blob: 8ab8b65e17413e4a015aab5e073728fd23e21b47 [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>
shli@kernel.org46d5b782014-12-15 12:57:02 +110057#include <linux/flex_array.h>
NeilBrowna9add5d2012-10-31 11:59:09 +110058#include <trace/events/block.h>
59
NeilBrown43b2e5d2009-03-31 14:33:13 +110060#include "md.h"
NeilBrownbff61972009-03-31 14:33:13 +110061#include "raid5.h"
Trela Maciej54071b32010-03-08 16:02:42 +110062#include "raid0.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110063#include "bitmap.h"
NeilBrown72626682005-09-09 16:23:54 -070064
Shaohua Li851c30c2013-08-28 14:30:16 +080065#define cpu_to_group(cpu) cpu_to_node(cpu)
66#define ANY_GROUP NUMA_NO_NODE
67
NeilBrown8e0e99b2014-10-02 13:45:00 +100068static bool devices_handle_discard_safely = false;
69module_param(devices_handle_discard_safely, bool, 0644);
70MODULE_PARM_DESC(devices_handle_discard_safely,
71 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
Shaohua Li851c30c2013-08-28 14:30:16 +080072static struct workqueue_struct *raid5_wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073/*
74 * Stripe cache
75 */
76
77#define NR_STRIPES 256
78#define STRIPE_SIZE PAGE_SIZE
79#define STRIPE_SHIFT (PAGE_SHIFT - 9)
80#define STRIPE_SECTORS (STRIPE_SIZE>>9)
81#define IO_THRESHOLD 1
Dan Williams8b3e6cd2008-04-28 02:15:53 -070082#define BYPASS_THRESHOLD 1
NeilBrownfccddba2006-01-06 00:20:33 -080083#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
Linus Torvalds1da177e2005-04-16 15:20:36 -070084#define HASH_MASK (NR_HASH - 1)
Shaohua Libfc90cb2013-08-29 15:40:32 +080085#define MAX_STRIPE_BATCH 8
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
NeilBrownd1688a62011-10-11 16:49:52 +110087static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
NeilBrowndb298e12011-10-07 14:23:00 +110088{
89 int hash = (sect >> STRIPE_SHIFT) & HASH_MASK;
90 return &conf->stripe_hashtbl[hash];
91}
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Shaohua Li566c09c2013-11-14 15:16:17 +110093static inline int stripe_hash_locks_hash(sector_t sect)
94{
95 return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
96}
97
98static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
99{
100 spin_lock_irq(conf->hash_locks + hash);
101 spin_lock(&conf->device_lock);
102}
103
104static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
105{
106 spin_unlock(&conf->device_lock);
107 spin_unlock_irq(conf->hash_locks + hash);
108}
109
110static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
111{
112 int i;
113 local_irq_disable();
114 spin_lock(conf->hash_locks);
115 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
116 spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
117 spin_lock(&conf->device_lock);
118}
119
120static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
121{
122 int i;
123 spin_unlock(&conf->device_lock);
124 for (i = NR_STRIPE_HASH_LOCKS; i; i--)
125 spin_unlock(conf->hash_locks + i - 1);
126 local_irq_enable();
127}
128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129/* bio's attached to a stripe+device for I/O are linked together in bi_sector
130 * order without overlap. There may be several bio's per stripe+device, and
131 * a bio could span several devices.
132 * When walking this list for a particular stripe+device, we must never proceed
133 * beyond a bio that extends past this device, as the next bio might no longer
134 * be valid.
NeilBrowndb298e12011-10-07 14:23:00 +1100135 * This function is used to determine the 'next' bio in the list, given the sector
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 * of the current stripe+device
137 */
NeilBrowndb298e12011-10-07 14:23:00 +1100138static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector)
139{
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800140 int sectors = bio_sectors(bio);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700141 if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS)
NeilBrowndb298e12011-10-07 14:23:00 +1100142 return bio->bi_next;
143 else
144 return NULL;
145}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Jens Axboe960e7392008-08-15 10:41:18 +0200147/*
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200148 * We maintain a biased count of active stripes in the bottom 16 bits of
149 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
Jens Axboe960e7392008-08-15 10:41:18 +0200150 */
Shaohua Lie7836bd62012-07-19 16:01:31 +1000151static inline int raid5_bi_processed_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200152{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000153 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
154 return (atomic_read(segments) >> 16) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200155}
156
Shaohua Lie7836bd62012-07-19 16:01:31 +1000157static inline int raid5_dec_bi_active_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200158{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000159 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
160 return atomic_sub_return(1, segments) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200161}
162
Shaohua Lie7836bd62012-07-19 16:01:31 +1000163static inline void raid5_inc_bi_active_stripes(struct bio *bio)
Jens Axboe960e7392008-08-15 10:41:18 +0200164{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000165 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
166 atomic_inc(segments);
Jens Axboe960e7392008-08-15 10:41:18 +0200167}
168
Shaohua Lie7836bd62012-07-19 16:01:31 +1000169static inline void raid5_set_bi_processed_stripes(struct bio *bio,
170 unsigned int cnt)
Jens Axboe960e7392008-08-15 10:41:18 +0200171{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000172 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
173 int old, new;
Jens Axboe960e7392008-08-15 10:41:18 +0200174
Shaohua Lie7836bd62012-07-19 16:01:31 +1000175 do {
176 old = atomic_read(segments);
177 new = (old & 0xffff) | (cnt << 16);
178 } while (atomic_cmpxchg(segments, old, new) != old);
Jens Axboe960e7392008-08-15 10:41:18 +0200179}
180
Shaohua Lie7836bd62012-07-19 16:01:31 +1000181static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt)
Jens Axboe960e7392008-08-15 10:41:18 +0200182{
Shaohua Lie7836bd62012-07-19 16:01:31 +1000183 atomic_t *segments = (atomic_t *)&bio->bi_phys_segments;
184 atomic_set(segments, cnt);
Jens Axboe960e7392008-08-15 10:41:18 +0200185}
186
NeilBrownd0dabf72009-03-31 14:39:38 +1100187/* Find first data disk in a raid6 stripe */
188static inline int raid6_d0(struct stripe_head *sh)
189{
NeilBrown67cc2b82009-03-31 14:39:38 +1100190 if (sh->ddf_layout)
191 /* ddf always start from first device */
192 return 0;
193 /* md starts just after Q block */
NeilBrownd0dabf72009-03-31 14:39:38 +1100194 if (sh->qd_idx == sh->disks - 1)
195 return 0;
196 else
197 return sh->qd_idx + 1;
198}
NeilBrown16a53ec2006-06-26 00:27:38 -0700199static inline int raid6_next_disk(int disk, int raid_disks)
200{
201 disk++;
202 return (disk < raid_disks) ? disk : 0;
203}
Dan Williamsa4456852007-07-09 11:56:43 -0700204
NeilBrownd0dabf72009-03-31 14:39:38 +1100205/* When walking through the disks in a raid5, starting at raid6_d0,
206 * We need to map each disk to a 'slot', where the data disks are slot
207 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
208 * is raid_disks-1. This help does that mapping.
209 */
NeilBrown67cc2b82009-03-31 14:39:38 +1100210static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
211 int *count, int syndrome_disks)
NeilBrownd0dabf72009-03-31 14:39:38 +1100212{
Dan Williams66295422009-10-19 18:09:32 -0700213 int slot = *count;
NeilBrown67cc2b82009-03-31 14:39:38 +1100214
NeilBrowne4424fe2009-10-16 16:27:34 +1100215 if (sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700216 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100217 if (idx == sh->pd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100218 return syndrome_disks;
NeilBrownd0dabf72009-03-31 14:39:38 +1100219 if (idx == sh->qd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100220 return syndrome_disks + 1;
NeilBrowne4424fe2009-10-16 16:27:34 +1100221 if (!sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700222 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100223 return slot;
224}
225
NeilBrown34a6f802015-08-14 12:07:57 +1000226static void return_io(struct bio_list *return_bi)
Dan Williamsa4456852007-07-09 11:56:43 -0700227{
NeilBrown34a6f802015-08-14 12:07:57 +1000228 struct bio *bi;
229 while ((bi = bio_list_pop(return_bi)) != NULL) {
Kent Overstreet4f024f32013-10-11 15:44:27 -0700230 bi->bi_iter.bi_size = 0;
Linus Torvalds0a82a8d2013-04-18 09:00:26 -0700231 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
232 bi, 0);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200233 bio_endio(bi);
Dan Williamsa4456852007-07-09 11:56:43 -0700234 }
235}
236
NeilBrownd1688a62011-10-11 16:49:52 +1100237static void print_raid5_conf (struct r5conf *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Dan Williams600aa102008-06-28 08:32:05 +1000239static int stripe_operations_active(struct stripe_head *sh)
240{
241 return sh->check_state || sh->reconstruct_state ||
242 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
243 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
244}
245
Shaohua Li851c30c2013-08-28 14:30:16 +0800246static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
247{
248 struct r5conf *conf = sh->raid_conf;
249 struct r5worker_group *group;
Shaohua Libfc90cb2013-08-29 15:40:32 +0800250 int thread_cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +0800251 int i, cpu = sh->cpu;
252
253 if (!cpu_online(cpu)) {
254 cpu = cpumask_any(cpu_online_mask);
255 sh->cpu = cpu;
256 }
257
258 if (list_empty(&sh->lru)) {
259 struct r5worker_group *group;
260 group = conf->worker_groups + cpu_to_group(cpu);
261 list_add_tail(&sh->lru, &group->handle_list);
Shaohua Libfc90cb2013-08-29 15:40:32 +0800262 group->stripes_cnt++;
263 sh->group = group;
Shaohua Li851c30c2013-08-28 14:30:16 +0800264 }
265
266 if (conf->worker_cnt_per_group == 0) {
267 md_wakeup_thread(conf->mddev->thread);
268 return;
269 }
270
271 group = conf->worker_groups + cpu_to_group(sh->cpu);
272
Shaohua Libfc90cb2013-08-29 15:40:32 +0800273 group->workers[0].working = true;
274 /* at least one worker should run to avoid race */
275 queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work);
276
277 thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1;
278 /* wakeup more workers */
279 for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) {
280 if (group->workers[i].working == false) {
281 group->workers[i].working = true;
282 queue_work_on(sh->cpu, raid5_wq,
283 &group->workers[i].work);
284 thread_cnt--;
285 }
286 }
Shaohua Li851c30c2013-08-28 14:30:16 +0800287}
288
Shaohua Li566c09c2013-11-14 15:16:17 +1100289static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
290 struct list_head *temp_inactive_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
Shaohua Li4eb788d2012-07-19 16:01:31 +1000292 BUG_ON(!list_empty(&sh->lru));
293 BUG_ON(atomic_read(&conf->active_stripes)==0);
294 if (test_bit(STRIPE_HANDLE, &sh->state)) {
295 if (test_bit(STRIPE_DELAYED, &sh->state) &&
Jes Sorensenad3ab8b2015-01-29 12:38:29 -0500296 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
Shaohua Li4eb788d2012-07-19 16:01:31 +1000297 list_add_tail(&sh->lru, &conf->delayed_list);
Jes Sorensenad3ab8b2015-01-29 12:38:29 -0500298 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
Shaohua Li4eb788d2012-07-19 16:01:31 +1000299 sh->bm_seq - conf->seq_write > 0)
300 list_add_tail(&sh->lru, &conf->bitmap_list);
301 else {
302 clear_bit(STRIPE_DELAYED, &sh->state);
303 clear_bit(STRIPE_BIT_DELAY, &sh->state);
Shaohua Li851c30c2013-08-28 14:30:16 +0800304 if (conf->worker_cnt_per_group == 0) {
305 list_add_tail(&sh->lru, &conf->handle_list);
306 } else {
307 raid5_wakeup_stripe_thread(sh);
308 return;
309 }
Shaohua Li4eb788d2012-07-19 16:01:31 +1000310 }
311 md_wakeup_thread(conf->mddev->thread);
312 } else {
313 BUG_ON(stripe_operations_active(sh));
314 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
315 if (atomic_dec_return(&conf->preread_active_stripes)
316 < IO_THRESHOLD)
317 md_wakeup_thread(conf->mddev->thread);
318 atomic_dec(&conf->active_stripes);
Shaohua Li566c09c2013-11-14 15:16:17 +1100319 if (!test_bit(STRIPE_EXPANDING, &sh->state))
320 list_add_tail(&sh->lru, temp_inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 }
322}
NeilBrownd0dabf72009-03-31 14:39:38 +1100323
Shaohua Li566c09c2013-11-14 15:16:17 +1100324static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
325 struct list_head *temp_inactive_list)
Shaohua Li4eb788d2012-07-19 16:01:31 +1000326{
327 if (atomic_dec_and_test(&sh->count))
Shaohua Li566c09c2013-11-14 15:16:17 +1100328 do_release_stripe(conf, sh, temp_inactive_list);
329}
330
331/*
332 * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
333 *
334 * Be careful: Only one task can add/delete stripes from temp_inactive_list at
335 * given time. Adding stripes only takes device lock, while deleting stripes
336 * only takes hash lock.
337 */
338static void release_inactive_stripe_list(struct r5conf *conf,
339 struct list_head *temp_inactive_list,
340 int hash)
341{
342 int size;
Shaohua Li6ab2a4b2016-02-25 16:24:42 -0800343 bool do_wakeup = false;
Shaohua Li566c09c2013-11-14 15:16:17 +1100344 unsigned long flags;
345
346 if (hash == NR_STRIPE_HASH_LOCKS) {
347 size = NR_STRIPE_HASH_LOCKS;
348 hash = NR_STRIPE_HASH_LOCKS - 1;
349 } else
350 size = 1;
351 while (size) {
352 struct list_head *list = &temp_inactive_list[size - 1];
353
354 /*
Shaohua Li6d036f72015-08-13 14:31:57 -0700355 * We don't hold any lock here yet, raid5_get_active_stripe() might
Shaohua Li566c09c2013-11-14 15:16:17 +1100356 * remove stripes from the list
357 */
358 if (!list_empty_careful(list)) {
359 spin_lock_irqsave(conf->hash_locks + hash, flags);
Shaohua Li4bda5562013-11-14 15:16:17 +1100360 if (list_empty(conf->inactive_list + hash) &&
361 !list_empty(list))
362 atomic_dec(&conf->empty_inactive_list_nr);
Shaohua Li566c09c2013-11-14 15:16:17 +1100363 list_splice_tail_init(list, conf->inactive_list + hash);
Shaohua Li6ab2a4b2016-02-25 16:24:42 -0800364 do_wakeup = true;
Shaohua Li566c09c2013-11-14 15:16:17 +1100365 spin_unlock_irqrestore(conf->hash_locks + hash, flags);
366 }
367 size--;
368 hash--;
369 }
370
371 if (do_wakeup) {
Shaohua Li6ab2a4b2016-02-25 16:24:42 -0800372 wake_up(&conf->wait_for_stripe);
Yuanhan Liub1b46482015-05-08 18:19:06 +1000373 if (atomic_read(&conf->active_stripes) == 0)
374 wake_up(&conf->wait_for_quiescent);
Shaohua Li566c09c2013-11-14 15:16:17 +1100375 if (conf->retry_read_aligned)
376 md_wakeup_thread(conf->mddev->thread);
377 }
Shaohua Li4eb788d2012-07-19 16:01:31 +1000378}
379
Shaohua Li773ca822013-08-27 17:50:39 +0800380/* should hold conf->device_lock already */
Shaohua Li566c09c2013-11-14 15:16:17 +1100381static int release_stripe_list(struct r5conf *conf,
382 struct list_head *temp_inactive_list)
Shaohua Li773ca822013-08-27 17:50:39 +0800383{
384 struct stripe_head *sh;
385 int count = 0;
386 struct llist_node *head;
387
388 head = llist_del_all(&conf->released_stripes);
Shaohua Lid265d9d2013-08-28 14:29:05 +0800389 head = llist_reverse_order(head);
Shaohua Li773ca822013-08-27 17:50:39 +0800390 while (head) {
Shaohua Li566c09c2013-11-14 15:16:17 +1100391 int hash;
392
Shaohua Li773ca822013-08-27 17:50:39 +0800393 sh = llist_entry(head, struct stripe_head, release_list);
394 head = llist_next(head);
395 /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
396 smp_mb();
397 clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
398 /*
399 * Don't worry the bit is set here, because if the bit is set
400 * again, the count is always > 1. This is true for
401 * STRIPE_ON_UNPLUG_LIST bit too.
402 */
Shaohua Li566c09c2013-11-14 15:16:17 +1100403 hash = sh->hash_lock_index;
404 __release_stripe(conf, sh, &temp_inactive_list[hash]);
Shaohua Li773ca822013-08-27 17:50:39 +0800405 count++;
406 }
407
408 return count;
409}
410
Shaohua Li6d036f72015-08-13 14:31:57 -0700411void raid5_release_stripe(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
NeilBrownd1688a62011-10-11 16:49:52 +1100413 struct r5conf *conf = sh->raid_conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 unsigned long flags;
Shaohua Li566c09c2013-11-14 15:16:17 +1100415 struct list_head list;
416 int hash;
Shaohua Li773ca822013-08-27 17:50:39 +0800417 bool wakeup;
NeilBrown16a53ec2006-06-26 00:27:38 -0700418
Eivind Sartocf170f32014-05-28 13:39:23 +1000419 /* Avoid release_list until the last reference.
420 */
421 if (atomic_add_unless(&sh->count, -1, 1))
422 return;
423
majianpengad4068d2013-11-14 15:16:15 +1100424 if (unlikely(!conf->mddev->thread) ||
425 test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
Shaohua Li773ca822013-08-27 17:50:39 +0800426 goto slow_path;
427 wakeup = llist_add(&sh->release_list, &conf->released_stripes);
428 if (wakeup)
429 md_wakeup_thread(conf->mddev->thread);
430 return;
431slow_path:
Shaohua Li4eb788d2012-07-19 16:01:31 +1000432 local_irq_save(flags);
Shaohua Li773ca822013-08-27 17:50:39 +0800433 /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
Shaohua Li4eb788d2012-07-19 16:01:31 +1000434 if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
Shaohua Li566c09c2013-11-14 15:16:17 +1100435 INIT_LIST_HEAD(&list);
436 hash = sh->hash_lock_index;
437 do_release_stripe(conf, sh, &list);
Shaohua Li4eb788d2012-07-19 16:01:31 +1000438 spin_unlock(&conf->device_lock);
Shaohua Li566c09c2013-11-14 15:16:17 +1100439 release_inactive_stripe_list(conf, &list, hash);
Shaohua Li4eb788d2012-07-19 16:01:31 +1000440 }
441 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443
NeilBrownfccddba2006-01-06 00:20:33 -0800444static inline void remove_hash(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
Dan Williams45b42332007-07-09 11:56:43 -0700446 pr_debug("remove_hash(), stripe %llu\n",
447 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
NeilBrownfccddba2006-01-06 00:20:33 -0800449 hlist_del_init(&sh->hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450}
451
NeilBrownd1688a62011-10-11 16:49:52 +1100452static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
NeilBrownfccddba2006-01-06 00:20:33 -0800454 struct hlist_head *hp = stripe_hash(conf, sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
Dan Williams45b42332007-07-09 11:56:43 -0700456 pr_debug("insert_hash(), stripe %llu\n",
457 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
NeilBrownfccddba2006-01-06 00:20:33 -0800459 hlist_add_head(&sh->hash, hp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462/* find an idle stripe, make sure it is unhashed, and return it. */
Shaohua Li566c09c2013-11-14 15:16:17 +1100463static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
465 struct stripe_head *sh = NULL;
466 struct list_head *first;
467
Shaohua Li566c09c2013-11-14 15:16:17 +1100468 if (list_empty(conf->inactive_list + hash))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 goto out;
Shaohua Li566c09c2013-11-14 15:16:17 +1100470 first = (conf->inactive_list + hash)->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 sh = list_entry(first, struct stripe_head, lru);
472 list_del_init(first);
473 remove_hash(sh);
474 atomic_inc(&conf->active_stripes);
Shaohua Li566c09c2013-11-14 15:16:17 +1100475 BUG_ON(hash != sh->hash_lock_index);
Shaohua Li4bda5562013-11-14 15:16:17 +1100476 if (list_empty(conf->inactive_list + hash))
477 atomic_inc(&conf->empty_inactive_list_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478out:
479 return sh;
480}
481
NeilBrowne4e11e32010-06-16 16:45:16 +1000482static void shrink_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
484 struct page *p;
485 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000486 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
NeilBrowne4e11e32010-06-16 16:45:16 +1000488 for (i = 0; i < num ; i++) {
Shaohua Lid592a992014-05-21 17:57:44 +0800489 WARN_ON(sh->dev[i].page != sh->dev[i].orig_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 p = sh->dev[i].page;
491 if (!p)
492 continue;
493 sh->dev[i].page = NULL;
NeilBrown2d1f3b52006-01-06 00:20:31 -0800494 put_page(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
496}
497
NeilBrowna9683a72015-02-25 12:02:51 +1100498static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
500 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000501 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
NeilBrowne4e11e32010-06-16 16:45:16 +1000503 for (i = 0; i < num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 struct page *page;
505
NeilBrowna9683a72015-02-25 12:02:51 +1100506 if (!(page = alloc_page(gfp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 return 1;
508 }
509 sh->dev[i].page = page;
Shaohua Lid592a992014-05-21 17:57:44 +0800510 sh->dev[i].orig_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512 return 0;
513}
514
NeilBrown784052e2009-03-31 15:19:07 +1100515static void raid5_build_block(struct stripe_head *sh, int i, int previous);
NeilBrownd1688a62011-10-11 16:49:52 +1100516static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
NeilBrown911d4ee2009-03-31 14:39:38 +1100517 struct stripe_head *sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
NeilBrownb5663ba2009-03-31 14:39:38 +1100519static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
NeilBrownd1688a62011-10-11 16:49:52 +1100521 struct r5conf *conf = sh->raid_conf;
Shaohua Li566c09c2013-11-14 15:16:17 +1100522 int i, seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200524 BUG_ON(atomic_read(&sh->count) != 0);
525 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
Dan Williams600aa102008-06-28 08:32:05 +1000526 BUG_ON(stripe_operations_active(sh));
shli@kernel.org59fc6302014-12-15 12:57:03 +1100527 BUG_ON(sh->batch_head);
Dan Williamsd84e0f12007-01-02 13:52:30 -0700528
Dan Williams45b42332007-07-09 11:56:43 -0700529 pr_debug("init_stripe called, stripe %llu\n",
Markus Stockhausenb8e6a152014-08-23 20:19:27 +1000530 (unsigned long long)sector);
Shaohua Li566c09c2013-11-14 15:16:17 +1100531retry:
532 seq = read_seqcount_begin(&conf->gen_lock);
NeilBrown86b42c72009-03-31 15:19:03 +1100533 sh->generation = conf->generation - previous;
NeilBrownb5663ba2009-03-31 14:39:38 +1100534 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 sh->sector = sector;
NeilBrown911d4ee2009-03-31 14:39:38 +1100536 stripe_set_idx(sector, conf, previous, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 sh->state = 0;
538
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800539 for (i = sh->disks; i--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 struct r5dev *dev = &sh->dev[i];
541
Dan Williamsd84e0f12007-01-02 13:52:30 -0700542 if (dev->toread || dev->read || dev->towrite || dev->written ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 test_bit(R5_LOCKED, &dev->flags)) {
Dan Williamsd84e0f12007-01-02 13:52:30 -0700544 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 (unsigned long long)sh->sector, i, dev->toread,
Dan Williamsd84e0f12007-01-02 13:52:30 -0700546 dev->read, dev->towrite, dev->written,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 test_bit(R5_LOCKED, &dev->flags));
NeilBrown8cfa7b02011-07-27 11:00:36 +1000548 WARN_ON(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
550 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +1100551 raid5_build_block(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 }
Shaohua Li566c09c2013-11-14 15:16:17 +1100553 if (read_seqcount_retry(&conf->gen_lock, seq))
554 goto retry;
shli@kernel.org7a87f432014-12-15 12:57:03 +1100555 sh->overwrite_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 insert_hash(conf, sh);
Shaohua Li851c30c2013-08-28 14:30:16 +0800557 sh->cpu = smp_processor_id();
shli@kernel.orgda41ba62014-12-15 12:57:03 +1100558 set_bit(STRIPE_BATCH_READY, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
560
NeilBrownd1688a62011-10-11 16:49:52 +1100561static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
NeilBrown86b42c72009-03-31 15:19:03 +1100562 short generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
564 struct stripe_head *sh;
565
Dan Williams45b42332007-07-09 11:56:43 -0700566 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800567 hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
NeilBrown86b42c72009-03-31 15:19:03 +1100568 if (sh->sector == sector && sh->generation == generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 return sh;
Dan Williams45b42332007-07-09 11:56:43 -0700570 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 return NULL;
572}
573
NeilBrown674806d2010-06-16 17:17:53 +1000574/*
575 * Need to check if array has failed when deciding whether to:
576 * - start an array
577 * - remove non-faulty devices
578 * - add a spare
579 * - allow a reshape
580 * This determination is simple when no reshape is happening.
581 * However if there is a reshape, we need to carefully check
582 * both the before and after sections.
583 * This is because some failed devices may only affect one
584 * of the two sections, and some non-in_sync devices may
585 * be insync in the section most affected by failed devices.
586 */
NeilBrown908f4fb2011-12-23 10:17:50 +1100587static int calc_degraded(struct r5conf *conf)
NeilBrown674806d2010-06-16 17:17:53 +1000588{
NeilBrown908f4fb2011-12-23 10:17:50 +1100589 int degraded, degraded2;
NeilBrown674806d2010-06-16 17:17:53 +1000590 int i;
NeilBrown674806d2010-06-16 17:17:53 +1000591
592 rcu_read_lock();
593 degraded = 0;
594 for (i = 0; i < conf->previous_raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100595 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowne5c86472012-09-19 12:52:30 +1000596 if (rdev && test_bit(Faulty, &rdev->flags))
597 rdev = rcu_dereference(conf->disks[i].replacement);
NeilBrown674806d2010-06-16 17:17:53 +1000598 if (!rdev || test_bit(Faulty, &rdev->flags))
599 degraded++;
600 else if (test_bit(In_sync, &rdev->flags))
601 ;
602 else
603 /* not in-sync or faulty.
604 * If the reshape increases the number of devices,
605 * this is being recovered by the reshape, so
606 * this 'previous' section is not in_sync.
607 * If the number of devices is being reduced however,
608 * the device can only be part of the array if
609 * we are reverting a reshape, so this section will
610 * be in-sync.
611 */
612 if (conf->raid_disks >= conf->previous_raid_disks)
613 degraded++;
614 }
615 rcu_read_unlock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100616 if (conf->raid_disks == conf->previous_raid_disks)
617 return degraded;
NeilBrown674806d2010-06-16 17:17:53 +1000618 rcu_read_lock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100619 degraded2 = 0;
NeilBrown674806d2010-06-16 17:17:53 +1000620 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100621 struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowne5c86472012-09-19 12:52:30 +1000622 if (rdev && test_bit(Faulty, &rdev->flags))
623 rdev = rcu_dereference(conf->disks[i].replacement);
NeilBrown674806d2010-06-16 17:17:53 +1000624 if (!rdev || test_bit(Faulty, &rdev->flags))
NeilBrown908f4fb2011-12-23 10:17:50 +1100625 degraded2++;
NeilBrown674806d2010-06-16 17:17:53 +1000626 else if (test_bit(In_sync, &rdev->flags))
627 ;
628 else
629 /* not in-sync or faulty.
630 * If reshape increases the number of devices, this
631 * section has already been recovered, else it
632 * almost certainly hasn't.
633 */
634 if (conf->raid_disks <= conf->previous_raid_disks)
NeilBrown908f4fb2011-12-23 10:17:50 +1100635 degraded2++;
NeilBrown674806d2010-06-16 17:17:53 +1000636 }
637 rcu_read_unlock();
NeilBrown908f4fb2011-12-23 10:17:50 +1100638 if (degraded2 > degraded)
639 return degraded2;
640 return degraded;
641}
642
643static int has_failed(struct r5conf *conf)
644{
645 int degraded;
646
647 if (conf->mddev->reshape_position == MaxSector)
648 return conf->mddev->degraded > conf->max_degraded;
649
650 degraded = calc_degraded(conf);
NeilBrown674806d2010-06-16 17:17:53 +1000651 if (degraded > conf->max_degraded)
652 return 1;
653 return 0;
654}
655
Shaohua Li6d036f72015-08-13 14:31:57 -0700656struct stripe_head *
657raid5_get_active_stripe(struct r5conf *conf, sector_t sector,
658 int previous, int noblock, int noquiesce)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
660 struct stripe_head *sh;
Shaohua Li566c09c2013-11-14 15:16:17 +1100661 int hash = stripe_hash_locks_hash(sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Dan Williams45b42332007-07-09 11:56:43 -0700663 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Shaohua Li566c09c2013-11-14 15:16:17 +1100665 spin_lock_irq(conf->hash_locks + hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
667 do {
Yuanhan Liub1b46482015-05-08 18:19:06 +1000668 wait_event_lock_irq(conf->wait_for_quiescent,
NeilBrowna8c906c2009-06-09 14:39:59 +1000669 conf->quiesce == 0 || noquiesce,
Shaohua Li566c09c2013-11-14 15:16:17 +1100670 *(conf->hash_locks + hash));
NeilBrown86b42c72009-03-31 15:19:03 +1100671 sh = __find_stripe(conf, sector, conf->generation - previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (!sh) {
NeilBrownedbe83a2015-02-26 12:47:56 +1100673 if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) {
Shaohua Li566c09c2013-11-14 15:16:17 +1100674 sh = get_free_stripe(conf, hash);
Shaohua Li713bc5c2015-05-28 17:33:47 -0700675 if (!sh && !test_bit(R5_DID_ALLOC,
676 &conf->cache_state))
NeilBrownedbe83a2015-02-26 12:47:56 +1100677 set_bit(R5_ALLOC_MORE,
678 &conf->cache_state);
679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 if (noblock && sh == NULL)
681 break;
682 if (!sh) {
NeilBrown54233992015-02-26 12:21:04 +1100683 set_bit(R5_INACTIVE_BLOCKED,
684 &conf->cache_state);
Shaohua Li6ab2a4b2016-02-25 16:24:42 -0800685 wait_event_lock_irq(
686 conf->wait_for_stripe,
Shaohua Li566c09c2013-11-14 15:16:17 +1100687 !list_empty(conf->inactive_list + hash) &&
688 (atomic_read(&conf->active_stripes)
689 < (conf->max_nr_stripes * 3 / 4)
NeilBrown54233992015-02-26 12:21:04 +1100690 || !test_bit(R5_INACTIVE_BLOCKED,
691 &conf->cache_state)),
Shaohua Li6ab2a4b2016-02-25 16:24:42 -0800692 *(conf->hash_locks + hash));
NeilBrown54233992015-02-26 12:21:04 +1100693 clear_bit(R5_INACTIVE_BLOCKED,
694 &conf->cache_state);
NeilBrown7da9d452014-01-22 11:45:03 +1100695 } else {
NeilBrownb5663ba2009-03-31 14:39:38 +1100696 init_stripe(sh, sector, previous);
NeilBrown7da9d452014-01-22 11:45:03 +1100697 atomic_inc(&sh->count);
698 }
Shaohua Lie240c182014-04-09 11:27:42 +0800699 } else if (!atomic_inc_not_zero(&sh->count)) {
NeilBrown6d183de2013-11-28 10:55:27 +1100700 spin_lock(&conf->device_lock);
Shaohua Lie240c182014-04-09 11:27:42 +0800701 if (!atomic_read(&sh->count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (!test_bit(STRIPE_HANDLE, &sh->state))
703 atomic_inc(&conf->active_stripes);
NeilBrown5af9bef2014-01-14 15:16:10 +1100704 BUG_ON(list_empty(&sh->lru) &&
705 !test_bit(STRIPE_EXPANDING, &sh->state));
NeilBrown16a53ec2006-06-26 00:27:38 -0700706 list_del_init(&sh->lru);
Shaohua Libfc90cb2013-08-29 15:40:32 +0800707 if (sh->group) {
708 sh->group->stripes_cnt--;
709 sh->group = NULL;
710 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
NeilBrown7da9d452014-01-22 11:45:03 +1100712 atomic_inc(&sh->count);
NeilBrown6d183de2013-11-28 10:55:27 +1100713 spin_unlock(&conf->device_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 }
715 } while (sh == NULL);
716
Shaohua Li566c09c2013-11-14 15:16:17 +1100717 spin_unlock_irq(conf->hash_locks + hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 return sh;
719}
720
shli@kernel.org7a87f432014-12-15 12:57:03 +1100721static bool is_full_stripe_write(struct stripe_head *sh)
722{
723 BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded));
724 return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded);
725}
726
shli@kernel.org59fc6302014-12-15 12:57:03 +1100727static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
728{
729 local_irq_disable();
730 if (sh1 > sh2) {
731 spin_lock(&sh2->stripe_lock);
732 spin_lock_nested(&sh1->stripe_lock, 1);
733 } else {
734 spin_lock(&sh1->stripe_lock);
735 spin_lock_nested(&sh2->stripe_lock, 1);
736 }
737}
738
739static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2)
740{
741 spin_unlock(&sh1->stripe_lock);
742 spin_unlock(&sh2->stripe_lock);
743 local_irq_enable();
744}
745
746/* Only freshly new full stripe normal write stripe can be added to a batch list */
747static bool stripe_can_batch(struct stripe_head *sh)
748{
Shaohua Li9c3e3332015-08-13 14:32:02 -0700749 struct r5conf *conf = sh->raid_conf;
750
751 if (conf->log)
752 return false;
shli@kernel.org59fc6302014-12-15 12:57:03 +1100753 return test_bit(STRIPE_BATCH_READY, &sh->state) &&
NeilBrownd0852df52015-05-27 08:43:45 +1000754 !test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
shli@kernel.org59fc6302014-12-15 12:57:03 +1100755 is_full_stripe_write(sh);
756}
757
758/* we only do back search */
759static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh)
760{
761 struct stripe_head *head;
762 sector_t head_sector, tmp_sec;
763 int hash;
764 int dd_idx;
765
shli@kernel.org59fc6302014-12-15 12:57:03 +1100766 /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */
767 tmp_sec = sh->sector;
768 if (!sector_div(tmp_sec, conf->chunk_sectors))
769 return;
770 head_sector = sh->sector - STRIPE_SECTORS;
771
772 hash = stripe_hash_locks_hash(head_sector);
773 spin_lock_irq(conf->hash_locks + hash);
774 head = __find_stripe(conf, head_sector, conf->generation);
775 if (head && !atomic_inc_not_zero(&head->count)) {
776 spin_lock(&conf->device_lock);
777 if (!atomic_read(&head->count)) {
778 if (!test_bit(STRIPE_HANDLE, &head->state))
779 atomic_inc(&conf->active_stripes);
780 BUG_ON(list_empty(&head->lru) &&
781 !test_bit(STRIPE_EXPANDING, &head->state));
782 list_del_init(&head->lru);
783 if (head->group) {
784 head->group->stripes_cnt--;
785 head->group = NULL;
786 }
787 }
788 atomic_inc(&head->count);
789 spin_unlock(&conf->device_lock);
790 }
791 spin_unlock_irq(conf->hash_locks + hash);
792
793 if (!head)
794 return;
795 if (!stripe_can_batch(head))
796 goto out;
797
798 lock_two_stripes(head, sh);
799 /* clear_batch_ready clear the flag */
800 if (!stripe_can_batch(head) || !stripe_can_batch(sh))
801 goto unlock_out;
802
803 if (sh->batch_head)
804 goto unlock_out;
805
806 dd_idx = 0;
807 while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx)
808 dd_idx++;
809 if (head->dev[dd_idx].towrite->bi_rw != sh->dev[dd_idx].towrite->bi_rw)
810 goto unlock_out;
811
812 if (head->batch_head) {
813 spin_lock(&head->batch_head->batch_lock);
814 /* This batch list is already running */
815 if (!stripe_can_batch(head)) {
816 spin_unlock(&head->batch_head->batch_lock);
817 goto unlock_out;
818 }
819
820 /*
821 * at this point, head's BATCH_READY could be cleared, but we
822 * can still add the stripe to batch list
823 */
824 list_add(&sh->batch_list, &head->batch_list);
825 spin_unlock(&head->batch_head->batch_lock);
826
827 sh->batch_head = head->batch_head;
828 } else {
829 head->batch_head = head;
830 sh->batch_head = head->batch_head;
831 spin_lock(&head->batch_lock);
832 list_add_tail(&sh->batch_list, &head->batch_list);
833 spin_unlock(&head->batch_lock);
834 }
835
836 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
837 if (atomic_dec_return(&conf->preread_active_stripes)
838 < IO_THRESHOLD)
839 md_wakeup_thread(conf->mddev->thread);
840
NeilBrown2b6b2452015-05-21 15:10:01 +1000841 if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) {
842 int seq = sh->bm_seq;
843 if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) &&
844 sh->batch_head->bm_seq > seq)
845 seq = sh->batch_head->bm_seq;
846 set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state);
847 sh->batch_head->bm_seq = seq;
848 }
849
shli@kernel.org59fc6302014-12-15 12:57:03 +1100850 atomic_inc(&sh->count);
851unlock_out:
852 unlock_two_stripes(head, sh);
853out:
Shaohua Li6d036f72015-08-13 14:31:57 -0700854 raid5_release_stripe(head);
shli@kernel.org59fc6302014-12-15 12:57:03 +1100855}
856
NeilBrown05616be2012-05-21 09:27:00 +1000857/* Determine if 'data_offset' or 'new_data_offset' should be used
858 * in this stripe_head.
859 */
860static int use_new_offset(struct r5conf *conf, struct stripe_head *sh)
861{
862 sector_t progress = conf->reshape_progress;
863 /* Need a memory barrier to make sure we see the value
864 * of conf->generation, or ->data_offset that was set before
865 * reshape_progress was updated.
866 */
867 smp_rmb();
868 if (progress == MaxSector)
869 return 0;
870 if (sh->generation == conf->generation - 1)
871 return 0;
872 /* We are in a reshape, and this is a new-generation stripe,
873 * so use new_data_offset.
874 */
875 return 1;
876}
877
NeilBrown6712ecf2007-09-27 12:47:43 +0200878static void
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200879raid5_end_read_request(struct bio *bi);
NeilBrown6712ecf2007-09-27 12:47:43 +0200880static void
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200881raid5_end_write_request(struct bio *bi);
Dan Williams91c00922007-01-02 13:52:30 -0700882
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000883static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
Dan Williams91c00922007-01-02 13:52:30 -0700884{
NeilBrownd1688a62011-10-11 16:49:52 +1100885 struct r5conf *conf = sh->raid_conf;
Dan Williams91c00922007-01-02 13:52:30 -0700886 int i, disks = sh->disks;
shli@kernel.org59fc6302014-12-15 12:57:03 +1100887 struct stripe_head *head_sh = sh;
Dan Williams91c00922007-01-02 13:52:30 -0700888
889 might_sleep();
890
Shaohua Lif6bed0e2015-08-13 14:31:59 -0700891 if (r5l_write_stripe(conf->log, sh) == 0)
892 return;
Dan Williams91c00922007-01-02 13:52:30 -0700893 for (i = disks; i--; ) {
894 int rw;
NeilBrown9a3e1102011-12-23 10:17:53 +1100895 int replace_only = 0;
NeilBrown977df362011-12-23 10:17:53 +1100896 struct bio *bi, *rbi;
897 struct md_rdev *rdev, *rrdev = NULL;
shli@kernel.org59fc6302014-12-15 12:57:03 +1100898
899 sh = head_sh;
Tejun Heoe9c74692010-09-03 11:56:18 +0200900 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
901 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
902 rw = WRITE_FUA;
903 else
904 rw = WRITE;
Shaohua Li9e4447682012-10-11 13:49:49 +1100905 if (test_bit(R5_Discard, &sh->dev[i].flags))
Shaohua Li620125f2012-10-11 13:49:05 +1100906 rw |= REQ_DISCARD;
Tejun Heoe9c74692010-09-03 11:56:18 +0200907 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
Dan Williams91c00922007-01-02 13:52:30 -0700908 rw = READ;
NeilBrown9a3e1102011-12-23 10:17:53 +1100909 else if (test_and_clear_bit(R5_WantReplace,
910 &sh->dev[i].flags)) {
911 rw = WRITE;
912 replace_only = 1;
913 } else
Dan Williams91c00922007-01-02 13:52:30 -0700914 continue;
Shaohua Libc0934f2012-05-22 13:55:05 +1000915 if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags))
916 rw |= REQ_SYNC;
Dan Williams91c00922007-01-02 13:52:30 -0700917
shli@kernel.org59fc6302014-12-15 12:57:03 +1100918again:
Dan Williams91c00922007-01-02 13:52:30 -0700919 bi = &sh->dev[i].req;
NeilBrown977df362011-12-23 10:17:53 +1100920 rbi = &sh->dev[i].rreq; /* For writing to replacement */
Dan Williams91c00922007-01-02 13:52:30 -0700921
Dan Williams91c00922007-01-02 13:52:30 -0700922 rcu_read_lock();
NeilBrown9a3e1102011-12-23 10:17:53 +1100923 rrdev = rcu_dereference(conf->disks[i].replacement);
NeilBrowndd054fc2011-12-23 10:17:53 +1100924 smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */
925 rdev = rcu_dereference(conf->disks[i].rdev);
926 if (!rdev) {
927 rdev = rrdev;
928 rrdev = NULL;
929 }
NeilBrown9a3e1102011-12-23 10:17:53 +1100930 if (rw & WRITE) {
931 if (replace_only)
932 rdev = NULL;
NeilBrowndd054fc2011-12-23 10:17:53 +1100933 if (rdev == rrdev)
934 /* We raced and saw duplicates */
935 rrdev = NULL;
NeilBrown9a3e1102011-12-23 10:17:53 +1100936 } else {
shli@kernel.org59fc6302014-12-15 12:57:03 +1100937 if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev)
NeilBrown9a3e1102011-12-23 10:17:53 +1100938 rdev = rrdev;
939 rrdev = NULL;
940 }
NeilBrown977df362011-12-23 10:17:53 +1100941
Dan Williams91c00922007-01-02 13:52:30 -0700942 if (rdev && test_bit(Faulty, &rdev->flags))
943 rdev = NULL;
944 if (rdev)
945 atomic_inc(&rdev->nr_pending);
NeilBrown977df362011-12-23 10:17:53 +1100946 if (rrdev && test_bit(Faulty, &rrdev->flags))
947 rrdev = NULL;
948 if (rrdev)
949 atomic_inc(&rrdev->nr_pending);
Dan Williams91c00922007-01-02 13:52:30 -0700950 rcu_read_unlock();
951
NeilBrown73e92e52011-07-28 11:39:22 +1000952 /* We have already checked bad blocks for reads. Now
NeilBrown977df362011-12-23 10:17:53 +1100953 * need to check for writes. We never accept write errors
954 * on the replacement, so we don't to check rrdev.
NeilBrown73e92e52011-07-28 11:39:22 +1000955 */
956 while ((rw & WRITE) && rdev &&
957 test_bit(WriteErrorSeen, &rdev->flags)) {
958 sector_t first_bad;
959 int bad_sectors;
960 int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
961 &first_bad, &bad_sectors);
962 if (!bad)
963 break;
964
965 if (bad < 0) {
966 set_bit(BlockedBadBlocks, &rdev->flags);
967 if (!conf->mddev->external &&
968 conf->mddev->flags) {
969 /* It is very unlikely, but we might
970 * still need to write out the
971 * bad block log - better give it
972 * a chance*/
973 md_check_recovery(conf->mddev);
974 }
majianpeng18507532012-07-03 12:11:54 +1000975 /*
976 * Because md_wait_for_blocked_rdev
977 * will dec nr_pending, we must
978 * increment it first.
