blob: 49bf5f8914353a1c7972beaa27d6b5caee1031cc [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>
Dan Williams07a3b412009-08-29 19:13:13 -070050#include <linux/async.h>
NeilBrownbff61972009-03-31 14:33:13 +110051#include <linux/seq_file.h>
Dan Williams36d1c642009-07-14 11:48:22 -070052#include <linux/cpu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090053#include <linux/slab.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110054#include "md.h"
NeilBrownbff61972009-03-31 14:33:13 +110055#include "raid5.h"
Trela Maciej54071b32010-03-08 16:02:42 +110056#include "raid0.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110057#include "bitmap.h"
NeilBrown72626682005-09-09 16:23:54 -070058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/*
60 * Stripe cache
61 */
62
63#define NR_STRIPES 256
64#define STRIPE_SIZE PAGE_SIZE
65#define STRIPE_SHIFT (PAGE_SHIFT - 9)
66#define STRIPE_SECTORS (STRIPE_SIZE>>9)
67#define IO_THRESHOLD 1
Dan Williams8b3e6cd2008-04-28 02:15:53 -070068#define BYPASS_THRESHOLD 1
NeilBrownfccddba2006-01-06 00:20:33 -080069#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#define HASH_MASK (NR_HASH - 1)
71
NeilBrownfccddba2006-01-06 00:20:33 -080072#define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74/* bio's attached to a stripe+device for I/O are linked together in bi_sector
75 * order without overlap. There may be several bio's per stripe+device, and
76 * a bio could span several devices.
77 * When walking this list for a particular stripe+device, we must never proceed
78 * beyond a bio that extends past this device, as the next bio might no longer
79 * be valid.
80 * This macro is used to determine the 'next' bio in the list, given the sector
81 * of the current stripe+device
82 */
83#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
84/*
85 * The following can be used to debug the driver
86 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#define RAID5_PARANOIA 1
88#if RAID5_PARANOIA && defined(CONFIG_SMP)
89# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
90#else
91# define CHECK_DEVLOCK()
92#endif
93
Dan Williams45b42332007-07-09 11:56:43 -070094#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -070095#define inline
96#define __inline__
97#endif
98
Bernd Schubert6be9d492008-05-23 13:04:34 -070099#define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
100
Jens Axboe960e7392008-08-15 10:41:18 +0200101/*
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200102 * We maintain a biased count of active stripes in the bottom 16 bits of
103 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
Jens Axboe960e7392008-08-15 10:41:18 +0200104 */
105static inline int raid5_bi_phys_segments(struct bio *bio)
106{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200107 return bio->bi_phys_segments & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200108}
109
110static inline int raid5_bi_hw_segments(struct bio *bio)
111{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200112 return (bio->bi_phys_segments >> 16) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200113}
114
115static inline int raid5_dec_bi_phys_segments(struct bio *bio)
116{
117 --bio->bi_phys_segments;
118 return raid5_bi_phys_segments(bio);
119}
120
121static inline int raid5_dec_bi_hw_segments(struct bio *bio)
122{
123 unsigned short val = raid5_bi_hw_segments(bio);
124
125 --val;
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200126 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
Jens Axboe960e7392008-08-15 10:41:18 +0200127 return val;
128}
129
130static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
131{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200132 bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16);
Jens Axboe960e7392008-08-15 10:41:18 +0200133}
134
NeilBrownd0dabf72009-03-31 14:39:38 +1100135/* Find first data disk in a raid6 stripe */
136static inline int raid6_d0(struct stripe_head *sh)
137{
NeilBrown67cc2b82009-03-31 14:39:38 +1100138 if (sh->ddf_layout)
139 /* ddf always start from first device */
140 return 0;
141 /* md starts just after Q block */
NeilBrownd0dabf72009-03-31 14:39:38 +1100142 if (sh->qd_idx == sh->disks - 1)
143 return 0;
144 else
145 return sh->qd_idx + 1;
146}
NeilBrown16a53ec2006-06-26 00:27:38 -0700147static inline int raid6_next_disk(int disk, int raid_disks)
148{
149 disk++;
150 return (disk < raid_disks) ? disk : 0;
151}
Dan Williamsa4456852007-07-09 11:56:43 -0700152
NeilBrownd0dabf72009-03-31 14:39:38 +1100153/* When walking through the disks in a raid5, starting at raid6_d0,
154 * We need to map each disk to a 'slot', where the data disks are slot
155 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
156 * is raid_disks-1. This help does that mapping.
157 */
NeilBrown67cc2b82009-03-31 14:39:38 +1100158static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
159 int *count, int syndrome_disks)
NeilBrownd0dabf72009-03-31 14:39:38 +1100160{
Dan Williams66295422009-10-19 18:09:32 -0700161 int slot = *count;
NeilBrown67cc2b82009-03-31 14:39:38 +1100162
NeilBrowne4424fe2009-10-16 16:27:34 +1100163 if (sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700164 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100165 if (idx == sh->pd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100166 return syndrome_disks;
NeilBrownd0dabf72009-03-31 14:39:38 +1100167 if (idx == sh->qd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100168 return syndrome_disks + 1;
NeilBrowne4424fe2009-10-16 16:27:34 +1100169 if (!sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700170 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100171 return slot;
172}
173
Dan Williamsa4456852007-07-09 11:56:43 -0700174static void return_io(struct bio *return_bi)
175{
176 struct bio *bi = return_bi;
177 while (bi) {
Dan Williamsa4456852007-07-09 11:56:43 -0700178
179 return_bi = bi->bi_next;
180 bi->bi_next = NULL;
181 bi->bi_size = 0;
Neil Brown0e13fe232008-06-28 08:31:20 +1000182 bio_endio(bi, 0);
Dan Williamsa4456852007-07-09 11:56:43 -0700183 bi = return_bi;
184 }
185}
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187static void print_raid5_conf (raid5_conf_t *conf);
188
Dan Williams600aa102008-06-28 08:32:05 +1000189static int stripe_operations_active(struct stripe_head *sh)
190{
191 return sh->check_state || sh->reconstruct_state ||
192 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
193 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
194}
195
Arjan van de Ven858119e2006-01-14 13:20:43 -0800196static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 if (atomic_dec_and_test(&sh->count)) {
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200199 BUG_ON(!list_empty(&sh->lru));
200 BUG_ON(atomic_read(&conf->active_stripes)==0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 if (test_bit(STRIPE_HANDLE, &sh->state)) {
NeilBrown482c0832011-04-18 18:25:42 +1000202 if (test_bit(STRIPE_DELAYED, &sh->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 list_add_tail(&sh->lru, &conf->delayed_list);
NeilBrown482c0832011-04-18 18:25:42 +1000204 else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
205 sh->bm_seq - conf->seq_write > 0)
NeilBrown72626682005-09-09 16:23:54 -0700206 list_add_tail(&sh->lru, &conf->bitmap_list);
NeilBrown482c0832011-04-18 18:25:42 +1000207 else {
NeilBrown72626682005-09-09 16:23:54 -0700208 clear_bit(STRIPE_BIT_DELAY, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 list_add_tail(&sh->lru, &conf->handle_list);
NeilBrown72626682005-09-09 16:23:54 -0700210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 md_wakeup_thread(conf->mddev->thread);
212 } else {
Dan Williams600aa102008-06-28 08:32:05 +1000213 BUG_ON(stripe_operations_active(sh));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
215 atomic_dec(&conf->preread_active_stripes);
216 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
217 md_wakeup_thread(conf->mddev->thread);
218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 atomic_dec(&conf->active_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800220 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
221 list_add_tail(&sh->lru, &conf->inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 wake_up(&conf->wait_for_stripe);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800223 if (conf->retry_read_aligned)
224 md_wakeup_thread(conf->mddev->thread);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 }
227 }
228}
NeilBrownd0dabf72009-03-31 14:39:38 +1100229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230static void release_stripe(struct stripe_head *sh)
231{
232 raid5_conf_t *conf = sh->raid_conf;
233 unsigned long flags;
NeilBrown16a53ec2006-06-26 00:27:38 -0700234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 spin_lock_irqsave(&conf->device_lock, flags);
236 __release_stripe(conf, sh);
237 spin_unlock_irqrestore(&conf->device_lock, flags);
238}
239
NeilBrownfccddba2006-01-06 00:20:33 -0800240static inline void remove_hash(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
Dan Williams45b42332007-07-09 11:56:43 -0700242 pr_debug("remove_hash(), stripe %llu\n",
243 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
NeilBrownfccddba2006-01-06 00:20:33 -0800245 hlist_del_init(&sh->hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
NeilBrown16a53ec2006-06-26 00:27:38 -0700248static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
NeilBrownfccddba2006-01-06 00:20:33 -0800250 struct hlist_head *hp = stripe_hash(conf, sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
Dan Williams45b42332007-07-09 11:56:43 -0700252 pr_debug("insert_hash(), stripe %llu\n",
253 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
255 CHECK_DEVLOCK();
NeilBrownfccddba2006-01-06 00:20:33 -0800256 hlist_add_head(&sh->hash, hp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
259
260/* find an idle stripe, make sure it is unhashed, and return it. */
261static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
262{
263 struct stripe_head *sh = NULL;
264 struct list_head *first;
265
266 CHECK_DEVLOCK();
267 if (list_empty(&conf->inactive_list))
268 goto out;
269 first = conf->inactive_list.next;
270 sh = list_entry(first, struct stripe_head, lru);
271 list_del_init(first);
272 remove_hash(sh);
273 atomic_inc(&conf->active_stripes);
274out:
275 return sh;
276}
277
NeilBrowne4e11e32010-06-16 16:45:16 +1000278static void shrink_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
280 struct page *p;
281 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000282 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
NeilBrowne4e11e32010-06-16 16:45:16 +1000284 for (i = 0; i < num ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 p = sh->dev[i].page;
286 if (!p)
287 continue;
288 sh->dev[i].page = NULL;
NeilBrown2d1f3b52006-01-06 00:20:31 -0800289 put_page(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 }
291}
292
NeilBrowne4e11e32010-06-16 16:45:16 +1000293static int grow_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000296 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
NeilBrowne4e11e32010-06-16 16:45:16 +1000298 for (i = 0; i < num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 struct page *page;
300
301 if (!(page = alloc_page(GFP_KERNEL))) {
302 return 1;
303 }
304 sh->dev[i].page = page;
305 }
306 return 0;
307}
308
NeilBrown784052e2009-03-31 15:19:07 +1100309static void raid5_build_block(struct stripe_head *sh, int i, int previous);
NeilBrown911d4ee2009-03-31 14:39:38 +1100310static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
311 struct stripe_head *sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
NeilBrownb5663ba2009-03-31 14:39:38 +1100313static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
315 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800316 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200318 BUG_ON(atomic_read(&sh->count) != 0);
319 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
Dan Williams600aa102008-06-28 08:32:05 +1000320 BUG_ON(stripe_operations_active(sh));
Dan Williamsd84e0f12007-01-02 13:52:30 -0700321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 CHECK_DEVLOCK();
Dan Williams45b42332007-07-09 11:56:43 -0700323 pr_debug("init_stripe called, stripe %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 (unsigned long long)sh->sector);
325
326 remove_hash(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -0700327
NeilBrown86b42c72009-03-31 15:19:03 +1100328 sh->generation = conf->generation - previous;
NeilBrownb5663ba2009-03-31 14:39:38 +1100329 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 sh->sector = sector;
NeilBrown911d4ee2009-03-31 14:39:38 +1100331 stripe_set_idx(sector, conf, previous, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 sh->state = 0;
333
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800334
335 for (i = sh->disks; i--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 struct r5dev *dev = &sh->dev[i];
337
Dan Williamsd84e0f12007-01-02 13:52:30 -0700338 if (dev->toread || dev->read || dev->towrite || dev->written ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 test_bit(R5_LOCKED, &dev->flags)) {
Dan Williamsd84e0f12007-01-02 13:52:30 -0700340 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 (unsigned long long)sh->sector, i, dev->toread,
Dan Williamsd84e0f12007-01-02 13:52:30 -0700342 dev->read, dev->towrite, dev->written,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 test_bit(R5_LOCKED, &dev->flags));
344 BUG();
345 }
346 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +1100347 raid5_build_block(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 }
349 insert_hash(conf, sh);
350}
351
NeilBrown86b42c72009-03-31 15:19:03 +1100352static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector,
353 short generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
355 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -0800356 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 CHECK_DEVLOCK();
Dan Williams45b42332007-07-09 11:56:43 -0700359 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
NeilBrownfccddba2006-01-06 00:20:33 -0800360 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
NeilBrown86b42c72009-03-31 15:19:03 +1100361 if (sh->sector == sector && sh->generation == generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return sh;
Dan Williams45b42332007-07-09 11:56:43 -0700363 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 return NULL;
365}
366
NeilBrown674806d2010-06-16 17:17:53 +1000367/*
368 * Need to check if array has failed when deciding whether to:
369 * - start an array
370 * - remove non-faulty devices
371 * - add a spare
372 * - allow a reshape
373 * This determination is simple when no reshape is happening.
374 * However if there is a reshape, we need to carefully check
375 * both the before and after sections.
376 * This is because some failed devices may only affect one
377 * of the two sections, and some non-in_sync devices may
378 * be insync in the section most affected by failed devices.
379 */
380static int has_failed(raid5_conf_t *conf)
381{
382 int degraded;
383 int i;
384 if (conf->mddev->reshape_position == MaxSector)
385 return conf->mddev->degraded > conf->max_degraded;
386
387 rcu_read_lock();
388 degraded = 0;
389 for (i = 0; i < conf->previous_raid_disks; i++) {
390 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
391 if (!rdev || test_bit(Faulty, &rdev->flags))
392 degraded++;
393 else if (test_bit(In_sync, &rdev->flags))
394 ;
395 else
396 /* not in-sync or faulty.
397 * If the reshape increases the number of devices,
398 * this is being recovered by the reshape, so
399 * this 'previous' section is not in_sync.
400 * If the number of devices is being reduced however,
401 * the device can only be part of the array if
402 * we are reverting a reshape, so this section will
403 * be in-sync.
404 */
405 if (conf->raid_disks >= conf->previous_raid_disks)
406 degraded++;
407 }
408 rcu_read_unlock();
409 if (degraded > conf->max_degraded)
410 return 1;
411 rcu_read_lock();
412 degraded = 0;
413 for (i = 0; i < conf->raid_disks; i++) {
414 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
415 if (!rdev || test_bit(Faulty, &rdev->flags))
416 degraded++;
417 else if (test_bit(In_sync, &rdev->flags))
418 ;
419 else
420 /* not in-sync or faulty.
421 * If reshape increases the number of devices, this
422 * section has already been recovered, else it
423 * almost certainly hasn't.
424 */
425 if (conf->raid_disks <= conf->previous_raid_disks)
426 degraded++;
427 }
428 rcu_read_unlock();
429 if (degraded > conf->max_degraded)
430 return 1;
431 return 0;
432}
433
NeilBrownb5663ba2009-03-31 14:39:38 +1100434static struct stripe_head *
435get_active_stripe(raid5_conf_t *conf, sector_t sector,
NeilBrowna8c906c2009-06-09 14:39:59 +1000436 int previous, int noblock, int noquiesce)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 struct stripe_head *sh;
439
Dan Williams45b42332007-07-09 11:56:43 -0700440 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
442 spin_lock_irq(&conf->device_lock);
443
444 do {
NeilBrown72626682005-09-09 16:23:54 -0700445 wait_event_lock_irq(conf->wait_for_stripe,
NeilBrowna8c906c2009-06-09 14:39:59 +1000446 conf->quiesce == 0 || noquiesce,
NeilBrown72626682005-09-09 16:23:54 -0700447 conf->device_lock, /* nothing */);
NeilBrown86b42c72009-03-31 15:19:03 +1100448 sh = __find_stripe(conf, sector, conf->generation - previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 if (!sh) {
450 if (!conf->inactive_blocked)
451 sh = get_free_stripe(conf);
452 if (noblock && sh == NULL)
453 break;
454 if (!sh) {
455 conf->inactive_blocked = 1;
456 wait_event_lock_irq(conf->wait_for_stripe,
457 !list_empty(&conf->inactive_list) &&
NeilBrown50368052005-12-12 02:39:17 -0800458 (atomic_read(&conf->active_stripes)
459 < (conf->max_nr_stripes *3/4)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 || !conf->inactive_blocked),
461 conf->device_lock,
NeilBrown7c13edc2011-04-18 18:25:43 +1000462 );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 conf->inactive_blocked = 0;
464 } else
NeilBrownb5663ba2009-03-31 14:39:38 +1100465 init_stripe(sh, sector, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 } else {
467 if (atomic_read(&sh->count)) {
NeilBrownab69ae12009-03-31 15:26:47 +1100468 BUG_ON(!list_empty(&sh->lru)
469 && !test_bit(STRIPE_EXPANDING, &sh->state));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 } else {
471 if (!test_bit(STRIPE_HANDLE, &sh->state))
472 atomic_inc(&conf->active_stripes);
NeilBrownff4e8d92006-07-10 04:44:16 -0700473 if (list_empty(&sh->lru) &&
474 !test_bit(STRIPE_EXPANDING, &sh->state))
NeilBrown16a53ec2006-06-26 00:27:38 -0700475 BUG();
476 list_del_init(&sh->lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
478 }
479 } while (sh == NULL);
480
481 if (sh)
482 atomic_inc(&sh->count);
483
484 spin_unlock_irq(&conf->device_lock);
485 return sh;
486}
487
NeilBrown6712ecf2007-09-27 12:47:43 +0200488static void
489raid5_end_read_request(struct bio *bi, int error);
490static void
491raid5_end_write_request(struct bio *bi, int error);
Dan Williams91c00922007-01-02 13:52:30 -0700492
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000493static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
Dan Williams91c00922007-01-02 13:52:30 -0700494{
495 raid5_conf_t *conf = sh->raid_conf;
496 int i, disks = sh->disks;
497
498 might_sleep();
499
500 for (i = disks; i--; ) {
501 int rw;
502 struct bio *bi;
503 mdk_rdev_t *rdev;
Tejun Heoe9c74692010-09-03 11:56:18 +0200504 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
505 if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags))
506 rw = WRITE_FUA;
507 else
508 rw = WRITE;
509 } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
Dan Williams91c00922007-01-02 13:52:30 -0700510 rw = READ;
511 else
512 continue;
513
514 bi = &sh->dev[i].req;
515
516 bi->bi_rw = rw;
517 if (rw == WRITE)
518 bi->bi_end_io = raid5_end_write_request;
519 else
520 bi->bi_end_io = raid5_end_read_request;
521
522 rcu_read_lock();
523 rdev = rcu_dereference(conf->disks[i].rdev);
524 if (rdev && test_bit(Faulty, &rdev->flags))
525 rdev = NULL;
526 if (rdev)
527 atomic_inc(&rdev->nr_pending);
528 rcu_read_unlock();
529
530 if (rdev) {
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000531 if (s->syncing || s->expanding || s->expanded)
Dan Williams91c00922007-01-02 13:52:30 -0700532 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
533
Dan Williams2b7497f2008-06-28 08:31:52 +1000534 set_bit(STRIPE_IO_STARTED, &sh->state);
535
Dan Williams91c00922007-01-02 13:52:30 -0700536 bi->bi_bdev = rdev->bdev;
537 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700538 __func__, (unsigned long long)sh->sector,
Dan Williams91c00922007-01-02 13:52:30 -0700539 bi->bi_rw, i);
540 atomic_inc(&sh->count);
541 bi->bi_sector = sh->sector + rdev->data_offset;
542 bi->bi_flags = 1 << BIO_UPTODATE;
543 bi->bi_vcnt = 1;
544 bi->bi_max_vecs = 1;
545 bi->bi_idx = 0;
546 bi->bi_io_vec = &sh->dev[i].vec;
547 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
548 bi->bi_io_vec[0].bv_offset = 0;
549 bi->bi_size = STRIPE_SIZE;
550 bi->bi_next = NULL;
551 if (rw == WRITE &&
552 test_bit(R5_ReWrite, &sh->dev[i].flags))
553 atomic_add(STRIPE_SECTORS,
554 &rdev->corrected_errors);
555 generic_make_request(bi);
556 } else {
557 if (rw == WRITE)
558 set_bit(STRIPE_DEGRADED, &sh->state);
559 pr_debug("skip op %ld on disc %d for sector %llu\n",
560 bi->bi_rw, i, (unsigned long long)sh->sector);
561 clear_bit(R5_LOCKED, &sh->dev[i].flags);
562 set_bit(STRIPE_HANDLE, &sh->state);
563 }
564 }
565}
566
567static struct dma_async_tx_descriptor *
568async_copy_data(int frombio, struct bio *bio, struct page *page,
569 sector_t sector, struct dma_async_tx_descriptor *tx)
570{
571 struct bio_vec *bvl;
572 struct page *bio_page;
573 int i;
574 int page_offset;
Dan Williamsa08abd82009-06-03 11:43:59 -0700575 struct async_submit_ctl submit;
Dan Williams0403e382009-09-08 17:42:50 -0700576 enum async_tx_flags flags = 0;
Dan Williams91c00922007-01-02 13:52:30 -0700577
578 if (bio->bi_sector >= sector)
579 page_offset = (signed)(bio->bi_sector - sector) * 512;
580 else
581 page_offset = (signed)(sector - bio->bi_sector) * -512;
Dan Williamsa08abd82009-06-03 11:43:59 -0700582
Dan Williams0403e382009-09-08 17:42:50 -0700583 if (frombio)
584 flags |= ASYNC_TX_FENCE;
585 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
586
Dan Williams91c00922007-01-02 13:52:30 -0700587 bio_for_each_segment(bvl, bio, i) {
588 int len = bio_iovec_idx(bio, i)->bv_len;
589 int clen;
590 int b_offset = 0;
591
592 if (page_offset < 0) {
593 b_offset = -page_offset;
594 page_offset += b_offset;
595 len -= b_offset;
596 }
597
598 if (len > 0 && page_offset + len > STRIPE_SIZE)
599 clen = STRIPE_SIZE - page_offset;
600 else
601 clen = len;
602
603 if (clen > 0) {
604 b_offset += bio_iovec_idx(bio, i)->bv_offset;
605 bio_page = bio_iovec_idx(bio, i)->bv_page;
606 if (frombio)
607 tx = async_memcpy(page, bio_page, page_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700608 b_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700609 else
610 tx = async_memcpy(bio_page, page, b_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700611 page_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700612 }
Dan Williamsa08abd82009-06-03 11:43:59 -0700613 /* chain the operations */
614 submit.depend_tx = tx;
615
Dan Williams91c00922007-01-02 13:52:30 -0700616 if (clen < len) /* hit end of page */
617 break;
618 page_offset += len;
619 }
620
621 return tx;
622}
623
624static void ops_complete_biofill(void *stripe_head_ref)
625{
626 struct stripe_head *sh = stripe_head_ref;
627 struct bio *return_bi = NULL;
628 raid5_conf_t *conf = sh->raid_conf;
Dan Williamse4d84902007-09-24 10:06:13 -0700629 int i;
Dan Williams91c00922007-01-02 13:52:30 -0700630
Harvey Harrisone46b2722008-04-28 02:15:50 -0700631 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700632 (unsigned long long)sh->sector);
633
634 /* clear completed biofills */
Dan Williams83de75c2008-06-28 08:31:58 +1000635 spin_lock_irq(&conf->device_lock);
Dan Williams91c00922007-01-02 13:52:30 -0700636 for (i = sh->disks; i--; ) {
637 struct r5dev *dev = &sh->dev[i];
Dan Williams91c00922007-01-02 13:52:30 -0700638
639 /* acknowledge completion of a biofill operation */
Dan Williamse4d84902007-09-24 10:06:13 -0700640 /* and check if we need to reply to a read request,
641 * new R5_Wantfill requests are held off until
Dan Williams83de75c2008-06-28 08:31:58 +1000642 * !STRIPE_BIOFILL_RUN
Dan Williamse4d84902007-09-24 10:06:13 -0700643 */
644 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -0700645 struct bio *rbi, *rbi2;
Dan Williams91c00922007-01-02 13:52:30 -0700646
Dan Williams91c00922007-01-02 13:52:30 -0700647 BUG_ON(!dev->read);
648 rbi = dev->read;
649 dev->read = NULL;
650 while (rbi && rbi->bi_sector <
651 dev->sector + STRIPE_SECTORS) {
652 rbi2 = r5_next_bio(rbi, dev->sector);
Jens Axboe960e7392008-08-15 10:41:18 +0200653 if (!raid5_dec_bi_phys_segments(rbi)) {
Dan Williams91c00922007-01-02 13:52:30 -0700654 rbi->bi_next = return_bi;
655 return_bi = rbi;
656 }
Dan Williams91c00922007-01-02 13:52:30 -0700657 rbi = rbi2;
658 }
659 }
660 }
Dan Williams83de75c2008-06-28 08:31:58 +1000661 spin_unlock_irq(&conf->device_lock);
662 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -0700663
664 return_io(return_bi);
665
Dan Williamse4d84902007-09-24 10:06:13 -0700666 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -0700667 release_stripe(sh);
668}
669
670static void ops_run_biofill(struct stripe_head *sh)
671{
672 struct dma_async_tx_descriptor *tx = NULL;
673 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsa08abd82009-06-03 11:43:59 -0700674 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700675 int i;
676
Harvey Harrisone46b2722008-04-28 02:15:50 -0700677 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700678 (unsigned long long)sh->sector);
679
680 for (i = sh->disks; i--; ) {
681 struct r5dev *dev = &sh->dev[i];
682 if (test_bit(R5_Wantfill, &dev->flags)) {
683 struct bio *rbi;
684 spin_lock_irq(&conf->device_lock);
685 dev->read = rbi = dev->toread;
686 dev->toread = NULL;
687 spin_unlock_irq(&conf->device_lock);
688 while (rbi && rbi->bi_sector <
689 dev->sector + STRIPE_SECTORS) {
690 tx = async_copy_data(0, rbi, dev->page,
691 dev->sector, tx);
692 rbi = r5_next_bio(rbi, dev->sector);
693 }
694 }
695 }
696
697 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -0700698 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
699 async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -0700700}
701
Dan Williams4e7d2c02009-08-29 19:13:11 -0700702static void mark_target_uptodate(struct stripe_head *sh, int target)
703{
704 struct r5dev *tgt;
705
706 if (target < 0)
707 return;
708
709 tgt = &sh->dev[target];
710 set_bit(R5_UPTODATE, &tgt->flags);
711 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
712 clear_bit(R5_Wantcompute, &tgt->flags);
713}
714
Dan Williamsac6b53b2009-07-14 13:40:19 -0700715static void ops_complete_compute(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -0700716{
717 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -0700718
Harvey Harrisone46b2722008-04-28 02:15:50 -0700719 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700720 (unsigned long long)sh->sector);
721
Dan Williamsac6b53b2009-07-14 13:40:19 -0700722 /* mark the computed target(s) as uptodate */
Dan Williams4e7d2c02009-08-29 19:13:11 -0700723 mark_target_uptodate(sh, sh->ops.target);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700724 mark_target_uptodate(sh, sh->ops.target2);
Dan Williams4e7d2c02009-08-29 19:13:11 -0700725
Dan Williamsecc65c92008-06-28 08:31:57 +1000726 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
727 if (sh->check_state == check_state_compute_run)
728 sh->check_state = check_state_compute_result;
Dan Williams91c00922007-01-02 13:52:30 -0700729 set_bit(STRIPE_HANDLE, &sh->state);
730 release_stripe(sh);
731}
732
Dan Williamsd6f38f32009-07-14 11:50:52 -0700733/* return a pointer to the address conversion region of the scribble buffer */
734static addr_conv_t *to_addr_conv(struct stripe_head *sh,
735 struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -0700736{
Dan Williamsd6f38f32009-07-14 11:50:52 -0700737 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
738}
739
740static struct dma_async_tx_descriptor *
741ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
742{
Dan Williams91c00922007-01-02 13:52:30 -0700743 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -0700744 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -0700745 int target = sh->ops.target;
746 struct r5dev *tgt = &sh->dev[target];
747 struct page *xor_dest = tgt->page;
748 int count = 0;
749 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -0700750 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700751 int i;
752
753 pr_debug("%s: stripe %llu block: %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700754 __func__, (unsigned long long)sh->sector, target);
Dan Williams91c00922007-01-02 13:52:30 -0700755 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
756
757 for (i = disks; i--; )
758 if (i != target)
759 xor_srcs[count++] = sh->dev[i].page;
760
761 atomic_inc(&sh->count);
762
Dan Williams0403e382009-09-08 17:42:50 -0700763 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700764 ops_complete_compute, sh, to_addr_conv(sh, percpu));
Dan Williams91c00922007-01-02 13:52:30 -0700765 if (unlikely(count == 1))
Dan Williamsa08abd82009-06-03 11:43:59 -0700766 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700767 else
Dan Williamsa08abd82009-06-03 11:43:59 -0700768 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700769
Dan Williams91c00922007-01-02 13:52:30 -0700770 return tx;
771}
772
Dan Williamsac6b53b2009-07-14 13:40:19 -0700773/* set_syndrome_sources - populate source buffers for gen_syndrome
774 * @srcs - (struct page *) array of size sh->disks
775 * @sh - stripe_head to parse
776 *
777 * Populates srcs in proper layout order for the stripe and returns the
778 * 'count' of sources to be used in a call to async_gen_syndrome. The P
779 * destination buffer is recorded in srcs[count] and the Q destination
780 * is recorded in srcs[count+1]].
