blob: bd4067a70834a3b10a741e125b8ac1f106487831 [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.
30 * conf->bm_write is the number of the last batch successfully written.
31 * conf->bm_flush is the number of the last batch that was closed to
32 * 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
35 * the number of the batch it will be in. This is bm_flush+1.
36 * When we are ready to do a write, if that batch hasn't been written yet,
37 * we plug the array and queue the stripe for later.
38 * When an unplug happens, we increment bm_flush, thus closing the current
39 * batch.
40 * When we notice that bm_flush > bm_write, we write out all pending updates
41 * to the bitmap, and advance bm_write to where bm_flush was.
42 * This may occasionally write a bit out twice, but is sure never to
43 * miss any bits.
44 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
NeilBrownbff61972009-03-31 14:33:13 +110046#include <linux/blkdev.h>
NeilBrownf6705572006-03-27 01:18:11 -080047#include <linux/kthread.h>
Dan Williamsf701d582009-03-31 15:09:39 +110048#include <linux/raid/pq.h>
Dan Williams91c00922007-01-02 13:52:30 -070049#include <linux/async_tx.h>
Dan Williams07a3b412009-08-29 19:13:13 -070050#include <linux/async.h>
NeilBrownbff61972009-03-31 14:33:13 +110051#include <linux/seq_file.h>
Dan Williams36d1c642009-07-14 11:48:22 -070052#include <linux/cpu.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090053#include <linux/slab.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110054#include "md.h"
NeilBrownbff61972009-03-31 14:33:13 +110055#include "raid5.h"
Trela Maciej54071b32010-03-08 16:02:42 +110056#include "raid0.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110057#include "bitmap.h"
NeilBrown72626682005-09-09 16:23:54 -070058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/*
60 * Stripe cache
61 */
62
63#define NR_STRIPES 256
64#define STRIPE_SIZE PAGE_SIZE
65#define STRIPE_SHIFT (PAGE_SHIFT - 9)
66#define STRIPE_SECTORS (STRIPE_SIZE>>9)
67#define IO_THRESHOLD 1
Dan Williams8b3e6cd2008-04-28 02:15:53 -070068#define BYPASS_THRESHOLD 1
NeilBrownfccddba2006-01-06 00:20:33 -080069#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#define HASH_MASK (NR_HASH - 1)
71
NeilBrownfccddba2006-01-06 00:20:33 -080072#define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74/* bio's attached to a stripe+device for I/O are linked together in bi_sector
75 * order without overlap. There may be several bio's per stripe+device, and
76 * a bio could span several devices.
77 * When walking this list for a particular stripe+device, we must never proceed
78 * beyond a bio that extends past this device, as the next bio might no longer
79 * be valid.
80 * This macro is used to determine the 'next' bio in the list, given the sector
81 * of the current stripe+device
82 */
83#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
84/*
85 * The following can be used to debug the driver
86 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#define RAID5_PARANOIA 1
88#if RAID5_PARANOIA && defined(CONFIG_SMP)
89# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
90#else
91# define CHECK_DEVLOCK()
92#endif
93
Dan Williams45b42332007-07-09 11:56:43 -070094#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -070095#define inline
96#define __inline__
97#endif
98
Bernd Schubert6be9d492008-05-23 13:04:34 -070099#define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
100
Jens Axboe960e7392008-08-15 10:41:18 +0200101/*
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200102 * We maintain a biased count of active stripes in the bottom 16 bits of
103 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
Jens Axboe960e7392008-08-15 10:41:18 +0200104 */
105static inline int raid5_bi_phys_segments(struct bio *bio)
106{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200107 return bio->bi_phys_segments & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200108}
109
110static inline int raid5_bi_hw_segments(struct bio *bio)
111{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200112 return (bio->bi_phys_segments >> 16) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200113}
114
115static inline int raid5_dec_bi_phys_segments(struct bio *bio)
116{
117 --bio->bi_phys_segments;
118 return raid5_bi_phys_segments(bio);
119}
120
121static inline int raid5_dec_bi_hw_segments(struct bio *bio)
122{
123 unsigned short val = raid5_bi_hw_segments(bio);
124
125 --val;
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200126 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
Jens Axboe960e7392008-08-15 10:41:18 +0200127 return val;
128}
129
130static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
131{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200132 bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16);
Jens Axboe960e7392008-08-15 10:41:18 +0200133}
134
NeilBrownd0dabf72009-03-31 14:39:38 +1100135/* Find first data disk in a raid6 stripe */
136static inline int raid6_d0(struct stripe_head *sh)
137{
NeilBrown67cc2b82009-03-31 14:39:38 +1100138 if (sh->ddf_layout)
139 /* ddf always start from first device */
140 return 0;
141 /* md starts just after Q block */
NeilBrownd0dabf72009-03-31 14:39:38 +1100142 if (sh->qd_idx == sh->disks - 1)
143 return 0;
144 else
145 return sh->qd_idx + 1;
146}
NeilBrown16a53ec2006-06-26 00:27:38 -0700147static inline int raid6_next_disk(int disk, int raid_disks)
148{
149 disk++;
150 return (disk < raid_disks) ? disk : 0;
151}
Dan Williamsa4456852007-07-09 11:56:43 -0700152
NeilBrownd0dabf72009-03-31 14:39:38 +1100153/* When walking through the disks in a raid5, starting at raid6_d0,
154 * We need to map each disk to a 'slot', where the data disks are slot
155 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
156 * is raid_disks-1. This help does that mapping.
157 */
NeilBrown67cc2b82009-03-31 14:39:38 +1100158static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
159 int *count, int syndrome_disks)
NeilBrownd0dabf72009-03-31 14:39:38 +1100160{
Dan Williams66295422009-10-19 18:09:32 -0700161 int slot = *count;
NeilBrown67cc2b82009-03-31 14:39:38 +1100162
NeilBrowne4424fe2009-10-16 16:27:34 +1100163 if (sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700164 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100165 if (idx == sh->pd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100166 return syndrome_disks;
NeilBrownd0dabf72009-03-31 14:39:38 +1100167 if (idx == sh->qd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100168 return syndrome_disks + 1;
NeilBrowne4424fe2009-10-16 16:27:34 +1100169 if (!sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700170 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100171 return slot;
172}
173
Dan Williamsa4456852007-07-09 11:56:43 -0700174static void return_io(struct bio *return_bi)
175{
176 struct bio *bi = return_bi;
177 while (bi) {
Dan Williamsa4456852007-07-09 11:56:43 -0700178
179 return_bi = bi->bi_next;
180 bi->bi_next = NULL;
181 bi->bi_size = 0;
Neil Brown0e13fe232008-06-28 08:31:20 +1000182 bio_endio(bi, 0);
Dan Williamsa4456852007-07-09 11:56:43 -0700183 bi = return_bi;
184 }
185}
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187static void print_raid5_conf (raid5_conf_t *conf);
188
Dan Williams600aa102008-06-28 08:32:05 +1000189static int stripe_operations_active(struct stripe_head *sh)
190{
191 return sh->check_state || sh->reconstruct_state ||
192 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
193 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
194}
195
Arjan van de Ven858119e2006-01-14 13:20:43 -0800196static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 if (atomic_dec_and_test(&sh->count)) {
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200199 BUG_ON(!list_empty(&sh->lru));
200 BUG_ON(atomic_read(&conf->active_stripes)==0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 if (test_bit(STRIPE_HANDLE, &sh->state)) {
NeilBrown7c785b72006-07-10 04:44:16 -0700202 if (test_bit(STRIPE_DELAYED, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 list_add_tail(&sh->lru, &conf->delayed_list);
NeilBrown7c785b72006-07-10 04:44:16 -0700204 blk_plug_device(conf->mddev->queue);
205 } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
NeilBrownae3c20c2006-07-10 04:44:17 -0700206 sh->bm_seq - conf->seq_write > 0) {
NeilBrown72626682005-09-09 16:23:54 -0700207 list_add_tail(&sh->lru, &conf->bitmap_list);
NeilBrown7c785b72006-07-10 04:44:16 -0700208 blk_plug_device(conf->mddev->queue);
209 } else {
NeilBrown72626682005-09-09 16:23:54 -0700210 clear_bit(STRIPE_BIT_DELAY, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 list_add_tail(&sh->lru, &conf->handle_list);
NeilBrown72626682005-09-09 16:23:54 -0700212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 md_wakeup_thread(conf->mddev->thread);
214 } else {
Dan Williams600aa102008-06-28 08:32:05 +1000215 BUG_ON(stripe_operations_active(sh));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
217 atomic_dec(&conf->preread_active_stripes);
218 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
219 md_wakeup_thread(conf->mddev->thread);
220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 atomic_dec(&conf->active_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800222 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
223 list_add_tail(&sh->lru, &conf->inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 wake_up(&conf->wait_for_stripe);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800225 if (conf->retry_read_aligned)
226 md_wakeup_thread(conf->mddev->thread);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 }
229 }
230}
NeilBrownd0dabf72009-03-31 14:39:38 +1100231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232static void release_stripe(struct stripe_head *sh)
233{
234 raid5_conf_t *conf = sh->raid_conf;
235 unsigned long flags;
NeilBrown16a53ec2006-06-26 00:27:38 -0700236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 spin_lock_irqsave(&conf->device_lock, flags);
238 __release_stripe(conf, sh);
239 spin_unlock_irqrestore(&conf->device_lock, flags);
240}
241
NeilBrownfccddba2006-01-06 00:20:33 -0800242static inline void remove_hash(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
Dan Williams45b42332007-07-09 11:56:43 -0700244 pr_debug("remove_hash(), stripe %llu\n",
245 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
NeilBrownfccddba2006-01-06 00:20:33 -0800247 hlist_del_init(&sh->hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
NeilBrown16a53ec2006-06-26 00:27:38 -0700250static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
NeilBrownfccddba2006-01-06 00:20:33 -0800252 struct hlist_head *hp = stripe_hash(conf, sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Dan Williams45b42332007-07-09 11:56:43 -0700254 pr_debug("insert_hash(), stripe %llu\n",
255 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257 CHECK_DEVLOCK();
NeilBrownfccddba2006-01-06 00:20:33 -0800258 hlist_add_head(&sh->hash, hp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
261
262/* find an idle stripe, make sure it is unhashed, and return it. */
263static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
264{
265 struct stripe_head *sh = NULL;
266 struct list_head *first;
267
268 CHECK_DEVLOCK();
269 if (list_empty(&conf->inactive_list))
270 goto out;
271 first = conf->inactive_list.next;
272 sh = list_entry(first, struct stripe_head, lru);
273 list_del_init(first);
274 remove_hash(sh);
275 atomic_inc(&conf->active_stripes);
276out:
277 return sh;
278}
279
NeilBrowne4e11e32010-06-16 16:45:16 +1000280static void shrink_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
282 struct page *p;
283 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000284 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
NeilBrowne4e11e32010-06-16 16:45:16 +1000286 for (i = 0; i < num ; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 p = sh->dev[i].page;
288 if (!p)
289 continue;
290 sh->dev[i].page = NULL;
NeilBrown2d1f3b52006-01-06 00:20:31 -0800291 put_page(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 }
293}
294
NeilBrowne4e11e32010-06-16 16:45:16 +1000295static int grow_buffers(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
297 int i;
NeilBrowne4e11e32010-06-16 16:45:16 +1000298 int num = sh->raid_conf->pool_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
NeilBrowne4e11e32010-06-16 16:45:16 +1000300 for (i = 0; i < num; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 struct page *page;
302
303 if (!(page = alloc_page(GFP_KERNEL))) {
304 return 1;
305 }
306 sh->dev[i].page = page;
307 }
308 return 0;
309}
310
NeilBrown784052e2009-03-31 15:19:07 +1100311static void raid5_build_block(struct stripe_head *sh, int i, int previous);
NeilBrown911d4ee2009-03-31 14:39:38 +1100312static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
313 struct stripe_head *sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
NeilBrownb5663ba2009-03-31 14:39:38 +1100315static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
317 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800318 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200320 BUG_ON(atomic_read(&sh->count) != 0);
321 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
Dan Williams600aa102008-06-28 08:32:05 +1000322 BUG_ON(stripe_operations_active(sh));
Dan Williamsd84e0f12007-01-02 13:52:30 -0700323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 CHECK_DEVLOCK();
Dan Williams45b42332007-07-09 11:56:43 -0700325 pr_debug("init_stripe called, stripe %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 (unsigned long long)sh->sector);
327
328 remove_hash(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -0700329
NeilBrown86b42c72009-03-31 15:19:03 +1100330 sh->generation = conf->generation - previous;
NeilBrownb5663ba2009-03-31 14:39:38 +1100331 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 sh->sector = sector;
NeilBrown911d4ee2009-03-31 14:39:38 +1100333 stripe_set_idx(sector, conf, previous, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 sh->state = 0;
335
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800336
337 for (i = sh->disks; i--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 struct r5dev *dev = &sh->dev[i];
339
Dan Williamsd84e0f12007-01-02 13:52:30 -0700340 if (dev->toread || dev->read || dev->towrite || dev->written ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 test_bit(R5_LOCKED, &dev->flags)) {
Dan Williamsd84e0f12007-01-02 13:52:30 -0700342 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 (unsigned long long)sh->sector, i, dev->toread,
Dan Williamsd84e0f12007-01-02 13:52:30 -0700344 dev->read, dev->towrite, dev->written,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 test_bit(R5_LOCKED, &dev->flags));
346 BUG();
347 }
348 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +1100349 raid5_build_block(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
351 insert_hash(conf, sh);
352}
353
NeilBrown86b42c72009-03-31 15:19:03 +1100354static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector,
355 short generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
357 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -0800358 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 CHECK_DEVLOCK();
Dan Williams45b42332007-07-09 11:56:43 -0700361 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
NeilBrownfccddba2006-01-06 00:20:33 -0800362 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
NeilBrown86b42c72009-03-31 15:19:03 +1100363 if (sh->sector == sector && sh->generation == generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 return sh;
Dan Williams45b42332007-07-09 11:56:43 -0700365 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 return NULL;
367}
368
NeilBrown674806d2010-06-16 17:17:53 +1000369/*
370 * Need to check if array has failed when deciding whether to:
371 * - start an array
372 * - remove non-faulty devices
373 * - add a spare
374 * - allow a reshape
375 * This determination is simple when no reshape is happening.
376 * However if there is a reshape, we need to carefully check
377 * both the before and after sections.
378 * This is because some failed devices may only affect one
379 * of the two sections, and some non-in_sync devices may
380 * be insync in the section most affected by failed devices.
381 */
382static int has_failed(raid5_conf_t *conf)
383{
384 int degraded;
385 int i;
386 if (conf->mddev->reshape_position == MaxSector)
387 return conf->mddev->degraded > conf->max_degraded;
388
389 rcu_read_lock();
390 degraded = 0;
391 for (i = 0; i < conf->previous_raid_disks; i++) {
392 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
393 if (!rdev || test_bit(Faulty, &rdev->flags))
394 degraded++;
395 else if (test_bit(In_sync, &rdev->flags))
396 ;
397 else
398 /* not in-sync or faulty.
399 * If the reshape increases the number of devices,
400 * this is being recovered by the reshape, so
401 * this 'previous' section is not in_sync.
402 * If the number of devices is being reduced however,
403 * the device can only be part of the array if
404 * we are reverting a reshape, so this section will
405 * be in-sync.
406 */
407 if (conf->raid_disks >= conf->previous_raid_disks)
408 degraded++;
409 }
410 rcu_read_unlock();
411 if (degraded > conf->max_degraded)
412 return 1;
413 rcu_read_lock();
414 degraded = 0;
415 for (i = 0; i < conf->raid_disks; i++) {
416 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
417 if (!rdev || test_bit(Faulty, &rdev->flags))
418 degraded++;
419 else if (test_bit(In_sync, &rdev->flags))
420 ;
421 else
422 /* not in-sync or faulty.
423 * If reshape increases the number of devices, this
424 * section has already been recovered, else it
425 * almost certainly hasn't.
426 */
427 if (conf->raid_disks <= conf->previous_raid_disks)
428 degraded++;
429 }
430 rcu_read_unlock();
431 if (degraded > conf->max_degraded)
432 return 1;
433 return 0;
434}
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436static void unplug_slaves(mddev_t *mddev);
Jens Axboe165125e2007-07-24 09:28:11 +0200437static void raid5_unplug_device(struct request_queue *q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
NeilBrownb5663ba2009-03-31 14:39:38 +1100439static struct stripe_head *
440get_active_stripe(raid5_conf_t *conf, sector_t sector,
NeilBrowna8c906c2009-06-09 14:39:59 +1000441 int previous, int noblock, int noquiesce)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
443 struct stripe_head *sh;
444
Dan Williams45b42332007-07-09 11:56:43 -0700445 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
447 spin_lock_irq(&conf->device_lock);
448
449 do {
NeilBrown72626682005-09-09 16:23:54 -0700450 wait_event_lock_irq(conf->wait_for_stripe,
NeilBrowna8c906c2009-06-09 14:39:59 +1000451 conf->quiesce == 0 || noquiesce,
NeilBrown72626682005-09-09 16:23:54 -0700452 conf->device_lock, /* nothing */);
NeilBrown86b42c72009-03-31 15:19:03 +1100453 sh = __find_stripe(conf, sector, conf->generation - previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 if (!sh) {
455 if (!conf->inactive_blocked)
456 sh = get_free_stripe(conf);
457 if (noblock && sh == NULL)
458 break;
459 if (!sh) {
460 conf->inactive_blocked = 1;
461 wait_event_lock_irq(conf->wait_for_stripe,
462 !list_empty(&conf->inactive_list) &&
NeilBrown50368052005-12-12 02:39:17 -0800463 (atomic_read(&conf->active_stripes)
464 < (conf->max_nr_stripes *3/4)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 || !conf->inactive_blocked),
466 conf->device_lock,
NeilBrownf4370782006-07-10 04:44:14 -0700467 raid5_unplug_device(conf->mddev->queue)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 );
469 conf->inactive_blocked = 0;
470 } else
NeilBrownb5663ba2009-03-31 14:39:38 +1100471 init_stripe(sh, sector, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 } else {
473 if (atomic_read(&sh->count)) {
NeilBrownab69ae12009-03-31 15:26:47 +1100474 BUG_ON(!list_empty(&sh->lru)
475 && !test_bit(STRIPE_EXPANDING, &sh->state));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 } else {
477 if (!test_bit(STRIPE_HANDLE, &sh->state))
478 atomic_inc(&conf->active_stripes);
NeilBrownff4e8d92006-07-10 04:44:16 -0700479 if (list_empty(&sh->lru) &&
480 !test_bit(STRIPE_EXPANDING, &sh->state))
NeilBrown16a53ec2006-06-26 00:27:38 -0700481 BUG();
482 list_del_init(&sh->lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 }
484 }
485 } while (sh == NULL);
486
487 if (sh)
488 atomic_inc(&sh->count);
489
490 spin_unlock_irq(&conf->device_lock);
491 return sh;
492}
493
NeilBrown6712ecf2007-09-27 12:47:43 +0200494static void
495raid5_end_read_request(struct bio *bi, int error);
496static void
497raid5_end_write_request(struct bio *bi, int error);
Dan Williams91c00922007-01-02 13:52:30 -0700498
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000499static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
Dan Williams91c00922007-01-02 13:52:30 -0700500{
501 raid5_conf_t *conf = sh->raid_conf;
502 int i, disks = sh->disks;
503
504 might_sleep();
505
506 for (i = disks; i--; ) {
507 int rw;
508 struct bio *bi;
509 mdk_rdev_t *rdev;
510 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags))
511 rw = WRITE;
512 else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
513 rw = READ;
514 else
515 continue;
516
517 bi = &sh->dev[i].req;
518
519 bi->bi_rw = rw;
520 if (rw == WRITE)
521 bi->bi_end_io = raid5_end_write_request;
522 else
523 bi->bi_end_io = raid5_end_read_request;
524
525 rcu_read_lock();
526 rdev = rcu_dereference(conf->disks[i].rdev);
527 if (rdev && test_bit(Faulty, &rdev->flags))
528 rdev = NULL;
529 if (rdev)
530 atomic_inc(&rdev->nr_pending);
531 rcu_read_unlock();
532
533 if (rdev) {
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000534 if (s->syncing || s->expanding || s->expanded)
Dan Williams91c00922007-01-02 13:52:30 -0700535 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
536
Dan Williams2b7497f2008-06-28 08:31:52 +1000537 set_bit(STRIPE_IO_STARTED, &sh->state);
538
Dan Williams91c00922007-01-02 13:52:30 -0700539 bi->bi_bdev = rdev->bdev;
540 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700541 __func__, (unsigned long long)sh->sector,
Dan Williams91c00922007-01-02 13:52:30 -0700542 bi->bi_rw, i);
543 atomic_inc(&sh->count);
544 bi->bi_sector = sh->sector + rdev->data_offset;
545 bi->bi_flags = 1 << BIO_UPTODATE;
546 bi->bi_vcnt = 1;
547 bi->bi_max_vecs = 1;
548 bi->bi_idx = 0;
549 bi->bi_io_vec = &sh->dev[i].vec;
550 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
551 bi->bi_io_vec[0].bv_offset = 0;
552 bi->bi_size = STRIPE_SIZE;
553 bi->bi_next = NULL;
554 if (rw == WRITE &&
555 test_bit(R5_ReWrite, &sh->dev[i].flags))
556 atomic_add(STRIPE_SECTORS,
557 &rdev->corrected_errors);
558 generic_make_request(bi);
559 } else {
560 if (rw == WRITE)
561 set_bit(STRIPE_DEGRADED, &sh->state);
562 pr_debug("skip op %ld on disc %d for sector %llu\n",
563 bi->bi_rw, i, (unsigned long long)sh->sector);
564 clear_bit(R5_LOCKED, &sh->dev[i].flags);
565 set_bit(STRIPE_HANDLE, &sh->state);
566 }
567 }
568}
569
570static struct dma_async_tx_descriptor *
571async_copy_data(int frombio, struct bio *bio, struct page *page,
572 sector_t sector, struct dma_async_tx_descriptor *tx)
573{
574 struct bio_vec *bvl;
575 struct page *bio_page;
576 int i;
577 int page_offset;
Dan Williamsa08abd82009-06-03 11:43:59 -0700578 struct async_submit_ctl submit;
Dan Williams0403e382009-09-08 17:42:50 -0700579 enum async_tx_flags flags = 0;
Dan Williams91c00922007-01-02 13:52:30 -0700580
581 if (bio->bi_sector >= sector)
582 page_offset = (signed)(bio->bi_sector - sector) * 512;
583 else
584 page_offset = (signed)(sector - bio->bi_sector) * -512;
Dan Williamsa08abd82009-06-03 11:43:59 -0700585
Dan Williams0403e382009-09-08 17:42:50 -0700586 if (frombio)
587 flags |= ASYNC_TX_FENCE;
588 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
589
Dan Williams91c00922007-01-02 13:52:30 -0700590 bio_for_each_segment(bvl, bio, i) {
591 int len = bio_iovec_idx(bio, i)->bv_len;
592 int clen;
593 int b_offset = 0;
594
595 if (page_offset < 0) {
596 b_offset = -page_offset;
597 page_offset += b_offset;
598 len -= b_offset;
599 }
600
601 if (len > 0 && page_offset + len > STRIPE_SIZE)
602 clen = STRIPE_SIZE - page_offset;
603 else
604 clen = len;
605
606 if (clen > 0) {
607 b_offset += bio_iovec_idx(bio, i)->bv_offset;
608 bio_page = bio_iovec_idx(bio, i)->bv_page;
609 if (frombio)
610 tx = async_memcpy(page, bio_page, page_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700611 b_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700612 else
613 tx = async_memcpy(bio_page, page, b_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700614 page_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700615 }
Dan Williamsa08abd82009-06-03 11:43:59 -0700616 /* chain the operations */
617 submit.depend_tx = tx;
618
Dan Williams91c00922007-01-02 13:52:30 -0700619 if (clen < len) /* hit end of page */
620 break;
621 page_offset += len;
622 }
623
624 return tx;
625}
626
627static void ops_complete_biofill(void *stripe_head_ref)
628{
629 struct stripe_head *sh = stripe_head_ref;
630 struct bio *return_bi = NULL;
631 raid5_conf_t *conf = sh->raid_conf;
Dan Williamse4d84902007-09-24 10:06:13 -0700632 int i;
Dan Williams91c00922007-01-02 13:52:30 -0700633
Harvey Harrisone46b2722008-04-28 02:15:50 -0700634 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700635 (unsigned long long)sh->sector);
636
637 /* clear completed biofills */
Dan Williams83de75c2008-06-28 08:31:58 +1000638 spin_lock_irq(&conf->device_lock);
Dan Williams91c00922007-01-02 13:52:30 -0700639 for (i = sh->disks; i--; ) {
640 struct r5dev *dev = &sh->dev[i];
Dan Williams91c00922007-01-02 13:52:30 -0700641
642 /* acknowledge completion of a biofill operation */
Dan Williamse4d84902007-09-24 10:06:13 -0700643 /* and check if we need to reply to a read request,
644 * new R5_Wantfill requests are held off until
Dan Williams83de75c2008-06-28 08:31:58 +1000645 * !STRIPE_BIOFILL_RUN
Dan Williamse4d84902007-09-24 10:06:13 -0700646 */
647 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -0700648 struct bio *rbi, *rbi2;
Dan Williams91c00922007-01-02 13:52:30 -0700649
Dan Williams91c00922007-01-02 13:52:30 -0700650 BUG_ON(!dev->read);
651 rbi = dev->read;
652 dev->read = NULL;
653 while (rbi && rbi->bi_sector <
654 dev->sector + STRIPE_SECTORS) {
655 rbi2 = r5_next_bio(rbi, dev->sector);
Jens Axboe960e7392008-08-15 10:41:18 +0200656 if (!raid5_dec_bi_phys_segments(rbi)) {
Dan Williams91c00922007-01-02 13:52:30 -0700657 rbi->bi_next = return_bi;
658 return_bi = rbi;
659 }
Dan Williams91c00922007-01-02 13:52:30 -0700660 rbi = rbi2;
661 }
662 }
663 }
Dan Williams83de75c2008-06-28 08:31:58 +1000664 spin_unlock_irq(&conf->device_lock);
665 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -0700666
667 return_io(return_bi);
668
Dan Williamse4d84902007-09-24 10:06:13 -0700669 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -0700670 release_stripe(sh);
671}
672
673static void ops_run_biofill(struct stripe_head *sh)
674{
675 struct dma_async_tx_descriptor *tx = NULL;
676 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsa08abd82009-06-03 11:43:59 -0700677 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700678 int i;
679
Harvey Harrisone46b2722008-04-28 02:15:50 -0700680 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700681 (unsigned long long)sh->sector);
682
683 for (i = sh->disks; i--; ) {
684 struct r5dev *dev = &sh->dev[i];
685 if (test_bit(R5_Wantfill, &dev->flags)) {
686 struct bio *rbi;
687 spin_lock_irq(&conf->device_lock);
688 dev->read = rbi = dev->toread;
689 dev->toread = NULL;
690 spin_unlock_irq(&conf->device_lock);
691 while (rbi && rbi->bi_sector <
692 dev->sector + STRIPE_SECTORS) {
693 tx = async_copy_data(0, rbi, dev->page,
694 dev->sector, tx);
695 rbi = r5_next_bio(rbi, dev->sector);
696 }
697 }
698 }
699
700 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -0700701 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
702 async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -0700703}
704
Dan Williams4e7d2c02009-08-29 19:13:11 -0700705static void mark_target_uptodate(struct stripe_head *sh, int target)
706{
707 struct r5dev *tgt;
708
709 if (target < 0)
710 return;
711
712 tgt = &sh->dev[target];
713 set_bit(R5_UPTODATE, &tgt->flags);
714 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
715 clear_bit(R5_Wantcompute, &tgt->flags);
716}
717
Dan Williamsac6b53b2009-07-14 13:40:19 -0700718static void ops_complete_compute(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -0700719{
720 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -0700721
Harvey Harrisone46b2722008-04-28 02:15:50 -0700722 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700723 (unsigned long long)sh->sector);
724
Dan Williamsac6b53b2009-07-14 13:40:19 -0700725 /* mark the computed target(s) as uptodate */
Dan Williams4e7d2c02009-08-29 19:13:11 -0700726 mark_target_uptodate(sh, sh->ops.target);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700727 mark_target_uptodate(sh, sh->ops.target2);
Dan Williams4e7d2c02009-08-29 19:13:11 -0700728
Dan Williamsecc65c92008-06-28 08:31:57 +1000729 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
730 if (sh->check_state == check_state_compute_run)
731 sh->check_state = check_state_compute_result;
Dan Williams91c00922007-01-02 13:52:30 -0700732 set_bit(STRIPE_HANDLE, &sh->state);
733 release_stripe(sh);
734}
735
Dan Williamsd6f38f32009-07-14 11:50:52 -0700736/* return a pointer to the address conversion region of the scribble buffer */
737static addr_conv_t *to_addr_conv(struct stripe_head *sh,
738 struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -0700739{
Dan Williamsd6f38f32009-07-14 11:50:52 -0700740 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
741}
742
743static struct dma_async_tx_descriptor *
744ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
745{
Dan Williams91c00922007-01-02 13:52:30 -0700746 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -0700747 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -0700748 int target = sh->ops.target;
749 struct r5dev *tgt = &sh->dev[target];
750 struct page *xor_dest = tgt->page;
751 int count = 0;
752 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -0700753 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700754 int i;
755
756 pr_debug("%s: stripe %llu block: %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700757 __func__, (unsigned long long)sh->sector, target);
Dan Williams91c00922007-01-02 13:52:30 -0700758 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
759
760 for (i = disks; i--; )
761 if (i != target)
762 xor_srcs[count++] = sh->dev[i].page;
763
764 atomic_inc(&sh->count);
765
Dan Williams0403e382009-09-08 17:42:50 -0700766 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700767 ops_complete_compute, sh, to_addr_conv(sh, percpu));
Dan Williams91c00922007-01-02 13:52:30 -0700768 if (unlikely(count == 1))
Dan Williamsa08abd82009-06-03 11:43:59 -0700769 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700770 else
Dan Williamsa08abd82009-06-03 11:43:59 -0700771 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700772
Dan Williams91c00922007-01-02 13:52:30 -0700773 return tx;
774}
775
Dan Williamsac6b53b2009-07-14 13:40:19 -0700776/* set_syndrome_sources - populate source buffers for gen_syndrome
777 * @srcs - (struct page *) array of size sh->disks
778 * @sh - stripe_head to parse
779 *
780 * Populates srcs in proper layout order for the stripe and returns the
781 * 'count' of sources to be used in a call to async_gen_syndrome. The P
782 * destination buffer is recorded in srcs[count] and the Q destination
783 * is recorded in srcs[count+1]].
