blob: c1b0d52bfcb0f7b014bbc48f4823e3b9b028dcce [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
NeilBrown8e0e99b2014-10-02 13:45:00 +100067static bool devices_handle_discard_safely = false;
68module_param(devices_handle_discard_safely, bool, 0644);
69MODULE_PARM_DESC(devices_handle_discard_safely,
70 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
Shaohua Li851c30c2013-08-28 14:30:16 +080071static struct workqueue_struct *raid5_wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072/*
73 * Stripe cache
74 */
75
76#define NR_STRIPES 256
77#define STRIPE_SIZE PAGE_SIZE
78#define STRIPE_SHIFT (PAGE_SHIFT - 9)
79#define STRIPE_SECTORS (STRIPE_SIZE>>9)
80#define IO_THRESHOLD 1
Dan Williams8b3e6cd2008-04-28 02:15:53 -070081#define BYPASS_THRESHOLD 1
NeilBrownfccddba2006-01-06 00:20:33 -080082#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
Linus Torvalds1da177e2005-04-16 15:20:36 -070083#define HASH_MASK (NR_HASH - 1)
Shaohua Libfc90cb2013-08-29 15:40:32 +080084#define MAX_STRIPE_BATCH 8
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
NeilBrownd1688a62011-10-11 16:49:52 +110086static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
NeilBrowndb298e12011-10-07 14:23:00 +110087{
88 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
89 return &conf->stripe_hashtbl[hash];
90}
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Shaohua Li566c09c2013-11-14 15:16:17 +110092static inline int stripe_hash_locks_hash(sector_t sect)
93{
94 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
95}
96
97static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
98{
99 spin_lock_irq(conf->hash_locks + hash);
100 spin_lock(&conf->device_lock);
101}
102
103static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
104{
105 spin_unlock(&conf->device_lock);
106 spin_unlock_irq(conf->hash_locks + hash);
107}
108
109static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
110{
111 int i;
112 local_irq_disable();
113 spin_lock(conf->hash_locks);
114 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
115 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
116 spin_lock(&conf->device_lock);
117}
118
119static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
120{
121 int i;
122 spin_unlock(&conf->device_lock);
123 for (i = NR_STRIPE_HASH_LOCKS; i; i--)
124 spin_unlock(conf->hash_locks + i - 1);
125 local_irq_enable();
126}
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128/* bio's attached to a stripe+device for I/O are linked together in bi_sector
129 * order without overlap. There may be several bio's per stripe+device, and
130 * a bio could span several devices.
131 * When walking this list for a particular stripe+device, we must never proceed
132 * beyond a bio that extends past this device, as the next bio might no longer
133 * be valid.
NeilBrowndb298e12011-10-07 14:23:00 +1100134 * This function is used to determine the 'next' bio in the list, given the sector
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 * of the current stripe+device
136 */
NeilBrowndb298e12011-10-07 14:23:00 +1100137static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
138{
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800139 int sectors = bio_sectors(bio);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700140 if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS)
NeilBrowndb298e12011-10-07 14:23:00 +1100141 return bio->bi_next;
142 else
143 return NULL;
144}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Jens Axboe960e7392008-08-15 10:41:18 +0200146/*
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200147 * We maintain a biased count of active stripes in the bottom 16 bits of
148 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
Jens Axboe960e7392008-08-15 10:41:18 +0200149 */
Shaohua Lie7836bd62012-07-19 16:01:31 +1000150static inline int raid5_bi_processed_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200151{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000152 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
153 return (atomic_read(segments) >> 16) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200154}
155
Shaohua Lie7836bd62012-07-19 16:01:31 +1000156static inline int raid5_dec_bi_active_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200157{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000158 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
159 return atomic_sub_return(1, segments) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200160}
161
Shaohua Lie7836bd62012-07-19 16:01:31 +1000162static inline void raid5_inc_bi_active_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200163{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000164 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
165 atomic_inc(segments);
Jens Axboe960e7392008-08-15 10:41:18 +0200166}
167
Shaohua Lie7836bd62012-07-19 16:01:31 +1000168static inline void raid5_set_bi_processed_stripes(struct bio *bio,
169 unsigned int cnt)
Jens Axboe960e7392008-08-15 10:41:18 +0200170{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000171 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
172 int old, new;
Jens Axboe960e7392008-08-15 10:41:18 +0200173
Shaohua Lie7836bd62012-07-19 16:01:31 +1000174 do {
175 old = atomic_read(segments);
176 new = (old & 0xffff) | (cnt << 16);
177 } while (atomic_cmpxchg(segments, old, new) != old);
Jens Axboe960e7392008-08-15 10:41:18 +0200178}
179
Shaohua Lie7836bd62012-07-19 16:01:31 +1000180static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt)
Jens Axboe960e7392008-08-15 10:41:18 +0200181{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000182 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
183 atomic_set(segments, cnt);
Jens Axboe960e7392008-08-15 10:41:18 +0200184}
185
NeilBrownd0dabf72009-03-31 14:39:38 +1100186/* Find first data disk in a raid6 stripe */
187static inline int raid6_d0(struct stripe_head *sh)
188{
NeilBrown67cc2b82009-03-31 14:39:38 +1100189 if (sh->ddf_layout)
190 /* ddf always start from first device */
191 return 0;
192 /* md starts just after Q block */
NeilBrownd0dabf72009-03-31 14:39:38 +1100193 if (sh->qd_idx == sh->disks - 1)
194 return 0;
195 else
196 return sh->qd_idx + 1;
197}
NeilBrown16a53ec2006-06-26 00:27:38 -0700198static inline int raid6_next_disk(int disk, int raid_disks)
199{
200 disk++;
201 return (disk < raid_disks) ? disk : 0;
202}
Dan Williamsa4456852007-07-09 11:56:43 -0700203
NeilBrownd0dabf72009-03-31 14:39:38 +1100204/* When walking through the disks in a raid5, starting at raid6_d0,
205 * We need to map each disk to a 'slot', where the data disks are slot
206 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
207 * is raid_disks-1. This help does that mapping.
208 */
NeilBrown67cc2b82009-03-31 14:39:38 +1100209static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
210 int *count, int syndrome_disks)
NeilBrownd0dabf72009-03-31 14:39:38 +1100211{
Dan Williams66295422009-10-19 18:09:32 -0700212 int slot = *count;
NeilBrown67cc2b82009-03-31 14:39:38 +1100213
NeilBrowne4424fe2009-10-16 16:27:34 +1100214 if (sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700215 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100216 if (idx == sh->pd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100217 return syndrome_disks;
NeilBrownd0dabf72009-03-31 14:39:38 +1100218 if (idx == sh->qd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100219 return syndrome_disks + 1;
NeilBrowne4424fe2009-10-16 16:27:34 +1100220 if (!sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700221 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100222 return slot;
223}
224
Dan Williamsa4456852007-07-09 11:56:43 -0700225static void return_io(struct bio *return_bi)
226{
227 struct bio *bi = return_bi;
228 while (bi) {
Dan Williamsa4456852007-07-09 11:56:43 -0700229
230 return_bi = bi->bi_next;
231 bi->bi_next = NULL;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700232 bi->bi_iter.bi_size = 0;
Linus Torvalds0a82a8d2013-04-18 09:00:26 -0700233 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
234 bi, 0);
Neil Brown0e13fe232008-06-28 08:31:20 +1000235 bio_endio(bi, 0);
Dan Williamsa4456852007-07-09 11:56:43 -0700236 bi = return_bi;
237 }
238}
239
NeilBrownd1688a62011-10-11 16:49:52 +1100240static void print_raid5_conf (struct r5conf *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Dan Williams600aa102008-06-28 08:32:05 +1000242static int stripe_operations_active(struct stripe_head *sh)
243{
244 return sh->check_state || sh->reconstruct_state ||
245 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
246 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
247}
248
Shaohua Li851c30c2013-08-28 14:30:16 +0800249static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
250{
251 struct r5conf *conf = sh->raid_conf;
252 struct r5worker_group *group;
Shaohua Libfc90cb2013-08-29 15:40:32 +0800253 int thread_cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +0800254 int i, cpu = sh->cpu;
255
256 if (!cpu_online(cpu)) {
257 cpu = cpumask_any(cpu_online_mask);
258 sh->cpu = cpu;
259 }
260
261 if (list_empty(&sh->lru)) {
262 struct r5worker_group *group;
263 group = conf->worker_groups + cpu_to_group(cpu);
264 list_add_tail(&sh->lru, &group->handle_list);
Shaohua Libfc90cb2013-08-29 15:40:32 +0800265 group->stripes_cnt++;
266 sh->group = group;
Shaohua Li851c30c2013-08-28 14:30:16 +0800267 }
268
269 if (conf->worker_cnt_per_group == 0) {
270 md_wakeup_thread(conf->mddev->thread);
271 return;
272 }
273
274 group = conf->worker_groups + cpu_to_group(sh->cpu);
275
Shaohua Libfc90cb2013-08-29 15:40:32 +0800276 group->workers[0].working = true;
277 /* at least one worker should run to avoid race */
278 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
279
280 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
281 /* wakeup more workers */
282 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
283 if (group->workers[i].working == false) {
284 group->workers[i].working = true;
285 queue_work_on(sh->cpu, raid5_wq,
286 &group->workers[i].work);
287 thread_cnt--;
288 }
289 }
Shaohua Li851c30c2013-08-28 14:30:16 +0800290}
291
Shaohua Li566c09c2013-11-14 15:16:17 +1100292static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
293 struct list_head *temp_inactive_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
Shaohua Li4eb788d2012-07-19 16:01:31 +1000295 BUG_ON(!list_empty(&sh->lru));
296 BUG_ON(atomic_read(&conf->active_stripes)==0);
297 if (test_bit(STRIPE_HANDLE, &sh->state)) {
298 if (test_bit(STRIPE_DELAYED, &sh->state) &&
NeilBrown67f45542014-05-28 13:39:22 +1000299 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Shaohua Li4eb788d2012-07-19 16:01:31 +1000300 list_add_tail(&sh->lru, &conf->delayed_list);
NeilBrown67f45542014-05-28 13:39:22 +1000301 if (atomic_read(&conf->preread_active_stripes)
302 < IO_THRESHOLD)
303 md_wakeup_thread(conf->mddev->thread);
304 } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
Shaohua Li4eb788d2012-07-19 16:01:31 +1000305 sh->bm_seq - conf->seq_write > 0)
306 list_add_tail(&sh->lru, &conf->bitmap_list);
307 else {
308 clear_bit(STRIPE_DELAYED, &sh->state);
309 clear_bit(STRIPE_BIT_DELAY, &sh->state);
Shaohua Li851c30c2013-08-28 14:30:16 +0800310 if (conf->worker_cnt_per_group == 0) {
311 list_add_tail(&sh->lru, &conf->handle_list);
312 } else {
313 raid5_wakeup_stripe_thread(sh);
314 return;
315 }
Shaohua Li4eb788d2012-07-19 16:01:31 +1000316 }
317 md_wakeup_thread(conf->mddev->thread);
318 } else {
319 BUG_ON(stripe_operations_active(sh));
320 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
321 if (atomic_dec_return(&conf->preread_active_stripes)
322 < IO_THRESHOLD)
323 md_wakeup_thread(conf->mddev->thread);
324 atomic_dec(&conf->active_stripes);
Shaohua Li566c09c2013-11-14 15:16:17 +1100325 if (!test_bit(STRIPE_EXPANDING, &sh->state))
326 list_add_tail(&sh->lru, temp_inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
328}
NeilBrownd0dabf72009-03-31 14:39:38 +1100329
Shaohua Li566c09c2013-11-14 15:16:17 +1100330static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
331 struct list_head *temp_inactive_list)
Shaohua Li4eb788d2012-07-19 16:01:31 +1000332{
333 if (atomic_dec_and_test(&sh->count))
Shaohua Li566c09c2013-11-14 15:16:17 +1100334 do_release_stripe(conf, sh, temp_inactive_list);
335}
336
337/*
338 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
339 *
340 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
341 * given time. Adding stripes only takes device lock, while deleting stripes
342 * only takes hash lock.
343 */
344static void release_inactive_stripe_list(struct r5conf *conf,
345 struct list_head *temp_inactive_list,
346 int hash)
347{
348 int size;
349 bool do_wakeup = false;
350 unsigned long flags;
351
352 if (hash == NR_STRIPE_HASH_LOCKS) {
353 size = NR_STRIPE_HASH_LOCKS;
354 hash = NR_STRIPE_HASH_LOCKS - 1;
355 } else
356 size = 1;
357 while (size) {
358 struct list_head *list = &temp_inactive_list[size - 1];
359
360 /*
361 * We don't hold any lock here yet, get_active_stripe() might
362 * remove stripes from the list
363 */
364 if (!list_empty_careful(list)) {
365 spin_lock_irqsave(conf->hash_locks + hash, flags);
Shaohua Li4bda5562013-11-14 15:16:17 +1100366 if (list_empty(conf->inactive_list + hash) &&
367 !list_empty(list))
368 atomic_dec(&conf->empty_inactive_list_nr);
Shaohua Li566c09c2013-11-14 15:16:17 +1100369 list_splice_tail_init(list, conf->inactive_list + hash);
370 do_wakeup = true;
371 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
372 }
373 size--;
374 hash--;
375 }
376
377 if (do_wakeup) {
378 wake_up(&conf->wait_for_stripe);
379 if (conf->retry_read_aligned)
380 md_wakeup_thread(conf->mddev->thread);
381 }
Shaohua Li4eb788d2012-07-19 16:01:31 +1000382}
383
Shaohua Li773ca822013-08-27 17:50:39 +0800384/* should hold conf->device_lock already */
Shaohua Li566c09c2013-11-14 15:16:17 +1100385static int release_stripe_list(struct r5conf *conf,
386 struct list_head *temp_inactive_list)
Shaohua Li773ca822013-08-27 17:50:39 +0800387{
388 struct stripe_head *sh;
389 int count = 0;
390 struct llist_node *head;
391
392 head = llist_del_all(&conf->released_stripes);
Shaohua Lid265d9d2013-08-28 14:29:05 +0800393 head = llist_reverse_order(head);
Shaohua Li773ca822013-08-27 17:50:39 +0800394 while (head) {
Shaohua Li566c09c2013-11-14 15:16:17 +1100395 int hash;
396
Shaohua Li773ca822013-08-27 17:50:39 +0800397 sh = llist_entry(head, struct stripe_head, release_list);
398 head = llist_next(head);
399 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
400 smp_mb();
401 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
402 /*
403 * Don't worry the bit is set here, because if the bit is set
404 * again, the count is always > 1. This is true for
405 * STRIPE_ON_UNPLUG_LIST bit too.
406 */
Shaohua Li566c09c2013-11-14 15:16:17 +1100407 hash = sh->hash_lock_index;
408 __release_stripe(conf, sh, &temp_inactive_list[hash]);
Shaohua Li773ca822013-08-27 17:50:39 +0800409 count++;
410 }
411
412 return count;
413}
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415static void release_stripe(struct stripe_head *sh)
416{
NeilBrownd1688a62011-10-11 16:49:52 +1100417 struct r5conf *conf = sh->raid_conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 unsigned long flags;
Shaohua Li566c09c2013-11-14 15:16:17 +1100419 struct list_head list;
420 int hash;
Shaohua Li773ca822013-08-27 17:50:39 +0800421 bool wakeup;
NeilBrown16a53ec2006-06-26 00:27:38 -0700422
Eivind Sartocf170f32014-05-28 13:39:23 +1000423 /* Avoid release_list until the last reference.
424 */
425 if (atomic_add_unless(&sh->count, -1, 1))
426 return;
427
majianpengad4068d2013-11-14 15:16:15 +1100428 if (unlikely(!conf->mddev->thread) ||
429 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
Shaohua Li773ca822013-08-27 17:50:39 +0800430 goto slow_path;
431 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
432 if (wakeup)
433 md_wakeup_thread(conf->mddev->thread);
434 return;
435slow_path:
Shaohua Li4eb788d2012-07-19 16:01:31 +1000436 local_irq_save(flags);
Shaohua Li773ca822013-08-27 17:50:39 +0800437 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
Shaohua Li4eb788d2012-07-19 16:01:31 +1000438 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
Shaohua Li566c09c2013-11-14 15:16:17 +1100439 INIT_LIST_HEAD(&list);
440 hash = sh->hash_lock_index;
441 do_release_stripe(conf, sh, &list);
Shaohua Li4eb788d2012-07-19 16:01:31 +1000442 spin_unlock(&conf->device_lock);
Shaohua Li566c09c2013-11-14 15:16:17 +1100443 release_inactive_stripe_list(conf, &list, hash);
Shaohua Li4eb788d2012-07-19 16:01:31 +1000444 }
445 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
NeilBrownfccddba2006-01-06 00:20:33 -0800448static inline void remove_hash(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
Dan Williams45b42332007-07-09 11:56:43 -0700450 pr_debug("remove_hash(), stripe %llu\n",
451 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
NeilBrownfccddba2006-01-06 00:20:33 -0800453 hlist_del_init(&sh->hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
NeilBrownd1688a62011-10-11 16:49:52 +1100456static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
NeilBrownfccddba2006-01-06 00:20:33 -0800458 struct hlist_head *hp = stripe_hash(conf, sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Dan Williams45b42332007-07-09 11:56:43 -0700460 pr_debug("insert_hash(), stripe %llu\n",
461 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
NeilBrownfccddba2006-01-06 00:20:33 -0800463 hlist_add_head(&sh->hash, hp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466/* find an idle stripe, make sure it is unhashed, and return it. */
Shaohua Li566c09c2013-11-14 15:16:17 +1100467static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
469 struct stripe_head *sh = NULL;
470 struct list_head *first;
471
Shaohua Li566c09c2013-11-14 15:16:17 +1100472 if (list_empty(conf->inactive_list + hash))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 goto out;
Shaohua Li566c09c2013-11-14 15:16:17 +1100474 first = (conf->inactive_list + hash)->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 sh = list_entry(first, struct stripe_head, lru);
476 list_del_init(first);
477 remove_hash(sh);
478 atomic_inc(&conf->active_stripes);
Shaohua Li566c09c2013-11-14 15:16:17 +1100479 BUG_ON(hash != sh->hash_lock_index);
Shaohua Li4bda5562013-11-14 15:16:17 +1100480 if (list_empty(conf->inactive_list + hash))
481 atomic_inc(&conf->empty_inactive_list_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482out:
483 return sh;
484}
485
NeilBrowne4e11e32010-06-16 16:45:16 +1000486static void shrink_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
488 struct page *p;
489 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000490 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
NeilBrowne4e11e32010-06-16 16:45:16 +1000492 for (i = 0; i < num ; i++) {
Shaohua Lid592a992014-05-21 17:57:44 +0800493 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 p = sh->dev[i].page;
495 if (!p)
496 continue;
497 sh->dev[i].page = NULL;
NeilBrown2d1f3b52006-01-06 00:20:31 -0800498 put_page(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
500}
501
NeilBrowne4e11e32010-06-16 16:45:16 +1000502static int grow_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503{
504 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000505 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
NeilBrowne4e11e32010-06-16 16:45:16 +1000507 for (i = 0; i < num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 struct page *page;
509
510 if (!(page = alloc_page(GFP_KERNEL))) {
511 return 1;
512 }
513 sh->dev[i].page = page;
Shaohua Lid592a992014-05-21 17:57:44 +0800514 sh->dev[i].orig_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
516 return 0;
517}
518
NeilBrown784052e2009-03-31 15:19:07 +1100519static void raid5_build_block(struct stripe_head *sh, int i, int previous);
NeilBrownd1688a62011-10-11 16:49:52 +1100520static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
NeilBrown911d4ee2009-03-31 14:39:38 +1100521 struct stripe_head *sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
NeilBrownb5663ba2009-03-31 14:39:38 +1100523static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
NeilBrownd1688a62011-10-11 16:49:52 +1100525 struct r5conf *conf = sh->raid_conf;
Shaohua Li566c09c2013-11-14 15:16:17 +1100526 int i, seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200528 BUG_ON(atomic_read(&sh->count) != 0);
529 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
Dan Williams600aa102008-06-28 08:32:05 +1000530 BUG_ON(stripe_operations_active(sh));
Dan Williamsd84e0f12007-01-02 13:52:30 -0700531
Dan Williams45b42332007-07-09 11:56:43 -0700532 pr_debug("init_stripe called, stripe %llu\n",
Markus Stockhausenb8e6a152014-08-23 20:19:27 +1000533 (unsigned long long)sector);
Shaohua Li566c09c2013-11-14 15:16:17 +1100534retry:
535 seq = read_seqcount_begin(&conf->gen_lock);
NeilBrown86b42c72009-03-31 15:19:03 +1100536 sh->generation = conf->generation - previous;
NeilBrownb5663ba2009-03-31 14:39:38 +1100537 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 sh->sector = sector;
NeilBrown911d4ee2009-03-31 14:39:38 +1100539 stripe_set_idx(sector, conf, previous, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 sh->state = 0;
541
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800542 for (i = sh->disks; i--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 struct r5dev *dev = &sh->dev[i];
544
Dan Williamsd84e0f12007-01-02 13:52:30 -0700545 if (dev->toread || dev->read || dev->towrite || dev->written ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 test_bit(R5_LOCKED, &dev->flags)) {
Dan Williamsd84e0f12007-01-02 13:52:30 -0700547 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 (unsigned long long)sh->sector, i, dev->toread,
Dan Williamsd84e0f12007-01-02 13:52:30 -0700549 dev->read, dev->towrite, dev->written,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 test_bit(R5_LOCKED, &dev->flags));
NeilBrown8cfa7b02011-07-27 11:00:36 +1000551 WARN_ON(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 }
553 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +1100554 raid5_build_block(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
Shaohua Li566c09c2013-11-14 15:16:17 +1100556 if (read_seqcount_retry(&conf->gen_lock, seq))
557 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 insert_hash(conf, sh);
Shaohua Li851c30c2013-08-28 14:30:16 +0800559 sh->cpu = smp_processor_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
NeilBrownd1688a62011-10-11 16:49:52 +1100562static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
NeilBrown86b42c72009-03-31 15:19:03 +1100563 short generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
565 struct stripe_head *sh;
566
Dan Williams45b42332007-07-09 11:56:43 -0700567 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800568 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
NeilBrown86b42c72009-03-31 15:19:03 +1100569 if (sh->sector == sector && sh->generation == generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 return sh;
Dan Williams45b42332007-07-09 11:56:43 -0700571 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 return NULL;
573}
574
NeilBrown674806d2010-06-16 17:17:53 +1000575/*
576 * Need to check if array has failed when deciding whether to:
577 * - start an array
578 * - remove non-faulty devices
579 * - add a spare
580 * - allow a reshape
581 * This determination is simple when no reshape is happening.
582 * However if there is a reshape, we need to carefully check
583 * both the before and after sections.
584 * This is because some failed devices may only affect one
585 * of the two sections, and some non-in_sync devices may
586 * be insync in the section most affected by failed devices.
587 */
NeilBrown908f4fb2011-12-23 10:17:50 +1100588static int calc_degraded(struct r5conf *conf)
NeilBrown674806d2010-06-16 17:17:53 +1000589{
NeilBrown908f4fb2011-12-23 10:17:50 +1100590 int degraded, degraded2;
NeilBrown674806d2010-06-16 17:17:53 +1000591 int i;
NeilBrown674806d2010-06-16 17:17:53 +1000592
593 rcu_read_lock();
594 degraded = 0;
595 for (i = 0; i < conf->previous_raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100596 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowne5c86472012-09-19 12:52:30 +1000597 if (rdev && test_bit(Faulty, &rdev->flags))
598 rdev = rcu_dereference(conf->disks[i].replacement);
NeilBrown674806d2010-06-16 17:17:53 +1000599 if (!rdev || test_bit(Faulty, &rdev->flags))
600 degraded++;
601 else if (test_bit(In_sync, &rdev->flags))
602 ;
603 else
604 /* not in-sync or faulty.
605 * If the reshape increases the number of devices,
606 * this is being recovered by the reshape, so
607 * this 'previous' section is not in_sync.
608 * If the number of devices is being reduced however,
609 * the device can only be part of the array if
610 * we are reverting a reshape, so this section will
611 * be in-sync.
612 */
613 if (conf->raid_disks >= conf->previous_raid_disks)
614 degraded++;
615 }
616 rcu_read_unlock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100617 if (conf->raid_disks == conf->previous_raid_disks)
618 return degraded;
NeilBrown674806d2010-06-16 17:17:53 +1000619 rcu_read_lock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100620 degraded2 = 0;
NeilBrown674806d2010-06-16 17:17:53 +1000621 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100622 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowne5c86472012-09-19 12:52:30 +1000623 if (rdev && test_bit(Faulty, &rdev->flags))
624 rdev = rcu_dereference(conf->disks[i].replacement);
NeilBrown674806d2010-06-16 17:17:53 +1000625 if (!rdev || test_bit(Faulty, &rdev->flags))
NeilBrown908f4fb2011-12-23 10:17:50 +1100626 degraded2++;
NeilBrown674806d2010-06-16 17:17:53 +1000627 else if (test_bit(In_sync, &rdev->flags))
628 ;
629 else
630 /* not in-sync or faulty.
631 * If reshape increases the number of devices, this
632 * section has already been recovered, else it
633 * almost certainly hasn't.
634 */
635 if (conf->raid_disks <= conf->previous_raid_disks)
NeilBrown908f4fb2011-12-23 10:17:50 +1100636 degraded2++;
NeilBrown674806d2010-06-16 17:17:53 +1000637 }
638 rcu_read_unlock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100639 if (degraded2 > degraded)
640 return degraded2;
641 return degraded;
642}
643
644static int has_failed(struct r5conf *conf)
645{
646 int degraded;
647
648 if (conf->mddev->reshape_position == MaxSector)
649 return conf->mddev->degraded > conf->max_degraded;
650
651 degraded = calc_degraded(conf);
NeilBrown674806d2010-06-16 17:17:53 +1000652 if (degraded > conf->max_degraded)
653 return 1;
654 return 0;
655}
656
NeilBrownb5663ba2009-03-31 14:39:38 +1100657static struct stripe_head *
NeilBrownd1688a62011-10-11 16:49:52 +1100658get_active_stripe(struct r5conf *conf, sector_t sector,
NeilBrowna8c906c2009-06-09 14:39:59 +1000659 int previous, int noblock, int noquiesce)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 struct stripe_head *sh;
Shaohua Li566c09c2013-11-14 15:16:17 +1100662 int hash = stripe_hash_locks_hash(sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Dan Williams45b42332007-07-09 11:56:43 -0700664 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Shaohua Li566c09c2013-11-14 15:16:17 +1100666 spin_lock_irq(conf->hash_locks + hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668 do {
NeilBrown72626682005-09-09 16:23:54 -0700669 wait_event_lock_irq(conf->wait_for_stripe,
NeilBrowna8c906c2009-06-09 14:39:59 +1000670 conf->quiesce == 0 || noquiesce,
Shaohua Li566c09c2013-11-14 15:16:17 +1100671 *(conf->hash_locks + hash));
NeilBrown86b42c72009-03-31 15:19:03 +1100672 sh = __find_stripe(conf, sector, conf->generation - previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (!sh) {
674 if (!conf->inactive_blocked)
Shaohua Li566c09c2013-11-14 15:16:17 +1100675 sh = get_free_stripe(conf, hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (noblock && sh == NULL)
677 break;
678 if (!sh) {
679 conf->inactive_blocked = 1;
Shaohua Li566c09c2013-11-14 15:16:17 +1100680 wait_event_lock_irq(
681 conf->wait_for_stripe,
682 !list_empty(conf->inactive_list + hash) &&
683 (atomic_read(&conf->active_stripes)
684 < (conf->max_nr_stripes * 3 / 4)
685 || !conf->inactive_blocked),
686 *(conf->hash_locks + hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 conf->inactive_blocked = 0;
NeilBrown7da9d452014-01-22 11:45:03 +1100688 } else {
NeilBrownb5663ba2009-03-31 14:39:38 +1100689 init_stripe(sh, sector, previous);
NeilBrown7da9d452014-01-22 11:45:03 +1100690 atomic_inc(&sh->count);
691 }
Shaohua Lie240c182014-04-09 11:27:42 +0800692 } else if (!atomic_inc_not_zero(&sh->count)) {
NeilBrown6d183de2013-11-28 10:55:27 +1100693 spin_lock(&conf->device_lock);
Shaohua Lie240c182014-04-09 11:27:42 +0800694 if (!atomic_read(&sh->count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 if (!test_bit(STRIPE_HANDLE, &sh->state))
696 atomic_inc(&conf->active_stripes);
NeilBrown5af9bef2014-01-14 15:16:10 +1100697 BUG_ON(list_empty(&sh->lru) &&
698 !test_bit(STRIPE_EXPANDING, &sh->state));
NeilBrown16a53ec2006-06-26 00:27:38 -0700699 list_del_init(&sh->lru);
Shaohua Libfc90cb2013-08-29 15:40:32 +0800700 if (sh->group) {
701 sh->group->stripes_cnt--;
702 sh->group = NULL;
703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 }
NeilBrown7da9d452014-01-22 11:45:03 +1100705 atomic_inc(&sh->count);
NeilBrown6d183de2013-11-28 10:55:27 +1100706 spin_unlock(&conf->device_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 }
708 } while (sh == NULL);
709
Shaohua Li566c09c2013-11-14 15:16:17 +1100710 spin_unlock_irq(conf->hash_locks + hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return sh;
712}
713
NeilBrown05616be2012-05-21 09:27:00 +1000714/* Determine if 'data_offset' or 'new_data_offset' should be used
715 * in this stripe_head.
716 */
717static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
718{
719 sector_t progress = conf->reshape_progress;
720 /* Need a memory barrier to make sure we see the value
721 * of conf->generation, or ->data_offset that was set before
722 * reshape_progress was updated.
723 */
724 smp_rmb();
725 if (progress == MaxSector)
726 return 0;
727 if (sh->generation == conf->generation - 1)
728 return 0;
729 /* We are in a reshape, and this is a new-generation stripe,
730 * so use new_data_offset.
731 */
732 return 1;
733}
734
NeilBrown6712ecf2007-09-27 12:47:43 +0200735static void
736raid5_end_read_request(struct bio *bi, int error);
737static void
738raid5_end_write_request(struct bio *bi, int error);
Dan Williams91c00922007-01-02 13:52:30 -0700739
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000740static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
Dan Williams91c00922007-01-02 13:52:30 -0700741{
NeilBrownd1688a62011-10-11 16:49:52 +1100742 struct r5conf *conf = sh->raid_conf;
Dan Williams91c00922007-01-02 13:52:30 -0700743 int i, disks = sh->disks;
744
745 might_sleep();
746
747 for (i = disks; i--; ) {
748 int rw;
NeilBrown9a3e1102011-12-23 10:17:53 +1100749 int replace_only = 0;
NeilBrown977df362011-12-23 10:17:53 +1100750 struct bio *bi, *rbi;
751 struct md_rdev *rdev, *rrdev = NULL;
Tejun Heoe9c74692010-09-03 11:56:18 +0200752 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
753 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
754 rw = WRITE_FUA;
755 else
756 rw = WRITE;
Shaohua Li9e4447682012-10-11 13:49:49 +1100757 if (test_bit(R5_Discard, &sh->dev[i].flags))
Shaohua Li620125f2012-10-11 13:49:05 +1100758 rw |= REQ_DISCARD;
Tejun Heoe9c74692010-09-03 11:56:18 +0200759 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
Dan Williams91c00922007-01-02 13:52:30 -0700760 rw = READ;
NeilBrown9a3e1102011-12-23 10:17:53 +1100761 else if (test_and_clear_bit(R5_WantReplace,
762 &sh->dev[i].flags)) {
763 rw = WRITE;
764 replace_only = 1;
765 } else
Dan Williams91c00922007-01-02 13:52:30 -0700766 continue;
Shaohua Libc0934f2012-05-22 13:55:05 +1000767 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
768 rw |= REQ_SYNC;
Dan Williams91c00922007-01-02 13:52:30 -0700769
770 bi = &sh->dev[i].req;
NeilBrown977df362011-12-23 10:17:53 +1100771 rbi = &sh->dev[i].rreq; /* For writing to replacement */
Dan Williams91c00922007-01-02 13:52:30 -0700772
Dan Williams91c00922007-01-02 13:52:30 -0700773 rcu_read_lock();
NeilBrown9a3e1102011-12-23 10:17:53 +1100774 rrdev = rcu_dereference(conf->disks[i].replacement);
NeilBrowndd054fc2011-12-23 10:17:53 +1100775 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
776 rdev = rcu_dereference(conf->disks[i].rdev);
777 if (!rdev) {
778 rdev = rrdev;
779 rrdev = NULL;
780 }
NeilBrown9a3e1102011-12-23 10:17:53 +1100781 if (rw & WRITE) {
782 if (replace_only)
783 rdev = NULL;
NeilBrowndd054fc2011-12-23 10:17:53 +1100784 if (rdev == rrdev)
785 /* We raced and saw duplicates */
786 rrdev = NULL;
NeilBrown9a3e1102011-12-23 10:17:53 +1100787 } else {
NeilBrowndd054fc2011-12-23 10:17:53 +1100788 if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev)
NeilBrown9a3e1102011-12-23 10:17:53 +1100789 rdev = rrdev;
790 rrdev = NULL;
791 }
NeilBrown977df362011-12-23 10:17:53 +1100792
Dan Williams91c00922007-01-02 13:52:30 -0700793 if (rdev && test_bit(Faulty, &rdev->flags))
794 rdev = NULL;
795 if (rdev)
796 atomic_inc(&rdev->nr_pending);
NeilBrown977df362011-12-23 10:17:53 +1100797 if (rrdev && test_bit(Faulty, &rrdev->flags))
798 rrdev = NULL;
799 if (rrdev)
800 atomic_inc(&rrdev->nr_pending);
Dan Williams91c00922007-01-02 13:52:30 -0700801 rcu_read_unlock();
802
NeilBrown73e92e52011-07-28 11:39:22 +1000803 /* We have already checked bad blocks for reads. Now
NeilBrown977df362011-12-23 10:17:53 +1100804 * need to check for writes. We never accept write errors
805 * on the replacement, so we don't to check rrdev.
NeilBrown73e92e52011-07-28 11:39:22 +1000806 */
807 while ((rw & WRITE) && rdev &&
808 test_bit(WriteErrorSeen, &rdev->flags)) {
809 sector_t first_bad;
810 int bad_sectors;
811 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
812 &first_bad, &bad_sectors);
813 if (!bad)
814 break;
815
816 if (bad < 0) {
817 set_bit(BlockedBadBlocks, &rdev->flags);
818 if (!conf->mddev->external &&
819 conf->mddev->flags) {
820 /* It is very unlikely, but we might
821 * still need to write out the
822 * bad block log - better give it
823 * a chance*/
824 md_check_recovery(conf->mddev);
825 }
majianpeng18507532012-07-03 12:11:54 +1000826 /*
827 * Because md_wait_for_blocked_rdev
828 * will dec nr_pending, we must
829 * increment it first.
