blob: eacf02a6ec5faa66f577ebff692b60867f038c88 [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>
NeilBrown43b2e5d2009-03-31 14:33:13 +110053#include "md.h"
NeilBrownbff61972009-03-31 14:33:13 +110054#include "raid5.h"
Trela Maciej54071b32010-03-08 16:02:42 +110055#include "raid0.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110056#include "bitmap.h"
NeilBrown72626682005-09-09 16:23:54 -070057
Linus Torvalds1da177e2005-04-16 15:20:36 -070058/*
59 * Stripe cache
60 */
61
62#define NR_STRIPES 256
63#define STRIPE_SIZE PAGE_SIZE
64#define STRIPE_SHIFT (PAGE_SHIFT - 9)
65#define STRIPE_SECTORS (STRIPE_SIZE>>9)
66#define IO_THRESHOLD 1
Dan Williams8b3e6cd2008-04-28 02:15:53 -070067#define BYPASS_THRESHOLD 1
NeilBrownfccddba2006-01-06 00:20:33 -080068#define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head))
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#define HASH_MASK (NR_HASH - 1)
70
NeilBrownfccddba2006-01-06 00:20:33 -080071#define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]))
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73/* bio's attached to a stripe+device for I/O are linked together in bi_sector
74 * order without overlap. There may be several bio's per stripe+device, and
75 * a bio could span several devices.
76 * When walking this list for a particular stripe+device, we must never proceed
77 * beyond a bio that extends past this device, as the next bio might no longer
78 * be valid.
79 * This macro is used to determine the 'next' bio in the list, given the sector
80 * of the current stripe+device
81 */
82#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL)
83/*
84 * The following can be used to debug the driver
85 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070086#define RAID5_PARANOIA 1
87#if RAID5_PARANOIA && defined(CONFIG_SMP)
88# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock)
89#else
90# define CHECK_DEVLOCK()
91#endif
92
Dan Williams45b42332007-07-09 11:56:43 -070093#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -070094#define inline
95#define __inline__
96#endif
97
Bernd Schubert6be9d492008-05-23 13:04:34 -070098#define printk_rl(args...) ((void) (printk_ratelimit() && printk(args)))
99
Jens Axboe960e7392008-08-15 10:41:18 +0200100/*
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200101 * We maintain a biased count of active stripes in the bottom 16 bits of
102 * bi_phys_segments, and a count of processed stripes in the upper 16 bits
Jens Axboe960e7392008-08-15 10:41:18 +0200103 */
104static inline int raid5_bi_phys_segments(struct bio *bio)
105{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200106 return bio->bi_phys_segments & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200107}
108
109static inline int raid5_bi_hw_segments(struct bio *bio)
110{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200111 return (bio->bi_phys_segments >> 16) & 0xffff;
Jens Axboe960e7392008-08-15 10:41:18 +0200112}
113
114static inline int raid5_dec_bi_phys_segments(struct bio *bio)
115{
116 --bio->bi_phys_segments;
117 return raid5_bi_phys_segments(bio);
118}
119
120static inline int raid5_dec_bi_hw_segments(struct bio *bio)
121{
122 unsigned short val = raid5_bi_hw_segments(bio);
123
124 --val;
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200125 bio->bi_phys_segments = (val << 16) | raid5_bi_phys_segments(bio);
Jens Axboe960e7392008-08-15 10:41:18 +0200126 return val;
127}
128
129static inline void raid5_set_bi_hw_segments(struct bio *bio, unsigned int cnt)
130{
Jens Axboe5b99c2f2008-08-15 10:56:11 +0200131 bio->bi_phys_segments = raid5_bi_phys_segments(bio) || (cnt << 16);
Jens Axboe960e7392008-08-15 10:41:18 +0200132}
133
NeilBrownd0dabf72009-03-31 14:39:38 +1100134/* Find first data disk in a raid6 stripe */
135static inline int raid6_d0(struct stripe_head *sh)
136{
NeilBrown67cc2b82009-03-31 14:39:38 +1100137 if (sh->ddf_layout)
138 /* ddf always start from first device */
139 return 0;
140 /* md starts just after Q block */
NeilBrownd0dabf72009-03-31 14:39:38 +1100141 if (sh->qd_idx == sh->disks - 1)
142 return 0;
143 else
144 return sh->qd_idx + 1;
145}
NeilBrown16a53ec2006-06-26 00:27:38 -0700146static inline int raid6_next_disk(int disk, int raid_disks)
147{
148 disk++;
149 return (disk < raid_disks) ? disk : 0;
150}
Dan Williamsa4456852007-07-09 11:56:43 -0700151
NeilBrownd0dabf72009-03-31 14:39:38 +1100152/* When walking through the disks in a raid5, starting at raid6_d0,
153 * We need to map each disk to a 'slot', where the data disks are slot
154 * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk
155 * is raid_disks-1. This help does that mapping.
156 */
NeilBrown67cc2b82009-03-31 14:39:38 +1100157static int raid6_idx_to_slot(int idx, struct stripe_head *sh,
158 int *count, int syndrome_disks)
NeilBrownd0dabf72009-03-31 14:39:38 +1100159{
Dan Williams66295422009-10-19 18:09:32 -0700160 int slot = *count;
NeilBrown67cc2b82009-03-31 14:39:38 +1100161
NeilBrowne4424fe2009-10-16 16:27:34 +1100162 if (sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700163 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100164 if (idx == sh->pd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100165 return syndrome_disks;
NeilBrownd0dabf72009-03-31 14:39:38 +1100166 if (idx == sh->qd_idx)
NeilBrown67cc2b82009-03-31 14:39:38 +1100167 return syndrome_disks + 1;
NeilBrowne4424fe2009-10-16 16:27:34 +1100168 if (!sh->ddf_layout)
Dan Williams66295422009-10-19 18:09:32 -0700169 (*count)++;
NeilBrownd0dabf72009-03-31 14:39:38 +1100170 return slot;
171}
172
Dan Williamsa4456852007-07-09 11:56:43 -0700173static void return_io(struct bio *return_bi)
174{
175 struct bio *bi = return_bi;
176 while (bi) {
Dan Williamsa4456852007-07-09 11:56:43 -0700177
178 return_bi = bi->bi_next;
179 bi->bi_next = NULL;
180 bi->bi_size = 0;
Neil Brown0e13fe232008-06-28 08:31:20 +1000181 bio_endio(bi, 0);
Dan Williamsa4456852007-07-09 11:56:43 -0700182 bi = return_bi;
183 }
184}
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186static void print_raid5_conf (raid5_conf_t *conf);
187
Dan Williams600aa102008-06-28 08:32:05 +1000188static int stripe_operations_active(struct stripe_head *sh)
189{
190 return sh->check_state || sh->reconstruct_state ||
191 test_bit(STRIPE_BIOFILL_RUN, &sh->state) ||
192 test_bit(STRIPE_COMPUTE_RUN, &sh->state);
193}
194
Arjan van de Ven858119e2006-01-14 13:20:43 -0800195static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
197 if (atomic_dec_and_test(&sh->count)) {
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200198 BUG_ON(!list_empty(&sh->lru));
199 BUG_ON(atomic_read(&conf->active_stripes)==0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (test_bit(STRIPE_HANDLE, &sh->state)) {
NeilBrown7c785b72006-07-10 04:44:16 -0700201 if (test_bit(STRIPE_DELAYED, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 list_add_tail(&sh->lru, &conf->delayed_list);
NeilBrown7c785b72006-07-10 04:44:16 -0700203 blk_plug_device(conf->mddev->queue);
204 } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) &&
NeilBrownae3c20c2006-07-10 04:44:17 -0700205 sh->bm_seq - conf->seq_write > 0) {
NeilBrown72626682005-09-09 16:23:54 -0700206 list_add_tail(&sh->lru, &conf->bitmap_list);
NeilBrown7c785b72006-07-10 04:44:16 -0700207 blk_plug_device(conf->mddev->queue);
208 } else {
NeilBrown72626682005-09-09 16:23:54 -0700209 clear_bit(STRIPE_BIT_DELAY, &sh->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 list_add_tail(&sh->lru, &conf->handle_list);
NeilBrown72626682005-09-09 16:23:54 -0700211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 md_wakeup_thread(conf->mddev->thread);
213 } else {
Dan Williams600aa102008-06-28 08:32:05 +1000214 BUG_ON(stripe_operations_active(sh));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
216 atomic_dec(&conf->preread_active_stripes);
217 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD)
218 md_wakeup_thread(conf->mddev->thread);
219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 atomic_dec(&conf->active_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800221 if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
222 list_add_tail(&sh->lru, &conf->inactive_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 wake_up(&conf->wait_for_stripe);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -0800224 if (conf->retry_read_aligned)
225 md_wakeup_thread(conf->mddev->thread);
NeilBrownccfcc3c2006-03-27 01:18:09 -0800226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 }
228 }
229}
NeilBrownd0dabf72009-03-31 14:39:38 +1100230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231static void release_stripe(struct stripe_head *sh)
232{
233 raid5_conf_t *conf = sh->raid_conf;
234 unsigned long flags;
NeilBrown16a53ec2006-06-26 00:27:38 -0700235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 spin_lock_irqsave(&conf->device_lock, flags);
237 __release_stripe(conf, sh);
238 spin_unlock_irqrestore(&conf->device_lock, flags);
239}
240
NeilBrownfccddba2006-01-06 00:20:33 -0800241static inline void remove_hash(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
Dan Williams45b42332007-07-09 11:56:43 -0700243 pr_debug("remove_hash(), stripe %llu\n",
244 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
NeilBrownfccddba2006-01-06 00:20:33 -0800246 hlist_del_init(&sh->hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
NeilBrown16a53ec2006-06-26 00:27:38 -0700249static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
NeilBrownfccddba2006-01-06 00:20:33 -0800251 struct hlist_head *hp = stripe_hash(conf, sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Dan Williams45b42332007-07-09 11:56:43 -0700253 pr_debug("insert_hash(), stripe %llu\n",
254 (unsigned long long)sh->sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256 CHECK_DEVLOCK();
NeilBrownfccddba2006-01-06 00:20:33 -0800257 hlist_add_head(&sh->hash, hp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259
260
261/* find an idle stripe, make sure it is unhashed, and return it. */
262static struct stripe_head *get_free_stripe(raid5_conf_t *conf)
263{
264 struct stripe_head *sh = NULL;
265 struct list_head *first;
266
267 CHECK_DEVLOCK();
268 if (list_empty(&conf->inactive_list))
269 goto out;
270 first = conf->inactive_list.next;
271 sh = list_entry(first, struct stripe_head, lru);
272 list_del_init(first);
273 remove_hash(sh);
274 atomic_inc(&conf->active_stripes);
275out:
276 return sh;
277}
278
279static void shrink_buffers(struct stripe_head *sh, int num)
280{
281 struct page *p;
282 int i;
283
284 for (i=0; i<num ; i++) {
285 p = sh->dev[i].page;
286 if (!p)
287 continue;
288 sh->dev[i].page = NULL;
NeilBrown2d1f3b52006-01-06 00:20:31 -0800289 put_page(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 }
291}
292
293static int grow_buffers(struct stripe_head *sh, int num)
294{
295 int i;
296
297 for (i=0; i<num; i++) {
298 struct page *page;
299
300 if (!(page = alloc_page(GFP_KERNEL))) {
301 return 1;
302 }
303 sh->dev[i].page = page;
304 }
305 return 0;
306}
307
NeilBrown784052e2009-03-31 15:19:07 +1100308static void raid5_build_block(struct stripe_head *sh, int i, int previous);
NeilBrown911d4ee2009-03-31 14:39:38 +1100309static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
310 struct stripe_head *sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
NeilBrownb5663ba2009-03-31 14:39:38 +1100312static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
314 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800315 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +0200317 BUG_ON(atomic_read(&sh->count) != 0);
318 BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
Dan Williams600aa102008-06-28 08:32:05 +1000319 BUG_ON(stripe_operations_active(sh));
Dan Williamsd84e0f12007-01-02 13:52:30 -0700320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 CHECK_DEVLOCK();
Dan Williams45b42332007-07-09 11:56:43 -0700322 pr_debug("init_stripe called, stripe %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 (unsigned long long)sh->sector);
324
325 remove_hash(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -0700326
NeilBrown86b42c72009-03-31 15:19:03 +1100327 sh->generation = conf->generation - previous;
NeilBrownb5663ba2009-03-31 14:39:38 +1100328 sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 sh->sector = sector;
NeilBrown911d4ee2009-03-31 14:39:38 +1100330 stripe_set_idx(sector, conf, previous, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 sh->state = 0;
332
NeilBrown7ecaa1e2006-03-27 01:18:08 -0800333
334 for (i = sh->disks; i--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 struct r5dev *dev = &sh->dev[i];
336
Dan Williamsd84e0f12007-01-02 13:52:30 -0700337 if (dev->toread || dev->read || dev->towrite || dev->written ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 test_bit(R5_LOCKED, &dev->flags)) {
Dan Williamsd84e0f12007-01-02 13:52:30 -0700339 printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 (unsigned long long)sh->sector, i, dev->toread,
Dan Williamsd84e0f12007-01-02 13:52:30 -0700341 dev->read, dev->towrite, dev->written,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 test_bit(R5_LOCKED, &dev->flags));
343 BUG();
344 }
345 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +1100346 raid5_build_block(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 }
348 insert_hash(conf, sh);
349}
350
NeilBrown86b42c72009-03-31 15:19:03 +1100351static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector,
352 short generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
354 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -0800355 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
357 CHECK_DEVLOCK();
Dan Williams45b42332007-07-09 11:56:43 -0700358 pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
NeilBrownfccddba2006-01-06 00:20:33 -0800359 hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash)
NeilBrown86b42c72009-03-31 15:19:03 +1100360 if (sh->sector == sector && sh->generation == generation)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 return sh;
Dan Williams45b42332007-07-09 11:56:43 -0700362 pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 return NULL;
364}
365
366static void unplug_slaves(mddev_t *mddev);
Jens Axboe165125e2007-07-24 09:28:11 +0200367static void raid5_unplug_device(struct request_queue *q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
NeilBrownb5663ba2009-03-31 14:39:38 +1100369static struct stripe_head *
370get_active_stripe(raid5_conf_t *conf, sector_t sector,
NeilBrowna8c906c2009-06-09 14:39:59 +1000371 int previous, int noblock, int noquiesce)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
373 struct stripe_head *sh;
374
Dan Williams45b42332007-07-09 11:56:43 -0700375 pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
377 spin_lock_irq(&conf->device_lock);
378
379 do {
NeilBrown72626682005-09-09 16:23:54 -0700380 wait_event_lock_irq(conf->wait_for_stripe,
NeilBrowna8c906c2009-06-09 14:39:59 +1000381 conf->quiesce == 0 || noquiesce,
NeilBrown72626682005-09-09 16:23:54 -0700382 conf->device_lock, /* nothing */);
NeilBrown86b42c72009-03-31 15:19:03 +1100383 sh = __find_stripe(conf, sector, conf->generation - previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 if (!sh) {
385 if (!conf->inactive_blocked)
386 sh = get_free_stripe(conf);
387 if (noblock && sh == NULL)
388 break;
389 if (!sh) {
390 conf->inactive_blocked = 1;
391 wait_event_lock_irq(conf->wait_for_stripe,
392 !list_empty(&conf->inactive_list) &&
NeilBrown50368052005-12-12 02:39:17 -0800393 (atomic_read(&conf->active_stripes)
394 < (conf->max_nr_stripes *3/4)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 || !conf->inactive_blocked),
396 conf->device_lock,
NeilBrownf4370782006-07-10 04:44:14 -0700397 raid5_unplug_device(conf->mddev->queue)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 );
399 conf->inactive_blocked = 0;
400 } else
NeilBrownb5663ba2009-03-31 14:39:38 +1100401 init_stripe(sh, sector, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 } else {
403 if (atomic_read(&sh->count)) {
NeilBrownab69ae12009-03-31 15:26:47 +1100404 BUG_ON(!list_empty(&sh->lru)
405 && !test_bit(STRIPE_EXPANDING, &sh->state));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 } else {
407 if (!test_bit(STRIPE_HANDLE, &sh->state))
408 atomic_inc(&conf->active_stripes);
NeilBrownff4e8d92006-07-10 04:44:16 -0700409 if (list_empty(&sh->lru) &&
410 !test_bit(STRIPE_EXPANDING, &sh->state))
NeilBrown16a53ec2006-06-26 00:27:38 -0700411 BUG();
412 list_del_init(&sh->lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
414 }
415 } while (sh == NULL);
416
417 if (sh)
418 atomic_inc(&sh->count);
419
420 spin_unlock_irq(&conf->device_lock);
421 return sh;
422}
423
NeilBrown6712ecf2007-09-27 12:47:43 +0200424static void
425raid5_end_read_request(struct bio *bi, int error);
426static void
427raid5_end_write_request(struct bio *bi, int error);
Dan Williams91c00922007-01-02 13:52:30 -0700428
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000429static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
Dan Williams91c00922007-01-02 13:52:30 -0700430{
431 raid5_conf_t *conf = sh->raid_conf;
432 int i, disks = sh->disks;
433
434 might_sleep();
435
436 for (i = disks; i--; ) {
437 int rw;
438 struct bio *bi;
439 mdk_rdev_t *rdev;
440 if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags))
441 rw = WRITE;
442 else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags))
443 rw = READ;
444 else
445 continue;
446
447 bi = &sh->dev[i].req;
448
449 bi->bi_rw = rw;
450 if (rw == WRITE)
451 bi->bi_end_io = raid5_end_write_request;
452 else
453 bi->bi_end_io = raid5_end_read_request;
454
455 rcu_read_lock();
456 rdev = rcu_dereference(conf->disks[i].rdev);
457 if (rdev && test_bit(Faulty, &rdev->flags))
458 rdev = NULL;
459 if (rdev)
460 atomic_inc(&rdev->nr_pending);
461 rcu_read_unlock();
462
463 if (rdev) {
Dan Williamsc4e5ac02008-06-28 08:31:53 +1000464 if (s->syncing || s->expanding || s->expanded)
Dan Williams91c00922007-01-02 13:52:30 -0700465 md_sync_acct(rdev->bdev, STRIPE_SECTORS);
466
Dan Williams2b7497f2008-06-28 08:31:52 +1000467 set_bit(STRIPE_IO_STARTED, &sh->state);
468
Dan Williams91c00922007-01-02 13:52:30 -0700469 bi->bi_bdev = rdev->bdev;
470 pr_debug("%s: for %llu schedule op %ld on disc %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700471 __func__, (unsigned long long)sh->sector,
Dan Williams91c00922007-01-02 13:52:30 -0700472 bi->bi_rw, i);
473 atomic_inc(&sh->count);
474 bi->bi_sector = sh->sector + rdev->data_offset;
475 bi->bi_flags = 1 << BIO_UPTODATE;
476 bi->bi_vcnt = 1;
477 bi->bi_max_vecs = 1;
478 bi->bi_idx = 0;
479 bi->bi_io_vec = &sh->dev[i].vec;
480 bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
481 bi->bi_io_vec[0].bv_offset = 0;
482 bi->bi_size = STRIPE_SIZE;
483 bi->bi_next = NULL;
484 if (rw == WRITE &&
485 test_bit(R5_ReWrite, &sh->dev[i].flags))
486 atomic_add(STRIPE_SECTORS,
487 &rdev->corrected_errors);
488 generic_make_request(bi);
489 } else {
490 if (rw == WRITE)
491 set_bit(STRIPE_DEGRADED, &sh->state);
492 pr_debug("skip op %ld on disc %d for sector %llu\n",
493 bi->bi_rw, i, (unsigned long long)sh->sector);
494 clear_bit(R5_LOCKED, &sh->dev[i].flags);
495 set_bit(STRIPE_HANDLE, &sh->state);
496 }
497 }
498}
499
500static struct dma_async_tx_descriptor *
501async_copy_data(int frombio, struct bio *bio, struct page *page,
502 sector_t sector, struct dma_async_tx_descriptor *tx)
503{
504 struct bio_vec *bvl;
505 struct page *bio_page;
506 int i;
507 int page_offset;
Dan Williamsa08abd82009-06-03 11:43:59 -0700508 struct async_submit_ctl submit;
Dan Williams0403e382009-09-08 17:42:50 -0700509 enum async_tx_flags flags = 0;
Dan Williams91c00922007-01-02 13:52:30 -0700510
511 if (bio->bi_sector >= sector)
512 page_offset = (signed)(bio->bi_sector - sector) * 512;
513 else
514 page_offset = (signed)(sector - bio->bi_sector) * -512;
Dan Williamsa08abd82009-06-03 11:43:59 -0700515
Dan Williams0403e382009-09-08 17:42:50 -0700516 if (frombio)
517 flags |= ASYNC_TX_FENCE;
518 init_async_submit(&submit, flags, tx, NULL, NULL, NULL);
519
Dan Williams91c00922007-01-02 13:52:30 -0700520 bio_for_each_segment(bvl, bio, i) {
521 int len = bio_iovec_idx(bio, i)->bv_len;
522 int clen;
523 int b_offset = 0;
524
525 if (page_offset < 0) {
526 b_offset = -page_offset;
527 page_offset += b_offset;
528 len -= b_offset;
529 }
530
531 if (len > 0 && page_offset + len > STRIPE_SIZE)
532 clen = STRIPE_SIZE - page_offset;
533 else
534 clen = len;
535
536 if (clen > 0) {
537 b_offset += bio_iovec_idx(bio, i)->bv_offset;
538 bio_page = bio_iovec_idx(bio, i)->bv_page;
539 if (frombio)
540 tx = async_memcpy(page, bio_page, page_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700541 b_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700542 else
543 tx = async_memcpy(bio_page, page, b_offset,
Dan Williamsa08abd82009-06-03 11:43:59 -0700544 page_offset, clen, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700545 }
Dan Williamsa08abd82009-06-03 11:43:59 -0700546 /* chain the operations */
547 submit.depend_tx = tx;
548
Dan Williams91c00922007-01-02 13:52:30 -0700549 if (clen < len) /* hit end of page */
550 break;
551 page_offset += len;
552 }
553
554 return tx;
555}
556
557static void ops_complete_biofill(void *stripe_head_ref)
558{
559 struct stripe_head *sh = stripe_head_ref;
560 struct bio *return_bi = NULL;
561 raid5_conf_t *conf = sh->raid_conf;
Dan Williamse4d84902007-09-24 10:06:13 -0700562 int i;
Dan Williams91c00922007-01-02 13:52:30 -0700563
Harvey Harrisone46b2722008-04-28 02:15:50 -0700564 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700565 (unsigned long long)sh->sector);
566
567 /* clear completed biofills */
Dan Williams83de75c2008-06-28 08:31:58 +1000568 spin_lock_irq(&conf->device_lock);
Dan Williams91c00922007-01-02 13:52:30 -0700569 for (i = sh->disks; i--; ) {
570 struct r5dev *dev = &sh->dev[i];
Dan Williams91c00922007-01-02 13:52:30 -0700571
572 /* acknowledge completion of a biofill operation */
Dan Williamse4d84902007-09-24 10:06:13 -0700573 /* and check if we need to reply to a read request,
574 * new R5_Wantfill requests are held off until
Dan Williams83de75c2008-06-28 08:31:58 +1000575 * !STRIPE_BIOFILL_RUN
Dan Williamse4d84902007-09-24 10:06:13 -0700576 */
577 if (test_and_clear_bit(R5_Wantfill, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -0700578 struct bio *rbi, *rbi2;
Dan Williams91c00922007-01-02 13:52:30 -0700579
Dan Williams91c00922007-01-02 13:52:30 -0700580 BUG_ON(!dev->read);
581 rbi = dev->read;
582 dev->read = NULL;
583 while (rbi && rbi->bi_sector <
584 dev->sector + STRIPE_SECTORS) {
585 rbi2 = r5_next_bio(rbi, dev->sector);
Jens Axboe960e7392008-08-15 10:41:18 +0200586 if (!raid5_dec_bi_phys_segments(rbi)) {
Dan Williams91c00922007-01-02 13:52:30 -0700587 rbi->bi_next = return_bi;
588 return_bi = rbi;
589 }
Dan Williams91c00922007-01-02 13:52:30 -0700590 rbi = rbi2;
591 }
592 }
593 }
Dan Williams83de75c2008-06-28 08:31:58 +1000594 spin_unlock_irq(&conf->device_lock);
595 clear_bit(STRIPE_BIOFILL_RUN, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -0700596
597 return_io(return_bi);
598
Dan Williamse4d84902007-09-24 10:06:13 -0700599 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williams91c00922007-01-02 13:52:30 -0700600 release_stripe(sh);
601}
602
603static void ops_run_biofill(struct stripe_head *sh)
604{
605 struct dma_async_tx_descriptor *tx = NULL;
606 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsa08abd82009-06-03 11:43:59 -0700607 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700608 int i;
609
Harvey Harrisone46b2722008-04-28 02:15:50 -0700610 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700611 (unsigned long long)sh->sector);
612
613 for (i = sh->disks; i--; ) {
614 struct r5dev *dev = &sh->dev[i];
615 if (test_bit(R5_Wantfill, &dev->flags)) {
616 struct bio *rbi;
617 spin_lock_irq(&conf->device_lock);
618 dev->read = rbi = dev->toread;
619 dev->toread = NULL;
620 spin_unlock_irq(&conf->device_lock);
621 while (rbi && rbi->bi_sector <
622 dev->sector + STRIPE_SECTORS) {
623 tx = async_copy_data(0, rbi, dev->page,
624 dev->sector, tx);
625 rbi = r5_next_bio(rbi, dev->sector);
626 }
627 }
628 }
629
630 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -0700631 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL);
632 async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -0700633}
634
Dan Williams4e7d2c02009-08-29 19:13:11 -0700635static void mark_target_uptodate(struct stripe_head *sh, int target)
636{
637 struct r5dev *tgt;
638
639 if (target < 0)
640 return;
641
642 tgt = &sh->dev[target];
643 set_bit(R5_UPTODATE, &tgt->flags);
644 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
645 clear_bit(R5_Wantcompute, &tgt->flags);
646}
647
Dan Williamsac6b53b2009-07-14 13:40:19 -0700648static void ops_complete_compute(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -0700649{
650 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -0700651
Harvey Harrisone46b2722008-04-28 02:15:50 -0700652 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700653 (unsigned long long)sh->sector);
654
Dan Williamsac6b53b2009-07-14 13:40:19 -0700655 /* mark the computed target(s) as uptodate */
Dan Williams4e7d2c02009-08-29 19:13:11 -0700656 mark_target_uptodate(sh, sh->ops.target);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700657 mark_target_uptodate(sh, sh->ops.target2);
Dan Williams4e7d2c02009-08-29 19:13:11 -0700658
Dan Williamsecc65c92008-06-28 08:31:57 +1000659 clear_bit(STRIPE_COMPUTE_RUN, &sh->state);
660 if (sh->check_state == check_state_compute_run)
661 sh->check_state = check_state_compute_result;
Dan Williams91c00922007-01-02 13:52:30 -0700662 set_bit(STRIPE_HANDLE, &sh->state);
663 release_stripe(sh);
664}
665
Dan Williamsd6f38f32009-07-14 11:50:52 -0700666/* return a pointer to the address conversion region of the scribble buffer */
667static addr_conv_t *to_addr_conv(struct stripe_head *sh,
668 struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -0700669{
Dan Williamsd6f38f32009-07-14 11:50:52 -0700670 return percpu->scribble + sizeof(struct page *) * (sh->disks + 2);
671}
672
673static struct dma_async_tx_descriptor *
674ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu)
675{
Dan Williams91c00922007-01-02 13:52:30 -0700676 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -0700677 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -0700678 int target = sh->ops.target;
679 struct r5dev *tgt = &sh->dev[target];
680 struct page *xor_dest = tgt->page;
681 int count = 0;
682 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -0700683 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700684 int i;
685
686 pr_debug("%s: stripe %llu block: %d\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -0700687 __func__, (unsigned long long)sh->sector, target);
Dan Williams91c00922007-01-02 13:52:30 -0700688 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
689
690 for (i = disks; i--; )
691 if (i != target)
692 xor_srcs[count++] = sh->dev[i].page;
693
694 atomic_inc(&sh->count);
695
Dan Williams0403e382009-09-08 17:42:50 -0700696 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700697 ops_complete_compute, sh, to_addr_conv(sh, percpu));
Dan Williams91c00922007-01-02 13:52:30 -0700698 if (unlikely(count == 1))
Dan Williamsa08abd82009-06-03 11:43:59 -0700699 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700700 else
Dan Williamsa08abd82009-06-03 11:43:59 -0700701 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700702
Dan Williams91c00922007-01-02 13:52:30 -0700703 return tx;
704}
705
Dan Williamsac6b53b2009-07-14 13:40:19 -0700706/* set_syndrome_sources - populate source buffers for gen_syndrome
707 * @srcs - (struct page *) array of size sh->disks
708 * @sh - stripe_head to parse
709 *
710 * Populates srcs in proper layout order for the stripe and returns the
711 * 'count' of sources to be used in a call to async_gen_syndrome. The P
712 * destination buffer is recorded in srcs[count] and the Q destination
713 * is recorded in srcs[count+1]].
