blob: 793dd76aeae045058efb9235d45fcd2b4d44eeeb [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
NeilBrowncc940152011-07-26 11:35:35 +10003001static void 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);
NeilBrowncc940152011-07-26 11:35:35 +10003083 return;
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);
Dan Williamse89f8962007-01-02 13:52:31 -07003179
NeilBrowncc940152011-07-26 11:35:35 +10003180 if (s->syncing && s->locked == 0
3181 && test_bit(STRIPE_INSYNC, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3183 clear_bit(STRIPE_SYNCING, &sh->state);
3184 }
NeilBrown4e5314b2005-11-08 21:39:22 -08003185
3186 /* If the failed drive is just a ReadError, then we might need to progress
3187 * the repair/check process
3188 */
NeilBrowncc940152011-07-26 11:35:35 +10003189 if (s->failed == 1 && !conf->mddev->ro &&
3190 test_bit(R5_ReadError, &sh->dev[s->failed_num[0]].flags)
3191 && !test_bit(R5_LOCKED, &sh->dev[s->failed_num[0]].flags)
3192 && test_bit(R5_UPTODATE, &sh->dev[s->failed_num[0]].flags)
NeilBrown4e5314b2005-11-08 21:39:22 -08003193 ) {
NeilBrowncc940152011-07-26 11:35:35 +10003194 dev = &sh->dev[s->failed_num[0]];
NeilBrown4e5314b2005-11-08 21:39:22 -08003195 if (!test_bit(R5_ReWrite, &dev->flags)) {
3196 set_bit(R5_Wantwrite, &dev->flags);
3197 set_bit(R5_ReWrite, &dev->flags);
3198 set_bit(R5_LOCKED, &dev->flags);
NeilBrowncc940152011-07-26 11:35:35 +10003199 s->locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003200 } else {
3201 /* let's read it back */
3202 set_bit(R5_Wantread, &dev->flags);
3203 set_bit(R5_LOCKED, &dev->flags);
NeilBrowncc940152011-07-26 11:35:35 +10003204 s->locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003205 }
3206 }
3207
Dan Williams600aa102008-06-28 08:32:05 +10003208 /* Finish reconstruct operations initiated by the expansion process */
3209 if (sh->reconstruct_state == reconstruct_state_result) {
NeilBrownab69ae12009-03-31 15:26:47 +11003210 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003211 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003212 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3213 /* sh cannot be written until sh2 has been read.
3214 * so arrange for sh to be delayed a little
3215 */
3216 set_bit(STRIPE_DELAYED, &sh->state);
3217 set_bit(STRIPE_HANDLE, &sh->state);
3218 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3219 &sh2->state))
3220 atomic_inc(&conf->preread_active_stripes);
3221 release_stripe(sh2);
NeilBrowncc940152011-07-26 11:35:35 +10003222 return;
NeilBrownab69ae12009-03-31 15:26:47 +11003223 }
3224 if (sh2)
3225 release_stripe(sh2);
3226
Dan Williams600aa102008-06-28 08:32:05 +10003227 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamsf0a50d32007-01-02 13:52:31 -07003228 clear_bit(STRIPE_EXPANDING, &sh->state);
Dan Williams23397882008-07-23 20:05:34 -07003229 for (i = conf->raid_disks; i--; ) {
Dan Williamsf0a50d32007-01-02 13:52:31 -07003230 set_bit(R5_Wantwrite, &sh->dev[i].flags);
Dan Williams23397882008-07-23 20:05:34 -07003231 set_bit(R5_LOCKED, &sh->dev[i].flags);
NeilBrowncc940152011-07-26 11:35:35 +10003232 s->locked++;
Dan Williams23397882008-07-23 20:05:34 -07003233 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07003234 }
3235
NeilBrowncc940152011-07-26 11:35:35 +10003236 if (s->expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
Dan Williams600aa102008-06-28 08:32:05 +10003237 !sh->reconstruct_state) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003238 /* Need to write out all blocks after computing parity */
3239 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003240 stripe_set_idx(sh->sector, conf, 0, sh);
NeilBrowncc940152011-07-26 11:35:35 +10003241 schedule_reconstruction(sh, s, 1, 1);
3242 } else if (s->expanded && !sh->reconstruct_state && s->locked == 0) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003243 clear_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrownf6705572006-03-27 01:18:11 -08003244 atomic_dec(&conf->reshape_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003245 wake_up(&conf->wait_for_overlap);
3246 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3247 }
3248
NeilBrowncc940152011-07-26 11:35:35 +10003249 if (s->expanding && s->locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003250 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
NeilBrown86c374b2011-07-27 11:00:36 +10003251 handle_stripe_expansion(conf, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252}
3253
NeilBrowncc940152011-07-26 11:35:35 +10003254static void handle_stripe6(struct stripe_head *sh, struct stripe_head_state *s)
NeilBrown16a53ec2006-06-26 00:27:38 -07003255{
NeilBrownbff61972009-03-31 14:33:13 +11003256 raid5_conf_t *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08003257 int disks = sh->disks;
NeilBrown34e04e82009-03-31 15:10:16 +11003258 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
NeilBrown16a53ec2006-06-26 00:27:38 -07003259 struct r5dev *dev, *pdev, *qdev;
NeilBrown16a53ec2006-06-26 00:27:38 -07003260
NeilBrown16a53ec2006-06-26 00:27:38 -07003261 /* Now to look around and see what can be done */
3262
3263 rcu_read_lock();
NeilBrownc4c16632011-07-26 11:34:20 +10003264 spin_lock_irq(&conf->device_lock);
NeilBrown16a53ec2006-06-26 00:27:38 -07003265 for (i=disks; i--; ) {
3266 mdk_rdev_t *rdev;
3267 dev = &sh->dev[i];
NeilBrown16a53ec2006-06-26 00:27:38 -07003268
Dan Williams45b42332007-07-09 11:56:43 -07003269 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown16a53ec2006-06-26 00:27:38 -07003270 i, dev->flags, dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003271 /* maybe we can reply to a read
3272 *
3273 * new wantfill requests are only permitted while
3274 * ops_complete_biofill is guaranteed to be inactive
3275 */
3276 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3277 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3278 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07003279
3280 /* now count some things */
NeilBrowncc940152011-07-26 11:35:35 +10003281 if (test_bit(R5_LOCKED, &dev->flags))
3282 s->locked++;
3283 if (test_bit(R5_UPTODATE, &dev->flags))
3284 s->uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003285 if (test_bit(R5_Wantcompute, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003286 s->compute++;
3287 BUG_ON(s->compute > 2);
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003288 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003289
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003290 if (test_bit(R5_Wantfill, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003291 s->to_fill++;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003292 } else if (dev->toread)
NeilBrowncc940152011-07-26 11:35:35 +10003293 s->to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003294 if (dev->towrite) {
NeilBrowncc940152011-07-26 11:35:35 +10003295 s->to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003296 if (!test_bit(R5_OVERWRITE, &dev->flags))
NeilBrowncc940152011-07-26 11:35:35 +10003297 s->non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003298 }
Dan Williamsa4456852007-07-09 11:56:43 -07003299 if (dev->written)
NeilBrowncc940152011-07-26 11:35:35 +10003300 s->written++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003301 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrowncc940152011-07-26 11:35:35 +10003302 if (s->blocked_rdev == NULL &&
NeilBrownac4090d2008-08-05 15:54:13 +10003303 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
NeilBrowncc940152011-07-26 11:35:35 +10003304 s->blocked_rdev = rdev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003305 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003306 }
NeilBrown415e72d2010-06-17 17:25:21 +10003307 clear_bit(R5_Insync, &dev->flags);
3308 if (!rdev)
3309 /* Not in-sync */;
3310 else if (test_bit(In_sync, &rdev->flags))
3311 set_bit(R5_Insync, &dev->flags);
3312 else {
3313 /* in sync if before recovery_offset */
3314 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3315 set_bit(R5_Insync, &dev->flags);
3316 }
3317 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003318 /* The ReadError flag will just be confusing now */
3319 clear_bit(R5_ReadError, &dev->flags);
3320 clear_bit(R5_ReWrite, &dev->flags);
3321 }
NeilBrown415e72d2010-06-17 17:25:21 +10003322 if (test_bit(R5_ReadError, &dev->flags))
3323 clear_bit(R5_Insync, &dev->flags);
3324 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrowncc940152011-07-26 11:35:35 +10003325 if (s->failed < 2)
3326 s->failed_num[s->failed] = i;
3327 s->failed++;
NeilBrown415e72d2010-06-17 17:25:21 +10003328 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003329 }
NeilBrownc4c16632011-07-26 11:34:20 +10003330 spin_unlock_irq(&conf->device_lock);
NeilBrown16a53ec2006-06-26 00:27:38 -07003331 rcu_read_unlock();
Dan Williams6bfe0b42008-04-30 00:52:32 -07003332
NeilBrowncc940152011-07-26 11:35:35 +10003333 if (unlikely(s->blocked_rdev)) {
3334 if (s->syncing || s->expanding || s->expanded ||
3335 s->to_write || s->written) {
NeilBrownac4090d2008-08-05 15:54:13 +10003336 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrowncc940152011-07-26 11:35:35 +10003337 return;
NeilBrownac4090d2008-08-05 15:54:13 +10003338 }
3339 /* There is nothing for the blocked_rdev to block */
NeilBrowncc940152011-07-26 11:35:35 +10003340 rdev_dec_pending(s->blocked_rdev, conf->mddev);
3341 s->blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003342 }
NeilBrownac4090d2008-08-05 15:54:13 +10003343
NeilBrowncc940152011-07-26 11:35:35 +10003344 if (s->to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3345 set_bit(STRIPE_OP_BIOFILL, &s->ops_request);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003346 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3347 }
3348
Dan Williams45b42332007-07-09 11:56:43 -07003349 pr_debug("locked=%d uptodate=%d to_read=%d"
NeilBrown16a53ec2006-06-26 00:27:38 -07003350 " to_write=%d failed=%d failed_num=%d,%d\n",
NeilBrowncc940152011-07-26 11:35:35 +10003351 s->locked, s->uptodate, s->to_read, s->to_write, s->failed,
3352 s->failed_num[0], s->failed_num[1]);
Dan Williamsa4456852007-07-09 11:56:43 -07003353 /* check if the array has lost >2 devices and, if so, some requests
3354 * might need to be failed
NeilBrown16a53ec2006-06-26 00:27:38 -07003355 */
NeilBrowncc940152011-07-26 11:35:35 +10003356 if (s->failed > 2 && s->to_read+s->to_write+s->written)
3357 handle_failed_stripe(conf, sh, s, disks, &s->return_bi);
3358 if (s->failed > 2 && s->syncing) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003359 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3360 clear_bit(STRIPE_SYNCING, &sh->state);
NeilBrowncc940152011-07-26 11:35:35 +10003361 s->syncing = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003362 }
3363
3364 /*
3365 * might be able to return some write requests if the parity blocks
3366 * are safe, or on a failed drive
3367 */
3368 pdev = &sh->dev[pd_idx];
NeilBrowncc940152011-07-26 11:35:35 +10003369 s->p_failed = (s->failed >= 1 && s->failed_num[0] == pd_idx)
3370 || (s->failed >= 2 && s->failed_num[1] == pd_idx);
NeilBrown34e04e82009-03-31 15:10:16 +11003371 qdev = &sh->dev[qd_idx];
NeilBrowncc940152011-07-26 11:35:35 +10003372 s->q_failed = (s->failed >= 1 && s->failed_num[0] == qd_idx)
3373 || (s->failed >= 2 && s->failed_num[1] == qd_idx);
NeilBrown16a53ec2006-06-26 00:27:38 -07003374
NeilBrowncc940152011-07-26 11:35:35 +10003375 if (s->written &&
3376 (s->p_failed || ((test_bit(R5_Insync, &pdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003377 && !test_bit(R5_LOCKED, &pdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003378 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
NeilBrowncc940152011-07-26 11:35:35 +10003379 (s->q_failed || ((test_bit(R5_Insync, &qdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003380 && !test_bit(R5_LOCKED, &qdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003381 && test_bit(R5_UPTODATE, &qdev->flags)))))
NeilBrowncc940152011-07-26 11:35:35 +10003382 handle_stripe_clean_event(conf, sh, disks, &s->return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003383
3384 /* Now we might consider reading some blocks, either to check/generate
3385 * parity, or to satisfy requests
3386 * or to load a block that is being partially written.
3387 */
NeilBrowncc940152011-07-26 11:35:35 +10003388 if (s->to_read || s->non_overwrite || (s->to_write && s->failed) ||
3389 (s->syncing && (s->uptodate + s->compute < disks)) || s->expanding)
3390 handle_stripe_fill6(sh, s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003391
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003392 /* Now we check to see if any write operations have recently
3393 * completed
3394 */
3395 if (sh->reconstruct_state == reconstruct_state_drain_result) {
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003396
3397 sh->reconstruct_state = reconstruct_state_idle;
3398 /* All the 'written' buffers and the parity blocks are ready to
3399 * be written back to disk
3400 */
3401 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3402 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags));
3403 for (i = disks; i--; ) {
3404 dev = &sh->dev[i];
3405 if (test_bit(R5_LOCKED, &dev->flags) &&
3406 (i == sh->pd_idx || i == qd_idx ||
3407 dev->written)) {
3408 pr_debug("Writing block %d\n", i);
3409 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3410 set_bit(R5_Wantwrite, &dev->flags);
3411 if (!test_bit(R5_Insync, &dev->flags) ||
3412 ((i == sh->pd_idx || i == qd_idx) &&
NeilBrowncc940152011-07-26 11:35:35 +10003413 s->failed == 0))
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003414 set_bit(STRIPE_INSYNC, &sh->state);
3415 }
3416 }
NeilBrown729a1862009-12-14 12:49:50 +11003417 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
NeilBrowncc940152011-07-26 11:35:35 +10003418 s->dec_preread_active = 1;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003419 }
3420
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07003421 /* Now to consider new write requests and what else, if anything
3422 * should be read. We do not handle new writes when:
3423 * 1/ A 'write' operation (copy+gen_syndrome) is already in flight.