784 */
785static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
786{
787 int disks = sh->disks;
788 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
789 int d0_idx = raid6_d0(sh);
790 int count;
791 int i;
792
793 for (i = 0; i < disks; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +1100794 srcs[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700795
796 count = 0;
797 i = d0_idx;
798 do {
799 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
800
801 srcs[slot] = sh->dev[i].page;
802 i = raid6_next_disk(i, disks);
803 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700804
NeilBrowne4424fe2009-10-16 16:27:34 +1100805 return syndrome_disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700806}
807
808static struct dma_async_tx_descriptor *
809ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
810{
811 int disks = sh->disks;
812 struct page **blocks = percpu->scribble;
813 int target;
814 int qd_idx = sh->qd_idx;
815 struct dma_async_tx_descriptor *tx;
816 struct async_submit_ctl submit;
817 struct r5dev *tgt;
818 struct page *dest;
819 int i;
820 int count;
821
822 if (sh->ops.target < 0)
823 target = sh->ops.target2;
824 else if (sh->ops.target2 < 0)
825 target = sh->ops.target;
826 else
827 /* we should only have one valid target */
828 BUG();
829 BUG_ON(target < 0);
830 pr_debug("%s: stripe %llu block: %d\n",
831 __func__, (unsigned long long)sh->sector, target);
832
833 tgt = &sh->dev[target];
834 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
835 dest = tgt->page;
836
837 atomic_inc(&sh->count);
838
839 if (target == qd_idx) {
840 count = set_syndrome_sources(blocks, sh);
841 blocks[count] = NULL; /* regenerating p is not necessary */
842 BUG_ON(blocks[count+1] != dest); /* q should already be set */
Dan Williams0403e382009-09-08 17:42:50 -0700843 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
844 ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700845 to_addr_conv(sh, percpu));
846 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
847 } else {
848 /* Compute any data- or p-drive using XOR */
849 count = 0;
850 for (i = disks; i-- ; ) {
851 if (i == target || i == qd_idx)
852 continue;
853 blocks[count++] = sh->dev[i].page;
854 }
855
Dan Williams0403e382009-09-08 17:42:50 -0700856 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
857 NULL, ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700858 to_addr_conv(sh, percpu));
859 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
860 }
861
862 return tx;
863}
864
865static struct dma_async_tx_descriptor *
866ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
867{
868 int i, count, disks = sh->disks;
869 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
870 int d0_idx = raid6_d0(sh);
871 int faila = -1, failb = -1;
872 int target = sh->ops.target;
873 int target2 = sh->ops.target2;
874 struct r5dev *tgt = &sh->dev[target];
875 struct r5dev *tgt2 = &sh->dev[target2];
876 struct dma_async_tx_descriptor *tx;
877 struct page **blocks = percpu->scribble;
878 struct async_submit_ctl submit;
879
880 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
881 __func__, (unsigned long long)sh->sector, target, target2);
882 BUG_ON(target < 0 || target2 < 0);
883 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
884 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
885
Dan Williams6c910a72009-09-16 12:24:54 -0700886 /* we need to open-code set_syndrome_sources to handle the
Dan Williamsac6b53b2009-07-14 13:40:19 -0700887 * slot number conversion for 'faila' and 'failb'
888 */
889 for (i = 0; i < disks ; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +1100890 blocks[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700891 count = 0;
892 i = d0_idx;
893 do {
894 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
895
896 blocks[slot] = sh->dev[i].page;
897
898 if (i == target)
899 faila = slot;
900 if (i == target2)
901 failb = slot;
902 i = raid6_next_disk(i, disks);
903 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700904
905 BUG_ON(faila == failb);
906 if (failb < faila)
907 swap(faila, failb);
908 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
909 __func__, (unsigned long long)sh->sector, faila, failb);
910
911 atomic_inc(&sh->count);
912
913 if (failb == syndrome_disks+1) {
914 /* Q disk is one of the missing disks */
915 if (faila == syndrome_disks) {
916 /* Missing P+Q, just recompute */
Dan Williams0403e382009-09-08 17:42:50 -0700917 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
918 ops_complete_compute, sh,
919 to_addr_conv(sh, percpu));
NeilBrowne4424fe2009-10-16 16:27:34 +1100920 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700921 STRIPE_SIZE, &submit);
922 } else {
923 struct page *dest;
924 int data_target;
925 int qd_idx = sh->qd_idx;
926
927 /* Missing D+Q: recompute D from P, then recompute Q */
928 if (target == qd_idx)
929 data_target = target2;
930 else
931 data_target = target;
932
933 count = 0;
934 for (i = disks; i-- ; ) {
935 if (i == data_target || i == qd_idx)
936 continue;
937 blocks[count++] = sh->dev[i].page;
938 }
939 dest = sh->dev[data_target].page;
Dan Williams0403e382009-09-08 17:42:50 -0700940 init_async_submit(&submit,
941 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
942 NULL, NULL, NULL,
943 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -0700944 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
945 &submit);
946
947 count = set_syndrome_sources(blocks, sh);
Dan Williams0403e382009-09-08 17:42:50 -0700948 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
949 ops_complete_compute, sh,
950 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -0700951 return async_gen_syndrome(blocks, 0, count+2,
952 STRIPE_SIZE, &submit);
953 }
Dan Williamsac6b53b2009-07-14 13:40:19 -0700954 } else {
Dan Williams6c910a72009-09-16 12:24:54 -0700955 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
956 ops_complete_compute, sh,
957 to_addr_conv(sh, percpu));
958 if (failb == syndrome_disks) {
959 /* We're missing D+P. */
960 return async_raid6_datap_recov(syndrome_disks+2,
961 STRIPE_SIZE, faila,
962 blocks, &submit);
963 } else {
964 /* We're missing D+D. */
965 return async_raid6_2data_recov(syndrome_disks+2,
966 STRIPE_SIZE, faila, failb,
967 blocks, &submit);
968 }
Dan Williamsac6b53b2009-07-14 13:40:19 -0700969 }
970}
971
972
Dan Williams91c00922007-01-02 13:52:30 -0700973static void ops_complete_prexor(void *stripe_head_ref)
974{
975 struct stripe_head *sh = stripe_head_ref;
976
Harvey Harrisone46b2722008-04-28 02:15:50 -0700977 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700978 (unsigned long long)sh->sector);
Dan Williams91c00922007-01-02 13:52:30 -0700979}
980
981static struct dma_async_tx_descriptor *
Dan Williamsd6f38f32009-07-14 11:50:52 -0700982ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
983 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -0700984{
Dan Williams91c00922007-01-02 13:52:30 -0700985 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -0700986 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -0700987 int count = 0, pd_idx = sh->pd_idx, i;
Dan Williamsa08abd82009-06-03 11:43:59 -0700988 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700989
990 /* existing parity data subtracted */
991 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
992
Harvey Harrisone46b2722008-04-28 02:15:50 -0700993 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700994 (unsigned long long)sh->sector);
995
996 for (i = disks; i--; ) {
997 struct r5dev *dev = &sh->dev[i];
998 /* Only process blocks that are known to be uptodate */
Dan Williamsd8ee0722008-06-28 08:32:06 +1000999 if (test_bit(R5_Wantdrain, &dev->flags))
Dan Williams91c00922007-01-02 13:52:30 -07001000 xor_srcs[count++] = dev->page;
1001 }
1002
Dan Williams0403e382009-09-08 17:42:50 -07001003 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001004 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001005 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001006
1007 return tx;
1008}
1009
1010static struct dma_async_tx_descriptor *
Dan Williamsd8ee0722008-06-28 08:32:06 +10001011ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001012{
1013 int disks = sh->disks;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001014 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001015
Harvey Harrisone46b2722008-04-28 02:15:50 -07001016 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001017 (unsigned long long)sh->sector);
1018
1019 for (i = disks; i--; ) {
1020 struct r5dev *dev = &sh->dev[i];
1021 struct bio *chosen;
Dan Williams91c00922007-01-02 13:52:30 -07001022
Dan Williamsd8ee0722008-06-28 08:32:06 +10001023 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -07001024 struct bio *wbi;
1025
1026 spin_lock(&sh->lock);
1027 chosen = dev->towrite;
1028 dev->towrite = NULL;
1029 BUG_ON(dev->written);
1030 wbi = dev->written = chosen;
1031 spin_unlock(&sh->lock);
1032
1033 while (wbi && wbi->bi_sector <
1034 dev->sector + STRIPE_SECTORS) {
1035 tx = async_copy_data(1, wbi, dev->page,
1036 dev->sector, tx);
1037 wbi = r5_next_bio(wbi, dev->sector);
1038 }
1039 }
1040 }
1041
1042 return tx;
1043}
1044
Dan Williamsac6b53b2009-07-14 13:40:19 -07001045static void ops_complete_reconstruct(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -07001046{
1047 struct stripe_head *sh = stripe_head_ref;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001048 int disks = sh->disks;
1049 int pd_idx = sh->pd_idx;
1050 int qd_idx = sh->qd_idx;
1051 int i;
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
1056 for (i = disks; i--; ) {
1057 struct r5dev *dev = &sh->dev[i];
Dan Williamsac6b53b2009-07-14 13:40:19 -07001058
1059 if (dev->written || i == pd_idx || i == qd_idx)
Dan Williams91c00922007-01-02 13:52:30 -07001060 set_bit(R5_UPTODATE, &dev->flags);
1061 }
1062
Dan Williamsd8ee0722008-06-28 08:32:06 +10001063 if (sh->reconstruct_state == reconstruct_state_drain_run)
1064 sh->reconstruct_state = reconstruct_state_drain_result;
1065 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
1066 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
1067 else {
1068 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
1069 sh->reconstruct_state = reconstruct_state_result;
1070 }
Dan Williams91c00922007-01-02 13:52:30 -07001071
1072 set_bit(STRIPE_HANDLE, &sh->state);
1073 release_stripe(sh);
1074}
1075
1076static void
Dan Williamsac6b53b2009-07-14 13:40:19 -07001077ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1078 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001079{
Dan Williams91c00922007-01-02 13:52:30 -07001080 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001081 struct page **xor_srcs = percpu->scribble;
Dan Williamsa08abd82009-06-03 11:43:59 -07001082 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001083 int count = 0, pd_idx = sh->pd_idx, i;
1084 struct page *xor_dest;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001085 int prexor = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001086 unsigned long flags;
Dan Williams91c00922007-01-02 13:52:30 -07001087
Harvey Harrisone46b2722008-04-28 02:15:50 -07001088 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001089 (unsigned long long)sh->sector);
1090
1091 /* check if prexor is active which means only process blocks
1092 * that are part of a read-modify-write (written)
1093 */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001094 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1095 prexor = 1;
Dan Williams91c00922007-01-02 13:52:30 -07001096 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1097 for (i = disks; i--; ) {
1098 struct r5dev *dev = &sh->dev[i];
1099 if (dev->written)
1100 xor_srcs[count++] = dev->page;
1101 }
1102 } else {
1103 xor_dest = sh->dev[pd_idx].page;
1104 for (i = disks; i--; ) {
1105 struct r5dev *dev = &sh->dev[i];
1106 if (i != pd_idx)
1107 xor_srcs[count++] = dev->page;
1108 }
1109 }
1110
Dan Williams91c00922007-01-02 13:52:30 -07001111 /* 1/ if we prexor'd then the dest is reused as a source
1112 * 2/ if we did not prexor then we are redoing the parity
1113 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1114 * for the synchronous xor case
1115 */
Dan Williams88ba2aa2009-04-09 16:16:18 -07001116 flags = ASYNC_TX_ACK |
Dan Williams91c00922007-01-02 13:52:30 -07001117 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1118
1119 atomic_inc(&sh->count);
1120
Dan Williamsac6b53b2009-07-14 13:40:19 -07001121 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001122 to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001123 if (unlikely(count == 1))
1124 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1125 else
1126 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001127}
1128
Dan Williamsac6b53b2009-07-14 13:40:19 -07001129static void
1130ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1131 struct dma_async_tx_descriptor *tx)
1132{
1133 struct async_submit_ctl submit;
1134 struct page **blocks = percpu->scribble;
1135 int count;
1136
1137 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1138
1139 count = set_syndrome_sources(blocks, sh);
1140
1141 atomic_inc(&sh->count);
1142
1143 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1144 sh, to_addr_conv(sh, percpu));
1145 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001146}
1147
1148static void ops_complete_check(void *stripe_head_ref)
1149{
1150 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001151
Harvey Harrisone46b2722008-04-28 02:15:50 -07001152 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001153 (unsigned long long)sh->sector);
1154
Dan Williamsecc65c92008-06-28 08:31:57 +10001155 sh->check_state = check_state_check_result;
Dan Williams91c00922007-01-02 13:52:30 -07001156 set_bit(STRIPE_HANDLE, &sh->state);
1157 release_stripe(sh);
1158}
1159
Dan Williamsac6b53b2009-07-14 13:40:19 -07001160static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001161{
Dan Williams91c00922007-01-02 13:52:30 -07001162 int disks = sh->disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001163 int pd_idx = sh->pd_idx;
1164 int qd_idx = sh->qd_idx;
1165 struct page *xor_dest;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001166 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001167 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001168 struct async_submit_ctl submit;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001169 int count;
1170 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001171
Harvey Harrisone46b2722008-04-28 02:15:50 -07001172 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001173 (unsigned long long)sh->sector);
1174
Dan Williamsac6b53b2009-07-14 13:40:19 -07001175 count = 0;
1176 xor_dest = sh->dev[pd_idx].page;
1177 xor_srcs[count++] = xor_dest;
Dan Williams91c00922007-01-02 13:52:30 -07001178 for (i = disks; i--; ) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001179 if (i == pd_idx || i == qd_idx)
1180 continue;
1181 xor_srcs[count++] = sh->dev[i].page;
Dan Williams91c00922007-01-02 13:52:30 -07001182 }
1183
Dan Williamsd6f38f32009-07-14 11:50:52 -07001184 init_async_submit(&submit, 0, NULL, NULL, NULL,
1185 to_addr_conv(sh, percpu));
Dan Williams099f53c2009-04-08 14:28:37 -07001186 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07001187 &sh->ops.zero_sum_result, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001188
Dan Williams91c00922007-01-02 13:52:30 -07001189 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001190 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1191 tx = async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001192}
1193
Dan Williamsac6b53b2009-07-14 13:40:19 -07001194static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1195{
1196 struct page **srcs = percpu->scribble;
1197 struct async_submit_ctl submit;
1198 int count;
1199
1200 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1201 (unsigned long long)sh->sector, checkp);
1202
1203 count = set_syndrome_sources(srcs, sh);
1204 if (!checkp)
1205 srcs[count] = NULL;
1206
1207 atomic_inc(&sh->count);
1208 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1209 sh, to_addr_conv(sh, percpu));
1210 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1211 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
1212}
1213
Dan Williams417b8d42009-10-16 16:25:22 +11001214static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
Dan Williams91c00922007-01-02 13:52:30 -07001215{
1216 int overlap_clear = 0, i, disks = sh->disks;
1217 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001218 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001219 int level = conf->level;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001220 struct raid5_percpu *percpu;
1221 unsigned long cpu;
Dan Williams91c00922007-01-02 13:52:30 -07001222
Dan Williamsd6f38f32009-07-14 11:50:52 -07001223 cpu = get_cpu();
1224 percpu = per_cpu_ptr(conf->percpu, cpu);
Dan Williams83de75c2008-06-28 08:31:58 +10001225 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
Dan Williams91c00922007-01-02 13:52:30 -07001226 ops_run_biofill(sh);
1227 overlap_clear++;
1228 }
1229
Dan Williams7b3a8712008-06-28 08:32:09 +10001230 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001231 if (level < 6)
1232 tx = ops_run_compute5(sh, percpu);
1233 else {
1234 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1235 tx = ops_run_compute6_1(sh, percpu);
1236 else
1237 tx = ops_run_compute6_2(sh, percpu);
1238 }
1239 /* terminate the chain if reconstruct is not set to be run */
1240 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
Dan Williams7b3a8712008-06-28 08:32:09 +10001241 async_tx_ack(tx);
1242 }
Dan Williams91c00922007-01-02 13:52:30 -07001243
Dan Williams600aa102008-06-28 08:32:05 +10001244 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
Dan Williamsd6f38f32009-07-14 11:50:52 -07001245 tx = ops_run_prexor(sh, percpu, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001246
Dan Williams600aa102008-06-28 08:32:05 +10001247 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
Dan Williamsd8ee0722008-06-28 08:32:06 +10001248 tx = ops_run_biodrain(sh, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001249 overlap_clear++;
1250 }
1251
Dan Williamsac6b53b2009-07-14 13:40:19 -07001252 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1253 if (level < 6)
1254 ops_run_reconstruct5(sh, percpu, tx);
1255 else
1256 ops_run_reconstruct6(sh, percpu, tx);
1257 }
Dan Williams91c00922007-01-02 13:52:30 -07001258
Dan Williamsac6b53b2009-07-14 13:40:19 -07001259 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1260 if (sh->check_state == check_state_run)
1261 ops_run_check_p(sh, percpu);
1262 else if (sh->check_state == check_state_run_q)
1263 ops_run_check_pq(sh, percpu, 0);
1264 else if (sh->check_state == check_state_run_pq)
1265 ops_run_check_pq(sh, percpu, 1);
1266 else
1267 BUG();
1268 }
Dan Williams91c00922007-01-02 13:52:30 -07001269
Dan Williams91c00922007-01-02 13:52:30 -07001270 if (overlap_clear)
1271 for (i = disks; i--; ) {
1272 struct r5dev *dev = &sh->dev[i];
1273 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1274 wake_up(&sh->raid_conf->wait_for_overlap);
1275 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07001276 put_cpu();
Dan Williams91c00922007-01-02 13:52:30 -07001277}
1278
Dan Williams417b8d42009-10-16 16:25:22 +11001279#ifdef CONFIG_MULTICORE_RAID456
1280static void async_run_ops(void *param, async_cookie_t cookie)
1281{
1282 struct stripe_head *sh = param;
1283 unsigned long ops_request = sh->ops.request;
1284
1285 clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state);
1286 wake_up(&sh->ops.wait_for_ops);
1287
1288 __raid_run_ops(sh, ops_request);
1289 release_stripe(sh);
1290}
1291
1292static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
1293{
1294 /* since handle_stripe can be called outside of raid5d context
1295 * we need to ensure sh->ops.request is de-staged before another
1296 * request arrives
1297 */
1298 wait_event(sh->ops.wait_for_ops,
1299 !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state));
1300 sh->ops.request = ops_request;
1301
1302 atomic_inc(&sh->count);
1303 async_schedule(async_run_ops, sh);
1304}
1305#else
1306#define raid_run_ops __raid_run_ops
1307#endif
1308
NeilBrown3f294f42005-11-08 21:39:25 -08001309static int grow_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310{
1311 struct stripe_head *sh;
NeilBrown3f294f42005-11-08 21:39:25 -08001312 sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL);
1313 if (!sh)
1314 return 0;
NeilBrowne4e11e32010-06-16 16:45:16 +10001315 memset(sh, 0, sizeof(*sh) + (conf->pool_size-1)*sizeof(struct r5dev));
NeilBrown3f294f42005-11-08 21:39:25 -08001316 sh->raid_conf = conf;
1317 spin_lock_init(&sh->lock);
Dan Williams417b8d42009-10-16 16:25:22 +11001318 #ifdef CONFIG_MULTICORE_RAID456
1319 init_waitqueue_head(&sh->ops.wait_for_ops);
1320 #endif
NeilBrown3f294f42005-11-08 21:39:25 -08001321
NeilBrowne4e11e32010-06-16 16:45:16 +10001322 if (grow_buffers(sh)) {
1323 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001324 kmem_cache_free(conf->slab_cache, sh);
1325 return 0;
1326 }
1327 /* we just created an active stripe so... */
1328 atomic_set(&sh->count, 1);
1329 atomic_inc(&conf->active_stripes);
1330 INIT_LIST_HEAD(&sh->lru);
1331 release_stripe(sh);
1332 return 1;
1333}
1334
1335static int grow_stripes(raid5_conf_t *conf, int num)
1336{
Christoph Lametere18b8902006-12-06 20:33:20 -08001337 struct kmem_cache *sc;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001338 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
NeilBrown245f46c2009-03-31 14:39:39 +11001340 sprintf(conf->cache_name[0],
1341 "raid%d-%s", conf->level, mdname(conf->mddev));
1342 sprintf(conf->cache_name[1],
1343 "raid%d-%s-alt", conf->level, mdname(conf->mddev));
NeilBrownad01c9e2006-03-27 01:18:07 -08001344 conf->active_name = 0;
1345 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001347 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 if (!sc)
1349 return 1;
1350 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001351 conf->pool_size = devs;
NeilBrown16a53ec2006-06-26 00:27:38 -07001352 while (num--)
NeilBrown3f294f42005-11-08 21:39:25 -08001353 if (!grow_one_stripe(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 return 0;
1356}
NeilBrown29269552006-03-27 01:18:10 -08001357
Dan Williamsd6f38f32009-07-14 11:50:52 -07001358/**
1359 * scribble_len - return the required size of the scribble region
1360 * @num - total number of disks in the array
1361 *
1362 * The size must be enough to contain:
1363 * 1/ a struct page pointer for each device in the array +2
1364 * 2/ room to convert each entry in (1) to its corresponding dma
1365 * (dma_map_page()) or page (page_address()) address.
1366 *
1367 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1368 * calculate over all devices (not just the data blocks), using zeros in place
1369 * of the P and Q blocks.
1370 */
1371static size_t scribble_len(int num)
1372{
1373 size_t len;
1374
1375 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1376
1377 return len;
1378}
1379
NeilBrownad01c9e2006-03-27 01:18:07 -08001380static int resize_stripes(raid5_conf_t *conf, int newsize)
1381{
1382 /* Make all the stripes able to hold 'newsize' devices.
1383 * New slots in each stripe get 'page' set to a new page.
1384 *
1385 * This happens in stages:
1386 * 1/ create a new kmem_cache and allocate the required number of
1387 * stripe_heads.
1388 * 2/ gather all the old stripe_heads and tranfer the pages across
1389 * to the new stripe_heads. This will have the side effect of
1390 * freezing the array as once all stripe_heads have been collected,
1391 * no IO will be possible. Old stripe heads are freed once their
1392 * pages have been transferred over, and the old kmem_cache is
1393 * freed when all stripes are done.
1394 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1395 * we simple return a failre status - no need to clean anything up.
1396 * 4/ allocate new pages for the new slots in the new stripe_heads.
1397 * If this fails, we don't bother trying the shrink the
1398 * stripe_heads down again, we just leave them as they are.
1399 * As each stripe_head is processed the new one is released into
1400 * active service.
1401 *
1402 * Once step2 is started, we cannot afford to wait for a write,
1403 * so we use GFP_NOIO allocations.
1404 */
1405 struct stripe_head *osh, *nsh;
1406 LIST_HEAD(newstripes);
1407 struct disk_info *ndisks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001408 unsigned long cpu;
Dan Williamsb5470dc2008-06-27 21:44:04 -07001409 int err;
Christoph Lametere18b8902006-12-06 20:33:20 -08001410 struct kmem_cache *sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001411 int i;
1412
1413 if (newsize <= conf->pool_size)
1414 return 0; /* never bother to shrink */
1415
Dan Williamsb5470dc2008-06-27 21:44:04 -07001416 err = md_allow_write(conf->mddev);
1417 if (err)
1418 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08001419
NeilBrownad01c9e2006-03-27 01:18:07 -08001420 /* Step 1 */
1421 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1422 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001423 0, 0, NULL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001424 if (!sc)
1425 return -ENOMEM;
1426
1427 for (i = conf->max_nr_stripes; i; i--) {
1428 nsh = kmem_cache_alloc(sc, GFP_KERNEL);
1429 if (!nsh)
1430 break;
1431
1432 memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev));
1433
1434 nsh->raid_conf = conf;
1435 spin_lock_init(&nsh->lock);
Dan Williams417b8d42009-10-16 16:25:22 +11001436 #ifdef CONFIG_MULTICORE_RAID456
1437 init_waitqueue_head(&nsh->ops.wait_for_ops);
1438 #endif
NeilBrownad01c9e2006-03-27 01:18:07 -08001439
1440 list_add(&nsh->lru, &newstripes);
1441 }
1442 if (i) {
1443 /* didn't get enough, give up */
1444 while (!list_empty(&newstripes)) {
1445 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1446 list_del(&nsh->lru);
1447 kmem_cache_free(sc, nsh);
1448 }
1449 kmem_cache_destroy(sc);
1450 return -ENOMEM;
1451 }
1452 /* Step 2 - Must use GFP_NOIO now.
1453 * OK, we have enough stripes, start collecting inactive
1454 * stripes and copying them over
1455 */
1456 list_for_each_entry(nsh, &newstripes, lru) {
1457 spin_lock_irq(&conf->device_lock);
1458 wait_event_lock_irq(conf->wait_for_stripe,
1459 !list_empty(&conf->inactive_list),
1460 conf->device_lock,
NeilBrownb3b46be2006-03-27 01:18:16 -08001461 unplug_slaves(conf->mddev)
NeilBrownad01c9e2006-03-27 01:18:07 -08001462 );
1463 osh = get_free_stripe(conf);
1464 spin_unlock_irq(&conf->device_lock);
1465 atomic_set(&nsh->count, 1);
1466 for(i=0; i<conf->pool_size; i++)
1467 nsh->dev[i].page = osh->dev[i].page;
1468 for( ; i<newsize; i++)
1469 nsh->dev[i].page = NULL;
1470 kmem_cache_free(conf->slab_cache, osh);
1471 }
1472 kmem_cache_destroy(conf->slab_cache);
1473
1474 /* Step 3.