979 */
980 atomic_inc(&rdev->nr_pending);
NeilBrown73e92e52011-07-28 11:39:22 +1000981 md_wait_for_blocked_rdev(rdev, conf->mddev);
982 } else {
983 /* Acknowledged bad block - skip the write */
984 rdev_dec_pending(rdev, conf->mddev);
985 rdev = NULL;
986 }
987 }
988
Dan Williams91c00922007-01-02 13:52:30 -0700989 if (rdev) {
NeilBrown9a3e1102011-12-23 10:17:53 +1100990 if (s->syncing || s->expanding || s->expanded
991 || s->replacing)
Dan Williams91c00922007-01-02 13:52:30 -0700992 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
993
Dan Williams2b7497f2008-06-28 08:31:52 +1000994 set_bit(STRIPE_IO_STARTED, &sh->state);
995
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700996 bio_reset(bi);
Dan Williams91c00922007-01-02 13:52:30 -0700997 bi->bi_bdev = rdev->bdev;
Kent Overstreet2f6db2a2012-09-11 12:26:38 -0700998 bi->bi_rw = rw;
999 bi->bi_end_io = (rw & WRITE)
1000 ? raid5_end_write_request
1001 : raid5_end_read_request;
1002 bi->bi_private = sh;
1003
Dan Williams91c00922007-01-02 13:52:30 -07001004 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07001005 __func__, (unsigned long long)sh->sector,
Dan Williams91c00922007-01-02 13:52:30 -07001006 bi->bi_rw, i);
1007 atomic_inc(&sh->count);
shli@kernel.org59fc6302014-12-15 12:57:03 +11001008 if (sh != head_sh)
1009 atomic_inc(&head_sh->count);
NeilBrown05616be2012-05-21 09:27:00 +10001010 if (use_new_offset(conf, sh))
Kent Overstreet4f024f32013-10-11 15:44:27 -07001011 bi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +10001012 + rdev->new_data_offset);
1013 else
Kent Overstreet4f024f32013-10-11 15:44:27 -07001014 bi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +10001015 + rdev->data_offset);
shli@kernel.org59fc6302014-12-15 12:57:03 +11001016 if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags))
majianpenge59aa232013-11-14 15:16:19 +11001017 bi->bi_rw |= REQ_NOMERGE;
majianpeng3f9e7c12012-07-31 10:04:21 +10001018
Shaohua Lid592a992014-05-21 17:57:44 +08001019 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1020 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1021 sh->dev[i].vec.bv_page = sh->dev[i].page;
Kent Overstreet4997b722013-05-30 08:44:39 +02001022 bi->bi_vcnt = 1;
Dan Williams91c00922007-01-02 13:52:30 -07001023 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1024 bi->bi_io_vec[0].bv_offset = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001025 bi->bi_iter.bi_size = STRIPE_SIZE;
Shaohua Li37c61ff2013-10-19 14:50:28 +08001026 /*
1027 * If this is discard request, set bi_vcnt 0. We don't
1028 * want to confuse SCSI because SCSI will replace payload
1029 */
1030 if (rw & REQ_DISCARD)
1031 bi->bi_vcnt = 0;
NeilBrown977df362011-12-23 10:17:53 +11001032 if (rrdev)
1033 set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags);
Jonathan Brassowe3620a32013-03-07 16:22:01 -06001034
1035 if (conf->mddev->gendisk)
1036 trace_block_bio_remap(bdev_get_queue(bi->bi_bdev),
1037 bi, disk_devt(conf->mddev->gendisk),
1038 sh->dev[i].sector);
Dan Williams91c00922007-01-02 13:52:30 -07001039 generic_make_request(bi);
NeilBrown977df362011-12-23 10:17:53 +11001040 }
1041 if (rrdev) {
NeilBrown9a3e1102011-12-23 10:17:53 +11001042 if (s->syncing || s->expanding || s->expanded
1043 || s->replacing)
NeilBrown977df362011-12-23 10:17:53 +11001044 md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
1045
1046 set_bit(STRIPE_IO_STARTED, &sh->state);
1047
Kent Overstreet2f6db2a2012-09-11 12:26:38 -07001048 bio_reset(rbi);
NeilBrown977df362011-12-23 10:17:53 +11001049 rbi->bi_bdev = rrdev->bdev;
Kent Overstreet2f6db2a2012-09-11 12:26:38 -07001050 rbi->bi_rw = rw;
1051 BUG_ON(!(rw & WRITE));
1052 rbi->bi_end_io = raid5_end_write_request;
1053 rbi->bi_private = sh;
1054
NeilBrown977df362011-12-23 10:17:53 +11001055 pr_debug("%s: for %llu schedule op %ld on "
1056 "replacement disc %d\n",
1057 __func__, (unsigned long long)sh->sector,
1058 rbi->bi_rw, i);
1059 atomic_inc(&sh->count);
shli@kernel.org59fc6302014-12-15 12:57:03 +11001060 if (sh != head_sh)
1061 atomic_inc(&head_sh->count);
NeilBrown05616be2012-05-21 09:27:00 +10001062 if (use_new_offset(conf, sh))
Kent Overstreet4f024f32013-10-11 15:44:27 -07001063 rbi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +10001064 + rrdev->new_data_offset);
1065 else
Kent Overstreet4f024f32013-10-11 15:44:27 -07001066 rbi->bi_iter.bi_sector = (sh->sector
NeilBrown05616be2012-05-21 09:27:00 +10001067 + rrdev->data_offset);
Shaohua Lid592a992014-05-21 17:57:44 +08001068 if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1069 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1070 sh->dev[i].rvec.bv_page = sh->dev[i].page;
Kent Overstreet4997b722013-05-30 08:44:39 +02001071 rbi->bi_vcnt = 1;
NeilBrown977df362011-12-23 10:17:53 +11001072 rbi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1073 rbi->bi_io_vec[0].bv_offset = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001074 rbi->bi_iter.bi_size = STRIPE_SIZE;
Shaohua Li37c61ff2013-10-19 14:50:28 +08001075 /*
1076 * If this is discard request, set bi_vcnt 0. We don't
1077 * want to confuse SCSI because SCSI will replace payload
1078 */
1079 if (rw & REQ_DISCARD)
1080 rbi->bi_vcnt = 0;
Jonathan Brassowe3620a32013-03-07 16:22:01 -06001081 if (conf->mddev->gendisk)
1082 trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev),
1083 rbi, disk_devt(conf->mddev->gendisk),
1084 sh->dev[i].sector);
NeilBrown977df362011-12-23 10:17:53 +11001085 generic_make_request(rbi);
1086 }
1087 if (!rdev && !rrdev) {
Namhyung Kimb0629622011-06-14 14:20:19 +10001088 if (rw & WRITE)
Dan Williams91c00922007-01-02 13:52:30 -07001089 set_bit(STRIPE_DEGRADED, &sh->state);
1090 pr_debug("skip op %ld on disc %d for sector %llu\n",
1091 bi->bi_rw, i, (unsigned long long)sh->sector);
1092 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1093 set_bit(STRIPE_HANDLE, &sh->state);
1094 }
shli@kernel.org59fc6302014-12-15 12:57:03 +11001095
1096 if (!head_sh->batch_head)
1097 continue;
1098 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1099 batch_list);
1100 if (sh != head_sh)
1101 goto again;
Dan Williams91c00922007-01-02 13:52:30 -07001102 }
1103}
1104
1105static struct dma_async_tx_descriptor *
Shaohua Lid592a992014-05-21 17:57:44 +08001106async_copy_data(int frombio, struct bio *bio, struct page **page,
1107 sector_t sector, struct dma_async_tx_descriptor *tx,
1108 struct stripe_head *sh)
Dan Williams91c00922007-01-02 13:52:30 -07001109{
Kent Overstreet79886132013-11-23 17:19:00 -08001110 struct bio_vec bvl;
1111 struct bvec_iter iter;
Dan Williams91c00922007-01-02 13:52:30 -07001112 struct page *bio_page;
Dan Williams91c00922007-01-02 13:52:30 -07001113 int page_offset;
Dan Williamsa08abd82009-06-03 11:43:59 -07001114 struct async_submit_ctl submit;
Dan Williams0403e382009-09-08 17:42:50 -07001115 enum async_tx_flags flags = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001116
Kent Overstreet4f024f32013-10-11 15:44:27 -07001117 if (bio->bi_iter.bi_sector >= sector)
1118 page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512;
Dan Williams91c00922007-01-02 13:52:30 -07001119 else
Kent Overstreet4f024f32013-10-11 15:44:27 -07001120 page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512;
Dan Williamsa08abd82009-06-03 11:43:59 -07001121
Dan Williams0403e382009-09-08 17:42:50 -07001122 if (frombio)
1123 flags |= ASYNC_TX_FENCE;
1124 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
1125
Kent Overstreet79886132013-11-23 17:19:00 -08001126 bio_for_each_segment(bvl, bio, iter) {
1127 int len = bvl.bv_len;
Dan Williams91c00922007-01-02 13:52:30 -07001128 int clen;
1129 int b_offset = 0;
1130
1131 if (page_offset < 0) {
1132 b_offset = -page_offset;
1133 page_offset += b_offset;
1134 len -= b_offset;
1135 }
1136
1137 if (len > 0 && page_offset + len > STRIPE_SIZE)
1138 clen = STRIPE_SIZE - page_offset;
1139 else
1140 clen = len;
1141
1142 if (clen > 0) {
Kent Overstreet79886132013-11-23 17:19:00 -08001143 b_offset += bvl.bv_offset;
1144 bio_page = bvl.bv_page;
Shaohua Lid592a992014-05-21 17:57:44 +08001145 if (frombio) {
1146 if (sh->raid_conf->skip_copy &&
1147 b_offset == 0 && page_offset == 0 &&
1148 clen == STRIPE_SIZE)
1149 *page = bio_page;
1150 else
1151 tx = async_memcpy(*page, bio_page, page_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -07001152 b_offset, clen, &submit);
Shaohua Lid592a992014-05-21 17:57:44 +08001153 } else
1154 tx = async_memcpy(bio_page, *page, b_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -07001155 page_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001156 }
Dan Williamsa08abd82009-06-03 11:43:59 -07001157 /* chain the operations */
1158 submit.depend_tx = tx;
1159
Dan Williams91c00922007-01-02 13:52:30 -07001160 if (clen < len) /* hit end of page */
1161 break;
1162 page_offset += len;
1163 }
1164
1165 return tx;
1166}
1167
1168static void ops_complete_biofill(void *stripe_head_ref)
1169{
1170 struct stripe_head *sh = stripe_head_ref;
NeilBrown34a6f802015-08-14 12:07:57 +10001171 struct bio_list return_bi = BIO_EMPTY_LIST;
Dan Williamse4d84902007-09-24 10:06:13 -07001172 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001173
Harvey Harrisone46b2722008-04-28 02:15:50 -07001174 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001175 (unsigned long long)sh->sector);
1176
1177 /* clear completed biofills */
1178 for (i = sh->disks; i--; ) {
1179 struct r5dev *dev = &sh->dev[i];
Dan Williams91c00922007-01-02 13:52:30 -07001180
1181 /* acknowledge completion of a biofill operation */
Dan Williamse4d84902007-09-24 10:06:13 -07001182 /* and check if we need to reply to a read request,
1183 * new R5_Wantfill requests are held off until
Dan Williams83de75c2008-06-28 08:31:58 +10001184 * !STRIPE_BIOFILL_RUN
Dan Williamse4d84902007-09-24 10:06:13 -07001185 */
1186 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001187 struct bio *rbi, *rbi2;
Dan Williams91c00922007-01-02 13:52:30 -07001188
Dan Williams91c00922007-01-02 13:52:30 -07001189 BUG_ON(!dev->read);
1190 rbi = dev->read;
1191 dev->read = NULL;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001192 while (rbi && rbi->bi_iter.bi_sector <
Dan Williams91c00922007-01-02 13:52:30 -07001193 dev->sector + STRIPE_SECTORS) {
1194 rbi2 = r5_next_bio(rbi, dev->sector);
NeilBrown34a6f802015-08-14 12:07:57 +10001195 if (!raid5_dec_bi_active_stripes(rbi))
1196 bio_list_add(&return_bi, rbi);
Dan Williams91c00922007-01-02 13:52:30 -07001197 rbi = rbi2;
1198 }
1199 }
1200 }
Dan Williams83de75c2008-06-28 08:31:58 +10001201 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -07001202
NeilBrown34a6f802015-08-14 12:07:57 +10001203 return_io(&return_bi);
Dan Williams91c00922007-01-02 13:52:30 -07001204
Dan Williamse4d84902007-09-24 10:06:13 -07001205 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07001206 raid5_release_stripe(sh);
Dan Williams91c00922007-01-02 13:52:30 -07001207}
1208
1209static void ops_run_biofill(struct stripe_head *sh)
1210{
1211 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa08abd82009-06-03 11:43:59 -07001212 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001213 int i;
1214
shli@kernel.org59fc6302014-12-15 12:57:03 +11001215 BUG_ON(sh->batch_head);
Harvey Harrisone46b2722008-04-28 02:15:50 -07001216 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001217 (unsigned long long)sh->sector);
1218
1219 for (i = sh->disks; i--; ) {
1220 struct r5dev *dev = &sh->dev[i];
1221 if (test_bit(R5_Wantfill, &dev->flags)) {
1222 struct bio *rbi;
Shaohua Lib17459c2012-07-19 16:01:31 +10001223 spin_lock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001224 dev->read = rbi = dev->toread;
1225 dev->toread = NULL;
Shaohua Lib17459c2012-07-19 16:01:31 +10001226 spin_unlock_irq(&sh->stripe_lock);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001227 while (rbi && rbi->bi_iter.bi_sector <
Dan Williams91c00922007-01-02 13:52:30 -07001228 dev->sector + STRIPE_SECTORS) {
Shaohua Lid592a992014-05-21 17:57:44 +08001229 tx = async_copy_data(0, rbi, &dev->page,
1230 dev->sector, tx, sh);
Dan Williams91c00922007-01-02 13:52:30 -07001231 rbi = r5_next_bio(rbi, dev->sector);
1232 }
1233 }
1234 }
1235
1236 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001237 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
1238 async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001239}
1240
Dan Williams4e7d2c02009-08-29 19:13:11 -07001241static void mark_target_uptodate(struct stripe_head *sh, int target)
1242{
1243 struct r5dev *tgt;
1244
1245 if (target < 0)
1246 return;
1247
1248 tgt = &sh->dev[target];
1249 set_bit(R5_UPTODATE, &tgt->flags);
1250 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1251 clear_bit(R5_Wantcompute, &tgt->flags);
1252}
1253
Dan Williamsac6b53b2009-07-14 13:40:19 -07001254static void ops_complete_compute(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001255{
1256 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001257
Harvey Harrisone46b2722008-04-28 02:15:50 -07001258 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001259 (unsigned long long)sh->sector);
1260
Dan Williamsac6b53b2009-07-14 13:40:19 -07001261 /* mark the computed target(s) as uptodate */
Dan Williams4e7d2c02009-08-29 19:13:11 -07001262 mark_target_uptodate(sh, sh->ops.target);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001263 mark_target_uptodate(sh, sh->ops.target2);
Dan Williams4e7d2c02009-08-29 19:13:11 -07001264
Dan Williamsecc65c92008-06-28 08:31:57 +10001265 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
1266 if (sh->check_state == check_state_compute_run)
1267 sh->check_state = check_state_compute_result;
Dan Williams91c00922007-01-02 13:52:30 -07001268 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07001269 raid5_release_stripe(sh);
Dan Williams91c00922007-01-02 13:52:30 -07001270}
1271
Dan Williamsd6f38f32009-07-14 11:50:52 -07001272/* return a pointer to the address conversion region of the scribble buffer */
1273static addr_conv_t *to_addr_conv(struct stripe_head *sh,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001274 struct raid5_percpu *percpu, int i)
Dan Williams91c00922007-01-02 13:52:30 -07001275{
shli@kernel.org46d5b782014-12-15 12:57:02 +11001276 void *addr;
1277
1278 addr = flex_array_get(percpu->scribble, i);
1279 return addr + sizeof(struct page *) * (sh->disks + 2);
1280}
1281
1282/* return a pointer to the address conversion region of the scribble buffer */
1283static struct page **to_addr_page(struct raid5_percpu *percpu, int i)
1284{
1285 void *addr;
1286
1287 addr = flex_array_get(percpu->scribble, i);
1288 return addr;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001289}
1290
1291static struct dma_async_tx_descriptor *
1292ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
1293{
Dan Williams91c00922007-01-02 13:52:30 -07001294 int disks = sh->disks;
shli@kernel.org46d5b782014-12-15 12:57:02 +11001295 struct page **xor_srcs = to_addr_page(percpu, 0);
Dan Williams91c00922007-01-02 13:52:30 -07001296 int target = sh->ops.target;
1297 struct r5dev *tgt = &sh->dev[target];
1298 struct page *xor_dest = tgt->page;
1299 int count = 0;
1300 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001301 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001302 int i;
1303
shli@kernel.org59fc6302014-12-15 12:57:03 +11001304 BUG_ON(sh->batch_head);
1305
Dan Williams91c00922007-01-02 13:52:30 -07001306 pr_debug("%s: stripe %llu block: %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07001307 __func__, (unsigned long long)sh->sector, target);
Dan Williams91c00922007-01-02 13:52:30 -07001308 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1309
1310 for (i = disks; i--; )
1311 if (i != target)
1312 xor_srcs[count++] = sh->dev[i].page;
1313
1314 atomic_inc(&sh->count);
1315
Dan Williams0403e382009-09-08 17:42:50 -07001316 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001317 ops_complete_compute, sh, to_addr_conv(sh, percpu, 0));
Dan Williams91c00922007-01-02 13:52:30 -07001318 if (unlikely(count == 1))
Dan Williamsa08abd82009-06-03 11:43:59 -07001319 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001320 else
Dan Williamsa08abd82009-06-03 11:43:59 -07001321 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001322
Dan Williams91c00922007-01-02 13:52:30 -07001323 return tx;
1324}
1325
Dan Williamsac6b53b2009-07-14 13:40:19 -07001326/* set_syndrome_sources - populate source buffers for gen_syndrome
1327 * @srcs - (struct page *) array of size sh->disks
1328 * @sh - stripe_head to parse
1329 *
1330 * Populates srcs in proper layout order for the stripe and returns the
1331 * 'count' of sources to be used in a call to async_gen_syndrome. The P
1332 * destination buffer is recorded in srcs[count] and the Q destination
1333 * is recorded in srcs[count+1]].
1334 */
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001335static int set_syndrome_sources(struct page **srcs,
1336 struct stripe_head *sh,
1337 int srctype)
Dan Williamsac6b53b2009-07-14 13:40:19 -07001338{
1339 int disks = sh->disks;
1340 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
1341 int d0_idx = raid6_d0(sh);
1342 int count;
1343 int i;
1344
1345 for (i = 0; i < disks; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +11001346 srcs[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001347
1348 count = 0;
1349 i = d0_idx;
1350 do {
1351 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001352 struct r5dev *dev = &sh->dev[i];
Dan Williamsac6b53b2009-07-14 13:40:19 -07001353
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001354 if (i == sh->qd_idx || i == sh->pd_idx ||
1355 (srctype == SYNDROME_SRC_ALL) ||
1356 (srctype == SYNDROME_SRC_WANT_DRAIN &&
1357 test_bit(R5_Wantdrain, &dev->flags)) ||
1358 (srctype == SYNDROME_SRC_WRITTEN &&
1359 dev->written))
1360 srcs[slot] = sh->dev[i].page;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001361 i = raid6_next_disk(i, disks);
1362 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001363
NeilBrowne4424fe2009-10-16 16:27:34 +11001364 return syndrome_disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001365}
1366
1367static struct dma_async_tx_descriptor *
1368ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
1369{
1370 int disks = sh->disks;
shli@kernel.org46d5b782014-12-15 12:57:02 +11001371 struct page **blocks = to_addr_page(percpu, 0);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001372 int target;
1373 int qd_idx = sh->qd_idx;
1374 struct dma_async_tx_descriptor *tx;
1375 struct async_submit_ctl submit;
1376 struct r5dev *tgt;
1377 struct page *dest;
1378 int i;
1379 int count;
1380
shli@kernel.org59fc6302014-12-15 12:57:03 +11001381 BUG_ON(sh->batch_head);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001382 if (sh->ops.target < 0)
1383 target = sh->ops.target2;
1384 else if (sh->ops.target2 < 0)
1385 target = sh->ops.target;
1386 else
1387 /* we should only have one valid target */
1388 BUG();
1389 BUG_ON(target < 0);
1390 pr_debug("%s: stripe %llu block: %d\n",
1391 __func__, (unsigned long long)sh->sector, target);
1392
1393 tgt = &sh->dev[target];
1394 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1395 dest = tgt->page;
1396
1397 atomic_inc(&sh->count);
1398
1399 if (target == qd_idx) {
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001400 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001401 blocks[count] = NULL; /* regenerating p is not necessary */
1402 BUG_ON(blocks[count+1] != dest); /* q should already be set */
Dan Williams0403e382009-09-08 17:42:50 -07001403 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1404 ops_complete_compute, sh,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001405 to_addr_conv(sh, percpu, 0));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001406 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1407 } else {
1408 /* Compute any data- or p-drive using XOR */
1409 count = 0;
1410 for (i = disks; i-- ; ) {
1411 if (i == target || i == qd_idx)
1412 continue;
1413 blocks[count++] = sh->dev[i].page;
1414 }
1415
Dan Williams0403e382009-09-08 17:42:50 -07001416 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1417 NULL, ops_complete_compute, sh,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001418 to_addr_conv(sh, percpu, 0));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001419 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
1420 }
1421
1422 return tx;
1423}
1424
1425static struct dma_async_tx_descriptor *
1426ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
1427{
1428 int i, count, disks = sh->disks;
1429 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
1430 int d0_idx = raid6_d0(sh);
1431 int faila = -1, failb = -1;
1432 int target = sh->ops.target;
1433 int target2 = sh->ops.target2;
1434 struct r5dev *tgt = &sh->dev[target];
1435 struct r5dev *tgt2 = &sh->dev[target2];
1436 struct dma_async_tx_descriptor *tx;
shli@kernel.org46d5b782014-12-15 12:57:02 +11001437 struct page **blocks = to_addr_page(percpu, 0);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001438 struct async_submit_ctl submit;
1439
shli@kernel.org59fc6302014-12-15 12:57:03 +11001440 BUG_ON(sh->batch_head);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001441 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
1442 __func__, (unsigned long long)sh->sector, target, target2);
1443 BUG_ON(target < 0 || target2 < 0);
1444 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
1445 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
1446
Dan Williams6c910a72009-09-16 12:24:54 -07001447 /* we need to open-code set_syndrome_sources to handle the
Dan Williamsac6b53b2009-07-14 13:40:19 -07001448 * slot number conversion for 'faila' and 'failb'
1449 */
1450 for (i = 0; i < disks ; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +11001451 blocks[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001452 count = 0;
1453 i = d0_idx;
1454 do {
1455 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
1456
1457 blocks[slot] = sh->dev[i].page;
1458
1459 if (i == target)
1460 faila = slot;
1461 if (i == target2)
1462 failb = slot;
1463 i = raid6_next_disk(i, disks);
1464 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001465
1466 BUG_ON(faila == failb);
1467 if (failb < faila)
1468 swap(faila, failb);
1469 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
1470 __func__, (unsigned long long)sh->sector, faila, failb);
1471
1472 atomic_inc(&sh->count);
1473
1474 if (failb == syndrome_disks+1) {
1475 /* Q disk is one of the missing disks */
1476 if (faila == syndrome_disks) {
1477 /* Missing P+Q, just recompute */
Dan Williams0403e382009-09-08 17:42:50 -07001478 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1479 ops_complete_compute, sh,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001480 to_addr_conv(sh, percpu, 0));
NeilBrowne4424fe2009-10-16 16:27:34 +11001481 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
Dan Williamsac6b53b2009-07-14 13:40:19 -07001482 STRIPE_SIZE, &submit);
1483 } else {
1484 struct page *dest;
1485 int data_target;
1486 int qd_idx = sh->qd_idx;
1487
1488 /* Missing D+Q: recompute D from P, then recompute Q */
1489 if (target == qd_idx)
1490 data_target = target2;
1491 else
1492 data_target = target;
1493
1494 count = 0;
1495 for (i = disks; i-- ; ) {
1496 if (i == data_target || i == qd_idx)
1497 continue;
1498 blocks[count++] = sh->dev[i].page;
1499 }
1500 dest = sh->dev[data_target].page;
Dan Williams0403e382009-09-08 17:42:50 -07001501 init_async_submit(&submit,
1502 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
1503 NULL, NULL, NULL,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001504 to_addr_conv(sh, percpu, 0));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001505 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
1506 &submit);
1507
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001508 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL);
Dan Williams0403e382009-09-08 17:42:50 -07001509 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
1510 ops_complete_compute, sh,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001511 to_addr_conv(sh, percpu, 0));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001512 return async_gen_syndrome(blocks, 0, count+2,
1513 STRIPE_SIZE, &submit);
1514 }
Dan Williamsac6b53b2009-07-14 13:40:19 -07001515 } else {
Dan Williams6c910a72009-09-16 12:24:54 -07001516 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
1517 ops_complete_compute, sh,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001518 to_addr_conv(sh, percpu, 0));
Dan Williams6c910a72009-09-16 12:24:54 -07001519 if (failb == syndrome_disks) {
1520 /* We're missing D+P. */
1521 return async_raid6_datap_recov(syndrome_disks+2,
1522 STRIPE_SIZE, faila,
1523 blocks, &submit);
1524 } else {
1525 /* We're missing D+D. */
1526 return async_raid6_2data_recov(syndrome_disks+2,
1527 STRIPE_SIZE, faila, failb,
1528 blocks, &submit);
1529 }
Dan Williamsac6b53b2009-07-14 13:40:19 -07001530 }
1531}
1532
Dan Williams91c00922007-01-02 13:52:30 -07001533static void ops_complete_prexor(void *stripe_head_ref)
1534{
1535 struct stripe_head *sh = stripe_head_ref;
1536
Harvey Harrisone46b2722008-04-28 02:15:50 -07001537 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001538 (unsigned long long)sh->sector);
Dan Williams91c00922007-01-02 13:52:30 -07001539}
1540
1541static struct dma_async_tx_descriptor *
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001542ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu,
1543 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001544{
Dan Williams91c00922007-01-02 13:52:30 -07001545 int disks = sh->disks;
shli@kernel.org46d5b782014-12-15 12:57:02 +11001546 struct page **xor_srcs = to_addr_page(percpu, 0);
Dan Williams91c00922007-01-02 13:52:30 -07001547 int count = 0, pd_idx = sh->pd_idx, i;
Dan Williamsa08abd82009-06-03 11:43:59 -07001548 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001549
1550 /* existing parity data subtracted */
1551 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1552
shli@kernel.org59fc6302014-12-15 12:57:03 +11001553 BUG_ON(sh->batch_head);
Harvey Harrisone46b2722008-04-28 02:15:50 -07001554 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001555 (unsigned long long)sh->sector);
1556
1557 for (i = disks; i--; ) {
1558 struct r5dev *dev = &sh->dev[i];
1559 /* Only process blocks that are known to be uptodate */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001560 if (test_bit(R5_Wantdrain, &dev->flags))
Dan Williams91c00922007-01-02 13:52:30 -07001561 xor_srcs[count++] = dev->page;
1562 }
1563
Dan Williams0403e382009-09-08 17:42:50 -07001564 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001565 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
Dan Williamsa08abd82009-06-03 11:43:59 -07001566 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001567
1568 return tx;
1569}
1570
1571static struct dma_async_tx_descriptor *
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001572ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu,
1573 struct dma_async_tx_descriptor *tx)
1574{
1575 struct page **blocks = to_addr_page(percpu, 0);
1576 int count;
1577 struct async_submit_ctl submit;
1578
1579 pr_debug("%s: stripe %llu\n", __func__,
1580 (unsigned long long)sh->sector);
1581
1582 count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN);
1583
1584 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx,
1585 ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0));
1586 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
1587
1588 return tx;
1589}
1590
1591static struct dma_async_tx_descriptor *
Dan Williamsd8ee0722008-06-28 08:32:06 +10001592ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001593{
1594 int disks = sh->disks;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001595 int i;
shli@kernel.org59fc6302014-12-15 12:57:03 +11001596 struct stripe_head *head_sh = sh;
Dan Williams91c00922007-01-02 13:52:30 -07001597
Harvey Harrisone46b2722008-04-28 02:15:50 -07001598 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001599 (unsigned long long)sh->sector);
1600
1601 for (i = disks; i--; ) {
shli@kernel.org59fc6302014-12-15 12:57:03 +11001602 struct r5dev *dev;
Dan Williams91c00922007-01-02 13:52:30 -07001603 struct bio *chosen;
Dan Williams91c00922007-01-02 13:52:30 -07001604
shli@kernel.org59fc6302014-12-15 12:57:03 +11001605 sh = head_sh;
1606 if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001607 struct bio *wbi;
1608
shli@kernel.org59fc6302014-12-15 12:57:03 +11001609again:
1610 dev = &sh->dev[i];
Shaohua Lib17459c2012-07-19 16:01:31 +10001611 spin_lock_irq(&sh->stripe_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001612 chosen = dev->towrite;
1613 dev->towrite = NULL;
shli@kernel.org7a87f432014-12-15 12:57:03 +11001614 sh->overwrite_disks = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001615 BUG_ON(dev->written);
1616 wbi = dev->written = chosen;
Shaohua Lib17459c2012-07-19 16:01:31 +10001617 spin_unlock_irq(&sh->stripe_lock);
Shaohua Lid592a992014-05-21 17:57:44 +08001618 WARN_ON(dev->page != dev->orig_page);
Dan Williams91c00922007-01-02 13:52:30 -07001619
Kent Overstreet4f024f32013-10-11 15:44:27 -07001620 while (wbi && wbi->bi_iter.bi_sector <
Dan Williams91c00922007-01-02 13:52:30 -07001621 dev->sector + STRIPE_SECTORS) {
Tejun Heoe9c74692010-09-03 11:56:18 +02001622 if (wbi->bi_rw & REQ_FUA)
1623 set_bit(R5_WantFUA, &dev->flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001624 if (wbi->bi_rw & REQ_SYNC)
1625 set_bit(R5_SyncIO, &dev->flags);
Shaohua Li9e4447682012-10-11 13:49:49 +11001626 if (wbi->bi_rw & REQ_DISCARD)
Shaohua Li620125f2012-10-11 13:49:05 +11001627 set_bit(R5_Discard, &dev->flags);
Shaohua Lid592a992014-05-21 17:57:44 +08001628 else {
1629 tx = async_copy_data(1, wbi, &dev->page,
1630 dev->sector, tx, sh);
1631 if (dev->page != dev->orig_page) {
1632 set_bit(R5_SkipCopy, &dev->flags);
1633 clear_bit(R5_UPTODATE, &dev->flags);
1634 clear_bit(R5_OVERWRITE, &dev->flags);
1635 }
1636 }
Dan Williams91c00922007-01-02 13:52:30 -07001637 wbi = r5_next_bio(wbi, dev->sector);
1638 }
shli@kernel.org59fc6302014-12-15 12:57:03 +11001639
1640 if (head_sh->batch_head) {
1641 sh = list_first_entry(&sh->batch_list,
1642 struct stripe_head,
1643 batch_list);
1644 if (sh == head_sh)
1645 continue;
1646 goto again;
1647 }
Dan Williams91c00922007-01-02 13:52:30 -07001648 }
1649 }
1650
1651 return tx;
1652}
1653
Dan Williamsac6b53b2009-07-14 13:40:19 -07001654static void ops_complete_reconstruct(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001655{
1656 struct stripe_head *sh = stripe_head_ref;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001657 int disks = sh->disks;
1658 int pd_idx = sh->pd_idx;
1659 int qd_idx = sh->qd_idx;
1660 int i;
Shaohua Li9e4447682012-10-11 13:49:49 +11001661 bool fua = false, sync = false, discard = false;
Dan Williams91c00922007-01-02 13:52:30 -07001662
Harvey Harrisone46b2722008-04-28 02:15:50 -07001663 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001664 (unsigned long long)sh->sector);
1665
Shaohua Libc0934f2012-05-22 13:55:05 +10001666 for (i = disks; i--; ) {
Tejun Heoe9c74692010-09-03 11:56:18 +02001667 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001668 sync |= test_bit(R5_SyncIO, &sh->dev[i].flags);
Shaohua Li9e4447682012-10-11 13:49:49 +11001669 discard |= test_bit(R5_Discard, &sh->dev[i].flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001670 }
Tejun Heoe9c74692010-09-03 11:56:18 +02001671
Dan Williams91c00922007-01-02 13:52:30 -07001672 for (i = disks; i--; ) {
1673 struct r5dev *dev = &sh->dev[i];
Dan Williamsac6b53b2009-07-14 13:40:19 -07001674
Tejun Heoe9c74692010-09-03 11:56:18 +02001675 if (dev->written || i == pd_idx || i == qd_idx) {
Shaohua Lid592a992014-05-21 17:57:44 +08001676 if (!discard && !test_bit(R5_SkipCopy, &dev->flags))
Shaohua Li9e4447682012-10-11 13:49:49 +11001677 set_bit(R5_UPTODATE, &dev->flags);
Tejun Heoe9c74692010-09-03 11:56:18 +02001678 if (fua)
1679 set_bit(R5_WantFUA, &dev->flags);
Shaohua Libc0934f2012-05-22 13:55:05 +10001680 if (sync)
1681 set_bit(R5_SyncIO, &dev->flags);
Tejun Heoe9c74692010-09-03 11:56:18 +02001682 }
Dan Williams91c00922007-01-02 13:52:30 -07001683 }
1684
Dan Williamsd8ee0722008-06-28 08:32:06 +10001685 if (sh->reconstruct_state == reconstruct_state_drain_run)
1686 sh->reconstruct_state = reconstruct_state_drain_result;
1687 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1688 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1689 else {
1690 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1691 sh->reconstruct_state = reconstruct_state_result;
1692 }
Dan Williams91c00922007-01-02 13:52:30 -07001693
1694 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07001695 raid5_release_stripe(sh);
Dan Williams91c00922007-01-02 13:52:30 -07001696}
1697
1698static void
Dan Williamsac6b53b2009-07-14 13:40:19 -07001699ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1700 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001701{
Dan Williams91c00922007-01-02 13:52:30 -07001702 int disks = sh->disks;
shli@kernel.org59fc6302014-12-15 12:57:03 +11001703 struct page **xor_srcs;
Dan Williamsa08abd82009-06-03 11:43:59 -07001704 struct async_submit_ctl submit;
shli@kernel.org59fc6302014-12-15 12:57:03 +11001705 int count, pd_idx = sh->pd_idx, i;
Dan Williams91c00922007-01-02 13:52:30 -07001706 struct page *xor_dest;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001707 int prexor = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001708 unsigned long flags;
shli@kernel.org59fc6302014-12-15 12:57:03 +11001709 int j = 0;
1710 struct stripe_head *head_sh = sh;
1711 int last_stripe;
Dan Williams91c00922007-01-02 13:52:30 -07001712
Harvey Harrisone46b2722008-04-28 02:15:50 -07001713 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001714 (unsigned long long)sh->sector);
1715
Shaohua Li620125f2012-10-11 13:49:05 +11001716 for (i = 0; i < sh->disks; i++) {
1717 if (pd_idx == i)
1718 continue;
1719 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1720 break;
1721 }
1722 if (i >= sh->disks) {
1723 atomic_inc(&sh->count);
Shaohua Li620125f2012-10-11 13:49:05 +11001724 set_bit(R5_Discard, &sh->dev[pd_idx].flags);
1725 ops_complete_reconstruct(sh);
1726 return;
1727 }
shli@kernel.org59fc6302014-12-15 12:57:03 +11001728again:
1729 count = 0;
1730 xor_srcs = to_addr_page(percpu, j);
Dan Williams91c00922007-01-02 13:52:30 -07001731 /* check if prexor is active which means only process blocks
1732 * that are part of a read-modify-write (written)
1733 */
shli@kernel.org59fc6302014-12-15 12:57:03 +11001734 if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
Dan Williamsd8ee0722008-06-28 08:32:06 +10001735 prexor = 1;
Dan Williams91c00922007-01-02 13:52:30 -07001736 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1737 for (i = disks; i--; ) {
1738 struct r5dev *dev = &sh->dev[i];
shli@kernel.org59fc6302014-12-15 12:57:03 +11001739 if (head_sh->dev[i].written)
Dan Williams91c00922007-01-02 13:52:30 -07001740 xor_srcs[count++] = dev->page;
1741 }
1742 } else {
1743 xor_dest = sh->dev[pd_idx].page;
1744 for (i = disks; i--; ) {
1745 struct r5dev *dev = &sh->dev[i];
1746 if (i != pd_idx)
1747 xor_srcs[count++] = dev->page;
1748 }
1749 }
1750
Dan Williams91c00922007-01-02 13:52:30 -07001751 /* 1/ if we prexor'd then the dest is reused as a source
1752 * 2/ if we did not prexor then we are redoing the parity
1753 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1754 * for the synchronous xor case
1755 */
shli@kernel.org59fc6302014-12-15 12:57:03 +11001756 last_stripe = !head_sh->batch_head ||
1757 list_first_entry(&sh->batch_list,
1758 struct stripe_head, batch_list) == head_sh;
1759 if (last_stripe) {
1760 flags = ASYNC_TX_ACK |
1761 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
Dan Williams91c00922007-01-02 13:52:30 -07001762
shli@kernel.org59fc6302014-12-15 12:57:03 +11001763 atomic_inc(&head_sh->count);
1764 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh,
1765 to_addr_conv(sh, percpu, j));
1766 } else {
1767 flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST;
1768 init_async_submit(&submit, flags, tx, NULL, NULL,
1769 to_addr_conv(sh, percpu, j));
1770 }
Dan Williams91c00922007-01-02 13:52:30 -07001771
Dan Williamsa08abd82009-06-03 11:43:59 -07001772 if (unlikely(count == 1))
1773 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1774 else
1775 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
shli@kernel.org59fc6302014-12-15 12:57:03 +11001776 if (!last_stripe) {
1777 j++;
1778 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1779 batch_list);
1780 goto again;
1781 }
Dan Williams91c00922007-01-02 13:52:30 -07001782}
1783
Dan Williamsac6b53b2009-07-14 13:40:19 -07001784static void
1785ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1786 struct dma_async_tx_descriptor *tx)
1787{
1788 struct async_submit_ctl submit;
shli@kernel.org59fc6302014-12-15 12:57:03 +11001789 struct page **blocks;
1790 int count, i, j = 0;
1791 struct stripe_head *head_sh = sh;
1792 int last_stripe;
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001793 int synflags;
1794 unsigned long txflags;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001795
1796 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1797
Shaohua Li620125f2012-10-11 13:49:05 +11001798 for (i = 0; i < sh->disks; i++) {
1799 if (sh->pd_idx == i || sh->qd_idx == i)
1800 continue;
1801 if (!test_bit(R5_Discard, &sh->dev[i].flags))
1802 break;
1803 }
1804 if (i >= sh->disks) {
1805 atomic_inc(&sh->count);
Shaohua Li620125f2012-10-11 13:49:05 +11001806 set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
1807 set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
1808 ops_complete_reconstruct(sh);
1809 return;
1810 }
1811
shli@kernel.org59fc6302014-12-15 12:57:03 +11001812again:
1813 blocks = to_addr_page(percpu, j);
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001814
1815 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1816 synflags = SYNDROME_SRC_WRITTEN;
1817 txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST;
1818 } else {
1819 synflags = SYNDROME_SRC_ALL;
1820 txflags = ASYNC_TX_ACK;
1821 }
1822
1823 count = set_syndrome_sources(blocks, sh, synflags);
shli@kernel.org59fc6302014-12-15 12:57:03 +11001824 last_stripe = !head_sh->batch_head ||
1825 list_first_entry(&sh->batch_list,
1826 struct stripe_head, batch_list) == head_sh;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001827
shli@kernel.org59fc6302014-12-15 12:57:03 +11001828 if (last_stripe) {
1829 atomic_inc(&head_sh->count);
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001830 init_async_submit(&submit, txflags, tx, ops_complete_reconstruct,
shli@kernel.org59fc6302014-12-15 12:57:03 +11001831 head_sh, to_addr_conv(sh, percpu, j));
1832 } else
1833 init_async_submit(&submit, 0, tx, NULL, NULL,
1834 to_addr_conv(sh, percpu, j));
Shaohua Li48769692015-05-13 09:30:08 -07001835 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
shli@kernel.org59fc6302014-12-15 12:57:03 +11001836 if (!last_stripe) {
1837 j++;
1838 sh = list_first_entry(&sh->batch_list, struct stripe_head,
1839 batch_list);
1840 goto again;
1841 }
Dan Williams91c00922007-01-02 13:52:30 -07001842}
1843
1844static void ops_complete_check(void *stripe_head_ref)
1845{
1846 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001847
Harvey Harrisone46b2722008-04-28 02:15:50 -07001848 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001849 (unsigned long long)sh->sector);
1850
Dan Williamsecc65c92008-06-28 08:31:57 +10001851 sh->check_state = check_state_check_result;
Dan Williams91c00922007-01-02 13:52:30 -07001852 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07001853 raid5_release_stripe(sh);
Dan Williams91c00922007-01-02 13:52:30 -07001854}
1855
Dan Williamsac6b53b2009-07-14 13:40:19 -07001856static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001857{
Dan Williams91c00922007-01-02 13:52:30 -07001858 int disks = sh->disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001859 int pd_idx = sh->pd_idx;
1860 int qd_idx = sh->qd_idx;
1861 struct page *xor_dest;
shli@kernel.org46d5b782014-12-15 12:57:02 +11001862 struct page **xor_srcs = to_addr_page(percpu, 0);
Dan Williams91c00922007-01-02 13:52:30 -07001863 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001864 struct async_submit_ctl submit;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001865 int count;
1866 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001867
Harvey Harrisone46b2722008-04-28 02:15:50 -07001868 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001869 (unsigned long long)sh->sector);
1870
shli@kernel.org59fc6302014-12-15 12:57:03 +11001871 BUG_ON(sh->batch_head);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001872 count = 0;
1873 xor_dest = sh->dev[pd_idx].page;
1874 xor_srcs[count++] = xor_dest;
Dan Williams91c00922007-01-02 13:52:30 -07001875 for (i = disks; i--; ) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001876 if (i == pd_idx || i == qd_idx)
1877 continue;
1878 xor_srcs[count++] = sh->dev[i].page;
Dan Williams91c00922007-01-02 13:52:30 -07001879 }
1880
Dan Williamsd6f38f32009-07-14 11:50:52 -07001881 init_async_submit(&submit, 0, NULL, NULL, NULL,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001882 to_addr_conv(sh, percpu, 0));
Dan Williams099f53c2009-04-08 14:28:37 -07001883 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07001884 &sh->ops.zero_sum_result, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001885
Dan Williams91c00922007-01-02 13:52:30 -07001886 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001887 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1888 tx = async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001889}
1890
Dan Williamsac6b53b2009-07-14 13:40:19 -07001891static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1892{
shli@kernel.org46d5b782014-12-15 12:57:02 +11001893 struct page **srcs = to_addr_page(percpu, 0);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001894 struct async_submit_ctl submit;
1895 int count;
1896
1897 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1898 (unsigned long long)sh->sector, checkp);
1899
shli@kernel.org59fc6302014-12-15 12:57:03 +11001900 BUG_ON(sh->batch_head);
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001901 count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL);
Dan Williamsac6b53b2009-07-14 13:40:19 -07001902 if (!checkp)
1903 srcs[count] = NULL;
1904
1905 atomic_inc(&sh->count);
1906 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
shli@kernel.org46d5b782014-12-15 12:57:02 +11001907 sh, to_addr_conv(sh, percpu, 0));
Dan Williamsac6b53b2009-07-14 13:40:19 -07001908 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1909 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
1910}
1911
NeilBrown51acbce2013-02-28 09:08:34 +11001912static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
Dan Williams91c00922007-01-02 13:52:30 -07001913{
1914 int overlap_clear = 0, i, disks = sh->disks;
1915 struct dma_async_tx_descriptor *tx = NULL;
NeilBrownd1688a62011-10-11 16:49:52 +11001916 struct r5conf *conf = sh->raid_conf;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001917 int level = conf->level;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001918 struct raid5_percpu *percpu;
1919 unsigned long cpu;
Dan Williams91c00922007-01-02 13:52:30 -07001920
Dan Williamsd6f38f32009-07-14 11:50:52 -07001921 cpu = get_cpu();
1922 percpu = per_cpu_ptr(conf->percpu, cpu);
Dan Williams83de75c2008-06-28 08:31:58 +10001923 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
Dan Williams91c00922007-01-02 13:52:30 -07001924 ops_run_biofill(sh);
1925 overlap_clear++;
1926 }
1927
Dan Williams7b3a8712008-06-28 08:32:09 +10001928 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001929 if (level < 6)
1930 tx = ops_run_compute5(sh, percpu);
1931 else {
1932 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1933 tx = ops_run_compute6_1(sh, percpu);
1934 else
1935 tx = ops_run_compute6_2(sh, percpu);
1936 }
1937 /* terminate the chain if reconstruct is not set to be run */
1938 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
Dan Williams7b3a8712008-06-28 08:32:09 +10001939 async_tx_ack(tx);
1940 }
Dan Williams91c00922007-01-02 13:52:30 -07001941
Markus Stockhausen584acdd2014-12-15 12:57:05 +11001942 if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
1943 if (level < 6)
1944 tx = ops_run_prexor5(sh, percpu, tx);
1945 else
1946 tx = ops_run_prexor6(sh, percpu, tx);
1947 }
Dan Williams91c00922007-01-02 13:52:30 -07001948
Dan Williams600aa102008-06-28 08:32:05 +10001949 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
Dan Williamsd8ee0722008-06-28 08:32:06 +10001950 tx = ops_run_biodrain(sh, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001951 overlap_clear++;
1952 }
1953
Dan Williamsac6b53b2009-07-14 13:40:19 -07001954 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1955 if (level < 6)
1956 ops_run_reconstruct5(sh, percpu, tx);
1957 else
1958 ops_run_reconstruct6(sh, percpu, tx);
1959 }
Dan Williams91c00922007-01-02 13:52:30 -07001960
Dan Williamsac6b53b2009-07-14 13:40:19 -07001961 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1962 if (sh->check_state == check_state_run)
1963 ops_run_check_p(sh, percpu);
1964 else if (sh->check_state == check_state_run_q)
1965 ops_run_check_pq(sh, percpu, 0);
1966 else if (sh->check_state == check_state_run_pq)
1967 ops_run_check_pq(sh, percpu, 1);
1968 else
1969 BUG();
1970 }
Dan Williams91c00922007-01-02 13:52:30 -07001971
shli@kernel.org59fc6302014-12-15 12:57:03 +11001972 if (overlap_clear && !sh->batch_head)
Dan Williams91c00922007-01-02 13:52:30 -07001973 for (i = disks; i--; ) {
1974 struct r5dev *dev = &sh->dev[i];
1975 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1976 wake_up(&sh->raid_conf->wait_for_overlap);
1977 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07001978 put_cpu();
Dan Williams91c00922007-01-02 13:52:30 -07001979}
1980
NeilBrownf18c1a32015-05-08 18:19:04 +10001981static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp)
1982{
1983 struct stripe_head *sh;
1984
1985 sh = kmem_cache_zalloc(sc, gfp);
1986 if (sh) {
1987 spin_lock_init(&sh->stripe_lock);
1988 spin_lock_init(&sh->batch_lock);
1989 INIT_LIST_HEAD(&sh->batch_list);
1990 INIT_LIST_HEAD(&sh->lru);
1991 atomic_set(&sh->count, 1);
1992 }
1993 return sh;
1994}
NeilBrown486f0642015-02-25 12:10:35 +11001995static int grow_one_stripe(struct r5conf *conf, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996{
1997 struct stripe_head *sh;
NeilBrownf18c1a32015-05-08 18:19:04 +10001998
1999 sh = alloc_stripe(conf->slab_cache, gfp);
NeilBrown3f294f42005-11-08 21:39:25 -08002000 if (!sh)
2001 return 0;
Namhyung Kim6ce32842011-07-18 17:38:50 +10002002
NeilBrown3f294f42005-11-08 21:39:25 -08002003 sh->raid_conf = conf;
NeilBrown3f294f42005-11-08 21:39:25 -08002004
NeilBrowna9683a72015-02-25 12:02:51 +11002005 if (grow_buffers(sh, gfp)) {
NeilBrowne4e11e32010-06-16 16:45:16 +10002006 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08002007 kmem_cache_free(conf->slab_cache, sh);
2008 return 0;
2009 }
NeilBrown486f0642015-02-25 12:10:35 +11002010 sh->hash_lock_index =
2011 conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
NeilBrown3f294f42005-11-08 21:39:25 -08002012 /* we just created an active stripe so... */
NeilBrown3f294f42005-11-08 21:39:25 -08002013 atomic_inc(&conf->active_stripes);
shli@kernel.org59fc6302014-12-15 12:57:03 +11002014
Shaohua Li6d036f72015-08-13 14:31:57 -07002015 raid5_release_stripe(sh);
NeilBrown486f0642015-02-25 12:10:35 +11002016 conf->max_nr_stripes++;
NeilBrown3f294f42005-11-08 21:39:25 -08002017 return 1;
2018}
2019
NeilBrownd1688a62011-10-11 16:49:52 +11002020static int grow_stripes(struct r5conf *conf, int num)
NeilBrown3f294f42005-11-08 21:39:25 -08002021{
Christoph Lametere18b8902006-12-06 20:33:20 -08002022 struct kmem_cache *sc;
NeilBrown5e5e3e72009-10-16 16:35:30 +11002023 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024
NeilBrownf4be6b42010-06-01 19:37:25 +10002025 if (conf->mddev->gendisk)
2026 sprintf(conf->cache_name[0],
2027 "raid%d-%s", conf->level, mdname(conf->mddev));
2028 else
2029 sprintf(conf->cache_name[0],
2030 "raid%d-%p", conf->level, conf->mddev);
2031 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
2032
NeilBrownad01c9e2006-03-27 01:18:07 -08002033 conf->active_name = 0;
2034 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09002036 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 if (!sc)
2038 return 1;
2039 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08002040 conf->pool_size = devs;
NeilBrown486f0642015-02-25 12:10:35 +11002041 while (num--)
2042 if (!grow_one_stripe(conf, GFP_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return 1;
NeilBrown486f0642015-02-25 12:10:35 +11002044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 return 0;
2046}
NeilBrown29269552006-03-27 01:18:10 -08002047
Dan Williamsd6f38f32009-07-14 11:50:52 -07002048/**
2049 * scribble_len - return the required size of the scribble region
2050 * @num - total number of disks in the array
2051 *
2052 * The size must be enough to contain:
2053 * 1/ a struct page pointer for each device in the array +2
2054 * 2/ room to convert each entry in (1) to its corresponding dma
2055 * (dma_map_page()) or page (page_address()) address.