3424 * 2/ A 'check' operation is in flight, as it may clobber the parity
3425 * block.
3426 */
NeilBrowncc940152011-07-26 11:35:35 +10003427 if (s->to_write && !sh->reconstruct_state && !sh->check_state)
3428 handle_stripe_dirtying6(conf, sh, s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003429
3430 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamsa4456852007-07-09 11:56:43 -07003431 * Any reads will already have been scheduled, so we just see if enough
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003432 * data is available. The parity check is held off while parity
3433 * dependent operations are in flight.
NeilBrown16a53ec2006-06-26 00:27:38 -07003434 */
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003435 if (sh->check_state ||
NeilBrowncc940152011-07-26 11:35:35 +10003436 (s->syncing && s->locked == 0 &&
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003437 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3438 !test_bit(STRIPE_INSYNC, &sh->state)))
NeilBrowncc940152011-07-26 11:35:35 +10003439 handle_parity_checks6(conf, sh, s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003440
NeilBrowncc940152011-07-26 11:35:35 +10003441 if (s->syncing && s->locked == 0
3442 && test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003443 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3444 clear_bit(STRIPE_SYNCING, &sh->state);
3445 }
3446
3447 /* If the failed drives are just a ReadError, then we might need
3448 * to progress the repair/check process
3449 */
NeilBrowncc940152011-07-26 11:35:35 +10003450 if (s->failed <= 2 && !conf->mddev->ro)
3451 for (i = 0; i < s->failed; i++) {
3452 dev = &sh->dev[s->failed_num[i]];
NeilBrown16a53ec2006-06-26 00:27:38 -07003453 if (test_bit(R5_ReadError, &dev->flags)
3454 && !test_bit(R5_LOCKED, &dev->flags)
3455 && test_bit(R5_UPTODATE, &dev->flags)
3456 ) {
3457 if (!test_bit(R5_ReWrite, &dev->flags)) {
3458 set_bit(R5_Wantwrite, &dev->flags);
3459 set_bit(R5_ReWrite, &dev->flags);
3460 set_bit(R5_LOCKED, &dev->flags);
NeilBrowncc940152011-07-26 11:35:35 +10003461 s->locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003462 } else {
3463 /* let's read it back */
3464 set_bit(R5_Wantread, &dev->flags);
3465 set_bit(R5_LOCKED, &dev->flags);
NeilBrowncc940152011-07-26 11:35:35 +10003466 s->locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003467 }
3468 }
3469 }
NeilBrownf4168852007-02-28 20:11:53 -08003470
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003471 /* Finish reconstruct operations initiated by the expansion process */
3472 if (sh->reconstruct_state == reconstruct_state_result) {
3473 sh->reconstruct_state = reconstruct_state_idle;
3474 clear_bit(STRIPE_EXPANDING, &sh->state);
3475 for (i = conf->raid_disks; i--; ) {
3476 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3477 set_bit(R5_LOCKED, &sh->dev[i].flags);
NeilBrowncc940152011-07-26 11:35:35 +10003478 s->locked++;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003479 }
3480 }
3481
NeilBrowncc940152011-07-26 11:35:35 +10003482 if (s->expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003483 !sh->reconstruct_state) {
NeilBrownab69ae12009-03-31 15:26:47 +11003484 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003485 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003486 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3487 /* sh cannot be written until sh2 has been read.
3488 * so arrange for sh to be delayed a little
3489 */
3490 set_bit(STRIPE_DELAYED, &sh->state);
3491 set_bit(STRIPE_HANDLE, &sh->state);
3492 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3493 &sh2->state))
3494 atomic_inc(&conf->preread_active_stripes);
3495 release_stripe(sh2);
NeilBrowncc940152011-07-26 11:35:35 +10003496 return;
NeilBrownab69ae12009-03-31 15:26:47 +11003497 }
3498 if (sh2)
3499 release_stripe(sh2);
3500
NeilBrownf4168852007-02-28 20:11:53 -08003501 /* Need to write out all blocks after computing P&Q */
3502 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003503 stripe_set_idx(sh->sector, conf, 0, sh);
NeilBrowncc940152011-07-26 11:35:35 +10003504 schedule_reconstruction(sh, s, 1, 1);
3505 } else if (s->expanded && !sh->reconstruct_state && s->locked == 0) {
NeilBrownf4168852007-02-28 20:11:53 -08003506 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3507 atomic_dec(&conf->reshape_stripes);
3508 wake_up(&conf->wait_for_overlap);
3509 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3510 }
3511
NeilBrowncc940152011-07-26 11:35:35 +10003512 if (s->expanding && s->locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003513 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
NeilBrown86c374b2011-07-27 11:00:36 +10003514 handle_stripe_expansion(conf, sh);
NeilBrowncc940152011-07-26 11:35:35 +10003515}
NeilBrownf4168852007-02-28 20:11:53 -08003516
NeilBrowncc940152011-07-26 11:35:35 +10003517static void handle_stripe(struct stripe_head *sh)
3518{
3519 struct stripe_head_state s;
3520 raid5_conf_t *conf = sh->raid_conf;
3521
3522 clear_bit(STRIPE_HANDLE, &sh->state);
3523 if (test_and_set_bit(STRIPE_ACTIVE, &sh->state)) {
3524 /* already being handled, ensure it gets handled
3525 * again when current action finishes */
3526 set_bit(STRIPE_HANDLE, &sh->state);
3527 return;
3528 }
3529
3530 if (test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) {
3531 set_bit(STRIPE_SYNCING, &sh->state);
3532 clear_bit(STRIPE_INSYNC, &sh->state);
3533 }
3534 clear_bit(STRIPE_DELAYED, &sh->state);
3535
3536 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
3537 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
3538 (unsigned long long)sh->sector, sh->state,
3539 atomic_read(&sh->count), sh->pd_idx, sh->qd_idx,
3540 sh->check_state, sh->reconstruct_state);
3541 memset(&s, 0, sizeof(s));
3542
3543 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3544 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3545 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
3546
3547 if (conf->level == 6)
3548 handle_stripe6(sh, &s);
3549 else
3550 handle_stripe5(sh, &s);
3551
NeilBrown16a53ec2006-06-26 00:27:38 -07003552
Dan Williams6bfe0b42008-04-30 00:52:32 -07003553 /* wait for this device to become unblocked */
NeilBrownc5709ef2011-07-26 11:35:20 +10003554 if (unlikely(s.blocked_rdev))
3555 md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003556
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003557 if (s.ops_request)
3558 raid_run_ops(sh, s.ops_request);
3559
Dan Williamsf0e43bc2008-06-28 08:31:55 +10003560 ops_run_io(sh, &s);
3561
NeilBrown729a1862009-12-14 12:49:50 +11003562
NeilBrownc5709ef2011-07-26 11:35:20 +10003563 if (s.dec_preread_active) {
NeilBrown729a1862009-12-14 12:49:50 +11003564 /* We delay this until after ops_run_io so that if make_request
Tejun Heoe9c74692010-09-03 11:56:18 +02003565 * is waiting on a flush, it won't continue until the writes
NeilBrown729a1862009-12-14 12:49:50 +11003566 * have actually been submitted.
3567 */
3568 atomic_dec(&conf->preread_active_stripes);
3569 if (atomic_read(&conf->preread_active_stripes) <
3570 IO_THRESHOLD)
3571 md_wakeup_thread(conf->mddev->thread);
3572 }
3573
NeilBrownc5709ef2011-07-26 11:35:20 +10003574 return_io(s.return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003575
NeilBrownc4c16632011-07-26 11:34:20 +10003576 clear_bit(STRIPE_ACTIVE, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07003577}
3578
Arjan van de Ven858119e2006-01-14 13:20:43 -08003579static void raid5_activate_delayed(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580{
3581 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3582 while (!list_empty(&conf->delayed_list)) {
3583 struct list_head *l = conf->delayed_list.next;
3584 struct stripe_head *sh;
3585 sh = list_entry(l, struct stripe_head, lru);
3586 list_del_init(l);
3587 clear_bit(STRIPE_DELAYED, &sh->state);
3588 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3589 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003590 list_add_tail(&sh->lru, &conf->hold_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 }
NeilBrown482c0832011-04-18 18:25:42 +10003592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593}
3594
Arjan van de Ven858119e2006-01-14 13:20:43 -08003595static void activate_bit_delay(raid5_conf_t *conf)
NeilBrown72626682005-09-09 16:23:54 -07003596{
3597 /* device_lock is held */
3598 struct list_head head;
3599 list_add(&head, &conf->bitmap_list);
3600 list_del_init(&conf->bitmap_list);
3601 while (!list_empty(&head)) {
3602 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3603 list_del_init(&sh->lru);
3604 atomic_inc(&sh->count);
3605 __release_stripe(conf, sh);
3606 }
3607}
3608
NeilBrown11d8a6e2010-07-26 11:57:07 +10003609int md_raid5_congested(mddev_t *mddev, int bits)
NeilBrownf022b2f2006-10-03 01:15:56 -07003610{
NeilBrown070ec552009-06-16 16:54:21 +10003611 raid5_conf_t *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07003612
3613 /* No difference between reads and writes. Just check
3614 * how busy the stripe_cache is
3615 */
NeilBrown3fa841d2009-09-23 18:10:29 +10003616
NeilBrownf022b2f2006-10-03 01:15:56 -07003617 if (conf->inactive_blocked)
3618 return 1;
3619 if (conf->quiesce)
3620 return 1;
3621 if (list_empty_careful(&conf->inactive_list))
3622 return 1;
3623
3624 return 0;
3625}
NeilBrown11d8a6e2010-07-26 11:57:07 +10003626EXPORT_SYMBOL_GPL(md_raid5_congested);
3627
3628static int raid5_congested(void *data, int bits)
3629{
3630 mddev_t *mddev = data;
3631
3632 return mddev_congested(mddev, bits) ||
3633 md_raid5_congested(mddev, bits);
3634}
NeilBrownf022b2f2006-10-03 01:15:56 -07003635
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003636/* We want read requests to align with chunks where possible,
3637 * but write requests don't need to.
3638 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003639static int raid5_mergeable_bvec(struct request_queue *q,
3640 struct bvec_merge_data *bvm,
3641 struct bio_vec *biovec)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003642{
3643 mddev_t *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003644 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003645 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +10003646 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003647 unsigned int bio_sectors = bvm->bi_size >> 9;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003648
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003649 if ((bvm->bi_rw & 1) == WRITE)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003650 return biovec->bv_len; /* always allow writes to be mergeable */
3651
Andre Noll664e7c42009-06-18 08:45:27 +10003652 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3653 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003654 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3655 if (max < 0) max = 0;
3656 if (max <= biovec->bv_len && bio_sectors == 0)
3657 return biovec->bv_len;
3658 else
3659 return max;
3660}
3661
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003662
3663static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3664{
3665 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
Andre Noll9d8f0362009-06-18 08:45:01 +10003666 unsigned int chunk_sectors = mddev->chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003667 unsigned int bio_sectors = bio->bi_size >> 9;
3668
Andre Noll664e7c42009-06-18 08:45:27 +10003669 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3670 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003671 return chunk_sectors >=
3672 ((sector & (chunk_sectors - 1)) + bio_sectors);
3673}
3674
3675/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003676 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3677 * later sampled by raid5d.
3678 */
3679static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3680{
3681 unsigned long flags;
3682
3683 spin_lock_irqsave(&conf->device_lock, flags);
3684
3685 bi->bi_next = conf->retry_read_aligned_list;
3686 conf->retry_read_aligned_list = bi;
3687
3688 spin_unlock_irqrestore(&conf->device_lock, flags);
3689 md_wakeup_thread(conf->mddev->thread);
3690}
3691
3692
3693static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3694{
3695 struct bio *bi;
3696
3697 bi = conf->retry_read_aligned;
3698 if (bi) {
3699 conf->retry_read_aligned = NULL;
3700 return bi;
3701 }
3702 bi = conf->retry_read_aligned_list;
3703 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08003704 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003705 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02003706 /*
3707 * this sets the active strip count to 1 and the processed
3708 * strip count to zero (upper 8 bits)
3709 */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003710 bi->bi_phys_segments = 1; /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003711 }
3712
3713 return bi;
3714}
3715
3716
3717/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003718 * The "raid5_align_endio" should check if the read succeeded and if it
3719 * did, call bio_endio on the original bio (having bio_put the new bio
3720 * first).
3721 * If the read failed..
3722 */
NeilBrown6712ecf2007-09-27 12:47:43 +02003723static void raid5_align_endio(struct bio *bi, int error)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003724{
3725 struct bio* raid_bi = bi->bi_private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003726 mddev_t *mddev;
3727 raid5_conf_t *conf;
3728 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3729 mdk_rdev_t *rdev;
3730
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003731 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003732
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003733 rdev = (void*)raid_bi->bi_next;
3734 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11003735 mddev = rdev->mddev;
3736 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003737
3738 rdev_dec_pending(rdev, conf->mddev);
3739
3740 if (!error && uptodate) {
NeilBrown6712ecf2007-09-27 12:47:43 +02003741 bio_endio(raid_bi, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003742 if (atomic_dec_and_test(&conf->active_aligned_reads))
3743 wake_up(&conf->wait_for_stripe);
NeilBrown6712ecf2007-09-27 12:47:43 +02003744 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003745 }
3746
3747
Dan Williams45b42332007-07-09 11:56:43 -07003748 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003749
3750 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003751}
3752
Neil Brown387bb172007-02-08 14:20:29 -08003753static int bio_fits_rdev(struct bio *bi)
3754{
Jens Axboe165125e2007-07-24 09:28:11 +02003755 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
Neil Brown387bb172007-02-08 14:20:29 -08003756
Martin K. Petersenae03bf62009-05-22 17:17:50 -04003757 if ((bi->bi_size>>9) > queue_max_sectors(q))
Neil Brown387bb172007-02-08 14:20:29 -08003758 return 0;
3759 blk_recount_segments(q, bi);
Martin K. Petersen8a783622010-02-26 00:20:39 -05003760 if (bi->bi_phys_segments > queue_max_segments(q))
Neil Brown387bb172007-02-08 14:20:29 -08003761 return 0;
3762
3763 if (q->merge_bvec_fn)
3764 /* it's too hard to apply the merge_bvec_fn at this stage,
3765 * just just give up
3766 */
3767 return 0;
3768
3769 return 1;
3770}
3771
3772
NeilBrown21a52c62010-04-01 15:02:13 +11003773static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003774{
NeilBrown070ec552009-06-16 16:54:21 +10003775 raid5_conf_t *conf = mddev->private;
NeilBrown8553fe72009-12-14 12:49:47 +11003776 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003777 struct bio* align_bi;
3778 mdk_rdev_t *rdev;
3779
3780 if (!in_chunk_boundary(mddev, raid_bio)) {
Dan Williams45b42332007-07-09 11:56:43 -07003781 pr_debug("chunk_aligned_read : non aligned\n");
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003782 return 0;
3783 }
3784 /*
NeilBrowna167f662010-10-26 18:31:13 +11003785 * use bio_clone_mddev to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003786 */
NeilBrowna167f662010-10-26 18:31:13 +11003787 align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003788 if (!align_bi)
3789 return 0;
3790 /*
3791 * set bi_end_io to a new function, and set bi_private to the
3792 * original bio.