1475 * At this point, we are holding all the stripes so the array
1476 * is completely stalled, so now is a good time to resize
Dan Williamsd6f38f32009-07-14 11:50:52 -07001477 * conf->disks and the scribble region
NeilBrownad01c9e2006-03-27 01:18:07 -08001478 */
1479 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1480 if (ndisks) {
1481 for (i=0; i<conf->raid_disks; i++)
1482 ndisks[i] = conf->disks[i];
1483 kfree(conf->disks);
1484 conf->disks = ndisks;
1485 } else
1486 err = -ENOMEM;
1487
Dan Williamsd6f38f32009-07-14 11:50:52 -07001488 get_online_cpus();
1489 conf->scribble_len = scribble_len(newsize);
1490 for_each_present_cpu(cpu) {
1491 struct raid5_percpu *percpu;
1492 void *scribble;
1493
1494 percpu = per_cpu_ptr(conf->percpu, cpu);
1495 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1496
1497 if (scribble) {
1498 kfree(percpu->scribble);
1499 percpu->scribble = scribble;
1500 } else {
1501 err = -ENOMEM;
1502 break;
1503 }
1504 }
1505 put_online_cpus();
1506
NeilBrownad01c9e2006-03-27 01:18:07 -08001507 /* Step 4, return new stripes to service */
1508 while(!list_empty(&newstripes)) {
1509 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1510 list_del_init(&nsh->lru);
Dan Williamsd6f38f32009-07-14 11:50:52 -07001511
NeilBrownad01c9e2006-03-27 01:18:07 -08001512 for (i=conf->raid_disks; i < newsize; i++)
1513 if (nsh->dev[i].page == NULL) {
1514 struct page *p = alloc_page(GFP_NOIO);
1515 nsh->dev[i].page = p;
1516 if (!p)
1517 err = -ENOMEM;
1518 }
1519 release_stripe(nsh);
1520 }
1521 /* critical section pass, GFP_NOIO no longer needed */
1522
1523 conf->slab_cache = sc;
1524 conf->active_name = 1-conf->active_name;
1525 conf->pool_size = newsize;
1526 return err;
1527}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
NeilBrown3f294f42005-11-08 21:39:25 -08001529static int drop_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
1531 struct stripe_head *sh;
1532
NeilBrown3f294f42005-11-08 21:39:25 -08001533 spin_lock_irq(&conf->device_lock);
1534 sh = get_free_stripe(conf);
1535 spin_unlock_irq(&conf->device_lock);
1536 if (!sh)
1537 return 0;
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02001538 BUG_ON(atomic_read(&sh->count));
NeilBrowne4e11e32010-06-16 16:45:16 +10001539 shrink_buffers(sh);
NeilBrown3f294f42005-11-08 21:39:25 -08001540 kmem_cache_free(conf->slab_cache, sh);
1541 atomic_dec(&conf->active_stripes);
1542 return 1;
1543}
1544
1545static void shrink_stripes(raid5_conf_t *conf)
1546{
1547 while (drop_one_stripe(conf))
1548 ;
1549
NeilBrown29fc7e32006-02-03 03:03:41 -08001550 if (conf->slab_cache)
1551 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 conf->slab_cache = NULL;
1553}
1554
NeilBrown6712ecf2007-09-27 12:47:43 +02001555static void raid5_end_read_request(struct bio * bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556{
NeilBrown99c0fb52009-03-31 14:39:38 +11001557 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001559 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownd6950432006-07-10 04:44:20 -07001561 char b[BDEVNAME_SIZE];
1562 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
1565 for (i=0 ; i<disks; i++)
1566 if (bi == &sh->dev[i].req)
1567 break;
1568
Dan Williams45b42332007-07-09 11:56:43 -07001569 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1570 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 uptodate);
1572 if (i == disks) {
1573 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001574 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 }
1576
1577 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -08001579 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrownd6950432006-07-10 04:44:20 -07001580 rdev = conf->disks[i].rdev;
NeilBrown0c55e022010-05-03 14:09:02 +10001581 printk_rl(KERN_INFO "md/raid:%s: read error corrected"
Bernd Schubert6be9d492008-05-23 13:04:34 -07001582 " (%lu sectors at %llu on %s)\n",
1583 mdname(conf->mddev), STRIPE_SECTORS,
1584 (unsigned long long)(sh->sector
1585 + rdev->data_offset),
1586 bdevname(rdev->bdev, b));
NeilBrown4e5314b2005-11-08 21:39:22 -08001587 clear_bit(R5_ReadError, &sh->dev[i].flags);
1588 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1589 }
NeilBrownba22dcb2005-11-08 21:39:31 -08001590 if (atomic_read(&conf->disks[i].rdev->read_errors))
1591 atomic_set(&conf->disks[i].rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 } else {
NeilBrownd6950432006-07-10 04:44:20 -07001593 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
NeilBrownba22dcb2005-11-08 21:39:31 -08001594 int retry = 0;
NeilBrownd6950432006-07-10 04:44:20 -07001595 rdev = conf->disks[i].rdev;
1596
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001598 atomic_inc(&rdev->read_errors);
Gabriele A. Trombetti7b0bb532010-04-28 11:51:17 +10001599 if (conf->mddev->degraded >= conf->max_degraded)
Bernd Schubert6be9d492008-05-23 13:04:34 -07001600 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001601 "md/raid:%s: read error not correctable "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001602 "(sector %llu on %s).\n",
1603 mdname(conf->mddev),
1604 (unsigned long long)(sh->sector
1605 + rdev->data_offset),
1606 bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001607 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
NeilBrown4e5314b2005-11-08 21:39:22 -08001608 /* Oh, no!!! */
Bernd Schubert6be9d492008-05-23 13:04:34 -07001609 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001610 "md/raid:%s: read error NOT corrected!! "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001611 "(sector %llu on %s).\n",
1612 mdname(conf->mddev),
1613 (unsigned long long)(sh->sector
1614 + rdev->data_offset),
1615 bdn);
NeilBrownd6950432006-07-10 04:44:20 -07001616 else if (atomic_read(&rdev->read_errors)
NeilBrownba22dcb2005-11-08 21:39:31 -08001617 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -08001618 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001619 "md/raid:%s: Too many read errors, failing device %s.\n",
NeilBrownd6950432006-07-10 04:44:20 -07001620 mdname(conf->mddev), bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001621 else
1622 retry = 1;
1623 if (retry)
1624 set_bit(R5_ReadError, &sh->dev[i].flags);
1625 else {
NeilBrown4e5314b2005-11-08 21:39:22 -08001626 clear_bit(R5_ReadError, &sh->dev[i].flags);
1627 clear_bit(R5_ReWrite, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001628 md_error(conf->mddev, rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -08001629 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 }
1631 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1633 set_bit(STRIPE_HANDLE, &sh->state);
1634 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635}
1636
NeilBrownd710e132008-10-13 11:55:12 +11001637static void raid5_end_write_request(struct bio *bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
NeilBrown99c0fb52009-03-31 14:39:38 +11001639 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001641 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1643
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 for (i=0 ; i<disks; i++)
1645 if (bi == &sh->dev[i].req)
1646 break;
1647
Dan Williams45b42332007-07-09 11:56:43 -07001648 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1650 uptodate);
1651 if (i == disks) {
1652 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001653 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 }
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 if (!uptodate)
1657 md_error(conf->mddev, conf->disks[i].rdev);
1658
1659 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1660
1661 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1662 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrownc04be0a2006-10-03 01:15:53 -07001663 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664}
1665
1666
NeilBrown784052e2009-03-31 15:19:07 +11001667static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
NeilBrown784052e2009-03-31 15:19:07 +11001669static void raid5_build_block(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
1671 struct r5dev *dev = &sh->dev[i];
1672
1673 bio_init(&dev->req);
1674 dev->req.bi_io_vec = &dev->vec;
1675 dev->req.bi_vcnt++;
1676 dev->req.bi_max_vecs++;
1677 dev->vec.bv_page = dev->page;
1678 dev->vec.bv_len = STRIPE_SIZE;
1679 dev->vec.bv_offset = 0;
1680
1681 dev->req.bi_sector = sh->sector;
1682 dev->req.bi_private = sh;
1683
1684 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +11001685 dev->sector = compute_blocknr(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686}
1687
1688static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1689{
1690 char b[BDEVNAME_SIZE];
H Hartley Sweeten7b928132010-03-08 16:02:40 +11001691 raid5_conf_t *conf = mddev->private;
NeilBrown0c55e022010-05-03 14:09:02 +10001692 pr_debug("raid456: error called\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693
NeilBrownb2d444d2005-11-08 21:39:31 -08001694 if (!test_bit(Faulty, &rdev->flags)) {
NeilBrown850b2b42006-10-03 01:15:46 -07001695 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001696 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1697 unsigned long flags;
1698 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 mddev->degraded++;
NeilBrownc04be0a2006-10-03 01:15:53 -07001700 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 /*
1702 * if recovery was running, make sure it aborts.
1703 */
NeilBrowndfc70642008-05-23 13:04:39 -07001704 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 }
NeilBrownb2d444d2005-11-08 21:39:31 -08001706 set_bit(Faulty, &rdev->flags);
NeilBrownd710e132008-10-13 11:55:12 +11001707 printk(KERN_ALERT
NeilBrown0c55e022010-05-03 14:09:02 +10001708 "md/raid:%s: Disk failure on %s, disabling device.\n"
1709 KERN_ALERT
1710 "md/raid:%s: Operation continuing on %d devices.\n",
1711 mdname(mddev),
1712 bdevname(rdev->bdev, b),
1713 mdname(mddev),
1714 conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 }
NeilBrown16a53ec2006-06-26 00:27:38 -07001716}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718/*
1719 * Input: a 'big' sector number,
1720 * Output: index of the data and parity disk, and the sector # in them.
1721 */
NeilBrown112bf892009-03-31 14:39:38 +11001722static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11001723 int previous, int *dd_idx,
1724 struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725{
NeilBrown6e3b96e2010-04-23 07:08:28 +10001726 sector_t stripe, stripe2;
NeilBrown35f2a592010-04-20 14:13:34 +10001727 sector_t chunk_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 unsigned int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11001729 int pd_idx, qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001730 int ddf_layout = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 sector_t new_sector;
NeilBrowne183eae2009-03-31 15:20:22 +11001732 int algorithm = previous ? conf->prev_algo
1733 : conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001734 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1735 : conf->chunk_sectors;
NeilBrown112bf892009-03-31 14:39:38 +11001736 int raid_disks = previous ? conf->previous_raid_disks
1737 : conf->raid_disks;
1738 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
1740 /* First compute the information on this sector */
1741
1742 /*
1743 * Compute the chunk number and the sector offset inside the chunk
1744 */
1745 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1746 chunk_number = r_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
1748 /*
1749 * Compute the stripe number
1750 */
NeilBrown35f2a592010-04-20 14:13:34 +10001751 stripe = chunk_number;
1752 *dd_idx = sector_div(stripe, data_disks);
NeilBrown6e3b96e2010-04-23 07:08:28 +10001753 stripe2 = stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 /*
1755 * Select the parity disk based on the user selected algorithm.
1756 */
NeilBrown911d4ee2009-03-31 14:39:38 +11001757 pd_idx = qd_idx = ~0;
NeilBrown16a53ec2006-06-26 00:27:38 -07001758 switch(conf->level) {
1759 case 4:
NeilBrown911d4ee2009-03-31 14:39:38 +11001760 pd_idx = data_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001761 break;
1762 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001763 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001765 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001766 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 (*dd_idx)++;
1768 break;
1769 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001770 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001771 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 (*dd_idx)++;
1773 break;
1774 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001775 pd_idx = data_disks - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001776 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 break;
1778 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001779 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001780 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001782 case ALGORITHM_PARITY_0:
1783 pd_idx = 0;
1784 (*dd_idx)++;
1785 break;
1786 case ALGORITHM_PARITY_N:
1787 pd_idx = data_disks;
1788 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001790 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001791 }
1792 break;
1793 case 6:
1794
NeilBrowne183eae2009-03-31 15:20:22 +11001795 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001796 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001797 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001798 qd_idx = pd_idx + 1;
1799 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001800 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001801 qd_idx = 0;
1802 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001803 (*dd_idx) += 2; /* D D P Q D */
1804 break;
1805 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001806 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001807 qd_idx = pd_idx + 1;
1808 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001809 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001810 qd_idx = 0;
1811 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001812 (*dd_idx) += 2; /* D D P Q D */
1813 break;
1814 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001815 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001816 qd_idx = (pd_idx + 1) % raid_disks;
1817 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001818 break;
1819 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001820 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown911d4ee2009-03-31 14:39:38 +11001821 qd_idx = (pd_idx + 1) % raid_disks;
1822 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001823 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001824
1825 case ALGORITHM_PARITY_0:
1826 pd_idx = 0;
1827 qd_idx = 1;
1828 (*dd_idx) += 2;
1829 break;
1830 case ALGORITHM_PARITY_N:
1831 pd_idx = data_disks;
1832 qd_idx = data_disks + 1;
1833 break;
1834
1835 case ALGORITHM_ROTATING_ZERO_RESTART:
1836 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1837 * of blocks for computing Q is different.
1838 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001839 pd_idx = sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001840 qd_idx = pd_idx + 1;
1841 if (pd_idx == raid_disks-1) {
1842 (*dd_idx)++; /* Q D D D P */
1843 qd_idx = 0;
1844 } else if (*dd_idx >= pd_idx)
1845 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001846 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001847 break;
1848
1849 case ALGORITHM_ROTATING_N_RESTART:
1850 /* Same a left_asymmetric, by first stripe is
1851 * D D D P Q rather than
1852 * Q D D D P
1853 */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001854 stripe2 += 1;
1855 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001856 qd_idx = pd_idx + 1;
1857 if (pd_idx == raid_disks-1) {
1858 (*dd_idx)++; /* Q D D D P */
1859 qd_idx = 0;
1860 } else if (*dd_idx >= pd_idx)
1861 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001862 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001863 break;
1864
1865 case ALGORITHM_ROTATING_N_CONTINUE:
1866 /* Same as left_symmetric but Q is before P */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001867 pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks);
NeilBrown99c0fb52009-03-31 14:39:38 +11001868 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1869 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
NeilBrown67cc2b82009-03-31 14:39:38 +11001870 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001871 break;
1872
1873 case ALGORITHM_LEFT_ASYMMETRIC_6:
1874 /* RAID5 left_asymmetric, with Q on last device */
NeilBrown6e3b96e2010-04-23 07:08:28 +10001875 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001876 if (*dd_idx >= pd_idx)
1877 (*dd_idx)++;
1878 qd_idx = raid_disks - 1;
1879 break;
1880
1881 case ALGORITHM_RIGHT_ASYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001882 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001883 if (*dd_idx >= pd_idx)
1884 (*dd_idx)++;
1885 qd_idx = raid_disks - 1;
1886 break;
1887
1888 case ALGORITHM_LEFT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001889 pd_idx = data_disks - sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001890 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1891 qd_idx = raid_disks - 1;
1892 break;
1893
1894 case ALGORITHM_RIGHT_SYMMETRIC_6:
NeilBrown6e3b96e2010-04-23 07:08:28 +10001895 pd_idx = sector_div(stripe2, raid_disks-1);
NeilBrown99c0fb52009-03-31 14:39:38 +11001896 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1897 qd_idx = raid_disks - 1;
1898 break;
1899
1900 case ALGORITHM_PARITY_0_6:
1901 pd_idx = 0;
1902 (*dd_idx)++;
1903 qd_idx = raid_disks - 1;
1904 break;
1905
NeilBrown16a53ec2006-06-26 00:27:38 -07001906 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001907 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001908 }
1909 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 }
1911
NeilBrown911d4ee2009-03-31 14:39:38 +11001912 if (sh) {
1913 sh->pd_idx = pd_idx;
1914 sh->qd_idx = qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001915 sh->ddf_layout = ddf_layout;
NeilBrown911d4ee2009-03-31 14:39:38 +11001916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 /*
1918 * Finally, compute the new sector number
1919 */
1920 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1921 return new_sector;
1922}
1923
1924
NeilBrown784052e2009-03-31 15:19:07 +11001925static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926{
1927 raid5_conf_t *conf = sh->raid_conf;
NeilBrownb875e532006-12-10 02:20:49 -08001928 int raid_disks = sh->disks;
1929 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 sector_t new_sector = sh->sector, check;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001931 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1932 : conf->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11001933 int algorithm = previous ? conf->prev_algo
1934 : conf->algorithm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 sector_t stripe;
1936 int chunk_offset;
NeilBrown35f2a592010-04-20 14:13:34 +10001937 sector_t chunk_number;
1938 int dummy1, dd_idx = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 sector_t r_sector;
NeilBrown911d4ee2009-03-31 14:39:38 +11001940 struct stripe_head sh2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
NeilBrown16a53ec2006-06-26 00:27:38 -07001942
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1944 stripe = new_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
NeilBrown16a53ec2006-06-26 00:27:38 -07001946 if (i == sh->pd_idx)
1947 return 0;
1948 switch(conf->level) {
1949 case 4: break;
1950 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001951 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 case ALGORITHM_LEFT_ASYMMETRIC:
1953 case ALGORITHM_RIGHT_ASYMMETRIC:
1954 if (i > sh->pd_idx)
1955 i--;
1956 break;
1957 case ALGORITHM_LEFT_SYMMETRIC:
1958 case ALGORITHM_RIGHT_SYMMETRIC:
1959 if (i < sh->pd_idx)
1960 i += raid_disks;
1961 i -= (sh->pd_idx + 1);
1962 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001963 case ALGORITHM_PARITY_0:
1964 i -= 1;
1965 break;
1966 case ALGORITHM_PARITY_N:
1967 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001969 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001970 }
1971 break;
1972 case 6:
NeilBrownd0dabf72009-03-31 14:39:38 +11001973 if (i == sh->qd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001974 return 0; /* It is the Q disk */
NeilBrowne183eae2009-03-31 15:20:22 +11001975 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001976 case ALGORITHM_LEFT_ASYMMETRIC:
1977 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown99c0fb52009-03-31 14:39:38 +11001978 case ALGORITHM_ROTATING_ZERO_RESTART:
1979 case ALGORITHM_ROTATING_N_RESTART:
1980 if (sh->pd_idx == raid_disks-1)
1981 i--; /* Q D D D P */
NeilBrown16a53ec2006-06-26 00:27:38 -07001982 else if (i > sh->pd_idx)
1983 i -= 2; /* D D P Q D */
1984 break;
1985 case ALGORITHM_LEFT_SYMMETRIC:
1986 case ALGORITHM_RIGHT_SYMMETRIC:
1987 if (sh->pd_idx == raid_disks-1)
1988 i--; /* Q D D D P */
1989 else {
1990 /* D D P Q D */
1991 if (i < sh->pd_idx)
1992 i += raid_disks;
1993 i -= (sh->pd_idx + 2);
1994 }
1995 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001996 case ALGORITHM_PARITY_0:
1997 i -= 2;
1998 break;
1999 case ALGORITHM_PARITY_N:
2000 break;
2001 case ALGORITHM_ROTATING_N_CONTINUE:
NeilBrowne4424fe2009-10-16 16:27:34 +11002002 /* Like left_symmetric, but P is before Q */
NeilBrown99c0fb52009-03-31 14:39:38 +11002003 if (sh->pd_idx == 0)
2004 i--; /* P D D D Q */
NeilBrowne4424fe2009-10-16 16:27:34 +11002005 else {
2006 /* D D Q P D */
2007 if (i < sh->pd_idx)
2008 i += raid_disks;
2009 i -= (sh->pd_idx + 1);
2010 }
NeilBrown99c0fb52009-03-31 14:39:38 +11002011 break;
2012 case ALGORITHM_LEFT_ASYMMETRIC_6:
2013 case ALGORITHM_RIGHT_ASYMMETRIC_6:
2014 if (i > sh->pd_idx)
2015 i--;
2016 break;
2017 case ALGORITHM_LEFT_SYMMETRIC_6:
2018 case ALGORITHM_RIGHT_SYMMETRIC_6:
2019 if (i < sh->pd_idx)
2020 i += data_disks + 1;
2021 i -= (sh->pd_idx + 1);
2022 break;
2023 case ALGORITHM_PARITY_0_6:
2024 i -= 1;
2025 break;
NeilBrown16a53ec2006-06-26 00:27:38 -07002026 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11002027 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07002028 }
2029 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 }
2031
2032 chunk_number = stripe * data_disks + i;
NeilBrown35f2a592010-04-20 14:13:34 +10002033 r_sector = chunk_number * sectors_per_chunk + chunk_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
NeilBrown112bf892009-03-31 14:39:38 +11002035 check = raid5_compute_sector(conf, r_sector,
NeilBrown784052e2009-03-31 15:19:07 +11002036 previous, &dummy1, &sh2);
NeilBrown911d4ee2009-03-31 14:39:38 +11002037 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
2038 || sh2.qd_idx != sh->qd_idx) {
NeilBrown0c55e022010-05-03 14:09:02 +10002039 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
2040 mdname(conf->mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 return 0;
2042 }
2043 return r_sector;
2044}
2045
2046
Dan Williams600aa102008-06-28 08:32:05 +10002047static void
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002048schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
Dan Williams600aa102008-06-28 08:32:05 +10002049 int rcw, int expand)
Dan Williamse33129d2007-01-02 13:52:30 -07002050{
2051 int i, pd_idx = sh->pd_idx, disks = sh->disks;
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002052 raid5_conf_t *conf = sh->raid_conf;
2053 int level = conf->level;
NeilBrown16a53ec2006-06-26 00:27:38 -07002054
Dan Williamse33129d2007-01-02 13:52:30 -07002055 if (rcw) {
2056 /* if we are not expanding this is a proper write request, and
2057 * there will be bios with new data to be drained into the
2058 * stripe cache
2059 */
2060 if (!expand) {
Dan Williams600aa102008-06-28 08:32:05 +10002061 sh->reconstruct_state = reconstruct_state_drain_run;
2062 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
2063 } else
2064 sh->reconstruct_state = reconstruct_state_run;
Dan Williamse33129d2007-01-02 13:52:30 -07002065
Dan Williamsac6b53b2009-07-14 13:40:19 -07002066 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002067
2068 for (i = disks; i--; ) {
2069 struct r5dev *dev = &sh->dev[i];
2070
2071 if (dev->towrite) {
2072 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsd8ee0722008-06-28 08:32:06 +10002073 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002074 if (!expand)
2075 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002076 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002077 }
2078 }
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002079 if (s->locked + conf->max_degraded == disks)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002080 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002081 atomic_inc(&conf->pending_full_writes);
Dan Williamse33129d2007-01-02 13:52:30 -07002082 } else {
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002083 BUG_ON(level == 6);
Dan Williamse33129d2007-01-02 13:52:30 -07002084 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2085 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2086
Dan Williamsd8ee0722008-06-28 08:32:06 +10002087 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
Dan Williams600aa102008-06-28 08:32:05 +10002088 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2089 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
Dan Williamsac6b53b2009-07-14 13:40:19 -07002090 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002091
2092 for (i = disks; i--; ) {
2093 struct r5dev *dev = &sh->dev[i];
2094 if (i == pd_idx)
2095 continue;
2096
Dan Williamse33129d2007-01-02 13:52:30 -07002097 if (dev->towrite &&
2098 (test_bit(R5_UPTODATE, &dev->flags) ||
Dan Williamsd8ee0722008-06-28 08:32:06 +10002099 test_bit(R5_Wantcompute, &dev->flags))) {
2100 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002101 set_bit(R5_LOCKED, &dev->flags);
2102 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002103 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002104 }
2105 }
2106 }
2107
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002108 /* keep the parity disk(s) locked while asynchronous operations
Dan Williamse33129d2007-01-02 13:52:30 -07002109 * are in flight
2110 */
2111 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2112 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
Dan Williams600aa102008-06-28 08:32:05 +10002113 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002114
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002115 if (level == 6) {
2116 int qd_idx = sh->qd_idx;
2117 struct r5dev *dev = &sh->dev[qd_idx];
2118
2119 set_bit(R5_LOCKED, &dev->flags);
2120 clear_bit(R5_UPTODATE, &dev->flags);
2121 s->locked++;
2122 }
2123
Dan Williams600aa102008-06-28 08:32:05 +10002124 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07002125 __func__, (unsigned long long)sh->sector,
Dan Williams600aa102008-06-28 08:32:05 +10002126 s->locked, s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002127}
NeilBrown16a53ec2006-06-26 00:27:38 -07002128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129/*
2130 * Each stripe/dev can have one or more bion attached.
NeilBrown16a53ec2006-06-26 00:27:38 -07002131 * toread/towrite point to the first in a chain.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 * The bi_next chain must be in order.
2133 */
2134static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2135{
2136 struct bio **bip;
2137 raid5_conf_t *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -07002138 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
Dan Williams45b42332007-07-09 11:56:43 -07002140 pr_debug("adding bh b#%llu to stripe s#%llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 (unsigned long long)bi->bi_sector,
2142 (unsigned long long)sh->sector);
2143
2144
2145 spin_lock(&sh->lock);
2146 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07002147 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 bip = &sh->dev[dd_idx].towrite;
NeilBrown72626682005-09-09 16:23:54 -07002149 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2150 firstwrite = 1;
2151 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 bip = &sh->dev[dd_idx].toread;
2153 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2154 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2155 goto overlap;
2156 bip = & (*bip)->bi_next;
2157 }
2158 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2159 goto overlap;
2160
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002161 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 if (*bip)
2163 bi->bi_next = *bip;
2164 *bip = bi;
Jens Axboe960e7392008-08-15 10:41:18 +02002165 bi->bi_phys_segments++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 spin_unlock_irq(&conf->device_lock);
2167 spin_unlock(&sh->lock);
2168
Dan Williams45b42332007-07-09 11:56:43 -07002169 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 (unsigned long long)bi->bi_sector,
2171 (unsigned long long)sh->sector, dd_idx);
2172
NeilBrown72626682005-09-09 16:23:54 -07002173 if (conf->mddev->bitmap && firstwrite) {
NeilBrown72626682005-09-09 16:23:54 -07002174 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2175 STRIPE_SECTORS, 0);
NeilBrownae3c20c2006-07-10 04:44:17 -07002176 sh->bm_seq = conf->seq_flush+1;
NeilBrown72626682005-09-09 16:23:54 -07002177 set_bit(STRIPE_BIT_DELAY, &sh->state);
2178 }
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 if (forwrite) {
2181 /* check if page is covered */
2182 sector_t sector = sh->dev[dd_idx].sector;
2183 for (bi=sh->dev[dd_idx].towrite;
2184 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2185 bi && bi->bi_sector <= sector;
2186 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2187 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2188 sector = bi->bi_sector + (bi->bi_size>>9);
2189 }
2190 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2191 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2192 }
2193 return 1;
2194
2195 overlap:
2196 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2197 spin_unlock_irq(&conf->device_lock);
2198 spin_unlock(&sh->lock);
2199 return 0;
2200}
2201
NeilBrown29269552006-03-27 01:18:10 -08002202static void end_reshape(raid5_conf_t *conf);
2203
NeilBrown911d4ee2009-03-31 14:39:38 +11002204static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
2205 struct stripe_head *sh)
NeilBrownccfcc3c2006-03-27 01:18:09 -08002206{
NeilBrown784052e2009-03-31 15:19:07 +11002207 int sectors_per_chunk =
Andre Noll09c9e5f2009-06-18 08:45:55 +10002208 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
NeilBrown911d4ee2009-03-31 14:39:38 +11002209 int dd_idx;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002210 int chunk_offset = sector_div(stripe, sectors_per_chunk);
NeilBrown112bf892009-03-31 14:39:38 +11002211 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002212
NeilBrown112bf892009-03-31 14:39:38 +11002213 raid5_compute_sector(conf,
2214 stripe * (disks - conf->max_degraded)
NeilBrownb875e532006-12-10 02:20:49 -08002215 *sectors_per_chunk + chunk_offset,
NeilBrown112bf892009-03-31 14:39:38 +11002216 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002217 &dd_idx, sh);
NeilBrownccfcc3c2006-03-27 01:18:09 -08002218}
2219
Dan Williamsa4456852007-07-09 11:56:43 -07002220static void
Dan Williams1fe797e2008-06-28 09:16:30 +10002221handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002222 struct stripe_head_state *s, int disks,
2223 struct bio **return_bi)
2224{
2225 int i;
2226 for (i = disks; i--; ) {
2227 struct bio *bi;
2228 int bitmap_end = 0;
2229
2230 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2231 mdk_rdev_t *rdev;
2232 rcu_read_lock();
2233 rdev = rcu_dereference(conf->disks[i].rdev);
2234 if (rdev && test_bit(In_sync, &rdev->flags))
2235 /* multiple read failures in one stripe */
2236 md_error(conf->mddev, rdev);
2237 rcu_read_unlock();
2238 }
2239 spin_lock_irq(&conf->device_lock);
2240 /* fail all writes first */
2241 bi = sh->dev[i].towrite;
2242 sh->dev[i].towrite = NULL;
2243 if (bi) {
2244 s->to_write--;
2245 bitmap_end = 1;
2246 }
2247
2248 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2249 wake_up(&conf->wait_for_overlap);
2250
2251 while (bi && bi->bi_sector <
2252 sh->dev[i].sector + STRIPE_SECTORS) {
2253 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2254 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002255 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002256 md_write_end(conf->mddev);
2257 bi->bi_next = *return_bi;
2258 *return_bi = bi;
2259 }
2260 bi = nextbi;
2261 }
2262 /* and fail all 'written' */
2263 bi = sh->dev[i].written;
2264 sh->dev[i].written = NULL;
2265 if (bi) bitmap_end = 1;
2266 while (bi && bi->bi_sector <
2267 sh->dev[i].sector + STRIPE_SECTORS) {
2268 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2269 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002270 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002271 md_write_end(conf->mddev);
2272 bi->bi_next = *return_bi;
2273 *return_bi = bi;
2274 }
2275 bi = bi2;
2276 }
2277
Dan Williamsb5e98d62007-01-02 13:52:31 -07002278 /* fail any reads if this device is non-operational and
2279 * the data has not reached the cache yet.
2280 */
2281 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2282 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2283 test_bit(R5_ReadError, &sh->dev[i].flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002284 bi = sh->dev[i].toread;
2285 sh->dev[i].toread = NULL;
2286 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2287 wake_up(&conf->wait_for_overlap);
2288 if (bi) s->to_read--;
2289 while (bi && bi->bi_sector <
2290 sh->dev[i].sector + STRIPE_SECTORS) {
2291 struct bio *nextbi =
2292 r5_next_bio(bi, sh->dev[i].sector);
2293 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002294 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002295 bi->bi_next = *return_bi;
2296 *return_bi = bi;
2297 }
2298 bi = nextbi;
2299 }
2300 }
2301 spin_unlock_irq(&conf->device_lock);
2302 if (bitmap_end)
2303 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2304 STRIPE_SECTORS, 0, 0);
2305 }
2306
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002307 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2308 if (atomic_dec_and_test(&conf->pending_full_writes))
2309 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002310}
2311
Dan Williams1fe797e2008-06-28 09:16:30 +10002312/* fetch_block5 - checks the given member device to see if its data needs
2313 * to be read or computed to satisfy a request.
