
1591
Dan Williamsd6f38f32009-07-14 11:50:52 -07001592 init_async_submit(&submit, 0, NULL, NULL, NULL,
1593 to_addr_conv(sh, percpu));
Dan Williams099f53c2009-04-08 14:28:37 -07001594 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07001595 &sh->ops.zero_sum_result, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001596
Dan Williams91c00922007-01-02 13:52:30 -07001597 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001598 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1599 tx = async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001600}
1601
Dan Williamsac6b53b2009-07-14 13:40:19 -07001602static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1603{
1604 struct page **srcs = percpu->scribble;
1605 struct async_submit_ctl submit;
1606 int count;
1607
1608 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1609 (unsigned long long)sh->sector, checkp);
1610
1611 count = set_syndrome_sources(srcs, sh);
1612 if (!checkp)
1613 srcs[count] = NULL;
1614
1615 atomic_inc(&sh->count);
1616 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1617 sh, to_addr_conv(sh, percpu));
1618 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1619 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
1620}
1621
NeilBrown51acbce2013-02-28 09:08:34 +11001622static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
Dan Williams91c00922007-01-02 13:52:30 -07001623{
1624 int overlap_clear = 0, i, disks = sh->disks;
1625 struct dma_async_tx_descriptor *tx = NULL;
NeilBrownd1688a62011-10-11 16:49:52 +11001626 struct r5conf *conf = sh->raid_conf;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001627 int level = conf->level;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001628 struct raid5_percpu *percpu;
1629 unsigned long cpu;
Dan Williams91c00922007-01-02 13:52:30 -07001630
Dan Williamsd6f38f32009-07-14 11:50:52 -07001631 cpu = get_cpu();
1632 percpu = per_cpu_ptr(conf->percpu, cpu);
Dan Williams83de75c2008-06-28 08:31:58 +10001633 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
Dan Williams91c00922007-01-02 13:52:30 -07001634 ops_run_biofill(sh);
1635 overlap_clear++;
1636 }
1637
Dan Williams7b3a8712008-06-28 08:32:09 +10001638 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001639 if (level < 6)
1640 tx = ops_run_compute5(sh, percpu);
1641 else {
1642 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1643 tx = ops_run_compute6_1(sh, percpu);
1644 else
1645 tx = ops_run_compute6_2(sh, percpu);
1646 }
1647 /* terminate the chain if reconstruct is not set to be run */
1648 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
Dan Williams7b3a8712008-06-28 08:32:09 +10001649 async_tx_ack(tx);
1650 }
Dan Williams91c00922007-01-02 13:52:30 -07001651
Dan Williams600aa102008-06-28 08:32:05 +10001652 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
Dan Williamsd6f38f32009-07-14 11:50:52 -07001653 tx = ops_run_prexor(sh, percpu, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001654
Dan Williams600aa102008-06-28 08:32:05 +10001655 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
Dan Williamsd8ee0722008-06-28 08:32:06 +10001656 tx = ops_run_biodrain(sh, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001657 overlap_clear++;
1658 }
1659
Dan Williamsac6b53b2009-07-14 13:40:19 -07001660 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1661 if (level < 6)
1662 ops_run_reconstruct5(sh, percpu, tx);
1663 else
1664 ops_run_reconstruct6(sh, percpu, tx);
1665 }
Dan Williams91c00922007-01-02 13:52:30 -07001666
Dan Williamsac6b53b2009-07-14 13:40:19 -07001667 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1668 if (sh->check_state == check_state_run)
1669 ops_run_check_p(sh, percpu);
1670 else if (sh->check_state == check_state_run_q)
1671 ops_run_check_pq(sh, percpu, 0);
1672 else if (sh->check_state == check_state_run_pq)
1673 ops_run_check_pq(sh, percpu, 1);
1674 else
1675 BUG();
1676 }
Dan Williams91c00922007-01-02 13:52:30 -07001677
Dan Williams91c00922007-01-02 13:52:30 -07001678 if (overlap_clear)
1679 for (i = disks; i--; ) {
1680 struct r5dev *dev = &sh->dev[i];
1681 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1682 wake_up(&sh->raid_conf->wait_for_overlap);
1683 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07001684 put_cpu();
Dan Williams91c00922007-01-02 13:52:30 -07001685}
1686
Shaohua Li566c09c2013-11-14 15:16:17 +11001687static int grow_one_stripe(struct r5conf *conf, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688{
1689 struct stripe_head *sh;
Namhyung Kim6ce32842011-07-18 17:38:50 +10001690 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
NeilBrown3f294f42005-11-08 21:39:25 -08001691 if (!sh)
1692 return 0;
Namhyung Kim6ce32842011-07-18 17:38:50 +10001693
NeilBrown3f294f42005-11-08 21:39:25 -08001694 sh->raid_conf = conf;
NeilBrown3f294f42005-11-08 21:39:25 -08001695
Shaohua Lib17459c2012-07-19 16:01:31 +10001696 spin_lock_init(&sh->stripe_lock);
1697
NeilBrowne4e11e32010-06-16 16:45:16 +10001698 if (grow_buffers(sh)) {
1699 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001700 kmem_cache_free(conf->slab_cache, sh);
1701 return 0;
1702 }
Shaohua Li566c09c2013-11-14 15:16:17 +11001703 sh->hash_lock_index = hash;
NeilBrown3f294f42005-11-08 21:39:25 -08001704 /* we just created an active stripe so... */
1705 atomic_set(&sh->count, 1);
1706 atomic_inc(&conf->active_stripes);
1707 INIT_LIST_HEAD(&sh->lru);
1708 release_stripe(sh);
1709 return 1;
1710}
1711
NeilBrownd1688a62011-10-11 16:49:52 +11001712static int grow_stripes(struct r5conf *conf, int num)
NeilBrown3f294f42005-11-08 21:39:25 -08001713{
Christoph Lametere18b8902006-12-06 20:33:20 -08001714 struct kmem_cache *sc;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001715 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Shaohua Li566c09c2013-11-14 15:16:17 +11001716 int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
NeilBrownf4be6b42010-06-01 19:37:25 +10001718 if (conf->mddev->gendisk)
1719 sprintf(conf->cache_name[0],
1720 "raid%d-%s", conf->level, mdname(conf->mddev));
1721 else
1722 sprintf(conf->cache_name[0],
1723 "raid%d-%p", conf->level, conf->mddev);
1724 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1725
NeilBrownad01c9e2006-03-27 01:18:07 -08001726 conf->active_name = 0;
1727 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001729 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 if (!sc)
1731 return 1;
1732 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001733 conf->pool_size = devs;
Shaohua Li566c09c2013-11-14 15:16:17 +11001734 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
1735 while (num--) {
1736 if (!grow_one_stripe(conf, hash))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 return 1;
Shaohua Li566c09c2013-11-14 15:16:17 +11001738 conf->max_nr_stripes++;
1739 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
1740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 return 0;
1742}
NeilBrown29269552006-03-27 01:18:10 -08001743
Dan Williamsd6f38f32009-07-14 11:50:52 -07001744/**
1745 * scribble_len - return the required size of the scribble region
1746 * @num - total number of disks in the array
1747 *
1748 * The size must be enough to contain:
1749 * 1/ a struct page pointer for each device in the array +2
1750 * 2/ room to convert each entry in (1) to its corresponding dma
1751 * (dma_map_page()) or page (page_address()) address.
1752 *
1753 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1754 * calculate over all devices (not just the data blocks), using zeros in place
1755 * of the P and Q blocks.
1756 */
1757static size_t scribble_len(int num)
1758{
1759 size_t len;
1760
1761 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1762
1763 return len;
1764}
1765
NeilBrownd1688a62011-10-11 16:49:52 +11001766static int resize_stripes(struct r5conf *conf, int newsize)
NeilBrownad01c9e2006-03-27 01:18:07 -08001767{
1768 /* Make all the stripes able to hold 'newsize' devices.
1769 * New slots in each stripe get 'page' set to a new page.
1770 *
1771 * This happens in stages:
1772 * 1/ create a new kmem_cache and allocate the required number of
1773 * stripe_heads.
Masanari Iida83f0d772012-10-30 00:18:08 +09001774 * 2/ gather all the old stripe_heads and transfer the pages across
NeilBrownad01c9e2006-03-27 01:18:07 -08001775 * to the new stripe_heads. This will have the side effect of
1776 * freezing the array as once all stripe_heads have been collected,
1777 * no IO will be possible. Old stripe heads are freed once their
1778 * pages have been transferred over, and the old kmem_cache is
1779 * freed when all stripes are done.
1780 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1781 * we simple return a failre status - no need to clean anything up.
1782 * 4/ allocate new pages for the new slots in the new stripe_heads.
1783 * If this fails, we don't bother trying the shrink the
1784 * stripe_heads down again, we just leave them as they are.
1785 * As each stripe_head is processed the new one is released into
1786 * active service.
1787 *
1788 * Once step2 is started, we cannot afford to wait for a write,
1789 * so we use GFP_NOIO allocations.
1790 */
1791 struct stripe_head *osh, *nsh;
1792 LIST_HEAD(newstripes);
1793 struct disk_info *ndisks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001794 unsigned long cpu;
Dan Williamsb5470dc2008-06-27 21:44:04 -07001795 int err;
Christoph Lametere18b8902006-12-06 20:33:20 -08001796 struct kmem_cache *sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001797 int i;
Shaohua Li566c09c2013-11-14 15:16:17 +11001798 int hash, cnt;
NeilBrownad01c9e2006-03-27 01:18:07 -08001799
1800 if (newsize <= conf->pool_size)
1801 return 0; /* never bother to shrink */
1802
Dan Williamsb5470dc2008-06-27 21:44:04 -07001803 err = md_allow_write(conf->mddev);
1804 if (err)
1805 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08001806
NeilBrownad01c9e2006-03-27 01:18:07 -08001807 /* Step 1 */
1808 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1809 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001810 0, 0, NULL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001811 if (!sc)
1812 return -ENOMEM;
1813
1814 for (i = conf->max_nr_stripes; i; i--) {
Namhyung Kim6ce32842011-07-18 17:38:50 +10001815 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001816 if (!nsh)
1817 break;
1818
NeilBrownad01c9e2006-03-27 01:18:07 -08001819 nsh->raid_conf = conf;
NeilBrowncb13ff62012-09-24 16:27:20 +10001820 spin_lock_init(&nsh->stripe_lock);
NeilBrownad01c9e2006-03-27 01:18:07 -08001821
1822 list_add(&nsh->lru, &newstripes);
1823 }
1824 if (i) {
1825 /* didn't get enough, give up */
1826 while (!list_empty(&newstripes)) {
1827 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1828 list_del(&nsh->lru);
1829 kmem_cache_free(sc, nsh);
1830 }
1831 kmem_cache_destroy(sc);
1832 return -ENOMEM;
1833 }
1834 /* Step 2 - Must use GFP_NOIO now.
1835 * OK, we have enough stripes, start collecting inactive
1836 * stripes and copying them over
1837 */
Shaohua Li566c09c2013-11-14 15:16:17 +11001838 hash = 0;
1839 cnt = 0;
NeilBrownad01c9e2006-03-27 01:18:07 -08001840 list_for_each_entry(nsh, &newstripes, lru) {
Shaohua Li566c09c2013-11-14 15:16:17 +11001841 lock_device_hash_lock(conf, hash);
1842 wait_event_cmd(conf->wait_for_stripe,
1843 !list_empty(conf->inactive_list + hash),
1844 unlock_device_hash_lock(conf, hash),
1845 lock_device_hash_lock(conf, hash));
1846 osh = get_free_stripe(conf, hash);
1847 unlock_device_hash_lock(conf, hash);
NeilBrownad01c9e2006-03-27 01:18:07 -08001848 atomic_set(&nsh->count, 1);
1849 for(i=0; i<conf->pool_size; i++)
1850 nsh->dev[i].page = osh->dev[i].page;
1851 for( ; i<newsize; i++)
1852 nsh->dev[i].page = NULL;
Shaohua Li566c09c2013-11-14 15:16:17 +11001853 nsh->hash_lock_index = hash;
NeilBrownad01c9e2006-03-27 01:18:07 -08001854 kmem_cache_free(conf->slab_cache, osh);
Shaohua Li566c09c2013-11-14 15:16:17 +11001855 cnt++;
1856 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
1857 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
1858 hash++;
1859 cnt = 0;
1860 }
NeilBrownad01c9e2006-03-27 01:18:07 -08001861 }
1862 kmem_cache_destroy(conf->slab_cache);
1863
1864 /* Step 3.
1865 * At this point, we are holding all the stripes so the array
1866 * is completely stalled, so now is a good time to resize
Dan Williamsd6f38f32009-07-14 11:50:52 -07001867 * conf->disks and the scribble region
NeilBrownad01c9e2006-03-27 01:18:07 -08001868 */
1869 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1870 if (ndisks) {
1871 for (i=0; i<conf->raid_disks; i++)
1872 ndisks[i] = conf->disks[i];
1873 kfree(conf->disks);
1874 conf->disks = ndisks;
1875 } else
1876 err = -ENOMEM;
1877
Dan Williamsd6f38f32009-07-14 11:50:52 -07001878 get_online_cpus();
1879 conf->scribble_len = scribble_len(newsize);
1880 for_each_present_cpu(cpu) {
1881 struct raid5_percpu *percpu;
1882 void *scribble;
1883
1884 percpu = per_cpu_ptr(conf->percpu, cpu);
1885 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1886
1887 if (scribble) {
1888 kfree(percpu->scribble);
1889 percpu->scribble = scribble;
1890 } else {
1891 err = -ENOMEM;
1892 break;
1893 }
1894 }
1895 put_online_cpus();
1896
NeilBrownad01c9e2006-03-27 01:18:07 -08001897 /* Step 4, return new stripes to service */
1898 while(!list_empty(&newstripes)) {
1899 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1900 list_del_init(&nsh->lru);
Dan Williamsd6f38f32009-07-14 11:50:52 -07001901
NeilBrownad01c9e2006-03-27 01:18:07 -08001902 for (i=conf->raid_disks; i < newsize; i++)
1903 if (nsh->dev[i].page == NULL) {
1904 struct page *p = alloc_page(GFP_NOIO);
1905 nsh->dev[i].page = p;
1906 if (!p)
1907 err = -ENOMEM;
1908 }
1909 release_stripe(nsh);
1910 }
1911 /* critical section pass, GFP_NOIO no longer needed */
1912
1913 conf->slab_cache = sc;
1914 conf->active_name = 1-conf->active_name;
1915 conf->pool_size = newsize;
1916 return err;
1917}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Shaohua Li566c09c2013-11-14 15:16:17 +11001919static int drop_one_stripe(struct r5conf *conf, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920{
1921 struct stripe_head *sh;
1922
Shaohua Li566c09c2013-11-14 15:16:17 +11001923 spin_lock_irq(conf->hash_locks + hash);
1924 sh = get_free_stripe(conf, hash);
1925 spin_unlock_irq(conf->hash_locks + hash);
NeilBrown3f294f42005-11-08 21:39:25 -08001926 if (!sh)
1927 return 0;
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02001928 BUG_ON(atomic_read(&sh->count));
NeilBrowne4e11e32010-06-16 16:45:16 +10001929 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001930 kmem_cache_free(conf->slab_cache, sh);
1931 atomic_dec(&conf->active_stripes);
1932 return 1;
1933}
1934
NeilBrownd1688a62011-10-11 16:49:52 +11001935static void shrink_stripes(struct r5conf *conf)
NeilBrown3f294f42005-11-08 21:39:25 -08001936{
Shaohua Li566c09c2013-11-14 15:16:17 +11001937 int hash;
1938 for (hash = 0; hash < NR_STRIPE_HASH_LOCKS; hash++)
1939 while (drop_one_stripe(conf, hash))
1940 ;
NeilBrown3f294f42005-11-08 21:39:25 -08001941
NeilBrown29fc7e32006-02-03 03:03:41 -08001942 if (conf->slab_cache)
1943 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 conf->slab_cache = NULL;
1945}
1946
NeilBrown6712ecf2007-09-27 12:47:43 +02001947static void raid5_end_read_request(struct bio * bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948{
NeilBrown99c0fb52009-03-31 14:39:38 +11001949 struct stripe_head *sh = bi->bi_private;
NeilBrownd1688a62011-10-11 16:49:52 +11001950 struct r5conf *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001951 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownd6950432006-07-10 04:44:20 -07001953 char b[BDEVNAME_SIZE];
NeilBrowndd054fc2011-12-23 10:17:53 +11001954 struct md_rdev *rdev = NULL;
NeilBrown05616be2012-05-21 09:27:00 +10001955 sector_t s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
1957 for (i=0 ; i<disks; i++)
1958 if (bi == &sh->dev[i].req)
1959 break;
1960
Dan Williams45b42332007-07-09 11:56:43 -07001961 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1962 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 uptodate);
1964 if (i == disks) {
1965 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001966 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 }
NeilBrown14a75d32011-12-23 10:17:52 +11001968 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
NeilBrowndd054fc2011-12-23 10:17:53 +11001969 /* If replacement finished while this request was outstanding,
1970 * 'replacement' might be NULL already.
1971 * In that case it moved down to 'rdev'.
1972 * rdev is not removed until all requests are finished.
1973 */
NeilBrown14a75d32011-12-23 10:17:52 +11001974 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11001975 if (!rdev)
NeilBrown14a75d32011-12-23 10:17:52 +11001976 rdev = conf->disks[i].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
NeilBrown05616be2012-05-21 09:27:00 +10001978 if (use_new_offset(conf, sh))
1979 s = sh->sector + rdev->new_data_offset;
1980 else
1981 s = sh->sector + rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -08001984 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11001985 /* Note that this cannot happen on a
1986 * replacement device. We just fail those on
1987 * any error
1988 */
Christian Dietrich8bda4702011-07-27 11:00:36 +10001989 printk_ratelimited(
1990 KERN_INFO
1991 "md/raid:%s: read error corrected"
1992 " (%lu sectors at %llu on %s)\n",
1993 mdname(conf->mddev), STRIPE_SECTORS,
NeilBrown05616be2012-05-21 09:27:00 +10001994 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10001995 bdevname(rdev->bdev, b));
Namhyung Kimddd51152011-07-27 11:00:36 +10001996 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
NeilBrown4e5314b2005-11-08 21:39:22 -08001997 clear_bit(R5_ReadError, &sh->dev[i].flags);
1998 clear_bit(R5_ReWrite, &sh->dev[i].flags);
majianpeng3f9e7c12012-07-31 10:04:21 +10001999 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2000 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2001
NeilBrown14a75d32011-12-23 10:17:52 +11002002 if (atomic_read(&rdev->read_errors))
2003 atomic_set(&rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 } else {
NeilBrown14a75d32011-12-23 10:17:52 +11002005 const char *bdn = bdevname(rdev->bdev, b);
NeilBrownba22dcb2005-11-08 21:39:31 -08002006 int retry = 0;
majianpeng2e8ac3032012-07-03 15:57:02 +10002007 int set_bad = 0;
NeilBrownd6950432006-07-10 04:44:20 -07002008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07002010 atomic_inc(&rdev->read_errors);
NeilBrown14a75d32011-12-23 10:17:52 +11002011 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
2012 printk_ratelimited(
2013 KERN_WARNING
2014 "md/raid:%s: read error on replacement device "
2015 "(sector %llu on %s).\n",
2016 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002017 (unsigned long long)s,
NeilBrown14a75d32011-12-23 10:17:52 +11002018 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002019 else if (conf->mddev->degraded >= conf->max_degraded) {
2020 set_bad = 1;
Christian Dietrich8bda4702011-07-27 11:00:36 +10002021 printk_ratelimited(
2022 KERN_WARNING
2023 "md/raid:%s: read error not correctable "
2024 "(sector %llu on %s).\n",
2025 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002026 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002027 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002028 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
NeilBrown4e5314b2005-11-08 21:39:22 -08002029 /* Oh, no!!! */
majianpeng2e8ac3032012-07-03 15:57:02 +10002030 set_bad = 1;
Christian Dietrich8bda4702011-07-27 11:00:36 +10002031 printk_ratelimited(
2032 KERN_WARNING
2033 "md/raid:%s: read error NOT corrected!! "
2034 "(sector %llu on %s).\n",
2035 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002036 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002037 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002038 } else if (atomic_read(&rdev->read_errors)
NeilBrownba22dcb2005-11-08 21:39:31 -08002039 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -08002040 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10002041 "md/raid:%s: Too many read errors, failing device %s.\n",
NeilBrownd6950432006-07-10 04:44:20 -07002042 mdname(conf->mddev), bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08002043 else
2044 retry = 1;
Bian Yuedfa1f62013-11-14 15:16:17 +11002045 if (set_bad && test_bit(In_sync, &rdev->flags)
2046 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2047 retry = 1;
NeilBrownba22dcb2005-11-08 21:39:31 -08002048 if (retry)
majianpeng3f9e7c12012-07-31 10:04:21 +10002049 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2050 set_bit(R5_ReadError, &sh->dev[i].flags);
2051 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2052 } else
2053 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
NeilBrownba22dcb2005-11-08 21:39:31 -08002054 else {
NeilBrown4e5314b2005-11-08 21:39:22 -08002055 clear_bit(R5_ReadError, &sh->dev[i].flags);
2056 clear_bit(R5_ReWrite, &sh->dev[i].flags);
majianpeng2e8ac3032012-07-03 15:57:02 +10002057 if (!(set_bad
2058 && test_bit(In_sync, &rdev->flags)
2059 && rdev_set_badblocks(
2060 rdev, sh->sector, STRIPE_SECTORS, 0)))
2061 md_error(conf->mddev, rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -08002062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 }
NeilBrown14a75d32011-12-23 10:17:52 +11002064 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2066 set_bit(STRIPE_HANDLE, &sh->state);
2067 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068}
2069
NeilBrownd710e132008-10-13 11:55:12 +11002070static void raid5_end_write_request(struct bio *bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071{
NeilBrown99c0fb52009-03-31 14:39:38 +11002072 struct stripe_head *sh = bi->bi_private;
NeilBrownd1688a62011-10-11 16:49:52 +11002073 struct r5conf *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08002074 int disks = sh->disks, i;
NeilBrown977df362011-12-23 10:17:53 +11002075 struct md_rdev *uninitialized_var(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownb84db562011-07-28 11:39:23 +10002077 sector_t first_bad;
2078 int bad_sectors;
NeilBrown977df362011-12-23 10:17:53 +11002079 int replacement = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
NeilBrown977df362011-12-23 10:17:53 +11002081 for (i = 0 ; i < disks; i++) {
2082 if (bi == &sh->dev[i].req) {
2083 rdev = conf->disks[i].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 break;
NeilBrown977df362011-12-23 10:17:53 +11002085 }
2086 if (bi == &sh->dev[i].rreq) {
2087 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11002088 if (rdev)
2089 replacement = 1;
2090 else
2091 /* rdev was removed and 'replacement'
2092 * replaced it. rdev is not removed
2093 * until all requests are finished.
2094 */
2095 rdev = conf->disks[i].rdev;
NeilBrown977df362011-12-23 10:17:53 +11002096 break;
2097 }
2098 }
Dan Williams45b42332007-07-09 11:56:43 -07002099 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
2101 uptodate);
2102 if (i == disks) {
2103 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02002104 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 }
2106
NeilBrown977df362011-12-23 10:17:53 +11002107 if (replacement) {
2108 if (!uptodate)
2109 md_error(conf->mddev, rdev);
2110 else if (is_badblock(rdev, sh->sector,
2111 STRIPE_SECTORS,
2112 &first_bad, &bad_sectors))
2113 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2114 } else {
2115 if (!uptodate) {
2116 set_bit(WriteErrorSeen, &rdev->flags);
2117 set_bit(R5_WriteError, &sh->dev[i].flags);
NeilBrown3a6de292011-12-23 10:17:54 +11002118 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2119 set_bit(MD_RECOVERY_NEEDED,
2120 &rdev->mddev->recovery);
NeilBrown977df362011-12-23 10:17:53 +11002121 } else if (is_badblock(rdev, sh->sector,
2122 STRIPE_SECTORS,
NeilBrownc0b32972013-04-24 11:42:42 +10002123 &first_bad, &bad_sectors)) {
NeilBrown977df362011-12-23 10:17:53 +11002124 set_bit(R5_MadeGood, &sh->dev[i].flags);
NeilBrownc0b32972013-04-24 11:42:42 +10002125 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2126 /* That was a successful write so make
2127 * sure it looks like we already did
2128 * a re-write.
2129 */
2130 set_bit(R5_ReWrite, &sh->dev[i].flags);
2131 }
NeilBrown977df362011-12-23 10:17:53 +11002132 }
2133 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
NeilBrown977df362011-12-23 10:17:53 +11002135 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2136 clear_bit(R5_LOCKED, &sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrownc04be0a2006-10-03 01:15:53 -07002138 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139}
2140
NeilBrown784052e2009-03-31 15:19:07 +11002141static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
NeilBrown784052e2009-03-31 15:19:07 +11002143static void raid5_build_block(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144{
2145 struct r5dev *dev = &sh->dev[i];
2146
2147 bio_init(&dev->req);
2148 dev->req.bi_io_vec = &dev->vec;
2149 dev->req.bi_vcnt++;
2150 dev->req.bi_max_vecs++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 dev->req.bi_private = sh;
NeilBrown995c4272011-12-23 10:17:52 +11002152 dev->vec.bv_page = dev->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
NeilBrown977df362011-12-23 10:17:53 +11002154 bio_init(&dev->rreq);
2155 dev->rreq.bi_io_vec = &dev->rvec;
2156 dev->rreq.bi_vcnt++;
2157 dev->rreq.bi_max_vecs++;
2158 dev->rreq.bi_private = sh;
2159 dev->rvec.bv_page = dev->page;
2160
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +11002162 dev->sector = compute_blocknr(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163}
2164
NeilBrownfd01b882011-10-11 16:47:53 +11002165static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166{
2167 char b[BDEVNAME_SIZE];
NeilBrownd1688a62011-10-11 16:49:52 +11002168 struct r5conf *conf = mddev->private;
NeilBrown908f4fb2011-12-23 10:17:50 +11002169 unsigned long flags;
NeilBrown0c55e022010-05-03 14:09:02 +10002170 pr_debug("raid456: error called\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
NeilBrown908f4fb2011-12-23 10:17:50 +11002172 spin_lock_irqsave(&conf->device_lock, flags);
2173 clear_bit(In_sync, &rdev->flags);
2174 mddev->degraded = calc_degraded(conf);
2175 spin_unlock_irqrestore(&conf->device_lock, flags);
2176 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2177
NeilBrownde393cd2011-07-28 11:31:48 +10002178 set_bit(Blocked, &rdev->flags);
NeilBrown6f8d0c72011-05-11 14:38:44 +10002179 set_bit(Faulty, &rdev->flags);
2180 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2181 printk(KERN_ALERT
2182 "md/raid:%s: Disk failure on %s, disabling device.\n"
2183 "md/raid:%s: Operation continuing on %d devices.\n",
2184 mdname(mddev),
2185 bdevname(rdev->bdev, b),
2186 mdname(mddev),
2187 conf->raid_disks - mddev->degraded);
NeilBrown16a53ec2006-06-26 00:27:38 -07002188}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
2190/*
2191 * Input: a 'big' sector number,
2192 * Output: index of the data and parity disk, and the sector # in them.
2193 */
NeilBrownd1688a62011-10-11 16:49:52 +11002194static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11002195 int previous, int *dd_idx,
2196 struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197{
NeilBrown6e3b96e2010-04-23 07:08:28 +10002198 sector_t stripe, stripe2;
NeilBrown35f2a592010-04-20 14:13:34 +10002199 sector_t chunk_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 unsigned int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11002201 int pd_idx, qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11002202 int ddf_layout = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 sector_t new_sector;
NeilBrowne183eae2009-03-31 15:20:22 +11002204 int algorithm = previous ? conf->prev_algo
2205 : conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10002206 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2207 : conf->chunk_sectors;
NeilBrown112bf892009-03-31 14:39:38 +11002208 int raid_disks = previous ? conf->previous_raid_disks
2209 : conf->raid_disks;
2210 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211
2212 /* First compute the information on this sector */
2213
2214 /*
2215 * Compute the chunk number and the sector offset inside the chunk
2216 */
2217 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2218 chunk_number = r_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219
2220 /*
2221 * Compute the stripe number
2222 */
NeilBrown35f2a592010-04-20 14:13:34 +10002223 stripe = chunk_number;
2224 *dd_idx = sector_div(stripe, data_disks);
NeilBrown6e3b96e2010-04-23 07:08:28 +10002225 stripe2 = stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 /*
2227 * Select the parity disk based on the user selected algorithm.
2228 */
NeilBrown84789552011-07-27 11:00:36 +10002229 pd_idx = qd_idx = -1;
NeilBrown16a53ec2006-06-26 00:27:38 -07002230 switch(conf->level) {
2231 case 4:
NeilBrown911d4ee2009-03-31 14:39:38 +11002232 pd_idx = data_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002233 break;
2234 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11002235 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002237 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002238 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 (*dd_idx)++;
2240 break;
2241 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002242 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002243 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 (*dd_idx)++;
2245 break;
2246 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002247 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002248 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 break;
2250 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002251 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002252 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002254 case ALGORITHM_PARITY_0:
2255 pd_idx = 0;
2256 (*dd_idx)++;
2257 break;
2258 case ALGORITHM_PARITY_N:
2259 pd_idx = data_disks;
2260 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002262 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002263 }
2264 break;
2265 case 6:
2266
NeilBrowne183eae2009-03-31 15:20:22 +11002267 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07002268 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002269 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002270 qd_idx = pd_idx + 1;
2271 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11002272 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11002273 qd_idx = 0;
2274 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002275 (*dd_idx) += 2; /* D D P Q D */
2276 break;
2277 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002278 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002279 qd_idx = pd_idx + 1;
2280 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11002281 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11002282 qd_idx = 0;
2283 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002284 (*dd_idx) += 2; /* D D P Q D */
2285 break;
2286 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002287 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002288 qd_idx = (pd_idx + 1) % raid_disks;
2289 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002290 break;
2291 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002292 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002293 qd_idx = (pd_idx + 1) % raid_disks;
2294 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002295 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002296
2297 case ALGORITHM_PARITY_0:
2298 pd_idx = 0;
2299 qd_idx = 1;
2300 (*dd_idx) += 2;
2301 break;
2302 case ALGORITHM_PARITY_N:
2303 pd_idx = data_disks;
2304 qd_idx = data_disks + 1;
2305 break;
2306
2307 case ALGORITHM_ROTATING_ZERO_RESTART:
2308 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2309 * of blocks for computing Q is different.