714 */
715static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh)
716{
717 int disks = sh->disks;
718 int syndrome_disks = sh->ddf_layout ? disks : (disks - 2);
719 int d0_idx = raid6_d0(sh);
720 int count;
721 int i;
722
723 for (i = 0; i < disks; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +1100724 srcs[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700725
726 count = 0;
727 i = d0_idx;
728 do {
729 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
730
731 srcs[slot] = sh->dev[i].page;
732 i = raid6_next_disk(i, disks);
733 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700734
NeilBrowne4424fe2009-10-16 16:27:34 +1100735 return syndrome_disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700736}
737
738static struct dma_async_tx_descriptor *
739ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu)
740{
741 int disks = sh->disks;
742 struct page **blocks = percpu->scribble;
743 int target;
744 int qd_idx = sh->qd_idx;
745 struct dma_async_tx_descriptor *tx;
746 struct async_submit_ctl submit;
747 struct r5dev *tgt;
748 struct page *dest;
749 int i;
750 int count;
751
752 if (sh->ops.target < 0)
753 target = sh->ops.target2;
754 else if (sh->ops.target2 < 0)
755 target = sh->ops.target;
756 else
757 /* we should only have one valid target */
758 BUG();
759 BUG_ON(target < 0);
760 pr_debug("%s: stripe %llu block: %d\n",
761 __func__, (unsigned long long)sh->sector, target);
762
763 tgt = &sh->dev[target];
764 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
765 dest = tgt->page;
766
767 atomic_inc(&sh->count);
768
769 if (target == qd_idx) {
770 count = set_syndrome_sources(blocks, sh);
771 blocks[count] = NULL; /* regenerating p is not necessary */
772 BUG_ON(blocks[count+1] != dest); /* q should already be set */
Dan Williams0403e382009-09-08 17:42:50 -0700773 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
774 ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700775 to_addr_conv(sh, percpu));
776 tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
777 } else {
778 /* Compute any data- or p-drive using XOR */
779 count = 0;
780 for (i = disks; i-- ; ) {
781 if (i == target || i == qd_idx)
782 continue;
783 blocks[count++] = sh->dev[i].page;
784 }
785
Dan Williams0403e382009-09-08 17:42:50 -0700786 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
787 NULL, ops_complete_compute, sh,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700788 to_addr_conv(sh, percpu));
789 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit);
790 }
791
792 return tx;
793}
794
795static struct dma_async_tx_descriptor *
796ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu)
797{
798 int i, count, disks = sh->disks;
799 int syndrome_disks = sh->ddf_layout ? disks : disks-2;
800 int d0_idx = raid6_d0(sh);
801 int faila = -1, failb = -1;
802 int target = sh->ops.target;
803 int target2 = sh->ops.target2;
804 struct r5dev *tgt = &sh->dev[target];
805 struct r5dev *tgt2 = &sh->dev[target2];
806 struct dma_async_tx_descriptor *tx;
807 struct page **blocks = percpu->scribble;
808 struct async_submit_ctl submit;
809
810 pr_debug("%s: stripe %llu block1: %d block2: %d\n",
811 __func__, (unsigned long long)sh->sector, target, target2);
812 BUG_ON(target < 0 || target2 < 0);
813 BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags));
814 BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags));
815
Dan Williams6c910a72009-09-16 12:24:54 -0700816 /* we need to open-code set_syndrome_sources to handle the
Dan Williamsac6b53b2009-07-14 13:40:19 -0700817 * slot number conversion for 'faila' and 'failb'
818 */
819 for (i = 0; i < disks ; i++)
NeilBrown5dd33c92009-10-16 16:40:25 +1100820 blocks[i] = NULL;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700821 count = 0;
822 i = d0_idx;
823 do {
824 int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks);
825
826 blocks[slot] = sh->dev[i].page;
827
828 if (i == target)
829 faila = slot;
830 if (i == target2)
831 failb = slot;
832 i = raid6_next_disk(i, disks);
833 } while (i != d0_idx);
Dan Williamsac6b53b2009-07-14 13:40:19 -0700834
835 BUG_ON(faila == failb);
836 if (failb < faila)
837 swap(faila, failb);
838 pr_debug("%s: stripe: %llu faila: %d failb: %d\n",
839 __func__, (unsigned long long)sh->sector, faila, failb);
840
841 atomic_inc(&sh->count);
842
843 if (failb == syndrome_disks+1) {
844 /* Q disk is one of the missing disks */
845 if (faila == syndrome_disks) {
846 /* Missing P+Q, just recompute */
Dan Williams0403e382009-09-08 17:42:50 -0700847 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
848 ops_complete_compute, sh,
849 to_addr_conv(sh, percpu));
NeilBrowne4424fe2009-10-16 16:27:34 +1100850 return async_gen_syndrome(blocks, 0, syndrome_disks+2,
Dan Williamsac6b53b2009-07-14 13:40:19 -0700851 STRIPE_SIZE, &submit);
852 } else {
853 struct page *dest;
854 int data_target;
855 int qd_idx = sh->qd_idx;
856
857 /* Missing D+Q: recompute D from P, then recompute Q */
858 if (target == qd_idx)
859 data_target = target2;
860 else
861 data_target = target;
862
863 count = 0;
864 for (i = disks; i-- ; ) {
865 if (i == data_target || i == qd_idx)
866 continue;
867 blocks[count++] = sh->dev[i].page;
868 }
869 dest = sh->dev[data_target].page;
Dan Williams0403e382009-09-08 17:42:50 -0700870 init_async_submit(&submit,
871 ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST,
872 NULL, NULL, NULL,
873 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -0700874 tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE,
875 &submit);
876
877 count = set_syndrome_sources(blocks, sh);
Dan Williams0403e382009-09-08 17:42:50 -0700878 init_async_submit(&submit, ASYNC_TX_FENCE, tx,
879 ops_complete_compute, sh,
880 to_addr_conv(sh, percpu));
Dan Williamsac6b53b2009-07-14 13:40:19 -0700881 return async_gen_syndrome(blocks, 0, count+2,
882 STRIPE_SIZE, &submit);
883 }
Dan Williamsac6b53b2009-07-14 13:40:19 -0700884 } else {
Dan Williams6c910a72009-09-16 12:24:54 -0700885 init_async_submit(&submit, ASYNC_TX_FENCE, NULL,
886 ops_complete_compute, sh,
887 to_addr_conv(sh, percpu));
888 if (failb == syndrome_disks) {
889 /* We're missing D+P. */
890 return async_raid6_datap_recov(syndrome_disks+2,
891 STRIPE_SIZE, faila,
892 blocks, &submit);
893 } else {
894 /* We're missing D+D. */
895 return async_raid6_2data_recov(syndrome_disks+2,
896 STRIPE_SIZE, faila, failb,
897 blocks, &submit);
898 }
Dan Williamsac6b53b2009-07-14 13:40:19 -0700899 }
900}
901
902
Dan Williams91c00922007-01-02 13:52:30 -0700903static void ops_complete_prexor(void *stripe_head_ref)
904{
905 struct stripe_head *sh = stripe_head_ref;
906
Harvey Harrisone46b2722008-04-28 02:15:50 -0700907 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700908 (unsigned long long)sh->sector);
Dan Williams91c00922007-01-02 13:52:30 -0700909}
910
911static struct dma_async_tx_descriptor *
Dan Williamsd6f38f32009-07-14 11:50:52 -0700912ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu,
913 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -0700914{
Dan Williams91c00922007-01-02 13:52:30 -0700915 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -0700916 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -0700917 int count = 0, pd_idx = sh->pd_idx, i;
Dan Williamsa08abd82009-06-03 11:43:59 -0700918 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -0700919
920 /* existing parity data subtracted */
921 struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
922
Harvey Harrisone46b2722008-04-28 02:15:50 -0700923 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700924 (unsigned long long)sh->sector);
925
926 for (i = disks; i--; ) {
927 struct r5dev *dev = &sh->dev[i];
928 /* Only process blocks that are known to be uptodate */
Dan Williamsd8ee0722008-06-28 08:32:06 +1000929 if (test_bit(R5_Wantdrain, &dev->flags))
Dan Williams91c00922007-01-02 13:52:30 -0700930 xor_srcs[count++] = dev->page;
931 }
932
Dan Williams0403e382009-09-08 17:42:50 -0700933 init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx,
Dan Williamsd6f38f32009-07-14 11:50:52 -0700934 ops_complete_prexor, sh, to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -0700935 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -0700936
937 return tx;
938}
939
940static struct dma_async_tx_descriptor *
Dan Williamsd8ee0722008-06-28 08:32:06 +1000941ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -0700942{
943 int disks = sh->disks;
Dan Williamsd8ee0722008-06-28 08:32:06 +1000944 int i;
Dan Williams91c00922007-01-02 13:52:30 -0700945
Harvey Harrisone46b2722008-04-28 02:15:50 -0700946 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700947 (unsigned long long)sh->sector);
948
949 for (i = disks; i--; ) {
950 struct r5dev *dev = &sh->dev[i];
951 struct bio *chosen;
Dan Williams91c00922007-01-02 13:52:30 -0700952
Dan Williamsd8ee0722008-06-28 08:32:06 +1000953 if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) {
Dan Williams91c00922007-01-02 13:52:30 -0700954 struct bio *wbi;
955
956 spin_lock(&sh->lock);
957 chosen = dev->towrite;
958 dev->towrite = NULL;
959 BUG_ON(dev->written);
960 wbi = dev->written = chosen;
961 spin_unlock(&sh->lock);
962
963 while (wbi && wbi->bi_sector <
964 dev->sector + STRIPE_SECTORS) {
965 tx = async_copy_data(1, wbi, dev->page,
966 dev->sector, tx);
967 wbi = r5_next_bio(wbi, dev->sector);
968 }
969 }
970 }
971
972 return tx;
973}
974
Dan Williamsac6b53b2009-07-14 13:40:19 -0700975static void ops_complete_reconstruct(void *stripe_head_ref)
Dan Williams91c00922007-01-02 13:52:30 -0700976{
977 struct stripe_head *sh = stripe_head_ref;
Dan Williamsac6b53b2009-07-14 13:40:19 -0700978 int disks = sh->disks;
979 int pd_idx = sh->pd_idx;
980 int qd_idx = sh->qd_idx;
981 int i;
Dan Williams91c00922007-01-02 13:52:30 -0700982
Harvey Harrisone46b2722008-04-28 02:15:50 -0700983 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -0700984 (unsigned long long)sh->sector);
985
986 for (i = disks; i--; ) {
987 struct r5dev *dev = &sh->dev[i];
Dan Williamsac6b53b2009-07-14 13:40:19 -0700988
989 if (dev->written || i == pd_idx || i == qd_idx)
Dan Williams91c00922007-01-02 13:52:30 -0700990 set_bit(R5_UPTODATE, &dev->flags);
991 }
992
Dan Williamsd8ee0722008-06-28 08:32:06 +1000993 if (sh->reconstruct_state == reconstruct_state_drain_run)
994 sh->reconstruct_state = reconstruct_state_drain_result;
995 else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run)
996 sh->reconstruct_state = reconstruct_state_prexor_drain_result;
997 else {
998 BUG_ON(sh->reconstruct_state != reconstruct_state_run);
999 sh->reconstruct_state = reconstruct_state_result;
1000 }
Dan Williams91c00922007-01-02 13:52:30 -07001001
1002 set_bit(STRIPE_HANDLE, &sh->state);
1003 release_stripe(sh);
1004}
1005
1006static void
Dan Williamsac6b53b2009-07-14 13:40:19 -07001007ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu,
1008 struct dma_async_tx_descriptor *tx)
Dan Williams91c00922007-01-02 13:52:30 -07001009{
Dan Williams91c00922007-01-02 13:52:30 -07001010 int disks = sh->disks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001011 struct page **xor_srcs = percpu->scribble;
Dan Williamsa08abd82009-06-03 11:43:59 -07001012 struct async_submit_ctl submit;
Dan Williams91c00922007-01-02 13:52:30 -07001013 int count = 0, pd_idx = sh->pd_idx, i;
1014 struct page *xor_dest;
Dan Williamsd8ee0722008-06-28 08:32:06 +10001015 int prexor = 0;
Dan Williams91c00922007-01-02 13:52:30 -07001016 unsigned long flags;
Dan Williams91c00922007-01-02 13:52:30 -07001017
Harvey Harrisone46b2722008-04-28 02:15:50 -07001018 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001019 (unsigned long long)sh->sector);
1020
1021 /* check if prexor is active which means only process blocks
1022 * that are part of a read-modify-write (written)
1023 */
Dan Williamsd8ee0722008-06-28 08:32:06 +10001024 if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
1025 prexor = 1;
Dan Williams91c00922007-01-02 13:52:30 -07001026 xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page;
1027 for (i = disks; i--; ) {
1028 struct r5dev *dev = &sh->dev[i];
1029 if (dev->written)
1030 xor_srcs[count++] = dev->page;
1031 }
1032 } else {
1033 xor_dest = sh->dev[pd_idx].page;
1034 for (i = disks; i--; ) {
1035 struct r5dev *dev = &sh->dev[i];
1036 if (i != pd_idx)
1037 xor_srcs[count++] = dev->page;
1038 }
1039 }
1040
Dan Williams91c00922007-01-02 13:52:30 -07001041 /* 1/ if we prexor'd then the dest is reused as a source
1042 * 2/ if we did not prexor then we are redoing the parity
1043 * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST
1044 * for the synchronous xor case
1045 */
Dan Williams88ba2aa2009-04-09 16:16:18 -07001046 flags = ASYNC_TX_ACK |
Dan Williams91c00922007-01-02 13:52:30 -07001047 (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST);
1048
1049 atomic_inc(&sh->count);
1050
Dan Williamsac6b53b2009-07-14 13:40:19 -07001051 init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh,
Dan Williamsd6f38f32009-07-14 11:50:52 -07001052 to_addr_conv(sh, percpu));
Dan Williamsa08abd82009-06-03 11:43:59 -07001053 if (unlikely(count == 1))
1054 tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit);
1055 else
1056 tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001057}
1058
Dan Williamsac6b53b2009-07-14 13:40:19 -07001059static void
1060ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu,
1061 struct dma_async_tx_descriptor *tx)
1062{
1063 struct async_submit_ctl submit;
1064 struct page **blocks = percpu->scribble;
1065 int count;
1066
1067 pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
1068
1069 count = set_syndrome_sources(blocks, sh);
1070
1071 atomic_inc(&sh->count);
1072
1073 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct,
1074 sh, to_addr_conv(sh, percpu));
1075 async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001076}
1077
1078static void ops_complete_check(void *stripe_head_ref)
1079{
1080 struct stripe_head *sh = stripe_head_ref;
Dan Williams91c00922007-01-02 13:52:30 -07001081
Harvey Harrisone46b2722008-04-28 02:15:50 -07001082 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001083 (unsigned long long)sh->sector);
1084
Dan Williamsecc65c92008-06-28 08:31:57 +10001085 sh->check_state = check_state_check_result;
Dan Williams91c00922007-01-02 13:52:30 -07001086 set_bit(STRIPE_HANDLE, &sh->state);
1087 release_stripe(sh);
1088}
1089
Dan Williamsac6b53b2009-07-14 13:40:19 -07001090static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu)
Dan Williams91c00922007-01-02 13:52:30 -07001091{
Dan Williams91c00922007-01-02 13:52:30 -07001092 int disks = sh->disks;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001093 int pd_idx = sh->pd_idx;
1094 int qd_idx = sh->qd_idx;
1095 struct page *xor_dest;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001096 struct page **xor_srcs = percpu->scribble;
Dan Williams91c00922007-01-02 13:52:30 -07001097 struct dma_async_tx_descriptor *tx;
Dan Williamsa08abd82009-06-03 11:43:59 -07001098 struct async_submit_ctl submit;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001099 int count;
1100 int i;
Dan Williams91c00922007-01-02 13:52:30 -07001101
Harvey Harrisone46b2722008-04-28 02:15:50 -07001102 pr_debug("%s: stripe %llu\n", __func__,
Dan Williams91c00922007-01-02 13:52:30 -07001103 (unsigned long long)sh->sector);
1104
Dan Williamsac6b53b2009-07-14 13:40:19 -07001105 count = 0;
1106 xor_dest = sh->dev[pd_idx].page;
1107 xor_srcs[count++] = xor_dest;
Dan Williams91c00922007-01-02 13:52:30 -07001108 for (i = disks; i--; ) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001109 if (i == pd_idx || i == qd_idx)
1110 continue;
1111 xor_srcs[count++] = sh->dev[i].page;
Dan Williams91c00922007-01-02 13:52:30 -07001112 }
1113
Dan Williamsd6f38f32009-07-14 11:50:52 -07001114 init_async_submit(&submit, 0, NULL, NULL, NULL,
1115 to_addr_conv(sh, percpu));
Dan Williams099f53c2009-04-08 14:28:37 -07001116 tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07001117 &sh->ops.zero_sum_result, &submit);
Dan Williams91c00922007-01-02 13:52:30 -07001118
Dan Williams91c00922007-01-02 13:52:30 -07001119 atomic_inc(&sh->count);
Dan Williamsa08abd82009-06-03 11:43:59 -07001120 init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL);
1121 tx = async_trigger_callback(&submit);
Dan Williams91c00922007-01-02 13:52:30 -07001122}
1123
Dan Williamsac6b53b2009-07-14 13:40:19 -07001124static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp)
1125{
1126 struct page **srcs = percpu->scribble;
1127 struct async_submit_ctl submit;
1128 int count;
1129
1130 pr_debug("%s: stripe %llu checkp: %d\n", __func__,
1131 (unsigned long long)sh->sector, checkp);
1132
1133 count = set_syndrome_sources(srcs, sh);
1134 if (!checkp)
1135 srcs[count] = NULL;
1136
1137 atomic_inc(&sh->count);
1138 init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check,
1139 sh, to_addr_conv(sh, percpu));
1140 async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE,
1141 &sh->ops.zero_sum_result, percpu->spare_page, &submit);
1142}
1143
Dan Williams417b8d42009-10-16 16:25:22 +11001144static void __raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
Dan Williams91c00922007-01-02 13:52:30 -07001145{
1146 int overlap_clear = 0, i, disks = sh->disks;
1147 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001148 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsac6b53b2009-07-14 13:40:19 -07001149 int level = conf->level;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001150 struct raid5_percpu *percpu;
1151 unsigned long cpu;
Dan Williams91c00922007-01-02 13:52:30 -07001152
Dan Williamsd6f38f32009-07-14 11:50:52 -07001153 cpu = get_cpu();
1154 percpu = per_cpu_ptr(conf->percpu, cpu);
Dan Williams83de75c2008-06-28 08:31:58 +10001155 if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) {
Dan Williams91c00922007-01-02 13:52:30 -07001156 ops_run_biofill(sh);
1157 overlap_clear++;
1158 }
1159
Dan Williams7b3a8712008-06-28 08:32:09 +10001160 if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) {
Dan Williamsac6b53b2009-07-14 13:40:19 -07001161 if (level < 6)
1162 tx = ops_run_compute5(sh, percpu);
1163 else {
1164 if (sh->ops.target2 < 0 || sh->ops.target < 0)
1165 tx = ops_run_compute6_1(sh, percpu);
1166 else
1167 tx = ops_run_compute6_2(sh, percpu);
1168 }
1169 /* terminate the chain if reconstruct is not set to be run */
1170 if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request))
Dan Williams7b3a8712008-06-28 08:32:09 +10001171 async_tx_ack(tx);
1172 }
Dan Williams91c00922007-01-02 13:52:30 -07001173
Dan Williams600aa102008-06-28 08:32:05 +10001174 if (test_bit(STRIPE_OP_PREXOR, &ops_request))
Dan Williamsd6f38f32009-07-14 11:50:52 -07001175 tx = ops_run_prexor(sh, percpu, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001176
Dan Williams600aa102008-06-28 08:32:05 +10001177 if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) {
Dan Williamsd8ee0722008-06-28 08:32:06 +10001178 tx = ops_run_biodrain(sh, tx);
Dan Williams91c00922007-01-02 13:52:30 -07001179 overlap_clear++;
1180 }
1181
Dan Williamsac6b53b2009-07-14 13:40:19 -07001182 if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) {
1183 if (level < 6)
1184 ops_run_reconstruct5(sh, percpu, tx);
1185 else
1186 ops_run_reconstruct6(sh, percpu, tx);
1187 }
Dan Williams91c00922007-01-02 13:52:30 -07001188
Dan Williamsac6b53b2009-07-14 13:40:19 -07001189 if (test_bit(STRIPE_OP_CHECK, &ops_request)) {
1190 if (sh->check_state == check_state_run)
1191 ops_run_check_p(sh, percpu);
1192 else if (sh->check_state == check_state_run_q)
1193 ops_run_check_pq(sh, percpu, 0);
1194 else if (sh->check_state == check_state_run_pq)
1195 ops_run_check_pq(sh, percpu, 1);
1196 else
1197 BUG();
1198 }
Dan Williams91c00922007-01-02 13:52:30 -07001199
Dan Williams91c00922007-01-02 13:52:30 -07001200 if (overlap_clear)
1201 for (i = disks; i--; ) {
1202 struct r5dev *dev = &sh->dev[i];
1203 if (test_and_clear_bit(R5_Overlap, &dev->flags))
1204 wake_up(&sh->raid_conf->wait_for_overlap);
1205 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07001206 put_cpu();
Dan Williams91c00922007-01-02 13:52:30 -07001207}
1208
Dan Williams417b8d42009-10-16 16:25:22 +11001209#ifdef CONFIG_MULTICORE_RAID456
1210static void async_run_ops(void *param, async_cookie_t cookie)
1211{
1212 struct stripe_head *sh = param;
1213 unsigned long ops_request = sh->ops.request;
1214
1215 clear_bit_unlock(STRIPE_OPS_REQ_PENDING, &sh->state);
1216 wake_up(&sh->ops.wait_for_ops);
1217
1218 __raid_run_ops(sh, ops_request);
1219 release_stripe(sh);
1220}
1221
1222static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
1223{
1224 /* since handle_stripe can be called outside of raid5d context
1225 * we need to ensure sh->ops.request is de-staged before another
1226 * request arrives
1227 */
1228 wait_event(sh->ops.wait_for_ops,
1229 !test_and_set_bit_lock(STRIPE_OPS_REQ_PENDING, &sh->state));
1230 sh->ops.request = ops_request;
1231
1232 atomic_inc(&sh->count);
1233 async_schedule(async_run_ops, sh);
1234}
1235#else
1236#define raid_run_ops __raid_run_ops
1237#endif
1238
NeilBrown3f294f42005-11-08 21:39:25 -08001239static int grow_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240{
1241 struct stripe_head *sh;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001242 int disks = max(conf->raid_disks, conf->previous_raid_disks);
NeilBrown3f294f42005-11-08 21:39:25 -08001243 sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL);
1244 if (!sh)
1245 return 0;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001246 memset(sh, 0, sizeof(*sh) + (disks-1)*sizeof(struct r5dev));
NeilBrown3f294f42005-11-08 21:39:25 -08001247 sh->raid_conf = conf;
1248 spin_lock_init(&sh->lock);
Dan Williams417b8d42009-10-16 16:25:22 +11001249 #ifdef CONFIG_MULTICORE_RAID456
1250 init_waitqueue_head(&sh->ops.wait_for_ops);
1251 #endif
NeilBrown3f294f42005-11-08 21:39:25 -08001252
NeilBrown5e5e3e72009-10-16 16:35:30 +11001253 if (grow_buffers(sh, disks)) {
1254 shrink_buffers(sh, disks);
NeilBrown3f294f42005-11-08 21:39:25 -08001255 kmem_cache_free(conf->slab_cache, sh);
1256 return 0;
1257 }
1258 /* we just created an active stripe so... */
1259 atomic_set(&sh->count, 1);
1260 atomic_inc(&conf->active_stripes);
1261 INIT_LIST_HEAD(&sh->lru);
1262 release_stripe(sh);
1263 return 1;
1264}
1265
1266static int grow_stripes(raid5_conf_t *conf, int num)
1267{
Christoph Lametere18b8902006-12-06 20:33:20 -08001268 struct kmem_cache *sc;
NeilBrown5e5e3e72009-10-16 16:35:30 +11001269 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
NeilBrown245f46c2009-03-31 14:39:39 +11001271 sprintf(conf->cache_name[0],
1272 "raid%d-%s", conf->level, mdname(conf->mddev));
1273 sprintf(conf->cache_name[1],
1274 "raid%d-%s-alt", conf->level, mdname(conf->mddev));
NeilBrownad01c9e2006-03-27 01:18:07 -08001275 conf->active_name = 0;
1276 sc = kmem_cache_create(conf->cache_name[conf->active_name],
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001278 0, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 if (!sc)
1280 return 1;
1281 conf->slab_cache = sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001282 conf->pool_size = devs;
NeilBrown16a53ec2006-06-26 00:27:38 -07001283 while (num--)
NeilBrown3f294f42005-11-08 21:39:25 -08001284 if (!grow_one_stripe(conf))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return 0;
1287}
NeilBrown29269552006-03-27 01:18:10 -08001288
Dan Williamsd6f38f32009-07-14 11:50:52 -07001289/**
1290 * scribble_len - return the required size of the scribble region
1291 * @num - total number of disks in the array
1292 *
1293 * The size must be enough to contain:
1294 * 1/ a struct page pointer for each device in the array +2
1295 * 2/ room to convert each entry in (1) to its corresponding dma
1296 * (dma_map_page()) or page (page_address()) address.