2314 *
2315 * Returns 1 when no more member devices need to be checked, otherwise returns
2316 * 0 to tell the loop in handle_stripe_fill5 to continue
Dan Williamsf38e1212007-01-02 13:52:30 -07002317 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002318static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
2319 int disk_idx, int disks)
Dan Williamsf38e1212007-01-02 13:52:30 -07002320{
2321 struct r5dev *dev = &sh->dev[disk_idx];
2322 struct r5dev *failed_dev = &sh->dev[s->failed_num];
2323
Dan Williamsf38e1212007-01-02 13:52:30 -07002324 /* is the data in this block needed, and can we get it? */
2325 if (!test_bit(R5_LOCKED, &dev->flags) &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002326 !test_bit(R5_UPTODATE, &dev->flags) &&
2327 (dev->toread ||
2328 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2329 s->syncing || s->expanding ||
2330 (s->failed &&
2331 (failed_dev->toread ||
2332 (failed_dev->towrite &&
2333 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002334 /* We would like to get this block, possibly by computing it,
2335 * otherwise read it if the backing disk is insync
Dan Williamsf38e1212007-01-02 13:52:30 -07002336 */
2337 if ((s->uptodate == disks - 1) &&
Dan Williamsecc65c92008-06-28 08:31:57 +10002338 (s->failed && disk_idx == s->failed_num)) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002339 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2340 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
Dan Williamsf38e1212007-01-02 13:52:30 -07002341 set_bit(R5_Wantcompute, &dev->flags);
2342 sh->ops.target = disk_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002343 sh->ops.target2 = -1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002344 s->req_compute = 1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002345 /* Careful: from this point on 'uptodate' is in the eye
Dan Williamsac6b53b2009-07-14 13:40:19 -07002346 * of raid_run_ops which services 'compute' operations
Dan Williamsf38e1212007-01-02 13:52:30 -07002347 * before writes. R5_Wantcompute flags a block that will
2348 * be R5_UPTODATE by the time it is needed for a
2349 * subsequent operation.
2350 */
2351 s->uptodate++;
Dan Williams1fe797e2008-06-28 09:16:30 +10002352 return 1; /* uptodate + compute == disks */
Dan Williams7a1fc532008-07-10 04:54:57 -07002353 } else if (test_bit(R5_Insync, &dev->flags)) {
Dan Williamsf38e1212007-01-02 13:52:30 -07002354 set_bit(R5_LOCKED, &dev->flags);
2355 set_bit(R5_Wantread, &dev->flags);
Dan Williamsf38e1212007-01-02 13:52:30 -07002356 s->locked++;
2357 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
2358 s->syncing);
2359 }
2360 }
2361
Dan Williams1fe797e2008-06-28 09:16:30 +10002362 return 0;
Dan Williamsf38e1212007-01-02 13:52:30 -07002363}
2364
Dan Williams1fe797e2008-06-28 09:16:30 +10002365/**
2366 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
2367 */
2368static void handle_stripe_fill5(struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002369 struct stripe_head_state *s, int disks)
2370{
2371 int i;
Dan Williamsf38e1212007-01-02 13:52:30 -07002372
Dan Williamsf38e1212007-01-02 13:52:30 -07002373 /* look for blocks to read/compute, skip this if a compute
2374 * is already in flight, or if the stripe contents are in the
2375 * midst of changing due to a write
2376 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002377 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002378 !sh->reconstruct_state)
Dan Williamsf38e1212007-01-02 13:52:30 -07002379 for (i = disks; i--; )
Dan Williams1fe797e2008-06-28 09:16:30 +10002380 if (fetch_block5(sh, s, i, disks))
Dan Williamsf38e1212007-01-02 13:52:30 -07002381 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002382 set_bit(STRIPE_HANDLE, &sh->state);
2383}
2384
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002385/* fetch_block6 - checks the given member device to see if its data needs
2386 * to be read or computed to satisfy a request.
2387 *
2388 * Returns 1 when no more member devices need to be checked, otherwise returns
2389 * 0 to tell the loop in handle_stripe_fill6 to continue
2390 */
2391static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s,
2392 struct r6_state *r6s, int disk_idx, int disks)
2393{
2394 struct r5dev *dev = &sh->dev[disk_idx];
2395 struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]],
2396 &sh->dev[r6s->failed_num[1]] };
2397
2398 if (!test_bit(R5_LOCKED, &dev->flags) &&
2399 !test_bit(R5_UPTODATE, &dev->flags) &&
2400 (dev->toread ||
2401 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2402 s->syncing || s->expanding ||
2403 (s->failed >= 1 &&
2404 (fdev[0]->toread || s->to_write)) ||
2405 (s->failed >= 2 &&
2406 (fdev[1]->toread || s->to_write)))) {
2407 /* we would like to get this block, possibly by computing it,
2408 * otherwise read it if the backing disk is insync
2409 */
2410 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2411 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2412 if ((s->uptodate == disks - 1) &&
2413 (s->failed && (disk_idx == r6s->failed_num[0] ||
2414 disk_idx == r6s->failed_num[1]))) {
2415 /* have disk failed, and we're requested to fetch it;
2416 * do compute it
2417 */
2418 pr_debug("Computing stripe %llu block %d\n",
2419 (unsigned long long)sh->sector, disk_idx);
2420 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2421 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2422 set_bit(R5_Wantcompute, &dev->flags);
2423 sh->ops.target = disk_idx;
2424 sh->ops.target2 = -1; /* no 2nd target */
2425 s->req_compute = 1;
2426 s->uptodate++;
2427 return 1;
2428 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2429 /* Computing 2-failure is *very* expensive; only
2430 * do it if failed >= 2
2431 */
2432 int other;
2433 for (other = disks; other--; ) {
2434 if (other == disk_idx)
2435 continue;
2436 if (!test_bit(R5_UPTODATE,
2437 &sh->dev[other].flags))
2438 break;
2439 }
2440 BUG_ON(other < 0);
2441 pr_debug("Computing stripe %llu blocks %d,%d\n",
2442 (unsigned long long)sh->sector,
2443 disk_idx, other);
2444 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2445 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2446 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2447 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2448 sh->ops.target = disk_idx;
2449 sh->ops.target2 = other;
2450 s->uptodate += 2;
2451 s->req_compute = 1;
2452 return 1;
2453 } else if (test_bit(R5_Insync, &dev->flags)) {
2454 set_bit(R5_LOCKED, &dev->flags);
2455 set_bit(R5_Wantread, &dev->flags);
2456 s->locked++;
2457 pr_debug("Reading block %d (sync=%d)\n",
2458 disk_idx, s->syncing);
2459 }
2460 }
2461
2462 return 0;
2463}
2464
2465/**
2466 * handle_stripe_fill6 - read or compute data to satisfy pending requests.
2467 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002468static void handle_stripe_fill6(struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002469 struct stripe_head_state *s, struct r6_state *r6s,
2470 int disks)
2471{
2472 int i;
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002473
2474 /* look for blocks to read/compute, skip this if a compute
2475 * is already in flight, or if the stripe contents are in the
2476 * midst of changing due to a write
2477 */
2478 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2479 !sh->reconstruct_state)
2480 for (i = disks; i--; )
2481 if (fetch_block6(sh, s, r6s, i, disks))
2482 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002483 set_bit(STRIPE_HANDLE, &sh->state);
2484}
2485
2486
Dan Williams1fe797e2008-06-28 09:16:30 +10002487/* handle_stripe_clean_event
Dan Williamsa4456852007-07-09 11:56:43 -07002488 * any written block on an uptodate or failed drive can be returned.
2489 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2490 * never LOCKED, so we don't need to test 'failed' directly.
2491 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002492static void handle_stripe_clean_event(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002493 struct stripe_head *sh, int disks, struct bio **return_bi)
2494{
2495 int i;
2496 struct r5dev *dev;
2497
2498 for (i = disks; i--; )
2499 if (sh->dev[i].written) {
2500 dev = &sh->dev[i];
2501 if (!test_bit(R5_LOCKED, &dev->flags) &&
2502 test_bit(R5_UPTODATE, &dev->flags)) {
2503 /* We can return any write requests */
2504 struct bio *wbi, *wbi2;
2505 int bitmap_end = 0;
Dan Williams45b42332007-07-09 11:56:43 -07002506 pr_debug("Return write for disc %d\n", i);
Dan Williamsa4456852007-07-09 11:56:43 -07002507 spin_lock_irq(&conf->device_lock);
2508 wbi = dev->written;
2509 dev->written = NULL;
2510 while (wbi && wbi->bi_sector <
2511 dev->sector + STRIPE_SECTORS) {
2512 wbi2 = r5_next_bio(wbi, dev->sector);
Jens Axboe960e7392008-08-15 10:41:18 +02002513 if (!raid5_dec_bi_phys_segments(wbi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002514 md_write_end(conf->mddev);
2515 wbi->bi_next = *return_bi;
2516 *return_bi = wbi;
2517 }
2518 wbi = wbi2;
2519 }
2520 if (dev->towrite == NULL)
2521 bitmap_end = 1;
2522 spin_unlock_irq(&conf->device_lock);
2523 if (bitmap_end)
2524 bitmap_endwrite(conf->mddev->bitmap,
2525 sh->sector,
2526 STRIPE_SECTORS,
2527 !test_bit(STRIPE_DEGRADED, &sh->state),
2528 0);
2529 }
2530 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002531
2532 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2533 if (atomic_dec_and_test(&conf->pending_full_writes))
2534 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002535}
2536
Dan Williams1fe797e2008-06-28 09:16:30 +10002537static void handle_stripe_dirtying5(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002538 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2539{
2540 int rmw = 0, rcw = 0, i;
2541 for (i = disks; i--; ) {
2542 /* would I have to read this buffer for read_modify_write */
2543 struct r5dev *dev = &sh->dev[i];
2544 if ((dev->towrite || i == sh->pd_idx) &&
2545 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002546 !(test_bit(R5_UPTODATE, &dev->flags) ||
2547 test_bit(R5_Wantcompute, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002548 if (test_bit(R5_Insync, &dev->flags))
2549 rmw++;
2550 else
2551 rmw += 2*disks; /* cannot read it */
2552 }
2553 /* Would I have to read this buffer for reconstruct_write */
2554 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2555 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002556 !(test_bit(R5_UPTODATE, &dev->flags) ||
2557 test_bit(R5_Wantcompute, &dev->flags))) {
2558 if (test_bit(R5_Insync, &dev->flags)) rcw++;
Dan Williamsa4456852007-07-09 11:56:43 -07002559 else
2560 rcw += 2*disks;
2561 }
2562 }
Dan Williams45b42332007-07-09 11:56:43 -07002563 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07002564 (unsigned long long)sh->sector, rmw, rcw);
2565 set_bit(STRIPE_HANDLE, &sh->state);
2566 if (rmw < rcw && rmw > 0)
2567 /* prefer read-modify-write, but need to get some data */
2568 for (i = disks; i--; ) {
2569 struct r5dev *dev = &sh->dev[i];
2570 if ((dev->towrite || i == sh->pd_idx) &&
2571 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002572 !(test_bit(R5_UPTODATE, &dev->flags) ||
2573 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002574 test_bit(R5_Insync, &dev->flags)) {
2575 if (
2576 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002577 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002578 "%d for r-m-w\n", i);
2579 set_bit(R5_LOCKED, &dev->flags);
2580 set_bit(R5_Wantread, &dev->flags);
2581 s->locked++;
2582 } else {
2583 set_bit(STRIPE_DELAYED, &sh->state);
2584 set_bit(STRIPE_HANDLE, &sh->state);
2585 }
2586 }
2587 }
2588 if (rcw <= rmw && rcw > 0)
2589 /* want reconstruct write, but need to get some data */
2590 for (i = disks; i--; ) {
2591 struct r5dev *dev = &sh->dev[i];
2592 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2593 i != sh->pd_idx &&
2594 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002595 !(test_bit(R5_UPTODATE, &dev->flags) ||
2596 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002597 test_bit(R5_Insync, &dev->flags)) {
2598 if (
2599 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002600 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002601 "%d for Reconstruct\n", i);
2602 set_bit(R5_LOCKED, &dev->flags);
2603 set_bit(R5_Wantread, &dev->flags);
2604 s->locked++;
2605 } else {
2606 set_bit(STRIPE_DELAYED, &sh->state);
2607 set_bit(STRIPE_HANDLE, &sh->state);
2608 }
2609 }
2610 }
2611 /* now if nothing is locked, and if we have enough data,
2612 * we can start a write request
2613 */
Dan Williamsf38e1212007-01-02 13:52:30 -07002614 /* since handle_stripe can be called at any time we need to handle the
2615 * case where a compute block operation has been submitted and then a
Dan Williamsac6b53b2009-07-14 13:40:19 -07002616 * subsequent call wants to start a write request. raid_run_ops only
2617 * handles the case where compute block and reconstruct are requested
Dan Williamsf38e1212007-01-02 13:52:30 -07002618 * simultaneously. If this is not the case then new writes need to be
2619 * held off until the compute completes.
2620 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002621 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2622 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2623 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002624 schedule_reconstruction(sh, s, rcw == 0, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002625}
2626
Dan Williams1fe797e2008-06-28 09:16:30 +10002627static void handle_stripe_dirtying6(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002628 struct stripe_head *sh, struct stripe_head_state *s,
2629 struct r6_state *r6s, int disks)
2630{
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002631 int rcw = 0, pd_idx = sh->pd_idx, i;
NeilBrown34e04e82009-03-31 15:10:16 +11002632 int qd_idx = sh->qd_idx;
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002633
2634 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002635 for (i = disks; i--; ) {
2636 struct r5dev *dev = &sh->dev[i];
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002637 /* check if we haven't enough data */
2638 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2639 i != pd_idx && i != qd_idx &&
2640 !test_bit(R5_LOCKED, &dev->flags) &&
2641 !(test_bit(R5_UPTODATE, &dev->flags) ||
2642 test_bit(R5_Wantcompute, &dev->flags))) {
2643 rcw++;
2644 if (!test_bit(R5_Insync, &dev->flags))
2645 continue; /* it's a failed drive */
2646
2647 if (
2648 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2649 pr_debug("Read_old stripe %llu "
2650 "block %d for Reconstruct\n",
2651 (unsigned long long)sh->sector, i);
2652 set_bit(R5_LOCKED, &dev->flags);
2653 set_bit(R5_Wantread, &dev->flags);
2654 s->locked++;
2655 } else {
2656 pr_debug("Request delayed stripe %llu "
2657 "block %d for Reconstruct\n",
2658 (unsigned long long)sh->sector, i);
2659 set_bit(STRIPE_DELAYED, &sh->state);
2660 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002661 }
2662 }
2663 }
Dan Williamsa4456852007-07-09 11:56:43 -07002664 /* now if nothing is locked, and if we have enough data, we can start a
2665 * write request
2666 */
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002667 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2668 s->locked == 0 && rcw == 0 &&
Dan Williamsa4456852007-07-09 11:56:43 -07002669 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002670 schedule_reconstruction(sh, s, 1, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002671 }
2672}
2673
2674static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2675 struct stripe_head_state *s, int disks)
2676{
Dan Williamsecc65c92008-06-28 08:31:57 +10002677 struct r5dev *dev = NULL;
Dan Williamse89f8962007-01-02 13:52:31 -07002678
Dan Williamsbd2ab672008-04-10 21:29:27 -07002679 set_bit(STRIPE_HANDLE, &sh->state);
2680
Dan Williamsecc65c92008-06-28 08:31:57 +10002681 switch (sh->check_state) {
2682 case check_state_idle:
2683 /* start a new check operation if there are no failures */
Dan Williamsbd2ab672008-04-10 21:29:27 -07002684 if (s->failed == 0) {
Dan Williamsbd2ab672008-04-10 21:29:27 -07002685 BUG_ON(s->uptodate != disks);
Dan Williamsecc65c92008-06-28 08:31:57 +10002686 sh->check_state = check_state_run;
2687 set_bit(STRIPE_OP_CHECK, &s->ops_request);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002688 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002689 s->uptodate--;
Dan Williamsecc65c92008-06-28 08:31:57 +10002690 break;
Dan Williamsbd2ab672008-04-10 21:29:27 -07002691 }
Dan Williamsa4456852007-07-09 11:56:43 -07002692 dev = &sh->dev[s->failed_num];
Dan Williamsecc65c92008-06-28 08:31:57 +10002693 /* fall through */
2694 case check_state_compute_result:
2695 sh->check_state = check_state_idle;
2696 if (!dev)
2697 dev = &sh->dev[sh->pd_idx];
2698
2699 /* check that a write has not made the stripe insync */
2700 if (test_bit(STRIPE_INSYNC, &sh->state))
2701 break;
2702
2703 /* either failed parity check, or recovery is happening */
Dan Williamsa4456852007-07-09 11:56:43 -07002704 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2705 BUG_ON(s->uptodate != disks);
2706
2707 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsecc65c92008-06-28 08:31:57 +10002708 s->locked++;
Dan Williamsa4456852007-07-09 11:56:43 -07002709 set_bit(R5_Wantwrite, &dev->flags);
Dan Williams830ea012007-01-02 13:52:31 -07002710
Dan Williamsa4456852007-07-09 11:56:43 -07002711 clear_bit(STRIPE_DEGRADED, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002712 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002713 break;
2714 case check_state_run:
2715 break; /* we will be called again upon completion */
2716 case check_state_check_result:
2717 sh->check_state = check_state_idle;
2718
2719 /* if a failure occurred during the check operation, leave
2720 * STRIPE_INSYNC not set and let the stripe be handled again
2721 */
2722 if (s->failed)
2723 break;
2724
2725 /* handle a successful check operation, if parity is correct
2726 * we are done. Otherwise update the mismatch count and repair
2727 * parity if !MD_RECOVERY_CHECK
2728 */
Dan Williamsad283ea2009-08-29 19:09:26 -07002729 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
Dan Williamsecc65c92008-06-28 08:31:57 +10002730 /* parity is correct (on disc,
2731 * not in buffer any more)
2732 */
2733 set_bit(STRIPE_INSYNC, &sh->state);
2734 else {
2735 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2736 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2737 /* don't try to repair!! */
2738 set_bit(STRIPE_INSYNC, &sh->state);
2739 else {
2740 sh->check_state = check_state_compute_run;
Dan Williams976ea8d2008-06-28 08:32:03 +10002741 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002742 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2743 set_bit(R5_Wantcompute,
2744 &sh->dev[sh->pd_idx].flags);
2745 sh->ops.target = sh->pd_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002746 sh->ops.target2 = -1;
Dan Williamsecc65c92008-06-28 08:31:57 +10002747 s->uptodate++;
2748 }
2749 }
2750 break;
2751 case check_state_compute_run:
2752 break;
2753 default:
2754 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2755 __func__, sh->check_state,
2756 (unsigned long long) sh->sector);
2757 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002758 }
2759}
2760
2761
2762static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williams36d1c642009-07-14 11:48:22 -07002763 struct stripe_head_state *s,
2764 struct r6_state *r6s, int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07002765{
Dan Williamsa4456852007-07-09 11:56:43 -07002766 int pd_idx = sh->pd_idx;
NeilBrown34e04e82009-03-31 15:10:16 +11002767 int qd_idx = sh->qd_idx;
Dan Williamsd82dfee2009-07-14 13:40:57 -07002768 struct r5dev *dev;
Dan Williamsa4456852007-07-09 11:56:43 -07002769
2770 set_bit(STRIPE_HANDLE, &sh->state);
2771
2772 BUG_ON(s->failed > 2);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002773
Dan Williamsa4456852007-07-09 11:56:43 -07002774 /* Want to check and possibly repair P and Q.
2775 * However there could be one 'failed' device, in which
2776 * case we can only check one of them, possibly using the
2777 * other to generate missing data
2778 */
2779
Dan Williamsd82dfee2009-07-14 13:40:57 -07002780 switch (sh->check_state) {
2781 case check_state_idle:
2782 /* start a new check operation if there are < 2 failures */
Dan Williamsa4456852007-07-09 11:56:43 -07002783 if (s->failed == r6s->q_failed) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002784 /* The only possible failed device holds Q, so it
Dan Williamsa4456852007-07-09 11:56:43 -07002785 * makes sense to check P (If anything else were failed,
2786 * we would have used P to recreate it).
2787 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002788 sh->check_state = check_state_run;
Dan Williamsa4456852007-07-09 11:56:43 -07002789 }
2790 if (!r6s->q_failed && s->failed < 2) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002791 /* Q is not failed, and we didn't use it to generate
Dan Williamsa4456852007-07-09 11:56:43 -07002792 * anything, so it makes sense to check it
2793 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002794 if (sh->check_state == check_state_run)
2795 sh->check_state = check_state_run_pq;
2796 else
2797 sh->check_state = check_state_run_q;
Dan Williamsa4456852007-07-09 11:56:43 -07002798 }
Dan Williams36d1c642009-07-14 11:48:22 -07002799
Dan Williamsd82dfee2009-07-14 13:40:57 -07002800 /* discard potentially stale zero_sum_result */
2801 sh->ops.zero_sum_result = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07002802
Dan Williamsd82dfee2009-07-14 13:40:57 -07002803 if (sh->check_state == check_state_run) {
2804 /* async_xor_zero_sum destroys the contents of P */
2805 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2806 s->uptodate--;
Dan Williamsa4456852007-07-09 11:56:43 -07002807 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002808 if (sh->check_state >= check_state_run &&
2809 sh->check_state <= check_state_run_pq) {
2810 /* async_syndrome_zero_sum preserves P and Q, so
2811 * no need to mark them !uptodate here
2812 */
2813 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2814 break;
2815 }
Dan Williams36d1c642009-07-14 11:48:22 -07002816
Dan Williamsd82dfee2009-07-14 13:40:57 -07002817 /* we have 2-disk failure */
2818 BUG_ON(s->failed != 2);
2819 /* fall through */
2820 case check_state_compute_result:
2821 sh->check_state = check_state_idle;
Dan Williams36d1c642009-07-14 11:48:22 -07002822
Dan Williamsd82dfee2009-07-14 13:40:57 -07002823 /* check that a write has not made the stripe insync */
2824 if (test_bit(STRIPE_INSYNC, &sh->state))
2825 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002826
2827 /* now write out any block on a failed drive,
Dan Williamsd82dfee2009-07-14 13:40:57 -07002828 * or P or Q if they were recomputed
Dan Williamsa4456852007-07-09 11:56:43 -07002829 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002830 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
Dan Williamsa4456852007-07-09 11:56:43 -07002831 if (s->failed == 2) {
2832 dev = &sh->dev[r6s->failed_num[1]];
2833 s->locked++;
2834 set_bit(R5_LOCKED, &dev->flags);
2835 set_bit(R5_Wantwrite, &dev->flags);
2836 }
2837 if (s->failed >= 1) {
2838 dev = &sh->dev[r6s->failed_num[0]];
2839 s->locked++;
2840 set_bit(R5_LOCKED, &dev->flags);
2841 set_bit(R5_Wantwrite, &dev->flags);
2842 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002843 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002844 dev = &sh->dev[pd_idx];
2845 s->locked++;
2846 set_bit(R5_LOCKED, &dev->flags);
2847 set_bit(R5_Wantwrite, &dev->flags);
2848 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002849 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002850 dev = &sh->dev[qd_idx];
2851 s->locked++;
2852 set_bit(R5_LOCKED, &dev->flags);
2853 set_bit(R5_Wantwrite, &dev->flags);
2854 }
2855 clear_bit(STRIPE_DEGRADED, &sh->state);
2856
2857 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002858 break;
2859 case check_state_run:
2860 case check_state_run_q:
2861 case check_state_run_pq:
2862 break; /* we will be called again upon completion */
2863 case check_state_check_result:
2864 sh->check_state = check_state_idle;
2865
2866 /* handle a successful check operation, if parity is correct
2867 * we are done. Otherwise update the mismatch count and repair
2868 * parity if !MD_RECOVERY_CHECK
2869 */
2870 if (sh->ops.zero_sum_result == 0) {
2871 /* both parities are correct */
2872 if (!s->failed)
2873 set_bit(STRIPE_INSYNC, &sh->state);
2874 else {
2875 /* in contrast to the raid5 case we can validate
2876 * parity, but still have a failure to write
2877 * back
2878 */
2879 sh->check_state = check_state_compute_result;
2880 /* Returning at this point means that we may go
2881 * off and bring p and/or q uptodate again so
2882 * we make sure to check zero_sum_result again
2883 * to verify if p or q need writeback
2884 */
2885 }
2886 } else {
2887 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2888 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2889 /* don't try to repair!! */
2890 set_bit(STRIPE_INSYNC, &sh->state);
2891 else {
2892 int *target = &sh->ops.target;
2893
2894 sh->ops.target = -1;
2895 sh->ops.target2 = -1;
2896 sh->check_state = check_state_compute_run;
2897 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2898 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2899 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
2900 set_bit(R5_Wantcompute,
2901 &sh->dev[pd_idx].flags);
2902 *target = pd_idx;
2903 target = &sh->ops.target2;
2904 s->uptodate++;
2905 }
2906 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
2907 set_bit(R5_Wantcompute,
2908 &sh->dev[qd_idx].flags);
2909 *target = qd_idx;
2910 s->uptodate++;
2911 }
2912 }
2913 }
2914 break;
2915 case check_state_compute_run:
2916 break;
2917 default:
2918 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2919 __func__, sh->check_state,
2920 (unsigned long long) sh->sector);
2921 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002922 }
2923}
2924
2925static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
2926 struct r6_state *r6s)
2927{
2928 int i;
2929
2930 /* We have read all the blocks in this stripe and now we need to
2931 * copy some of them into a target stripe for expand.
2932 */
Dan Williamsf0a50d32007-01-02 13:52:31 -07002933 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07002934 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2935 for (i = 0; i < sh->disks; i++)
NeilBrown34e04e82009-03-31 15:10:16 +11002936 if (i != sh->pd_idx && i != sh->qd_idx) {
NeilBrown911d4ee2009-03-31 14:39:38 +11002937 int dd_idx, j;
Dan Williamsa4456852007-07-09 11:56:43 -07002938 struct stripe_head *sh2;
Dan Williamsa08abd82009-06-03 11:43:59 -07002939 struct async_submit_ctl submit;
Dan Williamsa4456852007-07-09 11:56:43 -07002940
NeilBrown784052e2009-03-31 15:19:07 +11002941 sector_t bn = compute_blocknr(sh, i, 1);
NeilBrown911d4ee2009-03-31 14:39:38 +11002942 sector_t s = raid5_compute_sector(conf, bn, 0,
2943 &dd_idx, NULL);
NeilBrowna8c906c2009-06-09 14:39:59 +10002944 sh2 = get_active_stripe(conf, s, 0, 1, 1);
Dan Williamsa4456852007-07-09 11:56:43 -07002945 if (sh2 == NULL)
2946 /* so far only the early blocks of this stripe
2947 * have been requested. When later blocks
2948 * get requested, we will try again
2949 */
2950 continue;
2951 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2952 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2953 /* must have already done this block */
2954 release_stripe(sh2);
2955 continue;
2956 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07002957
2958 /* place all the copies on one channel */
Dan Williamsa08abd82009-06-03 11:43:59 -07002959 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002960 tx = async_memcpy(sh2->dev[dd_idx].page,
Dan Williams88ba2aa2009-04-09 16:16:18 -07002961 sh->dev[i].page, 0, 0, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07002962 &submit);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002963
Dan Williamsa4456852007-07-09 11:56:43 -07002964 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2965 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2966 for (j = 0; j < conf->raid_disks; j++)
2967 if (j != sh2->pd_idx &&
NeilBrownd0dabf72009-03-31 14:39:38 +11002968 (!r6s || j != sh2->qd_idx) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002969 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2970 break;
2971 if (j == conf->raid_disks) {
2972 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2973 set_bit(STRIPE_HANDLE, &sh2->state);
2974 }
2975 release_stripe(sh2);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002976
Dan Williamsa4456852007-07-09 11:56:43 -07002977 }
NeilBrowna2e08552007-09-11 15:23:36 -07002978 /* done submitting copies, wait for them to complete */
2979 if (tx) {
2980 async_tx_ack(tx);
2981 dma_wait_for_async_tx(tx);
2982 }
Dan Williamsa4456852007-07-09 11:56:43 -07002983}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984
Dan Williams6bfe0b42008-04-30 00:52:32 -07002985
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986/*
2987 * handle_stripe - do things to a stripe.
2988 *
2989 * We lock the stripe and then examine the state of various bits
2990 * to see what needs to be done.
2991 * Possible results:
2992 * return some read request which now have data
2993 * return some write requests which are safely on disc
2994 * schedule a read on some buffers
2995 * schedule a write of some buffers
2996 * return confirmation of parity correctness
2997 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 * buffers are taken off read_list or write_list, and bh_cache buffers
2999 * get BH_Lock set before the stripe lock is released.