2310 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002311 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002312 qd_idx = pd_idx + 1;
2313 if (pd_idx == raid_disks-1) {
2314 (*dd_idx)++; /* Q D D D P */
2315 qd_idx = 0;
2316 } else if (*dd_idx >= pd_idx)
2317 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11002318 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002319 break;
2320
2321 case ALGORITHM_ROTATING_N_RESTART:
2322 /* Same a left_asymmetric, by first stripe is
2323 * D D D P Q rather than
2324 * Q D D D P
2325 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002326 stripe2 += 1;
2327 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002328 qd_idx = pd_idx + 1;
2329 if (pd_idx == raid_disks-1) {
2330 (*dd_idx)++; /* Q D D D P */
2331 qd_idx = 0;
2332 } else if (*dd_idx >= pd_idx)
2333 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11002334 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002335 break;
2336
2337 case ALGORITHM_ROTATING_N_CONTINUE:
2338 /* Same as left_symmetric but Q is before P */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002339 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002340 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2341 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
NeilBrown67cc2b82009-03-31 14:39:38 +11002342 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002343 break;
2344
2345 case ALGORITHM_LEFT_ASYMMETRIC_6:
2346 /* RAID5 left_asymmetric, with Q on last device */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002347 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002348 if (*dd_idx >= pd_idx)
2349 (*dd_idx)++;
2350 qd_idx = raid_disks - 1;
2351 break;
2352
2353 case ALGORITHM_RIGHT_ASYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002354 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002355 if (*dd_idx >= pd_idx)
2356 (*dd_idx)++;
2357 qd_idx = raid_disks - 1;
2358 break;
2359
2360 case ALGORITHM_LEFT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002361 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002362 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2363 qd_idx = raid_disks - 1;
2364 break;
2365
2366 case ALGORITHM_RIGHT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002367 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002368 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2369 qd_idx = raid_disks - 1;
2370 break;
2371
2372 case ALGORITHM_PARITY_0_6:
2373 pd_idx = 0;
2374 (*dd_idx)++;
2375 qd_idx = raid_disks - 1;
2376 break;
2377
NeilBrown16a53ec2006-06-26 00:27:38 -07002378 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002379 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002380 }
2381 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 }
2383
NeilBrown911d4ee2009-03-31 14:39:38 +11002384 if (sh) {
2385 sh->pd_idx = pd_idx;
2386 sh->qd_idx = qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11002387 sh->ddf_layout = ddf_layout;
NeilBrown911d4ee2009-03-31 14:39:38 +11002388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 /*
2390 * Finally, compute the new sector number
2391 */
2392 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2393 return new_sector;
2394}
2395
2396
NeilBrown784052e2009-03-31 15:19:07 +11002397static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398{
NeilBrownd1688a62011-10-11 16:49:52 +11002399 struct r5conf *conf = sh->raid_conf;
NeilBrownb875e532006-12-10 02:20:49 -08002400 int raid_disks = sh->disks;
2401 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 sector_t new_sector = sh->sector, check;
Andre Noll09c9e5f2009-06-18 08:45:55 +10002403 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2404 : conf->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11002405 int algorithm = previous ? conf->prev_algo
2406 : conf->algorithm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 sector_t stripe;
2408 int chunk_offset;
NeilBrown35f2a592010-04-20 14:13:34 +10002409 sector_t chunk_number;
2410 int dummy1, dd_idx = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 sector_t r_sector;
NeilBrown911d4ee2009-03-31 14:39:38 +11002412 struct stripe_head sh2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413
NeilBrown16a53ec2006-06-26 00:27:38 -07002414
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2416 stripe = new_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417
NeilBrown16a53ec2006-06-26 00:27:38 -07002418 if (i == sh->pd_idx)
2419 return 0;
2420 switch(conf->level) {
2421 case 4: break;
2422 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11002423 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 case ALGORITHM_LEFT_ASYMMETRIC:
2425 case ALGORITHM_RIGHT_ASYMMETRIC:
2426 if (i > sh->pd_idx)
2427 i--;
2428 break;
2429 case ALGORITHM_LEFT_SYMMETRIC:
2430 case ALGORITHM_RIGHT_SYMMETRIC:
2431 if (i < sh->pd_idx)
2432 i += raid_disks;
2433 i -= (sh->pd_idx + 1);
2434 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002435 case ALGORITHM_PARITY_0:
2436 i -= 1;
2437 break;
2438 case ALGORITHM_PARITY_N:
2439 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002441 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002442 }
2443 break;
2444 case 6:
NeilBrownd0dabf72009-03-31 14:39:38 +11002445 if (i == sh->qd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002446 return 0; /* It is the Q disk */
NeilBrowne183eae2009-03-31 15:20:22 +11002447 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07002448 case ALGORITHM_LEFT_ASYMMETRIC:
2449 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown99c0fb52009-03-31 14:39:38 +11002450 case ALGORITHM_ROTATING_ZERO_RESTART:
2451 case ALGORITHM_ROTATING_N_RESTART:
2452 if (sh->pd_idx == raid_disks-1)
2453 i--; /* Q D D D P */
NeilBrown16a53ec2006-06-26 00:27:38 -07002454 else if (i > sh->pd_idx)
2455 i -= 2; /* D D P Q D */
2456 break;
2457 case ALGORITHM_LEFT_SYMMETRIC:
2458 case ALGORITHM_RIGHT_SYMMETRIC:
2459 if (sh->pd_idx == raid_disks-1)
2460 i--; /* Q D D D P */
2461 else {
2462 /* D D P Q D */
2463 if (i < sh->pd_idx)
2464 i += raid_disks;
2465 i -= (sh->pd_idx + 2);
2466 }
2467 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002468 case ALGORITHM_PARITY_0:
2469 i -= 2;
2470 break;
2471 case ALGORITHM_PARITY_N:
2472 break;
2473 case ALGORITHM_ROTATING_N_CONTINUE:
NeilBrowne4424fe2009-10-16 16:27:34 +11002474 /* Like left_symmetric, but P is before Q */
NeilBrown99c0fb52009-03-31 14:39:38 +11002475 if (sh->pd_idx == 0)
2476 i--; /* P D D D Q */
NeilBrowne4424fe2009-10-16 16:27:34 +11002477 else {
2478 /* D D Q P D */
2479 if (i < sh->pd_idx)
2480 i += raid_disks;
2481 i -= (sh->pd_idx + 1);
2482 }
NeilBrown99c0fb52009-03-31 14:39:38 +11002483 break;
2484 case ALGORITHM_LEFT_ASYMMETRIC_6:
2485 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2486 if (i > sh->pd_idx)
2487 i--;
2488 break;
2489 case ALGORITHM_LEFT_SYMMETRIC_6:
2490 case ALGORITHM_RIGHT_SYMMETRIC_6:
2491 if (i < sh->pd_idx)
2492 i += data_disks + 1;
2493 i -= (sh->pd_idx + 1);
2494 break;
2495 case ALGORITHM_PARITY_0_6:
2496 i -= 1;
2497 break;
NeilBrown16a53ec2006-06-26 00:27:38 -07002498 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002499 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002500 }
2501 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 }
2503
2504 chunk_number = stripe * data_disks + i;
NeilBrown35f2a592010-04-20 14:13:34 +10002505 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506
NeilBrown112bf892009-03-31 14:39:38 +11002507 check = raid5_compute_sector(conf, r_sector,
NeilBrown784052e2009-03-31 15:19:07 +11002508 previous, &dummy1, &sh2);
NeilBrown911d4ee2009-03-31 14:39:38 +11002509 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2510 || sh2.qd_idx != sh->qd_idx) {
NeilBrown0c55e022010-05-03 14:09:02 +10002511 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2512 mdname(conf->mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 return 0;
2514 }
2515 return r_sector;
2516}
2517
2518
Dan Williams600aa102008-06-28 08:32:05 +10002519static void
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002520schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
Dan Williams600aa102008-06-28 08:32:05 +10002521 int rcw, int expand)
Dan Williamse33129d2007-01-02 13:52:30 -07002522{
2523 int i, pd_idx = sh->pd_idx, disks = sh->disks;
NeilBrownd1688a62011-10-11 16:49:52 +11002524 struct r5conf *conf = sh->raid_conf;
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002525 int level = conf->level;
NeilBrown16a53ec2006-06-26 00:27:38 -07002526
Dan Williamse33129d2007-01-02 13:52:30 -07002527 if (rcw) {
Dan Williamse33129d2007-01-02 13:52:30 -07002528
2529 for (i = disks; i--; ) {
2530 struct r5dev *dev = &sh->dev[i];
2531
2532 if (dev->towrite) {
2533 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsd8ee0722008-06-28 08:32:06 +10002534 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002535 if (!expand)
2536 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002537 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002538 }
2539 }
NeilBrownce7d3632013-03-04 12:37:14 +11002540 /* if we are not expanding this is a proper write request, and
2541 * there will be bios with new data to be drained into the
2542 * stripe cache
2543 */
2544 if (!expand) {
2545 if (!s->locked)
2546 /* False alarm, nothing to do */
2547 return;
2548 sh->reconstruct_state = reconstruct_state_drain_run;
2549 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2550 } else
2551 sh->reconstruct_state = reconstruct_state_run;
2552
2553 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2554
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002555 if (s->locked + conf->max_degraded == disks)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002556 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002557 atomic_inc(&conf->pending_full_writes);
Dan Williamse33129d2007-01-02 13:52:30 -07002558 } else {
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002559 BUG_ON(level == 6);
Dan Williamse33129d2007-01-02 13:52:30 -07002560 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2561 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2562
Dan Williamse33129d2007-01-02 13:52:30 -07002563 for (i = disks; i--; ) {
2564 struct r5dev *dev = &sh->dev[i];
2565 if (i == pd_idx)
2566 continue;
2567
Dan Williamse33129d2007-01-02 13:52:30 -07002568 if (dev->towrite &&
2569 (test_bit(R5_UPTODATE, &dev->flags) ||
Dan Williamsd8ee0722008-06-28 08:32:06 +10002570 test_bit(R5_Wantcompute, &dev->flags))) {
2571 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002572 set_bit(R5_LOCKED, &dev->flags);
2573 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002574 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002575 }
2576 }
NeilBrownce7d3632013-03-04 12:37:14 +11002577 if (!s->locked)
2578 /* False alarm - nothing to do */
2579 return;
2580 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2581 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2582 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2583 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002584 }
2585
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002586 /* keep the parity disk(s) locked while asynchronous operations
Dan Williamse33129d2007-01-02 13:52:30 -07002587 * are in flight
2588 */
2589 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2590 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
Dan Williams600aa102008-06-28 08:32:05 +10002591 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002592
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002593 if (level == 6) {
2594 int qd_idx = sh->qd_idx;
2595 struct r5dev *dev = &sh->dev[qd_idx];
2596
2597 set_bit(R5_LOCKED, &dev->flags);
2598 clear_bit(R5_UPTODATE, &dev->flags);
2599 s->locked++;
2600 }
2601
Dan Williams600aa102008-06-28 08:32:05 +10002602 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07002603 __func__, (unsigned long long)sh->sector,
Dan Williams600aa102008-06-28 08:32:05 +10002604 s->locked, s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002605}
NeilBrown16a53ec2006-06-26 00:27:38 -07002606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607/*
2608 * Each stripe/dev can have one or more bion attached.
NeilBrown16a53ec2006-06-26 00:27:38 -07002609 * toread/towrite point to the first in a chain.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 * The bi_next chain must be in order.
2611 */
2612static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2613{
2614 struct bio **bip;
NeilBrownd1688a62011-10-11 16:49:52 +11002615 struct r5conf *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -07002616 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617
NeilBrowncbe47ec2011-07-26 11:20:35 +10002618 pr_debug("adding bi b#%llu to stripe s#%llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 (unsigned long long)bi->bi_sector,
2620 (unsigned long long)sh->sector);
2621
Shaohua Lib17459c2012-07-19 16:01:31 +10002622 /*
2623 * If several bio share a stripe. The bio bi_phys_segments acts as a
2624 * reference count to avoid race. The reference count should already be
2625 * increased before this function is called (for example, in
2626 * make_request()), so other bio sharing this stripe will not free the
2627 * stripe. If a stripe is owned by one stripe, the stripe lock will
2628 * protect it.
2629 */
2630 spin_lock_irq(&sh->stripe_lock);
NeilBrown72626682005-09-09 16:23:54 -07002631 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 bip = &sh->dev[dd_idx].towrite;
Shaohua Li7eaf7e82012-07-19 16:01:31 +10002633 if (*bip == NULL)
NeilBrown72626682005-09-09 16:23:54 -07002634 firstwrite = 1;
2635 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 bip = &sh->dev[dd_idx].toread;
2637 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
Kent Overstreetf73a1c72012-09-25 15:05:12 -07002638 if (bio_end_sector(*bip) > bi->bi_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 goto overlap;
2640 bip = & (*bip)->bi_next;
2641 }
Kent Overstreetf73a1c72012-09-25 15:05:12 -07002642 if (*bip && (*bip)->bi_sector < bio_end_sector(bi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 goto overlap;
2644
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002645 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 if (*bip)
2647 bi->bi_next = *bip;
2648 *bip = bi;
Shaohua Lie7836bd62012-07-19 16:01:31 +10002649 raid5_inc_bi_active_stripes(bi);
NeilBrown72626682005-09-09 16:23:54 -07002650
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 if (forwrite) {
2652 /* check if page is covered */
2653 sector_t sector = sh->dev[dd_idx].sector;
2654 for (bi=sh->dev[dd_idx].towrite;
2655 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2656 bi && bi->bi_sector <= sector;
2657 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
Kent Overstreetf73a1c72012-09-25 15:05:12 -07002658 if (bio_end_sector(bi) >= sector)
2659 sector = bio_end_sector(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 }
2661 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2662 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2663 }
NeilBrowncbe47ec2011-07-26 11:20:35 +10002664
2665 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2666 (unsigned long long)(*bip)->bi_sector,
2667 (unsigned long long)sh->sector, dd_idx);
NeilBrownb97390a2012-10-11 13:50:12 +11002668 spin_unlock_irq(&sh->stripe_lock);
NeilBrowncbe47ec2011-07-26 11:20:35 +10002669
2670 if (conf->mddev->bitmap && firstwrite) {
2671 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2672 STRIPE_SECTORS, 0);
2673 sh->bm_seq = conf->seq_flush+1;
2674 set_bit(STRIPE_BIT_DELAY, &sh->state);
2675 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 return 1;
2677
2678 overlap:
2679 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
Shaohua Lib17459c2012-07-19 16:01:31 +10002680 spin_unlock_irq(&sh->stripe_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 return 0;
2682}
2683
NeilBrownd1688a62011-10-11 16:49:52 +11002684static void end_reshape(struct r5conf *conf);
NeilBrown29269552006-03-27 01:18:10 -08002685
NeilBrownd1688a62011-10-11 16:49:52 +11002686static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002687 struct stripe_head *sh)
NeilBrownccfcc3c2006-03-27 01:18:09 -08002688{
NeilBrown784052e2009-03-31 15:19:07 +11002689 int sectors_per_chunk =
Andre Noll09c9e5f2009-06-18 08:45:55 +10002690 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
NeilBrown911d4ee2009-03-31 14:39:38 +11002691 int dd_idx;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002692 int chunk_offset = sector_div(stripe, sectors_per_chunk);
NeilBrown112bf892009-03-31 14:39:38 +11002693 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002694
NeilBrown112bf892009-03-31 14:39:38 +11002695 raid5_compute_sector(conf,
2696 stripe * (disks - conf->max_degraded)
NeilBrownb875e532006-12-10 02:20:49 -08002697 *sectors_per_chunk + chunk_offset,
NeilBrown112bf892009-03-31 14:39:38 +11002698 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002699 &dd_idx, sh);
NeilBrownccfcc3c2006-03-27 01:18:09 -08002700}
2701
Dan Williamsa4456852007-07-09 11:56:43 -07002702static void
NeilBrownd1688a62011-10-11 16:49:52 +11002703handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002704 struct stripe_head_state *s, int disks,
2705 struct bio **return_bi)
2706{
2707 int i;
2708 for (i = disks; i--; ) {
2709 struct bio *bi;
2710 int bitmap_end = 0;
2711
2712 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown3cb03002011-10-11 16:45:26 +11002713 struct md_rdev *rdev;
Dan Williamsa4456852007-07-09 11:56:43 -07002714 rcu_read_lock();
2715 rdev = rcu_dereference(conf->disks[i].rdev);
2716 if (rdev && test_bit(In_sync, &rdev->flags))
NeilBrown7f0da592011-07-28 11:39:22 +10002717 atomic_inc(&rdev->nr_pending);
2718 else
2719 rdev = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07002720 rcu_read_unlock();
NeilBrown7f0da592011-07-28 11:39:22 +10002721 if (rdev) {
2722 if (!rdev_set_badblocks(
2723 rdev,
2724 sh->sector,
2725 STRIPE_SECTORS, 0))
2726 md_error(conf->mddev, rdev);
2727 rdev_dec_pending(rdev, conf->mddev);
2728 }
Dan Williamsa4456852007-07-09 11:56:43 -07002729 }
Shaohua Lib17459c2012-07-19 16:01:31 +10002730 spin_lock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07002731 /* fail all writes first */
2732 bi = sh->dev[i].towrite;
2733 sh->dev[i].towrite = NULL;
Shaohua Lib17459c2012-07-19 16:01:31 +10002734 spin_unlock_irq(&sh->stripe_lock);
NeilBrown1ed850f2012-10-11 13:50:13 +11002735 if (bi)
Dan Williamsa4456852007-07-09 11:56:43 -07002736 bitmap_end = 1;
Dan Williamsa4456852007-07-09 11:56:43 -07002737
2738 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2739 wake_up(&conf->wait_for_overlap);
2740
2741 while (bi && bi->bi_sector <
2742 sh->dev[i].sector + STRIPE_SECTORS) {
2743 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2744 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Shaohua Lie7836bd62012-07-19 16:01:31 +10002745 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002746 md_write_end(conf->mddev);
2747 bi->bi_next = *return_bi;
2748 *return_bi = bi;
2749 }
2750 bi = nextbi;
2751 }
Shaohua Li7eaf7e82012-07-19 16:01:31 +10002752 if (bitmap_end)
2753 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2754 STRIPE_SECTORS, 0, 0);
2755 bitmap_end = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07002756 /* and fail all 'written' */
2757 bi = sh->dev[i].written;
2758 sh->dev[i].written = NULL;
2759 if (bi) bitmap_end = 1;
2760 while (bi && bi->bi_sector <
2761 sh->dev[i].sector + STRIPE_SECTORS) {
2762 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2763 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Shaohua Lie7836bd62012-07-19 16:01:31 +10002764 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002765 md_write_end(conf->mddev);
2766 bi->bi_next = *return_bi;
2767 *return_bi = bi;
2768 }
2769 bi = bi2;
2770 }
2771
Dan Williamsb5e98d62007-01-02 13:52:31 -07002772 /* fail any reads if this device is non-operational and
2773 * the data has not reached the cache yet.
2774 */
2775 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2776 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2777 test_bit(R5_ReadError, &sh->dev[i].flags))) {
NeilBrown143c4d02012-10-11 13:50:12 +11002778 spin_lock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07002779 bi = sh->dev[i].toread;
2780 sh->dev[i].toread = NULL;
NeilBrown143c4d02012-10-11 13:50:12 +11002781 spin_unlock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07002782 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2783 wake_up(&conf->wait_for_overlap);
Dan Williamsa4456852007-07-09 11:56:43 -07002784 while (bi && bi->bi_sector <
2785 sh->dev[i].sector + STRIPE_SECTORS) {
2786 struct bio *nextbi =
2787 r5_next_bio(bi, sh->dev[i].sector);
2788 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Shaohua Lie7836bd62012-07-19 16:01:31 +10002789 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002790 bi->bi_next = *return_bi;
2791 *return_bi = bi;
2792 }
2793 bi = nextbi;
2794 }
2795 }
Dan Williamsa4456852007-07-09 11:56:43 -07002796 if (bitmap_end)
2797 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2798 STRIPE_SECTORS, 0, 0);
NeilBrown8cfa7b02011-07-27 11:00:36 +10002799 /* If we were in the middle of a write the parity block might
2800 * still be locked - so just clear all R5_LOCKED flags
2801 */
2802 clear_bit(R5_LOCKED, &sh->dev[i].flags);
Dan Williamsa4456852007-07-09 11:56:43 -07002803 }
2804
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002805 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2806 if (atomic_dec_and_test(&conf->pending_full_writes))
2807 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002808}
2809
NeilBrown7f0da592011-07-28 11:39:22 +10002810static void
NeilBrownd1688a62011-10-11 16:49:52 +11002811handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
NeilBrown7f0da592011-07-28 11:39:22 +10002812 struct stripe_head_state *s)
2813{
2814 int abort = 0;
2815 int i;
2816
NeilBrown7f0da592011-07-28 11:39:22 +10002817 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11002818 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
2819 wake_up(&conf->wait_for_overlap);
NeilBrown7f0da592011-07-28 11:39:22 +10002820 s->syncing = 0;
NeilBrown9a3e1102011-12-23 10:17:53 +11002821 s->replacing = 0;
NeilBrown7f0da592011-07-28 11:39:22 +10002822 /* There is nothing more to do for sync/check/repair.
NeilBrown18b98372012-04-01 23:48:38 +10002823 * Don't even need to abort as that is handled elsewhere
2824 * if needed, and not always wanted e.g. if there is a known
2825 * bad block here.
NeilBrown9a3e1102011-12-23 10:17:53 +11002826 * For recover/replace we need to record a bad block on all
NeilBrown7f0da592011-07-28 11:39:22 +10002827 * non-sync devices, or abort the recovery
2828 */
NeilBrown18b98372012-04-01 23:48:38 +10002829 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
2830 /* During recovery devices cannot be removed, so
2831 * locking and refcounting of rdevs is not needed
2832 */
2833 for (i = 0; i < conf->raid_disks; i++) {
2834 struct md_rdev *rdev = conf->disks[i].rdev;
2835 if (rdev
2836 && !test_bit(Faulty, &rdev->flags)
2837 && !test_bit(In_sync, &rdev->flags)
2838 && !rdev_set_badblocks(rdev, sh->sector,
2839 STRIPE_SECTORS, 0))
2840 abort = 1;
2841 rdev = conf->disks[i].replacement;
2842 if (rdev
2843 && !test_bit(Faulty, &rdev->flags)
2844 && !test_bit(In_sync, &rdev->flags)
2845 && !rdev_set_badblocks(rdev, sh->sector,
2846 STRIPE_SECTORS, 0))
2847 abort = 1;
2848 }
2849 if (abort)
2850 conf->recovery_disabled =
2851 conf->mddev->recovery_disabled;
NeilBrown7f0da592011-07-28 11:39:22 +10002852 }
NeilBrown18b98372012-04-01 23:48:38 +10002853 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
NeilBrown7f0da592011-07-28 11:39:22 +10002854}
2855
NeilBrown9a3e1102011-12-23 10:17:53 +11002856static int want_replace(struct stripe_head *sh, int disk_idx)
2857{
2858 struct md_rdev *rdev;
2859 int rv = 0;
2860 /* Doing recovery so rcu locking not required */
2861 rdev = sh->raid_conf->disks[disk_idx].replacement;
2862 if (rdev
2863 && !test_bit(Faulty, &rdev->flags)
2864 && !test_bit(In_sync, &rdev->flags)
2865 && (rdev->recovery_offset <= sh->sector
2866 || rdev->mddev->recovery_cp <= sh->sector))
2867 rv = 1;
2868
2869 return rv;
2870}
2871
NeilBrown93b3dbc2011-07-27 11:00:36 +10002872/* fetch_block - checks the given member device to see if its data needs
Dan Williams1fe797e2008-06-28 09:16:30 +10002873 * to be read or computed to satisfy a request.
2874 *
2875 * Returns 1 when no more member devices need to be checked, otherwise returns
NeilBrown93b3dbc2011-07-27 11:00:36 +10002876 * 0 to tell the loop in handle_stripe_fill to continue
Dan Williamsf38e1212007-01-02 13:52:30 -07002877 */
NeilBrown93b3dbc2011-07-27 11:00:36 +10002878static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2879 int disk_idx, int disks)
Dan Williamsf38e1212007-01-02 13:52:30 -07002880{
2881 struct r5dev *dev = &sh->dev[disk_idx];
NeilBrown93b3dbc2011-07-27 11:00:36 +10002882 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2883 &sh->dev[s->failed_num[1]] };
Dan Williamsf38e1212007-01-02 13:52:30 -07002884
Dan Williamsf38e1212007-01-02 13:52:30 -07002885 /* is the data in this block needed, and can we get it? */
2886 if (!test_bit(R5_LOCKED, &dev->flags) &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002887 !test_bit(R5_UPTODATE, &dev->flags) &&
2888 (dev->toread ||
2889 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2890 s->syncing || s->expanding ||
NeilBrown9a3e1102011-12-23 10:17:53 +11002891 (s->replacing && want_replace(sh, disk_idx)) ||
NeilBrown5d35e092011-07-27 11:00:36 +10002892 (s->failed >= 1 && fdev[0]->toread) ||
2893 (s->failed >= 2 && fdev[1]->toread) ||
NeilBrown93b3dbc2011-07-27 11:00:36 +10002894 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
2895 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
2896 (sh->raid_conf->level == 6 && s->failed && s->to_write))) {
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002897 /* we would like to get this block, possibly by computing it,
2898 * otherwise read it if the backing disk is insync
2899 */
2900 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2901 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2902 if ((s->uptodate == disks - 1) &&
NeilBrownf2b3b442011-07-26 11:35:19 +10002903 (s->failed && (disk_idx == s->failed_num[0] ||
2904 disk_idx == s->failed_num[1]))) {
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002905 /* have disk failed, and we're requested to fetch it;
2906 * do compute it
2907 */
2908 pr_debug("Computing stripe %llu block %d\n",
2909 (unsigned long long)sh->sector, disk_idx);
2910 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2911 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2912 set_bit(R5_Wantcompute, &dev->flags);
2913 sh->ops.target = disk_idx;
2914 sh->ops.target2 = -1; /* no 2nd target */
2915 s->req_compute = 1;
NeilBrown93b3dbc2011-07-27 11:00:36 +10002916 /* Careful: from this point on 'uptodate' is in the eye
2917 * of raid_run_ops which services 'compute' operations
2918 * before writes. R5_Wantcompute flags a block that will
2919 * be R5_UPTODATE by the time it is needed for a
2920 * subsequent operation.
2921 */
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002922 s->uptodate++;
2923 return 1;
2924 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2925 /* Computing 2-failure is *very* expensive; only
2926 * do it if failed >= 2
2927 */
2928 int other;
2929 for (other = disks; other--; ) {
2930 if (other == disk_idx)
2931 continue;
2932 if (!test_bit(R5_UPTODATE,
2933 &sh->dev[other].flags))
2934 break;
2935 }
2936 BUG_ON(other < 0);
2937 pr_debug("Computing stripe %llu blocks %d,%d\n",
2938 (unsigned long long)sh->sector,
2939 disk_idx, other);
2940 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2941 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2942 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2943 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2944 sh->ops.target = disk_idx;
2945 sh->ops.target2 = other;
2946 s->uptodate += 2;
2947 s->req_compute = 1;
2948 return 1;
2949 } else if (test_bit(R5_Insync, &dev->flags)) {
2950 set_bit(R5_LOCKED, &dev->flags);
2951 set_bit(R5_Wantread, &dev->flags);
2952 s->locked++;
2953 pr_debug("Reading block %d (sync=%d)\n",
2954 disk_idx, s->syncing);
2955 }
2956 }
2957
2958 return 0;
2959}
2960
2961/**
NeilBrown93b3dbc2011-07-27 11:00:36 +10002962 * handle_stripe_fill - read or compute data to satisfy pending requests.
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002963 */
NeilBrown93b3dbc2011-07-27 11:00:36 +10002964static void handle_stripe_fill(struct stripe_head *sh,
2965 struct stripe_head_state *s,
2966 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07002967{
2968 int i;
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002969
2970 /* look for blocks to read/compute, skip this if a compute
2971 * is already in flight, or if the stripe contents are in the
2972 * midst of changing due to a write
2973 */
2974 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2975 !sh->reconstruct_state)
2976 for (i = disks; i--; )
NeilBrown93b3dbc2011-07-27 11:00:36 +10002977 if (fetch_block(sh, s, i, disks))
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002978 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002979 set_bit(STRIPE_HANDLE, &sh->state);
2980}
2981
2982
Dan Williams1fe797e2008-06-28 09:16:30 +10002983/* handle_stripe_clean_event
Dan Williamsa4456852007-07-09 11:56:43 -07002984 * any written block on an uptodate or failed drive can be returned.
2985 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2986 * never LOCKED, so we don't need to test 'failed' directly.