781 */
782static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
783{
784 int disks = sh->disks;
785 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
786 int d0_idx = raid6_d0(sh);
787 int count;
788 int i;
789
790 for (i = 0; i < disks; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +1100791 srcs[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700792
793 count = 0;
794 i = d0_idx;
795 do {
796 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
797
798 srcs[slot] = sh->dev[i].page;
799 i = raid6_next_disk(i, disks);
800 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700801
NeilBrowne4424fe2009-10-16 16:27:34 +1100802 return syndrome_disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700803}
804
805static struct dma_async_tx_descriptor *
806ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
807{
808 int disks = sh->disks;
809 struct page **blocks = percpu->scribble;
810 int target;
811 int qd_idx = sh->qd_idx;
812 struct dma_async_tx_descriptor *tx;
813 struct async_submit_ctl submit;
814 struct r5dev *tgt;
815 struct page *dest;
816 int i;
817 int count;
818
819 if (sh->ops.target < 0)
820 target = sh->ops.target2;
821 else if (sh->ops.target2 < 0)
822 target = sh->ops.target;
823 else
824 /* we should only have one valid target */
825 BUG();
826 BUG_ON(target < 0);
827 pr_debug("%s: stripe %llu block: %d\n",
828 __func__, (unsigned long long)sh->sector, target);
829
830 tgt = &sh->dev[target];
831 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
832 dest = tgt->page;
833
834 atomic_inc(&sh->count);
835
836 if (target == qd_idx) {
837 count = set_syndrome_sources(blocks, sh);
838 blocks[count] = NULL; /* regenerating p is not necessary */
839 BUG_ON(blocks[count+1] != dest); /* q should already be set */
Dan Williams0403e382009-09-08 17:42:50 -0700840 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
841 ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700842 to_addr_conv(sh, percpu));
843 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
844 } else {
845 /* Compute any data- or p-drive using XOR */
846 count = 0;
847 for (i = disks; i-- ; ) {
848 if (i == target || i == qd_idx)
849 continue;
850 blocks[count++] = sh->dev[i].page;
851 }
852
Dan Williams0403e382009-09-08 17:42:50 -0700853 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
854 NULL, ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700855 to_addr_conv(sh, percpu));
856 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
857 }
858
859 return tx;
860}
861
862static struct dma_async_tx_descriptor *
863ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
864{
865 int i, count, disks = sh->disks;
866 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
867 int d0_idx = raid6_d0(sh);
868 int faila = -1, failb = -1;
869 int target = sh->ops.target;
870 int target2 = sh->ops.target2;
871 struct r5dev *tgt = &sh->dev[target];
872 struct r5dev *tgt2 = &sh->dev[target2];
873 struct dma_async_tx_descriptor *tx;
874 struct page **blocks = percpu->scribble;
875 struct async_submit_ctl submit;
876
877 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
878 __func__, (unsigned long long)sh->sector, target, target2);
879 BUG_ON(target < 0 || target2 < 0);
880 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
881 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
882
Dan Williams6c910a72009-09-16 12:24:54 -0700883 /* we need to open-code set_syndrome_sources to handle the
Dan Williamsac6b53b2009-07-14 13:40:19 -0700884 * slot number conversion for 'faila' and 'failb'
885 */
886 for (i = 0; i < disks ; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +1100887 blocks[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700888 count = 0;
889 i = d0_idx;
890 do {
891 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
892
893 blocks[slot] = sh->dev[i].page;
894
895 if (i == target)
896 faila = slot;
897 if (i == target2)
898 failb = slot;
899 i = raid6_next_disk(i, disks);
900 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700901
902 BUG_ON(faila == failb);
903 if (failb < faila)
904 swap(faila, failb);
905 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
906 __func__, (unsigned long long)sh->sector, faila, failb);
907
908 atomic_inc(&sh->count);
909
910 if (failb == syndrome_disks+1) {
911 /* Q disk is one of the missing disks */
912 if (faila == syndrome_disks) {
913 /* Missing P+Q, just recompute */
Dan Williams0403e382009-09-08 17:42:50 -0700914 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
915 ops_complete_compute, sh,
916 to_addr_conv(sh, percpu));
NeilBrowne4424fe2009-10-16 16:27:34 +1100917 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700918 STRIPE_SIZE, &submit);
919 } else {
920 struct page *dest;
921 int data_target;
922 int qd_idx = sh->qd_idx;
923
924 /* Missing D+Q: recompute D from P, then recompute Q */
925 if (target == qd_idx)
926 data_target = target2;
927 else
928 data_target = target;
929
930 count = 0;
931 for (i = disks; i-- ; ) {
932 if (i == data_target || i == qd_idx)
933 continue;
934 blocks[count++] = sh->dev[i].page;
935 }
936 dest = sh->dev[data_target].page;
Dan Williams0403e382009-09-08 17:42:50 -0700937 init_async_submit(&submit,
938 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
939 NULL, NULL, NULL,
940 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -0700941 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
942 &submit);
943
944 count = set_syndrome_sources(blocks, sh);
Dan Williams0403e382009-09-08 17:42:50 -0700945 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
946 ops_complete_compute, sh,
947 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -0700948 return async_gen_syndrome(blocks, 0, count+2,
949 STRIPE_SIZE, &submit);
950 }
Dan Williamsac6b53b2009-07-14 13:40:19 -0700951 } else {
Dan Williams6c910a72009-09-16 12:24:54 -0700952 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
953 ops_complete_compute, sh,
954 to_addr_conv(sh, percpu));
955 if (failb == syndrome_disks) {
956 /* We're missing D+P. */
957 return async_raid6_datap_recov(syndrome_disks+2,
958 STRIPE_SIZE, faila,
959 blocks, &submit);
960 } else {
961 /* We're missing D+D. */
962 return async_raid6_2data_recov(syndrome_disks+2,
963 STRIPE_SIZE, faila, failb,
964 blocks, &submit);
965 }
Dan Williamsac6b53b2009-07-14 13:40:19 -0700966 }
967}
968
969
Dan Williams91c00922007-01-02 13:52:30 -0700970static void ops_complete_prexor(void *stripe_head_ref)
971{
972 struct stripe_head *sh = stripe_head_ref;
973
Harvey Harrisone46b2722008-04-28 02:15:50 -0700974 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700975 (unsigned long long)sh->sector);
Dan Williams91c00922007-01-02 13:52:30 -0700976}
977
978static struct dma_async_tx_descriptor *
Dan Williamsd6f38f32009-07-14 11:50:52 -0700979ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
980 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -0700981{
Dan Williams91c00922007-01-02 13:52:30 -0700982 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -0700983 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -0700984 int count = 0, pd_idx = sh->pd_idx, i;
Dan Williamsa08abd82009-06-03 11:43:59 -0700985 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700986
987 /* existing parity data subtracted */
988 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
989
Harvey Harrisone46b2722008-04-28 02:15:50 -0700990 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700991 (unsigned long long)sh->sector);
992
993 for (i = disks; i--; ) {
994 struct r5dev *dev = &sh->dev[i];
995 /* Only process blocks that are known to be uptodate */
Dan Williamsd8ee0722008-06-28 08:32:06 +1000996 if (test_bit(R5_Wantdrain, &dev->flags))
Dan Williams91c00922007-01-02 13:52:30 -0700997 xor_srcs[count++] = dev->page;
998 }
999
Dan Williams0403e382009-09-08 17:42:50 -07001000 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001001 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001002 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001003
1004 return tx;
1005}
1006
1007static struct dma_async_tx_descriptor *
Dan Williamsd8ee0722008-06-28 08:32:06 +10001008ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001009{
1010 int disks = sh->disks;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001011 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001012
Harvey Harrisone46b2722008-04-28 02:15:50 -07001013 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001014 (unsigned long long)sh->sector);
1015
1016 for (i = disks; i--; ) {
1017 struct r5dev *dev = &sh->dev[i];
1018 struct bio *chosen;
Dan Williams91c00922007-01-02 13:52:30 -07001019
Dan Williamsd8ee0722008-06-28 08:32:06 +10001020 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001021 struct bio *wbi;
1022
1023 spin_lock(&sh->lock);
1024 chosen = dev->towrite;
1025 dev->towrite = NULL;
1026 BUG_ON(dev->written);
1027 wbi = dev->written = chosen;
1028 spin_unlock(&sh->lock);
1029
1030 while (wbi && wbi->bi_sector <
1031 dev->sector + STRIPE_SECTORS) {
Tejun Heoe9c74692010-09-03 11:56:18 +02001032 if (wbi->bi_rw & REQ_FUA)
1033 set_bit(R5_WantFUA, &dev->flags);
Dan Williams91c00922007-01-02 13:52:30 -07001034 tx = async_copy_data(1, wbi, dev->page,
1035 dev->sector, tx);
1036 wbi = r5_next_bio(wbi, dev->sector);
1037 }
1038 }
1039 }
1040
1041 return tx;
1042}
1043
Dan Williamsac6b53b2009-07-14 13:40:19 -07001044static void ops_complete_reconstruct(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001045{
1046 struct stripe_head *sh = stripe_head_ref;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001047 int disks = sh->disks;
1048 int pd_idx = sh->pd_idx;
1049 int qd_idx = sh->qd_idx;
1050 int i;
Tejun Heoe9c74692010-09-03 11:56:18 +02001051 bool fua = false;
Dan Williams91c00922007-01-02 13:52:30 -07001052
Harvey Harrisone46b2722008-04-28 02:15:50 -07001053 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001054 (unsigned long long)sh->sector);
1055
Tejun Heoe9c74692010-09-03 11:56:18 +02001056 for (i = disks; i--; )
1057 fua |= test_bit(R5_WantFUA, &sh->dev[i].flags);
1058
Dan Williams91c00922007-01-02 13:52:30 -07001059 for (i = disks; i--; ) {
1060 struct r5dev *dev = &sh->dev[i];
Dan Williamsac6b53b2009-07-14 13:40:19 -07001061
Tejun Heoe9c74692010-09-03 11:56:18 +02001062 if (dev->written || i == pd_idx || i == qd_idx) {
Dan Williams91c00922007-01-02 13:52:30 -07001063 set_bit(R5_UPTODATE, &dev->flags);
Tejun Heoe9c74692010-09-03 11:56:18 +02001064 if (fua)
1065 set_bit(R5_WantFUA, &dev->flags);
1066 }
Dan Williams91c00922007-01-02 13:52:30 -07001067 }
1068
Dan Williamsd8ee0722008-06-28 08:32:06 +10001069 if (sh->reconstruct_state == reconstruct_state_drain_run)
1070 sh->reconstruct_state = reconstruct_state_drain_result;
1071 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1072 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1073 else {
1074 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1075 sh->reconstruct_state = reconstruct_state_result;
1076 }
Dan Williams91c00922007-01-02 13:52:30 -07001077
1078 set_bit(STRIPE_HANDLE, &sh->state);
1079 release_stripe(sh);
1080}
1081
1082static void
Dan Williamsac6b53b2009-07-14 13:40:19 -07001083ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1084 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001085{
Dan Williams91c00922007-01-02 13:52:30 -07001086 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001087 struct page **xor_srcs = percpu->scribble;
Dan Williamsa08abd82009-06-03 11:43:59 -07001088 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001089 int count = 0, pd_idx = sh->pd_idx, i;
1090 struct page *xor_dest;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001091 int prexor = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001092 unsigned long flags;
Dan Williams91c00922007-01-02 13:52:30 -07001093
Harvey Harrisone46b2722008-04-28 02:15:50 -07001094 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001095 (unsigned long long)sh->sector);
1096
1097 /* check if prexor is active which means only process blocks
1098 * that are part of a read-modify-write (written)
1099 */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001100 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1101 prexor = 1;
Dan Williams91c00922007-01-02 13:52:30 -07001102 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1103 for (i = disks; i--; ) {
1104 struct r5dev *dev = &sh->dev[i];
1105 if (dev->written)
1106 xor_srcs[count++] = dev->page;
1107 }
1108 } else {
1109 xor_dest = sh->dev[pd_idx].page;
1110 for (i = disks; i--; ) {
1111 struct r5dev *dev = &sh->dev[i];
1112 if (i != pd_idx)
1113 xor_srcs[count++] = dev->page;
1114 }
1115 }
1116
Dan Williams91c00922007-01-02 13:52:30 -07001117 /* 1/ if we prexor'd then the dest is reused as a source
1118 * 2/ if we did not prexor then we are redoing the parity
1119 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1120 * for the synchronous xor case
1121 */
Dan Williams88ba2aa2009-04-09 16:16:18 -07001122 flags = ASYNC_TX_ACK |
Dan Williams91c00922007-01-02 13:52:30 -07001123 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1124
1125 atomic_inc(&sh->count);
1126
Dan Williamsac6b53b2009-07-14 13:40:19 -07001127 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001128 to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001129 if (unlikely(count == 1))
1130 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1131 else
1132 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001133}
1134
Dan Williamsac6b53b2009-07-14 13:40:19 -07001135static void
1136ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1137 struct dma_async_tx_descriptor *tx)
1138{
1139 struct async_submit_ctl submit;
1140 struct page **blocks = percpu->scribble;
1141 int count;
1142
1143 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1144
1145 count = set_syndrome_sources(blocks, sh);
1146
1147 atomic_inc(&sh->count);
1148
1149 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1150 sh, to_addr_conv(sh, percpu));
1151 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001152}
1153
1154static void ops_complete_check(void *stripe_head_ref)
1155{
1156 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001157
Harvey Harrisone46b2722008-04-28 02:15:50 -07001158 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001159 (unsigned long long)sh->sector);
1160
Dan Williamsecc65c92008-06-28 08:31:57 +10001161 sh->check_state = check_state_check_result;
Dan Williams91c00922007-01-02 13:52:30 -07001162 set_bit(STRIPE_HANDLE, &sh->state);
1163 release_stripe(sh);
1164}
1165
Dan Williamsac6b53b2009-07-14 13:40:19 -07001166static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001167{
Dan Williams91c00922007-01-02 13:52:30 -07001168 int disks = sh->disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001169 int pd_idx = sh->pd_idx;
1170 int qd_idx = sh->qd_idx;
1171 struct page *xor_dest;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001172 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001173 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001174 struct async_submit_ctl submit;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001175 int count;
1176 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001177
Harvey Harrisone46b2722008-04-28 02:15:50 -07001178 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001179 (unsigned long long)sh->sector);
1180
Dan Williamsac6b53b2009-07-14 13:40:19 -07001181 count = 0;
1182 xor_dest = sh->dev[pd_idx].page;
1183 xor_srcs[count++] = xor_dest;
Dan Williams91c00922007-01-02 13:52:30 -07001184 for (i = disks; i--; ) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001185 if (i == pd_idx || i == qd_idx)
1186 continue;
1187 xor_srcs[count++] = sh->dev[i].page;
Dan Williams91c00922007-01-02 13:52:30 -07001188 }
1189
Dan Williamsd6f38f32009-07-14 11:50:52 -07001190 init_async_submit(&submit, 0, NULL, NULL, NULL,
1191 to_addr_conv(sh, percpu));
Dan Williams099f53c2009-04-08 14:28:37 -07001192 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07001193 &sh->ops.zero_sum_result, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001194
Dan Williams91c00922007-01-02 13:52:30 -07001195 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001196 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1197 tx = async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001198}
1199
Dan Williamsac6b53b2009-07-14 13:40:19 -07001200static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1201{
1202 struct page **srcs = percpu->scribble;
1203 struct async_submit_ctl submit;
1204 int count;
1205
1206 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1207 (unsigned long long)sh->sector, checkp);
1208
1209 count = set_syndrome_sources(srcs, sh);
1210 if (!checkp)
1211 srcs[count] = NULL;
1212
1213 atomic_inc(&sh->count);
1214 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1215 sh, to_addr_conv(sh, percpu));
1216 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1217 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
1218}
1219
Dan Williams417b8d42009-10-16 16:25:22 +11001220static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
Dan Williams91c00922007-01-02 13:52:30 -07001221{
1222 int overlap_clear = 0, i, disks = sh->disks;
1223 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001224 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001225 int level = conf->level;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001226 struct raid5_percpu *percpu;
1227 unsigned long cpu;
Dan Williams91c00922007-01-02 13:52:30 -07001228
Dan Williamsd6f38f32009-07-14 11:50:52 -07001229 cpu = get_cpu();
1230 percpu = per_cpu_ptr(conf->percpu, cpu);
Dan Williams83de75c2008-06-28 08:31:58 +10001231 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
Dan Williams91c00922007-01-02 13:52:30 -07001232 ops_run_biofill(sh);
1233 overlap_clear++;
1234 }
1235
Dan Williams7b3a8712008-06-28 08:32:09 +10001236 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001237 if (level < 6)
1238 tx = ops_run_compute5(sh, percpu);
1239 else {
1240 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1241 tx = ops_run_compute6_1(sh, percpu);
1242 else
1243 tx = ops_run_compute6_2(sh, percpu);
1244 }
1245 /* terminate the chain if reconstruct is not set to be run */
1246 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
Dan Williams7b3a8712008-06-28 08:32:09 +10001247 async_tx_ack(tx);
1248 }
Dan Williams91c00922007-01-02 13:52:30 -07001249
Dan Williams600aa102008-06-28 08:32:05 +10001250 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
Dan Williamsd6f38f32009-07-14 11:50:52 -07001251 tx = ops_run_prexor(sh, percpu, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001252
Dan Williams600aa102008-06-28 08:32:05 +10001253 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
Dan Williamsd8ee0722008-06-28 08:32:06 +10001254 tx = ops_run_biodrain(sh, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001255 overlap_clear++;
1256 }
1257
Dan Williamsac6b53b2009-07-14 13:40:19 -07001258 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1259 if (level < 6)
1260 ops_run_reconstruct5(sh, percpu, tx);
1261 else
1262 ops_run_reconstruct6(sh, percpu, tx);
1263 }
Dan Williams91c00922007-01-02 13:52:30 -07001264
Dan Williamsac6b53b2009-07-14 13:40:19 -07001265 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1266 if (sh->check_state == check_state_run)
1267 ops_run_check_p(sh, percpu);
1268 else if (sh->check_state == check_state_run_q)
1269 ops_run_check_pq(sh, percpu, 0);
1270 else if (sh->check_state == check_state_run_pq)
1271 ops_run_check_pq(sh, percpu, 1);
1272 else
1273 BUG();
1274 }
Dan Williams91c00922007-01-02 13:52:30 -07001275
Dan Williams91c00922007-01-02 13:52:30 -07001276 if (overlap_clear)
1277 for (i = disks; i--; ) {
1278 struct r5dev *dev = &sh->dev[i];
1279 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1280 wake_up(&sh->raid_conf->wait_for_overlap);
1281 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07001282 put_cpu();
Dan Williams91c00922007-01-02 13:52:30 -07001283}
1284
Dan Williams417b8d42009-10-16 16:25:22 +11001285#ifdef CONFIG_MULTICORE_RAID456
1286static void async_run_ops(void *param, async_cookie_t cookie)
1287{
1288 struct stripe_head *sh = param;
1289 unsigned long ops_request = sh->ops.request;
1290
1291 clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state);
1292 wake_up(&sh->ops.wait_for_ops);
1293
1294 __raid_run_ops(sh, ops_request);
1295 release_stripe(sh);
1296}
1297
1298static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
1299{
1300 /* since handle_stripe can be called outside of raid5d context
1301 * we need to ensure sh->ops.request is de-staged before another
1302 * request arrives
1303 */
1304 wait_event(sh->ops.wait_for_ops,
1305 !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state));
1306 sh->ops.request = ops_request;
1307
1308 atomic_inc(&sh->count);
1309 async_schedule(async_run_ops, sh);
1310}
1311#else
1312#define raid_run_ops __raid_run_ops
1313#endif
1314
NeilBrown3f294f42005-11-08 21:39:25 -08001315static int grow_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
1317 struct stripe_head *sh;
NeilBrown3f294f42005-11-08 21:39:25 -08001318 sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL);
1319 if (!sh)
1320 return 0;
NeilBrowne4e11e32010-06-16 16:45:16 +10001321 memset(sh, 0, sizeof(*sh) + (conf->pool_size-1)*sizeof(struct r5dev));
NeilBrown3f294f42005-11-08 21:39:25 -08001322 sh->raid_conf = conf;
1323 spin_lock_init(&sh->lock);
Dan Williams417b8d42009-10-16 16:25:22 +11001324 #ifdef CONFIG_MULTICORE_RAID456
1325 init_waitqueue_head(&sh->ops.wait_for_ops);
1326 #endif
NeilBrown3f294f42005-11-08 21:39:25 -08001327
NeilBrowne4e11e32010-06-16 16:45:16 +10001328 if (grow_buffers(sh)) {
1329 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001330 kmem_cache_free(conf->slab_cache, sh);
1331 return 0;
1332 }
1333 /* we just created an active stripe so... */
1334 atomic_set(&sh->count, 1);
1335 atomic_inc(&conf->active_stripes);
1336 INIT_LIST_HEAD(&sh->lru);
1337 release_stripe(sh);
1338 return 1;
1339}
1340
1341static int grow_stripes(raid5_conf_t *conf, int num)
1342{
Christoph Lametere18b8902006-12-06 20:33:20 -08001343 struct kmem_cache *sc;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001344 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
NeilBrownf4be6b42010-06-01 19:37:25 +10001346 if (conf->mddev->gendisk)
1347 sprintf(conf->cache_name[0],
1348 "raid%d-%s", conf->level, mdname(conf->mddev));
1349 else
1350 sprintf(conf->cache_name[0],
1351 "raid%d-%p", conf->level, conf->mddev);
1352 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1353
NeilBrownad01c9e2006-03-27 01:18:07 -08001354 conf->active_name = 0;
1355 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001357 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 if (!sc)
1359 return 1;
1360 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001361 conf->pool_size = devs;
NeilBrown16a53ec2006-06-26 00:27:38 -07001362 while (num--)
NeilBrown3f294f42005-11-08 21:39:25 -08001363 if (!grow_one_stripe(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 return 0;
1366}
NeilBrown29269552006-03-27 01:18:10 -08001367
Dan Williamsd6f38f32009-07-14 11:50:52 -07001368/**
1369 * scribble_len - return the required size of the scribble region
1370 * @num - total number of disks in the array
1371 *
1372 * The size must be enough to contain:
1373 * 1/ a struct page pointer for each device in the array +2
1374 * 2/ room to convert each entry in (1) to its corresponding dma
1375 * (dma_map_page()) or page (page_address()) address.
1376 *
1377 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1378 * calculate over all devices (not just the data blocks), using zeros in place
1379 * of the P and Q blocks.
1380 */
1381static size_t scribble_len(int num)
1382{
1383 size_t len;
1384
1385 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1386
1387 return len;
1388}
1389
NeilBrownad01c9e2006-03-27 01:18:07 -08001390static int resize_stripes(raid5_conf_t *conf, int newsize)
1391{
1392 /* Make all the stripes able to hold 'newsize' devices.
1393 * New slots in each stripe get 'page' set to a new page.
1394 *
1395 * This happens in stages:
1396 * 1/ create a new kmem_cache and allocate the required number of
1397 * stripe_heads.
1398 * 2/ gather all the old stripe_heads and tranfer the pages across
1399 * to the new stripe_heads. This will have the side effect of
1400 * freezing the array as once all stripe_heads have been collected,
1401 * no IO will be possible. Old stripe heads are freed once their
1402 * pages have been transferred over, and the old kmem_cache is
1403 * freed when all stripes are done.
1404 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1405 * we simple return a failre status - no need to clean anything up.
1406 * 4/ allocate new pages for the new slots in the new stripe_heads.
1407 * If this fails, we don't bother trying the shrink the
1408 * stripe_heads down again, we just leave them as they are.
1409 * As each stripe_head is processed the new one is released into
1410 * active service.
1411 *
1412 * Once step2 is started, we cannot afford to wait for a write,
1413 * so we use GFP_NOIO allocations.
1414 */
1415 struct stripe_head *osh, *nsh;
1416 LIST_HEAD(newstripes);
1417 struct disk_info *ndisks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001418 unsigned long cpu;
Dan Williamsb5470dc2008-06-27 21:44:04 -07001419 int err;
Christoph Lametere18b8902006-12-06 20:33:20 -08001420 struct kmem_cache *sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001421 int i;
1422
1423 if (newsize <= conf->pool_size)
1424 return 0; /* never bother to shrink */
1425
Dan Williamsb5470dc2008-06-27 21:44:04 -07001426 err = md_allow_write(conf->mddev);
1427 if (err)
1428 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08001429
NeilBrownad01c9e2006-03-27 01:18:07 -08001430 /* Step 1 */
1431 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1432 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001433 0, 0, NULL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001434 if (!sc)
1435 return -ENOMEM;
1436
1437 for (i = conf->max_nr_stripes; i; i--) {
1438 nsh = kmem_cache_alloc(sc, GFP_KERNEL);
1439 if (!nsh)
1440 break;
1441
1442 memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev));
1443
1444 nsh->raid_conf = conf;
1445 spin_lock_init(&nsh->lock);
Dan Williams417b8d42009-10-16 16:25:22 +11001446 #ifdef CONFIG_MULTICORE_RAID456
1447 init_waitqueue_head(&nsh->ops.wait_for_ops);
1448 #endif
NeilBrownad01c9e2006-03-27 01:18:07 -08001449
1450 list_add(&nsh->lru, &newstripes);
1451 }
1452 if (i) {
1453 /* didn't get enough, give up */
1454 while (!list_empty(&newstripes)) {
1455 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1456 list_del(&nsh->lru);
1457 kmem_cache_free(sc, nsh);
1458 }
1459 kmem_cache_destroy(sc);
1460 return -ENOMEM;
1461 }
1462 /* Step 2 - Must use GFP_NOIO now.
1463 * OK, we have enough stripes, start collecting inactive
1464 * stripes and copying them over
1465 */
1466 list_for_each_entry(nsh, &newstripes, lru) {
1467 spin_lock_irq(&conf->device_lock);
1468 wait_event_lock_irq(conf->wait_for_stripe,
1469 !list_empty(&conf->inactive_list),
1470 conf->device_lock,
NeilBrown482c0832011-04-18 18:25:42 +10001471 );
NeilBrownad01c9e2006-03-27 01:18:07 -08001472 osh = get_free_stripe(conf);
1473 spin_unlock_irq(&conf->device_lock);
1474 atomic_set(&nsh->count, 1);
1475 for(i=0; i<conf->pool_size; i++)
1476 nsh->dev[i].page = osh->dev[i].page;
1477 for( ; i<newsize; i++)
1478 nsh->dev[i].page = NULL;
1479 kmem_cache_free(conf->slab_cache, osh);
1480 }
1481 kmem_cache_destroy(conf->slab_cache);
1482
1483 /* Step 3.