2056 *
2057 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
2058 * calculate over all devices (not just the data blocks), using zeros in place
2059 * of the P and Q blocks.
2060 */
shli@kernel.org46d5b782014-12-15 12:57:02 +11002061static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags)
Dan Williamsd6f38f32009-07-14 11:50:52 -07002062{
shli@kernel.org46d5b782014-12-15 12:57:02 +11002063 struct flex_array *ret;
Dan Williamsd6f38f32009-07-14 11:50:52 -07002064 size_t len;
2065
2066 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
shli@kernel.org46d5b782014-12-15 12:57:02 +11002067 ret = flex_array_alloc(len, cnt, flags);
2068 if (!ret)
2069 return NULL;
2070 /* always prealloc all elements, so no locking is required */
2071 if (flex_array_prealloc(ret, 0, cnt, flags)) {
2072 flex_array_free(ret);
2073 return NULL;
2074 }
2075 return ret;
Dan Williamsd6f38f32009-07-14 11:50:52 -07002076}
2077
NeilBrown738a2732015-05-08 18:19:39 +10002078static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors)
2079{
2080 unsigned long cpu;
2081 int err = 0;
2082
Shaohua Li27a353c2016-02-24 17:38:28 -08002083 /*
2084 * Never shrink. And mddev_suspend() could deadlock if this is called
2085 * from raid5d. In that case, scribble_disks and scribble_sectors
2086 * should equal to new_disks and new_sectors
2087 */
2088 if (conf->scribble_disks >= new_disks &&
2089 conf->scribble_sectors >= new_sectors)
2090 return 0;
NeilBrown738a2732015-05-08 18:19:39 +10002091 mddev_suspend(conf->mddev);
2092 get_online_cpus();
2093 for_each_present_cpu(cpu) {
2094 struct raid5_percpu *percpu;
2095 struct flex_array *scribble;
2096
2097 percpu = per_cpu_ptr(conf->percpu, cpu);
2098 scribble = scribble_alloc(new_disks,
2099 new_sectors / STRIPE_SECTORS,
2100 GFP_NOIO);
2101
2102 if (scribble) {
2103 flex_array_free(percpu->scribble);
2104 percpu->scribble = scribble;
2105 } else {
2106 err = -ENOMEM;
2107 break;
2108 }
2109 }
2110 put_online_cpus();
2111 mddev_resume(conf->mddev);
Shaohua Li27a353c2016-02-24 17:38:28 -08002112 if (!err) {
2113 conf->scribble_disks = new_disks;
2114 conf->scribble_sectors = new_sectors;
2115 }
NeilBrown738a2732015-05-08 18:19:39 +10002116 return err;
2117}
2118
NeilBrownd1688a62011-10-11 16:49:52 +11002119static int resize_stripes(struct r5conf *conf, int newsize)
NeilBrownad01c9e2006-03-27 01:18:07 -08002120{
2121 /* Make all the stripes able to hold 'newsize' devices.
2122 * New slots in each stripe get 'page' set to a new page.
2123 *
2124 * This happens in stages:
2125 * 1/ create a new kmem_cache and allocate the required number of
2126 * stripe_heads.
Masanari Iida83f0d772012-10-30 00:18:08 +09002127 * 2/ gather all the old stripe_heads and transfer the pages across
NeilBrownad01c9e2006-03-27 01:18:07 -08002128 * to the new stripe_heads. This will have the side effect of
2129 * freezing the array as once all stripe_heads have been collected,
2130 * no IO will be possible. Old stripe heads are freed once their
2131 * pages have been transferred over, and the old kmem_cache is
2132 * freed when all stripes are done.
2133 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
2134 * we simple return a failre status - no need to clean anything up.
2135 * 4/ allocate new pages for the new slots in the new stripe_heads.
2136 * If this fails, we don't bother trying the shrink the
2137 * stripe_heads down again, we just leave them as they are.
2138 * As each stripe_head is processed the new one is released into
2139 * active service.
2140 *
2141 * Once step2 is started, we cannot afford to wait for a write,
2142 * so we use GFP_NOIO allocations.
2143 */
2144 struct stripe_head *osh, *nsh;
2145 LIST_HEAD(newstripes);
2146 struct disk_info *ndisks;
Dan Williamsb5470dc2008-06-27 21:44:04 -07002147 int err;
Christoph Lametere18b8902006-12-06 20:33:20 -08002148 struct kmem_cache *sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08002149 int i;
Shaohua Li566c09c2013-11-14 15:16:17 +11002150 int hash, cnt;
NeilBrownad01c9e2006-03-27 01:18:07 -08002151
2152 if (newsize <= conf->pool_size)
2153 return 0; /* never bother to shrink */
2154
Dan Williamsb5470dc2008-06-27 21:44:04 -07002155 err = md_allow_write(conf->mddev);
2156 if (err)
2157 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08002158
NeilBrownad01c9e2006-03-27 01:18:07 -08002159 /* Step 1 */
2160 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
2161 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09002162 0, 0, NULL);
NeilBrownad01c9e2006-03-27 01:18:07 -08002163 if (!sc)
2164 return -ENOMEM;
2165
NeilBrown2d5b5692015-07-06 12:49:23 +10002166 /* Need to ensure auto-resizing doesn't interfere */
2167 mutex_lock(&conf->cache_size_mutex);
2168
NeilBrownad01c9e2006-03-27 01:18:07 -08002169 for (i = conf->max_nr_stripes; i; i--) {
NeilBrownf18c1a32015-05-08 18:19:04 +10002170 nsh = alloc_stripe(sc, GFP_KERNEL);
NeilBrownad01c9e2006-03-27 01:18:07 -08002171 if (!nsh)
2172 break;
2173
NeilBrownad01c9e2006-03-27 01:18:07 -08002174 nsh->raid_conf = conf;
NeilBrownad01c9e2006-03-27 01:18:07 -08002175 list_add(&nsh->lru, &newstripes);
2176 }
2177 if (i) {
2178 /* didn't get enough, give up */
2179 while (!list_empty(&newstripes)) {
2180 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2181 list_del(&nsh->lru);
2182 kmem_cache_free(sc, nsh);
2183 }
2184 kmem_cache_destroy(sc);
NeilBrown2d5b5692015-07-06 12:49:23 +10002185 mutex_unlock(&conf->cache_size_mutex);
NeilBrownad01c9e2006-03-27 01:18:07 -08002186 return -ENOMEM;
2187 }
2188 /* Step 2 - Must use GFP_NOIO now.
2189 * OK, we have enough stripes, start collecting inactive
2190 * stripes and copying them over
2191 */
Shaohua Li566c09c2013-11-14 15:16:17 +11002192 hash = 0;
2193 cnt = 0;
NeilBrownad01c9e2006-03-27 01:18:07 -08002194 list_for_each_entry(nsh, &newstripes, lru) {
Shaohua Li566c09c2013-11-14 15:16:17 +11002195 lock_device_hash_lock(conf, hash);
Shaohua Li6ab2a4b2016-02-25 16:24:42 -08002196 wait_event_cmd(conf->wait_for_stripe,
Shaohua Li566c09c2013-11-14 15:16:17 +11002197 !list_empty(conf->inactive_list + hash),
2198 unlock_device_hash_lock(conf, hash),
2199 lock_device_hash_lock(conf, hash));
2200 osh = get_free_stripe(conf, hash);
2201 unlock_device_hash_lock(conf, hash);
NeilBrownf18c1a32015-05-08 18:19:04 +10002202
Shaohua Lid592a992014-05-21 17:57:44 +08002203 for(i=0; i<conf->pool_size; i++) {
NeilBrownad01c9e2006-03-27 01:18:07 -08002204 nsh->dev[i].page = osh->dev[i].page;
Shaohua Lid592a992014-05-21 17:57:44 +08002205 nsh->dev[i].orig_page = osh->dev[i].page;
2206 }
Shaohua Li566c09c2013-11-14 15:16:17 +11002207 nsh->hash_lock_index = hash;
NeilBrownad01c9e2006-03-27 01:18:07 -08002208 kmem_cache_free(conf->slab_cache, osh);
Shaohua Li566c09c2013-11-14 15:16:17 +11002209 cnt++;
2210 if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
2211 !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
2212 hash++;
2213 cnt = 0;
2214 }
NeilBrownad01c9e2006-03-27 01:18:07 -08002215 }
2216 kmem_cache_destroy(conf->slab_cache);
2217
2218 /* Step 3.
2219 * At this point, we are holding all the stripes so the array
2220 * is completely stalled, so now is a good time to resize
Dan Williamsd6f38f32009-07-14 11:50:52 -07002221 * conf->disks and the scribble region
NeilBrownad01c9e2006-03-27 01:18:07 -08002222 */
2223 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
2224 if (ndisks) {
2225 for (i=0; i<conf->raid_disks; i++)
2226 ndisks[i] = conf->disks[i];
2227 kfree(conf->disks);
2228 conf->disks = ndisks;
2229 } else
2230 err = -ENOMEM;
2231
NeilBrown2d5b5692015-07-06 12:49:23 +10002232 mutex_unlock(&conf->cache_size_mutex);
NeilBrownad01c9e2006-03-27 01:18:07 -08002233 /* Step 4, return new stripes to service */
2234 while(!list_empty(&newstripes)) {
2235 nsh = list_entry(newstripes.next, struct stripe_head, lru);
2236 list_del_init(&nsh->lru);
Dan Williamsd6f38f32009-07-14 11:50:52 -07002237
NeilBrownad01c9e2006-03-27 01:18:07 -08002238 for (i=conf->raid_disks; i < newsize; i++)
2239 if (nsh->dev[i].page == NULL) {
2240 struct page *p = alloc_page(GFP_NOIO);
2241 nsh->dev[i].page = p;
Shaohua Lid592a992014-05-21 17:57:44 +08002242 nsh->dev[i].orig_page = p;
NeilBrownad01c9e2006-03-27 01:18:07 -08002243 if (!p)
2244 err = -ENOMEM;
2245 }
Shaohua Li6d036f72015-08-13 14:31:57 -07002246 raid5_release_stripe(nsh);
NeilBrownad01c9e2006-03-27 01:18:07 -08002247 }
2248 /* critical section pass, GFP_NOIO no longer needed */
2249
2250 conf->slab_cache = sc;
2251 conf->active_name = 1-conf->active_name;
NeilBrown6e9eac22015-05-08 18:19:34 +10002252 if (!err)
2253 conf->pool_size = newsize;
NeilBrownad01c9e2006-03-27 01:18:07 -08002254 return err;
2255}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256
NeilBrown486f0642015-02-25 12:10:35 +11002257static int drop_one_stripe(struct r5conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258{
2259 struct stripe_head *sh;
NeilBrown49895bc2015-08-03 17:09:57 +10002260 int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261
Shaohua Li566c09c2013-11-14 15:16:17 +11002262 spin_lock_irq(conf->hash_locks + hash);
2263 sh = get_free_stripe(conf, hash);
2264 spin_unlock_irq(conf->hash_locks + hash);
NeilBrown3f294f42005-11-08 21:39:25 -08002265 if (!sh)
2266 return 0;
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002267 BUG_ON(atomic_read(&sh->count));
NeilBrowne4e11e32010-06-16 16:45:16 +10002268 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08002269 kmem_cache_free(conf->slab_cache, sh);
2270 atomic_dec(&conf->active_stripes);
NeilBrown486f0642015-02-25 12:10:35 +11002271 conf->max_nr_stripes--;
NeilBrown3f294f42005-11-08 21:39:25 -08002272 return 1;
2273}
2274
NeilBrownd1688a62011-10-11 16:49:52 +11002275static void shrink_stripes(struct r5conf *conf)
NeilBrown3f294f42005-11-08 21:39:25 -08002276{
NeilBrown486f0642015-02-25 12:10:35 +11002277 while (conf->max_nr_stripes &&
2278 drop_one_stripe(conf))
2279 ;
NeilBrown3f294f42005-11-08 21:39:25 -08002280
Julia Lawall644df1a2015-09-13 14:15:10 +02002281 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 conf->slab_cache = NULL;
2283}
2284
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002285static void raid5_end_read_request(struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286{
NeilBrown99c0fb52009-03-31 14:39:38 +11002287 struct stripe_head *sh = bi->bi_private;
NeilBrownd1688a62011-10-11 16:49:52 +11002288 struct r5conf *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08002289 int disks = sh->disks, i;
NeilBrownd6950432006-07-10 04:44:20 -07002290 char b[BDEVNAME_SIZE];
NeilBrowndd054fc2011-12-23 10:17:53 +11002291 struct md_rdev *rdev = NULL;
NeilBrown05616be2012-05-21 09:27:00 +10002292 sector_t s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293
2294 for (i=0 ; i<disks; i++)
2295 if (bi == &sh->dev[i].req)
2296 break;
2297
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002298 pr_debug("end_read_request %llu/%d, count: %d, error %d.\n",
Dan Williams45b42332007-07-09 11:56:43 -07002299 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002300 bi->bi_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 if (i == disks) {
2302 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02002303 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 }
NeilBrown14a75d32011-12-23 10:17:52 +11002305 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
NeilBrowndd054fc2011-12-23 10:17:53 +11002306 /* If replacement finished while this request was outstanding,
2307 * 'replacement' might be NULL already.
2308 * In that case it moved down to 'rdev'.
2309 * rdev is not removed until all requests are finished.
2310 */
NeilBrown14a75d32011-12-23 10:17:52 +11002311 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11002312 if (!rdev)
NeilBrown14a75d32011-12-23 10:17:52 +11002313 rdev = conf->disks[i].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
NeilBrown05616be2012-05-21 09:27:00 +10002315 if (use_new_offset(conf, sh))
2316 s = sh->sector + rdev->new_data_offset;
2317 else
2318 s = sh->sector + rdev->data_offset;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002319 if (!bi->bi_error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -08002321 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11002322 /* Note that this cannot happen on a
2323 * replacement device. We just fail those on
2324 * any error
2325 */
Christian Dietrich8bda4702011-07-27 11:00:36 +10002326 printk_ratelimited(
2327 KERN_INFO
2328 "md/raid:%s: read error corrected"
2329 " (%lu sectors at %llu on %s)\n",
2330 mdname(conf->mddev), STRIPE_SECTORS,
NeilBrown05616be2012-05-21 09:27:00 +10002331 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002332 bdevname(rdev->bdev, b));
Namhyung Kimddd51152011-07-27 11:00:36 +10002333 atomic_add(STRIPE_SECTORS, &rdev->corrected_errors);
NeilBrown4e5314b2005-11-08 21:39:22 -08002334 clear_bit(R5_ReadError, &sh->dev[i].flags);
2335 clear_bit(R5_ReWrite, &sh->dev[i].flags);
majianpeng3f9e7c12012-07-31 10:04:21 +10002336 } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2337 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2338
NeilBrown14a75d32011-12-23 10:17:52 +11002339 if (atomic_read(&rdev->read_errors))
2340 atomic_set(&rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 } else {
NeilBrown14a75d32011-12-23 10:17:52 +11002342 const char *bdn = bdevname(rdev->bdev, b);
NeilBrownba22dcb2005-11-08 21:39:31 -08002343 int retry = 0;
majianpeng2e8ac3032012-07-03 15:57:02 +10002344 int set_bad = 0;
NeilBrownd6950432006-07-10 04:44:20 -07002345
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07002347 atomic_inc(&rdev->read_errors);
NeilBrown14a75d32011-12-23 10:17:52 +11002348 if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
2349 printk_ratelimited(
2350 KERN_WARNING
2351 "md/raid:%s: read error on replacement device "
2352 "(sector %llu on %s).\n",
2353 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002354 (unsigned long long)s,
NeilBrown14a75d32011-12-23 10:17:52 +11002355 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002356 else if (conf->mddev->degraded >= conf->max_degraded) {
2357 set_bad = 1;
Christian Dietrich8bda4702011-07-27 11:00:36 +10002358 printk_ratelimited(
2359 KERN_WARNING
2360 "md/raid:%s: read error not correctable "
2361 "(sector %llu on %s).\n",
2362 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002363 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002364 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002365 } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) {
NeilBrown4e5314b2005-11-08 21:39:22 -08002366 /* Oh, no!!! */
majianpeng2e8ac3032012-07-03 15:57:02 +10002367 set_bad = 1;
Christian Dietrich8bda4702011-07-27 11:00:36 +10002368 printk_ratelimited(
2369 KERN_WARNING
2370 "md/raid:%s: read error NOT corrected!! "
2371 "(sector %llu on %s).\n",
2372 mdname(conf->mddev),
NeilBrown05616be2012-05-21 09:27:00 +10002373 (unsigned long long)s,
Christian Dietrich8bda4702011-07-27 11:00:36 +10002374 bdn);
majianpeng2e8ac3032012-07-03 15:57:02 +10002375 } else if (atomic_read(&rdev->read_errors)
NeilBrownba22dcb2005-11-08 21:39:31 -08002376 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -08002377 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10002378 "md/raid:%s: Too many read errors, failing device %s.\n",
NeilBrownd6950432006-07-10 04:44:20 -07002379 mdname(conf->mddev), bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08002380 else
2381 retry = 1;
Bian Yuedfa1f62013-11-14 15:16:17 +11002382 if (set_bad && test_bit(In_sync, &rdev->flags)
2383 && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2384 retry = 1;
NeilBrownba22dcb2005-11-08 21:39:31 -08002385 if (retry)
majianpeng3f9e7c12012-07-31 10:04:21 +10002386 if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
2387 set_bit(R5_ReadError, &sh->dev[i].flags);
2388 clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2389 } else
2390 set_bit(R5_ReadNoMerge, &sh->dev[i].flags);
NeilBrownba22dcb2005-11-08 21:39:31 -08002391 else {
NeilBrown4e5314b2005-11-08 21:39:22 -08002392 clear_bit(R5_ReadError, &sh->dev[i].flags);
2393 clear_bit(R5_ReWrite, &sh->dev[i].flags);
majianpeng2e8ac3032012-07-03 15:57:02 +10002394 if (!(set_bad
2395 && test_bit(In_sync, &rdev->flags)
2396 && rdev_set_badblocks(
2397 rdev, sh->sector, STRIPE_SECTORS, 0)))
2398 md_error(conf->mddev, rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -08002399 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 }
NeilBrown14a75d32011-12-23 10:17:52 +11002401 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 clear_bit(R5_LOCKED, &sh->dev[i].flags);
2403 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07002404 raid5_release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405}
2406
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002407static void raid5_end_write_request(struct bio *bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
NeilBrown99c0fb52009-03-31 14:39:38 +11002409 struct stripe_head *sh = bi->bi_private;
NeilBrownd1688a62011-10-11 16:49:52 +11002410 struct r5conf *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08002411 int disks = sh->disks, i;
NeilBrown977df362011-12-23 10:17:53 +11002412 struct md_rdev *uninitialized_var(rdev);
NeilBrownb84db562011-07-28 11:39:23 +10002413 sector_t first_bad;
2414 int bad_sectors;
NeilBrown977df362011-12-23 10:17:53 +11002415 int replacement = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
NeilBrown977df362011-12-23 10:17:53 +11002417 for (i = 0 ; i < disks; i++) {
2418 if (bi == &sh->dev[i].req) {
2419 rdev = conf->disks[i].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 break;
NeilBrown977df362011-12-23 10:17:53 +11002421 }
2422 if (bi == &sh->dev[i].rreq) {
2423 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11002424 if (rdev)
2425 replacement = 1;
2426 else
2427 /* rdev was removed and 'replacement'
2428 * replaced it. rdev is not removed
2429 * until all requests are finished.
2430 */
2431 rdev = conf->disks[i].rdev;
NeilBrown977df362011-12-23 10:17:53 +11002432 break;
2433 }
2434 }
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002435 pr_debug("end_write_request %llu/%d, count %d, error: %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002437 bi->bi_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (i == disks) {
2439 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02002440 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 }
2442
NeilBrown977df362011-12-23 10:17:53 +11002443 if (replacement) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002444 if (bi->bi_error)
NeilBrown977df362011-12-23 10:17:53 +11002445 md_error(conf->mddev, rdev);
2446 else if (is_badblock(rdev, sh->sector,
2447 STRIPE_SECTORS,
2448 &first_bad, &bad_sectors))
2449 set_bit(R5_MadeGoodRepl, &sh->dev[i].flags);
2450 } else {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002451 if (bi->bi_error) {
NeilBrown9f97e4b2014-01-16 09:35:38 +11002452 set_bit(STRIPE_DEGRADED, &sh->state);
NeilBrown977df362011-12-23 10:17:53 +11002453 set_bit(WriteErrorSeen, &rdev->flags);
2454 set_bit(R5_WriteError, &sh->dev[i].flags);
NeilBrown3a6de292011-12-23 10:17:54 +11002455 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2456 set_bit(MD_RECOVERY_NEEDED,
2457 &rdev->mddev->recovery);
NeilBrown977df362011-12-23 10:17:53 +11002458 } else if (is_badblock(rdev, sh->sector,
2459 STRIPE_SECTORS,
NeilBrownc0b32972013-04-24 11:42:42 +10002460 &first_bad, &bad_sectors)) {
NeilBrown977df362011-12-23 10:17:53 +11002461 set_bit(R5_MadeGood, &sh->dev[i].flags);
NeilBrownc0b32972013-04-24 11:42:42 +10002462 if (test_bit(R5_ReadError, &sh->dev[i].flags))
2463 /* That was a successful write so make
2464 * sure it looks like we already did
2465 * a re-write.
2466 */
2467 set_bit(R5_ReWrite, &sh->dev[i].flags);
2468 }
NeilBrown977df362011-12-23 10:17:53 +11002469 }
2470 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002472 if (sh->batch_head && bi->bi_error && !replacement)
shli@kernel.org72ac7332014-12-15 12:57:03 +11002473 set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state);
2474
NeilBrown977df362011-12-23 10:17:53 +11002475 if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags))
2476 clear_bit(R5_LOCKED, &sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07002478 raid5_release_stripe(sh);
shli@kernel.org59fc6302014-12-15 12:57:03 +11002479
2480 if (sh->batch_head && sh != sh->batch_head)
Shaohua Li6d036f72015-08-13 14:31:57 -07002481 raid5_release_stripe(sh->batch_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482}
2483
NeilBrown784052e2009-03-31 15:19:07 +11002484static void raid5_build_block(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485{
2486 struct r5dev *dev = &sh->dev[i];
2487
2488 bio_init(&dev->req);
2489 dev->req.bi_io_vec = &dev->vec;
Shaohua Lid592a992014-05-21 17:57:44 +08002490 dev->req.bi_max_vecs = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 dev->req.bi_private = sh;
2492
NeilBrown977df362011-12-23 10:17:53 +11002493 bio_init(&dev->rreq);
2494 dev->rreq.bi_io_vec = &dev->rvec;
Shaohua Lid592a992014-05-21 17:57:44 +08002495 dev->rreq.bi_max_vecs = 1;
NeilBrown977df362011-12-23 10:17:53 +11002496 dev->rreq.bi_private = sh;
NeilBrown977df362011-12-23 10:17:53 +11002497
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 dev->flags = 0;
Shaohua Li6d036f72015-08-13 14:31:57 -07002499 dev->sector = raid5_compute_blocknr(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500}
2501
Shaohua Li849674e2016-01-20 13:52:20 -08002502static void raid5_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503{
2504 char b[BDEVNAME_SIZE];
NeilBrownd1688a62011-10-11 16:49:52 +11002505 struct r5conf *conf = mddev->private;
NeilBrown908f4fb2011-12-23 10:17:50 +11002506 unsigned long flags;
NeilBrown0c55e022010-05-03 14:09:02 +10002507 pr_debug("raid456: error called\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508
NeilBrown908f4fb2011-12-23 10:17:50 +11002509 spin_lock_irqsave(&conf->device_lock, flags);
2510 clear_bit(In_sync, &rdev->flags);
2511 mddev->degraded = calc_degraded(conf);
2512 spin_unlock_irqrestore(&conf->device_lock, flags);
2513 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2514
NeilBrownde393cd2011-07-28 11:31:48 +10002515 set_bit(Blocked, &rdev->flags);
NeilBrown6f8d0c72011-05-11 14:38:44 +10002516 set_bit(Faulty, &rdev->flags);
2517 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownc3cce6c2015-08-14 12:47:33 +10002518 set_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown6f8d0c72011-05-11 14:38:44 +10002519 printk(KERN_ALERT
2520 "md/raid:%s: Disk failure on %s, disabling device.\n"
2521 "md/raid:%s: Operation continuing on %d devices.\n",
2522 mdname(mddev),
2523 bdevname(rdev->bdev, b),
2524 mdname(mddev),
2525 conf->raid_disks - mddev->degraded);
NeilBrown16a53ec2006-06-26 00:27:38 -07002526}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
2528/*
2529 * Input: a 'big' sector number,
2530 * Output: index of the data and parity disk, and the sector # in them.
2531 */
Shaohua Li6d036f72015-08-13 14:31:57 -07002532sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector,
2533 int previous, int *dd_idx,
2534 struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535{
NeilBrown6e3b96e2010-04-23 07:08:28 +10002536 sector_t stripe, stripe2;
NeilBrown35f2a592010-04-20 14:13:34 +10002537 sector_t chunk_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 unsigned int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11002539 int pd_idx, qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11002540 int ddf_layout = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 sector_t new_sector;
NeilBrowne183eae2009-03-31 15:20:22 +11002542 int algorithm = previous ? conf->prev_algo
2543 : conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10002544 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2545 : conf->chunk_sectors;
NeilBrown112bf892009-03-31 14:39:38 +11002546 int raid_disks = previous ? conf->previous_raid_disks
2547 : conf->raid_disks;
2548 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
2550 /* First compute the information on this sector */
2551
2552 /*
2553 * Compute the chunk number and the sector offset inside the chunk
2554 */
2555 chunk_offset = sector_div(r_sector, sectors_per_chunk);
2556 chunk_number = r_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
2558 /*
2559 * Compute the stripe number
2560 */
NeilBrown35f2a592010-04-20 14:13:34 +10002561 stripe = chunk_number;
2562 *dd_idx = sector_div(stripe, data_disks);
NeilBrown6e3b96e2010-04-23 07:08:28 +10002563 stripe2 = stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 /*
2565 * Select the parity disk based on the user selected algorithm.
2566 */
NeilBrown84789552011-07-27 11:00:36 +10002567 pd_idx = qd_idx = -1;
NeilBrown16a53ec2006-06-26 00:27:38 -07002568 switch(conf->level) {
2569 case 4:
NeilBrown911d4ee2009-03-31 14:39:38 +11002570 pd_idx = data_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002571 break;
2572 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11002573 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002575 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002576 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 (*dd_idx)++;
2578 break;
2579 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002580 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002581 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 (*dd_idx)++;
2583 break;
2584 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002585 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002586 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 break;
2588 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002589 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002590 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002592 case ALGORITHM_PARITY_0:
2593 pd_idx = 0;
2594 (*dd_idx)++;
2595 break;
2596 case ALGORITHM_PARITY_N:
2597 pd_idx = data_disks;
2598 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002600 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002601 }
2602 break;
2603 case 6:
2604
NeilBrowne183eae2009-03-31 15:20:22 +11002605 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07002606 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002607 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002608 qd_idx = pd_idx + 1;
2609 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11002610 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11002611 qd_idx = 0;
2612 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002613 (*dd_idx) += 2; /* D D P Q D */
2614 break;
2615 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002616 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002617 qd_idx = pd_idx + 1;
2618 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11002619 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11002620 qd_idx = 0;
2621 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002622 (*dd_idx) += 2; /* D D P Q D */
2623 break;
2624 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002625 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002626 qd_idx = (pd_idx + 1) % raid_disks;
2627 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002628 break;
2629 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002630 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11002631 qd_idx = (pd_idx + 1) % raid_disks;
2632 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07002633 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002634
2635 case ALGORITHM_PARITY_0:
2636 pd_idx = 0;
2637 qd_idx = 1;
2638 (*dd_idx) += 2;
2639 break;
2640 case ALGORITHM_PARITY_N:
2641 pd_idx = data_disks;
2642 qd_idx = data_disks + 1;
2643 break;
2644
2645 case ALGORITHM_ROTATING_ZERO_RESTART:
2646 /* Exactly the same as RIGHT_ASYMMETRIC, but or
2647 * of blocks for computing Q is different.
2648 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002649 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002650 qd_idx = pd_idx + 1;
2651 if (pd_idx == raid_disks-1) {
2652 (*dd_idx)++; /* Q D D D P */
2653 qd_idx = 0;
2654 } else if (*dd_idx >= pd_idx)
2655 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11002656 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002657 break;
2658
2659 case ALGORITHM_ROTATING_N_RESTART:
2660 /* Same a left_asymmetric, by first stripe is
2661 * D D D P Q rather than
2662 * Q D D D P
2663 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002664 stripe2 += 1;
2665 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002666 qd_idx = pd_idx + 1;
2667 if (pd_idx == raid_disks-1) {
2668 (*dd_idx)++; /* Q D D D P */
2669 qd_idx = 0;
2670 } else if (*dd_idx >= pd_idx)
2671 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11002672 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002673 break;
2674
2675 case ALGORITHM_ROTATING_N_CONTINUE:
2676 /* Same as left_symmetric but Q is before P */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002677 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11002678 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
2679 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
NeilBrown67cc2b82009-03-31 14:39:38 +11002680 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11002681 break;
2682
2683 case ALGORITHM_LEFT_ASYMMETRIC_6:
2684 /* RAID5 left_asymmetric, with Q on last device */
NeilBrown6e3b96e2010-04-23 07:08:28 +10002685 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002686 if (*dd_idx >= pd_idx)
2687 (*dd_idx)++;
2688 qd_idx = raid_disks - 1;
2689 break;
2690
2691 case ALGORITHM_RIGHT_ASYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002692 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002693 if (*dd_idx >= pd_idx)
2694 (*dd_idx)++;
2695 qd_idx = raid_disks - 1;
2696 break;
2697
2698 case ALGORITHM_LEFT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002699 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002700 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2701 qd_idx = raid_disks - 1;
2702 break;
2703
2704 case ALGORITHM_RIGHT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10002705 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11002706 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
2707 qd_idx = raid_disks - 1;
2708 break;
2709
2710 case ALGORITHM_PARITY_0_6:
2711 pd_idx = 0;
2712 (*dd_idx)++;
2713 qd_idx = raid_disks - 1;
2714 break;
2715
NeilBrown16a53ec2006-06-26 00:27:38 -07002716 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002717 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002718 }
2719 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 }
2721
NeilBrown911d4ee2009-03-31 14:39:38 +11002722 if (sh) {
2723 sh->pd_idx = pd_idx;
2724 sh->qd_idx = qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11002725 sh->ddf_layout = ddf_layout;
NeilBrown911d4ee2009-03-31 14:39:38 +11002726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 /*
2728 * Finally, compute the new sector number
2729 */
2730 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
2731 return new_sector;
2732}
2733
Shaohua Li6d036f72015-08-13 14:31:57 -07002734sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735{
NeilBrownd1688a62011-10-11 16:49:52 +11002736 struct r5conf *conf = sh->raid_conf;
NeilBrownb875e532006-12-10 02:20:49 -08002737 int raid_disks = sh->disks;
2738 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 sector_t new_sector = sh->sector, check;
Andre Noll09c9e5f2009-06-18 08:45:55 +10002740 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
2741 : conf->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11002742 int algorithm = previous ? conf->prev_algo
2743 : conf->algorithm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 sector_t stripe;
2745 int chunk_offset;
NeilBrown35f2a592010-04-20 14:13:34 +10002746 sector_t chunk_number;
2747 int dummy1, dd_idx = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 sector_t r_sector;
NeilBrown911d4ee2009-03-31 14:39:38 +11002749 struct stripe_head sh2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
2751 chunk_offset = sector_div(new_sector, sectors_per_chunk);
2752 stripe = new_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
NeilBrown16a53ec2006-06-26 00:27:38 -07002754 if (i == sh->pd_idx)
2755 return 0;
2756 switch(conf->level) {
2757 case 4: break;
2758 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11002759 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 case ALGORITHM_LEFT_ASYMMETRIC:
2761 case ALGORITHM_RIGHT_ASYMMETRIC:
2762 if (i > sh->pd_idx)
2763 i--;
2764 break;
2765 case ALGORITHM_LEFT_SYMMETRIC:
2766 case ALGORITHM_RIGHT_SYMMETRIC:
2767 if (i < sh->pd_idx)
2768 i += raid_disks;
2769 i -= (sh->pd_idx + 1);
2770 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002771 case ALGORITHM_PARITY_0:
2772 i -= 1;
2773 break;
2774 case ALGORITHM_PARITY_N:
2775 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002777 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002778 }
2779 break;
2780 case 6:
NeilBrownd0dabf72009-03-31 14:39:38 +11002781 if (i == sh->qd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07002782 return 0; /* It is the Q disk */
NeilBrowne183eae2009-03-31 15:20:22 +11002783 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07002784 case ALGORITHM_LEFT_ASYMMETRIC:
2785 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown99c0fb52009-03-31 14:39:38 +11002786 case ALGORITHM_ROTATING_ZERO_RESTART:
2787 case ALGORITHM_ROTATING_N_RESTART:
2788 if (sh->pd_idx == raid_disks-1)
2789 i--; /* Q D D D P */
NeilBrown16a53ec2006-06-26 00:27:38 -07002790 else if (i > sh->pd_idx)
2791 i -= 2; /* D D P Q D */
2792 break;
2793 case ALGORITHM_LEFT_SYMMETRIC:
2794 case ALGORITHM_RIGHT_SYMMETRIC:
2795 if (sh->pd_idx == raid_disks-1)
2796 i--; /* Q D D D P */
2797 else {
2798 /* D D P Q D */
2799 if (i < sh->pd_idx)
2800 i += raid_disks;
2801 i -= (sh->pd_idx + 2);
2802 }
2803 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002804 case ALGORITHM_PARITY_0:
2805 i -= 2;
2806 break;
2807 case ALGORITHM_PARITY_N:
2808 break;
2809 case ALGORITHM_ROTATING_N_CONTINUE:
NeilBrowne4424fe2009-10-16 16:27:34 +11002810 /* Like left_symmetric, but P is before Q */
NeilBrown99c0fb52009-03-31 14:39:38 +11002811 if (sh->pd_idx == 0)
2812 i--; /* P D D D Q */
NeilBrowne4424fe2009-10-16 16:27:34 +11002813 else {
2814 /* D D Q P D */
2815 if (i < sh->pd_idx)
2816 i += raid_disks;
2817 i -= (sh->pd_idx + 1);
2818 }
NeilBrown99c0fb52009-03-31 14:39:38 +11002819 break;
2820 case ALGORITHM_LEFT_ASYMMETRIC_6:
2821 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2822 if (i > sh->pd_idx)
2823 i--;
2824 break;
2825 case ALGORITHM_LEFT_SYMMETRIC_6:
2826 case ALGORITHM_RIGHT_SYMMETRIC_6:
2827 if (i < sh->pd_idx)
2828 i += data_disks + 1;
2829 i -= (sh->pd_idx + 1);
2830 break;
2831 case ALGORITHM_PARITY_0_6:
2832 i -= 1;
2833 break;
NeilBrown16a53ec2006-06-26 00:27:38 -07002834 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002835 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002836 }
2837 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 }
2839
2840 chunk_number = stripe * data_disks + i;
NeilBrown35f2a592010-04-20 14:13:34 +10002841 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842
NeilBrown112bf892009-03-31 14:39:38 +11002843 check = raid5_compute_sector(conf, r_sector,
NeilBrown784052e2009-03-31 15:19:07 +11002844 previous, &dummy1, &sh2);
NeilBrown911d4ee2009-03-31 14:39:38 +11002845 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2846 || sh2.qd_idx != sh->qd_idx) {
NeilBrown0c55e022010-05-03 14:09:02 +10002847 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2848 mdname(conf->mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 return 0;
2850 }
2851 return r_sector;
2852}
2853
Dan Williams600aa102008-06-28 08:32:05 +10002854static void
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002855schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
Dan Williams600aa102008-06-28 08:32:05 +10002856 int rcw, int expand)
Dan Williamse33129d2007-01-02 13:52:30 -07002857{
Markus Stockhausen584acdd2014-12-15 12:57:05 +11002858 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks;
NeilBrownd1688a62011-10-11 16:49:52 +11002859 struct r5conf *conf = sh->raid_conf;
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002860 int level = conf->level;
NeilBrown16a53ec2006-06-26 00:27:38 -07002861
Dan Williamse33129d2007-01-02 13:52:30 -07002862 if (rcw) {
Dan Williamse33129d2007-01-02 13:52:30 -07002863
2864 for (i = disks; i--; ) {
2865 struct r5dev *dev = &sh->dev[i];
2866
2867 if (dev->towrite) {
2868 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsd8ee0722008-06-28 08:32:06 +10002869 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002870 if (!expand)
2871 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002872 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002873 }
2874 }
NeilBrownce7d3632013-03-04 12:37:14 +11002875 /* if we are not expanding this is a proper write request, and
2876 * there will be bios with new data to be drained into the
2877 * stripe cache
2878 */
2879 if (!expand) {
2880 if (!s->locked)
2881 /* False alarm, nothing to do */
2882 return;
2883 sh->reconstruct_state = reconstruct_state_drain_run;
2884 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2885 } else
2886 sh->reconstruct_state = reconstruct_state_run;
2887
2888 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
2889
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002890 if (s->locked + conf->max_degraded == disks)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002891 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002892 atomic_inc(&conf->pending_full_writes);
Dan Williamse33129d2007-01-02 13:52:30 -07002893 } else {
2894 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2895 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
Markus Stockhausen584acdd2014-12-15 12:57:05 +11002896 BUG_ON(level == 6 &&
2897 (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) ||
2898 test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags))));
Dan Williamse33129d2007-01-02 13:52:30 -07002899
Dan Williamse33129d2007-01-02 13:52:30 -07002900 for (i = disks; i--; ) {
2901 struct r5dev *dev = &sh->dev[i];
Markus Stockhausen584acdd2014-12-15 12:57:05 +11002902 if (i == pd_idx || i == qd_idx)
Dan Williamse33129d2007-01-02 13:52:30 -07002903 continue;
2904
Dan Williamse33129d2007-01-02 13:52:30 -07002905 if (dev->towrite &&
2906 (test_bit(R5_UPTODATE, &dev->flags) ||
Dan Williamsd8ee0722008-06-28 08:32:06 +10002907 test_bit(R5_Wantcompute, &dev->flags))) {
2908 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002909 set_bit(R5_LOCKED, &dev->flags);
2910 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002911 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002912 }
2913 }
NeilBrownce7d3632013-03-04 12:37:14 +11002914 if (!s->locked)
2915 /* False alarm - nothing to do */
2916 return;
2917 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
2918 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2919 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2920 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002921 }
2922
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002923 /* keep the parity disk(s) locked while asynchronous operations
Dan Williamse33129d2007-01-02 13:52:30 -07002924 * are in flight
2925 */
2926 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2927 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
Dan Williams600aa102008-06-28 08:32:05 +10002928 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002929
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002930 if (level == 6) {
2931 int qd_idx = sh->qd_idx;
2932 struct r5dev *dev = &sh->dev[qd_idx];
2933
2934 set_bit(R5_LOCKED, &dev->flags);
2935 clear_bit(R5_UPTODATE, &dev->flags);
2936 s->locked++;
2937 }
2938
Dan Williams600aa102008-06-28 08:32:05 +10002939 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07002940 __func__, (unsigned long long)sh->sector,
Dan Williams600aa102008-06-28 08:32:05 +10002941 s->locked, s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002942}
NeilBrown16a53ec2006-06-26 00:27:38 -07002943
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944/*
2945 * Each stripe/dev can have one or more bion attached.
NeilBrown16a53ec2006-06-26 00:27:38 -07002946 * toread/towrite point to the first in a chain.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 * The bi_next chain must be in order.
2948 */
shli@kernel.orgda41ba62014-12-15 12:57:03 +11002949static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
2950 int forwrite, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951{
2952 struct bio **bip;
NeilBrownd1688a62011-10-11 16:49:52 +11002953 struct r5conf *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -07002954 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955
NeilBrowncbe47ec2011-07-26 11:20:35 +10002956 pr_debug("adding bi b#%llu to stripe s#%llu\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002957 (unsigned long long)bi->bi_iter.bi_sector,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 (unsigned long long)sh->sector);
2959
Shaohua Lib17459c2012-07-19 16:01:31 +10002960 /*
2961 * If several bio share a stripe. The bio bi_phys_segments acts as a
2962 * reference count to avoid race. The reference count should already be
2963 * increased before this function is called (for example, in
Shaohua Li849674e2016-01-20 13:52:20 -08002964 * raid5_make_request()), so other bio sharing this stripe will not free the
Shaohua Lib17459c2012-07-19 16:01:31 +10002965 * stripe. If a stripe is owned by one stripe, the stripe lock will
2966 * protect it.
2967 */
2968 spin_lock_irq(&sh->stripe_lock);
shli@kernel.org59fc6302014-12-15 12:57:03 +11002969 /* Don't allow new IO added to stripes in batch list */
2970 if (sh->batch_head)
2971 goto overlap;
NeilBrown72626682005-09-09 16:23:54 -07002972 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 bip = &sh->dev[dd_idx].towrite;
Shaohua Li7eaf7e82012-07-19 16:01:31 +10002974 if (*bip == NULL)
NeilBrown72626682005-09-09 16:23:54 -07002975 firstwrite = 1;
2976 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 bip = &sh->dev[dd_idx].toread;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002978 while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) {
2979 if (bio_end_sector(*bip) > bi->bi_iter.bi_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 goto overlap;
2981 bip = & (*bip)->bi_next;
2982 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07002983 if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 goto overlap;
2985
shli@kernel.orgda41ba62014-12-15 12:57:03 +11002986 if (!forwrite || previous)
2987 clear_bit(STRIPE_BATCH_READY, &sh->state);
2988
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002989 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 if (*bip)
2991 bi->bi_next = *bip;
2992 *bip = bi;
Shaohua Lie7836bd62012-07-19 16:01:31 +10002993 raid5_inc_bi_active_stripes(bi);
NeilBrown72626682005-09-09 16:23:54 -07002994
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 if (forwrite) {
2996 /* check if page is covered */
2997 sector_t sector = sh->dev[dd_idx].sector;
2998 for (bi=sh->dev[dd_idx].towrite;
2999 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
Kent Overstreet4f024f32013-10-11 15:44:27 -07003000 bi && bi->bi_iter.bi_sector <= sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
Kent Overstreetf73a1c72012-09-25 15:05:12 -07003002 if (bio_end_sector(bi) >= sector)
3003 sector = bio_end_sector(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 }
3005 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
shli@kernel.org7a87f432014-12-15 12:57:03 +11003006 if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags))
3007 sh->overwrite_disks++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 }
NeilBrowncbe47ec2011-07-26 11:20:35 +10003009
3010 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
Kent Overstreet4f024f32013-10-11 15:44:27 -07003011 (unsigned long long)(*bip)->bi_iter.bi_sector,
NeilBrowncbe47ec2011-07-26 11:20:35 +10003012 (unsigned long long)sh->sector, dd_idx);
3013
3014 if (conf->mddev->bitmap && firstwrite) {
NeilBrownd0852df52015-05-27 08:43:45 +10003015 /* Cannot hold spinlock over bitmap_startwrite,
3016 * but must ensure this isn't added to a batch until
3017 * we have added to the bitmap and set bm_seq.
3018 * So set STRIPE_BITMAP_PENDING to prevent
3019 * batching.
3020 * If multiple add_stripe_bio() calls race here they
3021 * much all set STRIPE_BITMAP_PENDING. So only the first one
3022 * to complete "bitmap_startwrite" gets to set
3023 * STRIPE_BIT_DELAY. This is important as once a stripe
3024 * is added to a batch, STRIPE_BIT_DELAY cannot be changed
3025 * any more.