2987 */
NeilBrownd1688a62011-10-11 16:49:52 +11002988static void handle_stripe_clean_event(struct r5conf *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002989 struct stripe_head *sh, int disks, struct bio **return_bi)
2990{
2991 int i;
2992 struct r5dev *dev;
NeilBrownf8dfcff2013-03-12 12:18:06 +11002993 int discard_pending = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07002994
2995 for (i = disks; i--; )
2996 if (sh->dev[i].written) {
2997 dev = &sh->dev[i];
2998 if (!test_bit(R5_LOCKED, &dev->flags) &&
Shaohua Li9e4447682012-10-11 13:49:49 +11002999 (test_bit(R5_UPTODATE, &dev->flags) ||
NeilBrownca64cae2012-11-21 16:33:40 +11003000 test_bit(R5_Discard, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07003001 /* We can return any write requests */
3002 struct bio *wbi, *wbi2;
Dan Williams45b42332007-07-09 11:56:43 -07003003 pr_debug("Return write for disc %d\n", i);
NeilBrownca64cae2012-11-21 16:33:40 +11003004 if (test_and_clear_bit(R5_Discard, &dev->flags))
3005 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003006 wbi = dev->written;
3007 dev->written = NULL;
3008 while (wbi && wbi->bi_sector <
3009 dev->sector + STRIPE_SECTORS) {
3010 wbi2 = r5_next_bio(wbi, dev->sector);
Shaohua Lie7836bd62012-07-19 16:01:31 +10003011 if (!raid5_dec_bi_active_stripes(wbi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003012 md_write_end(conf->mddev);
3013 wbi->bi_next = *return_bi;
3014 *return_bi = wbi;
3015 }
3016 wbi = wbi2;
3017 }
Shaohua Li7eaf7e82012-07-19 16:01:31 +10003018 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3019 STRIPE_SECTORS,
Dan Williamsa4456852007-07-09 11:56:43 -07003020 !test_bit(STRIPE_DEGRADED, &sh->state),
Shaohua Li7eaf7e82012-07-19 16:01:31 +10003021 0);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003022 } else if (test_bit(R5_Discard, &dev->flags))
3023 discard_pending = 1;
3024 }
3025 if (!discard_pending &&
3026 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
3027 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3028 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3029 if (sh->qd_idx >= 0) {
3030 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3031 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3032 }
3033 /* now that discard is done we can proceed with any sync */
3034 clear_bit(STRIPE_DISCARD, &sh->state);
Shaohua Lid47648f2013-10-19 14:51:42 +08003035 /*
3036 * SCSI discard will change some bio fields and the stripe has
3037 * no updated data, so remove it from hash list and the stripe
3038 * will be reinitialized
3039 */
3040 spin_lock_irq(&conf->device_lock);
3041 remove_hash(sh);
3042 spin_unlock_irq(&conf->device_lock);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003043 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3044 set_bit(STRIPE_HANDLE, &sh->state);
3045
3046 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003047
3048 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3049 if (atomic_dec_and_test(&conf->pending_full_writes))
3050 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07003051}
3052
NeilBrownd1688a62011-10-11 16:49:52 +11003053static void handle_stripe_dirtying(struct r5conf *conf,
NeilBrownc8ac1802011-07-27 11:00:36 +10003054 struct stripe_head *sh,
3055 struct stripe_head_state *s,
3056 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07003057{
3058 int rmw = 0, rcw = 0, i;
Alexander Lyakasa7854482012-10-11 13:50:12 +11003059 sector_t recovery_cp = conf->mddev->recovery_cp;
3060
3061 /* RAID6 requires 'rcw' in current implementation.
3062 * Otherwise, check whether resync is now happening or should start.
3063 * If yes, then the array is dirty (after unclean shutdown or
3064 * initial creation), so parity in some stripes might be inconsistent.
3065 * In this case, we need to always do reconstruct-write, to ensure
3066 * that in case of drive failure or read-error correction, we
3067 * generate correct data from the parity.
3068 */
3069 if (conf->max_degraded == 2 ||
3070 (recovery_cp < MaxSector && sh->sector >= recovery_cp)) {
3071 /* Calculate the real rcw later - for now make it
NeilBrownc8ac1802011-07-27 11:00:36 +10003072 * look like rcw is cheaper
3073 */
3074 rcw = 1; rmw = 2;
Alexander Lyakasa7854482012-10-11 13:50:12 +11003075 pr_debug("force RCW max_degraded=%u, recovery_cp=%llu sh->sector=%llu\n",
3076 conf->max_degraded, (unsigned long long)recovery_cp,
3077 (unsigned long long)sh->sector);
NeilBrownc8ac1802011-07-27 11:00:36 +10003078 } else for (i = disks; i--; ) {
Dan Williamsa4456852007-07-09 11:56:43 -07003079 /* would I have to read this buffer for read_modify_write */
3080 struct r5dev *dev = &sh->dev[i];
3081 if ((dev->towrite || i == sh->pd_idx) &&
3082 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003083 !(test_bit(R5_UPTODATE, &dev->flags) ||
3084 test_bit(R5_Wantcompute, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07003085 if (test_bit(R5_Insync, &dev->flags))
3086 rmw++;
3087 else
3088 rmw += 2*disks; /* cannot read it */
3089 }
3090 /* Would I have to read this buffer for reconstruct_write */
3091 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
3092 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003093 !(test_bit(R5_UPTODATE, &dev->flags) ||
3094 test_bit(R5_Wantcompute, &dev->flags))) {
3095 if (test_bit(R5_Insync, &dev->flags)) rcw++;
Dan Williamsa4456852007-07-09 11:56:43 -07003096 else
3097 rcw += 2*disks;
3098 }
3099 }
Dan Williams45b42332007-07-09 11:56:43 -07003100 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003101 (unsigned long long)sh->sector, rmw, rcw);
3102 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrowna9add5d2012-10-31 11:59:09 +11003103 if (rmw < rcw && rmw > 0) {
Dan Williamsa4456852007-07-09 11:56:43 -07003104 /* prefer read-modify-write, but need to get some data */
Jonathan Brassowe3620a32013-03-07 16:22:01 -06003105 if (conf->mddev->queue)
3106 blk_add_trace_msg(conf->mddev->queue,
3107 "raid5 rmw %llu %d",
3108 (unsigned long long)sh->sector, rmw);
Dan Williamsa4456852007-07-09 11:56:43 -07003109 for (i = disks; i--; ) {
3110 struct r5dev *dev = &sh->dev[i];
3111 if ((dev->towrite || i == sh->pd_idx) &&
3112 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003113 !(test_bit(R5_UPTODATE, &dev->flags) ||
3114 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07003115 test_bit(R5_Insync, &dev->flags)) {
3116 if (
3117 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07003118 pr_debug("Read_old block "
NeilBrowna9add5d2012-10-31 11:59:09 +11003119 "%d for r-m-w\n", i);
Dan Williamsa4456852007-07-09 11:56:43 -07003120 set_bit(R5_LOCKED, &dev->flags);
3121 set_bit(R5_Wantread, &dev->flags);
3122 s->locked++;
3123 } else {
3124 set_bit(STRIPE_DELAYED, &sh->state);
3125 set_bit(STRIPE_HANDLE, &sh->state);
3126 }
3127 }
3128 }
NeilBrowna9add5d2012-10-31 11:59:09 +11003129 }
NeilBrownc8ac1802011-07-27 11:00:36 +10003130 if (rcw <= rmw && rcw > 0) {
Dan Williamsa4456852007-07-09 11:56:43 -07003131 /* want reconstruct write, but need to get some data */
NeilBrowna9add5d2012-10-31 11:59:09 +11003132 int qread =0;
NeilBrownc8ac1802011-07-27 11:00:36 +10003133 rcw = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07003134 for (i = disks; i--; ) {
3135 struct r5dev *dev = &sh->dev[i];
3136 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
NeilBrownc8ac1802011-07-27 11:00:36 +10003137 i != sh->pd_idx && i != sh->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07003138 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003139 !(test_bit(R5_UPTODATE, &dev->flags) ||
NeilBrownc8ac1802011-07-27 11:00:36 +10003140 test_bit(R5_Wantcompute, &dev->flags))) {
3141 rcw++;
3142 if (!test_bit(R5_Insync, &dev->flags))
3143 continue; /* it's a failed drive */
Dan Williamsa4456852007-07-09 11:56:43 -07003144 if (
3145 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07003146 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07003147 "%d for Reconstruct\n", i);
3148 set_bit(R5_LOCKED, &dev->flags);
3149 set_bit(R5_Wantread, &dev->flags);
3150 s->locked++;
NeilBrowna9add5d2012-10-31 11:59:09 +11003151 qread++;
Dan Williamsa4456852007-07-09 11:56:43 -07003152 } else {
3153 set_bit(STRIPE_DELAYED, &sh->state);
3154 set_bit(STRIPE_HANDLE, &sh->state);
3155 }
3156 }
3157 }
Jonathan Brassowe3620a32013-03-07 16:22:01 -06003158 if (rcw && conf->mddev->queue)
NeilBrowna9add5d2012-10-31 11:59:09 +11003159 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3160 (unsigned long long)sh->sector,
3161 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
NeilBrownc8ac1802011-07-27 11:00:36 +10003162 }
Dan Williamsa4456852007-07-09 11:56:43 -07003163 /* now if nothing is locked, and if we have enough data,
3164 * we can start a write request
3165 */
Dan Williamsf38e1212007-01-02 13:52:30 -07003166 /* since handle_stripe can be called at any time we need to handle the
3167 * case where a compute block operation has been submitted and then a
Dan Williamsac6b53b2009-07-14 13:40:19 -07003168 * subsequent call wants to start a write request. raid_run_ops only
3169 * handles the case where compute block and reconstruct are requested
Dan Williamsf38e1212007-01-02 13:52:30 -07003170 * simultaneously. If this is not the case then new writes need to be
3171 * held off until the compute completes.
3172 */
Dan Williams976ea8d2008-06-28 08:32:03 +10003173 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3174 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
3175 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07003176 schedule_reconstruction(sh, s, rcw == 0, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07003177}
3178
NeilBrownd1688a62011-10-11 16:49:52 +11003179static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07003180 struct stripe_head_state *s, int disks)
3181{
Dan Williamsecc65c92008-06-28 08:31:57 +10003182 struct r5dev *dev = NULL;
Dan Williamse89f8962007-01-02 13:52:31 -07003183
Dan Williamsbd2ab672008-04-10 21:29:27 -07003184 set_bit(STRIPE_HANDLE, &sh->state);
3185
Dan Williamsecc65c92008-06-28 08:31:57 +10003186 switch (sh->check_state) {
3187 case check_state_idle:
3188 /* start a new check operation if there are no failures */
Dan Williamsbd2ab672008-04-10 21:29:27 -07003189 if (s->failed == 0) {
Dan Williamsbd2ab672008-04-10 21:29:27 -07003190 BUG_ON(s->uptodate != disks);
Dan Williamsecc65c92008-06-28 08:31:57 +10003191 sh->check_state = check_state_run;
3192 set_bit(STRIPE_OP_CHECK, &s->ops_request);
Dan Williamsbd2ab672008-04-10 21:29:27 -07003193 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
Dan Williamsbd2ab672008-04-10 21:29:27 -07003194 s->uptodate--;
Dan Williamsecc65c92008-06-28 08:31:57 +10003195 break;
Dan Williamsbd2ab672008-04-10 21:29:27 -07003196 }
NeilBrownf2b3b442011-07-26 11:35:19 +10003197 dev = &sh->dev[s->failed_num[0]];
Dan Williamsecc65c92008-06-28 08:31:57 +10003198 /* fall through */
3199 case check_state_compute_result:
3200 sh->check_state = check_state_idle;
3201 if (!dev)
3202 dev = &sh->dev[sh->pd_idx];
3203
3204 /* check that a write has not made the stripe insync */
3205 if (test_bit(STRIPE_INSYNC, &sh->state))
3206 break;
3207
3208 /* either failed parity check, or recovery is happening */
Dan Williamsa4456852007-07-09 11:56:43 -07003209 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3210 BUG_ON(s->uptodate != disks);
3211
3212 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsecc65c92008-06-28 08:31:57 +10003213 s->locked++;
Dan Williamsa4456852007-07-09 11:56:43 -07003214 set_bit(R5_Wantwrite, &dev->flags);
Dan Williams830ea012007-01-02 13:52:31 -07003215
Dan Williamsa4456852007-07-09 11:56:43 -07003216 clear_bit(STRIPE_DEGRADED, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003217 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10003218 break;
3219 case check_state_run:
3220 break; /* we will be called again upon completion */
3221 case check_state_check_result:
3222 sh->check_state = check_state_idle;
3223
3224 /* if a failure occurred during the check operation, leave
3225 * STRIPE_INSYNC not set and let the stripe be handled again
3226 */
3227 if (s->failed)
3228 break;
3229
3230 /* handle a successful check operation, if parity is correct
3231 * we are done. Otherwise update the mismatch count and repair
3232 * parity if !MD_RECOVERY_CHECK
3233 */
Dan Williamsad283ea2009-08-29 19:09:26 -07003234 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
Dan Williamsecc65c92008-06-28 08:31:57 +10003235 /* parity is correct (on disc,
3236 * not in buffer any more)
3237 */
3238 set_bit(STRIPE_INSYNC, &sh->state);
3239 else {
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11003240 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
Dan Williamsecc65c92008-06-28 08:31:57 +10003241 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3242 /* don't try to repair!! */
3243 set_bit(STRIPE_INSYNC, &sh->state);
3244 else {
3245 sh->check_state = check_state_compute_run;
Dan Williams976ea8d2008-06-28 08:32:03 +10003246 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10003247 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3248 set_bit(R5_Wantcompute,
3249 &sh->dev[sh->pd_idx].flags);
3250 sh->ops.target = sh->pd_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07003251 sh->ops.target2 = -1;
Dan Williamsecc65c92008-06-28 08:31:57 +10003252 s->uptodate++;
3253 }
3254 }
3255 break;
3256 case check_state_compute_run:
3257 break;
3258 default:
3259 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3260 __func__, sh->check_state,
3261 (unsigned long long) sh->sector);
3262 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07003263 }
3264}
3265
3266
NeilBrownd1688a62011-10-11 16:49:52 +11003267static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
Dan Williams36d1c642009-07-14 11:48:22 -07003268 struct stripe_head_state *s,
NeilBrownf2b3b442011-07-26 11:35:19 +10003269 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07003270{
Dan Williamsa4456852007-07-09 11:56:43 -07003271 int pd_idx = sh->pd_idx;
NeilBrown34e04e82009-03-31 15:10:16 +11003272 int qd_idx = sh->qd_idx;
Dan Williamsd82dfee2009-07-14 13:40:57 -07003273 struct r5dev *dev;
Dan Williamsa4456852007-07-09 11:56:43 -07003274
3275 set_bit(STRIPE_HANDLE, &sh->state);
3276
3277 BUG_ON(s->failed > 2);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003278
Dan Williamsa4456852007-07-09 11:56:43 -07003279 /* Want to check and possibly repair P and Q.
3280 * However there could be one 'failed' device, in which
3281 * case we can only check one of them, possibly using the
3282 * other to generate missing data
3283 */
3284
Dan Williamsd82dfee2009-07-14 13:40:57 -07003285 switch (sh->check_state) {
3286 case check_state_idle:
3287 /* start a new check operation if there are < 2 failures */
NeilBrownf2b3b442011-07-26 11:35:19 +10003288 if (s->failed == s->q_failed) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07003289 /* The only possible failed device holds Q, so it
Dan Williamsa4456852007-07-09 11:56:43 -07003290 * makes sense to check P (If anything else were failed,
3291 * we would have used P to recreate it).
3292 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003293 sh->check_state = check_state_run;
Dan Williamsa4456852007-07-09 11:56:43 -07003294 }
NeilBrownf2b3b442011-07-26 11:35:19 +10003295 if (!s->q_failed && s->failed < 2) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07003296 /* Q is not failed, and we didn't use it to generate
Dan Williamsa4456852007-07-09 11:56:43 -07003297 * anything, so it makes sense to check it
3298 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003299 if (sh->check_state == check_state_run)
3300 sh->check_state = check_state_run_pq;
3301 else
3302 sh->check_state = check_state_run_q;
Dan Williamsa4456852007-07-09 11:56:43 -07003303 }
Dan Williams36d1c642009-07-14 11:48:22 -07003304
Dan Williamsd82dfee2009-07-14 13:40:57 -07003305 /* discard potentially stale zero_sum_result */
3306 sh->ops.zero_sum_result = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07003307
Dan Williamsd82dfee2009-07-14 13:40:57 -07003308 if (sh->check_state == check_state_run) {
3309 /* async_xor_zero_sum destroys the contents of P */
3310 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3311 s->uptodate--;
Dan Williamsa4456852007-07-09 11:56:43 -07003312 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003313 if (sh->check_state >= check_state_run &&
3314 sh->check_state <= check_state_run_pq) {
3315 /* async_syndrome_zero_sum preserves P and Q, so
3316 * no need to mark them !uptodate here
3317 */
3318 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3319 break;
3320 }
Dan Williams36d1c642009-07-14 11:48:22 -07003321
Dan Williamsd82dfee2009-07-14 13:40:57 -07003322 /* we have 2-disk failure */
3323 BUG_ON(s->failed != 2);
3324 /* fall through */
3325 case check_state_compute_result:
3326 sh->check_state = check_state_idle;
Dan Williams36d1c642009-07-14 11:48:22 -07003327
Dan Williamsd82dfee2009-07-14 13:40:57 -07003328 /* check that a write has not made the stripe insync */
3329 if (test_bit(STRIPE_INSYNC, &sh->state))
3330 break;
Dan Williamsa4456852007-07-09 11:56:43 -07003331
3332 /* now write out any block on a failed drive,
Dan Williamsd82dfee2009-07-14 13:40:57 -07003333 * or P or Q if they were recomputed
Dan Williamsa4456852007-07-09 11:56:43 -07003334 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003335 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
Dan Williamsa4456852007-07-09 11:56:43 -07003336 if (s->failed == 2) {
NeilBrownf2b3b442011-07-26 11:35:19 +10003337 dev = &sh->dev[s->failed_num[1]];
Dan Williamsa4456852007-07-09 11:56:43 -07003338 s->locked++;
3339 set_bit(R5_LOCKED, &dev->flags);
3340 set_bit(R5_Wantwrite, &dev->flags);
3341 }
3342 if (s->failed >= 1) {
NeilBrownf2b3b442011-07-26 11:35:19 +10003343 dev = &sh->dev[s->failed_num[0]];
Dan Williamsa4456852007-07-09 11:56:43 -07003344 s->locked++;
3345 set_bit(R5_LOCKED, &dev->flags);
3346 set_bit(R5_Wantwrite, &dev->flags);
3347 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003348 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07003349 dev = &sh->dev[pd_idx];
3350 s->locked++;
3351 set_bit(R5_LOCKED, &dev->flags);
3352 set_bit(R5_Wantwrite, &dev->flags);
3353 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003354 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07003355 dev = &sh->dev[qd_idx];
3356 s->locked++;
3357 set_bit(R5_LOCKED, &dev->flags);
3358 set_bit(R5_Wantwrite, &dev->flags);
3359 }
3360 clear_bit(STRIPE_DEGRADED, &sh->state);
3361
3362 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003363 break;
3364 case check_state_run:
3365 case check_state_run_q:
3366 case check_state_run_pq:
3367 break; /* we will be called again upon completion */
3368 case check_state_check_result:
3369 sh->check_state = check_state_idle;
3370
3371 /* handle a successful check operation, if parity is correct
3372 * we are done. Otherwise update the mismatch count and repair
3373 * parity if !MD_RECOVERY_CHECK
3374 */
3375 if (sh->ops.zero_sum_result == 0) {
3376 /* both parities are correct */
3377 if (!s->failed)
3378 set_bit(STRIPE_INSYNC, &sh->state);
3379 else {
3380 /* in contrast to the raid5 case we can validate
3381 * parity, but still have a failure to write
3382 * back
3383 */
3384 sh->check_state = check_state_compute_result;
3385 /* Returning at this point means that we may go
3386 * off and bring p and/or q uptodate again so
3387 * we make sure to check zero_sum_result again
3388 * to verify if p or q need writeback
3389 */
3390 }
3391 } else {
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11003392 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003393 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3394 /* don't try to repair!! */
3395 set_bit(STRIPE_INSYNC, &sh->state);
3396 else {
3397 int *target = &sh->ops.target;
3398
3399 sh->ops.target = -1;
3400 sh->ops.target2 = -1;
3401 sh->check_state = check_state_compute_run;
3402 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3403 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3404 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3405 set_bit(R5_Wantcompute,
3406 &sh->dev[pd_idx].flags);
3407 *target = pd_idx;
3408 target = &sh->ops.target2;
3409 s->uptodate++;
3410 }
3411 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3412 set_bit(R5_Wantcompute,
3413 &sh->dev[qd_idx].flags);
3414 *target = qd_idx;
3415 s->uptodate++;
3416 }
3417 }
3418 }
3419 break;
3420 case check_state_compute_run:
3421 break;
3422 default:
3423 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3424 __func__, sh->check_state,
3425 (unsigned long long) sh->sector);
3426 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07003427 }
3428}
3429
NeilBrownd1688a62011-10-11 16:49:52 +11003430static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
Dan Williamsa4456852007-07-09 11:56:43 -07003431{
3432 int i;
3433
3434 /* We have read all the blocks in this stripe and now we need to
3435 * copy some of them into a target stripe for expand.
3436 */
Dan Williamsf0a50d32007-01-02 13:52:31 -07003437 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07003438 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3439 for (i = 0; i < sh->disks; i++)
NeilBrown34e04e82009-03-31 15:10:16 +11003440 if (i != sh->pd_idx && i != sh->qd_idx) {
NeilBrown911d4ee2009-03-31 14:39:38 +11003441 int dd_idx, j;
Dan Williamsa4456852007-07-09 11:56:43 -07003442 struct stripe_head *sh2;
Dan Williamsa08abd82009-06-03 11:43:59 -07003443 struct async_submit_ctl submit;
Dan Williamsa4456852007-07-09 11:56:43 -07003444
NeilBrown784052e2009-03-31 15:19:07 +11003445 sector_t bn = compute_blocknr(sh, i, 1);
NeilBrown911d4ee2009-03-31 14:39:38 +11003446 sector_t s = raid5_compute_sector(conf, bn, 0,
3447 &dd_idx, NULL);
NeilBrowna8c906c2009-06-09 14:39:59 +10003448 sh2 = get_active_stripe(conf, s, 0, 1, 1);
Dan Williamsa4456852007-07-09 11:56:43 -07003449 if (sh2 == NULL)
3450 /* so far only the early blocks of this stripe
3451 * have been requested. When later blocks
3452 * get requested, we will try again
3453 */
3454 continue;
3455 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3456 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3457 /* must have already done this block */
3458 release_stripe(sh2);
3459 continue;
3460 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07003461
3462 /* place all the copies on one channel */
Dan Williamsa08abd82009-06-03 11:43:59 -07003463 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003464 tx = async_memcpy(sh2->dev[dd_idx].page,
Dan Williams88ba2aa2009-04-09 16:16:18 -07003465 sh->dev[i].page, 0, 0, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07003466 &submit);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003467
Dan Williamsa4456852007-07-09 11:56:43 -07003468 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3469 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3470 for (j = 0; j < conf->raid_disks; j++)
3471 if (j != sh2->pd_idx &&
NeilBrown86c374b2011-07-27 11:00:36 +10003472 j != sh2->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07003473 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3474 break;
3475 if (j == conf->raid_disks) {
3476 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3477 set_bit(STRIPE_HANDLE, &sh2->state);
3478 }
3479 release_stripe(sh2);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003480
Dan Williamsa4456852007-07-09 11:56:43 -07003481 }
NeilBrowna2e08552007-09-11 15:23:36 -07003482 /* done submitting copies, wait for them to complete */
NeilBrown749586b2012-11-20 14:11:15 +11003483 async_tx_quiesce(&tx);
Dan Williamsa4456852007-07-09 11:56:43 -07003484}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485
3486/*
3487 * handle_stripe - do things to a stripe.
3488 *
NeilBrown9a3e1102011-12-23 10:17:53 +11003489 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3490 * state of various bits to see what needs to be done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 * Possible results:
NeilBrown9a3e1102011-12-23 10:17:53 +11003492 * return some read requests which now have data
3493 * return some write requests which are safely on storage
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494 * schedule a read on some buffers
3495 * schedule a write of some buffers
3496 * return confirmation of parity correctness
3497 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 */
Dan Williamsa4456852007-07-09 11:56:43 -07003499
NeilBrownacfe7262011-07-27 11:00:36 +10003500static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
NeilBrown16a53ec2006-06-26 00:27:38 -07003501{
NeilBrownd1688a62011-10-11 16:49:52 +11003502 struct r5conf *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08003503 int disks = sh->disks;
NeilBrown474af965fe2011-07-27 11:00:36 +10003504 struct r5dev *dev;
3505 int i;
NeilBrown9a3e1102011-12-23 10:17:53 +11003506 int do_recovery = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003507
NeilBrownacfe7262011-07-27 11:00:36 +10003508 memset(s, 0, sizeof(*s));
NeilBrown16a53ec2006-06-26 00:27:38 -07003509
NeilBrownacfe7262011-07-27 11:00:36 +10003510 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3511 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3512 s->failed_num[0] = -1;
3513 s->failed_num[1] = -1;
3514
3515 /* Now to look around and see what can be done */
NeilBrown16a53ec2006-06-26 00:27:38 -07003516 rcu_read_lock();
3517 for (i=disks; i--; ) {
NeilBrown3cb03002011-10-11 16:45:26 +11003518 struct md_rdev *rdev;
NeilBrown31c176e2011-07-28 11:39:22 +10003519 sector_t first_bad;
3520 int bad_sectors;
3521 int is_bad = 0;
NeilBrownacfe7262011-07-27 11:00:36 +10003522
NeilBrown16a53ec2006-06-26 00:27:38 -07003523 dev = &sh->dev[i];
NeilBrown16a53ec2006-06-26 00:27:38 -07003524
Dan Williams45b42332007-07-09 11:56:43 -07003525 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown9a3e1102011-12-23 10:17:53 +11003526 i, dev->flags,
3527 dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003528 /* maybe we can reply to a read
3529 *
3530 * new wantfill requests are only permitted while
3531 * ops_complete_biofill is guaranteed to be inactive
3532 */
3533 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3534 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3535 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07003536
3537 /* now count some things */
NeilBrowncc940152011-07-26 11:35:35 +10003538 if (test_bit(R5_LOCKED, &dev->flags))
3539 s->locked++;
3540 if (test_bit(R5_UPTODATE, &dev->flags))
3541 s->uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003542 if (test_bit(R5_Wantcompute, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003543 s->compute++;
3544 BUG_ON(s->compute > 2);
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003545 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003546
NeilBrownacfe7262011-07-27 11:00:36 +10003547 if (test_bit(R5_Wantfill, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003548 s->to_fill++;
NeilBrownacfe7262011-07-27 11:00:36 +10003549 else if (dev->toread)
NeilBrowncc940152011-07-26 11:35:35 +10003550 s->to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003551 if (dev->towrite) {
NeilBrowncc940152011-07-26 11:35:35 +10003552 s->to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003553 if (!test_bit(R5_OVERWRITE, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003554 s->non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003555 }
Dan Williamsa4456852007-07-09 11:56:43 -07003556 if (dev->written)
NeilBrowncc940152011-07-26 11:35:35 +10003557 s->written++;
NeilBrown14a75d32011-12-23 10:17:52 +11003558 /* Prefer to use the replacement for reads, but only
3559 * if it is recovered enough and has no bad blocks.
3560 */
3561 rdev = rcu_dereference(conf->disks[i].replacement);
3562 if (rdev && !test_bit(Faulty, &rdev->flags) &&
3563 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
3564 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3565 &first_bad, &bad_sectors))
3566 set_bit(R5_ReadRepl, &dev->flags);
3567 else {
NeilBrown9a3e1102011-12-23 10:17:53 +11003568 if (rdev)
3569 set_bit(R5_NeedReplace, &dev->flags);
NeilBrown14a75d32011-12-23 10:17:52 +11003570 rdev = rcu_dereference(conf->disks[i].rdev);
3571 clear_bit(R5_ReadRepl, &dev->flags);
3572 }
NeilBrown9283d8c2011-12-08 16:27:57 +11003573 if (rdev && test_bit(Faulty, &rdev->flags))
3574 rdev = NULL;
NeilBrown31c176e2011-07-28 11:39:22 +10003575 if (rdev) {
3576 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3577 &first_bad, &bad_sectors);
3578 if (s->blocked_rdev == NULL
3579 && (test_bit(Blocked, &rdev->flags)
3580 || is_bad < 0)) {
3581 if (is_bad < 0)
3582 set_bit(BlockedBadBlocks,
3583 &rdev->flags);
3584 s->blocked_rdev = rdev;
3585 atomic_inc(&rdev->nr_pending);
3586 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07003587 }
NeilBrown415e72d2010-06-17 17:25:21 +10003588 clear_bit(R5_Insync, &dev->flags);
3589 if (!rdev)
3590 /* Not in-sync */;
NeilBrown31c176e2011-07-28 11:39:22 +10003591 else if (is_bad) {
3592 /* also not in-sync */
NeilBrown18b98372012-04-01 23:48:38 +10003593 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
3594 test_bit(R5_UPTODATE, &dev->flags)) {
NeilBrown31c176e2011-07-28 11:39:22 +10003595 /* treat as in-sync, but with a read error
3596 * which we can now try to correct
3597 */
3598 set_bit(R5_Insync, &dev->flags);
3599 set_bit(R5_ReadError, &dev->flags);
3600 }
3601 } else if (test_bit(In_sync, &rdev->flags))
NeilBrown415e72d2010-06-17 17:25:21 +10003602 set_bit(R5_Insync, &dev->flags);
NeilBrown30d7a482011-12-23 09:57:00 +11003603 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
NeilBrown415e72d2010-06-17 17:25:21 +10003604 /* in sync if before recovery_offset */
NeilBrown30d7a482011-12-23 09:57:00 +11003605 set_bit(R5_Insync, &dev->flags);
3606 else if (test_bit(R5_UPTODATE, &dev->flags) &&
3607 test_bit(R5_Expanded, &dev->flags))
3608 /* If we've reshaped into here, we assume it is Insync.