1484 * At this point, we are holding all the stripes so the array
1485 * is completely stalled, so now is a good time to resize
Dan Williamsd6f38f32009-07-14 11:50:52 -07001486 * conf->disks and the scribble region
NeilBrownad01c9e2006-03-27 01:18:07 -08001487 */
1488 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1489 if (ndisks) {
1490 for (i=0; i<conf->raid_disks; i++)
1491 ndisks[i] = conf->disks[i];
1492 kfree(conf->disks);
1493 conf->disks = ndisks;
1494 } else
1495 err = -ENOMEM;
1496
Dan Williamsd6f38f32009-07-14 11:50:52 -07001497 get_online_cpus();
1498 conf->scribble_len = scribble_len(newsize);
1499 for_each_present_cpu(cpu) {
1500 struct raid5_percpu *percpu;
1501 void *scribble;
1502
1503 percpu = per_cpu_ptr(conf->percpu, cpu);
1504 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1505
1506 if (scribble) {
1507 kfree(percpu->scribble);
1508 percpu->scribble = scribble;
1509 } else {
1510 err = -ENOMEM;
1511 break;
1512 }
1513 }
1514 put_online_cpus();
1515
NeilBrownad01c9e2006-03-27 01:18:07 -08001516 /* Step 4, return new stripes to service */
1517 while(!list_empty(&newstripes)) {
1518 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1519 list_del_init(&nsh->lru);
Dan Williamsd6f38f32009-07-14 11:50:52 -07001520
NeilBrownad01c9e2006-03-27 01:18:07 -08001521 for (i=conf->raid_disks; i < newsize; i++)
1522 if (nsh->dev[i].page == NULL) {
1523 struct page *p = alloc_page(GFP_NOIO);
1524 nsh->dev[i].page = p;
1525 if (!p)
1526 err = -ENOMEM;
1527 }
1528 release_stripe(nsh);
1529 }
1530 /* critical section pass, GFP_NOIO no longer needed */
1531
1532 conf->slab_cache = sc;
1533 conf->active_name = 1-conf->active_name;
1534 conf->pool_size = newsize;
1535 return err;
1536}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
NeilBrown3f294f42005-11-08 21:39:25 -08001538static int drop_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539{
1540 struct stripe_head *sh;
1541
NeilBrown3f294f42005-11-08 21:39:25 -08001542 spin_lock_irq(&conf->device_lock);
1543 sh = get_free_stripe(conf);
1544 spin_unlock_irq(&conf->device_lock);
1545 if (!sh)
1546 return 0;
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02001547 BUG_ON(atomic_read(&sh->count));
NeilBrowne4e11e32010-06-16 16:45:16 +10001548 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001549 kmem_cache_free(conf->slab_cache, sh);
1550 atomic_dec(&conf->active_stripes);
1551 return 1;
1552}
1553
1554static void shrink_stripes(raid5_conf_t *conf)
1555{
1556 while (drop_one_stripe(conf))
1557 ;
1558
NeilBrown29fc7e32006-02-03 03:03:41 -08001559 if (conf->slab_cache)
1560 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 conf->slab_cache = NULL;
1562}
1563
NeilBrown6712ecf2007-09-27 12:47:43 +02001564static void raid5_end_read_request(struct bio * bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565{
NeilBrown99c0fb52009-03-31 14:39:38 +11001566 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001568 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownd6950432006-07-10 04:44:20 -07001570 char b[BDEVNAME_SIZE];
1571 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
1574 for (i=0 ; i<disks; i++)
1575 if (bi == &sh->dev[i].req)
1576 break;
1577
Dan Williams45b42332007-07-09 11:56:43 -07001578 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1579 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 uptodate);
1581 if (i == disks) {
1582 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001583 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 }
1585
1586 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -08001588 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrownd6950432006-07-10 04:44:20 -07001589 rdev = conf->disks[i].rdev;
NeilBrown0c55e022010-05-03 14:09:02 +10001590 printk_rl(KERN_INFO "md/raid:%s: read error corrected"
Bernd Schubert6be9d492008-05-23 13:04:34 -07001591 " (%lu sectors at %llu on %s)\n",
1592 mdname(conf->mddev), STRIPE_SECTORS,
1593 (unsigned long long)(sh->sector
1594 + rdev->data_offset),
1595 bdevname(rdev->bdev, b));
NeilBrown4e5314b2005-11-08 21:39:22 -08001596 clear_bit(R5_ReadError, &sh->dev[i].flags);
1597 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1598 }
NeilBrownba22dcb2005-11-08 21:39:31 -08001599 if (atomic_read(&conf->disks[i].rdev->read_errors))
1600 atomic_set(&conf->disks[i].rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 } else {
NeilBrownd6950432006-07-10 04:44:20 -07001602 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
NeilBrownba22dcb2005-11-08 21:39:31 -08001603 int retry = 0;
NeilBrownd6950432006-07-10 04:44:20 -07001604 rdev = conf->disks[i].rdev;
1605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001607 atomic_inc(&rdev->read_errors);
Gabriele A. Trombetti7b0bb532010-04-28 11:51:17 +10001608 if (conf->mddev->degraded >= conf->max_degraded)
Bernd Schubert6be9d492008-05-23 13:04:34 -07001609 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001610 "md/raid:%s: read error not correctable "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001611 "(sector %llu on %s).\n",
1612 mdname(conf->mddev),
1613 (unsigned long long)(sh->sector
1614 + rdev->data_offset),
1615 bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001616 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
NeilBrown4e5314b2005-11-08 21:39:22 -08001617 /* Oh, no!!! */
Bernd Schubert6be9d492008-05-23 13:04:34 -07001618 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001619 "md/raid:%s: read error NOT corrected!! "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001620 "(sector %llu on %s).\n",
1621 mdname(conf->mddev),
1622 (unsigned long long)(sh->sector
1623 + rdev->data_offset),
1624 bdn);
NeilBrownd6950432006-07-10 04:44:20 -07001625 else if (atomic_read(&rdev->read_errors)
NeilBrownba22dcb2005-11-08 21:39:31 -08001626 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -08001627 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001628 "md/raid:%s: Too many read errors, failing device %s.\n",
NeilBrownd6950432006-07-10 04:44:20 -07001629 mdname(conf->mddev), bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001630 else
1631 retry = 1;
1632 if (retry)
1633 set_bit(R5_ReadError, &sh->dev[i].flags);
1634 else {
NeilBrown4e5314b2005-11-08 21:39:22 -08001635 clear_bit(R5_ReadError, &sh->dev[i].flags);
1636 clear_bit(R5_ReWrite, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001637 md_error(conf->mddev, rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -08001638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 }
1640 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1642 set_bit(STRIPE_HANDLE, &sh->state);
1643 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644}
1645
NeilBrownd710e132008-10-13 11:55:12 +11001646static void raid5_end_write_request(struct bio *bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
NeilBrown99c0fb52009-03-31 14:39:38 +11001648 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001650 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1652
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 for (i=0 ; i<disks; i++)
1654 if (bi == &sh->dev[i].req)
1655 break;
1656
Dan Williams45b42332007-07-09 11:56:43 -07001657 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1659 uptodate);
1660 if (i == disks) {
1661 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001662 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
1664
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 if (!uptodate)
1666 md_error(conf->mddev, conf->disks[i].rdev);
1667
1668 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1669
1670 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1671 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrownc04be0a2006-10-03 01:15:53 -07001672 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
1674
1675
NeilBrown784052e2009-03-31 15:19:07 +11001676static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
NeilBrown784052e2009-03-31 15:19:07 +11001678static void raid5_build_block(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 struct r5dev *dev = &sh->dev[i];
1681
1682 bio_init(&dev->req);
1683 dev->req.bi_io_vec = &dev->vec;
1684 dev->req.bi_vcnt++;
1685 dev->req.bi_max_vecs++;
1686 dev->vec.bv_page = dev->page;
1687 dev->vec.bv_len = STRIPE_SIZE;
1688 dev->vec.bv_offset = 0;
1689
1690 dev->req.bi_sector = sh->sector;
1691 dev->req.bi_private = sh;
1692
1693 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +11001694 dev->sector = compute_blocknr(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695}
1696
1697static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1698{
1699 char b[BDEVNAME_SIZE];
H Hartley Sweeten7b928132010-03-08 16:02:40 +11001700 raid5_conf_t *conf = mddev->private;
NeilBrown0c55e022010-05-03 14:09:02 +10001701 pr_debug("raid456: error called\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
NeilBrownb2d444d2005-11-08 21:39:31 -08001703 if (!test_bit(Faulty, &rdev->flags)) {
NeilBrown850b2b42006-10-03 01:15:46 -07001704 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001705 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1706 unsigned long flags;
1707 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 mddev->degraded++;
NeilBrownc04be0a2006-10-03 01:15:53 -07001709 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 /*
1711 * if recovery was running, make sure it aborts.
1712 */
NeilBrowndfc70642008-05-23 13:04:39 -07001713 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 }
NeilBrownb2d444d2005-11-08 21:39:31 -08001715 set_bit(Faulty, &rdev->flags);
NeilBrownd710e132008-10-13 11:55:12 +11001716 printk(KERN_ALERT
NeilBrown0c55e022010-05-03 14:09:02 +10001717 "md/raid:%s: Disk failure on %s, disabling device.\n"
NeilBrown0c55e022010-05-03 14:09:02 +10001718 "md/raid:%s: Operation continuing on %d devices.\n",
1719 mdname(mddev),
1720 bdevname(rdev->bdev, b),
1721 mdname(mddev),
1722 conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 }
NeilBrown16a53ec2006-06-26 00:27:38 -07001724}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725
1726/*
1727 * Input: a 'big' sector number,
1728 * Output: index of the data and parity disk, and the sector # in them.
1729 */
NeilBrown112bf892009-03-31 14:39:38 +11001730static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11001731 int previous, int *dd_idx,
1732 struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733{
NeilBrown6e3b96e2010-04-23 07:08:28 +10001734 sector_t stripe, stripe2;
NeilBrown35f2a592010-04-20 14:13:34 +10001735 sector_t chunk_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 unsigned int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11001737 int pd_idx, qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001738 int ddf_layout = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 sector_t new_sector;
NeilBrowne183eae2009-03-31 15:20:22 +11001740 int algorithm = previous ? conf->prev_algo
1741 : conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001742 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1743 : conf->chunk_sectors;
NeilBrown112bf892009-03-31 14:39:38 +11001744 int raid_disks = previous ? conf->previous_raid_disks
1745 : conf->raid_disks;
1746 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
1748 /* First compute the information on this sector */
1749
1750 /*
1751 * Compute the chunk number and the sector offset inside the chunk
1752 */
1753 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1754 chunk_number = r_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
1756 /*
1757 * Compute the stripe number
1758 */
NeilBrown35f2a592010-04-20 14:13:34 +10001759 stripe = chunk_number;
1760 *dd_idx = sector_div(stripe, data_disks);
NeilBrown6e3b96e2010-04-23 07:08:28 +10001761 stripe2 = stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 /*
1763 * Select the parity disk based on the user selected algorithm.
1764 */
NeilBrown911d4ee2009-03-31 14:39:38 +11001765 pd_idx = qd_idx = ~0;
NeilBrown16a53ec2006-06-26 00:27:38 -07001766 switch(conf->level) {
1767 case 4:
NeilBrown911d4ee2009-03-31 14:39:38 +11001768 pd_idx = data_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001769 break;
1770 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001771 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001773 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001774 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 (*dd_idx)++;
1776 break;
1777 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001778 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001779 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 (*dd_idx)++;
1781 break;
1782 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001783 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001784 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 break;
1786 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001787 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001788 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001790 case ALGORITHM_PARITY_0:
1791 pd_idx = 0;
1792 (*dd_idx)++;
1793 break;
1794 case ALGORITHM_PARITY_N:
1795 pd_idx = data_disks;
1796 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001798 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001799 }
1800 break;
1801 case 6:
1802
NeilBrowne183eae2009-03-31 15:20:22 +11001803 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001804 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001805 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001806 qd_idx = pd_idx + 1;
1807 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001808 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001809 qd_idx = 0;
1810 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001811 (*dd_idx) += 2; /* D D P Q D */
1812 break;
1813 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001814 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001815 qd_idx = pd_idx + 1;
1816 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001817 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001818 qd_idx = 0;
1819 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001820 (*dd_idx) += 2; /* D D P Q D */
1821 break;
1822 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001823 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001824 qd_idx = (pd_idx + 1) % raid_disks;
1825 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001826 break;
1827 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001828 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001829 qd_idx = (pd_idx + 1) % raid_disks;
1830 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001831 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001832
1833 case ALGORITHM_PARITY_0:
1834 pd_idx = 0;
1835 qd_idx = 1;
1836 (*dd_idx) += 2;
1837 break;
1838 case ALGORITHM_PARITY_N:
1839 pd_idx = data_disks;
1840 qd_idx = data_disks + 1;
1841 break;
1842
1843 case ALGORITHM_ROTATING_ZERO_RESTART:
1844 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1845 * of blocks for computing Q is different.
1846 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001847 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001848 qd_idx = pd_idx + 1;
1849 if (pd_idx == raid_disks-1) {
1850 (*dd_idx)++; /* Q D D D P */
1851 qd_idx = 0;
1852 } else if (*dd_idx >= pd_idx)
1853 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001854 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001855 break;
1856
1857 case ALGORITHM_ROTATING_N_RESTART:
1858 /* Same a left_asymmetric, by first stripe is
1859 * D D D P Q rather than
1860 * Q D D D P
1861 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001862 stripe2 += 1;
1863 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001864 qd_idx = pd_idx + 1;
1865 if (pd_idx == raid_disks-1) {
1866 (*dd_idx)++; /* Q D D D P */
1867 qd_idx = 0;
1868 } else if (*dd_idx >= pd_idx)
1869 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001870 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001871 break;
1872
1873 case ALGORITHM_ROTATING_N_CONTINUE:
1874 /* Same as left_symmetric but Q is before P */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001875 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001876 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1877 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
NeilBrown67cc2b82009-03-31 14:39:38 +11001878 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001879 break;
1880
1881 case ALGORITHM_LEFT_ASYMMETRIC_6:
1882 /* RAID5 left_asymmetric, with Q on last device */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001883 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001884 if (*dd_idx >= pd_idx)
1885 (*dd_idx)++;
1886 qd_idx = raid_disks - 1;
1887 break;
1888
1889 case ALGORITHM_RIGHT_ASYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001890 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001891 if (*dd_idx >= pd_idx)
1892 (*dd_idx)++;
1893 qd_idx = raid_disks - 1;
1894 break;
1895
1896 case ALGORITHM_LEFT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001897 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001898 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1899 qd_idx = raid_disks - 1;
1900 break;
1901
1902 case ALGORITHM_RIGHT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001903 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001904 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1905 qd_idx = raid_disks - 1;
1906 break;
1907
1908 case ALGORITHM_PARITY_0_6:
1909 pd_idx = 0;
1910 (*dd_idx)++;
1911 qd_idx = raid_disks - 1;
1912 break;
1913
NeilBrown16a53ec2006-06-26 00:27:38 -07001914 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001915 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001916 }
1917 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 }
1919
NeilBrown911d4ee2009-03-31 14:39:38 +11001920 if (sh) {
1921 sh->pd_idx = pd_idx;
1922 sh->qd_idx = qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001923 sh->ddf_layout = ddf_layout;
NeilBrown911d4ee2009-03-31 14:39:38 +11001924 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 /*
1926 * Finally, compute the new sector number
1927 */
1928 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1929 return new_sector;
1930}
1931
1932
NeilBrown784052e2009-03-31 15:19:07 +11001933static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
1935 raid5_conf_t *conf = sh->raid_conf;
NeilBrownb875e532006-12-10 02:20:49 -08001936 int raid_disks = sh->disks;
1937 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 sector_t new_sector = sh->sector, check;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001939 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1940 : conf->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11001941 int algorithm = previous ? conf->prev_algo
1942 : conf->algorithm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 sector_t stripe;
1944 int chunk_offset;
NeilBrown35f2a592010-04-20 14:13:34 +10001945 sector_t chunk_number;
1946 int dummy1, dd_idx = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 sector_t r_sector;
NeilBrown911d4ee2009-03-31 14:39:38 +11001948 struct stripe_head sh2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
NeilBrown16a53ec2006-06-26 00:27:38 -07001950
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1952 stripe = new_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
NeilBrown16a53ec2006-06-26 00:27:38 -07001954 if (i == sh->pd_idx)
1955 return 0;
1956 switch(conf->level) {
1957 case 4: break;
1958 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001959 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 case ALGORITHM_LEFT_ASYMMETRIC:
1961 case ALGORITHM_RIGHT_ASYMMETRIC:
1962 if (i > sh->pd_idx)
1963 i--;
1964 break;
1965 case ALGORITHM_LEFT_SYMMETRIC:
1966 case ALGORITHM_RIGHT_SYMMETRIC:
1967 if (i < sh->pd_idx)
1968 i += raid_disks;
1969 i -= (sh->pd_idx + 1);
1970 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001971 case ALGORITHM_PARITY_0:
1972 i -= 1;
1973 break;
1974 case ALGORITHM_PARITY_N:
1975 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001977 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001978 }
1979 break;
1980 case 6:
NeilBrownd0dabf72009-03-31 14:39:38 +11001981 if (i == sh->qd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001982 return 0; /* It is the Q disk */
NeilBrowne183eae2009-03-31 15:20:22 +11001983 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001984 case ALGORITHM_LEFT_ASYMMETRIC:
1985 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown99c0fb52009-03-31 14:39:38 +11001986 case ALGORITHM_ROTATING_ZERO_RESTART:
1987 case ALGORITHM_ROTATING_N_RESTART:
1988 if (sh->pd_idx == raid_disks-1)
1989 i--; /* Q D D D P */
NeilBrown16a53ec2006-06-26 00:27:38 -07001990 else if (i > sh->pd_idx)
1991 i -= 2; /* D D P Q D */
1992 break;
1993 case ALGORITHM_LEFT_SYMMETRIC:
1994 case ALGORITHM_RIGHT_SYMMETRIC:
1995 if (sh->pd_idx == raid_disks-1)
1996 i--; /* Q D D D P */
1997 else {
1998 /* D D P Q D */
1999 if (i < sh->pd_idx)
2000 i += raid_disks;
2001 i -= (sh->pd_idx + 2);
2002 }
2003 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11002004 case ALGORITHM_PARITY_0:
2005 i -= 2;
2006 break;
2007 case ALGORITHM_PARITY_N:
2008 break;
2009 case ALGORITHM_ROTATING_N_CONTINUE:
NeilBrowne4424fe2009-10-16 16:27:34 +11002010 /* Like left_symmetric, but P is before Q */
NeilBrown99c0fb52009-03-31 14:39:38 +11002011 if (sh->pd_idx == 0)
2012 i--; /* P D D D Q */
NeilBrowne4424fe2009-10-16 16:27:34 +11002013 else {
2014 /* D D Q P D */
2015 if (i < sh->pd_idx)
2016 i += raid_disks;
2017 i -= (sh->pd_idx + 1);
2018 }
NeilBrown99c0fb52009-03-31 14:39:38 +11002019 break;
2020 case ALGORITHM_LEFT_ASYMMETRIC_6:
2021 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2022 if (i > sh->pd_idx)
2023 i--;
2024 break;
2025 case ALGORITHM_LEFT_SYMMETRIC_6:
2026 case ALGORITHM_RIGHT_SYMMETRIC_6:
2027 if (i < sh->pd_idx)
2028 i += data_disks + 1;
2029 i -= (sh->pd_idx + 1);
2030 break;
2031 case ALGORITHM_PARITY_0_6:
2032 i -= 1;
2033 break;
NeilBrown16a53ec2006-06-26 00:27:38 -07002034 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002035 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002036 }
2037 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 }
2039
2040 chunk_number = stripe * data_disks + i;
NeilBrown35f2a592010-04-20 14:13:34 +10002041 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
NeilBrown112bf892009-03-31 14:39:38 +11002043 check = raid5_compute_sector(conf, r_sector,
NeilBrown784052e2009-03-31 15:19:07 +11002044 previous, &dummy1, &sh2);
NeilBrown911d4ee2009-03-31 14:39:38 +11002045 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2046 || sh2.qd_idx != sh->qd_idx) {
NeilBrown0c55e022010-05-03 14:09:02 +10002047 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2048 mdname(conf->mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 return 0;
2050 }
2051 return r_sector;
2052}
2053
2054
Dan Williams600aa102008-06-28 08:32:05 +10002055static void
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002056schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
Dan Williams600aa102008-06-28 08:32:05 +10002057 int rcw, int expand)
Dan Williamse33129d2007-01-02 13:52:30 -07002058{
2059 int i, pd_idx = sh->pd_idx, disks = sh->disks;
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002060 raid5_conf_t *conf = sh->raid_conf;
2061 int level = conf->level;
NeilBrown16a53ec2006-06-26 00:27:38 -07002062
Dan Williamse33129d2007-01-02 13:52:30 -07002063 if (rcw) {
2064 /* if we are not expanding this is a proper write request, and
2065 * there will be bios with new data to be drained into the
2066 * stripe cache
2067 */
2068 if (!expand) {
Dan Williams600aa102008-06-28 08:32:05 +10002069 sh->reconstruct_state = reconstruct_state_drain_run;
2070 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2071 } else
2072 sh->reconstruct_state = reconstruct_state_run;
Dan Williamse33129d2007-01-02 13:52:30 -07002073
Dan Williamsac6b53b2009-07-14 13:40:19 -07002074 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002075
2076 for (i = disks; i--; ) {
2077 struct r5dev *dev = &sh->dev[i];
2078
2079 if (dev->towrite) {
2080 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsd8ee0722008-06-28 08:32:06 +10002081 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002082 if (!expand)
2083 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002084 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002085 }
2086 }
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002087 if (s->locked + conf->max_degraded == disks)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002088 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002089 atomic_inc(&conf->pending_full_writes);
Dan Williamse33129d2007-01-02 13:52:30 -07002090 } else {
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002091 BUG_ON(level == 6);
Dan Williamse33129d2007-01-02 13:52:30 -07002092 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2093 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2094
Dan Williamsd8ee0722008-06-28 08:32:06 +10002095 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
Dan Williams600aa102008-06-28 08:32:05 +10002096 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2097 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
Dan Williamsac6b53b2009-07-14 13:40:19 -07002098 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002099
2100 for (i = disks; i--; ) {
2101 struct r5dev *dev = &sh->dev[i];
2102 if (i == pd_idx)
2103 continue;
2104
Dan Williamse33129d2007-01-02 13:52:30 -07002105 if (dev->towrite &&
2106 (test_bit(R5_UPTODATE, &dev->flags) ||
Dan Williamsd8ee0722008-06-28 08:32:06 +10002107 test_bit(R5_Wantcompute, &dev->flags))) {
2108 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002109 set_bit(R5_LOCKED, &dev->flags);
2110 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002111 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002112 }
2113 }
2114 }
2115
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002116 /* keep the parity disk(s) locked while asynchronous operations
Dan Williamse33129d2007-01-02 13:52:30 -07002117 * are in flight
2118 */
2119 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2120 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
Dan Williams600aa102008-06-28 08:32:05 +10002121 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002122
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002123 if (level == 6) {
2124 int qd_idx = sh->qd_idx;
2125 struct r5dev *dev = &sh->dev[qd_idx];
2126
2127 set_bit(R5_LOCKED, &dev->flags);
2128 clear_bit(R5_UPTODATE, &dev->flags);
2129 s->locked++;
2130 }
2131
Dan Williams600aa102008-06-28 08:32:05 +10002132 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07002133 __func__, (unsigned long long)sh->sector,
Dan Williams600aa102008-06-28 08:32:05 +10002134 s->locked, s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002135}
NeilBrown16a53ec2006-06-26 00:27:38 -07002136
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137/*
2138 * Each stripe/dev can have one or more bion attached.
NeilBrown16a53ec2006-06-26 00:27:38 -07002139 * toread/towrite point to the first in a chain.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 * The bi_next chain must be in order.
2141 */
2142static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2143{
2144 struct bio **bip;
2145 raid5_conf_t *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -07002146 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
Dan Williams45b42332007-07-09 11:56:43 -07002148 pr_debug("adding bh b#%llu to stripe s#%llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 (unsigned long long)bi->bi_sector,
2150 (unsigned long long)sh->sector);
2151
2152
2153 spin_lock(&sh->lock);
2154 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07002155 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 bip = &sh->dev[dd_idx].towrite;
NeilBrown72626682005-09-09 16:23:54 -07002157 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2158 firstwrite = 1;
2159 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 bip = &sh->dev[dd_idx].toread;
2161 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2162 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2163 goto overlap;
2164 bip = & (*bip)->bi_next;
2165 }
2166 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2167 goto overlap;
2168
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002169 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 if (*bip)
2171 bi->bi_next = *bip;
2172 *bip = bi;
Jens Axboe960e7392008-08-15 10:41:18 +02002173 bi->bi_phys_segments++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 spin_unlock_irq(&conf->device_lock);
2175 spin_unlock(&sh->lock);
2176
Dan Williams45b42332007-07-09 11:56:43 -07002177 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 (unsigned long long)bi->bi_sector,
2179 (unsigned long long)sh->sector, dd_idx);
2180
NeilBrown72626682005-09-09 16:23:54 -07002181 if (conf->mddev->bitmap && firstwrite) {
NeilBrown72626682005-09-09 16:23:54 -07002182 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2183 STRIPE_SECTORS, 0);
NeilBrownae3c20c2006-07-10 04:44:17 -07002184 sh->bm_seq = conf->seq_flush+1;
NeilBrown72626682005-09-09 16:23:54 -07002185 set_bit(STRIPE_BIT_DELAY, &sh->state);
2186 }
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 if (forwrite) {
2189 /* check if page is covered */
2190 sector_t sector = sh->dev[dd_idx].sector;
2191 for (bi=sh->dev[dd_idx].towrite;
2192 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2193 bi && bi->bi_sector <= sector;
2194 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2195 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2196 sector = bi->bi_sector + (bi->bi_size>>9);
2197 }
2198 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2199 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2200 }
2201 return 1;
2202
2203 overlap:
2204 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2205 spin_unlock_irq(&conf->device_lock);
2206 spin_unlock(&sh->lock);
2207 return 0;
2208}
2209
NeilBrown29269552006-03-27 01:18:10 -08002210static void end_reshape(raid5_conf_t *conf);
2211
NeilBrown911d4ee2009-03-31 14:39:38 +11002212static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
2213 struct stripe_head *sh)
NeilBrownccfcc3c2006-03-27 01:18:09 -08002214{
NeilBrown784052e2009-03-31 15:19:07 +11002215 int sectors_per_chunk =
Andre Noll09c9e5f2009-06-18 08:45:55 +10002216 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
NeilBrown911d4ee2009-03-31 14:39:38 +11002217 int dd_idx;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002218 int chunk_offset = sector_div(stripe, sectors_per_chunk);
NeilBrown112bf892009-03-31 14:39:38 +11002219 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002220
NeilBrown112bf892009-03-31 14:39:38 +11002221 raid5_compute_sector(conf,
2222 stripe * (disks - conf->max_degraded)
NeilBrownb875e532006-12-10 02:20:49 -08002223 *sectors_per_chunk + chunk_offset,
NeilBrown112bf892009-03-31 14:39:38 +11002224 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002225 &dd_idx, sh);
NeilBrownccfcc3c2006-03-27 01:18:09 -08002226}
2227
Dan Williamsa4456852007-07-09 11:56:43 -07002228static void
Dan Williams1fe797e2008-06-28 09:16:30 +10002229handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002230 struct stripe_head_state *s, int disks,
2231 struct bio **return_bi)
2232{
2233 int i;
2234 for (i = disks; i--; ) {
2235 struct bio *bi;
2236 int bitmap_end = 0;
2237
2238 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2239 mdk_rdev_t *rdev;
2240 rcu_read_lock();
2241 rdev = rcu_dereference(conf->disks[i].rdev);
2242 if (rdev && test_bit(In_sync, &rdev->flags))
2243 /* multiple read failures in one stripe */
2244 md_error(conf->mddev, rdev);
2245 rcu_read_unlock();
2246 }
2247 spin_lock_irq(&conf->device_lock);
2248 /* fail all writes first */
2249 bi = sh->dev[i].towrite;
2250 sh->dev[i].towrite = NULL;
2251 if (bi) {
2252 s->to_write--;
2253 bitmap_end = 1;
2254 }
2255
2256 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2257 wake_up(&conf->wait_for_overlap);
2258
2259 while (bi && bi->bi_sector <
2260 sh->dev[i].sector + STRIPE_SECTORS) {
2261 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2262 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002263 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002264 md_write_end(conf->mddev);
2265 bi->bi_next = *return_bi;
2266 *return_bi = bi;
2267 }
2268 bi = nextbi;
2269 }
2270 /* and fail all 'written' */
2271 bi = sh->dev[i].written;
2272 sh->dev[i].written = NULL;
2273 if (bi) bitmap_end = 1;
2274 while (bi && bi->bi_sector <
2275 sh->dev[i].sector + STRIPE_SECTORS) {
2276 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2277 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002278 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002279 md_write_end(conf->mddev);
2280 bi->bi_next = *return_bi;
2281 *return_bi = bi;
2282 }
2283 bi = bi2;
2284 }
2285
Dan Williamsb5e98d62007-01-02 13:52:31 -07002286 /* fail any reads if this device is non-operational and
2287 * the data has not reached the cache yet.
2288 */
2289 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2290 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2291 test_bit(R5_ReadError, &sh->dev[i].flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002292 bi = sh->dev[i].toread;
2293 sh->dev[i].toread = NULL;
2294 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2295 wake_up(&conf->wait_for_overlap);
2296 if (bi) s->to_read--;
2297 while (bi && bi->bi_sector <
2298 sh->dev[i].sector + STRIPE_SECTORS) {
2299 struct bio *nextbi =
2300 r5_next_bio(bi, sh->dev[i].sector);
2301 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002302 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002303 bi->bi_next = *return_bi;
2304 *return_bi = bi;
2305 }
2306 bi = nextbi;
2307 }
2308 }
2309 spin_unlock_irq(&conf->device_lock);
2310 if (bitmap_end)
2311 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2312 STRIPE_SECTORS, 0, 0);
2313 }
2314
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002315 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2316 if (atomic_dec_and_test(&conf->pending_full_writes))
2317 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002318}
2319
Dan Williams1fe797e2008-06-28 09:16:30 +10002320/* fetch_block5 - checks the given member device to see if its data needs
2321 * to be read or computed to satisfy a request.