3793 */
3794 align_bi->bi_end_io = raid5_align_endio;
3795 align_bi->bi_private = raid_bio;
3796 /*
3797 * compute position
3798 */
NeilBrown112bf892009-03-31 14:39:38 +11003799 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3800 0,
NeilBrown911d4ee2009-03-31 14:39:38 +11003801 &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003802
3803 rcu_read_lock();
3804 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3805 if (rdev && test_bit(In_sync, &rdev->flags)) {
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003806 atomic_inc(&rdev->nr_pending);
3807 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003808 raid_bio->bi_next = (void*)rdev;
3809 align_bi->bi_bdev = rdev->bdev;
3810 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3811 align_bi->bi_sector += rdev->data_offset;
3812
Neil Brown387bb172007-02-08 14:20:29 -08003813 if (!bio_fits_rdev(align_bi)) {
3814 /* too big in some way */
3815 bio_put(align_bi);
3816 rdev_dec_pending(rdev, mddev);
3817 return 0;
3818 }
3819
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003820 spin_lock_irq(&conf->device_lock);
3821 wait_event_lock_irq(conf->wait_for_stripe,
3822 conf->quiesce == 0,
3823 conf->device_lock, /* nothing */);
3824 atomic_inc(&conf->active_aligned_reads);
3825 spin_unlock_irq(&conf->device_lock);
3826
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003827 generic_make_request(align_bi);
3828 return 1;
3829 } else {
3830 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003831 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003832 return 0;
3833 }
3834}
3835
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003836/* __get_priority_stripe - get the next stripe to process
3837 *
3838 * Full stripe writes are allowed to pass preread active stripes up until
3839 * the bypass_threshold is exceeded. In general the bypass_count
3840 * increments when the handle_list is handled before the hold_list; however, it
3841 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3842 * stripe with in flight i/o. The bypass_count will be reset when the
3843 * head of the hold_list has changed, i.e. the head was promoted to the
3844 * handle_list.
3845 */
3846static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3847{
3848 struct stripe_head *sh;
3849
3850 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3851 __func__,
3852 list_empty(&conf->handle_list) ? "empty" : "busy",
3853 list_empty(&conf->hold_list) ? "empty" : "busy",
3854 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3855
3856 if (!list_empty(&conf->handle_list)) {
3857 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3858
3859 if (list_empty(&conf->hold_list))
3860 conf->bypass_count = 0;
3861 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3862 if (conf->hold_list.next == conf->last_hold)
3863 conf->bypass_count++;
3864 else {
3865 conf->last_hold = conf->hold_list.next;
3866 conf->bypass_count -= conf->bypass_threshold;
3867 if (conf->bypass_count < 0)
3868 conf->bypass_count = 0;
3869 }
3870 }
3871 } else if (!list_empty(&conf->hold_list) &&
3872 ((conf->bypass_threshold &&
3873 conf->bypass_count > conf->bypass_threshold) ||
3874 atomic_read(&conf->pending_full_writes) == 0)) {
3875 sh = list_entry(conf->hold_list.next,
3876 typeof(*sh), lru);
3877 conf->bypass_count -= conf->bypass_threshold;
3878 if (conf->bypass_count < 0)
3879 conf->bypass_count = 0;
3880 } else
3881 return NULL;
3882
3883 list_del_init(&sh->lru);
3884 atomic_inc(&sh->count);
3885 BUG_ON(atomic_read(&sh->count) != 1);
3886 return sh;
3887}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003888
NeilBrown21a52c62010-04-01 15:02:13 +11003889static int make_request(mddev_t *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003890{
NeilBrown070ec552009-06-16 16:54:21 +10003891 raid5_conf_t *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11003892 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 sector_t new_sector;
3894 sector_t logical_sector, last_sector;
3895 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01003896 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11003897 int remaining;
NeilBrown7c13edc2011-04-18 18:25:43 +10003898 int plugged;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899
Tejun Heoe9c74692010-09-03 11:56:18 +02003900 if (unlikely(bi->bi_rw & REQ_FLUSH)) {
3901 md_flush_request(mddev, bi);
NeilBrowne5dcdd82005-09-09 16:23:41 -07003902 return 0;
3903 }
3904
NeilBrown3d310eb2005-06-21 17:17:26 -07003905 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07003906
NeilBrown802ba062006-12-13 00:34:13 -08003907 if (rw == READ &&
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003908 mddev->reshape_position == MaxSector &&
NeilBrown21a52c62010-04-01 15:02:13 +11003909 chunk_aligned_read(mddev,bi))
NeilBrown99c0fb52009-03-31 14:39:38 +11003910 return 0;
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003911
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3913 last_sector = bi->bi_sector + (bi->bi_size>>9);
3914 bi->bi_next = NULL;
3915 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07003916
NeilBrown7c13edc2011-04-18 18:25:43 +10003917 plugged = mddev_check_plugged(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3919 DEFINE_WAIT(w);
NeilBrown16a53ec2006-06-26 00:27:38 -07003920 int disks, data_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003921 int previous;
NeilBrownb578d552006-03-27 01:18:12 -08003922
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003923 retry:
NeilBrownb5663ba2009-03-31 14:39:38 +11003924 previous = 0;
NeilBrownb0f9ec02009-03-31 15:27:18 +11003925 disks = conf->raid_disks;
NeilBrownb578d552006-03-27 01:18:12 -08003926 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003927 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11003928 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08003929 * 64bit on a 32bit platform, and so it might be
3930 * possible to see a half-updated value
Jesper Juhlaeb878b2011-04-10 18:06:17 +02003931 * Of course reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08003932 * the lock is dropped, so once we get a reference
3933 * to the stripe that we think it is, we will have
3934 * to check again.
3935 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003936 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11003937 if (mddev->delta_disks < 0
3938 ? logical_sector < conf->reshape_progress
3939 : logical_sector >= conf->reshape_progress) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003940 disks = conf->previous_raid_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003941 previous = 1;
3942 } else {
NeilBrownfef9c612009-03-31 15:16:46 +11003943 if (mddev->delta_disks < 0
3944 ? logical_sector < conf->reshape_safe
3945 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08003946 spin_unlock_irq(&conf->device_lock);
3947 schedule();
3948 goto retry;
3949 }
3950 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003951 spin_unlock_irq(&conf->device_lock);
3952 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003953 data_disks = disks - conf->max_degraded;
3954
NeilBrown112bf892009-03-31 14:39:38 +11003955 new_sector = raid5_compute_sector(conf, logical_sector,
3956 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11003957 &dd_idx, NULL);
NeilBrown0c55e022010-05-03 14:09:02 +10003958 pr_debug("raid456: make_request, sector %llu logical %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 (unsigned long long)new_sector,
3960 (unsigned long long)logical_sector);
3961
NeilBrownb5663ba2009-03-31 14:39:38 +11003962 sh = get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10003963 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11003965 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003966 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08003967 * stripe, so we must do the range check again.
3968 * Expansion could still move past after this
3969 * test, but as we are holding a reference to
3970 * 'sh', we know that if that happens,
3971 * STRIPE_EXPANDING will get set and the expansion
3972 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003973 */
3974 int must_retry = 0;
3975 spin_lock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003976 if (mddev->delta_disks < 0
3977 ? logical_sector >= conf->reshape_progress
3978 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003979 /* mismatch, need to try again */
3980 must_retry = 1;
3981 spin_unlock_irq(&conf->device_lock);
3982 if (must_retry) {
3983 release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07003984 schedule();
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003985 goto retry;
3986 }
3987 }
NeilBrowne62e58a2009-07-01 13:15:35 +10003988
Namhyung Kimffd96e32011-07-18 17:38:51 +10003989 if (rw == WRITE &&
NeilBrowna5c308d2009-07-01 13:15:35 +10003990 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08003991 logical_sector < mddev->suspend_hi) {
3992 release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10003993 /* As the suspend_* range is controlled by
3994 * userspace, we want an interruptible
3995 * wait.
3996 */
3997 flush_signals(current);
3998 prepare_to_wait(&conf->wait_for_overlap,
3999 &w, TASK_INTERRUPTIBLE);
4000 if (logical_sector >= mddev->suspend_lo &&
4001 logical_sector < mddev->suspend_hi)
4002 schedule();
NeilBrowne464eaf2006-03-27 01:18:14 -08004003 goto retry;
4004 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004005
4006 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
Namhyung Kimffd96e32011-07-18 17:38:51 +10004007 !add_stripe_bio(sh, bi, dd_idx, rw)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004008 /* Stripe is busy expanding or
4009 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07004010 * and wait a while
4011 */
NeilBrown482c0832011-04-18 18:25:42 +10004012 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013 release_stripe(sh);
4014 schedule();
4015 goto retry;
4016 }
4017 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown6ed30032008-02-06 01:40:00 -08004018 set_bit(STRIPE_HANDLE, &sh->state);
4019 clear_bit(STRIPE_DELAYED, &sh->state);
Tejun Heoe9c74692010-09-03 11:56:18 +02004020 if ((bi->bi_rw & REQ_SYNC) &&
NeilBrown729a1862009-12-14 12:49:50 +11004021 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4022 atomic_inc(&conf->preread_active_stripes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004023 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024 } else {
4025 /* cannot get stripe for read-ahead, just give-up */
4026 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4027 finish_wait(&conf->wait_for_overlap, &w);
4028 break;
4029 }
4030
4031 }
NeilBrown7c13edc2011-04-18 18:25:43 +10004032 if (!plugged)
4033 md_wakeup_thread(mddev->thread);
4034
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004036 remaining = raid5_dec_bi_phys_segments(bi);
NeilBrownf6344752006-03-27 01:18:17 -08004037 spin_unlock_irq(&conf->device_lock);
4038 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039
NeilBrown16a53ec2006-06-26 00:27:38 -07004040 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02004042
Neil Brown0e13fe232008-06-28 08:31:20 +10004043 bio_endio(bi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044 }
NeilBrown729a1862009-12-14 12:49:50 +11004045
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046 return 0;
4047}
4048
Dan Williamsb522adc2009-03-31 15:00:31 +11004049static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
4050
NeilBrown52c03292006-06-26 00:27:43 -07004051static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052{
NeilBrown52c03292006-06-26 00:27:43 -07004053 /* reshaping is quite different to recovery/resync so it is
4054 * handled quite separately ... here.
4055 *
4056 * On each call to sync_request, we gather one chunk worth of
4057 * destination stripes and flag them as expanding.
4058 * Then we find all the source stripes and request reads.
4059 * As the reads complete, handle_stripe will copy the data
4060 * into the destination stripe and release that stripe.
4061 */
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004062 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08004064 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08004065 int raid_disks = conf->previous_raid_disks;
4066 int data_disks = raid_disks - conf->max_degraded;
4067 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07004068 int i;
4069 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11004070 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11004071 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11004072 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11004073 struct list_head stripes;
NeilBrown52c03292006-06-26 00:27:43 -07004074
NeilBrownfef9c612009-03-31 15:16:46 +11004075 if (sector_nr == 0) {
4076 /* If restarting in the middle, skip the initial sectors */
4077 if (mddev->delta_disks < 0 &&
4078 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4079 sector_nr = raid5_size(mddev, 0, 0)
4080 - conf->reshape_progress;
NeilBrowna6397552009-08-13 10:13:00 +10004081 } else if (mddev->delta_disks >= 0 &&
NeilBrownfef9c612009-03-31 15:16:46 +11004082 conf->reshape_progress > 0)
4083 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004084 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004085 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11004086 mddev->curr_resync_completed = sector_nr;
4087 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11004088 *skipped = 1;
4089 return sector_nr;
4090 }
NeilBrown52c03292006-06-26 00:27:43 -07004091 }
4092
NeilBrown7a661382009-03-31 15:21:40 +11004093 /* We need to process a full chunk at a time.
4094 * If old and new chunk sizes differ, we need to process the
4095 * largest of these
4096 */
Andre Noll664e7c42009-06-18 08:45:27 +10004097 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4098 reshape_sectors = mddev->new_chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004099 else
Andre Noll9d8f0362009-06-18 08:45:01 +10004100 reshape_sectors = mddev->chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004101
NeilBrown52c03292006-06-26 00:27:43 -07004102 /* we update the metadata when there is more than 3Meg
4103 * in the block range (that is rather arbitrary, should
4104 * probably be time based) or when the data about to be
4105 * copied would over-write the source of the data at
4106 * the front of the range.