3609 * We will shortly update recovery_offset to make
3610 * it official.
3611 */
3612 set_bit(R5_Insync, &dev->flags);
3613
Adam Kwolek5d8c71f2011-12-09 14:26:11 +11003614 if (rdev && test_bit(R5_WriteError, &dev->flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11003615 /* This flag does not apply to '.replacement'
3616 * only to .rdev, so make sure to check that*/
3617 struct md_rdev *rdev2 = rcu_dereference(
3618 conf->disks[i].rdev);
3619 if (rdev2 == rdev)
3620 clear_bit(R5_Insync, &dev->flags);
3621 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
NeilBrownbc2607f2011-07-28 11:39:22 +10003622 s->handle_bad_blocks = 1;
NeilBrown14a75d32011-12-23 10:17:52 +11003623 atomic_inc(&rdev2->nr_pending);
NeilBrownbc2607f2011-07-28 11:39:22 +10003624 } else
3625 clear_bit(R5_WriteError, &dev->flags);
3626 }
Adam Kwolek5d8c71f2011-12-09 14:26:11 +11003627 if (rdev && test_bit(R5_MadeGood, &dev->flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11003628 /* This flag does not apply to '.replacement'
3629 * only to .rdev, so make sure to check that*/
3630 struct md_rdev *rdev2 = rcu_dereference(
3631 conf->disks[i].rdev);
3632 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
NeilBrownb84db562011-07-28 11:39:23 +10003633 s->handle_bad_blocks = 1;
NeilBrown14a75d32011-12-23 10:17:52 +11003634 atomic_inc(&rdev2->nr_pending);
NeilBrownb84db562011-07-28 11:39:23 +10003635 } else
3636 clear_bit(R5_MadeGood, &dev->flags);
3637 }
NeilBrown977df362011-12-23 10:17:53 +11003638 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
3639 struct md_rdev *rdev2 = rcu_dereference(
3640 conf->disks[i].replacement);
3641 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
3642 s->handle_bad_blocks = 1;
3643 atomic_inc(&rdev2->nr_pending);
3644 } else
3645 clear_bit(R5_MadeGoodRepl, &dev->flags);
3646 }
NeilBrown415e72d2010-06-17 17:25:21 +10003647 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003648 /* The ReadError flag will just be confusing now */
3649 clear_bit(R5_ReadError, &dev->flags);
3650 clear_bit(R5_ReWrite, &dev->flags);
3651 }
NeilBrown415e72d2010-06-17 17:25:21 +10003652 if (test_bit(R5_ReadError, &dev->flags))
3653 clear_bit(R5_Insync, &dev->flags);
3654 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003655 if (s->failed < 2)
3656 s->failed_num[s->failed] = i;
3657 s->failed++;
NeilBrown9a3e1102011-12-23 10:17:53 +11003658 if (rdev && !test_bit(Faulty, &rdev->flags))
3659 do_recovery = 1;
NeilBrown415e72d2010-06-17 17:25:21 +10003660 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003661 }
NeilBrown9a3e1102011-12-23 10:17:53 +11003662 if (test_bit(STRIPE_SYNCING, &sh->state)) {
3663 /* If there is a failed device being replaced,
3664 * we must be recovering.
3665 * else if we are after recovery_cp, we must be syncing
majianpengc6d2e082012-04-02 01:16:59 +10003666 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
NeilBrown9a3e1102011-12-23 10:17:53 +11003667 * else we can only be replacing
3668 * sync and recovery both need to read all devices, and so
3669 * use the same flag.
3670 */
3671 if (do_recovery ||
majianpengc6d2e082012-04-02 01:16:59 +10003672 sh->sector >= conf->mddev->recovery_cp ||
3673 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
NeilBrown9a3e1102011-12-23 10:17:53 +11003674 s->syncing = 1;
3675 else
3676 s->replacing = 1;
3677 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003678 rcu_read_unlock();
NeilBrowncc940152011-07-26 11:35:35 +10003679}
NeilBrownf4168852007-02-28 20:11:53 -08003680
NeilBrowncc940152011-07-26 11:35:35 +10003681static void handle_stripe(struct stripe_head *sh)
3682{
3683 struct stripe_head_state s;
NeilBrownd1688a62011-10-11 16:49:52 +11003684 struct r5conf *conf = sh->raid_conf;
NeilBrown3687c062011-07-27 11:00:36 +10003685 int i;
NeilBrown84789552011-07-27 11:00:36 +10003686 int prexor;
3687 int disks = sh->disks;
NeilBrown474af965fe2011-07-27 11:00:36 +10003688 struct r5dev *pdev, *qdev;
NeilBrowncc940152011-07-26 11:35:35 +10003689
3690 clear_bit(STRIPE_HANDLE, &sh->state);
Dan Williams257a4b42011-11-08 16:22:06 +11003691 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
NeilBrowncc940152011-07-26 11:35:35 +10003692 /* already being handled, ensure it gets handled
3693 * again when current action finishes */
3694 set_bit(STRIPE_HANDLE, &sh->state);
3695 return;
3696 }
3697
NeilBrownf8dfcff2013-03-12 12:18:06 +11003698 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3699 spin_lock(&sh->stripe_lock);
3700 /* Cannot process 'sync' concurrently with 'discard' */
3701 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
3702 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3703 set_bit(STRIPE_SYNCING, &sh->state);
3704 clear_bit(STRIPE_INSYNC, &sh->state);
NeilBrownf94c0b62013-07-22 12:57:21 +10003705 clear_bit(STRIPE_REPLACED, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003706 }
3707 spin_unlock(&sh->stripe_lock);
NeilBrowncc940152011-07-26 11:35:35 +10003708 }
3709 clear_bit(STRIPE_DELAYED, &sh->state);
3710
3711 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3712 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3713 (unsigned long long)sh->sector, sh->state,
3714 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3715 sh->check_state, sh->reconstruct_state);
NeilBrowncc940152011-07-26 11:35:35 +10003716
NeilBrownacfe7262011-07-27 11:00:36 +10003717 analyse_stripe(sh, &s);
NeilBrownc5a31002011-07-27 11:00:36 +10003718
NeilBrownbc2607f2011-07-28 11:39:22 +10003719 if (s.handle_bad_blocks) {
3720 set_bit(STRIPE_HANDLE, &sh->state);
3721 goto finish;
3722 }
3723
NeilBrown474af965fe2011-07-27 11:00:36 +10003724 if (unlikely(s.blocked_rdev)) {
3725 if (s.syncing || s.expanding || s.expanded ||
NeilBrown9a3e1102011-12-23 10:17:53 +11003726 s.replacing || s.to_write || s.written) {
NeilBrown474af965fe2011-07-27 11:00:36 +10003727 set_bit(STRIPE_HANDLE, &sh->state);
3728 goto finish;
3729 }
3730 /* There is nothing for the blocked_rdev to block */
3731 rdev_dec_pending(s.blocked_rdev, conf->mddev);
3732 s.blocked_rdev = NULL;
3733 }
3734
3735 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3736 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3737 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3738 }
3739
3740 pr_debug("locked=%d uptodate=%d to_read=%d"
3741 " to_write=%d failed=%d failed_num=%d,%d\n",
3742 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3743 s.failed_num[0], s.failed_num[1]);
3744 /* check if the array has lost more than max_degraded devices and,
3745 * if so, some requests might need to be failed.
3746 */
NeilBrown9a3f5302011-11-08 16:22:01 +11003747 if (s.failed > conf->max_degraded) {
3748 sh->check_state = 0;
3749 sh->reconstruct_state = 0;
3750 if (s.to_read+s.to_write+s.written)
3751 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
NeilBrown9a3e1102011-12-23 10:17:53 +11003752 if (s.syncing + s.replacing)
NeilBrown9a3f5302011-11-08 16:22:01 +11003753 handle_failed_sync(conf, sh, &s);
3754 }
NeilBrown474af965fe2011-07-27 11:00:36 +10003755
NeilBrown84789552011-07-27 11:00:36 +10003756 /* Now we check to see if any write operations have recently
3757 * completed
3758 */
3759 prexor = 0;
3760 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
3761 prexor = 1;
3762 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3763 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
3764 sh->reconstruct_state = reconstruct_state_idle;
3765
3766 /* All the 'written' buffers and the parity block are ready to
3767 * be written back to disk
3768 */
Shaohua Li9e4447682012-10-11 13:49:49 +11003769 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
3770 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
NeilBrown84789552011-07-27 11:00:36 +10003771 BUG_ON(sh->qd_idx >= 0 &&
Shaohua Li9e4447682012-10-11 13:49:49 +11003772 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
3773 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
NeilBrown84789552011-07-27 11:00:36 +10003774 for (i = disks; i--; ) {
3775 struct r5dev *dev = &sh->dev[i];
3776 if (test_bit(R5_LOCKED, &dev->flags) &&
3777 (i == sh->pd_idx || i == sh->qd_idx ||
3778 dev->written)) {
3779 pr_debug("Writing block %d\n", i);
3780 set_bit(R5_Wantwrite, &dev->flags);
3781 if (prexor)
3782 continue;
3783 if (!test_bit(R5_Insync, &dev->flags) ||
3784 ((i == sh->pd_idx || i == sh->qd_idx) &&
3785 s.failed == 0))
3786 set_bit(STRIPE_INSYNC, &sh->state);
3787 }
3788 }
3789 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3790 s.dec_preread_active = 1;
3791 }
3792
NeilBrownef5b7c62012-11-22 09:13:36 +11003793 /*
3794 * might be able to return some write requests if the parity blocks
3795 * are safe, or on a failed drive
3796 */
3797 pdev = &sh->dev[sh->pd_idx];
3798 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
3799 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
3800 qdev = &sh->dev[sh->qd_idx];
3801 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
3802 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
3803 || conf->level < 6;
3804
3805 if (s.written &&
3806 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3807 && !test_bit(R5_LOCKED, &pdev->flags)
3808 && (test_bit(R5_UPTODATE, &pdev->flags) ||
3809 test_bit(R5_Discard, &pdev->flags))))) &&
3810 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3811 && !test_bit(R5_LOCKED, &qdev->flags)
3812 && (test_bit(R5_UPTODATE, &qdev->flags) ||
3813 test_bit(R5_Discard, &qdev->flags))))))
3814 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
3815
3816 /* Now we might consider reading some blocks, either to check/generate
3817 * parity, or to satisfy requests
3818 * or to load a block that is being partially written.
3819 */
3820 if (s.to_read || s.non_overwrite
3821 || (conf->level == 6 && s.to_write && s.failed)
3822 || (s.syncing && (s.uptodate + s.compute < disks))
3823 || s.replacing
3824 || s.expanding)
3825 handle_stripe_fill(sh, &s, disks);
3826
NeilBrown84789552011-07-27 11:00:36 +10003827 /* Now to consider new write requests and what else, if anything
3828 * should be read. We do not handle new writes when:
3829 * 1/ A 'write' operation (copy+xor) is already in flight.
3830 * 2/ A 'check' operation is in flight, as it may clobber the parity
3831 * block.
3832 */
3833 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
3834 handle_stripe_dirtying(conf, sh, &s, disks);
3835
3836 /* maybe we need to check and possibly fix the parity for this stripe
3837 * Any reads will already have been scheduled, so we just see if enough
3838 * data is available. The parity check is held off while parity
3839 * dependent operations are in flight.
3840 */
3841 if (sh->check_state ||
3842 (s.syncing && s.locked == 0 &&
3843 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3844 !test_bit(STRIPE_INSYNC, &sh->state))) {
3845 if (conf->level == 6)
3846 handle_parity_checks6(conf, sh, &s, disks);
3847 else
3848 handle_parity_checks5(conf, sh, &s, disks);
3849 }
NeilBrownc5a31002011-07-27 11:00:36 +10003850
NeilBrownf94c0b62013-07-22 12:57:21 +10003851 if ((s.replacing || s.syncing) && s.locked == 0
3852 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
3853 && !test_bit(STRIPE_REPLACED, &sh->state)) {
NeilBrown9a3e1102011-12-23 10:17:53 +11003854 /* Write out to replacement devices where possible */
3855 for (i = 0; i < conf->raid_disks; i++)
NeilBrownf94c0b62013-07-22 12:57:21 +10003856 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
3857 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
NeilBrown9a3e1102011-12-23 10:17:53 +11003858 set_bit(R5_WantReplace, &sh->dev[i].flags);
3859 set_bit(R5_LOCKED, &sh->dev[i].flags);
3860 s.locked++;
3861 }
NeilBrownf94c0b62013-07-22 12:57:21 +10003862 if (s.replacing)
3863 set_bit(STRIPE_INSYNC, &sh->state);
3864 set_bit(STRIPE_REPLACED, &sh->state);
NeilBrown9a3e1102011-12-23 10:17:53 +11003865 }
3866 if ((s.syncing || s.replacing) && s.locked == 0 &&
NeilBrownf94c0b62013-07-22 12:57:21 +10003867 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
NeilBrown9a3e1102011-12-23 10:17:53 +11003868 test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrownc5a31002011-07-27 11:00:36 +10003869 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3870 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003871 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3872 wake_up(&conf->wait_for_overlap);
NeilBrownc5a31002011-07-27 11:00:36 +10003873 }
3874
3875 /* If the failed drives are just a ReadError, then we might need
3876 * to progress the repair/check process
3877 */
3878 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3879 for (i = 0; i < s.failed; i++) {
3880 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3881 if (test_bit(R5_ReadError, &dev->flags)
3882 && !test_bit(R5_LOCKED, &dev->flags)
3883 && test_bit(R5_UPTODATE, &dev->flags)
3884 ) {
3885 if (!test_bit(R5_ReWrite, &dev->flags)) {
3886 set_bit(R5_Wantwrite, &dev->flags);
3887 set_bit(R5_ReWrite, &dev->flags);
3888 set_bit(R5_LOCKED, &dev->flags);
3889 s.locked++;
3890 } else {
3891 /* let's read it back */
3892 set_bit(R5_Wantread, &dev->flags);
3893 set_bit(R5_LOCKED, &dev->flags);
3894 s.locked++;
3895 }
3896 }
3897 }
3898
3899
NeilBrown3687c062011-07-27 11:00:36 +10003900 /* Finish reconstruct operations initiated by the expansion process */
3901 if (sh->reconstruct_state == reconstruct_state_result) {
3902 struct stripe_head *sh_src
3903 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3904 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3905 /* sh cannot be written until sh_src has been read.
3906 * so arrange for sh to be delayed a little
3907 */
3908 set_bit(STRIPE_DELAYED, &sh->state);
3909 set_bit(STRIPE_HANDLE, &sh->state);
3910 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3911 &sh_src->state))
3912 atomic_inc(&conf->preread_active_stripes);
3913 release_stripe(sh_src);
3914 goto finish;
3915 }
3916 if (sh_src)
3917 release_stripe(sh_src);
NeilBrown16a53ec2006-06-26 00:27:38 -07003918
NeilBrown3687c062011-07-27 11:00:36 +10003919 sh->reconstruct_state = reconstruct_state_idle;
3920 clear_bit(STRIPE_EXPANDING, &sh->state);
3921 for (i = conf->raid_disks; i--; ) {
3922 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3923 set_bit(R5_LOCKED, &sh->dev[i].flags);
3924 s.locked++;
3925 }
3926 }
3927
3928 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3929 !sh->reconstruct_state) {
3930 /* Need to write out all blocks after computing parity */
3931 sh->disks = conf->raid_disks;
3932 stripe_set_idx(sh->sector, conf, 0, sh);
3933 schedule_reconstruction(sh, &s, 1, 1);
3934 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3935 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3936 atomic_dec(&conf->reshape_stripes);
3937 wake_up(&conf->wait_for_overlap);
3938 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3939 }
3940
3941 if (s.expanding && s.locked == 0 &&
3942 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3943 handle_stripe_expansion(conf, sh);
3944
3945finish:
Dan Williams6bfe0b42008-04-30 00:52:32 -07003946 /* wait for this device to become unblocked */
NeilBrown5f066c62012-07-03 12:13:29 +10003947 if (unlikely(s.blocked_rdev)) {
3948 if (conf->mddev->external)
3949 md_wait_for_blocked_rdev(s.blocked_rdev,
3950 conf->mddev);
3951 else
3952 /* Internal metadata will immediately
3953 * be written by raid5d, so we don't
3954 * need to wait here.
3955 */
3956 rdev_dec_pending(s.blocked_rdev,
3957 conf->mddev);
3958 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07003959
NeilBrownbc2607f2011-07-28 11:39:22 +10003960 if (s.handle_bad_blocks)
3961 for (i = disks; i--; ) {
NeilBrown3cb03002011-10-11 16:45:26 +11003962 struct md_rdev *rdev;
NeilBrownbc2607f2011-07-28 11:39:22 +10003963 struct r5dev *dev = &sh->dev[i];
3964 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
3965 /* We own a safe reference to the rdev */
3966 rdev = conf->disks[i].rdev;
3967 if (!rdev_set_badblocks(rdev, sh->sector,
3968 STRIPE_SECTORS, 0))
3969 md_error(conf->mddev, rdev);
3970 rdev_dec_pending(rdev, conf->mddev);
3971 }
NeilBrownb84db562011-07-28 11:39:23 +10003972 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
3973 rdev = conf->disks[i].rdev;
3974 rdev_clear_badblocks(rdev, sh->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10003975 STRIPE_SECTORS, 0);
NeilBrownb84db562011-07-28 11:39:23 +10003976 rdev_dec_pending(rdev, conf->mddev);
3977 }
NeilBrown977df362011-12-23 10:17:53 +11003978 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
3979 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11003980 if (!rdev)
3981 /* rdev have been moved down */
3982 rdev = conf->disks[i].rdev;
NeilBrown977df362011-12-23 10:17:53 +11003983 rdev_clear_badblocks(rdev, sh->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10003984 STRIPE_SECTORS, 0);
NeilBrown977df362011-12-23 10:17:53 +11003985 rdev_dec_pending(rdev, conf->mddev);
3986 }
NeilBrownbc2607f2011-07-28 11:39:22 +10003987 }
3988
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003989 if (s.ops_request)
3990 raid_run_ops(sh, s.ops_request);
3991
Dan Williamsf0e43bc2008-06-28 08:31:55 +10003992 ops_run_io(sh, &s);
3993
NeilBrownc5709ef2011-07-26 11:35:20 +10003994 if (s.dec_preread_active) {
NeilBrown729a1862009-12-14 12:49:50 +11003995 /* We delay this until after ops_run_io so that if make_request
Tejun Heoe9c74692010-09-03 11:56:18 +02003996 * is waiting on a flush, it won't continue until the writes
NeilBrown729a1862009-12-14 12:49:50 +11003997 * have actually been submitted.
3998 */
3999 atomic_dec(&conf->preread_active_stripes);
4000 if (atomic_read(&conf->preread_active_stripes) <
4001 IO_THRESHOLD)
4002 md_wakeup_thread(conf->mddev->thread);
4003 }
4004
NeilBrownc5709ef2011-07-26 11:35:20 +10004005 return_io(s.return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07004006
Dan Williams257a4b42011-11-08 16:22:06 +11004007 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07004008}
4009
NeilBrownd1688a62011-10-11 16:49:52 +11004010static void raid5_activate_delayed(struct r5conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011{
4012 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4013 while (!list_empty(&conf->delayed_list)) {
4014 struct list_head *l = conf->delayed_list.next;
4015 struct stripe_head *sh;
4016 sh = list_entry(l, struct stripe_head, lru);
4017 list_del_init(l);
4018 clear_bit(STRIPE_DELAYED, &sh->state);
4019 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4020 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004021 list_add_tail(&sh->lru, &conf->hold_list);
Shaohua Li851c30c2013-08-28 14:30:16 +08004022 raid5_wakeup_stripe_thread(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 }
NeilBrown482c0832011-04-18 18:25:42 +10004024 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025}
4026
Shaohua Li566c09c2013-11-14 15:16:17 +11004027static void activate_bit_delay(struct r5conf *conf,
4028 struct list_head *temp_inactive_list)
NeilBrown72626682005-09-09 16:23:54 -07004029{
4030 /* device_lock is held */
4031 struct list_head head;
4032 list_add(&head, &conf->bitmap_list);
4033 list_del_init(&conf->bitmap_list);
4034 while (!list_empty(&head)) {
4035 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
Shaohua Li566c09c2013-11-14 15:16:17 +11004036 int hash;
NeilBrown72626682005-09-09 16:23:54 -07004037 list_del_init(&sh->lru);
4038 atomic_inc(&sh->count);
Shaohua Li566c09c2013-11-14 15:16:17 +11004039 hash = sh->hash_lock_index;
4040 __release_stripe(conf, sh, &temp_inactive_list[hash]);
NeilBrown72626682005-09-09 16:23:54 -07004041 }
4042}
4043
NeilBrownfd01b882011-10-11 16:47:53 +11004044int md_raid5_congested(struct mddev *mddev, int bits)
NeilBrownf022b2f2006-10-03 01:15:56 -07004045{
NeilBrownd1688a62011-10-11 16:49:52 +11004046 struct r5conf *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07004047
4048 /* No difference between reads and writes. Just check
4049 * how busy the stripe_cache is
4050 */
NeilBrown3fa841d2009-09-23 18:10:29 +10004051
NeilBrownf022b2f2006-10-03 01:15:56 -07004052 if (conf->inactive_blocked)
4053 return 1;
4054 if (conf->quiesce)
4055 return 1;
Shaohua Li4bda5562013-11-14 15:16:17 +11004056 if (atomic_read(&conf->empty_inactive_list_nr))
NeilBrownf022b2f2006-10-03 01:15:56 -07004057 return 1;
4058
4059 return 0;
4060}
NeilBrown11d8a6e2010-07-26 11:57:07 +10004061EXPORT_SYMBOL_GPL(md_raid5_congested);
4062
4063static int raid5_congested(void *data, int bits)
4064{
NeilBrownfd01b882011-10-11 16:47:53 +11004065 struct mddev *mddev = data;
NeilBrown11d8a6e2010-07-26 11:57:07 +10004066
4067 return mddev_congested(mddev, bits) ||
4068 md_raid5_congested(mddev, bits);
4069}
NeilBrownf022b2f2006-10-03 01:15:56 -07004070
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004071/* We want read requests to align with chunks where possible,
4072 * but write requests don't need to.
4073 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004074static int raid5_mergeable_bvec(struct request_queue *q,
4075 struct bvec_merge_data *bvm,
4076 struct bio_vec *biovec)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004077{
NeilBrownfd01b882011-10-11 16:47:53 +11004078 struct mddev *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004079 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004080 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +10004081 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004082 unsigned int bio_sectors = bvm->bi_size >> 9;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004083
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004084 if ((bvm->bi_rw & 1) == WRITE)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004085 return biovec->bv_len; /* always allow writes to be mergeable */
4086
Andre Noll664e7c42009-06-18 08:45:27 +10004087 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4088 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004089 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
4090 if (max < 0) max = 0;
4091 if (max <= biovec->bv_len && bio_sectors == 0)
4092 return biovec->bv_len;
4093 else
4094 return max;
4095}
4096
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004097
NeilBrownfd01b882011-10-11 16:47:53 +11004098static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004099{
4100 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
Andre Noll9d8f0362009-06-18 08:45:01 +10004101 unsigned int chunk_sectors = mddev->chunk_sectors;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08004102 unsigned int bio_sectors = bio_sectors(bio);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004103
Andre Noll664e7c42009-06-18 08:45:27 +10004104 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4105 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004106 return chunk_sectors >=
4107 ((sector & (chunk_sectors - 1)) + bio_sectors);
4108}
4109
4110/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004111 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4112 * later sampled by raid5d.
4113 */
NeilBrownd1688a62011-10-11 16:49:52 +11004114static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004115{
4116 unsigned long flags;
4117
4118 spin_lock_irqsave(&conf->device_lock, flags);
4119
4120 bi->bi_next = conf->retry_read_aligned_list;
4121 conf->retry_read_aligned_list = bi;
4122
4123 spin_unlock_irqrestore(&conf->device_lock, flags);
4124 md_wakeup_thread(conf->mddev->thread);
4125}
4126
4127
NeilBrownd1688a62011-10-11 16:49:52 +11004128static struct bio *remove_bio_from_retry(struct r5conf *conf)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004129{
4130 struct bio *bi;
4131
4132 bi = conf->retry_read_aligned;
4133 if (bi) {
4134 conf->retry_read_aligned = NULL;
4135 return bi;
4136 }
4137 bi = conf->retry_read_aligned_list;
4138 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08004139 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004140 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02004141 /*
4142 * this sets the active strip count to 1 and the processed
4143 * strip count to zero (upper 8 bits)
4144 */
Shaohua Lie7836bd62012-07-19 16:01:31 +10004145 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004146 }
4147
4148 return bi;
4149}
4150
4151
4152/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004153 * The "raid5_align_endio" should check if the read succeeded and if it
4154 * did, call bio_endio on the original bio (having bio_put the new bio
4155 * first).
4156 * If the read failed..
4157 */
NeilBrown6712ecf2007-09-27 12:47:43 +02004158static void raid5_align_endio(struct bio *bi, int error)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004159{
4160 struct bio* raid_bi = bi->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11004161 struct mddev *mddev;
NeilBrownd1688a62011-10-11 16:49:52 +11004162 struct r5conf *conf;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004163 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrown3cb03002011-10-11 16:45:26 +11004164 struct md_rdev *rdev;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004165
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004166 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004167
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004168 rdev = (void*)raid_bi->bi_next;
4169 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11004170 mddev = rdev->mddev;
4171 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004172
4173 rdev_dec_pending(rdev, conf->mddev);
4174
4175 if (!error && uptodate) {
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07004176 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4177 raid_bi, 0);
NeilBrown6712ecf2007-09-27 12:47:43 +02004178 bio_endio(raid_bi, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004179 if (atomic_dec_and_test(&conf->active_aligned_reads))
4180 wake_up(&conf->wait_for_stripe);
NeilBrown6712ecf2007-09-27 12:47:43 +02004181 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004182 }
4183
4184
Dan Williams45b42332007-07-09 11:56:43 -07004185 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004186
4187 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004188}
4189
Neil Brown387bb172007-02-08 14:20:29 -08004190static int bio_fits_rdev(struct bio *bi)
4191{
Jens Axboe165125e2007-07-24 09:28:11 +02004192 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
Neil Brown387bb172007-02-08 14:20:29 -08004193
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08004194 if (bio_sectors(bi) > queue_max_sectors(q))
Neil Brown387bb172007-02-08 14:20:29 -08004195 return 0;
4196 blk_recount_segments(q, bi);
Martin K. Petersen8a783622010-02-26 00:20:39 -05004197 if (bi->bi_phys_segments > queue_max_segments(q))
Neil Brown387bb172007-02-08 14:20:29 -08004198 return 0;
4199
4200 if (q->merge_bvec_fn)
4201 /* it's too hard to apply the merge_bvec_fn at this stage,
4202 * just just give up
4203 */
4204 return 0;
4205
4206 return 1;
4207}
4208
4209
NeilBrownfd01b882011-10-11 16:47:53 +11004210static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004211{
NeilBrownd1688a62011-10-11 16:49:52 +11004212 struct r5conf *conf = mddev->private;
NeilBrown8553fe7ec2009-12-14 12:49:47 +11004213 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004214 struct bio* align_bi;
NeilBrown3cb03002011-10-11 16:45:26 +11004215 struct md_rdev *rdev;
NeilBrown671488c2011-12-23 10:17:52 +11004216 sector_t end_sector;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004217
4218 if (!in_chunk_boundary(mddev, raid_bio)) {
Dan Williams45b42332007-07-09 11:56:43 -07004219 pr_debug("chunk_aligned_read : non aligned\n");
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004220 return 0;
4221 }
4222 /*
NeilBrowna167f662010-10-26 18:31:13 +11004223 * use bio_clone_mddev to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004224 */
NeilBrowna167f662010-10-26 18:31:13 +11004225 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004226 if (!align_bi)
4227 return 0;
4228 /*
4229 * set bi_end_io to a new function, and set bi_private to the
4230 * original bio.