2322 *
2323 * Returns 1 when no more member devices need to be checked, otherwise returns
2324 * 0 to tell the loop in handle_stripe_fill5 to continue
Dan Williamsf38e1212007-01-02 13:52:30 -07002325 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002326static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
2327 int disk_idx, int disks)
Dan Williamsf38e1212007-01-02 13:52:30 -07002328{
2329 struct r5dev *dev = &sh->dev[disk_idx];
2330 struct r5dev *failed_dev = &sh->dev[s->failed_num];
2331
Dan Williamsf38e1212007-01-02 13:52:30 -07002332 /* is the data in this block needed, and can we get it? */
2333 if (!test_bit(R5_LOCKED, &dev->flags) &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002334 !test_bit(R5_UPTODATE, &dev->flags) &&
2335 (dev->toread ||
2336 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2337 s->syncing || s->expanding ||
2338 (s->failed &&
2339 (failed_dev->toread ||
2340 (failed_dev->towrite &&
2341 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002342 /* We would like to get this block, possibly by computing it,
2343 * otherwise read it if the backing disk is insync
Dan Williamsf38e1212007-01-02 13:52:30 -07002344 */
2345 if ((s->uptodate == disks - 1) &&
Dan Williamsecc65c92008-06-28 08:31:57 +10002346 (s->failed && disk_idx == s->failed_num)) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002347 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2348 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
Dan Williamsf38e1212007-01-02 13:52:30 -07002349 set_bit(R5_Wantcompute, &dev->flags);
2350 sh->ops.target = disk_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002351 sh->ops.target2 = -1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002352 s->req_compute = 1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002353 /* Careful: from this point on 'uptodate' is in the eye
Dan Williamsac6b53b2009-07-14 13:40:19 -07002354 * of raid_run_ops which services 'compute' operations
Dan Williamsf38e1212007-01-02 13:52:30 -07002355 * before writes. R5_Wantcompute flags a block that will
2356 * be R5_UPTODATE by the time it is needed for a
2357 * subsequent operation.
2358 */
2359 s->uptodate++;
Dan Williams1fe797e2008-06-28 09:16:30 +10002360 return 1; /* uptodate + compute == disks */
Dan Williams7a1fc532008-07-10 04:54:57 -07002361 } else if (test_bit(R5_Insync, &dev->flags)) {
Dan Williamsf38e1212007-01-02 13:52:30 -07002362 set_bit(R5_LOCKED, &dev->flags);
2363 set_bit(R5_Wantread, &dev->flags);
Dan Williamsf38e1212007-01-02 13:52:30 -07002364 s->locked++;
2365 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
2366 s->syncing);
2367 }
2368 }
2369
Dan Williams1fe797e2008-06-28 09:16:30 +10002370 return 0;
Dan Williamsf38e1212007-01-02 13:52:30 -07002371}
2372
Dan Williams1fe797e2008-06-28 09:16:30 +10002373/**
2374 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
2375 */
2376static void handle_stripe_fill5(struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002377 struct stripe_head_state *s, int disks)
2378{
2379 int i;
Dan Williamsf38e1212007-01-02 13:52:30 -07002380
Dan Williamsf38e1212007-01-02 13:52:30 -07002381 /* look for blocks to read/compute, skip this if a compute
2382 * is already in flight, or if the stripe contents are in the
2383 * midst of changing due to a write
2384 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002385 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002386 !sh->reconstruct_state)
Dan Williamsf38e1212007-01-02 13:52:30 -07002387 for (i = disks; i--; )
Dan Williams1fe797e2008-06-28 09:16:30 +10002388 if (fetch_block5(sh, s, i, disks))
Dan Williamsf38e1212007-01-02 13:52:30 -07002389 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002390 set_bit(STRIPE_HANDLE, &sh->state);
2391}
2392
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002393/* fetch_block6 - checks the given member device to see if its data needs
2394 * to be read or computed to satisfy a request.
2395 *
2396 * Returns 1 when no more member devices need to be checked, otherwise returns
2397 * 0 to tell the loop in handle_stripe_fill6 to continue
2398 */
2399static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s,
2400 struct r6_state *r6s, int disk_idx, int disks)
2401{
2402 struct r5dev *dev = &sh->dev[disk_idx];
2403 struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]],
2404 &sh->dev[r6s->failed_num[1]] };
2405
2406 if (!test_bit(R5_LOCKED, &dev->flags) &&
2407 !test_bit(R5_UPTODATE, &dev->flags) &&
2408 (dev->toread ||
2409 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2410 s->syncing || s->expanding ||
2411 (s->failed >= 1 &&
2412 (fdev[0]->toread || s->to_write)) ||
2413 (s->failed >= 2 &&
2414 (fdev[1]->toread || s->to_write)))) {
2415 /* we would like to get this block, possibly by computing it,
2416 * otherwise read it if the backing disk is insync
2417 */
2418 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2419 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2420 if ((s->uptodate == disks - 1) &&
2421 (s->failed && (disk_idx == r6s->failed_num[0] ||
2422 disk_idx == r6s->failed_num[1]))) {
2423 /* have disk failed, and we're requested to fetch it;
2424 * do compute it
2425 */
2426 pr_debug("Computing stripe %llu block %d\n",
2427 (unsigned long long)sh->sector, disk_idx);
2428 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2429 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2430 set_bit(R5_Wantcompute, &dev->flags);
2431 sh->ops.target = disk_idx;
2432 sh->ops.target2 = -1; /* no 2nd target */
2433 s->req_compute = 1;
2434 s->uptodate++;
2435 return 1;
2436 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2437 /* Computing 2-failure is *very* expensive; only
2438 * do it if failed >= 2
2439 */
2440 int other;
2441 for (other = disks; other--; ) {
2442 if (other == disk_idx)
2443 continue;
2444 if (!test_bit(R5_UPTODATE,
2445 &sh->dev[other].flags))
2446 break;
2447 }
2448 BUG_ON(other < 0);
2449 pr_debug("Computing stripe %llu blocks %d,%d\n",
2450 (unsigned long long)sh->sector,
2451 disk_idx, other);
2452 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2453 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2454 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2455 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2456 sh->ops.target = disk_idx;
2457 sh->ops.target2 = other;
2458 s->uptodate += 2;
2459 s->req_compute = 1;
2460 return 1;
2461 } else if (test_bit(R5_Insync, &dev->flags)) {
2462 set_bit(R5_LOCKED, &dev->flags);
2463 set_bit(R5_Wantread, &dev->flags);
2464 s->locked++;
2465 pr_debug("Reading block %d (sync=%d)\n",
2466 disk_idx, s->syncing);
2467 }
2468 }
2469
2470 return 0;
2471}
2472
2473/**
2474 * handle_stripe_fill6 - read or compute data to satisfy pending requests.
2475 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002476static void handle_stripe_fill6(struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002477 struct stripe_head_state *s, struct r6_state *r6s,
2478 int disks)
2479{
2480 int i;
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002481
2482 /* look for blocks to read/compute, skip this if a compute
2483 * is already in flight, or if the stripe contents are in the
2484 * midst of changing due to a write
2485 */
2486 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2487 !sh->reconstruct_state)
2488 for (i = disks; i--; )
2489 if (fetch_block6(sh, s, r6s, i, disks))
2490 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002491 set_bit(STRIPE_HANDLE, &sh->state);
2492}
2493
2494
Dan Williams1fe797e2008-06-28 09:16:30 +10002495/* handle_stripe_clean_event
Dan Williamsa4456852007-07-09 11:56:43 -07002496 * any written block on an uptodate or failed drive can be returned.
2497 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2498 * never LOCKED, so we don't need to test 'failed' directly.
2499 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002500static void handle_stripe_clean_event(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002501 struct stripe_head *sh, int disks, struct bio **return_bi)
2502{
2503 int i;
2504 struct r5dev *dev;
2505
2506 for (i = disks; i--; )
2507 if (sh->dev[i].written) {
2508 dev = &sh->dev[i];
2509 if (!test_bit(R5_LOCKED, &dev->flags) &&
2510 test_bit(R5_UPTODATE, &dev->flags)) {
2511 /* We can return any write requests */
2512 struct bio *wbi, *wbi2;
2513 int bitmap_end = 0;
Dan Williams45b42332007-07-09 11:56:43 -07002514 pr_debug("Return write for disc %d\n", i);
Dan Williamsa4456852007-07-09 11:56:43 -07002515 spin_lock_irq(&conf->device_lock);
2516 wbi = dev->written;
2517 dev->written = NULL;
2518 while (wbi && wbi->bi_sector <
2519 dev->sector + STRIPE_SECTORS) {
2520 wbi2 = r5_next_bio(wbi, dev->sector);
Jens Axboe960e7392008-08-15 10:41:18 +02002521 if (!raid5_dec_bi_phys_segments(wbi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002522 md_write_end(conf->mddev);
2523 wbi->bi_next = *return_bi;
2524 *return_bi = wbi;
2525 }
2526 wbi = wbi2;
2527 }
2528 if (dev->towrite == NULL)
2529 bitmap_end = 1;
2530 spin_unlock_irq(&conf->device_lock);
2531 if (bitmap_end)
2532 bitmap_endwrite(conf->mddev->bitmap,
2533 sh->sector,
2534 STRIPE_SECTORS,
2535 !test_bit(STRIPE_DEGRADED, &sh->state),
2536 0);
2537 }
2538 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002539
2540 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2541 if (atomic_dec_and_test(&conf->pending_full_writes))
2542 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002543}
2544
Dan Williams1fe797e2008-06-28 09:16:30 +10002545static void handle_stripe_dirtying5(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002546 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2547{
2548 int rmw = 0, rcw = 0, i;
2549 for (i = disks; i--; ) {
2550 /* would I have to read this buffer for read_modify_write */
2551 struct r5dev *dev = &sh->dev[i];
2552 if ((dev->towrite || i == sh->pd_idx) &&
2553 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002554 !(test_bit(R5_UPTODATE, &dev->flags) ||
2555 test_bit(R5_Wantcompute, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002556 if (test_bit(R5_Insync, &dev->flags))
2557 rmw++;
2558 else
2559 rmw += 2*disks; /* cannot read it */
2560 }
2561 /* Would I have to read this buffer for reconstruct_write */
2562 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2563 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002564 !(test_bit(R5_UPTODATE, &dev->flags) ||
2565 test_bit(R5_Wantcompute, &dev->flags))) {
2566 if (test_bit(R5_Insync, &dev->flags)) rcw++;
Dan Williamsa4456852007-07-09 11:56:43 -07002567 else
2568 rcw += 2*disks;
2569 }
2570 }
Dan Williams45b42332007-07-09 11:56:43 -07002571 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07002572 (unsigned long long)sh->sector, rmw, rcw);
2573 set_bit(STRIPE_HANDLE, &sh->state);
2574 if (rmw < rcw && rmw > 0)
2575 /* prefer read-modify-write, but need to get some data */
2576 for (i = disks; i--; ) {
2577 struct r5dev *dev = &sh->dev[i];
2578 if ((dev->towrite || i == sh->pd_idx) &&
2579 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002580 !(test_bit(R5_UPTODATE, &dev->flags) ||
2581 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002582 test_bit(R5_Insync, &dev->flags)) {
2583 if (
2584 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002585 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002586 "%d for r-m-w\n", i);
2587 set_bit(R5_LOCKED, &dev->flags);
2588 set_bit(R5_Wantread, &dev->flags);
2589 s->locked++;
2590 } else {
2591 set_bit(STRIPE_DELAYED, &sh->state);
2592 set_bit(STRIPE_HANDLE, &sh->state);
2593 }
2594 }
2595 }
2596 if (rcw <= rmw && rcw > 0)
2597 /* want reconstruct write, but need to get some data */
2598 for (i = disks; i--; ) {
2599 struct r5dev *dev = &sh->dev[i];
2600 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2601 i != sh->pd_idx &&
2602 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002603 !(test_bit(R5_UPTODATE, &dev->flags) ||
2604 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002605 test_bit(R5_Insync, &dev->flags)) {
2606 if (
2607 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002608 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002609 "%d for Reconstruct\n", i);
2610 set_bit(R5_LOCKED, &dev->flags);
2611 set_bit(R5_Wantread, &dev->flags);
2612 s->locked++;
2613 } else {
2614 set_bit(STRIPE_DELAYED, &sh->state);
2615 set_bit(STRIPE_HANDLE, &sh->state);
2616 }
2617 }
2618 }
2619 /* now if nothing is locked, and if we have enough data,
2620 * we can start a write request
2621 */
Dan Williamsf38e1212007-01-02 13:52:30 -07002622 /* since handle_stripe can be called at any time we need to handle the
2623 * case where a compute block operation has been submitted and then a
Dan Williamsac6b53b2009-07-14 13:40:19 -07002624 * subsequent call wants to start a write request. raid_run_ops only
2625 * handles the case where compute block and reconstruct are requested
Dan Williamsf38e1212007-01-02 13:52:30 -07002626 * simultaneously. If this is not the case then new writes need to be
2627 * held off until the compute completes.
2628 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002629 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2630 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2631 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002632 schedule_reconstruction(sh, s, rcw == 0, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002633}
2634
Dan Williams1fe797e2008-06-28 09:16:30 +10002635static void handle_stripe_dirtying6(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002636 struct stripe_head *sh, struct stripe_head_state *s,
2637 struct r6_state *r6s, int disks)
2638{
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002639 int rcw = 0, pd_idx = sh->pd_idx, i;
NeilBrown34e04e82009-03-31 15:10:16 +11002640 int qd_idx = sh->qd_idx;
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002641
2642 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002643 for (i = disks; i--; ) {
2644 struct r5dev *dev = &sh->dev[i];
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002645 /* check if we haven't enough data */
2646 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2647 i != pd_idx && i != qd_idx &&
2648 !test_bit(R5_LOCKED, &dev->flags) &&
2649 !(test_bit(R5_UPTODATE, &dev->flags) ||
2650 test_bit(R5_Wantcompute, &dev->flags))) {
2651 rcw++;
2652 if (!test_bit(R5_Insync, &dev->flags))
2653 continue; /* it's a failed drive */
2654
2655 if (
2656 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2657 pr_debug("Read_old stripe %llu "
2658 "block %d for Reconstruct\n",
2659 (unsigned long long)sh->sector, i);
2660 set_bit(R5_LOCKED, &dev->flags);
2661 set_bit(R5_Wantread, &dev->flags);
2662 s->locked++;
2663 } else {
2664 pr_debug("Request delayed stripe %llu "
2665 "block %d for Reconstruct\n",
2666 (unsigned long long)sh->sector, i);
2667 set_bit(STRIPE_DELAYED, &sh->state);
2668 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002669 }
2670 }
2671 }
Dan Williamsa4456852007-07-09 11:56:43 -07002672 /* now if nothing is locked, and if we have enough data, we can start a
2673 * write request
2674 */
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002675 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2676 s->locked == 0 && rcw == 0 &&
Dan Williamsa4456852007-07-09 11:56:43 -07002677 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002678 schedule_reconstruction(sh, s, 1, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002679 }
2680}
2681
2682static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2683 struct stripe_head_state *s, int disks)
2684{
Dan Williamsecc65c92008-06-28 08:31:57 +10002685 struct r5dev *dev = NULL;
Dan Williamse89f8962007-01-02 13:52:31 -07002686
Dan Williamsbd2ab672008-04-10 21:29:27 -07002687 set_bit(STRIPE_HANDLE, &sh->state);
2688
Dan Williamsecc65c92008-06-28 08:31:57 +10002689 switch (sh->check_state) {
2690 case check_state_idle:
2691 /* start a new check operation if there are no failures */
Dan Williamsbd2ab672008-04-10 21:29:27 -07002692 if (s->failed == 0) {
Dan Williamsbd2ab672008-04-10 21:29:27 -07002693 BUG_ON(s->uptodate != disks);
Dan Williamsecc65c92008-06-28 08:31:57 +10002694 sh->check_state = check_state_run;
2695 set_bit(STRIPE_OP_CHECK, &s->ops_request);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002696 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002697 s->uptodate--;
Dan Williamsecc65c92008-06-28 08:31:57 +10002698 break;
Dan Williamsbd2ab672008-04-10 21:29:27 -07002699 }
Dan Williamsa4456852007-07-09 11:56:43 -07002700 dev = &sh->dev[s->failed_num];
Dan Williamsecc65c92008-06-28 08:31:57 +10002701 /* fall through */
2702 case check_state_compute_result:
2703 sh->check_state = check_state_idle;
2704 if (!dev)
2705 dev = &sh->dev[sh->pd_idx];
2706
2707 /* check that a write has not made the stripe insync */
2708 if (test_bit(STRIPE_INSYNC, &sh->state))
2709 break;
2710
2711 /* either failed parity check, or recovery is happening */
Dan Williamsa4456852007-07-09 11:56:43 -07002712 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2713 BUG_ON(s->uptodate != disks);
2714
2715 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsecc65c92008-06-28 08:31:57 +10002716 s->locked++;
Dan Williamsa4456852007-07-09 11:56:43 -07002717 set_bit(R5_Wantwrite, &dev->flags);
Dan Williams830ea012007-01-02 13:52:31 -07002718
Dan Williamsa4456852007-07-09 11:56:43 -07002719 clear_bit(STRIPE_DEGRADED, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002720 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002721 break;
2722 case check_state_run:
2723 break; /* we will be called again upon completion */
2724 case check_state_check_result:
2725 sh->check_state = check_state_idle;
2726
2727 /* if a failure occurred during the check operation, leave
2728 * STRIPE_INSYNC not set and let the stripe be handled again
2729 */
2730 if (s->failed)
2731 break;
2732
2733 /* handle a successful check operation, if parity is correct
2734 * we are done. Otherwise update the mismatch count and repair
2735 * parity if !MD_RECOVERY_CHECK
2736 */
Dan Williamsad283ea2009-08-29 19:09:26 -07002737 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
Dan Williamsecc65c92008-06-28 08:31:57 +10002738 /* parity is correct (on disc,
2739 * not in buffer any more)
2740 */
2741 set_bit(STRIPE_INSYNC, &sh->state);
2742 else {
2743 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2744 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2745 /* don't try to repair!! */
2746 set_bit(STRIPE_INSYNC, &sh->state);
2747 else {
2748 sh->check_state = check_state_compute_run;
Dan Williams976ea8d2008-06-28 08:32:03 +10002749 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002750 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2751 set_bit(R5_Wantcompute,
2752 &sh->dev[sh->pd_idx].flags);
2753 sh->ops.target = sh->pd_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002754 sh->ops.target2 = -1;
Dan Williamsecc65c92008-06-28 08:31:57 +10002755 s->uptodate++;
2756 }
2757 }
2758 break;
2759 case check_state_compute_run:
2760 break;
2761 default:
2762 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2763 __func__, sh->check_state,
2764 (unsigned long long) sh->sector);
2765 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002766 }
2767}
2768
2769
2770static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williams36d1c642009-07-14 11:48:22 -07002771 struct stripe_head_state *s,
2772 struct r6_state *r6s, int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07002773{
Dan Williamsa4456852007-07-09 11:56:43 -07002774 int pd_idx = sh->pd_idx;
NeilBrown34e04e82009-03-31 15:10:16 +11002775 int qd_idx = sh->qd_idx;
Dan Williamsd82dfee2009-07-14 13:40:57 -07002776 struct r5dev *dev;
Dan Williamsa4456852007-07-09 11:56:43 -07002777
2778 set_bit(STRIPE_HANDLE, &sh->state);
2779
2780 BUG_ON(s->failed > 2);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002781
Dan Williamsa4456852007-07-09 11:56:43 -07002782 /* Want to check and possibly repair P and Q.
2783 * However there could be one 'failed' device, in which
2784 * case we can only check one of them, possibly using the
2785 * other to generate missing data
2786 */
2787
Dan Williamsd82dfee2009-07-14 13:40:57 -07002788 switch (sh->check_state) {
2789 case check_state_idle:
2790 /* start a new check operation if there are < 2 failures */
Dan Williamsa4456852007-07-09 11:56:43 -07002791 if (s->failed == r6s->q_failed) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002792 /* The only possible failed device holds Q, so it
Dan Williamsa4456852007-07-09 11:56:43 -07002793 * makes sense to check P (If anything else were failed,
2794 * we would have used P to recreate it).
2795 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002796 sh->check_state = check_state_run;
Dan Williamsa4456852007-07-09 11:56:43 -07002797 }
2798 if (!r6s->q_failed && s->failed < 2) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002799 /* Q is not failed, and we didn't use it to generate
Dan Williamsa4456852007-07-09 11:56:43 -07002800 * anything, so it makes sense to check it
2801 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002802 if (sh->check_state == check_state_run)
2803 sh->check_state = check_state_run_pq;
2804 else
2805 sh->check_state = check_state_run_q;
Dan Williamsa4456852007-07-09 11:56:43 -07002806 }
Dan Williams36d1c642009-07-14 11:48:22 -07002807
Dan Williamsd82dfee2009-07-14 13:40:57 -07002808 /* discard potentially stale zero_sum_result */
2809 sh->ops.zero_sum_result = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07002810
Dan Williamsd82dfee2009-07-14 13:40:57 -07002811 if (sh->check_state == check_state_run) {
2812 /* async_xor_zero_sum destroys the contents of P */
2813 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2814 s->uptodate--;
Dan Williamsa4456852007-07-09 11:56:43 -07002815 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002816 if (sh->check_state >= check_state_run &&
2817 sh->check_state <= check_state_run_pq) {
2818 /* async_syndrome_zero_sum preserves P and Q, so
2819 * no need to mark them !uptodate here
2820 */
2821 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2822 break;
2823 }
Dan Williams36d1c642009-07-14 11:48:22 -07002824
Dan Williamsd82dfee2009-07-14 13:40:57 -07002825 /* we have 2-disk failure */
2826 BUG_ON(s->failed != 2);
2827 /* fall through */
2828 case check_state_compute_result:
2829 sh->check_state = check_state_idle;
Dan Williams36d1c642009-07-14 11:48:22 -07002830
Dan Williamsd82dfee2009-07-14 13:40:57 -07002831 /* check that a write has not made the stripe insync */
2832 if (test_bit(STRIPE_INSYNC, &sh->state))
2833 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002834
2835 /* now write out any block on a failed drive,
Dan Williamsd82dfee2009-07-14 13:40:57 -07002836 * or P or Q if they were recomputed
Dan Williamsa4456852007-07-09 11:56:43 -07002837 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002838 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
Dan Williamsa4456852007-07-09 11:56:43 -07002839 if (s->failed == 2) {
2840 dev = &sh->dev[r6s->failed_num[1]];
2841 s->locked++;
2842 set_bit(R5_LOCKED, &dev->flags);
2843 set_bit(R5_Wantwrite, &dev->flags);
2844 }
2845 if (s->failed >= 1) {
2846 dev = &sh->dev[r6s->failed_num[0]];
2847 s->locked++;
2848 set_bit(R5_LOCKED, &dev->flags);
2849 set_bit(R5_Wantwrite, &dev->flags);
2850 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002851 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002852 dev = &sh->dev[pd_idx];
2853 s->locked++;
2854 set_bit(R5_LOCKED, &dev->flags);
2855 set_bit(R5_Wantwrite, &dev->flags);
2856 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002857 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002858 dev = &sh->dev[qd_idx];
2859 s->locked++;
2860 set_bit(R5_LOCKED, &dev->flags);
2861 set_bit(R5_Wantwrite, &dev->flags);
2862 }
2863 clear_bit(STRIPE_DEGRADED, &sh->state);
2864
2865 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002866 break;
2867 case check_state_run:
2868 case check_state_run_q:
2869 case check_state_run_pq:
2870 break; /* we will be called again upon completion */
2871 case check_state_check_result:
2872 sh->check_state = check_state_idle;
2873
2874 /* handle a successful check operation, if parity is correct
2875 * we are done. Otherwise update the mismatch count and repair
2876 * parity if !MD_RECOVERY_CHECK
2877 */
2878 if (sh->ops.zero_sum_result == 0) {
2879 /* both parities are correct */
2880 if (!s->failed)
2881 set_bit(STRIPE_INSYNC, &sh->state);
2882 else {
2883 /* in contrast to the raid5 case we can validate
2884 * parity, but still have a failure to write
2885 * back
2886 */
2887 sh->check_state = check_state_compute_result;
2888 /* Returning at this point means that we may go
2889 * off and bring p and/or q uptodate again so
2890 * we make sure to check zero_sum_result again
2891 * to verify if p or q need writeback
2892 */
2893 }
2894 } else {
2895 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2896 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2897 /* don't try to repair!! */
2898 set_bit(STRIPE_INSYNC, &sh->state);
2899 else {
2900 int *target = &sh->ops.target;
2901
2902 sh->ops.target = -1;
2903 sh->ops.target2 = -1;
2904 sh->check_state = check_state_compute_run;
2905 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2906 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2907 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
2908 set_bit(R5_Wantcompute,
2909 &sh->dev[pd_idx].flags);
2910 *target = pd_idx;
2911 target = &sh->ops.target2;
2912 s->uptodate++;
2913 }
2914 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
2915 set_bit(R5_Wantcompute,
2916 &sh->dev[qd_idx].flags);
2917 *target = qd_idx;
2918 s->uptodate++;
2919 }
2920 }
2921 }
2922 break;
2923 case check_state_compute_run:
2924 break;
2925 default:
2926 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2927 __func__, sh->check_state,
2928 (unsigned long long) sh->sector);
2929 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002930 }
2931}
2932
2933static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
2934 struct r6_state *r6s)
2935{
2936 int i;
2937
2938 /* We have read all the blocks in this stripe and now we need to
2939 * copy some of them into a target stripe for expand.
2940 */
Dan Williamsf0a50d32007-01-02 13:52:31 -07002941 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07002942 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2943 for (i = 0; i < sh->disks; i++)
NeilBrown34e04e82009-03-31 15:10:16 +11002944 if (i != sh->pd_idx && i != sh->qd_idx) {
NeilBrown911d4ee2009-03-31 14:39:38 +11002945 int dd_idx, j;
Dan Williamsa4456852007-07-09 11:56:43 -07002946 struct stripe_head *sh2;
Dan Williamsa08abd82009-06-03 11:43:59 -07002947 struct async_submit_ctl submit;
Dan Williamsa4456852007-07-09 11:56:43 -07002948
NeilBrown784052e2009-03-31 15:19:07 +11002949 sector_t bn = compute_blocknr(sh, i, 1);
NeilBrown911d4ee2009-03-31 14:39:38 +11002950 sector_t s = raid5_compute_sector(conf, bn, 0,
2951 &dd_idx, NULL);
NeilBrowna8c906c2009-06-09 14:39:59 +10002952 sh2 = get_active_stripe(conf, s, 0, 1, 1);
Dan Williamsa4456852007-07-09 11:56:43 -07002953 if (sh2 == NULL)
2954 /* so far only the early blocks of this stripe
2955 * have been requested. When later blocks
2956 * get requested, we will try again
2957 */
2958 continue;
2959 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2960 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2961 /* must have already done this block */
2962 release_stripe(sh2);
2963 continue;
2964 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07002965
2966 /* place all the copies on one channel */
Dan Williamsa08abd82009-06-03 11:43:59 -07002967 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002968 tx = async_memcpy(sh2->dev[dd_idx].page,
Dan Williams88ba2aa2009-04-09 16:16:18 -07002969 sh->dev[i].page, 0, 0, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07002970 &submit);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002971
Dan Williamsa4456852007-07-09 11:56:43 -07002972 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2973 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2974 for (j = 0; j < conf->raid_disks; j++)
2975 if (j != sh2->pd_idx &&
NeilBrownd0dabf72009-03-31 14:39:38 +11002976 (!r6s || j != sh2->qd_idx) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002977 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2978 break;
2979 if (j == conf->raid_disks) {
2980 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2981 set_bit(STRIPE_HANDLE, &sh2->state);
2982 }
2983 release_stripe(sh2);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002984
Dan Williamsa4456852007-07-09 11:56:43 -07002985 }
NeilBrowna2e08552007-09-11 15:23:36 -07002986 /* done submitting copies, wait for them to complete */
2987 if (tx) {
2988 async_tx_ack(tx);
2989 dma_wait_for_async_tx(tx);
2990 }
Dan Williamsa4456852007-07-09 11:56:43 -07002991}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
Dan Williams6bfe0b42008-04-30 00:52:32 -07002993
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994/*
2995 * handle_stripe - do things to a stripe.
2996 *
2997 * We lock the stripe and then examine the state of various bits
2998 * to see what needs to be done.
2999 * Possible results:
3000 * return some read request which now have data
3001 * return some write requests which are safely on disc
3002 * schedule a read on some buffers
3003 * schedule a write of some buffers
3004 * return confirmation of parity correctness
3005 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 * buffers are taken off read_list or write_list, and bh_cache buffers
3007 * get BH_Lock set before the stripe lock is released.