NeilBrownfef9c612009-03-31 15:16:46 +11004107 * i.e. one new_stripe along from reshape_progress new_maps
4108 * to after where reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07004109 */
NeilBrownfef9c612009-03-31 15:16:46 +11004110 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004111 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11004112 readpos = conf->reshape_progress;
4113 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004114 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08004115 sector_div(safepos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004116 if (mddev->delta_disks < 0) {
NeilBrowned37d832009-05-27 21:39:05 +10004117 writepos -= min_t(sector_t, reshape_sectors, writepos);
NeilBrownc8f517c2009-03-31 15:28:40 +11004118 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004119 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11004120 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004121 writepos += reshape_sectors;
NeilBrowned37d832009-05-27 21:39:05 +10004122 readpos -= min_t(sector_t, reshape_sectors, readpos);
4123 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11004124 }
NeilBrown52c03292006-06-26 00:27:43 -07004125
NeilBrownc8f517c2009-03-31 15:28:40 +11004126 /* 'writepos' is the most advanced device address we might write.
4127 * 'readpos' is the least advanced device address we might read.
4128 * 'safepos' is the least address recorded in the metadata as having
4129 * been reshaped.
4130 * If 'readpos' is behind 'writepos', then there is no way that we can
4131 * ensure safety in the face of a crash - that must be done by userspace
4132 * making a backup of the data. So in that case there is no particular
4133 * rush to update metadata.
4134 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4135 * update the metadata to advance 'safepos' to match 'readpos' so that
4136 * we can be safe in the event of a crash.
4137 * So we insist on updating metadata if safepos is behind writepos and
4138 * readpos is beyond writepos.
4139 * In any case, update the metadata every 10 seconds.
4140 * Maybe that number should be configurable, but I'm not sure it is
4141 * worth it.... maybe it could be a multiple of safemode_delay???
4142 */
NeilBrownfef9c612009-03-31 15:16:46 +11004143 if ((mddev->delta_disks < 0
NeilBrownc8f517c2009-03-31 15:28:40 +11004144 ? (safepos > writepos && readpos < writepos)
4145 : (safepos < writepos && readpos > writepos)) ||
4146 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07004147 /* Cannot proceed until we've updated the superblock... */
4148 wait_event(conf->wait_for_overlap,
4149 atomic_read(&conf->reshape_stripes)==0);
NeilBrownfef9c612009-03-31 15:16:46 +11004150 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004151 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004152 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07004153 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07004154 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07004155 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrown52c03292006-06-26 00:27:43 -07004156 kthread_should_stop());
4157 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004158 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07004159 spin_unlock_irq(&conf->device_lock);
4160 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004161 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07004162 }
4163
NeilBrownec32a2b2009-03-31 15:17:38 +11004164 if (mddev->delta_disks < 0) {
4165 BUG_ON(conf->reshape_progress == 0);
4166 stripe_addr = writepos;
4167 BUG_ON((mddev->dev_sectors &
NeilBrown7a661382009-03-31 15:21:40 +11004168 ~((sector_t)reshape_sectors - 1))
4169 - reshape_sectors - stripe_addr
NeilBrownec32a2b2009-03-31 15:17:38 +11004170 != sector_nr);
4171 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004172 BUG_ON(writepos != sector_nr + reshape_sectors);
NeilBrownec32a2b2009-03-31 15:17:38 +11004173 stripe_addr = sector_nr;
4174 }
NeilBrownab69ae12009-03-31 15:26:47 +11004175 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11004176 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07004177 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10004178 int skipped_disk = 0;
NeilBrowna8c906c2009-06-09 14:39:59 +10004179 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004180 set_bit(STRIPE_EXPANDING, &sh->state);
4181 atomic_inc(&conf->reshape_stripes);
4182 /* If any of this stripe is beyond the end of the old
4183 * array, then we need to zero those blocks
4184 */
4185 for (j=sh->disks; j--;) {
4186 sector_t s;
4187 if (j == sh->pd_idx)
4188 continue;
NeilBrownf4168852007-02-28 20:11:53 -08004189 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11004190 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08004191 continue;
NeilBrown784052e2009-03-31 15:19:07 +11004192 s = compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11004193 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10004194 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07004195 continue;
4196 }
4197 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4198 set_bit(R5_Expanded, &sh->dev[j].flags);
4199 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4200 }
NeilBrowna9f326e2009-09-23 18:06:41 +10004201 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07004202 set_bit(STRIPE_EXPAND_READY, &sh->state);
4203 set_bit(STRIPE_HANDLE, &sh->state);
4204 }
NeilBrownab69ae12009-03-31 15:26:47 +11004205 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07004206 }
4207 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004208 if (mddev->delta_disks < 0)
NeilBrown7a661382009-03-31 15:21:40 +11004209 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11004210 else
NeilBrown7a661382009-03-31 15:21:40 +11004211 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07004212 spin_unlock_irq(&conf->device_lock);
4213 /* Ok, those stripe are ready. We can start scheduling
4214 * reads on the source stripes.
4215 * The source stripes are determined by mapping the first and last
4216 * block on the destination stripes.
4217 */
NeilBrown52c03292006-06-26 00:27:43 -07004218 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11004219 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11004220 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07004221 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10004222 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10004223 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11004224 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11004225 if (last_sector >= mddev->dev_sectors)
4226 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07004227 while (first_sector <= last_sector) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004228 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004229 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4230 set_bit(STRIPE_HANDLE, &sh->state);
4231 release_stripe(sh);
4232 first_sector += STRIPE_SECTORS;
4233 }
NeilBrownab69ae12009-03-31 15:26:47 +11004234 /* Now that the sources are clearly marked, we can release
4235 * the destination stripes
4236 */
4237 while (!list_empty(&stripes)) {
4238 sh = list_entry(stripes.next, struct stripe_head, lru);
4239 list_del_init(&sh->lru);
4240 release_stripe(sh);
4241 }
NeilBrownc6207272008-02-06 01:39:52 -08004242 /* If this takes us to the resync_max point where we have to pause,
4243 * then we need to write out the superblock.
4244 */
NeilBrown7a661382009-03-31 15:21:40 +11004245 sector_nr += reshape_sectors;
NeilBrownc03f6a12009-04-17 11:06:30 +10004246 if ((sector_nr - mddev->curr_resync_completed) * 2
4247 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08004248 /* Cannot proceed until we've updated the superblock... */
4249 wait_event(conf->wait_for_overlap,
4250 atomic_read(&conf->reshape_stripes) == 0);
NeilBrownfef9c612009-03-31 15:16:46 +11004251 mddev->reshape_position = conf->reshape_progress;
NeilBrown75d3da42011-01-14 09:14:34 +11004252 mddev->curr_resync_completed = sector_nr;
NeilBrownc8f517c2009-03-31 15:28:40 +11004253 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08004254 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4255 md_wakeup_thread(mddev->thread);
4256 wait_event(mddev->sb_wait,
4257 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4258 || kthread_should_stop());
4259 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004260 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08004261 spin_unlock_irq(&conf->device_lock);
4262 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004263 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08004264 }
NeilBrown7a661382009-03-31 15:21:40 +11004265 return reshape_sectors;
NeilBrown52c03292006-06-26 00:27:43 -07004266}
4267
4268/* FIXME go_faster isn't used */
4269static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
4270{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004271 raid5_conf_t *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07004272 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11004273 sector_t max_sector = mddev->dev_sectors;
NeilBrown57dab0b2010-10-19 10:03:39 +11004274 sector_t sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07004275 int still_degraded = 0;
4276 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004277
NeilBrown72626682005-09-09 16:23:54 -07004278 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279 /* just being told to finish up .. nothing much to do */
NeilBrowncea9c222009-03-31 15:15:05 +11004280
NeilBrown29269552006-03-27 01:18:10 -08004281 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4282 end_reshape(conf);
4283 return 0;
4284 }
NeilBrown72626682005-09-09 16:23:54 -07004285
4286 if (mddev->curr_resync < max_sector) /* aborted */
4287 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4288 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07004289 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07004290 conf->fullsync = 0;
4291 bitmap_close_sync(mddev->bitmap);
4292
Linus Torvalds1da177e2005-04-16 15:20:36 -07004293 return 0;
4294 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08004295
NeilBrown64bd6602009-08-03 10:59:58 +10004296 /* Allow raid5_quiesce to complete */
4297 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4298
NeilBrown52c03292006-06-26 00:27:43 -07004299 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4300 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08004301
NeilBrownc6207272008-02-06 01:39:52 -08004302 /* No need to check resync_max as we never do more than one
4303 * stripe, and as resync_max will always be on a chunk boundary,
4304 * if the check in md_do_sync didn't fire, there is no chance
4305 * of overstepping resync_max here
4306 */
4307
NeilBrown16a53ec2006-06-26 00:27:38 -07004308 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07004309 * to resync, then assert that we are finished, because there is
4310 * nothing we can do.
4311 */
NeilBrown3285edf2006-06-26 00:27:55 -07004312 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07004313 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004314 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07004315 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316 return rv;
4317 }
NeilBrown72626682005-09-09 16:23:54 -07004318 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrown3855ad92005-11-08 21:39:38 -08004319 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown72626682005-09-09 16:23:54 -07004320 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4321 /* we can skip this block, and probably more */
4322 sync_blocks /= STRIPE_SECTORS;
4323 *skipped = 1;
4324 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004326
NeilBrownb47490c2008-02-06 01:39:50 -08004327
4328 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4329
NeilBrowna8c906c2009-06-09 14:39:59 +10004330 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004331 if (sh == NULL) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004332 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07004334 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08004336 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004338 /* Need to check if array will still be degraded after recovery/resync
4339 * We don't need to check the 'failed' flag as when that gets set,
4340 * recovery aborts.
4341 */
NeilBrownf001a702009-06-09 14:30:31 +10004342 for (i = 0; i < conf->raid_disks; i++)
NeilBrown16a53ec2006-06-26 00:27:38 -07004343 if (conf->disks[i].rdev == NULL)
4344 still_degraded = 1;
4345
4346 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4347
NeilBrown83206d62011-07-26 11:19:49 +10004348 set_bit(STRIPE_SYNC_REQUESTED, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004349
NeilBrown14425772009-10-16 15:55:25 +11004350 handle_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004351 release_stripe(sh);
4352
4353 return STRIPE_SECTORS;
4354}
4355
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004356static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4357{
4358 /* We may not be able to submit a whole bio at once as there
4359 * may not be enough stripe_heads available.
4360 * We cannot pre-allocate enough stripe_heads as we may need
4361 * more than exist in the cache (if we allow ever large chunks).
4362 * So we do one stripe head at a time and record in
4363 * ->bi_hw_segments how many have been done.
4364 *
4365 * We *know* that this entire raid_bio is in one chunk, so
4366 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4367 */
4368 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11004369 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004370 sector_t sector, logical_sector, last_sector;
4371 int scnt = 0;
4372 int remaining;
4373 int handled = 0;
4374
4375 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11004376 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11004377 0, &dd_idx, NULL);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004378 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4379
4380 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08004381 logical_sector += STRIPE_SECTORS,
4382 sector += STRIPE_SECTORS,
4383 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004384
Jens Axboe960e7392008-08-15 10:41:18 +02004385 if (scnt < raid5_bi_hw_segments(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004386 /* already done this stripe */
4387 continue;
4388
NeilBrowna8c906c2009-06-09 14:39:59 +10004389 sh = get_active_stripe(conf, sector, 0, 1, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004390
4391 if (!sh) {
4392 /* failed to get a stripe - must wait */
Jens Axboe960e7392008-08-15 10:41:18 +02004393 raid5_set_bi_hw_segments(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004394 conf->retry_read_aligned = raid_bio;
4395 return handled;
4396 }
4397
4398 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
Neil Brown387bb172007-02-08 14:20:29 -08004399 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4400 release_stripe(sh);
Jens Axboe960e7392008-08-15 10:41:18 +02004401 raid5_set_bi_hw_segments(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08004402 conf->retry_read_aligned = raid_bio;
4403 return handled;
4404 }
4405
Dan Williams36d1c642009-07-14 11:48:22 -07004406 handle_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004407 release_stripe(sh);
4408 handled++;
4409 }
4410 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004411 remaining = raid5_dec_bi_phys_segments(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004412 spin_unlock_irq(&conf->device_lock);
Neil Brown0e13fe232008-06-28 08:31:20 +10004413 if (remaining == 0)
4414 bio_endio(raid_bio, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004415 if (atomic_dec_and_test(&conf->active_aligned_reads))
4416 wake_up(&conf->wait_for_stripe);
4417 return handled;
4418}
4419
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004420
Linus Torvalds1da177e2005-04-16 15:20:36 -07004421/*
4422 * This is our raid5 kernel thread.
4423 *
4424 * We scan the hash table for stripes which can be handled now.
4425 * During the scan, completed stripes are saved for us by the interrupt
4426 * handler, so that they will not have to wait for our next wakeup.