4231 */
4232 align_bi->bi_end_io = raid5_align_endio;
4233 align_bi->bi_private = raid_bio;
4234 /*
4235 * compute position
4236 */
NeilBrown112bf892009-03-31 14:39:38 +11004237 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
4238 0,
NeilBrown911d4ee2009-03-31 14:39:38 +11004239 &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004240
Kent Overstreetf73a1c72012-09-25 15:05:12 -07004241 end_sector = bio_end_sector(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004242 rcu_read_lock();
NeilBrown671488c2011-12-23 10:17:52 +11004243 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4244 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4245 rdev->recovery_offset < end_sector) {
4246 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4247 if (rdev &&
4248 (test_bit(Faulty, &rdev->flags) ||
4249 !(test_bit(In_sync, &rdev->flags) ||
4250 rdev->recovery_offset >= end_sector)))
4251 rdev = NULL;
4252 }
4253 if (rdev) {
NeilBrown31c176e2011-07-28 11:39:22 +10004254 sector_t first_bad;
4255 int bad_sectors;
4256
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004257 atomic_inc(&rdev->nr_pending);
4258 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004259 raid_bio->bi_next = (void*)rdev;
4260 align_bi->bi_bdev = rdev->bdev;
4261 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004262
NeilBrown31c176e2011-07-28 11:39:22 +10004263 if (!bio_fits_rdev(align_bi) ||
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08004264 is_badblock(rdev, align_bi->bi_sector, bio_sectors(align_bi),
NeilBrown31c176e2011-07-28 11:39:22 +10004265 &first_bad, &bad_sectors)) {
4266 /* too big in some way, or has a known bad block */
Neil Brown387bb172007-02-08 14:20:29 -08004267 bio_put(align_bi);
4268 rdev_dec_pending(rdev, mddev);
4269 return 0;
4270 }
4271
majianpeng6c0544e2012-06-12 08:31:10 +08004272 /* No reshape active, so we can trust rdev->data_offset */
4273 align_bi->bi_sector += rdev->data_offset;
4274
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004275 spin_lock_irq(&conf->device_lock);
4276 wait_event_lock_irq(conf->wait_for_stripe,
4277 conf->quiesce == 0,
Lukas Czernereed8c022012-11-30 11:42:40 +01004278 conf->device_lock);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004279 atomic_inc(&conf->active_aligned_reads);
4280 spin_unlock_irq(&conf->device_lock);
4281
Jonathan Brassowe3620a32013-03-07 16:22:01 -06004282 if (mddev->gendisk)
4283 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4284 align_bi, disk_devt(mddev->gendisk),
4285 raid_bio->bi_sector);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004286 generic_make_request(align_bi);
4287 return 1;
4288 } else {
4289 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004290 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004291 return 0;
4292 }
4293}
4294
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004295/* __get_priority_stripe - get the next stripe to process
4296 *
4297 * Full stripe writes are allowed to pass preread active stripes up until
4298 * the bypass_threshold is exceeded. In general the bypass_count
4299 * increments when the handle_list is handled before the hold_list; however, it
4300 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4301 * stripe with in flight i/o. The bypass_count will be reset when the
4302 * head of the hold_list has changed, i.e. the head was promoted to the
4303 * handle_list.
4304 */
Shaohua Li851c30c2013-08-28 14:30:16 +08004305static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004306{
Shaohua Li851c30c2013-08-28 14:30:16 +08004307 struct stripe_head *sh = NULL, *tmp;
4308 struct list_head *handle_list = NULL;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004309 struct r5worker_group *wg = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08004310
4311 if (conf->worker_cnt_per_group == 0) {
4312 handle_list = &conf->handle_list;
4313 } else if (group != ANY_GROUP) {
4314 handle_list = &conf->worker_groups[group].handle_list;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004315 wg = &conf->worker_groups[group];
Shaohua Li851c30c2013-08-28 14:30:16 +08004316 } else {
4317 int i;
4318 for (i = 0; i < conf->group_cnt; i++) {
4319 handle_list = &conf->worker_groups[i].handle_list;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004320 wg = &conf->worker_groups[i];
Shaohua Li851c30c2013-08-28 14:30:16 +08004321 if (!list_empty(handle_list))
4322 break;
4323 }
4324 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004325
4326 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4327 __func__,
Shaohua Li851c30c2013-08-28 14:30:16 +08004328 list_empty(handle_list) ? "empty" : "busy",
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004329 list_empty(&conf->hold_list) ? "empty" : "busy",
4330 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4331
Shaohua Li851c30c2013-08-28 14:30:16 +08004332 if (!list_empty(handle_list)) {
4333 sh = list_entry(handle_list->next, typeof(*sh), lru);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004334
4335 if (list_empty(&conf->hold_list))
4336 conf->bypass_count = 0;
4337 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4338 if (conf->hold_list.next == conf->last_hold)
4339 conf->bypass_count++;
4340 else {
4341 conf->last_hold = conf->hold_list.next;
4342 conf->bypass_count -= conf->bypass_threshold;
4343 if (conf->bypass_count < 0)
4344 conf->bypass_count = 0;
4345 }
4346 }
4347 } else if (!list_empty(&conf->hold_list) &&
4348 ((conf->bypass_threshold &&
4349 conf->bypass_count > conf->bypass_threshold) ||
4350 atomic_read(&conf->pending_full_writes) == 0)) {
Shaohua Li851c30c2013-08-28 14:30:16 +08004351
4352 list_for_each_entry(tmp, &conf->hold_list, lru) {
4353 if (conf->worker_cnt_per_group == 0 ||
4354 group == ANY_GROUP ||
4355 !cpu_online(tmp->cpu) ||
4356 cpu_to_group(tmp->cpu) == group) {
4357 sh = tmp;
4358 break;
4359 }
4360 }
4361
4362 if (sh) {
4363 conf->bypass_count -= conf->bypass_threshold;
4364 if (conf->bypass_count < 0)
4365 conf->bypass_count = 0;
4366 }
Shaohua Libfc90cb2013-08-29 15:40:32 +08004367 wg = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08004368 }
4369
4370 if (!sh)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004371 return NULL;
4372
Shaohua Libfc90cb2013-08-29 15:40:32 +08004373 if (wg) {
4374 wg->stripes_cnt--;
4375 sh->group = NULL;
4376 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004377 list_del_init(&sh->lru);
4378 atomic_inc(&sh->count);
4379 BUG_ON(atomic_read(&sh->count) != 1);
4380 return sh;
4381}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004382
Shaohua Li8811b592012-08-02 08:33:00 +10004383struct raid5_plug_cb {
4384 struct blk_plug_cb cb;
4385 struct list_head list;
Shaohua Li566c09c2013-11-14 15:16:17 +11004386 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Li8811b592012-08-02 08:33:00 +10004387};
4388
4389static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4390{
4391 struct raid5_plug_cb *cb = container_of(
4392 blk_cb, struct raid5_plug_cb, cb);
4393 struct stripe_head *sh;
4394 struct mddev *mddev = cb->cb.data;
4395 struct r5conf *conf = mddev->private;
NeilBrowna9add5d2012-10-31 11:59:09 +11004396 int cnt = 0;
Shaohua Li566c09c2013-11-14 15:16:17 +11004397 int hash;
Shaohua Li8811b592012-08-02 08:33:00 +10004398
4399 if (cb->list.next && !list_empty(&cb->list)) {
4400 spin_lock_irq(&conf->device_lock);
4401 while (!list_empty(&cb->list)) {
4402 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4403 list_del_init(&sh->lru);
4404 /*
4405 * avoid race release_stripe_plug() sees
4406 * STRIPE_ON_UNPLUG_LIST clear but the stripe
4407 * is still in our list
4408 */
4409 smp_mb__before_clear_bit();
4410 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
Shaohua Li773ca822013-08-27 17:50:39 +08004411 /*
4412 * STRIPE_ON_RELEASE_LIST could be set here. In that
4413 * case, the count is always > 1 here
4414 */
Shaohua Li566c09c2013-11-14 15:16:17 +11004415 hash = sh->hash_lock_index;
4416 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
NeilBrowna9add5d2012-10-31 11:59:09 +11004417 cnt++;
Shaohua Li8811b592012-08-02 08:33:00 +10004418 }
4419 spin_unlock_irq(&conf->device_lock);
4420 }
Shaohua Li566c09c2013-11-14 15:16:17 +11004421 release_inactive_stripe_list(conf, cb->temp_inactive_list,
4422 NR_STRIPE_HASH_LOCKS);
Jonathan Brassowe3620a32013-03-07 16:22:01 -06004423 if (mddev->queue)
4424 trace_block_unplug(mddev->queue, cnt, !from_schedule);
Shaohua Li8811b592012-08-02 08:33:00 +10004425 kfree(cb);
4426}
4427
4428static void release_stripe_plug(struct mddev *mddev,
4429 struct stripe_head *sh)
4430{
4431 struct blk_plug_cb *blk_cb = blk_check_plugged(
4432 raid5_unplug, mddev,
4433 sizeof(struct raid5_plug_cb));
4434 struct raid5_plug_cb *cb;
4435
4436 if (!blk_cb) {
4437 release_stripe(sh);
4438 return;
4439 }
4440
4441 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
4442
Shaohua Li566c09c2013-11-14 15:16:17 +11004443 if (cb->list.next == NULL) {
4444 int i;
Shaohua Li8811b592012-08-02 08:33:00 +10004445 INIT_LIST_HEAD(&cb->list);
Shaohua Li566c09c2013-11-14 15:16:17 +11004446 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
4447 INIT_LIST_HEAD(cb->temp_inactive_list + i);
4448 }
Shaohua Li8811b592012-08-02 08:33:00 +10004449
4450 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
4451 list_add_tail(&sh->lru, &cb->list);
4452 else
4453 release_stripe(sh);
4454}
4455
Shaohua Li620125f2012-10-11 13:49:05 +11004456static void make_discard_request(struct mddev *mddev, struct bio *bi)
4457{
4458 struct r5conf *conf = mddev->private;
4459 sector_t logical_sector, last_sector;
4460 struct stripe_head *sh;
4461 int remaining;
4462 int stripe_sectors;
4463
4464 if (mddev->reshape_position != MaxSector)
4465 /* Skip discard while reshape is happening */
4466 return;
4467
4468 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
4469 last_sector = bi->bi_sector + (bi->bi_size>>9);
4470
4471 bi->bi_next = NULL;
4472 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
4473
4474 stripe_sectors = conf->chunk_sectors *
4475 (conf->raid_disks - conf->max_degraded);
4476 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
4477 stripe_sectors);
4478 sector_div(last_sector, stripe_sectors);
4479
4480 logical_sector *= conf->chunk_sectors;
4481 last_sector *= conf->chunk_sectors;
4482
4483 for (; logical_sector < last_sector;
4484 logical_sector += STRIPE_SECTORS) {
4485 DEFINE_WAIT(w);
4486 int d;
4487 again:
4488 sh = get_active_stripe(conf, logical_sector, 0, 0, 0);
4489 prepare_to_wait(&conf->wait_for_overlap, &w,
4490 TASK_UNINTERRUPTIBLE);
NeilBrownf8dfcff2013-03-12 12:18:06 +11004491 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
4492 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4493 release_stripe(sh);
4494 schedule();
4495 goto again;
4496 }
4497 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
Shaohua Li620125f2012-10-11 13:49:05 +11004498 spin_lock_irq(&sh->stripe_lock);
4499 for (d = 0; d < conf->raid_disks; d++) {
4500 if (d == sh->pd_idx || d == sh->qd_idx)
4501 continue;
4502 if (sh->dev[d].towrite || sh->dev[d].toread) {
4503 set_bit(R5_Overlap, &sh->dev[d].flags);
4504 spin_unlock_irq(&sh->stripe_lock);
4505 release_stripe(sh);
4506 schedule();
4507 goto again;
4508 }
4509 }
NeilBrownf8dfcff2013-03-12 12:18:06 +11004510 set_bit(STRIPE_DISCARD, &sh->state);
Shaohua Li620125f2012-10-11 13:49:05 +11004511 finish_wait(&conf->wait_for_overlap, &w);
4512 for (d = 0; d < conf->raid_disks; d++) {
4513 if (d == sh->pd_idx || d == sh->qd_idx)
4514 continue;
4515 sh->dev[d].towrite = bi;
4516 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
4517 raid5_inc_bi_active_stripes(bi);
4518 }
4519 spin_unlock_irq(&sh->stripe_lock);
4520 if (conf->mddev->bitmap) {
4521 for (d = 0;
4522 d < conf->raid_disks - conf->max_degraded;
4523 d++)
4524 bitmap_startwrite(mddev->bitmap,
4525 sh->sector,
4526 STRIPE_SECTORS,
4527 0);
4528 sh->bm_seq = conf->seq_flush + 1;
4529 set_bit(STRIPE_BIT_DELAY, &sh->state);
4530 }
4531
4532 set_bit(STRIPE_HANDLE, &sh->state);
4533 clear_bit(STRIPE_DELAYED, &sh->state);
4534 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4535 atomic_inc(&conf->preread_active_stripes);
4536 release_stripe_plug(mddev, sh);
4537 }
4538
4539 remaining = raid5_dec_bi_active_stripes(bi);
4540 if (remaining == 0) {
4541 md_write_end(mddev);
4542 bio_endio(bi, 0);
4543 }
4544}
4545
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -07004546static void make_request(struct mddev *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004547{
NeilBrownd1688a62011-10-11 16:49:52 +11004548 struct r5conf *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11004549 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004550 sector_t new_sector;
4551 sector_t logical_sector, last_sector;
4552 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01004553 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11004554 int remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004555
Tejun Heoe9c74692010-09-03 11:56:18 +02004556 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
4557 md_flush_request(mddev, bi);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02004558 return;
NeilBrowne5dcdd82005-09-09 16:23:41 -07004559 }
4560
NeilBrown3d310eb2005-06-21 17:17:26 -07004561 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07004562
NeilBrown802ba062006-12-13 00:34:13 -08004563 if (rw == READ &&
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08004564 mddev->reshape_position == MaxSector &&
NeilBrown21a52c62010-04-01 15:02:13 +11004565 chunk_aligned_read(mddev,bi))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02004566 return;
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08004567
Shaohua Li620125f2012-10-11 13:49:05 +11004568 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
4569 make_discard_request(mddev, bi);
4570 return;
4571 }
4572
Linus Torvalds1da177e2005-04-16 15:20:36 -07004573 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07004574 last_sector = bio_end_sector(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004575 bi->bi_next = NULL;
4576 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07004577
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
4579 DEFINE_WAIT(w);
NeilBrownb5663ba2009-03-31 14:39:38 +11004580 int previous;
NeilBrownc46501b2013-08-27 15:52:13 +10004581 int seq;
NeilBrownb578d552006-03-27 01:18:12 -08004582
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004583 retry:
NeilBrownc46501b2013-08-27 15:52:13 +10004584 seq = read_seqcount_begin(&conf->gen_lock);
NeilBrownb5663ba2009-03-31 14:39:38 +11004585 previous = 0;
NeilBrownb578d552006-03-27 01:18:12 -08004586 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11004587 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11004588 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08004589 * 64bit on a 32bit platform, and so it might be
4590 * possible to see a half-updated value
Jesper Juhlaeb878b2011-04-10 18:06:17 +02004591 * Of course reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08004592 * the lock is dropped, so once we get a reference
4593 * to the stripe that we think it is, we will have
4594 * to check again.
4595 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004596 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10004597 if (mddev->reshape_backwards
NeilBrownfef9c612009-03-31 15:16:46 +11004598 ? logical_sector < conf->reshape_progress
4599 : logical_sector >= conf->reshape_progress) {
NeilBrownb5663ba2009-03-31 14:39:38 +11004600 previous = 1;
4601 } else {
NeilBrown2c810cd2012-05-21 09:27:00 +10004602 if (mddev->reshape_backwards
NeilBrownfef9c612009-03-31 15:16:46 +11004603 ? logical_sector < conf->reshape_safe
4604 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08004605 spin_unlock_irq(&conf->device_lock);
4606 schedule();
4607 goto retry;
4608 }
4609 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004610 spin_unlock_irq(&conf->device_lock);
4611 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004612
NeilBrown112bf892009-03-31 14:39:38 +11004613 new_sector = raid5_compute_sector(conf, logical_sector,
4614 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11004615 &dd_idx, NULL);
NeilBrown0c55e022010-05-03 14:09:02 +10004616 pr_debug("raid456: make_request, sector %llu logical %llu\n",
NeilBrownc46501b2013-08-27 15:52:13 +10004617 (unsigned long long)new_sector,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004618 (unsigned long long)logical_sector);
4619
NeilBrownb5663ba2009-03-31 14:39:38 +11004620 sh = get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10004621 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004622 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11004623 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004624 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08004625 * stripe, so we must do the range check again.
4626 * Expansion could still move past after this
4627 * test, but as we are holding a reference to
4628 * 'sh', we know that if that happens,
4629 * STRIPE_EXPANDING will get set and the expansion
4630 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004631 */
4632 int must_retry = 0;
4633 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10004634 if (mddev->reshape_backwards
NeilBrownb0f9ec02009-03-31 15:27:18 +11004635 ? logical_sector >= conf->reshape_progress
4636 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004637 /* mismatch, need to try again */
4638 must_retry = 1;
4639 spin_unlock_irq(&conf->device_lock);
4640 if (must_retry) {
4641 release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07004642 schedule();
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004643 goto retry;
4644 }
4645 }
NeilBrownc46501b2013-08-27 15:52:13 +10004646 if (read_seqcount_retry(&conf->gen_lock, seq)) {
4647 /* Might have got the wrong stripe_head
4648 * by accident
4649 */
4650 release_stripe(sh);
4651 goto retry;
4652 }
NeilBrowne62e58a2009-07-01 13:15:35 +10004653
Namhyung Kimffd96e32011-07-18 17:38:51 +10004654 if (rw == WRITE &&
NeilBrowna5c308d2009-07-01 13:15:35 +10004655 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08004656 logical_sector < mddev->suspend_hi) {
4657 release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10004658 /* As the suspend_* range is controlled by
4659 * userspace, we want an interruptible
4660 * wait.
4661 */
4662 flush_signals(current);
4663 prepare_to_wait(&conf->wait_for_overlap,
4664 &w, TASK_INTERRUPTIBLE);
4665 if (logical_sector >= mddev->suspend_lo &&
4666 logical_sector < mddev->suspend_hi)
4667 schedule();
NeilBrowne464eaf2006-03-27 01:18:14 -08004668 goto retry;
4669 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004670
4671 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
Namhyung Kimffd96e32011-07-18 17:38:51 +10004672 !add_stripe_bio(sh, bi, dd_idx, rw)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004673 /* Stripe is busy expanding or
4674 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07004675 * and wait a while
4676 */
NeilBrown482c0832011-04-18 18:25:42 +10004677 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004678 release_stripe(sh);
4679 schedule();
4680 goto retry;
4681 }
4682 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown6ed30032008-02-06 01:40:00 -08004683 set_bit(STRIPE_HANDLE, &sh->state);
4684 clear_bit(STRIPE_DELAYED, &sh->state);
NeilBrowna852d7b2012-09-19 12:48:30 +10004685 if ((bi->bi_rw & REQ_SYNC) &&
NeilBrown729a1862009-12-14 12:49:50 +11004686 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4687 atomic_inc(&conf->preread_active_stripes);
Shaohua Li8811b592012-08-02 08:33:00 +10004688 release_stripe_plug(mddev, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004689 } else {
4690 /* cannot get stripe for read-ahead, just give-up */
4691 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4692 finish_wait(&conf->wait_for_overlap, &w);
4693 break;
4694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004695 }
NeilBrown7c13edc2011-04-18 18:25:43 +10004696
Shaohua Lie7836bd62012-07-19 16:01:31 +10004697 remaining = raid5_dec_bi_active_stripes(bi);
NeilBrownf6344752006-03-27 01:18:17 -08004698 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699
NeilBrown16a53ec2006-06-26 00:27:38 -07004700 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004701 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02004702
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07004703 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
4704 bi, 0);
Neil Brown0e13fe232008-06-28 08:31:20 +10004705 bio_endio(bi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004707}
4708
NeilBrownfd01b882011-10-11 16:47:53 +11004709static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
Dan Williamsb522adc2009-03-31 15:00:31 +11004710
NeilBrownfd01b882011-10-11 16:47:53 +11004711static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004712{
NeilBrown52c03292006-06-26 00:27:43 -07004713 /* reshaping is quite different to recovery/resync so it is
4714 * handled quite separately ... here.
4715 *
4716 * On each call to sync_request, we gather one chunk worth of
4717 * destination stripes and flag them as expanding.
4718 * Then we find all the source stripes and request reads.
4719 * As the reads complete, handle_stripe will copy the data
4720 * into the destination stripe and release that stripe.
4721 */
NeilBrownd1688a62011-10-11 16:49:52 +11004722 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004723 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08004724 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08004725 int raid_disks = conf->previous_raid_disks;
4726 int data_disks = raid_disks - conf->max_degraded;
4727 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07004728 int i;
4729 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11004730 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11004731 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11004732 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11004733 struct list_head stripes;
NeilBrown52c03292006-06-26 00:27:43 -07004734
NeilBrownfef9c612009-03-31 15:16:46 +11004735 if (sector_nr == 0) {
4736 /* If restarting in the middle, skip the initial sectors */
NeilBrown2c810cd2012-05-21 09:27:00 +10004737 if (mddev->reshape_backwards &&
NeilBrownfef9c612009-03-31 15:16:46 +11004738 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4739 sector_nr = raid5_size(mddev, 0, 0)
4740 - conf->reshape_progress;
NeilBrown2c810cd2012-05-21 09:27:00 +10004741 } else if (!mddev->reshape_backwards &&
NeilBrownfef9c612009-03-31 15:16:46 +11004742 conf->reshape_progress > 0)
4743 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004744 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004745 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11004746 mddev->curr_resync_completed = sector_nr;
4747 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11004748 *skipped = 1;
4749 return sector_nr;
4750 }
NeilBrown52c03292006-06-26 00:27:43 -07004751 }
4752
NeilBrown7a661382009-03-31 15:21:40 +11004753 /* We need to process a full chunk at a time.
4754 * If old and new chunk sizes differ, we need to process the
4755 * largest of these
4756 */
Andre Noll664e7c42009-06-18 08:45:27 +10004757 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4758 reshape_sectors = mddev->new_chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004759 else
Andre Noll9d8f0362009-06-18 08:45:01 +10004760 reshape_sectors = mddev->chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004761
NeilBrownb5254dd2012-05-21 09:27:01 +10004762 /* We update the metadata at least every 10 seconds, or when
4763 * the data about to be copied would over-write the source of
4764 * the data at the front of the range. i.e. one new_stripe
4765 * along from reshape_progress new_maps to after where
4766 * reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07004767 */
NeilBrownfef9c612009-03-31 15:16:46 +11004768 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004769 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11004770 readpos = conf->reshape_progress;
4771 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004772 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08004773 sector_div(safepos, data_disks);
NeilBrown2c810cd2012-05-21 09:27:00 +10004774 if (mddev->reshape_backwards) {
NeilBrowned37d832009-05-27 21:39:05 +10004775 writepos -= min_t(sector_t, reshape_sectors, writepos);
NeilBrownc8f517c2009-03-31 15:28:40 +11004776 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004777 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11004778 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004779 writepos += reshape_sectors;
NeilBrowned37d832009-05-27 21:39:05 +10004780 readpos -= min_t(sector_t, reshape_sectors, readpos);
4781 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11004782 }
NeilBrown52c03292006-06-26 00:27:43 -07004783
NeilBrownb5254dd2012-05-21 09:27:01 +10004784 /* Having calculated the 'writepos' possibly use it
4785 * to set 'stripe_addr' which is where we will write to.
4786 */
4787 if (mddev->reshape_backwards) {
4788 BUG_ON(conf->reshape_progress == 0);
4789 stripe_addr = writepos;
4790 BUG_ON((mddev->dev_sectors &
4791 ~((sector_t)reshape_sectors - 1))
4792 - reshape_sectors - stripe_addr
4793 != sector_nr);
4794 } else {
4795 BUG_ON(writepos != sector_nr + reshape_sectors);
4796 stripe_addr = sector_nr;
4797 }
4798
NeilBrownc8f517c2009-03-31 15:28:40 +11004799 /* 'writepos' is the most advanced device address we might write.
4800 * 'readpos' is the least advanced device address we might read.
4801 * 'safepos' is the least address recorded in the metadata as having
4802 * been reshaped.
NeilBrownb5254dd2012-05-21 09:27:01 +10004803 * If there is a min_offset_diff, these are adjusted either by
4804 * increasing the safepos/readpos if diff is negative, or
4805 * increasing writepos if diff is positive.
4806 * If 'readpos' is then behind 'writepos', there is no way that we can
NeilBrownc8f517c2009-03-31 15:28:40 +11004807 * ensure safety in the face of a crash - that must be done by userspace
4808 * making a backup of the data. So in that case there is no particular
4809 * rush to update metadata.
4810 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4811 * update the metadata to advance 'safepos' to match 'readpos' so that
4812 * we can be safe in the event of a crash.
4813 * So we insist on updating metadata if safepos is behind writepos and
4814 * readpos is beyond writepos.
4815 * In any case, update the metadata every 10 seconds.
4816 * Maybe that number should be configurable, but I'm not sure it is
4817 * worth it.... maybe it could be a multiple of safemode_delay???
4818 */
NeilBrownb5254dd2012-05-21 09:27:01 +10004819 if (conf->min_offset_diff < 0) {
4820 safepos += -conf->min_offset_diff;
4821 readpos += -conf->min_offset_diff;
4822 } else
4823 writepos += conf->min_offset_diff;
4824
NeilBrown2c810cd2012-05-21 09:27:00 +10004825 if ((mddev->reshape_backwards
NeilBrownc8f517c2009-03-31 15:28:40 +11004826 ? (safepos > writepos && readpos < writepos)
4827 : (safepos < writepos && readpos > writepos)) ||
4828 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07004829 /* Cannot proceed until we've updated the superblock... */
4830 wait_event(conf->wait_for_overlap,
NeilBrownc91abf52013-11-19 12:02:01 +11004831 atomic_read(&conf->reshape_stripes)==0
4832 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4833 if (atomic_read(&conf->reshape_stripes) != 0)
4834 return 0;
NeilBrownfef9c612009-03-31 15:16:46 +11004835 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004836 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004837 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07004838 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07004839 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07004840 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrownc91abf52013-11-19 12:02:01 +11004841 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4842 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4843 return 0;
NeilBrown52c03292006-06-26 00:27:43 -07004844 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004845 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07004846 spin_unlock_irq(&conf->device_lock);
4847 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004848 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07004849 }
4850
NeilBrownab69ae12009-03-31 15:26:47 +11004851 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11004852 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07004853 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10004854 int skipped_disk = 0;
NeilBrowna8c906c2009-06-09 14:39:59 +10004855 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004856 set_bit(STRIPE_EXPANDING, &sh->state);
4857 atomic_inc(&conf->reshape_stripes);
4858 /* If any of this stripe is beyond the end of the old
4859 * array, then we need to zero those blocks
4860 */
4861 for (j=sh->disks; j--;) {
4862 sector_t s;
4863 if (j == sh->pd_idx)
4864 continue;
NeilBrownf4168852007-02-28 20:11:53 -08004865 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11004866 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08004867 continue;
NeilBrown784052e2009-03-31 15:19:07 +11004868 s = compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11004869 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10004870 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07004871 continue;
4872 }
4873 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4874 set_bit(R5_Expanded, &sh->dev[j].flags);
4875 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4876 }
NeilBrowna9f326e2009-09-23 18:06:41 +10004877 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07004878 set_bit(STRIPE_EXPAND_READY, &sh->state);
4879 set_bit(STRIPE_HANDLE, &sh->state);
4880 }
NeilBrownab69ae12009-03-31 15:26:47 +11004881 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07004882 }
4883 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10004884 if (mddev->reshape_backwards)
NeilBrown7a661382009-03-31 15:21:40 +11004885 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11004886 else
NeilBrown7a661382009-03-31 15:21:40 +11004887 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07004888 spin_unlock_irq(&conf->device_lock);
4889 /* Ok, those stripe are ready. We can start scheduling
4890 * reads on the source stripes.
4891 * The source stripes are determined by mapping the first and last
4892 * block on the destination stripes.
4893 */
NeilBrown52c03292006-06-26 00:27:43 -07004894 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11004895 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11004896 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07004897 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10004898 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10004899 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11004900 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11004901 if (last_sector >= mddev->dev_sectors)
4902 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07004903 while (first_sector <= last_sector) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004904 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004905 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4906 set_bit(STRIPE_HANDLE, &sh->state);
4907 release_stripe(sh);
4908 first_sector += STRIPE_SECTORS;
4909 }
NeilBrownab69ae12009-03-31 15:26:47 +11004910 /* Now that the sources are clearly marked, we can release
4911 * the destination stripes
4912 */
4913 while (!list_empty(&stripes)) {
4914 sh = list_entry(stripes.next, struct stripe_head, lru);
4915 list_del_init(&sh->lru);
4916 release_stripe(sh);
4917 }
NeilBrownc6207272008-02-06 01:39:52 -08004918 /* If this takes us to the resync_max point where we have to pause,
4919 * then we need to write out the superblock.
4920 */
NeilBrown7a661382009-03-31 15:21:40 +11004921 sector_nr += reshape_sectors;
NeilBrownc03f6a12009-04-17 11:06:30 +10004922 if ((sector_nr - mddev->curr_resync_completed) * 2
4923 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08004924 /* Cannot proceed until we've updated the superblock... */
4925 wait_event(conf->wait_for_overlap,
NeilBrownc91abf52013-11-19 12:02:01 +11004926 atomic_read(&conf->reshape_stripes) == 0
4927 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4928 if (atomic_read(&conf->reshape_stripes) != 0)
4929 goto ret;
NeilBrownfef9c612009-03-31 15:16:46 +11004930 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004931 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004932 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08004933 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4934 md_wakeup_thread(mddev->thread);
4935 wait_event(mddev->sb_wait,
4936 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
NeilBrownc91abf52013-11-19 12:02:01 +11004937 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4938 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4939 goto ret;
NeilBrownc6207272008-02-06 01:39:52 -08004940 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004941 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08004942 spin_unlock_irq(&conf->device_lock);
4943 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004944 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08004945 }
NeilBrownc91abf52013-11-19 12:02:01 +11004946ret:
NeilBrown7a661382009-03-31 15:21:40 +11004947 return reshape_sectors;
NeilBrown52c03292006-06-26 00:27:43 -07004948}
4949
4950/* FIXME go_faster isn't used */
NeilBrownfd01b882011-10-11 16:47:53 +11004951static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
NeilBrown52c03292006-06-26 00:27:43 -07004952{
NeilBrownd1688a62011-10-11 16:49:52 +11004953 struct r5conf *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07004954 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11004955 sector_t max_sector = mddev->dev_sectors;
NeilBrown57dab0b2010-10-19 10:03:39 +11004956 sector_t sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07004957 int still_degraded = 0;
4958 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004959
NeilBrown72626682005-09-09 16:23:54 -07004960 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004961 /* just being told to finish up .. nothing much to do */
NeilBrowncea9c222009-03-31 15:15:05 +11004962
NeilBrown29269552006-03-27 01:18:10 -08004963 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4964 end_reshape(conf);
4965 return 0;
4966 }
NeilBrown72626682005-09-09 16:23:54 -07004967
4968 if (mddev->curr_resync < max_sector) /* aborted */
4969 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4970 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07004971 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07004972 conf->fullsync = 0;
4973 bitmap_close_sync(mddev->bitmap);
4974
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975 return 0;
4976 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08004977
NeilBrown64bd6602009-08-03 10:59:58 +10004978 /* Allow raid5_quiesce to complete */
4979 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4980
NeilBrown52c03292006-06-26 00:27:43 -07004981 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4982 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08004983
NeilBrownc6207272008-02-06 01:39:52 -08004984 /* No need to check resync_max as we never do more than one
4985 * stripe, and as resync_max will always be on a chunk boundary,
4986 * if the check in md_do_sync didn't fire, there is no chance
4987 * of overstepping resync_max here
4988 */
4989
NeilBrown16a53ec2006-06-26 00:27:38 -07004990 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07004991 * to resync, then assert that we are finished, because there is
4992 * nothing we can do.