830 */
831 atomic_inc(&rdev->nr_pending);
NeilBrown73e92e52011-07-28 11:39:22 +1000832 md_wait_for_blocked_rdev(rdev, conf->mddev);
833 } else {
834 /* Acknowledged bad block - skip the write */
835 rdev_dec_pending(rdev, conf->mddev);
836 rdev = NULL;
837 }
838 }
839
Dan Williams91c00922007-01-02 13:52:30 -0700840 if (rdev) {
NeilBrown9a3e1102011-12-23 10:17:53 +1100841 if (s->syncing || s->expanding || s->expanded
842 || s->replacing)
Dan Williams91c00922007-01-02 13:52:30 -0700843 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
844
Dan Williams2b7497f2008-06-28 08:31:52 +1000845 set_bit(STRIPE_IO_STARTED, &sh->state);
846
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700847 bio_reset(bi);
Dan Williams91c00922007-01-02 13:52:30 -0700848 bi->bi_bdev = rdev->bdev;
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700849 bi->bi_rw = rw;
850 bi->bi_end_io = (rw & WRITE)
851 ? raid5_end_write_request
852 : raid5_end_read_request;
853 bi->bi_private = sh;
854
Dan Williams91c00922007-01-02 13:52:30 -0700855 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700856 __func__, (unsigned long long)sh->sector,
Dan Williams91c00922007-01-02 13:52:30 -0700857 bi->bi_rw, i);
858 atomic_inc(&sh->count);
NeilBrown05616be2012-05-21 09:27:00 +1000859 if (use_new_offset(conf, sh))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700860 bi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +1000861 + rdev->new_data_offset);
862 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700863 bi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +1000864 + rdev->data_offset);
majianpeng3f9e7c12012-07-31 10:04:21 +1000865 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
majianpenge59aa232013-11-14 15:16:19 +1100866 bi->bi_rw |= REQ_NOMERGE;
majianpeng3f9e7c12012-07-31 10:04:21 +1000867
Shaohua Lid592a992014-05-21 17:57:44 +0800868 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
869 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
870 sh->dev[i].vec.bv_page = sh->dev[i].page;
Kent Overstreet4997b722013-05-30 08:44:39 +0200871 bi->bi_vcnt = 1;
Dan Williams91c00922007-01-02 13:52:30 -0700872 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
873 bi->bi_io_vec[0].bv_offset = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700874 bi->bi_iter.bi_size = STRIPE_SIZE;
Shaohua Li37c61ff2013-10-19 14:50:28 +0800875 /*
876 * If this is discard request, set bi_vcnt 0. We don't
877 * want to confuse SCSI because SCSI will replace payload
878 */
879 if (rw & REQ_DISCARD)
880 bi->bi_vcnt = 0;
NeilBrown977df362011-12-23 10:17:53 +1100881 if (rrdev)
882 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
Jonathan Brassowe3620a32013-03-07 16:22:01 -0600883
884 if (conf->mddev->gendisk)
885 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
886 bi, disk_devt(conf->mddev->gendisk),
887 sh->dev[i].sector);
Dan Williams91c00922007-01-02 13:52:30 -0700888 generic_make_request(bi);
NeilBrown977df362011-12-23 10:17:53 +1100889 }
890 if (rrdev) {
NeilBrown9a3e1102011-12-23 10:17:53 +1100891 if (s->syncing || s->expanding || s->expanded
892 || s->replacing)
NeilBrown977df362011-12-23 10:17:53 +1100893 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
894
895 set_bit(STRIPE_IO_STARTED, &sh->state);
896
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700897 bio_reset(rbi);
NeilBrown977df362011-12-23 10:17:53 +1100898 rbi->bi_bdev = rrdev->bdev;
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700899 rbi->bi_rw = rw;
900 BUG_ON(!(rw & WRITE));
901 rbi->bi_end_io = raid5_end_write_request;
902 rbi->bi_private = sh;
903
NeilBrown977df362011-12-23 10:17:53 +1100904 pr_debug("%s: for %llu schedule op %ld on "
905 "replacement disc %d\n",
906 __func__, (unsigned long long)sh->sector,
907 rbi->bi_rw, i);
908 atomic_inc(&sh->count);
NeilBrown05616be2012-05-21 09:27:00 +1000909 if (use_new_offset(conf, sh))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700910 rbi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +1000911 + rrdev->new_data_offset);
912 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700913 rbi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +1000914 + rrdev->data_offset);
Shaohua Lid592a992014-05-21 17:57:44 +0800915 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
916 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
917 sh->dev[i].rvec.bv_page = sh->dev[i].page;
Kent Overstreet4997b722013-05-30 08:44:39 +0200918 rbi->bi_vcnt = 1;
NeilBrown977df362011-12-23 10:17:53 +1100919 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
920 rbi->bi_io_vec[0].bv_offset = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700921 rbi->bi_iter.bi_size = STRIPE_SIZE;
Shaohua Li37c61ff2013-10-19 14:50:28 +0800922 /*
923 * If this is discard request, set bi_vcnt 0. We don't
924 * want to confuse SCSI because SCSI will replace payload
925 */
926 if (rw & REQ_DISCARD)
927 rbi->bi_vcnt = 0;
Jonathan Brassowe3620a32013-03-07 16:22:01 -0600928 if (conf->mddev->gendisk)
929 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
930 rbi, disk_devt(conf->mddev->gendisk),
931 sh->dev[i].sector);
NeilBrown977df362011-12-23 10:17:53 +1100932 generic_make_request(rbi);
933 }
934 if (!rdev && !rrdev) {
Namhyung Kimb0629622011-06-14 14:20:19 +1000935 if (rw & WRITE)
Dan Williams91c00922007-01-02 13:52:30 -0700936 set_bit(STRIPE_DEGRADED, &sh->state);
937 pr_debug("skip op %ld on disc %d for sector %llu\n",
938 bi->bi_rw, i, (unsigned long long)sh->sector);
939 clear_bit(R5_LOCKED, &sh->dev[i].flags);
940 set_bit(STRIPE_HANDLE, &sh->state);
941 }
942 }
943}
944
945static struct dma_async_tx_descriptor *
Shaohua Lid592a992014-05-21 17:57:44 +0800946async_copy_data(int frombio, struct bio *bio, struct page **page,
947 sector_t sector, struct dma_async_tx_descriptor *tx,
948 struct stripe_head *sh)
Dan Williams91c00922007-01-02 13:52:30 -0700949{
Kent Overstreet79886132013-11-23 17:19:00 -0800950 struct bio_vec bvl;
951 struct bvec_iter iter;
Dan Williams91c00922007-01-02 13:52:30 -0700952 struct page *bio_page;
Dan Williams91c00922007-01-02 13:52:30 -0700953 int page_offset;
Dan Williamsa08abd82009-06-03 11:43:59 -0700954 struct async_submit_ctl submit;
Dan Williams0403e382009-09-08 17:42:50 -0700955 enum async_tx_flags flags = 0;
Dan Williams91c00922007-01-02 13:52:30 -0700956
Kent Overstreet4f024f32013-10-11 15:44:27 -0700957 if (bio->bi_iter.bi_sector >= sector)
958 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
Dan Williams91c00922007-01-02 13:52:30 -0700959 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700960 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
Dan Williamsa08abd82009-06-03 11:43:59 -0700961
Dan Williams0403e382009-09-08 17:42:50 -0700962 if (frombio)
963 flags |= ASYNC_TX_FENCE;
964 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
965
Kent Overstreet79886132013-11-23 17:19:00 -0800966 bio_for_each_segment(bvl, bio, iter) {
967 int len = bvl.bv_len;
Dan Williams91c00922007-01-02 13:52:30 -0700968 int clen;
969 int b_offset = 0;
970
971 if (page_offset < 0) {
972 b_offset = -page_offset;
973 page_offset += b_offset;
974 len -= b_offset;
975 }
976
977 if (len > 0 && page_offset + len > STRIPE_SIZE)
978 clen = STRIPE_SIZE - page_offset;
979 else
980 clen = len;
981
982 if (clen > 0) {
Kent Overstreet79886132013-11-23 17:19:00 -0800983 b_offset += bvl.bv_offset;
984 bio_page = bvl.bv_page;
Shaohua Lid592a992014-05-21 17:57:44 +0800985 if (frombio) {
986 if (sh->raid_conf->skip_copy &&
987 b_offset == 0 && page_offset == 0 &&
988 clen == STRIPE_SIZE)
989 *page = bio_page;
990 else
991 tx = async_memcpy(*page, bio_page, page_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700992 b_offset, clen, &submit);
Shaohua Lid592a992014-05-21 17:57:44 +0800993 } else
994 tx = async_memcpy(bio_page, *page, b_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700995 page_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700996 }
Dan Williamsa08abd82009-06-03 11:43:59 -0700997 /* chain the operations */
998 submit.depend_tx = tx;
999
Dan Williams91c00922007-01-02 13:52:30 -07001000 if (clen < len) /* hit end of page */
1001 break;
1002 page_offset += len;
1003 }
1004
1005 return tx;
1006}
1007
1008static void ops_complete_biofill(void *stripe_head_ref)
1009{
1010 struct stripe_head *sh = stripe_head_ref;
1011 struct bio *return_bi = NULL;
Dan Williamse4d84902007-09-24 10:06:13 -07001012 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001013
Harvey Harrisone46b2722008-04-28 02:15:50 -07001014 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001015 (unsigned long long)sh->sector);
1016
1017 /* clear completed biofills */
1018 for (i = sh->disks; i--; ) {
1019 struct r5dev *dev = &sh->dev[i];
Dan Williams91c00922007-01-02 13:52:30 -07001020
1021 /* acknowledge completion of a biofill operation */
Dan Williamse4d84902007-09-24 10:06:13 -07001022 /* and check if we need to reply to a read request,
1023 * new R5_Wantfill requests are held off until
Dan Williams83de75c2008-06-28 08:31:58 +10001024 * !STRIPE_BIOFILL_RUN
Dan Williamse4d84902007-09-24 10:06:13 -07001025 */
1026 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001027 struct bio *rbi, *rbi2;
Dan Williams91c00922007-01-02 13:52:30 -07001028
Dan Williams91c00922007-01-02 13:52:30 -07001029 BUG_ON(!dev->read);
1030 rbi = dev->read;
1031 dev->read = NULL;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001032 while (rbi && rbi->bi_iter.bi_sector <
Dan Williams91c00922007-01-02 13:52:30 -07001033 dev->sector + STRIPE_SECTORS) {
1034 rbi2 = r5_next_bio(rbi, dev->sector);
Shaohua Lie7836bd62012-07-19 16:01:31 +10001035 if (!raid5_dec_bi_active_stripes(rbi)) {
Dan Williams91c00922007-01-02 13:52:30 -07001036 rbi->bi_next = return_bi;
1037 return_bi = rbi;
1038 }
Dan Williams91c00922007-01-02 13:52:30 -07001039 rbi = rbi2;
1040 }
1041 }
1042 }
Dan Williams83de75c2008-06-28 08:31:58 +10001043 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -07001044
1045 return_io(return_bi);
1046
Dan Williamse4d84902007-09-24 10:06:13 -07001047 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -07001048 release_stripe(sh);
1049}
1050
1051static void ops_run_biofill(struct stripe_head *sh)
1052{
1053 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa08abd82009-06-03 11:43:59 -07001054 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001055 int i;
1056
Harvey Harrisone46b2722008-04-28 02:15:50 -07001057 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001058 (unsigned long long)sh->sector);
1059
1060 for (i = sh->disks; i--; ) {
1061 struct r5dev *dev = &sh->dev[i];
1062 if (test_bit(R5_Wantfill, &dev->flags)) {
1063 struct bio *rbi;
Shaohua Lib17459c2012-07-19 16:01:31 +10001064 spin_lock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001065 dev->read = rbi = dev->toread;
1066 dev->toread = NULL;
Shaohua Lib17459c2012-07-19 16:01:31 +10001067 spin_unlock_irq(&sh->stripe_lock);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001068 while (rbi && rbi->bi_iter.bi_sector <
Dan Williams91c00922007-01-02 13:52:30 -07001069 dev->sector + STRIPE_SECTORS) {
Shaohua Lid592a992014-05-21 17:57:44 +08001070 tx = async_copy_data(0, rbi, &dev->page,
1071 dev->sector, tx, sh);
Dan Williams91c00922007-01-02 13:52:30 -07001072 rbi = r5_next_bio(rbi, dev->sector);
1073 }
1074 }
1075 }
1076
1077 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001078 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1079 async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001080}
1081
Dan Williams4e7d2c02009-08-29 19:13:11 -07001082static void mark_target_uptodate(struct stripe_head *sh, int target)
1083{
1084 struct r5dev *tgt;
1085
1086 if (target < 0)
1087 return;
1088
1089 tgt = &sh->dev[target];
1090 set_bit(R5_UPTODATE, &tgt->flags);
1091 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1092 clear_bit(R5_Wantcompute, &tgt->flags);
1093}
1094
Dan Williamsac6b53b2009-07-14 13:40:19 -07001095static void ops_complete_compute(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001096{
1097 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001098
Harvey Harrisone46b2722008-04-28 02:15:50 -07001099 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001100 (unsigned long long)sh->sector);
1101
Dan Williamsac6b53b2009-07-14 13:40:19 -07001102 /* mark the computed target(s) as uptodate */
Dan Williams4e7d2c02009-08-29 19:13:11 -07001103 mark_target_uptodate(sh, sh->ops.target);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001104 mark_target_uptodate(sh, sh->ops.target2);
Dan Williams4e7d2c02009-08-29 19:13:11 -07001105
Dan Williamsecc65c92008-06-28 08:31:57 +10001106 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1107 if (sh->check_state == check_state_compute_run)
1108 sh->check_state = check_state_compute_result;
Dan Williams91c00922007-01-02 13:52:30 -07001109 set_bit(STRIPE_HANDLE, &sh->state);
1110 release_stripe(sh);
1111}
1112
Dan Williamsd6f38f32009-07-14 11:50:52 -07001113/* return a pointer to the address conversion region of the scribble buffer */
1114static addr_conv_t *to_addr_conv(struct stripe_head *sh,
1115 struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001116{
Dan Williamsd6f38f32009-07-14 11:50:52 -07001117 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
1118}
1119
1120static struct dma_async_tx_descriptor *
1121ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
1122{
Dan Williams91c00922007-01-02 13:52:30 -07001123 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001124 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001125 int target = sh->ops.target;
1126 struct r5dev *tgt = &sh->dev[target];
1127 struct page *xor_dest = tgt->page;
1128 int count = 0;
1129 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001130 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001131 int i;
1132
1133 pr_debug("%s: stripe %llu block: %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07001134 __func__, (unsigned long long)sh->sector, target);
Dan Williams91c00922007-01-02 13:52:30 -07001135 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1136
1137 for (i = disks; i--; )
1138 if (i != target)
1139 xor_srcs[count++] = sh->dev[i].page;
1140
1141 atomic_inc(&sh->count);
1142
Dan Williams0403e382009-09-08 17:42:50 -07001143 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001144 ops_complete_compute, sh, to_addr_conv(sh, percpu));
Dan Williams91c00922007-01-02 13:52:30 -07001145 if (unlikely(count == 1))
Dan Williamsa08abd82009-06-03 11:43:59 -07001146 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001147 else
Dan Williamsa08abd82009-06-03 11:43:59 -07001148 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001149
Dan Williams91c00922007-01-02 13:52:30 -07001150 return tx;
1151}
1152
Dan Williamsac6b53b2009-07-14 13:40:19 -07001153/* set_syndrome_sources - populate source buffers for gen_syndrome
1154 * @srcs - (struct page *) array of size sh->disks
1155 * @sh - stripe_head to parse
1156 *
1157 * Populates srcs in proper layout order for the stripe and returns the
1158 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1159 * destination buffer is recorded in srcs[count] and the Q destination
1160 * is recorded in srcs[count+1]].
1161 */
1162static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
1163{
1164 int disks = sh->disks;
1165 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1166 int d0_idx = raid6_d0(sh);
1167 int count;
1168 int i;
1169
1170 for (i = 0; i < disks; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +11001171 srcs[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001172
1173 count = 0;
1174 i = d0_idx;
1175 do {
1176 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1177
1178 srcs[slot] = sh->dev[i].page;
1179 i = raid6_next_disk(i, disks);
1180 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001181
NeilBrowne4424fe2009-10-16 16:27:34 +11001182 return syndrome_disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001183}
1184
1185static struct dma_async_tx_descriptor *
1186ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1187{
1188 int disks = sh->disks;
1189 struct page **blocks = percpu->scribble;
1190 int target;
1191 int qd_idx = sh->qd_idx;
1192 struct dma_async_tx_descriptor *tx;
1193 struct async_submit_ctl submit;
1194 struct r5dev *tgt;
1195 struct page *dest;
1196 int i;
1197 int count;
1198
1199 if (sh->ops.target < 0)
1200 target = sh->ops.target2;
1201 else if (sh->ops.target2 < 0)
1202 target = sh->ops.target;
1203 else
1204 /* we should only have one valid target */
1205 BUG();
1206 BUG_ON(target < 0);
1207 pr_debug("%s: stripe %llu block: %d\n",
1208 __func__, (unsigned long long)sh->sector, target);
1209
1210 tgt = &sh->dev[target];
1211 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1212 dest = tgt->page;
1213
1214 atomic_inc(&sh->count);
1215
1216 if (target == qd_idx) {
1217 count = set_syndrome_sources(blocks, sh);
1218 blocks[count] = NULL; /* regenerating p is not necessary */
1219 BUG_ON(blocks[count+1] != dest); /* q should already be set */
Dan Williams0403e382009-09-08 17:42:50 -07001220 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1221 ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001222 to_addr_conv(sh, percpu));
1223 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1224 } else {
1225 /* Compute any data- or p-drive using XOR */
1226 count = 0;
1227 for (i = disks; i-- ; ) {
1228 if (i == target || i == qd_idx)
1229 continue;
1230 blocks[count++] = sh->dev[i].page;
1231 }
1232
Dan Williams0403e382009-09-08 17:42:50 -07001233 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1234 NULL, ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001235 to_addr_conv(sh, percpu));
1236 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1237 }
1238
1239 return tx;
1240}
1241
1242static struct dma_async_tx_descriptor *
1243ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1244{
1245 int i, count, disks = sh->disks;
1246 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1247 int d0_idx = raid6_d0(sh);
1248 int faila = -1, failb = -1;
1249 int target = sh->ops.target;
1250 int target2 = sh->ops.target2;
1251 struct r5dev *tgt = &sh->dev[target];
1252 struct r5dev *tgt2 = &sh->dev[target2];
1253 struct dma_async_tx_descriptor *tx;
1254 struct page **blocks = percpu->scribble;
1255 struct async_submit_ctl submit;
1256
1257 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1258 __func__, (unsigned long long)sh->sector, target, target2);
1259 BUG_ON(target < 0 || target2 < 0);
1260 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1261 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1262
Dan Williams6c910a72009-09-16 12:24:54 -07001263 /* we need to open-code set_syndrome_sources to handle the
Dan Williamsac6b53b2009-07-14 13:40:19 -07001264 * slot number conversion for 'faila' and 'failb'
1265 */
1266 for (i = 0; i < disks ; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +11001267 blocks[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001268 count = 0;
1269 i = d0_idx;
1270 do {
1271 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1272
1273 blocks[slot] = sh->dev[i].page;
1274
1275 if (i == target)
1276 faila = slot;
1277 if (i == target2)
1278 failb = slot;
1279 i = raid6_next_disk(i, disks);
1280 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001281
1282 BUG_ON(faila == failb);
1283 if (failb < faila)
1284 swap(faila, failb);
1285 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1286 __func__, (unsigned long long)sh->sector, faila, failb);
1287
1288 atomic_inc(&sh->count);
1289
1290 if (failb == syndrome_disks+1) {
1291 /* Q disk is one of the missing disks */
1292 if (faila == syndrome_disks) {
1293 /* Missing P+Q, just recompute */
Dan Williams0403e382009-09-08 17:42:50 -07001294 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1295 ops_complete_compute, sh,
1296 to_addr_conv(sh, percpu));
NeilBrowne4424fe2009-10-16 16:27:34 +11001297 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001298 STRIPE_SIZE, &submit);
1299 } else {
1300 struct page *dest;
1301 int data_target;
1302 int qd_idx = sh->qd_idx;
1303
1304 /* Missing D+Q: recompute D from P, then recompute Q */
1305 if (target == qd_idx)
1306 data_target = target2;
1307 else
1308 data_target = target;
1309
1310 count = 0;
1311 for (i = disks; i-- ; ) {
1312 if (i == data_target || i == qd_idx)
1313 continue;
1314 blocks[count++] = sh->dev[i].page;
1315 }
1316 dest = sh->dev[data_target].page;
Dan Williams0403e382009-09-08 17:42:50 -07001317 init_async_submit(&submit,
1318 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1319 NULL, NULL, NULL,
1320 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001321 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1322 &submit);
1323
1324 count = set_syndrome_sources(blocks, sh);
Dan Williams0403e382009-09-08 17:42:50 -07001325 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1326 ops_complete_compute, sh,
1327 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001328 return async_gen_syndrome(blocks, 0, count+2,
1329 STRIPE_SIZE, &submit);
1330 }
Dan Williamsac6b53b2009-07-14 13:40:19 -07001331 } else {
Dan Williams6c910a72009-09-16 12:24:54 -07001332 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1333 ops_complete_compute, sh,
1334 to_addr_conv(sh, percpu));
1335 if (failb == syndrome_disks) {
1336 /* We're missing D+P. */
1337 return async_raid6_datap_recov(syndrome_disks+2,
1338 STRIPE_SIZE, faila,
1339 blocks, &submit);
1340 } else {
1341 /* We're missing D+D. */
1342 return async_raid6_2data_recov(syndrome_disks+2,
1343 STRIPE_SIZE, faila, failb,
1344 blocks, &submit);
1345 }
Dan Williamsac6b53b2009-07-14 13:40:19 -07001346 }
1347}
1348
Dan Williams91c00922007-01-02 13:52:30 -07001349static void ops_complete_prexor(void *stripe_head_ref)
1350{
1351 struct stripe_head *sh = stripe_head_ref;
1352
Harvey Harrisone46b2722008-04-28 02:15:50 -07001353 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001354 (unsigned long long)sh->sector);
Dan Williams91c00922007-01-02 13:52:30 -07001355}
1356
1357static struct dma_async_tx_descriptor *
Dan Williamsd6f38f32009-07-14 11:50:52 -07001358ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
1359 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001360{
Dan Williams91c00922007-01-02 13:52:30 -07001361 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001362 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001363 int count = 0, pd_idx = sh->pd_idx, i;
Dan Williamsa08abd82009-06-03 11:43:59 -07001364 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001365
1366 /* existing parity data subtracted */
1367 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1368
Harvey Harrisone46b2722008-04-28 02:15:50 -07001369 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001370 (unsigned long long)sh->sector);
1371
1372 for (i = disks; i--; ) {
1373 struct r5dev *dev = &sh->dev[i];
1374 /* Only process blocks that are known to be uptodate */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001375 if (test_bit(R5_Wantdrain, &dev->flags))
Dan Williams91c00922007-01-02 13:52:30 -07001376 xor_srcs[count++] = dev->page;
1377 }
1378
Dan Williams0403e382009-09-08 17:42:50 -07001379 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001380 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001381 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001382
1383 return tx;
1384}
1385
1386static struct dma_async_tx_descriptor *
Dan Williamsd8ee0722008-06-28 08:32:06 +10001387ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001388{
1389 int disks = sh->disks;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001390 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001391
Harvey Harrisone46b2722008-04-28 02:15:50 -07001392 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001393 (unsigned long long)sh->sector);
1394
1395 for (i = disks; i--; ) {
1396 struct r5dev *dev = &sh->dev[i];
1397 struct bio *chosen;
Dan Williams91c00922007-01-02 13:52:30 -07001398
Dan Williamsd8ee0722008-06-28 08:32:06 +10001399 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001400 struct bio *wbi;
1401
Shaohua Lib17459c2012-07-19 16:01:31 +10001402 spin_lock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001403 chosen = dev->towrite;
1404 dev->towrite = NULL;
1405 BUG_ON(dev->written);
1406 wbi = dev->written = chosen;
Shaohua Lib17459c2012-07-19 16:01:31 +10001407 spin_unlock_irq(&sh->stripe_lock);
Shaohua Lid592a992014-05-21 17:57:44 +08001408 WARN_ON(dev->page != dev->orig_page);
Dan Williams91c00922007-01-02 13:52:30 -07001409
Kent Overstreet4f024f32013-10-11 15:44:27 -07001410 while (wbi && wbi->bi_iter.bi_sector <
Dan Williams91c00922007-01-02 13:52:30 -07001411 dev->sector + STRIPE_SECTORS) {
Tejun Heoe9c74692010-09-03 11:56:18 +02001412 if (wbi->bi_rw & REQ_FUA)
1413 set_bit(R5_WantFUA, &dev->flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001414 if (wbi->bi_rw & REQ_SYNC)
1415 set_bit(R5_SyncIO, &dev->flags);
Shaohua Li9e4447682012-10-11 13:49:49 +11001416 if (wbi->bi_rw & REQ_DISCARD)
Shaohua Li620125f2012-10-11 13:49:05 +11001417 set_bit(R5_Discard, &dev->flags);
Shaohua Lid592a992014-05-21 17:57:44 +08001418 else {
1419 tx = async_copy_data(1, wbi, &dev->page,
1420 dev->sector, tx, sh);
1421 if (dev->page != dev->orig_page) {
1422 set_bit(R5_SkipCopy, &dev->flags);
1423 clear_bit(R5_UPTODATE, &dev->flags);
1424 clear_bit(R5_OVERWRITE, &dev->flags);
1425 }
1426 }
Dan Williams91c00922007-01-02 13:52:30 -07001427 wbi = r5_next_bio(wbi, dev->sector);
1428 }
1429 }
1430 }
1431
1432 return tx;
1433}
1434
Dan Williamsac6b53b2009-07-14 13:40:19 -07001435static void ops_complete_reconstruct(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001436{
1437 struct stripe_head *sh = stripe_head_ref;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001438 int disks = sh->disks;
1439 int pd_idx = sh->pd_idx;
1440 int qd_idx = sh->qd_idx;
1441 int i;
Shaohua Li9e4447682012-10-11 13:49:49 +11001442 bool fua = false, sync = false, discard = false;
Dan Williams91c00922007-01-02 13:52:30 -07001443
Harvey Harrisone46b2722008-04-28 02:15:50 -07001444 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001445 (unsigned long long)sh->sector);
1446
Shaohua Libc0934f2012-05-22 13:55:05 +10001447 for (i = disks; i--; ) {
Tejun Heoe9c74692010-09-03 11:56:18 +02001448 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001449 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
Shaohua Li9e4447682012-10-11 13:49:49 +11001450 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001451 }
Tejun Heoe9c74692010-09-03 11:56:18 +02001452
Dan Williams91c00922007-01-02 13:52:30 -07001453 for (i = disks; i--; ) {
1454 struct r5dev *dev = &sh->dev[i];
Dan Williamsac6b53b2009-07-14 13:40:19 -07001455
Tejun Heoe9c74692010-09-03 11:56:18 +02001456 if (dev->written || i == pd_idx || i == qd_idx) {
Shaohua Lid592a992014-05-21 17:57:44 +08001457 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
Shaohua Li9e4447682012-10-11 13:49:49 +11001458 set_bit(R5_UPTODATE, &dev->flags);
Tejun Heoe9c74692010-09-03 11:56:18 +02001459 if (fua)
1460 set_bit(R5_WantFUA, &dev->flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001461 if (sync)
1462 set_bit(R5_SyncIO, &dev->flags);
Tejun Heoe9c74692010-09-03 11:56:18 +02001463 }
Dan Williams91c00922007-01-02 13:52:30 -07001464 }
1465
Dan Williamsd8ee0722008-06-28 08:32:06 +10001466 if (sh->reconstruct_state == reconstruct_state_drain_run)
1467 sh->reconstruct_state = reconstruct_state_drain_result;
1468 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1469 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1470 else {
1471 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1472 sh->reconstruct_state = reconstruct_state_result;
1473 }
Dan Williams91c00922007-01-02 13:52:30 -07001474
1475 set_bit(STRIPE_HANDLE, &sh->state);
1476 release_stripe(sh);
1477}
1478
1479static void
Dan Williamsac6b53b2009-07-14 13:40:19 -07001480ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1481 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001482{
Dan Williams91c00922007-01-02 13:52:30 -07001483 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001484 struct page **xor_srcs = percpu->scribble;
Dan Williamsa08abd82009-06-03 11:43:59 -07001485 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001486 int count = 0, pd_idx = sh->pd_idx, i;
1487 struct page *xor_dest;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001488 int prexor = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001489 unsigned long flags;
Dan Williams91c00922007-01-02 13:52:30 -07001490
Harvey Harrisone46b2722008-04-28 02:15:50 -07001491 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001492 (unsigned long long)sh->sector);
1493
Shaohua Li620125f2012-10-11 13:49:05 +11001494 for (i = 0; i < sh->disks; i++) {
1495 if (pd_idx == i)
1496 continue;
1497 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1498 break;
1499 }
1500 if (i >= sh->disks) {
1501 atomic_inc(&sh->count);
Shaohua Li620125f2012-10-11 13:49:05 +11001502 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1503 ops_complete_reconstruct(sh);
1504 return;
1505 }
Dan Williams91c00922007-01-02 13:52:30 -07001506 /* check if prexor is active which means only process blocks
1507 * that are part of a read-modify-write (written)
1508 */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001509 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1510 prexor = 1;
Dan Williams91c00922007-01-02 13:52:30 -07001511 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1512 for (i = disks; i--; ) {
1513 struct r5dev *dev = &sh->dev[i];
1514 if (dev->written)
1515 xor_srcs[count++] = dev->page;
1516 }
1517 } else {
1518 xor_dest = sh->dev[pd_idx].page;
1519 for (i = disks; i--; ) {
1520 struct r5dev *dev = &sh->dev[i];
1521 if (i != pd_idx)
1522 xor_srcs[count++] = dev->page;
1523 }
1524 }
1525
Dan Williams91c00922007-01-02 13:52:30 -07001526 /* 1/ if we prexor'd then the dest is reused as a source
1527 * 2/ if we did not prexor then we are redoing the parity
1528 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1529 * for the synchronous xor case
1530 */
Dan Williams88ba2aa2009-04-09 16:16:18 -07001531 flags = ASYNC_TX_ACK |
Dan Williams91c00922007-01-02 13:52:30 -07001532 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1533
1534 atomic_inc(&sh->count);
1535
Dan Williamsac6b53b2009-07-14 13:40:19 -07001536 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001537 to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001538 if (unlikely(count == 1))
1539 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1540 else
1541 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001542}
1543
Dan Williamsac6b53b2009-07-14 13:40:19 -07001544static void
1545ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1546 struct dma_async_tx_descriptor *tx)
1547{
1548 struct async_submit_ctl submit;
1549 struct page **blocks = percpu->scribble;
Shaohua Li620125f2012-10-11 13:49:05 +11001550 int count, i;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001551
1552 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1553
Shaohua Li620125f2012-10-11 13:49:05 +11001554 for (i = 0; i < sh->disks; i++) {
1555 if (sh->pd_idx == i || sh->qd_idx == i)
1556 continue;
1557 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1558 break;
1559 }
1560 if (i >= sh->disks) {
1561 atomic_inc(&sh->count);
Shaohua Li620125f2012-10-11 13:49:05 +11001562 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1563 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1564 ops_complete_reconstruct(sh);
1565 return;
1566 }
1567
Dan Williamsac6b53b2009-07-14 13:40:19 -07001568 count = set_syndrome_sources(blocks, sh);
1569
1570 atomic_inc(&sh->count);
1571
1572 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1573 sh, to_addr_conv(sh, percpu));
1574 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001575}
1576
1577static void ops_complete_check(void *stripe_head_ref)
1578{
1579 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001580
Harvey Harrisone46b2722008-04-28 02:15:50 -07001581 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001582 (unsigned long long)sh->sector);
1583
Dan Williamsecc65c92008-06-28 08:31:57 +10001584 sh->check_state = check_state_check_result;
Dan Williams91c00922007-01-02 13:52:30 -07001585 set_bit(STRIPE_HANDLE, &sh->state);
1586 release_stripe(sh);
1587}
1588
Dan Williamsac6b53b2009-07-14 13:40:19 -07001589static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001590{
Dan Williams91c00922007-01-02 13:52:30 -07001591 int disks = sh->disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001592 int pd_idx = sh->pd_idx;
1593 int qd_idx = sh->qd_idx;
1594 struct page *xor_dest;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001595 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001596 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001597 struct async_submit_ctl submit;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001598 int count;
1599 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001600
Harvey Harrisone46b2722008-04-28 02:15:50 -07001601 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001602 (unsigned long long)sh->sector);
1603
Dan Williamsac6b53b2009-07-14 13:40:19 -07001604 count = 0;
1605 xor_dest = sh->dev[pd_idx].page;
1606 xor_srcs[count++] = xor_dest;
Dan Williams91c00922007-01-02 13:52:30 -07001607 for (i = disks; i--; ) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001608 if (i == pd_idx || i == qd_idx)
1609 continue;
1610 xor_srcs[count++] = sh->dev[i].page;
Dan Williams91c00922007-01-02 13:52:30 -07001611 }
1612
Dan Williamsd6f38f32009-07-14 11:50:52 -07001613 init_async_submit(&submit, 0, NULL, NULL, NULL,
1614 to_addr_conv(sh, percpu));
Dan Williams099f53c2009-04-08 14:28:37 -07001615 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07001616 &sh->ops.zero_sum_result, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001617
Dan Williams91c00922007-01-02 13:52:30 -07001618 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001619 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1620 tx = async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001621}
1622
Dan Williamsac6b53b2009-07-14 13:40:19 -07001623static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1624{
1625 struct page **srcs = percpu->scribble;
1626 struct async_submit_ctl submit;
1627 int count;
1628
1629 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1630 (unsigned long long)sh->sector, checkp);
1631
1632 count = set_syndrome_sources(srcs, sh);
1633 if (!checkp)
1634 srcs[count] = NULL;
1635
1636 atomic_inc(&sh->count);
1637 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1638 sh, to_addr_conv(sh, percpu));
1639 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1640 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
1641}
1642
NeilBrown51acbce2013-02-28 09:08:34 +11001643static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
Dan Williams91c00922007-01-02 13:52:30 -07001644{
1645 int overlap_clear = 0, i, disks = sh->disks;
1646 struct dma_async_tx_descriptor *tx = NULL;
NeilBrownd1688a62011-10-11 16:49:52 +11001647 struct r5conf *conf = sh->raid_conf;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001648 int level = conf->level;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001649 struct raid5_percpu *percpu;
1650 unsigned long cpu;
Dan Williams91c00922007-01-02 13:52:30 -07001651
Dan Williamsd6f38f32009-07-14 11:50:52 -07001652 cpu = get_cpu();
1653 percpu = per_cpu_ptr(conf->percpu, cpu);
Dan Williams83de75c2008-06-28 08:31:58 +10001654 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
Dan Williams91c00922007-01-02 13:52:30 -07001655 ops_run_biofill(sh);
1656 overlap_clear++;
1657 }
1658
Dan Williams7b3a8712008-06-28 08:32:09 +10001659 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001660 if (level < 6)
1661 tx = ops_run_compute5(sh, percpu);
1662 else {
1663 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1664 tx = ops_run_compute6_1(sh, percpu);
1665 else
1666 tx = ops_run_compute6_2(sh, percpu);
1667 }
1668 /* terminate the chain if reconstruct is not set to be run */
1669 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
Dan Williams7b3a8712008-06-28 08:32:09 +10001670 async_tx_ack(tx);
1671 }
Dan Williams91c00922007-01-02 13:52:30 -07001672
Dan Williams600aa102008-06-28 08:32:05 +10001673 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
Dan Williamsd6f38f32009-07-14 11:50:52 -07001674 tx = ops_run_prexor(sh, percpu, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001675
Dan Williams600aa102008-06-28 08:32:05 +10001676 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
Dan Williamsd8ee0722008-06-28 08:32:06 +10001677 tx = ops_run_biodrain(sh, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001678 overlap_clear++;
1679 }
1680
Dan Williamsac6b53b2009-07-14 13:40:19 -07001681 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1682 if (level < 6)
1683 ops_run_reconstruct5(sh, percpu, tx);
1684 else
1685 ops_run_reconstruct6(sh, percpu, tx);
1686 }
Dan Williams91c00922007-01-02 13:52:30 -07001687
Dan Williamsac6b53b2009-07-14 13:40:19 -07001688 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1689 if (sh->check_state == check_state_run)
1690 ops_run_check_p(sh, percpu);
1691 else if (sh->check_state == check_state_run_q)
1692 ops_run_check_pq(sh, percpu, 0);
1693 else if (sh->check_state == check_state_run_pq)
1694 ops_run_check_pq(sh, percpu, 1);
1695 else
1696 BUG();
1697 }
Dan Williams91c00922007-01-02 13:52:30 -07001698
Dan Williams91c00922007-01-02 13:52:30 -07001699 if (overlap_clear)
1700 for (i = disks; i--; ) {
1701 struct r5dev *dev = &sh->dev[i];
1702 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1703 wake_up(&sh->raid_conf->wait_for_overlap);
1704 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07001705 put_cpu();
Dan Williams91c00922007-01-02 13:52:30 -07001706}
1707
Shaohua Li566c09c2013-11-14 15:16:17 +11001708static int grow_one_stripe(struct r5conf *conf, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709{
1710 struct stripe_head *sh;
Namhyung Kim6ce32842011-07-18 17:38:50 +10001711 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
NeilBrown3f294f42005-11-08 21:39:25 -08001712 if (!sh)
1713 return 0;
Namhyung Kim6ce32842011-07-18 17:38:50 +10001714
NeilBrown3f294f42005-11-08 21:39:25 -08001715 sh->raid_conf = conf;
NeilBrown3f294f42005-11-08 21:39:25 -08001716
Shaohua Lib17459c2012-07-19 16:01:31 +10001717 spin_lock_init(&sh->stripe_lock);
1718
NeilBrowne4e11e32010-06-16 16:45:16 +10001719 if (grow_buffers(sh)) {
1720 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001721 kmem_cache_free(conf->slab_cache, sh);
1722 return 0;
1723 }
Shaohua Li566c09c2013-11-14 15:16:17 +11001724 sh->hash_lock_index = hash;
NeilBrown3f294f42005-11-08 21:39:25 -08001725 /* we just created an active stripe so... */
1726 atomic_set(&sh->count, 1);
1727 atomic_inc(&conf->active_stripes);
1728 INIT_LIST_HEAD(&sh->lru);
1729 release_stripe(sh);
1730 return 1;
1731}
1732
NeilBrownd1688a62011-10-11 16:49:52 +11001733static int grow_stripes(struct r5conf *conf, int num)
NeilBrown3f294f42005-11-08 21:39:25 -08001734{
Christoph Lametere18b8902006-12-06 20:33:20 -08001735 struct kmem_cache *sc;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001736 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Shaohua Li566c09c2013-11-14 15:16:17 +11001737 int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
NeilBrownf4be6b42010-06-01 19:37:25 +10001739 if (conf->mddev->gendisk)
1740 sprintf(conf->cache_name[0],
1741 "raid%d-%s", conf->level, mdname(conf->mddev));
1742 else
1743 sprintf(conf->cache_name[0],
1744 "raid%d-%p", conf->level, conf->mddev);
1745 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1746
NeilBrownad01c9e2006-03-27 01:18:07 -08001747 conf->active_name = 0;
1748 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001750 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 if (!sc)
1752 return 1;
1753 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001754 conf->pool_size = devs;
Shaohua Li566c09c2013-11-14 15:16:17 +11001755 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
1756 while (num--) {
1757 if (!grow_one_stripe(conf, hash))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 return 1;
Shaohua Li566c09c2013-11-14 15:16:17 +11001759 conf->max_nr_stripes++;
1760 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
1761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 return 0;
1763}
NeilBrown29269552006-03-27 01:18:10 -08001764
Dan Williamsd6f38f32009-07-14 11:50:52 -07001765/**
1766 * scribble_len - return the required size of the scribble region
1767 * @num - total number of disks in the array
1768 *
1769 * The size must be enough to contain:
1770 * 1/ a struct page pointer for each device in the array +2
1771 * 2/ room to convert each entry in (1) to its corresponding dma
1772 * (dma_map_page()) or page (page_address()) address.
1773 *
1774 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1775 * calculate over all devices (not just the data blocks), using zeros in place
1776 * of the P and Q blocks.
1777 */
1778static size_t scribble_len(int num)
1779{
1780 size_t len;
1781
1782 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1783
1784 return len;
1785}
1786
NeilBrownd1688a62011-10-11 16:49:52 +11001787static int resize_stripes(struct r5conf *conf, int newsize)
NeilBrownad01c9e2006-03-27 01:18:07 -08001788{
1789 /* Make all the stripes able to hold 'newsize' devices.
1790 * New slots in each stripe get 'page' set to a new page.
1791 *
1792 * This happens in stages:
1793 * 1/ create a new kmem_cache and allocate the required number of
1794 * stripe_heads.
Masanari Iida83f0d772012-10-30 00:18:08 +09001795 * 2/ gather all the old stripe_heads and transfer the pages across
NeilBrownad01c9e2006-03-27 01:18:07 -08001796 * to the new stripe_heads. This will have the side effect of
1797 * freezing the array as once all stripe_heads have been collected,
1798 * no IO will be possible. Old stripe heads are freed once their
1799 * pages have been transferred over, and the old kmem_cache is
1800 * freed when all stripes are done.
1801 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1802 * we simple return a failre status - no need to clean anything up.
1803 * 4/ allocate new pages for the new slots in the new stripe_heads.
1804 * If this fails, we don't bother trying the shrink the
1805 * stripe_heads down again, we just leave them as they are.
1806 * As each stripe_head is processed the new one is released into
1807 * active service.
1808 *
1809 * Once step2 is started, we cannot afford to wait for a write,
1810 * so we use GFP_NOIO allocations.