3026 */
3027 set_bit(STRIPE_BITMAP_PENDING, &sh->state);
3028 spin_unlock_irq(&sh->stripe_lock);
NeilBrowncbe47ec2011-07-26 11:20:35 +10003029 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
3030 STRIPE_SECTORS, 0);
NeilBrownd0852df52015-05-27 08:43:45 +10003031 spin_lock_irq(&sh->stripe_lock);
3032 clear_bit(STRIPE_BITMAP_PENDING, &sh->state);
3033 if (!sh->batch_head) {
3034 sh->bm_seq = conf->seq_flush+1;
3035 set_bit(STRIPE_BIT_DELAY, &sh->state);
3036 }
NeilBrowncbe47ec2011-07-26 11:20:35 +10003037 }
NeilBrownd0852df52015-05-27 08:43:45 +10003038 spin_unlock_irq(&sh->stripe_lock);
shli@kernel.org59fc6302014-12-15 12:57:03 +11003039
3040 if (stripe_can_batch(sh))
3041 stripe_add_to_batch_list(conf, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 return 1;
3043
3044 overlap:
3045 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
Shaohua Lib17459c2012-07-19 16:01:31 +10003046 spin_unlock_irq(&sh->stripe_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 return 0;
3048}
3049
NeilBrownd1688a62011-10-11 16:49:52 +11003050static void end_reshape(struct r5conf *conf);
NeilBrown29269552006-03-27 01:18:10 -08003051
NeilBrownd1688a62011-10-11 16:49:52 +11003052static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11003053 struct stripe_head *sh)
NeilBrownccfcc3c2006-03-27 01:18:09 -08003054{
NeilBrown784052e2009-03-31 15:19:07 +11003055 int sectors_per_chunk =
Andre Noll09c9e5f2009-06-18 08:45:55 +10003056 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
NeilBrown911d4ee2009-03-31 14:39:38 +11003057 int dd_idx;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07003058 int chunk_offset = sector_div(stripe, sectors_per_chunk);
NeilBrown112bf892009-03-31 14:39:38 +11003059 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07003060
NeilBrown112bf892009-03-31 14:39:38 +11003061 raid5_compute_sector(conf,
3062 stripe * (disks - conf->max_degraded)
NeilBrownb875e532006-12-10 02:20:49 -08003063 *sectors_per_chunk + chunk_offset,
NeilBrown112bf892009-03-31 14:39:38 +11003064 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11003065 &dd_idx, sh);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003066}
3067
Dan Williamsa4456852007-07-09 11:56:43 -07003068static void
NeilBrownd1688a62011-10-11 16:49:52 +11003069handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07003070 struct stripe_head_state *s, int disks,
NeilBrown34a6f802015-08-14 12:07:57 +10003071 struct bio_list *return_bi)
Dan Williamsa4456852007-07-09 11:56:43 -07003072{
3073 int i;
shli@kernel.org59fc6302014-12-15 12:57:03 +11003074 BUG_ON(sh->batch_head);
Dan Williamsa4456852007-07-09 11:56:43 -07003075 for (i = disks; i--; ) {
3076 struct bio *bi;
3077 int bitmap_end = 0;
3078
3079 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrown3cb03002011-10-11 16:45:26 +11003080 struct md_rdev *rdev;
Dan Williamsa4456852007-07-09 11:56:43 -07003081 rcu_read_lock();
3082 rdev = rcu_dereference(conf->disks[i].rdev);
3083 if (rdev && test_bit(In_sync, &rdev->flags))
NeilBrown7f0da592011-07-28 11:39:22 +10003084 atomic_inc(&rdev->nr_pending);
3085 else
3086 rdev = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07003087 rcu_read_unlock();
NeilBrown7f0da592011-07-28 11:39:22 +10003088 if (rdev) {
3089 if (!rdev_set_badblocks(
3090 rdev,
3091 sh->sector,
3092 STRIPE_SECTORS, 0))
3093 md_error(conf->mddev, rdev);
3094 rdev_dec_pending(rdev, conf->mddev);
3095 }
Dan Williamsa4456852007-07-09 11:56:43 -07003096 }
Shaohua Lib17459c2012-07-19 16:01:31 +10003097 spin_lock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07003098 /* fail all writes first */
3099 bi = sh->dev[i].towrite;
3100 sh->dev[i].towrite = NULL;
shli@kernel.org7a87f432014-12-15 12:57:03 +11003101 sh->overwrite_disks = 0;
Shaohua Lib17459c2012-07-19 16:01:31 +10003102 spin_unlock_irq(&sh->stripe_lock);
NeilBrown1ed850f2012-10-11 13:50:13 +11003103 if (bi)
Dan Williamsa4456852007-07-09 11:56:43 -07003104 bitmap_end = 1;
Dan Williamsa4456852007-07-09 11:56:43 -07003105
Shaohua Li0576b1c2015-08-13 14:32:00 -07003106 r5l_stripe_write_finished(sh);
3107
Dan Williamsa4456852007-07-09 11:56:43 -07003108 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3109 wake_up(&conf->wait_for_overlap);
3110
Kent Overstreet4f024f32013-10-11 15:44:27 -07003111 while (bi && bi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07003112 sh->dev[i].sector + STRIPE_SECTORS) {
3113 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003114
3115 bi->bi_error = -EIO;
Shaohua Lie7836bd62012-07-19 16:01:31 +10003116 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003117 md_write_end(conf->mddev);
NeilBrown34a6f802015-08-14 12:07:57 +10003118 bio_list_add(return_bi, bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003119 }
3120 bi = nextbi;
3121 }
Shaohua Li7eaf7e82012-07-19 16:01:31 +10003122 if (bitmap_end)
3123 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3124 STRIPE_SECTORS, 0, 0);
3125 bitmap_end = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07003126 /* and fail all 'written' */
3127 bi = sh->dev[i].written;
3128 sh->dev[i].written = NULL;
Shaohua Lid592a992014-05-21 17:57:44 +08003129 if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) {
3130 WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
3131 sh->dev[i].page = sh->dev[i].orig_page;
3132 }
3133
Dan Williamsa4456852007-07-09 11:56:43 -07003134 if (bi) bitmap_end = 1;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003135 while (bi && bi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07003136 sh->dev[i].sector + STRIPE_SECTORS) {
3137 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003138
3139 bi->bi_error = -EIO;
Shaohua Lie7836bd62012-07-19 16:01:31 +10003140 if (!raid5_dec_bi_active_stripes(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003141 md_write_end(conf->mddev);
NeilBrown34a6f802015-08-14 12:07:57 +10003142 bio_list_add(return_bi, bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003143 }
3144 bi = bi2;
3145 }
3146
Dan Williamsb5e98d62007-01-02 13:52:31 -07003147 /* fail any reads if this device is non-operational and
3148 * the data has not reached the cache yet.
3149 */
3150 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
Shaohua Li6e74a9c2015-10-08 21:54:08 -07003151 s->failed > conf->max_degraded &&
Dan Williamsb5e98d62007-01-02 13:52:31 -07003152 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
3153 test_bit(R5_ReadError, &sh->dev[i].flags))) {
NeilBrown143c4d02012-10-11 13:50:12 +11003154 spin_lock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07003155 bi = sh->dev[i].toread;
3156 sh->dev[i].toread = NULL;
NeilBrown143c4d02012-10-11 13:50:12 +11003157 spin_unlock_irq(&sh->stripe_lock);
Dan Williamsa4456852007-07-09 11:56:43 -07003158 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
3159 wake_up(&conf->wait_for_overlap);
Shaohua Liebda7802015-09-18 10:20:13 -07003160 if (bi)
3161 s->to_read--;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003162 while (bi && bi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07003163 sh->dev[i].sector + STRIPE_SECTORS) {
3164 struct bio *nextbi =
3165 r5_next_bio(bi, sh->dev[i].sector);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003166
3167 bi->bi_error = -EIO;
NeilBrown34a6f802015-08-14 12:07:57 +10003168 if (!raid5_dec_bi_active_stripes(bi))
3169 bio_list_add(return_bi, bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003170 bi = nextbi;
3171 }
3172 }
Dan Williamsa4456852007-07-09 11:56:43 -07003173 if (bitmap_end)
3174 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3175 STRIPE_SECTORS, 0, 0);
NeilBrown8cfa7b02011-07-27 11:00:36 +10003176 /* If we were in the middle of a write the parity block might
3177 * still be locked - so just clear all R5_LOCKED flags
3178 */
3179 clear_bit(R5_LOCKED, &sh->dev[i].flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003180 }
Shaohua Liebda7802015-09-18 10:20:13 -07003181 s->to_write = 0;
3182 s->written = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07003183
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003184 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3185 if (atomic_dec_and_test(&conf->pending_full_writes))
3186 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07003187}
3188
NeilBrown7f0da592011-07-28 11:39:22 +10003189static void
NeilBrownd1688a62011-10-11 16:49:52 +11003190handle_failed_sync(struct r5conf *conf, struct stripe_head *sh,
NeilBrown7f0da592011-07-28 11:39:22 +10003191 struct stripe_head_state *s)
3192{
3193 int abort = 0;
3194 int i;
3195
shli@kernel.org59fc6302014-12-15 12:57:03 +11003196 BUG_ON(sh->batch_head);
NeilBrown7f0da592011-07-28 11:39:22 +10003197 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003198 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
3199 wake_up(&conf->wait_for_overlap);
NeilBrown7f0da592011-07-28 11:39:22 +10003200 s->syncing = 0;
NeilBrown9a3e1102011-12-23 10:17:53 +11003201 s->replacing = 0;
NeilBrown7f0da592011-07-28 11:39:22 +10003202 /* There is nothing more to do for sync/check/repair.
NeilBrown18b98372012-04-01 23:48:38 +10003203 * Don't even need to abort as that is handled elsewhere
3204 * if needed, and not always wanted e.g. if there is a known
3205 * bad block here.
NeilBrown9a3e1102011-12-23 10:17:53 +11003206 * For recover/replace we need to record a bad block on all
NeilBrown7f0da592011-07-28 11:39:22 +10003207 * non-sync devices, or abort the recovery
3208 */
NeilBrown18b98372012-04-01 23:48:38 +10003209 if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) {
3210 /* During recovery devices cannot be removed, so
3211 * locking and refcounting of rdevs is not needed
3212 */
3213 for (i = 0; i < conf->raid_disks; i++) {
3214 struct md_rdev *rdev = conf->disks[i].rdev;
3215 if (rdev
3216 && !test_bit(Faulty, &rdev->flags)
3217 && !test_bit(In_sync, &rdev->flags)
3218 && !rdev_set_badblocks(rdev, sh->sector,
3219 STRIPE_SECTORS, 0))
3220 abort = 1;
3221 rdev = conf->disks[i].replacement;
3222 if (rdev
3223 && !test_bit(Faulty, &rdev->flags)
3224 && !test_bit(In_sync, &rdev->flags)
3225 && !rdev_set_badblocks(rdev, sh->sector,
3226 STRIPE_SECTORS, 0))
3227 abort = 1;
3228 }
3229 if (abort)
3230 conf->recovery_disabled =
3231 conf->mddev->recovery_disabled;
NeilBrown7f0da592011-07-28 11:39:22 +10003232 }
NeilBrown18b98372012-04-01 23:48:38 +10003233 md_done_sync(conf->mddev, STRIPE_SECTORS, !abort);
NeilBrown7f0da592011-07-28 11:39:22 +10003234}
3235
NeilBrown9a3e1102011-12-23 10:17:53 +11003236static int want_replace(struct stripe_head *sh, int disk_idx)
3237{
3238 struct md_rdev *rdev;
3239 int rv = 0;
3240 /* Doing recovery so rcu locking not required */
3241 rdev = sh->raid_conf->disks[disk_idx].replacement;
3242 if (rdev
3243 && !test_bit(Faulty, &rdev->flags)
3244 && !test_bit(In_sync, &rdev->flags)
3245 && (rdev->recovery_offset <= sh->sector
3246 || rdev->mddev->recovery_cp <= sh->sector))
3247 rv = 1;
3248
3249 return rv;
3250}
3251
NeilBrown93b3dbc2011-07-27 11:00:36 +10003252/* fetch_block - checks the given member device to see if its data needs
Dan Williams1fe797e2008-06-28 09:16:30 +10003253 * to be read or computed to satisfy a request.
3254 *
3255 * Returns 1 when no more member devices need to be checked, otherwise returns
NeilBrown93b3dbc2011-07-27 11:00:36 +10003256 * 0 to tell the loop in handle_stripe_fill to continue
Dan Williamsf38e1212007-01-02 13:52:30 -07003257 */
NeilBrown2c58f062015-02-02 11:32:23 +11003258
3259static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s,
3260 int disk_idx, int disks)
Dan Williamsf38e1212007-01-02 13:52:30 -07003261{
3262 struct r5dev *dev = &sh->dev[disk_idx];
NeilBrown93b3dbc2011-07-27 11:00:36 +10003263 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
3264 &sh->dev[s->failed_num[1]] };
NeilBrownea664c82015-02-02 14:03:28 +11003265 int i;
Dan Williamsf38e1212007-01-02 13:52:30 -07003266
NeilBrowna79cfe12015-02-02 11:37:59 +11003267
3268 if (test_bit(R5_LOCKED, &dev->flags) ||
3269 test_bit(R5_UPTODATE, &dev->flags))
3270 /* No point reading this as we already have it or have
3271 * decided to get it.
3272 */
3273 return 0;
3274
3275 if (dev->toread ||
3276 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)))
3277 /* We need this block to directly satisfy a request */
3278 return 1;
3279
3280 if (s->syncing || s->expanding ||
3281 (s->replacing && want_replace(sh, disk_idx)))
3282 /* When syncing, or expanding we read everything.
3283 * When replacing, we need the replaced block.
3284 */
3285 return 1;
3286
3287 if ((s->failed >= 1 && fdev[0]->toread) ||
3288 (s->failed >= 2 && fdev[1]->toread))
3289 /* If we want to read from a failed device, then
3290 * we need to actually read every other device.
3291 */
3292 return 1;
3293
NeilBrowna9d56952015-02-02 11:49:10 +11003294 /* Sometimes neither read-modify-write nor reconstruct-write
3295 * cycles can work. In those cases we read every block we
3296 * can. Then the parity-update is certain to have enough to
3297 * work with.
3298 * This can only be a problem when we need to write something,
3299 * and some device has failed. If either of those tests
3300 * fail we need look no further.
3301 */
3302 if (!s->failed || !s->to_write)
3303 return 0;
3304
3305 if (test_bit(R5_Insync, &dev->flags) &&
3306 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3307 /* Pre-reads at not permitted until after short delay
3308 * to gather multiple requests. However if this
3309 * device is no Insync, the block could only be be computed
3310 * and there is no need to delay that.
3311 */
3312 return 0;
NeilBrownea664c82015-02-02 14:03:28 +11003313
NeilBrown36707bb2015-09-24 15:25:36 +10003314 for (i = 0; i < s->failed && i < 2; i++) {
NeilBrownea664c82015-02-02 14:03:28 +11003315 if (fdev[i]->towrite &&
3316 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
3317 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3318 /* If we have a partial write to a failed
3319 * device, then we will need to reconstruct
3320 * the content of that device, so all other
3321 * devices must be read.
3322 */
3323 return 1;
3324 }
3325
3326 /* If we are forced to do a reconstruct-write, either because
3327 * the current RAID6 implementation only supports that, or
3328 * or because parity cannot be trusted and we are currently
3329 * recovering it, there is extra need to be careful.
3330 * If one of the devices that we would need to read, because
3331 * it is not being overwritten (and maybe not written at all)
3332 * is missing/faulty, then we need to read everything we can.
3333 */
3334 if (sh->raid_conf->level != 6 &&
3335 sh->sector < sh->raid_conf->mddev->recovery_cp)
3336 /* reconstruct-write isn't being forced */
3337 return 0;
NeilBrown36707bb2015-09-24 15:25:36 +10003338 for (i = 0; i < s->failed && i < 2; i++) {
NeilBrown10d82c52015-05-08 18:19:33 +10003339 if (s->failed_num[i] != sh->pd_idx &&
3340 s->failed_num[i] != sh->qd_idx &&
3341 !test_bit(R5_UPTODATE, &fdev[i]->flags) &&
NeilBrownea664c82015-02-02 14:03:28 +11003342 !test_bit(R5_OVERWRITE, &fdev[i]->flags))
3343 return 1;
3344 }
3345
NeilBrown2c58f062015-02-02 11:32:23 +11003346 return 0;
3347}
3348
3349static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s,
3350 int disk_idx, int disks)
3351{
3352 struct r5dev *dev = &sh->dev[disk_idx];
3353
3354 /* is the data in this block needed, and can we get it? */
3355 if (need_this_block(sh, s, disk_idx, disks)) {
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003356 /* we would like to get this block, possibly by computing it,
3357 * otherwise read it if the backing disk is insync
3358 */
3359 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
3360 BUG_ON(test_bit(R5_Wantread, &dev->flags));
NeilBrownb0c783b2015-05-08 18:19:32 +10003361 BUG_ON(sh->batch_head);
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003362 if ((s->uptodate == disks - 1) &&
NeilBrownf2b3b442011-07-26 11:35:19 +10003363 (s->failed && (disk_idx == s->failed_num[0] ||
3364 disk_idx == s->failed_num[1]))) {
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003365 /* have disk failed, and we're requested to fetch it;
3366 * do compute it
3367 */
3368 pr_debug("Computing stripe %llu block %d\n",
3369 (unsigned long long)sh->sector, disk_idx);
3370 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3371 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3372 set_bit(R5_Wantcompute, &dev->flags);
3373 sh->ops.target = disk_idx;
3374 sh->ops.target2 = -1; /* no 2nd target */
3375 s->req_compute = 1;
NeilBrown93b3dbc2011-07-27 11:00:36 +10003376 /* Careful: from this point on 'uptodate' is in the eye
3377 * of raid_run_ops which services 'compute' operations
3378 * before writes. R5_Wantcompute flags a block that will
3379 * be R5_UPTODATE by the time it is needed for a
3380 * subsequent operation.
3381 */
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003382 s->uptodate++;
3383 return 1;
3384 } else if (s->uptodate == disks-2 && s->failed >= 2) {
3385 /* Computing 2-failure is *very* expensive; only
3386 * do it if failed >= 2
3387 */
3388 int other;
3389 for (other = disks; other--; ) {
3390 if (other == disk_idx)
3391 continue;
3392 if (!test_bit(R5_UPTODATE,
3393 &sh->dev[other].flags))
3394 break;
3395 }
3396 BUG_ON(other < 0);
3397 pr_debug("Computing stripe %llu blocks %d,%d\n",
3398 (unsigned long long)sh->sector,
3399 disk_idx, other);
3400 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3401 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3402 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
3403 set_bit(R5_Wantcompute, &sh->dev[other].flags);
3404 sh->ops.target = disk_idx;
3405 sh->ops.target2 = other;
3406 s->uptodate += 2;
3407 s->req_compute = 1;
3408 return 1;
3409 } else if (test_bit(R5_Insync, &dev->flags)) {
3410 set_bit(R5_LOCKED, &dev->flags);
3411 set_bit(R5_Wantread, &dev->flags);
3412 s->locked++;
3413 pr_debug("Reading block %d (sync=%d)\n",
3414 disk_idx, s->syncing);
3415 }
3416 }
3417
3418 return 0;
3419}
3420
3421/**
NeilBrown93b3dbc2011-07-27 11:00:36 +10003422 * handle_stripe_fill - read or compute data to satisfy pending requests.
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003423 */
NeilBrown93b3dbc2011-07-27 11:00:36 +10003424static void handle_stripe_fill(struct stripe_head *sh,
3425 struct stripe_head_state *s,
3426 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07003427{
3428 int i;
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003429
3430 /* look for blocks to read/compute, skip this if a compute
3431 * is already in flight, or if the stripe contents are in the
3432 * midst of changing due to a write
3433 */
3434 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
3435 !sh->reconstruct_state)
3436 for (i = disks; i--; )
NeilBrown93b3dbc2011-07-27 11:00:36 +10003437 if (fetch_block(sh, s, i, disks))
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07003438 break;
Dan Williamsa4456852007-07-09 11:56:43 -07003439 set_bit(STRIPE_HANDLE, &sh->state);
3440}
3441
NeilBrown787b76f2015-05-21 12:56:41 +10003442static void break_stripe_batch_list(struct stripe_head *head_sh,
3443 unsigned long handle_flags);
Dan Williams1fe797e2008-06-28 09:16:30 +10003444/* handle_stripe_clean_event
Dan Williamsa4456852007-07-09 11:56:43 -07003445 * any written block on an uptodate or failed drive can be returned.
3446 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
3447 * never LOCKED, so we don't need to test 'failed' directly.
3448 */
NeilBrownd1688a62011-10-11 16:49:52 +11003449static void handle_stripe_clean_event(struct r5conf *conf,
NeilBrown34a6f802015-08-14 12:07:57 +10003450 struct stripe_head *sh, int disks, struct bio_list *return_bi)
Dan Williamsa4456852007-07-09 11:56:43 -07003451{
3452 int i;
3453 struct r5dev *dev;
NeilBrownf8dfcff2013-03-12 12:18:06 +11003454 int discard_pending = 0;
shli@kernel.org59fc6302014-12-15 12:57:03 +11003455 struct stripe_head *head_sh = sh;
3456 bool do_endio = false;
Dan Williamsa4456852007-07-09 11:56:43 -07003457
3458 for (i = disks; i--; )
3459 if (sh->dev[i].written) {
3460 dev = &sh->dev[i];
3461 if (!test_bit(R5_LOCKED, &dev->flags) &&
Shaohua Li9e4447682012-10-11 13:49:49 +11003462 (test_bit(R5_UPTODATE, &dev->flags) ||
Shaohua Lid592a992014-05-21 17:57:44 +08003463 test_bit(R5_Discard, &dev->flags) ||
3464 test_bit(R5_SkipCopy, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07003465 /* We can return any write requests */
3466 struct bio *wbi, *wbi2;
Dan Williams45b42332007-07-09 11:56:43 -07003467 pr_debug("Return write for disc %d\n", i);
NeilBrownca64cae2012-11-21 16:33:40 +11003468 if (test_and_clear_bit(R5_Discard, &dev->flags))
3469 clear_bit(R5_UPTODATE, &dev->flags);
Shaohua Lid592a992014-05-21 17:57:44 +08003470 if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) {
3471 WARN_ON(test_bit(R5_UPTODATE, &dev->flags));
Shaohua Lid592a992014-05-21 17:57:44 +08003472 }
shli@kernel.org59fc6302014-12-15 12:57:03 +11003473 do_endio = true;
3474
3475returnbi:
3476 dev->page = dev->orig_page;
Dan Williamsa4456852007-07-09 11:56:43 -07003477 wbi = dev->written;
3478 dev->written = NULL;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003479 while (wbi && wbi->bi_iter.bi_sector <
Dan Williamsa4456852007-07-09 11:56:43 -07003480 dev->sector + STRIPE_SECTORS) {
3481 wbi2 = r5_next_bio(wbi, dev->sector);
Shaohua Lie7836bd62012-07-19 16:01:31 +10003482 if (!raid5_dec_bi_active_stripes(wbi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003483 md_write_end(conf->mddev);
NeilBrown34a6f802015-08-14 12:07:57 +10003484 bio_list_add(return_bi, wbi);
Dan Williamsa4456852007-07-09 11:56:43 -07003485 }
3486 wbi = wbi2;
3487 }
Shaohua Li7eaf7e82012-07-19 16:01:31 +10003488 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
3489 STRIPE_SECTORS,
Dan Williamsa4456852007-07-09 11:56:43 -07003490 !test_bit(STRIPE_DEGRADED, &sh->state),
Shaohua Li7eaf7e82012-07-19 16:01:31 +10003491 0);
shli@kernel.org59fc6302014-12-15 12:57:03 +11003492 if (head_sh->batch_head) {
3493 sh = list_first_entry(&sh->batch_list,
3494 struct stripe_head,
3495 batch_list);
3496 if (sh != head_sh) {
3497 dev = &sh->dev[i];
3498 goto returnbi;
3499 }
3500 }
3501 sh = head_sh;
3502 dev = &sh->dev[i];
NeilBrownf8dfcff2013-03-12 12:18:06 +11003503 } else if (test_bit(R5_Discard, &dev->flags))
3504 discard_pending = 1;
Shaohua Lid592a992014-05-21 17:57:44 +08003505 WARN_ON(test_bit(R5_SkipCopy, &dev->flags));
3506 WARN_ON(dev->page != dev->orig_page);
NeilBrownf8dfcff2013-03-12 12:18:06 +11003507 }
Shaohua Lif6bed0e2015-08-13 14:31:59 -07003508
Shaohua Li0576b1c2015-08-13 14:32:00 -07003509 r5l_stripe_write_finished(sh);
3510
NeilBrownf8dfcff2013-03-12 12:18:06 +11003511 if (!discard_pending &&
3512 test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) {
Roman Gushchinb8a9d662015-10-31 10:53:50 +11003513 int hash;
NeilBrownf8dfcff2013-03-12 12:18:06 +11003514 clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags);
3515 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
3516 if (sh->qd_idx >= 0) {
3517 clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags);
3518 clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags);
3519 }
3520 /* now that discard is done we can proceed with any sync */
3521 clear_bit(STRIPE_DISCARD, &sh->state);
Shaohua Lid47648f2013-10-19 14:51:42 +08003522 /*
3523 * SCSI discard will change some bio fields and the stripe has
3524 * no updated data, so remove it from hash list and the stripe
3525 * will be reinitialized
3526 */
shli@kernel.org59fc6302014-12-15 12:57:03 +11003527unhash:
Roman Gushchinb8a9d662015-10-31 10:53:50 +11003528 hash = sh->hash_lock_index;
3529 spin_lock_irq(conf->hash_locks + hash);
Shaohua Lid47648f2013-10-19 14:51:42 +08003530 remove_hash(sh);
Roman Gushchinb8a9d662015-10-31 10:53:50 +11003531 spin_unlock_irq(conf->hash_locks + hash);
shli@kernel.org59fc6302014-12-15 12:57:03 +11003532 if (head_sh->batch_head) {
3533 sh = list_first_entry(&sh->batch_list,
3534 struct stripe_head, batch_list);
3535 if (sh != head_sh)
3536 goto unhash;
3537 }
shli@kernel.org59fc6302014-12-15 12:57:03 +11003538 sh = head_sh;
3539
NeilBrownf8dfcff2013-03-12 12:18:06 +11003540 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state))
3541 set_bit(STRIPE_HANDLE, &sh->state);
3542
3543 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003544
3545 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
3546 if (atomic_dec_and_test(&conf->pending_full_writes))
3547 md_wakeup_thread(conf->mddev->thread);
shli@kernel.org59fc6302014-12-15 12:57:03 +11003548
NeilBrown787b76f2015-05-21 12:56:41 +10003549 if (head_sh->batch_head && do_endio)
3550 break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
Dan Williamsa4456852007-07-09 11:56:43 -07003551}
3552
NeilBrownd1688a62011-10-11 16:49:52 +11003553static void handle_stripe_dirtying(struct r5conf *conf,
NeilBrownc8ac1802011-07-27 11:00:36 +10003554 struct stripe_head *sh,
3555 struct stripe_head_state *s,
3556 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07003557{
3558 int rmw = 0, rcw = 0, i;
Alexander Lyakasa7854482012-10-11 13:50:12 +11003559 sector_t recovery_cp = conf->mddev->recovery_cp;
3560
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003561 /* Check whether resync is now happening or should start.
Alexander Lyakasa7854482012-10-11 13:50:12 +11003562 * If yes, then the array is dirty (after unclean shutdown or
3563 * initial creation), so parity in some stripes might be inconsistent.
3564 * In this case, we need to always do reconstruct-write, to ensure
3565 * that in case of drive failure or read-error correction, we
3566 * generate correct data from the parity.
3567 */
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003568 if (conf->rmw_level == PARITY_DISABLE_RMW ||
NeilBrown26ac1072015-02-18 11:35:14 +11003569 (recovery_cp < MaxSector && sh->sector >= recovery_cp &&
3570 s->failed == 0)) {
Alexander Lyakasa7854482012-10-11 13:50:12 +11003571 /* Calculate the real rcw later - for now make it
NeilBrownc8ac1802011-07-27 11:00:36 +10003572 * look like rcw is cheaper
3573 */
3574 rcw = 1; rmw = 2;
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003575 pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n",
3576 conf->rmw_level, (unsigned long long)recovery_cp,
Alexander Lyakasa7854482012-10-11 13:50:12 +11003577 (unsigned long long)sh->sector);
NeilBrownc8ac1802011-07-27 11:00:36 +10003578 } else for (i = disks; i--; ) {
Dan Williamsa4456852007-07-09 11:56:43 -07003579 /* would I have to read this buffer for read_modify_write */
3580 struct r5dev *dev = &sh->dev[i];
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003581 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
Dan Williamsa4456852007-07-09 11:56:43 -07003582 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003583 !(test_bit(R5_UPTODATE, &dev->flags) ||
3584 test_bit(R5_Wantcompute, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07003585 if (test_bit(R5_Insync, &dev->flags))
3586 rmw++;
3587 else
3588 rmw += 2*disks; /* cannot read it */
3589 }
3590 /* Would I have to read this buffer for reconstruct_write */
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003591 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
3592 i != sh->pd_idx && i != sh->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07003593 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003594 !(test_bit(R5_UPTODATE, &dev->flags) ||
3595 test_bit(R5_Wantcompute, &dev->flags))) {
NeilBrown67f45542014-05-28 13:39:22 +10003596 if (test_bit(R5_Insync, &dev->flags))
3597 rcw++;
Dan Williamsa4456852007-07-09 11:56:43 -07003598 else
3599 rcw += 2*disks;
3600 }
3601 }
Dan Williams45b42332007-07-09 11:56:43 -07003602 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003603 (unsigned long long)sh->sector, rmw, rcw);
3604 set_bit(STRIPE_HANDLE, &sh->state);
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003605 if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_ENABLE_RMW)) && rmw > 0) {
Dan Williamsa4456852007-07-09 11:56:43 -07003606 /* prefer read-modify-write, but need to get some data */
Jonathan Brassowe3620a32013-03-07 16:22:01 -06003607 if (conf->mddev->queue)
3608 blk_add_trace_msg(conf->mddev->queue,
3609 "raid5 rmw %llu %d",
3610 (unsigned long long)sh->sector, rmw);
Dan Williamsa4456852007-07-09 11:56:43 -07003611 for (i = disks; i--; ) {
3612 struct r5dev *dev = &sh->dev[i];
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003613 if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) &&
Dan Williamsa4456852007-07-09 11:56:43 -07003614 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003615 !(test_bit(R5_UPTODATE, &dev->flags) ||
3616 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07003617 test_bit(R5_Insync, &dev->flags)) {
NeilBrown67f45542014-05-28 13:39:22 +10003618 if (test_bit(STRIPE_PREREAD_ACTIVE,
3619 &sh->state)) {
3620 pr_debug("Read_old block %d for r-m-w\n",
3621 i);
Dan Williamsa4456852007-07-09 11:56:43 -07003622 set_bit(R5_LOCKED, &dev->flags);
3623 set_bit(R5_Wantread, &dev->flags);
3624 s->locked++;
3625 } else {
3626 set_bit(STRIPE_DELAYED, &sh->state);
3627 set_bit(STRIPE_HANDLE, &sh->state);
3628 }
3629 }
3630 }
NeilBrowna9add5d2012-10-31 11:59:09 +11003631 }
Markus Stockhausen584acdd2014-12-15 12:57:05 +11003632 if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_ENABLE_RMW)) && rcw > 0) {
Dan Williamsa4456852007-07-09 11:56:43 -07003633 /* want reconstruct write, but need to get some data */
NeilBrowna9add5d2012-10-31 11:59:09 +11003634 int qread =0;
NeilBrownc8ac1802011-07-27 11:00:36 +10003635 rcw = 0;
Dan Williamsa4456852007-07-09 11:56:43 -07003636 for (i = disks; i--; ) {
3637 struct r5dev *dev = &sh->dev[i];
3638 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
NeilBrownc8ac1802011-07-27 11:00:36 +10003639 i != sh->pd_idx && i != sh->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07003640 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07003641 !(test_bit(R5_UPTODATE, &dev->flags) ||
NeilBrownc8ac1802011-07-27 11:00:36 +10003642 test_bit(R5_Wantcompute, &dev->flags))) {
3643 rcw++;
NeilBrown67f45542014-05-28 13:39:22 +10003644 if (test_bit(R5_Insync, &dev->flags) &&
3645 test_bit(STRIPE_PREREAD_ACTIVE,
3646 &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07003647 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07003648 "%d for Reconstruct\n", i);
3649 set_bit(R5_LOCKED, &dev->flags);
3650 set_bit(R5_Wantread, &dev->flags);
3651 s->locked++;
NeilBrowna9add5d2012-10-31 11:59:09 +11003652 qread++;
Dan Williamsa4456852007-07-09 11:56:43 -07003653 } else {
3654 set_bit(STRIPE_DELAYED, &sh->state);
3655 set_bit(STRIPE_HANDLE, &sh->state);
3656 }
3657 }
3658 }
Jonathan Brassowe3620a32013-03-07 16:22:01 -06003659 if (rcw && conf->mddev->queue)
NeilBrowna9add5d2012-10-31 11:59:09 +11003660 blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d",
3661 (unsigned long long)sh->sector,
3662 rcw, qread, test_bit(STRIPE_DELAYED, &sh->state));
NeilBrownc8ac1802011-07-27 11:00:36 +10003663 }
NeilBrownb1b02fe2015-02-02 10:44:29 +11003664
3665 if (rcw > disks && rmw > disks &&
3666 !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3667 set_bit(STRIPE_DELAYED, &sh->state);
3668
Dan Williamsa4456852007-07-09 11:56:43 -07003669 /* now if nothing is locked, and if we have enough data,
3670 * we can start a write request
3671 */
Dan Williamsf38e1212007-01-02 13:52:30 -07003672 /* since handle_stripe can be called at any time we need to handle the
3673 * case where a compute block operation has been submitted and then a
Dan Williamsac6b53b2009-07-14 13:40:19 -07003674 * subsequent call wants to start a write request. raid_run_ops only
3675 * handles the case where compute block and reconstruct are requested
Dan Williamsf38e1212007-01-02 13:52:30 -07003676 * simultaneously. If this is not the case then new writes need to be
3677 * held off until the compute completes.
3678 */
Dan Williams976ea8d2008-06-28 08:32:03 +10003679 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
3680 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
3681 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07003682 schedule_reconstruction(sh, s, rcw == 0, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07003683}
3684
NeilBrownd1688a62011-10-11 16:49:52 +11003685static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07003686 struct stripe_head_state *s, int disks)
3687{
Dan Williamsecc65c92008-06-28 08:31:57 +10003688 struct r5dev *dev = NULL;
Dan Williamse89f8962007-01-02 13:52:31 -07003689
shli@kernel.org59fc6302014-12-15 12:57:03 +11003690 BUG_ON(sh->batch_head);
Dan Williamsbd2ab672008-04-10 21:29:27 -07003691 set_bit(STRIPE_HANDLE, &sh->state);
3692
Dan Williamsecc65c92008-06-28 08:31:57 +10003693 switch (sh->check_state) {
3694 case check_state_idle:
3695 /* start a new check operation if there are no failures */
Dan Williamsbd2ab672008-04-10 21:29:27 -07003696 if (s->failed == 0) {
Dan Williamsbd2ab672008-04-10 21:29:27 -07003697 BUG_ON(s->uptodate != disks);
Dan Williamsecc65c92008-06-28 08:31:57 +10003698 sh->check_state = check_state_run;
3699 set_bit(STRIPE_OP_CHECK, &s->ops_request);
Dan Williamsbd2ab672008-04-10 21:29:27 -07003700 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
Dan Williamsbd2ab672008-04-10 21:29:27 -07003701 s->uptodate--;
Dan Williamsecc65c92008-06-28 08:31:57 +10003702 break;
Dan Williamsbd2ab672008-04-10 21:29:27 -07003703 }
NeilBrownf2b3b442011-07-26 11:35:19 +10003704 dev = &sh->dev[s->failed_num[0]];
Dan Williamsecc65c92008-06-28 08:31:57 +10003705 /* fall through */
3706 case check_state_compute_result:
3707 sh->check_state = check_state_idle;
3708 if (!dev)
3709 dev = &sh->dev[sh->pd_idx];
3710
3711 /* check that a write has not made the stripe insync */
3712 if (test_bit(STRIPE_INSYNC, &sh->state))
3713 break;
3714
3715 /* either failed parity check, or recovery is happening */
Dan Williamsa4456852007-07-09 11:56:43 -07003716 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3717 BUG_ON(s->uptodate != disks);
3718
3719 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsecc65c92008-06-28 08:31:57 +10003720 s->locked++;
Dan Williamsa4456852007-07-09 11:56:43 -07003721 set_bit(R5_Wantwrite, &dev->flags);
Dan Williams830ea012007-01-02 13:52:31 -07003722
Dan Williamsa4456852007-07-09 11:56:43 -07003723 clear_bit(STRIPE_DEGRADED, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003724 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10003725 break;
3726 case check_state_run:
3727 break; /* we will be called again upon completion */
3728 case check_state_check_result:
3729 sh->check_state = check_state_idle;
3730
3731 /* if a failure occurred during the check operation, leave
3732 * STRIPE_INSYNC not set and let the stripe be handled again
3733 */
3734 if (s->failed)
3735 break;
3736
3737 /* handle a successful check operation, if parity is correct
3738 * we are done. Otherwise update the mismatch count and repair
3739 * parity if !MD_RECOVERY_CHECK
3740 */
Dan Williamsad283ea2009-08-29 19:09:26 -07003741 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
Dan Williamsecc65c92008-06-28 08:31:57 +10003742 /* parity is correct (on disc,
3743 * not in buffer any more)
3744 */
3745 set_bit(STRIPE_INSYNC, &sh->state);
3746 else {
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11003747 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
Dan Williamsecc65c92008-06-28 08:31:57 +10003748 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3749 /* don't try to repair!! */
3750 set_bit(STRIPE_INSYNC, &sh->state);
3751 else {
3752 sh->check_state = check_state_compute_run;
Dan Williams976ea8d2008-06-28 08:32:03 +10003753 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10003754 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3755 set_bit(R5_Wantcompute,
3756 &sh->dev[sh->pd_idx].flags);
3757 sh->ops.target = sh->pd_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07003758 sh->ops.target2 = -1;
Dan Williamsecc65c92008-06-28 08:31:57 +10003759 s->uptodate++;
3760 }
3761 }
3762 break;
3763 case check_state_compute_run:
3764 break;
3765 default:
3766 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3767 __func__, sh->check_state,
3768 (unsigned long long) sh->sector);
3769 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07003770 }
3771}
3772
NeilBrownd1688a62011-10-11 16:49:52 +11003773static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh,
Dan Williams36d1c642009-07-14 11:48:22 -07003774 struct stripe_head_state *s,
NeilBrownf2b3b442011-07-26 11:35:19 +10003775 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07003776{
Dan Williamsa4456852007-07-09 11:56:43 -07003777 int pd_idx = sh->pd_idx;
NeilBrown34e04e82009-03-31 15:10:16 +11003778 int qd_idx = sh->qd_idx;
Dan Williamsd82dfee2009-07-14 13:40:57 -07003779 struct r5dev *dev;
Dan Williamsa4456852007-07-09 11:56:43 -07003780
shli@kernel.org59fc6302014-12-15 12:57:03 +11003781 BUG_ON(sh->batch_head);
Dan Williamsa4456852007-07-09 11:56:43 -07003782 set_bit(STRIPE_HANDLE, &sh->state);
3783
3784 BUG_ON(s->failed > 2);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003785
Dan Williamsa4456852007-07-09 11:56:43 -07003786 /* Want to check and possibly repair P and Q.
3787 * However there could be one 'failed' device, in which
3788 * case we can only check one of them, possibly using the
3789 * other to generate missing data
3790 */
3791
Dan Williamsd82dfee2009-07-14 13:40:57 -07003792 switch (sh->check_state) {
3793 case check_state_idle:
3794 /* start a new check operation if there are < 2 failures */
NeilBrownf2b3b442011-07-26 11:35:19 +10003795 if (s->failed == s->q_failed) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07003796 /* The only possible failed device holds Q, so it
Dan Williamsa4456852007-07-09 11:56:43 -07003797 * makes sense to check P (If anything else were failed,
3798 * we would have used P to recreate it).
3799 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003800 sh->check_state = check_state_run;
Dan Williamsa4456852007-07-09 11:56:43 -07003801 }
NeilBrownf2b3b442011-07-26 11:35:19 +10003802 if (!s->q_failed && s->failed < 2) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07003803 /* Q is not failed, and we didn't use it to generate
Dan Williamsa4456852007-07-09 11:56:43 -07003804 * anything, so it makes sense to check it
3805 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003806 if (sh->check_state == check_state_run)
3807 sh->check_state = check_state_run_pq;
3808 else
3809 sh->check_state = check_state_run_q;
Dan Williamsa4456852007-07-09 11:56:43 -07003810 }
Dan Williams36d1c642009-07-14 11:48:22 -07003811
Dan Williamsd82dfee2009-07-14 13:40:57 -07003812 /* discard potentially stale zero_sum_result */
3813 sh->ops.zero_sum_result = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07003814
Dan Williamsd82dfee2009-07-14 13:40:57 -07003815 if (sh->check_state == check_state_run) {
3816 /* async_xor_zero_sum destroys the contents of P */
3817 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
3818 s->uptodate--;
Dan Williamsa4456852007-07-09 11:56:43 -07003819 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003820 if (sh->check_state >= check_state_run &&
3821 sh->check_state <= check_state_run_pq) {
3822 /* async_syndrome_zero_sum preserves P and Q, so
3823 * no need to mark them !uptodate here
3824 */
3825 set_bit(STRIPE_OP_CHECK, &s->ops_request);
3826 break;
3827 }
Dan Williams36d1c642009-07-14 11:48:22 -07003828
Dan Williamsd82dfee2009-07-14 13:40:57 -07003829 /* we have 2-disk failure */
3830 BUG_ON(s->failed != 2);
3831 /* fall through */
3832 case check_state_compute_result:
3833 sh->check_state = check_state_idle;
Dan Williams36d1c642009-07-14 11:48:22 -07003834
Dan Williamsd82dfee2009-07-14 13:40:57 -07003835 /* check that a write has not made the stripe insync */
3836 if (test_bit(STRIPE_INSYNC, &sh->state))
3837 break;
Dan Williamsa4456852007-07-09 11:56:43 -07003838
3839 /* now write out any block on a failed drive,
Dan Williamsd82dfee2009-07-14 13:40:57 -07003840 * or P or Q if they were recomputed
Dan Williamsa4456852007-07-09 11:56:43 -07003841 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07003842 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
Dan Williamsa4456852007-07-09 11:56:43 -07003843 if (s->failed == 2) {
NeilBrownf2b3b442011-07-26 11:35:19 +10003844 dev = &sh->dev[s->failed_num[1]];
Dan Williamsa4456852007-07-09 11:56:43 -07003845 s->locked++;
3846 set_bit(R5_LOCKED, &dev->flags);
3847 set_bit(R5_Wantwrite, &dev->flags);
3848 }
3849 if (s->failed >= 1) {
NeilBrownf2b3b442011-07-26 11:35:19 +10003850 dev = &sh->dev[s->failed_num[0]];
Dan Williamsa4456852007-07-09 11:56:43 -07003851 s->locked++;
3852 set_bit(R5_LOCKED, &dev->flags);
3853 set_bit(R5_Wantwrite, &dev->flags);
3854 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003855 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07003856 dev = &sh->dev[pd_idx];
3857 s->locked++;
3858 set_bit(R5_LOCKED, &dev->flags);
3859 set_bit(R5_Wantwrite, &dev->flags);
3860 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07003861 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07003862 dev = &sh->dev[qd_idx];
3863 s->locked++;
3864 set_bit(R5_LOCKED, &dev->flags);
3865 set_bit(R5_Wantwrite, &dev->flags);
3866 }
3867 clear_bit(STRIPE_DEGRADED, &sh->state);
3868
3869 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003870 break;
3871 case check_state_run:
3872 case check_state_run_q:
3873 case check_state_run_pq:
3874 break; /* we will be called again upon completion */
3875 case check_state_check_result:
3876 sh->check_state = check_state_idle;
3877
3878 /* handle a successful check operation, if parity is correct
3879 * we are done. Otherwise update the mismatch count and repair
3880 * parity if !MD_RECOVERY_CHECK
3881 */
3882 if (sh->ops.zero_sum_result == 0) {
3883 /* both parities are correct */
3884 if (!s->failed)
3885 set_bit(STRIPE_INSYNC, &sh->state);
3886 else {
3887 /* in contrast to the raid5 case we can validate
3888 * parity, but still have a failure to write
3889 * back
3890 */
3891 sh->check_state = check_state_compute_result;
3892 /* Returning at this point means that we may go
3893 * off and bring p and/or q uptodate again so
3894 * we make sure to check zero_sum_result again
3895 * to verify if p or q need writeback
3896 */
3897 }
3898 } else {
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11003899 atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches);
Dan Williamsd82dfee2009-07-14 13:40:57 -07003900 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
3901 /* don't try to repair!! */
3902 set_bit(STRIPE_INSYNC, &sh->state);
3903 else {
3904 int *target = &sh->ops.target;
3905
3906 sh->ops.target = -1;
3907 sh->ops.target2 = -1;
3908 sh->check_state = check_state_compute_run;
3909 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
3910 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
3911 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
3912 set_bit(R5_Wantcompute,
3913 &sh->dev[pd_idx].flags);
3914 *target = pd_idx;
3915 target = &sh->ops.target2;
3916 s->uptodate++;
3917 }
3918 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
3919 set_bit(R5_Wantcompute,
3920 &sh->dev[qd_idx].flags);
3921 *target = qd_idx;
3922 s->uptodate++;
3923 }
3924 }
3925 }
3926 break;
3927 case check_state_compute_run:
3928 break;
3929 default:
3930 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
3931 __func__, sh->check_state,
3932 (unsigned long long) sh->sector);
3933 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07003934 }
3935}
3936
NeilBrownd1688a62011-10-11 16:49:52 +11003937static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh)
Dan Williamsa4456852007-07-09 11:56:43 -07003938{
3939 int i;
3940
3941 /* We have read all the blocks in this stripe and now we need to
3942 * copy some of them into a target stripe for expand.
3943 */
Dan Williamsf0a50d32007-01-02 13:52:31 -07003944 struct dma_async_tx_descriptor *tx = NULL;
shli@kernel.org59fc6302014-12-15 12:57:03 +11003945 BUG_ON(sh->batch_head);
Dan Williamsa4456852007-07-09 11:56:43 -07003946 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3947 for (i = 0; i < sh->disks; i++)
NeilBrown34e04e82009-03-31 15:10:16 +11003948 if (i != sh->pd_idx && i != sh->qd_idx) {
NeilBrown911d4ee2009-03-31 14:39:38 +11003949 int dd_idx, j;
Dan Williamsa4456852007-07-09 11:56:43 -07003950 struct stripe_head *sh2;
Dan Williamsa08abd82009-06-03 11:43:59 -07003951 struct async_submit_ctl submit;
Dan Williamsa4456852007-07-09 11:56:43 -07003952
Shaohua Li6d036f72015-08-13 14:31:57 -07003953 sector_t bn = raid5_compute_blocknr(sh, i, 1);
NeilBrown911d4ee2009-03-31 14:39:38 +11003954 sector_t s = raid5_compute_sector(conf, bn, 0,
3955 &dd_idx, NULL);
Shaohua Li6d036f72015-08-13 14:31:57 -07003956 sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1);
Dan Williamsa4456852007-07-09 11:56:43 -07003957 if (sh2 == NULL)
3958 /* so far only the early blocks of this stripe
3959 * have been requested. When later blocks
3960 * get requested, we will try again
3961 */
3962 continue;
3963 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
3964 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
3965 /* must have already done this block */
Shaohua Li6d036f72015-08-13 14:31:57 -07003966 raid5_release_stripe(sh2);
Dan Williamsa4456852007-07-09 11:56:43 -07003967 continue;
3968 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07003969
3970 /* place all the copies on one channel */
Dan Williamsa08abd82009-06-03 11:43:59 -07003971 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003972 tx = async_memcpy(sh2->dev[dd_idx].page,
Dan Williams88ba2aa2009-04-09 16:16:18 -07003973 sh->dev[i].page, 0, 0, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07003974 &submit);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003975
Dan Williamsa4456852007-07-09 11:56:43 -07003976 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
3977 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
3978 for (j = 0; j < conf->raid_disks; j++)
3979 if (j != sh2->pd_idx &&
NeilBrown86c374b2011-07-27 11:00:36 +10003980 j != sh2->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07003981 !test_bit(R5_Expanded, &sh2->dev[j].flags))
3982 break;
3983 if (j == conf->raid_disks) {
3984 set_bit(STRIPE_EXPAND_READY, &sh2->state);
3985 set_bit(STRIPE_HANDLE, &sh2->state);
3986 }
Shaohua Li6d036f72015-08-13 14:31:57 -07003987 raid5_release_stripe(sh2);
Dan Williamsf0a50d32007-01-02 13:52:31 -07003988
Dan Williamsa4456852007-07-09 11:56:43 -07003989 }
NeilBrowna2e08552007-09-11 15:23:36 -07003990 /* done submitting copies, wait for them to complete */
NeilBrown749586b2012-11-20 14:11:15 +11003991 async_tx_quiesce(&tx);
Dan Williamsa4456852007-07-09 11:56:43 -07003992}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993
3994/*
3995 * handle_stripe - do things to a stripe.