4993 */
NeilBrown3285edf2006-06-26 00:27:55 -07004994 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07004995 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004996 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07004997 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998 return rv;
4999 }
majianpeng6f608042013-04-24 11:42:41 +10005000 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5001 !conf->fullsync &&
5002 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5003 sync_blocks >= STRIPE_SECTORS) {
NeilBrown72626682005-09-09 16:23:54 -07005004 /* we can skip this block, and probably more */
5005 sync_blocks /= STRIPE_SECTORS;
5006 *skipped = 1;
5007 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005009
NeilBrownb47490c2008-02-06 01:39:50 -08005010 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
5011
NeilBrowna8c906c2009-06-09 14:39:59 +10005012 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013 if (sh == NULL) {
NeilBrowna8c906c2009-06-09 14:39:59 +10005014 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005015 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07005016 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07005017 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08005018 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005019 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005020 /* Need to check if array will still be degraded after recovery/resync
5021 * We don't need to check the 'failed' flag as when that gets set,
5022 * recovery aborts.
5023 */
NeilBrownf001a702009-06-09 14:30:31 +10005024 for (i = 0; i < conf->raid_disks; i++)
NeilBrown16a53ec2006-06-26 00:27:38 -07005025 if (conf->disks[i].rdev == NULL)
5026 still_degraded = 1;
5027
5028 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5029
NeilBrown83206d62011-07-26 11:19:49 +10005030 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005031
NeilBrown14425772009-10-16 15:55:25 +11005032 handle_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005033 release_stripe(sh);
5034
5035 return STRIPE_SECTORS;
5036}
5037
NeilBrownd1688a62011-10-11 16:49:52 +11005038static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005039{
5040 /* We may not be able to submit a whole bio at once as there
5041 * may not be enough stripe_heads available.
5042 * We cannot pre-allocate enough stripe_heads as we may need
5043 * more than exist in the cache (if we allow ever large chunks).
5044 * So we do one stripe head at a time and record in
5045 * ->bi_hw_segments how many have been done.
5046 *
5047 * We *know* that this entire raid_bio is in one chunk, so
5048 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5049 */
5050 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11005051 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005052 sector_t sector, logical_sector, last_sector;
5053 int scnt = 0;
5054 int remaining;
5055 int handled = 0;
5056
5057 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11005058 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11005059 0, &dd_idx, NULL);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07005060 last_sector = bio_end_sector(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005061
5062 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08005063 logical_sector += STRIPE_SECTORS,
5064 sector += STRIPE_SECTORS,
5065 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005066
Shaohua Lie7836bd62012-07-19 16:01:31 +10005067 if (scnt < raid5_bi_processed_stripes(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005068 /* already done this stripe */
5069 continue;
5070
NeilBrowna8c906c2009-06-09 14:39:59 +10005071 sh = get_active_stripe(conf, sector, 0, 1, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005072
5073 if (!sh) {
5074 /* failed to get a stripe - must wait */
Shaohua Lie7836bd62012-07-19 16:01:31 +10005075 raid5_set_bi_processed_stripes(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005076 conf->retry_read_aligned = raid_bio;
5077 return handled;
5078 }
5079
Neil Brown387bb172007-02-08 14:20:29 -08005080 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
5081 release_stripe(sh);
Shaohua Lie7836bd62012-07-19 16:01:31 +10005082 raid5_set_bi_processed_stripes(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08005083 conf->retry_read_aligned = raid_bio;
5084 return handled;
5085 }
5086
majianpeng3f9e7c12012-07-31 10:04:21 +10005087 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
Dan Williams36d1c642009-07-14 11:48:22 -07005088 handle_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005089 release_stripe(sh);
5090 handled++;
5091 }
Shaohua Lie7836bd62012-07-19 16:01:31 +10005092 remaining = raid5_dec_bi_active_stripes(raid_bio);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07005093 if (remaining == 0) {
5094 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5095 raid_bio, 0);
Neil Brown0e13fe232008-06-28 08:31:20 +10005096 bio_endio(raid_bio, 0);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07005097 }
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005098 if (atomic_dec_and_test(&conf->active_aligned_reads))
5099 wake_up(&conf->wait_for_stripe);
5100 return handled;
5101}
5102
Shaohua Libfc90cb2013-08-29 15:40:32 +08005103static int handle_active_stripes(struct r5conf *conf, int group,
Shaohua Li566c09c2013-11-14 15:16:17 +11005104 struct r5worker *worker,
5105 struct list_head *temp_inactive_list)
Shaohua Li46a06402012-08-02 08:33:15 +10005106{
5107 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
Shaohua Li566c09c2013-11-14 15:16:17 +11005108 int i, batch_size = 0, hash;
5109 bool release_inactive = false;
Shaohua Li46a06402012-08-02 08:33:15 +10005110
5111 while (batch_size < MAX_STRIPE_BATCH &&
Shaohua Li851c30c2013-08-28 14:30:16 +08005112 (sh = __get_priority_stripe(conf, group)) != NULL)
Shaohua Li46a06402012-08-02 08:33:15 +10005113 batch[batch_size++] = sh;
5114
Shaohua Li566c09c2013-11-14 15:16:17 +11005115 if (batch_size == 0) {
5116 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5117 if (!list_empty(temp_inactive_list + i))
5118 break;
5119 if (i == NR_STRIPE_HASH_LOCKS)
5120 return batch_size;
5121 release_inactive = true;
5122 }
Shaohua Li46a06402012-08-02 08:33:15 +10005123 spin_unlock_irq(&conf->device_lock);
5124
Shaohua Li566c09c2013-11-14 15:16:17 +11005125 release_inactive_stripe_list(conf, temp_inactive_list,
5126 NR_STRIPE_HASH_LOCKS);
5127
5128 if (release_inactive) {
5129 spin_lock_irq(&conf->device_lock);
5130 return 0;
5131 }
5132
Shaohua Li46a06402012-08-02 08:33:15 +10005133 for (i = 0; i < batch_size; i++)
5134 handle_stripe(batch[i]);
5135
5136 cond_resched();
5137
5138 spin_lock_irq(&conf->device_lock);
Shaohua Li566c09c2013-11-14 15:16:17 +11005139 for (i = 0; i < batch_size; i++) {
5140 hash = batch[i]->hash_lock_index;
5141 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5142 }
Shaohua Li46a06402012-08-02 08:33:15 +10005143 return batch_size;
5144}
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005145
Shaohua Li851c30c2013-08-28 14:30:16 +08005146static void raid5_do_work(struct work_struct *work)
5147{
5148 struct r5worker *worker = container_of(work, struct r5worker, work);
5149 struct r5worker_group *group = worker->group;
5150 struct r5conf *conf = group->conf;
5151 int group_id = group - conf->worker_groups;
5152 int handled;
5153 struct blk_plug plug;
5154
5155 pr_debug("+++ raid5worker active\n");
5156
5157 blk_start_plug(&plug);
5158 handled = 0;
5159 spin_lock_irq(&conf->device_lock);
5160 while (1) {
5161 int batch_size, released;
5162
Shaohua Li566c09c2013-11-14 15:16:17 +11005163 released = release_stripe_list(conf, worker->temp_inactive_list);
Shaohua Li851c30c2013-08-28 14:30:16 +08005164
Shaohua Li566c09c2013-11-14 15:16:17 +11005165 batch_size = handle_active_stripes(conf, group_id, worker,
5166 worker->temp_inactive_list);
Shaohua Libfc90cb2013-08-29 15:40:32 +08005167 worker->working = false;
Shaohua Li851c30c2013-08-28 14:30:16 +08005168 if (!batch_size && !released)
5169 break;
5170 handled += batch_size;
5171 }
5172 pr_debug("%d stripes handled\n", handled);
5173
5174 spin_unlock_irq(&conf->device_lock);
5175 blk_finish_plug(&plug);
5176
5177 pr_debug("--- raid5worker inactive\n");
5178}
5179
Linus Torvalds1da177e2005-04-16 15:20:36 -07005180/*
5181 * This is our raid5 kernel thread.
5182 *
5183 * We scan the hash table for stripes which can be handled now.
5184 * During the scan, completed stripes are saved for us by the interrupt
5185 * handler, so that they will not have to wait for our next wakeup.
5186 */
Shaohua Li4ed87312012-10-11 13:34:00 +11005187static void raid5d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005188{
Shaohua Li4ed87312012-10-11 13:34:00 +11005189 struct mddev *mddev = thread->mddev;
NeilBrownd1688a62011-10-11 16:49:52 +11005190 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005191 int handled;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005192 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005193
Dan Williams45b42332007-07-09 11:56:43 -07005194 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005195
5196 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005197
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005198 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005199 handled = 0;
5200 spin_lock_irq(&conf->device_lock);
5201 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005202 struct bio *bio;
Shaohua Li773ca822013-08-27 17:50:39 +08005203 int batch_size, released;
5204
Shaohua Li566c09c2013-11-14 15:16:17 +11005205 released = release_stripe_list(conf, conf->temp_inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005206
NeilBrown0021b7b2012-07-31 09:08:14 +02005207 if (
NeilBrown7c13edc2011-04-18 18:25:43 +10005208 !list_empty(&conf->bitmap_list)) {
5209 /* Now is a good time to flush some bitmap updates */
5210 conf->seq_flush++;
NeilBrown700e4322005-11-28 13:44:10 -08005211 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07005212 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08005213 spin_lock_irq(&conf->device_lock);
NeilBrown7c13edc2011-04-18 18:25:43 +10005214 conf->seq_write = conf->seq_flush;
Shaohua Li566c09c2013-11-14 15:16:17 +11005215 activate_bit_delay(conf, conf->temp_inactive_list);
NeilBrown72626682005-09-09 16:23:54 -07005216 }
NeilBrown0021b7b2012-07-31 09:08:14 +02005217 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07005218
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005219 while ((bio = remove_bio_from_retry(conf))) {
5220 int ok;
5221 spin_unlock_irq(&conf->device_lock);
5222 ok = retry_aligned_read(conf, bio);
5223 spin_lock_irq(&conf->device_lock);
5224 if (!ok)
5225 break;
5226 handled++;
5227 }
5228
Shaohua Li566c09c2013-11-14 15:16:17 +11005229 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
5230 conf->temp_inactive_list);
Shaohua Li773ca822013-08-27 17:50:39 +08005231 if (!batch_size && !released)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005232 break;
Shaohua Li46a06402012-08-02 08:33:15 +10005233 handled += batch_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005234
Shaohua Li46a06402012-08-02 08:33:15 +10005235 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
5236 spin_unlock_irq(&conf->device_lock);
NeilBrownde393cd2011-07-28 11:31:48 +10005237 md_check_recovery(mddev);
Shaohua Li46a06402012-08-02 08:33:15 +10005238 spin_lock_irq(&conf->device_lock);
5239 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005240 }
Dan Williams45b42332007-07-09 11:56:43 -07005241 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005242
5243 spin_unlock_irq(&conf->device_lock);
5244
Dan Williamsc9f21aa2008-07-23 12:05:51 -07005245 async_tx_issue_pending_all();
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005246 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005247
Dan Williams45b42332007-07-09 11:56:43 -07005248 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005249}
5250
NeilBrown3f294f42005-11-08 21:39:25 -08005251static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005252raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08005253{
NeilBrownd1688a62011-10-11 16:49:52 +11005254 struct r5conf *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08005255 if (conf)
5256 return sprintf(page, "%d\n", conf->max_nr_stripes);
5257 else
5258 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08005259}
5260
NeilBrownc41d4ac2010-06-01 19:37:24 +10005261int
NeilBrownfd01b882011-10-11 16:47:53 +11005262raid5_set_cache_size(struct mddev *mddev, int size)
NeilBrownc41d4ac2010-06-01 19:37:24 +10005263{
NeilBrownd1688a62011-10-11 16:49:52 +11005264 struct r5conf *conf = mddev->private;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005265 int err;
Shaohua Li566c09c2013-11-14 15:16:17 +11005266 int hash;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005267
5268 if (size <= 16 || size > 32768)
5269 return -EINVAL;
Shaohua Li566c09c2013-11-14 15:16:17 +11005270 hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005271 while (size < conf->max_nr_stripes) {
Shaohua Li566c09c2013-11-14 15:16:17 +11005272 if (drop_one_stripe(conf, hash))
NeilBrownc41d4ac2010-06-01 19:37:24 +10005273 conf->max_nr_stripes--;
5274 else
5275 break;
Shaohua Li566c09c2013-11-14 15:16:17 +11005276 hash--;
5277 if (hash < 0)
5278 hash = NR_STRIPE_HASH_LOCKS - 1;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005279 }
5280 err = md_allow_write(mddev);
5281 if (err)
5282 return err;
Shaohua Li566c09c2013-11-14 15:16:17 +11005283 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005284 while (size > conf->max_nr_stripes) {
Shaohua Li566c09c2013-11-14 15:16:17 +11005285 if (grow_one_stripe(conf, hash))
NeilBrownc41d4ac2010-06-01 19:37:24 +10005286 conf->max_nr_stripes++;
5287 else break;
Shaohua Li566c09c2013-11-14 15:16:17 +11005288 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005289 }
5290 return 0;
5291}
5292EXPORT_SYMBOL(raid5_set_cache_size);
5293
NeilBrown3f294f42005-11-08 21:39:25 -08005294static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005295raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08005296{
NeilBrownd1688a62011-10-11 16:49:52 +11005297 struct r5conf *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07005298 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07005299 int err;
5300
NeilBrown3f294f42005-11-08 21:39:25 -08005301 if (len >= PAGE_SIZE)
5302 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08005303 if (!conf)
5304 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08005305
Jingoo Hanb29bebd2013-06-01 16:15:16 +09005306 if (kstrtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08005307 return -EINVAL;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005308 err = raid5_set_cache_size(mddev, new);
Dan Williamsb5470dc2008-06-27 21:44:04 -07005309 if (err)
5310 return err;
NeilBrown3f294f42005-11-08 21:39:25 -08005311 return len;
5312}
NeilBrown007583c2005-11-08 21:39:30 -08005313
NeilBrown96de1e62005-11-08 21:39:39 -08005314static struct md_sysfs_entry
5315raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
5316 raid5_show_stripe_cache_size,
5317 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08005318
5319static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005320raid5_show_preread_threshold(struct mddev *mddev, char *page)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005321{
NeilBrownd1688a62011-10-11 16:49:52 +11005322 struct r5conf *conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005323 if (conf)
5324 return sprintf(page, "%d\n", conf->bypass_threshold);
5325 else
5326 return 0;
5327}
5328
5329static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005330raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005331{
NeilBrownd1688a62011-10-11 16:49:52 +11005332 struct r5conf *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07005333 unsigned long new;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005334 if (len >= PAGE_SIZE)
5335 return -EINVAL;
5336 if (!conf)
5337 return -ENODEV;
5338
Jingoo Hanb29bebd2013-06-01 16:15:16 +09005339 if (kstrtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005340 return -EINVAL;
Dan Williams4ef197d82008-04-28 02:15:54 -07005341 if (new > conf->max_nr_stripes)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005342 return -EINVAL;
5343 conf->bypass_threshold = new;
5344 return len;
5345}
5346
5347static struct md_sysfs_entry
5348raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
5349 S_IRUGO | S_IWUSR,
5350 raid5_show_preread_threshold,
5351 raid5_store_preread_threshold);
5352
5353static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005354stripe_cache_active_show(struct mddev *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08005355{
NeilBrownd1688a62011-10-11 16:49:52 +11005356 struct r5conf *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08005357 if (conf)
5358 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
5359 else
5360 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08005361}
5362
NeilBrown96de1e62005-11-08 21:39:39 -08005363static struct md_sysfs_entry
5364raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08005365
Shaohua Lib7214202013-08-27 17:50:42 +08005366static ssize_t
5367raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
5368{
5369 struct r5conf *conf = mddev->private;
5370 if (conf)
5371 return sprintf(page, "%d\n", conf->worker_cnt_per_group);
5372 else
5373 return 0;
5374}
5375
majianpeng60aaf932013-11-14 15:16:20 +11005376static int alloc_thread_groups(struct r5conf *conf, int cnt,
5377 int *group_cnt,
5378 int *worker_cnt_per_group,
5379 struct r5worker_group **worker_groups);
Shaohua Lib7214202013-08-27 17:50:42 +08005380static ssize_t
5381raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
5382{
5383 struct r5conf *conf = mddev->private;
5384 unsigned long new;
5385 int err;
majianpeng60aaf932013-11-14 15:16:20 +11005386 struct r5worker_group *new_groups, *old_groups;
5387 int group_cnt, worker_cnt_per_group;
Shaohua Lib7214202013-08-27 17:50:42 +08005388
5389 if (len >= PAGE_SIZE)
5390 return -EINVAL;
5391 if (!conf)
5392 return -ENODEV;
5393
5394 if (kstrtoul(page, 10, &new))
5395 return -EINVAL;
5396
5397 if (new == conf->worker_cnt_per_group)
5398 return len;
5399
5400 mddev_suspend(mddev);
5401
5402 old_groups = conf->worker_groups;
majianpengd206dcf2013-11-14 15:16:19 +11005403 if (old_groups)
5404 flush_workqueue(raid5_wq);
Shaohua Lib7214202013-08-27 17:50:42 +08005405
majianpeng60aaf932013-11-14 15:16:20 +11005406 err = alloc_thread_groups(conf, new,
5407 &group_cnt, &worker_cnt_per_group,
5408 &new_groups);
5409 if (!err) {
5410 spin_lock_irq(&conf->device_lock);
5411 conf->group_cnt = group_cnt;
5412 conf->worker_cnt_per_group = worker_cnt_per_group;
5413 conf->worker_groups = new_groups;
5414 spin_unlock_irq(&conf->device_lock);
5415
Shaohua Lib7214202013-08-27 17:50:42 +08005416 if (old_groups)
5417 kfree(old_groups[0].workers);
5418 kfree(old_groups);
5419 }
5420
5421 mddev_resume(mddev);
5422
5423 if (err)
5424 return err;
5425 return len;
5426}
5427
5428static struct md_sysfs_entry
5429raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
5430 raid5_show_group_thread_cnt,
5431 raid5_store_group_thread_cnt);
5432
NeilBrown007583c2005-11-08 21:39:30 -08005433static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08005434 &raid5_stripecache_size.attr,
5435 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005436 &raid5_preread_bypass_threshold.attr,
Shaohua Lib7214202013-08-27 17:50:42 +08005437 &raid5_group_thread_cnt.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08005438 NULL,
5439};
NeilBrown007583c2005-11-08 21:39:30 -08005440static struct attribute_group raid5_attrs_group = {
5441 .name = NULL,
5442 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08005443};
5444
majianpeng60aaf932013-11-14 15:16:20 +11005445static int alloc_thread_groups(struct r5conf *conf, int cnt,
5446 int *group_cnt,
5447 int *worker_cnt_per_group,
5448 struct r5worker_group **worker_groups)
Shaohua Li851c30c2013-08-28 14:30:16 +08005449{
Shaohua Li566c09c2013-11-14 15:16:17 +11005450 int i, j, k;
Shaohua Li851c30c2013-08-28 14:30:16 +08005451 ssize_t size;
5452 struct r5worker *workers;
5453
majianpeng60aaf932013-11-14 15:16:20 +11005454 *worker_cnt_per_group = cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +08005455 if (cnt == 0) {
majianpeng60aaf932013-11-14 15:16:20 +11005456 *group_cnt = 0;
5457 *worker_groups = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08005458 return 0;
5459 }
majianpeng60aaf932013-11-14 15:16:20 +11005460 *group_cnt = num_possible_nodes();
Shaohua Li851c30c2013-08-28 14:30:16 +08005461 size = sizeof(struct r5worker) * cnt;
majianpeng60aaf932013-11-14 15:16:20 +11005462 workers = kzalloc(size * *group_cnt, GFP_NOIO);
5463 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
5464 *group_cnt, GFP_NOIO);
5465 if (!*worker_groups || !workers) {
Shaohua Li851c30c2013-08-28 14:30:16 +08005466 kfree(workers);
majianpeng60aaf932013-11-14 15:16:20 +11005467 kfree(*worker_groups);
Shaohua Li851c30c2013-08-28 14:30:16 +08005468 return -ENOMEM;
5469 }
5470
majianpeng60aaf932013-11-14 15:16:20 +11005471 for (i = 0; i < *group_cnt; i++) {
Shaohua Li851c30c2013-08-28 14:30:16 +08005472 struct r5worker_group *group;
5473
majianpeng60aaf932013-11-14 15:16:20 +11005474 group = worker_groups[i];
Shaohua Li851c30c2013-08-28 14:30:16 +08005475 INIT_LIST_HEAD(&group->handle_list);
5476 group->conf = conf;
5477 group->workers = workers + i * cnt;
5478
5479 for (j = 0; j < cnt; j++) {
Shaohua Li566c09c2013-11-14 15:16:17 +11005480 struct r5worker *worker = group->workers + j;
5481 worker->group = group;
5482 INIT_WORK(&worker->work, raid5_do_work);
5483
5484 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
5485 INIT_LIST_HEAD(worker->temp_inactive_list + k);
Shaohua Li851c30c2013-08-28 14:30:16 +08005486 }
5487 }
5488
5489 return 0;
5490}
5491
5492static void free_thread_groups(struct r5conf *conf)
5493{
5494 if (conf->worker_groups)
5495 kfree(conf->worker_groups[0].workers);
5496 kfree(conf->worker_groups);
5497 conf->worker_groups = NULL;
5498}
5499
Dan Williams80c3a6c2009-03-17 18:10:40 -07005500static sector_t
NeilBrownfd01b882011-10-11 16:47:53 +11005501raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07005502{
NeilBrownd1688a62011-10-11 16:49:52 +11005503 struct r5conf *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07005504
5505 if (!sectors)
5506 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005507 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11005508 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11005509 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07005510
Andre Noll9d8f0362009-06-18 08:45:01 +10005511 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Andre Noll664e7c42009-06-18 08:45:27 +10005512 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07005513 return sectors * (raid_disks - conf->max_degraded);
5514}
5515
NeilBrownd1688a62011-10-11 16:49:52 +11005516static void raid5_free_percpu(struct r5conf *conf)
Dan Williams36d1c642009-07-14 11:48:22 -07005517{
5518 struct raid5_percpu *percpu;
5519 unsigned long cpu;
5520
5521 if (!conf->percpu)
5522 return;
5523
5524 get_online_cpus();
5525 for_each_possible_cpu(cpu) {
5526 percpu = per_cpu_ptr(conf->percpu, cpu);
5527 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07005528 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07005529 }
5530#ifdef CONFIG_HOTPLUG_CPU
5531 unregister_cpu_notifier(&conf->cpu_notify);
5532#endif
5533 put_online_cpus();
5534
5535 free_percpu(conf->percpu);
5536}
5537
NeilBrownd1688a62011-10-11 16:49:52 +11005538static void free_conf(struct r5conf *conf)
Dan Williams95fc17a2009-07-31 12:39:15 +10005539{
Shaohua Li851c30c2013-08-28 14:30:16 +08005540 free_thread_groups(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005541 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07005542 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005543 kfree(conf->disks);
5544 kfree(conf->stripe_hashtbl);
5545 kfree(conf);
5546}
5547
Dan Williams36d1c642009-07-14 11:48:22 -07005548#ifdef CONFIG_HOTPLUG_CPU
5549static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
5550 void *hcpu)
5551{
NeilBrownd1688a62011-10-11 16:49:52 +11005552 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
Dan Williams36d1c642009-07-14 11:48:22 -07005553 long cpu = (long)hcpu;
5554 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
5555
5556 switch (action) {
5557 case CPU_UP_PREPARE:
5558 case CPU_UP_PREPARE_FROZEN:
Dan Williamsd6f38f32009-07-14 11:50:52 -07005559 if (conf->level == 6 && !percpu->spare_page)
Dan Williams36d1c642009-07-14 11:48:22 -07005560 percpu->spare_page = alloc_page(GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07005561 if (!percpu->scribble)
5562 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
5563
5564 if (!percpu->scribble ||
5565 (conf->level == 6 && !percpu->spare_page)) {
5566 safe_put_page(percpu->spare_page);
5567 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07005568 pr_err("%s: failed memory allocation for cpu%ld\n",
5569 __func__, cpu);
Akinobu Mita55af6bb2010-05-26 14:43:35 -07005570 return notifier_from_errno(-ENOMEM);
Dan Williams36d1c642009-07-14 11:48:22 -07005571 }
5572 break;
5573 case CPU_DEAD:
5574 case CPU_DEAD_FROZEN:
5575 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07005576 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07005577 percpu->spare_page = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07005578 percpu->scribble = NULL;
Dan Williams36d1c642009-07-14 11:48:22 -07005579 break;
5580 default:
5581 break;
5582 }
5583 return NOTIFY_OK;
5584}
5585#endif
5586
NeilBrownd1688a62011-10-11 16:49:52 +11005587static int raid5_alloc_percpu(struct r5conf *conf)
Dan Williams36d1c642009-07-14 11:48:22 -07005588{
5589 unsigned long cpu;
5590 struct page *spare_page;
Tejun Heoa29d8b82010-02-02 14:39:15 +09005591 struct raid5_percpu __percpu *allcpus;
Dan Williamsd6f38f32009-07-14 11:50:52 -07005592 void *scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07005593 int err;
5594
Dan Williams36d1c642009-07-14 11:48:22 -07005595 allcpus = alloc_percpu(struct raid5_percpu);
5596 if (!allcpus)
5597 return -ENOMEM;
5598 conf->percpu = allcpus;
5599
5600 get_online_cpus();
5601 err = 0;
5602 for_each_present_cpu(cpu) {
Dan Williamsd6f38f32009-07-14 11:50:52 -07005603 if (conf->level == 6) {
5604 spare_page = alloc_page(GFP_KERNEL);
5605 if (!spare_page) {
5606 err = -ENOMEM;
5607 break;
5608 }
5609 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
5610 }
NeilBrown5e5e3e72009-10-16 16:35:30 +11005611 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07005612 if (!scribble) {
Dan Williams36d1c642009-07-14 11:48:22 -07005613 err = -ENOMEM;
5614 break;
5615 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07005616 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07005617 }
5618#ifdef CONFIG_HOTPLUG_CPU
5619 conf->cpu_notify.notifier_call = raid456_cpu_notify;
5620 conf->cpu_notify.priority = 0;
5621 if (err == 0)
5622 err = register_cpu_notifier(&conf->cpu_notify);
5623#endif
5624 put_online_cpus();
5625
5626 return err;
5627}
5628
NeilBrownd1688a62011-10-11 16:49:52 +11005629static struct r5conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005630{
NeilBrownd1688a62011-10-11 16:49:52 +11005631 struct r5conf *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005632 int raid_disk, memory, max_disks;
NeilBrown3cb03002011-10-11 16:45:26 +11005633 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005634 struct disk_info *disk;
NeilBrown02326052012-07-03 15:56:52 +10005635 char pers_name[6];
Shaohua Li566c09c2013-11-14 15:16:17 +11005636 int i;
majianpeng60aaf932013-11-14 15:16:20 +11005637 int group_cnt, worker_cnt_per_group;
5638 struct r5worker_group *new_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005639
NeilBrown91adb562009-03-31 14:39:39 +11005640 if (mddev->new_level != 5
5641 && mddev->new_level != 4
5642 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10005643 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11005644 mdname(mddev), mddev->new_level);
5645 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005646 }
NeilBrown91adb562009-03-31 14:39:39 +11005647 if ((mddev->new_level == 5
5648 && !algorithm_valid_raid5(mddev->new_layout)) ||
5649 (mddev->new_level == 6
5650 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10005651 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11005652 mdname(mddev), mddev->new_layout);
5653 return ERR_PTR(-EIO);
5654 }
5655 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10005656 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11005657 mdname(mddev), mddev->raid_disks);
5658 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11005659 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005660
Andre Noll664e7c42009-06-18 08:45:27 +10005661 if (!mddev->new_chunk_sectors ||
5662 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
5663 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10005664 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
5665 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11005666 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11005667 }
5668
NeilBrownd1688a62011-10-11 16:49:52 +11005669 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
NeilBrown91adb562009-03-31 14:39:39 +11005670 if (conf == NULL)
5671 goto abort;
Shaohua Li851c30c2013-08-28 14:30:16 +08005672 /* Don't enable multi-threading by default*/
majianpeng60aaf932013-11-14 15:16:20 +11005673 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
5674 &new_group)) {
5675 conf->group_cnt = group_cnt;
5676 conf->worker_cnt_per_group = worker_cnt_per_group;
5677 conf->worker_groups = new_group;
5678 } else
Shaohua Li851c30c2013-08-28 14:30:16 +08005679 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11005680 spin_lock_init(&conf->device_lock);
NeilBrownc46501b2013-08-27 15:52:13 +10005681 seqcount_init(&conf->gen_lock);
Dan Williamsf5efd452009-10-16 15:55:38 +11005682 init_waitqueue_head(&conf->wait_for_stripe);
5683 init_waitqueue_head(&conf->wait_for_overlap);
5684 INIT_LIST_HEAD(&conf->handle_list);
5685 INIT_LIST_HEAD(&conf->hold_list);
5686 INIT_LIST_HEAD(&conf->delayed_list);
5687 INIT_LIST_HEAD(&conf->bitmap_list);
Shaohua Li773ca822013-08-27 17:50:39 +08005688 init_llist_head(&conf->released_stripes);
Dan Williamsf5efd452009-10-16 15:55:38 +11005689 atomic_set(&conf->active_stripes, 0);
5690 atomic_set(&conf->preread_active_stripes, 0);
5691 atomic_set(&conf->active_aligned_reads, 0);
5692 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrownd890fa22011-10-26 11:54:39 +11005693 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown91adb562009-03-31 14:39:39 +11005694
5695 conf->raid_disks = mddev->raid_disks;
5696 if (mddev->reshape_position == MaxSector)
5697 conf->previous_raid_disks = mddev->raid_disks;
5698 else
5699 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005700 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
5701 conf->scribble_len = scribble_len(max_disks);
NeilBrown91adb562009-03-31 14:39:39 +11005702
NeilBrown5e5e3e72009-10-16 16:35:30 +11005703 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11005704 GFP_KERNEL);
5705 if (!conf->disks)
5706 goto abort;
5707
5708 conf->mddev = mddev;
5709
5710 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
5711 goto abort;
5712
Shaohua Li566c09c2013-11-14 15:16:17 +11005713 /* We init hash_locks[0] separately to that it can be used
5714 * as the reference lock in the spin_lock_nest_lock() call
5715 * in lock_all_device_hash_locks_irq in order to convince
5716 * lockdep that we know what we are doing.