1811 */
1812 struct stripe_head *osh, *nsh;
1813 LIST_HEAD(newstripes);
1814 struct disk_info *ndisks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001815 unsigned long cpu;
Dan Williamsb5470dc2008-06-27 21:44:04 -07001816 int err;
Christoph Lametere18b8902006-12-06 20:33:20 -08001817 struct kmem_cache *sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001818 int i;
Shaohua Li566c09c2013-11-14 15:16:17 +11001819 int hash, cnt;
NeilBrownad01c9e2006-03-27 01:18:07 -08001820
1821 if (newsize <= conf->pool_size)
1822 return 0; /* never bother to shrink */
1823
Dan Williamsb5470dc2008-06-27 21:44:04 -07001824 err = md_allow_write(conf->mddev);
1825 if (err)
1826 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08001827
NeilBrownad01c9e2006-03-27 01:18:07 -08001828 /* Step 1 */
1829 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1830 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001831 0, 0, NULL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001832 if (!sc)
1833 return -ENOMEM;
1834
1835 for (i = conf->max_nr_stripes; i; i--) {
Namhyung Kim6ce32842011-07-18 17:38:50 +10001836 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001837 if (!nsh)
1838 break;
1839
NeilBrownad01c9e2006-03-27 01:18:07 -08001840 nsh->raid_conf = conf;
NeilBrowncb13ff62012-09-24 16:27:20 +10001841 spin_lock_init(&nsh->stripe_lock);
NeilBrownad01c9e2006-03-27 01:18:07 -08001842
1843 list_add(&nsh->lru, &newstripes);
1844 }
1845 if (i) {
1846 /* didn't get enough, give up */
1847 while (!list_empty(&newstripes)) {
1848 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1849 list_del(&nsh->lru);
1850 kmem_cache_free(sc, nsh);
1851 }
1852 kmem_cache_destroy(sc);
1853 return -ENOMEM;
1854 }
1855 /* Step 2 - Must use GFP_NOIO now.
1856 * OK, we have enough stripes, start collecting inactive
1857 * stripes and copying them over
1858 */
Shaohua Li566c09c2013-11-14 15:16:17 +11001859 hash = 0;
1860 cnt = 0;
NeilBrownad01c9e2006-03-27 01:18:07 -08001861 list_for_each_entry(nsh, &newstripes, lru) {
Shaohua Li566c09c2013-11-14 15:16:17 +11001862 lock_device_hash_lock(conf, hash);
1863 wait_event_cmd(conf->wait_for_stripe,
1864 !list_empty(conf->inactive_list + hash),
1865 unlock_device_hash_lock(conf, hash),
1866 lock_device_hash_lock(conf, hash));
1867 osh = get_free_stripe(conf, hash);
1868 unlock_device_hash_lock(conf, hash);
NeilBrownad01c9e2006-03-27 01:18:07 -08001869 atomic_set(&nsh->count, 1);
Shaohua Lid592a992014-05-21 17:57:44 +08001870 for(i=0; i<conf->pool_size; i++) {
NeilBrownad01c9e2006-03-27 01:18:07 -08001871 nsh->dev[i].page = osh->dev[i].page;
Shaohua Lid592a992014-05-21 17:57:44 +08001872 nsh->dev[i].orig_page = osh->dev[i].page;
1873 }
NeilBrownad01c9e2006-03-27 01:18:07 -08001874 for( ; i<newsize; i++)
1875 nsh->dev[i].page = NULL;
Shaohua Li566c09c2013-11-14 15:16:17 +11001876 nsh->hash_lock_index = hash;
NeilBrownad01c9e2006-03-27 01:18:07 -08001877 kmem_cache_free(conf->slab_cache, osh);
Shaohua Li566c09c2013-11-14 15:16:17 +11001878 cnt++;
1879 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
1880 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
1881 hash++;
1882 cnt = 0;
1883 }
NeilBrownad01c9e2006-03-27 01:18:07 -08001884 }
1885 kmem_cache_destroy(conf->slab_cache);
1886
1887 /* Step 3.
1888 * At this point, we are holding all the stripes so the array
1889 * is completely stalled, so now is a good time to resize
Dan Williamsd6f38f32009-07-14 11:50:52 -07001890 * conf->disks and the scribble region
NeilBrownad01c9e2006-03-27 01:18:07 -08001891 */
1892 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1893 if (ndisks) {
1894 for (i=0; i<conf->raid_disks; i++)
1895 ndisks[i] = conf->disks[i];
1896 kfree(conf->disks);
1897 conf->disks = ndisks;
1898 } else
1899 err = -ENOMEM;
1900
Dan Williamsd6f38f32009-07-14 11:50:52 -07001901 get_online_cpus();
1902 conf->scribble_len = scribble_len(newsize);
1903 for_each_present_cpu(cpu) {
1904 struct raid5_percpu *percpu;
1905 void *scribble;
1906
1907 percpu = per_cpu_ptr(conf->percpu, cpu);
1908 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1909
1910 if (scribble) {
1911 kfree(percpu->scribble);
1912 percpu->scribble = scribble;
1913 } else {
1914 err = -ENOMEM;
1915 break;
1916 }
1917 }
1918 put_online_cpus();
1919
NeilBrownad01c9e2006-03-27 01:18:07 -08001920 /* Step 4, return new stripes to service */
1921 while(!list_empty(&newstripes)) {
1922 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1923 list_del_init(&nsh->lru);
Dan Williamsd6f38f32009-07-14 11:50:52 -07001924
NeilBrownad01c9e2006-03-27 01:18:07 -08001925 for (i=conf->raid_disks; i < newsize; i++)
1926 if (nsh->dev[i].page == NULL) {
1927 struct page *p = alloc_page(GFP_NOIO);
1928 nsh->dev[i].page = p;
Shaohua Lid592a992014-05-21 17:57:44 +08001929 nsh->dev[i].orig_page = p;
NeilBrownad01c9e2006-03-27 01:18:07 -08001930 if (!p)
1931 err = -ENOMEM;
1932 }
1933 release_stripe(nsh);
1934 }
1935 /* critical section pass, GFP_NOIO no longer needed */
1936
1937 conf->slab_cache = sc;
1938 conf->active_name = 1-conf->active_name;
1939 conf->pool_size = newsize;
1940 return err;
1941}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Shaohua Li566c09c2013-11-14 15:16:17 +11001943static int drop_one_stripe(struct r5conf *conf, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944{
1945 struct stripe_head *sh;
1946
Shaohua Li566c09c2013-11-14 15:16:17 +11001947 spin_lock_irq(conf->hash_locks + hash);
1948 sh = get_free_stripe(conf, hash);
1949 spin_unlock_irq(conf->hash_locks + hash);
NeilBrown3f294f42005-11-08 21:39:25 -08001950 if (!sh)
1951 return 0;
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02001952 BUG_ON(atomic_read(&sh->count));
NeilBrowne4e11e32010-06-16 16:45:16 +10001953 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001954 kmem_cache_free(conf->slab_cache, sh);
1955 atomic_dec(&conf->active_stripes);
1956 return 1;
1957}
1958
NeilBrownd1688a62011-10-11 16:49:52 +11001959static void shrink_stripes(struct r5conf *conf)
NeilBrown3f294f42005-11-08 21:39:25 -08001960{
Shaohua Li566c09c2013-11-14 15:16:17 +11001961 int hash;
1962 for (hash = 0; hash < NR_STRIPE_HASH_LOCKS; hash++)
1963 while (drop_one_stripe(conf, hash))
1964 ;
NeilBrown3f294f42005-11-08 21:39:25 -08001965
NeilBrown29fc7e32006-02-03 03:03:41 -08001966 if (conf->slab_cache)
1967 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 conf->slab_cache = NULL;
1969}
1970
NeilBrown6712ecf2007-09-27 12:47:43 +02001971static void raid5_end_read_request(struct bio * bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972{
NeilBrown99c0fb52009-03-31 14:39:38 +11001973 struct stripe_head *sh = bi->bi_private;
NeilBrownd1688a62011-10-11 16:49:52 +11001974 struct r5conf *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001975 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownd6950432006-07-10 04:44:20 -07001977 char b[BDEVNAME_SIZE];
NeilBrowndd054fc2011-12-23 10:17:53 +11001978 struct md_rdev *rdev = NULL;
NeilBrown05616be2012-05-21 09:27:00 +10001979 sector_t s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
1981 for (i=0 ; i<disks; i++)
1982 if (bi == &sh->dev[i].req)
1983 break;
1984
Dan Williams45b42332007-07-09 11:56:43 -07001985 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1986 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 uptodate);
1988 if (i == disks) {
1989 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001990 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 }
NeilBrown14a75d32011-12-23 10:17:52 +11001992 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
NeilBrowndd054fc2011-12-23 10:17:53 +11001993 /* If replacement finished while this request was outstanding,
1994 * 'replacement' might be NULL already.
1995 * In that case it moved down to 'rdev'.
1996 * rdev is not removed until all requests are finished.
1997 */
NeilBrown14a75d32011-12-23 10:17:52 +11001998 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11001999 if (!rdev)
NeilBrown14a75d32011-12-23 10:17:52 +11002000 rdev = conf->disks[i].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
NeilBrown05616be2012-05-21 09:27:00 +10002002 if (use_new_offset(conf, sh))
2003 s = sh->sector + rdev->new_data_offset;
2004 else
2005 s = sh->sector + rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -08002008 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11002009 /* Note that this cannot happen on a
2010 * replacement device. We just fail those on
2011 * any error
2012 */
Christian Dietrich8bda4702011-07-27 11:00:36 +10002013 printk_ratelimited(
2014 KERN_INFO
2015 "md/raid:%s: read error corrected"
2016 " (%lu sectors at %llu on %s)\n",
2017 mdname(conf->mddev), STRIPE_SECTORS,
NeilBrown05616be2012-05-21 09:27:00 +10002018 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002019 bdevname(rdev->bdev, b));
Namhyung Kimddd51152011-07-27 11:00:36 +10002020 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
NeilBrown4e5314b2005-11-08 21:39:22 -08002021 clear_bit(R5_ReadError, &sh->dev[i].flags);
2022 clear_bit(R5_ReWrite, &sh->dev[i].flags);
majianpeng3f9e7c12012-07-31 10:04:21 +10002023 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2024 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2025
NeilBrown14a75d32011-12-23 10:17:52 +11002026 if (atomic_read(&rdev->read_errors))
2027 atomic_set(&rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 } else {
NeilBrown14a75d32011-12-23 10:17:52 +11002029 const char *bdn = bdevname(rdev->bdev, b);
NeilBrownba22dcb2005-11-08 21:39:31 -08002030 int retry = 0;
majianpeng2e8ac3032012-07-03 15:57:02 +10002031 int set_bad = 0;
NeilBrownd6950432006-07-10 04:44:20 -07002032
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07002034 atomic_inc(&rdev->read_errors);
NeilBrown14a75d32011-12-23 10:17:52 +11002035 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
2036 printk_ratelimited(
2037 KERN_WARNING
2038 "md/raid:%s: read error on replacement device "
2039 "(sector %llu on %s).\n",
2040 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002041 (unsigned long long)s,
NeilBrown14a75d32011-12-23 10:17:52 +11002042 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002043 else if (conf->mddev->degraded >= conf->max_degraded) {
2044 set_bad = 1;
Christian Dietrich8bda4702011-07-27 11:00:36 +10002045 printk_ratelimited(
2046 KERN_WARNING
2047 "md/raid:%s: read error not correctable "
2048 "(sector %llu on %s).\n",
2049 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002050 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002051 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002052 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
NeilBrown4e5314b2005-11-08 21:39:22 -08002053 /* Oh, no!!! */
majianpeng2e8ac3032012-07-03 15:57:02 +10002054 set_bad = 1;
Christian Dietrich8bda4702011-07-27 11:00:36 +10002055 printk_ratelimited(
2056 KERN_WARNING
2057 "md/raid:%s: read error NOT corrected!! "
2058 "(sector %llu on %s).\n",
2059 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002060 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002061 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002062 } else if (atomic_read(&rdev->read_errors)
NeilBrownba22dcb2005-11-08 21:39:31 -08002063 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -08002064 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10002065 "md/raid:%s: Too many read errors, failing device %s.\n",
NeilBrownd6950432006-07-10 04:44:20 -07002066 mdname(conf->mddev), bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08002067 else
2068 retry = 1;
Bian Yuedfa1f62013-11-14 15:16:17 +11002069 if (set_bad && test_bit(In_sync, &rdev->flags)
2070 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2071 retry = 1;
NeilBrownba22dcb2005-11-08 21:39:31 -08002072 if (retry)
majianpeng3f9e7c12012-07-31 10:04:21 +10002073 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2074 set_bit(R5_ReadError, &sh->dev[i].flags);
2075 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2076 } else
2077 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
NeilBrownba22dcb2005-11-08 21:39:31 -08002078 else {
NeilBrown4e5314b2005-11-08 21:39:22 -08002079 clear_bit(R5_ReadError, &sh->dev[i].flags);
2080 clear_bit(R5_ReWrite, &sh->dev[i].flags);
majianpeng2e8ac3032012-07-03 15:57:02 +10002081 if (!(set_bad
2082 && test_bit(In_sync, &rdev->flags)
2083 && rdev_set_badblocks(
2084 rdev, sh->sector, STRIPE_SECTORS, 0)))
2085 md_error(conf->mddev, rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -08002086 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 }
NeilBrown14a75d32011-12-23 10:17:52 +11002088 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2090 set_bit(STRIPE_HANDLE, &sh->state);
2091 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092}
2093
NeilBrownd710e132008-10-13 11:55:12 +11002094static void raid5_end_write_request(struct bio *bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095{
NeilBrown99c0fb52009-03-31 14:39:38 +11002096 struct stripe_head *sh = bi->bi_private;
NeilBrownd1688a62011-10-11 16:49:52 +11002097 struct r5conf *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08002098 int disks = sh->disks, i;
NeilBrown977df362011-12-23 10:17:53 +11002099 struct md_rdev *uninitialized_var(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownb84db562011-07-28 11:39:23 +10002101 sector_t first_bad;
2102 int bad_sectors;
NeilBrown977df362011-12-23 10:17:53 +11002103 int replacement = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
NeilBrown977df362011-12-23 10:17:53 +11002105 for (i = 0 ; i < disks; i++) {
2106 if (bi == &sh->dev[i].req) {
2107 rdev = conf->disks[i].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 break;
NeilBrown977df362011-12-23 10:17:53 +11002109 }
2110 if (bi == &sh->dev[i].rreq) {
2111 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11002112 if (rdev)
2113 replacement = 1;
2114 else
2115 /* rdev was removed and 'replacement'
2116 * replaced it. rdev is not removed
2117 * until all requests are finished.
2118 */
2119 rdev = conf->disks[i].rdev;
NeilBrown977df362011-12-23 10:17:53 +11002120 break;
2121 }
2122 }
Dan Williams45b42332007-07-09 11:56:43 -07002123 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
2125 uptodate);
2126 if (i == disks) {
2127 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02002128 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 }
2130
NeilBrown977df362011-12-23 10:17:53 +11002131 if (replacement) {
2132 if (!uptodate)
2133 md_error(conf->mddev, rdev);
2134 else if (is_badblock(rdev, sh->sector,
2135 STRIPE_SECTORS,
2136 &first_bad, &bad_sectors))
2137 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2138 } else {
2139 if (!uptodate) {
NeilBrown9f97e4b2014-01-16 09:35:38 +11002140 set_bit(STRIPE_DEGRADED, &sh->state);
NeilBrown977df362011-12-23 10:17:53 +11002141 set_bit(WriteErrorSeen, &rdev->flags);
2142 set_bit(R5_WriteError, &sh->dev[i].flags);
NeilBrown3a6de292011-12-23 10:17:54 +11002143 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2144 set_bit(MD_RECOVERY_NEEDED,
2145 &rdev->mddev->recovery);
NeilBrown977df362011-12-23 10:17:53 +11002146 } else if (is_badblock(rdev, sh->sector,
2147 STRIPE_SECTORS,
NeilBrownc0b32972013-04-24 11:42:42 +10002148 &first_bad, &bad_sectors)) {
NeilBrown977df362011-12-23 10:17:53 +11002149 set_bit(R5_MadeGood, &sh->dev[i].flags);
NeilBrownc0b32972013-04-24 11:42:42 +10002150 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2151 /* That was a successful write so make
2152 * sure it looks like we already did
2153 * a re-write.
2154 */
2155 set_bit(R5_ReWrite, &sh->dev[i].flags);
2156 }
NeilBrown977df362011-12-23 10:17:53 +11002157 }
2158 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
NeilBrown977df362011-12-23 10:17:53 +11002160 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2161 clear_bit(R5_LOCKED, &sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrownc04be0a2006-10-03 01:15:53 -07002163 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164}
2165
NeilBrown784052e2009-03-31 15:19:07 +11002166static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
Shaohua Lid592a992014-05-21 17:57:44 +08002167
NeilBrown784052e2009-03-31 15:19:07 +11002168static void raid5_build_block(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169{
2170 struct r5dev *dev = &sh->dev[i];
2171
2172 bio_init(&dev->req);
2173 dev->req.bi_io_vec = &dev->vec;
Shaohua Lid592a992014-05-21 17:57:44 +08002174 dev->req.bi_max_vecs = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 dev->req.bi_private = sh;
2176
NeilBrown977df362011-12-23 10:17:53 +11002177 bio_init(&dev->rreq);
2178 dev->rreq.bi_io_vec = &dev->rvec;
Shaohua Lid592a992014-05-21 17:57:44 +08002179 dev->rreq.bi_max_vecs = 1;
NeilBrown977df362011-12-23 10:17:53 +11002180 dev->rreq.bi_private = sh;
NeilBrown977df362011-12-23 10:17:53 +11002181
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +11002183 dev->sector = compute_blocknr(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184}
2185
NeilBrownfd01b882011-10-11 16:47:53 +11002186static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187{
2188 char b[BDEVNAME_SIZE];
NeilBrownd1688a62011-10-11 16:49:52 +11002189 struct r5conf *conf = mddev->private;
NeilBrown908f4fb2011-12-23 10:17:50 +11002190 unsigned long flags;
NeilBrown0c55e022010-05-03 14:09:02 +10002191 pr_debug("raid456: error called\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
NeilBrown908f4fb2011-12-23 10:17:50 +11002193 spin_lock_irqsave(&conf->device_lock, flags);
2194 clear_bit(In_sync, &rdev->flags);
2195 mddev->degraded = calc_degraded(conf);
2196 spin_unlock_irqrestore(&conf->device_lock, flags);
2197 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2198
NeilBrownde393cd2011-07-28 11:31:48 +10002199 set_bit(Blocked, &rdev->flags);
NeilBrown6f8d0c72011-05-11 14:38:44 +10002200 set_bit(Faulty, &rdev->flags);
2201 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2202 printk(KERN_ALERT
2203 "md/raid:%s: Disk failure on %s, disabling device.\n"
2204 "md/raid:%s: Operation continuing on %d devices.\n",
2205 mdname(mddev),
2206 bdevname(rdev->bdev, b),
2207 mdname(mddev),
2208 conf->raid_disks - mddev->degraded);
NeilBrown16a53ec2006-06-26 00:27:38 -07002209}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
2211/*
2212 * Input: a 'big' sector number,
2213 * Output: index of the data and parity disk, and the sector # in them.
2214 */
NeilBrownd1688a62011-10-11 16:49:52 +11002215static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11002216 int previous, int *dd_idx,
2217 struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218{
NeilBrown6e3b96e2010-04-23 07:08:28 +10002219 sector_t stripe, stripe2;
NeilBrown35f2a592010-04-20 14:13:34 +10002220 sector_t chunk_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 unsigned int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11002222 int pd_idx, qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11002223 int ddf_layout = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 sector_t new_sector;
NeilBrowne183eae2009-03-31 15:20:22 +11002225 int algorithm = previous ? conf->prev_algo
2226 : conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10002227 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2228 : conf->chunk_sectors;
NeilBrown112bf892009-03-31 14:39:38 +11002229 int raid_disks = previous ? conf->previous_raid_disks
2230 : conf->raid_disks;
2231 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232
2233 /* First compute the information on this sector */
2234
2235 /*
2236 * Compute the chunk number and the sector offset inside the chunk
2237 */
2238 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2239 chunk_number = r_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
2241 /*
2242 * Compute the stripe number
2243 */
NeilBrown35f2a592010-04-20 14:13:34 +10002244 stripe = chunk_number;
2245 *dd_idx = sector_div(stripe, data_disks);
NeilBrown6e3b96e2010-04-23 07:08:28 +10002246 stripe2 = stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 /*
2248 * Select the parity disk based on the user selected algorithm.
2249 */
NeilBrown84789552011-07-27 11:00:36 +10002250 pd_idx = qd_idx = -1;
NeilBrown16a53ec2006-06-26 00:27:38 -07002251 switch(conf->level) {
2252 case 4:
NeilBrown911d4ee2009-03-31 14:39:38 +11002253 pd_idx = data_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002254 break;
2255 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11002256 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002258 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002259 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 (*dd_idx)++;
2261 break;
2262 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002263 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002264 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 (*dd_idx)++;
2266 break;
2267 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002268 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002269 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 break;
2271 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002272 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002273 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002275 case ALGORITHM_PARITY_0:
2276 pd_idx = 0;
2277 (*dd_idx)++;
2278 break;
2279 case ALGORITHM_PARITY_N:
2280 pd_idx = data_disks;
2281 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002283 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002284 }
2285 break;
2286 case 6:
2287
NeilBrowne183eae2009-03-31 15:20:22 +11002288 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07002289 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002290 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002291 qd_idx = pd_idx + 1;
2292 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11002293 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11002294 qd_idx = 0;
2295 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002296 (*dd_idx) += 2; /* D D P Q D */
2297 break;
2298 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002299 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002300 qd_idx = pd_idx + 1;
2301 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11002302 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11002303 qd_idx = 0;
2304 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002305 (*dd_idx) += 2; /* D D P Q D */
2306 break;
2307 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002308 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002309 qd_idx = (pd_idx + 1) % raid_disks;
2310 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002311 break;
2312 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002313 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002314 qd_idx = (pd_idx + 1) % raid_disks;
2315 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002316 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002317
2318 case ALGORITHM_PARITY_0:
2319 pd_idx = 0;
2320 qd_idx = 1;
2321 (*dd_idx) += 2;
2322 break;
2323 case ALGORITHM_PARITY_N:
2324 pd_idx = data_disks;
2325 qd_idx = data_disks + 1;
2326 break;
2327
2328 case ALGORITHM_ROTATING_ZERO_RESTART:
2329 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2330 * of blocks for computing Q is different.
2331 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002332 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002333 qd_idx = pd_idx + 1;
2334 if (pd_idx == raid_disks-1) {
2335 (*dd_idx)++; /* Q D D D P */
2336 qd_idx = 0;
2337 } else if (*dd_idx >= pd_idx)
2338 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11002339 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002340 break;
2341
2342 case ALGORITHM_ROTATING_N_RESTART:
2343 /* Same a left_asymmetric, by first stripe is
2344 * D D D P Q rather than
2345 * Q D D D P
2346 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002347 stripe2 += 1;
2348 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002349 qd_idx = pd_idx + 1;
2350 if (pd_idx == raid_disks-1) {
2351 (*dd_idx)++; /* Q D D D P */
2352 qd_idx = 0;
2353 } else if (*dd_idx >= pd_idx)
2354 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11002355 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002356 break;
2357
2358 case ALGORITHM_ROTATING_N_CONTINUE:
2359 /* Same as left_symmetric but Q is before P */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002360 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002361 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2362 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
NeilBrown67cc2b82009-03-31 14:39:38 +11002363 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002364 break;
2365
2366 case ALGORITHM_LEFT_ASYMMETRIC_6:
2367 /* RAID5 left_asymmetric, with Q on last device */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002368 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002369 if (*dd_idx >= pd_idx)
2370 (*dd_idx)++;
2371 qd_idx = raid_disks - 1;
2372 break;
2373
2374 case ALGORITHM_RIGHT_ASYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002375 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002376 if (*dd_idx >= pd_idx)
2377 (*dd_idx)++;
2378 qd_idx = raid_disks - 1;
2379 break;
2380
2381 case ALGORITHM_LEFT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002382 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002383 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2384 qd_idx = raid_disks - 1;
2385 break;
2386
2387 case ALGORITHM_RIGHT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002388 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002389 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2390 qd_idx = raid_disks - 1;
2391 break;
2392
2393 case ALGORITHM_PARITY_0_6:
2394 pd_idx = 0;
2395 (*dd_idx)++;
2396 qd_idx = raid_disks - 1;
2397 break;
2398
NeilBrown16a53ec2006-06-26 00:27:38 -07002399 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002400 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002401 }
2402 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 }
2404
NeilBrown911d4ee2009-03-31 14:39:38 +11002405 if (sh) {
2406 sh->pd_idx = pd_idx;
2407 sh->qd_idx = qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11002408 sh->ddf_layout = ddf_layout;
NeilBrown911d4ee2009-03-31 14:39:38 +11002409 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 /*
2411 * Finally, compute the new sector number
2412 */
2413 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2414 return new_sector;
2415}
2416
NeilBrown784052e2009-03-31 15:19:07 +11002417static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418{
NeilBrownd1688a62011-10-11 16:49:52 +11002419 struct r5conf *conf = sh->raid_conf;
NeilBrownb875e532006-12-10 02:20:49 -08002420 int raid_disks = sh->disks;
2421 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 sector_t new_sector = sh->sector, check;
Andre Noll09c9e5f2009-06-18 08:45:55 +10002423 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2424 : conf->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11002425 int algorithm = previous ? conf->prev_algo
2426 : conf->algorithm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 sector_t stripe;
2428 int chunk_offset;
NeilBrown35f2a592010-04-20 14:13:34 +10002429 sector_t chunk_number;
2430 int dummy1, dd_idx = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 sector_t r_sector;
NeilBrown911d4ee2009-03-31 14:39:38 +11002432 struct stripe_head sh2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
2434 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2435 stripe = new_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
NeilBrown16a53ec2006-06-26 00:27:38 -07002437 if (i == sh->pd_idx)
2438 return 0;
2439 switch(conf->level) {
2440 case 4: break;
2441 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11002442 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 case ALGORITHM_LEFT_ASYMMETRIC:
2444 case ALGORITHM_RIGHT_ASYMMETRIC:
2445 if (i > sh->pd_idx)
2446 i--;
2447 break;
2448 case ALGORITHM_LEFT_SYMMETRIC:
2449 case ALGORITHM_RIGHT_SYMMETRIC:
2450 if (i < sh->pd_idx)
2451 i += raid_disks;
2452 i -= (sh->pd_idx + 1);
2453 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002454 case ALGORITHM_PARITY_0:
2455 i -= 1;
2456 break;
2457 case ALGORITHM_PARITY_N:
2458 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002460 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002461 }
2462 break;
2463 case 6:
NeilBrownd0dabf72009-03-31 14:39:38 +11002464 if (i == sh->qd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002465 return 0; /* It is the Q disk */
NeilBrowne183eae2009-03-31 15:20:22 +11002466 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07002467 case ALGORITHM_LEFT_ASYMMETRIC:
2468 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown99c0fb52009-03-31 14:39:38 +11002469 case ALGORITHM_ROTATING_ZERO_RESTART:
2470 case ALGORITHM_ROTATING_N_RESTART:
2471 if (sh->pd_idx == raid_disks-1)
2472 i--; /* Q D D D P */
NeilBrown16a53ec2006-06-26 00:27:38 -07002473 else if (i > sh->pd_idx)
2474 i -= 2; /* D D P Q D */
2475 break;
2476 case ALGORITHM_LEFT_SYMMETRIC:
2477 case ALGORITHM_RIGHT_SYMMETRIC:
2478 if (sh->pd_idx == raid_disks-1)
2479 i--; /* Q D D D P */
2480 else {
2481 /* D D P Q D */
2482 if (i < sh->pd_idx)
2483 i += raid_disks;
2484 i -= (sh->pd_idx + 2);
2485 }
2486 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002487 case ALGORITHM_PARITY_0:
2488 i -= 2;
2489 break;
2490 case ALGORITHM_PARITY_N:
2491 break;
2492 case ALGORITHM_ROTATING_N_CONTINUE:
NeilBrowne4424fe2009-10-16 16:27:34 +11002493 /* Like left_symmetric, but P is before Q */
NeilBrown99c0fb52009-03-31 14:39:38 +11002494 if (sh->pd_idx == 0)
2495 i--; /* P D D D Q */
NeilBrowne4424fe2009-10-16 16:27:34 +11002496 else {
2497 /* D D Q P D */
2498 if (i < sh->pd_idx)
2499 i += raid_disks;
2500 i -= (sh->pd_idx + 1);
2501 }
NeilBrown99c0fb52009-03-31 14:39:38 +11002502 break;
2503 case ALGORITHM_LEFT_ASYMMETRIC_6:
2504 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2505 if (i > sh->pd_idx)
2506 i--;
2507 break;
2508 case ALGORITHM_LEFT_SYMMETRIC_6:
2509 case ALGORITHM_RIGHT_SYMMETRIC_6:
2510 if (i < sh->pd_idx)
2511 i += data_disks + 1;
2512 i -= (sh->pd_idx + 1);
2513 break;
2514 case ALGORITHM_PARITY_0_6:
2515 i -= 1;
2516 break;
NeilBrown16a53ec2006-06-26 00:27:38 -07002517 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002518 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002519 }
2520 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 }
2522
2523 chunk_number = stripe * data_disks + i;
NeilBrown35f2a592010-04-20 14:13:34 +10002524 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
NeilBrown112bf892009-03-31 14:39:38 +11002526 check = raid5_compute_sector(conf, r_sector,
NeilBrown784052e2009-03-31 15:19:07 +11002527 previous, &dummy1, &sh2);
NeilBrown911d4ee2009-03-31 14:39:38 +11002528 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2529 || sh2.qd_idx != sh->qd_idx) {
NeilBrown0c55e022010-05-03 14:09:02 +10002530 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2531 mdname(conf->mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 return 0;
2533 }
2534 return r_sector;
2535}
2536
Dan Williams600aa102008-06-28 08:32:05 +10002537static void
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002538schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
Dan Williams600aa102008-06-28 08:32:05 +10002539 int rcw, int expand)
Dan Williamse33129d2007-01-02 13:52:30 -07002540{
2541 int i, pd_idx = sh->pd_idx, disks = sh->disks;
NeilBrownd1688a62011-10-11 16:49:52 +11002542 struct r5conf *conf = sh->raid_conf;
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002543 int level = conf->level;
NeilBrown16a53ec2006-06-26 00:27:38 -07002544
Dan Williamse33129d2007-01-02 13:52:30 -07002545 if (rcw) {
Dan Williamse33129d2007-01-02 13:52:30 -07002546
2547 for (i = disks; i--; ) {
2548 struct r5dev *dev = &sh->dev[i];
2549
2550 if (dev->towrite) {
2551 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsd8ee0722008-06-28 08:32:06 +10002552 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002553 if (!expand)
2554 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002555 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002556 }
2557 }
NeilBrownce7d3632013-03-04 12:37:14 +11002558 /* if we are not expanding this is a proper write request, and
2559 * there will be bios with new data to be drained into the
2560 * stripe cache
2561 */
2562 if (!expand) {
2563 if (!s->locked)
2564 /* False alarm, nothing to do */
2565 return;
2566 sh->reconstruct_state = reconstruct_state_drain_run;
2567 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2568 } else
2569 sh->reconstruct_state = reconstruct_state_run;
2570
2571 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2572
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002573 if (s->locked + conf->max_degraded == disks)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002574 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002575 atomic_inc(&conf->pending_full_writes);
Dan Williamse33129d2007-01-02 13:52:30 -07002576 } else {
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002577 BUG_ON(level == 6);
Dan Williamse33129d2007-01-02 13:52:30 -07002578 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2579 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2580
Dan Williamse33129d2007-01-02 13:52:30 -07002581 for (i = disks; i--; ) {
2582 struct r5dev *dev = &sh->dev[i];
2583 if (i == pd_idx)
2584 continue;
2585
Dan Williamse33129d2007-01-02 13:52:30 -07002586 if (dev->towrite &&
2587 (test_bit(R5_UPTODATE, &dev->flags) ||
Dan Williamsd8ee0722008-06-28 08:32:06 +10002588 test_bit(R5_Wantcompute, &dev->flags))) {
2589 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002590 set_bit(R5_LOCKED, &dev->flags);
2591 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002592 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002593 }
2594 }
NeilBrownce7d3632013-03-04 12:37:14 +11002595 if (!s->locked)
2596 /* False alarm - nothing to do */
2597 return;
2598 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2599 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2600 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2601 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002602 }
2603
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002604 /* keep the parity disk(s) locked while asynchronous operations
Dan Williamse33129d2007-01-02 13:52:30 -07002605 * are in flight
2606 */
2607 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2608 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
Dan Williams600aa102008-06-28 08:32:05 +10002609 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002610
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002611 if (level == 6) {
2612 int qd_idx = sh->qd_idx;
2613 struct r5dev *dev = &sh->dev[qd_idx];
2614
2615 set_bit(R5_LOCKED, &dev->flags);
2616 clear_bit(R5_UPTODATE, &dev->flags);
2617 s->locked++;
2618 }
2619
Dan Williams600aa102008-06-28 08:32:05 +10002620 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07002621 __func__, (unsigned long long)sh->sector,
Dan Williams600aa102008-06-28 08:32:05 +10002622 s->locked, s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002623}
NeilBrown16a53ec2006-06-26 00:27:38 -07002624
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625/*
2626 * Each stripe/dev can have one or more bion attached.
NeilBrown16a53ec2006-06-26 00:27:38 -07002627 * toread/towrite point to the first in a chain.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 * The bi_next chain must be in order.
2629 */
2630static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2631{
2632 struct bio **bip;
NeilBrownd1688a62011-10-11 16:49:52 +11002633 struct r5conf *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -07002634 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635
NeilBrowncbe47ec2011-07-26 11:20:35 +10002636 pr_debug("adding bi b#%llu to stripe s#%llu\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002637 (unsigned long long)bi->bi_iter.bi_sector,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 (unsigned long long)sh->sector);
2639
Shaohua Lib17459c2012-07-19 16:01:31 +10002640 /*
2641 * If several bio share a stripe. The bio bi_phys_segments acts as a
2642 * reference count to avoid race. The reference count should already be
2643 * increased before this function is called (for example, in
2644 * make_request()), so other bio sharing this stripe will not free the
2645 * stripe. If a stripe is owned by one stripe, the stripe lock will
2646 * protect it.
2647 */
2648 spin_lock_irq(&sh->stripe_lock);
NeilBrown72626682005-09-09 16:23:54 -07002649 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 bip = &sh->dev[dd_idx].towrite;
Shaohua Li7eaf7e82012-07-19 16:01:31 +10002651 if (*bip == NULL)
NeilBrown72626682005-09-09 16:23:54 -07002652 firstwrite = 1;
2653 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 bip = &sh->dev[dd_idx].toread;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002655 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
2656 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 goto overlap;
2658 bip = & (*bip)->bi_next;
2659 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07002660 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 goto overlap;
2662
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002663 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 if (*bip)
2665 bi->bi_next = *bip;
2666 *bip = bi;
Shaohua Lie7836bd62012-07-19 16:01:31 +10002667 raid5_inc_bi_active_stripes(bi);
NeilBrown72626682005-09-09 16:23:54 -07002668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 if (forwrite) {
2670 /* check if page is covered */
2671 sector_t sector = sh->dev[dd_idx].sector;
2672 for (bi=sh->dev[dd_idx].towrite;
2673 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
Kent Overstreet4f024f32013-10-11 15:44:27 -07002674 bi && bi->bi_iter.bi_sector <= sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
Kent Overstreetf73a1c72012-09-25 15:05:12 -07002676 if (bio_end_sector(bi) >= sector)
2677 sector = bio_end_sector(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 }
2679 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2680 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2681 }
NeilBrowncbe47ec2011-07-26 11:20:35 +10002682
2683 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002684 (unsigned long long)(*bip)->bi_iter.bi_sector,
NeilBrowncbe47ec2011-07-26 11:20:35 +10002685 (unsigned long long)sh->sector, dd_idx);
NeilBrownb97390a2012-10-11 13:50:12 +11002686 spin_unlock_irq(&sh->stripe_lock);
NeilBrowncbe47ec2011-07-26 11:20:35 +10002687
2688 if (conf->mddev->bitmap && firstwrite) {
2689 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2690 STRIPE_SECTORS, 0);
2691 sh->bm_seq = conf->seq_flush+1;
2692 set_bit(STRIPE_BIT_DELAY, &sh->state);
2693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 return 1;
2695
2696 overlap:
2697 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
Shaohua Lib17459c2012-07-19 16:01:31 +10002698 spin_unlock_irq(&sh->stripe_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 return 0;
2700}
2701
NeilBrownd1688a62011-10-11 16:49:52 +11002702static void end_reshape(struct r5conf *conf);
NeilBrown29269552006-03-27 01:18:10 -08002703
NeilBrownd1688a62011-10-11 16:49:52 +11002704static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002705 struct stripe_head *sh)
NeilBrownccfcc3c2006-03-27 01:18:09 -08002706{
NeilBrown784052e2009-03-31 15:19:07 +11002707 int sectors_per_chunk =
Andre Noll09c9e5f2009-06-18 08:45:55 +10002708 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
NeilBrown911d4ee2009-03-31 14:39:38 +11002709 int dd_idx;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002710 int chunk_offset = sector_div(stripe, sectors_per_chunk);
NeilBrown112bf892009-03-31 14:39:38 +11002711 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002712
NeilBrown112bf892009-03-31 14:39:38 +11002713 raid5_compute_sector(conf,
2714 stripe * (disks - conf->max_degraded)
NeilBrownb875e532006-12-10 02:20:49 -08002715 *sectors_per_chunk + chunk_offset,
NeilBrown112bf892009-03-31 14:39:38 +11002716 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002717 &dd_idx, sh);
NeilBrownccfcc3c2006-03-27 01:18:09 -08002718}
2719
Dan Williamsa4456852007-07-09 11:56:43 -07002720static void
NeilBrownd1688a62011-10-11 16:49:52 +11002721handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002722 struct stripe_head_state *s, int disks,
2723 struct bio **return_bi)
2724{
2725 int i;
2726 for (i = disks; i--; ) {
2727 struct bio *bi;
2728 int bitmap_end = 0;
2729
2730 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown3cb03002011-10-11 16:45:26 +11002731 struct md_rdev *rdev;
Dan Williamsa4456852007-07-09 11:56:43 -07002732 rcu_read_lock();
2733 rdev = rcu_dereference(conf->disks[i].rdev);
2734 if (rdev && test_bit(In_sync, &rdev->flags))
NeilBrown7f0da592011-07-28 11:39:22 +10002735 atomic_inc(&rdev->nr_pending);
2736 else
2737 rdev = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07002738 rcu_read_unlock();
NeilBrown7f0da592011-07-28 11:39:22 +10002739 if (rdev) {
2740 if (!rdev_set_badblocks(
2741 rdev,
2742 sh->sector,
2743 STRIPE_SECTORS, 0))
2744 md_error(conf->mddev, rdev);
2745 rdev_dec_pending(rdev, conf->mddev);
2746 }
Dan Williamsa4456852007-07-09 11:56:43 -07002747 }
Shaohua Lib17459c2012-07-19 16:01:31 +10002748 spin_lock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07002749 /* fail all writes first */
2750 bi = sh->dev[i].towrite;
2751 sh->dev[i].towrite = NULL;
Shaohua Lib17459c2012-07-19 16:01:31 +10002752 spin_unlock_irq(&sh->stripe_lock);
NeilBrown1ed850f2012-10-11 13:50:13 +11002753 if (bi)
Dan Williamsa4456852007-07-09 11:56:43 -07002754 bitmap_end = 1;
Dan Williamsa4456852007-07-09 11:56:43 -07002755
2756 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2757 wake_up(&conf->wait_for_overlap);
2758
Kent Overstreet4f024f32013-10-11 15:44:27 -07002759 while (bi && bi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07002760 sh->dev[i].sector + STRIPE_SECTORS) {
2761 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2762 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Shaohua Lie7836bd62012-07-19 16:01:31 +10002763 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002764 md_write_end(conf->mddev);
2765 bi->bi_next = *return_bi;
2766 *return_bi = bi;
2767 }
2768 bi = nextbi;
2769 }
Shaohua Li7eaf7e82012-07-19 16:01:31 +10002770 if (bitmap_end)
2771 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2772 STRIPE_SECTORS, 0, 0);
2773 bitmap_end = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07002774 /* and fail all 'written' */
2775 bi = sh->dev[i].written;
2776 sh->dev[i].written = NULL;
Shaohua Lid592a992014-05-21 17:57:44 +08002777 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
2778 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
2779 sh->dev[i].page = sh->dev[i].orig_page;
2780 }
2781
Dan Williamsa4456852007-07-09 11:56:43 -07002782 if (bi) bitmap_end = 1;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002783 while (bi && bi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07002784 sh->dev[i].sector + STRIPE_SECTORS) {
2785 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2786 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Shaohua Lie7836bd62012-07-19 16:01:31 +10002787 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002788 md_write_end(conf->mddev);
2789 bi->bi_next = *return_bi;
2790 *return_bi = bi;
2791 }
2792 bi = bi2;
2793 }
2794
Dan Williamsb5e98d62007-01-02 13:52:31 -07002795 /* fail any reads if this device is non-operational and
2796 * the data has not reached the cache yet.