1297 *
1298 * Note: the +2 is for the destination buffers of the ddf/raid6 case where we
1299 * calculate over all devices (not just the data blocks), using zeros in place
1300 * of the P and Q blocks.
1301 */
1302static size_t scribble_len(int num)
1303{
1304 size_t len;
1305
1306 len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2);
1307
1308 return len;
1309}
1310
NeilBrownad01c9e2006-03-27 01:18:07 -08001311static int resize_stripes(raid5_conf_t *conf, int newsize)
1312{
1313 /* Make all the stripes able to hold 'newsize' devices.
1314 * New slots in each stripe get 'page' set to a new page.
1315 *
1316 * This happens in stages:
1317 * 1/ create a new kmem_cache and allocate the required number of
1318 * stripe_heads.
1319 * 2/ gather all the old stripe_heads and tranfer the pages across
1320 * to the new stripe_heads. This will have the side effect of
1321 * freezing the array as once all stripe_heads have been collected,
1322 * no IO will be possible. Old stripe heads are freed once their
1323 * pages have been transferred over, and the old kmem_cache is
1324 * freed when all stripes are done.
1325 * 3/ reallocate conf->disks to be suitable bigger. If this fails,
1326 * we simple return a failre status - no need to clean anything up.
1327 * 4/ allocate new pages for the new slots in the new stripe_heads.
1328 * If this fails, we don't bother trying the shrink the
1329 * stripe_heads down again, we just leave them as they are.
1330 * As each stripe_head is processed the new one is released into
1331 * active service.
1332 *
1333 * Once step2 is started, we cannot afford to wait for a write,
1334 * so we use GFP_NOIO allocations.
1335 */
1336 struct stripe_head *osh, *nsh;
1337 LIST_HEAD(newstripes);
1338 struct disk_info *ndisks;
Dan Williamsd6f38f32009-07-14 11:50:52 -07001339 unsigned long cpu;
Dan Williamsb5470dc2008-06-27 21:44:04 -07001340 int err;
Christoph Lametere18b8902006-12-06 20:33:20 -08001341 struct kmem_cache *sc;
NeilBrownad01c9e2006-03-27 01:18:07 -08001342 int i;
1343
1344 if (newsize <= conf->pool_size)
1345 return 0; /* never bother to shrink */
1346
Dan Williamsb5470dc2008-06-27 21:44:04 -07001347 err = md_allow_write(conf->mddev);
1348 if (err)
1349 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08001350
NeilBrownad01c9e2006-03-27 01:18:07 -08001351 /* Step 1 */
1352 sc = kmem_cache_create(conf->cache_name[1-conf->active_name],
1353 sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev),
Paul Mundt20c2df82007-07-20 10:11:58 +09001354 0, 0, NULL);
NeilBrownad01c9e2006-03-27 01:18:07 -08001355 if (!sc)
1356 return -ENOMEM;
1357
1358 for (i = conf->max_nr_stripes; i; i--) {
1359 nsh = kmem_cache_alloc(sc, GFP_KERNEL);
1360 if (!nsh)
1361 break;
1362
1363 memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev));
1364
1365 nsh->raid_conf = conf;
1366 spin_lock_init(&nsh->lock);
Dan Williams417b8d42009-10-16 16:25:22 +11001367 #ifdef CONFIG_MULTICORE_RAID456
1368 init_waitqueue_head(&nsh->ops.wait_for_ops);
1369 #endif
NeilBrownad01c9e2006-03-27 01:18:07 -08001370
1371 list_add(&nsh->lru, &newstripes);
1372 }
1373 if (i) {
1374 /* didn't get enough, give up */
1375 while (!list_empty(&newstripes)) {
1376 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1377 list_del(&nsh->lru);
1378 kmem_cache_free(sc, nsh);
1379 }
1380 kmem_cache_destroy(sc);
1381 return -ENOMEM;
1382 }
1383 /* Step 2 - Must use GFP_NOIO now.
1384 * OK, we have enough stripes, start collecting inactive
1385 * stripes and copying them over
1386 */
1387 list_for_each_entry(nsh, &newstripes, lru) {
1388 spin_lock_irq(&conf->device_lock);
1389 wait_event_lock_irq(conf->wait_for_stripe,
1390 !list_empty(&conf->inactive_list),
1391 conf->device_lock,
NeilBrownb3b46be2006-03-27 01:18:16 -08001392 unplug_slaves(conf->mddev)
NeilBrownad01c9e2006-03-27 01:18:07 -08001393 );
1394 osh = get_free_stripe(conf);
1395 spin_unlock_irq(&conf->device_lock);
1396 atomic_set(&nsh->count, 1);
1397 for(i=0; i<conf->pool_size; i++)
1398 nsh->dev[i].page = osh->dev[i].page;
1399 for( ; i<newsize; i++)
1400 nsh->dev[i].page = NULL;
1401 kmem_cache_free(conf->slab_cache, osh);
1402 }
1403 kmem_cache_destroy(conf->slab_cache);
1404
1405 /* Step 3.
1406 * At this point, we are holding all the stripes so the array
1407 * is completely stalled, so now is a good time to resize
Dan Williamsd6f38f32009-07-14 11:50:52 -07001408 * conf->disks and the scribble region
NeilBrownad01c9e2006-03-27 01:18:07 -08001409 */
1410 ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO);
1411 if (ndisks) {
1412 for (i=0; i<conf->raid_disks; i++)
1413 ndisks[i] = conf->disks[i];
1414 kfree(conf->disks);
1415 conf->disks = ndisks;
1416 } else
1417 err = -ENOMEM;
1418
Dan Williamsd6f38f32009-07-14 11:50:52 -07001419 get_online_cpus();
1420 conf->scribble_len = scribble_len(newsize);
1421 for_each_present_cpu(cpu) {
1422 struct raid5_percpu *percpu;
1423 void *scribble;
1424
1425 percpu = per_cpu_ptr(conf->percpu, cpu);
1426 scribble = kmalloc(conf->scribble_len, GFP_NOIO);
1427
1428 if (scribble) {
1429 kfree(percpu->scribble);
1430 percpu->scribble = scribble;
1431 } else {
1432 err = -ENOMEM;
1433 break;
1434 }
1435 }
1436 put_online_cpus();
1437
NeilBrownad01c9e2006-03-27 01:18:07 -08001438 /* Step 4, return new stripes to service */
1439 while(!list_empty(&newstripes)) {
1440 nsh = list_entry(newstripes.next, struct stripe_head, lru);
1441 list_del_init(&nsh->lru);
Dan Williamsd6f38f32009-07-14 11:50:52 -07001442
NeilBrownad01c9e2006-03-27 01:18:07 -08001443 for (i=conf->raid_disks; i < newsize; i++)
1444 if (nsh->dev[i].page == NULL) {
1445 struct page *p = alloc_page(GFP_NOIO);
1446 nsh->dev[i].page = p;
1447 if (!p)
1448 err = -ENOMEM;
1449 }
1450 release_stripe(nsh);
1451 }
1452 /* critical section pass, GFP_NOIO no longer needed */
1453
1454 conf->slab_cache = sc;
1455 conf->active_name = 1-conf->active_name;
1456 conf->pool_size = newsize;
1457 return err;
1458}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
NeilBrown3f294f42005-11-08 21:39:25 -08001460static int drop_one_stripe(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
1462 struct stripe_head *sh;
1463
NeilBrown3f294f42005-11-08 21:39:25 -08001464 spin_lock_irq(&conf->device_lock);
1465 sh = get_free_stripe(conf);
1466 spin_unlock_irq(&conf->device_lock);
1467 if (!sh)
1468 return 0;
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02001469 BUG_ON(atomic_read(&sh->count));
NeilBrownad01c9e2006-03-27 01:18:07 -08001470 shrink_buffers(sh, conf->pool_size);
NeilBrown3f294f42005-11-08 21:39:25 -08001471 kmem_cache_free(conf->slab_cache, sh);
1472 atomic_dec(&conf->active_stripes);
1473 return 1;
1474}
1475
1476static void shrink_stripes(raid5_conf_t *conf)
1477{
1478 while (drop_one_stripe(conf))
1479 ;
1480
NeilBrown29fc7e32006-02-03 03:03:41 -08001481 if (conf->slab_cache)
1482 kmem_cache_destroy(conf->slab_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 conf->slab_cache = NULL;
1484}
1485
NeilBrown6712ecf2007-09-27 12:47:43 +02001486static void raid5_end_read_request(struct bio * bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
NeilBrown99c0fb52009-03-31 14:39:38 +11001488 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001490 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
NeilBrownd6950432006-07-10 04:44:20 -07001492 char b[BDEVNAME_SIZE];
1493 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
1496 for (i=0 ; i<disks; i++)
1497 if (bi == &sh->dev[i].req)
1498 break;
1499
Dan Williams45b42332007-07-09 11:56:43 -07001500 pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n",
1501 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 uptodate);
1503 if (i == disks) {
1504 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001505 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 }
1507
1508 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 set_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrown4e5314b2005-11-08 21:39:22 -08001510 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
NeilBrownd6950432006-07-10 04:44:20 -07001511 rdev = conf->disks[i].rdev;
NeilBrown0c55e022010-05-03 14:09:02 +10001512 printk_rl(KERN_INFO "md/raid:%s: read error corrected"
Bernd Schubert6be9d492008-05-23 13:04:34 -07001513 " (%lu sectors at %llu on %s)\n",
1514 mdname(conf->mddev), STRIPE_SECTORS,
1515 (unsigned long long)(sh->sector
1516 + rdev->data_offset),
1517 bdevname(rdev->bdev, b));
NeilBrown4e5314b2005-11-08 21:39:22 -08001518 clear_bit(R5_ReadError, &sh->dev[i].flags);
1519 clear_bit(R5_ReWrite, &sh->dev[i].flags);
1520 }
NeilBrownba22dcb2005-11-08 21:39:31 -08001521 if (atomic_read(&conf->disks[i].rdev->read_errors))
1522 atomic_set(&conf->disks[i].rdev->read_errors, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 } else {
NeilBrownd6950432006-07-10 04:44:20 -07001524 const char *bdn = bdevname(conf->disks[i].rdev->bdev, b);
NeilBrownba22dcb2005-11-08 21:39:31 -08001525 int retry = 0;
NeilBrownd6950432006-07-10 04:44:20 -07001526 rdev = conf->disks[i].rdev;
1527
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 clear_bit(R5_UPTODATE, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001529 atomic_inc(&rdev->read_errors);
Gabriele A. Trombetti7b0bb532010-04-28 11:51:17 +10001530 if (conf->mddev->degraded >= conf->max_degraded)
Bernd Schubert6be9d492008-05-23 13:04:34 -07001531 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001532 "md/raid:%s: read error not correctable "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001533 "(sector %llu on %s).\n",
1534 mdname(conf->mddev),
1535 (unsigned long long)(sh->sector
1536 + rdev->data_offset),
1537 bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001538 else if (test_bit(R5_ReWrite, &sh->dev[i].flags))
NeilBrown4e5314b2005-11-08 21:39:22 -08001539 /* Oh, no!!! */
Bernd Schubert6be9d492008-05-23 13:04:34 -07001540 printk_rl(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001541 "md/raid:%s: read error NOT corrected!! "
Bernd Schubert6be9d492008-05-23 13:04:34 -07001542 "(sector %llu on %s).\n",
1543 mdname(conf->mddev),
1544 (unsigned long long)(sh->sector
1545 + rdev->data_offset),
1546 bdn);
NeilBrownd6950432006-07-10 04:44:20 -07001547 else if (atomic_read(&rdev->read_errors)
NeilBrownba22dcb2005-11-08 21:39:31 -08001548 > conf->max_nr_stripes)
NeilBrown14f8d262006-01-06 00:20:14 -08001549 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10001550 "md/raid:%s: Too many read errors, failing device %s.\n",
NeilBrownd6950432006-07-10 04:44:20 -07001551 mdname(conf->mddev), bdn);
NeilBrownba22dcb2005-11-08 21:39:31 -08001552 else
1553 retry = 1;
1554 if (retry)
1555 set_bit(R5_ReadError, &sh->dev[i].flags);
1556 else {
NeilBrown4e5314b2005-11-08 21:39:22 -08001557 clear_bit(R5_ReadError, &sh->dev[i].flags);
1558 clear_bit(R5_ReWrite, &sh->dev[i].flags);
NeilBrownd6950432006-07-10 04:44:20 -07001559 md_error(conf->mddev, rdev);
NeilBrownba22dcb2005-11-08 21:39:31 -08001560 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 }
1562 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1564 set_bit(STRIPE_HANDLE, &sh->state);
1565 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566}
1567
NeilBrownd710e132008-10-13 11:55:12 +11001568static void raid5_end_write_request(struct bio *bi, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
NeilBrown99c0fb52009-03-31 14:39:38 +11001570 struct stripe_head *sh = bi->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 raid5_conf_t *conf = sh->raid_conf;
NeilBrown7ecaa1e2006-03-27 01:18:08 -08001572 int disks = sh->disks, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
1574
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 for (i=0 ; i<disks; i++)
1576 if (bi == &sh->dev[i].req)
1577 break;
1578
Dan Williams45b42332007-07-09 11:56:43 -07001579 pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 (unsigned long long)sh->sector, i, atomic_read(&sh->count),
1581 uptodate);
1582 if (i == disks) {
1583 BUG();
NeilBrown6712ecf2007-09-27 12:47:43 +02001584 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 }
1586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 if (!uptodate)
1588 md_error(conf->mddev, conf->disks[i].rdev);
1589
1590 rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
1591
1592 clear_bit(R5_LOCKED, &sh->dev[i].flags);
1593 set_bit(STRIPE_HANDLE, &sh->state);
NeilBrownc04be0a2006-10-03 01:15:53 -07001594 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595}
1596
1597
NeilBrown784052e2009-03-31 15:19:07 +11001598static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
NeilBrown784052e2009-03-31 15:19:07 +11001600static void raid5_build_block(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
1602 struct r5dev *dev = &sh->dev[i];
1603
1604 bio_init(&dev->req);
1605 dev->req.bi_io_vec = &dev->vec;
1606 dev->req.bi_vcnt++;
1607 dev->req.bi_max_vecs++;
1608 dev->vec.bv_page = dev->page;
1609 dev->vec.bv_len = STRIPE_SIZE;
1610 dev->vec.bv_offset = 0;
1611
1612 dev->req.bi_sector = sh->sector;
1613 dev->req.bi_private = sh;
1614
1615 dev->flags = 0;
NeilBrown784052e2009-03-31 15:19:07 +11001616 dev->sector = compute_blocknr(sh, i, previous);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617}
1618
1619static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1620{
1621 char b[BDEVNAME_SIZE];
H Hartley Sweeten7b928132010-03-08 16:02:40 +11001622 raid5_conf_t *conf = mddev->private;
NeilBrown0c55e022010-05-03 14:09:02 +10001623 pr_debug("raid456: error called\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
NeilBrownb2d444d2005-11-08 21:39:31 -08001625 if (!test_bit(Faulty, &rdev->flags)) {
NeilBrown850b2b42006-10-03 01:15:46 -07001626 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001627 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1628 unsigned long flags;
1629 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 mddev->degraded++;
NeilBrownc04be0a2006-10-03 01:15:53 -07001631 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 /*
1633 * if recovery was running, make sure it aborts.
1634 */
NeilBrowndfc70642008-05-23 13:04:39 -07001635 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 }
NeilBrownb2d444d2005-11-08 21:39:31 -08001637 set_bit(Faulty, &rdev->flags);
NeilBrownd710e132008-10-13 11:55:12 +11001638 printk(KERN_ALERT
NeilBrown0c55e022010-05-03 14:09:02 +10001639 "md/raid:%s: Disk failure on %s, disabling device.\n"
1640 KERN_ALERT
1641 "md/raid:%s: Operation continuing on %d devices.\n",
1642 mdname(mddev),
1643 bdevname(rdev->bdev, b),
1644 mdname(mddev),
1645 conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 }
NeilBrown16a53ec2006-06-26 00:27:38 -07001647}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
1649/*
1650 * Input: a 'big' sector number,
1651 * Output: index of the data and parity disk, and the sector # in them.
1652 */
NeilBrown112bf892009-03-31 14:39:38 +11001653static sector_t raid5_compute_sector(raid5_conf_t *conf, sector_t r_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11001654 int previous, int *dd_idx,
1655 struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
1657 long stripe;
1658 unsigned long chunk_number;
1659 unsigned int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11001660 int pd_idx, qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001661 int ddf_layout = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 sector_t new_sector;
NeilBrowne183eae2009-03-31 15:20:22 +11001663 int algorithm = previous ? conf->prev_algo
1664 : conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001665 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1666 : conf->chunk_sectors;
NeilBrown112bf892009-03-31 14:39:38 +11001667 int raid_disks = previous ? conf->previous_raid_disks
1668 : conf->raid_disks;
1669 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
1671 /* First compute the information on this sector */
1672
1673 /*
1674 * Compute the chunk number and the sector offset inside the chunk
1675 */
1676 chunk_offset = sector_div(r_sector, sectors_per_chunk);
1677 chunk_number = r_sector;
1678 BUG_ON(r_sector != chunk_number);
1679
1680 /*
1681 * Compute the stripe number
1682 */
1683 stripe = chunk_number / data_disks;
1684
1685 /*
1686 * Compute the data disk and parity disk indexes inside the stripe
1687 */
1688 *dd_idx = chunk_number % data_disks;
1689
1690 /*
1691 * Select the parity disk based on the user selected algorithm.
1692 */
NeilBrown911d4ee2009-03-31 14:39:38 +11001693 pd_idx = qd_idx = ~0;
NeilBrown16a53ec2006-06-26 00:27:38 -07001694 switch(conf->level) {
1695 case 4:
NeilBrown911d4ee2009-03-31 14:39:38 +11001696 pd_idx = data_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001697 break;
1698 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001699 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001701 pd_idx = data_disks - stripe % raid_disks;
1702 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 (*dd_idx)++;
1704 break;
1705 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001706 pd_idx = stripe % raid_disks;
1707 if (*dd_idx >= pd_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 (*dd_idx)++;
1709 break;
1710 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001711 pd_idx = data_disks - stripe % raid_disks;
1712 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 break;
1714 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001715 pd_idx = stripe % raid_disks;
1716 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001718 case ALGORITHM_PARITY_0:
1719 pd_idx = 0;
1720 (*dd_idx)++;
1721 break;
1722 case ALGORITHM_PARITY_N:
1723 pd_idx = data_disks;
1724 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001726 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001727 }
1728 break;
1729 case 6:
1730
NeilBrowne183eae2009-03-31 15:20:22 +11001731 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001732 case ALGORITHM_LEFT_ASYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001733 pd_idx = raid_disks - 1 - (stripe % raid_disks);
1734 qd_idx = pd_idx + 1;
1735 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001736 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001737 qd_idx = 0;
1738 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001739 (*dd_idx) += 2; /* D D P Q D */
1740 break;
1741 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001742 pd_idx = stripe % raid_disks;
1743 qd_idx = pd_idx + 1;
1744 if (pd_idx == raid_disks-1) {
NeilBrown99c0fb52009-03-31 14:39:38 +11001745 (*dd_idx)++; /* Q D D D P */
NeilBrown911d4ee2009-03-31 14:39:38 +11001746 qd_idx = 0;
1747 } else if (*dd_idx >= pd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001748 (*dd_idx) += 2; /* D D P Q D */
1749 break;
1750 case ALGORITHM_LEFT_SYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001751 pd_idx = raid_disks - 1 - (stripe % raid_disks);
1752 qd_idx = (pd_idx + 1) % raid_disks;
1753 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001754 break;
1755 case ALGORITHM_RIGHT_SYMMETRIC:
NeilBrown911d4ee2009-03-31 14:39:38 +11001756 pd_idx = stripe % raid_disks;
1757 qd_idx = (pd_idx + 1) % raid_disks;
1758 *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks;
NeilBrown16a53ec2006-06-26 00:27:38 -07001759 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001760
1761 case ALGORITHM_PARITY_0:
1762 pd_idx = 0;
1763 qd_idx = 1;
1764 (*dd_idx) += 2;
1765 break;
1766 case ALGORITHM_PARITY_N:
1767 pd_idx = data_disks;
1768 qd_idx = data_disks + 1;
1769 break;
1770
1771 case ALGORITHM_ROTATING_ZERO_RESTART:
1772 /* Exactly the same as RIGHT_ASYMMETRIC, but or
1773 * of blocks for computing Q is different.