5717 */
5718 spin_lock_init(conf->hash_locks);
5719 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
5720 spin_lock_init(conf->hash_locks + i);
5721
5722 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5723 INIT_LIST_HEAD(conf->inactive_list + i);
5724
5725 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5726 INIT_LIST_HEAD(conf->temp_inactive_list + i);
5727
Dan Williams36d1c642009-07-14 11:48:22 -07005728 conf->level = mddev->new_level;
5729 if (raid5_alloc_percpu(conf) != 0)
5730 goto abort;
5731
NeilBrown0c55e022010-05-03 14:09:02 +10005732 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11005733
NeilBrowndafb20f2012-03-19 12:46:39 +11005734 rdev_for_each(rdev, mddev) {
NeilBrown91adb562009-03-31 14:39:39 +11005735 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005736 if (raid_disk >= max_disks
NeilBrown91adb562009-03-31 14:39:39 +11005737 || raid_disk < 0)
5738 continue;
5739 disk = conf->disks + raid_disk;
5740
NeilBrown17045f52011-12-23 10:17:53 +11005741 if (test_bit(Replacement, &rdev->flags)) {
5742 if (disk->replacement)
5743 goto abort;
5744 disk->replacement = rdev;
5745 } else {
5746 if (disk->rdev)
5747 goto abort;
5748 disk->rdev = rdev;
5749 }
NeilBrown91adb562009-03-31 14:39:39 +11005750
5751 if (test_bit(In_sync, &rdev->flags)) {
5752 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10005753 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
5754 " disk %d\n",
5755 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
Jonathan Brassowd6b212f2011-06-08 18:00:28 -05005756 } else if (rdev->saved_raid_disk != raid_disk)
NeilBrown91adb562009-03-31 14:39:39 +11005757 /* Cannot rely on bitmap to complete recovery */
5758 conf->fullsync = 1;
5759 }
5760
Andre Noll09c9e5f2009-06-18 08:45:55 +10005761 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown91adb562009-03-31 14:39:39 +11005762 conf->level = mddev->new_level;
5763 if (conf->level == 6)
5764 conf->max_degraded = 2;
5765 else
5766 conf->max_degraded = 1;
5767 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11005768 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11005769 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10005770 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11005771 conf->prev_algo = mddev->layout;
5772 }
NeilBrown91adb562009-03-31 14:39:39 +11005773
5774 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11005775 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
Shaohua Li4bda5562013-11-14 15:16:17 +11005776 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
Shaohua Li566c09c2013-11-14 15:16:17 +11005777 if (grow_stripes(conf, NR_STRIPES)) {
NeilBrown91adb562009-03-31 14:39:39 +11005778 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005779 "md/raid:%s: couldn't allocate %dkB for buffers\n",
5780 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11005781 goto abort;
5782 } else
NeilBrown0c55e022010-05-03 14:09:02 +10005783 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
5784 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11005785
NeilBrown02326052012-07-03 15:56:52 +10005786 sprintf(pers_name, "raid%d", mddev->new_level);
5787 conf->thread = md_register_thread(raid5d, mddev, pers_name);
NeilBrown91adb562009-03-31 14:39:39 +11005788 if (!conf->thread) {
5789 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005790 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11005791 mdname(mddev));
5792 goto abort;
5793 }
5794
5795 return conf;
5796
5797 abort:
5798 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10005799 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11005800 return ERR_PTR(-EIO);
5801 } else
5802 return ERR_PTR(-ENOMEM);
5803}
5804
NeilBrownc148ffd2009-11-13 17:47:00 +11005805
5806static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
5807{
5808 switch (algo) {
5809 case ALGORITHM_PARITY_0:
5810 if (raid_disk < max_degraded)
5811 return 1;
5812 break;
5813 case ALGORITHM_PARITY_N:
5814 if (raid_disk >= raid_disks - max_degraded)
5815 return 1;
5816 break;
5817 case ALGORITHM_PARITY_0_6:
5818 if (raid_disk == 0 ||
5819 raid_disk == raid_disks - 1)
5820 return 1;
5821 break;
5822 case ALGORITHM_LEFT_ASYMMETRIC_6:
5823 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5824 case ALGORITHM_LEFT_SYMMETRIC_6:
5825 case ALGORITHM_RIGHT_SYMMETRIC_6:
5826 if (raid_disk == raid_disks - 1)
5827 return 1;
5828 }
5829 return 0;
5830}
5831
NeilBrownfd01b882011-10-11 16:47:53 +11005832static int run(struct mddev *mddev)
NeilBrown91adb562009-03-31 14:39:39 +11005833{
NeilBrownd1688a62011-10-11 16:49:52 +11005834 struct r5conf *conf;
NeilBrown9f7c2222010-07-26 12:04:13 +10005835 int working_disks = 0;
NeilBrownc148ffd2009-11-13 17:47:00 +11005836 int dirty_parity_disks = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11005837 struct md_rdev *rdev;
NeilBrownc148ffd2009-11-13 17:47:00 +11005838 sector_t reshape_offset = 0;
NeilBrown17045f52011-12-23 10:17:53 +11005839 int i;
NeilBrownb5254dd2012-05-21 09:27:01 +10005840 long long min_offset_diff = 0;
5841 int first = 1;
NeilBrown91adb562009-03-31 14:39:39 +11005842
Andre Noll8c6ac862009-06-18 08:48:06 +10005843 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10005844 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10005845 " -- starting background reconstruction\n",
5846 mdname(mddev));
NeilBrownb5254dd2012-05-21 09:27:01 +10005847
5848 rdev_for_each(rdev, mddev) {
5849 long long diff;
5850 if (rdev->raid_disk < 0)
5851 continue;
5852 diff = (rdev->new_data_offset - rdev->data_offset);
5853 if (first) {
5854 min_offset_diff = diff;
5855 first = 0;
5856 } else if (mddev->reshape_backwards &&
5857 diff < min_offset_diff)
5858 min_offset_diff = diff;
5859 else if (!mddev->reshape_backwards &&
5860 diff > min_offset_diff)
5861 min_offset_diff = diff;
5862 }
5863
NeilBrownf6705572006-03-27 01:18:11 -08005864 if (mddev->reshape_position != MaxSector) {
5865 /* Check that we can continue the reshape.
NeilBrownb5254dd2012-05-21 09:27:01 +10005866 * Difficulties arise if the stripe we would write to
5867 * next is at or after the stripe we would read from next.
5868 * For a reshape that changes the number of devices, this
5869 * is only possible for a very short time, and mdadm makes
5870 * sure that time appears to have past before assembling
5871 * the array. So we fail if that time hasn't passed.
5872 * For a reshape that keeps the number of devices the same
5873 * mdadm must be monitoring the reshape can keeping the
5874 * critical areas read-only and backed up. It will start
5875 * the array in read-only mode, so we check for that.
NeilBrownf6705572006-03-27 01:18:11 -08005876 */
5877 sector_t here_new, here_old;
5878 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11005879 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrownf6705572006-03-27 01:18:11 -08005880
NeilBrown88ce4932009-03-31 15:24:23 +11005881 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10005882 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08005883 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08005884 mdname(mddev));
5885 return -EINVAL;
5886 }
NeilBrownf6705572006-03-27 01:18:11 -08005887 old_disks = mddev->raid_disks - mddev->delta_disks;
5888 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08005889 * further up in new geometry must map after here in old
5890 * geometry.
NeilBrownf6705572006-03-27 01:18:11 -08005891 */
5892 here_new = mddev->reshape_position;
Andre Noll664e7c42009-06-18 08:45:27 +10005893 if (sector_div(here_new, mddev->new_chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08005894 (mddev->raid_disks - max_degraded))) {
NeilBrown0c55e022010-05-03 14:09:02 +10005895 printk(KERN_ERR "md/raid:%s: reshape_position not "
5896 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005897 return -EINVAL;
5898 }
NeilBrownc148ffd2009-11-13 17:47:00 +11005899 reshape_offset = here_new * mddev->new_chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08005900 /* here_new is the stripe we will write to */
5901 here_old = mddev->reshape_position;
Andre Noll9d8f0362009-06-18 08:45:01 +10005902 sector_div(here_old, mddev->chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08005903 (old_disks-max_degraded));
5904 /* here_old is the first stripe that we might need to read
5905 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10005906 if (mddev->delta_disks == 0) {
NeilBrownb5254dd2012-05-21 09:27:01 +10005907 if ((here_new * mddev->new_chunk_sectors !=
5908 here_old * mddev->chunk_sectors)) {
5909 printk(KERN_ERR "md/raid:%s: reshape position is"
5910 " confused - aborting\n", mdname(mddev));
5911 return -EINVAL;
5912 }
NeilBrown67ac6012009-08-13 10:06:24 +10005913 /* We cannot be sure it is safe to start an in-place
NeilBrownb5254dd2012-05-21 09:27:01 +10005914 * reshape. It is only safe if user-space is monitoring
NeilBrown67ac6012009-08-13 10:06:24 +10005915 * and taking constant backups.
5916 * mdadm always starts a situation like this in
5917 * readonly mode so it can take control before
5918 * allowing any writes. So just check for that.
5919 */
NeilBrownb5254dd2012-05-21 09:27:01 +10005920 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
5921 abs(min_offset_diff) >= mddev->new_chunk_sectors)
5922 /* not really in-place - so OK */;
5923 else if (mddev->ro == 0) {
5924 printk(KERN_ERR "md/raid:%s: in-place reshape "
5925 "must be started in read-only mode "
5926 "- aborting\n",
NeilBrown0c55e022010-05-03 14:09:02 +10005927 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10005928 return -EINVAL;
5929 }
NeilBrown2c810cd2012-05-21 09:27:00 +10005930 } else if (mddev->reshape_backwards
NeilBrownb5254dd2012-05-21 09:27:01 +10005931 ? (here_new * mddev->new_chunk_sectors + min_offset_diff <=
NeilBrown67ac6012009-08-13 10:06:24 +10005932 here_old * mddev->chunk_sectors)
5933 : (here_new * mddev->new_chunk_sectors >=
NeilBrownb5254dd2012-05-21 09:27:01 +10005934 here_old * mddev->chunk_sectors + (-min_offset_diff))) {
NeilBrownf6705572006-03-27 01:18:11 -08005935 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10005936 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5937 "auto-recovery - aborting.\n",
5938 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005939 return -EINVAL;
5940 }
NeilBrown0c55e022010-05-03 14:09:02 +10005941 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5942 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005943 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08005944 } else {
NeilBrown91adb562009-03-31 14:39:39 +11005945 BUG_ON(mddev->level != mddev->new_level);
5946 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10005947 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11005948 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08005949 }
5950
NeilBrown245f46c2009-03-31 14:39:39 +11005951 if (mddev->private == NULL)
5952 conf = setup_conf(mddev);
5953 else
5954 conf = mddev->private;
5955
NeilBrown91adb562009-03-31 14:39:39 +11005956 if (IS_ERR(conf))
5957 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08005958
NeilBrownb5254dd2012-05-21 09:27:01 +10005959 conf->min_offset_diff = min_offset_diff;
NeilBrown91adb562009-03-31 14:39:39 +11005960 mddev->thread = conf->thread;
5961 conf->thread = NULL;
5962 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005963
NeilBrown17045f52011-12-23 10:17:53 +11005964 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
5965 i++) {
5966 rdev = conf->disks[i].rdev;
5967 if (!rdev && conf->disks[i].replacement) {
5968 /* The replacement is all we have yet */
5969 rdev = conf->disks[i].replacement;
5970 conf->disks[i].replacement = NULL;
5971 clear_bit(Replacement, &rdev->flags);
5972 conf->disks[i].rdev = rdev;
5973 }
5974 if (!rdev)
NeilBrownc148ffd2009-11-13 17:47:00 +11005975 continue;
NeilBrown17045f52011-12-23 10:17:53 +11005976 if (conf->disks[i].replacement &&
5977 conf->reshape_progress != MaxSector) {
5978 /* replacements and reshape simply do not mix. */
5979 printk(KERN_ERR "md: cannot handle concurrent "
5980 "replacement and reshape.\n");
5981 goto abort;
5982 }
NeilBrown2f115882010-06-17 17:41:03 +10005983 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown91adb562009-03-31 14:39:39 +11005984 working_disks++;
NeilBrown2f115882010-06-17 17:41:03 +10005985 continue;
5986 }
NeilBrownc148ffd2009-11-13 17:47:00 +11005987 /* This disc is not fully in-sync. However if it
5988 * just stored parity (beyond the recovery_offset),
5989 * when we don't need to be concerned about the
5990 * array being dirty.
5991 * When reshape goes 'backwards', we never have
5992 * partially completed devices, so we only need
5993 * to worry about reshape going forwards.
5994 */
5995 /* Hack because v0.91 doesn't store recovery_offset properly. */
5996 if (mddev->major_version == 0 &&
5997 mddev->minor_version > 90)
5998 rdev->recovery_offset = reshape_offset;
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07005999
NeilBrownc148ffd2009-11-13 17:47:00 +11006000 if (rdev->recovery_offset < reshape_offset) {
6001 /* We need to check old and new layout */
6002 if (!only_parity(rdev->raid_disk,
6003 conf->algorithm,
6004 conf->raid_disks,
6005 conf->max_degraded))
6006 continue;
6007 }
6008 if (!only_parity(rdev->raid_disk,
6009 conf->prev_algo,
6010 conf->previous_raid_disks,
6011 conf->max_degraded))
6012 continue;
6013 dirty_parity_disks++;
6014 }
NeilBrown91adb562009-03-31 14:39:39 +11006015
NeilBrown17045f52011-12-23 10:17:53 +11006016 /*
6017 * 0 for a fully functional array, 1 or 2 for a degraded array.
6018 */
NeilBrown908f4fb2011-12-23 10:17:50 +11006019 mddev->degraded = calc_degraded(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006020
NeilBrown674806d2010-06-16 17:17:53 +10006021 if (has_failed(conf)) {
NeilBrown0c55e022010-05-03 14:09:02 +10006022 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07006023 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07006024 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006025 goto abort;
6026 }
6027
NeilBrown91adb562009-03-31 14:39:39 +11006028 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10006029 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11006030 mddev->resync_max_sectors = mddev->dev_sectors;
6031
NeilBrownc148ffd2009-11-13 17:47:00 +11006032 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07006033 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006034 if (mddev->ok_start_degraded)
6035 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10006036 "md/raid:%s: starting dirty degraded array"
6037 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006038 mdname(mddev));
6039 else {
6040 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10006041 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006042 mdname(mddev));
6043 goto abort;
6044 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006045 }
6046
Linus Torvalds1da177e2005-04-16 15:20:36 -07006047 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10006048 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
6049 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11006050 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
6051 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006052 else
NeilBrown0c55e022010-05-03 14:09:02 +10006053 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
6054 " out of %d devices, algorithm %d\n",
6055 mdname(mddev), conf->level,
6056 mddev->raid_disks - mddev->degraded,
6057 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006058
6059 print_raid5_conf(conf);
6060
NeilBrownfef9c612009-03-31 15:16:46 +11006061 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11006062 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08006063 atomic_set(&conf->reshape_stripes, 0);
6064 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6065 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6066 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6067 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6068 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10006069 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08006070 }
6071
Linus Torvalds1da177e2005-04-16 15:20:36 -07006072
6073 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10006074 if (mddev->to_remove == &raid5_attrs_group)
6075 mddev->to_remove = NULL;
NeilBrown00bcb4a2010-06-01 19:37:23 +10006076 else if (mddev->kobj.sd &&
6077 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08006078 printk(KERN_WARNING
NeilBrown4a5add42010-06-01 19:37:28 +10006079 "raid5: failed to create sysfs attributes for %s\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08006080 mdname(mddev));
NeilBrown4a5add42010-06-01 19:37:28 +10006081 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6082
6083 if (mddev->queue) {
NeilBrown9f7c2222010-07-26 12:04:13 +10006084 int chunk_size;
Shaohua Li620125f2012-10-11 13:49:05 +11006085 bool discard_supported = true;
NeilBrown4a5add42010-06-01 19:37:28 +10006086 /* read-ahead size must cover two whole stripes, which
6087 * is 2 * (datadisks) * chunksize where 'n' is the
6088 * number of raid devices
6089 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006090 int data_disks = conf->previous_raid_disks - conf->max_degraded;
6091 int stripe = data_disks *
6092 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
6093 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6094 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
NeilBrown4a5add42010-06-01 19:37:28 +10006095
6096 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
NeilBrown11d8a6e2010-07-26 11:57:07 +10006097
6098 mddev->queue->backing_dev_info.congested_data = mddev;
6099 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
NeilBrown9f7c2222010-07-26 12:04:13 +10006100
6101 chunk_size = mddev->chunk_sectors << 9;
6102 blk_queue_io_min(mddev->queue, chunk_size);
6103 blk_queue_io_opt(mddev->queue, chunk_size *
6104 (conf->raid_disks - conf->max_degraded));
Shaohua Li620125f2012-10-11 13:49:05 +11006105 /*
6106 * We can only discard a whole stripe. It doesn't make sense to
6107 * discard data disk but write parity disk
6108 */
6109 stripe = stripe * PAGE_SIZE;
NeilBrown4ac68752012-11-19 13:11:26 +11006110 /* Round up to power of 2, as discard handling
6111 * currently assumes that */
6112 while ((stripe-1) & stripe)
6113 stripe = (stripe | (stripe-1)) + 1;
Shaohua Li620125f2012-10-11 13:49:05 +11006114 mddev->queue->limits.discard_alignment = stripe;
6115 mddev->queue->limits.discard_granularity = stripe;
6116 /*
6117 * unaligned part of discard request will be ignored, so can't
6118 * guarantee discard_zerors_data
6119 */
6120 mddev->queue->limits.discard_zeroes_data = 0;
NeilBrown9f7c2222010-07-26 12:04:13 +10006121
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07006122 blk_queue_max_write_same_sectors(mddev->queue, 0);
6123
NeilBrown05616be2012-05-21 09:27:00 +10006124 rdev_for_each(rdev, mddev) {
NeilBrown9f7c2222010-07-26 12:04:13 +10006125 disk_stack_limits(mddev->gendisk, rdev->bdev,
6126 rdev->data_offset << 9);
NeilBrown05616be2012-05-21 09:27:00 +10006127 disk_stack_limits(mddev->gendisk, rdev->bdev,
6128 rdev->new_data_offset << 9);
Shaohua Li620125f2012-10-11 13:49:05 +11006129 /*
6130 * discard_zeroes_data is required, otherwise data
6131 * could be lost. Consider a scenario: discard a stripe
6132 * (the stripe could be inconsistent if
6133 * discard_zeroes_data is 0); write one disk of the
6134 * stripe (the stripe could be inconsistent again
6135 * depending on which disks are used to calculate
6136 * parity); the disk is broken; The stripe data of this
6137 * disk is lost.
6138 */
6139 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
6140 !bdev_get_queue(rdev->bdev)->
6141 limits.discard_zeroes_data)
6142 discard_supported = false;
NeilBrown05616be2012-05-21 09:27:00 +10006143 }
Shaohua Li620125f2012-10-11 13:49:05 +11006144
6145 if (discard_supported &&
6146 mddev->queue->limits.max_discard_sectors >= stripe &&
6147 mddev->queue->limits.discard_granularity >= stripe)
6148 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
6149 mddev->queue);
6150 else
6151 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
6152 mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006153 }
6154
Linus Torvalds1da177e2005-04-16 15:20:36 -07006155 return 0;
6156abort:
NeilBrown01f96c02011-09-21 15:30:20 +10006157 md_unregister_thread(&mddev->thread);
NeilBrowne4f869d2011-10-07 14:22:49 +11006158 print_raid5_conf(conf);
6159 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006160 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10006161 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006162 return -EIO;
6163}
6164
NeilBrownfd01b882011-10-11 16:47:53 +11006165static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006166{
NeilBrownd1688a62011-10-11 16:49:52 +11006167 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006168
NeilBrown01f96c02011-09-21 15:30:20 +10006169 md_unregister_thread(&mddev->thread);
NeilBrown11d8a6e2010-07-26 11:57:07 +10006170 if (mddev->queue)
6171 mddev->queue->backing_dev_info.congested_fn = NULL;
Dan Williams95fc17a2009-07-31 12:39:15 +10006172 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10006173 mddev->private = NULL;
6174 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006175 return 0;
6176}
6177
NeilBrownfd01b882011-10-11 16:47:53 +11006178static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006179{
NeilBrownd1688a62011-10-11 16:49:52 +11006180 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006181 int i;
6182
Andre Noll9d8f0362009-06-18 08:45:01 +10006183 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
6184 mddev->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07006185 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006186 for (i = 0; i < conf->raid_disks; i++)
6187 seq_printf (seq, "%s",
6188 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08006189 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006190 seq_printf (seq, "]");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006191}
6192
NeilBrownd1688a62011-10-11 16:49:52 +11006193static void print_raid5_conf (struct r5conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006194{
6195 int i;
6196 struct disk_info *tmp;
6197
NeilBrown0c55e022010-05-03 14:09:02 +10006198 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006199 if (!conf) {
6200 printk("(conf==NULL)\n");
6201 return;
6202 }
NeilBrown0c55e022010-05-03 14:09:02 +10006203 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
6204 conf->raid_disks,
6205 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006206
6207 for (i = 0; i < conf->raid_disks; i++) {
6208 char b[BDEVNAME_SIZE];
6209 tmp = conf->disks + i;
6210 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10006211 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
6212 i, !test_bit(Faulty, &tmp->rdev->flags),
6213 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006214 }
6215}
6216
NeilBrownfd01b882011-10-11 16:47:53 +11006217static int raid5_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006218{
6219 int i;
NeilBrownd1688a62011-10-11 16:49:52 +11006220 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006221 struct disk_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10006222 int count = 0;
6223 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006224
6225 for (i = 0; i < conf->raid_disks; i++) {
6226 tmp = conf->disks + i;
NeilBrowndd054fc2011-12-23 10:17:53 +11006227 if (tmp->replacement
6228 && tmp->replacement->recovery_offset == MaxSector
6229 && !test_bit(Faulty, &tmp->replacement->flags)
6230 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
6231 /* Replacement has just become active. */
6232 if (!tmp->rdev
6233 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
6234 count++;
6235 if (tmp->rdev) {
6236 /* Replaced device not technically faulty,
6237 * but we need to be sure it gets removed
6238 * and never re-added.
6239 */
6240 set_bit(Faulty, &tmp->rdev->flags);
6241 sysfs_notify_dirent_safe(
6242 tmp->rdev->sysfs_state);
6243 }
6244 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
6245 } else if (tmp->rdev
NeilBrown70fffd02010-06-16 17:01:25 +10006246 && tmp->rdev->recovery_offset == MaxSector
NeilBrownb2d444d2005-11-08 21:39:31 -08006247 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07006248 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10006249 count++;
Jonathan Brassow43c73ca2011-01-14 09:14:33 +11006250 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006251 }
6252 }
NeilBrown6b965622010-08-18 11:56:59 +10006253 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown908f4fb2011-12-23 10:17:50 +11006254 mddev->degraded = calc_degraded(conf);
NeilBrown6b965622010-08-18 11:56:59 +10006255 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006256 print_raid5_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10006257 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006258}
6259
NeilBrownb8321b62011-12-23 10:17:51 +11006260static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006261{
NeilBrownd1688a62011-10-11 16:49:52 +11006262 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006263 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11006264 int number = rdev->raid_disk;
NeilBrown657e3e42011-12-23 10:17:52 +11006265 struct md_rdev **rdevp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006266 struct disk_info *p = conf->disks + number;
6267
6268 print_raid5_conf(conf);
NeilBrown657e3e42011-12-23 10:17:52 +11006269 if (rdev == p->rdev)
6270 rdevp = &p->rdev;
6271 else if (rdev == p->replacement)
6272 rdevp = &p->replacement;
6273 else
6274 return 0;
NeilBrownec32a2b2009-03-31 15:17:38 +11006275
NeilBrown657e3e42011-12-23 10:17:52 +11006276 if (number >= conf->raid_disks &&
6277 conf->reshape_progress == MaxSector)
6278 clear_bit(In_sync, &rdev->flags);
6279
6280 if (test_bit(In_sync, &rdev->flags) ||
6281 atomic_read(&rdev->nr_pending)) {
6282 err = -EBUSY;
6283 goto abort;
6284 }
6285 /* Only remove non-faulty devices if recovery
6286 * isn't possible.
6287 */
6288 if (!test_bit(Faulty, &rdev->flags) &&
6289 mddev->recovery_disabled != conf->recovery_disabled &&
6290 !has_failed(conf) &&
NeilBrowndd054fc2011-12-23 10:17:53 +11006291 (!p->replacement || p->replacement == rdev) &&
NeilBrown657e3e42011-12-23 10:17:52 +11006292 number < conf->raid_disks) {
6293 err = -EBUSY;
6294 goto abort;
6295 }
6296 *rdevp = NULL;
6297 synchronize_rcu();
6298 if (atomic_read(&rdev->nr_pending)) {
6299 /* lost the race, try later */
6300 err = -EBUSY;
6301 *rdevp = rdev;
NeilBrowndd054fc2011-12-23 10:17:53 +11006302 } else if (p->replacement) {
6303 /* We must have just cleared 'rdev' */
6304 p->rdev = p->replacement;
6305 clear_bit(Replacement, &p->replacement->flags);
6306 smp_mb(); /* Make sure other CPUs may see both as identical
6307 * but will never see neither - if they are careful
6308 */
6309 p->replacement = NULL;
6310 clear_bit(WantReplacement, &rdev->flags);
6311 } else
6312 /* We might have just removed the Replacement as faulty-
6313 * clear the bit just in case
6314 */
6315 clear_bit(WantReplacement, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006316abort:
6317
6318 print_raid5_conf(conf);
6319 return err;
6320}
6321
NeilBrownfd01b882011-10-11 16:47:53 +11006322static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006323{
NeilBrownd1688a62011-10-11 16:49:52 +11006324 struct r5conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10006325 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006326 int disk;
6327 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10006328 int first = 0;
6329 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006330
NeilBrown7f0da592011-07-28 11:39:22 +10006331 if (mddev->recovery_disabled == conf->recovery_disabled)
6332 return -EBUSY;
6333
NeilBrowndc10c642012-03-19 12:46:37 +11006334 if (rdev->saved_raid_disk < 0 && has_failed(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006335 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10006336 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006337
Neil Brown6c2fce22008-06-28 08:31:31 +10006338 if (rdev->raid_disk >= 0)
6339 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006340
6341 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07006342 * find the disk ... but prefer rdev->saved_raid_disk
6343 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006344 */
NeilBrown16a53ec2006-06-26 00:27:38 -07006345 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10006346 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07006347 conf->disks[rdev->saved_raid_disk].rdev == NULL)
NeilBrown5cfb22a2012-07-03 11:46:53 +10006348 first = rdev->saved_raid_disk;
6349
6350 for (disk = first; disk <= last; disk++) {
NeilBrown7bfec5f2011-12-23 10:17:53 +11006351 p = conf->disks + disk;
6352 if (p->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08006353 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006354 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10006355 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07006356 if (rdev->saved_raid_disk != disk)
6357 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08006358 rcu_assign_pointer(p->rdev, rdev);
NeilBrown5cfb22a2012-07-03 11:46:53 +10006359 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006360 }
NeilBrown5cfb22a2012-07-03 11:46:53 +10006361 }
6362 for (disk = first; disk <= last; disk++) {
6363 p = conf->disks + disk;
NeilBrown7bfec5f2011-12-23 10:17:53 +11006364 if (test_bit(WantReplacement, &p->rdev->flags) &&
6365 p->replacement == NULL) {
6366 clear_bit(In_sync, &rdev->flags);
6367 set_bit(Replacement, &rdev->flags);
6368 rdev->raid_disk = disk;
6369 err = 0;
6370 conf->fullsync = 1;
6371 rcu_assign_pointer(p->replacement, rdev);
6372 break;
6373 }
6374 }
NeilBrown5cfb22a2012-07-03 11:46:53 +10006375out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006376 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10006377 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006378}
6379
NeilBrownfd01b882011-10-11 16:47:53 +11006380static int raid5_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006381{
6382 /* no resync is happening, and there is enough space
6383 * on all devices, so we can resize.