2797 */
2798 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2799 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2800 test_bit(R5_ReadError, &sh->dev[i].flags))) {
NeilBrown143c4d02012-10-11 13:50:12 +11002801 spin_lock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07002802 bi = sh->dev[i].toread;
2803 sh->dev[i].toread = NULL;
NeilBrown143c4d02012-10-11 13:50:12 +11002804 spin_unlock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07002805 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2806 wake_up(&conf->wait_for_overlap);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002807 while (bi && bi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07002808 sh->dev[i].sector + STRIPE_SECTORS) {
2809 struct bio *nextbi =
2810 r5_next_bio(bi, sh->dev[i].sector);
2811 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Shaohua Lie7836bd62012-07-19 16:01:31 +10002812 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002813 bi->bi_next = *return_bi;
2814 *return_bi = bi;
2815 }
2816 bi = nextbi;
2817 }
2818 }
Dan Williamsa4456852007-07-09 11:56:43 -07002819 if (bitmap_end)
2820 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2821 STRIPE_SECTORS, 0, 0);
NeilBrown8cfa7b02011-07-27 11:00:36 +10002822 /* If we were in the middle of a write the parity block might
2823 * still be locked - so just clear all R5_LOCKED flags
2824 */
2825 clear_bit(R5_LOCKED, &sh->dev[i].flags);
Dan Williamsa4456852007-07-09 11:56:43 -07002826 }
2827
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002828 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2829 if (atomic_dec_and_test(&conf->pending_full_writes))
2830 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002831}
2832
NeilBrown7f0da592011-07-28 11:39:22 +10002833static void
NeilBrownd1688a62011-10-11 16:49:52 +11002834handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
NeilBrown7f0da592011-07-28 11:39:22 +10002835 struct stripe_head_state *s)
2836{
2837 int abort = 0;
2838 int i;
2839
NeilBrown7f0da592011-07-28 11:39:22 +10002840 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11002841 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
2842 wake_up(&conf->wait_for_overlap);
NeilBrown7f0da592011-07-28 11:39:22 +10002843 s->syncing = 0;
NeilBrown9a3e1102011-12-23 10:17:53 +11002844 s->replacing = 0;
NeilBrown7f0da592011-07-28 11:39:22 +10002845 /* There is nothing more to do for sync/check/repair.
NeilBrown18b98372012-04-01 23:48:38 +10002846 * Don't even need to abort as that is handled elsewhere
2847 * if needed, and not always wanted e.g. if there is a known
2848 * bad block here.
NeilBrown9a3e1102011-12-23 10:17:53 +11002849 * For recover/replace we need to record a bad block on all
NeilBrown7f0da592011-07-28 11:39:22 +10002850 * non-sync devices, or abort the recovery
2851 */
NeilBrown18b98372012-04-01 23:48:38 +10002852 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
2853 /* During recovery devices cannot be removed, so
2854 * locking and refcounting of rdevs is not needed
2855 */
2856 for (i = 0; i < conf->raid_disks; i++) {
2857 struct md_rdev *rdev = conf->disks[i].rdev;
2858 if (rdev
2859 && !test_bit(Faulty, &rdev->flags)
2860 && !test_bit(In_sync, &rdev->flags)
2861 && !rdev_set_badblocks(rdev, sh->sector,
2862 STRIPE_SECTORS, 0))
2863 abort = 1;
2864 rdev = conf->disks[i].replacement;
2865 if (rdev
2866 && !test_bit(Faulty, &rdev->flags)
2867 && !test_bit(In_sync, &rdev->flags)
2868 && !rdev_set_badblocks(rdev, sh->sector,
2869 STRIPE_SECTORS, 0))
2870 abort = 1;
2871 }
2872 if (abort)
2873 conf->recovery_disabled =
2874 conf->mddev->recovery_disabled;
NeilBrown7f0da592011-07-28 11:39:22 +10002875 }
NeilBrown18b98372012-04-01 23:48:38 +10002876 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
NeilBrown7f0da592011-07-28 11:39:22 +10002877}
2878
NeilBrown9a3e1102011-12-23 10:17:53 +11002879static int want_replace(struct stripe_head *sh, int disk_idx)
2880{
2881 struct md_rdev *rdev;
2882 int rv = 0;
2883 /* Doing recovery so rcu locking not required */
2884 rdev = sh->raid_conf->disks[disk_idx].replacement;
2885 if (rdev
2886 && !test_bit(Faulty, &rdev->flags)
2887 && !test_bit(In_sync, &rdev->flags)
2888 && (rdev->recovery_offset <= sh->sector
2889 || rdev->mddev->recovery_cp <= sh->sector))
2890 rv = 1;
2891
2892 return rv;
2893}
2894
NeilBrown93b3dbc2011-07-27 11:00:36 +10002895/* fetch_block - checks the given member device to see if its data needs
Dan Williams1fe797e2008-06-28 09:16:30 +10002896 * to be read or computed to satisfy a request.
2897 *
2898 * Returns 1 when no more member devices need to be checked, otherwise returns
NeilBrown93b3dbc2011-07-27 11:00:36 +10002899 * 0 to tell the loop in handle_stripe_fill to continue
Dan Williamsf38e1212007-01-02 13:52:30 -07002900 */
NeilBrown93b3dbc2011-07-27 11:00:36 +10002901static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
2902 int disk_idx, int disks)
Dan Williamsf38e1212007-01-02 13:52:30 -07002903{
2904 struct r5dev *dev = &sh->dev[disk_idx];
NeilBrown93b3dbc2011-07-27 11:00:36 +10002905 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2906 &sh->dev[s->failed_num[1]] };
Dan Williamsf38e1212007-01-02 13:52:30 -07002907
Dan Williamsf38e1212007-01-02 13:52:30 -07002908 /* is the data in this block needed, and can we get it? */
2909 if (!test_bit(R5_LOCKED, &dev->flags) &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002910 !test_bit(R5_UPTODATE, &dev->flags) &&
2911 (dev->toread ||
2912 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2913 s->syncing || s->expanding ||
NeilBrown9a3e1102011-12-23 10:17:53 +11002914 (s->replacing && want_replace(sh, disk_idx)) ||
NeilBrown5d35e092011-07-27 11:00:36 +10002915 (s->failed >= 1 && fdev[0]->toread) ||
2916 (s->failed >= 2 && fdev[1]->toread) ||
NeilBrown93b3dbc2011-07-27 11:00:36 +10002917 (sh->raid_conf->level <= 5 && s->failed && fdev[0]->towrite &&
NeilBrown67f45542014-05-28 13:39:22 +10002918 (!test_bit(R5_Insync, &dev->flags) || test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) &&
NeilBrown93b3dbc2011-07-27 11:00:36 +10002919 !test_bit(R5_OVERWRITE, &fdev[0]->flags)) ||
NeilBrown108cef32014-12-03 16:07:58 +11002920 ((sh->raid_conf->level == 6 ||
2921 sh->sector >= sh->raid_conf->mddev->recovery_cp)
2922 && s->failed && s->to_write &&
2923 (s->to_write - s->non_overwrite <
2924 sh->raid_conf->raid_disks - sh->raid_conf->max_degraded) &&
NeilBrown67f45542014-05-28 13:39:22 +10002925 (!test_bit(R5_Insync, &dev->flags) || test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))))) {
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002926 /* we would like to get this block, possibly by computing it,
2927 * otherwise read it if the backing disk is insync
2928 */
2929 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2930 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2931 if ((s->uptodate == disks - 1) &&
NeilBrownf2b3b442011-07-26 11:35:19 +10002932 (s->failed && (disk_idx == s->failed_num[0] ||
2933 disk_idx == s->failed_num[1]))) {
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002934 /* have disk failed, and we're requested to fetch it;
2935 * do compute it
2936 */
2937 pr_debug("Computing stripe %llu block %d\n",
2938 (unsigned long long)sh->sector, disk_idx);
2939 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2940 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2941 set_bit(R5_Wantcompute, &dev->flags);
2942 sh->ops.target = disk_idx;
2943 sh->ops.target2 = -1; /* no 2nd target */
2944 s->req_compute = 1;
NeilBrown93b3dbc2011-07-27 11:00:36 +10002945 /* Careful: from this point on 'uptodate' is in the eye
2946 * of raid_run_ops which services 'compute' operations
2947 * before writes. R5_Wantcompute flags a block that will
2948 * be R5_UPTODATE by the time it is needed for a
2949 * subsequent operation.
2950 */
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002951 s->uptodate++;
2952 return 1;
2953 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2954 /* Computing 2-failure is *very* expensive; only
2955 * do it if failed >= 2
2956 */
2957 int other;
2958 for (other = disks; other--; ) {
2959 if (other == disk_idx)
2960 continue;
2961 if (!test_bit(R5_UPTODATE,
2962 &sh->dev[other].flags))
2963 break;
2964 }
2965 BUG_ON(other < 0);
2966 pr_debug("Computing stripe %llu blocks %d,%d\n",
2967 (unsigned long long)sh->sector,
2968 disk_idx, other);
2969 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2970 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2971 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2972 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2973 sh->ops.target = disk_idx;
2974 sh->ops.target2 = other;
2975 s->uptodate += 2;
2976 s->req_compute = 1;
2977 return 1;
2978 } else if (test_bit(R5_Insync, &dev->flags)) {
2979 set_bit(R5_LOCKED, &dev->flags);
2980 set_bit(R5_Wantread, &dev->flags);
2981 s->locked++;
2982 pr_debug("Reading block %d (sync=%d)\n",
2983 disk_idx, s->syncing);
2984 }
2985 }
2986
2987 return 0;
2988}
2989
2990/**
NeilBrown93b3dbc2011-07-27 11:00:36 +10002991 * handle_stripe_fill - read or compute data to satisfy pending requests.
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002992 */
NeilBrown93b3dbc2011-07-27 11:00:36 +10002993static void handle_stripe_fill(struct stripe_head *sh,
2994 struct stripe_head_state *s,
2995 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07002996{
2997 int i;
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002998
2999 /* look for blocks to read/compute, skip this if a compute
3000 * is already in flight, or if the stripe contents are in the
3001 * midst of changing due to a write
3002 */
3003 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
3004 !sh->reconstruct_state)
3005 for (i = disks; i--; )
NeilBrown93b3dbc2011-07-27 11:00:36 +10003006 if (fetch_block(sh, s, i, disks))
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003007 break;
Dan Williamsa4456852007-07-09 11:56:43 -07003008 set_bit(STRIPE_HANDLE, &sh->state);
3009}
3010
Dan Williams1fe797e2008-06-28 09:16:30 +10003011/* handle_stripe_clean_event
Dan Williamsa4456852007-07-09 11:56:43 -07003012 * any written block on an uptodate or failed drive can be returned.
3013 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3014 * never LOCKED, so we don't need to test 'failed' directly.
3015 */
NeilBrownd1688a62011-10-11 16:49:52 +11003016static void handle_stripe_clean_event(struct r5conf *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07003017 struct stripe_head *sh, int disks, struct bio **return_bi)
3018{
3019 int i;
3020 struct r5dev *dev;
NeilBrownf8dfcff2013-03-12 12:18:06 +11003021 int discard_pending = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07003022
3023 for (i = disks; i--; )
3024 if (sh->dev[i].written) {
3025 dev = &sh->dev[i];
3026 if (!test_bit(R5_LOCKED, &dev->flags) &&
Shaohua Li9e4447682012-10-11 13:49:49 +11003027 (test_bit(R5_UPTODATE, &dev->flags) ||
Shaohua Lid592a992014-05-21 17:57:44 +08003028 test_bit(R5_Discard, &dev->flags) ||
3029 test_bit(R5_SkipCopy, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07003030 /* We can return any write requests */
3031 struct bio *wbi, *wbi2;
Dan Williams45b42332007-07-09 11:56:43 -07003032 pr_debug("Return write for disc %d\n", i);
NeilBrownca64cae2012-11-21 16:33:40 +11003033 if (test_and_clear_bit(R5_Discard, &dev->flags))
3034 clear_bit(R5_UPTODATE, &dev->flags);
Shaohua Lid592a992014-05-21 17:57:44 +08003035 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3036 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
3037 dev->page = dev->orig_page;
3038 }
Dan Williamsa4456852007-07-09 11:56:43 -07003039 wbi = dev->written;
3040 dev->written = NULL;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003041 while (wbi && wbi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07003042 dev->sector + STRIPE_SECTORS) {
3043 wbi2 = r5_next_bio(wbi, dev->sector);
Shaohua Lie7836bd62012-07-19 16:01:31 +10003044 if (!raid5_dec_bi_active_stripes(wbi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003045 md_write_end(conf->mddev);
3046 wbi->bi_next = *return_bi;
3047 *return_bi = wbi;
3048 }
3049 wbi = wbi2;
3050 }
Shaohua Li7eaf7e82012-07-19 16:01:31 +10003051 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3052 STRIPE_SECTORS,
Dan Williamsa4456852007-07-09 11:56:43 -07003053 !test_bit(STRIPE_DEGRADED, &sh->state),
Shaohua Li7eaf7e82012-07-19 16:01:31 +10003054 0);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003055 } else if (test_bit(R5_Discard, &dev->flags))
3056 discard_pending = 1;
Shaohua Lid592a992014-05-21 17:57:44 +08003057 WARN_ON(test_bit(R5_SkipCopy, &dev->flags));
3058 WARN_ON(dev->page != dev->orig_page);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003059 }
3060 if (!discard_pending &&
3061 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
3062 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3063 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3064 if (sh->qd_idx >= 0) {
3065 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3066 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3067 }
3068 /* now that discard is done we can proceed with any sync */
3069 clear_bit(STRIPE_DISCARD, &sh->state);
Shaohua Lid47648f2013-10-19 14:51:42 +08003070 /*
3071 * SCSI discard will change some bio fields and the stripe has
3072 * no updated data, so remove it from hash list and the stripe
3073 * will be reinitialized
3074 */
3075 spin_lock_irq(&conf->device_lock);
3076 remove_hash(sh);
3077 spin_unlock_irq(&conf->device_lock);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003078 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3079 set_bit(STRIPE_HANDLE, &sh->state);
3080
3081 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003082
3083 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3084 if (atomic_dec_and_test(&conf->pending_full_writes))
3085 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07003086}
3087
NeilBrownd1688a62011-10-11 16:49:52 +11003088static void handle_stripe_dirtying(struct r5conf *conf,
NeilBrownc8ac1802011-07-27 11:00:36 +10003089 struct stripe_head *sh,
3090 struct stripe_head_state *s,
3091 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07003092{
3093 int rmw = 0, rcw = 0, i;
Alexander Lyakasa7854482012-10-11 13:50:12 +11003094 sector_t recovery_cp = conf->mddev->recovery_cp;
3095
3096 /* RAID6 requires 'rcw' in current implementation.
3097 * Otherwise, check whether resync is now happening or should start.
3098 * If yes, then the array is dirty (after unclean shutdown or
3099 * initial creation), so parity in some stripes might be inconsistent.
3100 * In this case, we need to always do reconstruct-write, to ensure
3101 * that in case of drive failure or read-error correction, we
3102 * generate correct data from the parity.
3103 */
3104 if (conf->max_degraded == 2 ||
3105 (recovery_cp < MaxSector && sh->sector >= recovery_cp)) {
3106 /* Calculate the real rcw later - for now make it
NeilBrownc8ac1802011-07-27 11:00:36 +10003107 * look like rcw is cheaper
3108 */
3109 rcw = 1; rmw = 2;
Alexander Lyakasa7854482012-10-11 13:50:12 +11003110 pr_debug("force RCW max_degraded=%u, recovery_cp=%llu sh->sector=%llu\n",
3111 conf->max_degraded, (unsigned long long)recovery_cp,
3112 (unsigned long long)sh->sector);
NeilBrownc8ac1802011-07-27 11:00:36 +10003113 } else for (i = disks; i--; ) {
Dan Williamsa4456852007-07-09 11:56:43 -07003114 /* would I have to read this buffer for read_modify_write */
3115 struct r5dev *dev = &sh->dev[i];
3116 if ((dev->towrite || i == sh->pd_idx) &&
3117 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003118 !(test_bit(R5_UPTODATE, &dev->flags) ||
3119 test_bit(R5_Wantcompute, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07003120 if (test_bit(R5_Insync, &dev->flags))
3121 rmw++;
3122 else
3123 rmw += 2*disks; /* cannot read it */
3124 }
3125 /* Would I have to read this buffer for reconstruct_write */
3126 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
3127 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003128 !(test_bit(R5_UPTODATE, &dev->flags) ||
3129 test_bit(R5_Wantcompute, &dev->flags))) {
NeilBrown67f45542014-05-28 13:39:22 +10003130 if (test_bit(R5_Insync, &dev->flags))
3131 rcw++;
Dan Williamsa4456852007-07-09 11:56:43 -07003132 else
3133 rcw += 2*disks;
3134 }
3135 }
Dan Williams45b42332007-07-09 11:56:43 -07003136 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003137 (unsigned long long)sh->sector, rmw, rcw);
3138 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrowna9add5d2012-10-31 11:59:09 +11003139 if (rmw < rcw && rmw > 0) {
Dan Williamsa4456852007-07-09 11:56:43 -07003140 /* prefer read-modify-write, but need to get some data */
Jonathan Brassowe3620a32013-03-07 16:22:01 -06003141 if (conf->mddev->queue)
3142 blk_add_trace_msg(conf->mddev->queue,
3143 "raid5 rmw %llu %d",
3144 (unsigned long long)sh->sector, rmw);
Dan Williamsa4456852007-07-09 11:56:43 -07003145 for (i = disks; i--; ) {
3146 struct r5dev *dev = &sh->dev[i];
3147 if ((dev->towrite || i == sh->pd_idx) &&
3148 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003149 !(test_bit(R5_UPTODATE, &dev->flags) ||
3150 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07003151 test_bit(R5_Insync, &dev->flags)) {
NeilBrown67f45542014-05-28 13:39:22 +10003152 if (test_bit(STRIPE_PREREAD_ACTIVE,
3153 &sh->state)) {
3154 pr_debug("Read_old block %d for r-m-w\n",
3155 i);
Dan Williamsa4456852007-07-09 11:56:43 -07003156 set_bit(R5_LOCKED, &dev->flags);
3157 set_bit(R5_Wantread, &dev->flags);
3158 s->locked++;
3159 } else {
3160 set_bit(STRIPE_DELAYED, &sh->state);
3161 set_bit(STRIPE_HANDLE, &sh->state);
3162 }
3163 }
3164 }
NeilBrowna9add5d2012-10-31 11:59:09 +11003165 }
NeilBrownc8ac1802011-07-27 11:00:36 +10003166 if (rcw <= rmw && rcw > 0) {
Dan Williamsa4456852007-07-09 11:56:43 -07003167 /* want reconstruct write, but need to get some data */
NeilBrowna9add5d2012-10-31 11:59:09 +11003168 int qread =0;
NeilBrownc8ac1802011-07-27 11:00:36 +10003169 rcw = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07003170 for (i = disks; i--; ) {
3171 struct r5dev *dev = &sh->dev[i];
3172 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
NeilBrownc8ac1802011-07-27 11:00:36 +10003173 i != sh->pd_idx && i != sh->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07003174 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003175 !(test_bit(R5_UPTODATE, &dev->flags) ||
NeilBrownc8ac1802011-07-27 11:00:36 +10003176 test_bit(R5_Wantcompute, &dev->flags))) {
3177 rcw++;
NeilBrown67f45542014-05-28 13:39:22 +10003178 if (test_bit(R5_Insync, &dev->flags) &&
3179 test_bit(STRIPE_PREREAD_ACTIVE,
3180 &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07003181 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07003182 "%d for Reconstruct\n", i);
3183 set_bit(R5_LOCKED, &dev->flags);
3184 set_bit(R5_Wantread, &dev->flags);
3185 s->locked++;
NeilBrowna9add5d2012-10-31 11:59:09 +11003186 qread++;
Dan Williamsa4456852007-07-09 11:56:43 -07003187 } else {
3188 set_bit(STRIPE_DELAYED, &sh->state);
3189 set_bit(STRIPE_HANDLE, &sh->state);
3190 }
3191 }
3192 }
Jonathan Brassowe3620a32013-03-07 16:22:01 -06003193 if (rcw && conf->mddev->queue)
NeilBrowna9add5d2012-10-31 11:59:09 +11003194 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3195 (unsigned long long)sh->sector,
3196 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
NeilBrownc8ac1802011-07-27 11:00:36 +10003197 }
Dan Williamsa4456852007-07-09 11:56:43 -07003198 /* now if nothing is locked, and if we have enough data,
3199 * we can start a write request
3200 */
Dan Williamsf38e1212007-01-02 13:52:30 -07003201 /* since handle_stripe can be called at any time we need to handle the
3202 * case where a compute block operation has been submitted and then a
Dan Williamsac6b53b2009-07-14 13:40:19 -07003203 * subsequent call wants to start a write request. raid_run_ops only
3204 * handles the case where compute block and reconstruct are requested
Dan Williamsf38e1212007-01-02 13:52:30 -07003205 * simultaneously. If this is not the case then new writes need to be
3206 * held off until the compute completes.
3207 */
Dan Williams976ea8d2008-06-28 08:32:03 +10003208 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3209 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
3210 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07003211 schedule_reconstruction(sh, s, rcw == 0, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07003212}
3213
NeilBrownd1688a62011-10-11 16:49:52 +11003214static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07003215 struct stripe_head_state *s, int disks)
3216{
Dan Williamsecc65c92008-06-28 08:31:57 +10003217 struct r5dev *dev = NULL;
Dan Williamse89f8962007-01-02 13:52:31 -07003218
Dan Williamsbd2ab672008-04-10 21:29:27 -07003219 set_bit(STRIPE_HANDLE, &sh->state);
3220
Dan Williamsecc65c92008-06-28 08:31:57 +10003221 switch (sh->check_state) {
3222 case check_state_idle:
3223 /* start a new check operation if there are no failures */
Dan Williamsbd2ab672008-04-10 21:29:27 -07003224 if (s->failed == 0) {
Dan Williamsbd2ab672008-04-10 21:29:27 -07003225 BUG_ON(s->uptodate != disks);
Dan Williamsecc65c92008-06-28 08:31:57 +10003226 sh->check_state = check_state_run;
3227 set_bit(STRIPE_OP_CHECK, &s->ops_request);
Dan Williamsbd2ab672008-04-10 21:29:27 -07003228 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
Dan Williamsbd2ab672008-04-10 21:29:27 -07003229 s->uptodate--;
Dan Williamsecc65c92008-06-28 08:31:57 +10003230 break;
Dan Williamsbd2ab672008-04-10 21:29:27 -07003231 }
NeilBrownf2b3b442011-07-26 11:35:19 +10003232 dev = &sh->dev[s->failed_num[0]];
Dan Williamsecc65c92008-06-28 08:31:57 +10003233 /* fall through */
3234 case check_state_compute_result:
3235 sh->check_state = check_state_idle;
3236 if (!dev)
3237 dev = &sh->dev[sh->pd_idx];
3238
3239 /* check that a write has not made the stripe insync */
3240 if (test_bit(STRIPE_INSYNC, &sh->state))
3241 break;
3242
3243 /* either failed parity check, or recovery is happening */
Dan Williamsa4456852007-07-09 11:56:43 -07003244 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3245 BUG_ON(s->uptodate != disks);
3246
3247 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsecc65c92008-06-28 08:31:57 +10003248 s->locked++;
Dan Williamsa4456852007-07-09 11:56:43 -07003249 set_bit(R5_Wantwrite, &dev->flags);
Dan Williams830ea012007-01-02 13:52:31 -07003250
Dan Williamsa4456852007-07-09 11:56:43 -07003251 clear_bit(STRIPE_DEGRADED, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003252 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10003253 break;
3254 case check_state_run:
3255 break; /* we will be called again upon completion */
3256 case check_state_check_result:
3257 sh->check_state = check_state_idle;
3258
3259 /* if a failure occurred during the check operation, leave
3260 * STRIPE_INSYNC not set and let the stripe be handled again
3261 */
3262 if (s->failed)
3263 break;
3264
3265 /* handle a successful check operation, if parity is correct
3266 * we are done. Otherwise update the mismatch count and repair
3267 * parity if !MD_RECOVERY_CHECK
3268 */
Dan Williamsad283ea2009-08-29 19:09:26 -07003269 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
Dan Williamsecc65c92008-06-28 08:31:57 +10003270 /* parity is correct (on disc,
3271 * not in buffer any more)
3272 */
3273 set_bit(STRIPE_INSYNC, &sh->state);
3274 else {
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11003275 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
Dan Williamsecc65c92008-06-28 08:31:57 +10003276 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3277 /* don't try to repair!! */
3278 set_bit(STRIPE_INSYNC, &sh->state);
3279 else {
3280 sh->check_state = check_state_compute_run;
Dan Williams976ea8d2008-06-28 08:32:03 +10003281 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10003282 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3283 set_bit(R5_Wantcompute,
3284 &sh->dev[sh->pd_idx].flags);
3285 sh->ops.target = sh->pd_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07003286 sh->ops.target2 = -1;
Dan Williamsecc65c92008-06-28 08:31:57 +10003287 s->uptodate++;
3288 }
3289 }
3290 break;
3291 case check_state_compute_run:
3292 break;
3293 default:
3294 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3295 __func__, sh->check_state,
3296 (unsigned long long) sh->sector);
3297 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07003298 }
3299}
3300
NeilBrownd1688a62011-10-11 16:49:52 +11003301static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
Dan Williams36d1c642009-07-14 11:48:22 -07003302 struct stripe_head_state *s,
NeilBrownf2b3b442011-07-26 11:35:19 +10003303 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07003304{
Dan Williamsa4456852007-07-09 11:56:43 -07003305 int pd_idx = sh->pd_idx;
NeilBrown34e04e82009-03-31 15:10:16 +11003306 int qd_idx = sh->qd_idx;
Dan Williamsd82dfee2009-07-14 13:40:57 -07003307 struct r5dev *dev;
Dan Williamsa4456852007-07-09 11:56:43 -07003308
3309 set_bit(STRIPE_HANDLE, &sh->state);
3310
3311 BUG_ON(s->failed > 2);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003312
Dan Williamsa4456852007-07-09 11:56:43 -07003313 /* Want to check and possibly repair P and Q.
3314 * However there could be one 'failed' device, in which
3315 * case we can only check one of them, possibly using the
3316 * other to generate missing data
3317 */
3318
Dan Williamsd82dfee2009-07-14 13:40:57 -07003319 switch (sh->check_state) {
3320 case check_state_idle:
3321 /* start a new check operation if there are < 2 failures */
NeilBrownf2b3b442011-07-26 11:35:19 +10003322 if (s->failed == s->q_failed) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07003323 /* The only possible failed device holds Q, so it
Dan Williamsa4456852007-07-09 11:56:43 -07003324 * makes sense to check P (If anything else were failed,
3325 * we would have used P to recreate it).
3326 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003327 sh->check_state = check_state_run;
Dan Williamsa4456852007-07-09 11:56:43 -07003328 }
NeilBrownf2b3b442011-07-26 11:35:19 +10003329 if (!s->q_failed && s->failed < 2) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07003330 /* Q is not failed, and we didn't use it to generate
Dan Williamsa4456852007-07-09 11:56:43 -07003331 * anything, so it makes sense to check it
3332 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003333 if (sh->check_state == check_state_run)
3334 sh->check_state = check_state_run_pq;
3335 else
3336 sh->check_state = check_state_run_q;
Dan Williamsa4456852007-07-09 11:56:43 -07003337 }
Dan Williams36d1c642009-07-14 11:48:22 -07003338
Dan Williamsd82dfee2009-07-14 13:40:57 -07003339 /* discard potentially stale zero_sum_result */
3340 sh->ops.zero_sum_result = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07003341
Dan Williamsd82dfee2009-07-14 13:40:57 -07003342 if (sh->check_state == check_state_run) {
3343 /* async_xor_zero_sum destroys the contents of P */
3344 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3345 s->uptodate--;
Dan Williamsa4456852007-07-09 11:56:43 -07003346 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003347 if (sh->check_state >= check_state_run &&
3348 sh->check_state <= check_state_run_pq) {
3349 /* async_syndrome_zero_sum preserves P and Q, so
3350 * no need to mark them !uptodate here
3351 */
3352 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3353 break;
3354 }
Dan Williams36d1c642009-07-14 11:48:22 -07003355
Dan Williamsd82dfee2009-07-14 13:40:57 -07003356 /* we have 2-disk failure */
3357 BUG_ON(s->failed != 2);
3358 /* fall through */
3359 case check_state_compute_result:
3360 sh->check_state = check_state_idle;
Dan Williams36d1c642009-07-14 11:48:22 -07003361
Dan Williamsd82dfee2009-07-14 13:40:57 -07003362 /* check that a write has not made the stripe insync */
3363 if (test_bit(STRIPE_INSYNC, &sh->state))
3364 break;
Dan Williamsa4456852007-07-09 11:56:43 -07003365
3366 /* now write out any block on a failed drive,
Dan Williamsd82dfee2009-07-14 13:40:57 -07003367 * or P or Q if they were recomputed
Dan Williamsa4456852007-07-09 11:56:43 -07003368 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003369 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
Dan Williamsa4456852007-07-09 11:56:43 -07003370 if (s->failed == 2) {
NeilBrownf2b3b442011-07-26 11:35:19 +10003371 dev = &sh->dev[s->failed_num[1]];
Dan Williamsa4456852007-07-09 11:56:43 -07003372 s->locked++;
3373 set_bit(R5_LOCKED, &dev->flags);
3374 set_bit(R5_Wantwrite, &dev->flags);
3375 }
3376 if (s->failed >= 1) {
NeilBrownf2b3b442011-07-26 11:35:19 +10003377 dev = &sh->dev[s->failed_num[0]];
Dan Williamsa4456852007-07-09 11:56:43 -07003378 s->locked++;
3379 set_bit(R5_LOCKED, &dev->flags);
3380 set_bit(R5_Wantwrite, &dev->flags);
3381 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003382 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07003383 dev = &sh->dev[pd_idx];
3384 s->locked++;
3385 set_bit(R5_LOCKED, &dev->flags);
3386 set_bit(R5_Wantwrite, &dev->flags);
3387 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003388 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07003389 dev = &sh->dev[qd_idx];
3390 s->locked++;
3391 set_bit(R5_LOCKED, &dev->flags);
3392 set_bit(R5_Wantwrite, &dev->flags);
3393 }
3394 clear_bit(STRIPE_DEGRADED, &sh->state);
3395
3396 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003397 break;
3398 case check_state_run:
3399 case check_state_run_q:
3400 case check_state_run_pq:
3401 break; /* we will be called again upon completion */
3402 case check_state_check_result:
3403 sh->check_state = check_state_idle;
3404
3405 /* handle a successful check operation, if parity is correct
3406 * we are done. Otherwise update the mismatch count and repair
3407 * parity if !MD_RECOVERY_CHECK
3408 */
3409 if (sh->ops.zero_sum_result == 0) {
3410 /* both parities are correct */
3411 if (!s->failed)
3412 set_bit(STRIPE_INSYNC, &sh->state);
3413 else {
3414 /* in contrast to the raid5 case we can validate
3415 * parity, but still have a failure to write
3416 * back
3417 */
3418 sh->check_state = check_state_compute_result;
3419 /* Returning at this point means that we may go
3420 * off and bring p and/or q uptodate again so
3421 * we make sure to check zero_sum_result again
3422 * to verify if p or q need writeback
3423 */
3424 }
3425 } else {
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11003426 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003427 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3428 /* don't try to repair!! */
3429 set_bit(STRIPE_INSYNC, &sh->state);
3430 else {
3431 int *target = &sh->ops.target;
3432
3433 sh->ops.target = -1;
3434 sh->ops.target2 = -1;
3435 sh->check_state = check_state_compute_run;
3436 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3437 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3438 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3439 set_bit(R5_Wantcompute,
3440 &sh->dev[pd_idx].flags);
3441 *target = pd_idx;
3442 target = &sh->ops.target2;
3443 s->uptodate++;
3444 }
3445 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3446 set_bit(R5_Wantcompute,
3447 &sh->dev[qd_idx].flags);
3448 *target = qd_idx;
3449 s->uptodate++;
3450 }
3451 }
3452 }
3453 break;
3454 case check_state_compute_run:
3455 break;
3456 default:
3457 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3458 __func__, sh->check_state,
3459 (unsigned long long) sh->sector);
3460 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07003461 }
3462}
3463
NeilBrownd1688a62011-10-11 16:49:52 +11003464static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
Dan Williamsa4456852007-07-09 11:56:43 -07003465{
3466 int i;
3467
3468 /* We have read all the blocks in this stripe and now we need to
3469 * copy some of them into a target stripe for expand.
3470 */
Dan Williamsf0a50d32007-01-02 13:52:31 -07003471 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07003472 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3473 for (i = 0; i < sh->disks; i++)
NeilBrown34e04e82009-03-31 15:10:16 +11003474 if (i != sh->pd_idx && i != sh->qd_idx) {
NeilBrown911d4ee2009-03-31 14:39:38 +11003475 int dd_idx, j;
Dan Williamsa4456852007-07-09 11:56:43 -07003476 struct stripe_head *sh2;
Dan Williamsa08abd82009-06-03 11:43:59 -07003477 struct async_submit_ctl submit;
Dan Williamsa4456852007-07-09 11:56:43 -07003478
NeilBrown784052e2009-03-31 15:19:07 +11003479 sector_t bn = compute_blocknr(sh, i, 1);
NeilBrown911d4ee2009-03-31 14:39:38 +11003480 sector_t s = raid5_compute_sector(conf, bn, 0,
3481 &dd_idx, NULL);
NeilBrowna8c906c2009-06-09 14:39:59 +10003482 sh2 = get_active_stripe(conf, s, 0, 1, 1);
Dan Williamsa4456852007-07-09 11:56:43 -07003483 if (sh2 == NULL)
3484 /* so far only the early blocks of this stripe
3485 * have been requested. When later blocks
3486 * get requested, we will try again
3487 */
3488 continue;
3489 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3490 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3491 /* must have already done this block */
3492 release_stripe(sh2);
3493 continue;
3494 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07003495
3496 /* place all the copies on one channel */
Dan Williamsa08abd82009-06-03 11:43:59 -07003497 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003498 tx = async_memcpy(sh2->dev[dd_idx].page,
Dan Williams88ba2aa2009-04-09 16:16:18 -07003499 sh->dev[i].page, 0, 0, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07003500 &submit);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003501
Dan Williamsa4456852007-07-09 11:56:43 -07003502 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3503 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3504 for (j = 0; j < conf->raid_disks; j++)
3505 if (j != sh2->pd_idx &&
NeilBrown86c374b2011-07-27 11:00:36 +10003506 j != sh2->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07003507 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3508 break;
3509 if (j == conf->raid_disks) {
3510 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3511 set_bit(STRIPE_HANDLE, &sh2->state);
3512 }
3513 release_stripe(sh2);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003514
Dan Williamsa4456852007-07-09 11:56:43 -07003515 }
NeilBrowna2e08552007-09-11 15:23:36 -07003516 /* done submitting copies, wait for them to complete */
NeilBrown749586b2012-11-20 14:11:15 +11003517 async_tx_quiesce(&tx);
Dan Williamsa4456852007-07-09 11:56:43 -07003518}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
3520/*
3521 * handle_stripe - do things to a stripe.
3522 *
NeilBrown9a3e1102011-12-23 10:17:53 +11003523 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3524 * state of various bits to see what needs to be done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 * Possible results:
NeilBrown9a3e1102011-12-23 10:17:53 +11003526 * return some read requests which now have data
3527 * return some write requests which are safely on storage
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528 * schedule a read on some buffers
3529 * schedule a write of some buffers
3530 * return confirmation of parity correctness
3531 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 */
Dan Williamsa4456852007-07-09 11:56:43 -07003533
NeilBrownacfe7262011-07-27 11:00:36 +10003534static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
NeilBrown16a53ec2006-06-26 00:27:38 -07003535{
NeilBrownd1688a62011-10-11 16:49:52 +11003536 struct r5conf *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08003537 int disks = sh->disks;
NeilBrown474af965fe2011-07-27 11:00:36 +10003538 struct r5dev *dev;
3539 int i;
NeilBrown9a3e1102011-12-23 10:17:53 +11003540 int do_recovery = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003541
NeilBrownacfe7262011-07-27 11:00:36 +10003542 memset(s, 0, sizeof(*s));
NeilBrown16a53ec2006-06-26 00:27:38 -07003543
NeilBrownacfe7262011-07-27 11:00:36 +10003544 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3545 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3546 s->failed_num[0] = -1;
3547 s->failed_num[1] = -1;
3548
3549 /* Now to look around and see what can be done */
NeilBrown16a53ec2006-06-26 00:27:38 -07003550 rcu_read_lock();
3551 for (i=disks; i--; ) {
NeilBrown3cb03002011-10-11 16:45:26 +11003552 struct md_rdev *rdev;
NeilBrown31c176e2011-07-28 11:39:22 +10003553 sector_t first_bad;
3554 int bad_sectors;
3555 int is_bad = 0;
NeilBrownacfe7262011-07-27 11:00:36 +10003556
NeilBrown16a53ec2006-06-26 00:27:38 -07003557 dev = &sh->dev[i];
NeilBrown16a53ec2006-06-26 00:27:38 -07003558
Dan Williams45b42332007-07-09 11:56:43 -07003559 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown9a3e1102011-12-23 10:17:53 +11003560 i, dev->flags,
3561 dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003562 /* maybe we can reply to a read
3563 *
3564 * new wantfill requests are only permitted while
3565 * ops_complete_biofill is guaranteed to be inactive
3566 */
3567 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3568 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3569 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07003570
3571 /* now count some things */
NeilBrowncc940152011-07-26 11:35:35 +10003572 if (test_bit(R5_LOCKED, &dev->flags))
3573 s->locked++;
3574 if (test_bit(R5_UPTODATE, &dev->flags))
3575 s->uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003576 if (test_bit(R5_Wantcompute, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003577 s->compute++;
3578 BUG_ON(s->compute > 2);
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003579 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003580
NeilBrownacfe7262011-07-27 11:00:36 +10003581 if (test_bit(R5_Wantfill, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003582 s->to_fill++;
NeilBrownacfe7262011-07-27 11:00:36 +10003583 else if (dev->toread)
NeilBrowncc940152011-07-26 11:35:35 +10003584 s->to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003585 if (dev->towrite) {
NeilBrowncc940152011-07-26 11:35:35 +10003586 s->to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003587 if (!test_bit(R5_OVERWRITE, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003588 s->non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003589 }
Dan Williamsa4456852007-07-09 11:56:43 -07003590 if (dev->written)
NeilBrowncc940152011-07-26 11:35:35 +10003591 s->written++;
NeilBrown14a75d32011-12-23 10:17:52 +11003592 /* Prefer to use the replacement for reads, but only
3593 * if it is recovered enough and has no bad blocks.