4427 */
NeilBrown6ed30032008-02-06 01:40:00 -08004428static void raid5d(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004429{
4430 struct stripe_head *sh;
NeilBrown070ec552009-06-16 16:54:21 +10004431 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004432 int handled;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004433 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434
Dan Williams45b42332007-07-09 11:56:43 -07004435 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004436
4437 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004439 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004440 handled = 0;
4441 spin_lock_irq(&conf->device_lock);
4442 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004443 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004444
NeilBrown7c13edc2011-04-18 18:25:43 +10004445 if (atomic_read(&mddev->plug_cnt) == 0 &&
4446 !list_empty(&conf->bitmap_list)) {
4447 /* Now is a good time to flush some bitmap updates */
4448 conf->seq_flush++;
NeilBrown700e4322005-11-28 13:44:10 -08004449 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07004450 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08004451 spin_lock_irq(&conf->device_lock);
NeilBrown7c13edc2011-04-18 18:25:43 +10004452 conf->seq_write = conf->seq_flush;
NeilBrown72626682005-09-09 16:23:54 -07004453 activate_bit_delay(conf);
4454 }
NeilBrown7c13edc2011-04-18 18:25:43 +10004455 if (atomic_read(&mddev->plug_cnt) == 0)
4456 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07004457
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004458 while ((bio = remove_bio_from_retry(conf))) {
4459 int ok;
4460 spin_unlock_irq(&conf->device_lock);
4461 ok = retry_aligned_read(conf, bio);
4462 spin_lock_irq(&conf->device_lock);
4463 if (!ok)
4464 break;
4465 handled++;
4466 }
4467
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004468 sh = __get_priority_stripe(conf);
4469
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004470 if (!sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004472 spin_unlock_irq(&conf->device_lock);
4473
4474 handled++;
Dan Williams417b8d42009-10-16 16:25:22 +11004475 handle_stripe(sh);
4476 release_stripe(sh);
4477 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004478
4479 spin_lock_irq(&conf->device_lock);
4480 }
Dan Williams45b42332007-07-09 11:56:43 -07004481 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004482
4483 spin_unlock_irq(&conf->device_lock);
4484
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004485 async_tx_issue_pending_all();
NeilBrowne1dfa0a2011-04-18 18:25:41 +10004486 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004487
Dan Williams45b42332007-07-09 11:56:43 -07004488 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004489}
4490
NeilBrown3f294f42005-11-08 21:39:25 -08004491static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004492raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004493{
NeilBrown070ec552009-06-16 16:54:21 +10004494 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004495 if (conf)
4496 return sprintf(page, "%d\n", conf->max_nr_stripes);
4497 else
4498 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004499}
4500
NeilBrownc41d4ac2010-06-01 19:37:24 +10004501int
4502raid5_set_cache_size(mddev_t *mddev, int size)
4503{
4504 raid5_conf_t *conf = mddev->private;
4505 int err;
4506
4507 if (size <= 16 || size > 32768)
4508 return -EINVAL;
4509 while (size < conf->max_nr_stripes) {
4510 if (drop_one_stripe(conf))
4511 conf->max_nr_stripes--;
4512 else
4513 break;
4514 }
4515 err = md_allow_write(mddev);
4516 if (err)
4517 return err;
4518 while (size > conf->max_nr_stripes) {
4519 if (grow_one_stripe(conf))
4520 conf->max_nr_stripes++;
4521 else break;
4522 }
4523 return 0;
4524}
4525EXPORT_SYMBOL(raid5_set_cache_size);
4526
NeilBrown3f294f42005-11-08 21:39:25 -08004527static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004528raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08004529{
NeilBrown070ec552009-06-16 16:54:21 +10004530 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004531 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07004532 int err;
4533
NeilBrown3f294f42005-11-08 21:39:25 -08004534 if (len >= PAGE_SIZE)
4535 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08004536 if (!conf)
4537 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08004538
Dan Williams4ef197d82008-04-28 02:15:54 -07004539 if (strict_strtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08004540 return -EINVAL;
NeilBrownc41d4ac2010-06-01 19:37:24 +10004541 err = raid5_set_cache_size(mddev, new);
Dan Williamsb5470dc2008-06-27 21:44:04 -07004542 if (err)
4543 return err;
NeilBrown3f294f42005-11-08 21:39:25 -08004544 return len;
4545}
NeilBrown007583c2005-11-08 21:39:30 -08004546
NeilBrown96de1e62005-11-08 21:39:39 -08004547static struct md_sysfs_entry
4548raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4549 raid5_show_stripe_cache_size,
4550 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08004551
4552static ssize_t
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004553raid5_show_preread_threshold(mddev_t *mddev, char *page)
4554{
NeilBrown070ec552009-06-16 16:54:21 +10004555 raid5_conf_t *conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004556 if (conf)
4557 return sprintf(page, "%d\n", conf->bypass_threshold);
4558 else
4559 return 0;
4560}
4561
4562static ssize_t
4563raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4564{
NeilBrown070ec552009-06-16 16:54:21 +10004565 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004566 unsigned long new;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004567 if (len >= PAGE_SIZE)
4568 return -EINVAL;
4569 if (!conf)
4570 return -ENODEV;
4571
Dan Williams4ef197d82008-04-28 02:15:54 -07004572 if (strict_strtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004573 return -EINVAL;
Dan Williams4ef197d82008-04-28 02:15:54 -07004574 if (new > conf->max_nr_stripes)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004575 return -EINVAL;
4576 conf->bypass_threshold = new;
4577 return len;
4578}
4579
4580static struct md_sysfs_entry
4581raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4582 S_IRUGO | S_IWUSR,
4583 raid5_show_preread_threshold,
4584 raid5_store_preread_threshold);
4585
4586static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08004587stripe_cache_active_show(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004588{
NeilBrown070ec552009-06-16 16:54:21 +10004589 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004590 if (conf)
4591 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4592 else
4593 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004594}
4595
NeilBrown96de1e62005-11-08 21:39:39 -08004596static struct md_sysfs_entry
4597raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08004598
NeilBrown007583c2005-11-08 21:39:30 -08004599static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08004600 &raid5_stripecache_size.attr,
4601 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004602 &raid5_preread_bypass_threshold.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08004603 NULL,
4604};
NeilBrown007583c2005-11-08 21:39:30 -08004605static struct attribute_group raid5_attrs_group = {
4606 .name = NULL,
4607 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08004608};
4609
Dan Williams80c3a6c2009-03-17 18:10:40 -07004610static sector_t
4611raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4612{
NeilBrown070ec552009-06-16 16:54:21 +10004613 raid5_conf_t *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07004614
4615 if (!sectors)
4616 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004617 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11004618 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11004619 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004620
Andre Noll9d8f0362009-06-18 08:45:01 +10004621 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Andre Noll664e7c42009-06-18 08:45:27 +10004622 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004623 return sectors * (raid_disks - conf->max_degraded);
4624}
4625
Dan Williams36d1c642009-07-14 11:48:22 -07004626static void raid5_free_percpu(raid5_conf_t *conf)
4627{
4628 struct raid5_percpu *percpu;
4629 unsigned long cpu;
4630
4631 if (!conf->percpu)
4632 return;
4633
4634 get_online_cpus();
4635 for_each_possible_cpu(cpu) {
4636 percpu = per_cpu_ptr(conf->percpu, cpu);
4637 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004638 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004639 }
4640#ifdef CONFIG_HOTPLUG_CPU
4641 unregister_cpu_notifier(&conf->cpu_notify);
4642#endif
4643 put_online_cpus();
4644
4645 free_percpu(conf->percpu);
4646}
4647
Dan Williams95fc17a2009-07-31 12:39:15 +10004648static void free_conf(raid5_conf_t *conf)
4649{
4650 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07004651 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10004652 kfree(conf->disks);
4653 kfree(conf->stripe_hashtbl);
4654 kfree(conf);
4655}
4656
Dan Williams36d1c642009-07-14 11:48:22 -07004657#ifdef CONFIG_HOTPLUG_CPU
4658static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4659 void *hcpu)
4660{
4661 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4662 long cpu = (long)hcpu;
4663 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4664
4665 switch (action) {
4666 case CPU_UP_PREPARE:
4667 case CPU_UP_PREPARE_FROZEN:
Dan Williamsd6f38f32009-07-14 11:50:52 -07004668 if (conf->level == 6 && !percpu->spare_page)
Dan Williams36d1c642009-07-14 11:48:22 -07004669 percpu->spare_page = alloc_page(GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004670 if (!percpu->scribble)
4671 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4672
4673 if (!percpu->scribble ||
4674 (conf->level == 6 && !percpu->spare_page)) {
4675 safe_put_page(percpu->spare_page);
4676 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004677 pr_err("%s: failed memory allocation for cpu%ld\n",
4678 __func__, cpu);
Akinobu Mita55af6bb2010-05-26 14:43:35 -07004679 return notifier_from_errno(-ENOMEM);
Dan Williams36d1c642009-07-14 11:48:22 -07004680 }
4681 break;
4682 case CPU_DEAD:
4683 case CPU_DEAD_FROZEN:
4684 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004685 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004686 percpu->spare_page = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004687 percpu->scribble = NULL;
Dan Williams36d1c642009-07-14 11:48:22 -07004688 break;
4689 default:
4690 break;
4691 }
4692 return NOTIFY_OK;
4693}
4694#endif
4695
4696static int raid5_alloc_percpu(raid5_conf_t *conf)
4697{
4698 unsigned long cpu;
4699 struct page *spare_page;
Tejun Heoa29d8b82010-02-02 14:39:15 +09004700 struct raid5_percpu __percpu *allcpus;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004701 void *scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004702 int err;
4703
Dan Williams36d1c642009-07-14 11:48:22 -07004704 allcpus = alloc_percpu(struct raid5_percpu);
4705 if (!allcpus)
4706 return -ENOMEM;
4707 conf->percpu = allcpus;
4708
4709 get_online_cpus();
4710 err = 0;
4711 for_each_present_cpu(cpu) {
Dan Williamsd6f38f32009-07-14 11:50:52 -07004712 if (conf->level == 6) {
4713 spare_page = alloc_page(GFP_KERNEL);
4714 if (!spare_page) {
4715 err = -ENOMEM;
4716 break;
4717 }
4718 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4719 }
NeilBrown5e5e3e72009-10-16 16:35:30 +11004720 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004721 if (!scribble) {
Dan Williams36d1c642009-07-14 11:48:22 -07004722 err = -ENOMEM;
4723 break;
4724 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07004725 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004726 }
4727#ifdef CONFIG_HOTPLUG_CPU
4728 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4729 conf->cpu_notify.priority = 0;
4730 if (err == 0)
4731 err = register_cpu_notifier(&conf->cpu_notify);
4732#endif
4733 put_online_cpus();
4734
4735 return err;
4736}
4737
NeilBrown91adb562009-03-31 14:39:39 +11004738static raid5_conf_t *setup_conf(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004739{
4740 raid5_conf_t *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004741 int raid_disk, memory, max_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004742 mdk_rdev_t *rdev;
4743 struct disk_info *disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004744
NeilBrown91adb562009-03-31 14:39:39 +11004745 if (mddev->new_level != 5
4746 && mddev->new_level != 4
4747 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10004748 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004749 mdname(mddev), mddev->new_level);
4750 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004751 }
NeilBrown91adb562009-03-31 14:39:39 +11004752 if ((mddev->new_level == 5
4753 && !algorithm_valid_raid5(mddev->new_layout)) ||
4754 (mddev->new_level == 6
4755 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004756 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11004757 mdname(mddev), mddev->new_layout);
4758 return ERR_PTR(-EIO);
4759 }
4760 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10004761 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004762 mdname(mddev), mddev->raid_disks);
4763 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11004764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004765
Andre Noll664e7c42009-06-18 08:45:27 +10004766 if (!mddev->new_chunk_sectors ||
4767 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4768 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10004769 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4770 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11004771 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11004772 }
4773
NeilBrown91adb562009-03-31 14:39:39 +11004774 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4775 if (conf == NULL)
4776 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11004777 spin_lock_init(&conf->device_lock);
4778 init_waitqueue_head(&conf->wait_for_stripe);
4779 init_waitqueue_head(&conf->wait_for_overlap);
4780 INIT_LIST_HEAD(&conf->handle_list);
4781 INIT_LIST_HEAD(&conf->hold_list);
4782 INIT_LIST_HEAD(&conf->delayed_list);
4783 INIT_LIST_HEAD(&conf->bitmap_list);
4784 INIT_LIST_HEAD(&conf->inactive_list);
4785 atomic_set(&conf->active_stripes, 0);
4786 atomic_set(&conf->preread_active_stripes, 0);
4787 atomic_set(&conf->active_aligned_reads, 0);
4788 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrown91adb562009-03-31 14:39:39 +11004789
4790 conf->raid_disks = mddev->raid_disks;
4791 if (mddev->reshape_position == MaxSector)
4792 conf->previous_raid_disks = mddev->raid_disks;
4793 else
4794 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004795 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4796 conf->scribble_len = scribble_len(max_disks);
NeilBrown91adb562009-03-31 14:39:39 +11004797
NeilBrown5e5e3e72009-10-16 16:35:30 +11004798 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11004799 GFP_KERNEL);
4800 if (!conf->disks)
4801 goto abort;
4802
4803 conf->mddev = mddev;
4804
4805 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
4806 goto abort;
4807
Dan Williams36d1c642009-07-14 11:48:22 -07004808 conf->level = mddev->new_level;
4809 if (raid5_alloc_percpu(conf) != 0)
4810 goto abort;
4811
NeilBrown0c55e022010-05-03 14:09:02 +10004812 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11004813
4814 list_for_each_entry(rdev, &mddev->disks, same_set) {
4815 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004816 if (raid_disk >= max_disks
NeilBrown91adb562009-03-31 14:39:39 +11004817 || raid_disk < 0)
4818 continue;
4819 disk = conf->disks + raid_disk;
4820
4821 disk->rdev = rdev;
4822
4823 if (test_bit(In_sync, &rdev->flags)) {
4824 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10004825 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4826 " disk %d\n",
4827 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
Jonathan Brassowd6b212f2011-06-08 18:00:28 -05004828 } else if (rdev->saved_raid_disk != raid_disk)
NeilBrown91adb562009-03-31 14:39:39 +11004829 /* Cannot rely on bitmap to complete recovery */
4830 conf->fullsync = 1;
4831 }
4832
Andre Noll09c9e5f2009-06-18 08:45:55 +10004833 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown91adb562009-03-31 14:39:39 +11004834 conf->level = mddev->new_level;
4835 if (conf->level == 6)
4836 conf->max_degraded = 2;
4837 else
4838 conf->max_degraded = 1;
4839 conf->algorithm = mddev->new_layout;
4840 conf->max_nr_stripes = NR_STRIPES;
NeilBrownfef9c612009-03-31 15:16:46 +11004841 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11004842 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10004843 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11004844 conf->prev_algo = mddev->layout;
4845 }
NeilBrown91adb562009-03-31 14:39:39 +11004846
4847 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11004848 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
NeilBrown91adb562009-03-31 14:39:39 +11004849 if (grow_stripes(conf, conf->max_nr_stripes)) {
4850 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004851 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4852 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004853 goto abort;
4854 } else
NeilBrown0c55e022010-05-03 14:09:02 +10004855 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4856 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004857
NeilBrown0da3c612009-09-23 18:09:45 +10004858 conf->thread = md_register_thread(raid5d, mddev, NULL);
NeilBrown91adb562009-03-31 14:39:39 +11004859 if (!conf->thread) {
4860 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004861 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11004862 mdname(mddev));
4863 goto abort;
4864 }
4865
4866 return conf;
4867
4868 abort:
4869 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10004870 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11004871 return ERR_PTR(-EIO);
4872 } else
4873 return ERR_PTR(-ENOMEM);
4874}
4875
NeilBrownc148ffd2009-11-13 17:47:00 +11004876
4877static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4878{
4879 switch (algo) {
4880 case ALGORITHM_PARITY_0:
4881 if (raid_disk < max_degraded)
4882 return 1;
4883 break;
4884 case ALGORITHM_PARITY_N:
4885 if (raid_disk >= raid_disks - max_degraded)
4886 return 1;
4887 break;
4888 case ALGORITHM_PARITY_0_6:
4889 if (raid_disk == 0 ||
4890 raid_disk == raid_disks - 1)
4891 return 1;
4892 break;
4893 case ALGORITHM_LEFT_ASYMMETRIC_6:
4894 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4895 case ALGORITHM_LEFT_SYMMETRIC_6:
4896 case ALGORITHM_RIGHT_SYMMETRIC_6:
4897 if (raid_disk == raid_disks - 1)
4898 return 1;
4899 }
4900 return 0;
4901}
4902
NeilBrown91adb562009-03-31 14:39:39 +11004903static int run(mddev_t *mddev)
4904{
4905 raid5_conf_t *conf;
NeilBrown9f7c2222010-07-26 12:04:13 +10004906 int working_disks = 0;
NeilBrownc148ffd2009-11-13 17:47:00 +11004907 int dirty_parity_disks = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004908 mdk_rdev_t *rdev;
NeilBrownc148ffd2009-11-13 17:47:00 +11004909 sector_t reshape_offset = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004910
Andre Noll8c6ac862009-06-18 08:48:06 +10004911 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10004912 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10004913 " -- starting background reconstruction\n",
4914 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004915 if (mddev->reshape_position != MaxSector) {
4916 /* Check that we can continue the reshape.
