blob: 74a575bbd25c6dd1d9a4aa310102834b360fa2fd [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{
Namhyung Kim9b2dc8b2011-06-13 14:48:22 +0900132 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;
Namhyung Kimb0629622011-06-14 14:20:19 +1000517 if (rw & WRITE)
Dan Williams91c00922007-01-02 13:52:30 -0700518 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;
Namhyung Kimb0629622011-06-14 14:20:19 +1000551 if ((rw & WRITE) &&
Dan Williams91c00922007-01-02 13:52:30 -0700552 test_bit(R5_ReWrite, &sh->dev[i].flags))
553 atomic_add(STRIPE_SECTORS,
554 &rdev->corrected_errors);
555 generic_make_request(bi);
556 } else {
Namhyung Kimb0629622011-06-14 14:20:19 +1000557 if (rw & WRITE)
Dan Williams91c00922007-01-02 13:52:30 -0700558 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) {
Namhyung Kimfcde9072011-06-14 14:23:57 +1000588 int len = bvl->bv_len;
Dan Williams91c00922007-01-02 13:52:30 -0700589 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) {
Namhyung Kimfcde9072011-06-14 14:23:57 +1000604 b_offset += bvl->bv_offset;
605 bio_page = bvl->bv_page;
Dan Williams91c00922007-01-02 13:52:30 -0700606 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
NeilBrowncbe47ec2011-07-26 11:20:35 +10001023 spin_lock_irq(&sh->raid_conf->device_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001024 chosen = dev->towrite;
1025 dev->towrite = NULL;
1026 BUG_ON(dev->written);
1027 wbi = dev->written = chosen;
NeilBrowncbe47ec2011-07-26 11:20:35 +10001028 spin_unlock_irq(&sh->raid_conf->device_lock);
Dan Williams91c00922007-01-02 13:52:30 -07001029
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;
Namhyung Kim6ce32842011-07-18 17:38:50 +10001318 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
NeilBrown3f294f42005-11-08 21:39:25 -08001319 if (!sh)
1320 return 0;
Namhyung Kim6ce32842011-07-18 17:38:50 +10001321
NeilBrown3f294f42005-11-08 21:39:25 -08001322 sh->raid_conf = conf;
Dan Williams417b8d42009-10-16 16:25:22 +11001323 #ifdef CONFIG_MULTICORE_RAID456
1324 init_waitqueue_head(&sh->ops.wait_for_ops);
1325 #endif
NeilBrown3f294f42005-11-08 21:39:25 -08001326
NeilBrowne4e11e32010-06-16 16:45:16 +10001327 if (grow_buffers(sh)) {
1328 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001329 kmem_cache_free(conf->slab_cache, sh);
1330 return 0;
1331 }
1332 /* we just created an active stripe so... */
1333 atomic_set(&sh->count, 1);
1334 atomic_inc(&conf->active_stripes);
1335 INIT_LIST_HEAD(&sh->lru);
1336 release_stripe(sh);
1337 return 1;
1338}
1339
1340static int grow_stripes(raid5_conf_t *conf, int num)
1341{
Christoph Lametere18b8902006-12-06 20:33:20 -08001342 struct kmem_cache *sc;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001343 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
NeilBrownf4be6b42010-06-01 19:37:25 +10001345 if (conf->mddev->gendisk)
1346 sprintf(conf->cache_name[0],
1347 "raid%d-%s", conf->level, mdname(conf->mddev));
1348 else
1349 sprintf(conf->cache_name[0],
1350 "raid%d-%p", conf->level, conf->mddev);
1351 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1352
NeilBrownad01c9e2006-03-27 01:18:07 -08001353 conf->active_name = 0;
1354 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001356 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 if (!sc)
1358 return 1;
1359 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001360 conf->pool_size = devs;
NeilBrown16a53ec2006-06-26 00:27:38 -07001361 while (num--)
NeilBrown3f294f42005-11-08 21:39:25 -08001362 if (!grow_one_stripe(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 return 0;
1365}
NeilBrown29269552006-03-27 01:18:10 -08001366
Dan Williamsd6f38f32009-07-14 11:50:52 -07001367/**
1368 * scribble_len - return the required size of the scribble region
1369 * @num - total number of disks in the array
1370 *
1371 * The size must be enough to contain:
1372 * 1/ a struct page pointer for each device in the array +2
1373 * 2/ room to convert each entry in (1) to its corresponding dma
1374 * (dma_map_page()) or page (page_address()) address.
1375 *
1376 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1377 * calculate over all devices (not just the data blocks), using zeros in place
1378 * of the P and Q blocks.
1379 */
1380static size_t scribble_len(int num)
1381{
1382 size_t len;
1383
1384 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1385
1386 return len;
1387}
1388
NeilBrownad01c9e2006-03-27 01:18:07 -08001389static int resize_stripes(raid5_conf_t *conf, int newsize)
1390{
1391 /* Make all the stripes able to hold 'newsize' devices.
1392 * New slots in each stripe get 'page' set to a new page.
1393 *
1394 * This happens in stages:
1395 * 1/ create a new kmem_cache and allocate the required number of
1396 * stripe_heads.
1397 * 2/ gather all the old stripe_heads and tranfer the pages across
1398 * to the new stripe_heads. This will have the side effect of
1399 * freezing the array as once all stripe_heads have been collected,
1400 * no IO will be possible. Old stripe heads are freed once their
1401 * pages have been transferred over, and the old kmem_cache is
1402 * freed when all stripes are done.
1403 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1404 * we simple return a failre status - no need to clean anything up.
1405 * 4/ allocate new pages for the new slots in the new stripe_heads.
1406 * If this fails, we don't bother trying the shrink the
1407 * stripe_heads down again, we just leave them as they are.
1408 * As each stripe_head is processed the new one is released into
1409 * active service.
1410 *
1411 * Once step2 is started, we cannot afford to wait for a write,
1412 * so we use GFP_NOIO allocations.
1413 */
1414 struct stripe_head *osh, *nsh;
1415 LIST_HEAD(newstripes);
1416 struct disk_info *ndisks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001417 unsigned long cpu;
Dan Williamsb5470dc2008-06-27 21:44:04 -07001418 int err;
Christoph Lametere18b8902006-12-06 20:33:20 -08001419 struct kmem_cache *sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001420 int i;
1421
1422 if (newsize <= conf->pool_size)
1423 return 0; /* never bother to shrink */
1424
Dan Williamsb5470dc2008-06-27 21:44:04 -07001425 err = md_allow_write(conf->mddev);
1426 if (err)
1427 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08001428
NeilBrownad01c9e2006-03-27 01:18:07 -08001429 /* Step 1 */
1430 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1431 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001432 0, 0, NULL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001433 if (!sc)
1434 return -ENOMEM;
1435
1436 for (i = conf->max_nr_stripes; i; i--) {
Namhyung Kim6ce32842011-07-18 17:38:50 +10001437 nsh = kmem_cache_zalloc(sc, GFP_KERNEL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001438 if (!nsh)
1439 break;
1440
NeilBrownad01c9e2006-03-27 01:18:07 -08001441 nsh->raid_conf = conf;
Dan Williams417b8d42009-10-16 16:25:22 +11001442 #ifdef CONFIG_MULTICORE_RAID456
1443 init_waitqueue_head(&nsh->ops.wait_for_ops);
1444 #endif
NeilBrownad01c9e2006-03-27 01:18:07 -08001445
1446 list_add(&nsh->lru, &newstripes);
1447 }
1448 if (i) {
1449 /* didn't get enough, give up */
1450 while (!list_empty(&newstripes)) {
1451 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1452 list_del(&nsh->lru);
1453 kmem_cache_free(sc, nsh);
1454 }
1455 kmem_cache_destroy(sc);
1456 return -ENOMEM;
1457 }
1458 /* Step 2 - Must use GFP_NOIO now.
1459 * OK, we have enough stripes, start collecting inactive
1460 * stripes and copying them over
1461 */
1462 list_for_each_entry(nsh, &newstripes, lru) {
1463 spin_lock_irq(&conf->device_lock);
1464 wait_event_lock_irq(conf->wait_for_stripe,
1465 !list_empty(&conf->inactive_list),
1466 conf->device_lock,
NeilBrown482c0832011-04-18 18:25:42 +10001467 );
NeilBrownad01c9e2006-03-27 01:18:07 -08001468 osh = get_free_stripe(conf);
1469 spin_unlock_irq(&conf->device_lock);
1470 atomic_set(&nsh->count, 1);
1471 for(i=0; i<conf->pool_size; i++)
1472 nsh->dev[i].page = osh->dev[i].page;
1473 for( ; i<newsize; i++)
1474 nsh->dev[i].page = NULL;
1475 kmem_cache_free(conf->slab_cache, osh);
1476 }
1477 kmem_cache_destroy(conf->slab_cache);
1478
1479 /* Step 3.
1480 * At this point, we are holding all the stripes so the array
1481 * is completely stalled, so now is a good time to resize
Dan Williamsd6f38f32009-07-14 11:50:52 -07001482 * conf->disks and the scribble region
NeilBrownad01c9e2006-03-27 01:18:07 -08001483 */
1484 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1485 if (ndisks) {
1486 for (i=0; i<conf->raid_disks; i++)
1487 ndisks[i] = conf->disks[i];
1488 kfree(conf->disks);
1489 conf->disks = ndisks;
1490 } else
1491 err = -ENOMEM;
1492
Dan Williamsd6f38f32009-07-14 11:50:52 -07001493 get_online_cpus();
1494 conf->scribble_len = scribble_len(newsize);
1495 for_each_present_cpu(cpu) {
1496 struct raid5_percpu *percpu;
1497 void *scribble;
1498
1499 percpu = per_cpu_ptr(conf->percpu, cpu);
1500 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1501
1502 if (scribble) {
1503 kfree(percpu->scribble);
1504 percpu->scribble = scribble;
1505 } else {
1506 err = -ENOMEM;
1507 break;
1508 }
1509 }
1510 put_online_cpus();
1511
NeilBrownad01c9e2006-03-27 01:18:07 -08001512 /* Step 4, return new stripes to service */
1513 while(!list_empty(&newstripes)) {
1514 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1515 list_del_init(&nsh->lru);
Dan Williamsd6f38f32009-07-14 11:50:52 -07001516
NeilBrownad01c9e2006-03-27 01:18:07 -08001517 for (i=conf->raid_disks; i < newsize; i++)
1518 if (nsh->dev[i].page == NULL) {
1519 struct page *p = alloc_page(GFP_NOIO);
1520 nsh->dev[i].page = p;
1521 if (!p)
1522 err = -ENOMEM;
1523 }
1524 release_stripe(nsh);
1525 }
1526 /* critical section pass, GFP_NOIO no longer needed */
1527
1528 conf->slab_cache = sc;
1529 conf->active_name = 1-conf->active_name;
1530 conf->pool_size = newsize;
1531 return err;
1532}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
NeilBrown3f294f42005-11-08 21:39:25 -08001534static int drop_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535{
1536 struct stripe_head *sh;
1537
NeilBrown3f294f42005-11-08 21:39:25 -08001538 spin_lock_irq(&conf->device_lock);
1539 sh = get_free_stripe(conf);
1540 spin_unlock_irq(&conf->device_lock);
1541 if (!sh)
1542 return 0;
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02001543 BUG_ON(atomic_read(&sh->count));
NeilBrowne4e11e32010-06-16 16:45:16 +10001544 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001545 kmem_cache_free(conf->slab_cache, sh);
1546 atomic_dec(&conf->active_stripes);
1547 return 1;
1548}
1549
1550static void shrink_stripes(raid5_conf_t *conf)
1551{
1552 while (drop_one_stripe(conf))
1553 ;
1554
NeilBrown29fc7e32006-02-03 03:03:41 -08001555 if (conf->slab_cache)
1556 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 conf->slab_cache = NULL;
1558}
1559
NeilBrown6712ecf2007-09-27 12:47:43 +02001560static void raid5_end_read_request(struct bio * bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
NeilBrown99c0fb52009-03-31 14:39:38 +11001562 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001564 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownd6950432006-07-10 04:44:20 -07001566 char b[BDEVNAME_SIZE];
1567 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
1570 for (i=0 ; i<disks; i++)
1571 if (bi == &sh->dev[i].req)
1572 break;
1573
Dan Williams45b42332007-07-09 11:56:43 -07001574 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1575 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 uptodate);
1577 if (i == disks) {
1578 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001579 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581
1582 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -08001584 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrownd6950432006-07-10 04:44:20 -07001585 rdev = conf->disks[i].rdev;
NeilBrown0c55e022010-05-03 14:09:02 +10001586 printk_rl(KERN_INFO "md/raid:%s: read error corrected"
Bernd Schubert6be9d492008-05-23 13:04:34 -07001587 " (%lu sectors at %llu on %s)\n",
1588 mdname(conf->mddev), STRIPE_SECTORS,
1589 (unsigned long long)(sh->sector
1590 + rdev->data_offset),
1591 bdevname(rdev->bdev, b));
NeilBrown4e5314b2005-11-08 21:39:22 -08001592 clear_bit(R5_ReadError, &sh->dev[i].flags);
1593 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1594 }
NeilBrownba22dcb2005-11-08 21:39:31 -08001595 if (atomic_read(&conf->disks[i].rdev->read_errors))
1596 atomic_set(&conf->disks[i].rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 } else {
NeilBrownd6950432006-07-10 04:44:20 -07001598 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
NeilBrownba22dcb2005-11-08 21:39:31 -08001599 int retry = 0;
NeilBrownd6950432006-07-10 04:44:20 -07001600 rdev = conf->disks[i].rdev;
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001603 atomic_inc(&rdev->read_errors);
Gabriele A. Trombetti7b0bb532010-04-28 11:51:17 +10001604 if (conf->mddev->degraded >= conf->max_degraded)
Bernd Schubert6be9d492008-05-23 13:04:34 -07001605 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001606 "md/raid:%s: read error not correctable "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001607 "(sector %llu on %s).\n",
1608 mdname(conf->mddev),
1609 (unsigned long long)(sh->sector
1610 + rdev->data_offset),
1611 bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001612 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
NeilBrown4e5314b2005-11-08 21:39:22 -08001613 /* Oh, no!!! */
Bernd Schubert6be9d492008-05-23 13:04:34 -07001614 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001615 "md/raid:%s: read error NOT corrected!! "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001616 "(sector %llu on %s).\n",
1617 mdname(conf->mddev),
1618 (unsigned long long)(sh->sector
1619 + rdev->data_offset),
1620 bdn);
NeilBrownd6950432006-07-10 04:44:20 -07001621 else if (atomic_read(&rdev->read_errors)
NeilBrownba22dcb2005-11-08 21:39:31 -08001622 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -08001623 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001624 "md/raid:%s: Too many read errors, failing device %s.\n",
NeilBrownd6950432006-07-10 04:44:20 -07001625 mdname(conf->mddev), bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001626 else
1627 retry = 1;
1628 if (retry)
1629 set_bit(R5_ReadError, &sh->dev[i].flags);
1630 else {
NeilBrown4e5314b2005-11-08 21:39:22 -08001631 clear_bit(R5_ReadError, &sh->dev[i].flags);
1632 clear_bit(R5_ReWrite, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001633 md_error(conf->mddev, rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -08001634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
1636 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1638 set_bit(STRIPE_HANDLE, &sh->state);
1639 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640}
1641
NeilBrownd710e132008-10-13 11:55:12 +11001642static void raid5_end_write_request(struct bio *bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
NeilBrown99c0fb52009-03-31 14:39:38 +11001644 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001646 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 for (i=0 ; i<disks; i++)
1650 if (bi == &sh->dev[i].req)
1651 break;
1652
Dan Williams45b42332007-07-09 11:56:43 -07001653 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1655 uptodate);
1656 if (i == disks) {
1657 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001658 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 }
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (!uptodate)
1662 md_error(conf->mddev, conf->disks[i].rdev);
1663
1664 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1665
1666 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1667 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrownc04be0a2006-10-03 01:15:53 -07001668 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669}
1670
1671
NeilBrown784052e2009-03-31 15:19:07 +11001672static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
NeilBrown784052e2009-03-31 15:19:07 +11001674static void raid5_build_block(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
1676 struct r5dev *dev = &sh->dev[i];
1677
1678 bio_init(&dev->req);
1679 dev->req.bi_io_vec = &dev->vec;
1680 dev->req.bi_vcnt++;
1681 dev->req.bi_max_vecs++;
1682 dev->vec.bv_page = dev->page;
1683 dev->vec.bv_len = STRIPE_SIZE;
1684 dev->vec.bv_offset = 0;
1685
1686 dev->req.bi_sector = sh->sector;
1687 dev->req.bi_private = sh;
1688
1689 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +11001690 dev->sector = compute_blocknr(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691}
1692
1693static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1694{
1695 char b[BDEVNAME_SIZE];
H Hartley Sweeten7b928132010-03-08 16:02:40 +11001696 raid5_conf_t *conf = mddev->private;
NeilBrown0c55e022010-05-03 14:09:02 +10001697 pr_debug("raid456: error called\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
NeilBrown6f8d0c72011-05-11 14:38:44 +10001699 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1700 unsigned long flags;
1701 spin_lock_irqsave(&conf->device_lock, flags);
1702 mddev->degraded++;
1703 spin_unlock_irqrestore(&conf->device_lock, flags);
1704 /*
1705 * if recovery was running, make sure it aborts.
1706 */
1707 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
NeilBrown6f8d0c72011-05-11 14:38:44 +10001709 set_bit(Faulty, &rdev->flags);
1710 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1711 printk(KERN_ALERT
1712 "md/raid:%s: Disk failure on %s, disabling device.\n"
1713 "md/raid:%s: Operation continuing on %d devices.\n",
1714 mdname(mddev),
1715 bdevname(rdev->bdev, b),
1716 mdname(mddev),
1717 conf->raid_disks - mddev->degraded);
NeilBrown16a53ec2006-06-26 00:27:38 -07001718}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
1720/*
1721 * Input: a 'big' sector number,
1722 * Output: index of the data and parity disk, and the sector # in them.
1723 */
NeilBrown112bf892009-03-31 14:39:38 +11001724static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11001725 int previous, int *dd_idx,
1726 struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727{
NeilBrown6e3b96e2010-04-23 07:08:28 +10001728 sector_t stripe, stripe2;
NeilBrown35f2a592010-04-20 14:13:34 +10001729 sector_t chunk_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 unsigned int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11001731 int pd_idx, qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001732 int ddf_layout = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 sector_t new_sector;
NeilBrowne183eae2009-03-31 15:20:22 +11001734 int algorithm = previous ? conf->prev_algo
1735 : conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001736 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1737 : conf->chunk_sectors;
NeilBrown112bf892009-03-31 14:39:38 +11001738 int raid_disks = previous ? conf->previous_raid_disks
1739 : conf->raid_disks;
1740 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
1742 /* First compute the information on this sector */
1743
1744 /*
1745 * Compute the chunk number and the sector offset inside the chunk
1746 */
1747 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1748 chunk_number = r_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
1750 /*
1751 * Compute the stripe number
1752 */
NeilBrown35f2a592010-04-20 14:13:34 +10001753 stripe = chunk_number;
1754 *dd_idx = sector_div(stripe, data_disks);
NeilBrown6e3b96e2010-04-23 07:08:28 +10001755 stripe2 = stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 /*
1757 * Select the parity disk based on the user selected algorithm.
1758 */
NeilBrown911d4ee2009-03-31 14:39:38 +11001759 pd_idx = qd_idx = ~0;
NeilBrown16a53ec2006-06-26 00:27:38 -07001760 switch(conf->level) {
1761 case 4:
NeilBrown911d4ee2009-03-31 14:39:38 +11001762 pd_idx = data_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001763 break;
1764 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001765 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001767 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001768 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 (*dd_idx)++;
1770 break;
1771 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001772 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001773 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 (*dd_idx)++;
1775 break;
1776 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001777 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001778 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 break;
1780 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001781 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001782 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001784 case ALGORITHM_PARITY_0:
1785 pd_idx = 0;
1786 (*dd_idx)++;
1787 break;
1788 case ALGORITHM_PARITY_N:
1789 pd_idx = data_disks;
1790 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001792 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001793 }
1794 break;
1795 case 6:
1796
NeilBrowne183eae2009-03-31 15:20:22 +11001797 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001798 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001799 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001800 qd_idx = pd_idx + 1;
1801 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001802 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001803 qd_idx = 0;
1804 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001805 (*dd_idx) += 2; /* D D P Q D */
1806 break;
1807 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001808 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001809 qd_idx = pd_idx + 1;
1810 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001811 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001812 qd_idx = 0;
1813 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001814 (*dd_idx) += 2; /* D D P Q D */
1815 break;
1816 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001817 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001818 qd_idx = (pd_idx + 1) % raid_disks;
1819 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001820 break;
1821 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001822 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001823 qd_idx = (pd_idx + 1) % raid_disks;
1824 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001825 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001826
1827 case ALGORITHM_PARITY_0:
1828 pd_idx = 0;
1829 qd_idx = 1;
1830 (*dd_idx) += 2;
1831 break;
1832 case ALGORITHM_PARITY_N:
1833 pd_idx = data_disks;
1834 qd_idx = data_disks + 1;
1835 break;
1836
1837 case ALGORITHM_ROTATING_ZERO_RESTART:
1838 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1839 * of blocks for computing Q is different.