3594 */
3595 rdev = rcu_dereference(conf->disks[i].replacement);
3596 if (rdev && !test_bit(Faulty, &rdev->flags) &&
3597 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
3598 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3599 &first_bad, &bad_sectors))
3600 set_bit(R5_ReadRepl, &dev->flags);
3601 else {
NeilBrown9a3e1102011-12-23 10:17:53 +11003602 if (rdev)
3603 set_bit(R5_NeedReplace, &dev->flags);
NeilBrown14a75d32011-12-23 10:17:52 +11003604 rdev = rcu_dereference(conf->disks[i].rdev);
3605 clear_bit(R5_ReadRepl, &dev->flags);
3606 }
NeilBrown9283d8c2011-12-08 16:27:57 +11003607 if (rdev && test_bit(Faulty, &rdev->flags))
3608 rdev = NULL;
NeilBrown31c176e2011-07-28 11:39:22 +10003609 if (rdev) {
3610 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
3611 &first_bad, &bad_sectors);
3612 if (s->blocked_rdev == NULL
3613 && (test_bit(Blocked, &rdev->flags)
3614 || is_bad < 0)) {
3615 if (is_bad < 0)
3616 set_bit(BlockedBadBlocks,
3617 &rdev->flags);
3618 s->blocked_rdev = rdev;
3619 atomic_inc(&rdev->nr_pending);
3620 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07003621 }
NeilBrown415e72d2010-06-17 17:25:21 +10003622 clear_bit(R5_Insync, &dev->flags);
3623 if (!rdev)
3624 /* Not in-sync */;
NeilBrown31c176e2011-07-28 11:39:22 +10003625 else if (is_bad) {
3626 /* also not in-sync */
NeilBrown18b98372012-04-01 23:48:38 +10003627 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
3628 test_bit(R5_UPTODATE, &dev->flags)) {
NeilBrown31c176e2011-07-28 11:39:22 +10003629 /* treat as in-sync, but with a read error
3630 * which we can now try to correct
3631 */
3632 set_bit(R5_Insync, &dev->flags);
3633 set_bit(R5_ReadError, &dev->flags);
3634 }
3635 } else if (test_bit(In_sync, &rdev->flags))
NeilBrown415e72d2010-06-17 17:25:21 +10003636 set_bit(R5_Insync, &dev->flags);
NeilBrown30d7a482011-12-23 09:57:00 +11003637 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
NeilBrown415e72d2010-06-17 17:25:21 +10003638 /* in sync if before recovery_offset */
NeilBrown30d7a482011-12-23 09:57:00 +11003639 set_bit(R5_Insync, &dev->flags);
3640 else if (test_bit(R5_UPTODATE, &dev->flags) &&
3641 test_bit(R5_Expanded, &dev->flags))
3642 /* If we've reshaped into here, we assume it is Insync.
3643 * We will shortly update recovery_offset to make
3644 * it official.
3645 */
3646 set_bit(R5_Insync, &dev->flags);
3647
NeilBrown1cc03eb2014-01-06 13:19:42 +11003648 if (test_bit(R5_WriteError, &dev->flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11003649 /* This flag does not apply to '.replacement'
3650 * only to .rdev, so make sure to check that*/
3651 struct md_rdev *rdev2 = rcu_dereference(
3652 conf->disks[i].rdev);
3653 if (rdev2 == rdev)
3654 clear_bit(R5_Insync, &dev->flags);
3655 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
NeilBrownbc2607f2011-07-28 11:39:22 +10003656 s->handle_bad_blocks = 1;
NeilBrown14a75d32011-12-23 10:17:52 +11003657 atomic_inc(&rdev2->nr_pending);
NeilBrownbc2607f2011-07-28 11:39:22 +10003658 } else
3659 clear_bit(R5_WriteError, &dev->flags);
3660 }
NeilBrown1cc03eb2014-01-06 13:19:42 +11003661 if (test_bit(R5_MadeGood, &dev->flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11003662 /* This flag does not apply to '.replacement'
3663 * only to .rdev, so make sure to check that*/
3664 struct md_rdev *rdev2 = rcu_dereference(
3665 conf->disks[i].rdev);
3666 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
NeilBrownb84db562011-07-28 11:39:23 +10003667 s->handle_bad_blocks = 1;
NeilBrown14a75d32011-12-23 10:17:52 +11003668 atomic_inc(&rdev2->nr_pending);
NeilBrownb84db562011-07-28 11:39:23 +10003669 } else
3670 clear_bit(R5_MadeGood, &dev->flags);
3671 }
NeilBrown977df362011-12-23 10:17:53 +11003672 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
3673 struct md_rdev *rdev2 = rcu_dereference(
3674 conf->disks[i].replacement);
3675 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
3676 s->handle_bad_blocks = 1;
3677 atomic_inc(&rdev2->nr_pending);
3678 } else
3679 clear_bit(R5_MadeGoodRepl, &dev->flags);
3680 }
NeilBrown415e72d2010-06-17 17:25:21 +10003681 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003682 /* The ReadError flag will just be confusing now */
3683 clear_bit(R5_ReadError, &dev->flags);
3684 clear_bit(R5_ReWrite, &dev->flags);
3685 }
NeilBrown415e72d2010-06-17 17:25:21 +10003686 if (test_bit(R5_ReadError, &dev->flags))
3687 clear_bit(R5_Insync, &dev->flags);
3688 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003689 if (s->failed < 2)
3690 s->failed_num[s->failed] = i;
3691 s->failed++;
NeilBrown9a3e1102011-12-23 10:17:53 +11003692 if (rdev && !test_bit(Faulty, &rdev->flags))
3693 do_recovery = 1;
NeilBrown415e72d2010-06-17 17:25:21 +10003694 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003695 }
NeilBrown9a3e1102011-12-23 10:17:53 +11003696 if (test_bit(STRIPE_SYNCING, &sh->state)) {
3697 /* If there is a failed device being replaced,
3698 * we must be recovering.
3699 * else if we are after recovery_cp, we must be syncing
majianpengc6d2e082012-04-02 01:16:59 +10003700 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
NeilBrown9a3e1102011-12-23 10:17:53 +11003701 * else we can only be replacing
3702 * sync and recovery both need to read all devices, and so
3703 * use the same flag.
3704 */
3705 if (do_recovery ||
majianpengc6d2e082012-04-02 01:16:59 +10003706 sh->sector >= conf->mddev->recovery_cp ||
3707 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
NeilBrown9a3e1102011-12-23 10:17:53 +11003708 s->syncing = 1;
3709 else
3710 s->replacing = 1;
3711 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003712 rcu_read_unlock();
NeilBrowncc940152011-07-26 11:35:35 +10003713}
NeilBrownf4168852007-02-28 20:11:53 -08003714
NeilBrowncc940152011-07-26 11:35:35 +10003715static void handle_stripe(struct stripe_head *sh)
3716{
3717 struct stripe_head_state s;
NeilBrownd1688a62011-10-11 16:49:52 +11003718 struct r5conf *conf = sh->raid_conf;
NeilBrown3687c062011-07-27 11:00:36 +10003719 int i;
NeilBrown84789552011-07-27 11:00:36 +10003720 int prexor;
3721 int disks = sh->disks;
NeilBrown474af965fe2011-07-27 11:00:36 +10003722 struct r5dev *pdev, *qdev;
NeilBrowncc940152011-07-26 11:35:35 +10003723
3724 clear_bit(STRIPE_HANDLE, &sh->state);
Dan Williams257a4b42011-11-08 16:22:06 +11003725 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
NeilBrowncc940152011-07-26 11:35:35 +10003726 /* already being handled, ensure it gets handled
3727 * again when current action finishes */
3728 set_bit(STRIPE_HANDLE, &sh->state);
3729 return;
3730 }
3731
NeilBrownf8dfcff2013-03-12 12:18:06 +11003732 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3733 spin_lock(&sh->stripe_lock);
3734 /* Cannot process 'sync' concurrently with 'discard' */
3735 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
3736 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3737 set_bit(STRIPE_SYNCING, &sh->state);
3738 clear_bit(STRIPE_INSYNC, &sh->state);
NeilBrownf94c0b62013-07-22 12:57:21 +10003739 clear_bit(STRIPE_REPLACED, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003740 }
3741 spin_unlock(&sh->stripe_lock);
NeilBrowncc940152011-07-26 11:35:35 +10003742 }
3743 clear_bit(STRIPE_DELAYED, &sh->state);
3744
3745 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3746 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3747 (unsigned long long)sh->sector, sh->state,
3748 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3749 sh->check_state, sh->reconstruct_state);
NeilBrowncc940152011-07-26 11:35:35 +10003750
NeilBrownacfe7262011-07-27 11:00:36 +10003751 analyse_stripe(sh, &s);
NeilBrownc5a31002011-07-27 11:00:36 +10003752
NeilBrownbc2607f2011-07-28 11:39:22 +10003753 if (s.handle_bad_blocks) {
3754 set_bit(STRIPE_HANDLE, &sh->state);
3755 goto finish;
3756 }
3757
NeilBrown474af965fe2011-07-27 11:00:36 +10003758 if (unlikely(s.blocked_rdev)) {
3759 if (s.syncing || s.expanding || s.expanded ||
NeilBrown9a3e1102011-12-23 10:17:53 +11003760 s.replacing || s.to_write || s.written) {
NeilBrown474af965fe2011-07-27 11:00:36 +10003761 set_bit(STRIPE_HANDLE, &sh->state);
3762 goto finish;
3763 }
3764 /* There is nothing for the blocked_rdev to block */
3765 rdev_dec_pending(s.blocked_rdev, conf->mddev);
3766 s.blocked_rdev = NULL;
3767 }
3768
3769 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3770 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3771 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3772 }
3773
3774 pr_debug("locked=%d uptodate=%d to_read=%d"
3775 " to_write=%d failed=%d failed_num=%d,%d\n",
3776 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3777 s.failed_num[0], s.failed_num[1]);
3778 /* check if the array has lost more than max_degraded devices and,
3779 * if so, some requests might need to be failed.
3780 */
NeilBrown9a3f5302011-11-08 16:22:01 +11003781 if (s.failed > conf->max_degraded) {
3782 sh->check_state = 0;
3783 sh->reconstruct_state = 0;
3784 if (s.to_read+s.to_write+s.written)
3785 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
NeilBrown9a3e1102011-12-23 10:17:53 +11003786 if (s.syncing + s.replacing)
NeilBrown9a3f5302011-11-08 16:22:01 +11003787 handle_failed_sync(conf, sh, &s);
3788 }
NeilBrown474af965fe2011-07-27 11:00:36 +10003789
NeilBrown84789552011-07-27 11:00:36 +10003790 /* Now we check to see if any write operations have recently
3791 * completed
3792 */
3793 prexor = 0;
3794 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
3795 prexor = 1;
3796 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3797 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
3798 sh->reconstruct_state = reconstruct_state_idle;
3799
3800 /* All the 'written' buffers and the parity block are ready to
3801 * be written back to disk
3802 */
Shaohua Li9e4447682012-10-11 13:49:49 +11003803 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
3804 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
NeilBrown84789552011-07-27 11:00:36 +10003805 BUG_ON(sh->qd_idx >= 0 &&
Shaohua Li9e4447682012-10-11 13:49:49 +11003806 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
3807 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
NeilBrown84789552011-07-27 11:00:36 +10003808 for (i = disks; i--; ) {
3809 struct r5dev *dev = &sh->dev[i];
3810 if (test_bit(R5_LOCKED, &dev->flags) &&
3811 (i == sh->pd_idx || i == sh->qd_idx ||
3812 dev->written)) {
3813 pr_debug("Writing block %d\n", i);
3814 set_bit(R5_Wantwrite, &dev->flags);
3815 if (prexor)
3816 continue;
NeilBrown9c4bdf62014-08-13 09:57:07 +10003817 if (s.failed > 1)
3818 continue;
NeilBrown84789552011-07-27 11:00:36 +10003819 if (!test_bit(R5_Insync, &dev->flags) ||
3820 ((i == sh->pd_idx || i == sh->qd_idx) &&
3821 s.failed == 0))
3822 set_bit(STRIPE_INSYNC, &sh->state);
3823 }
3824 }
3825 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3826 s.dec_preread_active = 1;
3827 }
3828
NeilBrownef5b7c62012-11-22 09:13:36 +11003829 /*
3830 * might be able to return some write requests if the parity blocks
3831 * are safe, or on a failed drive
3832 */
3833 pdev = &sh->dev[sh->pd_idx];
3834 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
3835 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
3836 qdev = &sh->dev[sh->qd_idx];
3837 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
3838 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
3839 || conf->level < 6;
3840
3841 if (s.written &&
3842 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
3843 && !test_bit(R5_LOCKED, &pdev->flags)
3844 && (test_bit(R5_UPTODATE, &pdev->flags) ||
3845 test_bit(R5_Discard, &pdev->flags))))) &&
3846 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
3847 && !test_bit(R5_LOCKED, &qdev->flags)
3848 && (test_bit(R5_UPTODATE, &qdev->flags) ||
3849 test_bit(R5_Discard, &qdev->flags))))))
3850 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
3851
3852 /* Now we might consider reading some blocks, either to check/generate
3853 * parity, or to satisfy requests
3854 * or to load a block that is being partially written.
3855 */
3856 if (s.to_read || s.non_overwrite
3857 || (conf->level == 6 && s.to_write && s.failed)
3858 || (s.syncing && (s.uptodate + s.compute < disks))
3859 || s.replacing
3860 || s.expanding)
3861 handle_stripe_fill(sh, &s, disks);
3862
NeilBrown84789552011-07-27 11:00:36 +10003863 /* Now to consider new write requests and what else, if anything
3864 * should be read. We do not handle new writes when:
3865 * 1/ A 'write' operation (copy+xor) is already in flight.
3866 * 2/ A 'check' operation is in flight, as it may clobber the parity
3867 * block.
3868 */
3869 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
3870 handle_stripe_dirtying(conf, sh, &s, disks);
3871
3872 /* maybe we need to check and possibly fix the parity for this stripe
3873 * Any reads will already have been scheduled, so we just see if enough
3874 * data is available. The parity check is held off while parity
3875 * dependent operations are in flight.
3876 */
3877 if (sh->check_state ||
3878 (s.syncing && s.locked == 0 &&
3879 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3880 !test_bit(STRIPE_INSYNC, &sh->state))) {
3881 if (conf->level == 6)
3882 handle_parity_checks6(conf, sh, &s, disks);
3883 else
3884 handle_parity_checks5(conf, sh, &s, disks);
3885 }
NeilBrownc5a31002011-07-27 11:00:36 +10003886
NeilBrownf94c0b62013-07-22 12:57:21 +10003887 if ((s.replacing || s.syncing) && s.locked == 0
3888 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
3889 && !test_bit(STRIPE_REPLACED, &sh->state)) {
NeilBrown9a3e1102011-12-23 10:17:53 +11003890 /* Write out to replacement devices where possible */
3891 for (i = 0; i < conf->raid_disks; i++)
NeilBrownf94c0b62013-07-22 12:57:21 +10003892 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
3893 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
NeilBrown9a3e1102011-12-23 10:17:53 +11003894 set_bit(R5_WantReplace, &sh->dev[i].flags);
3895 set_bit(R5_LOCKED, &sh->dev[i].flags);
3896 s.locked++;
3897 }
NeilBrownf94c0b62013-07-22 12:57:21 +10003898 if (s.replacing)
3899 set_bit(STRIPE_INSYNC, &sh->state);
3900 set_bit(STRIPE_REPLACED, &sh->state);
NeilBrown9a3e1102011-12-23 10:17:53 +11003901 }
3902 if ((s.syncing || s.replacing) && s.locked == 0 &&
NeilBrownf94c0b62013-07-22 12:57:21 +10003903 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
NeilBrown9a3e1102011-12-23 10:17:53 +11003904 test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrownc5a31002011-07-27 11:00:36 +10003905 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3906 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003907 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3908 wake_up(&conf->wait_for_overlap);
NeilBrownc5a31002011-07-27 11:00:36 +10003909 }
3910
3911 /* If the failed drives are just a ReadError, then we might need
3912 * to progress the repair/check process
3913 */
3914 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3915 for (i = 0; i < s.failed; i++) {
3916 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3917 if (test_bit(R5_ReadError, &dev->flags)
3918 && !test_bit(R5_LOCKED, &dev->flags)
3919 && test_bit(R5_UPTODATE, &dev->flags)
3920 ) {
3921 if (!test_bit(R5_ReWrite, &dev->flags)) {
3922 set_bit(R5_Wantwrite, &dev->flags);
3923 set_bit(R5_ReWrite, &dev->flags);
3924 set_bit(R5_LOCKED, &dev->flags);
3925 s.locked++;
3926 } else {
3927 /* let's read it back */
3928 set_bit(R5_Wantread, &dev->flags);
3929 set_bit(R5_LOCKED, &dev->flags);
3930 s.locked++;
3931 }
3932 }
3933 }
3934
NeilBrown3687c062011-07-27 11:00:36 +10003935 /* Finish reconstruct operations initiated by the expansion process */
3936 if (sh->reconstruct_state == reconstruct_state_result) {
3937 struct stripe_head *sh_src
3938 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3939 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3940 /* sh cannot be written until sh_src has been read.
3941 * so arrange for sh to be delayed a little
3942 */
3943 set_bit(STRIPE_DELAYED, &sh->state);
3944 set_bit(STRIPE_HANDLE, &sh->state);
3945 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3946 &sh_src->state))
3947 atomic_inc(&conf->preread_active_stripes);
3948 release_stripe(sh_src);
3949 goto finish;
3950 }
3951 if (sh_src)
3952 release_stripe(sh_src);
NeilBrown16a53ec2006-06-26 00:27:38 -07003953
NeilBrown3687c062011-07-27 11:00:36 +10003954 sh->reconstruct_state = reconstruct_state_idle;
3955 clear_bit(STRIPE_EXPANDING, &sh->state);
3956 for (i = conf->raid_disks; i--; ) {
3957 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3958 set_bit(R5_LOCKED, &sh->dev[i].flags);
3959 s.locked++;
3960 }
3961 }
3962
3963 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3964 !sh->reconstruct_state) {
3965 /* Need to write out all blocks after computing parity */
3966 sh->disks = conf->raid_disks;
3967 stripe_set_idx(sh->sector, conf, 0, sh);
3968 schedule_reconstruction(sh, &s, 1, 1);
3969 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3970 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3971 atomic_dec(&conf->reshape_stripes);
3972 wake_up(&conf->wait_for_overlap);
3973 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3974 }
3975
3976 if (s.expanding && s.locked == 0 &&
3977 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3978 handle_stripe_expansion(conf, sh);
3979
3980finish:
Dan Williams6bfe0b42008-04-30 00:52:32 -07003981 /* wait for this device to become unblocked */
NeilBrown5f066c62012-07-03 12:13:29 +10003982 if (unlikely(s.blocked_rdev)) {
3983 if (conf->mddev->external)
3984 md_wait_for_blocked_rdev(s.blocked_rdev,
3985 conf->mddev);
3986 else
3987 /* Internal metadata will immediately
3988 * be written by raid5d, so we don't
3989 * need to wait here.
3990 */
3991 rdev_dec_pending(s.blocked_rdev,
3992 conf->mddev);
3993 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07003994
NeilBrownbc2607f2011-07-28 11:39:22 +10003995 if (s.handle_bad_blocks)
3996 for (i = disks; i--; ) {
NeilBrown3cb03002011-10-11 16:45:26 +11003997 struct md_rdev *rdev;
NeilBrownbc2607f2011-07-28 11:39:22 +10003998 struct r5dev *dev = &sh->dev[i];
3999 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4000 /* We own a safe reference to the rdev */
4001 rdev = conf->disks[i].rdev;
4002 if (!rdev_set_badblocks(rdev, sh->sector,
4003 STRIPE_SECTORS, 0))
4004 md_error(conf->mddev, rdev);
4005 rdev_dec_pending(rdev, conf->mddev);
4006 }
NeilBrownb84db562011-07-28 11:39:23 +10004007 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4008 rdev = conf->disks[i].rdev;
4009 rdev_clear_badblocks(rdev, sh->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10004010 STRIPE_SECTORS, 0);
NeilBrownb84db562011-07-28 11:39:23 +10004011 rdev_dec_pending(rdev, conf->mddev);
4012 }
NeilBrown977df362011-12-23 10:17:53 +11004013 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
4014 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11004015 if (!rdev)
4016 /* rdev have been moved down */
4017 rdev = conf->disks[i].rdev;
NeilBrown977df362011-12-23 10:17:53 +11004018 rdev_clear_badblocks(rdev, sh->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10004019 STRIPE_SECTORS, 0);
NeilBrown977df362011-12-23 10:17:53 +11004020 rdev_dec_pending(rdev, conf->mddev);
4021 }
NeilBrownbc2607f2011-07-28 11:39:22 +10004022 }
4023
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07004024 if (s.ops_request)
4025 raid_run_ops(sh, s.ops_request);
4026
Dan Williamsf0e43bc2008-06-28 08:31:55 +10004027 ops_run_io(sh, &s);
4028
NeilBrownc5709ef2011-07-26 11:35:20 +10004029 if (s.dec_preread_active) {
NeilBrown729a1862009-12-14 12:49:50 +11004030 /* We delay this until after ops_run_io so that if make_request
Tejun Heoe9c74692010-09-03 11:56:18 +02004031 * is waiting on a flush, it won't continue until the writes
NeilBrown729a1862009-12-14 12:49:50 +11004032 * have actually been submitted.
4033 */
4034 atomic_dec(&conf->preread_active_stripes);
4035 if (atomic_read(&conf->preread_active_stripes) <
4036 IO_THRESHOLD)
4037 md_wakeup_thread(conf->mddev->thread);
4038 }
4039
NeilBrownc5709ef2011-07-26 11:35:20 +10004040 return_io(s.return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07004041
Dan Williams257a4b42011-11-08 16:22:06 +11004042 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07004043}
4044
NeilBrownd1688a62011-10-11 16:49:52 +11004045static void raid5_activate_delayed(struct r5conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046{
4047 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4048 while (!list_empty(&conf->delayed_list)) {
4049 struct list_head *l = conf->delayed_list.next;
4050 struct stripe_head *sh;
4051 sh = list_entry(l, struct stripe_head, lru);
4052 list_del_init(l);
4053 clear_bit(STRIPE_DELAYED, &sh->state);
4054 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4055 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004056 list_add_tail(&sh->lru, &conf->hold_list);
Shaohua Li851c30c2013-08-28 14:30:16 +08004057 raid5_wakeup_stripe_thread(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058 }
NeilBrown482c0832011-04-18 18:25:42 +10004059 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060}
4061
Shaohua Li566c09c2013-11-14 15:16:17 +11004062static void activate_bit_delay(struct r5conf *conf,
4063 struct list_head *temp_inactive_list)
NeilBrown72626682005-09-09 16:23:54 -07004064{
4065 /* device_lock is held */
4066 struct list_head head;
4067 list_add(&head, &conf->bitmap_list);
4068 list_del_init(&conf->bitmap_list);
4069 while (!list_empty(&head)) {
4070 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
Shaohua Li566c09c2013-11-14 15:16:17 +11004071 int hash;
NeilBrown72626682005-09-09 16:23:54 -07004072 list_del_init(&sh->lru);
4073 atomic_inc(&sh->count);
Shaohua Li566c09c2013-11-14 15:16:17 +11004074 hash = sh->hash_lock_index;
4075 __release_stripe(conf, sh, &temp_inactive_list[hash]);
NeilBrown72626682005-09-09 16:23:54 -07004076 }
4077}
4078
NeilBrownfd01b882011-10-11 16:47:53 +11004079int md_raid5_congested(struct mddev *mddev, int bits)
NeilBrownf022b2f2006-10-03 01:15:56 -07004080{
NeilBrownd1688a62011-10-11 16:49:52 +11004081 struct r5conf *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07004082
4083 /* No difference between reads and writes. Just check
4084 * how busy the stripe_cache is
4085 */
NeilBrown3fa841d2009-09-23 18:10:29 +10004086
NeilBrownf022b2f2006-10-03 01:15:56 -07004087 if (conf->inactive_blocked)
4088 return 1;
4089 if (conf->quiesce)
4090 return 1;
Shaohua Li4bda5562013-11-14 15:16:17 +11004091 if (atomic_read(&conf->empty_inactive_list_nr))
NeilBrownf022b2f2006-10-03 01:15:56 -07004092 return 1;
4093
4094 return 0;
4095}
NeilBrown11d8a6e2010-07-26 11:57:07 +10004096EXPORT_SYMBOL_GPL(md_raid5_congested);
4097
4098static int raid5_congested(void *data, int bits)
4099{
NeilBrownfd01b882011-10-11 16:47:53 +11004100 struct mddev *mddev = data;
NeilBrown11d8a6e2010-07-26 11:57:07 +10004101
4102 return mddev_congested(mddev, bits) ||
4103 md_raid5_congested(mddev, bits);
4104}
NeilBrownf022b2f2006-10-03 01:15:56 -07004105
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004106/* We want read requests to align with chunks where possible,
4107 * but write requests don't need to.
4108 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004109static int raid5_mergeable_bvec(struct request_queue *q,
4110 struct bvec_merge_data *bvm,
4111 struct bio_vec *biovec)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004112{
NeilBrownfd01b882011-10-11 16:47:53 +11004113 struct mddev *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004114 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004115 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +10004116 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004117 unsigned int bio_sectors = bvm->bi_size >> 9;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004118
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02004119 if ((bvm->bi_rw & 1) == WRITE)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004120 return biovec->bv_len; /* always allow writes to be mergeable */
4121
Andre Noll664e7c42009-06-18 08:45:27 +10004122 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4123 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08004124 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
4125 if (max < 0) max = 0;
4126 if (max <= biovec->bv_len && bio_sectors == 0)
4127 return biovec->bv_len;
4128 else
4129 return max;
4130}
4131
NeilBrownfd01b882011-10-11 16:47:53 +11004132static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004133{
Kent Overstreet4f024f32013-10-11 15:44:27 -07004134 sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev);
Andre Noll9d8f0362009-06-18 08:45:01 +10004135 unsigned int chunk_sectors = mddev->chunk_sectors;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08004136 unsigned int bio_sectors = bio_sectors(bio);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004137
Andre Noll664e7c42009-06-18 08:45:27 +10004138 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
4139 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004140 return chunk_sectors >=
4141 ((sector & (chunk_sectors - 1)) + bio_sectors);
4142}
4143
4144/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004145 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4146 * later sampled by raid5d.
4147 */
NeilBrownd1688a62011-10-11 16:49:52 +11004148static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004149{
4150 unsigned long flags;
4151
4152 spin_lock_irqsave(&conf->device_lock, flags);
4153
4154 bi->bi_next = conf->retry_read_aligned_list;
4155 conf->retry_read_aligned_list = bi;
4156
4157 spin_unlock_irqrestore(&conf->device_lock, flags);
4158 md_wakeup_thread(conf->mddev->thread);
4159}
4160
NeilBrownd1688a62011-10-11 16:49:52 +11004161static struct bio *remove_bio_from_retry(struct r5conf *conf)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004162{
4163 struct bio *bi;
4164
4165 bi = conf->retry_read_aligned;
4166 if (bi) {
4167 conf->retry_read_aligned = NULL;
4168 return bi;
4169 }
4170 bi = conf->retry_read_aligned_list;
4171 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08004172 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004173 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02004174 /*
4175 * this sets the active strip count to 1 and the processed
4176 * strip count to zero (upper 8 bits)
4177 */
Shaohua Lie7836bd62012-07-19 16:01:31 +10004178 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004179 }
4180
4181 return bi;
4182}
4183
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004184/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004185 * The "raid5_align_endio" should check if the read succeeded and if it
4186 * did, call bio_endio on the original bio (having bio_put the new bio
4187 * first).
4188 * If the read failed..
4189 */
NeilBrown6712ecf2007-09-27 12:47:43 +02004190static void raid5_align_endio(struct bio *bi, int error)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004191{
4192 struct bio* raid_bi = bi->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11004193 struct mddev *mddev;
NeilBrownd1688a62011-10-11 16:49:52 +11004194 struct r5conf *conf;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004195 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrown3cb03002011-10-11 16:45:26 +11004196 struct md_rdev *rdev;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004197
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004198 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004199
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004200 rdev = (void*)raid_bi->bi_next;
4201 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11004202 mddev = rdev->mddev;
4203 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004204
4205 rdev_dec_pending(rdev, conf->mddev);
4206
4207 if (!error && uptodate) {
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07004208 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4209 raid_bi, 0);
NeilBrown6712ecf2007-09-27 12:47:43 +02004210 bio_endio(raid_bi, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004211 if (atomic_dec_and_test(&conf->active_aligned_reads))
4212 wake_up(&conf->wait_for_stripe);
NeilBrown6712ecf2007-09-27 12:47:43 +02004213 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004214 }
4215
Dan Williams45b42332007-07-09 11:56:43 -07004216 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004217
4218 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004219}
4220
Neil Brown387bb172007-02-08 14:20:29 -08004221static int bio_fits_rdev(struct bio *bi)
4222{
Jens Axboe165125e2007-07-24 09:28:11 +02004223 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
Neil Brown387bb172007-02-08 14:20:29 -08004224
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08004225 if (bio_sectors(bi) > queue_max_sectors(q))
Neil Brown387bb172007-02-08 14:20:29 -08004226 return 0;
4227 blk_recount_segments(q, bi);
Martin K. Petersen8a783622010-02-26 00:20:39 -05004228 if (bi->bi_phys_segments > queue_max_segments(q))
Neil Brown387bb172007-02-08 14:20:29 -08004229 return 0;
4230
4231 if (q->merge_bvec_fn)
4232 /* it's too hard to apply the merge_bvec_fn at this stage,
4233 * just just give up
4234 */
4235 return 0;
4236
4237 return 1;
4238}
4239
NeilBrownfd01b882011-10-11 16:47:53 +11004240static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004241{
NeilBrownd1688a62011-10-11 16:49:52 +11004242 struct r5conf *conf = mddev->private;
NeilBrown8553fe7ec2009-12-14 12:49:47 +11004243 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004244 struct bio* align_bi;
NeilBrown3cb03002011-10-11 16:45:26 +11004245 struct md_rdev *rdev;
NeilBrown671488c2011-12-23 10:17:52 +11004246 sector_t end_sector;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004247
4248 if (!in_chunk_boundary(mddev, raid_bio)) {
Dan Williams45b42332007-07-09 11:56:43 -07004249 pr_debug("chunk_aligned_read : non aligned\n");
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004250 return 0;
4251 }
4252 /*
NeilBrowna167f662010-10-26 18:31:13 +11004253 * use bio_clone_mddev to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004254 */
NeilBrowna167f662010-10-26 18:31:13 +11004255 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004256 if (!align_bi)
4257 return 0;
4258 /*
4259 * set bi_end_io to a new function, and set bi_private to the
4260 * original bio.
4261 */
4262 align_bi->bi_end_io = raid5_align_endio;
4263 align_bi->bi_private = raid_bio;
4264 /*
4265 * compute position
4266 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07004267 align_bi->bi_iter.bi_sector =
4268 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
4269 0, &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004270
Kent Overstreetf73a1c72012-09-25 15:05:12 -07004271 end_sector = bio_end_sector(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004272 rcu_read_lock();
NeilBrown671488c2011-12-23 10:17:52 +11004273 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4274 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4275 rdev->recovery_offset < end_sector) {
4276 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4277 if (rdev &&
4278 (test_bit(Faulty, &rdev->flags) ||
4279 !(test_bit(In_sync, &rdev->flags) ||
4280 rdev->recovery_offset >= end_sector)))
4281 rdev = NULL;
4282 }
4283 if (rdev) {
NeilBrown31c176e2011-07-28 11:39:22 +10004284 sector_t first_bad;
4285 int bad_sectors;
4286
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004287 atomic_inc(&rdev->nr_pending);
4288 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004289 raid_bio->bi_next = (void*)rdev;
4290 align_bi->bi_bdev = rdev->bdev;
NeilBrown3fd83712014-08-23 20:19:26 +10004291 __clear_bit(BIO_SEG_VALID, &align_bi->bi_flags);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004292
NeilBrown31c176e2011-07-28 11:39:22 +10004293 if (!bio_fits_rdev(align_bi) ||
Kent Overstreet4f024f32013-10-11 15:44:27 -07004294 is_badblock(rdev, align_bi->bi_iter.bi_sector,
4295 bio_sectors(align_bi),
NeilBrown31c176e2011-07-28 11:39:22 +10004296 &first_bad, &bad_sectors)) {
4297 /* too big in some way, or has a known bad block */
Neil Brown387bb172007-02-08 14:20:29 -08004298 bio_put(align_bi);
4299 rdev_dec_pending(rdev, mddev);
4300 return 0;
4301 }
4302
majianpeng6c0544e2012-06-12 08:31:10 +08004303 /* No reshape active, so we can trust rdev->data_offset */
Kent Overstreet4f024f32013-10-11 15:44:27 -07004304 align_bi->bi_iter.bi_sector += rdev->data_offset;
majianpeng6c0544e2012-06-12 08:31:10 +08004305
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004306 spin_lock_irq(&conf->device_lock);
4307 wait_event_lock_irq(conf->wait_for_stripe,
4308 conf->quiesce == 0,
Lukas Czernereed8c022012-11-30 11:42:40 +01004309 conf->device_lock);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004310 atomic_inc(&conf->active_aligned_reads);
4311 spin_unlock_irq(&conf->device_lock);
4312
Jonathan Brassowe3620a32013-03-07 16:22:01 -06004313 if (mddev->gendisk)
4314 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4315 align_bi, disk_devt(mddev->gendisk),
Kent Overstreet4f024f32013-10-11 15:44:27 -07004316 raid_bio->bi_iter.bi_sector);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004317 generic_make_request(align_bi);
4318 return 1;
4319 } else {
4320 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004321 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004322 return 0;
4323 }
4324}
4325
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004326/* __get_priority_stripe - get the next stripe to process
4327 *
4328 * Full stripe writes are allowed to pass preread active stripes up until
4329 * the bypass_threshold is exceeded. In general the bypass_count
4330 * increments when the handle_list is handled before the hold_list; however, it
4331 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4332 * stripe with in flight i/o. The bypass_count will be reset when the
4333 * head of the hold_list has changed, i.e. the head was promoted to the
4334 * handle_list.