1774 */
1775 pd_idx = stripe % raid_disks;
1776 qd_idx = pd_idx + 1;
1777 if (pd_idx == raid_disks-1) {
1778 (*dd_idx)++; /* Q D D D P */
1779 qd_idx = 0;
1780 } else if (*dd_idx >= pd_idx)
1781 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001782 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001783 break;
1784
1785 case ALGORITHM_ROTATING_N_RESTART:
1786 /* Same a left_asymmetric, by first stripe is
1787 * D D D P Q rather than
1788 * Q D D D P
1789 */
1790 pd_idx = raid_disks - 1 - ((stripe + 1) % raid_disks);
1791 qd_idx = pd_idx + 1;
1792 if (pd_idx == raid_disks-1) {
1793 (*dd_idx)++; /* Q D D D P */
1794 qd_idx = 0;
1795 } else if (*dd_idx >= pd_idx)
1796 (*dd_idx) += 2; /* D D P Q D */
NeilBrown67cc2b82009-03-31 14:39:38 +11001797 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001798 break;
1799
1800 case ALGORITHM_ROTATING_N_CONTINUE:
1801 /* Same as left_symmetric but Q is before P */
1802 pd_idx = raid_disks - 1 - (stripe % raid_disks);
1803 qd_idx = (pd_idx + raid_disks - 1) % raid_disks;
1804 *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks;
NeilBrown67cc2b82009-03-31 14:39:38 +11001805 ddf_layout = 1;
NeilBrown99c0fb52009-03-31 14:39:38 +11001806 break;
1807
1808 case ALGORITHM_LEFT_ASYMMETRIC_6:
1809 /* RAID5 left_asymmetric, with Q on last device */
1810 pd_idx = data_disks - stripe % (raid_disks-1);
1811 if (*dd_idx >= pd_idx)
1812 (*dd_idx)++;
1813 qd_idx = raid_disks - 1;
1814 break;
1815
1816 case ALGORITHM_RIGHT_ASYMMETRIC_6:
1817 pd_idx = stripe % (raid_disks-1);
1818 if (*dd_idx >= pd_idx)
1819 (*dd_idx)++;
1820 qd_idx = raid_disks - 1;
1821 break;
1822
1823 case ALGORITHM_LEFT_SYMMETRIC_6:
1824 pd_idx = data_disks - stripe % (raid_disks-1);
1825 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1826 qd_idx = raid_disks - 1;
1827 break;
1828
1829 case ALGORITHM_RIGHT_SYMMETRIC_6:
1830 pd_idx = stripe % (raid_disks-1);
1831 *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1);
1832 qd_idx = raid_disks - 1;
1833 break;
1834
1835 case ALGORITHM_PARITY_0_6:
1836 pd_idx = 0;
1837 (*dd_idx)++;
1838 qd_idx = raid_disks - 1;
1839 break;
1840
NeilBrown16a53ec2006-06-26 00:27:38 -07001841 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001842 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001843 }
1844 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 }
1846
NeilBrown911d4ee2009-03-31 14:39:38 +11001847 if (sh) {
1848 sh->pd_idx = pd_idx;
1849 sh->qd_idx = qd_idx;
NeilBrown67cc2b82009-03-31 14:39:38 +11001850 sh->ddf_layout = ddf_layout;
NeilBrown911d4ee2009-03-31 14:39:38 +11001851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 /*
1853 * Finally, compute the new sector number
1854 */
1855 new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset;
1856 return new_sector;
1857}
1858
1859
NeilBrown784052e2009-03-31 15:19:07 +11001860static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861{
1862 raid5_conf_t *conf = sh->raid_conf;
NeilBrownb875e532006-12-10 02:20:49 -08001863 int raid_disks = sh->disks;
1864 int data_disks = raid_disks - conf->max_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 sector_t new_sector = sh->sector, check;
Andre Noll09c9e5f2009-06-18 08:45:55 +10001866 int sectors_per_chunk = previous ? conf->prev_chunk_sectors
1867 : conf->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11001868 int algorithm = previous ? conf->prev_algo
1869 : conf->algorithm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 sector_t stripe;
1871 int chunk_offset;
NeilBrown911d4ee2009-03-31 14:39:38 +11001872 int chunk_number, dummy1, dd_idx = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 sector_t r_sector;
NeilBrown911d4ee2009-03-31 14:39:38 +11001874 struct stripe_head sh2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
NeilBrown16a53ec2006-06-26 00:27:38 -07001876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 chunk_offset = sector_div(new_sector, sectors_per_chunk);
1878 stripe = new_sector;
1879 BUG_ON(new_sector != stripe);
1880
NeilBrown16a53ec2006-06-26 00:27:38 -07001881 if (i == sh->pd_idx)
1882 return 0;
1883 switch(conf->level) {
1884 case 4: break;
1885 case 5:
NeilBrowne183eae2009-03-31 15:20:22 +11001886 switch (algorithm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 case ALGORITHM_LEFT_ASYMMETRIC:
1888 case ALGORITHM_RIGHT_ASYMMETRIC:
1889 if (i > sh->pd_idx)
1890 i--;
1891 break;
1892 case ALGORITHM_LEFT_SYMMETRIC:
1893 case ALGORITHM_RIGHT_SYMMETRIC:
1894 if (i < sh->pd_idx)
1895 i += raid_disks;
1896 i -= (sh->pd_idx + 1);
1897 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001898 case ALGORITHM_PARITY_0:
1899 i -= 1;
1900 break;
1901 case ALGORITHM_PARITY_N:
1902 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001904 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001905 }
1906 break;
1907 case 6:
NeilBrownd0dabf72009-03-31 14:39:38 +11001908 if (i == sh->qd_idx)
NeilBrown16a53ec2006-06-26 00:27:38 -07001909 return 0; /* It is the Q disk */
NeilBrowne183eae2009-03-31 15:20:22 +11001910 switch (algorithm) {
NeilBrown16a53ec2006-06-26 00:27:38 -07001911 case ALGORITHM_LEFT_ASYMMETRIC:
1912 case ALGORITHM_RIGHT_ASYMMETRIC:
NeilBrown99c0fb52009-03-31 14:39:38 +11001913 case ALGORITHM_ROTATING_ZERO_RESTART:
1914 case ALGORITHM_ROTATING_N_RESTART:
1915 if (sh->pd_idx == raid_disks-1)
1916 i--; /* Q D D D P */
NeilBrown16a53ec2006-06-26 00:27:38 -07001917 else if (i > sh->pd_idx)
1918 i -= 2; /* D D P Q D */
1919 break;
1920 case ALGORITHM_LEFT_SYMMETRIC:
1921 case ALGORITHM_RIGHT_SYMMETRIC:
1922 if (sh->pd_idx == raid_disks-1)
1923 i--; /* Q D D D P */
1924 else {
1925 /* D D P Q D */
1926 if (i < sh->pd_idx)
1927 i += raid_disks;
1928 i -= (sh->pd_idx + 2);
1929 }
1930 break;
NeilBrown99c0fb52009-03-31 14:39:38 +11001931 case ALGORITHM_PARITY_0:
1932 i -= 2;
1933 break;
1934 case ALGORITHM_PARITY_N:
1935 break;
1936 case ALGORITHM_ROTATING_N_CONTINUE:
NeilBrowne4424fe2009-10-16 16:27:34 +11001937 /* Like left_symmetric, but P is before Q */
NeilBrown99c0fb52009-03-31 14:39:38 +11001938 if (sh->pd_idx == 0)
1939 i--; /* P D D D Q */
NeilBrowne4424fe2009-10-16 16:27:34 +11001940 else {
1941 /* D D Q P D */
1942 if (i < sh->pd_idx)
1943 i += raid_disks;
1944 i -= (sh->pd_idx + 1);
1945 }
NeilBrown99c0fb52009-03-31 14:39:38 +11001946 break;
1947 case ALGORITHM_LEFT_ASYMMETRIC_6:
1948 case ALGORITHM_RIGHT_ASYMMETRIC_6:
1949 if (i > sh->pd_idx)
1950 i--;
1951 break;
1952 case ALGORITHM_LEFT_SYMMETRIC_6:
1953 case ALGORITHM_RIGHT_SYMMETRIC_6:
1954 if (i < sh->pd_idx)
1955 i += data_disks + 1;
1956 i -= (sh->pd_idx + 1);
1957 break;
1958 case ALGORITHM_PARITY_0_6:
1959 i -= 1;
1960 break;
NeilBrown16a53ec2006-06-26 00:27:38 -07001961 default:
NeilBrown99c0fb52009-03-31 14:39:38 +11001962 BUG();
NeilBrown16a53ec2006-06-26 00:27:38 -07001963 }
1964 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 }
1966
1967 chunk_number = stripe * data_disks + i;
1968 r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset;
1969
NeilBrown112bf892009-03-31 14:39:38 +11001970 check = raid5_compute_sector(conf, r_sector,
NeilBrown784052e2009-03-31 15:19:07 +11001971 previous, &dummy1, &sh2);
NeilBrown911d4ee2009-03-31 14:39:38 +11001972 if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx
1973 || sh2.qd_idx != sh->qd_idx) {
NeilBrown0c55e022010-05-03 14:09:02 +10001974 printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n",
1975 mdname(conf->mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 return 0;
1977 }
1978 return r_sector;
1979}
1980
1981
Dan Williams600aa102008-06-28 08:32:05 +10001982static void
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07001983schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
Dan Williams600aa102008-06-28 08:32:05 +10001984 int rcw, int expand)
Dan Williamse33129d2007-01-02 13:52:30 -07001985{
1986 int i, pd_idx = sh->pd_idx, disks = sh->disks;
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07001987 raid5_conf_t *conf = sh->raid_conf;
1988 int level = conf->level;
NeilBrown16a53ec2006-06-26 00:27:38 -07001989
Dan Williamse33129d2007-01-02 13:52:30 -07001990 if (rcw) {
1991 /* if we are not expanding this is a proper write request, and
1992 * there will be bios with new data to be drained into the
1993 * stripe cache
1994 */
1995 if (!expand) {
Dan Williams600aa102008-06-28 08:32:05 +10001996 sh->reconstruct_state = reconstruct_state_drain_run;
1997 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
1998 } else
1999 sh->reconstruct_state = reconstruct_state_run;
Dan Williamse33129d2007-01-02 13:52:30 -07002000
Dan Williamsac6b53b2009-07-14 13:40:19 -07002001 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002002
2003 for (i = disks; i--; ) {
2004 struct r5dev *dev = &sh->dev[i];
2005
2006 if (dev->towrite) {
2007 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsd8ee0722008-06-28 08:32:06 +10002008 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002009 if (!expand)
2010 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002011 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002012 }
2013 }
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002014 if (s->locked + conf->max_degraded == disks)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002015 if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002016 atomic_inc(&conf->pending_full_writes);
Dan Williamse33129d2007-01-02 13:52:30 -07002017 } else {
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002018 BUG_ON(level == 6);
Dan Williamse33129d2007-01-02 13:52:30 -07002019 BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) ||
2020 test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags)));
2021
Dan Williamsd8ee0722008-06-28 08:32:06 +10002022 sh->reconstruct_state = reconstruct_state_prexor_drain_run;
Dan Williams600aa102008-06-28 08:32:05 +10002023 set_bit(STRIPE_OP_PREXOR, &s->ops_request);
2024 set_bit(STRIPE_OP_BIODRAIN, &s->ops_request);
Dan Williamsac6b53b2009-07-14 13:40:19 -07002025 set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002026
2027 for (i = disks; i--; ) {
2028 struct r5dev *dev = &sh->dev[i];
2029 if (i == pd_idx)
2030 continue;
2031
Dan Williamse33129d2007-01-02 13:52:30 -07002032 if (dev->towrite &&
2033 (test_bit(R5_UPTODATE, &dev->flags) ||
Dan Williamsd8ee0722008-06-28 08:32:06 +10002034 test_bit(R5_Wantcompute, &dev->flags))) {
2035 set_bit(R5_Wantdrain, &dev->flags);
Dan Williamse33129d2007-01-02 13:52:30 -07002036 set_bit(R5_LOCKED, &dev->flags);
2037 clear_bit(R5_UPTODATE, &dev->flags);
Dan Williams600aa102008-06-28 08:32:05 +10002038 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002039 }
2040 }
2041 }
2042
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002043 /* keep the parity disk(s) locked while asynchronous operations
Dan Williamse33129d2007-01-02 13:52:30 -07002044 * are in flight
2045 */
2046 set_bit(R5_LOCKED, &sh->dev[pd_idx].flags);
2047 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
Dan Williams600aa102008-06-28 08:32:05 +10002048 s->locked++;
Dan Williamse33129d2007-01-02 13:52:30 -07002049
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002050 if (level == 6) {
2051 int qd_idx = sh->qd_idx;
2052 struct r5dev *dev = &sh->dev[qd_idx];
2053
2054 set_bit(R5_LOCKED, &dev->flags);
2055 clear_bit(R5_UPTODATE, &dev->flags);
2056 s->locked++;
2057 }
2058
Dan Williams600aa102008-06-28 08:32:05 +10002059 pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
Harvey Harrisone46b2722008-04-28 02:15:50 -07002060 __func__, (unsigned long long)sh->sector,
Dan Williams600aa102008-06-28 08:32:05 +10002061 s->locked, s->ops_request);
Dan Williamse33129d2007-01-02 13:52:30 -07002062}
NeilBrown16a53ec2006-06-26 00:27:38 -07002063
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064/*
2065 * Each stripe/dev can have one or more bion attached.
NeilBrown16a53ec2006-06-26 00:27:38 -07002066 * toread/towrite point to the first in a chain.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 * The bi_next chain must be in order.
2068 */
2069static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite)
2070{
2071 struct bio **bip;
2072 raid5_conf_t *conf = sh->raid_conf;
NeilBrown72626682005-09-09 16:23:54 -07002073 int firstwrite=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Dan Williams45b42332007-07-09 11:56:43 -07002075 pr_debug("adding bh b#%llu to stripe s#%llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 (unsigned long long)bi->bi_sector,
2077 (unsigned long long)sh->sector);
2078
2079
2080 spin_lock(&sh->lock);
2081 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07002082 if (forwrite) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 bip = &sh->dev[dd_idx].towrite;
NeilBrown72626682005-09-09 16:23:54 -07002084 if (*bip == NULL && sh->dev[dd_idx].written == NULL)
2085 firstwrite = 1;
2086 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 bip = &sh->dev[dd_idx].toread;
2088 while (*bip && (*bip)->bi_sector < bi->bi_sector) {
2089 if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector)
2090 goto overlap;
2091 bip = & (*bip)->bi_next;
2092 }
2093 if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9))
2094 goto overlap;
2095
Eric Sesterhenn78bafeb2006-04-02 13:31:42 +02002096 BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 if (*bip)
2098 bi->bi_next = *bip;
2099 *bip = bi;
Jens Axboe960e7392008-08-15 10:41:18 +02002100 bi->bi_phys_segments++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 spin_unlock_irq(&conf->device_lock);
2102 spin_unlock(&sh->lock);
2103
Dan Williams45b42332007-07-09 11:56:43 -07002104 pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 (unsigned long long)bi->bi_sector,
2106 (unsigned long long)sh->sector, dd_idx);
2107
NeilBrown72626682005-09-09 16:23:54 -07002108 if (conf->mddev->bitmap && firstwrite) {
NeilBrown72626682005-09-09 16:23:54 -07002109 bitmap_startwrite(conf->mddev->bitmap, sh->sector,
2110 STRIPE_SECTORS, 0);
NeilBrownae3c20c2006-07-10 04:44:17 -07002111 sh->bm_seq = conf->seq_flush+1;
NeilBrown72626682005-09-09 16:23:54 -07002112 set_bit(STRIPE_BIT_DELAY, &sh->state);
2113 }
2114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 if (forwrite) {
2116 /* check if page is covered */
2117 sector_t sector = sh->dev[dd_idx].sector;
2118 for (bi=sh->dev[dd_idx].towrite;
2119 sector < sh->dev[dd_idx].sector + STRIPE_SECTORS &&
2120 bi && bi->bi_sector <= sector;
2121 bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) {
2122 if (bi->bi_sector + (bi->bi_size>>9) >= sector)
2123 sector = bi->bi_sector + (bi->bi_size>>9);
2124 }
2125 if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS)
2126 set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags);
2127 }
2128 return 1;
2129
2130 overlap:
2131 set_bit(R5_Overlap, &sh->dev[dd_idx].flags);
2132 spin_unlock_irq(&conf->device_lock);
2133 spin_unlock(&sh->lock);
2134 return 0;
2135}
2136
NeilBrown29269552006-03-27 01:18:10 -08002137static void end_reshape(raid5_conf_t *conf);
2138
NeilBrown911d4ee2009-03-31 14:39:38 +11002139static void stripe_set_idx(sector_t stripe, raid5_conf_t *conf, int previous,
2140 struct stripe_head *sh)
NeilBrownccfcc3c2006-03-27 01:18:09 -08002141{
NeilBrown784052e2009-03-31 15:19:07 +11002142 int sectors_per_chunk =
Andre Noll09c9e5f2009-06-18 08:45:55 +10002143 previous ? conf->prev_chunk_sectors : conf->chunk_sectors;
NeilBrown911d4ee2009-03-31 14:39:38 +11002144 int dd_idx;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002145 int chunk_offset = sector_div(stripe, sectors_per_chunk);
NeilBrown112bf892009-03-31 14:39:38 +11002146 int disks = previous ? conf->previous_raid_disks : conf->raid_disks;
Coywolf Qi Hunt2d2063c2006-10-03 01:15:50 -07002147
NeilBrown112bf892009-03-31 14:39:38 +11002148 raid5_compute_sector(conf,
2149 stripe * (disks - conf->max_degraded)
NeilBrownb875e532006-12-10 02:20:49 -08002150 *sectors_per_chunk + chunk_offset,
NeilBrown112bf892009-03-31 14:39:38 +11002151 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11002152 &dd_idx, sh);
NeilBrownccfcc3c2006-03-27 01:18:09 -08002153}
2154
Dan Williamsa4456852007-07-09 11:56:43 -07002155static void
Dan Williams1fe797e2008-06-28 09:16:30 +10002156handle_failed_stripe(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002157 struct stripe_head_state *s, int disks,
2158 struct bio **return_bi)
2159{
2160 int i;
2161 for (i = disks; i--; ) {
2162 struct bio *bi;
2163 int bitmap_end = 0;
2164
2165 if (test_bit(R5_ReadError, &sh->dev[i].flags)) {
2166 mdk_rdev_t *rdev;
2167 rcu_read_lock();
2168 rdev = rcu_dereference(conf->disks[i].rdev);
2169 if (rdev && test_bit(In_sync, &rdev->flags))
2170 /* multiple read failures in one stripe */
2171 md_error(conf->mddev, rdev);
2172 rcu_read_unlock();
2173 }
2174 spin_lock_irq(&conf->device_lock);
2175 /* fail all writes first */
2176 bi = sh->dev[i].towrite;
2177 sh->dev[i].towrite = NULL;
2178 if (bi) {
2179 s->to_write--;
2180 bitmap_end = 1;
2181 }
2182
2183 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2184 wake_up(&conf->wait_for_overlap);
2185
2186 while (bi && bi->bi_sector <
2187 sh->dev[i].sector + STRIPE_SECTORS) {
2188 struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector);
2189 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002190 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002191 md_write_end(conf->mddev);
2192 bi->bi_next = *return_bi;
2193 *return_bi = bi;
2194 }
2195 bi = nextbi;
2196 }
2197 /* and fail all 'written' */
2198 bi = sh->dev[i].written;
2199 sh->dev[i].written = NULL;
2200 if (bi) bitmap_end = 1;
2201 while (bi && bi->bi_sector <
2202 sh->dev[i].sector + STRIPE_SECTORS) {
2203 struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector);
2204 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002205 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002206 md_write_end(conf->mddev);
2207 bi->bi_next = *return_bi;
2208 *return_bi = bi;
2209 }
2210 bi = bi2;
2211 }
2212
Dan Williamsb5e98d62007-01-02 13:52:31 -07002213 /* fail any reads if this device is non-operational and
2214 * the data has not reached the cache yet.
2215 */
2216 if (!test_bit(R5_Wantfill, &sh->dev[i].flags) &&
2217 (!test_bit(R5_Insync, &sh->dev[i].flags) ||
2218 test_bit(R5_ReadError, &sh->dev[i].flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002219 bi = sh->dev[i].toread;
2220 sh->dev[i].toread = NULL;
2221 if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
2222 wake_up(&conf->wait_for_overlap);
2223 if (bi) s->to_read--;
2224 while (bi && bi->bi_sector <
2225 sh->dev[i].sector + STRIPE_SECTORS) {
2226 struct bio *nextbi =
2227 r5_next_bio(bi, sh->dev[i].sector);
2228 clear_bit(BIO_UPTODATE, &bi->bi_flags);
Jens Axboe960e7392008-08-15 10:41:18 +02002229 if (!raid5_dec_bi_phys_segments(bi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002230 bi->bi_next = *return_bi;
2231 *return_bi = bi;
2232 }
2233 bi = nextbi;
2234 }
2235 }
2236 spin_unlock_irq(&conf->device_lock);
2237 if (bitmap_end)
2238 bitmap_endwrite(conf->mddev->bitmap, sh->sector,
2239 STRIPE_SECTORS, 0, 0);
2240 }
2241
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002242 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2243 if (atomic_dec_and_test(&conf->pending_full_writes))
2244 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002245}
2246
Dan Williams1fe797e2008-06-28 09:16:30 +10002247/* fetch_block5 - checks the given member device to see if its data needs
2248 * to be read or computed to satisfy a request.
2249 *
2250 * Returns 1 when no more member devices need to be checked, otherwise returns
2251 * 0 to tell the loop in handle_stripe_fill5 to continue
Dan Williamsf38e1212007-01-02 13:52:30 -07002252 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002253static int fetch_block5(struct stripe_head *sh, struct stripe_head_state *s,
2254 int disk_idx, int disks)
Dan Williamsf38e1212007-01-02 13:52:30 -07002255{
2256 struct r5dev *dev = &sh->dev[disk_idx];
2257 struct r5dev *failed_dev = &sh->dev[s->failed_num];
2258
Dan Williamsf38e1212007-01-02 13:52:30 -07002259 /* is the data in this block needed, and can we get it? */
2260 if (!test_bit(R5_LOCKED, &dev->flags) &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002261 !test_bit(R5_UPTODATE, &dev->flags) &&
2262 (dev->toread ||
2263 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2264 s->syncing || s->expanding ||
2265 (s->failed &&
2266 (failed_dev->toread ||
2267 (failed_dev->towrite &&
2268 !test_bit(R5_OVERWRITE, &failed_dev->flags)))))) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002269 /* We would like to get this block, possibly by computing it,
2270 * otherwise read it if the backing disk is insync
Dan Williamsf38e1212007-01-02 13:52:30 -07002271 */
2272 if ((s->uptodate == disks - 1) &&
Dan Williamsecc65c92008-06-28 08:31:57 +10002273 (s->failed && disk_idx == s->failed_num)) {
Dan Williams976ea8d2008-06-28 08:32:03 +10002274 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2275 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
Dan Williamsf38e1212007-01-02 13:52:30 -07002276 set_bit(R5_Wantcompute, &dev->flags);
2277 sh->ops.target = disk_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002278 sh->ops.target2 = -1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002279 s->req_compute = 1;
Dan Williamsf38e1212007-01-02 13:52:30 -07002280 /* Careful: from this point on 'uptodate' is in the eye
Dan Williamsac6b53b2009-07-14 13:40:19 -07002281 * of raid_run_ops which services 'compute' operations
Dan Williamsf38e1212007-01-02 13:52:30 -07002282 * before writes. R5_Wantcompute flags a block that will
2283 * be R5_UPTODATE by the time it is needed for a
2284 * subsequent operation.
2285 */
2286 s->uptodate++;
Dan Williams1fe797e2008-06-28 09:16:30 +10002287 return 1; /* uptodate + compute == disks */
Dan Williams7a1fc532008-07-10 04:54:57 -07002288 } else if (test_bit(R5_Insync, &dev->flags)) {
Dan Williamsf38e1212007-01-02 13:52:30 -07002289 set_bit(R5_LOCKED, &dev->flags);
2290 set_bit(R5_Wantread, &dev->flags);
Dan Williamsf38e1212007-01-02 13:52:30 -07002291 s->locked++;
2292 pr_debug("Reading block %d (sync=%d)\n", disk_idx,
2293 s->syncing);
2294 }
2295 }
2296
Dan Williams1fe797e2008-06-28 09:16:30 +10002297 return 0;
Dan Williamsf38e1212007-01-02 13:52:30 -07002298}
2299
Dan Williams1fe797e2008-06-28 09:16:30 +10002300/**
2301 * handle_stripe_fill5 - read or compute data to satisfy pending requests.
2302 */
2303static void handle_stripe_fill5(struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002304 struct stripe_head_state *s, int disks)
2305{
2306 int i;
Dan Williamsf38e1212007-01-02 13:52:30 -07002307
Dan Williamsf38e1212007-01-02 13:52:30 -07002308 /* look for blocks to read/compute, skip this if a compute
2309 * is already in flight, or if the stripe contents are in the
2310 * midst of changing due to a write
2311 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002312 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
Dan Williams1fe797e2008-06-28 09:16:30 +10002313 !sh->reconstruct_state)
Dan Williamsf38e1212007-01-02 13:52:30 -07002314 for (i = disks; i--; )
Dan Williams1fe797e2008-06-28 09:16:30 +10002315 if (fetch_block5(sh, s, i, disks))
Dan Williamsf38e1212007-01-02 13:52:30 -07002316 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002317 set_bit(STRIPE_HANDLE, &sh->state);
2318}
2319
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002320/* fetch_block6 - checks the given member device to see if its data needs
2321 * to be read or computed to satisfy a request.
2322 *
2323 * Returns 1 when no more member devices need to be checked, otherwise returns
2324 * 0 to tell the loop in handle_stripe_fill6 to continue
2325 */
2326static int fetch_block6(struct stripe_head *sh, struct stripe_head_state *s,
2327 struct r6_state *r6s, int disk_idx, int disks)
2328{
2329 struct r5dev *dev = &sh->dev[disk_idx];
2330 struct r5dev *fdev[2] = { &sh->dev[r6s->failed_num[0]],
2331 &sh->dev[r6s->failed_num[1]] };
2332
2333 if (!test_bit(R5_LOCKED, &dev->flags) &&
2334 !test_bit(R5_UPTODATE, &dev->flags) &&
2335 (dev->toread ||
2336 (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) ||
2337 s->syncing || s->expanding ||
2338 (s->failed >= 1 &&
2339 (fdev[0]->toread || s->to_write)) ||
2340 (s->failed >= 2 &&
2341 (fdev[1]->toread || s->to_write)))) {
2342 /* we would like to get this block, possibly by computing it,
2343 * otherwise read it if the backing disk is insync
2344 */
2345 BUG_ON(test_bit(R5_Wantcompute, &dev->flags));
2346 BUG_ON(test_bit(R5_Wantread, &dev->flags));
2347 if ((s->uptodate == disks - 1) &&
2348 (s->failed && (disk_idx == r6s->failed_num[0] ||
2349 disk_idx == r6s->failed_num[1]))) {
2350 /* have disk failed, and we're requested to fetch it;
2351 * do compute it
2352 */
2353 pr_debug("Computing stripe %llu block %d\n",
2354 (unsigned long long)sh->sector, disk_idx);
2355 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2356 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2357 set_bit(R5_Wantcompute, &dev->flags);
2358 sh->ops.target = disk_idx;
2359 sh->ops.target2 = -1; /* no 2nd target */
2360 s->req_compute = 1;
2361 s->uptodate++;
2362 return 1;
2363 } else if (s->uptodate == disks-2 && s->failed >= 2) {
2364 /* Computing 2-failure is *very* expensive; only
2365 * do it if failed >= 2
2366 */
2367 int other;
2368 for (other = disks; other--; ) {
2369 if (other == disk_idx)
2370 continue;
2371 if (!test_bit(R5_UPTODATE,
2372 &sh->dev[other].flags))
2373 break;
2374 }
2375 BUG_ON(other < 0);
2376 pr_debug("Computing stripe %llu blocks %d,%d\n",
2377 (unsigned long long)sh->sector,
2378 disk_idx, other);
2379 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2380 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2381 set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags);
2382 set_bit(R5_Wantcompute, &sh->dev[other].flags);
2383 sh->ops.target = disk_idx;
2384 sh->ops.target2 = other;
2385 s->uptodate += 2;
2386 s->req_compute = 1;
2387 return 1;
2388 } else if (test_bit(R5_Insync, &dev->flags)) {
2389 set_bit(R5_LOCKED, &dev->flags);
2390 set_bit(R5_Wantread, &dev->flags);
2391 s->locked++;
2392 pr_debug("Reading block %d (sync=%d)\n",
2393 disk_idx, s->syncing);
2394 }
2395 }
2396
2397 return 0;
2398}
2399
2400/**
2401 * handle_stripe_fill6 - read or compute data to satisfy pending requests.