1840 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001841 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001842 qd_idx = pd_idx + 1;
1843 if (pd_idx == raid_disks-1) {
1844 (*dd_idx)++; /* Q D D D P */
1845 qd_idx = 0;
1846 } else if (*dd_idx >= pd_idx)
1847 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001848 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001849 break;
1850
1851 case ALGORITHM_ROTATING_N_RESTART:
1852 /* Same a left_asymmetric, by first stripe is
1853 * D D D P Q rather than
1854 * Q D D D P
1855 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001856 stripe2 += 1;
1857 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001858 qd_idx = pd_idx + 1;
1859 if (pd_idx == raid_disks-1) {
1860 (*dd_idx)++; /* Q D D D P */
1861 qd_idx = 0;
1862 } else if (*dd_idx >= pd_idx)
1863 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001864 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001865 break;
1866
1867 case ALGORITHM_ROTATING_N_CONTINUE:
1868 /* Same as left_symmetric but Q is before P */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001869 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001870 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1871 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
NeilBrown67cc2b82009-03-31 14:39:38 +11001872 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001873 break;
1874
1875 case ALGORITHM_LEFT_ASYMMETRIC_6:
1876 /* RAID5 left_asymmetric, with Q on last device */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001877 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001878 if (*dd_idx >= pd_idx)
1879 (*dd_idx)++;
1880 qd_idx = raid_disks - 1;
1881 break;
1882
1883 case ALGORITHM_RIGHT_ASYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001884 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001885 if (*dd_idx >= pd_idx)
1886 (*dd_idx)++;
1887 qd_idx = raid_disks - 1;
1888 break;
1889
1890 case ALGORITHM_LEFT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001891 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001892 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1893 qd_idx = raid_disks - 1;
1894 break;
1895
1896 case ALGORITHM_RIGHT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001897 pd_idx = 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_PARITY_0_6:
1903 pd_idx = 0;
1904 (*dd_idx)++;
1905 qd_idx = raid_disks - 1;
1906 break;
1907
NeilBrown16a53ec2006-06-26 00:27:38 -07001908 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001909 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001910 }
1911 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
1913
NeilBrown911d4ee2009-03-31 14:39:38 +11001914 if (sh) {
1915 sh->pd_idx = pd_idx;
1916 sh->qd_idx = qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001917 sh->ddf_layout = ddf_layout;
NeilBrown911d4ee2009-03-31 14:39:38 +11001918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 /*
1920 * Finally, compute the new sector number
1921 */
1922 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1923 return new_sector;
1924}
1925
1926
NeilBrown784052e2009-03-31 15:19:07 +11001927static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928{
1929 raid5_conf_t *conf = sh->raid_conf;
NeilBrownb875e532006-12-10 02:20:49 -08001930 int raid_disks = sh->disks;
1931 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 sector_t new_sector = sh->sector, check;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001933 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1934 : conf->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11001935 int algorithm = previous ? conf->prev_algo
1936 : conf->algorithm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 sector_t stripe;
1938 int chunk_offset;
NeilBrown35f2a592010-04-20 14:13:34 +10001939 sector_t chunk_number;
1940 int dummy1, dd_idx = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 sector_t r_sector;
NeilBrown911d4ee2009-03-31 14:39:38 +11001942 struct stripe_head sh2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
NeilBrown16a53ec2006-06-26 00:27:38 -07001944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1946 stripe = new_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
NeilBrown16a53ec2006-06-26 00:27:38 -07001948 if (i == sh->pd_idx)
1949 return 0;
1950 switch(conf->level) {
1951 case 4: break;
1952 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001953 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 case ALGORITHM_LEFT_ASYMMETRIC:
1955 case ALGORITHM_RIGHT_ASYMMETRIC:
1956 if (i > sh->pd_idx)
1957 i--;
1958 break;
1959 case ALGORITHM_LEFT_SYMMETRIC:
1960 case ALGORITHM_RIGHT_SYMMETRIC:
1961 if (i < sh->pd_idx)
1962 i += raid_disks;
1963 i -= (sh->pd_idx + 1);
1964 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001965 case ALGORITHM_PARITY_0:
1966 i -= 1;
1967 break;
1968 case ALGORITHM_PARITY_N:
1969 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001971 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001972 }
1973 break;
1974 case 6:
NeilBrownd0dabf72009-03-31 14:39:38 +11001975 if (i == sh->qd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001976 return 0; /* It is the Q disk */
NeilBrowne183eae2009-03-31 15:20:22 +11001977 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001978 case ALGORITHM_LEFT_ASYMMETRIC:
1979 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown99c0fb52009-03-31 14:39:38 +11001980 case ALGORITHM_ROTATING_ZERO_RESTART:
1981 case ALGORITHM_ROTATING_N_RESTART:
1982 if (sh->pd_idx == raid_disks-1)
1983 i--; /* Q D D D P */
NeilBrown16a53ec2006-06-26 00:27:38 -07001984 else if (i > sh->pd_idx)
1985 i -= 2; /* D D P Q D */
1986 break;
1987 case ALGORITHM_LEFT_SYMMETRIC:
1988 case ALGORITHM_RIGHT_SYMMETRIC:
1989 if (sh->pd_idx == raid_disks-1)
1990 i--; /* Q D D D P */
1991 else {
1992 /* D D P Q D */
1993 if (i < sh->pd_idx)
1994 i += raid_disks;
1995 i -= (sh->pd_idx + 2);
1996 }
1997 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001998 case ALGORITHM_PARITY_0:
1999 i -= 2;
2000 break;
2001 case ALGORITHM_PARITY_N:
2002 break;
2003 case ALGORITHM_ROTATING_N_CONTINUE:
NeilBrowne4424fe2009-10-16 16:27:34 +11002004 /* Like left_symmetric, but P is before Q */
NeilBrown99c0fb52009-03-31 14:39:38 +11002005 if (sh->pd_idx == 0)
2006 i--; /* P D D D Q */
NeilBrowne4424fe2009-10-16 16:27:34 +11002007 else {
2008 /* D D Q P D */
2009 if (i < sh->pd_idx)
2010 i += raid_disks;
2011 i -= (sh->pd_idx + 1);
2012 }
NeilBrown99c0fb52009-03-31 14:39:38 +11002013 break;
2014 case ALGORITHM_LEFT_ASYMMETRIC_6:
2015 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2016 if (i > sh->pd_idx)
2017 i--;
2018 break;
2019 case ALGORITHM_LEFT_SYMMETRIC_6:
2020 case ALGORITHM_RIGHT_SYMMETRIC_6:
2021 if (i < sh->pd_idx)
2022 i += data_disks + 1;
2023 i -= (sh->pd_idx + 1);
2024 break;
2025 case ALGORITHM_PARITY_0_6:
2026 i -= 1;
2027 break;
NeilBrown16a53ec2006-06-26 00:27:38 -07002028 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002029 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002030 }
2031 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 }
2033
2034 chunk_number = stripe * data_disks + i;
NeilBrown35f2a592010-04-20 14:13:34 +10002035 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
NeilBrown112bf892009-03-31 14:39:38 +11002037 check = raid5_compute_sector(conf, r_sector,
NeilBrown784052e2009-03-31 15:19:07 +11002038 previous, &dummy1, &sh2);
NeilBrown911d4ee2009-03-31 14:39:38 +11002039 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2040 || sh2.qd_idx != sh->qd_idx) {
NeilBrown0c55e022010-05-03 14:09:02 +10002041 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2042 mdname(conf->mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return 0;
2044 }
2045 return r_sector;
2046}
2047
2048
Dan Williams600aa102008-06-28 08:32:05 +10002049static void
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002050schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
Dan Williams600aa102008-06-28 08:32:05 +10002051 int rcw, int expand)
Dan Williamse33129d2007-01-02 13:52:30 -07002052{
2053 int i, pd_idx = sh->pd_idx, disks = sh->disks;
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002054 raid5_conf_t *conf = sh->raid_conf;
2055 int level = conf->level;
NeilBrown16a53ec2006-06-26 00:27:38 -07002056
Dan Williamse33129d2007-01-02 13:52:30 -07002057 if (rcw) {
2058 /* if we are not expanding this is a proper write request, and
2059 * there will be bios with new data to be drained into the
2060 * stripe cache
2061 */
2062 if (!expand) {
Dan Williams600aa102008-06-28 08:32:05 +10002063 sh->reconstruct_state = reconstruct_state_drain_run;
2064 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2065 } else
2066 sh->reconstruct_state = reconstruct_state_run;
Dan Williamse33129d2007-01-02 13:52:30 -07002067
Dan Williamsac6b53b2009-07-14 13:40:19 -07002068 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002069
2070 for (i = disks; i--; ) {
2071 struct r5dev *dev = &sh->dev[i];
2072
2073 if (dev->towrite) {
2074 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsd8ee0722008-06-28 08:32:06 +10002075 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002076 if (!expand)
2077 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002078 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002079 }
2080 }
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002081 if (s->locked + conf->max_degraded == disks)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002082 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002083 atomic_inc(&conf->pending_full_writes);
Dan Williamse33129d2007-01-02 13:52:30 -07002084 } else {
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002085 BUG_ON(level == 6);
Dan Williamse33129d2007-01-02 13:52:30 -07002086 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2087 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2088
Dan Williamsd8ee0722008-06-28 08:32:06 +10002089 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
Dan Williams600aa102008-06-28 08:32:05 +10002090 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2091 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
Dan Williamsac6b53b2009-07-14 13:40:19 -07002092 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002093
2094 for (i = disks; i--; ) {
2095 struct r5dev *dev = &sh->dev[i];
2096 if (i == pd_idx)
2097 continue;
2098
Dan Williamse33129d2007-01-02 13:52:30 -07002099 if (dev->towrite &&
2100 (test_bit(R5_UPTODATE, &dev->flags) ||
Dan Williamsd8ee0722008-06-28 08:32:06 +10002101 test_bit(R5_Wantcompute, &dev->flags))) {
2102 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002103 set_bit(R5_LOCKED, &dev->flags);
2104 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002105 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002106 }
2107 }
2108 }
2109
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002110 /* keep the parity disk(s) locked while asynchronous operations
Dan Williamse33129d2007-01-02 13:52:30 -07002111 * are in flight
2112 */
2113 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2114 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
Dan Williams600aa102008-06-28 08:32:05 +10002115 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002116
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002117 if (level == 6) {
2118 int qd_idx = sh->qd_idx;
2119 struct r5dev *dev = &sh->dev[qd_idx];
2120
2121 set_bit(R5_LOCKED, &dev->flags);
2122 clear_bit(R5_UPTODATE, &dev->flags);
2123 s->locked++;
2124 }
2125
Dan Williams600aa102008-06-28 08:32:05 +10002126 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07002127 __func__, (unsigned long long)sh->sector,
Dan Williams600aa102008-06-28 08:32:05 +10002128 s->locked, s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002129}
NeilBrown16a53ec2006-06-26 00:27:38 -07002130
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131/*
2132 * Each stripe/dev can have one or more bion attached.
NeilBrown16a53ec2006-06-26 00:27:38 -07002133 * toread/towrite point to the first in a chain.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 * The bi_next chain must be in order.
2135 */
2136static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2137{
2138 struct bio **bip;
2139 raid5_conf_t *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -07002140 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
NeilBrowncbe47ec2011-07-26 11:20:35 +10002142 pr_debug("adding bi b#%llu to stripe s#%llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 (unsigned long long)bi->bi_sector,
2144 (unsigned long long)sh->sector);
2145
2146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07002148 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 bip = &sh->dev[dd_idx].towrite;
NeilBrown72626682005-09-09 16:23:54 -07002150 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2151 firstwrite = 1;
2152 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 bip = &sh->dev[dd_idx].toread;
2154 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2155 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2156 goto overlap;
2157 bip = & (*bip)->bi_next;
2158 }
2159 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2160 goto overlap;
2161
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002162 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 if (*bip)
2164 bi->bi_next = *bip;
2165 *bip = bi;
Jens Axboe960e7392008-08-15 10:41:18 +02002166 bi->bi_phys_segments++;
NeilBrown72626682005-09-09 16:23:54 -07002167
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 if (forwrite) {
2169 /* check if page is covered */
2170 sector_t sector = sh->dev[dd_idx].sector;
2171 for (bi=sh->dev[dd_idx].towrite;
2172 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2173 bi && bi->bi_sector <= sector;
2174 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2175 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2176 sector = bi->bi_sector + (bi->bi_size>>9);
2177 }
2178 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2179 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2180 }
NeilBrowncbe47ec2011-07-26 11:20:35 +10002181 spin_unlock_irq(&conf->device_lock);
NeilBrowncbe47ec2011-07-26 11:20:35 +10002182
2183 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
2184 (unsigned long long)(*bip)->bi_sector,
2185 (unsigned long long)sh->sector, dd_idx);
2186
2187 if (conf->mddev->bitmap && firstwrite) {
2188 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2189 STRIPE_SECTORS, 0);
2190 sh->bm_seq = conf->seq_flush+1;
2191 set_bit(STRIPE_BIT_DELAY, &sh->state);
2192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 return 1;
2194
2195 overlap:
2196 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2197 spin_unlock_irq(&conf->device_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 return 0;
2199}
2200
NeilBrown29269552006-03-27 01:18:10 -08002201static void end_reshape(raid5_conf_t *conf);
2202
NeilBrown911d4ee2009-03-31 14:39:38 +11002203static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
2204 struct stripe_head *sh)
NeilBrownccfcc3c2006-03-27 01:18:09 -08002205{
NeilBrown784052e2009-03-31 15:19:07 +11002206 int sectors_per_chunk =
Andre Noll09c9e5f2009-06-18 08:45:55 +10002207 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
NeilBrown911d4ee2009-03-31 14:39:38 +11002208 int dd_idx;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002209 int chunk_offset = sector_div(stripe, sectors_per_chunk);
NeilBrown112bf892009-03-31 14:39:38 +11002210 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002211
NeilBrown112bf892009-03-31 14:39:38 +11002212 raid5_compute_sector(conf,
2213 stripe * (disks - conf->max_degraded)
NeilBrownb875e532006-12-10 02:20:49 -08002214 *sectors_per_chunk + chunk_offset,
NeilBrown112bf892009-03-31 14:39:38 +11002215 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002216 &dd_idx, sh);
NeilBrownccfcc3c2006-03-27 01:18:09 -08002217}
2218
Dan Williamsa4456852007-07-09 11:56:43 -07002219static void
Dan Williams1fe797e2008-06-28 09:16:30 +10002220handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002221 struct stripe_head_state *s, int disks,
2222 struct bio **return_bi)
2223{
2224 int i;
2225 for (i = disks; i--; ) {
2226 struct bio *bi;
2227 int bitmap_end = 0;
2228
2229 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2230 mdk_rdev_t *rdev;
2231 rcu_read_lock();
2232 rdev = rcu_dereference(conf->disks[i].rdev);
2233 if (rdev && test_bit(In_sync, &rdev->flags))
2234 /* multiple read failures in one stripe */
2235 md_error(conf->mddev, rdev);
2236 rcu_read_unlock();
2237 }
2238 spin_lock_irq(&conf->device_lock);
2239 /* fail all writes first */
2240 bi = sh->dev[i].towrite;
2241 sh->dev[i].towrite = NULL;
2242 if (bi) {
2243 s->to_write--;
2244 bitmap_end = 1;
2245 }
2246
2247 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2248 wake_up(&conf->wait_for_overlap);
2249
2250 while (bi && bi->bi_sector <
2251 sh->dev[i].sector + STRIPE_SECTORS) {
2252 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2253 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002254 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002255 md_write_end(conf->mddev);
2256 bi->bi_next = *return_bi;
2257 *return_bi = bi;
2258 }
2259 bi = nextbi;
2260 }
2261 /* and fail all 'written' */
2262 bi = sh->dev[i].written;
2263 sh->dev[i].written = NULL;
2264 if (bi) bitmap_end = 1;
2265 while (bi && bi->bi_sector <
2266 sh->dev[i].sector + STRIPE_SECTORS) {
2267 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2268 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002269 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002270 md_write_end(conf->mddev);
2271 bi->bi_next = *return_bi;
2272 *return_bi = bi;
2273 }
2274 bi = bi2;
2275 }
2276
Dan Williamsb5e98d62007-01-02 13:52:31 -07002277 /* fail any reads if this device is non-operational and
2278 * the data has not reached the cache yet.
2279 */
2280 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2281 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2282 test_bit(R5_ReadError, &sh->dev[i].flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002283 bi = sh->dev[i].toread;
2284 sh->dev[i].toread = NULL;
2285 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2286 wake_up(&conf->wait_for_overlap);
2287 if (bi) s->to_read--;
2288 while (bi && bi->bi_sector <
2289 sh->dev[i].sector + STRIPE_SECTORS) {
2290 struct bio *nextbi =
2291 r5_next_bio(bi, sh->dev[i].sector);
2292 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002293 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002294 bi->bi_next = *return_bi;
2295 *return_bi = bi;
2296 }
2297 bi = nextbi;
2298 }
2299 }
2300 spin_unlock_irq(&conf->device_lock);
2301 if (bitmap_end)
2302 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2303 STRIPE_SECTORS, 0, 0);
2304 }
2305
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002306 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2307 if (atomic_dec_and_test(&conf->pending_full_writes))
2308 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002309}
2310
Dan Williams1fe797e2008-06-28 09:16:30 +10002311/* fetch_block5 - checks the given member device to see if its data needs
2312 * to be read or computed to satisfy a request.
2313 *
2314 * Returns 1 when no more member devices need to be checked, otherwise returns
2315 * 0 to tell the loop in handle_stripe_fill5 to continue
Dan Williamsf38e1212007-01-02 13:52:30 -07002316 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002317static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
2318 int disk_idx, int disks)
Dan Williamsf38e1212007-01-02 13:52:30 -07002319{
2320 struct r5dev *dev = &sh->dev[disk_idx];
NeilBrownf2b3b442011-07-26 11:35:19 +10002321 struct r5dev *failed_dev = &sh->dev[s->failed_num[0]];
Dan Williamsf38e1212007-01-02 13:52:30 -07002322
Dan Williamsf38e1212007-01-02 13:52:30 -07002323 /* is the data in this block needed, and can we get it? */
2324 if (!test_bit(R5_LOCKED, &dev->flags) &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002325 !test_bit(R5_UPTODATE, &dev->flags) &&
2326 (dev->toread ||
2327 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2328 s->syncing || s->expanding ||
2329 (s->failed &&
2330 (failed_dev->toread ||
2331 (failed_dev->towrite &&
2332 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002333 /* We would like to get this block, possibly by computing it,
2334 * otherwise read it if the backing disk is insync
Dan Williamsf38e1212007-01-02 13:52:30 -07002335 */
2336 if ((s->uptodate == disks - 1) &&
NeilBrownf2b3b442011-07-26 11:35:19 +10002337 (s->failed && disk_idx == s->failed_num[0])) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002338 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2339 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
Dan Williamsf38e1212007-01-02 13:52:30 -07002340 set_bit(R5_Wantcompute, &dev->flags);
2341 sh->ops.target = disk_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002342 sh->ops.target2 = -1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002343 s->req_compute = 1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002344 /* Careful: from this point on 'uptodate' is in the eye
Dan Williamsac6b53b2009-07-14 13:40:19 -07002345 * of raid_run_ops which services 'compute' operations
Dan Williamsf38e1212007-01-02 13:52:30 -07002346 * before writes. R5_Wantcompute flags a block that will
2347 * be R5_UPTODATE by the time it is needed for a
2348 * subsequent operation.
2349 */
2350 s->uptodate++;
Dan Williams1fe797e2008-06-28 09:16:30 +10002351 return 1; /* uptodate + compute == disks */
Dan Williams7a1fc532008-07-10 04:54:57 -07002352 } else if (test_bit(R5_Insync, &dev->flags)) {
Dan Williamsf38e1212007-01-02 13:52:30 -07002353 set_bit(R5_LOCKED, &dev->flags);
2354 set_bit(R5_Wantread, &dev->flags);
Dan Williamsf38e1212007-01-02 13:52:30 -07002355 s->locked++;
2356 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
2357 s->syncing);
2358 }
2359 }
2360
Dan Williams1fe797e2008-06-28 09:16:30 +10002361 return 0;
Dan Williamsf38e1212007-01-02 13:52:30 -07002362}
2363
Dan Williams1fe797e2008-06-28 09:16:30 +10002364/**
2365 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
2366 */
2367static void handle_stripe_fill5(struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002368 struct stripe_head_state *s, int disks)
2369{
2370 int i;
Dan Williamsf38e1212007-01-02 13:52:30 -07002371
Dan Williamsf38e1212007-01-02 13:52:30 -07002372 /* look for blocks to read/compute, skip this if a compute
2373 * is already in flight, or if the stripe contents are in the
2374 * midst of changing due to a write
2375 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002376 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002377 !sh->reconstruct_state)
Dan Williamsf38e1212007-01-02 13:52:30 -07002378 for (i = disks; i--; )
Dan Williams1fe797e2008-06-28 09:16:30 +10002379 if (fetch_block5(sh, s, i, disks))
Dan Williamsf38e1212007-01-02 13:52:30 -07002380 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002381 set_bit(STRIPE_HANDLE, &sh->state);
2382}
2383
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002384/* fetch_block6 - checks the given member device to see if its data needs
2385 * to be read or computed to satisfy a request.
2386 *
2387 * Returns 1 when no more member devices need to be checked, otherwise returns
2388 * 0 to tell the loop in handle_stripe_fill6 to continue
2389 */
2390static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s,
NeilBrownf2b3b442011-07-26 11:35:19 +10002391 int disk_idx, int disks)
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002392{
2393 struct r5dev *dev = &sh->dev[disk_idx];
NeilBrownf2b3b442011-07-26 11:35:19 +10002394 struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]],
2395 &sh->dev[s->failed_num[1]] };
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002396
2397 if (!test_bit(R5_LOCKED, &dev->flags) &&
2398 !test_bit(R5_UPTODATE, &dev->flags) &&
2399 (dev->toread ||
2400 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2401 s->syncing || s->expanding ||
2402 (s->failed >= 1 &&
2403 (fdev[0]->toread || s->to_write)) ||
2404 (s->failed >= 2 &&
2405 (fdev[1]->toread || s->to_write)))) {
2406 /* we would like to get this block, possibly by computing it,
2407 * otherwise read it if the backing disk is insync
2408 */
2409 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2410 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2411 if ((s->uptodate == disks - 1) &&
NeilBrownf2b3b442011-07-26 11:35:19 +10002412 (s->failed && (disk_idx == s->failed_num[0] ||
2413 disk_idx == s->failed_num[1]))) {
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002414 /* have disk failed, and we're requested to fetch it;
2415 * do compute it
2416 */
2417 pr_debug("Computing stripe %llu block %d\n",
2418 (unsigned long long)sh->sector, disk_idx);
2419 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2420 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2421 set_bit(R5_Wantcompute, &dev->flags);
2422 sh->ops.target = disk_idx;
2423 sh->ops.target2 = -1; /* no 2nd target */
2424 s->req_compute = 1;
2425 s->uptodate++;
2426 return 1;
2427 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2428 /* Computing 2-failure is *very* expensive; only
2429 * do it if failed >= 2
2430 */
2431 int other;
2432 for (other = disks; other--; ) {
2433 if (other == disk_idx)
2434 continue;
2435 if (!test_bit(R5_UPTODATE,
2436 &sh->dev[other].flags))
2437 break;
2438 }
2439 BUG_ON(other < 0);
2440 pr_debug("Computing stripe %llu blocks %d,%d\n",
2441 (unsigned long long)sh->sector,
2442 disk_idx, other);
2443 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2444 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2445 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2446 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2447 sh->ops.target = disk_idx;
2448 sh->ops.target2 = other;
2449 s->uptodate += 2;
2450 s->req_compute = 1;
2451 return 1;
2452 } else if (test_bit(R5_Insync, &dev->flags)) {
2453 set_bit(R5_LOCKED, &dev->flags);
2454 set_bit(R5_Wantread, &dev->flags);
2455 s->locked++;
2456 pr_debug("Reading block %d (sync=%d)\n",
2457 disk_idx, s->syncing);
2458 }
2459 }
2460
2461 return 0;
2462}
2463
2464/**
2465 * handle_stripe_fill6 - read or compute data to satisfy pending requests.
2466 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002467static void handle_stripe_fill6(struct stripe_head *sh,
NeilBrownf2b3b442011-07-26 11:35:19 +10002468 struct stripe_head_state *s,
Dan Williamsa4456852007-07-09 11:56:43 -07002469 int disks)
2470{
2471 int i;
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002472
2473 /* look for blocks to read/compute, skip this if a compute
2474 * is already in flight, or if the stripe contents are in the
2475 * midst of changing due to a write
2476 */
2477 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2478 !sh->reconstruct_state)
2479 for (i = disks; i--; )
NeilBrownf2b3b442011-07-26 11:35:19 +10002480 if (fetch_block6(sh, s, i, disks))
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002481 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002482 set_bit(STRIPE_HANDLE, &sh->state);
2483}
2484
2485
Dan Williams1fe797e2008-06-28 09:16:30 +10002486/* handle_stripe_clean_event
Dan Williamsa4456852007-07-09 11:56:43 -07002487 * any written block on an uptodate or failed drive can be returned.