4917 * Currently only disks can change, it must
4918 * increase, and we must be past the point where
4919 * a stripe over-writes itself
4920 */
4921 sector_t here_new, here_old;
4922 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11004923 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrownf6705572006-03-27 01:18:11 -08004924
NeilBrown88ce4932009-03-31 15:24:23 +11004925 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10004926 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08004927 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08004928 mdname(mddev));
4929 return -EINVAL;
4930 }
NeilBrownf6705572006-03-27 01:18:11 -08004931 old_disks = mddev->raid_disks - mddev->delta_disks;
4932 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08004933 * further up in new geometry must map after here in old
4934 * geometry.
NeilBrownf6705572006-03-27 01:18:11 -08004935 */
4936 here_new = mddev->reshape_position;
Andre Noll664e7c42009-06-18 08:45:27 +10004937 if (sector_div(here_new, mddev->new_chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004938 (mddev->raid_disks - max_degraded))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004939 printk(KERN_ERR "md/raid:%s: reshape_position not "
4940 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004941 return -EINVAL;
4942 }
NeilBrownc148ffd2009-11-13 17:47:00 +11004943 reshape_offset = here_new * mddev->new_chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08004944 /* here_new is the stripe we will write to */
4945 here_old = mddev->reshape_position;
Andre Noll9d8f0362009-06-18 08:45:01 +10004946 sector_div(here_old, mddev->chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004947 (old_disks-max_degraded));
4948 /* here_old is the first stripe that we might need to read
4949 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10004950 if (mddev->delta_disks == 0) {
4951 /* We cannot be sure it is safe to start an in-place
4952 * reshape. It is only safe if user-space if monitoring
4953 * and taking constant backups.
4954 * mdadm always starts a situation like this in
4955 * readonly mode so it can take control before
4956 * allowing any writes. So just check for that.
4957 */
4958 if ((here_new * mddev->new_chunk_sectors !=
4959 here_old * mddev->chunk_sectors) ||
4960 mddev->ro == 0) {
NeilBrown0c55e022010-05-03 14:09:02 +10004961 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
4962 " in read-only mode - aborting\n",
4963 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10004964 return -EINVAL;
4965 }
4966 } else if (mddev->delta_disks < 0
4967 ? (here_new * mddev->new_chunk_sectors <=
4968 here_old * mddev->chunk_sectors)
4969 : (here_new * mddev->new_chunk_sectors >=
4970 here_old * mddev->chunk_sectors)) {
NeilBrownf6705572006-03-27 01:18:11 -08004971 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10004972 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
4973 "auto-recovery - aborting.\n",
4974 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004975 return -EINVAL;
4976 }
NeilBrown0c55e022010-05-03 14:09:02 +10004977 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
4978 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004979 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08004980 } else {
NeilBrown91adb562009-03-31 14:39:39 +11004981 BUG_ON(mddev->level != mddev->new_level);
4982 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10004983 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11004984 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08004985 }
4986
NeilBrown245f46c2009-03-31 14:39:39 +11004987 if (mddev->private == NULL)
4988 conf = setup_conf(mddev);
4989 else
4990 conf = mddev->private;
4991
NeilBrown91adb562009-03-31 14:39:39 +11004992 if (IS_ERR(conf))
4993 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08004994
NeilBrown91adb562009-03-31 14:39:39 +11004995 mddev->thread = conf->thread;
4996 conf->thread = NULL;
4997 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998
Linus Torvalds1da177e2005-04-16 15:20:36 -07004999 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005000 * 0 for a fully functional array, 1 or 2 for a degraded array.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005001 */
NeilBrownc148ffd2009-11-13 17:47:00 +11005002 list_for_each_entry(rdev, &mddev->disks, same_set) {
5003 if (rdev->raid_disk < 0)
5004 continue;
NeilBrown2f115882010-06-17 17:41:03 +10005005 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown91adb562009-03-31 14:39:39 +11005006 working_disks++;
NeilBrown2f115882010-06-17 17:41:03 +10005007 continue;
5008 }
NeilBrownc148ffd2009-11-13 17:47:00 +11005009 /* This disc is not fully in-sync. However if it
5010 * just stored parity (beyond the recovery_offset),
5011 * when we don't need to be concerned about the
5012 * array being dirty.
5013 * When reshape goes 'backwards', we never have
5014 * partially completed devices, so we only need
5015 * to worry about reshape going forwards.
5016 */
5017 /* Hack because v0.91 doesn't store recovery_offset properly. */
5018 if (mddev->major_version == 0 &&
5019 mddev->minor_version > 90)
5020 rdev->recovery_offset = reshape_offset;
5021
NeilBrownc148ffd2009-11-13 17:47:00 +11005022 if (rdev->recovery_offset < reshape_offset) {
5023 /* We need to check old and new layout */
5024 if (!only_parity(rdev->raid_disk,
5025 conf->algorithm,
5026 conf->raid_disks,
5027 conf->max_degraded))
5028 continue;
5029 }
5030 if (!only_parity(rdev->raid_disk,
5031 conf->prev_algo,
5032 conf->previous_raid_disks,
5033 conf->max_degraded))
5034 continue;
5035 dirty_parity_disks++;
5036 }
NeilBrown91adb562009-03-31 14:39:39 +11005037
NeilBrown5e5e3e72009-10-16 16:35:30 +11005038 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
5039 - working_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005040
NeilBrown674806d2010-06-16 17:17:53 +10005041 if (has_failed(conf)) {
NeilBrown0c55e022010-05-03 14:09:02 +10005042 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07005043 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07005044 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005045 goto abort;
5046 }
5047
NeilBrown91adb562009-03-31 14:39:39 +11005048 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10005049 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11005050 mddev->resync_max_sectors = mddev->dev_sectors;
5051
NeilBrownc148ffd2009-11-13 17:47:00 +11005052 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07005053 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005054 if (mddev->ok_start_degraded)
5055 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10005056 "md/raid:%s: starting dirty degraded array"
5057 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005058 mdname(mddev));
5059 else {
5060 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005061 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005062 mdname(mddev));
5063 goto abort;
5064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005065 }
5066
Linus Torvalds1da177e2005-04-16 15:20:36 -07005067 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10005068 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5069 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11005070 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5071 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005072 else
NeilBrown0c55e022010-05-03 14:09:02 +10005073 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5074 " out of %d devices, algorithm %d\n",
5075 mdname(mddev), conf->level,
5076 mddev->raid_disks - mddev->degraded,
5077 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005078
5079 print_raid5_conf(conf);
5080
NeilBrownfef9c612009-03-31 15:16:46 +11005081 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11005082 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08005083 atomic_set(&conf->reshape_stripes, 0);
5084 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5085 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5086 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5087 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5088 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005089 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08005090 }
5091
Linus Torvalds1da177e2005-04-16 15:20:36 -07005092
5093 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10005094 if (mddev->to_remove == &raid5_attrs_group)
5095 mddev->to_remove = NULL;
NeilBrown00bcb4a2010-06-01 19:37:23 +10005096 else if (mddev->kobj.sd &&
5097 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08005098 printk(KERN_WARNING
NeilBrown4a5add42010-06-01 19:37:28 +10005099 "raid5: failed to create sysfs attributes for %s\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08005100 mdname(mddev));
NeilBrown4a5add42010-06-01 19:37:28 +10005101 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5102
5103 if (mddev->queue) {
NeilBrown9f7c2222010-07-26 12:04:13 +10005104 int chunk_size;
NeilBrown4a5add42010-06-01 19:37:28 +10005105 /* read-ahead size must cover two whole stripes, which
5106 * is 2 * (datadisks) * chunksize where 'n' is the
5107 * number of raid devices
5108 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005109 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5110 int stripe = data_disks *
5111 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
5112 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5113 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
NeilBrown4a5add42010-06-01 19:37:28 +10005114
5115 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
NeilBrown11d8a6e2010-07-26 11:57:07 +10005116
5117 mddev->queue->backing_dev_info.congested_data = mddev;
5118 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
NeilBrown9f7c2222010-07-26 12:04:13 +10005119
5120 chunk_size = mddev->chunk_sectors << 9;
5121 blk_queue_io_min(mddev->queue, chunk_size);
5122 blk_queue_io_opt(mddev->queue, chunk_size *
5123 (conf->raid_disks - conf->max_degraded));
5124
5125 list_for_each_entry(rdev, &mddev->disks, same_set)
5126 disk_stack_limits(mddev->gendisk, rdev->bdev,
5127 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005128 }
5129
Linus Torvalds1da177e2005-04-16 15:20:36 -07005130 return 0;
5131abort:
NeilBrowne0cf8f02009-03-31 14:39:39 +11005132 md_unregister_thread(mddev->thread);
NeilBrown91adb562009-03-31 14:39:39 +11005133 mddev->thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005134 if (conf) {
5135 print_raid5_conf(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005136 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005137 }
5138 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10005139 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005140 return -EIO;
5141}
5142
NeilBrown3f294f42005-11-08 21:39:25 -08005143static int stop(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005144{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005145 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005146
5147 md_unregister_thread(mddev->thread);
5148 mddev->thread = NULL;
NeilBrown11d8a6e2010-07-26 11:57:07 +10005149 if (mddev->queue)
5150 mddev->queue->backing_dev_info.congested_fn = NULL;
Dan Williams95fc17a2009-07-31 12:39:15 +10005151 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10005152 mddev->private = NULL;
5153 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005154 return 0;
5155}
5156
Dan Williams45b42332007-07-09 11:56:43 -07005157#ifdef DEBUG
NeilBrownd710e132008-10-13 11:55:12 +11005158static void print_sh(struct seq_file *seq, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005159{
5160 int i;
5161
NeilBrown16a53ec2006-06-26 00:27:38 -07005162 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
5163 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
5164 seq_printf(seq, "sh %llu, count %d.\n",
5165 (unsigned long long)sh->sector, atomic_read(&sh->count));
5166 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005167 for (i = 0; i < sh->disks; i++) {
NeilBrown16a53ec2006-06-26 00:27:38 -07005168 seq_printf(seq, "(cache%d: %p %ld) ",
5169 i, sh->dev[i].page, sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005170 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005171 seq_printf(seq, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005172}
5173
NeilBrownd710e132008-10-13 11:55:12 +11005174static void printall(struct seq_file *seq, raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005175{
5176 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -08005177 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005178 int i;
5179
5180 spin_lock_irq(&conf->device_lock);
5181 for (i = 0; i < NR_HASH; i++) {
NeilBrownfccddba2006-01-06 00:20:33 -08005182 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005183 if (sh->raid_conf != conf)
5184 continue;
NeilBrown16a53ec2006-06-26 00:27:38 -07005185 print_sh(seq, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005186 }
5187 }
5188 spin_unlock_irq(&conf->device_lock);
5189}
5190#endif
5191
NeilBrownd710e132008-10-13 11:55:12 +11005192static void status(struct seq_file *seq, mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005193{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005194 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005195 int i;
5196
Andre Noll9d8f0362009-06-18 08:45:01 +10005197 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5198 mddev->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07005199 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005200 for (i = 0; i < conf->raid_disks; i++)
5201 seq_printf (seq, "%s",
5202 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08005203 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005204 seq_printf (seq, "]");
Dan Williams45b42332007-07-09 11:56:43 -07005205#ifdef DEBUG
NeilBrown16a53ec2006-06-26 00:27:38 -07005206 seq_printf (seq, "\n");
5207 printall(seq, conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005208#endif
5209}
5210
5211static void print_raid5_conf (raid5_conf_t *conf)
5212{
5213 int i;
5214 struct disk_info *tmp;
5215
NeilBrown0c55e022010-05-03 14:09:02 +10005216 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005217 if (!conf) {
5218 printk("(conf==NULL)\n");
5219 return;
5220 }
NeilBrown0c55e022010-05-03 14:09:02 +10005221 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5222 conf->raid_disks,
5223 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005224
5225 for (i = 0; i < conf->raid_disks; i++) {
5226 char b[BDEVNAME_SIZE];
5227 tmp = conf->disks + i;
5228 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10005229 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5230 i, !test_bit(Faulty, &tmp->rdev->flags),
5231 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005232 }
5233}
5234
5235static int raid5_spare_active(mddev_t *mddev)
5236{
5237 int i;
5238 raid5_conf_t *conf = mddev->private;
5239 struct disk_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10005240 int count = 0;
5241 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005242
5243 for (i = 0; i < conf->raid_disks; i++) {
5244 tmp = conf->disks + i;
5245 if (tmp->rdev
NeilBrown70fffd02010-06-16 17:01:25 +10005246 && tmp->rdev->recovery_offset == MaxSector
NeilBrownb2d444d2005-11-08 21:39:31 -08005247 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07005248 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10005249 count++;
Jonathan Brassow43c73ca2011-01-14 09:14:33 +11005250 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005251 }
5252 }
NeilBrown6b965622010-08-18 11:56:59 +10005253 spin_lock_irqsave(&conf->device_lock, flags);
5254 mddev->degraded -= count;
5255 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005256 print_raid5_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10005257 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005258}
5259
5260static int raid5_remove_disk(mddev_t *mddev, int number)
5261{
5262 raid5_conf_t *conf = mddev->private;
5263 int err = 0;
5264 mdk_rdev_t *rdev;
5265 struct disk_info *p = conf->disks + number;
5266
5267 print_raid5_conf(conf);
5268 rdev = p->rdev;
5269 if (rdev) {
NeilBrownec32a2b2009-03-31 15:17:38 +11005270 if (number >= conf->raid_disks &&
5271 conf->reshape_progress == MaxSector)
5272 clear_bit(In_sync, &rdev->flags);
5273
NeilBrownb2d444d2005-11-08 21:39:31 -08005274 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07005275 atomic_read(&rdev->nr_pending)) {
5276 err = -EBUSY;
5277 goto abort;
5278 }
NeilBrowndfc70642008-05-23 13:04:39 -07005279 /* Only remove non-faulty devices if recovery
5280 * isn't possible.