4335 */
Shaohua Li851c30c2013-08-28 14:30:16 +08004336static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004337{
Shaohua Li851c30c2013-08-28 14:30:16 +08004338 struct stripe_head *sh = NULL, *tmp;
4339 struct list_head *handle_list = NULL;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004340 struct r5worker_group *wg = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08004341
4342 if (conf->worker_cnt_per_group == 0) {
4343 handle_list = &conf->handle_list;
4344 } else if (group != ANY_GROUP) {
4345 handle_list = &conf->worker_groups[group].handle_list;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004346 wg = &conf->worker_groups[group];
Shaohua Li851c30c2013-08-28 14:30:16 +08004347 } else {
4348 int i;
4349 for (i = 0; i < conf->group_cnt; i++) {
4350 handle_list = &conf->worker_groups[i].handle_list;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004351 wg = &conf->worker_groups[i];
Shaohua Li851c30c2013-08-28 14:30:16 +08004352 if (!list_empty(handle_list))
4353 break;
4354 }
4355 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004356
4357 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4358 __func__,
Shaohua Li851c30c2013-08-28 14:30:16 +08004359 list_empty(handle_list) ? "empty" : "busy",
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004360 list_empty(&conf->hold_list) ? "empty" : "busy",
4361 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4362
Shaohua Li851c30c2013-08-28 14:30:16 +08004363 if (!list_empty(handle_list)) {
4364 sh = list_entry(handle_list->next, typeof(*sh), lru);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004365
4366 if (list_empty(&conf->hold_list))
4367 conf->bypass_count = 0;
4368 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4369 if (conf->hold_list.next == conf->last_hold)
4370 conf->bypass_count++;
4371 else {
4372 conf->last_hold = conf->hold_list.next;
4373 conf->bypass_count -= conf->bypass_threshold;
4374 if (conf->bypass_count < 0)
4375 conf->bypass_count = 0;
4376 }
4377 }
4378 } else if (!list_empty(&conf->hold_list) &&
4379 ((conf->bypass_threshold &&
4380 conf->bypass_count > conf->bypass_threshold) ||
4381 atomic_read(&conf->pending_full_writes) == 0)) {
Shaohua Li851c30c2013-08-28 14:30:16 +08004382
4383 list_for_each_entry(tmp, &conf->hold_list, lru) {
4384 if (conf->worker_cnt_per_group == 0 ||
4385 group == ANY_GROUP ||
4386 !cpu_online(tmp->cpu) ||
4387 cpu_to_group(tmp->cpu) == group) {
4388 sh = tmp;
4389 break;
4390 }
4391 }
4392
4393 if (sh) {
4394 conf->bypass_count -= conf->bypass_threshold;
4395 if (conf->bypass_count < 0)
4396 conf->bypass_count = 0;
4397 }
Shaohua Libfc90cb2013-08-29 15:40:32 +08004398 wg = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08004399 }
4400
4401 if (!sh)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004402 return NULL;
4403
Shaohua Libfc90cb2013-08-29 15:40:32 +08004404 if (wg) {
4405 wg->stripes_cnt--;
4406 sh->group = NULL;
4407 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004408 list_del_init(&sh->lru);
Shaohua Lic7a6d352014-04-15 09:12:54 +08004409 BUG_ON(atomic_inc_return(&sh->count) != 1);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004410 return sh;
4411}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004412
Shaohua Li8811b592012-08-02 08:33:00 +10004413struct raid5_plug_cb {
4414 struct blk_plug_cb cb;
4415 struct list_head list;
Shaohua Li566c09c2013-11-14 15:16:17 +11004416 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Li8811b592012-08-02 08:33:00 +10004417};
4418
4419static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4420{
4421 struct raid5_plug_cb *cb = container_of(
4422 blk_cb, struct raid5_plug_cb, cb);
4423 struct stripe_head *sh;
4424 struct mddev *mddev = cb->cb.data;
4425 struct r5conf *conf = mddev->private;
NeilBrowna9add5d2012-10-31 11:59:09 +11004426 int cnt = 0;
Shaohua Li566c09c2013-11-14 15:16:17 +11004427 int hash;
Shaohua Li8811b592012-08-02 08:33:00 +10004428
4429 if (cb->list.next && !list_empty(&cb->list)) {
4430 spin_lock_irq(&conf->device_lock);
4431 while (!list_empty(&cb->list)) {
4432 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4433 list_del_init(&sh->lru);
4434 /*
4435 * avoid race release_stripe_plug() sees
4436 * STRIPE_ON_UNPLUG_LIST clear but the stripe
4437 * is still in our list
4438 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01004439 smp_mb__before_atomic();
Shaohua Li8811b592012-08-02 08:33:00 +10004440 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
Shaohua Li773ca822013-08-27 17:50:39 +08004441 /*
4442 * STRIPE_ON_RELEASE_LIST could be set here. In that
4443 * case, the count is always > 1 here
4444 */
Shaohua Li566c09c2013-11-14 15:16:17 +11004445 hash = sh->hash_lock_index;
4446 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
NeilBrowna9add5d2012-10-31 11:59:09 +11004447 cnt++;
Shaohua Li8811b592012-08-02 08:33:00 +10004448 }
4449 spin_unlock_irq(&conf->device_lock);
4450 }
Shaohua Li566c09c2013-11-14 15:16:17 +11004451 release_inactive_stripe_list(conf, cb->temp_inactive_list,
4452 NR_STRIPE_HASH_LOCKS);
Jonathan Brassowe3620a32013-03-07 16:22:01 -06004453 if (mddev->queue)
4454 trace_block_unplug(mddev->queue, cnt, !from_schedule);
Shaohua Li8811b592012-08-02 08:33:00 +10004455 kfree(cb);
4456}
4457
4458static void release_stripe_plug(struct mddev *mddev,
4459 struct stripe_head *sh)
4460{
4461 struct blk_plug_cb *blk_cb = blk_check_plugged(
4462 raid5_unplug, mddev,
4463 sizeof(struct raid5_plug_cb));
4464 struct raid5_plug_cb *cb;
4465
4466 if (!blk_cb) {
4467 release_stripe(sh);
4468 return;
4469 }
4470
4471 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
4472
Shaohua Li566c09c2013-11-14 15:16:17 +11004473 if (cb->list.next == NULL) {
4474 int i;
Shaohua Li8811b592012-08-02 08:33:00 +10004475 INIT_LIST_HEAD(&cb->list);
Shaohua Li566c09c2013-11-14 15:16:17 +11004476 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
4477 INIT_LIST_HEAD(cb->temp_inactive_list + i);
4478 }
Shaohua Li8811b592012-08-02 08:33:00 +10004479
4480 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
4481 list_add_tail(&sh->lru, &cb->list);
4482 else
4483 release_stripe(sh);
4484}
4485
Shaohua Li620125f2012-10-11 13:49:05 +11004486static void make_discard_request(struct mddev *mddev, struct bio *bi)
4487{
4488 struct r5conf *conf = mddev->private;
4489 sector_t logical_sector, last_sector;
4490 struct stripe_head *sh;
4491 int remaining;
4492 int stripe_sectors;
4493
4494 if (mddev->reshape_position != MaxSector)
4495 /* Skip discard while reshape is happening */
4496 return;
4497
Kent Overstreet4f024f32013-10-11 15:44:27 -07004498 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
4499 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
Shaohua Li620125f2012-10-11 13:49:05 +11004500
4501 bi->bi_next = NULL;
4502 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
4503
4504 stripe_sectors = conf->chunk_sectors *
4505 (conf->raid_disks - conf->max_degraded);
4506 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
4507 stripe_sectors);
4508 sector_div(last_sector, stripe_sectors);
4509
4510 logical_sector *= conf->chunk_sectors;
4511 last_sector *= conf->chunk_sectors;
4512
4513 for (; logical_sector < last_sector;
4514 logical_sector += STRIPE_SECTORS) {
4515 DEFINE_WAIT(w);
4516 int d;
4517 again:
4518 sh = get_active_stripe(conf, logical_sector, 0, 0, 0);
4519 prepare_to_wait(&conf->wait_for_overlap, &w,
4520 TASK_UNINTERRUPTIBLE);
NeilBrownf8dfcff2013-03-12 12:18:06 +11004521 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
4522 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4523 release_stripe(sh);
4524 schedule();
4525 goto again;
4526 }
4527 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
Shaohua Li620125f2012-10-11 13:49:05 +11004528 spin_lock_irq(&sh->stripe_lock);
4529 for (d = 0; d < conf->raid_disks; d++) {
4530 if (d == sh->pd_idx || d == sh->qd_idx)
4531 continue;
4532 if (sh->dev[d].towrite || sh->dev[d].toread) {
4533 set_bit(R5_Overlap, &sh->dev[d].flags);
4534 spin_unlock_irq(&sh->stripe_lock);
4535 release_stripe(sh);
4536 schedule();
4537 goto again;
4538 }
4539 }
NeilBrownf8dfcff2013-03-12 12:18:06 +11004540 set_bit(STRIPE_DISCARD, &sh->state);
Shaohua Li620125f2012-10-11 13:49:05 +11004541 finish_wait(&conf->wait_for_overlap, &w);
4542 for (d = 0; d < conf->raid_disks; d++) {
4543 if (d == sh->pd_idx || d == sh->qd_idx)
4544 continue;
4545 sh->dev[d].towrite = bi;
4546 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
4547 raid5_inc_bi_active_stripes(bi);
4548 }
4549 spin_unlock_irq(&sh->stripe_lock);
4550 if (conf->mddev->bitmap) {
4551 for (d = 0;
4552 d < conf->raid_disks - conf->max_degraded;
4553 d++)
4554 bitmap_startwrite(mddev->bitmap,
4555 sh->sector,
4556 STRIPE_SECTORS,
4557 0);
4558 sh->bm_seq = conf->seq_flush + 1;
4559 set_bit(STRIPE_BIT_DELAY, &sh->state);
4560 }
4561
4562 set_bit(STRIPE_HANDLE, &sh->state);
4563 clear_bit(STRIPE_DELAYED, &sh->state);
4564 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4565 atomic_inc(&conf->preread_active_stripes);
4566 release_stripe_plug(mddev, sh);
4567 }
4568
4569 remaining = raid5_dec_bi_active_stripes(bi);
4570 if (remaining == 0) {
4571 md_write_end(mddev);
4572 bio_endio(bi, 0);
4573 }
4574}
4575
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -07004576static void make_request(struct mddev *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004577{
NeilBrownd1688a62011-10-11 16:49:52 +11004578 struct r5conf *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11004579 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004580 sector_t new_sector;
4581 sector_t logical_sector, last_sector;
4582 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01004583 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11004584 int remaining;
Shaohua Li27c0f682014-04-09 11:25:47 +08004585 DEFINE_WAIT(w);
4586 bool do_prepare;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004587
Tejun Heoe9c74692010-09-03 11:56:18 +02004588 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
4589 md_flush_request(mddev, bi);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02004590 return;
NeilBrowne5dcdd82005-09-09 16:23:41 -07004591 }
4592
NeilBrown3d310eb2005-06-21 17:17:26 -07004593 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07004594
NeilBrown802ba062006-12-13 00:34:13 -08004595 if (rw == READ &&
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08004596 mddev->reshape_position == MaxSector &&
NeilBrown21a52c62010-04-01 15:02:13 +11004597 chunk_aligned_read(mddev,bi))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02004598 return;
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08004599
Shaohua Li620125f2012-10-11 13:49:05 +11004600 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
4601 make_discard_request(mddev, bi);
4602 return;
4603 }
4604
Kent Overstreet4f024f32013-10-11 15:44:27 -07004605 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07004606 last_sector = bio_end_sector(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004607 bi->bi_next = NULL;
4608 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07004609
Shaohua Li27c0f682014-04-09 11:25:47 +08004610 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004611 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
NeilBrownb5663ba2009-03-31 14:39:38 +11004612 int previous;
NeilBrownc46501b2013-08-27 15:52:13 +10004613 int seq;
NeilBrownb578d552006-03-27 01:18:12 -08004614
Shaohua Li27c0f682014-04-09 11:25:47 +08004615 do_prepare = false;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004616 retry:
NeilBrownc46501b2013-08-27 15:52:13 +10004617 seq = read_seqcount_begin(&conf->gen_lock);
NeilBrownb5663ba2009-03-31 14:39:38 +11004618 previous = 0;
Shaohua Li27c0f682014-04-09 11:25:47 +08004619 if (do_prepare)
4620 prepare_to_wait(&conf->wait_for_overlap, &w,
4621 TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11004622 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11004623 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08004624 * 64bit on a 32bit platform, and so it might be
4625 * possible to see a half-updated value
Jesper Juhlaeb878b2011-04-10 18:06:17 +02004626 * Of course reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08004627 * the lock is dropped, so once we get a reference
4628 * to the stripe that we think it is, we will have
4629 * to check again.
4630 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004631 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10004632 if (mddev->reshape_backwards
NeilBrownfef9c612009-03-31 15:16:46 +11004633 ? logical_sector < conf->reshape_progress
4634 : logical_sector >= conf->reshape_progress) {
NeilBrownb5663ba2009-03-31 14:39:38 +11004635 previous = 1;
4636 } else {
NeilBrown2c810cd2012-05-21 09:27:00 +10004637 if (mddev->reshape_backwards
NeilBrownfef9c612009-03-31 15:16:46 +11004638 ? logical_sector < conf->reshape_safe
4639 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08004640 spin_unlock_irq(&conf->device_lock);
4641 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08004642 do_prepare = true;
NeilBrownb578d552006-03-27 01:18:12 -08004643 goto retry;
4644 }
4645 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004646 spin_unlock_irq(&conf->device_lock);
4647 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004648
NeilBrown112bf892009-03-31 14:39:38 +11004649 new_sector = raid5_compute_sector(conf, logical_sector,
4650 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11004651 &dd_idx, NULL);
NeilBrown0c55e022010-05-03 14:09:02 +10004652 pr_debug("raid456: make_request, sector %llu logical %llu\n",
NeilBrownc46501b2013-08-27 15:52:13 +10004653 (unsigned long long)new_sector,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004654 (unsigned long long)logical_sector);
4655
NeilBrownb5663ba2009-03-31 14:39:38 +11004656 sh = get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10004657 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004658 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11004659 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004660 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08004661 * stripe, so we must do the range check again.
4662 * Expansion could still move past after this
4663 * test, but as we are holding a reference to
4664 * 'sh', we know that if that happens,
4665 * STRIPE_EXPANDING will get set and the expansion
4666 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004667 */
4668 int must_retry = 0;
4669 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10004670 if (mddev->reshape_backwards
NeilBrownb0f9ec02009-03-31 15:27:18 +11004671 ? logical_sector >= conf->reshape_progress
4672 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004673 /* mismatch, need to try again */
4674 must_retry = 1;
4675 spin_unlock_irq(&conf->device_lock);
4676 if (must_retry) {
4677 release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07004678 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08004679 do_prepare = true;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004680 goto retry;
4681 }
4682 }
NeilBrownc46501b2013-08-27 15:52:13 +10004683 if (read_seqcount_retry(&conf->gen_lock, seq)) {
4684 /* Might have got the wrong stripe_head
4685 * by accident
4686 */
4687 release_stripe(sh);
4688 goto retry;
4689 }
NeilBrowne62e58a2009-07-01 13:15:35 +10004690
Namhyung Kimffd96e32011-07-18 17:38:51 +10004691 if (rw == WRITE &&
NeilBrowna5c308d2009-07-01 13:15:35 +10004692 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08004693 logical_sector < mddev->suspend_hi) {
4694 release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10004695 /* As the suspend_* range is controlled by
4696 * userspace, we want an interruptible
4697 * wait.
4698 */
4699 flush_signals(current);
4700 prepare_to_wait(&conf->wait_for_overlap,
4701 &w, TASK_INTERRUPTIBLE);
4702 if (logical_sector >= mddev->suspend_lo &&
Shaohua Li27c0f682014-04-09 11:25:47 +08004703 logical_sector < mddev->suspend_hi) {
NeilBrowne62e58a2009-07-01 13:15:35 +10004704 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08004705 do_prepare = true;
4706 }
NeilBrowne464eaf2006-03-27 01:18:14 -08004707 goto retry;
4708 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004709
4710 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
Namhyung Kimffd96e32011-07-18 17:38:51 +10004711 !add_stripe_bio(sh, bi, dd_idx, rw)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004712 /* Stripe is busy expanding or
4713 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07004714 * and wait a while
4715 */
NeilBrown482c0832011-04-18 18:25:42 +10004716 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004717 release_stripe(sh);
4718 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08004719 do_prepare = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004720 goto retry;
4721 }
NeilBrown6ed30032008-02-06 01:40:00 -08004722 set_bit(STRIPE_HANDLE, &sh->state);
4723 clear_bit(STRIPE_DELAYED, &sh->state);
NeilBrowna852d7b2012-09-19 12:48:30 +10004724 if ((bi->bi_rw & REQ_SYNC) &&
NeilBrown729a1862009-12-14 12:49:50 +11004725 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4726 atomic_inc(&conf->preread_active_stripes);
Shaohua Li8811b592012-08-02 08:33:00 +10004727 release_stripe_plug(mddev, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004728 } else {
4729 /* cannot get stripe for read-ahead, just give-up */
4730 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004731 break;
4732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004733 }
Shaohua Li27c0f682014-04-09 11:25:47 +08004734 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown7c13edc2011-04-18 18:25:43 +10004735
Shaohua Lie7836bd62012-07-19 16:01:31 +10004736 remaining = raid5_dec_bi_active_stripes(bi);
NeilBrownf6344752006-03-27 01:18:17 -08004737 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004738
NeilBrown16a53ec2006-06-26 00:27:38 -07004739 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004740 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02004741
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07004742 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
4743 bi, 0);
Neil Brown0e13fe232008-06-28 08:31:20 +10004744 bio_endio(bi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004746}
4747
NeilBrownfd01b882011-10-11 16:47:53 +11004748static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
Dan Williamsb522adc2009-03-31 15:00:31 +11004749
NeilBrownfd01b882011-10-11 16:47:53 +11004750static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004751{
NeilBrown52c03292006-06-26 00:27:43 -07004752 /* reshaping is quite different to recovery/resync so it is
4753 * handled quite separately ... here.
4754 *
4755 * On each call to sync_request, we gather one chunk worth of
4756 * destination stripes and flag them as expanding.
4757 * Then we find all the source stripes and request reads.
4758 * As the reads complete, handle_stripe will copy the data
4759 * into the destination stripe and release that stripe.
4760 */
NeilBrownd1688a62011-10-11 16:49:52 +11004761 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004762 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08004763 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08004764 int raid_disks = conf->previous_raid_disks;
4765 int data_disks = raid_disks - conf->max_degraded;
4766 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07004767 int i;
4768 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11004769 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11004770 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11004771 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11004772 struct list_head stripes;
NeilBrown52c03292006-06-26 00:27:43 -07004773
NeilBrownfef9c612009-03-31 15:16:46 +11004774 if (sector_nr == 0) {
4775 /* If restarting in the middle, skip the initial sectors */
NeilBrown2c810cd2012-05-21 09:27:00 +10004776 if (mddev->reshape_backwards &&
NeilBrownfef9c612009-03-31 15:16:46 +11004777 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4778 sector_nr = raid5_size(mddev, 0, 0)
4779 - conf->reshape_progress;
NeilBrown2c810cd2012-05-21 09:27:00 +10004780 } else if (!mddev->reshape_backwards &&
NeilBrownfef9c612009-03-31 15:16:46 +11004781 conf->reshape_progress > 0)
4782 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004783 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004784 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11004785 mddev->curr_resync_completed = sector_nr;
4786 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11004787 *skipped = 1;
4788 return sector_nr;
4789 }
NeilBrown52c03292006-06-26 00:27:43 -07004790 }
4791
NeilBrown7a661382009-03-31 15:21:40 +11004792 /* We need to process a full chunk at a time.
4793 * If old and new chunk sizes differ, we need to process the
4794 * largest of these
4795 */
Andre Noll664e7c42009-06-18 08:45:27 +10004796 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4797 reshape_sectors = mddev->new_chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004798 else
Andre Noll9d8f0362009-06-18 08:45:01 +10004799 reshape_sectors = mddev->chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004800
NeilBrownb5254dd2012-05-21 09:27:01 +10004801 /* We update the metadata at least every 10 seconds, or when
4802 * the data about to be copied would over-write the source of
4803 * the data at the front of the range. i.e. one new_stripe
4804 * along from reshape_progress new_maps to after where
4805 * reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07004806 */
NeilBrownfef9c612009-03-31 15:16:46 +11004807 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004808 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11004809 readpos = conf->reshape_progress;
4810 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004811 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08004812 sector_div(safepos, data_disks);
NeilBrown2c810cd2012-05-21 09:27:00 +10004813 if (mddev->reshape_backwards) {
NeilBrowned37d832009-05-27 21:39:05 +10004814 writepos -= min_t(sector_t, reshape_sectors, writepos);
NeilBrownc8f517c2009-03-31 15:28:40 +11004815 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004816 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11004817 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004818 writepos += reshape_sectors;
NeilBrowned37d832009-05-27 21:39:05 +10004819 readpos -= min_t(sector_t, reshape_sectors, readpos);
4820 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11004821 }
NeilBrown52c03292006-06-26 00:27:43 -07004822
NeilBrownb5254dd2012-05-21 09:27:01 +10004823 /* Having calculated the 'writepos' possibly use it
4824 * to set 'stripe_addr' which is where we will write to.
4825 */
4826 if (mddev->reshape_backwards) {
4827 BUG_ON(conf->reshape_progress == 0);
4828 stripe_addr = writepos;
4829 BUG_ON((mddev->dev_sectors &
4830 ~((sector_t)reshape_sectors - 1))
4831 - reshape_sectors - stripe_addr
4832 != sector_nr);
4833 } else {
4834 BUG_ON(writepos != sector_nr + reshape_sectors);
4835 stripe_addr = sector_nr;
4836 }
4837
NeilBrownc8f517c2009-03-31 15:28:40 +11004838 /* 'writepos' is the most advanced device address we might write.
4839 * 'readpos' is the least advanced device address we might read.
4840 * 'safepos' is the least address recorded in the metadata as having
4841 * been reshaped.
NeilBrownb5254dd2012-05-21 09:27:01 +10004842 * If there is a min_offset_diff, these are adjusted either by
4843 * increasing the safepos/readpos if diff is negative, or
4844 * increasing writepos if diff is positive.
4845 * If 'readpos' is then behind 'writepos', there is no way that we can
NeilBrownc8f517c2009-03-31 15:28:40 +11004846 * ensure safety in the face of a crash - that must be done by userspace
4847 * making a backup of the data. So in that case there is no particular
4848 * rush to update metadata.
4849 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4850 * update the metadata to advance 'safepos' to match 'readpos' so that
4851 * we can be safe in the event of a crash.
4852 * So we insist on updating metadata if safepos is behind writepos and
4853 * readpos is beyond writepos.
4854 * In any case, update the metadata every 10 seconds.
4855 * Maybe that number should be configurable, but I'm not sure it is
4856 * worth it.... maybe it could be a multiple of safemode_delay???
4857 */
NeilBrownb5254dd2012-05-21 09:27:01 +10004858 if (conf->min_offset_diff < 0) {
4859 safepos += -conf->min_offset_diff;
4860 readpos += -conf->min_offset_diff;
4861 } else
4862 writepos += conf->min_offset_diff;
4863
NeilBrown2c810cd2012-05-21 09:27:00 +10004864 if ((mddev->reshape_backwards
NeilBrownc8f517c2009-03-31 15:28:40 +11004865 ? (safepos > writepos && readpos < writepos)
4866 : (safepos < writepos && readpos > writepos)) ||
4867 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07004868 /* Cannot proceed until we've updated the superblock... */
4869 wait_event(conf->wait_for_overlap,
NeilBrownc91abf52013-11-19 12:02:01 +11004870 atomic_read(&conf->reshape_stripes)==0
4871 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4872 if (atomic_read(&conf->reshape_stripes) != 0)
4873 return 0;
NeilBrownfef9c612009-03-31 15:16:46 +11004874 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004875 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004876 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07004877 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07004878 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07004879 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrownc91abf52013-11-19 12:02:01 +11004880 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4881 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4882 return 0;
NeilBrown52c03292006-06-26 00:27:43 -07004883 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004884 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07004885 spin_unlock_irq(&conf->device_lock);
4886 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004887 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07004888 }
4889
NeilBrownab69ae12009-03-31 15:26:47 +11004890 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11004891 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07004892 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10004893 int skipped_disk = 0;
NeilBrowna8c906c2009-06-09 14:39:59 +10004894 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004895 set_bit(STRIPE_EXPANDING, &sh->state);
4896 atomic_inc(&conf->reshape_stripes);
4897 /* If any of this stripe is beyond the end of the old
4898 * array, then we need to zero those blocks
4899 */
4900 for (j=sh->disks; j--;) {
4901 sector_t s;
4902 if (j == sh->pd_idx)
4903 continue;
NeilBrownf4168852007-02-28 20:11:53 -08004904 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11004905 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08004906 continue;
NeilBrown784052e2009-03-31 15:19:07 +11004907 s = compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11004908 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10004909 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07004910 continue;
4911 }
4912 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4913 set_bit(R5_Expanded, &sh->dev[j].flags);
4914 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4915 }
NeilBrowna9f326e2009-09-23 18:06:41 +10004916 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07004917 set_bit(STRIPE_EXPAND_READY, &sh->state);
4918 set_bit(STRIPE_HANDLE, &sh->state);
4919 }
NeilBrownab69ae12009-03-31 15:26:47 +11004920 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07004921 }
4922 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10004923 if (mddev->reshape_backwards)
NeilBrown7a661382009-03-31 15:21:40 +11004924 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11004925 else
NeilBrown7a661382009-03-31 15:21:40 +11004926 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07004927 spin_unlock_irq(&conf->device_lock);
4928 /* Ok, those stripe are ready. We can start scheduling
4929 * reads on the source stripes.
4930 * The source stripes are determined by mapping the first and last
4931 * block on the destination stripes.
4932 */
NeilBrown52c03292006-06-26 00:27:43 -07004933 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11004934 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11004935 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07004936 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10004937 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10004938 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11004939 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11004940 if (last_sector >= mddev->dev_sectors)
4941 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07004942 while (first_sector <= last_sector) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004943 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004944 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4945 set_bit(STRIPE_HANDLE, &sh->state);
4946 release_stripe(sh);
4947 first_sector += STRIPE_SECTORS;
4948 }
NeilBrownab69ae12009-03-31 15:26:47 +11004949 /* Now that the sources are clearly marked, we can release
4950 * the destination stripes
4951 */
4952 while (!list_empty(&stripes)) {
4953 sh = list_entry(stripes.next, struct stripe_head, lru);
4954 list_del_init(&sh->lru);
4955 release_stripe(sh);
4956 }
NeilBrownc6207272008-02-06 01:39:52 -08004957 /* If this takes us to the resync_max point where we have to pause,
4958 * then we need to write out the superblock.
4959 */
NeilBrown7a661382009-03-31 15:21:40 +11004960 sector_nr += reshape_sectors;
NeilBrownc03f6a12009-04-17 11:06:30 +10004961 if ((sector_nr - mddev->curr_resync_completed) * 2
4962 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08004963 /* Cannot proceed until we've updated the superblock... */
4964 wait_event(conf->wait_for_overlap,
NeilBrownc91abf52013-11-19 12:02:01 +11004965 atomic_read(&conf->reshape_stripes) == 0
4966 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4967 if (atomic_read(&conf->reshape_stripes) != 0)
4968 goto ret;
NeilBrownfef9c612009-03-31 15:16:46 +11004969 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004970 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004971 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08004972 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4973 md_wakeup_thread(mddev->thread);
4974 wait_event(mddev->sb_wait,
4975 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
NeilBrownc91abf52013-11-19 12:02:01 +11004976 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4977 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4978 goto ret;
NeilBrownc6207272008-02-06 01:39:52 -08004979 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004980 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08004981 spin_unlock_irq(&conf->device_lock);
4982 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004983 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08004984 }
NeilBrownc91abf52013-11-19 12:02:01 +11004985ret:
NeilBrown7a661382009-03-31 15:21:40 +11004986 return reshape_sectors;
NeilBrown52c03292006-06-26 00:27:43 -07004987}
4988
4989/* FIXME go_faster isn't used */
NeilBrownfd01b882011-10-11 16:47:53 +11004990static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
NeilBrown52c03292006-06-26 00:27:43 -07004991{
NeilBrownd1688a62011-10-11 16:49:52 +11004992 struct r5conf *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07004993 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11004994 sector_t max_sector = mddev->dev_sectors;
NeilBrown57dab0b2010-10-19 10:03:39 +11004995 sector_t sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07004996 int still_degraded = 0;
4997 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998
NeilBrown72626682005-09-09 16:23:54 -07004999 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005000 /* just being told to finish up .. nothing much to do */
NeilBrowncea9c222009-03-31 15:15:05 +11005001
NeilBrown29269552006-03-27 01:18:10 -08005002 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
5003 end_reshape(conf);
5004 return 0;
5005 }
NeilBrown72626682005-09-09 16:23:54 -07005006
5007 if (mddev->curr_resync < max_sector) /* aborted */
5008 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
5009 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07005010 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07005011 conf->fullsync = 0;
5012 bitmap_close_sync(mddev->bitmap);
5013
Linus Torvalds1da177e2005-04-16 15:20:36 -07005014 return 0;
5015 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08005016
NeilBrown64bd6602009-08-03 10:59:58 +10005017 /* Allow raid5_quiesce to complete */
5018 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
5019
NeilBrown52c03292006-06-26 00:27:43 -07005020 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5021 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08005022
NeilBrownc6207272008-02-06 01:39:52 -08005023 /* No need to check resync_max as we never do more than one
5024 * stripe, and as resync_max will always be on a chunk boundary,
5025 * if the check in md_do_sync didn't fire, there is no chance
5026 * of overstepping resync_max here
5027 */
5028
NeilBrown16a53ec2006-06-26 00:27:38 -07005029 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07005030 * to resync, then assert that we are finished, because there is
5031 * nothing we can do.
5032 */
NeilBrown3285edf2006-06-26 00:27:55 -07005033 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07005034 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11005035 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07005036 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005037 return rv;
5038 }
majianpeng6f608042013-04-24 11:42:41 +10005039 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5040 !conf->fullsync &&
5041 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5042 sync_blocks >= STRIPE_SECTORS) {
NeilBrown72626682005-09-09 16:23:54 -07005043 /* we can skip this block, and probably more */
5044 sync_blocks /= STRIPE_SECTORS;
5045 *skipped = 1;
5046 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005048
NeilBrownb47490c2008-02-06 01:39:50 -08005049 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
5050
NeilBrowna8c906c2009-06-09 14:39:59 +10005051 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005052 if (sh == NULL) {
NeilBrowna8c906c2009-06-09 14:39:59 +10005053 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005054 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07005055 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07005056 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08005057 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005058 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005059 /* Need to check if array will still be degraded after recovery/resync
5060 * We don't need to check the 'failed' flag as when that gets set,
5061 * recovery aborts.
5062 */
NeilBrownf001a702009-06-09 14:30:31 +10005063 for (i = 0; i < conf->raid_disks; i++)
NeilBrown16a53ec2006-06-26 00:27:38 -07005064 if (conf->disks[i].rdev == NULL)
5065 still_degraded = 1;
5066
5067 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5068
NeilBrown83206d62011-07-26 11:19:49 +10005069 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
Eivind Sarto053f5b62014-06-09 17:06:19 -07005070 set_bit(STRIPE_HANDLE, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005071
Linus Torvalds1da177e2005-04-16 15:20:36 -07005072 release_stripe(sh);
5073
5074 return STRIPE_SECTORS;
5075}
5076
NeilBrownd1688a62011-10-11 16:49:52 +11005077static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005078{
5079 /* We may not be able to submit a whole bio at once as there
5080 * may not be enough stripe_heads available.
5081 * We cannot pre-allocate enough stripe_heads as we may need
5082 * more than exist in the cache (if we allow ever large chunks).
5083 * So we do one stripe head at a time and record in
5084 * ->bi_hw_segments how many have been done.
5085 *
5086 * We *know* that this entire raid_bio is in one chunk, so
5087 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5088 */
5089 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11005090 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005091 sector_t sector, logical_sector, last_sector;
5092 int scnt = 0;
5093 int remaining;
5094 int handled = 0;
5095
Kent Overstreet4f024f32013-10-11 15:44:27 -07005096 logical_sector = raid_bio->bi_iter.bi_sector &
5097 ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11005098 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11005099 0, &dd_idx, NULL);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07005100 last_sector = bio_end_sector(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005101
5102 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08005103 logical_sector += STRIPE_SECTORS,
5104 sector += STRIPE_SECTORS,
5105 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005106
Shaohua Lie7836bd62012-07-19 16:01:31 +10005107 if (scnt < raid5_bi_processed_stripes(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005108 /* already done this stripe */
5109 continue;
5110
hui jiao2844dc32014-06-05 11:34:24 +08005111 sh = get_active_stripe(conf, sector, 0, 1, 1);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005112
5113 if (!sh) {
5114 /* failed to get a stripe - must wait */
Shaohua Lie7836bd62012-07-19 16:01:31 +10005115 raid5_set_bi_processed_stripes(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005116 conf->retry_read_aligned = raid_bio;
5117 return handled;
5118 }
5119
Neil Brown387bb172007-02-08 14:20:29 -08005120 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
5121 release_stripe(sh);
Shaohua Lie7836bd62012-07-19 16:01:31 +10005122 raid5_set_bi_processed_stripes(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08005123 conf->retry_read_aligned = raid_bio;
5124 return handled;
5125 }
5126
majianpeng3f9e7c12012-07-31 10:04:21 +10005127 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
Dan Williams36d1c642009-07-14 11:48:22 -07005128 handle_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005129 release_stripe(sh);
5130 handled++;
5131 }
Shaohua Lie7836bd62012-07-19 16:01:31 +10005132 remaining = raid5_dec_bi_active_stripes(raid_bio);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07005133 if (remaining == 0) {
5134 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5135 raid_bio, 0);
Neil Brown0e13fe232008-06-28 08:31:20 +10005136 bio_endio(raid_bio, 0);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07005137 }
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005138 if (atomic_dec_and_test(&conf->active_aligned_reads))
5139 wake_up(&conf->wait_for_stripe);
5140 return handled;
5141}
5142
Shaohua Libfc90cb2013-08-29 15:40:32 +08005143static int handle_active_stripes(struct r5conf *conf, int group,
Shaohua Li566c09c2013-11-14 15:16:17 +11005144 struct r5worker *worker,
5145 struct list_head *temp_inactive_list)
Shaohua Li46a06402012-08-02 08:33:15 +10005146{
5147 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
Shaohua Li566c09c2013-11-14 15:16:17 +11005148 int i, batch_size = 0, hash;
5149 bool release_inactive = false;
Shaohua Li46a06402012-08-02 08:33:15 +10005150
5151 while (batch_size < MAX_STRIPE_BATCH &&
Shaohua Li851c30c2013-08-28 14:30:16 +08005152 (sh = __get_priority_stripe(conf, group)) != NULL)
Shaohua Li46a06402012-08-02 08:33:15 +10005153 batch[batch_size++] = sh;
5154
Shaohua Li566c09c2013-11-14 15:16:17 +11005155 if (batch_size == 0) {
5156 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5157 if (!list_empty(temp_inactive_list + i))
5158 break;
5159 if (i == NR_STRIPE_HASH_LOCKS)
5160 return batch_size;
5161 release_inactive = true;
5162 }
Shaohua Li46a06402012-08-02 08:33:15 +10005163 spin_unlock_irq(&conf->device_lock);
5164
Shaohua Li566c09c2013-11-14 15:16:17 +11005165 release_inactive_stripe_list(conf, temp_inactive_list,
5166 NR_STRIPE_HASH_LOCKS);
5167
5168 if (release_inactive) {
5169 spin_lock_irq(&conf->device_lock);
5170 return 0;
5171 }
5172
Shaohua Li46a06402012-08-02 08:33:15 +10005173 for (i = 0; i < batch_size; i++)
5174 handle_stripe(batch[i]);
5175
5176 cond_resched();
5177
5178 spin_lock_irq(&conf->device_lock);
Shaohua Li566c09c2013-11-14 15:16:17 +11005179 for (i = 0; i < batch_size; i++) {
5180 hash = batch[i]->hash_lock_index;
5181 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5182 }
Shaohua Li46a06402012-08-02 08:33:15 +10005183 return batch_size;
5184}
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005185
Shaohua Li851c30c2013-08-28 14:30:16 +08005186static void raid5_do_work(struct work_struct *work)
5187{
5188 struct r5worker *worker = container_of(work, struct r5worker, work);
5189 struct r5worker_group *group = worker->group;
5190 struct r5conf *conf = group->conf;
5191 int group_id = group - conf->worker_groups;
5192 int handled;
5193 struct blk_plug plug;
5194
5195 pr_debug("+++ raid5worker active\n");
5196
5197 blk_start_plug(&plug);
5198 handled = 0;
5199 spin_lock_irq(&conf->device_lock);
5200 while (1) {
5201 int batch_size, released;
5202
Shaohua Li566c09c2013-11-14 15:16:17 +11005203 released = release_stripe_list(conf, worker->temp_inactive_list);
Shaohua Li851c30c2013-08-28 14:30:16 +08005204
Shaohua Li566c09c2013-11-14 15:16:17 +11005205 batch_size = handle_active_stripes(conf, group_id, worker,
5206 worker->temp_inactive_list);
Shaohua Libfc90cb2013-08-29 15:40:32 +08005207 worker->working = false;
Shaohua Li851c30c2013-08-28 14:30:16 +08005208 if (!batch_size && !released)
5209 break;
5210 handled += batch_size;
5211 }
5212 pr_debug("%d stripes handled\n", handled);
5213
5214 spin_unlock_irq(&conf->device_lock);
5215 blk_finish_plug(&plug);
5216
5217 pr_debug("--- raid5worker inactive\n");
5218}
5219
Linus Torvalds1da177e2005-04-16 15:20:36 -07005220/*
5221 * This is our raid5 kernel thread.
5222 *
5223 * We scan the hash table for stripes which can be handled now.
5224 * During the scan, completed stripes are saved for us by the interrupt
5225 * handler, so that they will not have to wait for our next wakeup.