2488 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2489 * never LOCKED, so we don't need to test 'failed' directly.
2490 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002491static void handle_stripe_clean_event(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002492 struct stripe_head *sh, int disks, struct bio **return_bi)
2493{
2494 int i;
2495 struct r5dev *dev;
2496
2497 for (i = disks; i--; )
2498 if (sh->dev[i].written) {
2499 dev = &sh->dev[i];
2500 if (!test_bit(R5_LOCKED, &dev->flags) &&
2501 test_bit(R5_UPTODATE, &dev->flags)) {
2502 /* We can return any write requests */
2503 struct bio *wbi, *wbi2;
2504 int bitmap_end = 0;
Dan Williams45b42332007-07-09 11:56:43 -07002505 pr_debug("Return write for disc %d\n", i);
Dan Williamsa4456852007-07-09 11:56:43 -07002506 spin_lock_irq(&conf->device_lock);
2507 wbi = dev->written;
2508 dev->written = NULL;
2509 while (wbi && wbi->bi_sector <
2510 dev->sector + STRIPE_SECTORS) {
2511 wbi2 = r5_next_bio(wbi, dev->sector);
Jens Axboe960e7392008-08-15 10:41:18 +02002512 if (!raid5_dec_bi_phys_segments(wbi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002513 md_write_end(conf->mddev);
2514 wbi->bi_next = *return_bi;
2515 *return_bi = wbi;
2516 }
2517 wbi = wbi2;
2518 }
2519 if (dev->towrite == NULL)
2520 bitmap_end = 1;
2521 spin_unlock_irq(&conf->device_lock);
2522 if (bitmap_end)
2523 bitmap_endwrite(conf->mddev->bitmap,
2524 sh->sector,
2525 STRIPE_SECTORS,
2526 !test_bit(STRIPE_DEGRADED, &sh->state),
2527 0);
2528 }
2529 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002530
2531 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2532 if (atomic_dec_and_test(&conf->pending_full_writes))
2533 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002534}
2535
Dan Williams1fe797e2008-06-28 09:16:30 +10002536static void handle_stripe_dirtying5(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002537 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2538{
2539 int rmw = 0, rcw = 0, i;
2540 for (i = disks; i--; ) {
2541 /* would I have to read this buffer for read_modify_write */
2542 struct r5dev *dev = &sh->dev[i];
2543 if ((dev->towrite || i == sh->pd_idx) &&
2544 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002545 !(test_bit(R5_UPTODATE, &dev->flags) ||
2546 test_bit(R5_Wantcompute, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002547 if (test_bit(R5_Insync, &dev->flags))
2548 rmw++;
2549 else
2550 rmw += 2*disks; /* cannot read it */
2551 }
2552 /* Would I have to read this buffer for reconstruct_write */
2553 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2554 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002555 !(test_bit(R5_UPTODATE, &dev->flags) ||
2556 test_bit(R5_Wantcompute, &dev->flags))) {
2557 if (test_bit(R5_Insync, &dev->flags)) rcw++;
Dan Williamsa4456852007-07-09 11:56:43 -07002558 else
2559 rcw += 2*disks;
2560 }
2561 }
Dan Williams45b42332007-07-09 11:56:43 -07002562 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07002563 (unsigned long long)sh->sector, rmw, rcw);
2564 set_bit(STRIPE_HANDLE, &sh->state);
2565 if (rmw < rcw && rmw > 0)
2566 /* prefer read-modify-write, but need to get some data */
2567 for (i = disks; i--; ) {
2568 struct r5dev *dev = &sh->dev[i];
2569 if ((dev->towrite || i == sh->pd_idx) &&
2570 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002571 !(test_bit(R5_UPTODATE, &dev->flags) ||
2572 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002573 test_bit(R5_Insync, &dev->flags)) {
2574 if (
2575 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002576 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002577 "%d for r-m-w\n", i);
2578 set_bit(R5_LOCKED, &dev->flags);
2579 set_bit(R5_Wantread, &dev->flags);
2580 s->locked++;
2581 } else {
2582 set_bit(STRIPE_DELAYED, &sh->state);
2583 set_bit(STRIPE_HANDLE, &sh->state);
2584 }
2585 }
2586 }
2587 if (rcw <= rmw && rcw > 0)
2588 /* want reconstruct write, but need to get some data */
2589 for (i = disks; i--; ) {
2590 struct r5dev *dev = &sh->dev[i];
2591 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2592 i != sh->pd_idx &&
2593 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002594 !(test_bit(R5_UPTODATE, &dev->flags) ||
2595 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002596 test_bit(R5_Insync, &dev->flags)) {
2597 if (
2598 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002599 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002600 "%d for Reconstruct\n", i);
2601 set_bit(R5_LOCKED, &dev->flags);
2602 set_bit(R5_Wantread, &dev->flags);
2603 s->locked++;
2604 } else {
2605 set_bit(STRIPE_DELAYED, &sh->state);
2606 set_bit(STRIPE_HANDLE, &sh->state);
2607 }
2608 }
2609 }
2610 /* now if nothing is locked, and if we have enough data,
2611 * we can start a write request
2612 */
Dan Williamsf38e1212007-01-02 13:52:30 -07002613 /* since handle_stripe can be called at any time we need to handle the
2614 * case where a compute block operation has been submitted and then a
Dan Williamsac6b53b2009-07-14 13:40:19 -07002615 * subsequent call wants to start a write request. raid_run_ops only
2616 * handles the case where compute block and reconstruct are requested
Dan Williamsf38e1212007-01-02 13:52:30 -07002617 * simultaneously. If this is not the case then new writes need to be
2618 * held off until the compute completes.
2619 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002620 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2621 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2622 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002623 schedule_reconstruction(sh, s, rcw == 0, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002624}
2625
Dan Williams1fe797e2008-06-28 09:16:30 +10002626static void handle_stripe_dirtying6(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002627 struct stripe_head *sh, struct stripe_head_state *s,
NeilBrownf2b3b442011-07-26 11:35:19 +10002628 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07002629{
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002630 int rcw = 0, pd_idx = sh->pd_idx, i;
NeilBrown34e04e82009-03-31 15:10:16 +11002631 int qd_idx = sh->qd_idx;
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002632
2633 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002634 for (i = disks; i--; ) {
2635 struct r5dev *dev = &sh->dev[i];
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002636 /* check if we haven't enough data */
2637 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2638 i != pd_idx && i != qd_idx &&
2639 !test_bit(R5_LOCKED, &dev->flags) &&
2640 !(test_bit(R5_UPTODATE, &dev->flags) ||
2641 test_bit(R5_Wantcompute, &dev->flags))) {
2642 rcw++;
2643 if (!test_bit(R5_Insync, &dev->flags))
2644 continue; /* it's a failed drive */
2645
2646 if (
2647 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2648 pr_debug("Read_old stripe %llu "
2649 "block %d for Reconstruct\n",
2650 (unsigned long long)sh->sector, i);
2651 set_bit(R5_LOCKED, &dev->flags);
2652 set_bit(R5_Wantread, &dev->flags);
2653 s->locked++;
2654 } else {
2655 pr_debug("Request delayed stripe %llu "
2656 "block %d for Reconstruct\n",
2657 (unsigned long long)sh->sector, i);
2658 set_bit(STRIPE_DELAYED, &sh->state);
2659 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002660 }
2661 }
2662 }
Dan Williamsa4456852007-07-09 11:56:43 -07002663 /* now if nothing is locked, and if we have enough data, we can start a
2664 * write request
2665 */
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002666 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2667 s->locked == 0 && rcw == 0 &&
Dan Williamsa4456852007-07-09 11:56:43 -07002668 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002669 schedule_reconstruction(sh, s, 1, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002670 }
2671}
2672
2673static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2674 struct stripe_head_state *s, int disks)
2675{
Dan Williamsecc65c92008-06-28 08:31:57 +10002676 struct r5dev *dev = NULL;
Dan Williamse89f8962007-01-02 13:52:31 -07002677
Dan Williamsbd2ab672008-04-10 21:29:27 -07002678 set_bit(STRIPE_HANDLE, &sh->state);
2679
Dan Williamsecc65c92008-06-28 08:31:57 +10002680 switch (sh->check_state) {
2681 case check_state_idle:
2682 /* start a new check operation if there are no failures */
Dan Williamsbd2ab672008-04-10 21:29:27 -07002683 if (s->failed == 0) {
Dan Williamsbd2ab672008-04-10 21:29:27 -07002684 BUG_ON(s->uptodate != disks);
Dan Williamsecc65c92008-06-28 08:31:57 +10002685 sh->check_state = check_state_run;
2686 set_bit(STRIPE_OP_CHECK, &s->ops_request);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002687 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002688 s->uptodate--;
Dan Williamsecc65c92008-06-28 08:31:57 +10002689 break;
Dan Williamsbd2ab672008-04-10 21:29:27 -07002690 }
NeilBrownf2b3b442011-07-26 11:35:19 +10002691 dev = &sh->dev[s->failed_num[0]];
Dan Williamsecc65c92008-06-28 08:31:57 +10002692 /* fall through */
2693 case check_state_compute_result:
2694 sh->check_state = check_state_idle;
2695 if (!dev)
2696 dev = &sh->dev[sh->pd_idx];
2697
2698 /* check that a write has not made the stripe insync */
2699 if (test_bit(STRIPE_INSYNC, &sh->state))
2700 break;
2701
2702 /* either failed parity check, or recovery is happening */
Dan Williamsa4456852007-07-09 11:56:43 -07002703 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2704 BUG_ON(s->uptodate != disks);
2705
2706 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsecc65c92008-06-28 08:31:57 +10002707 s->locked++;
Dan Williamsa4456852007-07-09 11:56:43 -07002708 set_bit(R5_Wantwrite, &dev->flags);
Dan Williams830ea012007-01-02 13:52:31 -07002709
Dan Williamsa4456852007-07-09 11:56:43 -07002710 clear_bit(STRIPE_DEGRADED, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002711 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002712 break;
2713 case check_state_run:
2714 break; /* we will be called again upon completion */
2715 case check_state_check_result:
2716 sh->check_state = check_state_idle;
2717
2718 /* if a failure occurred during the check operation, leave
2719 * STRIPE_INSYNC not set and let the stripe be handled again
2720 */
2721 if (s->failed)
2722 break;
2723
2724 /* handle a successful check operation, if parity is correct
2725 * we are done. Otherwise update the mismatch count and repair
2726 * parity if !MD_RECOVERY_CHECK
2727 */
Dan Williamsad283ea2009-08-29 19:09:26 -07002728 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
Dan Williamsecc65c92008-06-28 08:31:57 +10002729 /* parity is correct (on disc,
2730 * not in buffer any more)
2731 */
2732 set_bit(STRIPE_INSYNC, &sh->state);
2733 else {
2734 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2735 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2736 /* don't try to repair!! */
2737 set_bit(STRIPE_INSYNC, &sh->state);
2738 else {
2739 sh->check_state = check_state_compute_run;
Dan Williams976ea8d2008-06-28 08:32:03 +10002740 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002741 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2742 set_bit(R5_Wantcompute,
2743 &sh->dev[sh->pd_idx].flags);
2744 sh->ops.target = sh->pd_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002745 sh->ops.target2 = -1;
Dan Williamsecc65c92008-06-28 08:31:57 +10002746 s->uptodate++;
2747 }
2748 }
2749 break;
2750 case check_state_compute_run:
2751 break;
2752 default:
2753 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2754 __func__, sh->check_state,
2755 (unsigned long long) sh->sector);
2756 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002757 }
2758}
2759
2760
2761static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williams36d1c642009-07-14 11:48:22 -07002762 struct stripe_head_state *s,
NeilBrownf2b3b442011-07-26 11:35:19 +10002763 int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07002764{
Dan Williamsa4456852007-07-09 11:56:43 -07002765 int pd_idx = sh->pd_idx;
NeilBrown34e04e82009-03-31 15:10:16 +11002766 int qd_idx = sh->qd_idx;
Dan Williamsd82dfee2009-07-14 13:40:57 -07002767 struct r5dev *dev;
Dan Williamsa4456852007-07-09 11:56:43 -07002768
2769 set_bit(STRIPE_HANDLE, &sh->state);
2770
2771 BUG_ON(s->failed > 2);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002772
Dan Williamsa4456852007-07-09 11:56:43 -07002773 /* Want to check and possibly repair P and Q.
2774 * However there could be one 'failed' device, in which
2775 * case we can only check one of them, possibly using the
2776 * other to generate missing data
2777 */
2778
Dan Williamsd82dfee2009-07-14 13:40:57 -07002779 switch (sh->check_state) {
2780 case check_state_idle:
2781 /* start a new check operation if there are < 2 failures */
NeilBrownf2b3b442011-07-26 11:35:19 +10002782 if (s->failed == s->q_failed) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002783 /* The only possible failed device holds Q, so it
Dan Williamsa4456852007-07-09 11:56:43 -07002784 * makes sense to check P (If anything else were failed,
2785 * we would have used P to recreate it).
2786 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002787 sh->check_state = check_state_run;
Dan Williamsa4456852007-07-09 11:56:43 -07002788 }
NeilBrownf2b3b442011-07-26 11:35:19 +10002789 if (!s->q_failed && s->failed < 2) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002790 /* Q is not failed, and we didn't use it to generate
Dan Williamsa4456852007-07-09 11:56:43 -07002791 * anything, so it makes sense to check it
2792 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002793 if (sh->check_state == check_state_run)
2794 sh->check_state = check_state_run_pq;
2795 else
2796 sh->check_state = check_state_run_q;
Dan Williamsa4456852007-07-09 11:56:43 -07002797 }
Dan Williams36d1c642009-07-14 11:48:22 -07002798
Dan Williamsd82dfee2009-07-14 13:40:57 -07002799 /* discard potentially stale zero_sum_result */
2800 sh->ops.zero_sum_result = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07002801
Dan Williamsd82dfee2009-07-14 13:40:57 -07002802 if (sh->check_state == check_state_run) {
2803 /* async_xor_zero_sum destroys the contents of P */
2804 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2805 s->uptodate--;
Dan Williamsa4456852007-07-09 11:56:43 -07002806 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002807 if (sh->check_state >= check_state_run &&
2808 sh->check_state <= check_state_run_pq) {
2809 /* async_syndrome_zero_sum preserves P and Q, so
2810 * no need to mark them !uptodate here
2811 */
2812 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2813 break;
2814 }
Dan Williams36d1c642009-07-14 11:48:22 -07002815
Dan Williamsd82dfee2009-07-14 13:40:57 -07002816 /* we have 2-disk failure */
2817 BUG_ON(s->failed != 2);
2818 /* fall through */
2819 case check_state_compute_result:
2820 sh->check_state = check_state_idle;
Dan Williams36d1c642009-07-14 11:48:22 -07002821
Dan Williamsd82dfee2009-07-14 13:40:57 -07002822 /* check that a write has not made the stripe insync */
2823 if (test_bit(STRIPE_INSYNC, &sh->state))
2824 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002825
2826 /* now write out any block on a failed drive,
Dan Williamsd82dfee2009-07-14 13:40:57 -07002827 * or P or Q if they were recomputed
Dan Williamsa4456852007-07-09 11:56:43 -07002828 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002829 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
Dan Williamsa4456852007-07-09 11:56:43 -07002830 if (s->failed == 2) {
NeilBrownf2b3b442011-07-26 11:35:19 +10002831 dev = &sh->dev[s->failed_num[1]];
Dan Williamsa4456852007-07-09 11:56:43 -07002832 s->locked++;
2833 set_bit(R5_LOCKED, &dev->flags);
2834 set_bit(R5_Wantwrite, &dev->flags);
2835 }
2836 if (s->failed >= 1) {
NeilBrownf2b3b442011-07-26 11:35:19 +10002837 dev = &sh->dev[s->failed_num[0]];
Dan Williamsa4456852007-07-09 11:56:43 -07002838 s->locked++;
2839 set_bit(R5_LOCKED, &dev->flags);
2840 set_bit(R5_Wantwrite, &dev->flags);
2841 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002842 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002843 dev = &sh->dev[pd_idx];
2844 s->locked++;
2845 set_bit(R5_LOCKED, &dev->flags);
2846 set_bit(R5_Wantwrite, &dev->flags);
2847 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002848 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002849 dev = &sh->dev[qd_idx];
2850 s->locked++;
2851 set_bit(R5_LOCKED, &dev->flags);
2852 set_bit(R5_Wantwrite, &dev->flags);
2853 }
2854 clear_bit(STRIPE_DEGRADED, &sh->state);
2855
2856 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002857 break;
2858 case check_state_run:
2859 case check_state_run_q:
2860 case check_state_run_pq:
2861 break; /* we will be called again upon completion */
2862 case check_state_check_result:
2863 sh->check_state = check_state_idle;
2864
2865 /* handle a successful check operation, if parity is correct
2866 * we are done. Otherwise update the mismatch count and repair
2867 * parity if !MD_RECOVERY_CHECK
2868 */
2869 if (sh->ops.zero_sum_result == 0) {
2870 /* both parities are correct */
2871 if (!s->failed)
2872 set_bit(STRIPE_INSYNC, &sh->state);
2873 else {
2874 /* in contrast to the raid5 case we can validate
2875 * parity, but still have a failure to write
2876 * back
2877 */
2878 sh->check_state = check_state_compute_result;
2879 /* Returning at this point means that we may go
2880 * off and bring p and/or q uptodate again so
2881 * we make sure to check zero_sum_result again
2882 * to verify if p or q need writeback
2883 */
2884 }
2885 } else {
2886 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2887 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2888 /* don't try to repair!! */
2889 set_bit(STRIPE_INSYNC, &sh->state);
2890 else {
2891 int *target = &sh->ops.target;
2892
2893 sh->ops.target = -1;
2894 sh->ops.target2 = -1;
2895 sh->check_state = check_state_compute_run;
2896 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2897 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2898 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
2899 set_bit(R5_Wantcompute,
2900 &sh->dev[pd_idx].flags);
2901 *target = pd_idx;
2902 target = &sh->ops.target2;
2903 s->uptodate++;
2904 }
2905 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
2906 set_bit(R5_Wantcompute,
2907 &sh->dev[qd_idx].flags);
2908 *target = qd_idx;
2909 s->uptodate++;
2910 }
2911 }
2912 }
2913 break;
2914 case check_state_compute_run:
2915 break;
2916 default:
2917 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2918 __func__, sh->check_state,
2919 (unsigned long long) sh->sector);
2920 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002921 }
2922}
2923
NeilBrown86c374b2011-07-27 11:00:36 +10002924static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh)
Dan Williamsa4456852007-07-09 11:56:43 -07002925{
2926 int i;
2927
2928 /* We have read all the blocks in this stripe and now we need to
2929 * copy some of them into a target stripe for expand.
2930 */
Dan Williamsf0a50d32007-01-02 13:52:31 -07002931 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07002932 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2933 for (i = 0; i < sh->disks; i++)
NeilBrown34e04e82009-03-31 15:10:16 +11002934 if (i != sh->pd_idx && i != sh->qd_idx) {
NeilBrown911d4ee2009-03-31 14:39:38 +11002935 int dd_idx, j;
Dan Williamsa4456852007-07-09 11:56:43 -07002936 struct stripe_head *sh2;
Dan Williamsa08abd82009-06-03 11:43:59 -07002937 struct async_submit_ctl submit;
Dan Williamsa4456852007-07-09 11:56:43 -07002938
NeilBrown784052e2009-03-31 15:19:07 +11002939 sector_t bn = compute_blocknr(sh, i, 1);
NeilBrown911d4ee2009-03-31 14:39:38 +11002940 sector_t s = raid5_compute_sector(conf, bn, 0,
2941 &dd_idx, NULL);
NeilBrowna8c906c2009-06-09 14:39:59 +10002942 sh2 = get_active_stripe(conf, s, 0, 1, 1);
Dan Williamsa4456852007-07-09 11:56:43 -07002943 if (sh2 == NULL)
2944 /* so far only the early blocks of this stripe
2945 * have been requested. When later blocks
2946 * get requested, we will try again
2947 */
2948 continue;
2949 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2950 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2951 /* must have already done this block */
2952 release_stripe(sh2);
2953 continue;
2954 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07002955
2956 /* place all the copies on one channel */
Dan Williamsa08abd82009-06-03 11:43:59 -07002957 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002958 tx = async_memcpy(sh2->dev[dd_idx].page,
Dan Williams88ba2aa2009-04-09 16:16:18 -07002959 sh->dev[i].page, 0, 0, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07002960 &submit);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002961
Dan Williamsa4456852007-07-09 11:56:43 -07002962 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2963 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2964 for (j = 0; j < conf->raid_disks; j++)
2965 if (j != sh2->pd_idx &&
NeilBrown86c374b2011-07-27 11:00:36 +10002966 j != sh2->qd_idx &&
Dan Williamsa4456852007-07-09 11:56:43 -07002967 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2968 break;
2969 if (j == conf->raid_disks) {
2970 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2971 set_bit(STRIPE_HANDLE, &sh2->state);
2972 }
2973 release_stripe(sh2);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002974
Dan Williamsa4456852007-07-09 11:56:43 -07002975 }
NeilBrowna2e08552007-09-11 15:23:36 -07002976 /* done submitting copies, wait for them to complete */
2977 if (tx) {
2978 async_tx_ack(tx);
2979 dma_wait_for_async_tx(tx);
2980 }
Dan Williamsa4456852007-07-09 11:56:43 -07002981}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982
Dan Williams6bfe0b42008-04-30 00:52:32 -07002983
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984/*
2985 * handle_stripe - do things to a stripe.
2986 *
2987 * We lock the stripe and then examine the state of various bits
2988 * to see what needs to be done.
2989 * Possible results:
2990 * return some read request which now have data
2991 * return some write requests which are safely on disc
2992 * schedule a read on some buffers
2993 * schedule a write of some buffers
2994 * return confirmation of parity correctness
2995 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 * buffers are taken off read_list or write_list, and bh_cache buffers
2997 * get BH_Lock set before the stripe lock is released.