5281 */
5282 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown674806d2010-06-16 17:17:53 +10005283 !has_failed(conf) &&
NeilBrownec32a2b2009-03-31 15:17:38 +11005284 number < conf->raid_disks) {
NeilBrowndfc70642008-05-23 13:04:39 -07005285 err = -EBUSY;
5286 goto abort;
5287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005288 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07005289 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005290 if (atomic_read(&rdev->nr_pending)) {
5291 /* lost the race, try later */
5292 err = -EBUSY;
5293 p->rdev = rdev;
5294 }
5295 }
5296abort:
5297
5298 print_raid5_conf(conf);
5299 return err;
5300}
5301
5302static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
5303{
5304 raid5_conf_t *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10005305 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005306 int disk;
5307 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10005308 int first = 0;
5309 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005310
NeilBrown674806d2010-06-16 17:17:53 +10005311 if (has_failed(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005312 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10005313 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005314
Neil Brown6c2fce22008-06-28 08:31:31 +10005315 if (rdev->raid_disk >= 0)
5316 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005317
5318 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005319 * find the disk ... but prefer rdev->saved_raid_disk
5320 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005321 */
NeilBrown16a53ec2006-06-26 00:27:38 -07005322 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10005323 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07005324 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5325 disk = rdev->saved_raid_disk;
5326 else
Neil Brown6c2fce22008-06-28 08:31:31 +10005327 disk = first;
5328 for ( ; disk <= last ; disk++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005329 if ((p=conf->disks + disk)->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08005330 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005331 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10005332 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07005333 if (rdev->saved_raid_disk != disk)
5334 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08005335 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005336 break;
5337 }
5338 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10005339 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005340}
5341
5342static int raid5_resize(mddev_t *mddev, sector_t sectors)
5343{
5344 /* no resync is happening, and there is enough space
5345 * on all devices, so we can resize.
5346 * We need to make sure resync covers any new space.
5347 * If the array is shrinking we should possibly wait until
5348 * any io in the removed space completes, but it hardly seems
5349 * worth it.
5350 */
Andre Noll9d8f0362009-06-18 08:45:01 +10005351 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Dan Williams1f403622009-03-31 14:59:03 +11005352 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5353 mddev->raid_disks));
Dan Williamsb522adc2009-03-31 15:00:31 +11005354 if (mddev->array_sectors >
5355 raid5_size(mddev, sectors, mddev->raid_disks))
5356 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10005357 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005358 revalidate_disk(mddev->gendisk);
NeilBrownb0986362011-05-11 15:52:21 +10005359 if (sectors > mddev->dev_sectors &&
5360 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11005361 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005362 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5363 }
Andre Noll58c0fed2009-03-31 14:33:13 +11005364 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07005365 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005366 return 0;
5367}
5368
NeilBrown01ee22b2009-06-18 08:47:20 +10005369static int check_stripe_cache(mddev_t *mddev)
5370{
5371 /* Can only proceed if there are plenty of stripe_heads.
5372 * We need a minimum of one full stripe,, and for sensible progress
5373 * it is best to have about 4 times that.
5374 * If we require 4 times, then the default 256 4K stripe_heads will
5375 * allow for chunk sizes up to 256K, which is probably OK.
5376 * If the chunk size is greater, user-space should request more
5377 * stripe_heads first.
5378 */
5379 raid5_conf_t *conf = mddev->private;
5380 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5381 > conf->max_nr_stripes ||
5382 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5383 > conf->max_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10005384 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5385 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10005386 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5387 / STRIPE_SIZE)*4);
5388 return 0;
5389 }
5390 return 1;
5391}
5392
NeilBrown50ac1682009-06-18 08:47:55 +10005393static int check_reshape(mddev_t *mddev)
NeilBrown29269552006-03-27 01:18:10 -08005394{
NeilBrown070ec552009-06-16 16:54:21 +10005395 raid5_conf_t *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08005396
NeilBrown88ce4932009-03-31 15:24:23 +11005397 if (mddev->delta_disks == 0 &&
5398 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10005399 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10005400 return 0; /* nothing to do */
NeilBrowndba034e2008-08-05 15:54:13 +10005401 if (mddev->bitmap)
5402 /* Cannot grow a bitmap yet */
5403 return -EBUSY;
NeilBrown674806d2010-06-16 17:17:53 +10005404 if (has_failed(conf))
NeilBrownec32a2b2009-03-31 15:17:38 +11005405 return -EINVAL;
5406 if (mddev->delta_disks < 0) {
5407 /* We might be able to shrink, but the devices must
5408 * be made bigger first.
5409 * For raid6, 4 is the minimum size.
5410 * Otherwise 2 is the minimum
5411 */
5412 int min = 2;
5413 if (mddev->level == 6)
5414 min = 4;
5415 if (mddev->raid_disks + mddev->delta_disks < min)
5416 return -EINVAL;
5417 }
NeilBrown29269552006-03-27 01:18:10 -08005418
NeilBrown01ee22b2009-06-18 08:47:20 +10005419 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08005420 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08005421
NeilBrownec32a2b2009-03-31 15:17:38 +11005422 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
NeilBrown63c70c42006-03-27 01:18:13 -08005423}
5424
5425static int raid5_start_reshape(mddev_t *mddev)
5426{
NeilBrown070ec552009-06-16 16:54:21 +10005427 raid5_conf_t *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08005428 mdk_rdev_t *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08005429 int spares = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07005430 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08005431
NeilBrownf4168852007-02-28 20:11:53 -08005432 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08005433 return -EBUSY;
5434
NeilBrown01ee22b2009-06-18 08:47:20 +10005435 if (!check_stripe_cache(mddev))
5436 return -ENOSPC;
5437
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005438 list_for_each_entry(rdev, &mddev->disks, same_set)
NeilBrown469518a2011-01-31 11:57:43 +11005439 if (!test_bit(In_sync, &rdev->flags)
5440 && !test_bit(Faulty, &rdev->flags))
NeilBrown29269552006-03-27 01:18:10 -08005441 spares++;
NeilBrown63c70c42006-03-27 01:18:13 -08005442
NeilBrownf4168852007-02-28 20:11:53 -08005443 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08005444 /* Not enough devices even to make a degraded array
5445 * of that size
5446 */
5447 return -EINVAL;
5448
NeilBrownec32a2b2009-03-31 15:17:38 +11005449 /* Refuse to reduce size of the array. Any reductions in
5450 * array size must be through explicit setting of array_size
5451 * attribute.
5452 */
5453 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5454 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10005455 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11005456 "before number of disks\n", mdname(mddev));
5457 return -EINVAL;
5458 }
5459
NeilBrownf6705572006-03-27 01:18:11 -08005460 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08005461 spin_lock_irq(&conf->device_lock);
5462 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08005463 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005464 conf->prev_chunk_sectors = conf->chunk_sectors;
5465 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11005466 conf->prev_algo = conf->algorithm;
5467 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11005468 if (mddev->delta_disks < 0)
5469 conf->reshape_progress = raid5_size(mddev, 0, 0);
5470 else
5471 conf->reshape_progress = 0;
5472 conf->reshape_safe = conf->reshape_progress;
NeilBrown86b42c72009-03-31 15:19:03 +11005473 conf->generation++;
NeilBrown29269552006-03-27 01:18:10 -08005474 spin_unlock_irq(&conf->device_lock);
5475
5476 /* Add some new drives, as many as will fit.
5477 * We know there are enough to make the newly sized array work.
NeilBrown3424bf62010-06-17 17:48:26 +10005478 * Don't add devices if we are reducing the number of
5479 * devices in the array. This is because it is not possible
5480 * to correctly record the "partially reconstructed" state of
5481 * such devices during the reshape and confusion could result.
NeilBrown29269552006-03-27 01:18:10 -08005482 */
NeilBrown87a8dec2011-01-31 11:57:43 +11005483 if (mddev->delta_disks >= 0) {
5484 int added_devices = 0;
5485 list_for_each_entry(rdev, &mddev->disks, same_set)
5486 if (rdev->raid_disk < 0 &&
5487 !test_bit(Faulty, &rdev->flags)) {
5488 if (raid5_add_disk(mddev, rdev) == 0) {
5489 char nm[20];
5490 if (rdev->raid_disk
5491 >= conf->previous_raid_disks) {
5492 set_bit(In_sync, &rdev->flags);
5493 added_devices++;
5494 } else
5495 rdev->recovery_offset = 0;
5496 sprintf(nm, "rd%d", rdev->raid_disk);
5497 if (sysfs_create_link(&mddev->kobj,
5498 &rdev->kobj, nm))
5499 /* Failure here is OK */;
NeilBrown50da0842011-01-31 11:57:43 +11005500 }
NeilBrown87a8dec2011-01-31 11:57:43 +11005501 } else if (rdev->raid_disk >= conf->previous_raid_disks
5502 && !test_bit(Faulty, &rdev->flags)) {
5503 /* This is a spare that was manually added */
5504 set_bit(In_sync, &rdev->flags);
5505 added_devices++;
5506 }
NeilBrown29269552006-03-27 01:18:10 -08005507
NeilBrown87a8dec2011-01-31 11:57:43 +11005508 /* When a reshape changes the number of devices,
5509 * ->degraded is measured against the larger of the
5510 * pre and post number of devices.
5511 */
NeilBrownec32a2b2009-03-31 15:17:38 +11005512 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown9eb07c22010-02-09 12:31:47 +11005513 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
NeilBrownec32a2b2009-03-31 15:17:38 +11005514 - added_devices;
5515 spin_unlock_irqrestore(&conf->device_lock, flags);
5516 }
NeilBrown63c70c42006-03-27 01:18:13 -08005517 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10005518 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07005519 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08005520
NeilBrown29269552006-03-27 01:18:10 -08005521 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5522 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5523 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5524 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5525 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005526 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08005527 if (!mddev->sync_thread) {
5528 mddev->recovery = 0;
5529 spin_lock_irq(&conf->device_lock);
5530 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005531 conf->reshape_progress = MaxSector;
NeilBrown29269552006-03-27 01:18:10 -08005532 spin_unlock_irq(&conf->device_lock);
5533 return -EAGAIN;
5534 }
NeilBrownc8f517c2009-03-31 15:28:40 +11005535 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08005536 md_wakeup_thread(mddev->sync_thread);
5537 md_new_event(mddev);
5538 return 0;
5539}
NeilBrown29269552006-03-27 01:18:10 -08005540
NeilBrownec32a2b2009-03-31 15:17:38 +11005541/* This is called from the reshape thread and should make any
5542 * changes needed in 'conf'
5543 */
NeilBrown29269552006-03-27 01:18:10 -08005544static void end_reshape(raid5_conf_t *conf)
5545{
NeilBrown29269552006-03-27 01:18:10 -08005546
NeilBrownf6705572006-03-27 01:18:11 -08005547 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
Dan Williams80c3a6c2009-03-17 18:10:40 -07005548
NeilBrownf6705572006-03-27 01:18:11 -08005549 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11005550 conf->previous_raid_disks = conf->raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005551 conf->reshape_progress = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08005552 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11005553 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07005554
5555 /* read-ahead size must cover two whole stripes, which is
5556 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5557 */
NeilBrown4a5add42010-06-01 19:37:28 +10005558 if (conf->mddev->queue) {
NeilBrowncea9c222009-03-31 15:15:05 +11005559 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005560 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11005561 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07005562 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5563 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5564 }
NeilBrown29269552006-03-27 01:18:10 -08005565 }
NeilBrown29269552006-03-27 01:18:10 -08005566}
5567
NeilBrownec32a2b2009-03-31 15:17:38 +11005568/* This is called from the raid5d thread with mddev_lock held.
5569 * It makes config changes to the device.