2402 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002403static void handle_stripe_fill6(struct stripe_head *sh,
Dan Williamsa4456852007-07-09 11:56:43 -07002404 struct stripe_head_state *s, struct r6_state *r6s,
2405 int disks)
2406{
2407 int i;
Yuri Tikhonov5599bec2009-08-29 19:13:12 -07002408
2409 /* look for blocks to read/compute, skip this if a compute
2410 * is already in flight, or if the stripe contents are in the
2411 * midst of changing due to a write
2412 */
2413 if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state &&
2414 !sh->reconstruct_state)
2415 for (i = disks; i--; )
2416 if (fetch_block6(sh, s, r6s, i, disks))
2417 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002418 set_bit(STRIPE_HANDLE, &sh->state);
2419}
2420
2421
Dan Williams1fe797e2008-06-28 09:16:30 +10002422/* handle_stripe_clean_event
Dan Williamsa4456852007-07-09 11:56:43 -07002423 * any written block on an uptodate or failed drive can be returned.
2424 * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but
2425 * never LOCKED, so we don't need to test 'failed' directly.
2426 */
Dan Williams1fe797e2008-06-28 09:16:30 +10002427static void handle_stripe_clean_event(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002428 struct stripe_head *sh, int disks, struct bio **return_bi)
2429{
2430 int i;
2431 struct r5dev *dev;
2432
2433 for (i = disks; i--; )
2434 if (sh->dev[i].written) {
2435 dev = &sh->dev[i];
2436 if (!test_bit(R5_LOCKED, &dev->flags) &&
2437 test_bit(R5_UPTODATE, &dev->flags)) {
2438 /* We can return any write requests */
2439 struct bio *wbi, *wbi2;
2440 int bitmap_end = 0;
Dan Williams45b42332007-07-09 11:56:43 -07002441 pr_debug("Return write for disc %d\n", i);
Dan Williamsa4456852007-07-09 11:56:43 -07002442 spin_lock_irq(&conf->device_lock);
2443 wbi = dev->written;
2444 dev->written = NULL;
2445 while (wbi && wbi->bi_sector <
2446 dev->sector + STRIPE_SECTORS) {
2447 wbi2 = r5_next_bio(wbi, dev->sector);
Jens Axboe960e7392008-08-15 10:41:18 +02002448 if (!raid5_dec_bi_phys_segments(wbi)) {
Dan Williamsa4456852007-07-09 11:56:43 -07002449 md_write_end(conf->mddev);
2450 wbi->bi_next = *return_bi;
2451 *return_bi = wbi;
2452 }
2453 wbi = wbi2;
2454 }
2455 if (dev->towrite == NULL)
2456 bitmap_end = 1;
2457 spin_unlock_irq(&conf->device_lock);
2458 if (bitmap_end)
2459 bitmap_endwrite(conf->mddev->bitmap,
2460 sh->sector,
2461 STRIPE_SECTORS,
2462 !test_bit(STRIPE_DEGRADED, &sh->state),
2463 0);
2464 }
2465 }
Dan Williams8b3e6cd2008-04-28 02:15:53 -07002466
2467 if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state))
2468 if (atomic_dec_and_test(&conf->pending_full_writes))
2469 md_wakeup_thread(conf->mddev->thread);
Dan Williamsa4456852007-07-09 11:56:43 -07002470}
2471
Dan Williams1fe797e2008-06-28 09:16:30 +10002472static void handle_stripe_dirtying5(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002473 struct stripe_head *sh, struct stripe_head_state *s, int disks)
2474{
2475 int rmw = 0, rcw = 0, i;
2476 for (i = disks; i--; ) {
2477 /* would I have to read this buffer for read_modify_write */
2478 struct r5dev *dev = &sh->dev[i];
2479 if ((dev->towrite || i == sh->pd_idx) &&
2480 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002481 !(test_bit(R5_UPTODATE, &dev->flags) ||
2482 test_bit(R5_Wantcompute, &dev->flags))) {
Dan Williamsa4456852007-07-09 11:56:43 -07002483 if (test_bit(R5_Insync, &dev->flags))
2484 rmw++;
2485 else
2486 rmw += 2*disks; /* cannot read it */
2487 }
2488 /* Would I have to read this buffer for reconstruct_write */
2489 if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx &&
2490 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002491 !(test_bit(R5_UPTODATE, &dev->flags) ||
2492 test_bit(R5_Wantcompute, &dev->flags))) {
2493 if (test_bit(R5_Insync, &dev->flags)) rcw++;
Dan Williamsa4456852007-07-09 11:56:43 -07002494 else
2495 rcw += 2*disks;
2496 }
2497 }
Dan Williams45b42332007-07-09 11:56:43 -07002498 pr_debug("for sector %llu, rmw=%d rcw=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07002499 (unsigned long long)sh->sector, rmw, rcw);
2500 set_bit(STRIPE_HANDLE, &sh->state);
2501 if (rmw < rcw && rmw > 0)
2502 /* prefer read-modify-write, but need to get some data */
2503 for (i = disks; i--; ) {
2504 struct r5dev *dev = &sh->dev[i];
2505 if ((dev->towrite || i == sh->pd_idx) &&
2506 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002507 !(test_bit(R5_UPTODATE, &dev->flags) ||
2508 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002509 test_bit(R5_Insync, &dev->flags)) {
2510 if (
2511 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002512 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002513 "%d for r-m-w\n", i);
2514 set_bit(R5_LOCKED, &dev->flags);
2515 set_bit(R5_Wantread, &dev->flags);
2516 s->locked++;
2517 } else {
2518 set_bit(STRIPE_DELAYED, &sh->state);
2519 set_bit(STRIPE_HANDLE, &sh->state);
2520 }
2521 }
2522 }
2523 if (rcw <= rmw && rcw > 0)
2524 /* want reconstruct write, but need to get some data */
2525 for (i = disks; i--; ) {
2526 struct r5dev *dev = &sh->dev[i];
2527 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2528 i != sh->pd_idx &&
2529 !test_bit(R5_LOCKED, &dev->flags) &&
Dan Williamsf38e1212007-01-02 13:52:30 -07002530 !(test_bit(R5_UPTODATE, &dev->flags) ||
2531 test_bit(R5_Wantcompute, &dev->flags)) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002532 test_bit(R5_Insync, &dev->flags)) {
2533 if (
2534 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
Dan Williams45b42332007-07-09 11:56:43 -07002535 pr_debug("Read_old block "
Dan Williamsa4456852007-07-09 11:56:43 -07002536 "%d for Reconstruct\n", i);
2537 set_bit(R5_LOCKED, &dev->flags);
2538 set_bit(R5_Wantread, &dev->flags);
2539 s->locked++;
2540 } else {
2541 set_bit(STRIPE_DELAYED, &sh->state);
2542 set_bit(STRIPE_HANDLE, &sh->state);
2543 }
2544 }
2545 }
2546 /* now if nothing is locked, and if we have enough data,
2547 * we can start a write request
2548 */
Dan Williamsf38e1212007-01-02 13:52:30 -07002549 /* since handle_stripe can be called at any time we need to handle the
2550 * case where a compute block operation has been submitted and then a
Dan Williamsac6b53b2009-07-14 13:40:19 -07002551 * subsequent call wants to start a write request. raid_run_ops only
2552 * handles the case where compute block and reconstruct are requested
Dan Williamsf38e1212007-01-02 13:52:30 -07002553 * simultaneously. If this is not the case then new writes need to be
2554 * held off until the compute completes.
2555 */
Dan Williams976ea8d2008-06-28 08:32:03 +10002556 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2557 (s->locked == 0 && (rcw == 0 || rmw == 0) &&
2558 !test_bit(STRIPE_BIT_DELAY, &sh->state)))
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07002559 schedule_reconstruction(sh, s, rcw == 0, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002560}
2561
Dan Williams1fe797e2008-06-28 09:16:30 +10002562static void handle_stripe_dirtying6(raid5_conf_t *conf,
Dan Williamsa4456852007-07-09 11:56:43 -07002563 struct stripe_head *sh, struct stripe_head_state *s,
2564 struct r6_state *r6s, int disks)
2565{
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002566 int rcw = 0, pd_idx = sh->pd_idx, i;
NeilBrown34e04e82009-03-31 15:10:16 +11002567 int qd_idx = sh->qd_idx;
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002568
2569 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002570 for (i = disks; i--; ) {
2571 struct r5dev *dev = &sh->dev[i];
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002572 /* check if we haven't enough data */
2573 if (!test_bit(R5_OVERWRITE, &dev->flags) &&
2574 i != pd_idx && i != qd_idx &&
2575 !test_bit(R5_LOCKED, &dev->flags) &&
2576 !(test_bit(R5_UPTODATE, &dev->flags) ||
2577 test_bit(R5_Wantcompute, &dev->flags))) {
2578 rcw++;
2579 if (!test_bit(R5_Insync, &dev->flags))
2580 continue; /* it's a failed drive */
2581
2582 if (
2583 test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) {
2584 pr_debug("Read_old stripe %llu "
2585 "block %d for Reconstruct\n",
2586 (unsigned long long)sh->sector, i);
2587 set_bit(R5_LOCKED, &dev->flags);
2588 set_bit(R5_Wantread, &dev->flags);
2589 s->locked++;
2590 } else {
2591 pr_debug("Request delayed stripe %llu "
2592 "block %d for Reconstruct\n",
2593 (unsigned long long)sh->sector, i);
2594 set_bit(STRIPE_DELAYED, &sh->state);
2595 set_bit(STRIPE_HANDLE, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002596 }
2597 }
2598 }
Dan Williamsa4456852007-07-09 11:56:43 -07002599 /* now if nothing is locked, and if we have enough data, we can start a
2600 * write request
2601 */
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002602 if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) &&
2603 s->locked == 0 && rcw == 0 &&
Dan Williamsa4456852007-07-09 11:56:43 -07002604 !test_bit(STRIPE_BIT_DELAY, &sh->state)) {
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07002605 schedule_reconstruction(sh, s, 1, 0);
Dan Williamsa4456852007-07-09 11:56:43 -07002606 }
2607}
2608
2609static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh,
2610 struct stripe_head_state *s, int disks)
2611{
Dan Williamsecc65c92008-06-28 08:31:57 +10002612 struct r5dev *dev = NULL;
Dan Williamse89f8962007-01-02 13:52:31 -07002613
Dan Williamsbd2ab672008-04-10 21:29:27 -07002614 set_bit(STRIPE_HANDLE, &sh->state);
2615
Dan Williamsecc65c92008-06-28 08:31:57 +10002616 switch (sh->check_state) {
2617 case check_state_idle:
2618 /* start a new check operation if there are no failures */
Dan Williamsbd2ab672008-04-10 21:29:27 -07002619 if (s->failed == 0) {
Dan Williamsbd2ab672008-04-10 21:29:27 -07002620 BUG_ON(s->uptodate != disks);
Dan Williamsecc65c92008-06-28 08:31:57 +10002621 sh->check_state = check_state_run;
2622 set_bit(STRIPE_OP_CHECK, &s->ops_request);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002623 clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags);
Dan Williamsbd2ab672008-04-10 21:29:27 -07002624 s->uptodate--;
Dan Williamsecc65c92008-06-28 08:31:57 +10002625 break;
Dan Williamsbd2ab672008-04-10 21:29:27 -07002626 }
Dan Williamsa4456852007-07-09 11:56:43 -07002627 dev = &sh->dev[s->failed_num];
Dan Williamsecc65c92008-06-28 08:31:57 +10002628 /* fall through */
2629 case check_state_compute_result:
2630 sh->check_state = check_state_idle;
2631 if (!dev)
2632 dev = &sh->dev[sh->pd_idx];
2633
2634 /* check that a write has not made the stripe insync */
2635 if (test_bit(STRIPE_INSYNC, &sh->state))
2636 break;
2637
2638 /* either failed parity check, or recovery is happening */
Dan Williamsa4456852007-07-09 11:56:43 -07002639 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
2640 BUG_ON(s->uptodate != disks);
2641
2642 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsecc65c92008-06-28 08:31:57 +10002643 s->locked++;
Dan Williamsa4456852007-07-09 11:56:43 -07002644 set_bit(R5_Wantwrite, &dev->flags);
Dan Williams830ea012007-01-02 13:52:31 -07002645
Dan Williamsa4456852007-07-09 11:56:43 -07002646 clear_bit(STRIPE_DEGRADED, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07002647 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002648 break;
2649 case check_state_run:
2650 break; /* we will be called again upon completion */
2651 case check_state_check_result:
2652 sh->check_state = check_state_idle;
2653
2654 /* if a failure occurred during the check operation, leave
2655 * STRIPE_INSYNC not set and let the stripe be handled again
2656 */
2657 if (s->failed)
2658 break;
2659
2660 /* handle a successful check operation, if parity is correct
2661 * we are done. Otherwise update the mismatch count and repair
2662 * parity if !MD_RECOVERY_CHECK
2663 */
Dan Williamsad283ea2009-08-29 19:09:26 -07002664 if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0)
Dan Williamsecc65c92008-06-28 08:31:57 +10002665 /* parity is correct (on disc,
2666 * not in buffer any more)
2667 */
2668 set_bit(STRIPE_INSYNC, &sh->state);
2669 else {
2670 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2671 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2672 /* don't try to repair!! */
2673 set_bit(STRIPE_INSYNC, &sh->state);
2674 else {
2675 sh->check_state = check_state_compute_run;
Dan Williams976ea8d2008-06-28 08:32:03 +10002676 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
Dan Williamsecc65c92008-06-28 08:31:57 +10002677 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2678 set_bit(R5_Wantcompute,
2679 &sh->dev[sh->pd_idx].flags);
2680 sh->ops.target = sh->pd_idx;
Dan Williamsac6b53b2009-07-14 13:40:19 -07002681 sh->ops.target2 = -1;
Dan Williamsecc65c92008-06-28 08:31:57 +10002682 s->uptodate++;
2683 }
2684 }
2685 break;
2686 case check_state_compute_run:
2687 break;
2688 default:
2689 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2690 __func__, sh->check_state,
2691 (unsigned long long) sh->sector);
2692 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002693 }
2694}
2695
2696
2697static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh,
Dan Williams36d1c642009-07-14 11:48:22 -07002698 struct stripe_head_state *s,
2699 struct r6_state *r6s, int disks)
Dan Williamsa4456852007-07-09 11:56:43 -07002700{
Dan Williamsa4456852007-07-09 11:56:43 -07002701 int pd_idx = sh->pd_idx;
NeilBrown34e04e82009-03-31 15:10:16 +11002702 int qd_idx = sh->qd_idx;
Dan Williamsd82dfee2009-07-14 13:40:57 -07002703 struct r5dev *dev;
Dan Williamsa4456852007-07-09 11:56:43 -07002704
2705 set_bit(STRIPE_HANDLE, &sh->state);
2706
2707 BUG_ON(s->failed > 2);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002708
Dan Williamsa4456852007-07-09 11:56:43 -07002709 /* Want to check and possibly repair P and Q.
2710 * However there could be one 'failed' device, in which
2711 * case we can only check one of them, possibly using the
2712 * other to generate missing data
2713 */
2714
Dan Williamsd82dfee2009-07-14 13:40:57 -07002715 switch (sh->check_state) {
2716 case check_state_idle:
2717 /* start a new check operation if there are < 2 failures */
Dan Williamsa4456852007-07-09 11:56:43 -07002718 if (s->failed == r6s->q_failed) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002719 /* The only possible failed device holds Q, so it
Dan Williamsa4456852007-07-09 11:56:43 -07002720 * makes sense to check P (If anything else were failed,
2721 * we would have used P to recreate it).
2722 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002723 sh->check_state = check_state_run;
Dan Williamsa4456852007-07-09 11:56:43 -07002724 }
2725 if (!r6s->q_failed && s->failed < 2) {
Dan Williamsd82dfee2009-07-14 13:40:57 -07002726 /* Q is not failed, and we didn't use it to generate
Dan Williamsa4456852007-07-09 11:56:43 -07002727 * anything, so it makes sense to check it
2728 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002729 if (sh->check_state == check_state_run)
2730 sh->check_state = check_state_run_pq;
2731 else
2732 sh->check_state = check_state_run_q;
Dan Williamsa4456852007-07-09 11:56:43 -07002733 }
Dan Williams36d1c642009-07-14 11:48:22 -07002734
Dan Williamsd82dfee2009-07-14 13:40:57 -07002735 /* discard potentially stale zero_sum_result */
2736 sh->ops.zero_sum_result = 0;
Dan Williams36d1c642009-07-14 11:48:22 -07002737
Dan Williamsd82dfee2009-07-14 13:40:57 -07002738 if (sh->check_state == check_state_run) {
2739 /* async_xor_zero_sum destroys the contents of P */
2740 clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags);
2741 s->uptodate--;
Dan Williamsa4456852007-07-09 11:56:43 -07002742 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002743 if (sh->check_state >= check_state_run &&
2744 sh->check_state <= check_state_run_pq) {
2745 /* async_syndrome_zero_sum preserves P and Q, so
2746 * no need to mark them !uptodate here
2747 */
2748 set_bit(STRIPE_OP_CHECK, &s->ops_request);
2749 break;
2750 }
Dan Williams36d1c642009-07-14 11:48:22 -07002751
Dan Williamsd82dfee2009-07-14 13:40:57 -07002752 /* we have 2-disk failure */
2753 BUG_ON(s->failed != 2);
2754 /* fall through */
2755 case check_state_compute_result:
2756 sh->check_state = check_state_idle;
Dan Williams36d1c642009-07-14 11:48:22 -07002757
Dan Williamsd82dfee2009-07-14 13:40:57 -07002758 /* check that a write has not made the stripe insync */
2759 if (test_bit(STRIPE_INSYNC, &sh->state))
2760 break;
Dan Williamsa4456852007-07-09 11:56:43 -07002761
2762 /* now write out any block on a failed drive,
Dan Williamsd82dfee2009-07-14 13:40:57 -07002763 * or P or Q if they were recomputed
Dan Williamsa4456852007-07-09 11:56:43 -07002764 */
Dan Williamsd82dfee2009-07-14 13:40:57 -07002765 BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */
Dan Williamsa4456852007-07-09 11:56:43 -07002766 if (s->failed == 2) {
2767 dev = &sh->dev[r6s->failed_num[1]];
2768 s->locked++;
2769 set_bit(R5_LOCKED, &dev->flags);
2770 set_bit(R5_Wantwrite, &dev->flags);
2771 }
2772 if (s->failed >= 1) {
2773 dev = &sh->dev[r6s->failed_num[0]];
2774 s->locked++;
2775 set_bit(R5_LOCKED, &dev->flags);
2776 set_bit(R5_Wantwrite, &dev->flags);
2777 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002778 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002779 dev = &sh->dev[pd_idx];
2780 s->locked++;
2781 set_bit(R5_LOCKED, &dev->flags);
2782 set_bit(R5_Wantwrite, &dev->flags);
2783 }
Dan Williamsd82dfee2009-07-14 13:40:57 -07002784 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
Dan Williamsa4456852007-07-09 11:56:43 -07002785 dev = &sh->dev[qd_idx];
2786 s->locked++;
2787 set_bit(R5_LOCKED, &dev->flags);
2788 set_bit(R5_Wantwrite, &dev->flags);
2789 }
2790 clear_bit(STRIPE_DEGRADED, &sh->state);
2791
2792 set_bit(STRIPE_INSYNC, &sh->state);
Dan Williamsd82dfee2009-07-14 13:40:57 -07002793 break;
2794 case check_state_run:
2795 case check_state_run_q:
2796 case check_state_run_pq:
2797 break; /* we will be called again upon completion */
2798 case check_state_check_result:
2799 sh->check_state = check_state_idle;
2800
2801 /* handle a successful check operation, if parity is correct
2802 * we are done. Otherwise update the mismatch count and repair
2803 * parity if !MD_RECOVERY_CHECK
2804 */
2805 if (sh->ops.zero_sum_result == 0) {
2806 /* both parities are correct */
2807 if (!s->failed)
2808 set_bit(STRIPE_INSYNC, &sh->state);
2809 else {
2810 /* in contrast to the raid5 case we can validate
2811 * parity, but still have a failure to write
2812 * back
2813 */
2814 sh->check_state = check_state_compute_result;
2815 /* Returning at this point means that we may go
2816 * off and bring p and/or q uptodate again so
2817 * we make sure to check zero_sum_result again
2818 * to verify if p or q need writeback
2819 */
2820 }
2821 } else {
2822 conf->mddev->resync_mismatches += STRIPE_SECTORS;
2823 if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery))
2824 /* don't try to repair!! */
2825 set_bit(STRIPE_INSYNC, &sh->state);
2826 else {
2827 int *target = &sh->ops.target;
2828
2829 sh->ops.target = -1;
2830 sh->ops.target2 = -1;
2831 sh->check_state = check_state_compute_run;
2832 set_bit(STRIPE_COMPUTE_RUN, &sh->state);
2833 set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request);
2834 if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) {
2835 set_bit(R5_Wantcompute,
2836 &sh->dev[pd_idx].flags);
2837 *target = pd_idx;
2838 target = &sh->ops.target2;
2839 s->uptodate++;
2840 }
2841 if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) {
2842 set_bit(R5_Wantcompute,
2843 &sh->dev[qd_idx].flags);
2844 *target = qd_idx;
2845 s->uptodate++;
2846 }
2847 }
2848 }
2849 break;
2850 case check_state_compute_run:
2851 break;
2852 default:
2853 printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n",
2854 __func__, sh->check_state,
2855 (unsigned long long) sh->sector);
2856 BUG();
Dan Williamsa4456852007-07-09 11:56:43 -07002857 }
2858}
2859
2860static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh,
2861 struct r6_state *r6s)
2862{
2863 int i;
2864
2865 /* We have read all the blocks in this stripe and now we need to
2866 * copy some of them into a target stripe for expand.
2867 */
Dan Williamsf0a50d32007-01-02 13:52:31 -07002868 struct dma_async_tx_descriptor *tx = NULL;
Dan Williamsa4456852007-07-09 11:56:43 -07002869 clear_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2870 for (i = 0; i < sh->disks; i++)
NeilBrown34e04e82009-03-31 15:10:16 +11002871 if (i != sh->pd_idx && i != sh->qd_idx) {
NeilBrown911d4ee2009-03-31 14:39:38 +11002872 int dd_idx, j;
Dan Williamsa4456852007-07-09 11:56:43 -07002873 struct stripe_head *sh2;
Dan Williamsa08abd82009-06-03 11:43:59 -07002874 struct async_submit_ctl submit;
Dan Williamsa4456852007-07-09 11:56:43 -07002875
NeilBrown784052e2009-03-31 15:19:07 +11002876 sector_t bn = compute_blocknr(sh, i, 1);
NeilBrown911d4ee2009-03-31 14:39:38 +11002877 sector_t s = raid5_compute_sector(conf, bn, 0,
2878 &dd_idx, NULL);
NeilBrowna8c906c2009-06-09 14:39:59 +10002879 sh2 = get_active_stripe(conf, s, 0, 1, 1);
Dan Williamsa4456852007-07-09 11:56:43 -07002880 if (sh2 == NULL)
2881 /* so far only the early blocks of this stripe
2882 * have been requested. When later blocks
2883 * get requested, we will try again
2884 */
2885 continue;
2886 if (!test_bit(STRIPE_EXPANDING, &sh2->state) ||
2887 test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) {
2888 /* must have already done this block */
2889 release_stripe(sh2);
2890 continue;
2891 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07002892
2893 /* place all the copies on one channel */
Dan Williamsa08abd82009-06-03 11:43:59 -07002894 init_async_submit(&submit, 0, tx, NULL, NULL, NULL);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002895 tx = async_memcpy(sh2->dev[dd_idx].page,
Dan Williams88ba2aa2009-04-09 16:16:18 -07002896 sh->dev[i].page, 0, 0, STRIPE_SIZE,
Dan Williamsa08abd82009-06-03 11:43:59 -07002897 &submit);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002898
Dan Williamsa4456852007-07-09 11:56:43 -07002899 set_bit(R5_Expanded, &sh2->dev[dd_idx].flags);
2900 set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags);
2901 for (j = 0; j < conf->raid_disks; j++)
2902 if (j != sh2->pd_idx &&
NeilBrownd0dabf72009-03-31 14:39:38 +11002903 (!r6s || j != sh2->qd_idx) &&
Dan Williamsa4456852007-07-09 11:56:43 -07002904 !test_bit(R5_Expanded, &sh2->dev[j].flags))
2905 break;
2906 if (j == conf->raid_disks) {
2907 set_bit(STRIPE_EXPAND_READY, &sh2->state);
2908 set_bit(STRIPE_HANDLE, &sh2->state);
2909 }
2910 release_stripe(sh2);
Dan Williamsf0a50d32007-01-02 13:52:31 -07002911
Dan Williamsa4456852007-07-09 11:56:43 -07002912 }
NeilBrowna2e08552007-09-11 15:23:36 -07002913 /* done submitting copies, wait for them to complete */
2914 if (tx) {
2915 async_tx_ack(tx);
2916 dma_wait_for_async_tx(tx);
2917 }
Dan Williamsa4456852007-07-09 11:56:43 -07002918}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
Dan Williams6bfe0b42008-04-30 00:52:32 -07002920
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921/*
2922 * handle_stripe - do things to a stripe.
2923 *
2924 * We lock the stripe and then examine the state of various bits
2925 * to see what needs to be done.
2926 * Possible results:
2927 * return some read request which now have data
2928 * return some write requests which are safely on disc
2929 * schedule a read on some buffers
2930 * schedule a write of some buffers
2931 * return confirmation of parity correctness
2932 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 * buffers are taken off read_list or write_list, and bh_cache buffers
2934 * get BH_Lock set before the stripe lock is released.