2998 *
2999 */
Dan Williamsa4456852007-07-09 11:56:43 -07003000
NeilBrown3687c062011-07-27 11:00:36 +10003001static int handle_stripe5(struct stripe_head *sh, struct stripe_head_state *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002{
3003 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsa4456852007-07-09 11:56:43 -07003004 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 struct r5dev *dev;
Dan Williamse0a115e2008-06-05 22:45:52 -07003006 int prexor;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 /* Now to look around and see what can be done */
NeilBrown9910f162006-01-06 00:20:24 -08003009 rcu_read_lock();
NeilBrownc4c16632011-07-26 11:34:20 +10003010 spin_lock_irq(&conf->device_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 for (i=disks; i--; ) {
3012 mdk_rdev_t *rdev;
NeilBrowna9f326e2009-09-23 18:06:41 +10003013
3014 dev = &sh->dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015
Dan Williamsb5e98d62007-01-02 13:52:31 -07003016 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
3017 "written %p\n", i, dev->flags, dev->toread, dev->read,
3018 dev->towrite, dev->written);
3019
3020 /* maybe we can request a biofill operation
3021 *
3022 * new wantfill requests are only permitted while
Dan Williams83de75c2008-06-28 08:31:58 +10003023 * ops_complete_biofill is guaranteed to be inactive
Dan Williamsb5e98d62007-01-02 13:52:31 -07003024 */
3025 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
Dan Williams83de75c2008-06-28 08:31:58 +10003026 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
Dan Williamsb5e98d62007-01-02 13:52:31 -07003027 set_bit(R5_Wantfill, &dev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028
3029 /* now count some things */
NeilBrowncc940152011-07-26 11:35:35 +10003030 if (test_bit(R5_LOCKED, &dev->flags))
3031 s->locked++;
3032 if (test_bit(R5_UPTODATE, &dev->flags))
3033 s->uptodate++;
3034 if (test_bit(R5_Wantcompute, &dev->flags))
3035 s->compute++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036
Dan Williamsb5e98d62007-01-02 13:52:31 -07003037 if (test_bit(R5_Wantfill, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003038 s->to_fill++;
Dan Williamsb5e98d62007-01-02 13:52:31 -07003039 else if (dev->toread)
NeilBrowncc940152011-07-26 11:35:35 +10003040 s->to_read++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 if (dev->towrite) {
NeilBrowncc940152011-07-26 11:35:35 +10003042 s->to_write++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 if (!test_bit(R5_OVERWRITE, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003044 s->non_overwrite++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 }
Dan Williamsa4456852007-07-09 11:56:43 -07003046 if (dev->written)
NeilBrowncc940152011-07-26 11:35:35 +10003047 s->written++;
NeilBrown9910f162006-01-06 00:20:24 -08003048 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowncc940152011-07-26 11:35:35 +10003049 if (s->blocked_rdev == NULL &&
NeilBrownac4090d2008-08-05 15:54:13 +10003050 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
NeilBrowncc940152011-07-26 11:35:35 +10003051 s->blocked_rdev = rdev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003052 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003053 }
NeilBrown415e72d2010-06-17 17:25:21 +10003054 clear_bit(R5_Insync, &dev->flags);
3055 if (!rdev)
3056 /* Not in-sync */;
3057 else if (test_bit(In_sync, &rdev->flags))
3058 set_bit(R5_Insync, &dev->flags);
3059 else {
3060 /* could be in-sync depending on recovery/reshape status */
3061 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3062 set_bit(R5_Insync, &dev->flags);
3063 }
3064 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown14f8d262006-01-06 00:20:14 -08003065 /* The ReadError flag will just be confusing now */
NeilBrown4e5314b2005-11-08 21:39:22 -08003066 clear_bit(R5_ReadError, &dev->flags);
3067 clear_bit(R5_ReWrite, &dev->flags);
3068 }
NeilBrown415e72d2010-06-17 17:25:21 +10003069 if (test_bit(R5_ReadError, &dev->flags))
3070 clear_bit(R5_Insync, &dev->flags);
3071 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003072 s->failed++;
3073 s->failed_num[0] = i;
NeilBrown415e72d2010-06-17 17:25:21 +10003074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 }
NeilBrownc4c16632011-07-26 11:34:20 +10003076 spin_unlock_irq(&conf->device_lock);
NeilBrown9910f162006-01-06 00:20:24 -08003077 rcu_read_unlock();
Dan Williamsb5e98d62007-01-02 13:52:31 -07003078
NeilBrowncc940152011-07-26 11:35:35 +10003079 if (unlikely(s->blocked_rdev)) {
3080 if (s->syncing || s->expanding || s->expanded ||
3081 s->to_write || s->written) {
NeilBrownac4090d2008-08-05 15:54:13 +10003082 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrown3687c062011-07-27 11:00:36 +10003083 return 1;
NeilBrownac4090d2008-08-05 15:54:13 +10003084 }
3085 /* There is nothing for the blocked_rdev to block */
NeilBrowncc940152011-07-26 11:35:35 +10003086 rdev_dec_pending(s->blocked_rdev, conf->mddev);
3087 s->blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003088 }
3089
NeilBrowncc940152011-07-26 11:35:35 +10003090 if (s->to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3091 set_bit(STRIPE_OP_BIOFILL, &s->ops_request);
Dan Williams83de75c2008-06-28 08:31:58 +10003092 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3093 }
Dan Williamsb5e98d62007-01-02 13:52:31 -07003094
Dan Williams45b42332007-07-09 11:56:43 -07003095 pr_debug("locked=%d uptodate=%d to_read=%d"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 " to_write=%d failed=%d failed_num=%d\n",
NeilBrowncc940152011-07-26 11:35:35 +10003097 s->locked, s->uptodate, s->to_read, s->to_write,
3098 s->failed, s->failed_num[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 /* check if the array has lost two devices and, if so, some requests might
3100 * need to be failed
3101 */
NeilBrowncc940152011-07-26 11:35:35 +10003102 if (s->failed > 1 && s->to_read+s->to_write+s->written)
3103 handle_failed_stripe(conf, sh, s, disks, &s->return_bi);
3104 if (s->failed > 1 && s->syncing) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3106 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrowncc940152011-07-26 11:35:35 +10003107 s->syncing = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 }
3109
3110 /* might be able to return some write requests if the parity block
3111 * is safe, or on a failed drive
3112 */
3113 dev = &sh->dev[sh->pd_idx];
NeilBrowncc940152011-07-26 11:35:35 +10003114 if (s->written &&
3115 ((test_bit(R5_Insync, &dev->flags) &&
3116 !test_bit(R5_LOCKED, &dev->flags) &&
3117 test_bit(R5_UPTODATE, &dev->flags)) ||
3118 (s->failed == 1 && s->failed_num[0] == sh->pd_idx)))
3119 handle_stripe_clean_event(conf, sh, disks, &s->return_bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120
3121 /* Now we might consider reading some blocks, either to check/generate
3122 * parity, or to satisfy requests
3123 * or to load a block that is being partially written.
3124 */
NeilBrowncc940152011-07-26 11:35:35 +10003125 if (s->to_read || s->non_overwrite ||
3126 (s->syncing && (s->uptodate + s->compute < disks)) || s->expanding)
3127 handle_stripe_fill5(sh, s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128
Dan Williamse33129d2007-01-02 13:52:30 -07003129 /* Now we check to see if any write operations have recently
3130 * completed
3131 */
Dan Williamse0a115e2008-06-05 22:45:52 -07003132 prexor = 0;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003133 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
Dan Williamse0a115e2008-06-05 22:45:52 -07003134 prexor = 1;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003135 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3136 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
Dan Williams600aa102008-06-28 08:32:05 +10003137 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamse33129d2007-01-02 13:52:30 -07003138
3139 /* All the 'written' buffers and the parity block are ready to
3140 * be written back to disk
3141 */
3142 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3143 for (i = disks; i--; ) {
3144 dev = &sh->dev[i];
3145 if (test_bit(R5_LOCKED, &dev->flags) &&
3146 (i == sh->pd_idx || dev->written)) {
3147 pr_debug("Writing block %d\n", i);
3148 set_bit(R5_Wantwrite, &dev->flags);
Dan Williamse0a115e2008-06-05 22:45:52 -07003149 if (prexor)
3150 continue;
Dan Williamse33129d2007-01-02 13:52:30 -07003151 if (!test_bit(R5_Insync, &dev->flags) ||
NeilBrowncc940152011-07-26 11:35:35 +10003152 (i == sh->pd_idx && s->failed == 0))
Dan Williamse33129d2007-01-02 13:52:30 -07003153 set_bit(STRIPE_INSYNC, &sh->state);
3154 }
3155 }
NeilBrown729a1862009-12-14 12:49:50 +11003156 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
NeilBrowncc940152011-07-26 11:35:35 +10003157 s->dec_preread_active = 1;
Dan Williamse33129d2007-01-02 13:52:30 -07003158 }
3159
3160 /* Now to consider new write requests and what else, if anything
3161 * should be read. We do not handle new writes when:
3162 * 1/ A 'write' operation (copy+xor) is already in flight.
3163 * 2/ A 'check' operation is in flight, as it may clobber the parity
3164 * block.
3165 */
NeilBrowncc940152011-07-26 11:35:35 +10003166 if (s->to_write && !sh->reconstruct_state && !sh->check_state)
3167 handle_stripe_dirtying5(conf, sh, s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168
3169 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamse89f8962007-01-02 13:52:31 -07003170 * Any reads will already have been scheduled, so we just see if enough
3171 * data is available. The parity check is held off while parity
3172 * dependent operations are in flight.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 */
Dan Williamsecc65c92008-06-28 08:31:57 +10003174 if (sh->check_state ||
NeilBrowncc940152011-07-26 11:35:35 +10003175 (s->syncing && s->locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003176 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
Dan Williamsecc65c92008-06-28 08:31:57 +10003177 !test_bit(STRIPE_INSYNC, &sh->state)))
NeilBrowncc940152011-07-26 11:35:35 +10003178 handle_parity_checks5(conf, sh, s, disks);
NeilBrown3687c062011-07-27 11:00:36 +10003179 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180}
3181
NeilBrown3687c062011-07-27 11:00:36 +10003182static int handle_stripe6(struct stripe_head *sh, struct stripe_head_state *s)
NeilBrown16a53ec2006-06-26 00:27:38 -07003183{
NeilBrownbff61972009-03-31 14:33:13 +11003184 raid5_conf_t *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08003185 int disks = sh->disks;
NeilBrown34e04e82009-03-31 15:10:16 +11003186 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
NeilBrown16a53ec2006-06-26 00:27:38 -07003187 struct r5dev *dev, *pdev, *qdev;
NeilBrown16a53ec2006-06-26 00:27:38 -07003188
NeilBrown16a53ec2006-06-26 00:27:38 -07003189 /* Now to look around and see what can be done */
3190
3191 rcu_read_lock();
NeilBrownc4c16632011-07-26 11:34:20 +10003192 spin_lock_irq(&conf->device_lock);
NeilBrown16a53ec2006-06-26 00:27:38 -07003193 for (i=disks; i--; ) {
3194 mdk_rdev_t *rdev;
3195 dev = &sh->dev[i];
NeilBrown16a53ec2006-06-26 00:27:38 -07003196
Dan Williams45b42332007-07-09 11:56:43 -07003197 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown16a53ec2006-06-26 00:27:38 -07003198 i, dev->flags, dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003199 /* maybe we can reply to a read
3200 *
3201 * new wantfill requests are only permitted while
3202 * ops_complete_biofill is guaranteed to be inactive
3203 */
3204 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3205 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3206 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07003207
3208 /* now count some things */
NeilBrowncc940152011-07-26 11:35:35 +10003209 if (test_bit(R5_LOCKED, &dev->flags))
3210 s->locked++;
3211 if (test_bit(R5_UPTODATE, &dev->flags))
3212 s->uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003213 if (test_bit(R5_Wantcompute, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003214 s->compute++;
3215 BUG_ON(s->compute > 2);
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003216 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003217
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003218 if (test_bit(R5_Wantfill, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003219 s->to_fill++;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003220 } else if (dev->toread)
NeilBrowncc940152011-07-26 11:35:35 +10003221 s->to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003222 if (dev->towrite) {
NeilBrowncc940152011-07-26 11:35:35 +10003223 s->to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003224 if (!test_bit(R5_OVERWRITE, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003225 s->non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003226 }
Dan Williamsa4456852007-07-09 11:56:43 -07003227 if (dev->written)
NeilBrowncc940152011-07-26 11:35:35 +10003228 s->written++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003229 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowncc940152011-07-26 11:35:35 +10003230 if (s->blocked_rdev == NULL &&
NeilBrownac4090d2008-08-05 15:54:13 +10003231 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
NeilBrowncc940152011-07-26 11:35:35 +10003232 s->blocked_rdev = rdev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003233 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003234 }
NeilBrown415e72d2010-06-17 17:25:21 +10003235 clear_bit(R5_Insync, &dev->flags);
3236 if (!rdev)
3237 /* Not in-sync */;
3238 else if (test_bit(In_sync, &rdev->flags))
3239 set_bit(R5_Insync, &dev->flags);
3240 else {
3241 /* in sync if before recovery_offset */
3242 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3243 set_bit(R5_Insync, &dev->flags);
3244 }
3245 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003246 /* The ReadError flag will just be confusing now */
3247 clear_bit(R5_ReadError, &dev->flags);
3248 clear_bit(R5_ReWrite, &dev->flags);
3249 }
NeilBrown415e72d2010-06-17 17:25:21 +10003250 if (test_bit(R5_ReadError, &dev->flags))
3251 clear_bit(R5_Insync, &dev->flags);
3252 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003253 if (s->failed < 2)
3254 s->failed_num[s->failed] = i;
3255 s->failed++;
NeilBrown415e72d2010-06-17 17:25:21 +10003256 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003257 }
NeilBrownc4c16632011-07-26 11:34:20 +10003258 spin_unlock_irq(&conf->device_lock);
NeilBrown16a53ec2006-06-26 00:27:38 -07003259 rcu_read_unlock();
Dan Williams6bfe0b42008-04-30 00:52:32 -07003260
NeilBrowncc940152011-07-26 11:35:35 +10003261 if (unlikely(s->blocked_rdev)) {
3262 if (s->syncing || s->expanding || s->expanded ||
3263 s->to_write || s->written) {
NeilBrownac4090d2008-08-05 15:54:13 +10003264 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrown3687c062011-07-27 11:00:36 +10003265 return 1;
NeilBrownac4090d2008-08-05 15:54:13 +10003266 }
3267 /* There is nothing for the blocked_rdev to block */
NeilBrowncc940152011-07-26 11:35:35 +10003268 rdev_dec_pending(s->blocked_rdev, conf->mddev);
3269 s->blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003270 }
NeilBrownac4090d2008-08-05 15:54:13 +10003271
NeilBrowncc940152011-07-26 11:35:35 +10003272 if (s->to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3273 set_bit(STRIPE_OP_BIOFILL, &s->ops_request);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003274 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3275 }
3276
Dan Williams45b42332007-07-09 11:56:43 -07003277 pr_debug("locked=%d uptodate=%d to_read=%d"
NeilBrown16a53ec2006-06-26 00:27:38 -07003278 " to_write=%d failed=%d failed_num=%d,%d\n",
NeilBrowncc940152011-07-26 11:35:35 +10003279 s->locked, s->uptodate, s->to_read, s->to_write, s->failed,
3280 s->failed_num[0], s->failed_num[1]);
Dan Williamsa4456852007-07-09 11:56:43 -07003281 /* check if the array has lost >2 devices and, if so, some requests
3282 * might need to be failed
NeilBrown16a53ec2006-06-26 00:27:38 -07003283 */
NeilBrowncc940152011-07-26 11:35:35 +10003284 if (s->failed > 2 && s->to_read+s->to_write+s->written)
3285 handle_failed_stripe(conf, sh, s, disks, &s->return_bi);
3286 if (s->failed > 2 && s->syncing) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003287 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3288 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrowncc940152011-07-26 11:35:35 +10003289 s->syncing = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003290 }
3291
3292 /*
3293 * might be able to return some write requests if the parity blocks
3294 * are safe, or on a failed drive
3295 */
3296 pdev = &sh->dev[pd_idx];
NeilBrowncc940152011-07-26 11:35:35 +10003297 s->p_failed = (s->failed >= 1 && s->failed_num[0] == pd_idx)
3298 || (s->failed >= 2 && s->failed_num[1] == pd_idx);
NeilBrown34e04e82009-03-31 15:10:16 +11003299 qdev = &sh->dev[qd_idx];
NeilBrowncc940152011-07-26 11:35:35 +10003300 s->q_failed = (s->failed >= 1 && s->failed_num[0] == qd_idx)
3301 || (s->failed >= 2 && s->failed_num[1] == qd_idx);
NeilBrown16a53ec2006-06-26 00:27:38 -07003302
NeilBrowncc940152011-07-26 11:35:35 +10003303 if (s->written &&
3304 (s->p_failed || ((test_bit(R5_Insync, &pdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003305 && !test_bit(R5_LOCKED, &pdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003306 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
NeilBrowncc940152011-07-26 11:35:35 +10003307 (s->q_failed || ((test_bit(R5_Insync, &qdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003308 && !test_bit(R5_LOCKED, &qdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003309 && test_bit(R5_UPTODATE, &qdev->flags)))))
NeilBrowncc940152011-07-26 11:35:35 +10003310 handle_stripe_clean_event(conf, sh, disks, &s->return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003311
3312 /* Now we might consider reading some blocks, either to check/generate
3313 * parity, or to satisfy requests
3314 * or to load a block that is being partially written.
3315 */
NeilBrowncc940152011-07-26 11:35:35 +10003316 if (s->to_read || s->non_overwrite || (s->to_write && s->failed) ||
3317 (s->syncing && (s->uptodate + s->compute < disks)) || s->expanding)
3318 handle_stripe_fill6(sh, s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003319
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003320 /* Now we check to see if any write operations have recently
3321 * completed
3322 */
3323 if (sh->reconstruct_state == reconstruct_state_drain_result) {
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003324
3325 sh->reconstruct_state = reconstruct_state_idle;
3326 /* All the 'written' buffers and the parity blocks are ready to
3327 * be written back to disk
3328 */
3329 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3330 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags));
3331 for (i = disks; i--; ) {
3332 dev = &sh->dev[i];
3333 if (test_bit(R5_LOCKED, &dev->flags) &&
3334 (i == sh->pd_idx || i == qd_idx ||
3335 dev->written)) {
3336 pr_debug("Writing block %d\n", i);
3337 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3338 set_bit(R5_Wantwrite, &dev->flags);
3339 if (!test_bit(R5_Insync, &dev->flags) ||
3340 ((i == sh->pd_idx || i == qd_idx) &&
NeilBrowncc940152011-07-26 11:35:35 +10003341 s->failed == 0))
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003342 set_bit(STRIPE_INSYNC, &sh->state);
3343 }
3344 }
NeilBrown729a1862009-12-14 12:49:50 +11003345 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
NeilBrowncc940152011-07-26 11:35:35 +10003346 s->dec_preread_active = 1;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003347 }
3348
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07003349 /* Now to consider new write requests and what else, if anything
3350 * should be read. We do not handle new writes when:
3351 * 1/ A 'write' operation (copy+gen_syndrome) is already in flight.
3352 * 2/ A 'check' operation is in flight, as it may clobber the parity
3353 * block.
3354 */
NeilBrowncc940152011-07-26 11:35:35 +10003355 if (s->to_write && !sh->reconstruct_state && !sh->check_state)
3356 handle_stripe_dirtying6(conf, sh, s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003357
3358 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamsa4456852007-07-09 11:56:43 -07003359 * Any reads will already have been scheduled, so we just see if enough
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003360 * data is available. The parity check is held off while parity
3361 * dependent operations are in flight.
NeilBrown16a53ec2006-06-26 00:27:38 -07003362 */
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003363 if (sh->check_state ||
NeilBrowncc940152011-07-26 11:35:35 +10003364 (s->syncing && s->locked == 0 &&
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003365 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3366 !test_bit(STRIPE_INSYNC, &sh->state)))
NeilBrowncc940152011-07-26 11:35:35 +10003367 handle_parity_checks6(conf, sh, s, disks);
NeilBrown3687c062011-07-27 11:00:36 +10003368 return 0;
NeilBrowncc940152011-07-26 11:35:35 +10003369}
NeilBrownf4168852007-02-28 20:11:53 -08003370
NeilBrowncc940152011-07-26 11:35:35 +10003371static void handle_stripe(struct stripe_head *sh)
3372{
3373 struct stripe_head_state s;
NeilBrown3687c062011-07-27 11:00:36 +10003374 int done;
3375 int i;
NeilBrowncc940152011-07-26 11:35:35 +10003376 raid5_conf_t *conf = sh->raid_conf;
3377
3378 clear_bit(STRIPE_HANDLE, &sh->state);
3379 if (test_and_set_bit(STRIPE_ACTIVE, &sh->state)) {
3380 /* already being handled, ensure it gets handled
3381 * again when current action finishes */
3382 set_bit(STRIPE_HANDLE, &sh->state);
3383 return;
3384 }
3385
3386 if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3387 set_bit(STRIPE_SYNCING, &sh->state);
3388 clear_bit(STRIPE_INSYNC, &sh->state);
3389 }
3390 clear_bit(STRIPE_DELAYED, &sh->state);
3391
3392 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3393 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3394 (unsigned long long)sh->sector, sh->state,
3395 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3396 sh->check_state, sh->reconstruct_state);
3397 memset(&s, 0, sizeof(s));
3398
3399 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3400 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3401 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3402
3403 if (conf->level == 6)
NeilBrown3687c062011-07-27 11:00:36 +10003404 done = handle_stripe6(sh, &s);
NeilBrowncc940152011-07-26 11:35:35 +10003405 else
NeilBrown3687c062011-07-27 11:00:36 +10003406 done = handle_stripe5(sh, &s);
NeilBrowncc940152011-07-26 11:35:35 +10003407
NeilBrown3687c062011-07-27 11:00:36 +10003408 if (done)
3409 goto finish;
NeilBrownc5a31002011-07-27 11:00:36 +10003410
3411
3412 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
3413 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3414 clear_bit(STRIPE_SYNCING, &sh->state);
3415 }
3416
3417 /* If the failed drives are just a ReadError, then we might need
3418 * to progress the repair/check process
3419 */
3420 if (s.failed <= conf->max_degraded && !conf->mddev->ro)
3421 for (i = 0; i < s.failed; i++) {
3422 struct r5dev *dev = &sh->dev[s.failed_num[i]];
3423 if (test_bit(R5_ReadError, &dev->flags)
3424 && !test_bit(R5_LOCKED, &dev->flags)
3425 && test_bit(R5_UPTODATE, &dev->flags)
3426 ) {
3427 if (!test_bit(R5_ReWrite, &dev->flags)) {
3428 set_bit(R5_Wantwrite, &dev->flags);
3429 set_bit(R5_ReWrite, &dev->flags);
3430 set_bit(R5_LOCKED, &dev->flags);
3431 s.locked++;
3432 } else {
3433 /* let's read it back */
3434 set_bit(R5_Wantread, &dev->flags);
3435 set_bit(R5_LOCKED, &dev->flags);
3436 s.locked++;
3437 }
3438 }
3439 }
3440
3441
NeilBrown3687c062011-07-27 11:00:36 +10003442 /* Finish reconstruct operations initiated by the expansion process */
3443 if (sh->reconstruct_state == reconstruct_state_result) {
3444 struct stripe_head *sh_src
3445 = get_active_stripe(conf, sh->sector, 1, 1, 1);
3446 if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) {
3447 /* sh cannot be written until sh_src has been read.