3996 *
NeilBrown9a3e1102011-12-23 10:17:53 +11003997 * We lock the stripe by setting STRIPE_ACTIVE and then examine the
3998 * state of various bits to see what needs to be done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 * Possible results:
NeilBrown9a3e1102011-12-23 10:17:53 +11004000 * return some read requests which now have data
4001 * return some write requests which are safely on storage
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 * schedule a read on some buffers
4003 * schedule a write of some buffers
4004 * return confirmation of parity correctness
4005 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 */
Dan Williamsa4456852007-07-09 11:56:43 -07004007
NeilBrownacfe7262011-07-27 11:00:36 +10004008static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s)
NeilBrown16a53ec2006-06-26 00:27:38 -07004009{
NeilBrownd1688a62011-10-11 16:49:52 +11004010 struct r5conf *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08004011 int disks = sh->disks;
NeilBrown474af965fe2011-07-27 11:00:36 +10004012 struct r5dev *dev;
4013 int i;
NeilBrown9a3e1102011-12-23 10:17:53 +11004014 int do_recovery = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07004015
NeilBrownacfe7262011-07-27 11:00:36 +10004016 memset(s, 0, sizeof(*s));
NeilBrown16a53ec2006-06-26 00:27:38 -07004017
shli@kernel.orgdabc4ec2014-12-15 12:57:04 +11004018 s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head;
4019 s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head;
NeilBrownacfe7262011-07-27 11:00:36 +10004020 s->failed_num[0] = -1;
4021 s->failed_num[1] = -1;
Shaohua Li6e74a9c2015-10-08 21:54:08 -07004022 s->log_failed = r5l_log_disk_error(conf);
NeilBrownacfe7262011-07-27 11:00:36 +10004023
4024 /* Now to look around and see what can be done */
NeilBrown16a53ec2006-06-26 00:27:38 -07004025 rcu_read_lock();
4026 for (i=disks; i--; ) {
NeilBrown3cb03002011-10-11 16:45:26 +11004027 struct md_rdev *rdev;
NeilBrown31c176e2011-07-28 11:39:22 +10004028 sector_t first_bad;
4029 int bad_sectors;
4030 int is_bad = 0;
NeilBrownacfe7262011-07-27 11:00:36 +10004031
NeilBrown16a53ec2006-06-26 00:27:38 -07004032 dev = &sh->dev[i];
NeilBrown16a53ec2006-06-26 00:27:38 -07004033
Dan Williams45b42332007-07-09 11:56:43 -07004034 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown9a3e1102011-12-23 10:17:53 +11004035 i, dev->flags,
4036 dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07004037 /* maybe we can reply to a read
4038 *
4039 * new wantfill requests are only permitted while
4040 * ops_complete_biofill is guaranteed to be inactive
4041 */
4042 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
4043 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
4044 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07004045
4046 /* now count some things */
NeilBrowncc940152011-07-26 11:35:35 +10004047 if (test_bit(R5_LOCKED, &dev->flags))
4048 s->locked++;
4049 if (test_bit(R5_UPTODATE, &dev->flags))
4050 s->uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07004051 if (test_bit(R5_Wantcompute, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10004052 s->compute++;
4053 BUG_ON(s->compute > 2);
Dan Williams2d6e4ec2009-09-16 12:11:54 -07004054 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004055
NeilBrownacfe7262011-07-27 11:00:36 +10004056 if (test_bit(R5_Wantfill, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10004057 s->to_fill++;
NeilBrownacfe7262011-07-27 11:00:36 +10004058 else if (dev->toread)
NeilBrowncc940152011-07-26 11:35:35 +10004059 s->to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07004060 if (dev->towrite) {
NeilBrowncc940152011-07-26 11:35:35 +10004061 s->to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07004062 if (!test_bit(R5_OVERWRITE, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10004063 s->non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07004064 }
Dan Williamsa4456852007-07-09 11:56:43 -07004065 if (dev->written)
NeilBrowncc940152011-07-26 11:35:35 +10004066 s->written++;
NeilBrown14a75d32011-12-23 10:17:52 +11004067 /* Prefer to use the replacement for reads, but only
4068 * if it is recovered enough and has no bad blocks.
4069 */
4070 rdev = rcu_dereference(conf->disks[i].replacement);
4071 if (rdev && !test_bit(Faulty, &rdev->flags) &&
4072 rdev->recovery_offset >= sh->sector + STRIPE_SECTORS &&
4073 !is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4074 &first_bad, &bad_sectors))
4075 set_bit(R5_ReadRepl, &dev->flags);
4076 else {
NeilBrowne6030cb2015-07-17 13:26:23 +10004077 if (rdev && !test_bit(Faulty, &rdev->flags))
NeilBrown9a3e1102011-12-23 10:17:53 +11004078 set_bit(R5_NeedReplace, &dev->flags);
NeilBrowne6030cb2015-07-17 13:26:23 +10004079 else
4080 clear_bit(R5_NeedReplace, &dev->flags);
NeilBrown14a75d32011-12-23 10:17:52 +11004081 rdev = rcu_dereference(conf->disks[i].rdev);
4082 clear_bit(R5_ReadRepl, &dev->flags);
4083 }
NeilBrown9283d8c2011-12-08 16:27:57 +11004084 if (rdev && test_bit(Faulty, &rdev->flags))
4085 rdev = NULL;
NeilBrown31c176e2011-07-28 11:39:22 +10004086 if (rdev) {
4087 is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS,
4088 &first_bad, &bad_sectors);
4089 if (s->blocked_rdev == NULL
4090 && (test_bit(Blocked, &rdev->flags)
4091 || is_bad < 0)) {
4092 if (is_bad < 0)
4093 set_bit(BlockedBadBlocks,
4094 &rdev->flags);
4095 s->blocked_rdev = rdev;
4096 atomic_inc(&rdev->nr_pending);
4097 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07004098 }
NeilBrown415e72d2010-06-17 17:25:21 +10004099 clear_bit(R5_Insync, &dev->flags);
4100 if (!rdev)
4101 /* Not in-sync */;
NeilBrown31c176e2011-07-28 11:39:22 +10004102 else if (is_bad) {
4103 /* also not in-sync */
NeilBrown18b98372012-04-01 23:48:38 +10004104 if (!test_bit(WriteErrorSeen, &rdev->flags) &&
4105 test_bit(R5_UPTODATE, &dev->flags)) {
NeilBrown31c176e2011-07-28 11:39:22 +10004106 /* treat as in-sync, but with a read error
4107 * which we can now try to correct
4108 */
4109 set_bit(R5_Insync, &dev->flags);
4110 set_bit(R5_ReadError, &dev->flags);
4111 }
4112 } else if (test_bit(In_sync, &rdev->flags))
NeilBrown415e72d2010-06-17 17:25:21 +10004113 set_bit(R5_Insync, &dev->flags);
NeilBrown30d7a482011-12-23 09:57:00 +11004114 else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
NeilBrown415e72d2010-06-17 17:25:21 +10004115 /* in sync if before recovery_offset */
NeilBrown30d7a482011-12-23 09:57:00 +11004116 set_bit(R5_Insync, &dev->flags);
4117 else if (test_bit(R5_UPTODATE, &dev->flags) &&
4118 test_bit(R5_Expanded, &dev->flags))
4119 /* If we've reshaped into here, we assume it is Insync.
4120 * We will shortly update recovery_offset to make
4121 * it official.
4122 */
4123 set_bit(R5_Insync, &dev->flags);
4124
NeilBrown1cc03eb2014-01-06 13:19:42 +11004125 if (test_bit(R5_WriteError, &dev->flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11004126 /* This flag does not apply to '.replacement'
4127 * only to .rdev, so make sure to check that*/
4128 struct md_rdev *rdev2 = rcu_dereference(
4129 conf->disks[i].rdev);
4130 if (rdev2 == rdev)
4131 clear_bit(R5_Insync, &dev->flags);
4132 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
NeilBrownbc2607f2011-07-28 11:39:22 +10004133 s->handle_bad_blocks = 1;
NeilBrown14a75d32011-12-23 10:17:52 +11004134 atomic_inc(&rdev2->nr_pending);
NeilBrownbc2607f2011-07-28 11:39:22 +10004135 } else
4136 clear_bit(R5_WriteError, &dev->flags);
4137 }
NeilBrown1cc03eb2014-01-06 13:19:42 +11004138 if (test_bit(R5_MadeGood, &dev->flags)) {
NeilBrown14a75d32011-12-23 10:17:52 +11004139 /* This flag does not apply to '.replacement'
4140 * only to .rdev, so make sure to check that*/
4141 struct md_rdev *rdev2 = rcu_dereference(
4142 conf->disks[i].rdev);
4143 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
NeilBrownb84db562011-07-28 11:39:23 +10004144 s->handle_bad_blocks = 1;
NeilBrown14a75d32011-12-23 10:17:52 +11004145 atomic_inc(&rdev2->nr_pending);
NeilBrownb84db562011-07-28 11:39:23 +10004146 } else
4147 clear_bit(R5_MadeGood, &dev->flags);
4148 }
NeilBrown977df362011-12-23 10:17:53 +11004149 if (test_bit(R5_MadeGoodRepl, &dev->flags)) {
4150 struct md_rdev *rdev2 = rcu_dereference(
4151 conf->disks[i].replacement);
4152 if (rdev2 && !test_bit(Faulty, &rdev2->flags)) {
4153 s->handle_bad_blocks = 1;
4154 atomic_inc(&rdev2->nr_pending);
4155 } else
4156 clear_bit(R5_MadeGoodRepl, &dev->flags);
4157 }
NeilBrown415e72d2010-06-17 17:25:21 +10004158 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07004159 /* The ReadError flag will just be confusing now */
4160 clear_bit(R5_ReadError, &dev->flags);
4161 clear_bit(R5_ReWrite, &dev->flags);
4162 }
NeilBrown415e72d2010-06-17 17:25:21 +10004163 if (test_bit(R5_ReadError, &dev->flags))
4164 clear_bit(R5_Insync, &dev->flags);
4165 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10004166 if (s->failed < 2)
4167 s->failed_num[s->failed] = i;
4168 s->failed++;
NeilBrown9a3e1102011-12-23 10:17:53 +11004169 if (rdev && !test_bit(Faulty, &rdev->flags))
4170 do_recovery = 1;
NeilBrown415e72d2010-06-17 17:25:21 +10004171 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004172 }
NeilBrown9a3e1102011-12-23 10:17:53 +11004173 if (test_bit(STRIPE_SYNCING, &sh->state)) {
4174 /* If there is a failed device being replaced,
4175 * we must be recovering.
4176 * else if we are after recovery_cp, we must be syncing
majianpengc6d2e082012-04-02 01:16:59 +10004177 * else if MD_RECOVERY_REQUESTED is set, we also are syncing.
NeilBrown9a3e1102011-12-23 10:17:53 +11004178 * else we can only be replacing
4179 * sync and recovery both need to read all devices, and so
4180 * use the same flag.
4181 */
4182 if (do_recovery ||
majianpengc6d2e082012-04-02 01:16:59 +10004183 sh->sector >= conf->mddev->recovery_cp ||
4184 test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery)))
NeilBrown9a3e1102011-12-23 10:17:53 +11004185 s->syncing = 1;
4186 else
4187 s->replacing = 1;
4188 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004189 rcu_read_unlock();
NeilBrowncc940152011-07-26 11:35:35 +10004190}
NeilBrownf4168852007-02-28 20:11:53 -08004191
shli@kernel.org59fc6302014-12-15 12:57:03 +11004192static int clear_batch_ready(struct stripe_head *sh)
4193{
NeilBrownb15a9db2015-05-22 15:20:04 +10004194 /* Return '1' if this is a member of batch, or
4195 * '0' if it is a lone stripe or a head which can now be
4196 * handled.
4197 */
shli@kernel.org59fc6302014-12-15 12:57:03 +11004198 struct stripe_head *tmp;
4199 if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state))
NeilBrownb15a9db2015-05-22 15:20:04 +10004200 return (sh->batch_head && sh->batch_head != sh);
shli@kernel.org59fc6302014-12-15 12:57:03 +11004201 spin_lock(&sh->stripe_lock);
4202 if (!sh->batch_head) {
4203 spin_unlock(&sh->stripe_lock);
4204 return 0;
4205 }
4206
4207 /*
4208 * this stripe could be added to a batch list before we check
4209 * BATCH_READY, skips it
4210 */
4211 if (sh->batch_head != sh) {
4212 spin_unlock(&sh->stripe_lock);
4213 return 1;
4214 }
4215 spin_lock(&sh->batch_lock);
4216 list_for_each_entry(tmp, &sh->batch_list, batch_list)
4217 clear_bit(STRIPE_BATCH_READY, &tmp->state);
4218 spin_unlock(&sh->batch_lock);
4219 spin_unlock(&sh->stripe_lock);
4220
4221 /*
4222 * BATCH_READY is cleared, no new stripes can be added.
4223 * batch_list can be accessed without lock
4224 */
4225 return 0;
4226}
4227
NeilBrown3960ce72015-05-21 12:20:36 +10004228static void break_stripe_batch_list(struct stripe_head *head_sh,
4229 unsigned long handle_flags)
shli@kernel.org72ac7332014-12-15 12:57:03 +11004230{
NeilBrown4e3d62f2015-05-21 11:50:16 +10004231 struct stripe_head *sh, *next;
shli@kernel.org72ac7332014-12-15 12:57:03 +11004232 int i;
NeilBrownfb642b92015-05-21 12:00:47 +10004233 int do_wakeup = 0;
shli@kernel.org72ac7332014-12-15 12:57:03 +11004234
NeilBrownbb270512015-05-08 18:19:40 +10004235 list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) {
4236
shli@kernel.org72ac7332014-12-15 12:57:03 +11004237 list_del_init(&sh->batch_list);
4238
Shaohua Lifb3229d2016-03-09 10:08:38 -08004239 WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) |
NeilBrown1b956f72015-05-21 12:40:26 +10004240 (1 << STRIPE_SYNCING) |
4241 (1 << STRIPE_REPLACED) |
NeilBrown1b956f72015-05-21 12:40:26 +10004242 (1 << STRIPE_DELAYED) |
4243 (1 << STRIPE_BIT_DELAY) |
4244 (1 << STRIPE_FULL_WRITE) |
4245 (1 << STRIPE_BIOFILL_RUN) |
4246 (1 << STRIPE_COMPUTE_RUN) |
4247 (1 << STRIPE_OPS_REQ_PENDING) |
4248 (1 << STRIPE_DISCARD) |
4249 (1 << STRIPE_BATCH_READY) |
4250 (1 << STRIPE_BATCH_ERR) |
Shaohua Lifb3229d2016-03-09 10:08:38 -08004251 (1 << STRIPE_BITMAP_PENDING)),
4252 "stripe state: %lx\n", sh->state);
4253 WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) |
4254 (1 << STRIPE_REPLACED)),
4255 "head stripe state: %lx\n", head_sh->state);
NeilBrown1b956f72015-05-21 12:40:26 +10004256
4257 set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS |
NeilBrown550da242016-03-09 12:58:25 +11004258 (1 << STRIPE_PREREAD_ACTIVE) |
NeilBrown1b956f72015-05-21 12:40:26 +10004259 (1 << STRIPE_DEGRADED)),
4260 head_sh->state & (1 << STRIPE_INSYNC));
4261
shli@kernel.org72ac7332014-12-15 12:57:03 +11004262 sh->check_state = head_sh->check_state;
4263 sh->reconstruct_state = head_sh->reconstruct_state;
NeilBrownfb642b92015-05-21 12:00:47 +10004264 for (i = 0; i < sh->disks; i++) {
4265 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
4266 do_wakeup = 1;
shli@kernel.org72ac7332014-12-15 12:57:03 +11004267 sh->dev[i].flags = head_sh->dev[i].flags &
4268 (~((1 << R5_WriteError) | (1 << R5_Overlap)));
NeilBrownfb642b92015-05-21 12:00:47 +10004269 }
shli@kernel.org72ac7332014-12-15 12:57:03 +11004270 spin_lock_irq(&sh->stripe_lock);
4271 sh->batch_head = NULL;
4272 spin_unlock_irq(&sh->stripe_lock);
NeilBrown3960ce72015-05-21 12:20:36 +10004273 if (handle_flags == 0 ||
4274 sh->state & handle_flags)
4275 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07004276 raid5_release_stripe(sh);
shli@kernel.org72ac7332014-12-15 12:57:03 +11004277 }
NeilBrownfb642b92015-05-21 12:00:47 +10004278 spin_lock_irq(&head_sh->stripe_lock);
4279 head_sh->batch_head = NULL;
4280 spin_unlock_irq(&head_sh->stripe_lock);
4281 for (i = 0; i < head_sh->disks; i++)
4282 if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags))
4283 do_wakeup = 1;
NeilBrown3960ce72015-05-21 12:20:36 +10004284 if (head_sh->state & handle_flags)
4285 set_bit(STRIPE_HANDLE, &head_sh->state);
NeilBrownfb642b92015-05-21 12:00:47 +10004286
4287 if (do_wakeup)
4288 wake_up(&head_sh->raid_conf->wait_for_overlap);
shli@kernel.org72ac7332014-12-15 12:57:03 +11004289}
4290
NeilBrowncc940152011-07-26 11:35:35 +10004291static void handle_stripe(struct stripe_head *sh)
4292{
4293 struct stripe_head_state s;
NeilBrownd1688a62011-10-11 16:49:52 +11004294 struct r5conf *conf = sh->raid_conf;
NeilBrown3687c062011-07-27 11:00:36 +10004295 int i;
NeilBrown84789552011-07-27 11:00:36 +10004296 int prexor;
4297 int disks = sh->disks;
NeilBrown474af965fe2011-07-27 11:00:36 +10004298 struct r5dev *pdev, *qdev;
NeilBrowncc940152011-07-26 11:35:35 +10004299
4300 clear_bit(STRIPE_HANDLE, &sh->state);
Dan Williams257a4b42011-11-08 16:22:06 +11004301 if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) {
NeilBrowncc940152011-07-26 11:35:35 +10004302 /* already being handled, ensure it gets handled
4303 * again when current action finishes */
4304 set_bit(STRIPE_HANDLE, &sh->state);
4305 return;
4306 }
4307
shli@kernel.org59fc6302014-12-15 12:57:03 +11004308 if (clear_batch_ready(sh) ) {
4309 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
4310 return;
4311 }
4312
NeilBrown4e3d62f2015-05-21 11:50:16 +10004313 if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state))
NeilBrown3960ce72015-05-21 12:20:36 +10004314 break_stripe_batch_list(sh, 0);
shli@kernel.org72ac7332014-12-15 12:57:03 +11004315
shli@kernel.orgdabc4ec2014-12-15 12:57:04 +11004316 if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) {
NeilBrownf8dfcff2013-03-12 12:18:06 +11004317 spin_lock(&sh->stripe_lock);
4318 /* Cannot process 'sync' concurrently with 'discard' */
4319 if (!test_bit(STRIPE_DISCARD, &sh->state) &&
4320 test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
4321 set_bit(STRIPE_SYNCING, &sh->state);
4322 clear_bit(STRIPE_INSYNC, &sh->state);
NeilBrownf94c0b62013-07-22 12:57:21 +10004323 clear_bit(STRIPE_REPLACED, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11004324 }
4325 spin_unlock(&sh->stripe_lock);
NeilBrowncc940152011-07-26 11:35:35 +10004326 }
4327 clear_bit(STRIPE_DELAYED, &sh->state);
4328
4329 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
4330 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
4331 (unsigned long long)sh->sector, sh->state,
4332 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
4333 sh->check_state, sh->reconstruct_state);
NeilBrowncc940152011-07-26 11:35:35 +10004334
NeilBrownacfe7262011-07-27 11:00:36 +10004335 analyse_stripe(sh, &s);
NeilBrownc5a31002011-07-27 11:00:36 +10004336
Shaohua Lib70abcb2015-08-13 14:31:58 -07004337 if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
4338 goto finish;
4339
NeilBrownbc2607f2011-07-28 11:39:22 +10004340 if (s.handle_bad_blocks) {
4341 set_bit(STRIPE_HANDLE, &sh->state);
4342 goto finish;
4343 }
4344
NeilBrown474af965fe2011-07-27 11:00:36 +10004345 if (unlikely(s.blocked_rdev)) {
4346 if (s.syncing || s.expanding || s.expanded ||
NeilBrown9a3e1102011-12-23 10:17:53 +11004347 s.replacing || s.to_write || s.written) {
NeilBrown474af965fe2011-07-27 11:00:36 +10004348 set_bit(STRIPE_HANDLE, &sh->state);
4349 goto finish;
4350 }
4351 /* There is nothing for the blocked_rdev to block */
4352 rdev_dec_pending(s.blocked_rdev, conf->mddev);
4353 s.blocked_rdev = NULL;
4354 }
4355
4356 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
4357 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
4358 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
4359 }
4360
4361 pr_debug("locked=%d uptodate=%d to_read=%d"
4362 " to_write=%d failed=%d failed_num=%d,%d\n",
4363 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
4364 s.failed_num[0], s.failed_num[1]);
4365 /* check if the array has lost more than max_degraded devices and,
4366 * if so, some requests might need to be failed.
4367 */
Shaohua Li6e74a9c2015-10-08 21:54:08 -07004368 if (s.failed > conf->max_degraded || s.log_failed) {
NeilBrown9a3f5302011-11-08 16:22:01 +11004369 sh->check_state = 0;
4370 sh->reconstruct_state = 0;
NeilBrown626f2092015-05-22 14:03:10 +10004371 break_stripe_batch_list(sh, 0);
NeilBrown9a3f5302011-11-08 16:22:01 +11004372 if (s.to_read+s.to_write+s.written)
4373 handle_failed_stripe(conf, sh, &s, disks, &s.return_bi);
NeilBrown9a3e1102011-12-23 10:17:53 +11004374 if (s.syncing + s.replacing)
NeilBrown9a3f5302011-11-08 16:22:01 +11004375 handle_failed_sync(conf, sh, &s);
4376 }
NeilBrown474af965fe2011-07-27 11:00:36 +10004377
NeilBrown84789552011-07-27 11:00:36 +10004378 /* Now we check to see if any write operations have recently
4379 * completed
4380 */
4381 prexor = 0;
4382 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
4383 prexor = 1;
4384 if (sh->reconstruct_state == reconstruct_state_drain_result ||
4385 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
4386 sh->reconstruct_state = reconstruct_state_idle;
4387
4388 /* All the 'written' buffers and the parity block are ready to
4389 * be written back to disk
4390 */
Shaohua Li9e4447682012-10-11 13:49:49 +11004391 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) &&
4392 !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags));
NeilBrown84789552011-07-27 11:00:36 +10004393 BUG_ON(sh->qd_idx >= 0 &&
Shaohua Li9e4447682012-10-11 13:49:49 +11004394 !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) &&
4395 !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags));
NeilBrown84789552011-07-27 11:00:36 +10004396 for (i = disks; i--; ) {
4397 struct r5dev *dev = &sh->dev[i];
4398 if (test_bit(R5_LOCKED, &dev->flags) &&
4399 (i == sh->pd_idx || i == sh->qd_idx ||
4400 dev->written)) {
4401 pr_debug("Writing block %d\n", i);
4402 set_bit(R5_Wantwrite, &dev->flags);
4403 if (prexor)
4404 continue;
NeilBrown9c4bdf62014-08-13 09:57:07 +10004405 if (s.failed > 1)
4406 continue;
NeilBrown84789552011-07-27 11:00:36 +10004407 if (!test_bit(R5_Insync, &dev->flags) ||
4408 ((i == sh->pd_idx || i == sh->qd_idx) &&
4409 s.failed == 0))
4410 set_bit(STRIPE_INSYNC, &sh->state);
4411 }
4412 }
4413 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4414 s.dec_preread_active = 1;
4415 }
4416
NeilBrownef5b7c62012-11-22 09:13:36 +11004417 /*
4418 * might be able to return some write requests if the parity blocks
4419 * are safe, or on a failed drive
4420 */
4421 pdev = &sh->dev[sh->pd_idx];
4422 s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx)
4423 || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx);
4424 qdev = &sh->dev[sh->qd_idx];
4425 s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx)
4426 || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx)
4427 || conf->level < 6;
4428
4429 if (s.written &&
4430 (s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
4431 && !test_bit(R5_LOCKED, &pdev->flags)
4432 && (test_bit(R5_UPTODATE, &pdev->flags) ||
4433 test_bit(R5_Discard, &pdev->flags))))) &&
4434 (s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
4435 && !test_bit(R5_LOCKED, &qdev->flags)
4436 && (test_bit(R5_UPTODATE, &qdev->flags) ||
4437 test_bit(R5_Discard, &qdev->flags))))))
4438 handle_stripe_clean_event(conf, sh, disks, &s.return_bi);
4439
4440 /* Now we might consider reading some blocks, either to check/generate
4441 * parity, or to satisfy requests
4442 * or to load a block that is being partially written.
4443 */
4444 if (s.to_read || s.non_overwrite
4445 || (conf->level == 6 && s.to_write && s.failed)
4446 || (s.syncing && (s.uptodate + s.compute < disks))
4447 || s.replacing
4448 || s.expanding)
4449 handle_stripe_fill(sh, &s, disks);
4450
NeilBrown84789552011-07-27 11:00:36 +10004451 /* Now to consider new write requests and what else, if anything
4452 * should be read. We do not handle new writes when:
4453 * 1/ A 'write' operation (copy+xor) is already in flight.
4454 * 2/ A 'check' operation is in flight, as it may clobber the parity
4455 * block.
4456 */
4457 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
4458 handle_stripe_dirtying(conf, sh, &s, disks);
4459
4460 /* maybe we need to check and possibly fix the parity for this stripe
4461 * Any reads will already have been scheduled, so we just see if enough
4462 * data is available. The parity check is held off while parity
4463 * dependent operations are in flight.
4464 */
4465 if (sh->check_state ||
4466 (s.syncing && s.locked == 0 &&
4467 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
4468 !test_bit(STRIPE_INSYNC, &sh->state))) {
4469 if (conf->level == 6)
4470 handle_parity_checks6(conf, sh, &s, disks);
4471 else
4472 handle_parity_checks5(conf, sh, &s, disks);
4473 }
NeilBrownc5a31002011-07-27 11:00:36 +10004474
NeilBrownf94c0b62013-07-22 12:57:21 +10004475 if ((s.replacing || s.syncing) && s.locked == 0
4476 && !test_bit(STRIPE_COMPUTE_RUN, &sh->state)
4477 && !test_bit(STRIPE_REPLACED, &sh->state)) {
NeilBrown9a3e1102011-12-23 10:17:53 +11004478 /* Write out to replacement devices where possible */
4479 for (i = 0; i < conf->raid_disks; i++)
NeilBrownf94c0b62013-07-22 12:57:21 +10004480 if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) {
4481 WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags));
NeilBrown9a3e1102011-12-23 10:17:53 +11004482 set_bit(R5_WantReplace, &sh->dev[i].flags);
4483 set_bit(R5_LOCKED, &sh->dev[i].flags);
4484 s.locked++;
4485 }
NeilBrownf94c0b62013-07-22 12:57:21 +10004486 if (s.replacing)
4487 set_bit(STRIPE_INSYNC, &sh->state);
4488 set_bit(STRIPE_REPLACED, &sh->state);
NeilBrown9a3e1102011-12-23 10:17:53 +11004489 }
4490 if ((s.syncing || s.replacing) && s.locked == 0 &&
NeilBrownf94c0b62013-07-22 12:57:21 +10004491 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
NeilBrown9a3e1102011-12-23 10:17:53 +11004492 test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrownc5a31002011-07-27 11:00:36 +10004493 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4494 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrownf8dfcff2013-03-12 12:18:06 +11004495 if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags))
4496 wake_up(&conf->wait_for_overlap);
NeilBrownc5a31002011-07-27 11:00:36 +10004497 }
4498
4499 /* If the failed drives are just a ReadError, then we might need
4500 * to progress the repair/check process
4501 */
4502 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
4503 for (i = 0; i < s.failed; i++) {
4504 struct r5dev *dev = &sh->dev[s.failed_num[i]];
4505 if (test_bit(R5_ReadError, &dev->flags)
4506 && !test_bit(R5_LOCKED, &dev->flags)
4507 && test_bit(R5_UPTODATE, &dev->flags)
4508 ) {
4509 if (!test_bit(R5_ReWrite, &dev->flags)) {
4510 set_bit(R5_Wantwrite, &dev->flags);
4511 set_bit(R5_ReWrite, &dev->flags);
4512 set_bit(R5_LOCKED, &dev->flags);
4513 s.locked++;
4514 } else {
4515 /* let's read it back */
4516 set_bit(R5_Wantread, &dev->flags);
4517 set_bit(R5_LOCKED, &dev->flags);
4518 s.locked++;
4519 }
4520 }
4521 }
4522
NeilBrown3687c062011-07-27 11:00:36 +10004523 /* Finish reconstruct operations initiated by the expansion process */
4524 if (sh->reconstruct_state == reconstruct_state_result) {
4525 struct stripe_head *sh_src
Shaohua Li6d036f72015-08-13 14:31:57 -07004526 = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrown3687c062011-07-27 11:00:36 +10004527 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
4528 /* sh cannot be written until sh_src has been read.
4529 * so arrange for sh to be delayed a little
4530 */
4531 set_bit(STRIPE_DELAYED, &sh->state);
4532 set_bit(STRIPE_HANDLE, &sh->state);
4533 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
4534 &sh_src->state))
4535 atomic_inc(&conf->preread_active_stripes);
Shaohua Li6d036f72015-08-13 14:31:57 -07004536 raid5_release_stripe(sh_src);
NeilBrown3687c062011-07-27 11:00:36 +10004537 goto finish;
4538 }
4539 if (sh_src)
Shaohua Li6d036f72015-08-13 14:31:57 -07004540 raid5_release_stripe(sh_src);
NeilBrown16a53ec2006-06-26 00:27:38 -07004541
NeilBrown3687c062011-07-27 11:00:36 +10004542 sh->reconstruct_state = reconstruct_state_idle;
4543 clear_bit(STRIPE_EXPANDING, &sh->state);
4544 for (i = conf->raid_disks; i--; ) {
4545 set_bit(R5_Wantwrite, &sh->dev[i].flags);
4546 set_bit(R5_LOCKED, &sh->dev[i].flags);
4547 s.locked++;
4548 }
4549 }
4550
4551 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
4552 !sh->reconstruct_state) {
4553 /* Need to write out all blocks after computing parity */
4554 sh->disks = conf->raid_disks;
4555 stripe_set_idx(sh->sector, conf, 0, sh);
4556 schedule_reconstruction(sh, &s, 1, 1);
4557 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
4558 clear_bit(STRIPE_EXPAND_READY, &sh->state);
4559 atomic_dec(&conf->reshape_stripes);
4560 wake_up(&conf->wait_for_overlap);
4561 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
4562 }
4563
4564 if (s.expanding && s.locked == 0 &&
4565 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
4566 handle_stripe_expansion(conf, sh);
4567
4568finish:
Dan Williams6bfe0b42008-04-30 00:52:32 -07004569 /* wait for this device to become unblocked */
NeilBrown5f066c62012-07-03 12:13:29 +10004570 if (unlikely(s.blocked_rdev)) {
4571 if (conf->mddev->external)
4572 md_wait_for_blocked_rdev(s.blocked_rdev,
4573 conf->mddev);
4574 else
4575 /* Internal metadata will immediately
4576 * be written by raid5d, so we don't
4577 * need to wait here.
4578 */
4579 rdev_dec_pending(s.blocked_rdev,
4580 conf->mddev);
4581 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07004582
NeilBrownbc2607f2011-07-28 11:39:22 +10004583 if (s.handle_bad_blocks)
4584 for (i = disks; i--; ) {
NeilBrown3cb03002011-10-11 16:45:26 +11004585 struct md_rdev *rdev;
NeilBrownbc2607f2011-07-28 11:39:22 +10004586 struct r5dev *dev = &sh->dev[i];
4587 if (test_and_clear_bit(R5_WriteError, &dev->flags)) {
4588 /* We own a safe reference to the rdev */
4589 rdev = conf->disks[i].rdev;
4590 if (!rdev_set_badblocks(rdev, sh->sector,
4591 STRIPE_SECTORS, 0))
4592 md_error(conf->mddev, rdev);
4593 rdev_dec_pending(rdev, conf->mddev);
4594 }
NeilBrownb84db562011-07-28 11:39:23 +10004595 if (test_and_clear_bit(R5_MadeGood, &dev->flags)) {
4596 rdev = conf->disks[i].rdev;
4597 rdev_clear_badblocks(rdev, sh->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10004598 STRIPE_SECTORS, 0);
NeilBrownb84db562011-07-28 11:39:23 +10004599 rdev_dec_pending(rdev, conf->mddev);
4600 }
NeilBrown977df362011-12-23 10:17:53 +11004601 if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) {
4602 rdev = conf->disks[i].replacement;
NeilBrowndd054fc2011-12-23 10:17:53 +11004603 if (!rdev)
4604 /* rdev have been moved down */
4605 rdev = conf->disks[i].rdev;
NeilBrown977df362011-12-23 10:17:53 +11004606 rdev_clear_badblocks(rdev, sh->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10004607 STRIPE_SECTORS, 0);
NeilBrown977df362011-12-23 10:17:53 +11004608 rdev_dec_pending(rdev, conf->mddev);
4609 }
NeilBrownbc2607f2011-07-28 11:39:22 +10004610 }
4611
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07004612 if (s.ops_request)
4613 raid_run_ops(sh, s.ops_request);
4614
Dan Williamsf0e43bc2008-06-28 08:31:55 +10004615 ops_run_io(sh, &s);
4616
NeilBrownc5709ef2011-07-26 11:35:20 +10004617 if (s.dec_preread_active) {
NeilBrown729a1862009-12-14 12:49:50 +11004618 /* We delay this until after ops_run_io so that if make_request
Tejun Heoe9c74692010-09-03 11:56:18 +02004619 * is waiting on a flush, it won't continue until the writes
NeilBrown729a1862009-12-14 12:49:50 +11004620 * have actually been submitted.
4621 */
4622 atomic_dec(&conf->preread_active_stripes);
4623 if (atomic_read(&conf->preread_active_stripes) <
4624 IO_THRESHOLD)
4625 md_wakeup_thread(conf->mddev->thread);
4626 }
4627
NeilBrownc3cce6c2015-08-14 12:47:33 +10004628 if (!bio_list_empty(&s.return_bi)) {
4629 if (test_bit(MD_CHANGE_PENDING, &conf->mddev->flags)) {
4630 spin_lock_irq(&conf->device_lock);
4631 bio_list_merge(&conf->return_bi, &s.return_bi);
4632 spin_unlock_irq(&conf->device_lock);
4633 md_wakeup_thread(conf->mddev->thread);
4634 } else
4635 return_io(&s.return_bi);
4636 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004637
Dan Williams257a4b42011-11-08 16:22:06 +11004638 clear_bit_unlock(STRIPE_ACTIVE, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07004639}
4640
NeilBrownd1688a62011-10-11 16:49:52 +11004641static void raid5_activate_delayed(struct r5conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004642{
4643 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
4644 while (!list_empty(&conf->delayed_list)) {
4645 struct list_head *l = conf->delayed_list.next;
4646 struct stripe_head *sh;
4647 sh = list_entry(l, struct stripe_head, lru);
4648 list_del_init(l);
4649 clear_bit(STRIPE_DELAYED, &sh->state);
4650 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4651 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004652 list_add_tail(&sh->lru, &conf->hold_list);
Shaohua Li851c30c2013-08-28 14:30:16 +08004653 raid5_wakeup_stripe_thread(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004654 }
NeilBrown482c0832011-04-18 18:25:42 +10004655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004656}
4657
Shaohua Li566c09c2013-11-14 15:16:17 +11004658static void activate_bit_delay(struct r5conf *conf,
4659 struct list_head *temp_inactive_list)
NeilBrown72626682005-09-09 16:23:54 -07004660{
4661 /* device_lock is held */
4662 struct list_head head;
4663 list_add(&head, &conf->bitmap_list);
4664 list_del_init(&conf->bitmap_list);
4665 while (!list_empty(&head)) {
4666 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
Shaohua Li566c09c2013-11-14 15:16:17 +11004667 int hash;
NeilBrown72626682005-09-09 16:23:54 -07004668 list_del_init(&sh->lru);
4669 atomic_inc(&sh->count);
Shaohua Li566c09c2013-11-14 15:16:17 +11004670 hash = sh->hash_lock_index;
4671 __release_stripe(conf, sh, &temp_inactive_list[hash]);
NeilBrown72626682005-09-09 16:23:54 -07004672 }
4673}
4674
NeilBrown5c675f82014-12-15 12:56:56 +11004675static int raid5_congested(struct mddev *mddev, int bits)
NeilBrownf022b2f2006-10-03 01:15:56 -07004676{
NeilBrownd1688a62011-10-11 16:49:52 +11004677 struct r5conf *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07004678
4679 /* No difference between reads and writes. Just check
4680 * how busy the stripe_cache is
4681 */
NeilBrown3fa841d2009-09-23 18:10:29 +10004682
NeilBrown54233992015-02-26 12:21:04 +11004683 if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state))
NeilBrownf022b2f2006-10-03 01:15:56 -07004684 return 1;
4685 if (conf->quiesce)
4686 return 1;
Shaohua Li4bda5562013-11-14 15:16:17 +11004687 if (atomic_read(&conf->empty_inactive_list_nr))
NeilBrownf022b2f2006-10-03 01:15:56 -07004688 return 1;
4689
4690 return 0;
4691}
4692
NeilBrownfd01b882011-10-11 16:47:53 +11004693static int in_chunk_boundary(struct mddev *mddev, struct bio *bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004694{
NeilBrown3cb5edf2015-07-15 17:24:17 +10004695 struct r5conf *conf = mddev->private;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004696 sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev);
NeilBrown3cb5edf2015-07-15 17:24:17 +10004697 unsigned int chunk_sectors;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08004698 unsigned int bio_sectors = bio_sectors(bio);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004699
NeilBrown3cb5edf2015-07-15 17:24:17 +10004700 chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004701 return chunk_sectors >=
4702 ((sector & (chunk_sectors - 1)) + bio_sectors);
4703}
4704
4705/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004706 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
4707 * later sampled by raid5d.
4708 */
NeilBrownd1688a62011-10-11 16:49:52 +11004709static void add_bio_to_retry(struct bio *bi,struct r5conf *conf)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004710{
4711 unsigned long flags;
4712
4713 spin_lock_irqsave(&conf->device_lock, flags);
4714
4715 bi->bi_next = conf->retry_read_aligned_list;
4716 conf->retry_read_aligned_list = bi;
4717
4718 spin_unlock_irqrestore(&conf->device_lock, flags);
4719 md_wakeup_thread(conf->mddev->thread);
4720}
4721
NeilBrownd1688a62011-10-11 16:49:52 +11004722static struct bio *remove_bio_from_retry(struct r5conf *conf)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004723{
4724 struct bio *bi;
4725
4726 bi = conf->retry_read_aligned;
4727 if (bi) {
4728 conf->retry_read_aligned = NULL;
4729 return bi;
4730 }
4731 bi = conf->retry_read_aligned_list;
4732 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08004733 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004734 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02004735 /*
4736 * this sets the active strip count to 1 and the processed
4737 * strip count to zero (upper 8 bits)
4738 */
Shaohua Lie7836bd62012-07-19 16:01:31 +10004739 raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004740 }
4741
4742 return bi;
4743}
4744
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004745/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004746 * The "raid5_align_endio" should check if the read succeeded and if it
4747 * did, call bio_endio on the original bio (having bio_put the new bio
4748 * first).
4749 * If the read failed..
4750 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02004751static void raid5_align_endio(struct bio *bi)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004752{
4753 struct bio* raid_bi = bi->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11004754 struct mddev *mddev;
NeilBrownd1688a62011-10-11 16:49:52 +11004755 struct r5conf *conf;
NeilBrown3cb03002011-10-11 16:45:26 +11004756 struct md_rdev *rdev;
Sasha Levin9b81c842015-08-10 19:05:18 -04004757 int error = bi->bi_error;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004758
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004759 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004760
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004761 rdev = (void*)raid_bi->bi_next;
4762 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11004763 mddev = rdev->mddev;
4764 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004765
4766 rdev_dec_pending(rdev, conf->mddev);
4767
Sasha Levin9b81c842015-08-10 19:05:18 -04004768 if (!error) {
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07004769 trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev),
4770 raid_bi, 0);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02004771 bio_endio(raid_bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004772 if (atomic_dec_and_test(&conf->active_aligned_reads))
Yuanhan Liub1b46482015-05-08 18:19:06 +10004773 wake_up(&conf->wait_for_quiescent);
NeilBrown6712ecf2007-09-27 12:47:43 +02004774 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004775 }
4776
Dan Williams45b42332007-07-09 11:56:43 -07004777 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004778
4779 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004780}
4781
Ming Lin7ef6b122015-05-06 22:51:24 -07004782static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004783{
NeilBrownd1688a62011-10-11 16:49:52 +11004784 struct r5conf *conf = mddev->private;
NeilBrown8553fe7ec2009-12-14 12:49:47 +11004785 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004786 struct bio* align_bi;
NeilBrown3cb03002011-10-11 16:45:26 +11004787 struct md_rdev *rdev;
NeilBrown671488c2011-12-23 10:17:52 +11004788 sector_t end_sector;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004789
4790 if (!in_chunk_boundary(mddev, raid_bio)) {
Ming Lin7ef6b122015-05-06 22:51:24 -07004791 pr_debug("%s: non aligned\n", __func__);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004792 return 0;
4793 }
4794 /*
NeilBrowna167f662010-10-26 18:31:13 +11004795 * use bio_clone_mddev to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004796 */
NeilBrowna167f662010-10-26 18:31:13 +11004797 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004798 if (!align_bi)
4799 return 0;
4800 /*
4801 * set bi_end_io to a new function, and set bi_private to the
4802 * original bio.
4803 */
4804 align_bi->bi_end_io = raid5_align_endio;
4805 align_bi->bi_private = raid_bio;
4806 /*
4807 * compute position
4808 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07004809 align_bi->bi_iter.bi_sector =
4810 raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector,
4811 0, &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004812
Kent Overstreetf73a1c72012-09-25 15:05:12 -07004813 end_sector = bio_end_sector(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004814 rcu_read_lock();
NeilBrown671488c2011-12-23 10:17:52 +11004815 rdev = rcu_dereference(conf->disks[dd_idx].replacement);
4816 if (!rdev || test_bit(Faulty, &rdev->flags) ||
4817 rdev->recovery_offset < end_sector) {
4818 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
4819 if (rdev &&
4820 (test_bit(Faulty, &rdev->flags) ||
4821 !(test_bit(In_sync, &rdev->flags) ||
4822 rdev->recovery_offset >= end_sector)))
4823 rdev = NULL;
4824 }
4825 if (rdev) {
NeilBrown31c176e2011-07-28 11:39:22 +10004826 sector_t first_bad;
4827 int bad_sectors;
4828
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004829 atomic_inc(&rdev->nr_pending);
4830 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004831 raid_bio->bi_next = (void*)rdev;
4832 align_bi->bi_bdev = rdev->bdev;
Jens Axboeb7c44ed2015-07-24 12:37:59 -06004833 bio_clear_flag(align_bi, BIO_SEG_VALID);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004834
Kent Overstreet7140aaf2013-09-25 13:37:01 -07004835 if (is_badblock(rdev, align_bi->bi_iter.bi_sector,
Kent Overstreet4f024f32013-10-11 15:44:27 -07004836 bio_sectors(align_bi),
NeilBrown31c176e2011-07-28 11:39:22 +10004837 &first_bad, &bad_sectors)) {
Neil Brown387bb172007-02-08 14:20:29 -08004838 bio_put(align_bi);
4839 rdev_dec_pending(rdev, mddev);
4840 return 0;
4841 }
4842
majianpeng6c0544e2012-06-12 08:31:10 +08004843 /* No reshape active, so we can trust rdev->data_offset */
Kent Overstreet4f024f32013-10-11 15:44:27 -07004844 align_bi->bi_iter.bi_sector += rdev->data_offset;
majianpeng6c0544e2012-06-12 08:31:10 +08004845
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004846 spin_lock_irq(&conf->device_lock);
Yuanhan Liub1b46482015-05-08 18:19:06 +10004847 wait_event_lock_irq(conf->wait_for_quiescent,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004848 conf->quiesce == 0,
Lukas Czernereed8c022012-11-30 11:42:40 +01004849 conf->device_lock);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004850 atomic_inc(&conf->active_aligned_reads);
4851 spin_unlock_irq(&conf->device_lock);
4852
Jonathan Brassowe3620a32013-03-07 16:22:01 -06004853 if (mddev->gendisk)
4854 trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev),
4855 align_bi, disk_devt(mddev->gendisk),
Kent Overstreet4f024f32013-10-11 15:44:27 -07004856 raid_bio->bi_iter.bi_sector);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004857 generic_make_request(align_bi);
4858 return 1;
4859 } else {
4860 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004861 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004862 return 0;
4863 }
4864}
4865
Ming Lin7ef6b122015-05-06 22:51:24 -07004866static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio)
4867{
4868 struct bio *split;
4869
4870 do {
4871 sector_t sector = raid_bio->bi_iter.bi_sector;
4872 unsigned chunk_sects = mddev->chunk_sectors;
4873 unsigned sectors = chunk_sects - (sector & (chunk_sects-1));
4874
4875 if (sectors < bio_sectors(raid_bio)) {
4876 split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set);
4877 bio_chain(split, raid_bio);
4878 } else
4879 split = raid_bio;
4880
4881 if (!raid5_read_one_chunk(mddev, split)) {
4882 if (split != raid_bio)
4883 generic_make_request(raid_bio);
4884 return split;
4885 }
4886 } while (split != raid_bio);
4887
4888 return NULL;
4889}
4890
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004891/* __get_priority_stripe - get the next stripe to process
4892 *
4893 * Full stripe writes are allowed to pass preread active stripes up until
4894 * the bypass_threshold is exceeded. In general the bypass_count
4895 * increments when the handle_list is handled before the hold_list; however, it
4896 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
4897 * stripe with in flight i/o. The bypass_count will be reset when the
4898 * head of the hold_list has changed, i.e. the head was promoted to the
4899 * handle_list.