3000 *
3001 */
Dan Williamsa4456852007-07-09 11:56:43 -07003002
NeilBrown14425772009-10-16 15:55:25 +11003003static void handle_stripe5(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004{
3005 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsa4456852007-07-09 11:56:43 -07003006 int disks = sh->disks, i;
3007 struct bio *return_bi = NULL;
3008 struct stripe_head_state s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 struct r5dev *dev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003010 mdk_rdev_t *blocked_rdev = NULL;
Dan Williamse0a115e2008-06-05 22:45:52 -07003011 int prexor;
NeilBrown729a1862009-12-14 12:49:50 +11003012 int dec_preread_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013
Dan Williamsa4456852007-07-09 11:56:43 -07003014 memset(&s, 0, sizeof(s));
Dan Williams600aa102008-06-28 08:32:05 +10003015 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
3016 "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state,
3017 atomic_read(&sh->count), sh->pd_idx, sh->check_state,
3018 sh->reconstruct_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
3020 spin_lock(&sh->lock);
3021 clear_bit(STRIPE_HANDLE, &sh->state);
3022 clear_bit(STRIPE_DELAYED, &sh->state);
3023
Dan Williamsa4456852007-07-09 11:56:43 -07003024 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3025 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3026 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
Dan Williams83de75c2008-06-28 08:31:58 +10003027
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 /* Now to look around and see what can be done */
NeilBrown9910f162006-01-06 00:20:24 -08003029 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 for (i=disks; i--; ) {
3031 mdk_rdev_t *rdev;
NeilBrowna9f326e2009-09-23 18:06:41 +10003032
3033 dev = &sh->dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034
Dan Williamsb5e98d62007-01-02 13:52:31 -07003035 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
3036 "written %p\n", i, dev->flags, dev->toread, dev->read,
3037 dev->towrite, dev->written);
3038
3039 /* maybe we can request a biofill operation
3040 *
3041 * new wantfill requests are only permitted while
Dan Williams83de75c2008-06-28 08:31:58 +10003042 * ops_complete_biofill is guaranteed to be inactive
Dan Williamsb5e98d62007-01-02 13:52:31 -07003043 */
3044 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
Dan Williams83de75c2008-06-28 08:31:58 +10003045 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
Dan Williamsb5e98d62007-01-02 13:52:31 -07003046 set_bit(R5_Wantfill, &dev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
3048 /* now count some things */
Dan Williamsa4456852007-07-09 11:56:43 -07003049 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3050 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
Dan Williamsf38e1212007-01-02 13:52:30 -07003051 if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052
Dan Williamsb5e98d62007-01-02 13:52:31 -07003053 if (test_bit(R5_Wantfill, &dev->flags))
3054 s.to_fill++;
3055 else if (dev->toread)
Dan Williamsa4456852007-07-09 11:56:43 -07003056 s.to_read++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 if (dev->towrite) {
Dan Williamsa4456852007-07-09 11:56:43 -07003058 s.to_write++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 if (!test_bit(R5_OVERWRITE, &dev->flags))
Dan Williamsa4456852007-07-09 11:56:43 -07003060 s.non_overwrite++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 }
Dan Williamsa4456852007-07-09 11:56:43 -07003062 if (dev->written)
3063 s.written++;
NeilBrown9910f162006-01-06 00:20:24 -08003064 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownac4090d2008-08-05 15:54:13 +10003065 if (blocked_rdev == NULL &&
3066 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07003067 blocked_rdev = rdev;
3068 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003069 }
NeilBrown415e72d2010-06-17 17:25:21 +10003070 clear_bit(R5_Insync, &dev->flags);
3071 if (!rdev)
3072 /* Not in-sync */;
3073 else if (test_bit(In_sync, &rdev->flags))
3074 set_bit(R5_Insync, &dev->flags);
3075 else {
3076 /* could be in-sync depending on recovery/reshape status */
3077 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3078 set_bit(R5_Insync, &dev->flags);
3079 }
3080 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown14f8d262006-01-06 00:20:14 -08003081 /* The ReadError flag will just be confusing now */
NeilBrown4e5314b2005-11-08 21:39:22 -08003082 clear_bit(R5_ReadError, &dev->flags);
3083 clear_bit(R5_ReWrite, &dev->flags);
3084 }
NeilBrown415e72d2010-06-17 17:25:21 +10003085 if (test_bit(R5_ReadError, &dev->flags))
3086 clear_bit(R5_Insync, &dev->flags);
3087 if (!test_bit(R5_Insync, &dev->flags)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003088 s.failed++;
3089 s.failed_num = i;
NeilBrown415e72d2010-06-17 17:25:21 +10003090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 }
NeilBrown9910f162006-01-06 00:20:24 -08003092 rcu_read_unlock();
Dan Williamsb5e98d62007-01-02 13:52:31 -07003093
Dan Williams6bfe0b42008-04-30 00:52:32 -07003094 if (unlikely(blocked_rdev)) {
NeilBrownac4090d2008-08-05 15:54:13 +10003095 if (s.syncing || s.expanding || s.expanded ||
3096 s.to_write || s.written) {
3097 set_bit(STRIPE_HANDLE, &sh->state);
3098 goto unlock;
3099 }
3100 /* There is nothing for the blocked_rdev to block */
3101 rdev_dec_pending(blocked_rdev, conf->mddev);
3102 blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003103 }
3104
Dan Williams83de75c2008-06-28 08:31:58 +10003105 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3106 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3107 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3108 }
Dan Williamsb5e98d62007-01-02 13:52:31 -07003109
Dan Williams45b42332007-07-09 11:56:43 -07003110 pr_debug("locked=%d uptodate=%d to_read=%d"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 " to_write=%d failed=%d failed_num=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003112 s.locked, s.uptodate, s.to_read, s.to_write,
3113 s.failed, s.failed_num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 /* check if the array has lost two devices and, if so, some requests might
3115 * need to be failed
3116 */
Dan Williamsa4456852007-07-09 11:56:43 -07003117 if (s.failed > 1 && s.to_read+s.to_write+s.written)
Dan Williams1fe797e2008-06-28 09:16:30 +10003118 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003119 if (s.failed > 1 && s.syncing) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3121 clear_bit(STRIPE_SYNCING, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003122 s.syncing = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 }
3124
3125 /* might be able to return some write requests if the parity block
3126 * is safe, or on a failed drive
3127 */
3128 dev = &sh->dev[sh->pd_idx];
Dan Williamsa4456852007-07-09 11:56:43 -07003129 if ( s.written &&
3130 ((test_bit(R5_Insync, &dev->flags) &&
3131 !test_bit(R5_LOCKED, &dev->flags) &&
3132 test_bit(R5_UPTODATE, &dev->flags)) ||
3133 (s.failed == 1 && s.failed_num == sh->pd_idx)))
Dan Williams1fe797e2008-06-28 09:16:30 +10003134 handle_stripe_clean_event(conf, sh, disks, &return_bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135
3136 /* Now we might consider reading some blocks, either to check/generate
3137 * parity, or to satisfy requests
3138 * or to load a block that is being partially written.
3139 */
Dan Williamsa4456852007-07-09 11:56:43 -07003140 if (s.to_read || s.non_overwrite ||
Dan Williams976ea8d2008-06-28 08:32:03 +10003141 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
Dan Williams1fe797e2008-06-28 09:16:30 +10003142 handle_stripe_fill5(sh, &s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143
Dan Williamse33129d2007-01-02 13:52:30 -07003144 /* Now we check to see if any write operations have recently
3145 * completed
3146 */
Dan Williamse0a115e2008-06-05 22:45:52 -07003147 prexor = 0;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003148 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
Dan Williamse0a115e2008-06-05 22:45:52 -07003149 prexor = 1;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003150 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3151 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
Dan Williams600aa102008-06-28 08:32:05 +10003152 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamse33129d2007-01-02 13:52:30 -07003153
3154 /* All the 'written' buffers and the parity block are ready to
3155 * be written back to disk
3156 */
3157 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3158 for (i = disks; i--; ) {
3159 dev = &sh->dev[i];
3160 if (test_bit(R5_LOCKED, &dev->flags) &&
3161 (i == sh->pd_idx || dev->written)) {
3162 pr_debug("Writing block %d\n", i);
3163 set_bit(R5_Wantwrite, &dev->flags);
Dan Williamse0a115e2008-06-05 22:45:52 -07003164 if (prexor)
3165 continue;
Dan Williamse33129d2007-01-02 13:52:30 -07003166 if (!test_bit(R5_Insync, &dev->flags) ||
3167 (i == sh->pd_idx && s.failed == 0))
3168 set_bit(STRIPE_INSYNC, &sh->state);
3169 }
3170 }
NeilBrown729a1862009-12-14 12:49:50 +11003171 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3172 dec_preread_active = 1;
Dan Williamse33129d2007-01-02 13:52:30 -07003173 }
3174
3175 /* Now to consider new write requests and what else, if anything
3176 * should be read. We do not handle new writes when:
3177 * 1/ A 'write' operation (copy+xor) is already in flight.
3178 * 2/ A 'check' operation is in flight, as it may clobber the parity
3179 * block.
3180 */
Dan Williams600aa102008-06-28 08:32:05 +10003181 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
Dan Williams1fe797e2008-06-28 09:16:30 +10003182 handle_stripe_dirtying5(conf, sh, &s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183
3184 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamse89f8962007-01-02 13:52:31 -07003185 * Any reads will already have been scheduled, so we just see if enough
3186 * data is available. The parity check is held off while parity
3187 * dependent operations are in flight.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 */
Dan Williamsecc65c92008-06-28 08:31:57 +10003189 if (sh->check_state ||
3190 (s.syncing && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003191 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
Dan Williamsecc65c92008-06-28 08:31:57 +10003192 !test_bit(STRIPE_INSYNC, &sh->state)))
Dan Williamsa4456852007-07-09 11:56:43 -07003193 handle_parity_checks5(conf, sh, &s, disks);
Dan Williamse89f8962007-01-02 13:52:31 -07003194
Dan Williamsa4456852007-07-09 11:56:43 -07003195 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3197 clear_bit(STRIPE_SYNCING, &sh->state);
3198 }
NeilBrown4e5314b2005-11-08 21:39:22 -08003199
3200 /* If the failed drive is just a ReadError, then we might need to progress
3201 * the repair/check process
3202 */
Dan Williamsa4456852007-07-09 11:56:43 -07003203 if (s.failed == 1 && !conf->mddev->ro &&
3204 test_bit(R5_ReadError, &sh->dev[s.failed_num].flags)
3205 && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags)
3206 && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags)
NeilBrown4e5314b2005-11-08 21:39:22 -08003207 ) {
Dan Williamsa4456852007-07-09 11:56:43 -07003208 dev = &sh->dev[s.failed_num];
NeilBrown4e5314b2005-11-08 21:39:22 -08003209 if (!test_bit(R5_ReWrite, &dev->flags)) {
3210 set_bit(R5_Wantwrite, &dev->flags);
3211 set_bit(R5_ReWrite, &dev->flags);
3212 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003213 s.locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003214 } else {
3215 /* let's read it back */
3216 set_bit(R5_Wantread, &dev->flags);
3217 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003218 s.locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003219 }
3220 }
3221
Dan Williams600aa102008-06-28 08:32:05 +10003222 /* Finish reconstruct operations initiated by the expansion process */
3223 if (sh->reconstruct_state == reconstruct_state_result) {
NeilBrownab69ae12009-03-31 15:26:47 +11003224 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003225 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003226 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3227 /* sh cannot be written until sh2 has been read.
3228 * so arrange for sh to be delayed a little
3229 */
3230 set_bit(STRIPE_DELAYED, &sh->state);
3231 set_bit(STRIPE_HANDLE, &sh->state);
3232 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3233 &sh2->state))
3234 atomic_inc(&conf->preread_active_stripes);
3235 release_stripe(sh2);
3236 goto unlock;
3237 }
3238 if (sh2)
3239 release_stripe(sh2);
3240
Dan Williams600aa102008-06-28 08:32:05 +10003241 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamsf0a50d32007-01-02 13:52:31 -07003242 clear_bit(STRIPE_EXPANDING, &sh->state);
Dan Williams23397882008-07-23 20:05:34 -07003243 for (i = conf->raid_disks; i--; ) {
Dan Williamsf0a50d32007-01-02 13:52:31 -07003244 set_bit(R5_Wantwrite, &sh->dev[i].flags);
Dan Williams23397882008-07-23 20:05:34 -07003245 set_bit(R5_LOCKED, &sh->dev[i].flags);
Neil Brownefe31142008-06-28 08:31:14 +10003246 s.locked++;
Dan Williams23397882008-07-23 20:05:34 -07003247 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07003248 }
3249
3250 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
Dan Williams600aa102008-06-28 08:32:05 +10003251 !sh->reconstruct_state) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003252 /* Need to write out all blocks after computing parity */
3253 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003254 stripe_set_idx(sh->sector, conf, 0, sh);
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07003255 schedule_reconstruction(sh, &s, 1, 1);
Dan Williams600aa102008-06-28 08:32:05 +10003256 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003257 clear_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrownf6705572006-03-27 01:18:11 -08003258 atomic_dec(&conf->reshape_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003259 wake_up(&conf->wait_for_overlap);
3260 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3261 }
3262
Dan Williams0f94e872008-01-08 15:32:53 -08003263 if (s.expanding && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003264 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
Dan Williamsa4456852007-07-09 11:56:43 -07003265 handle_stripe_expansion(conf, sh, NULL);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003266
Dan Williams6bfe0b42008-04-30 00:52:32 -07003267 unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 spin_unlock(&sh->lock);
3269
Dan Williams6bfe0b42008-04-30 00:52:32 -07003270 /* wait for this device to become unblocked */
3271 if (unlikely(blocked_rdev))
3272 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3273
Dan Williams600aa102008-06-28 08:32:05 +10003274 if (s.ops_request)
Dan Williamsac6b53b2009-07-14 13:40:19 -07003275 raid_run_ops(sh, s.ops_request);
Dan Williamsd84e0f12007-01-02 13:52:30 -07003276
Dan Williamsc4e5ac02008-06-28 08:31:53 +10003277 ops_run_io(sh, &s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278
NeilBrown729a1862009-12-14 12:49:50 +11003279 if (dec_preread_active) {
3280 /* We delay this until after ops_run_io so that if make_request
3281 * is waiting on a barrier, it won't continue until the writes
3282 * have actually been submitted.
3283 */
3284 atomic_dec(&conf->preread_active_stripes);
3285 if (atomic_read(&conf->preread_active_stripes) <
3286 IO_THRESHOLD)
3287 md_wakeup_thread(conf->mddev->thread);
3288 }
Dan Williamsa4456852007-07-09 11:56:43 -07003289 return_io(return_bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290}
3291
NeilBrown14425772009-10-16 15:55:25 +11003292static void handle_stripe6(struct stripe_head *sh)
NeilBrown16a53ec2006-06-26 00:27:38 -07003293{
NeilBrownbff61972009-03-31 14:33:13 +11003294 raid5_conf_t *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08003295 int disks = sh->disks;
Dan Williamsa4456852007-07-09 11:56:43 -07003296 struct bio *return_bi = NULL;
NeilBrown34e04e82009-03-31 15:10:16 +11003297 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
Dan Williamsa4456852007-07-09 11:56:43 -07003298 struct stripe_head_state s;
3299 struct r6_state r6s;
NeilBrown16a53ec2006-06-26 00:27:38 -07003300 struct r5dev *dev, *pdev, *qdev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003301 mdk_rdev_t *blocked_rdev = NULL;
NeilBrown729a1862009-12-14 12:49:50 +11003302 int dec_preread_active = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003303
Dan Williams45b42332007-07-09 11:56:43 -07003304 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003305 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003306 (unsigned long long)sh->sector, sh->state,
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003307 atomic_read(&sh->count), pd_idx, qd_idx,
3308 sh->check_state, sh->reconstruct_state);
Dan Williamsa4456852007-07-09 11:56:43 -07003309 memset(&s, 0, sizeof(s));
NeilBrown16a53ec2006-06-26 00:27:38 -07003310
3311 spin_lock(&sh->lock);
3312 clear_bit(STRIPE_HANDLE, &sh->state);
3313 clear_bit(STRIPE_DELAYED, &sh->state);
3314
Dan Williamsa4456852007-07-09 11:56:43 -07003315 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3316 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3317 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07003318 /* Now to look around and see what can be done */
3319
3320 rcu_read_lock();
3321 for (i=disks; i--; ) {
3322 mdk_rdev_t *rdev;
3323 dev = &sh->dev[i];
NeilBrown16a53ec2006-06-26 00:27:38 -07003324
Dan Williams45b42332007-07-09 11:56:43 -07003325 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown16a53ec2006-06-26 00:27:38 -07003326 i, dev->flags, dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003327 /* maybe we can reply to a read
3328 *
3329 * new wantfill requests are only permitted while
3330 * ops_complete_biofill is guaranteed to be inactive
3331 */
3332 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3333 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3334 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07003335
3336 /* now count some things */
Dan Williamsa4456852007-07-09 11:56:43 -07003337 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3338 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003339 if (test_bit(R5_Wantcompute, &dev->flags)) {
3340 s.compute++;
3341 BUG_ON(s.compute > 2);
3342 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003343
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003344 if (test_bit(R5_Wantfill, &dev->flags)) {
3345 s.to_fill++;
3346 } else if (dev->toread)
Dan Williamsa4456852007-07-09 11:56:43 -07003347 s.to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003348 if (dev->towrite) {
Dan Williamsa4456852007-07-09 11:56:43 -07003349 s.to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003350 if (!test_bit(R5_OVERWRITE, &dev->flags))
Dan Williamsa4456852007-07-09 11:56:43 -07003351 s.non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003352 }
Dan Williamsa4456852007-07-09 11:56:43 -07003353 if (dev->written)
3354 s.written++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003355 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownac4090d2008-08-05 15:54:13 +10003356 if (blocked_rdev == NULL &&
3357 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07003358 blocked_rdev = rdev;
3359 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003360 }
NeilBrown415e72d2010-06-17 17:25:21 +10003361 clear_bit(R5_Insync, &dev->flags);
3362 if (!rdev)
3363 /* Not in-sync */;
3364 else if (test_bit(In_sync, &rdev->flags))
3365 set_bit(R5_Insync, &dev->flags);
3366 else {
3367 /* in sync if before recovery_offset */
3368 if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset)
3369 set_bit(R5_Insync, &dev->flags);
3370 }
3371 if (!test_bit(R5_Insync, &dev->flags)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003372 /* The ReadError flag will just be confusing now */
3373 clear_bit(R5_ReadError, &dev->flags);
3374 clear_bit(R5_ReWrite, &dev->flags);
3375 }
NeilBrown415e72d2010-06-17 17:25:21 +10003376 if (test_bit(R5_ReadError, &dev->flags))
3377 clear_bit(R5_Insync, &dev->flags);
3378 if (!test_bit(R5_Insync, &dev->flags)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003379 if (s.failed < 2)
3380 r6s.failed_num[s.failed] = i;
3381 s.failed++;
NeilBrown415e72d2010-06-17 17:25:21 +10003382 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003383 }
3384 rcu_read_unlock();
Dan Williams6bfe0b42008-04-30 00:52:32 -07003385
3386 if (unlikely(blocked_rdev)) {
NeilBrownac4090d2008-08-05 15:54:13 +10003387 if (s.syncing || s.expanding || s.expanded ||
3388 s.to_write || s.written) {
3389 set_bit(STRIPE_HANDLE, &sh->state);
3390 goto unlock;
3391 }
3392 /* There is nothing for the blocked_rdev to block */
3393 rdev_dec_pending(blocked_rdev, conf->mddev);
3394 blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003395 }
NeilBrownac4090d2008-08-05 15:54:13 +10003396
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003397 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3398 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3399 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3400 }
3401
Dan Williams45b42332007-07-09 11:56:43 -07003402 pr_debug("locked=%d uptodate=%d to_read=%d"
NeilBrown16a53ec2006-06-26 00:27:38 -07003403 " to_write=%d failed=%d failed_num=%d,%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003404 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3405 r6s.failed_num[0], r6s.failed_num[1]);
3406 /* check if the array has lost >2 devices and, if so, some requests
3407 * might need to be failed
NeilBrown16a53ec2006-06-26 00:27:38 -07003408 */
Dan Williamsa4456852007-07-09 11:56:43 -07003409 if (s.failed > 2 && s.to_read+s.to_write+s.written)
Dan Williams1fe797e2008-06-28 09:16:30 +10003410 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003411 if (s.failed > 2 && s.syncing) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003412 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3413 clear_bit(STRIPE_SYNCING, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003414 s.syncing = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003415 }
3416
3417 /*
3418 * might be able to return some write requests if the parity blocks
3419 * are safe, or on a failed drive
3420 */
3421 pdev = &sh->dev[pd_idx];
Dan Williamsa4456852007-07-09 11:56:43 -07003422 r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx)
3423 || (s.failed >= 2 && r6s.failed_num[1] == pd_idx);
NeilBrown34e04e82009-03-31 15:10:16 +11003424 qdev = &sh->dev[qd_idx];
3425 r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx)
3426 || (s.failed >= 2 && r6s.failed_num[1] == qd_idx);
NeilBrown16a53ec2006-06-26 00:27:38 -07003427
Dan Williamsa4456852007-07-09 11:56:43 -07003428 if ( s.written &&
3429 ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003430 && !test_bit(R5_LOCKED, &pdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003431 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
3432 ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003433 && !test_bit(R5_LOCKED, &qdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003434 && test_bit(R5_UPTODATE, &qdev->flags)))))
Dan Williams1fe797e2008-06-28 09:16:30 +10003435 handle_stripe_clean_event(conf, sh, disks, &return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003436
3437 /* Now we might consider reading some blocks, either to check/generate
3438 * parity, or to satisfy requests
3439 * or to load a block that is being partially written.
3440 */
Dan Williamsa4456852007-07-09 11:56:43 -07003441 if (s.to_read || s.non_overwrite || (s.to_write && s.failed) ||
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003442 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
Dan Williams1fe797e2008-06-28 09:16:30 +10003443 handle_stripe_fill6(sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003444
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003445 /* Now we check to see if any write operations have recently
3446 * completed
3447 */
3448 if (sh->reconstruct_state == reconstruct_state_drain_result) {
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003449
3450 sh->reconstruct_state = reconstruct_state_idle;
3451 /* All the 'written' buffers and the parity blocks are ready to
3452 * be written back to disk
3453 */
3454 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3455 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags));
3456 for (i = disks; i--; ) {
3457 dev = &sh->dev[i];
3458 if (test_bit(R5_LOCKED, &dev->flags) &&
3459 (i == sh->pd_idx || i == qd_idx ||
3460 dev->written)) {
3461 pr_debug("Writing block %d\n", i);
3462 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3463 set_bit(R5_Wantwrite, &dev->flags);
3464 if (!test_bit(R5_Insync, &dev->flags) ||
3465 ((i == sh->pd_idx || i == qd_idx) &&
3466 s.failed == 0))
3467 set_bit(STRIPE_INSYNC, &sh->state);
3468 }
3469 }
NeilBrown729a1862009-12-14 12:49:50 +11003470 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3471 dec_preread_active = 1;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003472 }
3473
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07003474 /* Now to consider new write requests and what else, if anything
3475 * should be read. We do not handle new writes when:
3476 * 1/ A 'write' operation (copy+gen_syndrome) is already in flight.
3477 * 2/ A 'check' operation is in flight, as it may clobber the parity
3478 * block.
3479 */
3480 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
Dan Williams1fe797e2008-06-28 09:16:30 +10003481 handle_stripe_dirtying6(conf, sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003482
3483 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamsa4456852007-07-09 11:56:43 -07003484 * Any reads will already have been scheduled, so we just see if enough
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003485 * data is available. The parity check is held off while parity
3486 * dependent operations are in flight.
NeilBrown16a53ec2006-06-26 00:27:38 -07003487 */
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003488 if (sh->check_state ||
3489 (s.syncing && s.locked == 0 &&
3490 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3491 !test_bit(STRIPE_INSYNC, &sh->state)))
Dan Williams36d1c642009-07-14 11:48:22 -07003492 handle_parity_checks6(conf, sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003493
Dan Williamsa4456852007-07-09 11:56:43 -07003494 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003495 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3496 clear_bit(STRIPE_SYNCING, &sh->state);
3497 }
3498
3499 /* If the failed drives are just a ReadError, then we might need
3500 * to progress the repair/check process
3501 */
Dan Williamsa4456852007-07-09 11:56:43 -07003502 if (s.failed <= 2 && !conf->mddev->ro)
3503 for (i = 0; i < s.failed; i++) {
3504 dev = &sh->dev[r6s.failed_num[i]];
NeilBrown16a53ec2006-06-26 00:27:38 -07003505 if (test_bit(R5_ReadError, &dev->flags)
3506 && !test_bit(R5_LOCKED, &dev->flags)
3507 && test_bit(R5_UPTODATE, &dev->flags)
3508 ) {
3509 if (!test_bit(R5_ReWrite, &dev->flags)) {
3510 set_bit(R5_Wantwrite, &dev->flags);
3511 set_bit(R5_ReWrite, &dev->flags);
3512 set_bit(R5_LOCKED, &dev->flags);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003513 s.locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003514 } else {
3515 /* let's read it back */
3516 set_bit(R5_Wantread, &dev->flags);
3517 set_bit(R5_LOCKED, &dev->flags);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003518 s.locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003519 }
3520 }
3521 }
NeilBrownf4168852007-02-28 20:11:53 -08003522
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003523 /* Finish reconstruct operations initiated by the expansion process */
3524 if (sh->reconstruct_state == reconstruct_state_result) {
3525 sh->reconstruct_state = reconstruct_state_idle;
3526 clear_bit(STRIPE_EXPANDING, &sh->state);
3527 for (i = conf->raid_disks; i--; ) {
3528 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3529 set_bit(R5_LOCKED, &sh->dev[i].flags);
3530 s.locked++;
3531 }
3532 }
3533
3534 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3535 !sh->reconstruct_state) {
NeilBrownab69ae12009-03-31 15:26:47 +11003536 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003537 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003538 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3539 /* sh cannot be written until sh2 has been read.
3540 * so arrange for sh to be delayed a little
3541 */
3542 set_bit(STRIPE_DELAYED, &sh->state);
3543 set_bit(STRIPE_HANDLE, &sh->state);
3544 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3545 &sh2->state))
3546 atomic_inc(&conf->preread_active_stripes);
3547 release_stripe(sh2);
3548 goto unlock;
3549 }
3550 if (sh2)
3551 release_stripe(sh2);
3552
NeilBrownf4168852007-02-28 20:11:53 -08003553 /* Need to write out all blocks after computing P&Q */
3554 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003555 stripe_set_idx(sh->sector, conf, 0, sh);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003556 schedule_reconstruction(sh, &s, 1, 1);
3557 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
NeilBrownf4168852007-02-28 20:11:53 -08003558 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3559 atomic_dec(&conf->reshape_stripes);
3560 wake_up(&conf->wait_for_overlap);
3561 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3562 }
3563
Dan Williams0f94e872008-01-08 15:32:53 -08003564 if (s.expanding && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003565 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
Dan Williamsa4456852007-07-09 11:56:43 -07003566 handle_stripe_expansion(conf, sh, &r6s);
NeilBrownf4168852007-02-28 20:11:53 -08003567
Dan Williams6bfe0b42008-04-30 00:52:32 -07003568 unlock:
NeilBrown16a53ec2006-06-26 00:27:38 -07003569 spin_unlock(&sh->lock);
3570
Dan Williams6bfe0b42008-04-30 00:52:32 -07003571 /* wait for this device to become unblocked */
3572 if (unlikely(blocked_rdev))
3573 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3574
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003575 if (s.ops_request)
3576 raid_run_ops(sh, s.ops_request);
3577
Dan Williamsf0e43bc2008-06-28 08:31:55 +10003578 ops_run_io(sh, &s);
3579
NeilBrown729a1862009-12-14 12:49:50 +11003580
3581 if (dec_preread_active) {
3582 /* We delay this until after ops_run_io so that if make_request
3583 * is waiting on a barrier, it won't continue until the writes
3584 * have actually been submitted.
3585 */
3586 atomic_dec(&conf->preread_active_stripes);
3587 if (atomic_read(&conf->preread_active_stripes) <
3588 IO_THRESHOLD)
3589 md_wakeup_thread(conf->mddev->thread);
3590 }
3591
Dan Williamsa4456852007-07-09 11:56:43 -07003592 return_io(return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003593}
3594
NeilBrown14425772009-10-16 15:55:25 +11003595static void handle_stripe(struct stripe_head *sh)
NeilBrown16a53ec2006-06-26 00:27:38 -07003596{
3597 if (sh->raid_conf->level == 6)
NeilBrown14425772009-10-16 15:55:25 +11003598 handle_stripe6(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -07003599 else
NeilBrown14425772009-10-16 15:55:25 +11003600 handle_stripe5(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -07003601}
3602
Arjan van de Ven858119e2006-01-14 13:20:43 -08003603static void raid5_activate_delayed(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604{
3605 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3606 while (!list_empty(&conf->delayed_list)) {
3607 struct list_head *l = conf->delayed_list.next;
3608 struct stripe_head *sh;
3609 sh = list_entry(l, struct stripe_head, lru);
3610 list_del_init(l);
3611 clear_bit(STRIPE_DELAYED, &sh->state);
3612 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3613 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003614 list_add_tail(&sh->lru, &conf->hold_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 }
NeilBrown6ed30032008-02-06 01:40:00 -08003616 } else
3617 blk_plug_device(conf->mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618}
3619
Arjan van de Ven858119e2006-01-14 13:20:43 -08003620static void activate_bit_delay(raid5_conf_t *conf)
NeilBrown72626682005-09-09 16:23:54 -07003621{
3622 /* device_lock is held */
3623 struct list_head head;
3624 list_add(&head, &conf->bitmap_list);
3625 list_del_init(&conf->bitmap_list);
3626 while (!list_empty(&head)) {
3627 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3628 list_del_init(&sh->lru);
3629 atomic_inc(&sh->count);
3630 __release_stripe(conf, sh);
3631 }
3632}
3633
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634static void unplug_slaves(mddev_t *mddev)
3635{
NeilBrown070ec552009-06-16 16:54:21 +10003636 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637 int i;
NeilBrown5e5e3e72009-10-16 16:35:30 +11003638 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639
3640 rcu_read_lock();
NeilBrown5e5e3e72009-10-16 16:35:30 +11003641 for (i = 0; i < devs; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -08003642 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -08003643 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
Jens Axboe165125e2007-07-24 09:28:11 +02003644 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645
3646 atomic_inc(&rdev->nr_pending);
3647 rcu_read_unlock();
3648
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05003649 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650
3651 rdev_dec_pending(rdev, mddev);
3652 rcu_read_lock();
3653 }
3654 }
3655 rcu_read_unlock();
3656}
3657
Jens Axboe165125e2007-07-24 09:28:11 +02003658static void raid5_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659{
3660 mddev_t *mddev = q->queuedata;
NeilBrown070ec552009-06-16 16:54:21 +10003661 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 unsigned long flags;
3663
3664 spin_lock_irqsave(&conf->device_lock, flags);
3665
NeilBrown72626682005-09-09 16:23:54 -07003666 if (blk_remove_plug(q)) {
3667 conf->seq_flush++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07003669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670 md_wakeup_thread(mddev->thread);
3671
3672 spin_unlock_irqrestore(&conf->device_lock, flags);
3673
3674 unplug_slaves(mddev);
3675}
3676
NeilBrownf022b2f2006-10-03 01:15:56 -07003677static int raid5_congested(void *data, int bits)
3678{
3679 mddev_t *mddev = data;
NeilBrown070ec552009-06-16 16:54:21 +10003680 raid5_conf_t *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07003681
3682 /* No difference between reads and writes. Just check
3683 * how busy the stripe_cache is
3684 */
NeilBrown3fa841d2009-09-23 18:10:29 +10003685
3686 if (mddev_congested(mddev, bits))
3687 return 1;
NeilBrownf022b2f2006-10-03 01:15:56 -07003688 if (conf->inactive_blocked)
3689 return 1;
3690 if (conf->quiesce)
3691 return 1;
3692 if (list_empty_careful(&conf->inactive_list))
3693 return 1;
3694
3695 return 0;
3696}
3697
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003698/* We want read requests to align with chunks where possible,
3699 * but write requests don't need to.