3448 * so arrange for sh to be delayed a little
3449 */
3450 set_bit(STRIPE_DELAYED, &sh->state);
3451 set_bit(STRIPE_HANDLE, &sh->state);
3452 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3453 &sh_src->state))
3454 atomic_inc(&conf->preread_active_stripes);
3455 release_stripe(sh_src);
3456 goto finish;
3457 }
3458 if (sh_src)
3459 release_stripe(sh_src);
NeilBrown16a53ec2006-06-26 00:27:38 -07003460
NeilBrown3687c062011-07-27 11:00:36 +10003461 sh->reconstruct_state = reconstruct_state_idle;
3462 clear_bit(STRIPE_EXPANDING, &sh->state);
3463 for (i = conf->raid_disks; i--; ) {
3464 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3465 set_bit(R5_LOCKED, &sh->dev[i].flags);
3466 s.locked++;
3467 }
3468 }
3469
3470 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3471 !sh->reconstruct_state) {
3472 /* Need to write out all blocks after computing parity */
3473 sh->disks = conf->raid_disks;
3474 stripe_set_idx(sh->sector, conf, 0, sh);
3475 schedule_reconstruction(sh, &s, 1, 1);
3476 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
3477 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3478 atomic_dec(&conf->reshape_stripes);
3479 wake_up(&conf->wait_for_overlap);
3480 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3481 }
3482
3483 if (s.expanding && s.locked == 0 &&
3484 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
3485 handle_stripe_expansion(conf, sh);
3486
3487finish:
Dan Williams6bfe0b42008-04-30 00:52:32 -07003488 /* wait for this device to become unblocked */
NeilBrownc5709ef2011-07-26 11:35:20 +10003489 if (unlikely(s.blocked_rdev))
3490 md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003491
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003492 if (s.ops_request)
3493 raid_run_ops(sh, s.ops_request);
3494
Dan Williamsf0e43bc2008-06-28 08:31:55 +10003495 ops_run_io(sh, &s);
3496
NeilBrown729a1862009-12-14 12:49:50 +11003497
NeilBrownc5709ef2011-07-26 11:35:20 +10003498 if (s.dec_preread_active) {
NeilBrown729a1862009-12-14 12:49:50 +11003499 /* We delay this until after ops_run_io so that if make_request
Tejun Heoe9c74692010-09-03 11:56:18 +02003500 * is waiting on a flush, it won't continue until the writes
NeilBrown729a1862009-12-14 12:49:50 +11003501 * have actually been submitted.
3502 */
3503 atomic_dec(&conf->preread_active_stripes);
3504 if (atomic_read(&conf->preread_active_stripes) <
3505 IO_THRESHOLD)
3506 md_wakeup_thread(conf->mddev->thread);
3507 }
3508
NeilBrownc5709ef2011-07-26 11:35:20 +10003509 return_io(s.return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003510
NeilBrownc4c16632011-07-26 11:34:20 +10003511 clear_bit(STRIPE_ACTIVE, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07003512}
3513
Arjan van de Ven858119e2006-01-14 13:20:43 -08003514static void raid5_activate_delayed(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515{
3516 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3517 while (!list_empty(&conf->delayed_list)) {
3518 struct list_head *l = conf->delayed_list.next;
3519 struct stripe_head *sh;
3520 sh = list_entry(l, struct stripe_head, lru);
3521 list_del_init(l);
3522 clear_bit(STRIPE_DELAYED, &sh->state);
3523 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3524 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003525 list_add_tail(&sh->lru, &conf->hold_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526 }
NeilBrown482c0832011-04-18 18:25:42 +10003527 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528}
3529
Arjan van de Ven858119e2006-01-14 13:20:43 -08003530static void activate_bit_delay(raid5_conf_t *conf)
NeilBrown72626682005-09-09 16:23:54 -07003531{
3532 /* device_lock is held */
3533 struct list_head head;
3534 list_add(&head, &conf->bitmap_list);
3535 list_del_init(&conf->bitmap_list);
3536 while (!list_empty(&head)) {
3537 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3538 list_del_init(&sh->lru);
3539 atomic_inc(&sh->count);
3540 __release_stripe(conf, sh);
3541 }
3542}
3543
NeilBrown11d8a6e2010-07-26 11:57:07 +10003544int md_raid5_congested(mddev_t *mddev, int bits)
NeilBrownf022b2f2006-10-03 01:15:56 -07003545{
NeilBrown070ec552009-06-16 16:54:21 +10003546 raid5_conf_t *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07003547
3548 /* No difference between reads and writes. Just check
3549 * how busy the stripe_cache is
3550 */
NeilBrown3fa841d2009-09-23 18:10:29 +10003551
NeilBrownf022b2f2006-10-03 01:15:56 -07003552 if (conf->inactive_blocked)
3553 return 1;
3554 if (conf->quiesce)
3555 return 1;
3556 if (list_empty_careful(&conf->inactive_list))
3557 return 1;
3558
3559 return 0;
3560}
NeilBrown11d8a6e2010-07-26 11:57:07 +10003561EXPORT_SYMBOL_GPL(md_raid5_congested);
3562
3563static int raid5_congested(void *data, int bits)
3564{
3565 mddev_t *mddev = data;
3566
3567 return mddev_congested(mddev, bits) ||
3568 md_raid5_congested(mddev, bits);
3569}
NeilBrownf022b2f2006-10-03 01:15:56 -07003570
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003571/* We want read requests to align with chunks where possible,
3572 * but write requests don't need to.
3573 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003574static int raid5_mergeable_bvec(struct request_queue *q,
3575 struct bvec_merge_data *bvm,
3576 struct bio_vec *biovec)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003577{
3578 mddev_t *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003579 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003580 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +10003581 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003582 unsigned int bio_sectors = bvm->bi_size >> 9;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003583
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003584 if ((bvm->bi_rw & 1) == WRITE)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003585 return biovec->bv_len; /* always allow writes to be mergeable */
3586
Andre Noll664e7c42009-06-18 08:45:27 +10003587 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3588 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003589 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3590 if (max < 0) max = 0;
3591 if (max <= biovec->bv_len && bio_sectors == 0)
3592 return biovec->bv_len;
3593 else
3594 return max;
3595}
3596
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003597
3598static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3599{
3600 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
Andre Noll9d8f0362009-06-18 08:45:01 +10003601 unsigned int chunk_sectors = mddev->chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003602 unsigned int bio_sectors = bio->bi_size >> 9;
3603
Andre Noll664e7c42009-06-18 08:45:27 +10003604 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3605 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003606 return chunk_sectors >=
3607 ((sector & (chunk_sectors - 1)) + bio_sectors);
3608}
3609
3610/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003611 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3612 * later sampled by raid5d.
3613 */
3614static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3615{
3616 unsigned long flags;
3617
3618 spin_lock_irqsave(&conf->device_lock, flags);
3619
3620 bi->bi_next = conf->retry_read_aligned_list;
3621 conf->retry_read_aligned_list = bi;
3622
3623 spin_unlock_irqrestore(&conf->device_lock, flags);
3624 md_wakeup_thread(conf->mddev->thread);
3625}
3626
3627
3628static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3629{
3630 struct bio *bi;
3631
3632 bi = conf->retry_read_aligned;
3633 if (bi) {
3634 conf->retry_read_aligned = NULL;
3635 return bi;
3636 }
3637 bi = conf->retry_read_aligned_list;
3638 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08003639 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003640 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02003641 /*
3642 * this sets the active strip count to 1 and the processed
3643 * strip count to zero (upper 8 bits)
3644 */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003645 bi->bi_phys_segments = 1; /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003646 }
3647
3648 return bi;
3649}
3650
3651
3652/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003653 * The "raid5_align_endio" should check if the read succeeded and if it
3654 * did, call bio_endio on the original bio (having bio_put the new bio
3655 * first).
3656 * If the read failed..
3657 */
NeilBrown6712ecf2007-09-27 12:47:43 +02003658static void raid5_align_endio(struct bio *bi, int error)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003659{
3660 struct bio* raid_bi = bi->bi_private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003661 mddev_t *mddev;
3662 raid5_conf_t *conf;
3663 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3664 mdk_rdev_t *rdev;
3665
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003666 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003667
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003668 rdev = (void*)raid_bi->bi_next;
3669 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11003670 mddev = rdev->mddev;
3671 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003672
3673 rdev_dec_pending(rdev, conf->mddev);
3674
3675 if (!error && uptodate) {
NeilBrown6712ecf2007-09-27 12:47:43 +02003676 bio_endio(raid_bi, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003677 if (atomic_dec_and_test(&conf->active_aligned_reads))
3678 wake_up(&conf->wait_for_stripe);
NeilBrown6712ecf2007-09-27 12:47:43 +02003679 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003680 }
3681
3682
Dan Williams45b42332007-07-09 11:56:43 -07003683 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003684
3685 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003686}
3687
Neil Brown387bb172007-02-08 14:20:29 -08003688static int bio_fits_rdev(struct bio *bi)
3689{
Jens Axboe165125e2007-07-24 09:28:11 +02003690 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
Neil Brown387bb172007-02-08 14:20:29 -08003691
Martin K. Petersenae03bf62009-05-22 17:17:50 -04003692 if ((bi->bi_size>>9) > queue_max_sectors(q))
Neil Brown387bb172007-02-08 14:20:29 -08003693 return 0;
3694 blk_recount_segments(q, bi);
Martin K. Petersen8a783622010-02-26 00:20:39 -05003695 if (bi->bi_phys_segments > queue_max_segments(q))
Neil Brown387bb172007-02-08 14:20:29 -08003696 return 0;
3697
3698 if (q->merge_bvec_fn)
3699 /* it's too hard to apply the merge_bvec_fn at this stage,
3700 * just just give up
3701 */
3702 return 0;
3703
3704 return 1;
3705}
3706
3707
NeilBrown21a52c62010-04-01 15:02:13 +11003708static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003709{
NeilBrown070ec552009-06-16 16:54:21 +10003710 raid5_conf_t *conf = mddev->private;
NeilBrown8553fe72009-12-14 12:49:47 +11003711 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003712 struct bio* align_bi;
3713 mdk_rdev_t *rdev;
3714
3715 if (!in_chunk_boundary(mddev, raid_bio)) {
Dan Williams45b42332007-07-09 11:56:43 -07003716 pr_debug("chunk_aligned_read : non aligned\n");
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003717 return 0;
3718 }
3719 /*
NeilBrowna167f662010-10-26 18:31:13 +11003720 * use bio_clone_mddev to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003721 */
NeilBrowna167f662010-10-26 18:31:13 +11003722 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003723 if (!align_bi)
3724 return 0;
3725 /*
3726 * set bi_end_io to a new function, and set bi_private to the
3727 * original bio.
3728 */
3729 align_bi->bi_end_io = raid5_align_endio;
3730 align_bi->bi_private = raid_bio;
3731 /*
3732 * compute position
3733 */
NeilBrown112bf892009-03-31 14:39:38 +11003734 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3735 0,
NeilBrown911d4ee2009-03-31 14:39:38 +11003736 &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003737
3738 rcu_read_lock();
3739 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3740 if (rdev && test_bit(In_sync, &rdev->flags)) {
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003741 atomic_inc(&rdev->nr_pending);
3742 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003743 raid_bio->bi_next = (void*)rdev;
3744 align_bi->bi_bdev = rdev->bdev;
3745 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3746 align_bi->bi_sector += rdev->data_offset;
3747
Neil Brown387bb172007-02-08 14:20:29 -08003748 if (!bio_fits_rdev(align_bi)) {
3749 /* too big in some way */
3750 bio_put(align_bi);
3751 rdev_dec_pending(rdev, mddev);
3752 return 0;
3753 }
3754
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003755 spin_lock_irq(&conf->device_lock);
3756 wait_event_lock_irq(conf->wait_for_stripe,
3757 conf->quiesce == 0,
3758 conf->device_lock, /* nothing */);
3759 atomic_inc(&conf->active_aligned_reads);
3760 spin_unlock_irq(&conf->device_lock);
3761
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003762 generic_make_request(align_bi);
3763 return 1;
3764 } else {
3765 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003766 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003767 return 0;
3768 }
3769}
3770
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003771/* __get_priority_stripe - get the next stripe to process
3772 *
3773 * Full stripe writes are allowed to pass preread active stripes up until
3774 * the bypass_threshold is exceeded. In general the bypass_count
3775 * increments when the handle_list is handled before the hold_list; however, it
3776 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3777 * stripe with in flight i/o. The bypass_count will be reset when the
3778 * head of the hold_list has changed, i.e. the head was promoted to the
3779 * handle_list.
3780 */
3781static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3782{
3783 struct stripe_head *sh;
3784
3785 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3786 __func__,
3787 list_empty(&conf->handle_list) ? "empty" : "busy",
3788 list_empty(&conf->hold_list) ? "empty" : "busy",
3789 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3790
3791 if (!list_empty(&conf->handle_list)) {
3792 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3793
3794 if (list_empty(&conf->hold_list))
3795 conf->bypass_count = 0;
3796 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3797 if (conf->hold_list.next == conf->last_hold)
3798 conf->bypass_count++;
3799 else {
3800 conf->last_hold = conf->hold_list.next;
3801 conf->bypass_count -= conf->bypass_threshold;
3802 if (conf->bypass_count < 0)
3803 conf->bypass_count = 0;
3804 }
3805 }
3806 } else if (!list_empty(&conf->hold_list) &&
3807 ((conf->bypass_threshold &&
3808 conf->bypass_count > conf->bypass_threshold) ||
3809 atomic_read(&conf->pending_full_writes) == 0)) {
3810 sh = list_entry(conf->hold_list.next,
3811 typeof(*sh), lru);
3812 conf->bypass_count -= conf->bypass_threshold;
3813 if (conf->bypass_count < 0)
3814 conf->bypass_count = 0;
3815 } else
3816 return NULL;
3817
3818 list_del_init(&sh->lru);
3819 atomic_inc(&sh->count);
3820 BUG_ON(atomic_read(&sh->count) != 1);
3821 return sh;
3822}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003823
NeilBrown21a52c62010-04-01 15:02:13 +11003824static int make_request(mddev_t *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825{
NeilBrown070ec552009-06-16 16:54:21 +10003826 raid5_conf_t *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11003827 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 sector_t new_sector;
3829 sector_t logical_sector, last_sector;
3830 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01003831 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11003832 int remaining;
NeilBrown7c13edc2011-04-18 18:25:43 +10003833 int plugged;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003834
Tejun Heoe9c74692010-09-03 11:56:18 +02003835 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
3836 md_flush_request(mddev, bi);
NeilBrowne5dcdd82005-09-09 16:23:41 -07003837 return 0;
3838 }
3839
NeilBrown3d310eb2005-06-21 17:17:26 -07003840 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07003841
NeilBrown802ba062006-12-13 00:34:13 -08003842 if (rw == READ &&
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003843 mddev->reshape_position == MaxSector &&
NeilBrown21a52c62010-04-01 15:02:13 +11003844 chunk_aligned_read(mddev,bi))
NeilBrown99c0fb52009-03-31 14:39:38 +11003845 return 0;
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003846
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3848 last_sector = bi->bi_sector + (bi->bi_size>>9);
3849 bi->bi_next = NULL;
3850 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07003851
NeilBrown7c13edc2011-04-18 18:25:43 +10003852 plugged = mddev_check_plugged(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3854 DEFINE_WAIT(w);
NeilBrown16a53ec2006-06-26 00:27:38 -07003855 int disks, data_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003856 int previous;
NeilBrownb578d552006-03-27 01:18:12 -08003857
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003858 retry:
NeilBrownb5663ba2009-03-31 14:39:38 +11003859 previous = 0;
NeilBrownb0f9ec02009-03-31 15:27:18 +11003860 disks = conf->raid_disks;
NeilBrownb578d552006-03-27 01:18:12 -08003861 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003862 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11003863 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08003864 * 64bit on a 32bit platform, and so it might be
3865 * possible to see a half-updated value
Jesper Juhlaeb878b2011-04-10 18:06:17 +02003866 * Of course reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08003867 * the lock is dropped, so once we get a reference
3868 * to the stripe that we think it is, we will have
3869 * to check again.
3870 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003871 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11003872 if (mddev->delta_disks < 0
3873 ? logical_sector < conf->reshape_progress
3874 : logical_sector >= conf->reshape_progress) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003875 disks = conf->previous_raid_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003876 previous = 1;
3877 } else {
NeilBrownfef9c612009-03-31 15:16:46 +11003878 if (mddev->delta_disks < 0
3879 ? logical_sector < conf->reshape_safe
3880 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08003881 spin_unlock_irq(&conf->device_lock);
3882 schedule();
3883 goto retry;
3884 }
3885 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003886 spin_unlock_irq(&conf->device_lock);
3887 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003888 data_disks = disks - conf->max_degraded;
3889
NeilBrown112bf892009-03-31 14:39:38 +11003890 new_sector = raid5_compute_sector(conf, logical_sector,
3891 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11003892 &dd_idx, NULL);
NeilBrown0c55e022010-05-03 14:09:02 +10003893 pr_debug("raid456: make_request, sector %llu logical %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894 (unsigned long long)new_sector,
3895 (unsigned long long)logical_sector);
3896
NeilBrownb5663ba2009-03-31 14:39:38 +11003897 sh = get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10003898 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11003900 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003901 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08003902 * stripe, so we must do the range check again.
3903 * Expansion could still move past after this
3904 * test, but as we are holding a reference to
3905 * 'sh', we know that if that happens,
3906 * STRIPE_EXPANDING will get set and the expansion
3907 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003908 */
3909 int must_retry = 0;
3910 spin_lock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003911 if (mddev->delta_disks < 0
3912 ? logical_sector >= conf->reshape_progress
3913 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003914 /* mismatch, need to try again */
3915 must_retry = 1;
3916 spin_unlock_irq(&conf->device_lock);
3917 if (must_retry) {
3918 release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07003919 schedule();
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003920 goto retry;
3921 }
3922 }
NeilBrowne62e58a2009-07-01 13:15:35 +10003923
Namhyung Kimffd96e32011-07-18 17:38:51 +10003924 if (rw == WRITE &&
NeilBrowna5c308d2009-07-01 13:15:35 +10003925 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08003926 logical_sector < mddev->suspend_hi) {
3927 release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10003928 /* As the suspend_* range is controlled by
3929 * userspace, we want an interruptible
3930 * wait.
3931 */
3932 flush_signals(current);
3933 prepare_to_wait(&conf->wait_for_overlap,
3934 &w, TASK_INTERRUPTIBLE);
3935 if (logical_sector >= mddev->suspend_lo &&
3936 logical_sector < mddev->suspend_hi)
3937 schedule();
NeilBrowne464eaf2006-03-27 01:18:14 -08003938 goto retry;
3939 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003940
3941 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
Namhyung Kimffd96e32011-07-18 17:38:51 +10003942 !add_stripe_bio(sh, bi, dd_idx, rw)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003943 /* Stripe is busy expanding or
3944 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945 * and wait a while
3946 */
NeilBrown482c0832011-04-18 18:25:42 +10003947 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 release_stripe(sh);
3949 schedule();
3950 goto retry;
3951 }
3952 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown6ed30032008-02-06 01:40:00 -08003953 set_bit(STRIPE_HANDLE, &sh->state);
3954 clear_bit(STRIPE_DELAYED, &sh->state);
Tejun Heoe9c74692010-09-03 11:56:18 +02003955 if ((bi->bi_rw & REQ_SYNC) &&
NeilBrown729a1862009-12-14 12:49:50 +11003956 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3957 atomic_inc(&conf->preread_active_stripes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 } else {
3960 /* cannot get stripe for read-ahead, just give-up */
3961 clear_bit(BIO_UPTODATE, &bi->bi_flags);
3962 finish_wait(&conf->wait_for_overlap, &w);
3963 break;
3964 }
3965
3966 }
NeilBrown7c13edc2011-04-18 18:25:43 +10003967 if (!plugged)
3968 md_wakeup_thread(mddev->thread);
3969
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02003971 remaining = raid5_dec_bi_phys_segments(bi);
NeilBrownf6344752006-03-27 01:18:17 -08003972 spin_unlock_irq(&conf->device_lock);
3973 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974
NeilBrown16a53ec2006-06-26 00:27:38 -07003975 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02003977
Neil Brown0e13fe232008-06-28 08:31:20 +10003978 bio_endio(bi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 }
NeilBrown729a1862009-12-14 12:49:50 +11003980
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981 return 0;
3982}
3983
Dan Williamsb522adc2009-03-31 15:00:31 +11003984static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
3985
NeilBrown52c03292006-06-26 00:27:43 -07003986static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987{
NeilBrown52c03292006-06-26 00:27:43 -07003988 /* reshaping is quite different to recovery/resync so it is
3989 * handled quite separately ... here.
3990 *
3991 * On each call to sync_request, we gather one chunk worth of
3992 * destination stripes and flag them as expanding.
3993 * Then we find all the source stripes and request reads.
3994 * As the reads complete, handle_stripe will copy the data
3995 * into the destination stripe and release that stripe.
3996 */
H Hartley Sweeten7b928132010-03-08 16:02:40 +11003997 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08003999 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08004000 int raid_disks = conf->previous_raid_disks;
4001 int data_disks = raid_disks - conf->max_degraded;
4002 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07004003 int i;
4004 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11004005 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11004006 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11004007 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11004008 struct list_head stripes;
NeilBrown52c03292006-06-26 00:27:43 -07004009
NeilBrownfef9c612009-03-31 15:16:46 +11004010 if (sector_nr == 0) {
4011 /* If restarting in the middle, skip the initial sectors */
4012 if (mddev->delta_disks < 0 &&
4013 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4014 sector_nr = raid5_size(mddev, 0, 0)
4015 - conf->reshape_progress;
NeilBrowna6397552009-08-13 10:13:00 +10004016 } else if (mddev->delta_disks >= 0 &&
NeilBrownfef9c612009-03-31 15:16:46 +11004017 conf->reshape_progress > 0)
4018 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004019 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004020 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11004021 mddev->curr_resync_completed = sector_nr;
4022 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11004023 *skipped = 1;
4024 return sector_nr;
4025 }
NeilBrown52c03292006-06-26 00:27:43 -07004026 }
4027
NeilBrown7a661382009-03-31 15:21:40 +11004028 /* We need to process a full chunk at a time.
4029 * If old and new chunk sizes differ, we need to process the
4030 * largest of these
4031 */
Andre Noll664e7c42009-06-18 08:45:27 +10004032 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4033 reshape_sectors = mddev->new_chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004034 else
Andre Noll9d8f0362009-06-18 08:45:01 +10004035 reshape_sectors = mddev->chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004036
NeilBrown52c03292006-06-26 00:27:43 -07004037 /* we update the metadata when there is more than 3Meg
4038 * in the block range (that is rather arbitrary, should
4039 * probably be time based) or when the data about to be
4040 * copied would over-write the source of the data at
4041 * the front of the range.