3008 *
3009 */
Dan Williamsa4456852007-07-09 11:56:43 -07003010
NeilBrown14425772009-10-16 15:55:25 +11003011static void handle_stripe5(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012{
3013 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsa4456852007-07-09 11:56:43 -07003014 int disks = sh->disks, i;
3015 struct bio *return_bi = NULL;
3016 struct stripe_head_state s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 struct r5dev *dev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003018 mdk_rdev_t *blocked_rdev = NULL;
Dan Williamse0a115e2008-06-05 22:45:52 -07003019 int prexor;
NeilBrown729a1862009-12-14 12:49:50 +11003020 int dec_preread_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021
Dan Williamsa4456852007-07-09 11:56:43 -07003022 memset(&s, 0, sizeof(s));
Dan Williams600aa102008-06-28 08:32:05 +10003023 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
3024 "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state,
3025 atomic_read(&sh->count), sh->pd_idx, sh->check_state,
3026 sh->reconstruct_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027
3028 spin_lock(&sh->lock);
3029 clear_bit(STRIPE_HANDLE, &sh->state);
3030 clear_bit(STRIPE_DELAYED, &sh->state);
3031
Dan Williamsa4456852007-07-09 11:56:43 -07003032 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3033 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3034 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
Dan Williams83de75c2008-06-28 08:31:58 +10003035
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 /* Now to look around and see what can be done */
NeilBrown9910f162006-01-06 00:20:24 -08003037 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 for (i=disks; i--; ) {
3039 mdk_rdev_t *rdev;
NeilBrowna9f326e2009-09-23 18:06:41 +10003040
3041 dev = &sh->dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042
Dan Williamsb5e98d62007-01-02 13:52:31 -07003043 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
3044 "written %p\n", i, dev->flags, dev->toread, dev->read,
3045 dev->towrite, dev->written);
3046
3047 /* maybe we can request a biofill operation
3048 *
3049 * new wantfill requests are only permitted while
Dan Williams83de75c2008-06-28 08:31:58 +10003050 * ops_complete_biofill is guaranteed to be inactive
Dan Williamsb5e98d62007-01-02 13:52:31 -07003051 */
3052 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
Dan Williams83de75c2008-06-28 08:31:58 +10003053 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
Dan Williamsb5e98d62007-01-02 13:52:31 -07003054 set_bit(R5_Wantfill, &dev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055
3056 /* now count some things */
Dan Williamsa4456852007-07-09 11:56:43 -07003057 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3058 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
Dan Williamsf38e1212007-01-02 13:52:30 -07003059 if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
Dan Williamsb5e98d62007-01-02 13:52:31 -07003061 if (test_bit(R5_Wantfill, &dev->flags))
3062 s.to_fill++;
3063 else if (dev->toread)
Dan Williamsa4456852007-07-09 11:56:43 -07003064 s.to_read++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 if (dev->towrite) {
Dan Williamsa4456852007-07-09 11:56:43 -07003066 s.to_write++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 if (!test_bit(R5_OVERWRITE, &dev->flags))
Dan Williamsa4456852007-07-09 11:56:43 -07003068 s.non_overwrite++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 }
Dan Williamsa4456852007-07-09 11:56:43 -07003070 if (dev->written)
3071 s.written++;
NeilBrown9910f162006-01-06 00:20:24 -08003072 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownac4090d2008-08-05 15:54:13 +10003073 if (blocked_rdev == NULL &&
3074 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07003075 blocked_rdev = rdev;
3076 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003077 }
NeilBrown415e72d2010-06-17 17:25:21 +10003078 clear_bit(R5_Insync, &dev->flags);
3079 if (!rdev)
3080 /* Not in-sync */;
3081 else if (test_bit(In_sync, &rdev->flags))
3082 set_bit(R5_Insync, &dev->flags);
3083 else {
3084 /* could be in-sync depending on recovery/reshape status */
3085 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3086 set_bit(R5_Insync, &dev->flags);
3087 }
3088 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown14f8d262006-01-06 00:20:14 -08003089 /* The ReadError flag will just be confusing now */
NeilBrown4e5314b2005-11-08 21:39:22 -08003090 clear_bit(R5_ReadError, &dev->flags);
3091 clear_bit(R5_ReWrite, &dev->flags);
3092 }
NeilBrown415e72d2010-06-17 17:25:21 +10003093 if (test_bit(R5_ReadError, &dev->flags))
3094 clear_bit(R5_Insync, &dev->flags);
3095 if (!test_bit(R5_Insync, &dev->flags)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003096 s.failed++;
3097 s.failed_num = i;
NeilBrown415e72d2010-06-17 17:25:21 +10003098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 }
NeilBrown9910f162006-01-06 00:20:24 -08003100 rcu_read_unlock();
Dan Williamsb5e98d62007-01-02 13:52:31 -07003101
Dan Williams6bfe0b42008-04-30 00:52:32 -07003102 if (unlikely(blocked_rdev)) {
NeilBrownac4090d2008-08-05 15:54:13 +10003103 if (s.syncing || s.expanding || s.expanded ||
3104 s.to_write || s.written) {
3105 set_bit(STRIPE_HANDLE, &sh->state);
3106 goto unlock;
3107 }
3108 /* There is nothing for the blocked_rdev to block */
3109 rdev_dec_pending(blocked_rdev, conf->mddev);
3110 blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003111 }
3112
Dan Williams83de75c2008-06-28 08:31:58 +10003113 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3114 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3115 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3116 }
Dan Williamsb5e98d62007-01-02 13:52:31 -07003117
Dan Williams45b42332007-07-09 11:56:43 -07003118 pr_debug("locked=%d uptodate=%d to_read=%d"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 " to_write=%d failed=%d failed_num=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003120 s.locked, s.uptodate, s.to_read, s.to_write,
3121 s.failed, s.failed_num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 /* check if the array has lost two devices and, if so, some requests might
3123 * need to be failed
3124 */
Dan Williamsa4456852007-07-09 11:56:43 -07003125 if (s.failed > 1 && s.to_read+s.to_write+s.written)
Dan Williams1fe797e2008-06-28 09:16:30 +10003126 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003127 if (s.failed > 1 && s.syncing) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3129 clear_bit(STRIPE_SYNCING, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003130 s.syncing = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 }
3132
3133 /* might be able to return some write requests if the parity block
3134 * is safe, or on a failed drive
3135 */
3136 dev = &sh->dev[sh->pd_idx];
Dan Williamsa4456852007-07-09 11:56:43 -07003137 if ( s.written &&
3138 ((test_bit(R5_Insync, &dev->flags) &&
3139 !test_bit(R5_LOCKED, &dev->flags) &&
3140 test_bit(R5_UPTODATE, &dev->flags)) ||
3141 (s.failed == 1 && s.failed_num == sh->pd_idx)))
Dan Williams1fe797e2008-06-28 09:16:30 +10003142 handle_stripe_clean_event(conf, sh, disks, &return_bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143
3144 /* Now we might consider reading some blocks, either to check/generate
3145 * parity, or to satisfy requests
3146 * or to load a block that is being partially written.
3147 */
Dan Williamsa4456852007-07-09 11:56:43 -07003148 if (s.to_read || s.non_overwrite ||
Dan Williams976ea8d2008-06-28 08:32:03 +10003149 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
Dan Williams1fe797e2008-06-28 09:16:30 +10003150 handle_stripe_fill5(sh, &s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151
Dan Williamse33129d2007-01-02 13:52:30 -07003152 /* Now we check to see if any write operations have recently
3153 * completed
3154 */
Dan Williamse0a115e2008-06-05 22:45:52 -07003155 prexor = 0;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003156 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
Dan Williamse0a115e2008-06-05 22:45:52 -07003157 prexor = 1;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003158 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3159 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
Dan Williams600aa102008-06-28 08:32:05 +10003160 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamse33129d2007-01-02 13:52:30 -07003161
3162 /* All the 'written' buffers and the parity block are ready to
3163 * be written back to disk
3164 */
3165 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3166 for (i = disks; i--; ) {
3167 dev = &sh->dev[i];
3168 if (test_bit(R5_LOCKED, &dev->flags) &&
3169 (i == sh->pd_idx || dev->written)) {
3170 pr_debug("Writing block %d\n", i);
3171 set_bit(R5_Wantwrite, &dev->flags);
Dan Williamse0a115e2008-06-05 22:45:52 -07003172 if (prexor)
3173 continue;
Dan Williamse33129d2007-01-02 13:52:30 -07003174 if (!test_bit(R5_Insync, &dev->flags) ||
3175 (i == sh->pd_idx && s.failed == 0))
3176 set_bit(STRIPE_INSYNC, &sh->state);
3177 }
3178 }
NeilBrown729a1862009-12-14 12:49:50 +11003179 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3180 dec_preread_active = 1;
Dan Williamse33129d2007-01-02 13:52:30 -07003181 }
3182
3183 /* Now to consider new write requests and what else, if anything
3184 * should be read. We do not handle new writes when:
3185 * 1/ A 'write' operation (copy+xor) is already in flight.
3186 * 2/ A 'check' operation is in flight, as it may clobber the parity
3187 * block.
3188 */
Dan Williams600aa102008-06-28 08:32:05 +10003189 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
Dan Williams1fe797e2008-06-28 09:16:30 +10003190 handle_stripe_dirtying5(conf, sh, &s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191
3192 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamse89f8962007-01-02 13:52:31 -07003193 * Any reads will already have been scheduled, so we just see if enough
3194 * data is available. The parity check is held off while parity
3195 * dependent operations are in flight.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 */
Dan Williamsecc65c92008-06-28 08:31:57 +10003197 if (sh->check_state ||
3198 (s.syncing && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003199 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
Dan Williamsecc65c92008-06-28 08:31:57 +10003200 !test_bit(STRIPE_INSYNC, &sh->state)))
Dan Williamsa4456852007-07-09 11:56:43 -07003201 handle_parity_checks5(conf, sh, &s, disks);
Dan Williamse89f8962007-01-02 13:52:31 -07003202
Dan Williamsa4456852007-07-09 11:56:43 -07003203 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3205 clear_bit(STRIPE_SYNCING, &sh->state);
3206 }
NeilBrown4e5314b2005-11-08 21:39:22 -08003207
3208 /* If the failed drive is just a ReadError, then we might need to progress
3209 * the repair/check process
3210 */
Dan Williamsa4456852007-07-09 11:56:43 -07003211 if (s.failed == 1 && !conf->mddev->ro &&
3212 test_bit(R5_ReadError, &sh->dev[s.failed_num].flags)
3213 && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags)
3214 && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags)
NeilBrown4e5314b2005-11-08 21:39:22 -08003215 ) {
Dan Williamsa4456852007-07-09 11:56:43 -07003216 dev = &sh->dev[s.failed_num];
NeilBrown4e5314b2005-11-08 21:39:22 -08003217 if (!test_bit(R5_ReWrite, &dev->flags)) {
3218 set_bit(R5_Wantwrite, &dev->flags);
3219 set_bit(R5_ReWrite, &dev->flags);
3220 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003221 s.locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003222 } else {
3223 /* let's read it back */
3224 set_bit(R5_Wantread, &dev->flags);
3225 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003226 s.locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003227 }
3228 }
3229
Dan Williams600aa102008-06-28 08:32:05 +10003230 /* Finish reconstruct operations initiated by the expansion process */
3231 if (sh->reconstruct_state == reconstruct_state_result) {
NeilBrownab69ae12009-03-31 15:26:47 +11003232 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003233 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003234 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3235 /* sh cannot be written until sh2 has been read.
3236 * so arrange for sh to be delayed a little
3237 */
3238 set_bit(STRIPE_DELAYED, &sh->state);
3239 set_bit(STRIPE_HANDLE, &sh->state);
3240 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3241 &sh2->state))
3242 atomic_inc(&conf->preread_active_stripes);
3243 release_stripe(sh2);
3244 goto unlock;
3245 }
3246 if (sh2)
3247 release_stripe(sh2);
3248
Dan Williams600aa102008-06-28 08:32:05 +10003249 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamsf0a50d32007-01-02 13:52:31 -07003250 clear_bit(STRIPE_EXPANDING, &sh->state);
Dan Williams23397882008-07-23 20:05:34 -07003251 for (i = conf->raid_disks; i--; ) {
Dan Williamsf0a50d32007-01-02 13:52:31 -07003252 set_bit(R5_Wantwrite, &sh->dev[i].flags);
Dan Williams23397882008-07-23 20:05:34 -07003253 set_bit(R5_LOCKED, &sh->dev[i].flags);
Neil Brownefe31142008-06-28 08:31:14 +10003254 s.locked++;
Dan Williams23397882008-07-23 20:05:34 -07003255 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07003256 }
3257
3258 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
Dan Williams600aa102008-06-28 08:32:05 +10003259 !sh->reconstruct_state) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003260 /* Need to write out all blocks after computing parity */
3261 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003262 stripe_set_idx(sh->sector, conf, 0, sh);
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07003263 schedule_reconstruction(sh, &s, 1, 1);
Dan Williams600aa102008-06-28 08:32:05 +10003264 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003265 clear_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrownf6705572006-03-27 01:18:11 -08003266 atomic_dec(&conf->reshape_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003267 wake_up(&conf->wait_for_overlap);
3268 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3269 }
3270
Dan Williams0f94e872008-01-08 15:32:53 -08003271 if (s.expanding && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003272 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
Dan Williamsa4456852007-07-09 11:56:43 -07003273 handle_stripe_expansion(conf, sh, NULL);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003274
Dan Williams6bfe0b42008-04-30 00:52:32 -07003275 unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 spin_unlock(&sh->lock);
3277
Dan Williams6bfe0b42008-04-30 00:52:32 -07003278 /* wait for this device to become unblocked */
3279 if (unlikely(blocked_rdev))
3280 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3281
Dan Williams600aa102008-06-28 08:32:05 +10003282 if (s.ops_request)
Dan Williamsac6b53b2009-07-14 13:40:19 -07003283 raid_run_ops(sh, s.ops_request);
Dan Williamsd84e0f12007-01-02 13:52:30 -07003284
Dan Williamsc4e5ac02008-06-28 08:31:53 +10003285 ops_run_io(sh, &s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286
NeilBrown729a1862009-12-14 12:49:50 +11003287 if (dec_preread_active) {
3288 /* We delay this until after ops_run_io so that if make_request
Tejun Heoe9c74692010-09-03 11:56:18 +02003289 * is waiting on a flush, it won't continue until the writes
NeilBrown729a1862009-12-14 12:49:50 +11003290 * have actually been submitted.
3291 */
3292 atomic_dec(&conf->preread_active_stripes);
3293 if (atomic_read(&conf->preread_active_stripes) <
3294 IO_THRESHOLD)
3295 md_wakeup_thread(conf->mddev->thread);
3296 }
Dan Williamsa4456852007-07-09 11:56:43 -07003297 return_io(return_bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298}
3299
NeilBrown14425772009-10-16 15:55:25 +11003300static void handle_stripe6(struct stripe_head *sh)
NeilBrown16a53ec2006-06-26 00:27:38 -07003301{
NeilBrownbff61972009-03-31 14:33:13 +11003302 raid5_conf_t *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08003303 int disks = sh->disks;
Dan Williamsa4456852007-07-09 11:56:43 -07003304 struct bio *return_bi = NULL;
NeilBrown34e04e82009-03-31 15:10:16 +11003305 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
Dan Williamsa4456852007-07-09 11:56:43 -07003306 struct stripe_head_state s;
3307 struct r6_state r6s;
NeilBrown16a53ec2006-06-26 00:27:38 -07003308 struct r5dev *dev, *pdev, *qdev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003309 mdk_rdev_t *blocked_rdev = NULL;
NeilBrown729a1862009-12-14 12:49:50 +11003310 int dec_preread_active = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003311
Dan Williams45b42332007-07-09 11:56:43 -07003312 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003313 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003314 (unsigned long long)sh->sector, sh->state,
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003315 atomic_read(&sh->count), pd_idx, qd_idx,
3316 sh->check_state, sh->reconstruct_state);
Dan Williamsa4456852007-07-09 11:56:43 -07003317 memset(&s, 0, sizeof(s));
NeilBrown16a53ec2006-06-26 00:27:38 -07003318
3319 spin_lock(&sh->lock);
3320 clear_bit(STRIPE_HANDLE, &sh->state);
3321 clear_bit(STRIPE_DELAYED, &sh->state);
3322
Dan Williamsa4456852007-07-09 11:56:43 -07003323 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3324 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3325 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07003326 /* Now to look around and see what can be done */
3327
3328 rcu_read_lock();
3329 for (i=disks; i--; ) {
3330 mdk_rdev_t *rdev;
3331 dev = &sh->dev[i];
NeilBrown16a53ec2006-06-26 00:27:38 -07003332
Dan Williams45b42332007-07-09 11:56:43 -07003333 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown16a53ec2006-06-26 00:27:38 -07003334 i, dev->flags, dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003335 /* maybe we can reply to a read
3336 *
3337 * new wantfill requests are only permitted while
3338 * ops_complete_biofill is guaranteed to be inactive
3339 */
3340 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3341 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3342 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07003343
3344 /* now count some things */
Dan Williamsa4456852007-07-09 11:56:43 -07003345 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3346 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003347 if (test_bit(R5_Wantcompute, &dev->flags)) {
3348 s.compute++;
3349 BUG_ON(s.compute > 2);
3350 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003351
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003352 if (test_bit(R5_Wantfill, &dev->flags)) {
3353 s.to_fill++;
3354 } else if (dev->toread)
Dan Williamsa4456852007-07-09 11:56:43 -07003355 s.to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003356 if (dev->towrite) {
Dan Williamsa4456852007-07-09 11:56:43 -07003357 s.to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003358 if (!test_bit(R5_OVERWRITE, &dev->flags))
Dan Williamsa4456852007-07-09 11:56:43 -07003359 s.non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003360 }
Dan Williamsa4456852007-07-09 11:56:43 -07003361 if (dev->written)
3362 s.written++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003363 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownac4090d2008-08-05 15:54:13 +10003364 if (blocked_rdev == NULL &&
3365 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07003366 blocked_rdev = rdev;
3367 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003368 }
NeilBrown415e72d2010-06-17 17:25:21 +10003369 clear_bit(R5_Insync, &dev->flags);
3370 if (!rdev)
3371 /* Not in-sync */;
3372 else if (test_bit(In_sync, &rdev->flags))
3373 set_bit(R5_Insync, &dev->flags);
3374 else {
3375 /* in sync if before recovery_offset */
3376 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3377 set_bit(R5_Insync, &dev->flags);
3378 }
3379 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003380 /* The ReadError flag will just be confusing now */
3381 clear_bit(R5_ReadError, &dev->flags);
3382 clear_bit(R5_ReWrite, &dev->flags);
3383 }
NeilBrown415e72d2010-06-17 17:25:21 +10003384 if (test_bit(R5_ReadError, &dev->flags))
3385 clear_bit(R5_Insync, &dev->flags);
3386 if (!test_bit(R5_Insync, &dev->flags)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003387 if (s.failed < 2)
3388 r6s.failed_num[s.failed] = i;
3389 s.failed++;
NeilBrown415e72d2010-06-17 17:25:21 +10003390 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003391 }
3392 rcu_read_unlock();
Dan Williams6bfe0b42008-04-30 00:52:32 -07003393
3394 if (unlikely(blocked_rdev)) {
NeilBrownac4090d2008-08-05 15:54:13 +10003395 if (s.syncing || s.expanding || s.expanded ||
3396 s.to_write || s.written) {
3397 set_bit(STRIPE_HANDLE, &sh->state);
3398 goto unlock;
3399 }
3400 /* There is nothing for the blocked_rdev to block */
3401 rdev_dec_pending(blocked_rdev, conf->mddev);
3402 blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003403 }
NeilBrownac4090d2008-08-05 15:54:13 +10003404
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003405 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3406 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3407 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3408 }
3409
Dan Williams45b42332007-07-09 11:56:43 -07003410 pr_debug("locked=%d uptodate=%d to_read=%d"
NeilBrown16a53ec2006-06-26 00:27:38 -07003411 " to_write=%d failed=%d failed_num=%d,%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003412 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3413 r6s.failed_num[0], r6s.failed_num[1]);
3414 /* check if the array has lost >2 devices and, if so, some requests
3415 * might need to be failed
NeilBrown16a53ec2006-06-26 00:27:38 -07003416 */
Dan Williamsa4456852007-07-09 11:56:43 -07003417 if (s.failed > 2 && s.to_read+s.to_write+s.written)
Dan Williams1fe797e2008-06-28 09:16:30 +10003418 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003419 if (s.failed > 2 && s.syncing) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003420 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3421 clear_bit(STRIPE_SYNCING, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003422 s.syncing = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003423 }
3424
3425 /*
3426 * might be able to return some write requests if the parity blocks
3427 * are safe, or on a failed drive
3428 */
3429 pdev = &sh->dev[pd_idx];
Dan Williamsa4456852007-07-09 11:56:43 -07003430 r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx)
3431 || (s.failed >= 2 && r6s.failed_num[1] == pd_idx);
NeilBrown34e04e82009-03-31 15:10:16 +11003432 qdev = &sh->dev[qd_idx];
3433 r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx)
3434 || (s.failed >= 2 && r6s.failed_num[1] == qd_idx);
NeilBrown16a53ec2006-06-26 00:27:38 -07003435
Dan Williamsa4456852007-07-09 11:56:43 -07003436 if ( s.written &&
3437 ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003438 && !test_bit(R5_LOCKED, &pdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003439 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
3440 ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003441 && !test_bit(R5_LOCKED, &qdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003442 && test_bit(R5_UPTODATE, &qdev->flags)))))
Dan Williams1fe797e2008-06-28 09:16:30 +10003443 handle_stripe_clean_event(conf, sh, disks, &return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003444
3445 /* Now we might consider reading some blocks, either to check/generate
3446 * parity, or to satisfy requests
3447 * or to load a block that is being partially written.
3448 */
Dan Williamsa4456852007-07-09 11:56:43 -07003449 if (s.to_read || s.non_overwrite || (s.to_write && s.failed) ||
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003450 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
Dan Williams1fe797e2008-06-28 09:16:30 +10003451 handle_stripe_fill6(sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003452
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003453 /* Now we check to see if any write operations have recently
3454 * completed
3455 */
3456 if (sh->reconstruct_state == reconstruct_state_drain_result) {
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003457
3458 sh->reconstruct_state = reconstruct_state_idle;
3459 /* All the 'written' buffers and the parity blocks are ready to
3460 * be written back to disk
3461 */
3462 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3463 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags));
3464 for (i = disks; i--; ) {
3465 dev = &sh->dev[i];
3466 if (test_bit(R5_LOCKED, &dev->flags) &&
3467 (i == sh->pd_idx || i == qd_idx ||
3468 dev->written)) {
3469 pr_debug("Writing block %d\n", i);
3470 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3471 set_bit(R5_Wantwrite, &dev->flags);
3472 if (!test_bit(R5_Insync, &dev->flags) ||
3473 ((i == sh->pd_idx || i == qd_idx) &&
3474 s.failed == 0))
3475 set_bit(STRIPE_INSYNC, &sh->state);
3476 }
3477 }
NeilBrown729a1862009-12-14 12:49:50 +11003478 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3479 dec_preread_active = 1;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003480 }
3481
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07003482 /* Now to consider new write requests and what else, if anything
3483 * should be read. We do not handle new writes when:
3484 * 1/ A 'write' operation (copy+gen_syndrome) is already in flight.
3485 * 2/ A 'check' operation is in flight, as it may clobber the parity
3486 * block.
3487 */
3488 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
Dan Williams1fe797e2008-06-28 09:16:30 +10003489 handle_stripe_dirtying6(conf, sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003490
3491 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamsa4456852007-07-09 11:56:43 -07003492 * Any reads will already have been scheduled, so we just see if enough
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003493 * data is available. The parity check is held off while parity
3494 * dependent operations are in flight.
NeilBrown16a53ec2006-06-26 00:27:38 -07003495 */
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003496 if (sh->check_state ||
3497 (s.syncing && s.locked == 0 &&
3498 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3499 !test_bit(STRIPE_INSYNC, &sh->state)))
Dan Williams36d1c642009-07-14 11:48:22 -07003500 handle_parity_checks6(conf, sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003501
Dan Williamsa4456852007-07-09 11:56:43 -07003502 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003503 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3504 clear_bit(STRIPE_SYNCING, &sh->state);
3505 }
3506
3507 /* If the failed drives are just a ReadError, then we might need
3508 * to progress the repair/check process
3509 */
Dan Williamsa4456852007-07-09 11:56:43 -07003510 if (s.failed <= 2 && !conf->mddev->ro)
3511 for (i = 0; i < s.failed; i++) {
3512 dev = &sh->dev[r6s.failed_num[i]];
NeilBrown16a53ec2006-06-26 00:27:38 -07003513 if (test_bit(R5_ReadError, &dev->flags)
3514 && !test_bit(R5_LOCKED, &dev->flags)
3515 && test_bit(R5_UPTODATE, &dev->flags)
3516 ) {
3517 if (!test_bit(R5_ReWrite, &dev->flags)) {
3518 set_bit(R5_Wantwrite, &dev->flags);
3519 set_bit(R5_ReWrite, &dev->flags);
3520 set_bit(R5_LOCKED, &dev->flags);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003521 s.locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003522 } else {
3523 /* let's read it back */
3524 set_bit(R5_Wantread, &dev->flags);
3525 set_bit(R5_LOCKED, &dev->flags);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003526 s.locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003527 }
3528 }
3529 }
NeilBrownf4168852007-02-28 20:11:53 -08003530
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003531 /* Finish reconstruct operations initiated by the expansion process */
3532 if (sh->reconstruct_state == reconstruct_state_result) {
3533 sh->reconstruct_state = reconstruct_state_idle;
3534 clear_bit(STRIPE_EXPANDING, &sh->state);
3535 for (i = conf->raid_disks; i--; ) {
3536 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3537 set_bit(R5_LOCKED, &sh->dev[i].flags);
3538 s.locked++;
3539 }
3540 }
3541
3542 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3543 !sh->reconstruct_state) {
NeilBrownab69ae12009-03-31 15:26:47 +11003544 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003545 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003546 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3547 /* sh cannot be written until sh2 has been read.
3548 * so arrange for sh to be delayed a little
3549 */
3550 set_bit(STRIPE_DELAYED, &sh->state);
3551 set_bit(STRIPE_HANDLE, &sh->state);
3552 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3553 &sh2->state))
3554 atomic_inc(&conf->preread_active_stripes);
3555 release_stripe(sh2);
3556 goto unlock;
3557 }
3558 if (sh2)
3559 release_stripe(sh2);
3560
NeilBrownf4168852007-02-28 20:11:53 -08003561 /* Need to write out all blocks after computing P&Q */
3562 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003563 stripe_set_idx(sh->sector, conf, 0, sh);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003564 schedule_reconstruction(sh, &s, 1, 1);
3565 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
NeilBrownf4168852007-02-28 20:11:53 -08003566 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3567 atomic_dec(&conf->reshape_stripes);
3568 wake_up(&conf->wait_for_overlap);
3569 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3570 }
3571
Dan Williams0f94e872008-01-08 15:32:53 -08003572 if (s.expanding && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003573 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
Dan Williamsa4456852007-07-09 11:56:43 -07003574 handle_stripe_expansion(conf, sh, &r6s);
NeilBrownf4168852007-02-28 20:11:53 -08003575
Dan Williams6bfe0b42008-04-30 00:52:32 -07003576 unlock:
NeilBrown16a53ec2006-06-26 00:27:38 -07003577 spin_unlock(&sh->lock);
3578
Dan Williams6bfe0b42008-04-30 00:52:32 -07003579 /* wait for this device to become unblocked */
3580 if (unlikely(blocked_rdev))
3581 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3582
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003583 if (s.ops_request)
3584 raid_run_ops(sh, s.ops_request);
3585
Dan Williamsf0e43bc2008-06-28 08:31:55 +10003586 ops_run_io(sh, &s);
3587
NeilBrown729a1862009-12-14 12:49:50 +11003588
3589 if (dec_preread_active) {
3590 /* We delay this until after ops_run_io so that if make_request
Tejun Heoe9c74692010-09-03 11:56:18 +02003591 * is waiting on a flush, it won't continue until the writes
NeilBrown729a1862009-12-14 12:49:50 +11003592 * have actually been submitted.
3593 */
3594 atomic_dec(&conf->preread_active_stripes);
3595 if (atomic_read(&conf->preread_active_stripes) <
3596 IO_THRESHOLD)
3597 md_wakeup_thread(conf->mddev->thread);
3598 }
3599
Dan Williamsa4456852007-07-09 11:56:43 -07003600 return_io(return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003601}
3602
NeilBrown14425772009-10-16 15:55:25 +11003603static void handle_stripe(struct stripe_head *sh)
NeilBrown16a53ec2006-06-26 00:27:38 -07003604{
3605 if (sh->raid_conf->level == 6)
NeilBrown14425772009-10-16 15:55:25 +11003606 handle_stripe6(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -07003607 else
NeilBrown14425772009-10-16 15:55:25 +11003608 handle_stripe5(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -07003609}
3610
Arjan van de Ven858119e2006-01-14 13:20:43 -08003611static void raid5_activate_delayed(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612{
3613 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3614 while (!list_empty(&conf->delayed_list)) {
3615 struct list_head *l = conf->delayed_list.next;
3616 struct stripe_head *sh;
3617 sh = list_entry(l, struct stripe_head, lru);
3618 list_del_init(l);
3619 clear_bit(STRIPE_DELAYED, &sh->state);
3620 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3621 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003622 list_add_tail(&sh->lru, &conf->hold_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 }
NeilBrown482c0832011-04-18 18:25:42 +10003624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625}
3626
Arjan van de Ven858119e2006-01-14 13:20:43 -08003627static void activate_bit_delay(raid5_conf_t *conf)
NeilBrown72626682005-09-09 16:23:54 -07003628{
3629 /* device_lock is held */
3630 struct list_head head;
3631 list_add(&head, &conf->bitmap_list);
3632 list_del_init(&conf->bitmap_list);
3633 while (!list_empty(&head)) {
3634 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3635 list_del_init(&sh->lru);
3636 atomic_inc(&sh->count);
3637 __release_stripe(conf, sh);
3638 }
3639}
3640
NeilBrown11d8a6e2010-07-26 11:57:07 +10003641int md_raid5_congested(mddev_t *mddev, int bits)
NeilBrownf022b2f2006-10-03 01:15:56 -07003642{
NeilBrown070ec552009-06-16 16:54:21 +10003643 raid5_conf_t *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07003644
3645 /* No difference between reads and writes. Just check
3646 * how busy the stripe_cache is
3647 */
NeilBrown3fa841d2009-09-23 18:10:29 +10003648
NeilBrownf022b2f2006-10-03 01:15:56 -07003649 if (conf->inactive_blocked)
3650 return 1;
3651 if (conf->quiesce)
3652 return 1;
3653 if (list_empty_careful(&conf->inactive_list))
3654 return 1;
3655
3656 return 0;
3657}
NeilBrown11d8a6e2010-07-26 11:57:07 +10003658EXPORT_SYMBOL_GPL(md_raid5_congested);
3659
3660static int raid5_congested(void *data, int bits)
3661{
3662 mddev_t *mddev = data;
3663
3664 return mddev_congested(mddev, bits) ||
3665 md_raid5_congested(mddev, bits);
3666}
NeilBrownf022b2f2006-10-03 01:15:56 -07003667
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003668/* We want read requests to align with chunks where possible,
3669 * but write requests don't need to.