5226 */
Shaohua Li4ed87312012-10-11 13:34:00 +11005227static void raid5d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005228{
Shaohua Li4ed87312012-10-11 13:34:00 +11005229 struct mddev *mddev = thread->mddev;
NeilBrownd1688a62011-10-11 16:49:52 +11005230 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005231 int handled;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005232 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005233
Dan Williams45b42332007-07-09 11:56:43 -07005234 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005235
5236 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005237
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005238 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005239 handled = 0;
5240 spin_lock_irq(&conf->device_lock);
5241 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005242 struct bio *bio;
Shaohua Li773ca822013-08-27 17:50:39 +08005243 int batch_size, released;
5244
Shaohua Li566c09c2013-11-14 15:16:17 +11005245 released = release_stripe_list(conf, conf->temp_inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005246
NeilBrown0021b7b2012-07-31 09:08:14 +02005247 if (
NeilBrown7c13edc2011-04-18 18:25:43 +10005248 !list_empty(&conf->bitmap_list)) {
5249 /* Now is a good time to flush some bitmap updates */
5250 conf->seq_flush++;
NeilBrown700e4322005-11-28 13:44:10 -08005251 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07005252 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08005253 spin_lock_irq(&conf->device_lock);
NeilBrown7c13edc2011-04-18 18:25:43 +10005254 conf->seq_write = conf->seq_flush;
Shaohua Li566c09c2013-11-14 15:16:17 +11005255 activate_bit_delay(conf, conf->temp_inactive_list);
NeilBrown72626682005-09-09 16:23:54 -07005256 }
NeilBrown0021b7b2012-07-31 09:08:14 +02005257 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07005258
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005259 while ((bio = remove_bio_from_retry(conf))) {
5260 int ok;
5261 spin_unlock_irq(&conf->device_lock);
5262 ok = retry_aligned_read(conf, bio);
5263 spin_lock_irq(&conf->device_lock);
5264 if (!ok)
5265 break;
5266 handled++;
5267 }
5268
Shaohua Li566c09c2013-11-14 15:16:17 +11005269 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
5270 conf->temp_inactive_list);
Shaohua Li773ca822013-08-27 17:50:39 +08005271 if (!batch_size && !released)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272 break;
Shaohua Li46a06402012-08-02 08:33:15 +10005273 handled += batch_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005274
Shaohua Li46a06402012-08-02 08:33:15 +10005275 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
5276 spin_unlock_irq(&conf->device_lock);
NeilBrownde393cd2011-07-28 11:31:48 +10005277 md_check_recovery(mddev);
Shaohua Li46a06402012-08-02 08:33:15 +10005278 spin_lock_irq(&conf->device_lock);
5279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005280 }
Dan Williams45b42332007-07-09 11:56:43 -07005281 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005282
5283 spin_unlock_irq(&conf->device_lock);
5284
Dan Williamsc9f21aa2008-07-23 12:05:51 -07005285 async_tx_issue_pending_all();
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005286 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005287
Dan Williams45b42332007-07-09 11:56:43 -07005288 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005289}
5290
NeilBrown3f294f42005-11-08 21:39:25 -08005291static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005292raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08005293{
NeilBrownd1688a62011-10-11 16:49:52 +11005294 struct r5conf *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08005295 if (conf)
5296 return sprintf(page, "%d\n", conf->max_nr_stripes);
5297 else
5298 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08005299}
5300
NeilBrownc41d4ac2010-06-01 19:37:24 +10005301int
NeilBrownfd01b882011-10-11 16:47:53 +11005302raid5_set_cache_size(struct mddev *mddev, int size)
NeilBrownc41d4ac2010-06-01 19:37:24 +10005303{
NeilBrownd1688a62011-10-11 16:49:52 +11005304 struct r5conf *conf = mddev->private;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005305 int err;
Shaohua Li566c09c2013-11-14 15:16:17 +11005306 int hash;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005307
5308 if (size <= 16 || size > 32768)
5309 return -EINVAL;
Shaohua Li566c09c2013-11-14 15:16:17 +11005310 hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005311 while (size < conf->max_nr_stripes) {
Shaohua Li566c09c2013-11-14 15:16:17 +11005312 if (drop_one_stripe(conf, hash))
NeilBrownc41d4ac2010-06-01 19:37:24 +10005313 conf->max_nr_stripes--;
5314 else
5315 break;
Shaohua Li566c09c2013-11-14 15:16:17 +11005316 hash--;
5317 if (hash < 0)
5318 hash = NR_STRIPE_HASH_LOCKS - 1;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005319 }
5320 err = md_allow_write(mddev);
5321 if (err)
5322 return err;
Shaohua Li566c09c2013-11-14 15:16:17 +11005323 hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005324 while (size > conf->max_nr_stripes) {
Shaohua Li566c09c2013-11-14 15:16:17 +11005325 if (grow_one_stripe(conf, hash))
NeilBrownc41d4ac2010-06-01 19:37:24 +10005326 conf->max_nr_stripes++;
5327 else break;
Shaohua Li566c09c2013-11-14 15:16:17 +11005328 hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005329 }
5330 return 0;
5331}
5332EXPORT_SYMBOL(raid5_set_cache_size);
5333
NeilBrown3f294f42005-11-08 21:39:25 -08005334static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005335raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08005336{
NeilBrownd1688a62011-10-11 16:49:52 +11005337 struct r5conf *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07005338 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07005339 int err;
5340
NeilBrown3f294f42005-11-08 21:39:25 -08005341 if (len >= PAGE_SIZE)
5342 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08005343 if (!conf)
5344 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08005345
Jingoo Hanb29bebd2013-06-01 16:15:16 +09005346 if (kstrtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08005347 return -EINVAL;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005348 err = raid5_set_cache_size(mddev, new);
Dan Williamsb5470dc2008-06-27 21:44:04 -07005349 if (err)
5350 return err;
NeilBrown3f294f42005-11-08 21:39:25 -08005351 return len;
5352}
NeilBrown007583c2005-11-08 21:39:30 -08005353
NeilBrown96de1e62005-11-08 21:39:39 -08005354static struct md_sysfs_entry
5355raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
5356 raid5_show_stripe_cache_size,
5357 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08005358
5359static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005360raid5_show_preread_threshold(struct mddev *mddev, char *page)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005361{
NeilBrownd1688a62011-10-11 16:49:52 +11005362 struct r5conf *conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005363 if (conf)
5364 return sprintf(page, "%d\n", conf->bypass_threshold);
5365 else
5366 return 0;
5367}
5368
5369static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005370raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005371{
NeilBrownd1688a62011-10-11 16:49:52 +11005372 struct r5conf *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07005373 unsigned long new;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005374 if (len >= PAGE_SIZE)
5375 return -EINVAL;
5376 if (!conf)
5377 return -ENODEV;
5378
Jingoo Hanb29bebd2013-06-01 16:15:16 +09005379 if (kstrtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005380 return -EINVAL;
Dan Williams4ef197d82008-04-28 02:15:54 -07005381 if (new > conf->max_nr_stripes)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005382 return -EINVAL;
5383 conf->bypass_threshold = new;
5384 return len;
5385}
5386
5387static struct md_sysfs_entry
5388raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
5389 S_IRUGO | S_IWUSR,
5390 raid5_show_preread_threshold,
5391 raid5_store_preread_threshold);
5392
5393static ssize_t
Shaohua Lid592a992014-05-21 17:57:44 +08005394raid5_show_skip_copy(struct mddev *mddev, char *page)
5395{
5396 struct r5conf *conf = mddev->private;
5397 if (conf)
5398 return sprintf(page, "%d\n", conf->skip_copy);
5399 else
5400 return 0;
5401}
5402
5403static ssize_t
5404raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
5405{
5406 struct r5conf *conf = mddev->private;
5407 unsigned long new;
5408 if (len >= PAGE_SIZE)
5409 return -EINVAL;
5410 if (!conf)
5411 return -ENODEV;
5412
5413 if (kstrtoul(page, 10, &new))
5414 return -EINVAL;
5415 new = !!new;
5416 if (new == conf->skip_copy)
5417 return len;
5418
5419 mddev_suspend(mddev);
5420 conf->skip_copy = new;
5421 if (new)
5422 mddev->queue->backing_dev_info.capabilities |=
5423 BDI_CAP_STABLE_WRITES;
5424 else
5425 mddev->queue->backing_dev_info.capabilities &=
5426 ~BDI_CAP_STABLE_WRITES;
5427 mddev_resume(mddev);
5428 return len;
5429}
5430
5431static struct md_sysfs_entry
5432raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
5433 raid5_show_skip_copy,
5434 raid5_store_skip_copy);
5435
Shaohua Lid592a992014-05-21 17:57:44 +08005436static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005437stripe_cache_active_show(struct mddev *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08005438{
NeilBrownd1688a62011-10-11 16:49:52 +11005439 struct r5conf *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08005440 if (conf)
5441 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
5442 else
5443 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08005444}
5445
NeilBrown96de1e62005-11-08 21:39:39 -08005446static struct md_sysfs_entry
5447raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08005448
Shaohua Lib7214202013-08-27 17:50:42 +08005449static ssize_t
5450raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
5451{
5452 struct r5conf *conf = mddev->private;
5453 if (conf)
5454 return sprintf(page, "%d\n", conf->worker_cnt_per_group);
5455 else
5456 return 0;
5457}
5458
majianpeng60aaf932013-11-14 15:16:20 +11005459static int alloc_thread_groups(struct r5conf *conf, int cnt,
5460 int *group_cnt,
5461 int *worker_cnt_per_group,
5462 struct r5worker_group **worker_groups);
Shaohua Lib7214202013-08-27 17:50:42 +08005463static ssize_t
5464raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
5465{
5466 struct r5conf *conf = mddev->private;
5467 unsigned long new;
5468 int err;
majianpeng60aaf932013-11-14 15:16:20 +11005469 struct r5worker_group *new_groups, *old_groups;
5470 int group_cnt, worker_cnt_per_group;
Shaohua Lib7214202013-08-27 17:50:42 +08005471
5472 if (len >= PAGE_SIZE)
5473 return -EINVAL;
5474 if (!conf)
5475 return -ENODEV;
5476
5477 if (kstrtoul(page, 10, &new))
5478 return -EINVAL;
5479
5480 if (new == conf->worker_cnt_per_group)
5481 return len;
5482
5483 mddev_suspend(mddev);
5484
5485 old_groups = conf->worker_groups;
majianpengd206dcf2013-11-14 15:16:19 +11005486 if (old_groups)
5487 flush_workqueue(raid5_wq);
Shaohua Lib7214202013-08-27 17:50:42 +08005488
majianpeng60aaf932013-11-14 15:16:20 +11005489 err = alloc_thread_groups(conf, new,
5490 &group_cnt, &worker_cnt_per_group,
5491 &new_groups);
5492 if (!err) {
5493 spin_lock_irq(&conf->device_lock);
5494 conf->group_cnt = group_cnt;
5495 conf->worker_cnt_per_group = worker_cnt_per_group;
5496 conf->worker_groups = new_groups;
5497 spin_unlock_irq(&conf->device_lock);
5498
Shaohua Lib7214202013-08-27 17:50:42 +08005499 if (old_groups)
5500 kfree(old_groups[0].workers);
5501 kfree(old_groups);
5502 }
5503
5504 mddev_resume(mddev);
5505
5506 if (err)
5507 return err;
5508 return len;
5509}
5510
5511static struct md_sysfs_entry
5512raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
5513 raid5_show_group_thread_cnt,
5514 raid5_store_group_thread_cnt);
5515
NeilBrown007583c2005-11-08 21:39:30 -08005516static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08005517 &raid5_stripecache_size.attr,
5518 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07005519 &raid5_preread_bypass_threshold.attr,
Shaohua Lib7214202013-08-27 17:50:42 +08005520 &raid5_group_thread_cnt.attr,
Shaohua Lid592a992014-05-21 17:57:44 +08005521 &raid5_skip_copy.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08005522 NULL,
5523};
NeilBrown007583c2005-11-08 21:39:30 -08005524static struct attribute_group raid5_attrs_group = {
5525 .name = NULL,
5526 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08005527};
5528
majianpeng60aaf932013-11-14 15:16:20 +11005529static int alloc_thread_groups(struct r5conf *conf, int cnt,
5530 int *group_cnt,
5531 int *worker_cnt_per_group,
5532 struct r5worker_group **worker_groups)
Shaohua Li851c30c2013-08-28 14:30:16 +08005533{
Shaohua Li566c09c2013-11-14 15:16:17 +11005534 int i, j, k;
Shaohua Li851c30c2013-08-28 14:30:16 +08005535 ssize_t size;
5536 struct r5worker *workers;
5537
majianpeng60aaf932013-11-14 15:16:20 +11005538 *worker_cnt_per_group = cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +08005539 if (cnt == 0) {
majianpeng60aaf932013-11-14 15:16:20 +11005540 *group_cnt = 0;
5541 *worker_groups = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08005542 return 0;
5543 }
majianpeng60aaf932013-11-14 15:16:20 +11005544 *group_cnt = num_possible_nodes();
Shaohua Li851c30c2013-08-28 14:30:16 +08005545 size = sizeof(struct r5worker) * cnt;
majianpeng60aaf932013-11-14 15:16:20 +11005546 workers = kzalloc(size * *group_cnt, GFP_NOIO);
5547 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
5548 *group_cnt, GFP_NOIO);
5549 if (!*worker_groups || !workers) {
Shaohua Li851c30c2013-08-28 14:30:16 +08005550 kfree(workers);
majianpeng60aaf932013-11-14 15:16:20 +11005551 kfree(*worker_groups);
Shaohua Li851c30c2013-08-28 14:30:16 +08005552 return -ENOMEM;
5553 }
5554
majianpeng60aaf932013-11-14 15:16:20 +11005555 for (i = 0; i < *group_cnt; i++) {
Shaohua Li851c30c2013-08-28 14:30:16 +08005556 struct r5worker_group *group;
5557
NeilBrown0c775d52013-11-25 11:12:43 +11005558 group = &(*worker_groups)[i];
Shaohua Li851c30c2013-08-28 14:30:16 +08005559 INIT_LIST_HEAD(&group->handle_list);
5560 group->conf = conf;
5561 group->workers = workers + i * cnt;
5562
5563 for (j = 0; j < cnt; j++) {
Shaohua Li566c09c2013-11-14 15:16:17 +11005564 struct r5worker *worker = group->workers + j;
5565 worker->group = group;
5566 INIT_WORK(&worker->work, raid5_do_work);
5567
5568 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
5569 INIT_LIST_HEAD(worker->temp_inactive_list + k);
Shaohua Li851c30c2013-08-28 14:30:16 +08005570 }
5571 }
5572
5573 return 0;
5574}
5575
5576static void free_thread_groups(struct r5conf *conf)
5577{
5578 if (conf->worker_groups)
5579 kfree(conf->worker_groups[0].workers);
5580 kfree(conf->worker_groups);
5581 conf->worker_groups = NULL;
5582}
5583
Dan Williams80c3a6c2009-03-17 18:10:40 -07005584static sector_t
NeilBrownfd01b882011-10-11 16:47:53 +11005585raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07005586{
NeilBrownd1688a62011-10-11 16:49:52 +11005587 struct r5conf *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07005588
5589 if (!sectors)
5590 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005591 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11005592 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11005593 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07005594
Andre Noll9d8f0362009-06-18 08:45:01 +10005595 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Andre Noll664e7c42009-06-18 08:45:27 +10005596 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07005597 return sectors * (raid_disks - conf->max_degraded);
5598}
5599
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305600static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
5601{
5602 safe_put_page(percpu->spare_page);
5603 kfree(percpu->scribble);
5604 percpu->spare_page = NULL;
5605 percpu->scribble = NULL;
5606}
5607
5608static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
5609{
5610 if (conf->level == 6 && !percpu->spare_page)
5611 percpu->spare_page = alloc_page(GFP_KERNEL);
5612 if (!percpu->scribble)
5613 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
5614
5615 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
5616 free_scratch_buffer(conf, percpu);
5617 return -ENOMEM;
5618 }
5619
5620 return 0;
5621}
5622
NeilBrownd1688a62011-10-11 16:49:52 +11005623static void raid5_free_percpu(struct r5conf *conf)
Dan Williams36d1c642009-07-14 11:48:22 -07005624{
Dan Williams36d1c642009-07-14 11:48:22 -07005625 unsigned long cpu;
5626
5627 if (!conf->percpu)
5628 return;
5629
Dan Williams36d1c642009-07-14 11:48:22 -07005630#ifdef CONFIG_HOTPLUG_CPU
5631 unregister_cpu_notifier(&conf->cpu_notify);
5632#endif
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305633
5634 get_online_cpus();
5635 for_each_possible_cpu(cpu)
5636 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
Dan Williams36d1c642009-07-14 11:48:22 -07005637 put_online_cpus();
5638
5639 free_percpu(conf->percpu);
5640}
5641
NeilBrownd1688a62011-10-11 16:49:52 +11005642static void free_conf(struct r5conf *conf)
Dan Williams95fc17a2009-07-31 12:39:15 +10005643{
Shaohua Li851c30c2013-08-28 14:30:16 +08005644 free_thread_groups(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005645 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07005646 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005647 kfree(conf->disks);
5648 kfree(conf->stripe_hashtbl);
5649 kfree(conf);
5650}
5651
Dan Williams36d1c642009-07-14 11:48:22 -07005652#ifdef CONFIG_HOTPLUG_CPU
5653static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
5654 void *hcpu)
5655{
NeilBrownd1688a62011-10-11 16:49:52 +11005656 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
Dan Williams36d1c642009-07-14 11:48:22 -07005657 long cpu = (long)hcpu;
5658 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
5659
5660 switch (action) {
5661 case CPU_UP_PREPARE:
5662 case CPU_UP_PREPARE_FROZEN:
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305663 if (alloc_scratch_buffer(conf, percpu)) {
Dan Williams36d1c642009-07-14 11:48:22 -07005664 pr_err("%s: failed memory allocation for cpu%ld\n",
5665 __func__, cpu);
Akinobu Mita55af6bb2010-05-26 14:43:35 -07005666 return notifier_from_errno(-ENOMEM);
Dan Williams36d1c642009-07-14 11:48:22 -07005667 }
5668 break;
5669 case CPU_DEAD:
5670 case CPU_DEAD_FROZEN:
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305671 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
Dan Williams36d1c642009-07-14 11:48:22 -07005672 break;
5673 default:
5674 break;
5675 }
5676 return NOTIFY_OK;
5677}
5678#endif
5679
NeilBrownd1688a62011-10-11 16:49:52 +11005680static int raid5_alloc_percpu(struct r5conf *conf)
Dan Williams36d1c642009-07-14 11:48:22 -07005681{
5682 unsigned long cpu;
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305683 int err = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07005684
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305685 conf->percpu = alloc_percpu(struct raid5_percpu);
5686 if (!conf->percpu)
Dan Williams36d1c642009-07-14 11:48:22 -07005687 return -ENOMEM;
Dan Williams36d1c642009-07-14 11:48:22 -07005688
Dan Williams36d1c642009-07-14 11:48:22 -07005689#ifdef CONFIG_HOTPLUG_CPU
5690 conf->cpu_notify.notifier_call = raid456_cpu_notify;
5691 conf->cpu_notify.priority = 0;
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305692 err = register_cpu_notifier(&conf->cpu_notify);
5693 if (err)
5694 return err;
Dan Williams36d1c642009-07-14 11:48:22 -07005695#endif
Oleg Nesterov789b5e02014-02-06 03:42:45 +05305696
5697 get_online_cpus();
5698 for_each_present_cpu(cpu) {
5699 err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
5700 if (err) {
5701 pr_err("%s: failed memory allocation for cpu%ld\n",
5702 __func__, cpu);
5703 break;
5704 }
5705 }
Dan Williams36d1c642009-07-14 11:48:22 -07005706 put_online_cpus();
5707
5708 return err;
5709}
5710
NeilBrownd1688a62011-10-11 16:49:52 +11005711static struct r5conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005712{
NeilBrownd1688a62011-10-11 16:49:52 +11005713 struct r5conf *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005714 int raid_disk, memory, max_disks;
NeilBrown3cb03002011-10-11 16:45:26 +11005715 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005716 struct disk_info *disk;
NeilBrown02326052012-07-03 15:56:52 +10005717 char pers_name[6];
Shaohua Li566c09c2013-11-14 15:16:17 +11005718 int i;
majianpeng60aaf932013-11-14 15:16:20 +11005719 int group_cnt, worker_cnt_per_group;
5720 struct r5worker_group *new_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005721
NeilBrown91adb562009-03-31 14:39:39 +11005722 if (mddev->new_level != 5
5723 && mddev->new_level != 4
5724 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10005725 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11005726 mdname(mddev), mddev->new_level);
5727 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005728 }
NeilBrown91adb562009-03-31 14:39:39 +11005729 if ((mddev->new_level == 5
5730 && !algorithm_valid_raid5(mddev->new_layout)) ||
5731 (mddev->new_level == 6
5732 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10005733 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11005734 mdname(mddev), mddev->new_layout);
5735 return ERR_PTR(-EIO);
5736 }
5737 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10005738 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11005739 mdname(mddev), mddev->raid_disks);
5740 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11005741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005742
Andre Noll664e7c42009-06-18 08:45:27 +10005743 if (!mddev->new_chunk_sectors ||
5744 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
5745 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10005746 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
5747 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11005748 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11005749 }
5750
NeilBrownd1688a62011-10-11 16:49:52 +11005751 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
NeilBrown91adb562009-03-31 14:39:39 +11005752 if (conf == NULL)
5753 goto abort;
Shaohua Li851c30c2013-08-28 14:30:16 +08005754 /* Don't enable multi-threading by default*/
majianpeng60aaf932013-11-14 15:16:20 +11005755 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
5756 &new_group)) {
5757 conf->group_cnt = group_cnt;
5758 conf->worker_cnt_per_group = worker_cnt_per_group;
5759 conf->worker_groups = new_group;
5760 } else
Shaohua Li851c30c2013-08-28 14:30:16 +08005761 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11005762 spin_lock_init(&conf->device_lock);
NeilBrownc46501b2013-08-27 15:52:13 +10005763 seqcount_init(&conf->gen_lock);
Dan Williamsf5efd452009-10-16 15:55:38 +11005764 init_waitqueue_head(&conf->wait_for_stripe);
5765 init_waitqueue_head(&conf->wait_for_overlap);
5766 INIT_LIST_HEAD(&conf->handle_list);
5767 INIT_LIST_HEAD(&conf->hold_list);
5768 INIT_LIST_HEAD(&conf->delayed_list);
5769 INIT_LIST_HEAD(&conf->bitmap_list);
Shaohua Li773ca822013-08-27 17:50:39 +08005770 init_llist_head(&conf->released_stripes);
Dan Williamsf5efd452009-10-16 15:55:38 +11005771 atomic_set(&conf->active_stripes, 0);
5772 atomic_set(&conf->preread_active_stripes, 0);
5773 atomic_set(&conf->active_aligned_reads, 0);
5774 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrownd890fa22011-10-26 11:54:39 +11005775 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown91adb562009-03-31 14:39:39 +11005776
5777 conf->raid_disks = mddev->raid_disks;
5778 if (mddev->reshape_position == MaxSector)
5779 conf->previous_raid_disks = mddev->raid_disks;
5780 else
5781 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005782 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
5783 conf->scribble_len = scribble_len(max_disks);
NeilBrown91adb562009-03-31 14:39:39 +11005784
NeilBrown5e5e3e72009-10-16 16:35:30 +11005785 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11005786 GFP_KERNEL);
5787 if (!conf->disks)
5788 goto abort;
5789
5790 conf->mddev = mddev;
5791
5792 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
5793 goto abort;
5794
Shaohua Li566c09c2013-11-14 15:16:17 +11005795 /* We init hash_locks[0] separately to that it can be used
5796 * as the reference lock in the spin_lock_nest_lock() call
5797 * in lock_all_device_hash_locks_irq in order to convince
5798 * lockdep that we know what we are doing.
5799 */
5800 spin_lock_init(conf->hash_locks);
5801 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
5802 spin_lock_init(conf->hash_locks + i);
5803
5804 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5805 INIT_LIST_HEAD(conf->inactive_list + i);
5806
5807 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5808 INIT_LIST_HEAD(conf->temp_inactive_list + i);
5809
Dan Williams36d1c642009-07-14 11:48:22 -07005810 conf->level = mddev->new_level;
5811 if (raid5_alloc_percpu(conf) != 0)
5812 goto abort;
5813
NeilBrown0c55e022010-05-03 14:09:02 +10005814 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11005815
NeilBrowndafb20f2012-03-19 12:46:39 +11005816 rdev_for_each(rdev, mddev) {
NeilBrown91adb562009-03-31 14:39:39 +11005817 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11005818 if (raid_disk >= max_disks
NeilBrown91adb562009-03-31 14:39:39 +11005819 || raid_disk < 0)
5820 continue;
5821 disk = conf->disks + raid_disk;
5822
NeilBrown17045f52011-12-23 10:17:53 +11005823 if (test_bit(Replacement, &rdev->flags)) {
5824 if (disk->replacement)
5825 goto abort;
5826 disk->replacement = rdev;
5827 } else {
5828 if (disk->rdev)
5829 goto abort;
5830 disk->rdev = rdev;
5831 }
NeilBrown91adb562009-03-31 14:39:39 +11005832
5833 if (test_bit(In_sync, &rdev->flags)) {
5834 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10005835 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
5836 " disk %d\n",
5837 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
Jonathan Brassowd6b212f2011-06-08 18:00:28 -05005838 } else if (rdev->saved_raid_disk != raid_disk)
NeilBrown91adb562009-03-31 14:39:39 +11005839 /* Cannot rely on bitmap to complete recovery */
5840 conf->fullsync = 1;
5841 }
5842
Andre Noll09c9e5f2009-06-18 08:45:55 +10005843 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown91adb562009-03-31 14:39:39 +11005844 conf->level = mddev->new_level;
5845 if (conf->level == 6)
5846 conf->max_degraded = 2;
5847 else
5848 conf->max_degraded = 1;
5849 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11005850 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11005851 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10005852 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11005853 conf->prev_algo = mddev->layout;
5854 }
NeilBrown91adb562009-03-31 14:39:39 +11005855
5856 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11005857 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
Shaohua Li4bda5562013-11-14 15:16:17 +11005858 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
Shaohua Li566c09c2013-11-14 15:16:17 +11005859 if (grow_stripes(conf, NR_STRIPES)) {
NeilBrown91adb562009-03-31 14:39:39 +11005860 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005861 "md/raid:%s: couldn't allocate %dkB for buffers\n",
5862 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11005863 goto abort;
5864 } else
NeilBrown0c55e022010-05-03 14:09:02 +10005865 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
5866 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11005867
NeilBrown02326052012-07-03 15:56:52 +10005868 sprintf(pers_name, "raid%d", mddev->new_level);
5869 conf->thread = md_register_thread(raid5d, mddev, pers_name);
NeilBrown91adb562009-03-31 14:39:39 +11005870 if (!conf->thread) {
5871 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005872 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11005873 mdname(mddev));
5874 goto abort;
5875 }
5876
5877 return conf;
5878
5879 abort:
5880 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10005881 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11005882 return ERR_PTR(-EIO);
5883 } else
5884 return ERR_PTR(-ENOMEM);
5885}
5886
NeilBrownc148ffd2009-11-13 17:47:00 +11005887static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
5888{
5889 switch (algo) {
5890 case ALGORITHM_PARITY_0:
5891 if (raid_disk < max_degraded)
5892 return 1;
5893 break;
5894 case ALGORITHM_PARITY_N:
5895 if (raid_disk >= raid_disks - max_degraded)
5896 return 1;
5897 break;
5898 case ALGORITHM_PARITY_0_6:
NeilBrownf72ffdd2014-09-30 14:23:59 +10005899 if (raid_disk == 0 ||
NeilBrownc148ffd2009-11-13 17:47:00 +11005900 raid_disk == raid_disks - 1)
5901 return 1;
5902 break;
5903 case ALGORITHM_LEFT_ASYMMETRIC_6:
5904 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5905 case ALGORITHM_LEFT_SYMMETRIC_6:
5906 case ALGORITHM_RIGHT_SYMMETRIC_6:
5907 if (raid_disk == raid_disks - 1)
5908 return 1;
5909 }
5910 return 0;
5911}
5912
NeilBrownfd01b882011-10-11 16:47:53 +11005913static int run(struct mddev *mddev)
NeilBrown91adb562009-03-31 14:39:39 +11005914{
NeilBrownd1688a62011-10-11 16:49:52 +11005915 struct r5conf *conf;
NeilBrown9f7c2222010-07-26 12:04:13 +10005916 int working_disks = 0;
NeilBrownc148ffd2009-11-13 17:47:00 +11005917 int dirty_parity_disks = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11005918 struct md_rdev *rdev;
NeilBrownc148ffd2009-11-13 17:47:00 +11005919 sector_t reshape_offset = 0;
NeilBrown17045f52011-12-23 10:17:53 +11005920 int i;
NeilBrownb5254dd2012-05-21 09:27:01 +10005921 long long min_offset_diff = 0;
5922 int first = 1;
NeilBrown91adb562009-03-31 14:39:39 +11005923
Andre Noll8c6ac862009-06-18 08:48:06 +10005924 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10005925 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10005926 " -- starting background reconstruction\n",
5927 mdname(mddev));
NeilBrownb5254dd2012-05-21 09:27:01 +10005928
5929 rdev_for_each(rdev, mddev) {
5930 long long diff;
5931 if (rdev->raid_disk < 0)
5932 continue;
5933 diff = (rdev->new_data_offset - rdev->data_offset);
5934 if (first) {
5935 min_offset_diff = diff;
5936 first = 0;
5937 } else if (mddev->reshape_backwards &&
5938 diff < min_offset_diff)
5939 min_offset_diff = diff;
5940 else if (!mddev->reshape_backwards &&
5941 diff > min_offset_diff)
5942 min_offset_diff = diff;
5943 }
5944
NeilBrownf6705572006-03-27 01:18:11 -08005945 if (mddev->reshape_position != MaxSector) {
5946 /* Check that we can continue the reshape.
NeilBrownb5254dd2012-05-21 09:27:01 +10005947 * Difficulties arise if the stripe we would write to
5948 * next is at or after the stripe we would read from next.
5949 * For a reshape that changes the number of devices, this
5950 * is only possible for a very short time, and mdadm makes
5951 * sure that time appears to have past before assembling
5952 * the array. So we fail if that time hasn't passed.
5953 * For a reshape that keeps the number of devices the same
5954 * mdadm must be monitoring the reshape can keeping the
5955 * critical areas read-only and backed up. It will start
5956 * the array in read-only mode, so we check for that.
NeilBrownf6705572006-03-27 01:18:11 -08005957 */
5958 sector_t here_new, here_old;
5959 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11005960 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrownf6705572006-03-27 01:18:11 -08005961
NeilBrown88ce4932009-03-31 15:24:23 +11005962 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10005963 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08005964 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08005965 mdname(mddev));
5966 return -EINVAL;
5967 }
NeilBrownf6705572006-03-27 01:18:11 -08005968 old_disks = mddev->raid_disks - mddev->delta_disks;
5969 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08005970 * further up in new geometry must map after here in old
5971 * geometry.
NeilBrownf6705572006-03-27 01:18:11 -08005972 */
5973 here_new = mddev->reshape_position;
Andre Noll664e7c42009-06-18 08:45:27 +10005974 if (sector_div(here_new, mddev->new_chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08005975 (mddev->raid_disks - max_degraded))) {
NeilBrown0c55e022010-05-03 14:09:02 +10005976 printk(KERN_ERR "md/raid:%s: reshape_position not "
5977 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005978 return -EINVAL;
5979 }
NeilBrownc148ffd2009-11-13 17:47:00 +11005980 reshape_offset = here_new * mddev->new_chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08005981 /* here_new is the stripe we will write to */
5982 here_old = mddev->reshape_position;
Andre Noll9d8f0362009-06-18 08:45:01 +10005983 sector_div(here_old, mddev->chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08005984 (old_disks-max_degraded));
5985 /* here_old is the first stripe that we might need to read
5986 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10005987 if (mddev->delta_disks == 0) {
NeilBrownb5254dd2012-05-21 09:27:01 +10005988 if ((here_new * mddev->new_chunk_sectors !=
5989 here_old * mddev->chunk_sectors)) {
5990 printk(KERN_ERR "md/raid:%s: reshape position is"
5991 " confused - aborting\n", mdname(mddev));
5992 return -EINVAL;
5993 }
NeilBrown67ac6012009-08-13 10:06:24 +10005994 /* We cannot be sure it is safe to start an in-place
NeilBrownb5254dd2012-05-21 09:27:01 +10005995 * reshape. It is only safe if user-space is monitoring
NeilBrown67ac6012009-08-13 10:06:24 +10005996 * and taking constant backups.
5997 * mdadm always starts a situation like this in
5998 * readonly mode so it can take control before
5999 * allowing any writes. So just check for that.
6000 */
NeilBrownb5254dd2012-05-21 09:27:01 +10006001 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
6002 abs(min_offset_diff) >= mddev->new_chunk_sectors)
6003 /* not really in-place - so OK */;
6004 else if (mddev->ro == 0) {
6005 printk(KERN_ERR "md/raid:%s: in-place reshape "
6006 "must be started in read-only mode "
6007 "- aborting\n",
NeilBrown0c55e022010-05-03 14:09:02 +10006008 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10006009 return -EINVAL;
6010 }
NeilBrown2c810cd2012-05-21 09:27:00 +10006011 } else if (mddev->reshape_backwards
NeilBrownb5254dd2012-05-21 09:27:01 +10006012 ? (here_new * mddev->new_chunk_sectors + min_offset_diff <=
NeilBrown67ac6012009-08-13 10:06:24 +10006013 here_old * mddev->chunk_sectors)
6014 : (here_new * mddev->new_chunk_sectors >=
NeilBrownb5254dd2012-05-21 09:27:01 +10006015 here_old * mddev->chunk_sectors + (-min_offset_diff))) {
NeilBrownf6705572006-03-27 01:18:11 -08006016 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10006017 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
6018 "auto-recovery - aborting.\n",
6019 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08006020 return -EINVAL;
6021 }
NeilBrown0c55e022010-05-03 14:09:02 +10006022 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
6023 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08006024 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08006025 } else {
NeilBrown91adb562009-03-31 14:39:39 +11006026 BUG_ON(mddev->level != mddev->new_level);
6027 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10006028 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11006029 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08006030 }
6031
NeilBrown245f46c2009-03-31 14:39:39 +11006032 if (mddev->private == NULL)
6033 conf = setup_conf(mddev);
6034 else
6035 conf = mddev->private;
6036
NeilBrown91adb562009-03-31 14:39:39 +11006037 if (IS_ERR(conf))
6038 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08006039
NeilBrownb5254dd2012-05-21 09:27:01 +10006040 conf->min_offset_diff = min_offset_diff;
NeilBrown91adb562009-03-31 14:39:39 +11006041 mddev->thread = conf->thread;
6042 conf->thread = NULL;
6043 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006044
NeilBrown17045f52011-12-23 10:17:53 +11006045 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
6046 i++) {
6047 rdev = conf->disks[i].rdev;
6048 if (!rdev && conf->disks[i].replacement) {
6049 /* The replacement is all we have yet */
6050 rdev = conf->disks[i].replacement;
6051 conf->disks[i].replacement = NULL;
6052 clear_bit(Replacement, &rdev->flags);
6053 conf->disks[i].rdev = rdev;
6054 }
6055 if (!rdev)
NeilBrownc148ffd2009-11-13 17:47:00 +11006056 continue;
NeilBrown17045f52011-12-23 10:17:53 +11006057 if (conf->disks[i].replacement &&
6058 conf->reshape_progress != MaxSector) {
6059 /* replacements and reshape simply do not mix. */
6060 printk(KERN_ERR "md: cannot handle concurrent "
6061 "replacement and reshape.\n");
6062 goto abort;
6063 }
NeilBrown2f115882010-06-17 17:41:03 +10006064 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown91adb562009-03-31 14:39:39 +11006065 working_disks++;
NeilBrown2f115882010-06-17 17:41:03 +10006066 continue;
6067 }
NeilBrownc148ffd2009-11-13 17:47:00 +11006068 /* This disc is not fully in-sync. However if it
6069 * just stored parity (beyond the recovery_offset),
6070 * when we don't need to be concerned about the
6071 * array being dirty.
6072 * When reshape goes 'backwards', we never have
6073 * partially completed devices, so we only need
6074 * to worry about reshape going forwards.
6075 */
6076 /* Hack because v0.91 doesn't store recovery_offset properly. */
6077 if (mddev->major_version == 0 &&
6078 mddev->minor_version > 90)
6079 rdev->recovery_offset = reshape_offset;
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07006080
NeilBrownc148ffd2009-11-13 17:47:00 +11006081 if (rdev->recovery_offset < reshape_offset) {
6082 /* We need to check old and new layout */
6083 if (!only_parity(rdev->raid_disk,
6084 conf->algorithm,
6085 conf->raid_disks,
6086 conf->max_degraded))
6087 continue;
6088 }
6089 if (!only_parity(rdev->raid_disk,
6090 conf->prev_algo,
6091 conf->previous_raid_disks,
6092 conf->max_degraded))
6093 continue;
6094 dirty_parity_disks++;
6095 }
NeilBrown91adb562009-03-31 14:39:39 +11006096
NeilBrown17045f52011-12-23 10:17:53 +11006097 /*
6098 * 0 for a fully functional array, 1 or 2 for a degraded array.
6099 */
NeilBrown908f4fb2011-12-23 10:17:50 +11006100 mddev->degraded = calc_degraded(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006101
NeilBrown674806d2010-06-16 17:17:53 +10006102 if (has_failed(conf)) {
NeilBrown0c55e022010-05-03 14:09:02 +10006103 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07006104 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07006105 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006106 goto abort;
6107 }
6108
NeilBrown91adb562009-03-31 14:39:39 +11006109 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10006110 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11006111 mddev->resync_max_sectors = mddev->dev_sectors;
6112
NeilBrownc148ffd2009-11-13 17:47:00 +11006113 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07006114 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006115 if (mddev->ok_start_degraded)
6116 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10006117 "md/raid:%s: starting dirty degraded array"
6118 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006119 mdname(mddev));
6120 else {
6121 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10006122 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006123 mdname(mddev));
6124 goto abort;
6125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006126 }
6127
Linus Torvalds1da177e2005-04-16 15:20:36 -07006128 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10006129 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
6130 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11006131 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
6132 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006133 else
NeilBrown0c55e022010-05-03 14:09:02 +10006134 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
6135 " out of %d devices, algorithm %d\n",
6136 mdname(mddev), conf->level,
6137 mddev->raid_disks - mddev->degraded,
6138 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006139
6140 print_raid5_conf(conf);
6141
NeilBrownfef9c612009-03-31 15:16:46 +11006142 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11006143 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08006144 atomic_set(&conf->reshape_stripes, 0);
6145 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6146 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6147 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6148 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6149 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10006150 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08006151 }
6152
Linus Torvalds1da177e2005-04-16 15:20:36 -07006153 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10006154 if (mddev->to_remove == &raid5_attrs_group)
6155 mddev->to_remove = NULL;
NeilBrown00bcb4a2010-06-01 19:37:23 +10006156 else if (mddev->kobj.sd &&
6157 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08006158 printk(KERN_WARNING
NeilBrown4a5add42010-06-01 19:37:28 +10006159 "raid5: failed to create sysfs attributes for %s\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08006160 mdname(mddev));
NeilBrown4a5add42010-06-01 19:37:28 +10006161 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6162
6163 if (mddev->queue) {
NeilBrown9f7c2222010-07-26 12:04:13 +10006164 int chunk_size;
Shaohua Li620125f2012-10-11 13:49:05 +11006165 bool discard_supported = true;
NeilBrown4a5add42010-06-01 19:37:28 +10006166 /* read-ahead size must cover two whole stripes, which
6167 * is 2 * (datadisks) * chunksize where 'n' is the
6168 * number of raid devices
6169 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006170 int data_disks = conf->previous_raid_disks - conf->max_degraded;
6171 int stripe = data_disks *
6172 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
6173 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6174 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
NeilBrown4a5add42010-06-01 19:37:28 +10006175
6176 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
NeilBrown11d8a6e2010-07-26 11:57:07 +10006177
6178 mddev->queue->backing_dev_info.congested_data = mddev;
6179 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
NeilBrown9f7c2222010-07-26 12:04:13 +10006180
6181 chunk_size = mddev->chunk_sectors << 9;
6182 blk_queue_io_min(mddev->queue, chunk_size);
6183 blk_queue_io_opt(mddev->queue, chunk_size *
6184 (conf->raid_disks - conf->max_degraded));
Kent Overstreetc78afc62013-07-11 22:39:53 -07006185 mddev->queue->limits.raid_partial_stripes_expensive = 1;
Shaohua Li620125f2012-10-11 13:49:05 +11006186 /*
6187 * We can only discard a whole stripe. It doesn't make sense to
6188 * discard data disk but write parity disk
6189 */
6190 stripe = stripe * PAGE_SIZE;
NeilBrown4ac68752012-11-19 13:11:26 +11006191 /* Round up to power of 2, as discard handling
6192 * currently assumes that */
6193 while ((stripe-1) & stripe)
6194 stripe = (stripe | (stripe-1)) + 1;
Shaohua Li620125f2012-10-11 13:49:05 +11006195 mddev->queue->limits.discard_alignment = stripe;
6196 mddev->queue->limits.discard_granularity = stripe;
6197 /*
6198 * unaligned part of discard request will be ignored, so can't
NeilBrown8e0e99b2014-10-02 13:45:00 +10006199 * guarantee discard_zeroes_data
Shaohua Li620125f2012-10-11 13:49:05 +11006200 */
6201 mddev->queue->limits.discard_zeroes_data = 0;
NeilBrown9f7c2222010-07-26 12:04:13 +10006202
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07006203 blk_queue_max_write_same_sectors(mddev->queue, 0);
6204
NeilBrown05616be2012-05-21 09:27:00 +10006205 rdev_for_each(rdev, mddev) {
NeilBrown9f7c2222010-07-26 12:04:13 +10006206 disk_stack_limits(mddev->gendisk, rdev->bdev,
6207 rdev->data_offset << 9);
NeilBrown05616be2012-05-21 09:27:00 +10006208 disk_stack_limits(mddev->gendisk, rdev->bdev,
6209 rdev->new_data_offset << 9);
Shaohua Li620125f2012-10-11 13:49:05 +11006210 /*
6211 * discard_zeroes_data is required, otherwise data
6212 * could be lost. Consider a scenario: discard a stripe
6213 * (the stripe could be inconsistent if
6214 * discard_zeroes_data is 0); write one disk of the
6215 * stripe (the stripe could be inconsistent again
6216 * depending on which disks are used to calculate
6217 * parity); the disk is broken; The stripe data of this
6218 * disk is lost.