NeilBrownfef9c612009-03-31 15:16:46 +11004042 * i.e. one new_stripe along from reshape_progress new_maps
4043 * to after where reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07004044 */
NeilBrownfef9c612009-03-31 15:16:46 +11004045 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004046 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11004047 readpos = conf->reshape_progress;
4048 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004049 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08004050 sector_div(safepos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004051 if (mddev->delta_disks < 0) {
NeilBrowned37d832009-05-27 21:39:05 +10004052 writepos -= min_t(sector_t, reshape_sectors, writepos);
NeilBrownc8f517c2009-03-31 15:28:40 +11004053 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004054 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11004055 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004056 writepos += reshape_sectors;
NeilBrowned37d832009-05-27 21:39:05 +10004057 readpos -= min_t(sector_t, reshape_sectors, readpos);
4058 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11004059 }
NeilBrown52c03292006-06-26 00:27:43 -07004060
NeilBrownc8f517c2009-03-31 15:28:40 +11004061 /* 'writepos' is the most advanced device address we might write.
4062 * 'readpos' is the least advanced device address we might read.
4063 * 'safepos' is the least address recorded in the metadata as having
4064 * been reshaped.
4065 * If 'readpos' is behind 'writepos', then there is no way that we can
4066 * ensure safety in the face of a crash - that must be done by userspace
4067 * making a backup of the data. So in that case there is no particular
4068 * rush to update metadata.
4069 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4070 * update the metadata to advance 'safepos' to match 'readpos' so that
4071 * we can be safe in the event of a crash.
4072 * So we insist on updating metadata if safepos is behind writepos and
4073 * readpos is beyond writepos.
4074 * In any case, update the metadata every 10 seconds.
4075 * Maybe that number should be configurable, but I'm not sure it is
4076 * worth it.... maybe it could be a multiple of safemode_delay???
4077 */
NeilBrownfef9c612009-03-31 15:16:46 +11004078 if ((mddev->delta_disks < 0
NeilBrownc8f517c2009-03-31 15:28:40 +11004079 ? (safepos > writepos && readpos < writepos)
4080 : (safepos < writepos && readpos > writepos)) ||
4081 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07004082 /* Cannot proceed until we've updated the superblock... */
4083 wait_event(conf->wait_for_overlap,
4084 atomic_read(&conf->reshape_stripes)==0);
NeilBrownfef9c612009-03-31 15:16:46 +11004085 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004086 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004087 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07004088 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07004089 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07004090 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrown52c03292006-06-26 00:27:43 -07004091 kthread_should_stop());
4092 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004093 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07004094 spin_unlock_irq(&conf->device_lock);
4095 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004096 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07004097 }
4098
NeilBrownec32a2b2009-03-31 15:17:38 +11004099 if (mddev->delta_disks < 0) {
4100 BUG_ON(conf->reshape_progress == 0);
4101 stripe_addr = writepos;
4102 BUG_ON((mddev->dev_sectors &
NeilBrown7a661382009-03-31 15:21:40 +11004103 ~((sector_t)reshape_sectors - 1))
4104 - reshape_sectors - stripe_addr
NeilBrownec32a2b2009-03-31 15:17:38 +11004105 != sector_nr);
4106 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004107 BUG_ON(writepos != sector_nr + reshape_sectors);
NeilBrownec32a2b2009-03-31 15:17:38 +11004108 stripe_addr = sector_nr;
4109 }
NeilBrownab69ae12009-03-31 15:26:47 +11004110 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11004111 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07004112 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10004113 int skipped_disk = 0;
NeilBrowna8c906c2009-06-09 14:39:59 +10004114 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004115 set_bit(STRIPE_EXPANDING, &sh->state);
4116 atomic_inc(&conf->reshape_stripes);
4117 /* If any of this stripe is beyond the end of the old
4118 * array, then we need to zero those blocks
4119 */
4120 for (j=sh->disks; j--;) {
4121 sector_t s;
4122 if (j == sh->pd_idx)
4123 continue;
NeilBrownf4168852007-02-28 20:11:53 -08004124 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11004125 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08004126 continue;
NeilBrown784052e2009-03-31 15:19:07 +11004127 s = compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11004128 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10004129 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07004130 continue;
4131 }
4132 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4133 set_bit(R5_Expanded, &sh->dev[j].flags);
4134 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4135 }
NeilBrowna9f326e2009-09-23 18:06:41 +10004136 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07004137 set_bit(STRIPE_EXPAND_READY, &sh->state);
4138 set_bit(STRIPE_HANDLE, &sh->state);
4139 }
NeilBrownab69ae12009-03-31 15:26:47 +11004140 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07004141 }
4142 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004143 if (mddev->delta_disks < 0)
NeilBrown7a661382009-03-31 15:21:40 +11004144 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11004145 else
NeilBrown7a661382009-03-31 15:21:40 +11004146 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07004147 spin_unlock_irq(&conf->device_lock);
4148 /* Ok, those stripe are ready. We can start scheduling
4149 * reads on the source stripes.
4150 * The source stripes are determined by mapping the first and last
4151 * block on the destination stripes.
4152 */
NeilBrown52c03292006-06-26 00:27:43 -07004153 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11004154 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11004155 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07004156 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10004157 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10004158 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11004159 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11004160 if (last_sector >= mddev->dev_sectors)
4161 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07004162 while (first_sector <= last_sector) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004163 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004164 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4165 set_bit(STRIPE_HANDLE, &sh->state);
4166 release_stripe(sh);
4167 first_sector += STRIPE_SECTORS;
4168 }
NeilBrownab69ae12009-03-31 15:26:47 +11004169 /* Now that the sources are clearly marked, we can release
4170 * the destination stripes
4171 */
4172 while (!list_empty(&stripes)) {
4173 sh = list_entry(stripes.next, struct stripe_head, lru);
4174 list_del_init(&sh->lru);
4175 release_stripe(sh);
4176 }
NeilBrownc6207272008-02-06 01:39:52 -08004177 /* If this takes us to the resync_max point where we have to pause,
4178 * then we need to write out the superblock.
4179 */
NeilBrown7a661382009-03-31 15:21:40 +11004180 sector_nr += reshape_sectors;
NeilBrownc03f6a12009-04-17 11:06:30 +10004181 if ((sector_nr - mddev->curr_resync_completed) * 2
4182 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08004183 /* Cannot proceed until we've updated the superblock... */
4184 wait_event(conf->wait_for_overlap,
4185 atomic_read(&conf->reshape_stripes) == 0);
NeilBrownfef9c612009-03-31 15:16:46 +11004186 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004187 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004188 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08004189 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4190 md_wakeup_thread(mddev->thread);
4191 wait_event(mddev->sb_wait,
4192 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4193 || kthread_should_stop());
4194 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004195 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08004196 spin_unlock_irq(&conf->device_lock);
4197 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004198 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08004199 }
NeilBrown7a661382009-03-31 15:21:40 +11004200 return reshape_sectors;
NeilBrown52c03292006-06-26 00:27:43 -07004201}
4202
4203/* FIXME go_faster isn't used */
4204static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
4205{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004206 raid5_conf_t *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07004207 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11004208 sector_t max_sector = mddev->dev_sectors;
NeilBrown57dab0b2010-10-19 10:03:39 +11004209 sector_t sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07004210 int still_degraded = 0;
4211 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004212
NeilBrown72626682005-09-09 16:23:54 -07004213 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214 /* just being told to finish up .. nothing much to do */
NeilBrowncea9c222009-03-31 15:15:05 +11004215
NeilBrown29269552006-03-27 01:18:10 -08004216 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4217 end_reshape(conf);
4218 return 0;
4219 }
NeilBrown72626682005-09-09 16:23:54 -07004220
4221 if (mddev->curr_resync < max_sector) /* aborted */
4222 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4223 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07004224 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07004225 conf->fullsync = 0;
4226 bitmap_close_sync(mddev->bitmap);
4227
Linus Torvalds1da177e2005-04-16 15:20:36 -07004228 return 0;
4229 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08004230
NeilBrown64bd6602009-08-03 10:59:58 +10004231 /* Allow raid5_quiesce to complete */
4232 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4233
NeilBrown52c03292006-06-26 00:27:43 -07004234 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4235 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08004236
NeilBrownc6207272008-02-06 01:39:52 -08004237 /* No need to check resync_max as we never do more than one
4238 * stripe, and as resync_max will always be on a chunk boundary,
4239 * if the check in md_do_sync didn't fire, there is no chance
4240 * of overstepping resync_max here
4241 */
4242
NeilBrown16a53ec2006-06-26 00:27:38 -07004243 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07004244 * to resync, then assert that we are finished, because there is
4245 * nothing we can do.
4246 */
NeilBrown3285edf2006-06-26 00:27:55 -07004247 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07004248 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004249 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07004250 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004251 return rv;
4252 }
NeilBrown72626682005-09-09 16:23:54 -07004253 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrown3855ad92005-11-08 21:39:38 -08004254 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown72626682005-09-09 16:23:54 -07004255 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4256 /* we can skip this block, and probably more */
4257 sync_blocks /= STRIPE_SECTORS;
4258 *skipped = 1;
4259 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261
NeilBrownb47490c2008-02-06 01:39:50 -08004262
4263 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4264
NeilBrowna8c906c2009-06-09 14:39:59 +10004265 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004266 if (sh == NULL) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004267 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004268 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07004269 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07004270 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08004271 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004272 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004273 /* Need to check if array will still be degraded after recovery/resync
4274 * We don't need to check the 'failed' flag as when that gets set,
4275 * recovery aborts.
4276 */
NeilBrownf001a702009-06-09 14:30:31 +10004277 for (i = 0; i < conf->raid_disks; i++)
NeilBrown16a53ec2006-06-26 00:27:38 -07004278 if (conf->disks[i].rdev == NULL)
4279 still_degraded = 1;
4280
4281 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4282
NeilBrown83206d62011-07-26 11:19:49 +10004283 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004284
NeilBrown14425772009-10-16 15:55:25 +11004285 handle_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004286 release_stripe(sh);
4287
4288 return STRIPE_SECTORS;
4289}
4290
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004291static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4292{
4293 /* We may not be able to submit a whole bio at once as there
4294 * may not be enough stripe_heads available.
4295 * We cannot pre-allocate enough stripe_heads as we may need
4296 * more than exist in the cache (if we allow ever large chunks).
4297 * So we do one stripe head at a time and record in
4298 * ->bi_hw_segments how many have been done.
4299 *
4300 * We *know* that this entire raid_bio is in one chunk, so
4301 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4302 */
4303 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11004304 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004305 sector_t sector, logical_sector, last_sector;
4306 int scnt = 0;
4307 int remaining;
4308 int handled = 0;
4309
4310 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11004311 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11004312 0, &dd_idx, NULL);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004313 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4314
4315 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08004316 logical_sector += STRIPE_SECTORS,
4317 sector += STRIPE_SECTORS,
4318 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004319
Jens Axboe960e7392008-08-15 10:41:18 +02004320 if (scnt < raid5_bi_hw_segments(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004321 /* already done this stripe */
4322 continue;
4323
NeilBrowna8c906c2009-06-09 14:39:59 +10004324 sh = get_active_stripe(conf, sector, 0, 1, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004325
4326 if (!sh) {
4327 /* failed to get a stripe - must wait */
Jens Axboe960e7392008-08-15 10:41:18 +02004328 raid5_set_bi_hw_segments(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004329 conf->retry_read_aligned = raid_bio;
4330 return handled;
4331 }
4332
4333 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
Neil Brown387bb172007-02-08 14:20:29 -08004334 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4335 release_stripe(sh);
Jens Axboe960e7392008-08-15 10:41:18 +02004336 raid5_set_bi_hw_segments(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08004337 conf->retry_read_aligned = raid_bio;
4338 return handled;
4339 }
4340
Dan Williams36d1c642009-07-14 11:48:22 -07004341 handle_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004342 release_stripe(sh);
4343 handled++;
4344 }
4345 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004346 remaining = raid5_dec_bi_phys_segments(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004347 spin_unlock_irq(&conf->device_lock);
Neil Brown0e13fe232008-06-28 08:31:20 +10004348 if (remaining == 0)
4349 bio_endio(raid_bio, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004350 if (atomic_dec_and_test(&conf->active_aligned_reads))
4351 wake_up(&conf->wait_for_stripe);
4352 return handled;
4353}
4354
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004355
Linus Torvalds1da177e2005-04-16 15:20:36 -07004356/*
4357 * This is our raid5 kernel thread.
4358 *
4359 * We scan the hash table for stripes which can be handled now.
4360 * During the scan, completed stripes are saved for us by the interrupt
4361 * handler, so that they will not have to wait for our next wakeup.
4362 */
NeilBrown6ed30032008-02-06 01:40:00 -08004363static void raid5d(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004364{
4365 struct stripe_head *sh;
NeilBrown070ec552009-06-16 16:54:21 +10004366 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004367 int handled;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004368 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004369
Dan Williams45b42332007-07-09 11:56:43 -07004370 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004371
4372 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004373
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004374 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004375 handled = 0;
4376 spin_lock_irq(&conf->device_lock);
4377 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004378 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004379
NeilBrown7c13edc2011-04-18 18:25:43 +10004380 if (atomic_read(&mddev->plug_cnt) == 0 &&
4381 !list_empty(&conf->bitmap_list)) {
4382 /* Now is a good time to flush some bitmap updates */
4383 conf->seq_flush++;
NeilBrown700e4322005-11-28 13:44:10 -08004384 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07004385 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08004386 spin_lock_irq(&conf->device_lock);
NeilBrown7c13edc2011-04-18 18:25:43 +10004387 conf->seq_write = conf->seq_flush;
NeilBrown72626682005-09-09 16:23:54 -07004388 activate_bit_delay(conf);
4389 }
NeilBrown7c13edc2011-04-18 18:25:43 +10004390 if (atomic_read(&mddev->plug_cnt) == 0)
4391 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07004392
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004393 while ((bio = remove_bio_from_retry(conf))) {
4394 int ok;
4395 spin_unlock_irq(&conf->device_lock);
4396 ok = retry_aligned_read(conf, bio);
4397 spin_lock_irq(&conf->device_lock);
4398 if (!ok)
4399 break;
4400 handled++;
4401 }
4402
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004403 sh = __get_priority_stripe(conf);
4404
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004405 if (!sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004407 spin_unlock_irq(&conf->device_lock);
4408
4409 handled++;
Dan Williams417b8d42009-10-16 16:25:22 +11004410 handle_stripe(sh);
4411 release_stripe(sh);
4412 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004413
4414 spin_lock_irq(&conf->device_lock);
4415 }
Dan Williams45b42332007-07-09 11:56:43 -07004416 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417
4418 spin_unlock_irq(&conf->device_lock);
4419
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004420 async_tx_issue_pending_all();
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004421 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004422
Dan Williams45b42332007-07-09 11:56:43 -07004423 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004424}
4425
NeilBrown3f294f42005-11-08 21:39:25 -08004426static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004427raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004428{
NeilBrown070ec552009-06-16 16:54:21 +10004429 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004430 if (conf)
4431 return sprintf(page, "%d\n", conf->max_nr_stripes);
4432 else
4433 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004434}
4435
NeilBrownc41d4ac2010-06-01 19:37:24 +10004436int
4437raid5_set_cache_size(mddev_t *mddev, int size)
4438{
4439 raid5_conf_t *conf = mddev->private;
4440 int err;
4441
4442 if (size <= 16 || size > 32768)
4443 return -EINVAL;
4444 while (size < conf->max_nr_stripes) {
4445 if (drop_one_stripe(conf))
4446 conf->max_nr_stripes--;
4447 else
4448 break;
4449 }
4450 err = md_allow_write(mddev);
4451 if (err)
4452 return err;
4453 while (size > conf->max_nr_stripes) {
4454 if (grow_one_stripe(conf))
4455 conf->max_nr_stripes++;
4456 else break;
4457 }
4458 return 0;
4459}
4460EXPORT_SYMBOL(raid5_set_cache_size);
4461
NeilBrown3f294f42005-11-08 21:39:25 -08004462static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004463raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08004464{
NeilBrown070ec552009-06-16 16:54:21 +10004465 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004466 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07004467 int err;
4468
NeilBrown3f294f42005-11-08 21:39:25 -08004469 if (len >= PAGE_SIZE)
4470 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08004471 if (!conf)
4472 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08004473
Dan Williams4ef197d82008-04-28 02:15:54 -07004474 if (strict_strtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08004475 return -EINVAL;
NeilBrownc41d4ac2010-06-01 19:37:24 +10004476 err = raid5_set_cache_size(mddev, new);
Dan Williamsb5470dc2008-06-27 21:44:04 -07004477 if (err)
4478 return err;
NeilBrown3f294f42005-11-08 21:39:25 -08004479 return len;
4480}
NeilBrown007583c2005-11-08 21:39:30 -08004481
NeilBrown96de1e62005-11-08 21:39:39 -08004482static struct md_sysfs_entry
4483raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4484 raid5_show_stripe_cache_size,
4485 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08004486
4487static ssize_t
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004488raid5_show_preread_threshold(mddev_t *mddev, char *page)
4489{
NeilBrown070ec552009-06-16 16:54:21 +10004490 raid5_conf_t *conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004491 if (conf)
4492 return sprintf(page, "%d\n", conf->bypass_threshold);
4493 else
4494 return 0;
4495}
4496
4497static ssize_t
4498raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4499{
NeilBrown070ec552009-06-16 16:54:21 +10004500 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004501 unsigned long new;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004502 if (len >= PAGE_SIZE)
4503 return -EINVAL;
4504 if (!conf)
4505 return -ENODEV;
4506
Dan Williams4ef197d82008-04-28 02:15:54 -07004507 if (strict_strtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004508 return -EINVAL;
Dan Williams4ef197d82008-04-28 02:15:54 -07004509 if (new > conf->max_nr_stripes)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004510 return -EINVAL;
4511 conf->bypass_threshold = new;
4512 return len;
4513}
4514
4515static struct md_sysfs_entry
4516raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4517 S_IRUGO | S_IWUSR,
4518 raid5_show_preread_threshold,
4519 raid5_store_preread_threshold);
4520
4521static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08004522stripe_cache_active_show(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004523{
NeilBrown070ec552009-06-16 16:54:21 +10004524 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004525 if (conf)
4526 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4527 else
4528 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004529}
4530
NeilBrown96de1e62005-11-08 21:39:39 -08004531static struct md_sysfs_entry
4532raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08004533
NeilBrown007583c2005-11-08 21:39:30 -08004534static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08004535 &raid5_stripecache_size.attr,
4536 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004537 &raid5_preread_bypass_threshold.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08004538 NULL,
4539};
NeilBrown007583c2005-11-08 21:39:30 -08004540static struct attribute_group raid5_attrs_group = {
4541 .name = NULL,
4542 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08004543};
4544
Dan Williams80c3a6c2009-03-17 18:10:40 -07004545static sector_t
4546raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4547{
NeilBrown070ec552009-06-16 16:54:21 +10004548 raid5_conf_t *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07004549
4550 if (!sectors)
4551 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004552 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11004553 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11004554 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004555
Andre Noll9d8f0362009-06-18 08:45:01 +10004556 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Andre Noll664e7c42009-06-18 08:45:27 +10004557 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004558 return sectors * (raid_disks - conf->max_degraded);
4559}
4560
Dan Williams36d1c642009-07-14 11:48:22 -07004561static void raid5_free_percpu(raid5_conf_t *conf)
4562{
4563 struct raid5_percpu *percpu;
4564 unsigned long cpu;
4565
4566 if (!conf->percpu)
4567 return;
4568
4569 get_online_cpus();
4570 for_each_possible_cpu(cpu) {
4571 percpu = per_cpu_ptr(conf->percpu, cpu);
4572 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004573 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004574 }
4575#ifdef CONFIG_HOTPLUG_CPU
4576 unregister_cpu_notifier(&conf->cpu_notify);
4577#endif
4578 put_online_cpus();
4579
4580 free_percpu(conf->percpu);
4581}
4582
Dan Williams95fc17a2009-07-31 12:39:15 +10004583static void free_conf(raid5_conf_t *conf)
4584{
4585 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07004586 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10004587 kfree(conf->disks);
4588 kfree(conf->stripe_hashtbl);
4589 kfree(conf);
4590}
4591
Dan Williams36d1c642009-07-14 11:48:22 -07004592#ifdef CONFIG_HOTPLUG_CPU
4593static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4594 void *hcpu)
4595{
4596 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4597 long cpu = (long)hcpu;
4598 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4599
4600 switch (action) {
4601 case CPU_UP_PREPARE:
4602 case CPU_UP_PREPARE_FROZEN:
Dan Williamsd6f38f32009-07-14 11:50:52 -07004603 if (conf->level == 6 && !percpu->spare_page)
Dan Williams36d1c642009-07-14 11:48:22 -07004604 percpu->spare_page = alloc_page(GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004605 if (!percpu->scribble)
4606 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4607
4608 if (!percpu->scribble ||
4609 (conf->level == 6 && !percpu->spare_page)) {
4610 safe_put_page(percpu->spare_page);
4611 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004612 pr_err("%s: failed memory allocation for cpu%ld\n",
4613 __func__, cpu);
Akinobu Mita55af6bb2010-05-26 14:43:35 -07004614 return notifier_from_errno(-ENOMEM);
Dan Williams36d1c642009-07-14 11:48:22 -07004615 }
4616 break;
4617 case CPU_DEAD:
4618 case CPU_DEAD_FROZEN:
4619 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004620 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004621 percpu->spare_page = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004622 percpu->scribble = NULL;
Dan Williams36d1c642009-07-14 11:48:22 -07004623 break;
4624 default:
4625 break;
4626 }
4627 return NOTIFY_OK;
4628}
4629#endif
4630
4631static int raid5_alloc_percpu(raid5_conf_t *conf)
4632{
4633 unsigned long cpu;
4634 struct page *spare_page;
Tejun Heoa29d8b82010-02-02 14:39:15 +09004635 struct raid5_percpu __percpu *allcpus;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004636 void *scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004637 int err;
4638
Dan Williams36d1c642009-07-14 11:48:22 -07004639 allcpus = alloc_percpu(struct raid5_percpu);
4640 if (!allcpus)
4641 return -ENOMEM;
4642 conf->percpu = allcpus;
4643
4644 get_online_cpus();
4645 err = 0;
4646 for_each_present_cpu(cpu) {
Dan Williamsd6f38f32009-07-14 11:50:52 -07004647 if (conf->level == 6) {
4648 spare_page = alloc_page(GFP_KERNEL);
4649 if (!spare_page) {
4650 err = -ENOMEM;
4651 break;
4652 }
4653 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4654 }
NeilBrown5e5e3e72009-10-16 16:35:30 +11004655 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004656 if (!scribble) {
Dan Williams36d1c642009-07-14 11:48:22 -07004657 err = -ENOMEM;
4658 break;
4659 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07004660 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004661 }
4662#ifdef CONFIG_HOTPLUG_CPU
4663 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4664 conf->cpu_notify.priority = 0;
4665 if (err == 0)
4666 err = register_cpu_notifier(&conf->cpu_notify);
4667#endif
4668 put_online_cpus();
4669
4670 return err;
4671}
4672
NeilBrown91adb562009-03-31 14:39:39 +11004673static raid5_conf_t *setup_conf(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004674{
4675 raid5_conf_t *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004676 int raid_disk, memory, max_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004677 mdk_rdev_t *rdev;
4678 struct disk_info *disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004679
NeilBrown91adb562009-03-31 14:39:39 +11004680 if (mddev->new_level != 5
4681 && mddev->new_level != 4
4682 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10004683 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004684 mdname(mddev), mddev->new_level);
4685 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004686 }
NeilBrown91adb562009-03-31 14:39:39 +11004687 if ((mddev->new_level == 5
4688 && !algorithm_valid_raid5(mddev->new_layout)) ||
4689 (mddev->new_level == 6
4690 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004691 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11004692 mdname(mddev), mddev->new_layout);
4693 return ERR_PTR(-EIO);
4694 }
4695 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10004696 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004697 mdname(mddev), mddev->raid_disks);
4698 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11004699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004700
Andre Noll664e7c42009-06-18 08:45:27 +10004701 if (!mddev->new_chunk_sectors ||
4702 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4703 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10004704 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4705 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11004706 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11004707 }
4708
NeilBrown91adb562009-03-31 14:39:39 +11004709 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4710 if (conf == NULL)
4711 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11004712 spin_lock_init(&conf->device_lock);
4713 init_waitqueue_head(&conf->wait_for_stripe);
4714 init_waitqueue_head(&conf->wait_for_overlap);
4715 INIT_LIST_HEAD(&conf->handle_list);
4716 INIT_LIST_HEAD(&conf->hold_list);
4717 INIT_LIST_HEAD(&conf->delayed_list);
4718 INIT_LIST_HEAD(&conf->bitmap_list);
4719 INIT_LIST_HEAD(&conf->inactive_list);
4720 atomic_set(&conf->active_stripes, 0);
4721 atomic_set(&conf->preread_active_stripes, 0);
4722 atomic_set(&conf->active_aligned_reads, 0);
4723 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrown91adb562009-03-31 14:39:39 +11004724
4725 conf->raid_disks = mddev->raid_disks;
4726 if (mddev->reshape_position == MaxSector)
4727 conf->previous_raid_disks = mddev->raid_disks;
4728 else
4729 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004730 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4731 conf->scribble_len = scribble_len(max_disks);
NeilBrown91adb562009-03-31 14:39:39 +11004732
NeilBrown5e5e3e72009-10-16 16:35:30 +11004733 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11004734 GFP_KERNEL);
4735 if (!conf->disks)
4736 goto abort;
4737
4738 conf->mddev = mddev;
4739
4740 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
4741 goto abort;
4742
Dan Williams36d1c642009-07-14 11:48:22 -07004743 conf->level = mddev->new_level;
4744 if (raid5_alloc_percpu(conf) != 0)
4745 goto abort;
4746
NeilBrown0c55e022010-05-03 14:09:02 +10004747 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11004748
4749 list_for_each_entry(rdev, &mddev->disks, same_set) {
4750 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004751 if (raid_disk >= max_disks
NeilBrown91adb562009-03-31 14:39:39 +11004752 || raid_disk < 0)
4753 continue;
4754 disk = conf->disks + raid_disk;
4755
4756 disk->rdev = rdev;
4757
4758 if (test_bit(In_sync, &rdev->flags)) {
4759 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10004760 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4761 " disk %d\n",
4762 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
Jonathan Brassowd6b212f2011-06-08 18:00:28 -05004763 } else if (rdev->saved_raid_disk != raid_disk)
NeilBrown91adb562009-03-31 14:39:39 +11004764 /* Cannot rely on bitmap to complete recovery */
4765 conf->fullsync = 1;
4766 }
4767
Andre Noll09c9e5f2009-06-18 08:45:55 +10004768 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown91adb562009-03-31 14:39:39 +11004769 conf->level = mddev->new_level;
4770 if (conf->level == 6)
4771 conf->max_degraded = 2;
4772 else
4773 conf->max_degraded = 1;
4774 conf->algorithm = mddev->new_layout;
4775 conf->max_nr_stripes = NR_STRIPES;
NeilBrownfef9c612009-03-31 15:16:46 +11004776 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11004777 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10004778 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11004779 conf->prev_algo = mddev->layout;
4780 }
NeilBrown91adb562009-03-31 14:39:39 +11004781
4782 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11004783 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
NeilBrown91adb562009-03-31 14:39:39 +11004784 if (grow_stripes(conf, conf->max_nr_stripes)) {
4785 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004786 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4787 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004788 goto abort;
4789 } else
NeilBrown0c55e022010-05-03 14:09:02 +10004790 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4791 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004792
NeilBrown0da3c612009-09-23 18:09:45 +10004793 conf->thread = md_register_thread(raid5d, mddev, NULL);
NeilBrown91adb562009-03-31 14:39:39 +11004794 if (!conf->thread) {
4795 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004796 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11004797 mdname(mddev));
4798 goto abort;
4799 }
4800
4801 return conf;
4802
4803 abort:
4804 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10004805 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11004806 return ERR_PTR(-EIO);
4807 } else
4808 return ERR_PTR(-ENOMEM);
4809}
4810
NeilBrownc148ffd2009-11-13 17:47:00 +11004811
4812static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4813{
4814 switch (algo) {
4815 case ALGORITHM_PARITY_0:
4816 if (raid_disk < max_degraded)
4817 return 1;
4818 break;
4819 case ALGORITHM_PARITY_N:
4820 if (raid_disk >= raid_disks - max_degraded)
4821 return 1;
4822 break;
4823 case ALGORITHM_PARITY_0_6:
4824 if (raid_disk == 0 ||
4825 raid_disk == raid_disks - 1)
4826 return 1;
4827 break;
4828 case ALGORITHM_LEFT_ASYMMETRIC_6:
4829 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4830 case ALGORITHM_LEFT_SYMMETRIC_6:
4831 case ALGORITHM_RIGHT_SYMMETRIC_6:
4832 if (raid_disk == raid_disks - 1)
4833 return 1;
4834 }
4835 return 0;
4836}
4837
NeilBrown91adb562009-03-31 14:39:39 +11004838static int run(mddev_t *mddev)
4839{
4840 raid5_conf_t *conf;
NeilBrown9f7c2222010-07-26 12:04:13 +10004841 int working_disks = 0;
NeilBrownc148ffd2009-11-13 17:47:00 +11004842 int dirty_parity_disks = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004843 mdk_rdev_t *rdev;
NeilBrownc148ffd2009-11-13 17:47:00 +11004844 sector_t reshape_offset = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004845
Andre Noll8c6ac862009-06-18 08:48:06 +10004846 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10004847 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10004848 " -- starting background reconstruction\n",
4849 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004850 if (mddev->reshape_position != MaxSector) {
4851 /* Check that we can continue the reshape.