3670 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003671static int raid5_mergeable_bvec(struct request_queue *q,
3672 struct bvec_merge_data *bvm,
3673 struct bio_vec *biovec)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003674{
3675 mddev_t *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003676 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003677 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +10003678 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003679 unsigned int bio_sectors = bvm->bi_size >> 9;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003680
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003681 if ((bvm->bi_rw & 1) == WRITE)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003682 return biovec->bv_len; /* always allow writes to be mergeable */
3683
Andre Noll664e7c42009-06-18 08:45:27 +10003684 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3685 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003686 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3687 if (max < 0) max = 0;
3688 if (max <= biovec->bv_len && bio_sectors == 0)
3689 return biovec->bv_len;
3690 else
3691 return max;
3692}
3693
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003694
3695static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3696{
3697 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
Andre Noll9d8f0362009-06-18 08:45:01 +10003698 unsigned int chunk_sectors = mddev->chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003699 unsigned int bio_sectors = bio->bi_size >> 9;
3700
Andre Noll664e7c42009-06-18 08:45:27 +10003701 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3702 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003703 return chunk_sectors >=
3704 ((sector & (chunk_sectors - 1)) + bio_sectors);
3705}
3706
3707/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003708 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3709 * later sampled by raid5d.
3710 */
3711static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3712{
3713 unsigned long flags;
3714
3715 spin_lock_irqsave(&conf->device_lock, flags);
3716
3717 bi->bi_next = conf->retry_read_aligned_list;
3718 conf->retry_read_aligned_list = bi;
3719
3720 spin_unlock_irqrestore(&conf->device_lock, flags);
3721 md_wakeup_thread(conf->mddev->thread);
3722}
3723
3724
3725static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3726{
3727 struct bio *bi;
3728
3729 bi = conf->retry_read_aligned;
3730 if (bi) {
3731 conf->retry_read_aligned = NULL;
3732 return bi;
3733 }
3734 bi = conf->retry_read_aligned_list;
3735 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08003736 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003737 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02003738 /*
3739 * this sets the active strip count to 1 and the processed
3740 * strip count to zero (upper 8 bits)
3741 */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003742 bi->bi_phys_segments = 1; /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003743 }
3744
3745 return bi;
3746}
3747
3748
3749/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003750 * The "raid5_align_endio" should check if the read succeeded and if it
3751 * did, call bio_endio on the original bio (having bio_put the new bio
3752 * first).
3753 * If the read failed..
3754 */
NeilBrown6712ecf2007-09-27 12:47:43 +02003755static void raid5_align_endio(struct bio *bi, int error)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003756{
3757 struct bio* raid_bi = bi->bi_private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003758 mddev_t *mddev;
3759 raid5_conf_t *conf;
3760 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3761 mdk_rdev_t *rdev;
3762
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003763 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003764
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003765 rdev = (void*)raid_bi->bi_next;
3766 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11003767 mddev = rdev->mddev;
3768 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003769
3770 rdev_dec_pending(rdev, conf->mddev);
3771
3772 if (!error && uptodate) {
NeilBrown6712ecf2007-09-27 12:47:43 +02003773 bio_endio(raid_bi, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003774 if (atomic_dec_and_test(&conf->active_aligned_reads))
3775 wake_up(&conf->wait_for_stripe);
NeilBrown6712ecf2007-09-27 12:47:43 +02003776 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003777 }
3778
3779
Dan Williams45b42332007-07-09 11:56:43 -07003780 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003781
3782 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003783}
3784
Neil Brown387bb172007-02-08 14:20:29 -08003785static int bio_fits_rdev(struct bio *bi)
3786{
Jens Axboe165125e2007-07-24 09:28:11 +02003787 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
Neil Brown387bb172007-02-08 14:20:29 -08003788
Martin K. Petersenae03bf62009-05-22 17:17:50 -04003789 if ((bi->bi_size>>9) > queue_max_sectors(q))
Neil Brown387bb172007-02-08 14:20:29 -08003790 return 0;
3791 blk_recount_segments(q, bi);
Martin K. Petersen8a783622010-02-26 00:20:39 -05003792 if (bi->bi_phys_segments > queue_max_segments(q))
Neil Brown387bb172007-02-08 14:20:29 -08003793 return 0;
3794
3795 if (q->merge_bvec_fn)
3796 /* it's too hard to apply the merge_bvec_fn at this stage,
3797 * just just give up
3798 */
3799 return 0;
3800
3801 return 1;
3802}
3803
3804
NeilBrown21a52c62010-04-01 15:02:13 +11003805static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003806{
NeilBrown070ec552009-06-16 16:54:21 +10003807 raid5_conf_t *conf = mddev->private;
NeilBrown8553fe7ec2009-12-14 12:49:47 +11003808 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003809 struct bio* align_bi;
3810 mdk_rdev_t *rdev;
3811
3812 if (!in_chunk_boundary(mddev, raid_bio)) {
Dan Williams45b42332007-07-09 11:56:43 -07003813 pr_debug("chunk_aligned_read : non aligned\n");
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003814 return 0;
3815 }
3816 /*
NeilBrowna167f662010-10-26 18:31:13 +11003817 * use bio_clone_mddev to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003818 */
NeilBrowna167f662010-10-26 18:31:13 +11003819 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003820 if (!align_bi)
3821 return 0;
3822 /*
3823 * set bi_end_io to a new function, and set bi_private to the
3824 * original bio.
3825 */
3826 align_bi->bi_end_io = raid5_align_endio;
3827 align_bi->bi_private = raid_bio;
3828 /*
3829 * compute position
3830 */
NeilBrown112bf892009-03-31 14:39:38 +11003831 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3832 0,
NeilBrown911d4ee2009-03-31 14:39:38 +11003833 &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003834
3835 rcu_read_lock();
3836 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3837 if (rdev && test_bit(In_sync, &rdev->flags)) {
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003838 atomic_inc(&rdev->nr_pending);
3839 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003840 raid_bio->bi_next = (void*)rdev;
3841 align_bi->bi_bdev = rdev->bdev;
3842 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3843 align_bi->bi_sector += rdev->data_offset;
3844
Neil Brown387bb172007-02-08 14:20:29 -08003845 if (!bio_fits_rdev(align_bi)) {
3846 /* too big in some way */
3847 bio_put(align_bi);
3848 rdev_dec_pending(rdev, mddev);
3849 return 0;
3850 }
3851
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003852 spin_lock_irq(&conf->device_lock);
3853 wait_event_lock_irq(conf->wait_for_stripe,
3854 conf->quiesce == 0,
3855 conf->device_lock, /* nothing */);
3856 atomic_inc(&conf->active_aligned_reads);
3857 spin_unlock_irq(&conf->device_lock);
3858
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003859 generic_make_request(align_bi);
3860 return 1;
3861 } else {
3862 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003863 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003864 return 0;
3865 }
3866}
3867
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003868/* __get_priority_stripe - get the next stripe to process
3869 *
3870 * Full stripe writes are allowed to pass preread active stripes up until
3871 * the bypass_threshold is exceeded. In general the bypass_count
3872 * increments when the handle_list is handled before the hold_list; however, it
3873 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3874 * stripe with in flight i/o. The bypass_count will be reset when the
3875 * head of the hold_list has changed, i.e. the head was promoted to the
3876 * handle_list.
3877 */
3878static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3879{
3880 struct stripe_head *sh;
3881
3882 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3883 __func__,
3884 list_empty(&conf->handle_list) ? "empty" : "busy",
3885 list_empty(&conf->hold_list) ? "empty" : "busy",
3886 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3887
3888 if (!list_empty(&conf->handle_list)) {
3889 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3890
3891 if (list_empty(&conf->hold_list))
3892 conf->bypass_count = 0;
3893 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3894 if (conf->hold_list.next == conf->last_hold)
3895 conf->bypass_count++;
3896 else {
3897 conf->last_hold = conf->hold_list.next;
3898 conf->bypass_count -= conf->bypass_threshold;
3899 if (conf->bypass_count < 0)
3900 conf->bypass_count = 0;
3901 }
3902 }
3903 } else if (!list_empty(&conf->hold_list) &&
3904 ((conf->bypass_threshold &&
3905 conf->bypass_count > conf->bypass_threshold) ||
3906 atomic_read(&conf->pending_full_writes) == 0)) {
3907 sh = list_entry(conf->hold_list.next,
3908 typeof(*sh), lru);
3909 conf->bypass_count -= conf->bypass_threshold;
3910 if (conf->bypass_count < 0)
3911 conf->bypass_count = 0;
3912 } else
3913 return NULL;
3914
3915 list_del_init(&sh->lru);
3916 atomic_inc(&sh->count);
3917 BUG_ON(atomic_read(&sh->count) != 1);
3918 return sh;
3919}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003920
NeilBrown21a52c62010-04-01 15:02:13 +11003921static int make_request(mddev_t *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922{
NeilBrown070ec552009-06-16 16:54:21 +10003923 raid5_conf_t *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11003924 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925 sector_t new_sector;
3926 sector_t logical_sector, last_sector;
3927 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01003928 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11003929 int remaining;
NeilBrown7c13edc2011-04-18 18:25:43 +10003930 int plugged;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931
Tejun Heoe9c74692010-09-03 11:56:18 +02003932 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
3933 md_flush_request(mddev, bi);
NeilBrowne5dcdd82005-09-09 16:23:41 -07003934 return 0;
3935 }
3936
NeilBrown3d310eb2005-06-21 17:17:26 -07003937 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07003938
NeilBrown802ba062006-12-13 00:34:13 -08003939 if (rw == READ &&
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003940 mddev->reshape_position == MaxSector &&
NeilBrown21a52c62010-04-01 15:02:13 +11003941 chunk_aligned_read(mddev,bi))
NeilBrown99c0fb52009-03-31 14:39:38 +11003942 return 0;
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003943
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3945 last_sector = bi->bi_sector + (bi->bi_size>>9);
3946 bi->bi_next = NULL;
3947 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07003948
NeilBrown7c13edc2011-04-18 18:25:43 +10003949 plugged = mddev_check_plugged(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3951 DEFINE_WAIT(w);
NeilBrown16a53ec2006-06-26 00:27:38 -07003952 int disks, data_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003953 int previous;
NeilBrownb578d552006-03-27 01:18:12 -08003954
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003955 retry:
NeilBrownb5663ba2009-03-31 14:39:38 +11003956 previous = 0;
NeilBrownb0f9ec02009-03-31 15:27:18 +11003957 disks = conf->raid_disks;
NeilBrownb578d552006-03-27 01:18:12 -08003958 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003959 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11003960 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08003961 * 64bit on a 32bit platform, and so it might be
3962 * possible to see a half-updated value
NeilBrownfef9c612009-03-31 15:16:46 +11003963 * Ofcourse reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08003964 * the lock is dropped, so once we get a reference
3965 * to the stripe that we think it is, we will have
3966 * to check again.
3967 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003968 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11003969 if (mddev->delta_disks < 0
3970 ? logical_sector < conf->reshape_progress
3971 : logical_sector >= conf->reshape_progress) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003972 disks = conf->previous_raid_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003973 previous = 1;
3974 } else {
NeilBrownfef9c612009-03-31 15:16:46 +11003975 if (mddev->delta_disks < 0
3976 ? logical_sector < conf->reshape_safe
3977 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08003978 spin_unlock_irq(&conf->device_lock);
3979 schedule();
3980 goto retry;
3981 }
3982 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003983 spin_unlock_irq(&conf->device_lock);
3984 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003985 data_disks = disks - conf->max_degraded;
3986
NeilBrown112bf892009-03-31 14:39:38 +11003987 new_sector = raid5_compute_sector(conf, logical_sector,
3988 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11003989 &dd_idx, NULL);
NeilBrown0c55e022010-05-03 14:09:02 +10003990 pr_debug("raid456: make_request, sector %llu logical %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 (unsigned long long)new_sector,
3992 (unsigned long long)logical_sector);
3993
NeilBrownb5663ba2009-03-31 14:39:38 +11003994 sh = get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10003995 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11003997 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003998 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08003999 * stripe, so we must do the range check again.
4000 * Expansion could still move past after this
4001 * test, but as we are holding a reference to
4002 * 'sh', we know that if that happens,
4003 * STRIPE_EXPANDING will get set and the expansion
4004 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004005 */
4006 int must_retry = 0;
4007 spin_lock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11004008 if (mddev->delta_disks < 0
4009 ? logical_sector >= conf->reshape_progress
4010 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004011 /* mismatch, need to try again */
4012 must_retry = 1;
4013 spin_unlock_irq(&conf->device_lock);
4014 if (must_retry) {
4015 release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07004016 schedule();
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004017 goto retry;
4018 }
4019 }
NeilBrowne62e58a2009-07-01 13:15:35 +10004020
NeilBrowna5c308d2009-07-01 13:15:35 +10004021 if (bio_data_dir(bi) == WRITE &&
4022 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08004023 logical_sector < mddev->suspend_hi) {
4024 release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10004025 /* As the suspend_* range is controlled by
4026 * userspace, we want an interruptible
4027 * wait.
4028 */
4029 flush_signals(current);
4030 prepare_to_wait(&conf->wait_for_overlap,
4031 &w, TASK_INTERRUPTIBLE);
4032 if (logical_sector >= mddev->suspend_lo &&
4033 logical_sector < mddev->suspend_hi)
4034 schedule();
NeilBrowne464eaf2006-03-27 01:18:14 -08004035 goto retry;
4036 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004037
4038 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
4039 !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) {
4040 /* Stripe is busy expanding or
4041 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 * and wait a while
4043 */
NeilBrown482c0832011-04-18 18:25:42 +10004044 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 release_stripe(sh);
4046 schedule();
4047 goto retry;
4048 }
4049 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown6ed30032008-02-06 01:40:00 -08004050 set_bit(STRIPE_HANDLE, &sh->state);
4051 clear_bit(STRIPE_DELAYED, &sh->state);
Tejun Heoe9c74692010-09-03 11:56:18 +02004052 if ((bi->bi_rw & REQ_SYNC) &&
NeilBrown729a1862009-12-14 12:49:50 +11004053 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4054 atomic_inc(&conf->preread_active_stripes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056 } else {
4057 /* cannot get stripe for read-ahead, just give-up */
4058 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4059 finish_wait(&conf->wait_for_overlap, &w);
4060 break;
4061 }
4062
4063 }
NeilBrown7c13edc2011-04-18 18:25:43 +10004064 if (!plugged)
4065 md_wakeup_thread(mddev->thread);
4066
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004068 remaining = raid5_dec_bi_phys_segments(bi);
NeilBrownf6344752006-03-27 01:18:17 -08004069 spin_unlock_irq(&conf->device_lock);
4070 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071
NeilBrown16a53ec2006-06-26 00:27:38 -07004072 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02004074
Neil Brown0e13fe232008-06-28 08:31:20 +10004075 bio_endio(bi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076 }
NeilBrown729a1862009-12-14 12:49:50 +11004077
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078 return 0;
4079}
4080
Dan Williamsb522adc2009-03-31 15:00:31 +11004081static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
4082
NeilBrown52c03292006-06-26 00:27:43 -07004083static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084{
NeilBrown52c03292006-06-26 00:27:43 -07004085 /* reshaping is quite different to recovery/resync so it is
4086 * handled quite separately ... here.
4087 *
4088 * On each call to sync_request, we gather one chunk worth of
4089 * destination stripes and flag them as expanding.
4090 * Then we find all the source stripes and request reads.
4091 * As the reads complete, handle_stripe will copy the data
4092 * into the destination stripe and release that stripe.
4093 */
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004094 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004095 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08004096 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08004097 int raid_disks = conf->previous_raid_disks;
4098 int data_disks = raid_disks - conf->max_degraded;
4099 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07004100 int i;
4101 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11004102 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11004103 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11004104 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11004105 struct list_head stripes;
NeilBrown52c03292006-06-26 00:27:43 -07004106
NeilBrownfef9c612009-03-31 15:16:46 +11004107 if (sector_nr == 0) {
4108 /* If restarting in the middle, skip the initial sectors */
4109 if (mddev->delta_disks < 0 &&
4110 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4111 sector_nr = raid5_size(mddev, 0, 0)
4112 - conf->reshape_progress;
NeilBrowna6397552009-08-13 10:13:00 +10004113 } else if (mddev->delta_disks >= 0 &&
NeilBrownfef9c612009-03-31 15:16:46 +11004114 conf->reshape_progress > 0)
4115 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004116 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004117 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11004118 mddev->curr_resync_completed = sector_nr;
4119 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11004120 *skipped = 1;
4121 return sector_nr;
4122 }
NeilBrown52c03292006-06-26 00:27:43 -07004123 }
4124
NeilBrown7a661382009-03-31 15:21:40 +11004125 /* We need to process a full chunk at a time.
4126 * If old and new chunk sizes differ, we need to process the
4127 * largest of these
4128 */
Andre Noll664e7c42009-06-18 08:45:27 +10004129 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4130 reshape_sectors = mddev->new_chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004131 else
Andre Noll9d8f0362009-06-18 08:45:01 +10004132 reshape_sectors = mddev->chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004133
NeilBrown52c03292006-06-26 00:27:43 -07004134 /* we update the metadata when there is more than 3Meg
4135 * in the block range (that is rather arbitrary, should
4136 * probably be time based) or when the data about to be
4137 * copied would over-write the source of the data at
4138 * the front of the range.
NeilBrownfef9c612009-03-31 15:16:46 +11004139 * i.e. one new_stripe along from reshape_progress new_maps
4140 * to after where reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07004141 */
NeilBrownfef9c612009-03-31 15:16:46 +11004142 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004143 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11004144 readpos = conf->reshape_progress;
4145 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004146 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08004147 sector_div(safepos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004148 if (mddev->delta_disks < 0) {
NeilBrowned37d832009-05-27 21:39:05 +10004149 writepos -= min_t(sector_t, reshape_sectors, writepos);
NeilBrownc8f517c2009-03-31 15:28:40 +11004150 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004151 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11004152 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004153 writepos += reshape_sectors;
NeilBrowned37d832009-05-27 21:39:05 +10004154 readpos -= min_t(sector_t, reshape_sectors, readpos);
4155 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11004156 }
NeilBrown52c03292006-06-26 00:27:43 -07004157
NeilBrownc8f517c2009-03-31 15:28:40 +11004158 /* 'writepos' is the most advanced device address we might write.
4159 * 'readpos' is the least advanced device address we might read.
4160 * 'safepos' is the least address recorded in the metadata as having
4161 * been reshaped.
4162 * If 'readpos' is behind 'writepos', then there is no way that we can
4163 * ensure safety in the face of a crash - that must be done by userspace
4164 * making a backup of the data. So in that case there is no particular
4165 * rush to update metadata.
4166 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4167 * update the metadata to advance 'safepos' to match 'readpos' so that
4168 * we can be safe in the event of a crash.
4169 * So we insist on updating metadata if safepos is behind writepos and
4170 * readpos is beyond writepos.
4171 * In any case, update the metadata every 10 seconds.
4172 * Maybe that number should be configurable, but I'm not sure it is
4173 * worth it.... maybe it could be a multiple of safemode_delay???
4174 */
NeilBrownfef9c612009-03-31 15:16:46 +11004175 if ((mddev->delta_disks < 0
NeilBrownc8f517c2009-03-31 15:28:40 +11004176 ? (safepos > writepos && readpos < writepos)
4177 : (safepos < writepos && readpos > writepos)) ||
4178 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07004179 /* Cannot proceed until we've updated the superblock... */
4180 wait_event(conf->wait_for_overlap,
4181 atomic_read(&conf->reshape_stripes)==0);
NeilBrownfef9c612009-03-31 15:16:46 +11004182 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004183 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004184 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07004185 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07004186 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07004187 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrown52c03292006-06-26 00:27:43 -07004188 kthread_should_stop());
4189 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004190 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07004191 spin_unlock_irq(&conf->device_lock);
4192 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004193 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07004194 }
4195
NeilBrownec32a2b2009-03-31 15:17:38 +11004196 if (mddev->delta_disks < 0) {
4197 BUG_ON(conf->reshape_progress == 0);
4198 stripe_addr = writepos;
4199 BUG_ON((mddev->dev_sectors &
NeilBrown7a661382009-03-31 15:21:40 +11004200 ~((sector_t)reshape_sectors - 1))
4201 - reshape_sectors - stripe_addr
NeilBrownec32a2b2009-03-31 15:17:38 +11004202 != sector_nr);
4203 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004204 BUG_ON(writepos != sector_nr + reshape_sectors);
NeilBrownec32a2b2009-03-31 15:17:38 +11004205 stripe_addr = sector_nr;
4206 }
NeilBrownab69ae12009-03-31 15:26:47 +11004207 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11004208 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07004209 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10004210 int skipped_disk = 0;
NeilBrowna8c906c2009-06-09 14:39:59 +10004211 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004212 set_bit(STRIPE_EXPANDING, &sh->state);
4213 atomic_inc(&conf->reshape_stripes);
4214 /* If any of this stripe is beyond the end of the old
4215 * array, then we need to zero those blocks
4216 */
4217 for (j=sh->disks; j--;) {
4218 sector_t s;
4219 if (j == sh->pd_idx)
4220 continue;
NeilBrownf4168852007-02-28 20:11:53 -08004221 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11004222 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08004223 continue;
NeilBrown784052e2009-03-31 15:19:07 +11004224 s = compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11004225 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10004226 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07004227 continue;
4228 }
4229 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4230 set_bit(R5_Expanded, &sh->dev[j].flags);
4231 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4232 }
NeilBrowna9f326e2009-09-23 18:06:41 +10004233 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07004234 set_bit(STRIPE_EXPAND_READY, &sh->state);
4235 set_bit(STRIPE_HANDLE, &sh->state);
4236 }
NeilBrownab69ae12009-03-31 15:26:47 +11004237 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07004238 }
4239 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004240 if (mddev->delta_disks < 0)
NeilBrown7a661382009-03-31 15:21:40 +11004241 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11004242 else
NeilBrown7a661382009-03-31 15:21:40 +11004243 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07004244 spin_unlock_irq(&conf->device_lock);
4245 /* Ok, those stripe are ready. We can start scheduling
4246 * reads on the source stripes.
4247 * The source stripes are determined by mapping the first and last
4248 * block on the destination stripes.
4249 */
NeilBrown52c03292006-06-26 00:27:43 -07004250 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11004251 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11004252 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07004253 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10004254 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10004255 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11004256 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11004257 if (last_sector >= mddev->dev_sectors)
4258 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07004259 while (first_sector <= last_sector) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004260 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004261 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4262 set_bit(STRIPE_HANDLE, &sh->state);
4263 release_stripe(sh);
4264 first_sector += STRIPE_SECTORS;
4265 }
NeilBrownab69ae12009-03-31 15:26:47 +11004266 /* Now that the sources are clearly marked, we can release
4267 * the destination stripes
4268 */
4269 while (!list_empty(&stripes)) {
4270 sh = list_entry(stripes.next, struct stripe_head, lru);
4271 list_del_init(&sh->lru);
4272 release_stripe(sh);
4273 }
NeilBrownc6207272008-02-06 01:39:52 -08004274 /* If this takes us to the resync_max point where we have to pause,
4275 * then we need to write out the superblock.
4276 */
NeilBrown7a661382009-03-31 15:21:40 +11004277 sector_nr += reshape_sectors;
NeilBrownc03f6a12009-04-17 11:06:30 +10004278 if ((sector_nr - mddev->curr_resync_completed) * 2
4279 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08004280 /* Cannot proceed until we've updated the superblock... */
4281 wait_event(conf->wait_for_overlap,
4282 atomic_read(&conf->reshape_stripes) == 0);
NeilBrownfef9c612009-03-31 15:16:46 +11004283 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004284 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004285 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08004286 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4287 md_wakeup_thread(mddev->thread);
4288 wait_event(mddev->sb_wait,
4289 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4290 || kthread_should_stop());
4291 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004292 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08004293 spin_unlock_irq(&conf->device_lock);
4294 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004295 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08004296 }
NeilBrown7a661382009-03-31 15:21:40 +11004297 return reshape_sectors;
NeilBrown52c03292006-06-26 00:27:43 -07004298}
4299
4300/* FIXME go_faster isn't used */
4301static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
4302{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004303 raid5_conf_t *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07004304 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11004305 sector_t max_sector = mddev->dev_sectors;
NeilBrown57dab0b2010-10-19 10:03:39 +11004306 sector_t sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07004307 int still_degraded = 0;
4308 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004309
NeilBrown72626682005-09-09 16:23:54 -07004310 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004311 /* just being told to finish up .. nothing much to do */
NeilBrowncea9c222009-03-31 15:15:05 +11004312
NeilBrown29269552006-03-27 01:18:10 -08004313 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4314 end_reshape(conf);
4315 return 0;
4316 }
NeilBrown72626682005-09-09 16:23:54 -07004317
4318 if (mddev->curr_resync < max_sector) /* aborted */
4319 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4320 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07004321 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07004322 conf->fullsync = 0;
4323 bitmap_close_sync(mddev->bitmap);
4324
Linus Torvalds1da177e2005-04-16 15:20:36 -07004325 return 0;
4326 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08004327
NeilBrown64bd6602009-08-03 10:59:58 +10004328 /* Allow raid5_quiesce to complete */
4329 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4330
NeilBrown52c03292006-06-26 00:27:43 -07004331 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4332 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08004333
NeilBrownc6207272008-02-06 01:39:52 -08004334 /* No need to check resync_max as we never do more than one
4335 * stripe, and as resync_max will always be on a chunk boundary,
4336 * if the check in md_do_sync didn't fire, there is no chance
4337 * of overstepping resync_max here
4338 */
4339
NeilBrown16a53ec2006-06-26 00:27:38 -07004340 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07004341 * to resync, then assert that we are finished, because there is
4342 * nothing we can do.
4343 */
NeilBrown3285edf2006-06-26 00:27:55 -07004344 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07004345 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004346 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07004347 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004348 return rv;
4349 }
NeilBrown72626682005-09-09 16:23:54 -07004350 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrown3855ad92005-11-08 21:39:38 -08004351 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown72626682005-09-09 16:23:54 -07004352 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4353 /* we can skip this block, and probably more */
4354 sync_blocks /= STRIPE_SECTORS;
4355 *skipped = 1;
4356 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4357 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004358
NeilBrownb47490c2008-02-06 01:39:50 -08004359
4360 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4361
NeilBrowna8c906c2009-06-09 14:39:59 +10004362 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004363 if (sh == NULL) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004364 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004365 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07004366 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08004368 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004369 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004370 /* Need to check if array will still be degraded after recovery/resync
4371 * We don't need to check the 'failed' flag as when that gets set,
4372 * recovery aborts.
4373 */
NeilBrownf001a702009-06-09 14:30:31 +10004374 for (i = 0; i < conf->raid_disks; i++)
NeilBrown16a53ec2006-06-26 00:27:38 -07004375 if (conf->disks[i].rdev == NULL)
4376 still_degraded = 1;
4377
4378 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4379
4380 spin_lock(&sh->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004381 set_bit(STRIPE_SYNCING, &sh->state);
4382 clear_bit(STRIPE_INSYNC, &sh->state);
4383 spin_unlock(&sh->lock);
4384
NeilBrown14425772009-10-16 15:55:25 +11004385 handle_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004386 release_stripe(sh);
4387
4388 return STRIPE_SECTORS;
4389}
4390
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004391static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4392{
4393 /* We may not be able to submit a whole bio at once as there
4394 * may not be enough stripe_heads available.
4395 * We cannot pre-allocate enough stripe_heads as we may need
4396 * more than exist in the cache (if we allow ever large chunks).
4397 * So we do one stripe head at a time and record in
4398 * ->bi_hw_segments how many have been done.