6219 */
6220 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
6221 !bdev_get_queue(rdev->bdev)->
6222 limits.discard_zeroes_data)
6223 discard_supported = false;
NeilBrown8e0e99b2014-10-02 13:45:00 +10006224 /* Unfortunately, discard_zeroes_data is not currently
6225 * a guarantee - just a hint. So we only allow DISCARD
6226 * if the sysadmin has confirmed that only safe devices
6227 * are in use by setting a module parameter.
6228 */
6229 if (!devices_handle_discard_safely) {
6230 if (discard_supported) {
6231 pr_info("md/raid456: discard support disabled due to uncertainty.\n");
6232 pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n");
6233 }
6234 discard_supported = false;
6235 }
NeilBrown05616be2012-05-21 09:27:00 +10006236 }
Shaohua Li620125f2012-10-11 13:49:05 +11006237
6238 if (discard_supported &&
6239 mddev->queue->limits.max_discard_sectors >= stripe &&
6240 mddev->queue->limits.discard_granularity >= stripe)
6241 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
6242 mddev->queue);
6243 else
6244 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
6245 mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006246 }
6247
Linus Torvalds1da177e2005-04-16 15:20:36 -07006248 return 0;
6249abort:
NeilBrown01f96c02011-09-21 15:30:20 +10006250 md_unregister_thread(&mddev->thread);
NeilBrowne4f869d2011-10-07 14:22:49 +11006251 print_raid5_conf(conf);
6252 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006253 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10006254 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006255 return -EIO;
6256}
6257
NeilBrownfd01b882011-10-11 16:47:53 +11006258static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006259{
NeilBrownd1688a62011-10-11 16:49:52 +11006260 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006261
NeilBrown01f96c02011-09-21 15:30:20 +10006262 md_unregister_thread(&mddev->thread);
NeilBrown11d8a6e2010-07-26 11:57:07 +10006263 if (mddev->queue)
6264 mddev->queue->backing_dev_info.congested_fn = NULL;
Dan Williams95fc17a2009-07-31 12:39:15 +10006265 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10006266 mddev->private = NULL;
6267 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006268 return 0;
6269}
6270
NeilBrownfd01b882011-10-11 16:47:53 +11006271static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006272{
NeilBrownd1688a62011-10-11 16:49:52 +11006273 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006274 int i;
6275
Andre Noll9d8f0362009-06-18 08:45:01 +10006276 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
6277 mddev->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07006278 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006279 for (i = 0; i < conf->raid_disks; i++)
6280 seq_printf (seq, "%s",
6281 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08006282 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006283 seq_printf (seq, "]");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006284}
6285
NeilBrownd1688a62011-10-11 16:49:52 +11006286static void print_raid5_conf (struct r5conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006287{
6288 int i;
6289 struct disk_info *tmp;
6290
NeilBrown0c55e022010-05-03 14:09:02 +10006291 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006292 if (!conf) {
6293 printk("(conf==NULL)\n");
6294 return;
6295 }
NeilBrown0c55e022010-05-03 14:09:02 +10006296 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
6297 conf->raid_disks,
6298 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006299
6300 for (i = 0; i < conf->raid_disks; i++) {
6301 char b[BDEVNAME_SIZE];
6302 tmp = conf->disks + i;
6303 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10006304 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
6305 i, !test_bit(Faulty, &tmp->rdev->flags),
6306 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006307 }
6308}
6309
NeilBrownfd01b882011-10-11 16:47:53 +11006310static int raid5_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006311{
6312 int i;
NeilBrownd1688a62011-10-11 16:49:52 +11006313 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006314 struct disk_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10006315 int count = 0;
6316 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006317
6318 for (i = 0; i < conf->raid_disks; i++) {
6319 tmp = conf->disks + i;
NeilBrowndd054fc2011-12-23 10:17:53 +11006320 if (tmp->replacement
6321 && tmp->replacement->recovery_offset == MaxSector
6322 && !test_bit(Faulty, &tmp->replacement->flags)
6323 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
6324 /* Replacement has just become active. */
6325 if (!tmp->rdev
6326 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
6327 count++;
6328 if (tmp->rdev) {
6329 /* Replaced device not technically faulty,
6330 * but we need to be sure it gets removed
6331 * and never re-added.
6332 */
6333 set_bit(Faulty, &tmp->rdev->flags);
6334 sysfs_notify_dirent_safe(
6335 tmp->rdev->sysfs_state);
6336 }
6337 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
6338 } else if (tmp->rdev
NeilBrown70fffd02010-06-16 17:01:25 +10006339 && tmp->rdev->recovery_offset == MaxSector
NeilBrownb2d444d2005-11-08 21:39:31 -08006340 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07006341 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10006342 count++;
Jonathan Brassow43c73ca2011-01-14 09:14:33 +11006343 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006344 }
6345 }
NeilBrown6b965622010-08-18 11:56:59 +10006346 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown908f4fb2011-12-23 10:17:50 +11006347 mddev->degraded = calc_degraded(conf);
NeilBrown6b965622010-08-18 11:56:59 +10006348 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006349 print_raid5_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10006350 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006351}
6352
NeilBrownb8321b62011-12-23 10:17:51 +11006353static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006354{
NeilBrownd1688a62011-10-11 16:49:52 +11006355 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006356 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11006357 int number = rdev->raid_disk;
NeilBrown657e3e42011-12-23 10:17:52 +11006358 struct md_rdev **rdevp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006359 struct disk_info *p = conf->disks + number;
6360
6361 print_raid5_conf(conf);
NeilBrown657e3e42011-12-23 10:17:52 +11006362 if (rdev == p->rdev)
6363 rdevp = &p->rdev;
6364 else if (rdev == p->replacement)
6365 rdevp = &p->replacement;
6366 else
6367 return 0;
NeilBrownec32a2b2009-03-31 15:17:38 +11006368
NeilBrown657e3e42011-12-23 10:17:52 +11006369 if (number >= conf->raid_disks &&
6370 conf->reshape_progress == MaxSector)
6371 clear_bit(In_sync, &rdev->flags);
6372
6373 if (test_bit(In_sync, &rdev->flags) ||
6374 atomic_read(&rdev->nr_pending)) {
6375 err = -EBUSY;
6376 goto abort;
6377 }
6378 /* Only remove non-faulty devices if recovery
6379 * isn't possible.
6380 */
6381 if (!test_bit(Faulty, &rdev->flags) &&
6382 mddev->recovery_disabled != conf->recovery_disabled &&
6383 !has_failed(conf) &&
NeilBrowndd054fc2011-12-23 10:17:53 +11006384 (!p->replacement || p->replacement == rdev) &&
NeilBrown657e3e42011-12-23 10:17:52 +11006385 number < conf->raid_disks) {
6386 err = -EBUSY;
6387 goto abort;
6388 }
6389 *rdevp = NULL;
6390 synchronize_rcu();
6391 if (atomic_read(&rdev->nr_pending)) {
6392 /* lost the race, try later */
6393 err = -EBUSY;
6394 *rdevp = rdev;
NeilBrowndd054fc2011-12-23 10:17:53 +11006395 } else if (p->replacement) {
6396 /* We must have just cleared 'rdev' */
6397 p->rdev = p->replacement;
6398 clear_bit(Replacement, &p->replacement->flags);
6399 smp_mb(); /* Make sure other CPUs may see both as identical
6400 * but will never see neither - if they are careful
6401 */
6402 p->replacement = NULL;
6403 clear_bit(WantReplacement, &rdev->flags);
6404 } else
6405 /* We might have just removed the Replacement as faulty-
6406 * clear the bit just in case
6407 */
6408 clear_bit(WantReplacement, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006409abort:
6410
6411 print_raid5_conf(conf);
6412 return err;
6413}
6414
NeilBrownfd01b882011-10-11 16:47:53 +11006415static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006416{
NeilBrownd1688a62011-10-11 16:49:52 +11006417 struct r5conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10006418 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006419 int disk;
6420 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10006421 int first = 0;
6422 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006423
NeilBrown7f0da592011-07-28 11:39:22 +10006424 if (mddev->recovery_disabled == conf->recovery_disabled)
6425 return -EBUSY;
6426
NeilBrowndc10c642012-03-19 12:46:37 +11006427 if (rdev->saved_raid_disk < 0 && has_failed(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006428 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10006429 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006430
Neil Brown6c2fce22008-06-28 08:31:31 +10006431 if (rdev->raid_disk >= 0)
6432 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006433
6434 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07006435 * find the disk ... but prefer rdev->saved_raid_disk
6436 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006437 */
NeilBrown16a53ec2006-06-26 00:27:38 -07006438 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10006439 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07006440 conf->disks[rdev->saved_raid_disk].rdev == NULL)
NeilBrown5cfb22a2012-07-03 11:46:53 +10006441 first = rdev->saved_raid_disk;
6442
6443 for (disk = first; disk <= last; disk++) {
NeilBrown7bfec5f2011-12-23 10:17:53 +11006444 p = conf->disks + disk;
6445 if (p->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08006446 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006447 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10006448 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07006449 if (rdev->saved_raid_disk != disk)
6450 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08006451 rcu_assign_pointer(p->rdev, rdev);
NeilBrown5cfb22a2012-07-03 11:46:53 +10006452 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006453 }
NeilBrown5cfb22a2012-07-03 11:46:53 +10006454 }
6455 for (disk = first; disk <= last; disk++) {
6456 p = conf->disks + disk;
NeilBrown7bfec5f2011-12-23 10:17:53 +11006457 if (test_bit(WantReplacement, &p->rdev->flags) &&
6458 p->replacement == NULL) {
6459 clear_bit(In_sync, &rdev->flags);
6460 set_bit(Replacement, &rdev->flags);
6461 rdev->raid_disk = disk;
6462 err = 0;
6463 conf->fullsync = 1;
6464 rcu_assign_pointer(p->replacement, rdev);
6465 break;
6466 }
6467 }
NeilBrown5cfb22a2012-07-03 11:46:53 +10006468out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006469 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10006470 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006471}
6472
NeilBrownfd01b882011-10-11 16:47:53 +11006473static int raid5_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006474{
6475 /* no resync is happening, and there is enough space
6476 * on all devices, so we can resize.
6477 * We need to make sure resync covers any new space.
6478 * If the array is shrinking we should possibly wait until
6479 * any io in the removed space completes, but it hardly seems
6480 * worth it.
6481 */
NeilBrowna4a61252012-05-22 13:55:27 +10006482 sector_t newsize;
Andre Noll9d8f0362009-06-18 08:45:01 +10006483 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
NeilBrowna4a61252012-05-22 13:55:27 +10006484 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
6485 if (mddev->external_size &&
6486 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11006487 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10006488 if (mddev->bitmap) {
6489 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
6490 if (ret)
6491 return ret;
6492 }
6493 md_set_array_sectors(mddev, newsize);
Andre Nollf233ea52008-07-21 17:05:22 +10006494 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10006495 revalidate_disk(mddev->gendisk);
NeilBrownb0986362011-05-11 15:52:21 +10006496 if (sectors > mddev->dev_sectors &&
6497 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11006498 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006499 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6500 }
Andre Noll58c0fed2009-03-31 14:33:13 +11006501 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07006502 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006503 return 0;
6504}
6505
NeilBrownfd01b882011-10-11 16:47:53 +11006506static int check_stripe_cache(struct mddev *mddev)
NeilBrown01ee22b2009-06-18 08:47:20 +10006507{
6508 /* Can only proceed if there are plenty of stripe_heads.
6509 * We need a minimum of one full stripe,, and for sensible progress
6510 * it is best to have about 4 times that.
6511 * If we require 4 times, then the default 256 4K stripe_heads will
6512 * allow for chunk sizes up to 256K, which is probably OK.
6513 * If the chunk size is greater, user-space should request more
6514 * stripe_heads first.
6515 */
NeilBrownd1688a62011-10-11 16:49:52 +11006516 struct r5conf *conf = mddev->private;
NeilBrown01ee22b2009-06-18 08:47:20 +10006517 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
6518 > conf->max_nr_stripes ||
6519 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
6520 > conf->max_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10006521 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
6522 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10006523 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
6524 / STRIPE_SIZE)*4);
6525 return 0;
6526 }
6527 return 1;
6528}
6529
NeilBrownfd01b882011-10-11 16:47:53 +11006530static int check_reshape(struct mddev *mddev)
NeilBrown29269552006-03-27 01:18:10 -08006531{
NeilBrownd1688a62011-10-11 16:49:52 +11006532 struct r5conf *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08006533
NeilBrown88ce4932009-03-31 15:24:23 +11006534 if (mddev->delta_disks == 0 &&
6535 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10006536 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10006537 return 0; /* nothing to do */
NeilBrown674806d2010-06-16 17:17:53 +10006538 if (has_failed(conf))
NeilBrownec32a2b2009-03-31 15:17:38 +11006539 return -EINVAL;
NeilBrownfdcfbbb2013-07-04 16:38:16 +10006540 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
NeilBrownec32a2b2009-03-31 15:17:38 +11006541 /* We might be able to shrink, but the devices must
6542 * be made bigger first.
6543 * For raid6, 4 is the minimum size.
6544 * Otherwise 2 is the minimum
6545 */
6546 int min = 2;
6547 if (mddev->level == 6)
6548 min = 4;
6549 if (mddev->raid_disks + mddev->delta_disks < min)
6550 return -EINVAL;
6551 }
NeilBrown29269552006-03-27 01:18:10 -08006552
NeilBrown01ee22b2009-06-18 08:47:20 +10006553 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08006554 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08006555
NeilBrowne56108d62012-10-11 14:24:13 +11006556 return resize_stripes(conf, (conf->previous_raid_disks
6557 + mddev->delta_disks));
NeilBrown63c70c42006-03-27 01:18:13 -08006558}
6559
NeilBrownfd01b882011-10-11 16:47:53 +11006560static int raid5_start_reshape(struct mddev *mddev)
NeilBrown63c70c42006-03-27 01:18:13 -08006561{
NeilBrownd1688a62011-10-11 16:49:52 +11006562 struct r5conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11006563 struct md_rdev *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08006564 int spares = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07006565 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08006566
NeilBrownf4168852007-02-28 20:11:53 -08006567 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08006568 return -EBUSY;
6569
NeilBrown01ee22b2009-06-18 08:47:20 +10006570 if (!check_stripe_cache(mddev))
6571 return -ENOSPC;
6572
NeilBrown30b67642012-05-22 13:55:28 +10006573 if (has_failed(conf))
6574 return -EINVAL;
6575
NeilBrownc6563a82012-05-21 09:27:00 +10006576 rdev_for_each(rdev, mddev) {
NeilBrown469518a2011-01-31 11:57:43 +11006577 if (!test_bit(In_sync, &rdev->flags)
6578 && !test_bit(Faulty, &rdev->flags))
NeilBrown29269552006-03-27 01:18:10 -08006579 spares++;
NeilBrownc6563a82012-05-21 09:27:00 +10006580 }
NeilBrown63c70c42006-03-27 01:18:13 -08006581
NeilBrownf4168852007-02-28 20:11:53 -08006582 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08006583 /* Not enough devices even to make a degraded array
6584 * of that size
6585 */
6586 return -EINVAL;
6587
NeilBrownec32a2b2009-03-31 15:17:38 +11006588 /* Refuse to reduce size of the array. Any reductions in
6589 * array size must be through explicit setting of array_size
6590 * attribute.
6591 */
6592 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
6593 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10006594 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11006595 "before number of disks\n", mdname(mddev));
6596 return -EINVAL;
6597 }
6598
NeilBrownf6705572006-03-27 01:18:11 -08006599 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08006600 spin_lock_irq(&conf->device_lock);
NeilBrownc46501b2013-08-27 15:52:13 +10006601 write_seqcount_begin(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006602 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08006603 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10006604 conf->prev_chunk_sectors = conf->chunk_sectors;
6605 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11006606 conf->prev_algo = conf->algorithm;
6607 conf->algorithm = mddev->new_layout;
NeilBrown05616be2012-05-21 09:27:00 +10006608 conf->generation++;
6609 /* Code that selects data_offset needs to see the generation update
6610 * if reshape_progress has been set - so a memory barrier needed.
6611 */
6612 smp_mb();
NeilBrown2c810cd2012-05-21 09:27:00 +10006613 if (mddev->reshape_backwards)
NeilBrownfef9c612009-03-31 15:16:46 +11006614 conf->reshape_progress = raid5_size(mddev, 0, 0);
6615 else
6616 conf->reshape_progress = 0;
6617 conf->reshape_safe = conf->reshape_progress;
NeilBrownc46501b2013-08-27 15:52:13 +10006618 write_seqcount_end(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006619 spin_unlock_irq(&conf->device_lock);
6620
NeilBrown4d77e3b2013-08-27 15:57:47 +10006621 /* Now make sure any requests that proceeded on the assumption
6622 * the reshape wasn't running - like Discard or Read - have
6623 * completed.
6624 */
6625 mddev_suspend(mddev);
6626 mddev_resume(mddev);
6627
NeilBrown29269552006-03-27 01:18:10 -08006628 /* Add some new drives, as many as will fit.
6629 * We know there are enough to make the newly sized array work.
NeilBrown3424bf62010-06-17 17:48:26 +10006630 * Don't add devices if we are reducing the number of
6631 * devices in the array. This is because it is not possible
6632 * to correctly record the "partially reconstructed" state of
6633 * such devices during the reshape and confusion could result.
NeilBrown29269552006-03-27 01:18:10 -08006634 */
NeilBrown87a8dec2011-01-31 11:57:43 +11006635 if (mddev->delta_disks >= 0) {
NeilBrowndafb20f2012-03-19 12:46:39 +11006636 rdev_for_each(rdev, mddev)
NeilBrown87a8dec2011-01-31 11:57:43 +11006637 if (rdev->raid_disk < 0 &&
6638 !test_bit(Faulty, &rdev->flags)) {
6639 if (raid5_add_disk(mddev, rdev) == 0) {
NeilBrown87a8dec2011-01-31 11:57:43 +11006640 if (rdev->raid_disk
NeilBrown9d4c7d82012-03-13 11:21:21 +11006641 >= conf->previous_raid_disks)
NeilBrown87a8dec2011-01-31 11:57:43 +11006642 set_bit(In_sync, &rdev->flags);
NeilBrown9d4c7d82012-03-13 11:21:21 +11006643 else
NeilBrown87a8dec2011-01-31 11:57:43 +11006644 rdev->recovery_offset = 0;
Namhyung Kim36fad852011-07-27 11:00:36 +10006645
6646 if (sysfs_link_rdev(mddev, rdev))
NeilBrown87a8dec2011-01-31 11:57:43 +11006647 /* Failure here is OK */;
NeilBrown50da0842011-01-31 11:57:43 +11006648 }
NeilBrown87a8dec2011-01-31 11:57:43 +11006649 } else if (rdev->raid_disk >= conf->previous_raid_disks
6650 && !test_bit(Faulty, &rdev->flags)) {
6651 /* This is a spare that was manually added */
6652 set_bit(In_sync, &rdev->flags);
NeilBrown87a8dec2011-01-31 11:57:43 +11006653 }
NeilBrown29269552006-03-27 01:18:10 -08006654
NeilBrown87a8dec2011-01-31 11:57:43 +11006655 /* When a reshape changes the number of devices,
6656 * ->degraded is measured against the larger of the
6657 * pre and post number of devices.
6658 */
NeilBrownec32a2b2009-03-31 15:17:38 +11006659 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown908f4fb2011-12-23 10:17:50 +11006660 mddev->degraded = calc_degraded(conf);
NeilBrownec32a2b2009-03-31 15:17:38 +11006661 spin_unlock_irqrestore(&conf->device_lock, flags);
6662 }
NeilBrown63c70c42006-03-27 01:18:13 -08006663 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10006664 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07006665 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08006666
NeilBrown29269552006-03-27 01:18:10 -08006667 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6668 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6669 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6670 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6671 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10006672 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08006673 if (!mddev->sync_thread) {
6674 mddev->recovery = 0;
6675 spin_lock_irq(&conf->device_lock);
NeilBrownba8805b2013-11-14 15:16:15 +11006676 write_seqcount_begin(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006677 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownba8805b2013-11-14 15:16:15 +11006678 mddev->new_chunk_sectors =
6679 conf->chunk_sectors = conf->prev_chunk_sectors;
6680 mddev->new_layout = conf->algorithm = conf->prev_algo;
NeilBrown05616be2012-05-21 09:27:00 +10006681 rdev_for_each(rdev, mddev)
6682 rdev->new_data_offset = rdev->data_offset;
6683 smp_wmb();
NeilBrownba8805b2013-11-14 15:16:15 +11006684 conf->generation --;
NeilBrownfef9c612009-03-31 15:16:46 +11006685 conf->reshape_progress = MaxSector;
NeilBrown1e3fa9b2012-03-13 11:21:18 +11006686 mddev->reshape_position = MaxSector;
NeilBrownba8805b2013-11-14 15:16:15 +11006687 write_seqcount_end(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08006688 spin_unlock_irq(&conf->device_lock);
6689 return -EAGAIN;
6690 }
NeilBrownc8f517c2009-03-31 15:28:40 +11006691 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08006692 md_wakeup_thread(mddev->sync_thread);
6693 md_new_event(mddev);
6694 return 0;
6695}
NeilBrown29269552006-03-27 01:18:10 -08006696
NeilBrownec32a2b2009-03-31 15:17:38 +11006697/* This is called from the reshape thread and should make any
6698 * changes needed in 'conf'
6699 */
NeilBrownd1688a62011-10-11 16:49:52 +11006700static void end_reshape(struct r5conf *conf)
NeilBrown29269552006-03-27 01:18:10 -08006701{
NeilBrown29269552006-03-27 01:18:10 -08006702
NeilBrownf6705572006-03-27 01:18:11 -08006703 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
NeilBrown05616be2012-05-21 09:27:00 +10006704 struct md_rdev *rdev;
Dan Williams80c3a6c2009-03-17 18:10:40 -07006705
NeilBrownf6705572006-03-27 01:18:11 -08006706 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11006707 conf->previous_raid_disks = conf->raid_disks;
NeilBrown05616be2012-05-21 09:27:00 +10006708 rdev_for_each(rdev, conf->mddev)
6709 rdev->data_offset = rdev->new_data_offset;
6710 smp_wmb();
NeilBrownfef9c612009-03-31 15:16:46 +11006711 conf->reshape_progress = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08006712 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11006713 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07006714
6715 /* read-ahead size must cover two whole stripes, which is
6716 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
6717 */
NeilBrown4a5add42010-06-01 19:37:28 +10006718 if (conf->mddev->queue) {
NeilBrowncea9c222009-03-31 15:15:05 +11006719 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10006720 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11006721 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07006722 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6723 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
6724 }
NeilBrown29269552006-03-27 01:18:10 -08006725 }
NeilBrown29269552006-03-27 01:18:10 -08006726}
6727
NeilBrownec32a2b2009-03-31 15:17:38 +11006728/* This is called from the raid5d thread with mddev_lock held.
6729 * It makes config changes to the device.
6730 */
NeilBrownfd01b882011-10-11 16:47:53 +11006731static void raid5_finish_reshape(struct mddev *mddev)
NeilBrowncea9c222009-03-31 15:15:05 +11006732{
NeilBrownd1688a62011-10-11 16:49:52 +11006733 struct r5conf *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11006734
6735 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6736
NeilBrownec32a2b2009-03-31 15:17:38 +11006737 if (mddev->delta_disks > 0) {
6738 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6739 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10006740 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11006741 } else {
6742 int d;
NeilBrown908f4fb2011-12-23 10:17:50 +11006743 spin_lock_irq(&conf->device_lock);
6744 mddev->degraded = calc_degraded(conf);
6745 spin_unlock_irq(&conf->device_lock);
NeilBrownec32a2b2009-03-31 15:17:38 +11006746 for (d = conf->raid_disks ;
6747 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10006748 d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11006749 struct md_rdev *rdev = conf->disks[d].rdev;
NeilBrownda7613b2012-05-22 13:55:33 +10006750 if (rdev)
6751 clear_bit(In_sync, &rdev->flags);
6752 rdev = conf->disks[d].replacement;
6753 if (rdev)
6754 clear_bit(In_sync, &rdev->flags);
NeilBrown1a67dde2009-08-13 10:41:49 +10006755 }
NeilBrowncea9c222009-03-31 15:15:05 +11006756 }
NeilBrown88ce4932009-03-31 15:24:23 +11006757 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10006758 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11006759 mddev->reshape_position = MaxSector;
6760 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10006761 mddev->reshape_backwards = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11006762 }
6763}
6764
NeilBrownfd01b882011-10-11 16:47:53 +11006765static void raid5_quiesce(struct mddev *mddev, int state)
NeilBrown72626682005-09-09 16:23:54 -07006766{
NeilBrownd1688a62011-10-11 16:49:52 +11006767 struct r5conf *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07006768
6769 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08006770 case 2: /* resume for a suspend */
6771 wake_up(&conf->wait_for_overlap);
6772 break;
6773
NeilBrown72626682005-09-09 16:23:54 -07006774 case 1: /* stop all writes */
Shaohua Li566c09c2013-11-14 15:16:17 +11006775 lock_all_device_hash_locks_irq(conf);
NeilBrown64bd6602009-08-03 10:59:58 +10006776 /* '2' tells resync/reshape to pause so that all
6777 * active stripes can drain
6778 */
6779 conf->quiesce = 2;
Shaohua Li566c09c2013-11-14 15:16:17 +11006780 wait_event_cmd(conf->wait_for_stripe,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08006781 atomic_read(&conf->active_stripes) == 0 &&
6782 atomic_read(&conf->active_aligned_reads) == 0,
Shaohua Li566c09c2013-11-14 15:16:17 +11006783 unlock_all_device_hash_locks_irq(conf),
6784 lock_all_device_hash_locks_irq(conf));
NeilBrown64bd6602009-08-03 10:59:58 +10006785 conf->quiesce = 1;
Shaohua Li566c09c2013-11-14 15:16:17 +11006786 unlock_all_device_hash_locks_irq(conf);
NeilBrown64bd6602009-08-03 10:59:58 +10006787 /* allow reshape to continue */
6788 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07006789 break;
6790
6791 case 0: /* re-enable writes */
Shaohua Li566c09c2013-11-14 15:16:17 +11006792 lock_all_device_hash_locks_irq(conf);
NeilBrown72626682005-09-09 16:23:54 -07006793 conf->quiesce = 0;
6794 wake_up(&conf->wait_for_stripe);
NeilBrowne464eaf2006-03-27 01:18:14 -08006795 wake_up(&conf->wait_for_overlap);
Shaohua Li566c09c2013-11-14 15:16:17 +11006796 unlock_all_device_hash_locks_irq(conf);
NeilBrown72626682005-09-09 16:23:54 -07006797 break;
6798 }
NeilBrown72626682005-09-09 16:23:54 -07006799}
NeilBrownb15c2e52006-01-06 00:20:16 -08006800
NeilBrownfd01b882011-10-11 16:47:53 +11006801static void *raid45_takeover_raid0(struct mddev *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11006802{
NeilBrowne373ab12011-10-11 16:48:59 +11006803 struct r0conf *raid0_conf = mddev->private;
Randy Dunlapd76c8422011-04-21 09:07:26 -07006804 sector_t sectors;
Trela Maciej54071b32010-03-08 16:02:42 +11006805
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006806 /* for raid0 takeover only one zone is supported */
NeilBrowne373ab12011-10-11 16:48:59 +11006807 if (raid0_conf->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10006808 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
6809 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006810 return ERR_PTR(-EINVAL);
6811 }
6812
NeilBrowne373ab12011-10-11 16:48:59 +11006813 sectors = raid0_conf->strip_zone[0].zone_end;
6814 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
NeilBrown3b71bd92011-04-20 15:38:18 +10006815 mddev->dev_sectors = sectors;
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006816 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11006817 mddev->new_layout = ALGORITHM_PARITY_N;
6818 mddev->new_chunk_sectors = mddev->chunk_sectors;
6819 mddev->raid_disks += 1;
6820 mddev->delta_disks = 1;
6821 /* make sure it will be not marked as dirty */
6822 mddev->recovery_cp = MaxSector;
6823
6824 return setup_conf(mddev);
6825}
6826
NeilBrownfd01b882011-10-11 16:47:53 +11006827static void *raid5_takeover_raid1(struct mddev *mddev)
NeilBrownd562b0c2009-03-31 14:39:39 +11006828{
6829 int chunksect;
6830
6831 if (mddev->raid_disks != 2 ||
6832 mddev->degraded > 1)
6833 return ERR_PTR(-EINVAL);
6834
6835 /* Should check if there are write-behind devices? */
6836
6837 chunksect = 64*2; /* 64K by default */
6838
6839 /* The array must be an exact multiple of chunksize */
6840 while (chunksect && (mddev->array_sectors & (chunksect-1)))
6841 chunksect >>= 1;
6842
6843 if ((chunksect<<9) < STRIPE_SIZE)
6844 /* array size does not allow a suitable chunk size */
6845 return ERR_PTR(-EINVAL);
6846
6847 mddev->new_level = 5;
6848 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10006849 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11006850
6851 return setup_conf(mddev);
6852}
6853
NeilBrownfd01b882011-10-11 16:47:53 +11006854static void *raid5_takeover_raid6(struct mddev *mddev)
NeilBrownfc9739c2009-03-31 14:57:20 +11006855{
6856 int new_layout;
6857
6858 switch (mddev->layout) {
6859 case ALGORITHM_LEFT_ASYMMETRIC_6:
6860 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
6861 break;
6862 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6863 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
6864 break;
6865 case ALGORITHM_LEFT_SYMMETRIC_6:
6866 new_layout = ALGORITHM_LEFT_SYMMETRIC;
6867 break;
6868 case ALGORITHM_RIGHT_SYMMETRIC_6:
6869 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
6870 break;
6871 case ALGORITHM_PARITY_0_6:
6872 new_layout = ALGORITHM_PARITY_0;
6873 break;
6874 case ALGORITHM_PARITY_N:
6875 new_layout = ALGORITHM_PARITY_N;
6876 break;
6877 default:
6878 return ERR_PTR(-EINVAL);
6879 }
6880 mddev->new_level = 5;
6881 mddev->new_layout = new_layout;
6882 mddev->delta_disks = -1;
6883 mddev->raid_disks -= 1;
6884 return setup_conf(mddev);
6885}
6886
NeilBrownfd01b882011-10-11 16:47:53 +11006887static int raid5_check_reshape(struct mddev *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11006888{
NeilBrown88ce4932009-03-31 15:24:23 +11006889 /* For a 2-drive array, the layout and chunk size can be changed
6890 * immediately as not restriping is needed.
6891 * For larger arrays we record the new value - after validation
6892 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11006893 */
NeilBrownd1688a62011-10-11 16:49:52 +11006894 struct r5conf *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10006895 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11006896
NeilBrown597a7112009-06-18 08:47:42 +10006897 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11006898 return -EINVAL;
6899 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10006900 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11006901 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006902 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11006903 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006904 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11006905 /* not factor of array size */
6906 return -EINVAL;
6907 }
6908
6909 /* They look valid */
6910
NeilBrown88ce4932009-03-31 15:24:23 +11006911 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10006912 /* can make the change immediately */
6913 if (mddev->new_layout >= 0) {
6914 conf->algorithm = mddev->new_layout;
6915 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11006916 }
6917 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10006918 conf->chunk_sectors = new_chunk ;
6919 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11006920 }
6921 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6922 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11006923 }
NeilBrown50ac1682009-06-18 08:47:55 +10006924 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11006925}
6926
NeilBrownfd01b882011-10-11 16:47:53 +11006927static int raid6_check_reshape(struct mddev *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11006928{
NeilBrown597a7112009-06-18 08:47:42 +10006929 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10006930
NeilBrown597a7112009-06-18 08:47:42 +10006931 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11006932 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11006933 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10006934 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11006935 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006936 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11006937 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006938 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11006939 /* not factor of array size */
6940 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11006941 }
NeilBrown88ce4932009-03-31 15:24:23 +11006942
6943 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10006944 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11006945}
6946
NeilBrownfd01b882011-10-11 16:47:53 +11006947static void *raid5_takeover(struct mddev *mddev)
NeilBrownd562b0c2009-03-31 14:39:39 +11006948{
6949 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006950 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11006951 * raid1 - if there are two drives. We need to know the chunk size
6952 * raid4 - trivial - just use a raid4 layout.
6953 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11006954 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006955 if (mddev->level == 0)
6956 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11006957 if (mddev->level == 1)
6958 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11006959 if (mddev->level == 4) {
6960 mddev->new_layout = ALGORITHM_PARITY_N;
6961 mddev->new_level = 5;
6962 return setup_conf(mddev);
6963 }
NeilBrownfc9739c2009-03-31 14:57:20 +11006964 if (mddev->level == 6)
6965 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11006966
6967 return ERR_PTR(-EINVAL);
6968}
6969
NeilBrownfd01b882011-10-11 16:47:53 +11006970static void *raid4_takeover(struct mddev *mddev)
NeilBrowna78d38a2010-03-22 16:53:49 +11006971{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006972 /* raid4 can take over:
6973 * raid0 - if there is only one strip zone
6974 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11006975 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07006976 if (mddev->level == 0)
6977 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11006978 if (mddev->level == 5 &&
6979 mddev->layout == ALGORITHM_PARITY_N) {
6980 mddev->new_layout = 0;
6981 mddev->new_level = 4;
6982 return setup_conf(mddev);
6983 }
6984 return ERR_PTR(-EINVAL);
6985}
NeilBrownd562b0c2009-03-31 14:39:39 +11006986
NeilBrown84fc4b52011-10-11 16:49:58 +11006987static struct md_personality raid5_personality;
NeilBrown245f46c2009-03-31 14:39:39 +11006988
NeilBrownfd01b882011-10-11 16:47:53 +11006989static void *raid6_takeover(struct mddev *mddev)
NeilBrown245f46c2009-03-31 14:39:39 +11006990{
6991 /* Currently can only take over a raid5. We map the
6992 * personality to an equivalent raid6 personality
6993 * with the Q block at the end.
6994 */
6995 int new_layout;
6996
6997 if (mddev->pers != &raid5_personality)
6998 return ERR_PTR(-EINVAL);
6999 if (mddev->degraded > 1)
7000 return ERR_PTR(-EINVAL);
7001 if (mddev->raid_disks > 253)
7002 return ERR_PTR(-EINVAL);
7003 if (mddev->raid_disks < 3)
7004 return ERR_PTR(-EINVAL);
7005
7006 switch (mddev->layout) {
7007 case ALGORITHM_LEFT_ASYMMETRIC:
7008 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
7009 break;
7010 case ALGORITHM_RIGHT_ASYMMETRIC:
7011 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
7012 break;
7013 case ALGORITHM_LEFT_SYMMETRIC:
7014 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
7015 break;
7016 case ALGORITHM_RIGHT_SYMMETRIC:
7017 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
7018 break;
7019 case ALGORITHM_PARITY_0:
7020 new_layout = ALGORITHM_PARITY_0_6;
7021 break;
7022 case ALGORITHM_PARITY_N:
7023 new_layout = ALGORITHM_PARITY_N;
7024 break;
7025 default:
7026 return ERR_PTR(-EINVAL);
7027 }
7028 mddev->new_level = 6;
7029 mddev->new_layout = new_layout;
7030 mddev->delta_disks = 1;
7031 mddev->raid_disks += 1;
7032 return setup_conf(mddev);
7033}
7034
NeilBrown84fc4b52011-10-11 16:49:58 +11007035static struct md_personality raid6_personality =
NeilBrown16a53ec2006-06-26 00:27:38 -07007036{
7037 .name = "raid6",
7038 .level = 6,
7039 .owner = THIS_MODULE,
7040 .make_request = make_request,
7041 .run = run,
7042 .stop = stop,
7043 .status = status,
7044 .error_handler = error,
7045 .hot_add_disk = raid5_add_disk,
7046 .hot_remove_disk= raid5_remove_disk,
7047 .spare_active = raid5_spare_active,
7048 .sync_request = sync_request,
7049 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07007050 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10007051 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08007052 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11007053 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07007054 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11007055 .takeover = raid6_takeover,
NeilBrown16a53ec2006-06-26 00:27:38 -07007056};
NeilBrown84fc4b52011-10-11 16:49:58 +11007057static struct md_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07007058{
7059 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08007060 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007061 .owner = THIS_MODULE,
7062 .make_request = make_request,
7063 .run = run,
7064 .stop = stop,
7065 .status = status,
7066 .error_handler = error,
7067 .hot_add_disk = raid5_add_disk,
7068 .hot_remove_disk= raid5_remove_disk,
7069 .spare_active = raid5_spare_active,
7070 .sync_request = sync_request,
7071 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07007072 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08007073 .check_reshape = raid5_check_reshape,
7074 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11007075 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07007076 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11007077 .takeover = raid5_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007078};
7079
NeilBrown84fc4b52011-10-11 16:49:58 +11007080static struct md_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07007081{
NeilBrown2604b702006-01-06 00:20:36 -08007082 .name = "raid4",
7083 .level = 4,
7084 .owner = THIS_MODULE,
7085 .make_request = make_request,
7086 .run = run,
7087 .stop = stop,
7088 .status = status,
7089 .error_handler = error,
7090 .hot_add_disk = raid5_add_disk,
7091 .hot_remove_disk= raid5_remove_disk,
7092 .spare_active = raid5_spare_active,
7093 .sync_request = sync_request,
7094 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07007095 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08007096 .check_reshape = raid5_check_reshape,
7097 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11007098 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08007099 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11007100 .takeover = raid4_takeover,
NeilBrown2604b702006-01-06 00:20:36 -08007101};
7102
7103static int __init raid5_init(void)
7104{
Shaohua Li851c30c2013-08-28 14:30:16 +08007105 raid5_wq = alloc_workqueue("raid5wq",
7106 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
7107 if (!raid5_wq)
7108 return -ENOMEM;
NeilBrown16a53ec2006-06-26 00:27:38 -07007109 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08007110 register_md_personality(&raid5_personality);
7111 register_md_personality(&raid4_personality);
7112 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007113}
7114
NeilBrown2604b702006-01-06 00:20:36 -08007115static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007116{
NeilBrown16a53ec2006-06-26 00:27:38 -07007117 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08007118 unregister_md_personality(&raid5_personality);
7119 unregister_md_personality(&raid4_personality);
Shaohua Li851c30c2013-08-28 14:30:16 +08007120 destroy_workqueue(raid5_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007121}
7122
7123module_init(raid5_init);
7124module_exit(raid5_exit);
7125MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11007126MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007127MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08007128MODULE_ALIAS("md-raid5");
7129MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08007130MODULE_ALIAS("md-level-5");
7131MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07007132MODULE_ALIAS("md-personality-8"); /* RAID6 */
7133MODULE_ALIAS("md-raid6");
7134MODULE_ALIAS("md-level-6");
7135
7136/* This used to be two separate modules, they were: */
7137MODULE_ALIAS("raid5");
7138MODULE_ALIAS("raid6");