3700 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003701static int raid5_mergeable_bvec(struct request_queue *q,
3702 struct bvec_merge_data *bvm,
3703 struct bio_vec *biovec)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003704{
3705 mddev_t *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003706 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003707 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +10003708 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003709 unsigned int bio_sectors = bvm->bi_size >> 9;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003710
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003711 if ((bvm->bi_rw & 1) == WRITE)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003712 return biovec->bv_len; /* always allow writes to be mergeable */
3713
Andre Noll664e7c42009-06-18 08:45:27 +10003714 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3715 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003716 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3717 if (max < 0) max = 0;
3718 if (max <= biovec->bv_len && bio_sectors == 0)
3719 return biovec->bv_len;
3720 else
3721 return max;
3722}
3723
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003724
3725static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3726{
3727 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
Andre Noll9d8f0362009-06-18 08:45:01 +10003728 unsigned int chunk_sectors = mddev->chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003729 unsigned int bio_sectors = bio->bi_size >> 9;
3730
Andre Noll664e7c42009-06-18 08:45:27 +10003731 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3732 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003733 return chunk_sectors >=
3734 ((sector & (chunk_sectors - 1)) + bio_sectors);
3735}
3736
3737/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003738 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3739 * later sampled by raid5d.
3740 */
3741static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3742{
3743 unsigned long flags;
3744
3745 spin_lock_irqsave(&conf->device_lock, flags);
3746
3747 bi->bi_next = conf->retry_read_aligned_list;
3748 conf->retry_read_aligned_list = bi;
3749
3750 spin_unlock_irqrestore(&conf->device_lock, flags);
3751 md_wakeup_thread(conf->mddev->thread);
3752}
3753
3754
3755static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3756{
3757 struct bio *bi;
3758
3759 bi = conf->retry_read_aligned;
3760 if (bi) {
3761 conf->retry_read_aligned = NULL;
3762 return bi;
3763 }
3764 bi = conf->retry_read_aligned_list;
3765 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08003766 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003767 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02003768 /*
3769 * this sets the active strip count to 1 and the processed
3770 * strip count to zero (upper 8 bits)
3771 */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003772 bi->bi_phys_segments = 1; /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003773 }
3774
3775 return bi;
3776}
3777
3778
3779/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003780 * The "raid5_align_endio" should check if the read succeeded and if it
3781 * did, call bio_endio on the original bio (having bio_put the new bio
3782 * first).
3783 * If the read failed..
3784 */
NeilBrown6712ecf2007-09-27 12:47:43 +02003785static void raid5_align_endio(struct bio *bi, int error)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003786{
3787 struct bio* raid_bi = bi->bi_private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003788 mddev_t *mddev;
3789 raid5_conf_t *conf;
3790 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3791 mdk_rdev_t *rdev;
3792
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003793 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003794
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003795 rdev = (void*)raid_bi->bi_next;
3796 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11003797 mddev = rdev->mddev;
3798 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003799
3800 rdev_dec_pending(rdev, conf->mddev);
3801
3802 if (!error && uptodate) {
NeilBrown6712ecf2007-09-27 12:47:43 +02003803 bio_endio(raid_bi, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003804 if (atomic_dec_and_test(&conf->active_aligned_reads))
3805 wake_up(&conf->wait_for_stripe);
NeilBrown6712ecf2007-09-27 12:47:43 +02003806 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003807 }
3808
3809
Dan Williams45b42332007-07-09 11:56:43 -07003810 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003811
3812 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003813}
3814
Neil Brown387bb172007-02-08 14:20:29 -08003815static int bio_fits_rdev(struct bio *bi)
3816{
Jens Axboe165125e2007-07-24 09:28:11 +02003817 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
Neil Brown387bb172007-02-08 14:20:29 -08003818
Martin K. Petersenae03bf62009-05-22 17:17:50 -04003819 if ((bi->bi_size>>9) > queue_max_sectors(q))
Neil Brown387bb172007-02-08 14:20:29 -08003820 return 0;
3821 blk_recount_segments(q, bi);
Martin K. Petersen8a783622010-02-26 00:20:39 -05003822 if (bi->bi_phys_segments > queue_max_segments(q))
Neil Brown387bb172007-02-08 14:20:29 -08003823 return 0;
3824
3825 if (q->merge_bvec_fn)
3826 /* it's too hard to apply the merge_bvec_fn at this stage,
3827 * just just give up
3828 */
3829 return 0;
3830
3831 return 1;
3832}
3833
3834
NeilBrown21a52c62010-04-01 15:02:13 +11003835static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003836{
NeilBrown070ec552009-06-16 16:54:21 +10003837 raid5_conf_t *conf = mddev->private;
NeilBrown8553fe7ec2009-12-14 12:49:47 +11003838 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003839 struct bio* align_bi;
3840 mdk_rdev_t *rdev;
3841
3842 if (!in_chunk_boundary(mddev, raid_bio)) {
Dan Williams45b42332007-07-09 11:56:43 -07003843 pr_debug("chunk_aligned_read : non aligned\n");
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003844 return 0;
3845 }
3846 /*
NeilBrown99c0fb52009-03-31 14:39:38 +11003847 * use bio_clone to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003848 */
3849 align_bi = bio_clone(raid_bio, GFP_NOIO);
3850 if (!align_bi)
3851 return 0;
3852 /*
3853 * set bi_end_io to a new function, and set bi_private to the
3854 * original bio.
3855 */
3856 align_bi->bi_end_io = raid5_align_endio;
3857 align_bi->bi_private = raid_bio;
3858 /*
3859 * compute position
3860 */
NeilBrown112bf892009-03-31 14:39:38 +11003861 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3862 0,
NeilBrown911d4ee2009-03-31 14:39:38 +11003863 &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003864
3865 rcu_read_lock();
3866 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3867 if (rdev && test_bit(In_sync, &rdev->flags)) {
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003868 atomic_inc(&rdev->nr_pending);
3869 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003870 raid_bio->bi_next = (void*)rdev;
3871 align_bi->bi_bdev = rdev->bdev;
3872 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3873 align_bi->bi_sector += rdev->data_offset;
3874
Neil Brown387bb172007-02-08 14:20:29 -08003875 if (!bio_fits_rdev(align_bi)) {
3876 /* too big in some way */
3877 bio_put(align_bi);
3878 rdev_dec_pending(rdev, mddev);
3879 return 0;
3880 }
3881
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003882 spin_lock_irq(&conf->device_lock);
3883 wait_event_lock_irq(conf->wait_for_stripe,
3884 conf->quiesce == 0,
3885 conf->device_lock, /* nothing */);
3886 atomic_inc(&conf->active_aligned_reads);
3887 spin_unlock_irq(&conf->device_lock);
3888
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003889 generic_make_request(align_bi);
3890 return 1;
3891 } else {
3892 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003893 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003894 return 0;
3895 }
3896}
3897
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003898/* __get_priority_stripe - get the next stripe to process
3899 *
3900 * Full stripe writes are allowed to pass preread active stripes up until
3901 * the bypass_threshold is exceeded. In general the bypass_count
3902 * increments when the handle_list is handled before the hold_list; however, it
3903 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3904 * stripe with in flight i/o. The bypass_count will be reset when the
3905 * head of the hold_list has changed, i.e. the head was promoted to the
3906 * handle_list.
3907 */
3908static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3909{
3910 struct stripe_head *sh;
3911
3912 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3913 __func__,
3914 list_empty(&conf->handle_list) ? "empty" : "busy",
3915 list_empty(&conf->hold_list) ? "empty" : "busy",
3916 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3917
3918 if (!list_empty(&conf->handle_list)) {
3919 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3920
3921 if (list_empty(&conf->hold_list))
3922 conf->bypass_count = 0;
3923 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3924 if (conf->hold_list.next == conf->last_hold)
3925 conf->bypass_count++;
3926 else {
3927 conf->last_hold = conf->hold_list.next;
3928 conf->bypass_count -= conf->bypass_threshold;
3929 if (conf->bypass_count < 0)
3930 conf->bypass_count = 0;
3931 }
3932 }
3933 } else if (!list_empty(&conf->hold_list) &&
3934 ((conf->bypass_threshold &&
3935 conf->bypass_count > conf->bypass_threshold) ||
3936 atomic_read(&conf->pending_full_writes) == 0)) {
3937 sh = list_entry(conf->hold_list.next,
3938 typeof(*sh), lru);
3939 conf->bypass_count -= conf->bypass_threshold;
3940 if (conf->bypass_count < 0)
3941 conf->bypass_count = 0;
3942 } else
3943 return NULL;
3944
3945 list_del_init(&sh->lru);
3946 atomic_inc(&sh->count);
3947 BUG_ON(atomic_read(&sh->count) != 1);
3948 return sh;
3949}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003950
NeilBrown21a52c62010-04-01 15:02:13 +11003951static int make_request(mddev_t *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952{
NeilBrown070ec552009-06-16 16:54:21 +10003953 raid5_conf_t *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11003954 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 sector_t new_sector;
3956 sector_t logical_sector, last_sector;
3957 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01003958 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11003959 int remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960
Jens Axboe1f98a132009-09-11 14:32:04 +02003961 if (unlikely(bio_rw_flagged(bi, BIO_RW_BARRIER))) {
NeilBrowna2826aa2009-12-14 12:49:49 +11003962 /* Drain all pending writes. We only really need
3963 * to ensure they have been submitted, but this is
3964 * easier.
3965 */
3966 mddev->pers->quiesce(mddev, 1);
3967 mddev->pers->quiesce(mddev, 0);
3968 md_barrier_request(mddev, bi);
NeilBrowne5dcdd82005-09-09 16:23:41 -07003969 return 0;
3970 }
3971
NeilBrown3d310eb2005-06-21 17:17:26 -07003972 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07003973
NeilBrown802ba062006-12-13 00:34:13 -08003974 if (rw == READ &&
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003975 mddev->reshape_position == MaxSector &&
NeilBrown21a52c62010-04-01 15:02:13 +11003976 chunk_aligned_read(mddev,bi))
NeilBrown99c0fb52009-03-31 14:39:38 +11003977 return 0;
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003978
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3980 last_sector = bi->bi_sector + (bi->bi_size>>9);
3981 bi->bi_next = NULL;
3982 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07003983
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3985 DEFINE_WAIT(w);
NeilBrown16a53ec2006-06-26 00:27:38 -07003986 int disks, data_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003987 int previous;
NeilBrownb578d552006-03-27 01:18:12 -08003988
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003989 retry:
NeilBrownb5663ba2009-03-31 14:39:38 +11003990 previous = 0;
NeilBrownb0f9ec02009-03-31 15:27:18 +11003991 disks = conf->raid_disks;
NeilBrownb578d552006-03-27 01:18:12 -08003992 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003993 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11003994 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08003995 * 64bit on a 32bit platform, and so it might be
3996 * possible to see a half-updated value
NeilBrownfef9c612009-03-31 15:16:46 +11003997 * Ofcourse reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08003998 * the lock is dropped, so once we get a reference
3999 * to the stripe that we think it is, we will have
4000 * to check again.
4001 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004002 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004003 if (mddev->delta_disks < 0
4004 ? logical_sector < conf->reshape_progress
4005 : logical_sector >= conf->reshape_progress) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004006 disks = conf->previous_raid_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11004007 previous = 1;
4008 } else {
NeilBrownfef9c612009-03-31 15:16:46 +11004009 if (mddev->delta_disks < 0
4010 ? logical_sector < conf->reshape_safe
4011 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08004012 spin_unlock_irq(&conf->device_lock);
4013 schedule();
4014 goto retry;
4015 }
4016 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004017 spin_unlock_irq(&conf->device_lock);
4018 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004019 data_disks = disks - conf->max_degraded;
4020
NeilBrown112bf892009-03-31 14:39:38 +11004021 new_sector = raid5_compute_sector(conf, logical_sector,
4022 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11004023 &dd_idx, NULL);
NeilBrown0c55e022010-05-03 14:09:02 +10004024 pr_debug("raid456: make_request, sector %llu logical %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 (unsigned long long)new_sector,
4026 (unsigned long long)logical_sector);
4027
NeilBrownb5663ba2009-03-31 14:39:38 +11004028 sh = get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10004029 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004030 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11004031 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004032 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08004033 * stripe, so we must do the range check again.
4034 * Expansion could still move past after this
4035 * test, but as we are holding a reference to
4036 * 'sh', we know that if that happens,
4037 * STRIPE_EXPANDING will get set and the expansion
4038 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004039 */
4040 int must_retry = 0;
4041 spin_lock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11004042 if (mddev->delta_disks < 0
4043 ? logical_sector >= conf->reshape_progress
4044 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004045 /* mismatch, need to try again */
4046 must_retry = 1;
4047 spin_unlock_irq(&conf->device_lock);
4048 if (must_retry) {
4049 release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07004050 schedule();
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004051 goto retry;
4052 }
4053 }
NeilBrowne62e58a2009-07-01 13:15:35 +10004054
NeilBrowna5c308d2009-07-01 13:15:35 +10004055 if (bio_data_dir(bi) == WRITE &&
4056 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08004057 logical_sector < mddev->suspend_hi) {
4058 release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10004059 /* As the suspend_* range is controlled by
4060 * userspace, we want an interruptible
4061 * wait.
4062 */
4063 flush_signals(current);
4064 prepare_to_wait(&conf->wait_for_overlap,
4065 &w, TASK_INTERRUPTIBLE);
4066 if (logical_sector >= mddev->suspend_lo &&
4067 logical_sector < mddev->suspend_hi)
4068 schedule();
NeilBrowne464eaf2006-03-27 01:18:14 -08004069 goto retry;
4070 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08004071
4072 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
4073 !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) {
4074 /* Stripe is busy expanding or
4075 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07004076 * and wait a while
4077 */
4078 raid5_unplug_device(mddev->queue);
4079 release_stripe(sh);
4080 schedule();
4081 goto retry;
4082 }
4083 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown6ed30032008-02-06 01:40:00 -08004084 set_bit(STRIPE_HANDLE, &sh->state);
4085 clear_bit(STRIPE_DELAYED, &sh->state);
NeilBrown729a1862009-12-14 12:49:50 +11004086 if (mddev->barrier &&
4087 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4088 atomic_inc(&conf->preread_active_stripes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 } else {
4091 /* cannot get stripe for read-ahead, just give-up */
4092 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4093 finish_wait(&conf->wait_for_overlap, &w);
4094 break;
4095 }
4096
4097 }
4098 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004099 remaining = raid5_dec_bi_phys_segments(bi);
NeilBrownf6344752006-03-27 01:18:17 -08004100 spin_unlock_irq(&conf->device_lock);
4101 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102
NeilBrown16a53ec2006-06-26 00:27:38 -07004103 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02004105
Neil Brown0e13fe232008-06-28 08:31:20 +10004106 bio_endio(bi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107 }
NeilBrown729a1862009-12-14 12:49:50 +11004108
4109 if (mddev->barrier) {
4110 /* We need to wait for the stripes to all be handled.
4111 * So: wait for preread_active_stripes to drop to 0.
4112 */
4113 wait_event(mddev->thread->wqueue,
4114 atomic_read(&conf->preread_active_stripes) == 0);
4115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116 return 0;
4117}
4118
Dan Williamsb522adc2009-03-31 15:00:31 +11004119static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
4120
NeilBrown52c03292006-06-26 00:27:43 -07004121static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122{
NeilBrown52c03292006-06-26 00:27:43 -07004123 /* reshaping is quite different to recovery/resync so it is
4124 * handled quite separately ... here.
4125 *
4126 * On each call to sync_request, we gather one chunk worth of
4127 * destination stripes and flag them as expanding.
4128 * Then we find all the source stripes and request reads.
4129 * As the reads complete, handle_stripe will copy the data
4130 * into the destination stripe and release that stripe.
4131 */
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004132 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004133 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08004134 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08004135 int raid_disks = conf->previous_raid_disks;
4136 int data_disks = raid_disks - conf->max_degraded;
4137 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07004138 int i;
4139 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11004140 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11004141 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11004142 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11004143 struct list_head stripes;
NeilBrown52c03292006-06-26 00:27:43 -07004144
NeilBrownfef9c612009-03-31 15:16:46 +11004145 if (sector_nr == 0) {
4146 /* If restarting in the middle, skip the initial sectors */
4147 if (mddev->delta_disks < 0 &&
4148 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4149 sector_nr = raid5_size(mddev, 0, 0)
4150 - conf->reshape_progress;
NeilBrowna6397552009-08-13 10:13:00 +10004151 } else if (mddev->delta_disks >= 0 &&
NeilBrownfef9c612009-03-31 15:16:46 +11004152 conf->reshape_progress > 0)
4153 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004154 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004155 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11004156 mddev->curr_resync_completed = sector_nr;
4157 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11004158 *skipped = 1;
4159 return sector_nr;
4160 }
NeilBrown52c03292006-06-26 00:27:43 -07004161 }
4162
NeilBrown7a661382009-03-31 15:21:40 +11004163 /* We need to process a full chunk at a time.
4164 * If old and new chunk sizes differ, we need to process the
4165 * largest of these
4166 */
Andre Noll664e7c42009-06-18 08:45:27 +10004167 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4168 reshape_sectors = mddev->new_chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004169 else
Andre Noll9d8f0362009-06-18 08:45:01 +10004170 reshape_sectors = mddev->chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004171
NeilBrown52c03292006-06-26 00:27:43 -07004172 /* we update the metadata when there is more than 3Meg
4173 * in the block range (that is rather arbitrary, should
4174 * probably be time based) or when the data about to be
4175 * copied would over-write the source of the data at
4176 * the front of the range.
NeilBrownfef9c612009-03-31 15:16:46 +11004177 * i.e. one new_stripe along from reshape_progress new_maps
4178 * to after where reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07004179 */
NeilBrownfef9c612009-03-31 15:16:46 +11004180 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004181 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11004182 readpos = conf->reshape_progress;
4183 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004184 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08004185 sector_div(safepos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004186 if (mddev->delta_disks < 0) {
NeilBrowned37d832009-05-27 21:39:05 +10004187 writepos -= min_t(sector_t, reshape_sectors, writepos);
NeilBrownc8f517c2009-03-31 15:28:40 +11004188 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004189 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11004190 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004191 writepos += reshape_sectors;
NeilBrowned37d832009-05-27 21:39:05 +10004192 readpos -= min_t(sector_t, reshape_sectors, readpos);
4193 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11004194 }
NeilBrown52c03292006-06-26 00:27:43 -07004195
NeilBrownc8f517c2009-03-31 15:28:40 +11004196 /* 'writepos' is the most advanced device address we might write.
4197 * 'readpos' is the least advanced device address we might read.
4198 * 'safepos' is the least address recorded in the metadata as having
4199 * been reshaped.
4200 * If 'readpos' is behind 'writepos', then there is no way that we can
4201 * ensure safety in the face of a crash - that must be done by userspace
4202 * making a backup of the data. So in that case there is no particular
4203 * rush to update metadata.
4204 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4205 * update the metadata to advance 'safepos' to match 'readpos' so that
4206 * we can be safe in the event of a crash.
4207 * So we insist on updating metadata if safepos is behind writepos and
4208 * readpos is beyond writepos.
4209 * In any case, update the metadata every 10 seconds.
4210 * Maybe that number should be configurable, but I'm not sure it is
4211 * worth it.... maybe it could be a multiple of safemode_delay???
4212 */
NeilBrownfef9c612009-03-31 15:16:46 +11004213 if ((mddev->delta_disks < 0
NeilBrownc8f517c2009-03-31 15:28:40 +11004214 ? (safepos > writepos && readpos < writepos)
4215 : (safepos < writepos && readpos > writepos)) ||
4216 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07004217 /* Cannot proceed until we've updated the superblock... */
4218 wait_event(conf->wait_for_overlap,
4219 atomic_read(&conf->reshape_stripes)==0);
NeilBrownfef9c612009-03-31 15:16:46 +11004220 mddev->reshape_position = conf->reshape_progress;
NeilBrownacb180b2009-04-14 16:28:34 +10004221 mddev->curr_resync_completed = mddev->curr_resync;
NeilBrownc8f517c2009-03-31 15:28:40 +11004222 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07004223 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07004224 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07004225 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrown52c03292006-06-26 00:27:43 -07004226 kthread_should_stop());
4227 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004228 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07004229 spin_unlock_irq(&conf->device_lock);
4230 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004231 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07004232 }
4233
NeilBrownec32a2b2009-03-31 15:17:38 +11004234 if (mddev->delta_disks < 0) {
4235 BUG_ON(conf->reshape_progress == 0);
4236 stripe_addr = writepos;
4237 BUG_ON((mddev->dev_sectors &
NeilBrown7a661382009-03-31 15:21:40 +11004238 ~((sector_t)reshape_sectors - 1))
4239 - reshape_sectors - stripe_addr
NeilBrownec32a2b2009-03-31 15:17:38 +11004240 != sector_nr);
4241 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004242 BUG_ON(writepos != sector_nr + reshape_sectors);
NeilBrownec32a2b2009-03-31 15:17:38 +11004243 stripe_addr = sector_nr;
4244 }
NeilBrownab69ae12009-03-31 15:26:47 +11004245 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11004246 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07004247 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10004248 int skipped_disk = 0;
NeilBrowna8c906c2009-06-09 14:39:59 +10004249 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004250 set_bit(STRIPE_EXPANDING, &sh->state);
4251 atomic_inc(&conf->reshape_stripes);
4252 /* If any of this stripe is beyond the end of the old
4253 * array, then we need to zero those blocks
4254 */
4255 for (j=sh->disks; j--;) {
4256 sector_t s;
4257 if (j == sh->pd_idx)
4258 continue;
NeilBrownf4168852007-02-28 20:11:53 -08004259 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11004260 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08004261 continue;
NeilBrown784052e2009-03-31 15:19:07 +11004262 s = compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11004263 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10004264 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07004265 continue;
4266 }
4267 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4268 set_bit(R5_Expanded, &sh->dev[j].flags);
4269 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4270 }
NeilBrowna9f326e2009-09-23 18:06:41 +10004271 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07004272 set_bit(STRIPE_EXPAND_READY, &sh->state);
4273 set_bit(STRIPE_HANDLE, &sh->state);
4274 }
NeilBrownab69ae12009-03-31 15:26:47 +11004275 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07004276 }
4277 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004278 if (mddev->delta_disks < 0)
NeilBrown7a661382009-03-31 15:21:40 +11004279 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11004280 else
NeilBrown7a661382009-03-31 15:21:40 +11004281 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07004282 spin_unlock_irq(&conf->device_lock);
4283 /* Ok, those stripe are ready. We can start scheduling
4284 * reads on the source stripes.
4285 * The source stripes are determined by mapping the first and last
4286 * block on the destination stripes.
4287 */
NeilBrown52c03292006-06-26 00:27:43 -07004288 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11004289 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11004290 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07004291 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10004292 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10004293 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11004294 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11004295 if (last_sector >= mddev->dev_sectors)
4296 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07004297 while (first_sector <= last_sector) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004298 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004299 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4300 set_bit(STRIPE_HANDLE, &sh->state);
4301 release_stripe(sh);
4302 first_sector += STRIPE_SECTORS;
4303 }
NeilBrownab69ae12009-03-31 15:26:47 +11004304 /* Now that the sources are clearly marked, we can release
4305 * the destination stripes
4306 */
4307 while (!list_empty(&stripes)) {
4308 sh = list_entry(stripes.next, struct stripe_head, lru);
4309 list_del_init(&sh->lru);
4310 release_stripe(sh);
4311 }
NeilBrownc6207272008-02-06 01:39:52 -08004312 /* If this takes us to the resync_max point where we have to pause,
4313 * then we need to write out the superblock.
4314 */
NeilBrown7a661382009-03-31 15:21:40 +11004315 sector_nr += reshape_sectors;
NeilBrownc03f6a12009-04-17 11:06:30 +10004316 if ((sector_nr - mddev->curr_resync_completed) * 2
4317 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08004318 /* Cannot proceed until we've updated the superblock... */
4319 wait_event(conf->wait_for_overlap,
4320 atomic_read(&conf->reshape_stripes) == 0);
NeilBrownfef9c612009-03-31 15:16:46 +11004321 mddev->reshape_position = conf->reshape_progress;
NeilBrown48606a92009-06-18 09:14:12 +10004322 mddev->curr_resync_completed = mddev->curr_resync + reshape_sectors;
NeilBrownc8f517c2009-03-31 15:28:40 +11004323 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08004324 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4325 md_wakeup_thread(mddev->thread);
4326 wait_event(mddev->sb_wait,
4327 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4328 || kthread_should_stop());
4329 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004330 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08004331 spin_unlock_irq(&conf->device_lock);
4332 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004333 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08004334 }
NeilBrown7a661382009-03-31 15:21:40 +11004335 return reshape_sectors;
NeilBrown52c03292006-06-26 00:27:43 -07004336}
4337
4338/* FIXME go_faster isn't used */
4339static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
4340{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004341 raid5_conf_t *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07004342 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11004343 sector_t max_sector = mddev->dev_sectors;
NeilBrown72626682005-09-09 16:23:54 -07004344 int sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07004345 int still_degraded = 0;
4346 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004347
NeilBrown72626682005-09-09 16:23:54 -07004348 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004349 /* just being told to finish up .. nothing much to do */
4350 unplug_slaves(mddev);
NeilBrowncea9c222009-03-31 15:15:05 +11004351
NeilBrown29269552006-03-27 01:18:10 -08004352 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4353 end_reshape(conf);
4354 return 0;
4355 }
NeilBrown72626682005-09-09 16:23:54 -07004356
4357 if (mddev->curr_resync < max_sector) /* aborted */
4358 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4359 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07004360 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07004361 conf->fullsync = 0;
4362 bitmap_close_sync(mddev->bitmap);
4363
Linus Torvalds1da177e2005-04-16 15:20:36 -07004364 return 0;
4365 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08004366
NeilBrown64bd6602009-08-03 10:59:58 +10004367 /* Allow raid5_quiesce to complete */
4368 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4369
NeilBrown52c03292006-06-26 00:27:43 -07004370 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4371 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08004372
NeilBrownc6207272008-02-06 01:39:52 -08004373 /* No need to check resync_max as we never do more than one
4374 * stripe, and as resync_max will always be on a chunk boundary,
4375 * if the check in md_do_sync didn't fire, there is no chance
4376 * of overstepping resync_max here
4377 */
4378
NeilBrown16a53ec2006-06-26 00:27:38 -07004379 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07004380 * to resync, then assert that we are finished, because there is
4381 * nothing we can do.
4382 */
NeilBrown3285edf2006-06-26 00:27:55 -07004383 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07004384 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004385 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07004386 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387 return rv;
4388 }
NeilBrown72626682005-09-09 16:23:54 -07004389 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrown3855ad92005-11-08 21:39:38 -08004390 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown72626682005-09-09 16:23:54 -07004391 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4392 /* we can skip this block, and probably more */
4393 sync_blocks /= STRIPE_SECTORS;
4394 *skipped = 1;
4395 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004397
NeilBrownb47490c2008-02-06 01:39:50 -08004398
4399 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4400
NeilBrowna8c906c2009-06-09 14:39:59 +10004401 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004402 if (sh == NULL) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004403 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004404 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07004405 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07004406 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08004407 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004408 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004409 /* Need to check if array will still be degraded after recovery/resync
4410 * We don't need to check the 'failed' flag as when that gets set,
4411 * recovery aborts.
4412 */
NeilBrownf001a702009-06-09 14:30:31 +10004413 for (i = 0; i < conf->raid_disks; i++)
NeilBrown16a53ec2006-06-26 00:27:38 -07004414 if (conf->disks[i].rdev == NULL)
4415 still_degraded = 1;
4416
4417 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4418
4419 spin_lock(&sh->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 set_bit(STRIPE_SYNCING, &sh->state);
4421 clear_bit(STRIPE_INSYNC, &sh->state);
4422 spin_unlock(&sh->lock);
4423
NeilBrown14425772009-10-16 15:55:25 +11004424 handle_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004425 release_stripe(sh);
4426
4427 return STRIPE_SECTORS;
4428}
4429
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004430static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4431{
4432 /* We may not be able to submit a whole bio at once as there
4433 * may not be enough stripe_heads available.