4399 *
4400 * We *know* that this entire raid_bio is in one chunk, so
4401 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4402 */
4403 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11004404 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004405 sector_t sector, logical_sector, last_sector;
4406 int scnt = 0;
4407 int remaining;
4408 int handled = 0;
4409
4410 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11004411 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11004412 0, &dd_idx, NULL);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004413 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4414
4415 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08004416 logical_sector += STRIPE_SECTORS,
4417 sector += STRIPE_SECTORS,
4418 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004419
Jens Axboe960e7392008-08-15 10:41:18 +02004420 if (scnt < raid5_bi_hw_segments(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004421 /* already done this stripe */
4422 continue;
4423
NeilBrowna8c906c2009-06-09 14:39:59 +10004424 sh = get_active_stripe(conf, sector, 0, 1, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004425
4426 if (!sh) {
4427 /* failed to get a stripe - must wait */
Jens Axboe960e7392008-08-15 10:41:18 +02004428 raid5_set_bi_hw_segments(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004429 conf->retry_read_aligned = raid_bio;
4430 return handled;
4431 }
4432
4433 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
Neil Brown387bb172007-02-08 14:20:29 -08004434 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4435 release_stripe(sh);
Jens Axboe960e7392008-08-15 10:41:18 +02004436 raid5_set_bi_hw_segments(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08004437 conf->retry_read_aligned = raid_bio;
4438 return handled;
4439 }
4440
Dan Williams36d1c642009-07-14 11:48:22 -07004441 handle_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004442 release_stripe(sh);
4443 handled++;
4444 }
4445 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004446 remaining = raid5_dec_bi_phys_segments(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004447 spin_unlock_irq(&conf->device_lock);
Neil Brown0e13fe232008-06-28 08:31:20 +10004448 if (remaining == 0)
4449 bio_endio(raid_bio, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004450 if (atomic_dec_and_test(&conf->active_aligned_reads))
4451 wake_up(&conf->wait_for_stripe);
4452 return handled;
4453}
4454
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004455
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456/*
4457 * This is our raid5 kernel thread.
4458 *
4459 * We scan the hash table for stripes which can be handled now.
4460 * During the scan, completed stripes are saved for us by the interrupt
4461 * handler, so that they will not have to wait for our next wakeup.
4462 */
NeilBrown6ed30032008-02-06 01:40:00 -08004463static void raid5d(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004464{
4465 struct stripe_head *sh;
NeilBrown070ec552009-06-16 16:54:21 +10004466 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004467 int handled;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004468 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004469
Dan Williams45b42332007-07-09 11:56:43 -07004470 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471
4472 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004473
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004474 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004475 handled = 0;
4476 spin_lock_irq(&conf->device_lock);
4477 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004478 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004479
NeilBrown7c13edc2011-04-18 18:25:43 +10004480 if (atomic_read(&mddev->plug_cnt) == 0 &&
4481 !list_empty(&conf->bitmap_list)) {
4482 /* Now is a good time to flush some bitmap updates */
4483 conf->seq_flush++;
NeilBrown700e4322005-11-28 13:44:10 -08004484 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07004485 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08004486 spin_lock_irq(&conf->device_lock);
NeilBrown7c13edc2011-04-18 18:25:43 +10004487 conf->seq_write = conf->seq_flush;
NeilBrown72626682005-09-09 16:23:54 -07004488 activate_bit_delay(conf);
4489 }
NeilBrown7c13edc2011-04-18 18:25:43 +10004490 if (atomic_read(&mddev->plug_cnt) == 0)
4491 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07004492
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004493 while ((bio = remove_bio_from_retry(conf))) {
4494 int ok;
4495 spin_unlock_irq(&conf->device_lock);
4496 ok = retry_aligned_read(conf, bio);
4497 spin_lock_irq(&conf->device_lock);
4498 if (!ok)
4499 break;
4500 handled++;
4501 }
4502
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004503 sh = __get_priority_stripe(conf);
4504
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004505 if (!sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004506 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004507 spin_unlock_irq(&conf->device_lock);
4508
4509 handled++;
Dan Williams417b8d42009-10-16 16:25:22 +11004510 handle_stripe(sh);
4511 release_stripe(sh);
4512 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004513
4514 spin_lock_irq(&conf->device_lock);
4515 }
Dan Williams45b42332007-07-09 11:56:43 -07004516 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004517
4518 spin_unlock_irq(&conf->device_lock);
4519
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004520 async_tx_issue_pending_all();
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004521 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004522
Dan Williams45b42332007-07-09 11:56:43 -07004523 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004524}
4525
NeilBrown3f294f42005-11-08 21:39:25 -08004526static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004527raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004528{
NeilBrown070ec552009-06-16 16:54:21 +10004529 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004530 if (conf)
4531 return sprintf(page, "%d\n", conf->max_nr_stripes);
4532 else
4533 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004534}
4535
NeilBrownc41d4ac2010-06-01 19:37:24 +10004536int
4537raid5_set_cache_size(mddev_t *mddev, int size)
4538{
4539 raid5_conf_t *conf = mddev->private;
4540 int err;
4541
4542 if (size <= 16 || size > 32768)
4543 return -EINVAL;
4544 while (size < conf->max_nr_stripes) {
4545 if (drop_one_stripe(conf))
4546 conf->max_nr_stripes--;
4547 else
4548 break;
4549 }
4550 err = md_allow_write(mddev);
4551 if (err)
4552 return err;
4553 while (size > conf->max_nr_stripes) {
4554 if (grow_one_stripe(conf))
4555 conf->max_nr_stripes++;
4556 else break;
4557 }
4558 return 0;
4559}
4560EXPORT_SYMBOL(raid5_set_cache_size);
4561
NeilBrown3f294f42005-11-08 21:39:25 -08004562static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004563raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08004564{
NeilBrown070ec552009-06-16 16:54:21 +10004565 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004566 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07004567 int err;
4568
NeilBrown3f294f42005-11-08 21:39:25 -08004569 if (len >= PAGE_SIZE)
4570 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08004571 if (!conf)
4572 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08004573
Dan Williams4ef197d82008-04-28 02:15:54 -07004574 if (strict_strtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08004575 return -EINVAL;
NeilBrownc41d4ac2010-06-01 19:37:24 +10004576 err = raid5_set_cache_size(mddev, new);
Dan Williamsb5470dc2008-06-27 21:44:04 -07004577 if (err)
4578 return err;
NeilBrown3f294f42005-11-08 21:39:25 -08004579 return len;
4580}
NeilBrown007583c2005-11-08 21:39:30 -08004581
NeilBrown96de1e62005-11-08 21:39:39 -08004582static struct md_sysfs_entry
4583raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4584 raid5_show_stripe_cache_size,
4585 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08004586
4587static ssize_t
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004588raid5_show_preread_threshold(mddev_t *mddev, char *page)
4589{
NeilBrown070ec552009-06-16 16:54:21 +10004590 raid5_conf_t *conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004591 if (conf)
4592 return sprintf(page, "%d\n", conf->bypass_threshold);
4593 else
4594 return 0;
4595}
4596
4597static ssize_t
4598raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4599{
NeilBrown070ec552009-06-16 16:54:21 +10004600 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004601 unsigned long new;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004602 if (len >= PAGE_SIZE)
4603 return -EINVAL;
4604 if (!conf)
4605 return -ENODEV;
4606
Dan Williams4ef197d82008-04-28 02:15:54 -07004607 if (strict_strtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004608 return -EINVAL;
Dan Williams4ef197d82008-04-28 02:15:54 -07004609 if (new > conf->max_nr_stripes)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004610 return -EINVAL;
4611 conf->bypass_threshold = new;
4612 return len;
4613}
4614
4615static struct md_sysfs_entry
4616raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4617 S_IRUGO | S_IWUSR,
4618 raid5_show_preread_threshold,
4619 raid5_store_preread_threshold);
4620
4621static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08004622stripe_cache_active_show(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004623{
NeilBrown070ec552009-06-16 16:54:21 +10004624 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004625 if (conf)
4626 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4627 else
4628 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004629}
4630
NeilBrown96de1e62005-11-08 21:39:39 -08004631static struct md_sysfs_entry
4632raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08004633
NeilBrown007583c2005-11-08 21:39:30 -08004634static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08004635 &raid5_stripecache_size.attr,
4636 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004637 &raid5_preread_bypass_threshold.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08004638 NULL,
4639};
NeilBrown007583c2005-11-08 21:39:30 -08004640static struct attribute_group raid5_attrs_group = {
4641 .name = NULL,
4642 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08004643};
4644
Dan Williams80c3a6c2009-03-17 18:10:40 -07004645static sector_t
4646raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4647{
NeilBrown070ec552009-06-16 16:54:21 +10004648 raid5_conf_t *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07004649
4650 if (!sectors)
4651 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004652 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11004653 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11004654 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004655
Andre Noll9d8f0362009-06-18 08:45:01 +10004656 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Andre Noll664e7c42009-06-18 08:45:27 +10004657 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004658 return sectors * (raid_disks - conf->max_degraded);
4659}
4660
Dan Williams36d1c642009-07-14 11:48:22 -07004661static void raid5_free_percpu(raid5_conf_t *conf)
4662{
4663 struct raid5_percpu *percpu;
4664 unsigned long cpu;
4665
4666 if (!conf->percpu)
4667 return;
4668
4669 get_online_cpus();
4670 for_each_possible_cpu(cpu) {
4671 percpu = per_cpu_ptr(conf->percpu, cpu);
4672 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004673 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004674 }
4675#ifdef CONFIG_HOTPLUG_CPU
4676 unregister_cpu_notifier(&conf->cpu_notify);
4677#endif
4678 put_online_cpus();
4679
4680 free_percpu(conf->percpu);
4681}
4682
Dan Williams95fc17a2009-07-31 12:39:15 +10004683static void free_conf(raid5_conf_t *conf)
4684{
4685 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07004686 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10004687 kfree(conf->disks);
4688 kfree(conf->stripe_hashtbl);
4689 kfree(conf);
4690}
4691
Dan Williams36d1c642009-07-14 11:48:22 -07004692#ifdef CONFIG_HOTPLUG_CPU
4693static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4694 void *hcpu)
4695{
4696 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4697 long cpu = (long)hcpu;
4698 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4699
4700 switch (action) {
4701 case CPU_UP_PREPARE:
4702 case CPU_UP_PREPARE_FROZEN:
Dan Williamsd6f38f32009-07-14 11:50:52 -07004703 if (conf->level == 6 && !percpu->spare_page)
Dan Williams36d1c642009-07-14 11:48:22 -07004704 percpu->spare_page = alloc_page(GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004705 if (!percpu->scribble)
4706 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4707
4708 if (!percpu->scribble ||
4709 (conf->level == 6 && !percpu->spare_page)) {
4710 safe_put_page(percpu->spare_page);
4711 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004712 pr_err("%s: failed memory allocation for cpu%ld\n",
4713 __func__, cpu);
Akinobu Mita55af6bb2010-05-26 14:43:35 -07004714 return notifier_from_errno(-ENOMEM);
Dan Williams36d1c642009-07-14 11:48:22 -07004715 }
4716 break;
4717 case CPU_DEAD:
4718 case CPU_DEAD_FROZEN:
4719 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004720 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004721 percpu->spare_page = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004722 percpu->scribble = NULL;
Dan Williams36d1c642009-07-14 11:48:22 -07004723 break;
4724 default:
4725 break;
4726 }
4727 return NOTIFY_OK;
4728}
4729#endif
4730
4731static int raid5_alloc_percpu(raid5_conf_t *conf)
4732{
4733 unsigned long cpu;
4734 struct page *spare_page;
Tejun Heoa29d8b82010-02-02 14:39:15 +09004735 struct raid5_percpu __percpu *allcpus;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004736 void *scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004737 int err;
4738
Dan Williams36d1c642009-07-14 11:48:22 -07004739 allcpus = alloc_percpu(struct raid5_percpu);
4740 if (!allcpus)
4741 return -ENOMEM;
4742 conf->percpu = allcpus;
4743
4744 get_online_cpus();
4745 err = 0;
4746 for_each_present_cpu(cpu) {
Dan Williamsd6f38f32009-07-14 11:50:52 -07004747 if (conf->level == 6) {
4748 spare_page = alloc_page(GFP_KERNEL);
4749 if (!spare_page) {
4750 err = -ENOMEM;
4751 break;
4752 }
4753 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4754 }
NeilBrown5e5e3e72009-10-16 16:35:30 +11004755 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004756 if (!scribble) {
Dan Williams36d1c642009-07-14 11:48:22 -07004757 err = -ENOMEM;
4758 break;
4759 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07004760 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004761 }
4762#ifdef CONFIG_HOTPLUG_CPU
4763 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4764 conf->cpu_notify.priority = 0;
4765 if (err == 0)
4766 err = register_cpu_notifier(&conf->cpu_notify);
4767#endif
4768 put_online_cpus();
4769
4770 return err;
4771}
4772
NeilBrown91adb562009-03-31 14:39:39 +11004773static raid5_conf_t *setup_conf(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004774{
4775 raid5_conf_t *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004776 int raid_disk, memory, max_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004777 mdk_rdev_t *rdev;
4778 struct disk_info *disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004779
NeilBrown91adb562009-03-31 14:39:39 +11004780 if (mddev->new_level != 5
4781 && mddev->new_level != 4
4782 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10004783 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004784 mdname(mddev), mddev->new_level);
4785 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004786 }
NeilBrown91adb562009-03-31 14:39:39 +11004787 if ((mddev->new_level == 5
4788 && !algorithm_valid_raid5(mddev->new_layout)) ||
4789 (mddev->new_level == 6
4790 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004791 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11004792 mdname(mddev), mddev->new_layout);
4793 return ERR_PTR(-EIO);
4794 }
4795 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10004796 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004797 mdname(mddev), mddev->raid_disks);
4798 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11004799 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004800
Andre Noll664e7c42009-06-18 08:45:27 +10004801 if (!mddev->new_chunk_sectors ||
4802 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4803 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10004804 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4805 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11004806 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11004807 }
4808
NeilBrown91adb562009-03-31 14:39:39 +11004809 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4810 if (conf == NULL)
4811 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11004812 spin_lock_init(&conf->device_lock);
4813 init_waitqueue_head(&conf->wait_for_stripe);
4814 init_waitqueue_head(&conf->wait_for_overlap);
4815 INIT_LIST_HEAD(&conf->handle_list);
4816 INIT_LIST_HEAD(&conf->hold_list);
4817 INIT_LIST_HEAD(&conf->delayed_list);
4818 INIT_LIST_HEAD(&conf->bitmap_list);
4819 INIT_LIST_HEAD(&conf->inactive_list);
4820 atomic_set(&conf->active_stripes, 0);
4821 atomic_set(&conf->preread_active_stripes, 0);
4822 atomic_set(&conf->active_aligned_reads, 0);
4823 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrown91adb562009-03-31 14:39:39 +11004824
4825 conf->raid_disks = mddev->raid_disks;
4826 if (mddev->reshape_position == MaxSector)
4827 conf->previous_raid_disks = mddev->raid_disks;
4828 else
4829 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004830 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4831 conf->scribble_len = scribble_len(max_disks);
NeilBrown91adb562009-03-31 14:39:39 +11004832
NeilBrown5e5e3e72009-10-16 16:35:30 +11004833 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11004834 GFP_KERNEL);
4835 if (!conf->disks)
4836 goto abort;
4837
4838 conf->mddev = mddev;
4839
4840 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
4841 goto abort;
4842
Dan Williams36d1c642009-07-14 11:48:22 -07004843 conf->level = mddev->new_level;
4844 if (raid5_alloc_percpu(conf) != 0)
4845 goto abort;
4846
NeilBrown0c55e022010-05-03 14:09:02 +10004847 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11004848
4849 list_for_each_entry(rdev, &mddev->disks, same_set) {
4850 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004851 if (raid_disk >= max_disks
NeilBrown91adb562009-03-31 14:39:39 +11004852 || raid_disk < 0)
4853 continue;
4854 disk = conf->disks + raid_disk;
4855
4856 disk->rdev = rdev;
4857
4858 if (test_bit(In_sync, &rdev->flags)) {
4859 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10004860 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4861 " disk %d\n",
4862 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
NeilBrown91adb562009-03-31 14:39:39 +11004863 } else
4864 /* Cannot rely on bitmap to complete recovery */
4865 conf->fullsync = 1;
4866 }
4867
Andre Noll09c9e5f2009-06-18 08:45:55 +10004868 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown91adb562009-03-31 14:39:39 +11004869 conf->level = mddev->new_level;
4870 if (conf->level == 6)
4871 conf->max_degraded = 2;
4872 else
4873 conf->max_degraded = 1;
4874 conf->algorithm = mddev->new_layout;
4875 conf->max_nr_stripes = NR_STRIPES;
NeilBrownfef9c612009-03-31 15:16:46 +11004876 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11004877 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10004878 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11004879 conf->prev_algo = mddev->layout;
4880 }
NeilBrown91adb562009-03-31 14:39:39 +11004881
4882 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11004883 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
NeilBrown91adb562009-03-31 14:39:39 +11004884 if (grow_stripes(conf, conf->max_nr_stripes)) {
4885 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004886 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4887 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004888 goto abort;
4889 } else
NeilBrown0c55e022010-05-03 14:09:02 +10004890 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4891 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004892
NeilBrown0da3c612009-09-23 18:09:45 +10004893 conf->thread = md_register_thread(raid5d, mddev, NULL);
NeilBrown91adb562009-03-31 14:39:39 +11004894 if (!conf->thread) {
4895 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004896 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11004897 mdname(mddev));
4898 goto abort;
4899 }
4900
4901 return conf;
4902
4903 abort:
4904 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10004905 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11004906 return ERR_PTR(-EIO);
4907 } else
4908 return ERR_PTR(-ENOMEM);
4909}
4910
NeilBrownc148ffd2009-11-13 17:47:00 +11004911
4912static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4913{
4914 switch (algo) {
4915 case ALGORITHM_PARITY_0:
4916 if (raid_disk < max_degraded)
4917 return 1;
4918 break;
4919 case ALGORITHM_PARITY_N:
4920 if (raid_disk >= raid_disks - max_degraded)
4921 return 1;
4922 break;
4923 case ALGORITHM_PARITY_0_6:
4924 if (raid_disk == 0 ||
4925 raid_disk == raid_disks - 1)
4926 return 1;
4927 break;
4928 case ALGORITHM_LEFT_ASYMMETRIC_6:
4929 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4930 case ALGORITHM_LEFT_SYMMETRIC_6:
4931 case ALGORITHM_RIGHT_SYMMETRIC_6:
4932 if (raid_disk == raid_disks - 1)
4933 return 1;
4934 }
4935 return 0;
4936}
4937
NeilBrown91adb562009-03-31 14:39:39 +11004938static int run(mddev_t *mddev)
4939{
4940 raid5_conf_t *conf;
NeilBrown9f7c2222010-07-26 12:04:13 +10004941 int working_disks = 0;
NeilBrownc148ffd2009-11-13 17:47:00 +11004942 int dirty_parity_disks = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004943 mdk_rdev_t *rdev;
NeilBrownc148ffd2009-11-13 17:47:00 +11004944 sector_t reshape_offset = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004945
Andre Noll8c6ac862009-06-18 08:48:06 +10004946 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10004947 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10004948 " -- starting background reconstruction\n",
4949 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004950 if (mddev->reshape_position != MaxSector) {
4951 /* Check that we can continue the reshape.
4952 * Currently only disks can change, it must
4953 * increase, and we must be past the point where
4954 * a stripe over-writes itself
4955 */
4956 sector_t here_new, here_old;
4957 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11004958 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrownf6705572006-03-27 01:18:11 -08004959
NeilBrown88ce4932009-03-31 15:24:23 +11004960 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10004961 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08004962 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08004963 mdname(mddev));
4964 return -EINVAL;
4965 }
NeilBrownf6705572006-03-27 01:18:11 -08004966 old_disks = mddev->raid_disks - mddev->delta_disks;
4967 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08004968 * further up in new geometry must map after here in old
4969 * geometry.
NeilBrownf6705572006-03-27 01:18:11 -08004970 */
4971 here_new = mddev->reshape_position;
Andre Noll664e7c42009-06-18 08:45:27 +10004972 if (sector_div(here_new, mddev->new_chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004973 (mddev->raid_disks - max_degraded))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004974 printk(KERN_ERR "md/raid:%s: reshape_position not "
4975 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004976 return -EINVAL;
4977 }
NeilBrownc148ffd2009-11-13 17:47:00 +11004978 reshape_offset = here_new * mddev->new_chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08004979 /* here_new is the stripe we will write to */
4980 here_old = mddev->reshape_position;
Andre Noll9d8f0362009-06-18 08:45:01 +10004981 sector_div(here_old, mddev->chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004982 (old_disks-max_degraded));
4983 /* here_old is the first stripe that we might need to read
4984 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10004985 if (mddev->delta_disks == 0) {
4986 /* We cannot be sure it is safe to start an in-place
4987 * reshape. It is only safe if user-space if monitoring
4988 * and taking constant backups.
4989 * mdadm always starts a situation like this in
4990 * readonly mode so it can take control before
4991 * allowing any writes. So just check for that.
4992 */
4993 if ((here_new * mddev->new_chunk_sectors !=
4994 here_old * mddev->chunk_sectors) ||
4995 mddev->ro == 0) {
NeilBrown0c55e022010-05-03 14:09:02 +10004996 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
4997 " in read-only mode - aborting\n",
4998 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10004999 return -EINVAL;
5000 }
5001 } else if (mddev->delta_disks < 0
5002 ? (here_new * mddev->new_chunk_sectors <=
5003 here_old * mddev->chunk_sectors)
5004 : (here_new * mddev->new_chunk_sectors >=
5005 here_old * mddev->chunk_sectors)) {
NeilBrownf6705572006-03-27 01:18:11 -08005006 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10005007 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5008 "auto-recovery - aborting.\n",
5009 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005010 return -EINVAL;
5011 }
NeilBrown0c55e022010-05-03 14:09:02 +10005012 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5013 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005014 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08005015 } else {
NeilBrown91adb562009-03-31 14:39:39 +11005016 BUG_ON(mddev->level != mddev->new_level);
5017 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10005018 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11005019 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08005020 }
5021
NeilBrown245f46c2009-03-31 14:39:39 +11005022 if (mddev->private == NULL)
5023 conf = setup_conf(mddev);
5024 else
5025 conf = mddev->private;
5026
NeilBrown91adb562009-03-31 14:39:39 +11005027 if (IS_ERR(conf))
5028 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08005029
NeilBrown91adb562009-03-31 14:39:39 +11005030 mddev->thread = conf->thread;
5031 conf->thread = NULL;
5032 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005033
Linus Torvalds1da177e2005-04-16 15:20:36 -07005034 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005035 * 0 for a fully functional array, 1 or 2 for a degraded array.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005036 */
NeilBrownc148ffd2009-11-13 17:47:00 +11005037 list_for_each_entry(rdev, &mddev->disks, same_set) {
5038 if (rdev->raid_disk < 0)
5039 continue;
NeilBrown2f115882010-06-17 17:41:03 +10005040 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown91adb562009-03-31 14:39:39 +11005041 working_disks++;
NeilBrown2f115882010-06-17 17:41:03 +10005042 continue;
5043 }
NeilBrownc148ffd2009-11-13 17:47:00 +11005044 /* This disc is not fully in-sync. However if it
5045 * just stored parity (beyond the recovery_offset),
5046 * when we don't need to be concerned about the
5047 * array being dirty.
5048 * When reshape goes 'backwards', we never have
5049 * partially completed devices, so we only need
5050 * to worry about reshape going forwards.
5051 */
5052 /* Hack because v0.91 doesn't store recovery_offset properly. */
5053 if (mddev->major_version == 0 &&
5054 mddev->minor_version > 90)
5055 rdev->recovery_offset = reshape_offset;
5056
NeilBrownc148ffd2009-11-13 17:47:00 +11005057 if (rdev->recovery_offset < reshape_offset) {
5058 /* We need to check old and new layout */
5059 if (!only_parity(rdev->raid_disk,
5060 conf->algorithm,
5061 conf->raid_disks,
5062 conf->max_degraded))
5063 continue;
5064 }
5065 if (!only_parity(rdev->raid_disk,
5066 conf->prev_algo,
5067 conf->previous_raid_disks,
5068 conf->max_degraded))
5069 continue;
5070 dirty_parity_disks++;
5071 }
NeilBrown91adb562009-03-31 14:39:39 +11005072
NeilBrown5e5e3e72009-10-16 16:35:30 +11005073 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
5074 - working_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005075
NeilBrown674806d2010-06-16 17:17:53 +10005076 if (has_failed(conf)) {
NeilBrown0c55e022010-05-03 14:09:02 +10005077 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07005078 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07005079 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005080 goto abort;
5081 }
5082
NeilBrown91adb562009-03-31 14:39:39 +11005083 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10005084 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11005085 mddev->resync_max_sectors = mddev->dev_sectors;
5086
NeilBrownc148ffd2009-11-13 17:47:00 +11005087 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07005088 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005089 if (mddev->ok_start_degraded)
5090 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10005091 "md/raid:%s: starting dirty degraded array"
5092 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005093 mdname(mddev));
5094 else {
5095 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005096 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005097 mdname(mddev));
5098 goto abort;
5099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005100 }
5101
Linus Torvalds1da177e2005-04-16 15:20:36 -07005102 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10005103 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5104 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11005105 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5106 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005107 else
NeilBrown0c55e022010-05-03 14:09:02 +10005108 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5109 " out of %d devices, algorithm %d\n",
5110 mdname(mddev), conf->level,
5111 mddev->raid_disks - mddev->degraded,
5112 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005113
5114 print_raid5_conf(conf);
5115
NeilBrownfef9c612009-03-31 15:16:46 +11005116 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11005117 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08005118 atomic_set(&conf->reshape_stripes, 0);
5119 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5120 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5121 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5122 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5123 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005124 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08005125 }
5126
Linus Torvalds1da177e2005-04-16 15:20:36 -07005127
5128 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10005129 if (mddev->to_remove == &raid5_attrs_group)
5130 mddev->to_remove = NULL;
NeilBrown00bcb4a2010-06-01 19:37:23 +10005131 else if (mddev->kobj.sd &&
5132 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08005133 printk(KERN_WARNING
NeilBrown4a5add42010-06-01 19:37:28 +10005134 "raid5: failed to create sysfs attributes for %s\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08005135 mdname(mddev));
NeilBrown4a5add42010-06-01 19:37:28 +10005136 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5137
5138 if (mddev->queue) {
NeilBrown9f7c2222010-07-26 12:04:13 +10005139 int chunk_size;
NeilBrown4a5add42010-06-01 19:37:28 +10005140 /* read-ahead size must cover two whole stripes, which
5141 * is 2 * (datadisks) * chunksize where 'n' is the
5142 * number of raid devices
5143 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005144 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5145 int stripe = data_disks *
5146 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
5147 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5148 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
NeilBrown4a5add42010-06-01 19:37:28 +10005149
5150 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
NeilBrown11d8a6e2010-07-26 11:57:07 +10005151
5152 mddev->queue->backing_dev_info.congested_data = mddev;
5153 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
NeilBrown9f7c2222010-07-26 12:04:13 +10005154
5155 chunk_size = mddev->chunk_sectors << 9;
5156 blk_queue_io_min(mddev->queue, chunk_size);
5157 blk_queue_io_opt(mddev->queue, chunk_size *
5158 (conf->raid_disks - conf->max_degraded));
5159
5160 list_for_each_entry(rdev, &mddev->disks, same_set)
5161 disk_stack_limits(mddev->gendisk, rdev->bdev,
5162 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005163 }
5164
Linus Torvalds1da177e2005-04-16 15:20:36 -07005165 return 0;
5166abort:
NeilBrowne0cf8f02009-03-31 14:39:39 +11005167 md_unregister_thread(mddev->thread);
NeilBrown91adb562009-03-31 14:39:39 +11005168 mddev->thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005169 if (conf) {
5170 print_raid5_conf(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005171 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005172 }
5173 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10005174 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005175 return -EIO;
5176}
5177
NeilBrown3f294f42005-11-08 21:39:25 -08005178static int stop(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005179{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005180 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005181
5182 md_unregister_thread(mddev->thread);
5183 mddev->thread = NULL;
NeilBrown11d8a6e2010-07-26 11:57:07 +10005184 if (mddev->queue)
5185 mddev->queue->backing_dev_info.congested_fn = NULL;
Dan Williams95fc17a2009-07-31 12:39:15 +10005186 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10005187 mddev->private = NULL;
5188 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005189 return 0;
5190}
5191
Dan Williams45b42332007-07-09 11:56:43 -07005192#ifdef DEBUG
NeilBrownd710e132008-10-13 11:55:12 +11005193static void print_sh(struct seq_file *seq, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005194{
5195 int i;
5196
NeilBrown16a53ec2006-06-26 00:27:38 -07005197 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
5198 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
5199 seq_printf(seq, "sh %llu, count %d.\n",
5200 (unsigned long long)sh->sector, atomic_read(&sh->count));
5201 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005202 for (i = 0; i < sh->disks; i++) {
NeilBrown16a53ec2006-06-26 00:27:38 -07005203 seq_printf(seq, "(cache%d: %p %ld) ",
5204 i, sh->dev[i].page, sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005205 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005206 seq_printf(seq, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005207}
5208
NeilBrownd710e132008-10-13 11:55:12 +11005209static void printall(struct seq_file *seq, raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005210{
5211 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -08005212 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005213 int i;
5214
5215 spin_lock_irq(&conf->device_lock);
5216 for (i = 0; i < NR_HASH; i++) {
NeilBrownfccddba2006-01-06 00:20:33 -08005217 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005218 if (sh->raid_conf != conf)
5219 continue;
NeilBrown16a53ec2006-06-26 00:27:38 -07005220 print_sh(seq, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005221 }
5222 }
5223 spin_unlock_irq(&conf->device_lock);
5224}
5225#endif
5226
NeilBrownd710e132008-10-13 11:55:12 +11005227static void status(struct seq_file *seq, mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005228{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005229 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005230 int i;
5231
Andre Noll9d8f0362009-06-18 08:45:01 +10005232 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5233 mddev->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07005234 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005235 for (i = 0; i < conf->raid_disks; i++)
5236 seq_printf (seq, "%s",
5237 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08005238 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005239 seq_printf (seq, "]");
Dan Williams45b42332007-07-09 11:56:43 -07005240#ifdef DEBUG
NeilBrown16a53ec2006-06-26 00:27:38 -07005241 seq_printf (seq, "\n");
5242 printall(seq, conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005243#endif
5244}
5245
5246static void print_raid5_conf (raid5_conf_t *conf)
5247{
5248 int i;
5249 struct disk_info *tmp;
5250
NeilBrown0c55e022010-05-03 14:09:02 +10005251 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005252 if (!conf) {
5253 printk("(conf==NULL)\n");
5254 return;
5255 }
NeilBrown0c55e022010-05-03 14:09:02 +10005256 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5257 conf->raid_disks,
5258 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005259
5260 for (i = 0; i < conf->raid_disks; i++) {
5261 char b[BDEVNAME_SIZE];
5262 tmp = conf->disks + i;
5263 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10005264 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5265 i, !test_bit(Faulty, &tmp->rdev->flags),
5266 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005267 }
5268}
5269
5270static int raid5_spare_active(mddev_t *mddev)
5271{
5272 int i;
5273 raid5_conf_t *conf = mddev->private;
5274 struct disk_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10005275 int count = 0;
5276 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005277
5278 for (i = 0; i < conf->raid_disks; i++) {
5279 tmp = conf->disks + i;
5280 if (tmp->rdev
NeilBrown70fffd02010-06-16 17:01:25 +10005281 && tmp->rdev->recovery_offset == MaxSector
NeilBrownb2d444d2005-11-08 21:39:31 -08005282 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07005283 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10005284 count++;
Jonathan Brassow43c73ca2011-01-14 09:14:33 +11005285 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005286 }
5287 }
NeilBrown6b965622010-08-18 11:56:59 +10005288 spin_lock_irqsave(&conf->device_lock, flags);
5289 mddev->degraded -= count;
5290 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005291 print_raid5_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10005292 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005293}
5294
5295static int raid5_remove_disk(mddev_t *mddev, int number)
5296{
5297 raid5_conf_t *conf = mddev->private;
5298 int err = 0;
5299 mdk_rdev_t *rdev;
5300 struct disk_info *p = conf->disks + number;
5301
5302 print_raid5_conf(conf);
5303 rdev = p->rdev;
5304 if (rdev) {
NeilBrownec32a2b2009-03-31 15:17:38 +11005305 if (number >= conf->raid_disks &&
5306 conf->reshape_progress == MaxSector)
5307 clear_bit(In_sync, &rdev->flags);
5308
NeilBrownb2d444d2005-11-08 21:39:31 -08005309 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07005310 atomic_read(&rdev->nr_pending)) {
5311 err = -EBUSY;
5312 goto abort;
5313 }
NeilBrowndfc70642008-05-23 13:04:39 -07005314 /* Only remove non-faulty devices if recovery
5315 * isn't possible.