5570 */
NeilBrowncea9c222009-03-31 15:15:05 +11005571static void raid5_finish_reshape(mddev_t *mddev)
5572{
NeilBrown070ec552009-06-16 16:54:21 +10005573 raid5_conf_t *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11005574
5575 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5576
NeilBrownec32a2b2009-03-31 15:17:38 +11005577 if (mddev->delta_disks > 0) {
5578 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5579 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005580 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11005581 } else {
5582 int d;
NeilBrownec32a2b2009-03-31 15:17:38 +11005583 mddev->degraded = conf->raid_disks;
5584 for (d = 0; d < conf->raid_disks ; d++)
5585 if (conf->disks[d].rdev &&
5586 test_bit(In_sync,
5587 &conf->disks[d].rdev->flags))
5588 mddev->degraded--;
5589 for (d = conf->raid_disks ;
5590 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10005591 d++) {
5592 mdk_rdev_t *rdev = conf->disks[d].rdev;
5593 if (rdev && raid5_remove_disk(mddev, d) == 0) {
5594 char nm[20];
5595 sprintf(nm, "rd%d", rdev->raid_disk);
5596 sysfs_remove_link(&mddev->kobj, nm);
5597 rdev->raid_disk = -1;
5598 }
5599 }
NeilBrowncea9c222009-03-31 15:15:05 +11005600 }
NeilBrown88ce4932009-03-31 15:24:23 +11005601 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005602 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11005603 mddev->reshape_position = MaxSector;
5604 mddev->delta_disks = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11005605 }
5606}
5607
NeilBrown72626682005-09-09 16:23:54 -07005608static void raid5_quiesce(mddev_t *mddev, int state)
5609{
NeilBrown070ec552009-06-16 16:54:21 +10005610 raid5_conf_t *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07005611
5612 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08005613 case 2: /* resume for a suspend */
5614 wake_up(&conf->wait_for_overlap);
5615 break;
5616
NeilBrown72626682005-09-09 16:23:54 -07005617 case 1: /* stop all writes */
5618 spin_lock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005619 /* '2' tells resync/reshape to pause so that all
5620 * active stripes can drain
5621 */
5622 conf->quiesce = 2;
NeilBrown72626682005-09-09 16:23:54 -07005623 wait_event_lock_irq(conf->wait_for_stripe,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005624 atomic_read(&conf->active_stripes) == 0 &&
5625 atomic_read(&conf->active_aligned_reads) == 0,
NeilBrown72626682005-09-09 16:23:54 -07005626 conf->device_lock, /* nothing */);
NeilBrown64bd6602009-08-03 10:59:58 +10005627 conf->quiesce = 1;
NeilBrown72626682005-09-09 16:23:54 -07005628 spin_unlock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005629 /* allow reshape to continue */
5630 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005631 break;
5632
5633 case 0: /* re-enable writes */
5634 spin_lock_irq(&conf->device_lock);
5635 conf->quiesce = 0;
5636 wake_up(&conf->wait_for_stripe);
NeilBrowne464eaf2006-03-27 01:18:14 -08005637 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005638 spin_unlock_irq(&conf->device_lock);
5639 break;
5640 }
NeilBrown72626682005-09-09 16:23:54 -07005641}
NeilBrownb15c2e52006-01-06 00:20:16 -08005642
NeilBrownd562b0c2009-03-31 14:39:39 +11005643
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005644static void *raid45_takeover_raid0(mddev_t *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11005645{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005646 struct raid0_private_data *raid0_priv = mddev->private;
Randy Dunlapd76c8422011-04-21 09:07:26 -07005647 sector_t sectors;
Trela Maciej54071b32010-03-08 16:02:42 +11005648
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005649 /* for raid0 takeover only one zone is supported */
5650 if (raid0_priv->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10005651 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5652 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005653 return ERR_PTR(-EINVAL);
5654 }
5655
NeilBrown3b71bd92011-04-20 15:38:18 +10005656 sectors = raid0_priv->strip_zone[0].zone_end;
5657 sector_div(sectors, raid0_priv->strip_zone[0].nb_dev);
5658 mddev->dev_sectors = sectors;
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005659 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11005660 mddev->new_layout = ALGORITHM_PARITY_N;
5661 mddev->new_chunk_sectors = mddev->chunk_sectors;
5662 mddev->raid_disks += 1;
5663 mddev->delta_disks = 1;
5664 /* make sure it will be not marked as dirty */
5665 mddev->recovery_cp = MaxSector;
5666
5667 return setup_conf(mddev);
5668}
5669
5670
NeilBrownd562b0c2009-03-31 14:39:39 +11005671static void *raid5_takeover_raid1(mddev_t *mddev)
5672{
5673 int chunksect;
5674
5675 if (mddev->raid_disks != 2 ||
5676 mddev->degraded > 1)
5677 return ERR_PTR(-EINVAL);
5678
5679 /* Should check if there are write-behind devices? */
5680
5681 chunksect = 64*2; /* 64K by default */
5682
5683 /* The array must be an exact multiple of chunksize */
5684 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5685 chunksect >>= 1;
5686
5687 if ((chunksect<<9) < STRIPE_SIZE)
5688 /* array size does not allow a suitable chunk size */
5689 return ERR_PTR(-EINVAL);
5690
5691 mddev->new_level = 5;
5692 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10005693 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11005694
5695 return setup_conf(mddev);
5696}
5697
NeilBrownfc9739c2009-03-31 14:57:20 +11005698static void *raid5_takeover_raid6(mddev_t *mddev)
5699{
5700 int new_layout;
5701
5702 switch (mddev->layout) {
5703 case ALGORITHM_LEFT_ASYMMETRIC_6:
5704 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5705 break;
5706 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5707 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5708 break;
5709 case ALGORITHM_LEFT_SYMMETRIC_6:
5710 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5711 break;
5712 case ALGORITHM_RIGHT_SYMMETRIC_6:
5713 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5714 break;
5715 case ALGORITHM_PARITY_0_6:
5716 new_layout = ALGORITHM_PARITY_0;
5717 break;
5718 case ALGORITHM_PARITY_N:
5719 new_layout = ALGORITHM_PARITY_N;
5720 break;
5721 default:
5722 return ERR_PTR(-EINVAL);
5723 }
5724 mddev->new_level = 5;
5725 mddev->new_layout = new_layout;
5726 mddev->delta_disks = -1;
5727 mddev->raid_disks -= 1;
5728 return setup_conf(mddev);
5729}
5730
NeilBrownd562b0c2009-03-31 14:39:39 +11005731
NeilBrown50ac1682009-06-18 08:47:55 +10005732static int raid5_check_reshape(mddev_t *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11005733{
NeilBrown88ce4932009-03-31 15:24:23 +11005734 /* For a 2-drive array, the layout and chunk size can be changed
5735 * immediately as not restriping is needed.
5736 * For larger arrays we record the new value - after validation
5737 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11005738 */
NeilBrown070ec552009-06-16 16:54:21 +10005739 raid5_conf_t *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10005740 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11005741
NeilBrown597a7112009-06-18 08:47:42 +10005742 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11005743 return -EINVAL;
5744 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005745 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11005746 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005747 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11005748 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005749 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11005750 /* not factor of array size */
5751 return -EINVAL;
5752 }
5753
5754 /* They look valid */
5755
NeilBrown88ce4932009-03-31 15:24:23 +11005756 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10005757 /* can make the change immediately */
5758 if (mddev->new_layout >= 0) {
5759 conf->algorithm = mddev->new_layout;
5760 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11005761 }
5762 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10005763 conf->chunk_sectors = new_chunk ;
5764 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11005765 }
5766 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5767 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11005768 }
NeilBrown50ac1682009-06-18 08:47:55 +10005769 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11005770}
5771
NeilBrown50ac1682009-06-18 08:47:55 +10005772static int raid6_check_reshape(mddev_t *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11005773{
NeilBrown597a7112009-06-18 08:47:42 +10005774 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10005775
NeilBrown597a7112009-06-18 08:47:42 +10005776 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11005777 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005778 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005779 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11005780 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005781 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11005782 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005783 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11005784 /* not factor of array size */
5785 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005786 }
NeilBrown88ce4932009-03-31 15:24:23 +11005787
5788 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10005789 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11005790}
5791
NeilBrownd562b0c2009-03-31 14:39:39 +11005792static void *raid5_takeover(mddev_t *mddev)
5793{
5794 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005795 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005796 * raid1 - if there are two drives. We need to know the chunk size
5797 * raid4 - trivial - just use a raid4 layout.
5798 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005799 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005800 if (mddev->level == 0)
5801 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11005802 if (mddev->level == 1)
5803 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11005804 if (mddev->level == 4) {
5805 mddev->new_layout = ALGORITHM_PARITY_N;
5806 mddev->new_level = 5;
5807 return setup_conf(mddev);
5808 }
NeilBrownfc9739c2009-03-31 14:57:20 +11005809 if (mddev->level == 6)
5810 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11005811
5812 return ERR_PTR(-EINVAL);
5813}
5814
NeilBrowna78d38a2010-03-22 16:53:49 +11005815static void *raid4_takeover(mddev_t *mddev)
5816{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005817 /* raid4 can take over:
5818 * raid0 - if there is only one strip zone
5819 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11005820 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005821 if (mddev->level == 0)
5822 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11005823 if (mddev->level == 5 &&
5824 mddev->layout == ALGORITHM_PARITY_N) {
5825 mddev->new_layout = 0;
5826 mddev->new_level = 4;
5827 return setup_conf(mddev);
5828 }
5829 return ERR_PTR(-EINVAL);
5830}
NeilBrownd562b0c2009-03-31 14:39:39 +11005831
NeilBrown245f46c2009-03-31 14:39:39 +11005832static struct mdk_personality raid5_personality;
5833
5834static void *raid6_takeover(mddev_t *mddev)
5835{
5836 /* Currently can only take over a raid5. We map the
5837 * personality to an equivalent raid6 personality
5838 * with the Q block at the end.
5839 */
5840 int new_layout;
5841
5842 if (mddev->pers != &raid5_personality)
5843 return ERR_PTR(-EINVAL);
5844 if (mddev->degraded > 1)
5845 return ERR_PTR(-EINVAL);
5846 if (mddev->raid_disks > 253)
5847 return ERR_PTR(-EINVAL);
5848 if (mddev->raid_disks < 3)
5849 return ERR_PTR(-EINVAL);
5850
5851 switch (mddev->layout) {
5852 case ALGORITHM_LEFT_ASYMMETRIC:
5853 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5854 break;
5855 case ALGORITHM_RIGHT_ASYMMETRIC:
5856 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5857 break;
5858 case ALGORITHM_LEFT_SYMMETRIC:
5859 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5860 break;
5861 case ALGORITHM_RIGHT_SYMMETRIC:
5862 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5863 break;
5864 case ALGORITHM_PARITY_0:
5865 new_layout = ALGORITHM_PARITY_0_6;
5866 break;
5867 case ALGORITHM_PARITY_N:
5868 new_layout = ALGORITHM_PARITY_N;
5869 break;
5870 default:
5871 return ERR_PTR(-EINVAL);
5872 }
5873 mddev->new_level = 6;
5874 mddev->new_layout = new_layout;
5875 mddev->delta_disks = 1;
5876 mddev->raid_disks += 1;
5877 return setup_conf(mddev);
5878}
5879
5880
NeilBrown16a53ec2006-06-26 00:27:38 -07005881static struct mdk_personality raid6_personality =
5882{
5883 .name = "raid6",
5884 .level = 6,
5885 .owner = THIS_MODULE,
5886 .make_request = make_request,
5887 .run = run,
5888 .stop = stop,
5889 .status = status,
5890 .error_handler = error,
5891 .hot_add_disk = raid5_add_disk,
5892 .hot_remove_disk= raid5_remove_disk,
5893 .spare_active = raid5_spare_active,
5894 .sync_request = sync_request,
5895 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005896 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10005897 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08005898 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005899 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07005900 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11005901 .takeover = raid6_takeover,
NeilBrown16a53ec2006-06-26 00:27:38 -07005902};
NeilBrown2604b702006-01-06 00:20:36 -08005903static struct mdk_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005904{
5905 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08005906 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005907 .owner = THIS_MODULE,
5908 .make_request = make_request,
5909 .run = run,
5910 .stop = stop,
5911 .status = status,
5912 .error_handler = error,
5913 .hot_add_disk = raid5_add_disk,
5914 .hot_remove_disk= raid5_remove_disk,
5915 .spare_active = raid5_spare_active,
5916 .sync_request = sync_request,
5917 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005918 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08005919 .check_reshape = raid5_check_reshape,
5920 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005921 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07005922 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11005923 .takeover = raid5_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005924};
5925
NeilBrown2604b702006-01-06 00:20:36 -08005926static struct mdk_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005927{
NeilBrown2604b702006-01-06 00:20:36 -08005928 .name = "raid4",
5929 .level = 4,
5930 .owner = THIS_MODULE,
5931 .make_request = make_request,
5932 .run = run,
5933 .stop = stop,
5934 .status = status,
5935 .error_handler = error,
5936 .hot_add_disk = raid5_add_disk,
5937 .hot_remove_disk= raid5_remove_disk,
5938 .spare_active = raid5_spare_active,
5939 .sync_request = sync_request,
5940 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005941 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08005942 .check_reshape = raid5_check_reshape,
5943 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005944 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08005945 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11005946 .takeover = raid4_takeover,
NeilBrown2604b702006-01-06 00:20:36 -08005947};
5948
5949static int __init raid5_init(void)
5950{
NeilBrown16a53ec2006-06-26 00:27:38 -07005951 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005952 register_md_personality(&raid5_personality);
5953 register_md_personality(&raid4_personality);
5954 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005955}
5956
NeilBrown2604b702006-01-06 00:20:36 -08005957static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005958{
NeilBrown16a53ec2006-06-26 00:27:38 -07005959 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005960 unregister_md_personality(&raid5_personality);
5961 unregister_md_personality(&raid4_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005962}
5963
5964module_init(raid5_init);
5965module_exit(raid5_exit);
5966MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11005967MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005968MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08005969MODULE_ALIAS("md-raid5");
5970MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08005971MODULE_ALIAS("md-level-5");
5972MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07005973MODULE_ALIAS("md-personality-8"); /* RAID6 */
5974MODULE_ALIAS("md-raid6");
5975MODULE_ALIAS("md-level-6");
5976
5977/* This used to be two separate modules, they were: */
5978MODULE_ALIAS("raid5");
5979MODULE_ALIAS("raid6");