4434 * We cannot pre-allocate enough stripe_heads as we may need
4435 * more than exist in the cache (if we allow ever large chunks).
4436 * So we do one stripe head at a time and record in
4437 * ->bi_hw_segments how many have been done.
4438 *
4439 * We *know* that this entire raid_bio is in one chunk, so
4440 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4441 */
4442 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11004443 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004444 sector_t sector, logical_sector, last_sector;
4445 int scnt = 0;
4446 int remaining;
4447 int handled = 0;
4448
4449 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11004450 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11004451 0, &dd_idx, NULL);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004452 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4453
4454 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08004455 logical_sector += STRIPE_SECTORS,
4456 sector += STRIPE_SECTORS,
4457 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004458
Jens Axboe960e7392008-08-15 10:41:18 +02004459 if (scnt < raid5_bi_hw_segments(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004460 /* already done this stripe */
4461 continue;
4462
NeilBrowna8c906c2009-06-09 14:39:59 +10004463 sh = get_active_stripe(conf, sector, 0, 1, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004464
4465 if (!sh) {
4466 /* failed to get a stripe - must wait */
Jens Axboe960e7392008-08-15 10:41:18 +02004467 raid5_set_bi_hw_segments(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004468 conf->retry_read_aligned = raid_bio;
4469 return handled;
4470 }
4471
4472 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
Neil Brown387bb172007-02-08 14:20:29 -08004473 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4474 release_stripe(sh);
Jens Axboe960e7392008-08-15 10:41:18 +02004475 raid5_set_bi_hw_segments(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08004476 conf->retry_read_aligned = raid_bio;
4477 return handled;
4478 }
4479
Dan Williams36d1c642009-07-14 11:48:22 -07004480 handle_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004481 release_stripe(sh);
4482 handled++;
4483 }
4484 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004485 remaining = raid5_dec_bi_phys_segments(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004486 spin_unlock_irq(&conf->device_lock);
Neil Brown0e13fe232008-06-28 08:31:20 +10004487 if (remaining == 0)
4488 bio_endio(raid_bio, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004489 if (atomic_dec_and_test(&conf->active_aligned_reads))
4490 wake_up(&conf->wait_for_stripe);
4491 return handled;
4492}
4493
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004494
Linus Torvalds1da177e2005-04-16 15:20:36 -07004495/*
4496 * This is our raid5 kernel thread.
4497 *
4498 * We scan the hash table for stripes which can be handled now.
4499 * During the scan, completed stripes are saved for us by the interrupt
4500 * handler, so that they will not have to wait for our next wakeup.
4501 */
NeilBrown6ed30032008-02-06 01:40:00 -08004502static void raid5d(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004503{
4504 struct stripe_head *sh;
NeilBrown070ec552009-06-16 16:54:21 +10004505 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004506 int handled;
4507
Dan Williams45b42332007-07-09 11:56:43 -07004508 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004509
4510 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004511
4512 handled = 0;
4513 spin_lock_irq(&conf->device_lock);
4514 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004515 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004516
NeilBrownae3c20c2006-07-10 04:44:17 -07004517 if (conf->seq_flush != conf->seq_write) {
NeilBrown72626682005-09-09 16:23:54 -07004518 int seq = conf->seq_flush;
NeilBrown700e4322005-11-28 13:44:10 -08004519 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07004520 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08004521 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07004522 conf->seq_write = seq;
4523 activate_bit_delay(conf);
4524 }
4525
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004526 while ((bio = remove_bio_from_retry(conf))) {
4527 int ok;
4528 spin_unlock_irq(&conf->device_lock);
4529 ok = retry_aligned_read(conf, bio);
4530 spin_lock_irq(&conf->device_lock);
4531 if (!ok)
4532 break;
4533 handled++;
4534 }
4535
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004536 sh = __get_priority_stripe(conf);
4537
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004538 if (!sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004539 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004540 spin_unlock_irq(&conf->device_lock);
4541
4542 handled++;
Dan Williams417b8d42009-10-16 16:25:22 +11004543 handle_stripe(sh);
4544 release_stripe(sh);
4545 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546
4547 spin_lock_irq(&conf->device_lock);
4548 }
Dan Williams45b42332007-07-09 11:56:43 -07004549 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004550
4551 spin_unlock_irq(&conf->device_lock);
4552
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004553 async_tx_issue_pending_all();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004554 unplug_slaves(mddev);
4555
Dan Williams45b42332007-07-09 11:56:43 -07004556 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004557}
4558
NeilBrown3f294f42005-11-08 21:39:25 -08004559static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004560raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004561{
NeilBrown070ec552009-06-16 16:54:21 +10004562 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004563 if (conf)
4564 return sprintf(page, "%d\n", conf->max_nr_stripes);
4565 else
4566 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004567}
4568
NeilBrownc41d4ac2010-06-01 19:37:24 +10004569int
4570raid5_set_cache_size(mddev_t *mddev, int size)
4571{
4572 raid5_conf_t *conf = mddev->private;
4573 int err;
4574
4575 if (size <= 16 || size > 32768)
4576 return -EINVAL;
4577 while (size < conf->max_nr_stripes) {
4578 if (drop_one_stripe(conf))
4579 conf->max_nr_stripes--;
4580 else
4581 break;
4582 }
4583 err = md_allow_write(mddev);
4584 if (err)
4585 return err;
4586 while (size > conf->max_nr_stripes) {
4587 if (grow_one_stripe(conf))
4588 conf->max_nr_stripes++;
4589 else break;
4590 }
4591 return 0;
4592}
4593EXPORT_SYMBOL(raid5_set_cache_size);
4594
NeilBrown3f294f42005-11-08 21:39:25 -08004595static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004596raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08004597{
NeilBrown070ec552009-06-16 16:54:21 +10004598 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004599 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07004600 int err;
4601
NeilBrown3f294f42005-11-08 21:39:25 -08004602 if (len >= PAGE_SIZE)
4603 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08004604 if (!conf)
4605 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08004606
Dan Williams4ef197d82008-04-28 02:15:54 -07004607 if (strict_strtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08004608 return -EINVAL;
NeilBrownc41d4ac2010-06-01 19:37:24 +10004609 err = raid5_set_cache_size(mddev, new);
Dan Williamsb5470dc2008-06-27 21:44:04 -07004610 if (err)
4611 return err;
NeilBrown3f294f42005-11-08 21:39:25 -08004612 return len;
4613}
NeilBrown007583c2005-11-08 21:39:30 -08004614
NeilBrown96de1e62005-11-08 21:39:39 -08004615static struct md_sysfs_entry
4616raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4617 raid5_show_stripe_cache_size,
4618 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08004619
4620static ssize_t
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004621raid5_show_preread_threshold(mddev_t *mddev, char *page)
4622{
NeilBrown070ec552009-06-16 16:54:21 +10004623 raid5_conf_t *conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004624 if (conf)
4625 return sprintf(page, "%d\n", conf->bypass_threshold);
4626 else
4627 return 0;
4628}
4629
4630static ssize_t
4631raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4632{
NeilBrown070ec552009-06-16 16:54:21 +10004633 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004634 unsigned long new;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004635 if (len >= PAGE_SIZE)
4636 return -EINVAL;
4637 if (!conf)
4638 return -ENODEV;
4639
Dan Williams4ef197d82008-04-28 02:15:54 -07004640 if (strict_strtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004641 return -EINVAL;
Dan Williams4ef197d82008-04-28 02:15:54 -07004642 if (new > conf->max_nr_stripes)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004643 return -EINVAL;
4644 conf->bypass_threshold = new;
4645 return len;
4646}
4647
4648static struct md_sysfs_entry
4649raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4650 S_IRUGO | S_IWUSR,
4651 raid5_show_preread_threshold,
4652 raid5_store_preread_threshold);
4653
4654static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08004655stripe_cache_active_show(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004656{
NeilBrown070ec552009-06-16 16:54:21 +10004657 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004658 if (conf)
4659 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4660 else
4661 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004662}
4663
NeilBrown96de1e62005-11-08 21:39:39 -08004664static struct md_sysfs_entry
4665raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08004666
NeilBrown007583c2005-11-08 21:39:30 -08004667static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08004668 &raid5_stripecache_size.attr,
4669 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004670 &raid5_preread_bypass_threshold.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08004671 NULL,
4672};
NeilBrown007583c2005-11-08 21:39:30 -08004673static struct attribute_group raid5_attrs_group = {
4674 .name = NULL,
4675 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08004676};
4677
Dan Williams80c3a6c2009-03-17 18:10:40 -07004678static sector_t
4679raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4680{
NeilBrown070ec552009-06-16 16:54:21 +10004681 raid5_conf_t *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07004682
4683 if (!sectors)
4684 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004685 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11004686 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11004687 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004688
Andre Noll9d8f0362009-06-18 08:45:01 +10004689 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Andre Noll664e7c42009-06-18 08:45:27 +10004690 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004691 return sectors * (raid_disks - conf->max_degraded);
4692}
4693
Dan Williams36d1c642009-07-14 11:48:22 -07004694static void raid5_free_percpu(raid5_conf_t *conf)
4695{
4696 struct raid5_percpu *percpu;
4697 unsigned long cpu;
4698
4699 if (!conf->percpu)
4700 return;
4701
4702 get_online_cpus();
4703 for_each_possible_cpu(cpu) {
4704 percpu = per_cpu_ptr(conf->percpu, cpu);
4705 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004706 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004707 }
4708#ifdef CONFIG_HOTPLUG_CPU
4709 unregister_cpu_notifier(&conf->cpu_notify);
4710#endif
4711 put_online_cpus();
4712
4713 free_percpu(conf->percpu);
4714}
4715
Dan Williams95fc17a2009-07-31 12:39:15 +10004716static void free_conf(raid5_conf_t *conf)
4717{
4718 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07004719 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10004720 kfree(conf->disks);
4721 kfree(conf->stripe_hashtbl);
4722 kfree(conf);
4723}
4724
Dan Williams36d1c642009-07-14 11:48:22 -07004725#ifdef CONFIG_HOTPLUG_CPU
4726static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4727 void *hcpu)
4728{
4729 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4730 long cpu = (long)hcpu;
4731 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4732
4733 switch (action) {
4734 case CPU_UP_PREPARE:
4735 case CPU_UP_PREPARE_FROZEN:
Dan Williamsd6f38f32009-07-14 11:50:52 -07004736 if (conf->level == 6 && !percpu->spare_page)
Dan Williams36d1c642009-07-14 11:48:22 -07004737 percpu->spare_page = alloc_page(GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004738 if (!percpu->scribble)
4739 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4740
4741 if (!percpu->scribble ||
4742 (conf->level == 6 && !percpu->spare_page)) {
4743 safe_put_page(percpu->spare_page);
4744 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004745 pr_err("%s: failed memory allocation for cpu%ld\n",
4746 __func__, cpu);
Akinobu Mita55af6bb2010-05-26 14:43:35 -07004747 return notifier_from_errno(-ENOMEM);
Dan Williams36d1c642009-07-14 11:48:22 -07004748 }
4749 break;
4750 case CPU_DEAD:
4751 case CPU_DEAD_FROZEN:
4752 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004753 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004754 percpu->spare_page = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004755 percpu->scribble = NULL;
Dan Williams36d1c642009-07-14 11:48:22 -07004756 break;
4757 default:
4758 break;
4759 }
4760 return NOTIFY_OK;
4761}
4762#endif
4763
4764static int raid5_alloc_percpu(raid5_conf_t *conf)
4765{
4766 unsigned long cpu;
4767 struct page *spare_page;
Tejun Heoa29d8b82010-02-02 14:39:15 +09004768 struct raid5_percpu __percpu *allcpus;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004769 void *scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004770 int err;
4771
Dan Williams36d1c642009-07-14 11:48:22 -07004772 allcpus = alloc_percpu(struct raid5_percpu);
4773 if (!allcpus)
4774 return -ENOMEM;
4775 conf->percpu = allcpus;
4776
4777 get_online_cpus();
4778 err = 0;
4779 for_each_present_cpu(cpu) {
Dan Williamsd6f38f32009-07-14 11:50:52 -07004780 if (conf->level == 6) {
4781 spare_page = alloc_page(GFP_KERNEL);
4782 if (!spare_page) {
4783 err = -ENOMEM;
4784 break;
4785 }
4786 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4787 }
NeilBrown5e5e3e72009-10-16 16:35:30 +11004788 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004789 if (!scribble) {
Dan Williams36d1c642009-07-14 11:48:22 -07004790 err = -ENOMEM;
4791 break;
4792 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07004793 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004794 }
4795#ifdef CONFIG_HOTPLUG_CPU
4796 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4797 conf->cpu_notify.priority = 0;
4798 if (err == 0)
4799 err = register_cpu_notifier(&conf->cpu_notify);
4800#endif
4801 put_online_cpus();
4802
4803 return err;
4804}
4805
NeilBrown91adb562009-03-31 14:39:39 +11004806static raid5_conf_t *setup_conf(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004807{
4808 raid5_conf_t *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004809 int raid_disk, memory, max_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004810 mdk_rdev_t *rdev;
4811 struct disk_info *disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004812
NeilBrown91adb562009-03-31 14:39:39 +11004813 if (mddev->new_level != 5
4814 && mddev->new_level != 4
4815 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10004816 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004817 mdname(mddev), mddev->new_level);
4818 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004819 }
NeilBrown91adb562009-03-31 14:39:39 +11004820 if ((mddev->new_level == 5
4821 && !algorithm_valid_raid5(mddev->new_layout)) ||
4822 (mddev->new_level == 6
4823 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004824 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11004825 mdname(mddev), mddev->new_layout);
4826 return ERR_PTR(-EIO);
4827 }
4828 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10004829 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004830 mdname(mddev), mddev->raid_disks);
4831 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11004832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004833
Andre Noll664e7c42009-06-18 08:45:27 +10004834 if (!mddev->new_chunk_sectors ||
4835 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4836 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10004837 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4838 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11004839 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11004840 }
4841
NeilBrown91adb562009-03-31 14:39:39 +11004842 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4843 if (conf == NULL)
4844 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11004845 spin_lock_init(&conf->device_lock);
4846 init_waitqueue_head(&conf->wait_for_stripe);
4847 init_waitqueue_head(&conf->wait_for_overlap);
4848 INIT_LIST_HEAD(&conf->handle_list);
4849 INIT_LIST_HEAD(&conf->hold_list);
4850 INIT_LIST_HEAD(&conf->delayed_list);
4851 INIT_LIST_HEAD(&conf->bitmap_list);
4852 INIT_LIST_HEAD(&conf->inactive_list);
4853 atomic_set(&conf->active_stripes, 0);
4854 atomic_set(&conf->preread_active_stripes, 0);
4855 atomic_set(&conf->active_aligned_reads, 0);
4856 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrown91adb562009-03-31 14:39:39 +11004857
4858 conf->raid_disks = mddev->raid_disks;
4859 if (mddev->reshape_position == MaxSector)
4860 conf->previous_raid_disks = mddev->raid_disks;
4861 else
4862 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004863 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4864 conf->scribble_len = scribble_len(max_disks);
NeilBrown91adb562009-03-31 14:39:39 +11004865
NeilBrown5e5e3e72009-10-16 16:35:30 +11004866 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11004867 GFP_KERNEL);
4868 if (!conf->disks)
4869 goto abort;
4870
4871 conf->mddev = mddev;
4872
4873 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
4874 goto abort;
4875
Dan Williams36d1c642009-07-14 11:48:22 -07004876 conf->level = mddev->new_level;
4877 if (raid5_alloc_percpu(conf) != 0)
4878 goto abort;
4879
NeilBrown0c55e022010-05-03 14:09:02 +10004880 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11004881
4882 list_for_each_entry(rdev, &mddev->disks, same_set) {
4883 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004884 if (raid_disk >= max_disks
NeilBrown91adb562009-03-31 14:39:39 +11004885 || raid_disk < 0)
4886 continue;
4887 disk = conf->disks + raid_disk;
4888
4889 disk->rdev = rdev;
4890
4891 if (test_bit(In_sync, &rdev->flags)) {
4892 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10004893 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4894 " disk %d\n",
4895 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
NeilBrown91adb562009-03-31 14:39:39 +11004896 } else
4897 /* Cannot rely on bitmap to complete recovery */
4898 conf->fullsync = 1;
4899 }
4900
Andre Noll09c9e5f2009-06-18 08:45:55 +10004901 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown91adb562009-03-31 14:39:39 +11004902 conf->level = mddev->new_level;
4903 if (conf->level == 6)
4904 conf->max_degraded = 2;
4905 else
4906 conf->max_degraded = 1;
4907 conf->algorithm = mddev->new_layout;
4908 conf->max_nr_stripes = NR_STRIPES;
NeilBrownfef9c612009-03-31 15:16:46 +11004909 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11004910 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10004911 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11004912 conf->prev_algo = mddev->layout;
4913 }
NeilBrown91adb562009-03-31 14:39:39 +11004914
4915 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11004916 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
NeilBrown91adb562009-03-31 14:39:39 +11004917 if (grow_stripes(conf, conf->max_nr_stripes)) {
4918 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004919 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4920 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004921 goto abort;
4922 } else
NeilBrown0c55e022010-05-03 14:09:02 +10004923 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4924 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004925
NeilBrown0da3c612009-09-23 18:09:45 +10004926 conf->thread = md_register_thread(raid5d, mddev, NULL);
NeilBrown91adb562009-03-31 14:39:39 +11004927 if (!conf->thread) {
4928 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004929 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11004930 mdname(mddev));
4931 goto abort;
4932 }
4933
4934 return conf;
4935
4936 abort:
4937 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10004938 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11004939 return ERR_PTR(-EIO);
4940 } else
4941 return ERR_PTR(-ENOMEM);
4942}
4943
NeilBrownc148ffd2009-11-13 17:47:00 +11004944
4945static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4946{
4947 switch (algo) {
4948 case ALGORITHM_PARITY_0:
4949 if (raid_disk < max_degraded)
4950 return 1;
4951 break;
4952 case ALGORITHM_PARITY_N:
4953 if (raid_disk >= raid_disks - max_degraded)
4954 return 1;
4955 break;
4956 case ALGORITHM_PARITY_0_6:
4957 if (raid_disk == 0 ||
4958 raid_disk == raid_disks - 1)
4959 return 1;
4960 break;
4961 case ALGORITHM_LEFT_ASYMMETRIC_6:
4962 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4963 case ALGORITHM_LEFT_SYMMETRIC_6:
4964 case ALGORITHM_RIGHT_SYMMETRIC_6:
4965 if (raid_disk == raid_disks - 1)
4966 return 1;
4967 }
4968 return 0;
4969}
4970
NeilBrown91adb562009-03-31 14:39:39 +11004971static int run(mddev_t *mddev)
4972{
4973 raid5_conf_t *conf;
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10004974 int working_disks = 0, chunk_size;
NeilBrownc148ffd2009-11-13 17:47:00 +11004975 int dirty_parity_disks = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004976 mdk_rdev_t *rdev;
NeilBrownc148ffd2009-11-13 17:47:00 +11004977 sector_t reshape_offset = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004978
Andre Noll8c6ac862009-06-18 08:48:06 +10004979 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10004980 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10004981 " -- starting background reconstruction\n",
4982 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004983 if (mddev->reshape_position != MaxSector) {
4984 /* Check that we can continue the reshape.
4985 * Currently only disks can change, it must
4986 * increase, and we must be past the point where
4987 * a stripe over-writes itself
4988 */
4989 sector_t here_new, here_old;
4990 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11004991 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrownf6705572006-03-27 01:18:11 -08004992
NeilBrown88ce4932009-03-31 15:24:23 +11004993 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10004994 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08004995 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08004996 mdname(mddev));
4997 return -EINVAL;
4998 }
NeilBrownf6705572006-03-27 01:18:11 -08004999 old_disks = mddev->raid_disks - mddev->delta_disks;
5000 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08005001 * further up in new geometry must map after here in old
5002 * geometry.
NeilBrownf6705572006-03-27 01:18:11 -08005003 */
5004 here_new = mddev->reshape_position;
Andre Noll664e7c42009-06-18 08:45:27 +10005005 if (sector_div(here_new, mddev->new_chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08005006 (mddev->raid_disks - max_degraded))) {
NeilBrown0c55e022010-05-03 14:09:02 +10005007 printk(KERN_ERR "md/raid:%s: reshape_position not "
5008 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005009 return -EINVAL;
5010 }
NeilBrownc148ffd2009-11-13 17:47:00 +11005011 reshape_offset = here_new * mddev->new_chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08005012 /* here_new is the stripe we will write to */
5013 here_old = mddev->reshape_position;
Andre Noll9d8f0362009-06-18 08:45:01 +10005014 sector_div(here_old, mddev->chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08005015 (old_disks-max_degraded));
5016 /* here_old is the first stripe that we might need to read
5017 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10005018 if (mddev->delta_disks == 0) {
5019 /* We cannot be sure it is safe to start an in-place
5020 * reshape. It is only safe if user-space if monitoring
5021 * and taking constant backups.
5022 * mdadm always starts a situation like this in
5023 * readonly mode so it can take control before
5024 * allowing any writes. So just check for that.
5025 */
5026 if ((here_new * mddev->new_chunk_sectors !=
5027 here_old * mddev->chunk_sectors) ||
5028 mddev->ro == 0) {
NeilBrown0c55e022010-05-03 14:09:02 +10005029 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
5030 " in read-only mode - aborting\n",
5031 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10005032 return -EINVAL;
5033 }
5034 } else if (mddev->delta_disks < 0
5035 ? (here_new * mddev->new_chunk_sectors <=
5036 here_old * mddev->chunk_sectors)
5037 : (here_new * mddev->new_chunk_sectors >=
5038 here_old * mddev->chunk_sectors)) {
NeilBrownf6705572006-03-27 01:18:11 -08005039 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10005040 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
5041 "auto-recovery - aborting.\n",
5042 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005043 return -EINVAL;
5044 }
NeilBrown0c55e022010-05-03 14:09:02 +10005045 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
5046 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08005047 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08005048 } else {
NeilBrown91adb562009-03-31 14:39:39 +11005049 BUG_ON(mddev->level != mddev->new_level);
5050 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10005051 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11005052 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08005053 }
5054
NeilBrown245f46c2009-03-31 14:39:39 +11005055 if (mddev->private == NULL)
5056 conf = setup_conf(mddev);
5057 else
5058 conf = mddev->private;
5059
NeilBrown91adb562009-03-31 14:39:39 +11005060 if (IS_ERR(conf))
5061 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08005062
NeilBrown91adb562009-03-31 14:39:39 +11005063 mddev->thread = conf->thread;
5064 conf->thread = NULL;
5065 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005066
Linus Torvalds1da177e2005-04-16 15:20:36 -07005067 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005068 * 0 for a fully functional array, 1 or 2 for a degraded array.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005069 */
NeilBrownc148ffd2009-11-13 17:47:00 +11005070 list_for_each_entry(rdev, &mddev->disks, same_set) {
5071 if (rdev->raid_disk < 0)
5072 continue;
NeilBrown2f115882010-06-17 17:41:03 +10005073 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown91adb562009-03-31 14:39:39 +11005074 working_disks++;
NeilBrown2f115882010-06-17 17:41:03 +10005075 continue;
5076 }
NeilBrownc148ffd2009-11-13 17:47:00 +11005077 /* This disc is not fully in-sync. However if it
5078 * just stored parity (beyond the recovery_offset),
5079 * when we don't need to be concerned about the
5080 * array being dirty.
5081 * When reshape goes 'backwards', we never have
5082 * partially completed devices, so we only need
5083 * to worry about reshape going forwards.
5084 */
5085 /* Hack because v0.91 doesn't store recovery_offset properly. */
5086 if (mddev->major_version == 0 &&
5087 mddev->minor_version > 90)
5088 rdev->recovery_offset = reshape_offset;
5089
NeilBrownc148ffd2009-11-13 17:47:00 +11005090 if (rdev->recovery_offset < reshape_offset) {
5091 /* We need to check old and new layout */
5092 if (!only_parity(rdev->raid_disk,
5093 conf->algorithm,
5094 conf->raid_disks,
5095 conf->max_degraded))
5096 continue;
5097 }
5098 if (!only_parity(rdev->raid_disk,
5099 conf->prev_algo,
5100 conf->previous_raid_disks,
5101 conf->max_degraded))
5102 continue;
5103 dirty_parity_disks++;
5104 }
NeilBrown91adb562009-03-31 14:39:39 +11005105
NeilBrown5e5e3e72009-10-16 16:35:30 +11005106 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
5107 - working_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005108
NeilBrown674806d2010-06-16 17:17:53 +10005109 if (has_failed(conf)) {
NeilBrown0c55e022010-05-03 14:09:02 +10005110 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07005111 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07005112 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005113 goto abort;
5114 }
5115
NeilBrown91adb562009-03-31 14:39:39 +11005116 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10005117 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11005118 mddev->resync_max_sectors = mddev->dev_sectors;
5119
NeilBrownc148ffd2009-11-13 17:47:00 +11005120 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07005121 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005122 if (mddev->ok_start_degraded)
5123 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10005124 "md/raid:%s: starting dirty degraded array"
5125 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005126 mdname(mddev));
5127 else {
5128 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005129 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005130 mdname(mddev));
5131 goto abort;
5132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005133 }
5134
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10005136 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5137 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11005138 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5139 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005140 else
NeilBrown0c55e022010-05-03 14:09:02 +10005141 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5142 " out of %d devices, algorithm %d\n",
5143 mdname(mddev), conf->level,
5144 mddev->raid_disks - mddev->degraded,
5145 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005146
5147 print_raid5_conf(conf);
5148
NeilBrownfef9c612009-03-31 15:16:46 +11005149 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11005150 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08005151 atomic_set(&conf->reshape_stripes, 0);
5152 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5153 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5154 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5155 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5156 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005157 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08005158 }
5159
Linus Torvalds1da177e2005-04-16 15:20:36 -07005160 /* read-ahead size must cover two whole stripes, which is
NeilBrown16a53ec2006-06-26 00:27:38 -07005161 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
Linus Torvalds1da177e2005-04-16 15:20:36 -07005162 */
5163 {
NeilBrown16a53ec2006-06-26 00:27:38 -07005164 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5165 int stripe = data_disks *
Andre Noll9d8f0362009-06-18 08:45:01 +10005166 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005167 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5168 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5169 }
5170
5171 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10005172 if (mddev->to_remove == &raid5_attrs_group)
5173 mddev->to_remove = NULL;
NeilBrown00bcb4a2010-06-01 19:37:23 +10005174 else if (mddev->kobj.sd &&
5175 sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08005176 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10005177 "md/raid:%s: failed to create sysfs attributes.\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08005178 mdname(mddev));
NeilBrown7a5febe2005-05-16 21:53:16 -07005179
NeilBrown91adb562009-03-31 14:39:39 +11005180 mddev->queue->queue_lock = &conf->device_lock;
5181
NeilBrown7a5febe2005-05-16 21:53:16 -07005182 mddev->queue->unplug_fn = raid5_unplug_device;
NeilBrownf022b2f2006-10-03 01:15:56 -07005183 mddev->queue->backing_dev_info.congested_data = mddev;
NeilBrown041ae522007-03-26 21:32:14 -08005184 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
NeilBrownf022b2f2006-10-03 01:15:56 -07005185
Dan Williams1f403622009-03-31 14:59:03 +11005186 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
NeilBrown7a5febe2005-05-16 21:53:16 -07005187
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08005188 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10005189 chunk_size = mddev->chunk_sectors << 9;
5190 blk_queue_io_min(mddev->queue, chunk_size);
5191 blk_queue_io_opt(mddev->queue, chunk_size *
5192 (conf->raid_disks - conf->max_degraded));
5193
5194 list_for_each_entry(rdev, &mddev->disks, same_set)
5195 disk_stack_limits(mddev->gendisk, rdev->bdev,
5196 rdev->data_offset << 9);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08005197
Linus Torvalds1da177e2005-04-16 15:20:36 -07005198 return 0;
5199abort:
NeilBrowne0cf8f02009-03-31 14:39:39 +11005200 md_unregister_thread(mddev->thread);
NeilBrown91adb562009-03-31 14:39:39 +11005201 mddev->thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005202 if (conf) {
5203 print_raid5_conf(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005204 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005205 }
5206 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10005207 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005208 return -EIO;
5209}
5210
NeilBrown3f294f42005-11-08 21:39:25 -08005211static int stop(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005212{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005213 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005214
5215 md_unregister_thread(mddev->thread);
5216 mddev->thread = NULL;
NeilBrown041ae522007-03-26 21:32:14 -08005217 mddev->queue->backing_dev_info.congested_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005218 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
Dan Williams95fc17a2009-07-31 12:39:15 +10005219 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10005220 mddev->private = NULL;
5221 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005222 return 0;
5223}
5224
Dan Williams45b42332007-07-09 11:56:43 -07005225#ifdef DEBUG
NeilBrownd710e132008-10-13 11:55:12 +11005226static void print_sh(struct seq_file *seq, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005227{
5228 int i;
5229
NeilBrown16a53ec2006-06-26 00:27:38 -07005230 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
5231 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
5232 seq_printf(seq, "sh %llu, count %d.\n",
5233 (unsigned long long)sh->sector, atomic_read(&sh->count));
5234 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005235 for (i = 0; i < sh->disks; i++) {
NeilBrown16a53ec2006-06-26 00:27:38 -07005236 seq_printf(seq, "(cache%d: %p %ld) ",
5237 i, sh->dev[i].page, sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005238 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005239 seq_printf(seq, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005240}
5241
NeilBrownd710e132008-10-13 11:55:12 +11005242static void printall(struct seq_file *seq, raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005243{
5244 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -08005245 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005246 int i;
5247
5248 spin_lock_irq(&conf->device_lock);
5249 for (i = 0; i < NR_HASH; i++) {
NeilBrownfccddba2006-01-06 00:20:33 -08005250 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005251 if (sh->raid_conf != conf)
5252 continue;
NeilBrown16a53ec2006-06-26 00:27:38 -07005253 print_sh(seq, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005254 }
5255 }
5256 spin_unlock_irq(&conf->device_lock);
5257}
5258#endif
5259
NeilBrownd710e132008-10-13 11:55:12 +11005260static void status(struct seq_file *seq, mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005261{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005262 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005263 int i;
5264
Andre Noll9d8f0362009-06-18 08:45:01 +10005265 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5266 mddev->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07005267 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005268 for (i = 0; i < conf->raid_disks; i++)
5269 seq_printf (seq, "%s",
5270 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08005271 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005272 seq_printf (seq, "]");
Dan Williams45b42332007-07-09 11:56:43 -07005273#ifdef DEBUG
NeilBrown16a53ec2006-06-26 00:27:38 -07005274 seq_printf (seq, "\n");
5275 printall(seq, conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005276#endif
5277}
5278
5279static void print_raid5_conf (raid5_conf_t *conf)
5280{
5281 int i;
5282 struct disk_info *tmp;
5283
NeilBrown0c55e022010-05-03 14:09:02 +10005284 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005285 if (!conf) {
5286 printk("(conf==NULL)\n");
5287 return;
5288 }
NeilBrown0c55e022010-05-03 14:09:02 +10005289 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5290 conf->raid_disks,
5291 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005292
5293 for (i = 0; i < conf->raid_disks; i++) {
5294 char b[BDEVNAME_SIZE];
5295 tmp = conf->disks + i;
5296 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10005297 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5298 i, !test_bit(Faulty, &tmp->rdev->flags),
5299 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005300 }
5301}
5302
5303static int raid5_spare_active(mddev_t *mddev)
5304{
5305 int i;
5306 raid5_conf_t *conf = mddev->private;
5307 struct disk_info *tmp;
5308
5309 for (i = 0; i < conf->raid_disks; i++) {
5310 tmp = conf->disks + i;
5311 if (tmp->rdev
NeilBrown70fffd02010-06-16 17:01:25 +10005312 && tmp->rdev->recovery_offset == MaxSector
NeilBrownb2d444d2005-11-08 21:39:31 -08005313 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07005314 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
5315 unsigned long flags;
5316 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005317 mddev->degraded--;
NeilBrownc04be0a2006-10-03 01:15:53 -07005318 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005319 }
5320 }
5321 print_raid5_conf(conf);
5322 return 0;
5323}
5324
5325static int raid5_remove_disk(mddev_t *mddev, int number)
5326{
5327 raid5_conf_t *conf = mddev->private;
5328 int err = 0;
5329 mdk_rdev_t *rdev;
5330 struct disk_info *p = conf->disks + number;
5331
5332 print_raid5_conf(conf);
5333 rdev = p->rdev;
5334 if (rdev) {
NeilBrownec32a2b2009-03-31 15:17:38 +11005335 if (number >= conf->raid_disks &&
5336 conf->reshape_progress == MaxSector)
5337 clear_bit(In_sync, &rdev->flags);
5338
NeilBrownb2d444d2005-11-08 21:39:31 -08005339 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07005340 atomic_read(&rdev->nr_pending)) {
5341 err = -EBUSY;
5342 goto abort;
5343 }
NeilBrowndfc70642008-05-23 13:04:39 -07005344 /* Only remove non-faulty devices if recovery
5345 * isn't possible.