4900 */
Shaohua Li851c30c2013-08-28 14:30:16 +08004901static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004902{
Shaohua Li851c30c2013-08-28 14:30:16 +08004903 struct stripe_head *sh = NULL, *tmp;
4904 struct list_head *handle_list = NULL;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004905 struct r5worker_group *wg = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08004906
4907 if (conf->worker_cnt_per_group == 0) {
4908 handle_list = &conf->handle_list;
4909 } else if (group != ANY_GROUP) {
4910 handle_list = &conf->worker_groups[group].handle_list;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004911 wg = &conf->worker_groups[group];
Shaohua Li851c30c2013-08-28 14:30:16 +08004912 } else {
4913 int i;
4914 for (i = 0; i < conf->group_cnt; i++) {
4915 handle_list = &conf->worker_groups[i].handle_list;
Shaohua Libfc90cb2013-08-29 15:40:32 +08004916 wg = &conf->worker_groups[i];
Shaohua Li851c30c2013-08-28 14:30:16 +08004917 if (!list_empty(handle_list))
4918 break;
4919 }
4920 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004921
4922 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
4923 __func__,
Shaohua Li851c30c2013-08-28 14:30:16 +08004924 list_empty(handle_list) ? "empty" : "busy",
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004925 list_empty(&conf->hold_list) ? "empty" : "busy",
4926 atomic_read(&conf->pending_full_writes), conf->bypass_count);
4927
Shaohua Li851c30c2013-08-28 14:30:16 +08004928 if (!list_empty(handle_list)) {
4929 sh = list_entry(handle_list->next, typeof(*sh), lru);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004930
4931 if (list_empty(&conf->hold_list))
4932 conf->bypass_count = 0;
4933 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
4934 if (conf->hold_list.next == conf->last_hold)
4935 conf->bypass_count++;
4936 else {
4937 conf->last_hold = conf->hold_list.next;
4938 conf->bypass_count -= conf->bypass_threshold;
4939 if (conf->bypass_count < 0)
4940 conf->bypass_count = 0;
4941 }
4942 }
4943 } else if (!list_empty(&conf->hold_list) &&
4944 ((conf->bypass_threshold &&
4945 conf->bypass_count > conf->bypass_threshold) ||
4946 atomic_read(&conf->pending_full_writes) == 0)) {
Shaohua Li851c30c2013-08-28 14:30:16 +08004947
4948 list_for_each_entry(tmp, &conf->hold_list, lru) {
4949 if (conf->worker_cnt_per_group == 0 ||
4950 group == ANY_GROUP ||
4951 !cpu_online(tmp->cpu) ||
4952 cpu_to_group(tmp->cpu) == group) {
4953 sh = tmp;
4954 break;
4955 }
4956 }
4957
4958 if (sh) {
4959 conf->bypass_count -= conf->bypass_threshold;
4960 if (conf->bypass_count < 0)
4961 conf->bypass_count = 0;
4962 }
Shaohua Libfc90cb2013-08-29 15:40:32 +08004963 wg = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08004964 }
4965
4966 if (!sh)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004967 return NULL;
4968
Shaohua Libfc90cb2013-08-29 15:40:32 +08004969 if (wg) {
4970 wg->stripes_cnt--;
4971 sh->group = NULL;
4972 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004973 list_del_init(&sh->lru);
Shaohua Lic7a6d352014-04-15 09:12:54 +08004974 BUG_ON(atomic_inc_return(&sh->count) != 1);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004975 return sh;
4976}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08004977
Shaohua Li8811b592012-08-02 08:33:00 +10004978struct raid5_plug_cb {
4979 struct blk_plug_cb cb;
4980 struct list_head list;
Shaohua Li566c09c2013-11-14 15:16:17 +11004981 struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
Shaohua Li8811b592012-08-02 08:33:00 +10004982};
4983
4984static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
4985{
4986 struct raid5_plug_cb *cb = container_of(
4987 blk_cb, struct raid5_plug_cb, cb);
4988 struct stripe_head *sh;
4989 struct mddev *mddev = cb->cb.data;
4990 struct r5conf *conf = mddev->private;
NeilBrowna9add5d2012-10-31 11:59:09 +11004991 int cnt = 0;
Shaohua Li566c09c2013-11-14 15:16:17 +11004992 int hash;
Shaohua Li8811b592012-08-02 08:33:00 +10004993
4994 if (cb->list.next && !list_empty(&cb->list)) {
4995 spin_lock_irq(&conf->device_lock);
4996 while (!list_empty(&cb->list)) {
4997 sh = list_first_entry(&cb->list, struct stripe_head, lru);
4998 list_del_init(&sh->lru);
4999 /*
5000 * avoid race release_stripe_plug() sees
5001 * STRIPE_ON_UNPLUG_LIST clear but the stripe
5002 * is still in our list
5003 */
Peter Zijlstra4e857c52014-03-17 18:06:10 +01005004 smp_mb__before_atomic();
Shaohua Li8811b592012-08-02 08:33:00 +10005005 clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
Shaohua Li773ca822013-08-27 17:50:39 +08005006 /*
5007 * STRIPE_ON_RELEASE_LIST could be set here. In that
5008 * case, the count is always > 1 here
5009 */
Shaohua Li566c09c2013-11-14 15:16:17 +11005010 hash = sh->hash_lock_index;
5011 __release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
NeilBrowna9add5d2012-10-31 11:59:09 +11005012 cnt++;
Shaohua Li8811b592012-08-02 08:33:00 +10005013 }
5014 spin_unlock_irq(&conf->device_lock);
5015 }
Shaohua Li566c09c2013-11-14 15:16:17 +11005016 release_inactive_stripe_list(conf, cb->temp_inactive_list,
5017 NR_STRIPE_HASH_LOCKS);
Jonathan Brassowe3620a32013-03-07 16:22:01 -06005018 if (mddev->queue)
5019 trace_block_unplug(mddev->queue, cnt, !from_schedule);
Shaohua Li8811b592012-08-02 08:33:00 +10005020 kfree(cb);
5021}
5022
5023static void release_stripe_plug(struct mddev *mddev,
5024 struct stripe_head *sh)
5025{
5026 struct blk_plug_cb *blk_cb = blk_check_plugged(
5027 raid5_unplug, mddev,
5028 sizeof(struct raid5_plug_cb));
5029 struct raid5_plug_cb *cb;
5030
5031 if (!blk_cb) {
Shaohua Li6d036f72015-08-13 14:31:57 -07005032 raid5_release_stripe(sh);
Shaohua Li8811b592012-08-02 08:33:00 +10005033 return;
5034 }
5035
5036 cb = container_of(blk_cb, struct raid5_plug_cb, cb);
5037
Shaohua Li566c09c2013-11-14 15:16:17 +11005038 if (cb->list.next == NULL) {
5039 int i;
Shaohua Li8811b592012-08-02 08:33:00 +10005040 INIT_LIST_HEAD(&cb->list);
Shaohua Li566c09c2013-11-14 15:16:17 +11005041 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5042 INIT_LIST_HEAD(cb->temp_inactive_list + i);
5043 }
Shaohua Li8811b592012-08-02 08:33:00 +10005044
5045 if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
5046 list_add_tail(&sh->lru, &cb->list);
5047 else
Shaohua Li6d036f72015-08-13 14:31:57 -07005048 raid5_release_stripe(sh);
Shaohua Li8811b592012-08-02 08:33:00 +10005049}
5050
Shaohua Li620125f2012-10-11 13:49:05 +11005051static void make_discard_request(struct mddev *mddev, struct bio *bi)
5052{
5053 struct r5conf *conf = mddev->private;
5054 sector_t logical_sector, last_sector;
5055 struct stripe_head *sh;
5056 int remaining;
5057 int stripe_sectors;
5058
5059 if (mddev->reshape_position != MaxSector)
5060 /* Skip discard while reshape is happening */
5061 return;
5062
Kent Overstreet4f024f32013-10-11 15:44:27 -07005063 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
5064 last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9);
Shaohua Li620125f2012-10-11 13:49:05 +11005065
5066 bi->bi_next = NULL;
5067 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
5068
5069 stripe_sectors = conf->chunk_sectors *
5070 (conf->raid_disks - conf->max_degraded);
5071 logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector,
5072 stripe_sectors);
5073 sector_div(last_sector, stripe_sectors);
5074
5075 logical_sector *= conf->chunk_sectors;
5076 last_sector *= conf->chunk_sectors;
5077
5078 for (; logical_sector < last_sector;
5079 logical_sector += STRIPE_SECTORS) {
5080 DEFINE_WAIT(w);
5081 int d;
5082 again:
Shaohua Li6d036f72015-08-13 14:31:57 -07005083 sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0);
Shaohua Li620125f2012-10-11 13:49:05 +11005084 prepare_to_wait(&conf->wait_for_overlap, &w,
5085 TASK_UNINTERRUPTIBLE);
NeilBrownf8dfcff2013-03-12 12:18:06 +11005086 set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
5087 if (test_bit(STRIPE_SYNCING, &sh->state)) {
Shaohua Li6d036f72015-08-13 14:31:57 -07005088 raid5_release_stripe(sh);
NeilBrownf8dfcff2013-03-12 12:18:06 +11005089 schedule();
5090 goto again;
5091 }
5092 clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags);
Shaohua Li620125f2012-10-11 13:49:05 +11005093 spin_lock_irq(&sh->stripe_lock);
5094 for (d = 0; d < conf->raid_disks; d++) {
5095 if (d == sh->pd_idx || d == sh->qd_idx)
5096 continue;
5097 if (sh->dev[d].towrite || sh->dev[d].toread) {
5098 set_bit(R5_Overlap, &sh->dev[d].flags);
5099 spin_unlock_irq(&sh->stripe_lock);
Shaohua Li6d036f72015-08-13 14:31:57 -07005100 raid5_release_stripe(sh);
Shaohua Li620125f2012-10-11 13:49:05 +11005101 schedule();
5102 goto again;
5103 }
5104 }
NeilBrownf8dfcff2013-03-12 12:18:06 +11005105 set_bit(STRIPE_DISCARD, &sh->state);
Shaohua Li620125f2012-10-11 13:49:05 +11005106 finish_wait(&conf->wait_for_overlap, &w);
shli@kernel.org7a87f432014-12-15 12:57:03 +11005107 sh->overwrite_disks = 0;
Shaohua Li620125f2012-10-11 13:49:05 +11005108 for (d = 0; d < conf->raid_disks; d++) {
5109 if (d == sh->pd_idx || d == sh->qd_idx)
5110 continue;
5111 sh->dev[d].towrite = bi;
5112 set_bit(R5_OVERWRITE, &sh->dev[d].flags);
5113 raid5_inc_bi_active_stripes(bi);
shli@kernel.org7a87f432014-12-15 12:57:03 +11005114 sh->overwrite_disks++;
Shaohua Li620125f2012-10-11 13:49:05 +11005115 }
5116 spin_unlock_irq(&sh->stripe_lock);
5117 if (conf->mddev->bitmap) {
5118 for (d = 0;
5119 d < conf->raid_disks - conf->max_degraded;
5120 d++)
5121 bitmap_startwrite(mddev->bitmap,
5122 sh->sector,
5123 STRIPE_SECTORS,
5124 0);
5125 sh->bm_seq = conf->seq_flush + 1;
5126 set_bit(STRIPE_BIT_DELAY, &sh->state);
5127 }
5128
5129 set_bit(STRIPE_HANDLE, &sh->state);
5130 clear_bit(STRIPE_DELAYED, &sh->state);
5131 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5132 atomic_inc(&conf->preread_active_stripes);
5133 release_stripe_plug(mddev, sh);
5134 }
5135
5136 remaining = raid5_dec_bi_active_stripes(bi);
5137 if (remaining == 0) {
5138 md_write_end(mddev);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02005139 bio_endio(bi);
Shaohua Li620125f2012-10-11 13:49:05 +11005140 }
5141}
5142
Shaohua Li849674e2016-01-20 13:52:20 -08005143static void raid5_make_request(struct mddev *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005144{
NeilBrownd1688a62011-10-11 16:49:52 +11005145 struct r5conf *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11005146 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005147 sector_t new_sector;
5148 sector_t logical_sector, last_sector;
5149 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01005150 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11005151 int remaining;
Shaohua Li27c0f682014-04-09 11:25:47 +08005152 DEFINE_WAIT(w);
5153 bool do_prepare;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005154
Tejun Heoe9c74692010-09-03 11:56:18 +02005155 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
Shaohua Li828cbe92015-09-02 13:49:49 -07005156 int ret = r5l_handle_flush_request(conf->log, bi);
5157
5158 if (ret == 0)
5159 return;
5160 if (ret == -ENODEV) {
5161 md_flush_request(mddev, bi);
5162 return;
5163 }
5164 /* ret == -EAGAIN, fallback */
NeilBrowne5dcdd82005-09-09 16:23:41 -07005165 }
5166
NeilBrown3d310eb2005-06-21 17:17:26 -07005167 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07005168
Eric Mei9ffc8f72015-03-18 23:39:11 -06005169 /*
5170 * If array is degraded, better not do chunk aligned read because
5171 * later we might have to read it again in order to reconstruct
5172 * data on failed drives.
5173 */
5174 if (rw == READ && mddev->degraded == 0 &&
Ming Lin7ef6b122015-05-06 22:51:24 -07005175 mddev->reshape_position == MaxSector) {
5176 bi = chunk_aligned_read(mddev, bi);
5177 if (!bi)
5178 return;
5179 }
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08005180
Shaohua Li620125f2012-10-11 13:49:05 +11005181 if (unlikely(bi->bi_rw & REQ_DISCARD)) {
5182 make_discard_request(mddev, bi);
5183 return;
5184 }
5185
Kent Overstreet4f024f32013-10-11 15:44:27 -07005186 logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07005187 last_sector = bio_end_sector(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005188 bi->bi_next = NULL;
5189 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07005190
Shaohua Li27c0f682014-04-09 11:25:47 +08005191 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005192 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
NeilBrownb5663ba2009-03-31 14:39:38 +11005193 int previous;
NeilBrownc46501b2013-08-27 15:52:13 +10005194 int seq;
NeilBrownb578d552006-03-27 01:18:12 -08005195
Shaohua Li27c0f682014-04-09 11:25:47 +08005196 do_prepare = false;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005197 retry:
NeilBrownc46501b2013-08-27 15:52:13 +10005198 seq = read_seqcount_begin(&conf->gen_lock);
NeilBrownb5663ba2009-03-31 14:39:38 +11005199 previous = 0;
Shaohua Li27c0f682014-04-09 11:25:47 +08005200 if (do_prepare)
5201 prepare_to_wait(&conf->wait_for_overlap, &w,
5202 TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11005203 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11005204 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08005205 * 64bit on a 32bit platform, and so it might be
5206 * possible to see a half-updated value
Jesper Juhlaeb878b2011-04-10 18:06:17 +02005207 * Of course reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08005208 * the lock is dropped, so once we get a reference
5209 * to the stripe that we think it is, we will have
5210 * to check again.
5211 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005212 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10005213 if (mddev->reshape_backwards
NeilBrownfef9c612009-03-31 15:16:46 +11005214 ? logical_sector < conf->reshape_progress
5215 : logical_sector >= conf->reshape_progress) {
NeilBrownb5663ba2009-03-31 14:39:38 +11005216 previous = 1;
5217 } else {
NeilBrown2c810cd2012-05-21 09:27:00 +10005218 if (mddev->reshape_backwards
NeilBrownfef9c612009-03-31 15:16:46 +11005219 ? logical_sector < conf->reshape_safe
5220 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08005221 spin_unlock_irq(&conf->device_lock);
5222 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08005223 do_prepare = true;
NeilBrownb578d552006-03-27 01:18:12 -08005224 goto retry;
5225 }
5226 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005227 spin_unlock_irq(&conf->device_lock);
5228 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005229
NeilBrown112bf892009-03-31 14:39:38 +11005230 new_sector = raid5_compute_sector(conf, logical_sector,
5231 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11005232 &dd_idx, NULL);
Shaohua Li849674e2016-01-20 13:52:20 -08005233 pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n",
NeilBrownc46501b2013-08-27 15:52:13 +10005234 (unsigned long long)new_sector,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005235 (unsigned long long)logical_sector);
5236
Shaohua Li6d036f72015-08-13 14:31:57 -07005237 sh = raid5_get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10005238 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005239 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11005240 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005241 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08005242 * stripe, so we must do the range check again.
5243 * Expansion could still move past after this
5244 * test, but as we are holding a reference to
5245 * 'sh', we know that if that happens,
5246 * STRIPE_EXPANDING will get set and the expansion
5247 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005248 */
5249 int must_retry = 0;
5250 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10005251 if (mddev->reshape_backwards
NeilBrownb0f9ec02009-03-31 15:27:18 +11005252 ? logical_sector >= conf->reshape_progress
5253 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005254 /* mismatch, need to try again */
5255 must_retry = 1;
5256 spin_unlock_irq(&conf->device_lock);
5257 if (must_retry) {
Shaohua Li6d036f72015-08-13 14:31:57 -07005258 raid5_release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07005259 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08005260 do_prepare = true;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005261 goto retry;
5262 }
5263 }
NeilBrownc46501b2013-08-27 15:52:13 +10005264 if (read_seqcount_retry(&conf->gen_lock, seq)) {
5265 /* Might have got the wrong stripe_head
5266 * by accident
5267 */
Shaohua Li6d036f72015-08-13 14:31:57 -07005268 raid5_release_stripe(sh);
NeilBrownc46501b2013-08-27 15:52:13 +10005269 goto retry;
5270 }
NeilBrowne62e58a2009-07-01 13:15:35 +10005271
Namhyung Kimffd96e32011-07-18 17:38:51 +10005272 if (rw == WRITE &&
NeilBrowna5c308d2009-07-01 13:15:35 +10005273 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08005274 logical_sector < mddev->suspend_hi) {
Shaohua Li6d036f72015-08-13 14:31:57 -07005275 raid5_release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10005276 /* As the suspend_* range is controlled by
5277 * userspace, we want an interruptible
5278 * wait.
5279 */
5280 flush_signals(current);
5281 prepare_to_wait(&conf->wait_for_overlap,
5282 &w, TASK_INTERRUPTIBLE);
5283 if (logical_sector >= mddev->suspend_lo &&
Shaohua Li27c0f682014-04-09 11:25:47 +08005284 logical_sector < mddev->suspend_hi) {
NeilBrowne62e58a2009-07-01 13:15:35 +10005285 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08005286 do_prepare = true;
5287 }
NeilBrowne464eaf2006-03-27 01:18:14 -08005288 goto retry;
5289 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005290
5291 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
shli@kernel.orgda41ba62014-12-15 12:57:03 +11005292 !add_stripe_bio(sh, bi, dd_idx, rw, previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005293 /* Stripe is busy expanding or
5294 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07005295 * and wait a while
5296 */
NeilBrown482c0832011-04-18 18:25:42 +10005297 md_wakeup_thread(mddev->thread);
Shaohua Li6d036f72015-08-13 14:31:57 -07005298 raid5_release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005299 schedule();
Shaohua Li27c0f682014-04-09 11:25:47 +08005300 do_prepare = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005301 goto retry;
5302 }
NeilBrown6ed30032008-02-06 01:40:00 -08005303 set_bit(STRIPE_HANDLE, &sh->state);
5304 clear_bit(STRIPE_DELAYED, &sh->state);
shli@kernel.org59fc6302014-12-15 12:57:03 +11005305 if ((!sh->batch_head || sh == sh->batch_head) &&
5306 (bi->bi_rw & REQ_SYNC) &&
NeilBrown729a1862009-12-14 12:49:50 +11005307 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
5308 atomic_inc(&conf->preread_active_stripes);
Shaohua Li8811b592012-08-02 08:33:00 +10005309 release_stripe_plug(mddev, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005310 } else {
5311 /* cannot get stripe for read-ahead, just give-up */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02005312 bi->bi_error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005313 break;
5314 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005315 }
Shaohua Li27c0f682014-04-09 11:25:47 +08005316 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown7c13edc2011-04-18 18:25:43 +10005317
Shaohua Lie7836bd62012-07-19 16:01:31 +10005318 remaining = raid5_dec_bi_active_stripes(bi);
NeilBrownf6344752006-03-27 01:18:17 -08005319 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005320
NeilBrown16a53ec2006-06-26 00:27:38 -07005321 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07005322 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02005323
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07005324 trace_block_bio_complete(bdev_get_queue(bi->bi_bdev),
5325 bi, 0);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02005326 bio_endio(bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005328}
5329
NeilBrownfd01b882011-10-11 16:47:53 +11005330static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks);
Dan Williamsb522adc2009-03-31 15:00:31 +11005331
NeilBrownfd01b882011-10-11 16:47:53 +11005332static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005333{
NeilBrown52c03292006-06-26 00:27:43 -07005334 /* reshaping is quite different to recovery/resync so it is
5335 * handled quite separately ... here.
5336 *
5337 * On each call to sync_request, we gather one chunk worth of
5338 * destination stripes and flag them as expanding.
5339 * Then we find all the source stripes and request reads.
5340 * As the reads complete, handle_stripe will copy the data
5341 * into the destination stripe and release that stripe.
5342 */
NeilBrownd1688a62011-10-11 16:49:52 +11005343 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005344 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08005345 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08005346 int raid_disks = conf->previous_raid_disks;
5347 int data_disks = raid_disks - conf->max_degraded;
5348 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07005349 int i;
5350 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11005351 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11005352 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11005353 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11005354 struct list_head stripes;
NeilBrown92140482015-07-06 12:28:45 +10005355 sector_t retn;
NeilBrown52c03292006-06-26 00:27:43 -07005356
NeilBrownfef9c612009-03-31 15:16:46 +11005357 if (sector_nr == 0) {
5358 /* If restarting in the middle, skip the initial sectors */
NeilBrown2c810cd2012-05-21 09:27:00 +10005359 if (mddev->reshape_backwards &&
NeilBrownfef9c612009-03-31 15:16:46 +11005360 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
5361 sector_nr = raid5_size(mddev, 0, 0)
5362 - conf->reshape_progress;
NeilBrown6cbd8142015-07-24 13:30:32 +10005363 } else if (mddev->reshape_backwards &&
5364 conf->reshape_progress == MaxSector) {
5365 /* shouldn't happen, but just in case, finish up.*/
5366 sector_nr = MaxSector;
NeilBrown2c810cd2012-05-21 09:27:00 +10005367 } else if (!mddev->reshape_backwards &&
NeilBrownfef9c612009-03-31 15:16:46 +11005368 conf->reshape_progress > 0)
5369 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08005370 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11005371 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11005372 mddev->curr_resync_completed = sector_nr;
5373 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11005374 *skipped = 1;
NeilBrown92140482015-07-06 12:28:45 +10005375 retn = sector_nr;
5376 goto finish;
NeilBrownfef9c612009-03-31 15:16:46 +11005377 }
NeilBrown52c03292006-06-26 00:27:43 -07005378 }
5379
NeilBrown7a661382009-03-31 15:21:40 +11005380 /* We need to process a full chunk at a time.
5381 * If old and new chunk sizes differ, we need to process the
5382 * largest of these
5383 */
NeilBrown3cb5edf2015-07-15 17:24:17 +10005384
5385 reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors);
NeilBrown7a661382009-03-31 15:21:40 +11005386
NeilBrownb5254dd2012-05-21 09:27:01 +10005387 /* We update the metadata at least every 10 seconds, or when
5388 * the data about to be copied would over-write the source of
5389 * the data at the front of the range. i.e. one new_stripe
5390 * along from reshape_progress new_maps to after where
5391 * reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07005392 */
NeilBrownfef9c612009-03-31 15:16:46 +11005393 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08005394 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11005395 readpos = conf->reshape_progress;
5396 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11005397 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08005398 sector_div(safepos, data_disks);
NeilBrown2c810cd2012-05-21 09:27:00 +10005399 if (mddev->reshape_backwards) {
NeilBrownc74c0d72015-07-15 17:54:15 +10005400 BUG_ON(writepos < reshape_sectors);
5401 writepos -= reshape_sectors;
NeilBrownc8f517c2009-03-31 15:28:40 +11005402 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11005403 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11005404 } else {
NeilBrown7a661382009-03-31 15:21:40 +11005405 writepos += reshape_sectors;
NeilBrownc74c0d72015-07-15 17:54:15 +10005406 /* readpos and safepos are worst-case calculations.
5407 * A negative number is overly pessimistic, and causes
5408 * obvious problems for unsigned storage. So clip to 0.
5409 */
NeilBrowned37d832009-05-27 21:39:05 +10005410 readpos -= min_t(sector_t, reshape_sectors, readpos);
5411 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11005412 }
NeilBrown52c03292006-06-26 00:27:43 -07005413
NeilBrownb5254dd2012-05-21 09:27:01 +10005414 /* Having calculated the 'writepos' possibly use it
5415 * to set 'stripe_addr' which is where we will write to.
5416 */
5417 if (mddev->reshape_backwards) {
5418 BUG_ON(conf->reshape_progress == 0);
5419 stripe_addr = writepos;
5420 BUG_ON((mddev->dev_sectors &
5421 ~((sector_t)reshape_sectors - 1))
5422 - reshape_sectors - stripe_addr
5423 != sector_nr);
5424 } else {
5425 BUG_ON(writepos != sector_nr + reshape_sectors);
5426 stripe_addr = sector_nr;
5427 }
5428
NeilBrownc8f517c2009-03-31 15:28:40 +11005429 /* 'writepos' is the most advanced device address we might write.
5430 * 'readpos' is the least advanced device address we might read.
5431 * 'safepos' is the least address recorded in the metadata as having
5432 * been reshaped.
NeilBrownb5254dd2012-05-21 09:27:01 +10005433 * If there is a min_offset_diff, these are adjusted either by
5434 * increasing the safepos/readpos if diff is negative, or
5435 * increasing writepos if diff is positive.
5436 * If 'readpos' is then behind 'writepos', there is no way that we can
NeilBrownc8f517c2009-03-31 15:28:40 +11005437 * ensure safety in the face of a crash - that must be done by userspace
5438 * making a backup of the data. So in that case there is no particular
5439 * rush to update metadata.
5440 * Otherwise if 'safepos' is behind 'writepos', then we really need to
5441 * update the metadata to advance 'safepos' to match 'readpos' so that
5442 * we can be safe in the event of a crash.
5443 * So we insist on updating metadata if safepos is behind writepos and
5444 * readpos is beyond writepos.
5445 * In any case, update the metadata every 10 seconds.
5446 * Maybe that number should be configurable, but I'm not sure it is
5447 * worth it.... maybe it could be a multiple of safemode_delay???
5448 */
NeilBrownb5254dd2012-05-21 09:27:01 +10005449 if (conf->min_offset_diff < 0) {
5450 safepos += -conf->min_offset_diff;
5451 readpos += -conf->min_offset_diff;
5452 } else
5453 writepos += conf->min_offset_diff;
5454
NeilBrown2c810cd2012-05-21 09:27:00 +10005455 if ((mddev->reshape_backwards
NeilBrownc8f517c2009-03-31 15:28:40 +11005456 ? (safepos > writepos && readpos < writepos)
5457 : (safepos < writepos && readpos > writepos)) ||
5458 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07005459 /* Cannot proceed until we've updated the superblock... */
5460 wait_event(conf->wait_for_overlap,
NeilBrownc91abf52013-11-19 12:02:01 +11005461 atomic_read(&conf->reshape_stripes)==0
5462 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5463 if (atomic_read(&conf->reshape_stripes) != 0)
5464 return 0;
NeilBrownfef9c612009-03-31 15:16:46 +11005465 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11005466 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11005467 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07005468 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07005469 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07005470 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrownc91abf52013-11-19 12:02:01 +11005471 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5472 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5473 return 0;
NeilBrown52c03292006-06-26 00:27:43 -07005474 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11005475 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07005476 spin_unlock_irq(&conf->device_lock);
5477 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10005478 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07005479 }
5480
NeilBrownab69ae12009-03-31 15:26:47 +11005481 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11005482 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07005483 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10005484 int skipped_disk = 0;
Shaohua Li6d036f72015-08-13 14:31:57 -07005485 sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07005486 set_bit(STRIPE_EXPANDING, &sh->state);
5487 atomic_inc(&conf->reshape_stripes);
5488 /* If any of this stripe is beyond the end of the old
5489 * array, then we need to zero those blocks
5490 */
5491 for (j=sh->disks; j--;) {
5492 sector_t s;
5493 if (j == sh->pd_idx)
5494 continue;
NeilBrownf4168852007-02-28 20:11:53 -08005495 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11005496 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08005497 continue;
Shaohua Li6d036f72015-08-13 14:31:57 -07005498 s = raid5_compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11005499 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10005500 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07005501 continue;
5502 }
5503 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
5504 set_bit(R5_Expanded, &sh->dev[j].flags);
5505 set_bit(R5_UPTODATE, &sh->dev[j].flags);
5506 }
NeilBrowna9f326e2009-09-23 18:06:41 +10005507 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07005508 set_bit(STRIPE_EXPAND_READY, &sh->state);
5509 set_bit(STRIPE_HANDLE, &sh->state);
5510 }
NeilBrownab69ae12009-03-31 15:26:47 +11005511 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07005512 }
5513 spin_lock_irq(&conf->device_lock);
NeilBrown2c810cd2012-05-21 09:27:00 +10005514 if (mddev->reshape_backwards)
NeilBrown7a661382009-03-31 15:21:40 +11005515 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005516 else
NeilBrown7a661382009-03-31 15:21:40 +11005517 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07005518 spin_unlock_irq(&conf->device_lock);
5519 /* Ok, those stripe are ready. We can start scheduling
5520 * reads on the source stripes.
5521 * The source stripes are determined by mapping the first and last
5522 * block on the destination stripes.
5523 */
NeilBrown52c03292006-06-26 00:27:43 -07005524 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11005525 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11005526 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07005527 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10005528 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10005529 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11005530 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11005531 if (last_sector >= mddev->dev_sectors)
5532 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07005533 while (first_sector <= last_sector) {
Shaohua Li6d036f72015-08-13 14:31:57 -07005534 sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07005535 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
5536 set_bit(STRIPE_HANDLE, &sh->state);
Shaohua Li6d036f72015-08-13 14:31:57 -07005537 raid5_release_stripe(sh);
NeilBrown52c03292006-06-26 00:27:43 -07005538 first_sector += STRIPE_SECTORS;
5539 }
NeilBrownab69ae12009-03-31 15:26:47 +11005540 /* Now that the sources are clearly marked, we can release
5541 * the destination stripes
5542 */
5543 while (!list_empty(&stripes)) {
5544 sh = list_entry(stripes.next, struct stripe_head, lru);
5545 list_del_init(&sh->lru);
Shaohua Li6d036f72015-08-13 14:31:57 -07005546 raid5_release_stripe(sh);
NeilBrownab69ae12009-03-31 15:26:47 +11005547 }
NeilBrownc6207272008-02-06 01:39:52 -08005548 /* If this takes us to the resync_max point where we have to pause,
5549 * then we need to write out the superblock.
5550 */
NeilBrown7a661382009-03-31 15:21:40 +11005551 sector_nr += reshape_sectors;
NeilBrown92140482015-07-06 12:28:45 +10005552 retn = reshape_sectors;
5553finish:
NeilBrownc5e19d92015-07-17 12:06:02 +10005554 if (mddev->curr_resync_completed > mddev->resync_max ||
5555 (sector_nr - mddev->curr_resync_completed) * 2
NeilBrownc03f6a12009-04-17 11:06:30 +10005556 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08005557 /* Cannot proceed until we've updated the superblock... */
5558 wait_event(conf->wait_for_overlap,
NeilBrownc91abf52013-11-19 12:02:01 +11005559 atomic_read(&conf->reshape_stripes) == 0
5560 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5561 if (atomic_read(&conf->reshape_stripes) != 0)
5562 goto ret;
NeilBrownfef9c612009-03-31 15:16:46 +11005563 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11005564 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11005565 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08005566 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5567 md_wakeup_thread(mddev->thread);
5568 wait_event(mddev->sb_wait,
5569 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
NeilBrownc91abf52013-11-19 12:02:01 +11005570 || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
5571 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5572 goto ret;
NeilBrownc6207272008-02-06 01:39:52 -08005573 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11005574 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08005575 spin_unlock_irq(&conf->device_lock);
5576 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10005577 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08005578 }
NeilBrownc91abf52013-11-19 12:02:01 +11005579ret:
NeilBrown92140482015-07-06 12:28:45 +10005580 return retn;
NeilBrown52c03292006-06-26 00:27:43 -07005581}
5582
Shaohua Li849674e2016-01-20 13:52:20 -08005583static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr,
5584 int *skipped)
NeilBrown52c03292006-06-26 00:27:43 -07005585{
NeilBrownd1688a62011-10-11 16:49:52 +11005586 struct r5conf *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07005587 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11005588 sector_t max_sector = mddev->dev_sectors;
NeilBrown57dab0b2010-10-19 10:03:39 +11005589 sector_t sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07005590 int still_degraded = 0;
5591 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005592
NeilBrown72626682005-09-09 16:23:54 -07005593 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005594 /* just being told to finish up .. nothing much to do */
NeilBrowncea9c222009-03-31 15:15:05 +11005595
NeilBrown29269552006-03-27 01:18:10 -08005596 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
5597 end_reshape(conf);
5598 return 0;
5599 }
NeilBrown72626682005-09-09 16:23:54 -07005600
5601 if (mddev->curr_resync < max_sector) /* aborted */
5602 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
5603 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07005604 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07005605 conf->fullsync = 0;
5606 bitmap_close_sync(mddev->bitmap);
5607
Linus Torvalds1da177e2005-04-16 15:20:36 -07005608 return 0;
5609 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08005610
NeilBrown64bd6602009-08-03 10:59:58 +10005611 /* Allow raid5_quiesce to complete */
5612 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
5613
NeilBrown52c03292006-06-26 00:27:43 -07005614 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5615 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08005616
NeilBrownc6207272008-02-06 01:39:52 -08005617 /* No need to check resync_max as we never do more than one
5618 * stripe, and as resync_max will always be on a chunk boundary,
5619 * if the check in md_do_sync didn't fire, there is no chance
5620 * of overstepping resync_max here
5621 */
5622
NeilBrown16a53ec2006-06-26 00:27:38 -07005623 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07005624 * to resync, then assert that we are finished, because there is
5625 * nothing we can do.
5626 */
NeilBrown3285edf2006-06-26 00:27:55 -07005627 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07005628 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11005629 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07005630 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005631 return rv;
5632 }
majianpeng6f608042013-04-24 11:42:41 +10005633 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
5634 !conf->fullsync &&
5635 !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
5636 sync_blocks >= STRIPE_SECTORS) {
NeilBrown72626682005-09-09 16:23:54 -07005637 /* we can skip this block, and probably more */
5638 sync_blocks /= STRIPE_SECTORS;
5639 *skipped = 1;
5640 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
5641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005642
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10005643 bitmap_cond_end_sync(mddev->bitmap, sector_nr, false);
NeilBrownb47490c2008-02-06 01:39:50 -08005644
Shaohua Li6d036f72015-08-13 14:31:57 -07005645 sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005646 if (sh == NULL) {
Shaohua Li6d036f72015-08-13 14:31:57 -07005647 sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005648 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07005649 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07005650 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08005651 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005652 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005653 /* Need to check if array will still be degraded after recovery/resync
Eric Mei16d9cfa2015-01-06 09:35:02 -08005654 * Note in case of > 1 drive failures it's possible we're rebuilding
5655 * one drive while leaving another faulty drive in array.
NeilBrown16a53ec2006-06-26 00:27:38 -07005656 */
Eric Mei16d9cfa2015-01-06 09:35:02 -08005657 rcu_read_lock();
5658 for (i = 0; i < conf->raid_disks; i++) {
5659 struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev);
5660
5661 if (rdev == NULL || test_bit(Faulty, &rdev->flags))
NeilBrown16a53ec2006-06-26 00:27:38 -07005662 still_degraded = 1;
Eric Mei16d9cfa2015-01-06 09:35:02 -08005663 }
5664 rcu_read_unlock();
NeilBrown16a53ec2006-06-26 00:27:38 -07005665
5666 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
5667
NeilBrown83206d62011-07-26 11:19:49 +10005668 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
Eivind Sarto053f5b62014-06-09 17:06:19 -07005669 set_bit(STRIPE_HANDLE, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005670
Shaohua Li6d036f72015-08-13 14:31:57 -07005671 raid5_release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005672
5673 return STRIPE_SECTORS;
5674}
5675
NeilBrownd1688a62011-10-11 16:49:52 +11005676static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005677{
5678 /* We may not be able to submit a whole bio at once as there
5679 * may not be enough stripe_heads available.
5680 * We cannot pre-allocate enough stripe_heads as we may need
5681 * more than exist in the cache (if we allow ever large chunks).
5682 * So we do one stripe head at a time and record in
5683 * ->bi_hw_segments how many have been done.
5684 *
5685 * We *know* that this entire raid_bio is in one chunk, so
5686 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
5687 */
5688 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11005689 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005690 sector_t sector, logical_sector, last_sector;
5691 int scnt = 0;
5692 int remaining;
5693 int handled = 0;
5694
Kent Overstreet4f024f32013-10-11 15:44:27 -07005695 logical_sector = raid_bio->bi_iter.bi_sector &
5696 ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11005697 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11005698 0, &dd_idx, NULL);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07005699 last_sector = bio_end_sector(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005700
5701 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08005702 logical_sector += STRIPE_SECTORS,
5703 sector += STRIPE_SECTORS,
5704 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005705
Shaohua Lie7836bd62012-07-19 16:01:31 +10005706 if (scnt < raid5_bi_processed_stripes(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005707 /* already done this stripe */
5708 continue;
5709
Shaohua Li6d036f72015-08-13 14:31:57 -07005710 sh = raid5_get_active_stripe(conf, sector, 0, 1, 1);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005711
5712 if (!sh) {
5713 /* failed to get a stripe - must wait */
Shaohua Lie7836bd62012-07-19 16:01:31 +10005714 raid5_set_bi_processed_stripes(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005715 conf->retry_read_aligned = raid_bio;
5716 return handled;
5717 }
5718
shli@kernel.orgda41ba62014-12-15 12:57:03 +11005719 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) {
Shaohua Li6d036f72015-08-13 14:31:57 -07005720 raid5_release_stripe(sh);
Shaohua Lie7836bd62012-07-19 16:01:31 +10005721 raid5_set_bi_processed_stripes(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08005722 conf->retry_read_aligned = raid_bio;
5723 return handled;
5724 }
5725
majianpeng3f9e7c12012-07-31 10:04:21 +10005726 set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags);
Dan Williams36d1c642009-07-14 11:48:22 -07005727 handle_stripe(sh);
Shaohua Li6d036f72015-08-13 14:31:57 -07005728 raid5_release_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005729 handled++;
5730 }
Shaohua Lie7836bd62012-07-19 16:01:31 +10005731 remaining = raid5_dec_bi_active_stripes(raid_bio);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07005732 if (remaining == 0) {
5733 trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev),
5734 raid_bio, 0);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02005735 bio_endio(raid_bio);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -07005736 }
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005737 if (atomic_dec_and_test(&conf->active_aligned_reads))
Yuanhan Liub1b46482015-05-08 18:19:06 +10005738 wake_up(&conf->wait_for_quiescent);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005739 return handled;
5740}
5741
Shaohua Libfc90cb2013-08-29 15:40:32 +08005742static int handle_active_stripes(struct r5conf *conf, int group,
Shaohua Li566c09c2013-11-14 15:16:17 +11005743 struct r5worker *worker,
5744 struct list_head *temp_inactive_list)
Shaohua Li46a06402012-08-02 08:33:15 +10005745{
5746 struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
Shaohua Li566c09c2013-11-14 15:16:17 +11005747 int i, batch_size = 0, hash;
5748 bool release_inactive = false;
Shaohua Li46a06402012-08-02 08:33:15 +10005749
5750 while (batch_size < MAX_STRIPE_BATCH &&
Shaohua Li851c30c2013-08-28 14:30:16 +08005751 (sh = __get_priority_stripe(conf, group)) != NULL)
Shaohua Li46a06402012-08-02 08:33:15 +10005752 batch[batch_size++] = sh;
5753
Shaohua Li566c09c2013-11-14 15:16:17 +11005754 if (batch_size == 0) {
5755 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
5756 if (!list_empty(temp_inactive_list + i))
5757 break;
Shaohua Lia8c34f92015-09-02 13:49:46 -07005758 if (i == NR_STRIPE_HASH_LOCKS) {
5759 spin_unlock_irq(&conf->device_lock);
5760 r5l_flush_stripe_to_raid(conf->log);
5761 spin_lock_irq(&conf->device_lock);
Shaohua Li566c09c2013-11-14 15:16:17 +11005762 return batch_size;
Shaohua Lia8c34f92015-09-02 13:49:46 -07005763 }
Shaohua Li566c09c2013-11-14 15:16:17 +11005764 release_inactive = true;
5765 }
Shaohua Li46a06402012-08-02 08:33:15 +10005766 spin_unlock_irq(&conf->device_lock);
5767
Shaohua Li566c09c2013-11-14 15:16:17 +11005768 release_inactive_stripe_list(conf, temp_inactive_list,
5769 NR_STRIPE_HASH_LOCKS);
5770
Shaohua Lia8c34f92015-09-02 13:49:46 -07005771 r5l_flush_stripe_to_raid(conf->log);
Shaohua Li566c09c2013-11-14 15:16:17 +11005772 if (release_inactive) {
5773 spin_lock_irq(&conf->device_lock);
5774 return 0;
5775 }
5776
Shaohua Li46a06402012-08-02 08:33:15 +10005777 for (i = 0; i < batch_size; i++)
5778 handle_stripe(batch[i]);
Shaohua Lif6bed0e2015-08-13 14:31:59 -07005779 r5l_write_stripe_run(conf->log);
Shaohua Li46a06402012-08-02 08:33:15 +10005780
5781 cond_resched();
5782
5783 spin_lock_irq(&conf->device_lock);
Shaohua Li566c09c2013-11-14 15:16:17 +11005784 for (i = 0; i < batch_size; i++) {
5785 hash = batch[i]->hash_lock_index;
5786 __release_stripe(conf, batch[i], &temp_inactive_list[hash]);
5787 }
Shaohua Li46a06402012-08-02 08:33:15 +10005788 return batch_size;
5789}
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005790
Shaohua Li851c30c2013-08-28 14:30:16 +08005791static void raid5_do_work(struct work_struct *work)
5792{
5793 struct r5worker *worker = container_of(work, struct r5worker, work);
5794 struct r5worker_group *group = worker->group;
5795 struct r5conf *conf = group->conf;
5796 int group_id = group - conf->worker_groups;
5797 int handled;
5798 struct blk_plug plug;
5799
5800 pr_debug("+++ raid5worker active\n");
5801
5802 blk_start_plug(&plug);
5803 handled = 0;
5804 spin_lock_irq(&conf->device_lock);
5805 while (1) {
5806 int batch_size, released;
5807
Shaohua Li566c09c2013-11-14 15:16:17 +11005808 released = release_stripe_list(conf, worker->temp_inactive_list);
Shaohua Li851c30c2013-08-28 14:30:16 +08005809
Shaohua Li566c09c2013-11-14 15:16:17 +11005810 batch_size = handle_active_stripes(conf, group_id, worker,
5811 worker->temp_inactive_list);
Shaohua Libfc90cb2013-08-29 15:40:32 +08005812 worker->working = false;
Shaohua Li851c30c2013-08-28 14:30:16 +08005813 if (!batch_size && !released)
5814 break;
5815 handled += batch_size;
5816 }
5817 pr_debug("%d stripes handled\n", handled);
5818
5819 spin_unlock_irq(&conf->device_lock);
5820 blk_finish_plug(&plug);
5821
5822 pr_debug("--- raid5worker inactive\n");
5823}
5824
Linus Torvalds1da177e2005-04-16 15:20:36 -07005825/*
5826 * This is our raid5 kernel thread.
5827 *
5828 * We scan the hash table for stripes which can be handled now.
5829 * During the scan, completed stripes are saved for us by the interrupt
5830 * handler, so that they will not have to wait for our next wakeup.