4852 * Currently only disks can change, it must
4853 * increase, and we must be past the point where
4854 * a stripe over-writes itself
4855 */
4856 sector_t here_new, here_old;
4857 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11004858 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrownf6705572006-03-27 01:18:11 -08004859
NeilBrown88ce4932009-03-31 15:24:23 +11004860 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10004861 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08004862 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08004863 mdname(mddev));
4864 return -EINVAL;
4865 }
NeilBrownf6705572006-03-27 01:18:11 -08004866 old_disks = mddev->raid_disks - mddev->delta_disks;
4867 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08004868 * further up in new geometry must map after here in old
4869 * geometry.
NeilBrownf6705572006-03-27 01:18:11 -08004870 */
4871 here_new = mddev->reshape_position;
Andre Noll664e7c42009-06-18 08:45:27 +10004872 if (sector_div(here_new, mddev->new_chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004873 (mddev->raid_disks - max_degraded))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004874 printk(KERN_ERR "md/raid:%s: reshape_position not "
4875 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004876 return -EINVAL;
4877 }
NeilBrownc148ffd2009-11-13 17:47:00 +11004878 reshape_offset = here_new * mddev->new_chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08004879 /* here_new is the stripe we will write to */
4880 here_old = mddev->reshape_position;
Andre Noll9d8f0362009-06-18 08:45:01 +10004881 sector_div(here_old, mddev->chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004882 (old_disks-max_degraded));
4883 /* here_old is the first stripe that we might need to read
4884 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10004885 if (mddev->delta_disks == 0) {
4886 /* We cannot be sure it is safe to start an in-place
4887 * reshape. It is only safe if user-space if monitoring
4888 * and taking constant backups.
4889 * mdadm always starts a situation like this in
4890 * readonly mode so it can take control before
4891 * allowing any writes. So just check for that.
4892 */
4893 if ((here_new * mddev->new_chunk_sectors !=
4894 here_old * mddev->chunk_sectors) ||
4895 mddev->ro == 0) {
NeilBrown0c55e022010-05-03 14:09:02 +10004896 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
4897 " in read-only mode - aborting\n",
4898 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10004899 return -EINVAL;
4900 }
4901 } else if (mddev->delta_disks < 0
4902 ? (here_new * mddev->new_chunk_sectors <=
4903 here_old * mddev->chunk_sectors)
4904 : (here_new * mddev->new_chunk_sectors >=
4905 here_old * mddev->chunk_sectors)) {
NeilBrownf6705572006-03-27 01:18:11 -08004906 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10004907 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
4908 "auto-recovery - aborting.\n",
4909 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004910 return -EINVAL;
4911 }
NeilBrown0c55e022010-05-03 14:09:02 +10004912 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
4913 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004914 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08004915 } else {
NeilBrown91adb562009-03-31 14:39:39 +11004916 BUG_ON(mddev->level != mddev->new_level);
4917 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10004918 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11004919 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08004920 }
4921
NeilBrown245f46c2009-03-31 14:39:39 +11004922 if (mddev->private == NULL)
4923 conf = setup_conf(mddev);
4924 else
4925 conf = mddev->private;
4926
NeilBrown91adb562009-03-31 14:39:39 +11004927 if (IS_ERR(conf))
4928 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08004929
NeilBrown91adb562009-03-31 14:39:39 +11004930 mddev->thread = conf->thread;
4931 conf->thread = NULL;
4932 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004933
Linus Torvalds1da177e2005-04-16 15:20:36 -07004934 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07004935 * 0 for a fully functional array, 1 or 2 for a degraded array.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004936 */
NeilBrownc148ffd2009-11-13 17:47:00 +11004937 list_for_each_entry(rdev, &mddev->disks, same_set) {
4938 if (rdev->raid_disk < 0)
4939 continue;
NeilBrown2f115882010-06-17 17:41:03 +10004940 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown91adb562009-03-31 14:39:39 +11004941 working_disks++;
NeilBrown2f115882010-06-17 17:41:03 +10004942 continue;
4943 }
NeilBrownc148ffd2009-11-13 17:47:00 +11004944 /* This disc is not fully in-sync. However if it
4945 * just stored parity (beyond the recovery_offset),
4946 * when we don't need to be concerned about the
4947 * array being dirty.
4948 * When reshape goes 'backwards', we never have
4949 * partially completed devices, so we only need
4950 * to worry about reshape going forwards.
4951 */
4952 /* Hack because v0.91 doesn't store recovery_offset properly. */
4953 if (mddev->major_version == 0 &&
4954 mddev->minor_version > 90)
4955 rdev->recovery_offset = reshape_offset;
4956
NeilBrownc148ffd2009-11-13 17:47:00 +11004957 if (rdev->recovery_offset < reshape_offset) {
4958 /* We need to check old and new layout */
4959 if (!only_parity(rdev->raid_disk,
4960 conf->algorithm,
4961 conf->raid_disks,
4962 conf->max_degraded))
4963 continue;
4964 }
4965 if (!only_parity(rdev->raid_disk,
4966 conf->prev_algo,
4967 conf->previous_raid_disks,
4968 conf->max_degraded))
4969 continue;
4970 dirty_parity_disks++;
4971 }
NeilBrown91adb562009-03-31 14:39:39 +11004972
NeilBrown5e5e3e72009-10-16 16:35:30 +11004973 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
4974 - working_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004975
NeilBrown674806d2010-06-16 17:17:53 +10004976 if (has_failed(conf)) {
NeilBrown0c55e022010-05-03 14:09:02 +10004977 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07004978 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07004979 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004980 goto abort;
4981 }
4982
NeilBrown91adb562009-03-31 14:39:39 +11004983 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10004984 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11004985 mddev->resync_max_sectors = mddev->dev_sectors;
4986
NeilBrownc148ffd2009-11-13 17:47:00 +11004987 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07004988 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08004989 if (mddev->ok_start_degraded)
4990 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10004991 "md/raid:%s: starting dirty degraded array"
4992 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08004993 mdname(mddev));
4994 else {
4995 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004996 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08004997 mdname(mddev));
4998 goto abort;
4999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005000 }
5001
Linus Torvalds1da177e2005-04-16 15:20:36 -07005002 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10005003 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5004 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11005005 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5006 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005007 else
NeilBrown0c55e022010-05-03 14:09:02 +10005008 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5009 " out of %d devices, algorithm %d\n",
5010 mdname(mddev), conf->level,
5011 mddev->raid_disks - mddev->degraded,
5012 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013
5014 print_raid5_conf(conf);
5015
NeilBrownfef9c612009-03-31 15:16:46 +11005016 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11005017 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08005018 atomic_set(&conf->reshape_stripes, 0);
5019 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5020 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5021 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5022 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5023 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005024 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08005025 }
5026
Linus Torvalds1da177e2005-04-16 15:20:36 -07005027
5028 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10005029 if (mddev->to_remove == &raid5_attrs_group)
5030 mddev->to_remove = NULL;
NeilBrown00bcb4a2010-06-01 19:37:23 +10005031 else if (mddev->kobj.sd &&
5032 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08005033 printk(KERN_WARNING
NeilBrown4a5add42010-06-01 19:37:28 +10005034 "raid5: failed to create sysfs attributes for %s\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08005035 mdname(mddev));
NeilBrown4a5add42010-06-01 19:37:28 +10005036 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5037
5038 if (mddev->queue) {
NeilBrown9f7c2222010-07-26 12:04:13 +10005039 int chunk_size;
NeilBrown4a5add42010-06-01 19:37:28 +10005040 /* read-ahead size must cover two whole stripes, which
5041 * is 2 * (datadisks) * chunksize where 'n' is the
5042 * number of raid devices
5043 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005044 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5045 int stripe = data_disks *
5046 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
5047 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5048 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
NeilBrown4a5add42010-06-01 19:37:28 +10005049
5050 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
NeilBrown11d8a6e2010-07-26 11:57:07 +10005051
5052 mddev->queue->backing_dev_info.congested_data = mddev;
5053 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
NeilBrown9f7c2222010-07-26 12:04:13 +10005054
5055 chunk_size = mddev->chunk_sectors << 9;
5056 blk_queue_io_min(mddev->queue, chunk_size);
5057 blk_queue_io_opt(mddev->queue, chunk_size *
5058 (conf->raid_disks - conf->max_degraded));
5059
5060 list_for_each_entry(rdev, &mddev->disks, same_set)
5061 disk_stack_limits(mddev->gendisk, rdev->bdev,
5062 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005063 }
5064
Linus Torvalds1da177e2005-04-16 15:20:36 -07005065 return 0;
5066abort:
NeilBrowne0cf8f02009-03-31 14:39:39 +11005067 md_unregister_thread(mddev->thread);
NeilBrown91adb562009-03-31 14:39:39 +11005068 mddev->thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005069 if (conf) {
5070 print_raid5_conf(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005071 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005072 }
5073 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10005074 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005075 return -EIO;
5076}
5077
NeilBrown3f294f42005-11-08 21:39:25 -08005078static int stop(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005079{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005080 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005081
5082 md_unregister_thread(mddev->thread);
5083 mddev->thread = NULL;
NeilBrown11d8a6e2010-07-26 11:57:07 +10005084 if (mddev->queue)
5085 mddev->queue->backing_dev_info.congested_fn = NULL;
Dan Williams95fc17a2009-07-31 12:39:15 +10005086 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10005087 mddev->private = NULL;
5088 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005089 return 0;
5090}
5091
Dan Williams45b42332007-07-09 11:56:43 -07005092#ifdef DEBUG
NeilBrownd710e132008-10-13 11:55:12 +11005093static void print_sh(struct seq_file *seq, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005094{
5095 int i;
5096
NeilBrown16a53ec2006-06-26 00:27:38 -07005097 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
5098 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
5099 seq_printf(seq, "sh %llu, count %d.\n",
5100 (unsigned long long)sh->sector, atomic_read(&sh->count));
5101 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005102 for (i = 0; i < sh->disks; i++) {
NeilBrown16a53ec2006-06-26 00:27:38 -07005103 seq_printf(seq, "(cache%d: %p %ld) ",
5104 i, sh->dev[i].page, sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005105 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005106 seq_printf(seq, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005107}
5108
NeilBrownd710e132008-10-13 11:55:12 +11005109static void printall(struct seq_file *seq, raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005110{
5111 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -08005112 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005113 int i;
5114
5115 spin_lock_irq(&conf->device_lock);
5116 for (i = 0; i < NR_HASH; i++) {
NeilBrownfccddba2006-01-06 00:20:33 -08005117 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005118 if (sh->raid_conf != conf)
5119 continue;
NeilBrown16a53ec2006-06-26 00:27:38 -07005120 print_sh(seq, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005121 }
5122 }
5123 spin_unlock_irq(&conf->device_lock);
5124}
5125#endif
5126
NeilBrownd710e132008-10-13 11:55:12 +11005127static void status(struct seq_file *seq, mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005128{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005129 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005130 int i;
5131
Andre Noll9d8f0362009-06-18 08:45:01 +10005132 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5133 mddev->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07005134 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135 for (i = 0; i < conf->raid_disks; i++)
5136 seq_printf (seq, "%s",
5137 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08005138 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005139 seq_printf (seq, "]");
Dan Williams45b42332007-07-09 11:56:43 -07005140#ifdef DEBUG
NeilBrown16a53ec2006-06-26 00:27:38 -07005141 seq_printf (seq, "\n");
5142 printall(seq, conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005143#endif
5144}
5145
5146static void print_raid5_conf (raid5_conf_t *conf)
5147{
5148 int i;
5149 struct disk_info *tmp;
5150
NeilBrown0c55e022010-05-03 14:09:02 +10005151 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005152 if (!conf) {
5153 printk("(conf==NULL)\n");
5154 return;
5155 }
NeilBrown0c55e022010-05-03 14:09:02 +10005156 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5157 conf->raid_disks,
5158 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005159
5160 for (i = 0; i < conf->raid_disks; i++) {
5161 char b[BDEVNAME_SIZE];
5162 tmp = conf->disks + i;
5163 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10005164 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5165 i, !test_bit(Faulty, &tmp->rdev->flags),
5166 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005167 }
5168}
5169
5170static int raid5_spare_active(mddev_t *mddev)
5171{
5172 int i;
5173 raid5_conf_t *conf = mddev->private;
5174 struct disk_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10005175 int count = 0;
5176 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005177
5178 for (i = 0; i < conf->raid_disks; i++) {
5179 tmp = conf->disks + i;
5180 if (tmp->rdev
NeilBrown70fffd02010-06-16 17:01:25 +10005181 && tmp->rdev->recovery_offset == MaxSector
NeilBrownb2d444d2005-11-08 21:39:31 -08005182 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07005183 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10005184 count++;
Jonathan Brassow43c73ca2011-01-14 09:14:33 +11005185 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005186 }
5187 }
NeilBrown6b965622010-08-18 11:56:59 +10005188 spin_lock_irqsave(&conf->device_lock, flags);
5189 mddev->degraded -= count;
5190 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005191 print_raid5_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10005192 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005193}
5194
5195static int raid5_remove_disk(mddev_t *mddev, int number)
5196{
5197 raid5_conf_t *conf = mddev->private;
5198 int err = 0;
5199 mdk_rdev_t *rdev;
5200 struct disk_info *p = conf->disks + number;
5201
5202 print_raid5_conf(conf);
5203 rdev = p->rdev;
5204 if (rdev) {
NeilBrownec32a2b2009-03-31 15:17:38 +11005205 if (number >= conf->raid_disks &&
5206 conf->reshape_progress == MaxSector)
5207 clear_bit(In_sync, &rdev->flags);
5208
NeilBrownb2d444d2005-11-08 21:39:31 -08005209 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07005210 atomic_read(&rdev->nr_pending)) {
5211 err = -EBUSY;
5212 goto abort;
5213 }
NeilBrowndfc70642008-05-23 13:04:39 -07005214 /* Only remove non-faulty devices if recovery
5215 * isn't possible.
5216 */
5217 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown674806d2010-06-16 17:17:53 +10005218 !has_failed(conf) &&
NeilBrownec32a2b2009-03-31 15:17:38 +11005219 number < conf->raid_disks) {
NeilBrowndfc70642008-05-23 13:04:39 -07005220 err = -EBUSY;
5221 goto abort;
5222 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005223 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07005224 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005225 if (atomic_read(&rdev->nr_pending)) {
5226 /* lost the race, try later */
5227 err = -EBUSY;
5228 p->rdev = rdev;
5229 }
5230 }
5231abort:
5232
5233 print_raid5_conf(conf);
5234 return err;
5235}
5236
5237static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
5238{
5239 raid5_conf_t *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10005240 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005241 int disk;
5242 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10005243 int first = 0;
5244 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005245
NeilBrown674806d2010-06-16 17:17:53 +10005246 if (has_failed(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005247 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10005248 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005249
Neil Brown6c2fce22008-06-28 08:31:31 +10005250 if (rdev->raid_disk >= 0)
5251 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005252
5253 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005254 * find the disk ... but prefer rdev->saved_raid_disk
5255 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005256 */
NeilBrown16a53ec2006-06-26 00:27:38 -07005257 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10005258 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07005259 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5260 disk = rdev->saved_raid_disk;
5261 else
Neil Brown6c2fce22008-06-28 08:31:31 +10005262 disk = first;
5263 for ( ; disk <= last ; disk++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005264 if ((p=conf->disks + disk)->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08005265 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005266 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10005267 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07005268 if (rdev->saved_raid_disk != disk)
5269 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08005270 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005271 break;
5272 }
5273 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10005274 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005275}
5276
5277static int raid5_resize(mddev_t *mddev, sector_t sectors)
5278{
5279 /* no resync is happening, and there is enough space
5280 * on all devices, so we can resize.
5281 * We need to make sure resync covers any new space.
5282 * If the array is shrinking we should possibly wait until
5283 * any io in the removed space completes, but it hardly seems
5284 * worth it.
5285 */
Andre Noll9d8f0362009-06-18 08:45:01 +10005286 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Dan Williams1f403622009-03-31 14:59:03 +11005287 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5288 mddev->raid_disks));
Dan Williamsb522adc2009-03-31 15:00:31 +11005289 if (mddev->array_sectors >
5290 raid5_size(mddev, sectors, mddev->raid_disks))
5291 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10005292 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005293 revalidate_disk(mddev->gendisk);
NeilBrownb0986362011-05-11 15:52:21 +10005294 if (sectors > mddev->dev_sectors &&
5295 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11005296 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005297 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5298 }
Andre Noll58c0fed2009-03-31 14:33:13 +11005299 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07005300 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005301 return 0;
5302}
5303
NeilBrown01ee22b2009-06-18 08:47:20 +10005304static int check_stripe_cache(mddev_t *mddev)
5305{
5306 /* Can only proceed if there are plenty of stripe_heads.