2935 *
2936 */
Dan Williamsa4456852007-07-09 11:56:43 -07002937
NeilBrown14425772009-10-16 15:55:25 +11002938static void handle_stripe5(struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939{
2940 raid5_conf_t *conf = sh->raid_conf;
Dan Williamsa4456852007-07-09 11:56:43 -07002941 int disks = sh->disks, i;
2942 struct bio *return_bi = NULL;
2943 struct stripe_head_state s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 struct r5dev *dev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07002945 mdk_rdev_t *blocked_rdev = NULL;
Dan Williamse0a115e2008-06-05 22:45:52 -07002946 int prexor;
NeilBrown729a1862009-12-14 12:49:50 +11002947 int dec_preread_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948
Dan Williamsa4456852007-07-09 11:56:43 -07002949 memset(&s, 0, sizeof(s));
Dan Williams600aa102008-06-28 08:32:05 +10002950 pr_debug("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d check:%d "
2951 "reconstruct:%d\n", (unsigned long long)sh->sector, sh->state,
2952 atomic_read(&sh->count), sh->pd_idx, sh->check_state,
2953 sh->reconstruct_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954
2955 spin_lock(&sh->lock);
2956 clear_bit(STRIPE_HANDLE, &sh->state);
2957 clear_bit(STRIPE_DELAYED, &sh->state);
2958
Dan Williamsa4456852007-07-09 11:56:43 -07002959 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
2960 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
2961 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
Dan Williams83de75c2008-06-28 08:31:58 +10002962
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 /* Now to look around and see what can be done */
NeilBrown9910f162006-01-06 00:20:24 -08002964 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 for (i=disks; i--; ) {
2966 mdk_rdev_t *rdev;
NeilBrowna9f326e2009-09-23 18:06:41 +10002967
2968 dev = &sh->dev[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 clear_bit(R5_Insync, &dev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
Dan Williamsb5e98d62007-01-02 13:52:31 -07002971 pr_debug("check %d: state 0x%lx toread %p read %p write %p "
2972 "written %p\n", i, dev->flags, dev->toread, dev->read,
2973 dev->towrite, dev->written);
2974
2975 /* maybe we can request a biofill operation
2976 *
2977 * new wantfill requests are only permitted while
Dan Williams83de75c2008-06-28 08:31:58 +10002978 * ops_complete_biofill is guaranteed to be inactive
Dan Williamsb5e98d62007-01-02 13:52:31 -07002979 */
2980 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
Dan Williams83de75c2008-06-28 08:31:58 +10002981 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
Dan Williamsb5e98d62007-01-02 13:52:31 -07002982 set_bit(R5_Wantfill, &dev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983
2984 /* now count some things */
Dan Williamsa4456852007-07-09 11:56:43 -07002985 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
2986 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
Dan Williamsf38e1212007-01-02 13:52:30 -07002987 if (test_bit(R5_Wantcompute, &dev->flags)) s.compute++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
Dan Williamsb5e98d62007-01-02 13:52:31 -07002989 if (test_bit(R5_Wantfill, &dev->flags))
2990 s.to_fill++;
2991 else if (dev->toread)
Dan Williamsa4456852007-07-09 11:56:43 -07002992 s.to_read++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 if (dev->towrite) {
Dan Williamsa4456852007-07-09 11:56:43 -07002994 s.to_write++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 if (!test_bit(R5_OVERWRITE, &dev->flags))
Dan Williamsa4456852007-07-09 11:56:43 -07002996 s.non_overwrite++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 }
Dan Williamsa4456852007-07-09 11:56:43 -07002998 if (dev->written)
2999 s.written++;
NeilBrown9910f162006-01-06 00:20:24 -08003000 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownac4090d2008-08-05 15:54:13 +10003001 if (blocked_rdev == NULL &&
3002 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07003003 blocked_rdev = rdev;
3004 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003005 }
NeilBrownb2d444d2005-11-08 21:39:31 -08003006 if (!rdev || !test_bit(In_sync, &rdev->flags)) {
NeilBrown14f8d262006-01-06 00:20:14 -08003007 /* The ReadError flag will just be confusing now */
NeilBrown4e5314b2005-11-08 21:39:22 -08003008 clear_bit(R5_ReadError, &dev->flags);
3009 clear_bit(R5_ReWrite, &dev->flags);
3010 }
NeilBrownb2d444d2005-11-08 21:39:31 -08003011 if (!rdev || !test_bit(In_sync, &rdev->flags)
NeilBrown4e5314b2005-11-08 21:39:22 -08003012 || test_bit(R5_ReadError, &dev->flags)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003013 s.failed++;
3014 s.failed_num = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 } else
3016 set_bit(R5_Insync, &dev->flags);
3017 }
NeilBrown9910f162006-01-06 00:20:24 -08003018 rcu_read_unlock();
Dan Williamsb5e98d62007-01-02 13:52:31 -07003019
Dan Williams6bfe0b42008-04-30 00:52:32 -07003020 if (unlikely(blocked_rdev)) {
NeilBrownac4090d2008-08-05 15:54:13 +10003021 if (s.syncing || s.expanding || s.expanded ||
3022 s.to_write || s.written) {
3023 set_bit(STRIPE_HANDLE, &sh->state);
3024 goto unlock;
3025 }
3026 /* There is nothing for the blocked_rdev to block */
3027 rdev_dec_pending(blocked_rdev, conf->mddev);
3028 blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003029 }
3030
Dan Williams83de75c2008-06-28 08:31:58 +10003031 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3032 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3033 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3034 }
Dan Williamsb5e98d62007-01-02 13:52:31 -07003035
Dan Williams45b42332007-07-09 11:56:43 -07003036 pr_debug("locked=%d uptodate=%d to_read=%d"
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 " to_write=%d failed=%d failed_num=%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003038 s.locked, s.uptodate, s.to_read, s.to_write,
3039 s.failed, s.failed_num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 /* check if the array has lost two devices and, if so, some requests might
3041 * need to be failed
3042 */
Dan Williamsa4456852007-07-09 11:56:43 -07003043 if (s.failed > 1 && s.to_read+s.to_write+s.written)
Dan Williams1fe797e2008-06-28 09:16:30 +10003044 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003045 if (s.failed > 1 && s.syncing) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3047 clear_bit(STRIPE_SYNCING, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003048 s.syncing = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 }
3050
3051 /* might be able to return some write requests if the parity block
3052 * is safe, or on a failed drive
3053 */
3054 dev = &sh->dev[sh->pd_idx];
Dan Williamsa4456852007-07-09 11:56:43 -07003055 if ( s.written &&
3056 ((test_bit(R5_Insync, &dev->flags) &&
3057 !test_bit(R5_LOCKED, &dev->flags) &&
3058 test_bit(R5_UPTODATE, &dev->flags)) ||
3059 (s.failed == 1 && s.failed_num == sh->pd_idx)))
Dan Williams1fe797e2008-06-28 09:16:30 +10003060 handle_stripe_clean_event(conf, sh, disks, &return_bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061
3062 /* Now we might consider reading some blocks, either to check/generate
3063 * parity, or to satisfy requests
3064 * or to load a block that is being partially written.
3065 */
Dan Williamsa4456852007-07-09 11:56:43 -07003066 if (s.to_read || s.non_overwrite ||
Dan Williams976ea8d2008-06-28 08:32:03 +10003067 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
Dan Williams1fe797e2008-06-28 09:16:30 +10003068 handle_stripe_fill5(sh, &s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069
Dan Williamse33129d2007-01-02 13:52:30 -07003070 /* Now we check to see if any write operations have recently
3071 * completed
3072 */
Dan Williamse0a115e2008-06-05 22:45:52 -07003073 prexor = 0;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003074 if (sh->reconstruct_state == reconstruct_state_prexor_drain_result)
Dan Williamse0a115e2008-06-05 22:45:52 -07003075 prexor = 1;
Dan Williamsd8ee0722008-06-28 08:32:06 +10003076 if (sh->reconstruct_state == reconstruct_state_drain_result ||
3077 sh->reconstruct_state == reconstruct_state_prexor_drain_result) {
Dan Williams600aa102008-06-28 08:32:05 +10003078 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamse33129d2007-01-02 13:52:30 -07003079
3080 /* All the 'written' buffers and the parity block are ready to
3081 * be written back to disk
3082 */
3083 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3084 for (i = disks; i--; ) {
3085 dev = &sh->dev[i];
3086 if (test_bit(R5_LOCKED, &dev->flags) &&
3087 (i == sh->pd_idx || dev->written)) {
3088 pr_debug("Writing block %d\n", i);
3089 set_bit(R5_Wantwrite, &dev->flags);
Dan Williamse0a115e2008-06-05 22:45:52 -07003090 if (prexor)
3091 continue;
Dan Williamse33129d2007-01-02 13:52:30 -07003092 if (!test_bit(R5_Insync, &dev->flags) ||
3093 (i == sh->pd_idx && s.failed == 0))
3094 set_bit(STRIPE_INSYNC, &sh->state);
3095 }
3096 }
NeilBrown729a1862009-12-14 12:49:50 +11003097 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3098 dec_preread_active = 1;
Dan Williamse33129d2007-01-02 13:52:30 -07003099 }
3100
3101 /* Now to consider new write requests and what else, if anything
3102 * should be read. We do not handle new writes when:
3103 * 1/ A 'write' operation (copy+xor) is already in flight.
3104 * 2/ A 'check' operation is in flight, as it may clobber the parity
3105 * block.
3106 */
Dan Williams600aa102008-06-28 08:32:05 +10003107 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
Dan Williams1fe797e2008-06-28 09:16:30 +10003108 handle_stripe_dirtying5(conf, sh, &s, disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
3110 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamse89f8962007-01-02 13:52:31 -07003111 * Any reads will already have been scheduled, so we just see if enough
3112 * data is available. The parity check is held off while parity
3113 * dependent operations are in flight.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 */
Dan Williamsecc65c92008-06-28 08:31:57 +10003115 if (sh->check_state ||
3116 (s.syncing && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003117 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
Dan Williamsecc65c92008-06-28 08:31:57 +10003118 !test_bit(STRIPE_INSYNC, &sh->state)))
Dan Williamsa4456852007-07-09 11:56:43 -07003119 handle_parity_checks5(conf, sh, &s, disks);
Dan Williamse89f8962007-01-02 13:52:31 -07003120
Dan Williamsa4456852007-07-09 11:56:43 -07003121 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3123 clear_bit(STRIPE_SYNCING, &sh->state);
3124 }
NeilBrown4e5314b2005-11-08 21:39:22 -08003125
3126 /* If the failed drive is just a ReadError, then we might need to progress
3127 * the repair/check process
3128 */
Dan Williamsa4456852007-07-09 11:56:43 -07003129 if (s.failed == 1 && !conf->mddev->ro &&
3130 test_bit(R5_ReadError, &sh->dev[s.failed_num].flags)
3131 && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags)
3132 && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags)
NeilBrown4e5314b2005-11-08 21:39:22 -08003133 ) {
Dan Williamsa4456852007-07-09 11:56:43 -07003134 dev = &sh->dev[s.failed_num];
NeilBrown4e5314b2005-11-08 21:39:22 -08003135 if (!test_bit(R5_ReWrite, &dev->flags)) {
3136 set_bit(R5_Wantwrite, &dev->flags);
3137 set_bit(R5_ReWrite, &dev->flags);
3138 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003139 s.locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003140 } else {
3141 /* let's read it back */
3142 set_bit(R5_Wantread, &dev->flags);
3143 set_bit(R5_LOCKED, &dev->flags);
Dan Williamsa4456852007-07-09 11:56:43 -07003144 s.locked++;
NeilBrown4e5314b2005-11-08 21:39:22 -08003145 }
3146 }
3147
Dan Williams600aa102008-06-28 08:32:05 +10003148 /* Finish reconstruct operations initiated by the expansion process */
3149 if (sh->reconstruct_state == reconstruct_state_result) {
NeilBrownab69ae12009-03-31 15:26:47 +11003150 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003151 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003152 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3153 /* sh cannot be written until sh2 has been read.
3154 * so arrange for sh to be delayed a little
3155 */
3156 set_bit(STRIPE_DELAYED, &sh->state);
3157 set_bit(STRIPE_HANDLE, &sh->state);
3158 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3159 &sh2->state))
3160 atomic_inc(&conf->preread_active_stripes);
3161 release_stripe(sh2);
3162 goto unlock;
3163 }
3164 if (sh2)
3165 release_stripe(sh2);
3166
Dan Williams600aa102008-06-28 08:32:05 +10003167 sh->reconstruct_state = reconstruct_state_idle;
Dan Williamsf0a50d32007-01-02 13:52:31 -07003168 clear_bit(STRIPE_EXPANDING, &sh->state);
Dan Williams23397882008-07-23 20:05:34 -07003169 for (i = conf->raid_disks; i--; ) {
Dan Williamsf0a50d32007-01-02 13:52:31 -07003170 set_bit(R5_Wantwrite, &sh->dev[i].flags);
Dan Williams23397882008-07-23 20:05:34 -07003171 set_bit(R5_LOCKED, &sh->dev[i].flags);
Neil Brownefe31142008-06-28 08:31:14 +10003172 s.locked++;
Dan Williams23397882008-07-23 20:05:34 -07003173 }
Dan Williamsf0a50d32007-01-02 13:52:31 -07003174 }
3175
3176 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
Dan Williams600aa102008-06-28 08:32:05 +10003177 !sh->reconstruct_state) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003178 /* Need to write out all blocks after computing parity */
3179 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003180 stripe_set_idx(sh->sector, conf, 0, sh);
Yuri Tikhonovc0f7bdd2009-08-29 19:13:12 -07003181 schedule_reconstruction(sh, &s, 1, 1);
Dan Williams600aa102008-06-28 08:32:05 +10003182 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08003183 clear_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrownf6705572006-03-27 01:18:11 -08003184 atomic_dec(&conf->reshape_stripes);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003185 wake_up(&conf->wait_for_overlap);
3186 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3187 }
3188
Dan Williams0f94e872008-01-08 15:32:53 -08003189 if (s.expanding && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003190 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
Dan Williamsa4456852007-07-09 11:56:43 -07003191 handle_stripe_expansion(conf, sh, NULL);
NeilBrownccfcc3c2006-03-27 01:18:09 -08003192
Dan Williams6bfe0b42008-04-30 00:52:32 -07003193 unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194 spin_unlock(&sh->lock);
3195
Dan Williams6bfe0b42008-04-30 00:52:32 -07003196 /* wait for this device to become unblocked */
3197 if (unlikely(blocked_rdev))
3198 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3199
Dan Williams600aa102008-06-28 08:32:05 +10003200 if (s.ops_request)
Dan Williamsac6b53b2009-07-14 13:40:19 -07003201 raid_run_ops(sh, s.ops_request);
Dan Williamsd84e0f12007-01-02 13:52:30 -07003202
Dan Williamsc4e5ac02008-06-28 08:31:53 +10003203 ops_run_io(sh, &s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204
NeilBrown729a1862009-12-14 12:49:50 +11003205 if (dec_preread_active) {
3206 /* We delay this until after ops_run_io so that if make_request
3207 * is waiting on a barrier, it won't continue until the writes
3208 * have actually been submitted.
3209 */
3210 atomic_dec(&conf->preread_active_stripes);
3211 if (atomic_read(&conf->preread_active_stripes) <
3212 IO_THRESHOLD)
3213 md_wakeup_thread(conf->mddev->thread);
3214 }
Dan Williamsa4456852007-07-09 11:56:43 -07003215 return_io(return_bi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216}
3217
NeilBrown14425772009-10-16 15:55:25 +11003218static void handle_stripe6(struct stripe_head *sh)
NeilBrown16a53ec2006-06-26 00:27:38 -07003219{
NeilBrownbff61972009-03-31 14:33:13 +11003220 raid5_conf_t *conf = sh->raid_conf;
NeilBrownf4168852007-02-28 20:11:53 -08003221 int disks = sh->disks;
Dan Williamsa4456852007-07-09 11:56:43 -07003222 struct bio *return_bi = NULL;
NeilBrown34e04e82009-03-31 15:10:16 +11003223 int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx;
Dan Williamsa4456852007-07-09 11:56:43 -07003224 struct stripe_head_state s;
3225 struct r6_state r6s;
NeilBrown16a53ec2006-06-26 00:27:38 -07003226 struct r5dev *dev, *pdev, *qdev;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003227 mdk_rdev_t *blocked_rdev = NULL;
NeilBrown729a1862009-12-14 12:49:50 +11003228 int dec_preread_active = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003229
Dan Williams45b42332007-07-09 11:56:43 -07003230 pr_debug("handling stripe %llu, state=%#lx cnt=%d, "
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003231 "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003232 (unsigned long long)sh->sector, sh->state,
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003233 atomic_read(&sh->count), pd_idx, qd_idx,
3234 sh->check_state, sh->reconstruct_state);
Dan Williamsa4456852007-07-09 11:56:43 -07003235 memset(&s, 0, sizeof(s));
NeilBrown16a53ec2006-06-26 00:27:38 -07003236
3237 spin_lock(&sh->lock);
3238 clear_bit(STRIPE_HANDLE, &sh->state);
3239 clear_bit(STRIPE_DELAYED, &sh->state);
3240
Dan Williamsa4456852007-07-09 11:56:43 -07003241 s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
3242 s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
3243 s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);
NeilBrown16a53ec2006-06-26 00:27:38 -07003244 /* Now to look around and see what can be done */
3245
3246 rcu_read_lock();
3247 for (i=disks; i--; ) {
3248 mdk_rdev_t *rdev;
3249 dev = &sh->dev[i];
3250 clear_bit(R5_Insync, &dev->flags);
3251
Dan Williams45b42332007-07-09 11:56:43 -07003252 pr_debug("check %d: state 0x%lx read %p write %p written %p\n",
NeilBrown16a53ec2006-06-26 00:27:38 -07003253 i, dev->flags, dev->toread, dev->towrite, dev->written);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003254 /* maybe we can reply to a read
3255 *
3256 * new wantfill requests are only permitted while
3257 * ops_complete_biofill is guaranteed to be inactive
3258 */
3259 if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread &&
3260 !test_bit(STRIPE_BIOFILL_RUN, &sh->state))
3261 set_bit(R5_Wantfill, &dev->flags);
NeilBrown16a53ec2006-06-26 00:27:38 -07003262
3263 /* now count some things */
Dan Williamsa4456852007-07-09 11:56:43 -07003264 if (test_bit(R5_LOCKED, &dev->flags)) s.locked++;
3265 if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++;
Dan Williams2d6e4ec2009-09-16 12:11:54 -07003266 if (test_bit(R5_Wantcompute, &dev->flags)) {
3267 s.compute++;
3268 BUG_ON(s.compute > 2);
3269 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003270
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003271 if (test_bit(R5_Wantfill, &dev->flags)) {
3272 s.to_fill++;
3273 } else if (dev->toread)
Dan Williamsa4456852007-07-09 11:56:43 -07003274 s.to_read++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003275 if (dev->towrite) {
Dan Williamsa4456852007-07-09 11:56:43 -07003276 s.to_write++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003277 if (!test_bit(R5_OVERWRITE, &dev->flags))
Dan Williamsa4456852007-07-09 11:56:43 -07003278 s.non_overwrite++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003279 }
Dan Williamsa4456852007-07-09 11:56:43 -07003280 if (dev->written)
3281 s.written++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003282 rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownac4090d2008-08-05 15:54:13 +10003283 if (blocked_rdev == NULL &&
3284 rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07003285 blocked_rdev = rdev;
3286 atomic_inc(&rdev->nr_pending);
Dan Williams6bfe0b42008-04-30 00:52:32 -07003287 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003288 if (!rdev || !test_bit(In_sync, &rdev->flags)) {
3289 /* The ReadError flag will just be confusing now */
3290 clear_bit(R5_ReadError, &dev->flags);
3291 clear_bit(R5_ReWrite, &dev->flags);
3292 }
3293 if (!rdev || !test_bit(In_sync, &rdev->flags)
3294 || test_bit(R5_ReadError, &dev->flags)) {
Dan Williamsa4456852007-07-09 11:56:43 -07003295 if (s.failed < 2)
3296 r6s.failed_num[s.failed] = i;
3297 s.failed++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003298 } else
3299 set_bit(R5_Insync, &dev->flags);
3300 }
3301 rcu_read_unlock();
Dan Williams6bfe0b42008-04-30 00:52:32 -07003302
3303 if (unlikely(blocked_rdev)) {
NeilBrownac4090d2008-08-05 15:54:13 +10003304 if (s.syncing || s.expanding || s.expanded ||
3305 s.to_write || s.written) {
3306 set_bit(STRIPE_HANDLE, &sh->state);
3307 goto unlock;
3308 }
3309 /* There is nothing for the blocked_rdev to block */
3310 rdev_dec_pending(blocked_rdev, conf->mddev);
3311 blocked_rdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003312 }
NeilBrownac4090d2008-08-05 15:54:13 +10003313
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003314 if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) {
3315 set_bit(STRIPE_OP_BIOFILL, &s.ops_request);
3316 set_bit(STRIPE_BIOFILL_RUN, &sh->state);
3317 }
3318
Dan Williams45b42332007-07-09 11:56:43 -07003319 pr_debug("locked=%d uptodate=%d to_read=%d"
NeilBrown16a53ec2006-06-26 00:27:38 -07003320 " to_write=%d failed=%d failed_num=%d,%d\n",
Dan Williamsa4456852007-07-09 11:56:43 -07003321 s.locked, s.uptodate, s.to_read, s.to_write, s.failed,
3322 r6s.failed_num[0], r6s.failed_num[1]);
3323 /* check if the array has lost >2 devices and, if so, some requests
3324 * might need to be failed
NeilBrown16a53ec2006-06-26 00:27:38 -07003325 */
Dan Williamsa4456852007-07-09 11:56:43 -07003326 if (s.failed > 2 && s.to_read+s.to_write+s.written)
Dan Williams1fe797e2008-06-28 09:16:30 +10003327 handle_failed_stripe(conf, sh, &s, disks, &return_bi);
Dan Williamsa4456852007-07-09 11:56:43 -07003328 if (s.failed > 2 && s.syncing) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003329 md_done_sync(conf->mddev, STRIPE_SECTORS,0);
3330 clear_bit(STRIPE_SYNCING, &sh->state);
Dan Williamsa4456852007-07-09 11:56:43 -07003331 s.syncing = 0;
NeilBrown16a53ec2006-06-26 00:27:38 -07003332 }
3333
3334 /*
3335 * might be able to return some write requests if the parity blocks
3336 * are safe, or on a failed drive
3337 */
3338 pdev = &sh->dev[pd_idx];
Dan Williamsa4456852007-07-09 11:56:43 -07003339 r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx)
3340 || (s.failed >= 2 && r6s.failed_num[1] == pd_idx);
NeilBrown34e04e82009-03-31 15:10:16 +11003341 qdev = &sh->dev[qd_idx];
3342 r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == qd_idx)
3343 || (s.failed >= 2 && r6s.failed_num[1] == qd_idx);
NeilBrown16a53ec2006-06-26 00:27:38 -07003344
Dan Williamsa4456852007-07-09 11:56:43 -07003345 if ( s.written &&
3346 ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003347 && !test_bit(R5_LOCKED, &pdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003348 && test_bit(R5_UPTODATE, &pdev->flags)))) &&
3349 ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags)
NeilBrown16a53ec2006-06-26 00:27:38 -07003350 && !test_bit(R5_LOCKED, &qdev->flags)
Dan Williamsa4456852007-07-09 11:56:43 -07003351 && test_bit(R5_UPTODATE, &qdev->flags)))))
Dan Williams1fe797e2008-06-28 09:16:30 +10003352 handle_stripe_clean_event(conf, sh, disks, &return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003353
3354 /* Now we might consider reading some blocks, either to check/generate
3355 * parity, or to satisfy requests
3356 * or to load a block that is being partially written.
3357 */
Dan Williamsa4456852007-07-09 11:56:43 -07003358 if (s.to_read || s.non_overwrite || (s.to_write && s.failed) ||
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003359 (s.syncing && (s.uptodate + s.compute < disks)) || s.expanding)
Dan Williams1fe797e2008-06-28 09:16:30 +10003360 handle_stripe_fill6(sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003361
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003362 /* Now we check to see if any write operations have recently
3363 * completed
3364 */
3365 if (sh->reconstruct_state == reconstruct_state_drain_result) {
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003366
3367 sh->reconstruct_state = reconstruct_state_idle;
3368 /* All the 'written' buffers and the parity blocks are ready to
3369 * be written back to disk
3370 */
3371 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags));
3372 BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags));
3373 for (i = disks; i--; ) {
3374 dev = &sh->dev[i];
3375 if (test_bit(R5_LOCKED, &dev->flags) &&
3376 (i == sh->pd_idx || i == qd_idx ||
3377 dev->written)) {
3378 pr_debug("Writing block %d\n", i);
3379 BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));
3380 set_bit(R5_Wantwrite, &dev->flags);
3381 if (!test_bit(R5_Insync, &dev->flags) ||
3382 ((i == sh->pd_idx || i == qd_idx) &&
3383 s.failed == 0))
3384 set_bit(STRIPE_INSYNC, &sh->state);
3385 }
3386 }
NeilBrown729a1862009-12-14 12:49:50 +11003387 if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3388 dec_preread_active = 1;
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003389 }
3390
Yuri Tikhonova9b39a72009-08-29 19:13:12 -07003391 /* Now to consider new write requests and what else, if anything
3392 * should be read. We do not handle new writes when:
3393 * 1/ A 'write' operation (copy+gen_syndrome) is already in flight.
3394 * 2/ A 'check' operation is in flight, as it may clobber the parity
3395 * block.
3396 */
3397 if (s.to_write && !sh->reconstruct_state && !sh->check_state)
Dan Williams1fe797e2008-06-28 09:16:30 +10003398 handle_stripe_dirtying6(conf, sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003399
3400 /* maybe we need to check and possibly fix the parity for this stripe
Dan Williamsa4456852007-07-09 11:56:43 -07003401 * Any reads will already have been scheduled, so we just see if enough
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003402 * data is available. The parity check is held off while parity
3403 * dependent operations are in flight.