5831 */
Shaohua Li4ed87312012-10-11 13:34:00 +11005832static void raid5d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005833{
Shaohua Li4ed87312012-10-11 13:34:00 +11005834 struct mddev *mddev = thread->mddev;
NeilBrownd1688a62011-10-11 16:49:52 +11005835 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005836 int handled;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005837 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005838
Dan Williams45b42332007-07-09 11:56:43 -07005839 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005840
5841 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005842
NeilBrownc3cce6c2015-08-14 12:47:33 +10005843 if (!bio_list_empty(&conf->return_bi) &&
5844 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
5845 struct bio_list tmp = BIO_EMPTY_LIST;
5846 spin_lock_irq(&conf->device_lock);
5847 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
5848 bio_list_merge(&tmp, &conf->return_bi);
5849 bio_list_init(&conf->return_bi);
5850 }
5851 spin_unlock_irq(&conf->device_lock);
5852 return_io(&tmp);
5853 }
5854
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005855 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005856 handled = 0;
5857 spin_lock_irq(&conf->device_lock);
5858 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005859 struct bio *bio;
Shaohua Li773ca822013-08-27 17:50:39 +08005860 int batch_size, released;
5861
Shaohua Li566c09c2013-11-14 15:16:17 +11005862 released = release_stripe_list(conf, conf->temp_inactive_list);
NeilBrownedbe83a2015-02-26 12:47:56 +11005863 if (released)
5864 clear_bit(R5_DID_ALLOC, &conf->cache_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005865
NeilBrown0021b7b2012-07-31 09:08:14 +02005866 if (
NeilBrown7c13edc2011-04-18 18:25:43 +10005867 !list_empty(&conf->bitmap_list)) {
5868 /* Now is a good time to flush some bitmap updates */
5869 conf->seq_flush++;
NeilBrown700e4322005-11-28 13:44:10 -08005870 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07005871 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08005872 spin_lock_irq(&conf->device_lock);
NeilBrown7c13edc2011-04-18 18:25:43 +10005873 conf->seq_write = conf->seq_flush;
Shaohua Li566c09c2013-11-14 15:16:17 +11005874 activate_bit_delay(conf, conf->temp_inactive_list);
NeilBrown72626682005-09-09 16:23:54 -07005875 }
NeilBrown0021b7b2012-07-31 09:08:14 +02005876 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07005877
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005878 while ((bio = remove_bio_from_retry(conf))) {
5879 int ok;
5880 spin_unlock_irq(&conf->device_lock);
5881 ok = retry_aligned_read(conf, bio);
5882 spin_lock_irq(&conf->device_lock);
5883 if (!ok)
5884 break;
5885 handled++;
5886 }
5887
Shaohua Li566c09c2013-11-14 15:16:17 +11005888 batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
5889 conf->temp_inactive_list);
Shaohua Li773ca822013-08-27 17:50:39 +08005890 if (!batch_size && !released)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005891 break;
Shaohua Li46a06402012-08-02 08:33:15 +10005892 handled += batch_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005893
Shaohua Li46a06402012-08-02 08:33:15 +10005894 if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) {
5895 spin_unlock_irq(&conf->device_lock);
NeilBrownde393cd2011-07-28 11:31:48 +10005896 md_check_recovery(mddev);
Shaohua Li46a06402012-08-02 08:33:15 +10005897 spin_lock_irq(&conf->device_lock);
5898 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005899 }
Dan Williams45b42332007-07-09 11:56:43 -07005900 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005901
5902 spin_unlock_irq(&conf->device_lock);
NeilBrown2d5b5692015-07-06 12:49:23 +10005903 if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) &&
5904 mutex_trylock(&conf->cache_size_mutex)) {
NeilBrownedbe83a2015-02-26 12:47:56 +11005905 grow_one_stripe(conf, __GFP_NOWARN);
5906 /* Set flag even if allocation failed. This helps
5907 * slow down allocation requests when mem is short
5908 */
5909 set_bit(R5_DID_ALLOC, &conf->cache_state);
NeilBrown2d5b5692015-07-06 12:49:23 +10005910 mutex_unlock(&conf->cache_size_mutex);
NeilBrownedbe83a2015-02-26 12:47:56 +11005911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005912
Shaohua Li0576b1c2015-08-13 14:32:00 -07005913 r5l_flush_stripe_to_raid(conf->log);
5914
Dan Williamsc9f21aa2008-07-23 12:05:51 -07005915 async_tx_issue_pending_all();
NeilBrowne1dfa0a2011-04-18 18:25:41 +10005916 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005917
Dan Williams45b42332007-07-09 11:56:43 -07005918 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005919}
5920
NeilBrown3f294f42005-11-08 21:39:25 -08005921static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005922raid5_show_stripe_cache_size(struct mddev *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08005923{
NeilBrown7b1485b2014-12-15 12:56:59 +11005924 struct r5conf *conf;
5925 int ret = 0;
5926 spin_lock(&mddev->lock);
5927 conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08005928 if (conf)
NeilBrownedbe83a2015-02-26 12:47:56 +11005929 ret = sprintf(page, "%d\n", conf->min_nr_stripes);
NeilBrown7b1485b2014-12-15 12:56:59 +11005930 spin_unlock(&mddev->lock);
5931 return ret;
NeilBrown3f294f42005-11-08 21:39:25 -08005932}
5933
NeilBrownc41d4ac2010-06-01 19:37:24 +10005934int
NeilBrownfd01b882011-10-11 16:47:53 +11005935raid5_set_cache_size(struct mddev *mddev, int size)
NeilBrownc41d4ac2010-06-01 19:37:24 +10005936{
NeilBrownd1688a62011-10-11 16:49:52 +11005937 struct r5conf *conf = mddev->private;
NeilBrownc41d4ac2010-06-01 19:37:24 +10005938 int err;
5939
5940 if (size <= 16 || size > 32768)
5941 return -EINVAL;
NeilBrown486f0642015-02-25 12:10:35 +11005942
NeilBrownedbe83a2015-02-26 12:47:56 +11005943 conf->min_nr_stripes = size;
NeilBrown2d5b5692015-07-06 12:49:23 +10005944 mutex_lock(&conf->cache_size_mutex);
NeilBrown486f0642015-02-25 12:10:35 +11005945 while (size < conf->max_nr_stripes &&
5946 drop_one_stripe(conf))
5947 ;
NeilBrown2d5b5692015-07-06 12:49:23 +10005948 mutex_unlock(&conf->cache_size_mutex);
NeilBrown486f0642015-02-25 12:10:35 +11005949
NeilBrownedbe83a2015-02-26 12:47:56 +11005950
NeilBrownc41d4ac2010-06-01 19:37:24 +10005951 err = md_allow_write(mddev);
5952 if (err)
5953 return err;
NeilBrown486f0642015-02-25 12:10:35 +11005954
NeilBrown2d5b5692015-07-06 12:49:23 +10005955 mutex_lock(&conf->cache_size_mutex);
NeilBrown486f0642015-02-25 12:10:35 +11005956 while (size > conf->max_nr_stripes)
5957 if (!grow_one_stripe(conf, GFP_KERNEL))
5958 break;
NeilBrown2d5b5692015-07-06 12:49:23 +10005959 mutex_unlock(&conf->cache_size_mutex);
NeilBrown486f0642015-02-25 12:10:35 +11005960
NeilBrownc41d4ac2010-06-01 19:37:24 +10005961 return 0;
5962}
5963EXPORT_SYMBOL(raid5_set_cache_size);
5964
NeilBrown3f294f42005-11-08 21:39:25 -08005965static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11005966raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08005967{
NeilBrown67918752014-12-15 12:57:01 +11005968 struct r5conf *conf;
Dan Williams4ef197d82008-04-28 02:15:54 -07005969 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07005970 int err;
5971
NeilBrown3f294f42005-11-08 21:39:25 -08005972 if (len >= PAGE_SIZE)
5973 return -EINVAL;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09005974 if (kstrtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08005975 return -EINVAL;
NeilBrown67918752014-12-15 12:57:01 +11005976 err = mddev_lock(mddev);
Dan Williamsb5470dc2008-06-27 21:44:04 -07005977 if (err)
5978 return err;
NeilBrown67918752014-12-15 12:57:01 +11005979 conf = mddev->private;
5980 if (!conf)
5981 err = -ENODEV;
5982 else
5983 err = raid5_set_cache_size(mddev, new);
5984 mddev_unlock(mddev);
5985
5986 return err ?: len;
NeilBrown3f294f42005-11-08 21:39:25 -08005987}
NeilBrown007583c2005-11-08 21:39:30 -08005988
NeilBrown96de1e62005-11-08 21:39:39 -08005989static struct md_sysfs_entry
5990raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
5991 raid5_show_stripe_cache_size,
5992 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08005993
5994static ssize_t
Markus Stockhausend06f1912014-12-15 12:57:05 +11005995raid5_show_rmw_level(struct mddev *mddev, char *page)
5996{
5997 struct r5conf *conf = mddev->private;
5998 if (conf)
5999 return sprintf(page, "%d\n", conf->rmw_level);
6000 else
6001 return 0;
6002}
6003
6004static ssize_t
6005raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len)
6006{
6007 struct r5conf *conf = mddev->private;
6008 unsigned long new;
6009
6010 if (!conf)
6011 return -ENODEV;
6012
6013 if (len >= PAGE_SIZE)
6014 return -EINVAL;
6015
6016 if (kstrtoul(page, 10, &new))
6017 return -EINVAL;
6018
6019 if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome)
6020 return -EINVAL;
6021
6022 if (new != PARITY_DISABLE_RMW &&
6023 new != PARITY_ENABLE_RMW &&
6024 new != PARITY_PREFER_RMW)
6025 return -EINVAL;
6026
6027 conf->rmw_level = new;
6028 return len;
6029}
6030
6031static struct md_sysfs_entry
6032raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR,
6033 raid5_show_rmw_level,
6034 raid5_store_rmw_level);
6035
6036
6037static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11006038raid5_show_preread_threshold(struct mddev *mddev, char *page)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006039{
NeilBrown7b1485b2014-12-15 12:56:59 +11006040 struct r5conf *conf;
6041 int ret = 0;
6042 spin_lock(&mddev->lock);
6043 conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006044 if (conf)
NeilBrown7b1485b2014-12-15 12:56:59 +11006045 ret = sprintf(page, "%d\n", conf->bypass_threshold);
6046 spin_unlock(&mddev->lock);
6047 return ret;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006048}
6049
6050static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11006051raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006052{
NeilBrown67918752014-12-15 12:57:01 +11006053 struct r5conf *conf;
Dan Williams4ef197d82008-04-28 02:15:54 -07006054 unsigned long new;
NeilBrown67918752014-12-15 12:57:01 +11006055 int err;
6056
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006057 if (len >= PAGE_SIZE)
6058 return -EINVAL;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09006059 if (kstrtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006060 return -EINVAL;
NeilBrown67918752014-12-15 12:57:01 +11006061
6062 err = mddev_lock(mddev);
6063 if (err)
6064 return err;
6065 conf = mddev->private;
6066 if (!conf)
6067 err = -ENODEV;
NeilBrownedbe83a2015-02-26 12:47:56 +11006068 else if (new > conf->min_nr_stripes)
NeilBrown67918752014-12-15 12:57:01 +11006069 err = -EINVAL;
6070 else
6071 conf->bypass_threshold = new;
6072 mddev_unlock(mddev);
6073 return err ?: len;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006074}
6075
6076static struct md_sysfs_entry
6077raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
6078 S_IRUGO | S_IWUSR,
6079 raid5_show_preread_threshold,
6080 raid5_store_preread_threshold);
6081
6082static ssize_t
Shaohua Lid592a992014-05-21 17:57:44 +08006083raid5_show_skip_copy(struct mddev *mddev, char *page)
6084{
NeilBrown7b1485b2014-12-15 12:56:59 +11006085 struct r5conf *conf;
6086 int ret = 0;
6087 spin_lock(&mddev->lock);
6088 conf = mddev->private;
Shaohua Lid592a992014-05-21 17:57:44 +08006089 if (conf)
NeilBrown7b1485b2014-12-15 12:56:59 +11006090 ret = sprintf(page, "%d\n", conf->skip_copy);
6091 spin_unlock(&mddev->lock);
6092 return ret;
Shaohua Lid592a992014-05-21 17:57:44 +08006093}
6094
6095static ssize_t
6096raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len)
6097{
NeilBrown67918752014-12-15 12:57:01 +11006098 struct r5conf *conf;
Shaohua Lid592a992014-05-21 17:57:44 +08006099 unsigned long new;
NeilBrown67918752014-12-15 12:57:01 +11006100 int err;
6101
Shaohua Lid592a992014-05-21 17:57:44 +08006102 if (len >= PAGE_SIZE)
6103 return -EINVAL;
Shaohua Lid592a992014-05-21 17:57:44 +08006104 if (kstrtoul(page, 10, &new))
6105 return -EINVAL;
6106 new = !!new;
Shaohua Lid592a992014-05-21 17:57:44 +08006107
NeilBrown67918752014-12-15 12:57:01 +11006108 err = mddev_lock(mddev);
6109 if (err)
6110 return err;
6111 conf = mddev->private;
6112 if (!conf)
6113 err = -ENODEV;
6114 else if (new != conf->skip_copy) {
6115 mddev_suspend(mddev);
6116 conf->skip_copy = new;
6117 if (new)
6118 mddev->queue->backing_dev_info.capabilities |=
6119 BDI_CAP_STABLE_WRITES;
6120 else
6121 mddev->queue->backing_dev_info.capabilities &=
6122 ~BDI_CAP_STABLE_WRITES;
6123 mddev_resume(mddev);
6124 }
6125 mddev_unlock(mddev);
6126 return err ?: len;
Shaohua Lid592a992014-05-21 17:57:44 +08006127}
6128
6129static struct md_sysfs_entry
6130raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR,
6131 raid5_show_skip_copy,
6132 raid5_store_skip_copy);
6133
Shaohua Lid592a992014-05-21 17:57:44 +08006134static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11006135stripe_cache_active_show(struct mddev *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08006136{
NeilBrownd1688a62011-10-11 16:49:52 +11006137 struct r5conf *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08006138 if (conf)
6139 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
6140 else
6141 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08006142}
6143
NeilBrown96de1e62005-11-08 21:39:39 -08006144static struct md_sysfs_entry
6145raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08006146
Shaohua Lib7214202013-08-27 17:50:42 +08006147static ssize_t
6148raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
6149{
NeilBrown7b1485b2014-12-15 12:56:59 +11006150 struct r5conf *conf;
6151 int ret = 0;
6152 spin_lock(&mddev->lock);
6153 conf = mddev->private;
Shaohua Lib7214202013-08-27 17:50:42 +08006154 if (conf)
NeilBrown7b1485b2014-12-15 12:56:59 +11006155 ret = sprintf(page, "%d\n", conf->worker_cnt_per_group);
6156 spin_unlock(&mddev->lock);
6157 return ret;
Shaohua Lib7214202013-08-27 17:50:42 +08006158}
6159
majianpeng60aaf932013-11-14 15:16:20 +11006160static int alloc_thread_groups(struct r5conf *conf, int cnt,
6161 int *group_cnt,
6162 int *worker_cnt_per_group,
6163 struct r5worker_group **worker_groups);
Shaohua Lib7214202013-08-27 17:50:42 +08006164static ssize_t
6165raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
6166{
NeilBrown67918752014-12-15 12:57:01 +11006167 struct r5conf *conf;
Shaohua Lib7214202013-08-27 17:50:42 +08006168 unsigned long new;
6169 int err;
majianpeng60aaf932013-11-14 15:16:20 +11006170 struct r5worker_group *new_groups, *old_groups;
6171 int group_cnt, worker_cnt_per_group;
Shaohua Lib7214202013-08-27 17:50:42 +08006172
6173 if (len >= PAGE_SIZE)
6174 return -EINVAL;
Shaohua Lib7214202013-08-27 17:50:42 +08006175 if (kstrtoul(page, 10, &new))
6176 return -EINVAL;
6177
NeilBrown67918752014-12-15 12:57:01 +11006178 err = mddev_lock(mddev);
Shaohua Lib7214202013-08-27 17:50:42 +08006179 if (err)
6180 return err;
NeilBrown67918752014-12-15 12:57:01 +11006181 conf = mddev->private;
6182 if (!conf)
6183 err = -ENODEV;
6184 else if (new != conf->worker_cnt_per_group) {
6185 mddev_suspend(mddev);
6186
6187 old_groups = conf->worker_groups;
6188 if (old_groups)
6189 flush_workqueue(raid5_wq);
6190
6191 err = alloc_thread_groups(conf, new,
6192 &group_cnt, &worker_cnt_per_group,
6193 &new_groups);
6194 if (!err) {
6195 spin_lock_irq(&conf->device_lock);
6196 conf->group_cnt = group_cnt;
6197 conf->worker_cnt_per_group = worker_cnt_per_group;
6198 conf->worker_groups = new_groups;
6199 spin_unlock_irq(&conf->device_lock);
6200
6201 if (old_groups)
6202 kfree(old_groups[0].workers);
6203 kfree(old_groups);
6204 }
6205 mddev_resume(mddev);
6206 }
6207 mddev_unlock(mddev);
6208
6209 return err ?: len;
Shaohua Lib7214202013-08-27 17:50:42 +08006210}
6211
6212static struct md_sysfs_entry
6213raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
6214 raid5_show_group_thread_cnt,
6215 raid5_store_group_thread_cnt);
6216
NeilBrown007583c2005-11-08 21:39:30 -08006217static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08006218 &raid5_stripecache_size.attr,
6219 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07006220 &raid5_preread_bypass_threshold.attr,
Shaohua Lib7214202013-08-27 17:50:42 +08006221 &raid5_group_thread_cnt.attr,
Shaohua Lid592a992014-05-21 17:57:44 +08006222 &raid5_skip_copy.attr,
Markus Stockhausend06f1912014-12-15 12:57:05 +11006223 &raid5_rmw_level.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08006224 NULL,
6225};
NeilBrown007583c2005-11-08 21:39:30 -08006226static struct attribute_group raid5_attrs_group = {
6227 .name = NULL,
6228 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08006229};
6230
majianpeng60aaf932013-11-14 15:16:20 +11006231static int alloc_thread_groups(struct r5conf *conf, int cnt,
6232 int *group_cnt,
6233 int *worker_cnt_per_group,
6234 struct r5worker_group **worker_groups)
Shaohua Li851c30c2013-08-28 14:30:16 +08006235{
Shaohua Li566c09c2013-11-14 15:16:17 +11006236 int i, j, k;
Shaohua Li851c30c2013-08-28 14:30:16 +08006237 ssize_t size;
6238 struct r5worker *workers;
6239
majianpeng60aaf932013-11-14 15:16:20 +11006240 *worker_cnt_per_group = cnt;
Shaohua Li851c30c2013-08-28 14:30:16 +08006241 if (cnt == 0) {
majianpeng60aaf932013-11-14 15:16:20 +11006242 *group_cnt = 0;
6243 *worker_groups = NULL;
Shaohua Li851c30c2013-08-28 14:30:16 +08006244 return 0;
6245 }
majianpeng60aaf932013-11-14 15:16:20 +11006246 *group_cnt = num_possible_nodes();
Shaohua Li851c30c2013-08-28 14:30:16 +08006247 size = sizeof(struct r5worker) * cnt;
majianpeng60aaf932013-11-14 15:16:20 +11006248 workers = kzalloc(size * *group_cnt, GFP_NOIO);
6249 *worker_groups = kzalloc(sizeof(struct r5worker_group) *
6250 *group_cnt, GFP_NOIO);
6251 if (!*worker_groups || !workers) {
Shaohua Li851c30c2013-08-28 14:30:16 +08006252 kfree(workers);
majianpeng60aaf932013-11-14 15:16:20 +11006253 kfree(*worker_groups);
Shaohua Li851c30c2013-08-28 14:30:16 +08006254 return -ENOMEM;
6255 }
6256
majianpeng60aaf932013-11-14 15:16:20 +11006257 for (i = 0; i < *group_cnt; i++) {
Shaohua Li851c30c2013-08-28 14:30:16 +08006258 struct r5worker_group *group;
6259
NeilBrown0c775d52013-11-25 11:12:43 +11006260 group = &(*worker_groups)[i];
Shaohua Li851c30c2013-08-28 14:30:16 +08006261 INIT_LIST_HEAD(&group->handle_list);
6262 group->conf = conf;
6263 group->workers = workers + i * cnt;
6264
6265 for (j = 0; j < cnt; j++) {
Shaohua Li566c09c2013-11-14 15:16:17 +11006266 struct r5worker *worker = group->workers + j;
6267 worker->group = group;
6268 INIT_WORK(&worker->work, raid5_do_work);
6269
6270 for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
6271 INIT_LIST_HEAD(worker->temp_inactive_list + k);
Shaohua Li851c30c2013-08-28 14:30:16 +08006272 }
6273 }
6274
6275 return 0;
6276}
6277
6278static void free_thread_groups(struct r5conf *conf)
6279{
6280 if (conf->worker_groups)
6281 kfree(conf->worker_groups[0].workers);
6282 kfree(conf->worker_groups);
6283 conf->worker_groups = NULL;
6284}
6285
Dan Williams80c3a6c2009-03-17 18:10:40 -07006286static sector_t
NeilBrownfd01b882011-10-11 16:47:53 +11006287raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07006288{
NeilBrownd1688a62011-10-11 16:49:52 +11006289 struct r5conf *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07006290
6291 if (!sectors)
6292 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11006293 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11006294 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11006295 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07006296
NeilBrown3cb5edf2015-07-15 17:24:17 +10006297 sectors &= ~((sector_t)conf->chunk_sectors - 1);
6298 sectors &= ~((sector_t)conf->prev_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07006299 return sectors * (raid_disks - conf->max_degraded);
6300}
6301
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306302static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6303{
6304 safe_put_page(percpu->spare_page);
shli@kernel.org46d5b782014-12-15 12:57:02 +11006305 if (percpu->scribble)
6306 flex_array_free(percpu->scribble);
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306307 percpu->spare_page = NULL;
6308 percpu->scribble = NULL;
6309}
6310
6311static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu)
6312{
6313 if (conf->level == 6 && !percpu->spare_page)
6314 percpu->spare_page = alloc_page(GFP_KERNEL);
6315 if (!percpu->scribble)
shli@kernel.org46d5b782014-12-15 12:57:02 +11006316 percpu->scribble = scribble_alloc(max(conf->raid_disks,
NeilBrown738a2732015-05-08 18:19:39 +10006317 conf->previous_raid_disks),
6318 max(conf->chunk_sectors,
6319 conf->prev_chunk_sectors)
6320 / STRIPE_SECTORS,
6321 GFP_KERNEL);
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306322
6323 if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) {
6324 free_scratch_buffer(conf, percpu);
6325 return -ENOMEM;
6326 }
6327
6328 return 0;
6329}
6330
NeilBrownd1688a62011-10-11 16:49:52 +11006331static void raid5_free_percpu(struct r5conf *conf)
Dan Williams36d1c642009-07-14 11:48:22 -07006332{
Dan Williams36d1c642009-07-14 11:48:22 -07006333 unsigned long cpu;
6334
6335 if (!conf->percpu)
6336 return;
6337
Dan Williams36d1c642009-07-14 11:48:22 -07006338#ifdef CONFIG_HOTPLUG_CPU
6339 unregister_cpu_notifier(&conf->cpu_notify);
6340#endif
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306341
6342 get_online_cpus();
6343 for_each_possible_cpu(cpu)
6344 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
Dan Williams36d1c642009-07-14 11:48:22 -07006345 put_online_cpus();
6346
6347 free_percpu(conf->percpu);
6348}
6349
NeilBrownd1688a62011-10-11 16:49:52 +11006350static void free_conf(struct r5conf *conf)
Dan Williams95fc17a2009-07-31 12:39:15 +10006351{
Shaohua Li5c7e81c2015-08-13 14:32:04 -07006352 if (conf->log)
6353 r5l_exit_log(conf->log);
NeilBrownedbe83a2015-02-26 12:47:56 +11006354 if (conf->shrinker.seeks)
6355 unregister_shrinker(&conf->shrinker);
Shaohua Li5c7e81c2015-08-13 14:32:04 -07006356
Shaohua Li851c30c2013-08-28 14:30:16 +08006357 free_thread_groups(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10006358 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07006359 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10006360 kfree(conf->disks);
6361 kfree(conf->stripe_hashtbl);
6362 kfree(conf);
6363}
6364
Dan Williams36d1c642009-07-14 11:48:22 -07006365#ifdef CONFIG_HOTPLUG_CPU
6366static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
6367 void *hcpu)
6368{
NeilBrownd1688a62011-10-11 16:49:52 +11006369 struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify);
Dan Williams36d1c642009-07-14 11:48:22 -07006370 long cpu = (long)hcpu;
6371 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
6372
6373 switch (action) {
6374 case CPU_UP_PREPARE:
6375 case CPU_UP_PREPARE_FROZEN:
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306376 if (alloc_scratch_buffer(conf, percpu)) {
Dan Williams36d1c642009-07-14 11:48:22 -07006377 pr_err("%s: failed memory allocation for cpu%ld\n",
6378 __func__, cpu);
Akinobu Mita55af6bb2010-05-26 14:43:35 -07006379 return notifier_from_errno(-ENOMEM);
Dan Williams36d1c642009-07-14 11:48:22 -07006380 }
6381 break;
6382 case CPU_DEAD:
6383 case CPU_DEAD_FROZEN:
Anna-Maria Gleixner1d034e62016-03-16 09:25:23 +01006384 case CPU_UP_CANCELED:
6385 case CPU_UP_CANCELED_FROZEN:
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306386 free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
Dan Williams36d1c642009-07-14 11:48:22 -07006387 break;
6388 default:
6389 break;
6390 }
6391 return NOTIFY_OK;
6392}
6393#endif
6394
NeilBrownd1688a62011-10-11 16:49:52 +11006395static int raid5_alloc_percpu(struct r5conf *conf)
Dan Williams36d1c642009-07-14 11:48:22 -07006396{
6397 unsigned long cpu;
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306398 int err = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07006399
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306400 conf->percpu = alloc_percpu(struct raid5_percpu);
6401 if (!conf->percpu)
Dan Williams36d1c642009-07-14 11:48:22 -07006402 return -ENOMEM;
Dan Williams36d1c642009-07-14 11:48:22 -07006403
Dan Williams36d1c642009-07-14 11:48:22 -07006404#ifdef CONFIG_HOTPLUG_CPU
6405 conf->cpu_notify.notifier_call = raid456_cpu_notify;
6406 conf->cpu_notify.priority = 0;
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306407 err = register_cpu_notifier(&conf->cpu_notify);
6408 if (err)
6409 return err;
Dan Williams36d1c642009-07-14 11:48:22 -07006410#endif
Oleg Nesterov789b5e02014-02-06 03:42:45 +05306411
6412 get_online_cpus();
6413 for_each_present_cpu(cpu) {
6414 err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu));
6415 if (err) {
6416 pr_err("%s: failed memory allocation for cpu%ld\n",
6417 __func__, cpu);
6418 break;
6419 }
6420 }
Dan Williams36d1c642009-07-14 11:48:22 -07006421 put_online_cpus();
6422
Shaohua Li27a353c2016-02-24 17:38:28 -08006423 if (!err) {
6424 conf->scribble_disks = max(conf->raid_disks,
6425 conf->previous_raid_disks);
6426 conf->scribble_sectors = max(conf->chunk_sectors,
6427 conf->prev_chunk_sectors);
6428 }
Dan Williams36d1c642009-07-14 11:48:22 -07006429 return err;
6430}
6431
NeilBrownedbe83a2015-02-26 12:47:56 +11006432static unsigned long raid5_cache_scan(struct shrinker *shrink,
6433 struct shrink_control *sc)
6434{
6435 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
NeilBrown2d5b5692015-07-06 12:49:23 +10006436 unsigned long ret = SHRINK_STOP;
6437
6438 if (mutex_trylock(&conf->cache_size_mutex)) {
6439 ret= 0;
NeilBrown49895bc2015-08-03 17:09:57 +10006440 while (ret < sc->nr_to_scan &&
6441 conf->max_nr_stripes > conf->min_nr_stripes) {
NeilBrown2d5b5692015-07-06 12:49:23 +10006442 if (drop_one_stripe(conf) == 0) {
6443 ret = SHRINK_STOP;
6444 break;
6445 }
6446 ret++;
6447 }
6448 mutex_unlock(&conf->cache_size_mutex);
NeilBrownedbe83a2015-02-26 12:47:56 +11006449 }
6450 return ret;
6451}
6452
6453static unsigned long raid5_cache_count(struct shrinker *shrink,
6454 struct shrink_control *sc)
6455{
6456 struct r5conf *conf = container_of(shrink, struct r5conf, shrinker);
6457
6458 if (conf->max_nr_stripes < conf->min_nr_stripes)
6459 /* unlikely, but not impossible */
6460 return 0;
6461 return conf->max_nr_stripes - conf->min_nr_stripes;
6462}
6463
NeilBrownd1688a62011-10-11 16:49:52 +11006464static struct r5conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006465{
NeilBrownd1688a62011-10-11 16:49:52 +11006466 struct r5conf *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11006467 int raid_disk, memory, max_disks;
NeilBrown3cb03002011-10-11 16:45:26 +11006468 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006469 struct disk_info *disk;
NeilBrown02326052012-07-03 15:56:52 +10006470 char pers_name[6];
Shaohua Li566c09c2013-11-14 15:16:17 +11006471 int i;
majianpeng60aaf932013-11-14 15:16:20 +11006472 int group_cnt, worker_cnt_per_group;
6473 struct r5worker_group *new_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006474
NeilBrown91adb562009-03-31 14:39:39 +11006475 if (mddev->new_level != 5
6476 && mddev->new_level != 4
6477 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10006478 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11006479 mdname(mddev), mddev->new_level);
6480 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006481 }
NeilBrown91adb562009-03-31 14:39:39 +11006482 if ((mddev->new_level == 5
6483 && !algorithm_valid_raid5(mddev->new_layout)) ||
6484 (mddev->new_level == 6
6485 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10006486 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11006487 mdname(mddev), mddev->new_layout);
6488 return ERR_PTR(-EIO);
6489 }
6490 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10006491 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11006492 mdname(mddev), mddev->raid_disks);
6493 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11006494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006495
Andre Noll664e7c42009-06-18 08:45:27 +10006496 if (!mddev->new_chunk_sectors ||
6497 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
6498 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10006499 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
6500 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11006501 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11006502 }
6503
NeilBrownd1688a62011-10-11 16:49:52 +11006504 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
NeilBrown91adb562009-03-31 14:39:39 +11006505 if (conf == NULL)
6506 goto abort;
Shaohua Li851c30c2013-08-28 14:30:16 +08006507 /* Don't enable multi-threading by default*/
majianpeng60aaf932013-11-14 15:16:20 +11006508 if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
6509 &new_group)) {
6510 conf->group_cnt = group_cnt;
6511 conf->worker_cnt_per_group = worker_cnt_per_group;
6512 conf->worker_groups = new_group;
6513 } else
Shaohua Li851c30c2013-08-28 14:30:16 +08006514 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11006515 spin_lock_init(&conf->device_lock);
NeilBrownc46501b2013-08-27 15:52:13 +10006516 seqcount_init(&conf->gen_lock);
NeilBrown2d5b5692015-07-06 12:49:23 +10006517 mutex_init(&conf->cache_size_mutex);
Yuanhan Liub1b46482015-05-08 18:19:06 +10006518 init_waitqueue_head(&conf->wait_for_quiescent);
Shaohua Li6ab2a4b2016-02-25 16:24:42 -08006519 init_waitqueue_head(&conf->wait_for_stripe);
Dan Williamsf5efd452009-10-16 15:55:38 +11006520 init_waitqueue_head(&conf->wait_for_overlap);
6521 INIT_LIST_HEAD(&conf->handle_list);
6522 INIT_LIST_HEAD(&conf->hold_list);
6523 INIT_LIST_HEAD(&conf->delayed_list);
6524 INIT_LIST_HEAD(&conf->bitmap_list);
NeilBrownc3cce6c2015-08-14 12:47:33 +10006525 bio_list_init(&conf->return_bi);
Shaohua Li773ca822013-08-27 17:50:39 +08006526 init_llist_head(&conf->released_stripes);
Dan Williamsf5efd452009-10-16 15:55:38 +11006527 atomic_set(&conf->active_stripes, 0);
6528 atomic_set(&conf->preread_active_stripes, 0);
6529 atomic_set(&conf->active_aligned_reads, 0);
6530 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrownd890fa22011-10-26 11:54:39 +11006531 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown91adb562009-03-31 14:39:39 +11006532
6533 conf->raid_disks = mddev->raid_disks;
6534 if (mddev->reshape_position == MaxSector)
6535 conf->previous_raid_disks = mddev->raid_disks;
6536 else
6537 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11006538 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
NeilBrown91adb562009-03-31 14:39:39 +11006539
NeilBrown5e5e3e72009-10-16 16:35:30 +11006540 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11006541 GFP_KERNEL);
6542 if (!conf->disks)
6543 goto abort;
6544
6545 conf->mddev = mddev;
6546
6547 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
6548 goto abort;
6549
Shaohua Li566c09c2013-11-14 15:16:17 +11006550 /* We init hash_locks[0] separately to that it can be used
6551 * as the reference lock in the spin_lock_nest_lock() call
6552 * in lock_all_device_hash_locks_irq in order to convince
6553 * lockdep that we know what we are doing.
6554 */
6555 spin_lock_init(conf->hash_locks);
6556 for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
6557 spin_lock_init(conf->hash_locks + i);
6558
6559 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6560 INIT_LIST_HEAD(conf->inactive_list + i);
6561
6562 for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
6563 INIT_LIST_HEAD(conf->temp_inactive_list + i);
6564
Dan Williams36d1c642009-07-14 11:48:22 -07006565 conf->level = mddev->new_level;
shli@kernel.org46d5b782014-12-15 12:57:02 +11006566 conf->chunk_sectors = mddev->new_chunk_sectors;
Dan Williams36d1c642009-07-14 11:48:22 -07006567 if (raid5_alloc_percpu(conf) != 0)
6568 goto abort;
6569
NeilBrown0c55e022010-05-03 14:09:02 +10006570 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11006571
NeilBrowndafb20f2012-03-19 12:46:39 +11006572 rdev_for_each(rdev, mddev) {
NeilBrown91adb562009-03-31 14:39:39 +11006573 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11006574 if (raid_disk >= max_disks
Shaohua Lif2076e72015-10-08 21:54:12 -07006575 || raid_disk < 0 || test_bit(Journal, &rdev->flags))
NeilBrown91adb562009-03-31 14:39:39 +11006576 continue;
6577 disk = conf->disks + raid_disk;
6578
NeilBrown17045f52011-12-23 10:17:53 +11006579 if (test_bit(Replacement, &rdev->flags)) {
6580 if (disk->replacement)
6581 goto abort;
6582 disk->replacement = rdev;
6583 } else {
6584 if (disk->rdev)
6585 goto abort;
6586 disk->rdev = rdev;
6587 }
NeilBrown91adb562009-03-31 14:39:39 +11006588
6589 if (test_bit(In_sync, &rdev->flags)) {
6590 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10006591 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
6592 " disk %d\n",
6593 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
Jonathan Brassowd6b212f2011-06-08 18:00:28 -05006594 } else if (rdev->saved_raid_disk != raid_disk)
NeilBrown91adb562009-03-31 14:39:39 +11006595 /* Cannot rely on bitmap to complete recovery */
6596 conf->fullsync = 1;
6597 }
6598
NeilBrown91adb562009-03-31 14:39:39 +11006599 conf->level = mddev->new_level;
Markus Stockhausen584acdd2014-12-15 12:57:05 +11006600 if (conf->level == 6) {
NeilBrown91adb562009-03-31 14:39:39 +11006601 conf->max_degraded = 2;
Markus Stockhausen584acdd2014-12-15 12:57:05 +11006602 if (raid6_call.xor_syndrome)
6603 conf->rmw_level = PARITY_ENABLE_RMW;
6604 else
6605 conf->rmw_level = PARITY_DISABLE_RMW;
6606 } else {
NeilBrown91adb562009-03-31 14:39:39 +11006607 conf->max_degraded = 1;
Markus Stockhausen584acdd2014-12-15 12:57:05 +11006608 conf->rmw_level = PARITY_ENABLE_RMW;
6609 }
NeilBrown91adb562009-03-31 14:39:39 +11006610 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11006611 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11006612 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10006613 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11006614 conf->prev_algo = mddev->layout;
NeilBrown5cac6bc2015-07-17 12:17:50 +10006615 } else {
6616 conf->prev_chunk_sectors = conf->chunk_sectors;
6617 conf->prev_algo = conf->algorithm;
NeilBrowne183eae2009-03-31 15:20:22 +11006618 }
NeilBrown91adb562009-03-31 14:39:39 +11006619
NeilBrownedbe83a2015-02-26 12:47:56 +11006620 conf->min_nr_stripes = NR_STRIPES;
6621 memory = conf->min_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11006622 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
Shaohua Li4bda5562013-11-14 15:16:17 +11006623 atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
NeilBrownedbe83a2015-02-26 12:47:56 +11006624 if (grow_stripes(conf, conf->min_nr_stripes)) {
NeilBrown91adb562009-03-31 14:39:39 +11006625 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10006626 "md/raid:%s: couldn't allocate %dkB for buffers\n",
6627 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11006628 goto abort;
6629 } else
NeilBrown0c55e022010-05-03 14:09:02 +10006630 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
6631 mdname(mddev), memory);
NeilBrownedbe83a2015-02-26 12:47:56 +11006632 /*
6633 * Losing a stripe head costs more than the time to refill it,
6634 * it reduces the queue depth and so can hurt throughput.
6635 * So set it rather large, scaled by number of devices.
6636 */
6637 conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4;
6638 conf->shrinker.scan_objects = raid5_cache_scan;
6639 conf->shrinker.count_objects = raid5_cache_count;
6640 conf->shrinker.batch = 128;
6641 conf->shrinker.flags = 0;
6642 register_shrinker(&conf->shrinker);
NeilBrown91adb562009-03-31 14:39:39 +11006643
NeilBrown02326052012-07-03 15:56:52 +10006644 sprintf(pers_name, "raid%d", mddev->new_level);
6645 conf->thread = md_register_thread(raid5d, mddev, pers_name);
NeilBrown91adb562009-03-31 14:39:39 +11006646 if (!conf->thread) {
6647 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10006648 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11006649 mdname(mddev));
6650 goto abort;
6651 }
6652
6653 return conf;
6654
6655 abort:
6656 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10006657 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11006658 return ERR_PTR(-EIO);
6659 } else
6660 return ERR_PTR(-ENOMEM);
6661}
6662
NeilBrownc148ffd2009-11-13 17:47:00 +11006663static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
6664{
6665 switch (algo) {
6666 case ALGORITHM_PARITY_0:
6667 if (raid_disk < max_degraded)
6668 return 1;
6669 break;
6670 case ALGORITHM_PARITY_N:
6671 if (raid_disk >= raid_disks - max_degraded)
6672 return 1;
6673 break;
6674 case ALGORITHM_PARITY_0_6:
NeilBrownf72ffdd2014-09-30 14:23:59 +10006675 if (raid_disk == 0 ||
NeilBrownc148ffd2009-11-13 17:47:00 +11006676 raid_disk == raid_disks - 1)
6677 return 1;
6678 break;
6679 case ALGORITHM_LEFT_ASYMMETRIC_6:
6680 case ALGORITHM_RIGHT_ASYMMETRIC_6:
6681 case ALGORITHM_LEFT_SYMMETRIC_6:
6682 case ALGORITHM_RIGHT_SYMMETRIC_6:
6683 if (raid_disk == raid_disks - 1)
6684 return 1;
6685 }
6686 return 0;
6687}
6688
Shaohua Li849674e2016-01-20 13:52:20 -08006689static int raid5_run(struct mddev *mddev)
NeilBrown91adb562009-03-31 14:39:39 +11006690{
NeilBrownd1688a62011-10-11 16:49:52 +11006691 struct r5conf *conf;
NeilBrown9f7c2222010-07-26 12:04:13 +10006692 int working_disks = 0;
NeilBrownc148ffd2009-11-13 17:47:00 +11006693 int dirty_parity_disks = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11006694 struct md_rdev *rdev;
Shaohua Li713cf5a2015-08-13 14:32:03 -07006695 struct md_rdev *journal_dev = NULL;
NeilBrownc148ffd2009-11-13 17:47:00 +11006696 sector_t reshape_offset = 0;
NeilBrown17045f52011-12-23 10:17:53 +11006697 int i;
NeilBrownb5254dd2012-05-21 09:27:01 +10006698 long long min_offset_diff = 0;
6699 int first = 1;
NeilBrown91adb562009-03-31 14:39:39 +11006700
Andre Noll8c6ac862009-06-18 08:48:06 +10006701 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10006702 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10006703 " -- starting background reconstruction\n",
6704 mdname(mddev));
NeilBrownb5254dd2012-05-21 09:27:01 +10006705
6706 rdev_for_each(rdev, mddev) {
6707 long long diff;
Shaohua Li713cf5a2015-08-13 14:32:03 -07006708
Shaohua Lif2076e72015-10-08 21:54:12 -07006709 if (test_bit(Journal, &rdev->flags)) {
Shaohua Li713cf5a2015-08-13 14:32:03 -07006710 journal_dev = rdev;
Shaohua Lif2076e72015-10-08 21:54:12 -07006711 continue;
6712 }
NeilBrownb5254dd2012-05-21 09:27:01 +10006713 if (rdev->raid_disk < 0)
6714 continue;
6715 diff = (rdev->new_data_offset - rdev->data_offset);
6716 if (first) {
6717 min_offset_diff = diff;
6718 first = 0;
6719 } else if (mddev->reshape_backwards &&
6720 diff < min_offset_diff)
6721 min_offset_diff = diff;
6722 else if (!mddev->reshape_backwards &&
6723 diff > min_offset_diff)
6724 min_offset_diff = diff;
6725 }
6726
NeilBrownf6705572006-03-27 01:18:11 -08006727 if (mddev->reshape_position != MaxSector) {
6728 /* Check that we can continue the reshape.
NeilBrownb5254dd2012-05-21 09:27:01 +10006729 * Difficulties arise if the stripe we would write to
6730 * next is at or after the stripe we would read from next.
6731 * For a reshape that changes the number of devices, this
6732 * is only possible for a very short time, and mdadm makes
6733 * sure that time appears to have past before assembling
6734 * the array. So we fail if that time hasn't passed.
6735 * For a reshape that keeps the number of devices the same
6736 * mdadm must be monitoring the reshape can keeping the
6737 * critical areas read-only and backed up. It will start
6738 * the array in read-only mode, so we check for that.
NeilBrownf6705572006-03-27 01:18:11 -08006739 */
6740 sector_t here_new, here_old;
6741 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11006742 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrown05256d92015-07-15 17:36:21 +10006743 int chunk_sectors;
6744 int new_data_disks;
NeilBrownf6705572006-03-27 01:18:11 -08006745
Shaohua Li713cf5a2015-08-13 14:32:03 -07006746 if (journal_dev) {
6747 printk(KERN_ERR "md/raid:%s: don't support reshape with journal - aborting.\n",
6748 mdname(mddev));
6749 return -EINVAL;
6750 }
6751
NeilBrown88ce4932009-03-31 15:24:23 +11006752 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10006753 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08006754 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08006755 mdname(mddev));
6756 return -EINVAL;
6757 }
NeilBrownf6705572006-03-27 01:18:11 -08006758 old_disks = mddev->raid_disks - mddev->delta_disks;
6759 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08006760 * further up in new geometry must map after here in old
6761 * geometry.
NeilBrown05256d92015-07-15 17:36:21 +10006762 * If the chunk sizes are different, then as we perform reshape
6763 * in units of the largest of the two, reshape_position needs
6764 * be a multiple of the largest chunk size times new data disks.
NeilBrownf6705572006-03-27 01:18:11 -08006765 */
6766 here_new = mddev->reshape_position;
NeilBrown05256d92015-07-15 17:36:21 +10006767 chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors);
6768 new_data_disks = mddev->raid_disks - max_degraded;
6769 if (sector_div(here_new, chunk_sectors * new_data_disks)) {
NeilBrown0c55e022010-05-03 14:09:02 +10006770 printk(KERN_ERR "md/raid:%s: reshape_position not "
6771 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08006772 return -EINVAL;
6773 }
NeilBrown05256d92015-07-15 17:36:21 +10006774 reshape_offset = here_new * chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08006775 /* here_new is the stripe we will write to */
6776 here_old = mddev->reshape_position;
NeilBrown05256d92015-07-15 17:36:21 +10006777 sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
NeilBrownf4168852007-02-28 20:11:53 -08006778 /* here_old is the first stripe that we might need to read
6779 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10006780 if (mddev->delta_disks == 0) {
6781 /* We cannot be sure it is safe to start an in-place
NeilBrownb5254dd2012-05-21 09:27:01 +10006782 * reshape. It is only safe if user-space is monitoring
NeilBrown67ac6012009-08-13 10:06:24 +10006783 * and taking constant backups.
6784 * mdadm always starts a situation like this in
6785 * readonly mode so it can take control before
6786 * allowing any writes. So just check for that.
6787 */
NeilBrownb5254dd2012-05-21 09:27:01 +10006788 if (abs(min_offset_diff) >= mddev->chunk_sectors &&
6789 abs(min_offset_diff) >= mddev->new_chunk_sectors)
6790 /* not really in-place - so OK */;
6791 else if (mddev->ro == 0) {
6792 printk(KERN_ERR "md/raid:%s: in-place reshape "
6793 "must be started in read-only mode "
6794 "- aborting\n",
NeilBrown0c55e022010-05-03 14:09:02 +10006795 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10006796 return -EINVAL;
6797 }
NeilBrown2c810cd2012-05-21 09:27:00 +10006798 } else if (mddev->reshape_backwards
NeilBrown05256d92015-07-15 17:36:21 +10006799 ? (here_new * chunk_sectors + min_offset_diff <=
6800 here_old * chunk_sectors)
6801 : (here_new * chunk_sectors >=
6802 here_old * chunk_sectors + (-min_offset_diff))) {
NeilBrownf6705572006-03-27 01:18:11 -08006803 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10006804 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
6805 "auto-recovery - aborting.\n",
6806 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08006807 return -EINVAL;
6808 }
NeilBrown0c55e022010-05-03 14:09:02 +10006809 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
6810 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08006811 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08006812 } else {
NeilBrown91adb562009-03-31 14:39:39 +11006813 BUG_ON(mddev->level != mddev->new_level);
6814 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10006815 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11006816 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08006817 }
6818
NeilBrown245f46c2009-03-31 14:39:39 +11006819 if (mddev->private == NULL)
6820 conf = setup_conf(mddev);
6821 else
6822 conf = mddev->private;
6823
NeilBrown91adb562009-03-31 14:39:39 +11006824 if (IS_ERR(conf))
6825 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08006826
Shaohua Li7dde2ad2015-10-08 21:54:10 -07006827 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !journal_dev) {
6828 printk(KERN_ERR "md/raid:%s: journal disk is missing, force array readonly\n",
6829 mdname(mddev));
6830 mddev->ro = 1;
6831 set_disk_ro(mddev->gendisk, 1);
6832 }
6833
NeilBrownb5254dd2012-05-21 09:27:01 +10006834 conf->min_offset_diff = min_offset_diff;
NeilBrown91adb562009-03-31 14:39:39 +11006835 mddev->thread = conf->thread;
6836 conf->thread = NULL;
6837 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006838
NeilBrown17045f52011-12-23 10:17:53 +11006839 for (i = 0; i < conf->raid_disks && conf->previous_raid_disks;
6840 i++) {
6841 rdev = conf->disks[i].rdev;
6842 if (!rdev && conf->disks[i].replacement) {
6843 /* The replacement is all we have yet */
6844 rdev = conf->disks[i].replacement;
6845 conf->disks[i].replacement = NULL;
6846 clear_bit(Replacement, &rdev->flags);
6847 conf->disks[i].rdev = rdev;
6848 }
6849 if (!rdev)
NeilBrownc148ffd2009-11-13 17:47:00 +11006850 continue;
NeilBrown17045f52011-12-23 10:17:53 +11006851 if (conf->disks[i].replacement &&
6852 conf->reshape_progress != MaxSector) {
6853 /* replacements and reshape simply do not mix. */
6854 printk(KERN_ERR "md: cannot handle concurrent "
6855 "replacement and reshape.\n");
6856 goto abort;
6857 }
NeilBrown2f115882010-06-17 17:41:03 +10006858 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown91adb562009-03-31 14:39:39 +11006859 working_disks++;
NeilBrown2f115882010-06-17 17:41:03 +10006860 continue;
6861 }
NeilBrownc148ffd2009-11-13 17:47:00 +11006862 /* This disc is not fully in-sync. However if it
6863 * just stored parity (beyond the recovery_offset),
6864 * when we don't need to be concerned about the
6865 * array being dirty.
6866 * When reshape goes 'backwards', we never have
6867 * partially completed devices, so we only need
6868 * to worry about reshape going forwards.
6869 */
6870 /* Hack because v0.91 doesn't store recovery_offset properly. */
6871 if (mddev->major_version == 0 &&
6872 mddev->minor_version > 90)
6873 rdev->recovery_offset = reshape_offset;
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07006874
NeilBrownc148ffd2009-11-13 17:47:00 +11006875 if (rdev->recovery_offset < reshape_offset) {
6876 /* We need to check old and new layout */
6877 if (!only_parity(rdev->raid_disk,
6878 conf->algorithm,
6879 conf->raid_disks,
6880 conf->max_degraded))
6881 continue;
6882 }
6883 if (!only_parity(rdev->raid_disk,
6884 conf->prev_algo,
6885 conf->previous_raid_disks,
6886 conf->max_degraded))
6887 continue;
6888 dirty_parity_disks++;
6889 }
NeilBrown91adb562009-03-31 14:39:39 +11006890
NeilBrown17045f52011-12-23 10:17:53 +11006891 /*
6892 * 0 for a fully functional array, 1 or 2 for a degraded array.