5316 */
5317 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown674806d2010-06-16 17:17:53 +10005318 !has_failed(conf) &&
NeilBrownec32a2b2009-03-31 15:17:38 +11005319 number < conf->raid_disks) {
NeilBrowndfc70642008-05-23 13:04:39 -07005320 err = -EBUSY;
5321 goto abort;
5322 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005323 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07005324 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005325 if (atomic_read(&rdev->nr_pending)) {
5326 /* lost the race, try later */
5327 err = -EBUSY;
5328 p->rdev = rdev;
5329 }
5330 }
5331abort:
5332
5333 print_raid5_conf(conf);
5334 return err;
5335}
5336
5337static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
5338{
5339 raid5_conf_t *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10005340 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005341 int disk;
5342 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10005343 int first = 0;
5344 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005345
NeilBrown674806d2010-06-16 17:17:53 +10005346 if (has_failed(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005347 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10005348 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005349
Neil Brown6c2fce22008-06-28 08:31:31 +10005350 if (rdev->raid_disk >= 0)
5351 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005352
5353 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005354 * find the disk ... but prefer rdev->saved_raid_disk
5355 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005356 */
NeilBrown16a53ec2006-06-26 00:27:38 -07005357 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10005358 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07005359 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5360 disk = rdev->saved_raid_disk;
5361 else
Neil Brown6c2fce22008-06-28 08:31:31 +10005362 disk = first;
5363 for ( ; disk <= last ; disk++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005364 if ((p=conf->disks + disk)->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08005365 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005366 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10005367 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07005368 if (rdev->saved_raid_disk != disk)
5369 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08005370 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005371 break;
5372 }
5373 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10005374 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005375}
5376
5377static int raid5_resize(mddev_t *mddev, sector_t sectors)
5378{
5379 /* no resync is happening, and there is enough space
5380 * on all devices, so we can resize.
5381 * We need to make sure resync covers any new space.
5382 * If the array is shrinking we should possibly wait until
5383 * any io in the removed space completes, but it hardly seems
5384 * worth it.
5385 */
Andre Noll9d8f0362009-06-18 08:45:01 +10005386 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Dan Williams1f403622009-03-31 14:59:03 +11005387 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5388 mddev->raid_disks));
Dan Williamsb522adc2009-03-31 15:00:31 +11005389 if (mddev->array_sectors >
5390 raid5_size(mddev, sectors, mddev->raid_disks))
5391 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10005392 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005393 revalidate_disk(mddev->gendisk);
Andre Noll58c0fed2009-03-31 14:33:13 +11005394 if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) {
5395 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005396 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5397 }
Andre Noll58c0fed2009-03-31 14:33:13 +11005398 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07005399 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005400 return 0;
5401}
5402
NeilBrown01ee22b2009-06-18 08:47:20 +10005403static int check_stripe_cache(mddev_t *mddev)
5404{
5405 /* Can only proceed if there are plenty of stripe_heads.
5406 * We need a minimum of one full stripe,, and for sensible progress
5407 * it is best to have about 4 times that.
5408 * If we require 4 times, then the default 256 4K stripe_heads will
5409 * allow for chunk sizes up to 256K, which is probably OK.
5410 * If the chunk size is greater, user-space should request more
5411 * stripe_heads first.
5412 */
5413 raid5_conf_t *conf = mddev->private;
5414 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5415 > conf->max_nr_stripes ||
5416 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5417 > conf->max_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10005418 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5419 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10005420 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5421 / STRIPE_SIZE)*4);
5422 return 0;
5423 }
5424 return 1;
5425}
5426
NeilBrown50ac1682009-06-18 08:47:55 +10005427static int check_reshape(mddev_t *mddev)
NeilBrown29269552006-03-27 01:18:10 -08005428{
NeilBrown070ec552009-06-16 16:54:21 +10005429 raid5_conf_t *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08005430
NeilBrown88ce4932009-03-31 15:24:23 +11005431 if (mddev->delta_disks == 0 &&
5432 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10005433 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10005434 return 0; /* nothing to do */
NeilBrowndba034e2008-08-05 15:54:13 +10005435 if (mddev->bitmap)
5436 /* Cannot grow a bitmap yet */
5437 return -EBUSY;
NeilBrown674806d2010-06-16 17:17:53 +10005438 if (has_failed(conf))
NeilBrownec32a2b2009-03-31 15:17:38 +11005439 return -EINVAL;
5440 if (mddev->delta_disks < 0) {
5441 /* We might be able to shrink, but the devices must
5442 * be made bigger first.
5443 * For raid6, 4 is the minimum size.
5444 * Otherwise 2 is the minimum
5445 */
5446 int min = 2;
5447 if (mddev->level == 6)
5448 min = 4;
5449 if (mddev->raid_disks + mddev->delta_disks < min)
5450 return -EINVAL;
5451 }
NeilBrown29269552006-03-27 01:18:10 -08005452
NeilBrown01ee22b2009-06-18 08:47:20 +10005453 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08005454 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08005455
NeilBrownec32a2b2009-03-31 15:17:38 +11005456 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
NeilBrown63c70c42006-03-27 01:18:13 -08005457}
5458
5459static int raid5_start_reshape(mddev_t *mddev)
5460{
NeilBrown070ec552009-06-16 16:54:21 +10005461 raid5_conf_t *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08005462 mdk_rdev_t *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08005463 int spares = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07005464 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08005465
NeilBrownf4168852007-02-28 20:11:53 -08005466 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08005467 return -EBUSY;
5468
NeilBrown01ee22b2009-06-18 08:47:20 +10005469 if (!check_stripe_cache(mddev))
5470 return -ENOSPC;
5471
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005472 list_for_each_entry(rdev, &mddev->disks, same_set)
NeilBrown469518a2011-01-31 11:57:43 +11005473 if (!test_bit(In_sync, &rdev->flags)
5474 && !test_bit(Faulty, &rdev->flags))
NeilBrown29269552006-03-27 01:18:10 -08005475 spares++;
NeilBrown63c70c42006-03-27 01:18:13 -08005476
NeilBrownf4168852007-02-28 20:11:53 -08005477 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08005478 /* Not enough devices even to make a degraded array
5479 * of that size
5480 */
5481 return -EINVAL;
5482
NeilBrownec32a2b2009-03-31 15:17:38 +11005483 /* Refuse to reduce size of the array. Any reductions in
5484 * array size must be through explicit setting of array_size
5485 * attribute.
5486 */
5487 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5488 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10005489 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11005490 "before number of disks\n", mdname(mddev));
5491 return -EINVAL;
5492 }
5493
NeilBrownf6705572006-03-27 01:18:11 -08005494 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08005495 spin_lock_irq(&conf->device_lock);
5496 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08005497 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005498 conf->prev_chunk_sectors = conf->chunk_sectors;
5499 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11005500 conf->prev_algo = conf->algorithm;
5501 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11005502 if (mddev->delta_disks < 0)
5503 conf->reshape_progress = raid5_size(mddev, 0, 0);
5504 else
5505 conf->reshape_progress = 0;
5506 conf->reshape_safe = conf->reshape_progress;
NeilBrown86b42c72009-03-31 15:19:03 +11005507 conf->generation++;
NeilBrown29269552006-03-27 01:18:10 -08005508 spin_unlock_irq(&conf->device_lock);
5509
5510 /* Add some new drives, as many as will fit.
5511 * We know there are enough to make the newly sized array work.
NeilBrown3424bf62010-06-17 17:48:26 +10005512 * Don't add devices if we are reducing the number of
5513 * devices in the array. This is because it is not possible
5514 * to correctly record the "partially reconstructed" state of
5515 * such devices during the reshape and confusion could result.
NeilBrown29269552006-03-27 01:18:10 -08005516 */
NeilBrown87a8dec2011-01-31 11:57:43 +11005517 if (mddev->delta_disks >= 0) {
5518 int added_devices = 0;
5519 list_for_each_entry(rdev, &mddev->disks, same_set)
5520 if (rdev->raid_disk < 0 &&
5521 !test_bit(Faulty, &rdev->flags)) {
5522 if (raid5_add_disk(mddev, rdev) == 0) {
5523 char nm[20];
5524 if (rdev->raid_disk
5525 >= conf->previous_raid_disks) {
5526 set_bit(In_sync, &rdev->flags);
5527 added_devices++;
5528 } else
5529 rdev->recovery_offset = 0;
5530 sprintf(nm, "rd%d", rdev->raid_disk);
5531 if (sysfs_create_link(&mddev->kobj,
5532 &rdev->kobj, nm))
5533 /* Failure here is OK */;
NeilBrown50da0842011-01-31 11:57:43 +11005534 }
NeilBrown87a8dec2011-01-31 11:57:43 +11005535 } else if (rdev->raid_disk >= conf->previous_raid_disks
5536 && !test_bit(Faulty, &rdev->flags)) {
5537 /* This is a spare that was manually added */
5538 set_bit(In_sync, &rdev->flags);
5539 added_devices++;
5540 }
NeilBrown29269552006-03-27 01:18:10 -08005541
NeilBrown87a8dec2011-01-31 11:57:43 +11005542 /* When a reshape changes the number of devices,
5543 * ->degraded is measured against the larger of the
5544 * pre and post number of devices.
5545 */
NeilBrownec32a2b2009-03-31 15:17:38 +11005546 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown9eb07c22010-02-09 12:31:47 +11005547 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
NeilBrownec32a2b2009-03-31 15:17:38 +11005548 - added_devices;
5549 spin_unlock_irqrestore(&conf->device_lock, flags);
5550 }
NeilBrown63c70c42006-03-27 01:18:13 -08005551 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10005552 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07005553 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08005554
NeilBrown29269552006-03-27 01:18:10 -08005555 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5556 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5557 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5558 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5559 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005560 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08005561 if (!mddev->sync_thread) {
5562 mddev->recovery = 0;
5563 spin_lock_irq(&conf->device_lock);
5564 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005565 conf->reshape_progress = MaxSector;
NeilBrown29269552006-03-27 01:18:10 -08005566 spin_unlock_irq(&conf->device_lock);
5567 return -EAGAIN;
5568 }
NeilBrownc8f517c2009-03-31 15:28:40 +11005569 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08005570 md_wakeup_thread(mddev->sync_thread);
5571 md_new_event(mddev);
5572 return 0;
5573}
NeilBrown29269552006-03-27 01:18:10 -08005574
NeilBrownec32a2b2009-03-31 15:17:38 +11005575/* This is called from the reshape thread and should make any
5576 * changes needed in 'conf'
5577 */
NeilBrown29269552006-03-27 01:18:10 -08005578static void end_reshape(raid5_conf_t *conf)
5579{
NeilBrown29269552006-03-27 01:18:10 -08005580
NeilBrownf6705572006-03-27 01:18:11 -08005581 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
Dan Williams80c3a6c2009-03-17 18:10:40 -07005582
NeilBrownf6705572006-03-27 01:18:11 -08005583 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11005584 conf->previous_raid_disks = conf->raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005585 conf->reshape_progress = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08005586 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11005587 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07005588
5589 /* read-ahead size must cover two whole stripes, which is
5590 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5591 */
NeilBrown4a5add42010-06-01 19:37:28 +10005592 if (conf->mddev->queue) {
NeilBrowncea9c222009-03-31 15:15:05 +11005593 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005594 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11005595 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07005596 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5597 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5598 }
NeilBrown29269552006-03-27 01:18:10 -08005599 }
NeilBrown29269552006-03-27 01:18:10 -08005600}
5601
NeilBrownec32a2b2009-03-31 15:17:38 +11005602/* This is called from the raid5d thread with mddev_lock held.
5603 * It makes config changes to the device.
5604 */
NeilBrowncea9c222009-03-31 15:15:05 +11005605static void raid5_finish_reshape(mddev_t *mddev)
5606{
NeilBrown070ec552009-06-16 16:54:21 +10005607 raid5_conf_t *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11005608
5609 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5610
NeilBrownec32a2b2009-03-31 15:17:38 +11005611 if (mddev->delta_disks > 0) {
5612 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5613 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005614 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11005615 } else {
5616 int d;
NeilBrownec32a2b2009-03-31 15:17:38 +11005617 mddev->degraded = conf->raid_disks;
5618 for (d = 0; d < conf->raid_disks ; d++)
5619 if (conf->disks[d].rdev &&
5620 test_bit(In_sync,
5621 &conf->disks[d].rdev->flags))
5622 mddev->degraded--;
5623 for (d = conf->raid_disks ;
5624 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10005625 d++) {
5626 mdk_rdev_t *rdev = conf->disks[d].rdev;
5627 if (rdev && raid5_remove_disk(mddev, d) == 0) {
5628 char nm[20];
5629 sprintf(nm, "rd%d", rdev->raid_disk);
5630 sysfs_remove_link(&mddev->kobj, nm);
5631 rdev->raid_disk = -1;
5632 }
5633 }
NeilBrowncea9c222009-03-31 15:15:05 +11005634 }
NeilBrown88ce4932009-03-31 15:24:23 +11005635 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005636 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11005637 mddev->reshape_position = MaxSector;
5638 mddev->delta_disks = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11005639 }
5640}
5641
NeilBrown72626682005-09-09 16:23:54 -07005642static void raid5_quiesce(mddev_t *mddev, int state)
5643{
NeilBrown070ec552009-06-16 16:54:21 +10005644 raid5_conf_t *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07005645
5646 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08005647 case 2: /* resume for a suspend */
5648 wake_up(&conf->wait_for_overlap);
5649 break;
5650
NeilBrown72626682005-09-09 16:23:54 -07005651 case 1: /* stop all writes */
5652 spin_lock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005653 /* '2' tells resync/reshape to pause so that all
5654 * active stripes can drain
5655 */
5656 conf->quiesce = 2;
NeilBrown72626682005-09-09 16:23:54 -07005657 wait_event_lock_irq(conf->wait_for_stripe,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005658 atomic_read(&conf->active_stripes) == 0 &&
5659 atomic_read(&conf->active_aligned_reads) == 0,
NeilBrown72626682005-09-09 16:23:54 -07005660 conf->device_lock, /* nothing */);
NeilBrown64bd6602009-08-03 10:59:58 +10005661 conf->quiesce = 1;
NeilBrown72626682005-09-09 16:23:54 -07005662 spin_unlock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005663 /* allow reshape to continue */
5664 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005665 break;
5666
5667 case 0: /* re-enable writes */
5668 spin_lock_irq(&conf->device_lock);
5669 conf->quiesce = 0;
5670 wake_up(&conf->wait_for_stripe);
NeilBrowne464eaf2006-03-27 01:18:14 -08005671 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005672 spin_unlock_irq(&conf->device_lock);
5673 break;
5674 }
NeilBrown72626682005-09-09 16:23:54 -07005675}
NeilBrownb15c2e52006-01-06 00:20:16 -08005676
NeilBrownd562b0c2009-03-31 14:39:39 +11005677
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005678static void *raid45_takeover_raid0(mddev_t *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11005679{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005680 struct raid0_private_data *raid0_priv = mddev->private;
Randy Dunlapd76c8422011-04-21 09:07:26 -07005681 sector_t sectors;
Trela Maciej54071b32010-03-08 16:02:42 +11005682
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005683 /* for raid0 takeover only one zone is supported */
5684 if (raid0_priv->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10005685 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5686 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005687 return ERR_PTR(-EINVAL);
5688 }
5689
NeilBrown3b71bd92011-04-20 15:38:18 +10005690 sectors = raid0_priv->strip_zone[0].zone_end;
5691 sector_div(sectors, raid0_priv->strip_zone[0].nb_dev);
5692 mddev->dev_sectors = sectors;
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005693 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11005694 mddev->new_layout = ALGORITHM_PARITY_N;
5695 mddev->new_chunk_sectors = mddev->chunk_sectors;
5696 mddev->raid_disks += 1;
5697 mddev->delta_disks = 1;
5698 /* make sure it will be not marked as dirty */
5699 mddev->recovery_cp = MaxSector;
5700
5701 return setup_conf(mddev);
5702}
5703
5704
NeilBrownd562b0c2009-03-31 14:39:39 +11005705static void *raid5_takeover_raid1(mddev_t *mddev)
5706{
5707 int chunksect;
5708
5709 if (mddev->raid_disks != 2 ||
5710 mddev->degraded > 1)
5711 return ERR_PTR(-EINVAL);
5712
5713 /* Should check if there are write-behind devices? */
5714
5715 chunksect = 64*2; /* 64K by default */
5716
5717 /* The array must be an exact multiple of chunksize */
5718 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5719 chunksect >>= 1;
5720
5721 if ((chunksect<<9) < STRIPE_SIZE)
5722 /* array size does not allow a suitable chunk size */
5723 return ERR_PTR(-EINVAL);
5724
5725 mddev->new_level = 5;
5726 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10005727 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11005728
5729 return setup_conf(mddev);
5730}
5731
NeilBrownfc9739c2009-03-31 14:57:20 +11005732static void *raid5_takeover_raid6(mddev_t *mddev)
5733{
5734 int new_layout;
5735
5736 switch (mddev->layout) {
5737 case ALGORITHM_LEFT_ASYMMETRIC_6:
5738 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5739 break;
5740 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5741 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5742 break;
5743 case ALGORITHM_LEFT_SYMMETRIC_6:
5744 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5745 break;
5746 case ALGORITHM_RIGHT_SYMMETRIC_6:
5747 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5748 break;
5749 case ALGORITHM_PARITY_0_6:
5750 new_layout = ALGORITHM_PARITY_0;
5751 break;
5752 case ALGORITHM_PARITY_N:
5753 new_layout = ALGORITHM_PARITY_N;
5754 break;
5755 default:
5756 return ERR_PTR(-EINVAL);
5757 }
5758 mddev->new_level = 5;
5759 mddev->new_layout = new_layout;
5760 mddev->delta_disks = -1;
5761 mddev->raid_disks -= 1;
5762 return setup_conf(mddev);
5763}
5764
NeilBrownd562b0c2009-03-31 14:39:39 +11005765
NeilBrown50ac1682009-06-18 08:47:55 +10005766static int raid5_check_reshape(mddev_t *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11005767{
NeilBrown88ce4932009-03-31 15:24:23 +11005768 /* For a 2-drive array, the layout and chunk size can be changed
5769 * immediately as not restriping is needed.
5770 * For larger arrays we record the new value - after validation
5771 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11005772 */
NeilBrown070ec552009-06-16 16:54:21 +10005773 raid5_conf_t *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10005774 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11005775
NeilBrown597a7112009-06-18 08:47:42 +10005776 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11005777 return -EINVAL;
5778 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005779 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11005780 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005781 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11005782 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005783 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11005784 /* not factor of array size */
5785 return -EINVAL;
5786 }
5787
5788 /* They look valid */
5789
NeilBrown88ce4932009-03-31 15:24:23 +11005790 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10005791 /* can make the change immediately */
5792 if (mddev->new_layout >= 0) {
5793 conf->algorithm = mddev->new_layout;
5794 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11005795 }
5796 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10005797 conf->chunk_sectors = new_chunk ;
5798 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11005799 }
5800 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5801 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11005802 }
NeilBrown50ac1682009-06-18 08:47:55 +10005803 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11005804}
5805
NeilBrown50ac1682009-06-18 08:47:55 +10005806static int raid6_check_reshape(mddev_t *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11005807{
NeilBrown597a7112009-06-18 08:47:42 +10005808 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10005809
NeilBrown597a7112009-06-18 08:47:42 +10005810 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11005811 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005812 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005813 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11005814 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005815 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11005816 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005817 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11005818 /* not factor of array size */
5819 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005820 }
NeilBrown88ce4932009-03-31 15:24:23 +11005821
5822 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10005823 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11005824}
5825
NeilBrownd562b0c2009-03-31 14:39:39 +11005826static void *raid5_takeover(mddev_t *mddev)
5827{
5828 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005829 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005830 * raid1 - if there are two drives. We need to know the chunk size
5831 * raid4 - trivial - just use a raid4 layout.
5832 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005833 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005834 if (mddev->level == 0)
5835 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11005836 if (mddev->level == 1)
5837 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11005838 if (mddev->level == 4) {
5839 mddev->new_layout = ALGORITHM_PARITY_N;
5840 mddev->new_level = 5;
5841 return setup_conf(mddev);
5842 }
NeilBrownfc9739c2009-03-31 14:57:20 +11005843 if (mddev->level == 6)
5844 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11005845
5846 return ERR_PTR(-EINVAL);
5847}
5848
NeilBrowna78d38a2010-03-22 16:53:49 +11005849static void *raid4_takeover(mddev_t *mddev)
5850{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005851 /* raid4 can take over:
5852 * raid0 - if there is only one strip zone
5853 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11005854 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005855 if (mddev->level == 0)
5856 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11005857 if (mddev->level == 5 &&
5858 mddev->layout == ALGORITHM_PARITY_N) {
5859 mddev->new_layout = 0;
5860 mddev->new_level = 4;
5861 return setup_conf(mddev);
5862 }
5863 return ERR_PTR(-EINVAL);
5864}
NeilBrownd562b0c2009-03-31 14:39:39 +11005865
NeilBrown245f46c2009-03-31 14:39:39 +11005866static struct mdk_personality raid5_personality;
5867
5868static void *raid6_takeover(mddev_t *mddev)
5869{
5870 /* Currently can only take over a raid5. We map the
5871 * personality to an equivalent raid6 personality
5872 * with the Q block at the end.
5873 */
5874 int new_layout;
5875
5876 if (mddev->pers != &raid5_personality)
5877 return ERR_PTR(-EINVAL);
5878 if (mddev->degraded > 1)
5879 return ERR_PTR(-EINVAL);
5880 if (mddev->raid_disks > 253)
5881 return ERR_PTR(-EINVAL);
5882 if (mddev->raid_disks < 3)
5883 return ERR_PTR(-EINVAL);
5884
5885 switch (mddev->layout) {
5886 case ALGORITHM_LEFT_ASYMMETRIC:
5887 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5888 break;
5889 case ALGORITHM_RIGHT_ASYMMETRIC:
5890 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5891 break;
5892 case ALGORITHM_LEFT_SYMMETRIC:
5893 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5894 break;
5895 case ALGORITHM_RIGHT_SYMMETRIC:
5896 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5897 break;
5898 case ALGORITHM_PARITY_0:
5899 new_layout = ALGORITHM_PARITY_0_6;
5900 break;
5901 case ALGORITHM_PARITY_N:
5902 new_layout = ALGORITHM_PARITY_N;
5903 break;
5904 default:
5905 return ERR_PTR(-EINVAL);
5906 }
5907 mddev->new_level = 6;
5908 mddev->new_layout = new_layout;
5909 mddev->delta_disks = 1;
5910 mddev->raid_disks += 1;
5911 return setup_conf(mddev);
5912}
5913
5914
NeilBrown16a53ec2006-06-26 00:27:38 -07005915static struct mdk_personality raid6_personality =
5916{
5917 .name = "raid6",
5918 .level = 6,
5919 .owner = THIS_MODULE,
5920 .make_request = make_request,
5921 .run = run,
5922 .stop = stop,
5923 .status = status,
5924 .error_handler = error,
5925 .hot_add_disk = raid5_add_disk,
5926 .hot_remove_disk= raid5_remove_disk,
5927 .spare_active = raid5_spare_active,
5928 .sync_request = sync_request,
5929 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005930 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10005931 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08005932 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005933 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07005934 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11005935 .takeover = raid6_takeover,
NeilBrown16a53ec2006-06-26 00:27:38 -07005936};
NeilBrown2604b702006-01-06 00:20:36 -08005937static struct mdk_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005938{
5939 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08005940 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005941 .owner = THIS_MODULE,
5942 .make_request = make_request,
5943 .run = run,
5944 .stop = stop,
5945 .status = status,
5946 .error_handler = error,
5947 .hot_add_disk = raid5_add_disk,
5948 .hot_remove_disk= raid5_remove_disk,
5949 .spare_active = raid5_spare_active,
5950 .sync_request = sync_request,
5951 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005952 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08005953 .check_reshape = raid5_check_reshape,
5954 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005955 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07005956 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11005957 .takeover = raid5_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005958};
5959
NeilBrown2604b702006-01-06 00:20:36 -08005960static struct mdk_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005961{
NeilBrown2604b702006-01-06 00:20:36 -08005962 .name = "raid4",
5963 .level = 4,
5964 .owner = THIS_MODULE,
5965 .make_request = make_request,
5966 .run = run,
5967 .stop = stop,
5968 .status = status,
5969 .error_handler = error,
5970 .hot_add_disk = raid5_add_disk,
5971 .hot_remove_disk= raid5_remove_disk,
5972 .spare_active = raid5_spare_active,
5973 .sync_request = sync_request,
5974 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005975 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08005976 .check_reshape = raid5_check_reshape,
5977 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005978 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08005979 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11005980 .takeover = raid4_takeover,
NeilBrown2604b702006-01-06 00:20:36 -08005981};
5982
5983static int __init raid5_init(void)
5984{
NeilBrown16a53ec2006-06-26 00:27:38 -07005985 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005986 register_md_personality(&raid5_personality);
5987 register_md_personality(&raid4_personality);
5988 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005989}
5990
NeilBrown2604b702006-01-06 00:20:36 -08005991static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005992{
NeilBrown16a53ec2006-06-26 00:27:38 -07005993 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005994 unregister_md_personality(&raid5_personality);
5995 unregister_md_personality(&raid4_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005996}
5997
5998module_init(raid5_init);
5999module_exit(raid5_exit);
6000MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11006001MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006002MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08006003MODULE_ALIAS("md-raid5");
6004MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08006005MODULE_ALIAS("md-level-5");
6006MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07006007MODULE_ALIAS("md-personality-8"); /* RAID6 */
6008MODULE_ALIAS("md-raid6");
6009MODULE_ALIAS("md-level-6");
6010
6011/* This used to be two separate modules, they were: */
6012MODULE_ALIAS("raid5");
6013MODULE_ALIAS("raid6");