NeilBrown16a53ec2006-06-26 00:27:38 -07003404 */
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003405 if (sh->check_state ||
3406 (s.syncing && s.locked == 0 &&
3407 !test_bit(STRIPE_COMPUTE_RUN, &sh->state) &&
3408 !test_bit(STRIPE_INSYNC, &sh->state)))
Dan Williams36d1c642009-07-14 11:48:22 -07003409 handle_parity_checks6(conf, sh, &s, &r6s, disks);
NeilBrown16a53ec2006-06-26 00:27:38 -07003410
Dan Williamsa4456852007-07-09 11:56:43 -07003411 if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) {
NeilBrown16a53ec2006-06-26 00:27:38 -07003412 md_done_sync(conf->mddev, STRIPE_SECTORS,1);
3413 clear_bit(STRIPE_SYNCING, &sh->state);
3414 }
3415
3416 /* If the failed drives are just a ReadError, then we might need
3417 * to progress the repair/check process
3418 */
Dan Williamsa4456852007-07-09 11:56:43 -07003419 if (s.failed <= 2 && !conf->mddev->ro)
3420 for (i = 0; i < s.failed; i++) {
3421 dev = &sh->dev[r6s.failed_num[i]];
NeilBrown16a53ec2006-06-26 00:27:38 -07003422 if (test_bit(R5_ReadError, &dev->flags)
3423 && !test_bit(R5_LOCKED, &dev->flags)
3424 && test_bit(R5_UPTODATE, &dev->flags)
3425 ) {
3426 if (!test_bit(R5_ReWrite, &dev->flags)) {
3427 set_bit(R5_Wantwrite, &dev->flags);
3428 set_bit(R5_ReWrite, &dev->flags);
3429 set_bit(R5_LOCKED, &dev->flags);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003430 s.locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003431 } else {
3432 /* let's read it back */
3433 set_bit(R5_Wantread, &dev->flags);
3434 set_bit(R5_LOCKED, &dev->flags);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003435 s.locked++;
NeilBrown16a53ec2006-06-26 00:27:38 -07003436 }
3437 }
3438 }
NeilBrownf4168852007-02-28 20:11:53 -08003439
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003440 /* Finish reconstruct operations initiated by the expansion process */
3441 if (sh->reconstruct_state == reconstruct_state_result) {
3442 sh->reconstruct_state = reconstruct_state_idle;
3443 clear_bit(STRIPE_EXPANDING, &sh->state);
3444 for (i = conf->raid_disks; i--; ) {
3445 set_bit(R5_Wantwrite, &sh->dev[i].flags);
3446 set_bit(R5_LOCKED, &sh->dev[i].flags);
3447 s.locked++;
3448 }
3449 }
3450
3451 if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) &&
3452 !sh->reconstruct_state) {
NeilBrownab69ae12009-03-31 15:26:47 +11003453 struct stripe_head *sh2
NeilBrowna8c906c2009-06-09 14:39:59 +10003454 = get_active_stripe(conf, sh->sector, 1, 1, 1);
NeilBrownab69ae12009-03-31 15:26:47 +11003455 if (sh2 && test_bit(STRIPE_EXPAND_SOURCE, &sh2->state)) {
3456 /* sh cannot be written until sh2 has been read.
3457 * so arrange for sh to be delayed a little
3458 */
3459 set_bit(STRIPE_DELAYED, &sh->state);
3460 set_bit(STRIPE_HANDLE, &sh->state);
3461 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE,
3462 &sh2->state))
3463 atomic_inc(&conf->preread_active_stripes);
3464 release_stripe(sh2);
3465 goto unlock;
3466 }
3467 if (sh2)
3468 release_stripe(sh2);
3469
NeilBrownf4168852007-02-28 20:11:53 -08003470 /* Need to write out all blocks after computing P&Q */
3471 sh->disks = conf->raid_disks;
NeilBrown911d4ee2009-03-31 14:39:38 +11003472 stripe_set_idx(sh->sector, conf, 0, sh);
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003473 schedule_reconstruction(sh, &s, 1, 1);
3474 } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) {
NeilBrownf4168852007-02-28 20:11:53 -08003475 clear_bit(STRIPE_EXPAND_READY, &sh->state);
3476 atomic_dec(&conf->reshape_stripes);
3477 wake_up(&conf->wait_for_overlap);
3478 md_done_sync(conf->mddev, STRIPE_SECTORS, 1);
3479 }
3480
Dan Williams0f94e872008-01-08 15:32:53 -08003481 if (s.expanding && s.locked == 0 &&
Dan Williams976ea8d2008-06-28 08:32:03 +10003482 !test_bit(STRIPE_COMPUTE_RUN, &sh->state))
Dan Williamsa4456852007-07-09 11:56:43 -07003483 handle_stripe_expansion(conf, sh, &r6s);
NeilBrownf4168852007-02-28 20:11:53 -08003484
Dan Williams6bfe0b42008-04-30 00:52:32 -07003485 unlock:
NeilBrown16a53ec2006-06-26 00:27:38 -07003486 spin_unlock(&sh->lock);
3487
Dan Williams6bfe0b42008-04-30 00:52:32 -07003488 /* wait for this device to become unblocked */
3489 if (unlikely(blocked_rdev))
3490 md_wait_for_blocked_rdev(blocked_rdev, conf->mddev);
3491
Yuri Tikhonov6c0069c2009-08-29 19:13:13 -07003492 if (s.ops_request)
3493 raid_run_ops(sh, s.ops_request);
3494
Dan Williamsf0e43bc2008-06-28 08:31:55 +10003495 ops_run_io(sh, &s);
3496
NeilBrown729a1862009-12-14 12:49:50 +11003497
3498 if (dec_preread_active) {
3499 /* We delay this until after ops_run_io so that if make_request
3500 * is waiting on a barrier, it won't continue until the writes
3501 * have actually been submitted.
3502 */
3503 atomic_dec(&conf->preread_active_stripes);
3504 if (atomic_read(&conf->preread_active_stripes) <
3505 IO_THRESHOLD)
3506 md_wakeup_thread(conf->mddev->thread);
3507 }
3508
Dan Williamsa4456852007-07-09 11:56:43 -07003509 return_io(return_bi);
NeilBrown16a53ec2006-06-26 00:27:38 -07003510}
3511
NeilBrown14425772009-10-16 15:55:25 +11003512static void handle_stripe(struct stripe_head *sh)
NeilBrown16a53ec2006-06-26 00:27:38 -07003513{
3514 if (sh->raid_conf->level == 6)
NeilBrown14425772009-10-16 15:55:25 +11003515 handle_stripe6(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -07003516 else
NeilBrown14425772009-10-16 15:55:25 +11003517 handle_stripe5(sh);
NeilBrown16a53ec2006-06-26 00:27:38 -07003518}
3519
Arjan van de Ven858119e2006-01-14 13:20:43 -08003520static void raid5_activate_delayed(raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521{
3522 if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) {
3523 while (!list_empty(&conf->delayed_list)) {
3524 struct list_head *l = conf->delayed_list.next;
3525 struct stripe_head *sh;
3526 sh = list_entry(l, struct stripe_head, lru);
3527 list_del_init(l);
3528 clear_bit(STRIPE_DELAYED, &sh->state);
3529 if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
3530 atomic_inc(&conf->preread_active_stripes);
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003531 list_add_tail(&sh->lru, &conf->hold_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 }
NeilBrown6ed30032008-02-06 01:40:00 -08003533 } else
3534 blk_plug_device(conf->mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535}
3536
Arjan van de Ven858119e2006-01-14 13:20:43 -08003537static void activate_bit_delay(raid5_conf_t *conf)
NeilBrown72626682005-09-09 16:23:54 -07003538{
3539 /* device_lock is held */
3540 struct list_head head;
3541 list_add(&head, &conf->bitmap_list);
3542 list_del_init(&conf->bitmap_list);
3543 while (!list_empty(&head)) {
3544 struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
3545 list_del_init(&sh->lru);
3546 atomic_inc(&sh->count);
3547 __release_stripe(conf, sh);
3548 }
3549}
3550
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551static void unplug_slaves(mddev_t *mddev)
3552{
NeilBrown070ec552009-06-16 16:54:21 +10003553 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554 int i;
NeilBrown5e5e3e72009-10-16 16:35:30 +11003555 int devs = max(conf->raid_disks, conf->previous_raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556
3557 rcu_read_lock();
NeilBrown5e5e3e72009-10-16 16:35:30 +11003558 for (i = 0; i < devs; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -08003559 mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -08003560 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
Jens Axboe165125e2007-07-24 09:28:11 +02003561 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562
3563 atomic_inc(&rdev->nr_pending);
3564 rcu_read_unlock();
3565
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05003566 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567
3568 rdev_dec_pending(rdev, mddev);
3569 rcu_read_lock();
3570 }
3571 }
3572 rcu_read_unlock();
3573}
3574
Jens Axboe165125e2007-07-24 09:28:11 +02003575static void raid5_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576{
3577 mddev_t *mddev = q->queuedata;
NeilBrown070ec552009-06-16 16:54:21 +10003578 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 unsigned long flags;
3580
3581 spin_lock_irqsave(&conf->device_lock, flags);
3582
NeilBrown72626682005-09-09 16:23:54 -07003583 if (blk_remove_plug(q)) {
3584 conf->seq_flush++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585 raid5_activate_delayed(conf);
NeilBrown72626682005-09-09 16:23:54 -07003586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 md_wakeup_thread(mddev->thread);
3588
3589 spin_unlock_irqrestore(&conf->device_lock, flags);
3590
3591 unplug_slaves(mddev);
3592}
3593
NeilBrownf022b2f2006-10-03 01:15:56 -07003594static int raid5_congested(void *data, int bits)
3595{
3596 mddev_t *mddev = data;
NeilBrown070ec552009-06-16 16:54:21 +10003597 raid5_conf_t *conf = mddev->private;
NeilBrownf022b2f2006-10-03 01:15:56 -07003598
3599 /* No difference between reads and writes. Just check
3600 * how busy the stripe_cache is
3601 */
NeilBrown3fa841d2009-09-23 18:10:29 +10003602
3603 if (mddev_congested(mddev, bits))
3604 return 1;
NeilBrownf022b2f2006-10-03 01:15:56 -07003605 if (conf->inactive_blocked)
3606 return 1;
3607 if (conf->quiesce)
3608 return 1;
3609 if (list_empty_careful(&conf->inactive_list))
3610 return 1;
3611
3612 return 0;
3613}
3614
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003615/* We want read requests to align with chunks where possible,
3616 * but write requests don't need to.
3617 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003618static int raid5_mergeable_bvec(struct request_queue *q,
3619 struct bvec_merge_data *bvm,
3620 struct bio_vec *biovec)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003621{
3622 mddev_t *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003623 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003624 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +10003625 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003626 unsigned int bio_sectors = bvm->bi_size >> 9;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003627
Alasdair G Kergoncc371e62008-07-03 09:53:43 +02003628 if ((bvm->bi_rw & 1) == WRITE)
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003629 return biovec->bv_len; /* always allow writes to be mergeable */
3630
Andre Noll664e7c42009-06-18 08:45:27 +10003631 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3632 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08003633 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
3634 if (max < 0) max = 0;
3635 if (max <= biovec->bv_len && bio_sectors == 0)
3636 return biovec->bv_len;
3637 else
3638 return max;
3639}
3640
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003641
3642static int in_chunk_boundary(mddev_t *mddev, struct bio *bio)
3643{
3644 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
Andre Noll9d8f0362009-06-18 08:45:01 +10003645 unsigned int chunk_sectors = mddev->chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003646 unsigned int bio_sectors = bio->bi_size >> 9;
3647
Andre Noll664e7c42009-06-18 08:45:27 +10003648 if (mddev->new_chunk_sectors < mddev->chunk_sectors)
3649 chunk_sectors = mddev->new_chunk_sectors;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003650 return chunk_sectors >=
3651 ((sector & (chunk_sectors - 1)) + bio_sectors);
3652}
3653
3654/*
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003655 * add bio to the retry LIFO ( in O(1) ... we are in interrupt )
3656 * later sampled by raid5d.
3657 */
3658static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf)
3659{
3660 unsigned long flags;
3661
3662 spin_lock_irqsave(&conf->device_lock, flags);
3663
3664 bi->bi_next = conf->retry_read_aligned_list;
3665 conf->retry_read_aligned_list = bi;
3666
3667 spin_unlock_irqrestore(&conf->device_lock, flags);
3668 md_wakeup_thread(conf->mddev->thread);
3669}
3670
3671
3672static struct bio *remove_bio_from_retry(raid5_conf_t *conf)
3673{
3674 struct bio *bi;
3675
3676 bi = conf->retry_read_aligned;
3677 if (bi) {
3678 conf->retry_read_aligned = NULL;
3679 return bi;
3680 }
3681 bi = conf->retry_read_aligned_list;
3682 if(bi) {
Neil Brown387bb172007-02-08 14:20:29 -08003683 conf->retry_read_aligned_list = bi->bi_next;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003684 bi->bi_next = NULL;
Jens Axboe960e7392008-08-15 10:41:18 +02003685 /*
3686 * this sets the active strip count to 1 and the processed
3687 * strip count to zero (upper 8 bits)
3688 */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003689 bi->bi_phys_segments = 1; /* biased count of active stripes */
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003690 }
3691
3692 return bi;
3693}
3694
3695
3696/*
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003697 * The "raid5_align_endio" should check if the read succeeded and if it
3698 * did, call bio_endio on the original bio (having bio_put the new bio
3699 * first).
3700 * If the read failed..
3701 */
NeilBrown6712ecf2007-09-27 12:47:43 +02003702static void raid5_align_endio(struct bio *bi, int error)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003703{
3704 struct bio* raid_bi = bi->bi_private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003705 mddev_t *mddev;
3706 raid5_conf_t *conf;
3707 int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags);
3708 mdk_rdev_t *rdev;
3709
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003710 bio_put(bi);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003711
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003712 rdev = (void*)raid_bi->bi_next;
3713 raid_bi->bi_next = NULL;
NeilBrown2b7f2222010-03-25 16:06:03 +11003714 mddev = rdev->mddev;
3715 conf = mddev->private;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003716
3717 rdev_dec_pending(rdev, conf->mddev);
3718
3719 if (!error && uptodate) {
NeilBrown6712ecf2007-09-27 12:47:43 +02003720 bio_endio(raid_bi, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003721 if (atomic_dec_and_test(&conf->active_aligned_reads))
3722 wake_up(&conf->wait_for_stripe);
NeilBrown6712ecf2007-09-27 12:47:43 +02003723 return;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003724 }
3725
3726
Dan Williams45b42332007-07-09 11:56:43 -07003727 pr_debug("raid5_align_endio : io error...handing IO for a retry\n");
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003728
3729 add_bio_to_retry(raid_bi, conf);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003730}
3731
Neil Brown387bb172007-02-08 14:20:29 -08003732static int bio_fits_rdev(struct bio *bi)
3733{
Jens Axboe165125e2007-07-24 09:28:11 +02003734 struct request_queue *q = bdev_get_queue(bi->bi_bdev);
Neil Brown387bb172007-02-08 14:20:29 -08003735
Martin K. Petersenae03bf62009-05-22 17:17:50 -04003736 if ((bi->bi_size>>9) > queue_max_sectors(q))
Neil Brown387bb172007-02-08 14:20:29 -08003737 return 0;
3738 blk_recount_segments(q, bi);
Martin K. Petersen8a783622010-02-26 00:20:39 -05003739 if (bi->bi_phys_segments > queue_max_segments(q))
Neil Brown387bb172007-02-08 14:20:29 -08003740 return 0;
3741
3742 if (q->merge_bvec_fn)
3743 /* it's too hard to apply the merge_bvec_fn at this stage,
3744 * just just give up
3745 */
3746 return 0;
3747
3748 return 1;
3749}
3750
3751
NeilBrown21a52c62010-04-01 15:02:13 +11003752static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003753{
NeilBrown070ec552009-06-16 16:54:21 +10003754 raid5_conf_t *conf = mddev->private;
NeilBrown8553fe7ec2009-12-14 12:49:47 +11003755 int dd_idx;
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003756 struct bio* align_bi;
3757 mdk_rdev_t *rdev;
3758
3759 if (!in_chunk_boundary(mddev, raid_bio)) {
Dan Williams45b42332007-07-09 11:56:43 -07003760 pr_debug("chunk_aligned_read : non aligned\n");
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003761 return 0;
3762 }
3763 /*
NeilBrown99c0fb52009-03-31 14:39:38 +11003764 * use bio_clone to make a copy of the bio
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003765 */
3766 align_bi = bio_clone(raid_bio, GFP_NOIO);
3767 if (!align_bi)
3768 return 0;
3769 /*
3770 * set bi_end_io to a new function, and set bi_private to the
3771 * original bio.
3772 */
3773 align_bi->bi_end_io = raid5_align_endio;
3774 align_bi->bi_private = raid_bio;
3775 /*
3776 * compute position
3777 */
NeilBrown112bf892009-03-31 14:39:38 +11003778 align_bi->bi_sector = raid5_compute_sector(conf, raid_bio->bi_sector,
3779 0,
NeilBrown911d4ee2009-03-31 14:39:38 +11003780 &dd_idx, NULL);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003781
3782 rcu_read_lock();
3783 rdev = rcu_dereference(conf->disks[dd_idx].rdev);
3784 if (rdev && test_bit(In_sync, &rdev->flags)) {
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003785 atomic_inc(&rdev->nr_pending);
3786 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003787 raid_bio->bi_next = (void*)rdev;
3788 align_bi->bi_bdev = rdev->bdev;
3789 align_bi->bi_flags &= ~(1 << BIO_SEG_VALID);
3790 align_bi->bi_sector += rdev->data_offset;
3791
Neil Brown387bb172007-02-08 14:20:29 -08003792 if (!bio_fits_rdev(align_bi)) {
3793 /* too big in some way */
3794 bio_put(align_bi);
3795 rdev_dec_pending(rdev, mddev);
3796 return 0;
3797 }
3798
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003799 spin_lock_irq(&conf->device_lock);
3800 wait_event_lock_irq(conf->wait_for_stripe,
3801 conf->quiesce == 0,
3802 conf->device_lock, /* nothing */);
3803 atomic_inc(&conf->active_aligned_reads);
3804 spin_unlock_irq(&conf->device_lock);
3805
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003806 generic_make_request(align_bi);
3807 return 1;
3808 } else {
3809 rcu_read_unlock();
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08003810 bio_put(align_bi);
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003811 return 0;
3812 }
3813}
3814
Dan Williams8b3e6cd2008-04-28 02:15:53 -07003815/* __get_priority_stripe - get the next stripe to process
3816 *
3817 * Full stripe writes are allowed to pass preread active stripes up until
3818 * the bypass_threshold is exceeded. In general the bypass_count
3819 * increments when the handle_list is handled before the hold_list; however, it
3820 * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a
3821 * stripe with in flight i/o. The bypass_count will be reset when the
3822 * head of the hold_list has changed, i.e. the head was promoted to the
3823 * handle_list.
3824 */
3825static struct stripe_head *__get_priority_stripe(raid5_conf_t *conf)
3826{
3827 struct stripe_head *sh;
3828
3829 pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n",
3830 __func__,
3831 list_empty(&conf->handle_list) ? "empty" : "busy",
3832 list_empty(&conf->hold_list) ? "empty" : "busy",
3833 atomic_read(&conf->pending_full_writes), conf->bypass_count);
3834
3835 if (!list_empty(&conf->handle_list)) {
3836 sh = list_entry(conf->handle_list.next, typeof(*sh), lru);
3837
3838 if (list_empty(&conf->hold_list))
3839 conf->bypass_count = 0;
3840 else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) {
3841 if (conf->hold_list.next == conf->last_hold)
3842 conf->bypass_count++;
3843 else {
3844 conf->last_hold = conf->hold_list.next;
3845 conf->bypass_count -= conf->bypass_threshold;
3846 if (conf->bypass_count < 0)
3847 conf->bypass_count = 0;
3848 }
3849 }
3850 } else if (!list_empty(&conf->hold_list) &&
3851 ((conf->bypass_threshold &&
3852 conf->bypass_count > conf->bypass_threshold) ||
3853 atomic_read(&conf->pending_full_writes) == 0)) {
3854 sh = list_entry(conf->hold_list.next,
3855 typeof(*sh), lru);
3856 conf->bypass_count -= conf->bypass_threshold;
3857 if (conf->bypass_count < 0)
3858 conf->bypass_count = 0;
3859 } else
3860 return NULL;
3861
3862 list_del_init(&sh->lru);
3863 atomic_inc(&sh->count);
3864 BUG_ON(atomic_read(&sh->count) != 1);
3865 return sh;
3866}
Raz Ben-Jehuda(caro)f6796232006-12-10 02:20:46 -08003867
NeilBrown21a52c62010-04-01 15:02:13 +11003868static int make_request(mddev_t *mddev, struct bio * bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869{
NeilBrown070ec552009-06-16 16:54:21 +10003870 raid5_conf_t *conf = mddev->private;
NeilBrown911d4ee2009-03-31 14:39:38 +11003871 int dd_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 sector_t new_sector;
3873 sector_t logical_sector, last_sector;
3874 struct stripe_head *sh;
Jens Axboea3623572005-11-01 09:26:16 +01003875 const int rw = bio_data_dir(bi);
NeilBrown49077322010-03-25 16:20:56 +11003876 int remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877
Jens Axboe1f98a132009-09-11 14:32:04 +02003878 if (unlikely(bio_rw_flagged(bi, BIO_RW_BARRIER))) {
NeilBrowna2826aa2009-12-14 12:49:49 +11003879 /* Drain all pending writes. We only really need
3880 * to ensure they have been submitted, but this is
3881 * easier.
3882 */
3883 mddev->pers->quiesce(mddev, 1);
3884 mddev->pers->quiesce(mddev, 0);
3885 md_barrier_request(mddev, bi);
NeilBrowne5dcdd82005-09-09 16:23:41 -07003886 return 0;
3887 }
3888
NeilBrown3d310eb2005-06-21 17:17:26 -07003889 md_write_start(mddev, bi);
NeilBrown06d91a52005-06-21 17:17:12 -07003890
NeilBrown802ba062006-12-13 00:34:13 -08003891 if (rw == READ &&
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003892 mddev->reshape_position == MaxSector &&
NeilBrown21a52c62010-04-01 15:02:13 +11003893 chunk_aligned_read(mddev,bi))
NeilBrown99c0fb52009-03-31 14:39:38 +11003894 return 0;
Raz Ben-Jehuda(caro)52488612006-12-10 02:20:48 -08003895
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
3897 last_sector = bi->bi_sector + (bi->bi_size>>9);
3898 bi->bi_next = NULL;
3899 bi->bi_phys_segments = 1; /* over-loaded to count active stripes */
NeilBrown06d91a52005-06-21 17:17:12 -07003900
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
3902 DEFINE_WAIT(w);
NeilBrown16a53ec2006-06-26 00:27:38 -07003903 int disks, data_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003904 int previous;
NeilBrownb578d552006-03-27 01:18:12 -08003905
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003906 retry:
NeilBrownb5663ba2009-03-31 14:39:38 +11003907 previous = 0;
NeilBrownb0f9ec02009-03-31 15:27:18 +11003908 disks = conf->raid_disks;
NeilBrownb578d552006-03-27 01:18:12 -08003909 prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003910 if (unlikely(conf->reshape_progress != MaxSector)) {
NeilBrownfef9c612009-03-31 15:16:46 +11003911 /* spinlock is needed as reshape_progress may be
NeilBrowndf8e7f72006-03-27 01:18:15 -08003912 * 64bit on a 32bit platform, and so it might be
3913 * possible to see a half-updated value
NeilBrownfef9c612009-03-31 15:16:46 +11003914 * Ofcourse reshape_progress could change after
NeilBrowndf8e7f72006-03-27 01:18:15 -08003915 * the lock is dropped, so once we get a reference
3916 * to the stripe that we think it is, we will have
3917 * to check again.
3918 */
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003919 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11003920 if (mddev->delta_disks < 0
3921 ? logical_sector < conf->reshape_progress
3922 : logical_sector >= conf->reshape_progress) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003923 disks = conf->previous_raid_disks;
NeilBrownb5663ba2009-03-31 14:39:38 +11003924 previous = 1;
3925 } else {
NeilBrownfef9c612009-03-31 15:16:46 +11003926 if (mddev->delta_disks < 0
3927 ? logical_sector < conf->reshape_safe
3928 : logical_sector >= conf->reshape_safe) {
NeilBrownb578d552006-03-27 01:18:12 -08003929 spin_unlock_irq(&conf->device_lock);
3930 schedule();
3931 goto retry;
3932 }
3933 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003934 spin_unlock_irq(&conf->device_lock);
3935 }
NeilBrown16a53ec2006-06-26 00:27:38 -07003936 data_disks = disks - conf->max_degraded;
3937
NeilBrown112bf892009-03-31 14:39:38 +11003938 new_sector = raid5_compute_sector(conf, logical_sector,
3939 previous,
NeilBrown911d4ee2009-03-31 14:39:38 +11003940 &dd_idx, NULL);
NeilBrown0c55e022010-05-03 14:09:02 +10003941 pr_debug("raid456: make_request, sector %llu logical %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942 (unsigned long long)new_sector,
3943 (unsigned long long)logical_sector);
3944
NeilBrownb5663ba2009-03-31 14:39:38 +11003945 sh = get_active_stripe(conf, new_sector, previous,
NeilBrowna8c906c2009-06-09 14:39:59 +10003946 (bi->bi_rw&RWA_MASK), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947 if (sh) {
NeilBrownb0f9ec02009-03-31 15:27:18 +11003948 if (unlikely(previous)) {
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003949 /* expansion might have moved on while waiting for a
NeilBrowndf8e7f72006-03-27 01:18:15 -08003950 * stripe, so we must do the range check again.
3951 * Expansion could still move past after this
3952 * test, but as we are holding a reference to
3953 * 'sh', we know that if that happens,
3954 * STRIPE_EXPANDING will get set and the expansion
3955 * won't proceed until we finish with the stripe.