6384 * We need to make sure resync covers any new space.
6385 * If the array is shrinking we should possibly wait until
6386 * any io in the removed space completes, but it hardly seems
6387 * worth it.
6388 */
NeilBrowna4a61252012-05-22 13:55:27 +10006389 sector_t newsize;
Andre Noll9d8f0362009-06-18 08:45:01 +10006390 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
NeilBrowna4a61252012-05-22 13:55:27 +10006391 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
6392 if (mddev->external_size &&
6393 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11006394 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10006395 if (mddev->bitmap) {
6396 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
6397 if (ret)
6398 return ret;
6399 }
6400 md_set_array_sectors(mddev, newsize);
Andre Nollf233ea52008-07-21 17:05:22 +10006401 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10006402 revalidate_disk(mddev->gendisk);
NeilBrownb0986362011-05-11 15:52:21 +10006403 if (sectors > mddev->dev_sectors &&
6404 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11006405 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006406 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6407 }
Andre Noll58c0fed2009-03-31 14:33:13 +11006408 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07006409 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006410 return 0;
6411}
6412
NeilBrownfd01b882011-10-11 16:47:53 +11006413static int check_stripe_cache(struct mddev *mddev)
NeilBrown01ee22b2009-06-18 08:47:20 +10006414{
6415 /* Can only proceed if there are plenty of stripe_heads.
6416 * We need a minimum of one full stripe,, and for sensible progress
6417 * it is best to have about 4 times that.
6418 * If we require 4 times, then the default 256 4K stripe_heads will
6419 * allow for chunk sizes up to 256K, which is probably OK.
6420 * If the chunk size is greater, user-space should request more
6421 * stripe_heads first.
6422 */
NeilBrownd1688a62011-10-11 16:49:52 +11006423 struct r5conf *conf = mddev->private;
NeilBrown01ee22b2009-06-18 08:47:20 +10006424 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
6425 > conf->max_nr_stripes ||
6426 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
6427 > conf->max_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10006428 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
6429 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10006430 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
6431 / STRIPE_SIZE)*4);
6432 return 0;
6433 }
6434 return 1;
6435}
6436
NeilBrownfd01b882011-10-11 16:47:53 +11006437static int check_reshape(struct mddev *mddev)
NeilBrown29269552006-03-27 01:18:10 -08006438{
NeilBrownd1688a62011-10-11 16:49:52 +11006439 struct r5conf *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08006440
NeilBrown88ce4932009-03-31 15:24:23 +11006441 if (mddev->delta_disks == 0 &&
6442 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10006443 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10006444 return 0; /* nothing to do */
NeilBrown674806d2010-06-16 17:17:53 +10006445 if (has_failed(conf))
NeilBrownec32a2b2009-03-31 15:17:38 +11006446 return -EINVAL;
NeilBrownfdcfbbb2013-07-04 16:38:16 +10006447 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
NeilBrownec32a2b2009-03-31 15:17:38 +11006448 /* We might be able to shrink, but the devices must
6449 * be made bigger first.
6450 * For raid6, 4 is the minimum size.
6451 * Otherwise 2 is the minimum
6452 */
6453 int min = 2;
6454 if (mddev->level == 6)
6455 min = 4;
6456 if (mddev->raid_disks + mddev->delta_disks < min)
6457 return -EINVAL;
6458 }
NeilBrown29269552006-03-27 01:18:10 -08006459
NeilBrown01ee22b2009-06-18 08:47:20 +10006460 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08006461 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08006462
NeilBrowne56108d62012-10-11 14:24:13 +11006463 return resize_stripes(conf, (conf->previous_raid_disks
6464 + mddev->delta_disks));
NeilBrown63c70c42006-03-27 01:18:13 -08006465}
6466
NeilBrownfd01b882011-10-11 16:47:53 +11006467static int raid5_start_reshape(struct mddev *mddev)
NeilBrown63c70c42006-03-27 01:18:13 -08006468{
NeilBrownd1688a62011-10-11 16:49:52 +11006469 struct r5conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11006470 struct md_rdev *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08006471 int spares = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07006472 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08006473
NeilBrownf4168852007-02-28 20:11:53 -08006474 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08006475 return -EBUSY;
6476
NeilBrown01ee22b2009-06-18 08:47:20 +10006477 if (!check_stripe_cache(mddev))
6478 return -ENOSPC;
6479
NeilBrown30b67642012-05-22 13:55:28 +10006480 if (has_failed(conf))
6481 return -EINVAL;
6482
NeilBrownc6563a82012-05-21 09:27:00 +10006483 rdev_for_each(rdev, mddev) {
NeilBrown469518a2011-01-31 11:57:43 +11006484 if (!test_bit(In_sync, &rdev->flags)
6485 && !test_bit(Faulty, &rdev->flags))
NeilBrown29269552006-03-27 01:18:10 -08006486 spares++;
NeilBrownc6563a82012-05-21 09:27:00 +10006487 }
NeilBrown63c70c42006-03-27 01:18:13 -08006488
NeilBrownf4168852007-02-28 20:11:53 -08006489 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08006490 /* Not enough devices even to make a degraded array
6491 * of that size
6492 */
6493 return -EINVAL;
6494
NeilBrownec32a2b2009-03-31 15:17:38 +11006495 /* Refuse to reduce size of the array. Any reductions in
6496 * array size must be through explicit setting of array_size
6497 * attribute.
6498 */
6499 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
6500 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10006501 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11006502 "before number of disks\n", mdname(mddev));
6503 return -EINVAL;
6504 }
6505
NeilBrownf6705572006-03-27 01:18:11 -08006506 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08006507 spin_lock_irq(&conf->device_lock);
NeilBrownc46501b2013-08-27 15:52:13 +10006508 write_seqcount_begin(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006509 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08006510 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10006511 conf->prev_chunk_sectors = conf->chunk_sectors;
6512 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11006513 conf->prev_algo = conf->algorithm;
6514 conf->algorithm = mddev->new_layout;
NeilBrown05616be2012-05-21 09:27:00 +10006515 conf->generation++;
6516 /* Code that selects data_offset needs to see the generation update
6517 * if reshape_progress has been set - so a memory barrier needed.
6518 */
6519 smp_mb();
NeilBrown2c810cd2012-05-21 09:27:00 +10006520 if (mddev->reshape_backwards)
NeilBrownfef9c612009-03-31 15:16:46 +11006521 conf->reshape_progress = raid5_size(mddev, 0, 0);
6522 else
6523 conf->reshape_progress = 0;
6524 conf->reshape_safe = conf->reshape_progress;
NeilBrownc46501b2013-08-27 15:52:13 +10006525 write_seqcount_end(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006526 spin_unlock_irq(&conf->device_lock);
6527
NeilBrown4d77e3b2013-08-27 15:57:47 +10006528 /* Now make sure any requests that proceeded on the assumption
6529 * the reshape wasn't running - like Discard or Read - have
6530 * completed.
6531 */
6532 mddev_suspend(mddev);
6533 mddev_resume(mddev);
6534
NeilBrown29269552006-03-27 01:18:10 -08006535 /* Add some new drives, as many as will fit.
6536 * We know there are enough to make the newly sized array work.
NeilBrown3424bf62010-06-17 17:48:26 +10006537 * Don't add devices if we are reducing the number of
6538 * devices in the array. This is because it is not possible
6539 * to correctly record the "partially reconstructed" state of
6540 * such devices during the reshape and confusion could result.
NeilBrown29269552006-03-27 01:18:10 -08006541 */
NeilBrown87a8dec2011-01-31 11:57:43 +11006542 if (mddev->delta_disks >= 0) {
NeilBrowndafb20f2012-03-19 12:46:39 +11006543 rdev_for_each(rdev, mddev)
NeilBrown87a8dec2011-01-31 11:57:43 +11006544 if (rdev->raid_disk < 0 &&
6545 !test_bit(Faulty, &rdev->flags)) {
6546 if (raid5_add_disk(mddev, rdev) == 0) {
NeilBrown87a8dec2011-01-31 11:57:43 +11006547 if (rdev->raid_disk
NeilBrown9d4c7d82012-03-13 11:21:21 +11006548 >= conf->previous_raid_disks)
NeilBrown87a8dec2011-01-31 11:57:43 +11006549 set_bit(In_sync, &rdev->flags);
NeilBrown9d4c7d82012-03-13 11:21:21 +11006550 else
NeilBrown87a8dec2011-01-31 11:57:43 +11006551 rdev->recovery_offset = 0;
Namhyung Kim36fad852011-07-27 11:00:36 +10006552
6553 if (sysfs_link_rdev(mddev, rdev))
NeilBrown87a8dec2011-01-31 11:57:43 +11006554 /* Failure here is OK */;
NeilBrown50da0842011-01-31 11:57:43 +11006555 }
NeilBrown87a8dec2011-01-31 11:57:43 +11006556 } else if (rdev->raid_disk >= conf->previous_raid_disks
6557 && !test_bit(Faulty, &rdev->flags)) {
6558 /* This is a spare that was manually added */
6559 set_bit(In_sync, &rdev->flags);
NeilBrown87a8dec2011-01-31 11:57:43 +11006560 }
NeilBrown29269552006-03-27 01:18:10 -08006561
NeilBrown87a8dec2011-01-31 11:57:43 +11006562 /* When a reshape changes the number of devices,
6563 * ->degraded is measured against the larger of the
6564 * pre and post number of devices.
6565 */
NeilBrownec32a2b2009-03-31 15:17:38 +11006566 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown908f4fb2011-12-23 10:17:50 +11006567 mddev->degraded = calc_degraded(conf);
NeilBrownec32a2b2009-03-31 15:17:38 +11006568 spin_unlock_irqrestore(&conf->device_lock, flags);
6569 }
NeilBrown63c70c42006-03-27 01:18:13 -08006570 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10006571 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07006572 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08006573
NeilBrown29269552006-03-27 01:18:10 -08006574 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6575 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6576 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6577 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6578 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10006579 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08006580 if (!mddev->sync_thread) {
6581 mddev->recovery = 0;
6582 spin_lock_irq(&conf->device_lock);
NeilBrownba8805b2013-11-14 15:16:15 +11006583 write_seqcount_begin(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006584 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownba8805b2013-11-14 15:16:15 +11006585 mddev->new_chunk_sectors =
6586 conf->chunk_sectors = conf->prev_chunk_sectors;
6587 mddev->new_layout = conf->algorithm = conf->prev_algo;
NeilBrown05616be2012-05-21 09:27:00 +10006588 rdev_for_each(rdev, mddev)
6589 rdev->new_data_offset = rdev->data_offset;
6590 smp_wmb();
NeilBrownba8805b2013-11-14 15:16:15 +11006591 conf->generation --;
NeilBrownfef9c612009-03-31 15:16:46 +11006592 conf->reshape_progress = MaxSector;
NeilBrown1e3fa9b2012-03-13 11:21:18 +11006593 mddev->reshape_position = MaxSector;
NeilBrownba8805b2013-11-14 15:16:15 +11006594 write_seqcount_end(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006595 spin_unlock_irq(&conf->device_lock);
6596 return -EAGAIN;
6597 }
NeilBrownc8f517c2009-03-31 15:28:40 +11006598 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08006599 md_wakeup_thread(mddev->sync_thread);
6600 md_new_event(mddev);
6601 return 0;
6602}
NeilBrown29269552006-03-27 01:18:10 -08006603
NeilBrownec32a2b2009-03-31 15:17:38 +11006604/* This is called from the reshape thread and should make any
6605 * changes needed in 'conf'
6606 */
NeilBrownd1688a62011-10-11 16:49:52 +11006607static void end_reshape(struct r5conf *conf)
NeilBrown29269552006-03-27 01:18:10 -08006608{
NeilBrown29269552006-03-27 01:18:10 -08006609
NeilBrownf6705572006-03-27 01:18:11 -08006610 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
NeilBrown05616be2012-05-21 09:27:00 +10006611 struct md_rdev *rdev;
Dan Williams80c3a6c2009-03-17 18:10:40 -07006612
NeilBrownf6705572006-03-27 01:18:11 -08006613 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11006614 conf->previous_raid_disks = conf->raid_disks;
NeilBrown05616be2012-05-21 09:27:00 +10006615 rdev_for_each(rdev, conf->mddev)
6616 rdev->data_offset = rdev->new_data_offset;
6617 smp_wmb();
NeilBrownfef9c612009-03-31 15:16:46 +11006618 conf->reshape_progress = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08006619 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11006620 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07006621
6622 /* read-ahead size must cover two whole stripes, which is
6623 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
6624 */
NeilBrown4a5add42010-06-01 19:37:28 +10006625 if (conf->mddev->queue) {
NeilBrowncea9c222009-03-31 15:15:05 +11006626 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10006627 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11006628 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07006629 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6630 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
6631 }
NeilBrown29269552006-03-27 01:18:10 -08006632 }
NeilBrown29269552006-03-27 01:18:10 -08006633}
6634
NeilBrownec32a2b2009-03-31 15:17:38 +11006635/* This is called from the raid5d thread with mddev_lock held.
6636 * It makes config changes to the device.
6637 */
NeilBrownfd01b882011-10-11 16:47:53 +11006638static void raid5_finish_reshape(struct mddev *mddev)
NeilBrowncea9c222009-03-31 15:15:05 +11006639{
NeilBrownd1688a62011-10-11 16:49:52 +11006640 struct r5conf *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11006641
6642 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6643
NeilBrownec32a2b2009-03-31 15:17:38 +11006644 if (mddev->delta_disks > 0) {
6645 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6646 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10006647 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11006648 } else {
6649 int d;
NeilBrown908f4fb2011-12-23 10:17:50 +11006650 spin_lock_irq(&conf->device_lock);
6651 mddev->degraded = calc_degraded(conf);
6652 spin_unlock_irq(&conf->device_lock);
NeilBrownec32a2b2009-03-31 15:17:38 +11006653 for (d = conf->raid_disks ;
6654 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10006655 d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11006656 struct md_rdev *rdev = conf->disks[d].rdev;
NeilBrownda7613b2012-05-22 13:55:33 +10006657 if (rdev)
6658 clear_bit(In_sync, &rdev->flags);
6659 rdev = conf->disks[d].replacement;
6660 if (rdev)
6661 clear_bit(In_sync, &rdev->flags);
NeilBrown1a67dde2009-08-13 10:41:49 +10006662 }
NeilBrowncea9c222009-03-31 15:15:05 +11006663 }
NeilBrown88ce4932009-03-31 15:24:23 +11006664 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10006665 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11006666 mddev->reshape_position = MaxSector;
6667 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10006668 mddev->reshape_backwards = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11006669 }
6670}
6671
NeilBrownfd01b882011-10-11 16:47:53 +11006672static void raid5_quiesce(struct mddev *mddev, int state)
NeilBrown72626682005-09-09 16:23:54 -07006673{
NeilBrownd1688a62011-10-11 16:49:52 +11006674 struct r5conf *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07006675
6676 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08006677 case 2: /* resume for a suspend */
6678 wake_up(&conf->wait_for_overlap);
6679 break;
6680
NeilBrown72626682005-09-09 16:23:54 -07006681 case 1: /* stop all writes */
Shaohua Li566c09c2013-11-14 15:16:17 +11006682 lock_all_device_hash_locks_irq(conf);
NeilBrown64bd6602009-08-03 10:59:58 +10006683 /* '2' tells resync/reshape to pause so that all
6684 * active stripes can drain
6685 */
6686 conf->quiesce = 2;
Shaohua Li566c09c2013-11-14 15:16:17 +11006687 wait_event_cmd(conf->wait_for_stripe,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08006688 atomic_read(&conf->active_stripes) == 0 &&
6689 atomic_read(&conf->active_aligned_reads) == 0,
Shaohua Li566c09c2013-11-14 15:16:17 +11006690 unlock_all_device_hash_locks_irq(conf),
6691 lock_all_device_hash_locks_irq(conf));
NeilBrown64bd6602009-08-03 10:59:58 +10006692 conf->quiesce = 1;
Shaohua Li566c09c2013-11-14 15:16:17 +11006693 unlock_all_device_hash_locks_irq(conf);
NeilBrown64bd6602009-08-03 10:59:58 +10006694 /* allow reshape to continue */
6695 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07006696 break;
6697
6698 case 0: /* re-enable writes */
Shaohua Li566c09c2013-11-14 15:16:17 +11006699 lock_all_device_hash_locks_irq(conf);
NeilBrown72626682005-09-09 16:23:54 -07006700 conf->quiesce = 0;
6701 wake_up(&conf->wait_for_stripe);
NeilBrowne464eaf2006-03-27 01:18:14 -08006702 wake_up(&conf->wait_for_overlap);
Shaohua Li566c09c2013-11-14 15:16:17 +11006703 unlock_all_device_hash_locks_irq(conf);
NeilBrown72626682005-09-09 16:23:54 -07006704 break;
6705 }
NeilBrown72626682005-09-09 16:23:54 -07006706}
NeilBrownb15c2e52006-01-06 00:20:16 -08006707
NeilBrownd562b0c2009-03-31 14:39:39 +11006708
NeilBrownfd01b882011-10-11 16:47:53 +11006709static void *raid45_takeover_raid0(struct mddev *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11006710{
NeilBrowne373ab12011-10-11 16:48:59 +11006711 struct r0conf *raid0_conf = mddev->private;
Randy Dunlapd76c8422011-04-21 09:07:26 -07006712 sector_t sectors;
Trela Maciej54071b32010-03-08 16:02:42 +11006713
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006714 /* for raid0 takeover only one zone is supported */
NeilBrowne373ab12011-10-11 16:48:59 +11006715 if (raid0_conf->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10006716 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
6717 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006718 return ERR_PTR(-EINVAL);
6719 }
6720
NeilBrowne373ab12011-10-11 16:48:59 +11006721 sectors = raid0_conf->strip_zone[0].zone_end;
6722 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
NeilBrown3b71bd92011-04-20 15:38:18 +10006723 mddev->dev_sectors = sectors;
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006724 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11006725 mddev->new_layout = ALGORITHM_PARITY_N;
6726 mddev->new_chunk_sectors = mddev->chunk_sectors;
6727 mddev->raid_disks += 1;
6728 mddev->delta_disks = 1;
6729 /* make sure it will be not marked as dirty */
6730 mddev->recovery_cp = MaxSector;
6731
6732 return setup_conf(mddev);
6733}
6734
6735
NeilBrownfd01b882011-10-11 16:47:53 +11006736static void *raid5_takeover_raid1(struct mddev *mddev)
NeilBrownd562b0c2009-03-31 14:39:39 +11006737{
6738 int chunksect;
6739
6740 if (mddev->raid_disks != 2 ||
6741 mddev->degraded > 1)
6742 return ERR_PTR(-EINVAL);
6743
6744 /* Should check if there are write-behind devices? */
6745
6746 chunksect = 64*2; /* 64K by default */
6747
6748 /* The array must be an exact multiple of chunksize */
6749 while (chunksect && (mddev->array_sectors & (chunksect-1)))
6750 chunksect >>= 1;
6751
6752 if ((chunksect<<9) < STRIPE_SIZE)
6753 /* array size does not allow a suitable chunk size */
6754 return ERR_PTR(-EINVAL);
6755
6756 mddev->new_level = 5;
6757 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10006758 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11006759
6760 return setup_conf(mddev);
6761}
6762
NeilBrownfd01b882011-10-11 16:47:53 +11006763static void *raid5_takeover_raid6(struct mddev *mddev)
NeilBrownfc9739c2009-03-31 14:57:20 +11006764{
6765 int new_layout;
6766
6767 switch (mddev->layout) {
6768 case ALGORITHM_LEFT_ASYMMETRIC_6:
6769 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
6770 break;
6771 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6772 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
6773 break;
6774 case ALGORITHM_LEFT_SYMMETRIC_6:
6775 new_layout = ALGORITHM_LEFT_SYMMETRIC;
6776 break;
6777 case ALGORITHM_RIGHT_SYMMETRIC_6:
6778 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
6779 break;
6780 case ALGORITHM_PARITY_0_6:
6781 new_layout = ALGORITHM_PARITY_0;
6782 break;
6783 case ALGORITHM_PARITY_N:
6784 new_layout = ALGORITHM_PARITY_N;
6785 break;
6786 default:
6787 return ERR_PTR(-EINVAL);
6788 }
6789 mddev->new_level = 5;
6790 mddev->new_layout = new_layout;
6791 mddev->delta_disks = -1;
6792 mddev->raid_disks -= 1;
6793 return setup_conf(mddev);
6794}
6795
NeilBrownd562b0c2009-03-31 14:39:39 +11006796
NeilBrownfd01b882011-10-11 16:47:53 +11006797static int raid5_check_reshape(struct mddev *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11006798{
NeilBrown88ce4932009-03-31 15:24:23 +11006799 /* For a 2-drive array, the layout and chunk size can be changed
6800 * immediately as not restriping is needed.
6801 * For larger arrays we record the new value - after validation
6802 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11006803 */
NeilBrownd1688a62011-10-11 16:49:52 +11006804 struct r5conf *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10006805 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11006806
NeilBrown597a7112009-06-18 08:47:42 +10006807 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11006808 return -EINVAL;
6809 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10006810 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11006811 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006812 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11006813 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006814 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11006815 /* not factor of array size */
6816 return -EINVAL;
6817 }
6818
6819 /* They look valid */
6820
NeilBrown88ce4932009-03-31 15:24:23 +11006821 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10006822 /* can make the change immediately */
6823 if (mddev->new_layout >= 0) {
6824 conf->algorithm = mddev->new_layout;
6825 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11006826 }
6827 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10006828 conf->chunk_sectors = new_chunk ;
6829 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11006830 }
6831 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6832 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11006833 }
NeilBrown50ac1682009-06-18 08:47:55 +10006834 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11006835}
6836
NeilBrownfd01b882011-10-11 16:47:53 +11006837static int raid6_check_reshape(struct mddev *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11006838{
NeilBrown597a7112009-06-18 08:47:42 +10006839 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10006840
NeilBrown597a7112009-06-18 08:47:42 +10006841 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11006842 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11006843 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10006844 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11006845 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006846 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11006847 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006848 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11006849 /* not factor of array size */
6850 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11006851 }
NeilBrown88ce4932009-03-31 15:24:23 +11006852
6853 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10006854 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11006855}
6856
NeilBrownfd01b882011-10-11 16:47:53 +11006857static void *raid5_takeover(struct mddev *mddev)
NeilBrownd562b0c2009-03-31 14:39:39 +11006858{
6859 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006860 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11006861 * raid1 - if there are two drives. We need to know the chunk size
6862 * raid4 - trivial - just use a raid4 layout.
6863 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11006864 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006865 if (mddev->level == 0)
6866 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11006867 if (mddev->level == 1)
6868 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11006869 if (mddev->level == 4) {
6870 mddev->new_layout = ALGORITHM_PARITY_N;
6871 mddev->new_level = 5;
6872 return setup_conf(mddev);
6873 }
NeilBrownfc9739c2009-03-31 14:57:20 +11006874 if (mddev->level == 6)
6875 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11006876
6877 return ERR_PTR(-EINVAL);
6878}
6879
NeilBrownfd01b882011-10-11 16:47:53 +11006880static void *raid4_takeover(struct mddev *mddev)
NeilBrowna78d38a2010-03-22 16:53:49 +11006881{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006882 /* raid4 can take over:
6883 * raid0 - if there is only one strip zone
6884 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11006885 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006886 if (mddev->level == 0)
6887 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11006888 if (mddev->level == 5 &&
6889 mddev->layout == ALGORITHM_PARITY_N) {
6890 mddev->new_layout = 0;
6891 mddev->new_level = 4;
6892 return setup_conf(mddev);
6893 }
6894 return ERR_PTR(-EINVAL);
6895}
NeilBrownd562b0c2009-03-31 14:39:39 +11006896
NeilBrown84fc4b52011-10-11 16:49:58 +11006897static struct md_personality raid5_personality;
NeilBrown245f46c2009-03-31 14:39:39 +11006898
NeilBrownfd01b882011-10-11 16:47:53 +11006899static void *raid6_takeover(struct mddev *mddev)
NeilBrown245f46c2009-03-31 14:39:39 +11006900{
6901 /* Currently can only take over a raid5. We map the
6902 * personality to an equivalent raid6 personality
6903 * with the Q block at the end.
6904 */
6905 int new_layout;
6906
6907 if (mddev->pers != &raid5_personality)
6908 return ERR_PTR(-EINVAL);
6909 if (mddev->degraded > 1)
6910 return ERR_PTR(-EINVAL);
6911 if (mddev->raid_disks > 253)
6912 return ERR_PTR(-EINVAL);
6913 if (mddev->raid_disks < 3)
6914 return ERR_PTR(-EINVAL);
6915
6916 switch (mddev->layout) {
6917 case ALGORITHM_LEFT_ASYMMETRIC:
6918 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
6919 break;
6920 case ALGORITHM_RIGHT_ASYMMETRIC:
6921 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
6922 break;
6923 case ALGORITHM_LEFT_SYMMETRIC:
6924 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
6925 break;
6926 case ALGORITHM_RIGHT_SYMMETRIC:
6927 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
6928 break;
6929 case ALGORITHM_PARITY_0:
6930 new_layout = ALGORITHM_PARITY_0_6;
6931 break;
6932 case ALGORITHM_PARITY_N:
6933 new_layout = ALGORITHM_PARITY_N;
6934 break;
6935 default:
6936 return ERR_PTR(-EINVAL);
6937 }
6938 mddev->new_level = 6;
6939 mddev->new_layout = new_layout;
6940 mddev->delta_disks = 1;
6941 mddev->raid_disks += 1;
6942 return setup_conf(mddev);
6943}
6944
6945
NeilBrown84fc4b52011-10-11 16:49:58 +11006946static struct md_personality raid6_personality =
NeilBrown16a53ec2006-06-26 00:27:38 -07006947{
6948 .name = "raid6",
6949 .level = 6,
6950 .owner = THIS_MODULE,
6951 .make_request = make_request,
6952 .run = run,
6953 .stop = stop,
6954 .status = status,
6955 .error_handler = error,
6956 .hot_add_disk = raid5_add_disk,
6957 .hot_remove_disk= raid5_remove_disk,
6958 .spare_active = raid5_spare_active,
6959 .sync_request = sync_request,
6960 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07006961 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10006962 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08006963 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11006964 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07006965 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11006966 .takeover = raid6_takeover,
NeilBrown16a53ec2006-06-26 00:27:38 -07006967};
NeilBrown84fc4b52011-10-11 16:49:58 +11006968static struct md_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07006969{
6970 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08006971 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006972 .owner = THIS_MODULE,
6973 .make_request = make_request,
6974 .run = run,
6975 .stop = stop,
6976 .status = status,
6977 .error_handler = error,
6978 .hot_add_disk = raid5_add_disk,
6979 .hot_remove_disk= raid5_remove_disk,
6980 .spare_active = raid5_spare_active,
6981 .sync_request = sync_request,
6982 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07006983 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08006984 .check_reshape = raid5_check_reshape,
6985 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11006986 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07006987 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11006988 .takeover = raid5_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006989};
6990
NeilBrown84fc4b52011-10-11 16:49:58 +11006991static struct md_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07006992{
NeilBrown2604b702006-01-06 00:20:36 -08006993 .name = "raid4",
6994 .level = 4,
6995 .owner = THIS_MODULE,
6996 .make_request = make_request,
6997 .run = run,
6998 .stop = stop,
6999 .status = status,
7000 .error_handler = error,
7001 .hot_add_disk = raid5_add_disk,
7002 .hot_remove_disk= raid5_remove_disk,
7003 .spare_active = raid5_spare_active,
7004 .sync_request = sync_request,
7005 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07007006 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08007007 .check_reshape = raid5_check_reshape,
7008 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11007009 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08007010 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11007011 .takeover = raid4_takeover,
NeilBrown2604b702006-01-06 00:20:36 -08007012};
7013
7014static int __init raid5_init(void)
7015{
Shaohua Li851c30c2013-08-28 14:30:16 +08007016 raid5_wq = alloc_workqueue("raid5wq",
7017 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
7018 if (!raid5_wq)
7019 return -ENOMEM;
NeilBrown16a53ec2006-06-26 00:27:38 -07007020 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08007021 register_md_personality(&raid5_personality);
7022 register_md_personality(&raid4_personality);
7023 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007024}
7025
NeilBrown2604b702006-01-06 00:20:36 -08007026static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007027{
NeilBrown16a53ec2006-06-26 00:27:38 -07007028 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08007029 unregister_md_personality(&raid5_personality);
7030 unregister_md_personality(&raid4_personality);
Shaohua Li851c30c2013-08-28 14:30:16 +08007031 destroy_workqueue(raid5_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007032}
7033
7034module_init(raid5_init);
7035module_exit(raid5_exit);
7036MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11007037MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007038MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08007039MODULE_ALIAS("md-raid5");
7040MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08007041MODULE_ALIAS("md-level-5");
7042MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07007043MODULE_ALIAS("md-personality-8"); /* RAID6 */
7044MODULE_ALIAS("md-raid6");
7045MODULE_ALIAS("md-level-6");
7046
7047/* This used to be two separate modules, they were: */
7048MODULE_ALIAS("raid5");
7049MODULE_ALIAS("raid6");