5346 */
5347 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown674806d2010-06-16 17:17:53 +10005348 !has_failed(conf) &&
NeilBrownec32a2b2009-03-31 15:17:38 +11005349 number < conf->raid_disks) {
NeilBrowndfc70642008-05-23 13:04:39 -07005350 err = -EBUSY;
5351 goto abort;
5352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005353 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07005354 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005355 if (atomic_read(&rdev->nr_pending)) {
5356 /* lost the race, try later */
5357 err = -EBUSY;
5358 p->rdev = rdev;
5359 }
5360 }
5361abort:
5362
5363 print_raid5_conf(conf);
5364 return err;
5365}
5366
5367static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
5368{
5369 raid5_conf_t *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10005370 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005371 int disk;
5372 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10005373 int first = 0;
5374 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005375
NeilBrown674806d2010-06-16 17:17:53 +10005376 if (has_failed(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005377 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10005378 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005379
Neil Brown6c2fce22008-06-28 08:31:31 +10005380 if (rdev->raid_disk >= 0)
5381 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005382
5383 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005384 * find the disk ... but prefer rdev->saved_raid_disk
5385 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005386 */
NeilBrown16a53ec2006-06-26 00:27:38 -07005387 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10005388 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07005389 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5390 disk = rdev->saved_raid_disk;
5391 else
Neil Brown6c2fce22008-06-28 08:31:31 +10005392 disk = first;
5393 for ( ; disk <= last ; disk++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005394 if ((p=conf->disks + disk)->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08005395 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005396 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10005397 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07005398 if (rdev->saved_raid_disk != disk)
5399 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08005400 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005401 break;
5402 }
5403 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10005404 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005405}
5406
5407static int raid5_resize(mddev_t *mddev, sector_t sectors)
5408{
5409 /* no resync is happening, and there is enough space
5410 * on all devices, so we can resize.
5411 * We need to make sure resync covers any new space.
5412 * If the array is shrinking we should possibly wait until
5413 * any io in the removed space completes, but it hardly seems
5414 * worth it.
5415 */
Andre Noll9d8f0362009-06-18 08:45:01 +10005416 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Dan Williams1f403622009-03-31 14:59:03 +11005417 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5418 mddev->raid_disks));
Dan Williamsb522adc2009-03-31 15:00:31 +11005419 if (mddev->array_sectors >
5420 raid5_size(mddev, sectors, mddev->raid_disks))
5421 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10005422 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005423 revalidate_disk(mddev->gendisk);
Andre Noll58c0fed2009-03-31 14:33:13 +11005424 if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) {
5425 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005426 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5427 }
Andre Noll58c0fed2009-03-31 14:33:13 +11005428 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07005429 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005430 return 0;
5431}
5432
NeilBrown01ee22b2009-06-18 08:47:20 +10005433static int check_stripe_cache(mddev_t *mddev)
5434{
5435 /* Can only proceed if there are plenty of stripe_heads.
5436 * We need a minimum of one full stripe,, and for sensible progress
5437 * it is best to have about 4 times that.
5438 * If we require 4 times, then the default 256 4K stripe_heads will
5439 * allow for chunk sizes up to 256K, which is probably OK.
5440 * If the chunk size is greater, user-space should request more
5441 * stripe_heads first.
5442 */
5443 raid5_conf_t *conf = mddev->private;
5444 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5445 > conf->max_nr_stripes ||
5446 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5447 > conf->max_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10005448 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5449 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10005450 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5451 / STRIPE_SIZE)*4);
5452 return 0;
5453 }
5454 return 1;
5455}
5456
NeilBrown50ac1682009-06-18 08:47:55 +10005457static int check_reshape(mddev_t *mddev)
NeilBrown29269552006-03-27 01:18:10 -08005458{
NeilBrown070ec552009-06-16 16:54:21 +10005459 raid5_conf_t *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08005460
NeilBrown88ce4932009-03-31 15:24:23 +11005461 if (mddev->delta_disks == 0 &&
5462 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10005463 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10005464 return 0; /* nothing to do */
NeilBrowndba034e2008-08-05 15:54:13 +10005465 if (mddev->bitmap)
5466 /* Cannot grow a bitmap yet */
5467 return -EBUSY;
NeilBrown674806d2010-06-16 17:17:53 +10005468 if (has_failed(conf))
NeilBrownec32a2b2009-03-31 15:17:38 +11005469 return -EINVAL;
5470 if (mddev->delta_disks < 0) {
5471 /* We might be able to shrink, but the devices must
5472 * be made bigger first.
5473 * For raid6, 4 is the minimum size.
5474 * Otherwise 2 is the minimum
5475 */
5476 int min = 2;
5477 if (mddev->level == 6)
5478 min = 4;
5479 if (mddev->raid_disks + mddev->delta_disks < min)
5480 return -EINVAL;
5481 }
NeilBrown29269552006-03-27 01:18:10 -08005482
NeilBrown01ee22b2009-06-18 08:47:20 +10005483 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08005484 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08005485
NeilBrownec32a2b2009-03-31 15:17:38 +11005486 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
NeilBrown63c70c42006-03-27 01:18:13 -08005487}
5488
5489static int raid5_start_reshape(mddev_t *mddev)
5490{
NeilBrown070ec552009-06-16 16:54:21 +10005491 raid5_conf_t *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08005492 mdk_rdev_t *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08005493 int spares = 0;
5494 int added_devices = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07005495 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08005496
NeilBrownf4168852007-02-28 20:11:53 -08005497 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08005498 return -EBUSY;
5499
NeilBrown01ee22b2009-06-18 08:47:20 +10005500 if (!check_stripe_cache(mddev))
5501 return -ENOSPC;
5502
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005503 list_for_each_entry(rdev, &mddev->disks, same_set)
NeilBrown29269552006-03-27 01:18:10 -08005504 if (rdev->raid_disk < 0 &&
5505 !test_bit(Faulty, &rdev->flags))
5506 spares++;
NeilBrown63c70c42006-03-27 01:18:13 -08005507
NeilBrownf4168852007-02-28 20:11:53 -08005508 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08005509 /* Not enough devices even to make a degraded array
5510 * of that size
5511 */
5512 return -EINVAL;
5513
NeilBrownec32a2b2009-03-31 15:17:38 +11005514 /* Refuse to reduce size of the array. Any reductions in
5515 * array size must be through explicit setting of array_size
5516 * attribute.
5517 */
5518 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5519 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10005520 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11005521 "before number of disks\n", mdname(mddev));
5522 return -EINVAL;
5523 }
5524
NeilBrownf6705572006-03-27 01:18:11 -08005525 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08005526 spin_lock_irq(&conf->device_lock);
5527 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08005528 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005529 conf->prev_chunk_sectors = conf->chunk_sectors;
5530 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11005531 conf->prev_algo = conf->algorithm;
5532 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11005533 if (mddev->delta_disks < 0)
5534 conf->reshape_progress = raid5_size(mddev, 0, 0);
5535 else
5536 conf->reshape_progress = 0;
5537 conf->reshape_safe = conf->reshape_progress;
NeilBrown86b42c72009-03-31 15:19:03 +11005538 conf->generation++;
NeilBrown29269552006-03-27 01:18:10 -08005539 spin_unlock_irq(&conf->device_lock);
5540
5541 /* Add some new drives, as many as will fit.
5542 * We know there are enough to make the newly sized array work.
NeilBrown3424bf62010-06-17 17:48:26 +10005543 * Don't add devices if we are reducing the number of
5544 * devices in the array. This is because it is not possible
5545 * to correctly record the "partially reconstructed" state of
5546 * such devices during the reshape and confusion could result.
NeilBrown29269552006-03-27 01:18:10 -08005547 */
NeilBrown3424bf62010-06-17 17:48:26 +10005548 if (mddev->delta_disks >= 0)
5549 list_for_each_entry(rdev, &mddev->disks, same_set)
NeilBrown29269552006-03-27 01:18:10 -08005550 if (rdev->raid_disk < 0 &&
5551 !test_bit(Faulty, &rdev->flags)) {
Neil Brown199050e2008-06-28 08:31:33 +10005552 if (raid5_add_disk(mddev, rdev) == 0) {
NeilBrown29269552006-03-27 01:18:10 -08005553 char nm[20];
NeilBrown9eb07c22010-02-09 12:31:47 +11005554 if (rdev->raid_disk >= conf->previous_raid_disks) {
NeilBrown7ef90142009-11-13 17:40:51 +11005555 set_bit(In_sync, &rdev->flags);
NeilBrown9eb07c22010-02-09 12:31:47 +11005556 added_devices++;
5557 } else
NeilBrown7ef90142009-11-13 17:40:51 +11005558 rdev->recovery_offset = 0;
NeilBrown29269552006-03-27 01:18:10 -08005559 sprintf(nm, "rd%d", rdev->raid_disk);
NeilBrown5e55e2f2007-03-26 21:32:14 -08005560 if (sysfs_create_link(&mddev->kobj,
5561 &rdev->kobj, nm))
NeilBrown00bcb4a2010-06-01 19:37:23 +10005562 /* Failure here is OK */;
NeilBrown29269552006-03-27 01:18:10 -08005563 } else
5564 break;
5565 }
5566
NeilBrown9eb07c22010-02-09 12:31:47 +11005567 /* When a reshape changes the number of devices, ->degraded
NeilBrown3424bf62010-06-17 17:48:26 +10005568 * is measured against the larger of the pre and post number of
NeilBrown9eb07c22010-02-09 12:31:47 +11005569 * devices.*/
NeilBrownec32a2b2009-03-31 15:17:38 +11005570 if (mddev->delta_disks > 0) {
5571 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown9eb07c22010-02-09 12:31:47 +11005572 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
NeilBrownec32a2b2009-03-31 15:17:38 +11005573 - added_devices;
5574 spin_unlock_irqrestore(&conf->device_lock, flags);
5575 }
NeilBrown63c70c42006-03-27 01:18:13 -08005576 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10005577 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07005578 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08005579
NeilBrown29269552006-03-27 01:18:10 -08005580 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5581 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5582 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5583 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5584 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005585 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08005586 if (!mddev->sync_thread) {
5587 mddev->recovery = 0;
5588 spin_lock_irq(&conf->device_lock);
5589 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005590 conf->reshape_progress = MaxSector;
NeilBrown29269552006-03-27 01:18:10 -08005591 spin_unlock_irq(&conf->device_lock);
5592 return -EAGAIN;
5593 }
NeilBrownc8f517c2009-03-31 15:28:40 +11005594 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08005595 md_wakeup_thread(mddev->sync_thread);
5596 md_new_event(mddev);
5597 return 0;
5598}
NeilBrown29269552006-03-27 01:18:10 -08005599
NeilBrownec32a2b2009-03-31 15:17:38 +11005600/* This is called from the reshape thread and should make any
5601 * changes needed in 'conf'
5602 */
NeilBrown29269552006-03-27 01:18:10 -08005603static void end_reshape(raid5_conf_t *conf)
5604{
NeilBrown29269552006-03-27 01:18:10 -08005605
NeilBrownf6705572006-03-27 01:18:11 -08005606 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
Dan Williams80c3a6c2009-03-17 18:10:40 -07005607
NeilBrownf6705572006-03-27 01:18:11 -08005608 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11005609 conf->previous_raid_disks = conf->raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005610 conf->reshape_progress = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08005611 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11005612 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07005613
5614 /* read-ahead size must cover two whole stripes, which is
5615 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5616 */
5617 {
NeilBrowncea9c222009-03-31 15:15:05 +11005618 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005619 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11005620 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07005621 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5622 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5623 }
NeilBrown29269552006-03-27 01:18:10 -08005624 }
NeilBrown29269552006-03-27 01:18:10 -08005625}
5626
NeilBrownec32a2b2009-03-31 15:17:38 +11005627/* This is called from the raid5d thread with mddev_lock held.
5628 * It makes config changes to the device.
5629 */
NeilBrowncea9c222009-03-31 15:15:05 +11005630static void raid5_finish_reshape(mddev_t *mddev)
5631{
NeilBrown070ec552009-06-16 16:54:21 +10005632 raid5_conf_t *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11005633
5634 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5635
NeilBrownec32a2b2009-03-31 15:17:38 +11005636 if (mddev->delta_disks > 0) {
5637 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5638 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005639 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11005640 } else {
5641 int d;
NeilBrownec32a2b2009-03-31 15:17:38 +11005642 mddev->degraded = conf->raid_disks;
5643 for (d = 0; d < conf->raid_disks ; d++)
5644 if (conf->disks[d].rdev &&
5645 test_bit(In_sync,
5646 &conf->disks[d].rdev->flags))
5647 mddev->degraded--;
5648 for (d = conf->raid_disks ;
5649 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10005650 d++) {
5651 mdk_rdev_t *rdev = conf->disks[d].rdev;
5652 if (rdev && raid5_remove_disk(mddev, d) == 0) {
5653 char nm[20];
5654 sprintf(nm, "rd%d", rdev->raid_disk);
5655 sysfs_remove_link(&mddev->kobj, nm);
5656 rdev->raid_disk = -1;
5657 }
5658 }
NeilBrowncea9c222009-03-31 15:15:05 +11005659 }
NeilBrown88ce4932009-03-31 15:24:23 +11005660 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005661 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11005662 mddev->reshape_position = MaxSector;
5663 mddev->delta_disks = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11005664 }
5665}
5666
NeilBrown72626682005-09-09 16:23:54 -07005667static void raid5_quiesce(mddev_t *mddev, int state)
5668{
NeilBrown070ec552009-06-16 16:54:21 +10005669 raid5_conf_t *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07005670
5671 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08005672 case 2: /* resume for a suspend */
5673 wake_up(&conf->wait_for_overlap);
5674 break;
5675
NeilBrown72626682005-09-09 16:23:54 -07005676 case 1: /* stop all writes */
5677 spin_lock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005678 /* '2' tells resync/reshape to pause so that all
5679 * active stripes can drain
5680 */
5681 conf->quiesce = 2;
NeilBrown72626682005-09-09 16:23:54 -07005682 wait_event_lock_irq(conf->wait_for_stripe,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005683 atomic_read(&conf->active_stripes) == 0 &&
5684 atomic_read(&conf->active_aligned_reads) == 0,
NeilBrown72626682005-09-09 16:23:54 -07005685 conf->device_lock, /* nothing */);
NeilBrown64bd6602009-08-03 10:59:58 +10005686 conf->quiesce = 1;
NeilBrown72626682005-09-09 16:23:54 -07005687 spin_unlock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005688 /* allow reshape to continue */
5689 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005690 break;
5691
5692 case 0: /* re-enable writes */
5693 spin_lock_irq(&conf->device_lock);
5694 conf->quiesce = 0;
5695 wake_up(&conf->wait_for_stripe);
NeilBrowne464eaf2006-03-27 01:18:14 -08005696 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005697 spin_unlock_irq(&conf->device_lock);
5698 break;
5699 }
NeilBrown72626682005-09-09 16:23:54 -07005700}
NeilBrownb15c2e52006-01-06 00:20:16 -08005701
NeilBrownd562b0c2009-03-31 14:39:39 +11005702
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005703static void *raid45_takeover_raid0(mddev_t *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11005704{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005705 struct raid0_private_data *raid0_priv = mddev->private;
Trela Maciej54071b32010-03-08 16:02:42 +11005706
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005707 /* for raid0 takeover only one zone is supported */
5708 if (raid0_priv->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10005709 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5710 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005711 return ERR_PTR(-EINVAL);
5712 }
5713
5714 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11005715 mddev->new_layout = ALGORITHM_PARITY_N;
5716 mddev->new_chunk_sectors = mddev->chunk_sectors;
5717 mddev->raid_disks += 1;
5718 mddev->delta_disks = 1;
5719 /* make sure it will be not marked as dirty */
5720 mddev->recovery_cp = MaxSector;
5721
5722 return setup_conf(mddev);
5723}
5724
5725
NeilBrownd562b0c2009-03-31 14:39:39 +11005726static void *raid5_takeover_raid1(mddev_t *mddev)
5727{
5728 int chunksect;
5729
5730 if (mddev->raid_disks != 2 ||
5731 mddev->degraded > 1)
5732 return ERR_PTR(-EINVAL);
5733
5734 /* Should check if there are write-behind devices? */
5735
5736 chunksect = 64*2; /* 64K by default */
5737
5738 /* The array must be an exact multiple of chunksize */
5739 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5740 chunksect >>= 1;
5741
5742 if ((chunksect<<9) < STRIPE_SIZE)
5743 /* array size does not allow a suitable chunk size */
5744 return ERR_PTR(-EINVAL);
5745
5746 mddev->new_level = 5;
5747 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10005748 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11005749
5750 return setup_conf(mddev);
5751}
5752
NeilBrownfc9739c2009-03-31 14:57:20 +11005753static void *raid5_takeover_raid6(mddev_t *mddev)
5754{
5755 int new_layout;
5756
5757 switch (mddev->layout) {
5758 case ALGORITHM_LEFT_ASYMMETRIC_6:
5759 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5760 break;
5761 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5762 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5763 break;
5764 case ALGORITHM_LEFT_SYMMETRIC_6:
5765 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5766 break;
5767 case ALGORITHM_RIGHT_SYMMETRIC_6:
5768 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5769 break;
5770 case ALGORITHM_PARITY_0_6:
5771 new_layout = ALGORITHM_PARITY_0;
5772 break;
5773 case ALGORITHM_PARITY_N:
5774 new_layout = ALGORITHM_PARITY_N;
5775 break;
5776 default:
5777 return ERR_PTR(-EINVAL);
5778 }
5779 mddev->new_level = 5;
5780 mddev->new_layout = new_layout;
5781 mddev->delta_disks = -1;
5782 mddev->raid_disks -= 1;
5783 return setup_conf(mddev);
5784}
5785
NeilBrownd562b0c2009-03-31 14:39:39 +11005786
NeilBrown50ac1682009-06-18 08:47:55 +10005787static int raid5_check_reshape(mddev_t *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11005788{
NeilBrown88ce4932009-03-31 15:24:23 +11005789 /* For a 2-drive array, the layout and chunk size can be changed
5790 * immediately as not restriping is needed.
5791 * For larger arrays we record the new value - after validation
5792 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11005793 */
NeilBrown070ec552009-06-16 16:54:21 +10005794 raid5_conf_t *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10005795 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11005796
NeilBrown597a7112009-06-18 08:47:42 +10005797 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11005798 return -EINVAL;
5799 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005800 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11005801 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005802 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11005803 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005804 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11005805 /* not factor of array size */
5806 return -EINVAL;
5807 }
5808
5809 /* They look valid */
5810
NeilBrown88ce4932009-03-31 15:24:23 +11005811 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10005812 /* can make the change immediately */
5813 if (mddev->new_layout >= 0) {
5814 conf->algorithm = mddev->new_layout;
5815 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11005816 }
5817 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10005818 conf->chunk_sectors = new_chunk ;
5819 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11005820 }
5821 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5822 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11005823 }
NeilBrown50ac1682009-06-18 08:47:55 +10005824 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11005825}
5826
NeilBrown50ac1682009-06-18 08:47:55 +10005827static int raid6_check_reshape(mddev_t *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11005828{
NeilBrown597a7112009-06-18 08:47:42 +10005829 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10005830
NeilBrown597a7112009-06-18 08:47:42 +10005831 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11005832 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005833 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005834 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11005835 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005836 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11005837 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005838 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11005839 /* not factor of array size */
5840 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005841 }
NeilBrown88ce4932009-03-31 15:24:23 +11005842
5843 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10005844 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11005845}
5846
NeilBrownd562b0c2009-03-31 14:39:39 +11005847static void *raid5_takeover(mddev_t *mddev)
5848{
5849 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005850 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005851 * raid1 - if there are two drives. We need to know the chunk size
5852 * raid4 - trivial - just use a raid4 layout.
5853 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005854 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005855 if (mddev->level == 0)
5856 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11005857 if (mddev->level == 1)
5858 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11005859 if (mddev->level == 4) {
5860 mddev->new_layout = ALGORITHM_PARITY_N;
5861 mddev->new_level = 5;
5862 return setup_conf(mddev);
5863 }
NeilBrownfc9739c2009-03-31 14:57:20 +11005864 if (mddev->level == 6)
5865 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11005866
5867 return ERR_PTR(-EINVAL);
5868}
5869
NeilBrowna78d38a2010-03-22 16:53:49 +11005870static void *raid4_takeover(mddev_t *mddev)
5871{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005872 /* raid4 can take over:
5873 * raid0 - if there is only one strip zone
5874 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11005875 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005876 if (mddev->level == 0)
5877 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11005878 if (mddev->level == 5 &&
5879 mddev->layout == ALGORITHM_PARITY_N) {
5880 mddev->new_layout = 0;
5881 mddev->new_level = 4;
5882 return setup_conf(mddev);
5883 }
5884 return ERR_PTR(-EINVAL);
5885}
NeilBrownd562b0c2009-03-31 14:39:39 +11005886
NeilBrown245f46c2009-03-31 14:39:39 +11005887static struct mdk_personality raid5_personality;
5888
5889static void *raid6_takeover(mddev_t *mddev)
5890{
5891 /* Currently can only take over a raid5. We map the
5892 * personality to an equivalent raid6 personality
5893 * with the Q block at the end.
5894 */
5895 int new_layout;
5896
5897 if (mddev->pers != &raid5_personality)
5898 return ERR_PTR(-EINVAL);
5899 if (mddev->degraded > 1)
5900 return ERR_PTR(-EINVAL);
5901 if (mddev->raid_disks > 253)
5902 return ERR_PTR(-EINVAL);
5903 if (mddev->raid_disks < 3)
5904 return ERR_PTR(-EINVAL);
5905
5906 switch (mddev->layout) {
5907 case ALGORITHM_LEFT_ASYMMETRIC:
5908 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5909 break;
5910 case ALGORITHM_RIGHT_ASYMMETRIC:
5911 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5912 break;
5913 case ALGORITHM_LEFT_SYMMETRIC:
5914 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5915 break;
5916 case ALGORITHM_RIGHT_SYMMETRIC:
5917 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5918 break;
5919 case ALGORITHM_PARITY_0:
5920 new_layout = ALGORITHM_PARITY_0_6;
5921 break;
5922 case ALGORITHM_PARITY_N:
5923 new_layout = ALGORITHM_PARITY_N;
5924 break;
5925 default:
5926 return ERR_PTR(-EINVAL);
5927 }
5928 mddev->new_level = 6;
5929 mddev->new_layout = new_layout;
5930 mddev->delta_disks = 1;
5931 mddev->raid_disks += 1;
5932 return setup_conf(mddev);
5933}
5934
5935
NeilBrown16a53ec2006-06-26 00:27:38 -07005936static struct mdk_personality raid6_personality =
5937{
5938 .name = "raid6",
5939 .level = 6,
5940 .owner = THIS_MODULE,
5941 .make_request = make_request,
5942 .run = run,
5943 .stop = stop,
5944 .status = status,
5945 .error_handler = error,
5946 .hot_add_disk = raid5_add_disk,
5947 .hot_remove_disk= raid5_remove_disk,
5948 .spare_active = raid5_spare_active,
5949 .sync_request = sync_request,
5950 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005951 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10005952 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08005953 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005954 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07005955 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11005956 .takeover = raid6_takeover,
NeilBrown16a53ec2006-06-26 00:27:38 -07005957};
NeilBrown2604b702006-01-06 00:20:36 -08005958static struct mdk_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005959{
5960 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08005961 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005962 .owner = THIS_MODULE,
5963 .make_request = make_request,
5964 .run = run,
5965 .stop = stop,
5966 .status = status,
5967 .error_handler = error,
5968 .hot_add_disk = raid5_add_disk,
5969 .hot_remove_disk= raid5_remove_disk,
5970 .spare_active = raid5_spare_active,
5971 .sync_request = sync_request,
5972 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005973 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08005974 .check_reshape = raid5_check_reshape,
5975 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005976 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07005977 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11005978 .takeover = raid5_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005979};
5980
NeilBrown2604b702006-01-06 00:20:36 -08005981static struct mdk_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005982{
NeilBrown2604b702006-01-06 00:20:36 -08005983 .name = "raid4",
5984 .level = 4,
5985 .owner = THIS_MODULE,
5986 .make_request = make_request,
5987 .run = run,
5988 .stop = stop,
5989 .status = status,
5990 .error_handler = error,
5991 .hot_add_disk = raid5_add_disk,
5992 .hot_remove_disk= raid5_remove_disk,
5993 .spare_active = raid5_spare_active,
5994 .sync_request = sync_request,
5995 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005996 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08005997 .check_reshape = raid5_check_reshape,
5998 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005999 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08006000 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11006001 .takeover = raid4_takeover,
NeilBrown2604b702006-01-06 00:20:36 -08006002};
6003
6004static int __init raid5_init(void)
6005{
NeilBrown16a53ec2006-06-26 00:27:38 -07006006 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08006007 register_md_personality(&raid5_personality);
6008 register_md_personality(&raid4_personality);
6009 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006010}
6011
NeilBrown2604b702006-01-06 00:20:36 -08006012static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006013{
NeilBrown16a53ec2006-06-26 00:27:38 -07006014 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08006015 unregister_md_personality(&raid5_personality);
6016 unregister_md_personality(&raid4_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006017}
6018
6019module_init(raid5_init);
6020module_exit(raid5_exit);
6021MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11006022MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006023MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08006024MODULE_ALIAS("md-raid5");
6025MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08006026MODULE_ALIAS("md-level-5");
6027MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07006028MODULE_ALIAS("md-personality-8"); /* RAID6 */
6029MODULE_ALIAS("md-raid6");
6030MODULE_ALIAS("md-level-6");
6031
6032/* This used to be two separate modules, they were: */
6033MODULE_ALIAS("raid5");
6034MODULE_ALIAS("raid6");