6893 */
NeilBrown908f4fb2011-12-23 10:17:50 +11006894 mddev->degraded = calc_degraded(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006895
NeilBrown674806d2010-06-16 17:17:53 +10006896 if (has_failed(conf)) {
NeilBrown0c55e022010-05-03 14:09:02 +10006897 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07006898 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07006899 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006900 goto abort;
6901 }
6902
NeilBrown91adb562009-03-31 14:39:39 +11006903 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10006904 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11006905 mddev->resync_max_sectors = mddev->dev_sectors;
6906
NeilBrownc148ffd2009-11-13 17:47:00 +11006907 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07006908 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006909 if (mddev->ok_start_degraded)
6910 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10006911 "md/raid:%s: starting dirty degraded array"
6912 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006913 mdname(mddev));
6914 else {
6915 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10006916 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08006917 mdname(mddev));
6918 goto abort;
6919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006920 }
6921
Linus Torvalds1da177e2005-04-16 15:20:36 -07006922 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10006923 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
6924 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11006925 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
6926 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006927 else
NeilBrown0c55e022010-05-03 14:09:02 +10006928 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
6929 " out of %d devices, algorithm %d\n",
6930 mdname(mddev), conf->level,
6931 mddev->raid_disks - mddev->degraded,
6932 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006933
6934 print_raid5_conf(conf);
6935
NeilBrownfef9c612009-03-31 15:16:46 +11006936 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11006937 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08006938 atomic_set(&conf->reshape_stripes, 0);
6939 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6940 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6941 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6942 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6943 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10006944 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08006945 }
6946
Linus Torvalds1da177e2005-04-16 15:20:36 -07006947 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10006948 if (mddev->to_remove == &raid5_attrs_group)
6949 mddev->to_remove = NULL;
NeilBrown00bcb4a2010-06-01 19:37:23 +10006950 else if (mddev->kobj.sd &&
6951 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08006952 printk(KERN_WARNING
NeilBrown4a5add42010-06-01 19:37:28 +10006953 "raid5: failed to create sysfs attributes for %s\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08006954 mdname(mddev));
NeilBrown4a5add42010-06-01 19:37:28 +10006955 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
6956
6957 if (mddev->queue) {
NeilBrown9f7c2222010-07-26 12:04:13 +10006958 int chunk_size;
Shaohua Li620125f2012-10-11 13:49:05 +11006959 bool discard_supported = true;
NeilBrown4a5add42010-06-01 19:37:28 +10006960 /* read-ahead size must cover two whole stripes, which
6961 * is 2 * (datadisks) * chunksize where 'n' is the
6962 * number of raid devices
6963 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006964 int data_disks = conf->previous_raid_disks - conf->max_degraded;
6965 int stripe = data_disks *
6966 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
6967 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
6968 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
NeilBrown4a5add42010-06-01 19:37:28 +10006969
NeilBrown9f7c2222010-07-26 12:04:13 +10006970 chunk_size = mddev->chunk_sectors << 9;
6971 blk_queue_io_min(mddev->queue, chunk_size);
6972 blk_queue_io_opt(mddev->queue, chunk_size *
6973 (conf->raid_disks - conf->max_degraded));
Kent Overstreetc78afc62013-07-11 22:39:53 -07006974 mddev->queue->limits.raid_partial_stripes_expensive = 1;
Shaohua Li620125f2012-10-11 13:49:05 +11006975 /*
6976 * We can only discard a whole stripe. It doesn't make sense to
6977 * discard data disk but write parity disk
6978 */
6979 stripe = stripe * PAGE_SIZE;
NeilBrown4ac68752012-11-19 13:11:26 +11006980 /* Round up to power of 2, as discard handling
6981 * currently assumes that */
6982 while ((stripe-1) & stripe)
6983 stripe = (stripe | (stripe-1)) + 1;
Shaohua Li620125f2012-10-11 13:49:05 +11006984 mddev->queue->limits.discard_alignment = stripe;
6985 mddev->queue->limits.discard_granularity = stripe;
6986 /*
6987 * unaligned part of discard request will be ignored, so can't
NeilBrown8e0e99b2014-10-02 13:45:00 +10006988 * guarantee discard_zeroes_data
Shaohua Li620125f2012-10-11 13:49:05 +11006989 */
6990 mddev->queue->limits.discard_zeroes_data = 0;
NeilBrown9f7c2222010-07-26 12:04:13 +10006991
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07006992 blk_queue_max_write_same_sectors(mddev->queue, 0);
6993
NeilBrown05616be2012-05-21 09:27:00 +10006994 rdev_for_each(rdev, mddev) {
NeilBrown9f7c2222010-07-26 12:04:13 +10006995 disk_stack_limits(mddev->gendisk, rdev->bdev,
6996 rdev->data_offset << 9);
NeilBrown05616be2012-05-21 09:27:00 +10006997 disk_stack_limits(mddev->gendisk, rdev->bdev,
6998 rdev->new_data_offset << 9);
Shaohua Li620125f2012-10-11 13:49:05 +11006999 /*
7000 * discard_zeroes_data is required, otherwise data
7001 * could be lost. Consider a scenario: discard a stripe
7002 * (the stripe could be inconsistent if
7003 * discard_zeroes_data is 0); write one disk of the
7004 * stripe (the stripe could be inconsistent again
7005 * depending on which disks are used to calculate
7006 * parity); the disk is broken; The stripe data of this
7007 * disk is lost.
7008 */
7009 if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) ||
7010 !bdev_get_queue(rdev->bdev)->
7011 limits.discard_zeroes_data)
7012 discard_supported = false;
NeilBrown8e0e99b2014-10-02 13:45:00 +10007013 /* Unfortunately, discard_zeroes_data is not currently
7014 * a guarantee - just a hint. So we only allow DISCARD
7015 * if the sysadmin has confirmed that only safe devices
7016 * are in use by setting a module parameter.
7017 */
7018 if (!devices_handle_discard_safely) {
7019 if (discard_supported) {
7020 pr_info("md/raid456: discard support disabled due to uncertainty.\n");
7021 pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n");
7022 }
7023 discard_supported = false;
7024 }
NeilBrown05616be2012-05-21 09:27:00 +10007025 }
Shaohua Li620125f2012-10-11 13:49:05 +11007026
7027 if (discard_supported &&
Jes Sorensene7597e62016-02-16 16:44:24 -05007028 mddev->queue->limits.max_discard_sectors >= (stripe >> 9) &&
7029 mddev->queue->limits.discard_granularity >= stripe)
Shaohua Li620125f2012-10-11 13:49:05 +11007030 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
7031 mddev->queue);
7032 else
7033 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
7034 mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007035 }
7036
Shaohua Li5c7e81c2015-08-13 14:32:04 -07007037 if (journal_dev) {
7038 char b[BDEVNAME_SIZE];
7039
7040 printk(KERN_INFO"md/raid:%s: using device %s as journal\n",
7041 mdname(mddev), bdevname(journal_dev->bdev, b));
7042 r5l_init_log(conf, journal_dev);
7043 }
7044
Linus Torvalds1da177e2005-04-16 15:20:36 -07007045 return 0;
7046abort:
NeilBrown01f96c02011-09-21 15:30:20 +10007047 md_unregister_thread(&mddev->thread);
NeilBrowne4f869d2011-10-07 14:22:49 +11007048 print_raid5_conf(conf);
7049 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007050 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10007051 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007052 return -EIO;
7053}
7054
NeilBrownafa0f552014-12-15 12:56:58 +11007055static void raid5_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007056{
NeilBrownafa0f552014-12-15 12:56:58 +11007057 struct r5conf *conf = priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007058
Dan Williams95fc17a2009-07-31 12:39:15 +10007059 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10007060 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007061}
7062
Shaohua Li849674e2016-01-20 13:52:20 -08007063static void raid5_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007064{
NeilBrownd1688a62011-10-11 16:49:52 +11007065 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007066 int i;
7067
Andre Noll9d8f0362009-06-18 08:45:01 +10007068 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
NeilBrown3cb5edf2015-07-15 17:24:17 +10007069 conf->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07007070 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007071 for (i = 0; i < conf->raid_disks; i++)
7072 seq_printf (seq, "%s",
7073 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08007074 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007075 seq_printf (seq, "]");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007076}
7077
NeilBrownd1688a62011-10-11 16:49:52 +11007078static void print_raid5_conf (struct r5conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007079{
7080 int i;
7081 struct disk_info *tmp;
7082
NeilBrown0c55e022010-05-03 14:09:02 +10007083 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007084 if (!conf) {
7085 printk("(conf==NULL)\n");
7086 return;
7087 }
NeilBrown0c55e022010-05-03 14:09:02 +10007088 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
7089 conf->raid_disks,
7090 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007091
7092 for (i = 0; i < conf->raid_disks; i++) {
7093 char b[BDEVNAME_SIZE];
7094 tmp = conf->disks + i;
7095 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10007096 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
7097 i, !test_bit(Faulty, &tmp->rdev->flags),
7098 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007099 }
7100}
7101
NeilBrownfd01b882011-10-11 16:47:53 +11007102static int raid5_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007103{
7104 int i;
NeilBrownd1688a62011-10-11 16:49:52 +11007105 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007106 struct disk_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10007107 int count = 0;
7108 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007109
7110 for (i = 0; i < conf->raid_disks; i++) {
7111 tmp = conf->disks + i;
NeilBrowndd054fc2011-12-23 10:17:53 +11007112 if (tmp->replacement
7113 && tmp->replacement->recovery_offset == MaxSector
7114 && !test_bit(Faulty, &tmp->replacement->flags)
7115 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
7116 /* Replacement has just become active. */
7117 if (!tmp->rdev
7118 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
7119 count++;
7120 if (tmp->rdev) {
7121 /* Replaced device not technically faulty,
7122 * but we need to be sure it gets removed
7123 * and never re-added.
7124 */
7125 set_bit(Faulty, &tmp->rdev->flags);
7126 sysfs_notify_dirent_safe(
7127 tmp->rdev->sysfs_state);
7128 }
7129 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
7130 } else if (tmp->rdev
NeilBrown70fffd02010-06-16 17:01:25 +10007131 && tmp->rdev->recovery_offset == MaxSector
NeilBrownb2d444d2005-11-08 21:39:31 -08007132 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07007133 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10007134 count++;
Jonathan Brassow43c73ca2011-01-14 09:14:33 +11007135 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007136 }
7137 }
NeilBrown6b965622010-08-18 11:56:59 +10007138 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown908f4fb2011-12-23 10:17:50 +11007139 mddev->degraded = calc_degraded(conf);
NeilBrown6b965622010-08-18 11:56:59 +10007140 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007141 print_raid5_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10007142 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007143}
7144
NeilBrownb8321b62011-12-23 10:17:51 +11007145static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007146{
NeilBrownd1688a62011-10-11 16:49:52 +11007147 struct r5conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007148 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11007149 int number = rdev->raid_disk;
NeilBrown657e3e42011-12-23 10:17:52 +11007150 struct md_rdev **rdevp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007151 struct disk_info *p = conf->disks + number;
7152
7153 print_raid5_conf(conf);
Shaohua Lif6b6ec52015-12-21 10:51:02 +11007154 if (test_bit(Journal, &rdev->flags) && conf->log) {
7155 struct r5l_log *log;
Shaohua Lic2bb6242015-10-08 21:54:07 -07007156 /*
Shaohua Lif6b6ec52015-12-21 10:51:02 +11007157 * we can't wait pending write here, as this is called in
7158 * raid5d, wait will deadlock.
Shaohua Lic2bb6242015-10-08 21:54:07 -07007159 */
Shaohua Lif6b6ec52015-12-21 10:51:02 +11007160 if (atomic_read(&mddev->writes_pending))
7161 return -EBUSY;
7162 log = conf->log;
7163 conf->log = NULL;
7164 synchronize_rcu();
7165 r5l_exit_log(log);
7166 return 0;
Shaohua Lic2bb6242015-10-08 21:54:07 -07007167 }
NeilBrown657e3e42011-12-23 10:17:52 +11007168 if (rdev == p->rdev)
7169 rdevp = &p->rdev;
7170 else if (rdev == p->replacement)
7171 rdevp = &p->replacement;
7172 else
7173 return 0;
NeilBrownec32a2b2009-03-31 15:17:38 +11007174
NeilBrown657e3e42011-12-23 10:17:52 +11007175 if (number >= conf->raid_disks &&
7176 conf->reshape_progress == MaxSector)
7177 clear_bit(In_sync, &rdev->flags);
7178
7179 if (test_bit(In_sync, &rdev->flags) ||
7180 atomic_read(&rdev->nr_pending)) {
7181 err = -EBUSY;
7182 goto abort;
7183 }
7184 /* Only remove non-faulty devices if recovery
7185 * isn't possible.
7186 */
7187 if (!test_bit(Faulty, &rdev->flags) &&
7188 mddev->recovery_disabled != conf->recovery_disabled &&
7189 !has_failed(conf) &&
NeilBrowndd054fc2011-12-23 10:17:53 +11007190 (!p->replacement || p->replacement == rdev) &&
NeilBrown657e3e42011-12-23 10:17:52 +11007191 number < conf->raid_disks) {
7192 err = -EBUSY;
7193 goto abort;
7194 }
7195 *rdevp = NULL;
7196 synchronize_rcu();
7197 if (atomic_read(&rdev->nr_pending)) {
7198 /* lost the race, try later */
7199 err = -EBUSY;
7200 *rdevp = rdev;
NeilBrowndd054fc2011-12-23 10:17:53 +11007201 } else if (p->replacement) {
7202 /* We must have just cleared 'rdev' */
7203 p->rdev = p->replacement;
7204 clear_bit(Replacement, &p->replacement->flags);
7205 smp_mb(); /* Make sure other CPUs may see both as identical
7206 * but will never see neither - if they are careful
7207 */
7208 p->replacement = NULL;
7209 clear_bit(WantReplacement, &rdev->flags);
7210 } else
7211 /* We might have just removed the Replacement as faulty-
7212 * clear the bit just in case
7213 */
7214 clear_bit(WantReplacement, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007215abort:
7216
7217 print_raid5_conf(conf);
7218 return err;
7219}
7220
NeilBrownfd01b882011-10-11 16:47:53 +11007221static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007222{
NeilBrownd1688a62011-10-11 16:49:52 +11007223 struct r5conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10007224 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007225 int disk;
7226 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10007227 int first = 0;
7228 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007229
Shaohua Lif6b6ec52015-12-21 10:51:02 +11007230 if (test_bit(Journal, &rdev->flags)) {
7231 char b[BDEVNAME_SIZE];
7232 if (conf->log)
7233 return -EBUSY;
7234
7235 rdev->raid_disk = 0;
7236 /*
7237 * The array is in readonly mode if journal is missing, so no
7238 * write requests running. We should be safe
7239 */
7240 r5l_init_log(conf, rdev);
7241 printk(KERN_INFO"md/raid:%s: using device %s as journal\n",
7242 mdname(mddev), bdevname(rdev->bdev, b));
7243 return 0;
7244 }
NeilBrown7f0da592011-07-28 11:39:22 +10007245 if (mddev->recovery_disabled == conf->recovery_disabled)
7246 return -EBUSY;
7247
NeilBrowndc10c642012-03-19 12:46:37 +11007248 if (rdev->saved_raid_disk < 0 && has_failed(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07007249 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10007250 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007251
Neil Brown6c2fce22008-06-28 08:31:31 +10007252 if (rdev->raid_disk >= 0)
7253 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007254
7255 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07007256 * find the disk ... but prefer rdev->saved_raid_disk
7257 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07007258 */
NeilBrown16a53ec2006-06-26 00:27:38 -07007259 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10007260 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07007261 conf->disks[rdev->saved_raid_disk].rdev == NULL)
NeilBrown5cfb22a2012-07-03 11:46:53 +10007262 first = rdev->saved_raid_disk;
7263
7264 for (disk = first; disk <= last; disk++) {
NeilBrown7bfec5f2011-12-23 10:17:53 +11007265 p = conf->disks + disk;
7266 if (p->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08007267 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007268 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10007269 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07007270 if (rdev->saved_raid_disk != disk)
7271 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08007272 rcu_assign_pointer(p->rdev, rdev);
NeilBrown5cfb22a2012-07-03 11:46:53 +10007273 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007274 }
NeilBrown5cfb22a2012-07-03 11:46:53 +10007275 }
7276 for (disk = first; disk <= last; disk++) {
7277 p = conf->disks + disk;
NeilBrown7bfec5f2011-12-23 10:17:53 +11007278 if (test_bit(WantReplacement, &p->rdev->flags) &&
7279 p->replacement == NULL) {
7280 clear_bit(In_sync, &rdev->flags);
7281 set_bit(Replacement, &rdev->flags);
7282 rdev->raid_disk = disk;
7283 err = 0;
7284 conf->fullsync = 1;
7285 rcu_assign_pointer(p->replacement, rdev);
7286 break;
7287 }
7288 }
NeilBrown5cfb22a2012-07-03 11:46:53 +10007289out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07007290 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10007291 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007292}
7293
NeilBrownfd01b882011-10-11 16:47:53 +11007294static int raid5_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007295{
7296 /* no resync is happening, and there is enough space
7297 * on all devices, so we can resize.
7298 * We need to make sure resync covers any new space.
7299 * If the array is shrinking we should possibly wait until
7300 * any io in the removed space completes, but it hardly seems
7301 * worth it.
7302 */
NeilBrowna4a61252012-05-22 13:55:27 +10007303 sector_t newsize;
NeilBrown3cb5edf2015-07-15 17:24:17 +10007304 struct r5conf *conf = mddev->private;
7305
Shaohua Li713cf5a2015-08-13 14:32:03 -07007306 if (conf->log)
7307 return -EINVAL;
NeilBrown3cb5edf2015-07-15 17:24:17 +10007308 sectors &= ~((sector_t)conf->chunk_sectors - 1);
NeilBrowna4a61252012-05-22 13:55:27 +10007309 newsize = raid5_size(mddev, sectors, mddev->raid_disks);
7310 if (mddev->external_size &&
7311 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11007312 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10007313 if (mddev->bitmap) {
7314 int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0);
7315 if (ret)
7316 return ret;
7317 }
7318 md_set_array_sectors(mddev, newsize);
Andre Nollf233ea52008-07-21 17:05:22 +10007319 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10007320 revalidate_disk(mddev->gendisk);
NeilBrownb0986362011-05-11 15:52:21 +10007321 if (sectors > mddev->dev_sectors &&
7322 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11007323 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007324 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7325 }
Andre Noll58c0fed2009-03-31 14:33:13 +11007326 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07007327 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007328 return 0;
7329}
7330
NeilBrownfd01b882011-10-11 16:47:53 +11007331static int check_stripe_cache(struct mddev *mddev)
NeilBrown01ee22b2009-06-18 08:47:20 +10007332{
7333 /* Can only proceed if there are plenty of stripe_heads.
7334 * We need a minimum of one full stripe,, and for sensible progress
7335 * it is best to have about 4 times that.
7336 * If we require 4 times, then the default 256 4K stripe_heads will
7337 * allow for chunk sizes up to 256K, which is probably OK.
7338 * If the chunk size is greater, user-space should request more
7339 * stripe_heads first.
7340 */
NeilBrownd1688a62011-10-11 16:49:52 +11007341 struct r5conf *conf = mddev->private;
NeilBrown01ee22b2009-06-18 08:47:20 +10007342 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
NeilBrownedbe83a2015-02-26 12:47:56 +11007343 > conf->min_nr_stripes ||
NeilBrown01ee22b2009-06-18 08:47:20 +10007344 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
NeilBrownedbe83a2015-02-26 12:47:56 +11007345 > conf->min_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10007346 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
7347 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10007348 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
7349 / STRIPE_SIZE)*4);
7350 return 0;
7351 }
7352 return 1;
7353}
7354
NeilBrownfd01b882011-10-11 16:47:53 +11007355static int check_reshape(struct mddev *mddev)
NeilBrown29269552006-03-27 01:18:10 -08007356{
NeilBrownd1688a62011-10-11 16:49:52 +11007357 struct r5conf *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08007358
Shaohua Li713cf5a2015-08-13 14:32:03 -07007359 if (conf->log)
7360 return -EINVAL;
NeilBrown88ce4932009-03-31 15:24:23 +11007361 if (mddev->delta_disks == 0 &&
7362 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10007363 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10007364 return 0; /* nothing to do */
NeilBrown674806d2010-06-16 17:17:53 +10007365 if (has_failed(conf))
NeilBrownec32a2b2009-03-31 15:17:38 +11007366 return -EINVAL;
NeilBrownfdcfbbb2013-07-04 16:38:16 +10007367 if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) {
NeilBrownec32a2b2009-03-31 15:17:38 +11007368 /* We might be able to shrink, but the devices must
7369 * be made bigger first.
7370 * For raid6, 4 is the minimum size.
7371 * Otherwise 2 is the minimum
7372 */
7373 int min = 2;
7374 if (mddev->level == 6)
7375 min = 4;
7376 if (mddev->raid_disks + mddev->delta_disks < min)
7377 return -EINVAL;
7378 }
NeilBrown29269552006-03-27 01:18:10 -08007379
NeilBrown01ee22b2009-06-18 08:47:20 +10007380 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08007381 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08007382
NeilBrown738a2732015-05-08 18:19:39 +10007383 if (mddev->new_chunk_sectors > mddev->chunk_sectors ||
7384 mddev->delta_disks > 0)
7385 if (resize_chunks(conf,
7386 conf->previous_raid_disks
7387 + max(0, mddev->delta_disks),
7388 max(mddev->new_chunk_sectors,
7389 mddev->chunk_sectors)
7390 ) < 0)
7391 return -ENOMEM;
NeilBrowne56108d62012-10-11 14:24:13 +11007392 return resize_stripes(conf, (conf->previous_raid_disks
7393 + mddev->delta_disks));
NeilBrown63c70c42006-03-27 01:18:13 -08007394}
7395
NeilBrownfd01b882011-10-11 16:47:53 +11007396static int raid5_start_reshape(struct mddev *mddev)
NeilBrown63c70c42006-03-27 01:18:13 -08007397{
NeilBrownd1688a62011-10-11 16:49:52 +11007398 struct r5conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11007399 struct md_rdev *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08007400 int spares = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07007401 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08007402
NeilBrownf4168852007-02-28 20:11:53 -08007403 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08007404 return -EBUSY;
7405
NeilBrown01ee22b2009-06-18 08:47:20 +10007406 if (!check_stripe_cache(mddev))
7407 return -ENOSPC;
7408
NeilBrown30b67642012-05-22 13:55:28 +10007409 if (has_failed(conf))
7410 return -EINVAL;
7411
NeilBrownc6563a82012-05-21 09:27:00 +10007412 rdev_for_each(rdev, mddev) {
NeilBrown469518a2011-01-31 11:57:43 +11007413 if (!test_bit(In_sync, &rdev->flags)
7414 && !test_bit(Faulty, &rdev->flags))
NeilBrown29269552006-03-27 01:18:10 -08007415 spares++;
NeilBrownc6563a82012-05-21 09:27:00 +10007416 }
NeilBrown63c70c42006-03-27 01:18:13 -08007417
NeilBrownf4168852007-02-28 20:11:53 -08007418 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08007419 /* Not enough devices even to make a degraded array
7420 * of that size
7421 */
7422 return -EINVAL;
7423
NeilBrownec32a2b2009-03-31 15:17:38 +11007424 /* Refuse to reduce size of the array. Any reductions in
7425 * array size must be through explicit setting of array_size
7426 * attribute.
7427 */
7428 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
7429 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10007430 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11007431 "before number of disks\n", mdname(mddev));
7432 return -EINVAL;
7433 }
7434
NeilBrownf6705572006-03-27 01:18:11 -08007435 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08007436 spin_lock_irq(&conf->device_lock);
NeilBrownc46501b2013-08-27 15:52:13 +10007437 write_seqcount_begin(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08007438 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08007439 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10007440 conf->prev_chunk_sectors = conf->chunk_sectors;
7441 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11007442 conf->prev_algo = conf->algorithm;
7443 conf->algorithm = mddev->new_layout;
NeilBrown05616be2012-05-21 09:27:00 +10007444 conf->generation++;
7445 /* Code that selects data_offset needs to see the generation update
7446 * if reshape_progress has been set - so a memory barrier needed.
7447 */
7448 smp_mb();
NeilBrown2c810cd2012-05-21 09:27:00 +10007449 if (mddev->reshape_backwards)
NeilBrownfef9c612009-03-31 15:16:46 +11007450 conf->reshape_progress = raid5_size(mddev, 0, 0);
7451 else
7452 conf->reshape_progress = 0;
7453 conf->reshape_safe = conf->reshape_progress;
NeilBrownc46501b2013-08-27 15:52:13 +10007454 write_seqcount_end(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08007455 spin_unlock_irq(&conf->device_lock);
7456
NeilBrown4d77e3b2013-08-27 15:57:47 +10007457 /* Now make sure any requests that proceeded on the assumption
7458 * the reshape wasn't running - like Discard or Read - have
7459 * completed.
7460 */
7461 mddev_suspend(mddev);
7462 mddev_resume(mddev);
7463
NeilBrown29269552006-03-27 01:18:10 -08007464 /* Add some new drives, as many as will fit.
7465 * We know there are enough to make the newly sized array work.
NeilBrown3424bf62010-06-17 17:48:26 +10007466 * Don't add devices if we are reducing the number of
7467 * devices in the array. This is because it is not possible
7468 * to correctly record the "partially reconstructed" state of
7469 * such devices during the reshape and confusion could result.
NeilBrown29269552006-03-27 01:18:10 -08007470 */
NeilBrown87a8dec2011-01-31 11:57:43 +11007471 if (mddev->delta_disks >= 0) {
NeilBrowndafb20f2012-03-19 12:46:39 +11007472 rdev_for_each(rdev, mddev)
NeilBrown87a8dec2011-01-31 11:57:43 +11007473 if (rdev->raid_disk < 0 &&
7474 !test_bit(Faulty, &rdev->flags)) {
7475 if (raid5_add_disk(mddev, rdev) == 0) {
NeilBrown87a8dec2011-01-31 11:57:43 +11007476 if (rdev->raid_disk
NeilBrown9d4c7d82012-03-13 11:21:21 +11007477 >= conf->previous_raid_disks)
NeilBrown87a8dec2011-01-31 11:57:43 +11007478 set_bit(In_sync, &rdev->flags);
NeilBrown9d4c7d82012-03-13 11:21:21 +11007479 else
NeilBrown87a8dec2011-01-31 11:57:43 +11007480 rdev->recovery_offset = 0;
Namhyung Kim36fad852011-07-27 11:00:36 +10007481
7482 if (sysfs_link_rdev(mddev, rdev))
NeilBrown87a8dec2011-01-31 11:57:43 +11007483 /* Failure here is OK */;
NeilBrown50da0842011-01-31 11:57:43 +11007484 }
NeilBrown87a8dec2011-01-31 11:57:43 +11007485 } else if (rdev->raid_disk >= conf->previous_raid_disks
7486 && !test_bit(Faulty, &rdev->flags)) {
7487 /* This is a spare that was manually added */
7488 set_bit(In_sync, &rdev->flags);
NeilBrown87a8dec2011-01-31 11:57:43 +11007489 }
NeilBrown29269552006-03-27 01:18:10 -08007490
NeilBrown87a8dec2011-01-31 11:57:43 +11007491 /* When a reshape changes the number of devices,
7492 * ->degraded is measured against the larger of the
7493 * pre and post number of devices.
7494 */
NeilBrownec32a2b2009-03-31 15:17:38 +11007495 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown908f4fb2011-12-23 10:17:50 +11007496 mddev->degraded = calc_degraded(conf);
NeilBrownec32a2b2009-03-31 15:17:38 +11007497 spin_unlock_irqrestore(&conf->device_lock, flags);
7498 }
NeilBrown63c70c42006-03-27 01:18:13 -08007499 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10007500 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07007501 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08007502
NeilBrown29269552006-03-27 01:18:10 -08007503 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7504 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
NeilBrownea358cd2015-06-12 20:05:04 +10007505 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
NeilBrown29269552006-03-27 01:18:10 -08007506 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7507 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7508 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10007509 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08007510 if (!mddev->sync_thread) {
7511 mddev->recovery = 0;
7512 spin_lock_irq(&conf->device_lock);
NeilBrownba8805b2013-11-14 15:16:15 +11007513 write_seqcount_begin(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08007514 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownba8805b2013-11-14 15:16:15 +11007515 mddev->new_chunk_sectors =
7516 conf->chunk_sectors = conf->prev_chunk_sectors;
7517 mddev->new_layout = conf->algorithm = conf->prev_algo;
NeilBrown05616be2012-05-21 09:27:00 +10007518 rdev_for_each(rdev, mddev)
7519 rdev->new_data_offset = rdev->data_offset;
7520 smp_wmb();
NeilBrownba8805b2013-11-14 15:16:15 +11007521 conf->generation --;
NeilBrownfef9c612009-03-31 15:16:46 +11007522 conf->reshape_progress = MaxSector;
NeilBrown1e3fa9b2012-03-13 11:21:18 +11007523 mddev->reshape_position = MaxSector;
NeilBrownba8805b2013-11-14 15:16:15 +11007524 write_seqcount_end(&conf->gen_lock);
NeilBrown29269552006-03-27 01:18:10 -08007525 spin_unlock_irq(&conf->device_lock);
7526 return -EAGAIN;
7527 }
NeilBrownc8f517c2009-03-31 15:28:40 +11007528 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08007529 md_wakeup_thread(mddev->sync_thread);
7530 md_new_event(mddev);
7531 return 0;
7532}
NeilBrown29269552006-03-27 01:18:10 -08007533
NeilBrownec32a2b2009-03-31 15:17:38 +11007534/* This is called from the reshape thread and should make any
7535 * changes needed in 'conf'
7536 */
NeilBrownd1688a62011-10-11 16:49:52 +11007537static void end_reshape(struct r5conf *conf)
NeilBrown29269552006-03-27 01:18:10 -08007538{
NeilBrown29269552006-03-27 01:18:10 -08007539
NeilBrownf6705572006-03-27 01:18:11 -08007540 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
NeilBrown05616be2012-05-21 09:27:00 +10007541 struct md_rdev *rdev;
Dan Williams80c3a6c2009-03-17 18:10:40 -07007542
NeilBrownf6705572006-03-27 01:18:11 -08007543 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11007544 conf->previous_raid_disks = conf->raid_disks;
NeilBrown05616be2012-05-21 09:27:00 +10007545 rdev_for_each(rdev, conf->mddev)
7546 rdev->data_offset = rdev->new_data_offset;
7547 smp_wmb();
NeilBrownfef9c612009-03-31 15:16:46 +11007548 conf->reshape_progress = MaxSector;
NeilBrown6cbd8142015-07-24 13:30:32 +10007549 conf->mddev->reshape_position = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08007550 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11007551 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07007552
7553 /* read-ahead size must cover two whole stripes, which is
7554 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
7555 */
NeilBrown4a5add42010-06-01 19:37:28 +10007556 if (conf->mddev->queue) {
NeilBrowncea9c222009-03-31 15:15:05 +11007557 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10007558 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11007559 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07007560 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
7561 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
7562 }
NeilBrown29269552006-03-27 01:18:10 -08007563 }
NeilBrown29269552006-03-27 01:18:10 -08007564}
7565
NeilBrownec32a2b2009-03-31 15:17:38 +11007566/* This is called from the raid5d thread with mddev_lock held.
7567 * It makes config changes to the device.
7568 */
NeilBrownfd01b882011-10-11 16:47:53 +11007569static void raid5_finish_reshape(struct mddev *mddev)
NeilBrowncea9c222009-03-31 15:15:05 +11007570{
NeilBrownd1688a62011-10-11 16:49:52 +11007571 struct r5conf *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11007572
7573 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7574
NeilBrownec32a2b2009-03-31 15:17:38 +11007575 if (mddev->delta_disks > 0) {
7576 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
7577 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10007578 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11007579 } else {
7580 int d;
NeilBrown908f4fb2011-12-23 10:17:50 +11007581 spin_lock_irq(&conf->device_lock);
7582 mddev->degraded = calc_degraded(conf);
7583 spin_unlock_irq(&conf->device_lock);
NeilBrownec32a2b2009-03-31 15:17:38 +11007584 for (d = conf->raid_disks ;
7585 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10007586 d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11007587 struct md_rdev *rdev = conf->disks[d].rdev;
NeilBrownda7613b2012-05-22 13:55:33 +10007588 if (rdev)
7589 clear_bit(In_sync, &rdev->flags);
7590 rdev = conf->disks[d].replacement;
7591 if (rdev)
7592 clear_bit(In_sync, &rdev->flags);
NeilBrown1a67dde2009-08-13 10:41:49 +10007593 }
NeilBrowncea9c222009-03-31 15:15:05 +11007594 }
NeilBrown88ce4932009-03-31 15:24:23 +11007595 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10007596 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11007597 mddev->reshape_position = MaxSector;
7598 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10007599 mddev->reshape_backwards = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11007600 }
7601}
7602
NeilBrownfd01b882011-10-11 16:47:53 +11007603static void raid5_quiesce(struct mddev *mddev, int state)
NeilBrown72626682005-09-09 16:23:54 -07007604{
NeilBrownd1688a62011-10-11 16:49:52 +11007605 struct r5conf *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07007606
7607 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08007608 case 2: /* resume for a suspend */
7609 wake_up(&conf->wait_for_overlap);
7610 break;
7611
NeilBrown72626682005-09-09 16:23:54 -07007612 case 1: /* stop all writes */
Shaohua Li566c09c2013-11-14 15:16:17 +11007613 lock_all_device_hash_locks_irq(conf);
NeilBrown64bd6602009-08-03 10:59:58 +10007614 /* '2' tells resync/reshape to pause so that all
7615 * active stripes can drain
7616 */
7617 conf->quiesce = 2;
Yuanhan Liub1b46482015-05-08 18:19:06 +10007618 wait_event_cmd(conf->wait_for_quiescent,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08007619 atomic_read(&conf->active_stripes) == 0 &&
7620 atomic_read(&conf->active_aligned_reads) == 0,
Shaohua Li566c09c2013-11-14 15:16:17 +11007621 unlock_all_device_hash_locks_irq(conf),
7622 lock_all_device_hash_locks_irq(conf));
NeilBrown64bd6602009-08-03 10:59:58 +10007623 conf->quiesce = 1;
Shaohua Li566c09c2013-11-14 15:16:17 +11007624 unlock_all_device_hash_locks_irq(conf);
NeilBrown64bd6602009-08-03 10:59:58 +10007625 /* allow reshape to continue */
7626 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07007627 break;
7628
7629 case 0: /* re-enable writes */
Shaohua Li566c09c2013-11-14 15:16:17 +11007630 lock_all_device_hash_locks_irq(conf);
NeilBrown72626682005-09-09 16:23:54 -07007631 conf->quiesce = 0;
Yuanhan Liub1b46482015-05-08 18:19:06 +10007632 wake_up(&conf->wait_for_quiescent);
NeilBrowne464eaf2006-03-27 01:18:14 -08007633 wake_up(&conf->wait_for_overlap);
Shaohua Li566c09c2013-11-14 15:16:17 +11007634 unlock_all_device_hash_locks_irq(conf);
NeilBrown72626682005-09-09 16:23:54 -07007635 break;
7636 }
Shaohua Lie6c033f2015-10-04 09:20:12 -07007637 r5l_quiesce(conf->log, state);
NeilBrown72626682005-09-09 16:23:54 -07007638}
NeilBrownb15c2e52006-01-06 00:20:16 -08007639
NeilBrownfd01b882011-10-11 16:47:53 +11007640static void *raid45_takeover_raid0(struct mddev *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11007641{
NeilBrowne373ab12011-10-11 16:48:59 +11007642 struct r0conf *raid0_conf = mddev->private;
Randy Dunlapd76c8422011-04-21 09:07:26 -07007643 sector_t sectors;
Trela Maciej54071b32010-03-08 16:02:42 +11007644
Dan Williamsf1b29bc2010-05-01 18:09:05 -07007645 /* for raid0 takeover only one zone is supported */
NeilBrowne373ab12011-10-11 16:48:59 +11007646 if (raid0_conf->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10007647 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
7648 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07007649 return ERR_PTR(-EINVAL);
7650 }
7651
NeilBrowne373ab12011-10-11 16:48:59 +11007652 sectors = raid0_conf->strip_zone[0].zone_end;
7653 sector_div(sectors, raid0_conf->strip_zone[0].nb_dev);
NeilBrown3b71bd92011-04-20 15:38:18 +10007654 mddev->dev_sectors = sectors;
Dan Williamsf1b29bc2010-05-01 18:09:05 -07007655 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11007656 mddev->new_layout = ALGORITHM_PARITY_N;
7657 mddev->new_chunk_sectors = mddev->chunk_sectors;
7658 mddev->raid_disks += 1;
7659 mddev->delta_disks = 1;
7660 /* make sure it will be not marked as dirty */
7661 mddev->recovery_cp = MaxSector;
7662
7663 return setup_conf(mddev);
7664}
7665
NeilBrownfd01b882011-10-11 16:47:53 +11007666static void *raid5_takeover_raid1(struct mddev *mddev)
NeilBrownd562b0c2009-03-31 14:39:39 +11007667{
7668 int chunksect;
7669
7670 if (mddev->raid_disks != 2 ||
7671 mddev->degraded > 1)
7672 return ERR_PTR(-EINVAL);
7673
7674 /* Should check if there are write-behind devices? */
7675
7676 chunksect = 64*2; /* 64K by default */
7677
7678 /* The array must be an exact multiple of chunksize */
7679 while (chunksect && (mddev->array_sectors & (chunksect-1)))
7680 chunksect >>= 1;
7681
7682 if ((chunksect<<9) < STRIPE_SIZE)
7683 /* array size does not allow a suitable chunk size */
7684 return ERR_PTR(-EINVAL);
7685
7686 mddev->new_level = 5;
7687 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10007688 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11007689
7690 return setup_conf(mddev);
7691}
7692
NeilBrownfd01b882011-10-11 16:47:53 +11007693static void *raid5_takeover_raid6(struct mddev *mddev)
NeilBrownfc9739c2009-03-31 14:57:20 +11007694{
7695 int new_layout;
7696
7697 switch (mddev->layout) {
7698 case ALGORITHM_LEFT_ASYMMETRIC_6:
7699 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
7700 break;
7701 case ALGORITHM_RIGHT_ASYMMETRIC_6:
7702 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
7703 break;
7704 case ALGORITHM_LEFT_SYMMETRIC_6:
7705 new_layout = ALGORITHM_LEFT_SYMMETRIC;
7706 break;
7707 case ALGORITHM_RIGHT_SYMMETRIC_6:
7708 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
7709 break;
7710 case ALGORITHM_PARITY_0_6:
7711 new_layout = ALGORITHM_PARITY_0;
7712 break;
7713 case ALGORITHM_PARITY_N:
7714 new_layout = ALGORITHM_PARITY_N;
7715 break;
7716 default:
7717 return ERR_PTR(-EINVAL);
7718 }
7719 mddev->new_level = 5;
7720 mddev->new_layout = new_layout;
7721 mddev->delta_disks = -1;
7722 mddev->raid_disks -= 1;
7723 return setup_conf(mddev);
7724}
7725
NeilBrownfd01b882011-10-11 16:47:53 +11007726static int raid5_check_reshape(struct mddev *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11007727{
NeilBrown88ce4932009-03-31 15:24:23 +11007728 /* For a 2-drive array, the layout and chunk size can be changed
7729 * immediately as not restriping is needed.
7730 * For larger arrays we record the new value - after validation
7731 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11007732 */
NeilBrownd1688a62011-10-11 16:49:52 +11007733 struct r5conf *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10007734 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11007735
NeilBrown597a7112009-06-18 08:47:42 +10007736 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11007737 return -EINVAL;
7738 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10007739 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11007740 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10007741 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11007742 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10007743 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11007744 /* not factor of array size */
7745 return -EINVAL;
7746 }
7747
7748 /* They look valid */
7749
NeilBrown88ce4932009-03-31 15:24:23 +11007750 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10007751 /* can make the change immediately */
7752 if (mddev->new_layout >= 0) {
7753 conf->algorithm = mddev->new_layout;
7754 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11007755 }
7756 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10007757 conf->chunk_sectors = new_chunk ;
7758 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11007759 }
7760 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7761 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11007762 }
NeilBrown50ac1682009-06-18 08:47:55 +10007763 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11007764}
7765
NeilBrownfd01b882011-10-11 16:47:53 +11007766static int raid6_check_reshape(struct mddev *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11007767{
NeilBrown597a7112009-06-18 08:47:42 +10007768 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10007769
NeilBrown597a7112009-06-18 08:47:42 +10007770 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11007771 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11007772 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10007773 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11007774 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10007775 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11007776 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10007777 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11007778 /* not factor of array size */
7779 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11007780 }
NeilBrown88ce4932009-03-31 15:24:23 +11007781
7782 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10007783 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11007784}
7785
NeilBrownfd01b882011-10-11 16:47:53 +11007786static void *raid5_takeover(struct mddev *mddev)
NeilBrownd562b0c2009-03-31 14:39:39 +11007787{
7788 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07007789 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11007790 * raid1 - if there are two drives. We need to know the chunk size
7791 * raid4 - trivial - just use a raid4 layout.
7792 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11007793 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07007794 if (mddev->level == 0)
7795 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11007796 if (mddev->level == 1)
7797 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11007798 if (mddev->level == 4) {
7799 mddev->new_layout = ALGORITHM_PARITY_N;
7800 mddev->new_level = 5;
7801 return setup_conf(mddev);
7802 }
NeilBrownfc9739c2009-03-31 14:57:20 +11007803 if (mddev->level == 6)
7804 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11007805
7806 return ERR_PTR(-EINVAL);
7807}
7808
NeilBrownfd01b882011-10-11 16:47:53 +11007809static void *raid4_takeover(struct mddev *mddev)
NeilBrowna78d38a2010-03-22 16:53:49 +11007810{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07007811 /* raid4 can take over:
7812 * raid0 - if there is only one strip zone
7813 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11007814 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07007815 if (mddev->level == 0)
7816 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11007817 if (mddev->level == 5 &&
7818 mddev->layout == ALGORITHM_PARITY_N) {
7819 mddev->new_layout = 0;
7820 mddev->new_level = 4;
7821 return setup_conf(mddev);
7822 }
7823 return ERR_PTR(-EINVAL);
7824}
NeilBrownd562b0c2009-03-31 14:39:39 +11007825
NeilBrown84fc4b52011-10-11 16:49:58 +11007826static struct md_personality raid5_personality;
NeilBrown245f46c2009-03-31 14:39:39 +11007827
NeilBrownfd01b882011-10-11 16:47:53 +11007828static void *raid6_takeover(struct mddev *mddev)
NeilBrown245f46c2009-03-31 14:39:39 +11007829{
7830 /* Currently can only take over a raid5. We map the
7831 * personality to an equivalent raid6 personality
7832 * with the Q block at the end.
7833 */
7834 int new_layout;
7835
7836 if (mddev->pers != &raid5_personality)
7837 return ERR_PTR(-EINVAL);
7838 if (mddev->degraded > 1)
7839 return ERR_PTR(-EINVAL);
7840 if (mddev->raid_disks > 253)
7841 return ERR_PTR(-EINVAL);
7842 if (mddev->raid_disks < 3)
7843 return ERR_PTR(-EINVAL);
7844
7845 switch (mddev->layout) {
7846 case ALGORITHM_LEFT_ASYMMETRIC:
7847 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
7848 break;
7849 case ALGORITHM_RIGHT_ASYMMETRIC:
7850 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
7851 break;
7852 case ALGORITHM_LEFT_SYMMETRIC:
7853 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
7854 break;
7855 case ALGORITHM_RIGHT_SYMMETRIC:
7856 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
7857 break;
7858 case ALGORITHM_PARITY_0:
7859 new_layout = ALGORITHM_PARITY_0_6;
7860 break;
7861 case ALGORITHM_PARITY_N:
7862 new_layout = ALGORITHM_PARITY_N;
7863 break;
7864 default:
7865 return ERR_PTR(-EINVAL);
7866 }
7867 mddev->new_level = 6;
7868 mddev->new_layout = new_layout;
7869 mddev->delta_disks = 1;
7870 mddev->raid_disks += 1;
7871 return setup_conf(mddev);
7872}
7873
NeilBrown84fc4b52011-10-11 16:49:58 +11007874static struct md_personality raid6_personality =
NeilBrown16a53ec2006-06-26 00:27:38 -07007875{
7876 .name = "raid6",
7877 .level = 6,
7878 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08007879 .make_request = raid5_make_request,
7880 .run = raid5_run,
NeilBrownafa0f552014-12-15 12:56:58 +11007881 .free = raid5_free,
Shaohua Li849674e2016-01-20 13:52:20 -08007882 .status = raid5_status,
7883 .error_handler = raid5_error,
NeilBrown16a53ec2006-06-26 00:27:38 -07007884 .hot_add_disk = raid5_add_disk,
7885 .hot_remove_disk= raid5_remove_disk,
7886 .spare_active = raid5_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08007887 .sync_request = raid5_sync_request,
NeilBrown16a53ec2006-06-26 00:27:38 -07007888 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07007889 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10007890 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08007891 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11007892 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07007893 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11007894 .takeover = raid6_takeover,
NeilBrown5c675f82014-12-15 12:56:56 +11007895 .congested = raid5_congested,
NeilBrown16a53ec2006-06-26 00:27:38 -07007896};
NeilBrown84fc4b52011-10-11 16:49:58 +11007897static struct md_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07007898{
7899 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08007900 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007901 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08007902 .make_request = raid5_make_request,
7903 .run = raid5_run,
NeilBrownafa0f552014-12-15 12:56:58 +11007904 .free = raid5_free,
Shaohua Li849674e2016-01-20 13:52:20 -08007905 .status = raid5_status,
7906 .error_handler = raid5_error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007907 .hot_add_disk = raid5_add_disk,
7908 .hot_remove_disk= raid5_remove_disk,
7909 .spare_active = raid5_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08007910 .sync_request = raid5_sync_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007911 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07007912 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08007913 .check_reshape = raid5_check_reshape,
7914 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11007915 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07007916 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11007917 .takeover = raid5_takeover,
NeilBrown5c675f82014-12-15 12:56:56 +11007918 .congested = raid5_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007919};
7920
NeilBrown84fc4b52011-10-11 16:49:58 +11007921static struct md_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07007922{
NeilBrown2604b702006-01-06 00:20:36 -08007923 .name = "raid4",
7924 .level = 4,
7925 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08007926 .make_request = raid5_make_request,
7927 .run = raid5_run,
NeilBrownafa0f552014-12-15 12:56:58 +11007928 .free = raid5_free,
Shaohua Li849674e2016-01-20 13:52:20 -08007929 .status = raid5_status,
7930 .error_handler = raid5_error,
NeilBrown2604b702006-01-06 00:20:36 -08007931 .hot_add_disk = raid5_add_disk,
7932 .hot_remove_disk= raid5_remove_disk,
7933 .spare_active = raid5_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08007934 .sync_request = raid5_sync_request,
NeilBrown2604b702006-01-06 00:20:36 -08007935 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07007936 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08007937 .check_reshape = raid5_check_reshape,
7938 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11007939 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08007940 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11007941 .takeover = raid4_takeover,
NeilBrown5c675f82014-12-15 12:56:56 +11007942 .congested = raid5_congested,
NeilBrown2604b702006-01-06 00:20:36 -08007943};
7944
7945static int __init raid5_init(void)
7946{
Shaohua Li851c30c2013-08-28 14:30:16 +08007947 raid5_wq = alloc_workqueue("raid5wq",
7948 WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0);
7949 if (!raid5_wq)
7950 return -ENOMEM;
NeilBrown16a53ec2006-06-26 00:27:38 -07007951 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08007952 register_md_personality(&raid5_personality);
7953 register_md_personality(&raid4_personality);
7954 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007955}
7956
NeilBrown2604b702006-01-06 00:20:36 -08007957static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007958{
NeilBrown16a53ec2006-06-26 00:27:38 -07007959 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08007960 unregister_md_personality(&raid5_personality);
7961 unregister_md_personality(&raid4_personality);
Shaohua Li851c30c2013-08-28 14:30:16 +08007962 destroy_workqueue(raid5_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007963}
7964
7965module_init(raid5_init);
7966module_exit(raid5_exit);
7967MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11007968MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007969MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08007970MODULE_ALIAS("md-raid5");
7971MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08007972MODULE_ALIAS("md-level-5");
7973MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07007974MODULE_ALIAS("md-personality-8"); /* RAID6 */
7975MODULE_ALIAS("md-raid6");
7976MODULE_ALIAS("md-level-6");
7977
7978/* This used to be two separate modules, they were: */
7979MODULE_ALIAS("raid5");
7980MODULE_ALIAS("raid6");