5307 * We need a minimum of one full stripe,, and for sensible progress
5308 * it is best to have about 4 times that.
5309 * If we require 4 times, then the default 256 4K stripe_heads will
5310 * allow for chunk sizes up to 256K, which is probably OK.
5311 * If the chunk size is greater, user-space should request more
5312 * stripe_heads first.
5313 */
5314 raid5_conf_t *conf = mddev->private;
5315 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5316 > conf->max_nr_stripes ||
5317 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5318 > conf->max_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10005319 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5320 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10005321 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5322 / STRIPE_SIZE)*4);
5323 return 0;
5324 }
5325 return 1;
5326}
5327
NeilBrown50ac1682009-06-18 08:47:55 +10005328static int check_reshape(mddev_t *mddev)
NeilBrown29269552006-03-27 01:18:10 -08005329{
NeilBrown070ec552009-06-16 16:54:21 +10005330 raid5_conf_t *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08005331
NeilBrown88ce4932009-03-31 15:24:23 +11005332 if (mddev->delta_disks == 0 &&
5333 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10005334 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10005335 return 0; /* nothing to do */
NeilBrowndba034e2008-08-05 15:54:13 +10005336 if (mddev->bitmap)
5337 /* Cannot grow a bitmap yet */
5338 return -EBUSY;
NeilBrown674806d2010-06-16 17:17:53 +10005339 if (has_failed(conf))
NeilBrownec32a2b2009-03-31 15:17:38 +11005340 return -EINVAL;
5341 if (mddev->delta_disks < 0) {
5342 /* We might be able to shrink, but the devices must
5343 * be made bigger first.
5344 * For raid6, 4 is the minimum size.
5345 * Otherwise 2 is the minimum
5346 */
5347 int min = 2;
5348 if (mddev->level == 6)
5349 min = 4;
5350 if (mddev->raid_disks + mddev->delta_disks < min)
5351 return -EINVAL;
5352 }
NeilBrown29269552006-03-27 01:18:10 -08005353
NeilBrown01ee22b2009-06-18 08:47:20 +10005354 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08005355 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08005356
NeilBrownec32a2b2009-03-31 15:17:38 +11005357 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
NeilBrown63c70c42006-03-27 01:18:13 -08005358}
5359
5360static int raid5_start_reshape(mddev_t *mddev)
5361{
NeilBrown070ec552009-06-16 16:54:21 +10005362 raid5_conf_t *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08005363 mdk_rdev_t *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08005364 int spares = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07005365 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08005366
NeilBrownf4168852007-02-28 20:11:53 -08005367 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08005368 return -EBUSY;
5369
NeilBrown01ee22b2009-06-18 08:47:20 +10005370 if (!check_stripe_cache(mddev))
5371 return -ENOSPC;
5372
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005373 list_for_each_entry(rdev, &mddev->disks, same_set)
NeilBrown469518a2011-01-31 11:57:43 +11005374 if (!test_bit(In_sync, &rdev->flags)
5375 && !test_bit(Faulty, &rdev->flags))
NeilBrown29269552006-03-27 01:18:10 -08005376 spares++;
NeilBrown63c70c42006-03-27 01:18:13 -08005377
NeilBrownf4168852007-02-28 20:11:53 -08005378 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08005379 /* Not enough devices even to make a degraded array
5380 * of that size
5381 */
5382 return -EINVAL;
5383
NeilBrownec32a2b2009-03-31 15:17:38 +11005384 /* Refuse to reduce size of the array. Any reductions in
5385 * array size must be through explicit setting of array_size
5386 * attribute.
5387 */
5388 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5389 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10005390 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11005391 "before number of disks\n", mdname(mddev));
5392 return -EINVAL;
5393 }
5394
NeilBrownf6705572006-03-27 01:18:11 -08005395 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08005396 spin_lock_irq(&conf->device_lock);
5397 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08005398 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005399 conf->prev_chunk_sectors = conf->chunk_sectors;
5400 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11005401 conf->prev_algo = conf->algorithm;
5402 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11005403 if (mddev->delta_disks < 0)
5404 conf->reshape_progress = raid5_size(mddev, 0, 0);
5405 else
5406 conf->reshape_progress = 0;
5407 conf->reshape_safe = conf->reshape_progress;
NeilBrown86b42c72009-03-31 15:19:03 +11005408 conf->generation++;
NeilBrown29269552006-03-27 01:18:10 -08005409 spin_unlock_irq(&conf->device_lock);
5410
5411 /* Add some new drives, as many as will fit.
5412 * We know there are enough to make the newly sized array work.
NeilBrown3424bf62010-06-17 17:48:26 +10005413 * Don't add devices if we are reducing the number of
5414 * devices in the array. This is because it is not possible
5415 * to correctly record the "partially reconstructed" state of
5416 * such devices during the reshape and confusion could result.
NeilBrown29269552006-03-27 01:18:10 -08005417 */
NeilBrown87a8dec2011-01-31 11:57:43 +11005418 if (mddev->delta_disks >= 0) {
5419 int added_devices = 0;
5420 list_for_each_entry(rdev, &mddev->disks, same_set)
5421 if (rdev->raid_disk < 0 &&
5422 !test_bit(Faulty, &rdev->flags)) {
5423 if (raid5_add_disk(mddev, rdev) == 0) {
5424 char nm[20];
5425 if (rdev->raid_disk
5426 >= conf->previous_raid_disks) {
5427 set_bit(In_sync, &rdev->flags);
5428 added_devices++;
5429 } else
5430 rdev->recovery_offset = 0;
5431 sprintf(nm, "rd%d", rdev->raid_disk);
5432 if (sysfs_create_link(&mddev->kobj,
5433 &rdev->kobj, nm))
5434 /* Failure here is OK */;
NeilBrown50da0842011-01-31 11:57:43 +11005435 }
NeilBrown87a8dec2011-01-31 11:57:43 +11005436 } else if (rdev->raid_disk >= conf->previous_raid_disks
5437 && !test_bit(Faulty, &rdev->flags)) {
5438 /* This is a spare that was manually added */
5439 set_bit(In_sync, &rdev->flags);
5440 added_devices++;
5441 }
NeilBrown29269552006-03-27 01:18:10 -08005442
NeilBrown87a8dec2011-01-31 11:57:43 +11005443 /* When a reshape changes the number of devices,
5444 * ->degraded is measured against the larger of the
5445 * pre and post number of devices.
5446 */
NeilBrownec32a2b2009-03-31 15:17:38 +11005447 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown9eb07c22010-02-09 12:31:47 +11005448 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
NeilBrownec32a2b2009-03-31 15:17:38 +11005449 - added_devices;
5450 spin_unlock_irqrestore(&conf->device_lock, flags);
5451 }
NeilBrown63c70c42006-03-27 01:18:13 -08005452 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10005453 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07005454 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08005455
NeilBrown29269552006-03-27 01:18:10 -08005456 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5457 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5458 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5459 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5460 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005461 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08005462 if (!mddev->sync_thread) {
5463 mddev->recovery = 0;
5464 spin_lock_irq(&conf->device_lock);
5465 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005466 conf->reshape_progress = MaxSector;
NeilBrown29269552006-03-27 01:18:10 -08005467 spin_unlock_irq(&conf->device_lock);
5468 return -EAGAIN;
5469 }
NeilBrownc8f517c2009-03-31 15:28:40 +11005470 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08005471 md_wakeup_thread(mddev->sync_thread);
5472 md_new_event(mddev);
5473 return 0;
5474}
NeilBrown29269552006-03-27 01:18:10 -08005475
NeilBrownec32a2b2009-03-31 15:17:38 +11005476/* This is called from the reshape thread and should make any
5477 * changes needed in 'conf'
5478 */
NeilBrown29269552006-03-27 01:18:10 -08005479static void end_reshape(raid5_conf_t *conf)
5480{
NeilBrown29269552006-03-27 01:18:10 -08005481
NeilBrownf6705572006-03-27 01:18:11 -08005482 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
Dan Williams80c3a6c2009-03-17 18:10:40 -07005483
NeilBrownf6705572006-03-27 01:18:11 -08005484 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11005485 conf->previous_raid_disks = conf->raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005486 conf->reshape_progress = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08005487 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11005488 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07005489
5490 /* read-ahead size must cover two whole stripes, which is
5491 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5492 */
NeilBrown4a5add42010-06-01 19:37:28 +10005493 if (conf->mddev->queue) {
NeilBrowncea9c222009-03-31 15:15:05 +11005494 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005495 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11005496 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07005497 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5498 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5499 }
NeilBrown29269552006-03-27 01:18:10 -08005500 }
NeilBrown29269552006-03-27 01:18:10 -08005501}
5502
NeilBrownec32a2b2009-03-31 15:17:38 +11005503/* This is called from the raid5d thread with mddev_lock held.
5504 * It makes config changes to the device.
5505 */
NeilBrowncea9c222009-03-31 15:15:05 +11005506static void raid5_finish_reshape(mddev_t *mddev)
5507{
NeilBrown070ec552009-06-16 16:54:21 +10005508 raid5_conf_t *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11005509
5510 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5511
NeilBrownec32a2b2009-03-31 15:17:38 +11005512 if (mddev->delta_disks > 0) {
5513 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5514 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005515 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11005516 } else {
5517 int d;
NeilBrownec32a2b2009-03-31 15:17:38 +11005518 mddev->degraded = conf->raid_disks;
5519 for (d = 0; d < conf->raid_disks ; d++)
5520 if (conf->disks[d].rdev &&
5521 test_bit(In_sync,
5522 &conf->disks[d].rdev->flags))
5523 mddev->degraded--;
5524 for (d = conf->raid_disks ;
5525 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10005526 d++) {
5527 mdk_rdev_t *rdev = conf->disks[d].rdev;
5528 if (rdev && raid5_remove_disk(mddev, d) == 0) {
5529 char nm[20];
5530 sprintf(nm, "rd%d", rdev->raid_disk);
5531 sysfs_remove_link(&mddev->kobj, nm);
5532 rdev->raid_disk = -1;
5533 }
5534 }
NeilBrowncea9c222009-03-31 15:15:05 +11005535 }
NeilBrown88ce4932009-03-31 15:24:23 +11005536 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005537 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11005538 mddev->reshape_position = MaxSector;
5539 mddev->delta_disks = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11005540 }
5541}
5542
NeilBrown72626682005-09-09 16:23:54 -07005543static void raid5_quiesce(mddev_t *mddev, int state)
5544{
NeilBrown070ec552009-06-16 16:54:21 +10005545 raid5_conf_t *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07005546
5547 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08005548 case 2: /* resume for a suspend */
5549 wake_up(&conf->wait_for_overlap);
5550 break;
5551
NeilBrown72626682005-09-09 16:23:54 -07005552 case 1: /* stop all writes */
5553 spin_lock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005554 /* '2' tells resync/reshape to pause so that all
5555 * active stripes can drain
5556 */
5557 conf->quiesce = 2;
NeilBrown72626682005-09-09 16:23:54 -07005558 wait_event_lock_irq(conf->wait_for_stripe,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005559 atomic_read(&conf->active_stripes) == 0 &&
5560 atomic_read(&conf->active_aligned_reads) == 0,
NeilBrown72626682005-09-09 16:23:54 -07005561 conf->device_lock, /* nothing */);
NeilBrown64bd6602009-08-03 10:59:58 +10005562 conf->quiesce = 1;
NeilBrown72626682005-09-09 16:23:54 -07005563 spin_unlock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005564 /* allow reshape to continue */
5565 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005566 break;
5567
5568 case 0: /* re-enable writes */
5569 spin_lock_irq(&conf->device_lock);
5570 conf->quiesce = 0;
5571 wake_up(&conf->wait_for_stripe);
NeilBrowne464eaf2006-03-27 01:18:14 -08005572 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005573 spin_unlock_irq(&conf->device_lock);
5574 break;
5575 }
NeilBrown72626682005-09-09 16:23:54 -07005576}
NeilBrownb15c2e52006-01-06 00:20:16 -08005577
NeilBrownd562b0c2009-03-31 14:39:39 +11005578
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005579static void *raid45_takeover_raid0(mddev_t *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11005580{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005581 struct raid0_private_data *raid0_priv = mddev->private;
Randy Dunlapd76c8422011-04-21 09:07:26 -07005582 sector_t sectors;
Trela Maciej54071b32010-03-08 16:02:42 +11005583
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005584 /* for raid0 takeover only one zone is supported */
5585 if (raid0_priv->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10005586 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5587 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005588 return ERR_PTR(-EINVAL);
5589 }
5590
NeilBrown3b71bd92011-04-20 15:38:18 +10005591 sectors = raid0_priv->strip_zone[0].zone_end;
5592 sector_div(sectors, raid0_priv->strip_zone[0].nb_dev);
5593 mddev->dev_sectors = sectors;
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005594 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11005595 mddev->new_layout = ALGORITHM_PARITY_N;
5596 mddev->new_chunk_sectors = mddev->chunk_sectors;
5597 mddev->raid_disks += 1;
5598 mddev->delta_disks = 1;
5599 /* make sure it will be not marked as dirty */
5600 mddev->recovery_cp = MaxSector;
5601
5602 return setup_conf(mddev);
5603}
5604
5605
NeilBrownd562b0c2009-03-31 14:39:39 +11005606static void *raid5_takeover_raid1(mddev_t *mddev)
5607{
5608 int chunksect;
5609
5610 if (mddev->raid_disks != 2 ||
5611 mddev->degraded > 1)
5612 return ERR_PTR(-EINVAL);
5613
5614 /* Should check if there are write-behind devices? */
5615
5616 chunksect = 64*2; /* 64K by default */
5617
5618 /* The array must be an exact multiple of chunksize */
5619 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5620 chunksect >>= 1;
5621
5622 if ((chunksect<<9) < STRIPE_SIZE)
5623 /* array size does not allow a suitable chunk size */
5624 return ERR_PTR(-EINVAL);
5625
5626 mddev->new_level = 5;
5627 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10005628 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11005629
5630 return setup_conf(mddev);
5631}
5632
NeilBrownfc9739c2009-03-31 14:57:20 +11005633static void *raid5_takeover_raid6(mddev_t *mddev)
5634{
5635 int new_layout;
5636
5637 switch (mddev->layout) {
5638 case ALGORITHM_LEFT_ASYMMETRIC_6:
5639 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5640 break;
5641 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5642 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5643 break;
5644 case ALGORITHM_LEFT_SYMMETRIC_6:
5645 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5646 break;
5647 case ALGORITHM_RIGHT_SYMMETRIC_6:
5648 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5649 break;
5650 case ALGORITHM_PARITY_0_6:
5651 new_layout = ALGORITHM_PARITY_0;
5652 break;
5653 case ALGORITHM_PARITY_N:
5654 new_layout = ALGORITHM_PARITY_N;
5655 break;
5656 default:
5657 return ERR_PTR(-EINVAL);
5658 }
5659 mddev->new_level = 5;
5660 mddev->new_layout = new_layout;
5661 mddev->delta_disks = -1;
5662 mddev->raid_disks -= 1;
5663 return setup_conf(mddev);
5664}
5665
NeilBrownd562b0c2009-03-31 14:39:39 +11005666
NeilBrown50ac1682009-06-18 08:47:55 +10005667static int raid5_check_reshape(mddev_t *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11005668{
NeilBrown88ce4932009-03-31 15:24:23 +11005669 /* For a 2-drive array, the layout and chunk size can be changed
5670 * immediately as not restriping is needed.
5671 * For larger arrays we record the new value - after validation
5672 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11005673 */
NeilBrown070ec552009-06-16 16:54:21 +10005674 raid5_conf_t *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10005675 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11005676
NeilBrown597a7112009-06-18 08:47:42 +10005677 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11005678 return -EINVAL;
5679 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005680 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11005681 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005682 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11005683 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005684 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11005685 /* not factor of array size */
5686 return -EINVAL;
5687 }
5688
5689 /* They look valid */
5690
NeilBrown88ce4932009-03-31 15:24:23 +11005691 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10005692 /* can make the change immediately */
5693 if (mddev->new_layout >= 0) {
5694 conf->algorithm = mddev->new_layout;
5695 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11005696 }
5697 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10005698 conf->chunk_sectors = new_chunk ;
5699 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11005700 }
5701 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5702 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11005703 }
NeilBrown50ac1682009-06-18 08:47:55 +10005704 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11005705}
5706
NeilBrown50ac1682009-06-18 08:47:55 +10005707static int raid6_check_reshape(mddev_t *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11005708{
NeilBrown597a7112009-06-18 08:47:42 +10005709 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10005710
NeilBrown597a7112009-06-18 08:47:42 +10005711 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11005712 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005713 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005714 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11005715 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005716 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11005717 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005718 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11005719 /* not factor of array size */
5720 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005721 }
NeilBrown88ce4932009-03-31 15:24:23 +11005722
5723 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10005724 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11005725}
5726
NeilBrownd562b0c2009-03-31 14:39:39 +11005727static void *raid5_takeover(mddev_t *mddev)
5728{
5729 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005730 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005731 * raid1 - if there are two drives. We need to know the chunk size
5732 * raid4 - trivial - just use a raid4 layout.
5733 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005734 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005735 if (mddev->level == 0)
5736 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11005737 if (mddev->level == 1)
5738 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11005739 if (mddev->level == 4) {
5740 mddev->new_layout = ALGORITHM_PARITY_N;
5741 mddev->new_level = 5;
5742 return setup_conf(mddev);
5743 }
NeilBrownfc9739c2009-03-31 14:57:20 +11005744 if (mddev->level == 6)
5745 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11005746
5747 return ERR_PTR(-EINVAL);
5748}
5749
NeilBrowna78d38a2010-03-22 16:53:49 +11005750static void *raid4_takeover(mddev_t *mddev)
5751{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005752 /* raid4 can take over:
5753 * raid0 - if there is only one strip zone
5754 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11005755 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005756 if (mddev->level == 0)
5757 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11005758 if (mddev->level == 5 &&
5759 mddev->layout == ALGORITHM_PARITY_N) {
5760 mddev->new_layout = 0;
5761 mddev->new_level = 4;
5762 return setup_conf(mddev);
5763 }
5764 return ERR_PTR(-EINVAL);
5765}
NeilBrownd562b0c2009-03-31 14:39:39 +11005766
NeilBrown245f46c2009-03-31 14:39:39 +11005767static struct mdk_personality raid5_personality;
5768
5769static void *raid6_takeover(mddev_t *mddev)
5770{
5771 /* Currently can only take over a raid5. We map the
5772 * personality to an equivalent raid6 personality
5773 * with the Q block at the end.
5774 */
5775 int new_layout;
5776
5777 if (mddev->pers != &raid5_personality)
5778 return ERR_PTR(-EINVAL);
5779 if (mddev->degraded > 1)
5780 return ERR_PTR(-EINVAL);
5781 if (mddev->raid_disks > 253)
5782 return ERR_PTR(-EINVAL);
5783 if (mddev->raid_disks < 3)
5784 return ERR_PTR(-EINVAL);
5785
5786 switch (mddev->layout) {
5787 case ALGORITHM_LEFT_ASYMMETRIC:
5788 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5789 break;
5790 case ALGORITHM_RIGHT_ASYMMETRIC:
5791 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5792 break;
5793 case ALGORITHM_LEFT_SYMMETRIC:
5794 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5795 break;
5796 case ALGORITHM_RIGHT_SYMMETRIC:
5797 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5798 break;
5799 case ALGORITHM_PARITY_0:
5800 new_layout = ALGORITHM_PARITY_0_6;
5801 break;
5802 case ALGORITHM_PARITY_N:
5803 new_layout = ALGORITHM_PARITY_N;
5804 break;
5805 default:
5806 return ERR_PTR(-EINVAL);
5807 }
5808 mddev->new_level = 6;
5809 mddev->new_layout = new_layout;
5810 mddev->delta_disks = 1;
5811 mddev->raid_disks += 1;
5812 return setup_conf(mddev);
5813}
5814
5815
NeilBrown16a53ec2006-06-26 00:27:38 -07005816static struct mdk_personality raid6_personality =
5817{
5818 .name = "raid6",
5819 .level = 6,
5820 .owner = THIS_MODULE,
5821 .make_request = make_request,
5822 .run = run,
5823 .stop = stop,
5824 .status = status,
5825 .error_handler = error,
5826 .hot_add_disk = raid5_add_disk,
5827 .hot_remove_disk= raid5_remove_disk,
5828 .spare_active = raid5_spare_active,
5829 .sync_request = sync_request,
5830 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005831 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10005832 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08005833 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005834 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07005835 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11005836 .takeover = raid6_takeover,
NeilBrown16a53ec2006-06-26 00:27:38 -07005837};
NeilBrown2604b702006-01-06 00:20:36 -08005838static struct mdk_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005839{
5840 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08005841 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005842 .owner = THIS_MODULE,
5843 .make_request = make_request,
5844 .run = run,
5845 .stop = stop,
5846 .status = status,
5847 .error_handler = error,
5848 .hot_add_disk = raid5_add_disk,
5849 .hot_remove_disk= raid5_remove_disk,
5850 .spare_active = raid5_spare_active,
5851 .sync_request = sync_request,
5852 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005853 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08005854 .check_reshape = raid5_check_reshape,
5855 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005856 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07005857 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11005858 .takeover = raid5_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005859};
5860
NeilBrown2604b702006-01-06 00:20:36 -08005861static struct mdk_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005862{
NeilBrown2604b702006-01-06 00:20:36 -08005863 .name = "raid4",
5864 .level = 4,
5865 .owner = THIS_MODULE,
5866 .make_request = make_request,
5867 .run = run,
5868 .stop = stop,
5869 .status = status,
5870 .error_handler = error,
5871 .hot_add_disk = raid5_add_disk,
5872 .hot_remove_disk= raid5_remove_disk,
5873 .spare_active = raid5_spare_active,
5874 .sync_request = sync_request,
5875 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005876 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08005877 .check_reshape = raid5_check_reshape,
5878 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005879 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08005880 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11005881 .takeover = raid4_takeover,
NeilBrown2604b702006-01-06 00:20:36 -08005882};
5883
5884static int __init raid5_init(void)
5885{
NeilBrown16a53ec2006-06-26 00:27:38 -07005886 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005887 register_md_personality(&raid5_personality);
5888 register_md_personality(&raid4_personality);
5889 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005890}
5891
NeilBrown2604b702006-01-06 00:20:36 -08005892static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005893{
NeilBrown16a53ec2006-06-26 00:27:38 -07005894 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005895 unregister_md_personality(&raid5_personality);
5896 unregister_md_personality(&raid4_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005897}
5898
5899module_init(raid5_init);
5900module_exit(raid5_exit);
5901MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11005902MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005903MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08005904MODULE_ALIAS("md-raid5");
5905MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08005906MODULE_ALIAS("md-level-5");
5907MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07005908MODULE_ALIAS("md-personality-8"); /* RAID6 */
5909MODULE_ALIAS("md-raid6");
5910MODULE_ALIAS("md-level-6");
5911
5912/* This used to be two separate modules, they were: */
5913MODULE_ALIAS("raid5");
5914MODULE_ALIAS("raid6");