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003956 */
3957 int must_retry = 0;
3958 spin_lock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11003959 if (mddev->delta_disks < 0
3960 ? logical_sector >= conf->reshape_progress
3961 : logical_sector < conf->reshape_progress)
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003962 /* mismatch, need to try again */
3963 must_retry = 1;
3964 spin_unlock_irq(&conf->device_lock);
3965 if (must_retry) {
3966 release_stripe(sh);
Dan Williams7a3ab902009-06-16 16:00:33 -07003967 schedule();
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003968 goto retry;
3969 }
3970 }
NeilBrowne62e58a2009-07-01 13:15:35 +10003971
NeilBrowna5c308d2009-07-01 13:15:35 +10003972 if (bio_data_dir(bi) == WRITE &&
3973 logical_sector >= mddev->suspend_lo &&
NeilBrowne464eaf2006-03-27 01:18:14 -08003974 logical_sector < mddev->suspend_hi) {
3975 release_stripe(sh);
NeilBrowne62e58a2009-07-01 13:15:35 +10003976 /* As the suspend_* range is controlled by
3977 * userspace, we want an interruptible
3978 * wait.
3979 */
3980 flush_signals(current);
3981 prepare_to_wait(&conf->wait_for_overlap,
3982 &w, TASK_INTERRUPTIBLE);
3983 if (logical_sector >= mddev->suspend_lo &&
3984 logical_sector < mddev->suspend_hi)
3985 schedule();
NeilBrowne464eaf2006-03-27 01:18:14 -08003986 goto retry;
3987 }
NeilBrown7ecaa1e2006-03-27 01:18:08 -08003988
3989 if (test_bit(STRIPE_EXPANDING, &sh->state) ||
3990 !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) {
3991 /* Stripe is busy expanding or
3992 * add failed due to overlap. Flush everything
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 * and wait a while
3994 */
3995 raid5_unplug_device(mddev->queue);
3996 release_stripe(sh);
3997 schedule();
3998 goto retry;
3999 }
4000 finish_wait(&conf->wait_for_overlap, &w);
NeilBrown6ed30032008-02-06 01:40:00 -08004001 set_bit(STRIPE_HANDLE, &sh->state);
4002 clear_bit(STRIPE_DELAYED, &sh->state);
NeilBrown729a1862009-12-14 12:49:50 +11004003 if (mddev->barrier &&
4004 !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
4005 atomic_inc(&conf->preread_active_stripes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006 release_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004007 } else {
4008 /* cannot get stripe for read-ahead, just give-up */
4009 clear_bit(BIO_UPTODATE, &bi->bi_flags);
4010 finish_wait(&conf->wait_for_overlap, &w);
4011 break;
4012 }
4013
4014 }
4015 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004016 remaining = raid5_dec_bi_phys_segments(bi);
NeilBrownf6344752006-03-27 01:18:17 -08004017 spin_unlock_irq(&conf->device_lock);
4018 if (remaining == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019
NeilBrown16a53ec2006-06-26 00:27:38 -07004020 if ( rw == WRITE )
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +02004022
Neil Brown0e13fe232008-06-28 08:31:20 +10004023 bio_endio(bi, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024 }
NeilBrown729a1862009-12-14 12:49:50 +11004025
4026 if (mddev->barrier) {
4027 /* We need to wait for the stripes to all be handled.
4028 * So: wait for preread_active_stripes to drop to 0.
4029 */
4030 wait_event(mddev->thread->wqueue,
4031 atomic_read(&conf->preread_active_stripes) == 0);
4032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033 return 0;
4034}
4035
Dan Williamsb522adc2009-03-31 15:00:31 +11004036static sector_t raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks);
4037
NeilBrown52c03292006-06-26 00:27:43 -07004038static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039{
NeilBrown52c03292006-06-26 00:27:43 -07004040 /* reshaping is quite different to recovery/resync so it is
4041 * handled quite separately ... here.
4042 *
4043 * On each call to sync_request, we gather one chunk worth of
4044 * destination stripes and flag them as expanding.
4045 * Then we find all the source stripes and request reads.
4046 * As the reads complete, handle_stripe will copy the data
4047 * into the destination stripe and release that stripe.
4048 */
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004049 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050 struct stripe_head *sh;
NeilBrownccfcc3c2006-03-27 01:18:09 -08004051 sector_t first_sector, last_sector;
NeilBrownf4168852007-02-28 20:11:53 -08004052 int raid_disks = conf->previous_raid_disks;
4053 int data_disks = raid_disks - conf->max_degraded;
4054 int new_data_disks = conf->raid_disks - conf->max_degraded;
NeilBrown52c03292006-06-26 00:27:43 -07004055 int i;
4056 int dd_idx;
NeilBrownc8f517c2009-03-31 15:28:40 +11004057 sector_t writepos, readpos, safepos;
NeilBrownec32a2b2009-03-31 15:17:38 +11004058 sector_t stripe_addr;
NeilBrown7a661382009-03-31 15:21:40 +11004059 int reshape_sectors;
NeilBrownab69ae12009-03-31 15:26:47 +11004060 struct list_head stripes;
NeilBrown52c03292006-06-26 00:27:43 -07004061
NeilBrownfef9c612009-03-31 15:16:46 +11004062 if (sector_nr == 0) {
4063 /* If restarting in the middle, skip the initial sectors */
4064 if (mddev->delta_disks < 0 &&
4065 conf->reshape_progress < raid5_size(mddev, 0, 0)) {
4066 sector_nr = raid5_size(mddev, 0, 0)
4067 - conf->reshape_progress;
NeilBrowna6397552009-08-13 10:13:00 +10004068 } else if (mddev->delta_disks >= 0 &&
NeilBrownfef9c612009-03-31 15:16:46 +11004069 conf->reshape_progress > 0)
4070 sector_nr = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004071 sector_div(sector_nr, new_data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004072 if (sector_nr) {
NeilBrown8dee7212009-11-06 14:59:29 +11004073 mddev->curr_resync_completed = sector_nr;
4074 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownfef9c612009-03-31 15:16:46 +11004075 *skipped = 1;
4076 return sector_nr;
4077 }
NeilBrown52c03292006-06-26 00:27:43 -07004078 }
4079
NeilBrown7a661382009-03-31 15:21:40 +11004080 /* We need to process a full chunk at a time.
4081 * If old and new chunk sizes differ, we need to process the
4082 * largest of these
4083 */
Andre Noll664e7c42009-06-18 08:45:27 +10004084 if (mddev->new_chunk_sectors > mddev->chunk_sectors)
4085 reshape_sectors = mddev->new_chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004086 else
Andre Noll9d8f0362009-06-18 08:45:01 +10004087 reshape_sectors = mddev->chunk_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004088
NeilBrown52c03292006-06-26 00:27:43 -07004089 /* we update the metadata when there is more than 3Meg
4090 * in the block range (that is rather arbitrary, should
4091 * probably be time based) or when the data about to be
4092 * copied would over-write the source of the data at
4093 * the front of the range.
NeilBrownfef9c612009-03-31 15:16:46 +11004094 * i.e. one new_stripe along from reshape_progress new_maps
4095 * to after where reshape_safe old_maps to
NeilBrown52c03292006-06-26 00:27:43 -07004096 */
NeilBrownfef9c612009-03-31 15:16:46 +11004097 writepos = conf->reshape_progress;
NeilBrownf4168852007-02-28 20:11:53 -08004098 sector_div(writepos, new_data_disks);
NeilBrownc8f517c2009-03-31 15:28:40 +11004099 readpos = conf->reshape_progress;
4100 sector_div(readpos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004101 safepos = conf->reshape_safe;
NeilBrownf4168852007-02-28 20:11:53 -08004102 sector_div(safepos, data_disks);
NeilBrownfef9c612009-03-31 15:16:46 +11004103 if (mddev->delta_disks < 0) {
NeilBrowned37d832009-05-27 21:39:05 +10004104 writepos -= min_t(sector_t, reshape_sectors, writepos);
NeilBrownc8f517c2009-03-31 15:28:40 +11004105 readpos += reshape_sectors;
NeilBrown7a661382009-03-31 15:21:40 +11004106 safepos += reshape_sectors;
NeilBrownfef9c612009-03-31 15:16:46 +11004107 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004108 writepos += reshape_sectors;
NeilBrowned37d832009-05-27 21:39:05 +10004109 readpos -= min_t(sector_t, reshape_sectors, readpos);
4110 safepos -= min_t(sector_t, reshape_sectors, safepos);
NeilBrownfef9c612009-03-31 15:16:46 +11004111 }
NeilBrown52c03292006-06-26 00:27:43 -07004112
NeilBrownc8f517c2009-03-31 15:28:40 +11004113 /* 'writepos' is the most advanced device address we might write.
4114 * 'readpos' is the least advanced device address we might read.
4115 * 'safepos' is the least address recorded in the metadata as having
4116 * been reshaped.
4117 * If 'readpos' is behind 'writepos', then there is no way that we can
4118 * ensure safety in the face of a crash - that must be done by userspace
4119 * making a backup of the data. So in that case there is no particular
4120 * rush to update metadata.
4121 * Otherwise if 'safepos' is behind 'writepos', then we really need to
4122 * update the metadata to advance 'safepos' to match 'readpos' so that
4123 * we can be safe in the event of a crash.
4124 * So we insist on updating metadata if safepos is behind writepos and
4125 * readpos is beyond writepos.
4126 * In any case, update the metadata every 10 seconds.
4127 * Maybe that number should be configurable, but I'm not sure it is
4128 * worth it.... maybe it could be a multiple of safemode_delay???
4129 */
NeilBrownfef9c612009-03-31 15:16:46 +11004130 if ((mddev->delta_disks < 0
NeilBrownc8f517c2009-03-31 15:28:40 +11004131 ? (safepos > writepos && readpos < writepos)
4132 : (safepos < writepos && readpos > writepos)) ||
4133 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
NeilBrown52c03292006-06-26 00:27:43 -07004134 /* Cannot proceed until we've updated the superblock... */
4135 wait_event(conf->wait_for_overlap,
4136 atomic_read(&conf->reshape_stripes)==0);
NeilBrownfef9c612009-03-31 15:16:46 +11004137 mddev->reshape_position = conf->reshape_progress;
NeilBrownacb180b2009-04-14 16:28:34 +10004138 mddev->curr_resync_completed = mddev->curr_resync;
NeilBrownc8f517c2009-03-31 15:28:40 +11004139 conf->reshape_checkpoint = jiffies;
NeilBrown850b2b42006-10-03 01:15:46 -07004140 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown52c03292006-06-26 00:27:43 -07004141 md_wakeup_thread(mddev->thread);
NeilBrown850b2b42006-10-03 01:15:46 -07004142 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrown52c03292006-06-26 00:27:43 -07004143 kthread_should_stop());
4144 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004145 conf->reshape_safe = mddev->reshape_position;
NeilBrown52c03292006-06-26 00:27:43 -07004146 spin_unlock_irq(&conf->device_lock);
4147 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004148 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown52c03292006-06-26 00:27:43 -07004149 }
4150
NeilBrownec32a2b2009-03-31 15:17:38 +11004151 if (mddev->delta_disks < 0) {
4152 BUG_ON(conf->reshape_progress == 0);
4153 stripe_addr = writepos;
4154 BUG_ON((mddev->dev_sectors &
NeilBrown7a661382009-03-31 15:21:40 +11004155 ~((sector_t)reshape_sectors - 1))
4156 - reshape_sectors - stripe_addr
NeilBrownec32a2b2009-03-31 15:17:38 +11004157 != sector_nr);
4158 } else {
NeilBrown7a661382009-03-31 15:21:40 +11004159 BUG_ON(writepos != sector_nr + reshape_sectors);
NeilBrownec32a2b2009-03-31 15:17:38 +11004160 stripe_addr = sector_nr;
4161 }
NeilBrownab69ae12009-03-31 15:26:47 +11004162 INIT_LIST_HEAD(&stripes);
NeilBrown7a661382009-03-31 15:21:40 +11004163 for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) {
NeilBrown52c03292006-06-26 00:27:43 -07004164 int j;
NeilBrowna9f326e2009-09-23 18:06:41 +10004165 int skipped_disk = 0;
NeilBrowna8c906c2009-06-09 14:39:59 +10004166 sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004167 set_bit(STRIPE_EXPANDING, &sh->state);
4168 atomic_inc(&conf->reshape_stripes);
4169 /* If any of this stripe is beyond the end of the old
4170 * array, then we need to zero those blocks
4171 */
4172 for (j=sh->disks; j--;) {
4173 sector_t s;
4174 if (j == sh->pd_idx)
4175 continue;
NeilBrownf4168852007-02-28 20:11:53 -08004176 if (conf->level == 6 &&
NeilBrownd0dabf72009-03-31 14:39:38 +11004177 j == sh->qd_idx)
NeilBrownf4168852007-02-28 20:11:53 -08004178 continue;
NeilBrown784052e2009-03-31 15:19:07 +11004179 s = compute_blocknr(sh, j, 0);
Dan Williamsb522adc2009-03-31 15:00:31 +11004180 if (s < raid5_size(mddev, 0, 0)) {
NeilBrowna9f326e2009-09-23 18:06:41 +10004181 skipped_disk = 1;
NeilBrown52c03292006-06-26 00:27:43 -07004182 continue;
4183 }
4184 memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE);
4185 set_bit(R5_Expanded, &sh->dev[j].flags);
4186 set_bit(R5_UPTODATE, &sh->dev[j].flags);
4187 }
NeilBrowna9f326e2009-09-23 18:06:41 +10004188 if (!skipped_disk) {
NeilBrown52c03292006-06-26 00:27:43 -07004189 set_bit(STRIPE_EXPAND_READY, &sh->state);
4190 set_bit(STRIPE_HANDLE, &sh->state);
4191 }
NeilBrownab69ae12009-03-31 15:26:47 +11004192 list_add(&sh->lru, &stripes);
NeilBrown52c03292006-06-26 00:27:43 -07004193 }
4194 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004195 if (mddev->delta_disks < 0)
NeilBrown7a661382009-03-31 15:21:40 +11004196 conf->reshape_progress -= reshape_sectors * new_data_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11004197 else
NeilBrown7a661382009-03-31 15:21:40 +11004198 conf->reshape_progress += reshape_sectors * new_data_disks;
NeilBrown52c03292006-06-26 00:27:43 -07004199 spin_unlock_irq(&conf->device_lock);
4200 /* Ok, those stripe are ready. We can start scheduling
4201 * reads on the source stripes.
4202 * The source stripes are determined by mapping the first and last
4203 * block on the destination stripes.
4204 */
NeilBrown52c03292006-06-26 00:27:43 -07004205 first_sector =
NeilBrownec32a2b2009-03-31 15:17:38 +11004206 raid5_compute_sector(conf, stripe_addr*(new_data_disks),
NeilBrown911d4ee2009-03-31 14:39:38 +11004207 1, &dd_idx, NULL);
NeilBrown52c03292006-06-26 00:27:43 -07004208 last_sector =
NeilBrown0e6e0272009-06-09 16:32:22 +10004209 raid5_compute_sector(conf, ((stripe_addr+reshape_sectors)
Andre Noll09c9e5f2009-06-18 08:45:55 +10004210 * new_data_disks - 1),
NeilBrown911d4ee2009-03-31 14:39:38 +11004211 1, &dd_idx, NULL);
Andre Noll58c0fed2009-03-31 14:33:13 +11004212 if (last_sector >= mddev->dev_sectors)
4213 last_sector = mddev->dev_sectors - 1;
NeilBrown52c03292006-06-26 00:27:43 -07004214 while (first_sector <= last_sector) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004215 sh = get_active_stripe(conf, first_sector, 1, 0, 1);
NeilBrown52c03292006-06-26 00:27:43 -07004216 set_bit(STRIPE_EXPAND_SOURCE, &sh->state);
4217 set_bit(STRIPE_HANDLE, &sh->state);
4218 release_stripe(sh);
4219 first_sector += STRIPE_SECTORS;
4220 }
NeilBrownab69ae12009-03-31 15:26:47 +11004221 /* Now that the sources are clearly marked, we can release
4222 * the destination stripes
4223 */
4224 while (!list_empty(&stripes)) {
4225 sh = list_entry(stripes.next, struct stripe_head, lru);
4226 list_del_init(&sh->lru);
4227 release_stripe(sh);
4228 }
NeilBrownc6207272008-02-06 01:39:52 -08004229 /* If this takes us to the resync_max point where we have to pause,
4230 * then we need to write out the superblock.
4231 */
NeilBrown7a661382009-03-31 15:21:40 +11004232 sector_nr += reshape_sectors;
NeilBrownc03f6a12009-04-17 11:06:30 +10004233 if ((sector_nr - mddev->curr_resync_completed) * 2
4234 >= mddev->resync_max - mddev->curr_resync_completed) {
NeilBrownc6207272008-02-06 01:39:52 -08004235 /* Cannot proceed until we've updated the superblock... */
4236 wait_event(conf->wait_for_overlap,
4237 atomic_read(&conf->reshape_stripes) == 0);
NeilBrownfef9c612009-03-31 15:16:46 +11004238 mddev->reshape_position = conf->reshape_progress;
NeilBrown48606a92009-06-18 09:14:12 +10004239 mddev->curr_resync_completed = mddev->curr_resync + reshape_sectors;
NeilBrownc8f517c2009-03-31 15:28:40 +11004240 conf->reshape_checkpoint = jiffies;
NeilBrownc6207272008-02-06 01:39:52 -08004241 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4242 md_wakeup_thread(mddev->thread);
4243 wait_event(mddev->sb_wait,
4244 !test_bit(MD_CHANGE_DEVS, &mddev->flags)
4245 || kthread_should_stop());
4246 spin_lock_irq(&conf->device_lock);
NeilBrownfef9c612009-03-31 15:16:46 +11004247 conf->reshape_safe = mddev->reshape_position;
NeilBrownc6207272008-02-06 01:39:52 -08004248 spin_unlock_irq(&conf->device_lock);
4249 wake_up(&conf->wait_for_overlap);
NeilBrownacb180b2009-04-14 16:28:34 +10004250 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrownc6207272008-02-06 01:39:52 -08004251 }
NeilBrown7a661382009-03-31 15:21:40 +11004252 return reshape_sectors;
NeilBrown52c03292006-06-26 00:27:43 -07004253}
4254
4255/* FIXME go_faster isn't used */
4256static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
4257{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11004258 raid5_conf_t *conf = mddev->private;
NeilBrown52c03292006-06-26 00:27:43 -07004259 struct stripe_head *sh;
Andre Noll58c0fed2009-03-31 14:33:13 +11004260 sector_t max_sector = mddev->dev_sectors;
NeilBrown72626682005-09-09 16:23:54 -07004261 int sync_blocks;
NeilBrown16a53ec2006-06-26 00:27:38 -07004262 int still_degraded = 0;
4263 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264
NeilBrown72626682005-09-09 16:23:54 -07004265 if (sector_nr >= max_sector) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004266 /* just being told to finish up .. nothing much to do */
4267 unplug_slaves(mddev);
NeilBrowncea9c222009-03-31 15:15:05 +11004268
NeilBrown29269552006-03-27 01:18:10 -08004269 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
4270 end_reshape(conf);
4271 return 0;
4272 }
NeilBrown72626682005-09-09 16:23:54 -07004273
4274 if (mddev->curr_resync < max_sector) /* aborted */
4275 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
4276 &sync_blocks, 1);
NeilBrown16a53ec2006-06-26 00:27:38 -07004277 else /* completed sync */
NeilBrown72626682005-09-09 16:23:54 -07004278 conf->fullsync = 0;
4279 bitmap_close_sync(mddev->bitmap);
4280
Linus Torvalds1da177e2005-04-16 15:20:36 -07004281 return 0;
4282 }
NeilBrownccfcc3c2006-03-27 01:18:09 -08004283
NeilBrown64bd6602009-08-03 10:59:58 +10004284 /* Allow raid5_quiesce to complete */
4285 wait_event(conf->wait_for_overlap, conf->quiesce != 2);
4286
NeilBrown52c03292006-06-26 00:27:43 -07004287 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4288 return reshape_request(mddev, sector_nr, skipped);
NeilBrownf6705572006-03-27 01:18:11 -08004289
NeilBrownc6207272008-02-06 01:39:52 -08004290 /* No need to check resync_max as we never do more than one
4291 * stripe, and as resync_max will always be on a chunk boundary,
4292 * if the check in md_do_sync didn't fire, there is no chance
4293 * of overstepping resync_max here
4294 */
4295
NeilBrown16a53ec2006-06-26 00:27:38 -07004296 /* if there is too many failed drives and we are trying
Linus Torvalds1da177e2005-04-16 15:20:36 -07004297 * to resync, then assert that we are finished, because there is
4298 * nothing we can do.
4299 */
NeilBrown3285edf2006-06-26 00:27:55 -07004300 if (mddev->degraded >= conf->max_degraded &&
NeilBrown16a53ec2006-06-26 00:27:38 -07004301 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004302 sector_t rv = mddev->dev_sectors - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07004303 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304 return rv;
4305 }
NeilBrown72626682005-09-09 16:23:54 -07004306 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrown3855ad92005-11-08 21:39:38 -08004307 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown72626682005-09-09 16:23:54 -07004308 !conf->fullsync && sync_blocks >= STRIPE_SECTORS) {
4309 /* we can skip this block, and probably more */
4310 sync_blocks /= STRIPE_SECTORS;
4311 *skipped = 1;
4312 return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */
4313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004314
NeilBrownb47490c2008-02-06 01:39:50 -08004315
4316 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
4317
NeilBrowna8c906c2009-06-09 14:39:59 +10004318 sh = get_active_stripe(conf, sector_nr, 0, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004319 if (sh == NULL) {
NeilBrowna8c906c2009-06-09 14:39:59 +10004320 sh = get_active_stripe(conf, sector_nr, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004321 /* make sure we don't swamp the stripe cache if someone else
NeilBrown16a53ec2006-06-26 00:27:38 -07004322 * is trying to get access
Linus Torvalds1da177e2005-04-16 15:20:36 -07004323 */
Nishanth Aravamudan66c006a2005-11-07 01:01:17 -08004324 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004325 }
NeilBrown16a53ec2006-06-26 00:27:38 -07004326 /* Need to check if array will still be degraded after recovery/resync
4327 * We don't need to check the 'failed' flag as when that gets set,
4328 * recovery aborts.
4329 */
NeilBrownf001a702009-06-09 14:30:31 +10004330 for (i = 0; i < conf->raid_disks; i++)
NeilBrown16a53ec2006-06-26 00:27:38 -07004331 if (conf->disks[i].rdev == NULL)
4332 still_degraded = 1;
4333
4334 bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded);
4335
4336 spin_lock(&sh->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004337 set_bit(STRIPE_SYNCING, &sh->state);
4338 clear_bit(STRIPE_INSYNC, &sh->state);
4339 spin_unlock(&sh->lock);
4340
NeilBrown14425772009-10-16 15:55:25 +11004341 handle_stripe(sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004342 release_stripe(sh);
4343
4344 return STRIPE_SECTORS;
4345}
4346
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004347static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio)
4348{
4349 /* We may not be able to submit a whole bio at once as there
4350 * may not be enough stripe_heads available.
4351 * We cannot pre-allocate enough stripe_heads as we may need
4352 * more than exist in the cache (if we allow ever large chunks).
4353 * So we do one stripe head at a time and record in
4354 * ->bi_hw_segments how many have been done.
4355 *
4356 * We *know* that this entire raid_bio is in one chunk, so
4357 * it will be only one 'dd_idx' and only need one call to raid5_compute_sector.
4358 */
4359 struct stripe_head *sh;
NeilBrown911d4ee2009-03-31 14:39:38 +11004360 int dd_idx;
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004361 sector_t sector, logical_sector, last_sector;
4362 int scnt = 0;
4363 int remaining;
4364 int handled = 0;
4365
4366 logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1);
NeilBrown112bf892009-03-31 14:39:38 +11004367 sector = raid5_compute_sector(conf, logical_sector,
NeilBrown911d4ee2009-03-31 14:39:38 +11004368 0, &dd_idx, NULL);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004369 last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9);
4370
4371 for (; logical_sector < last_sector;
Neil Brown387bb172007-02-08 14:20:29 -08004372 logical_sector += STRIPE_SECTORS,
4373 sector += STRIPE_SECTORS,
4374 scnt++) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004375
Jens Axboe960e7392008-08-15 10:41:18 +02004376 if (scnt < raid5_bi_hw_segments(raid_bio))
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004377 /* already done this stripe */
4378 continue;
4379
NeilBrowna8c906c2009-06-09 14:39:59 +10004380 sh = get_active_stripe(conf, sector, 0, 1, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004381
4382 if (!sh) {
4383 /* failed to get a stripe - must wait */
Jens Axboe960e7392008-08-15 10:41:18 +02004384 raid5_set_bi_hw_segments(raid_bio, scnt);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004385 conf->retry_read_aligned = raid_bio;
4386 return handled;
4387 }
4388
4389 set_bit(R5_ReadError, &sh->dev[dd_idx].flags);
Neil Brown387bb172007-02-08 14:20:29 -08004390 if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) {
4391 release_stripe(sh);
Jens Axboe960e7392008-08-15 10:41:18 +02004392 raid5_set_bi_hw_segments(raid_bio, scnt);
Neil Brown387bb172007-02-08 14:20:29 -08004393 conf->retry_read_aligned = raid_bio;
4394 return handled;
4395 }
4396
Dan Williams36d1c642009-07-14 11:48:22 -07004397 handle_stripe(sh);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004398 release_stripe(sh);
4399 handled++;
4400 }
4401 spin_lock_irq(&conf->device_lock);
Jens Axboe960e7392008-08-15 10:41:18 +02004402 remaining = raid5_dec_bi_phys_segments(raid_bio);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004403 spin_unlock_irq(&conf->device_lock);
Neil Brown0e13fe232008-06-28 08:31:20 +10004404 if (remaining == 0)
4405 bio_endio(raid_bio, 0);
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004406 if (atomic_dec_and_test(&conf->active_aligned_reads))
4407 wake_up(&conf->wait_for_stripe);
4408 return handled;
4409}
4410
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004411
Linus Torvalds1da177e2005-04-16 15:20:36 -07004412/*
4413 * This is our raid5 kernel thread.
4414 *
4415 * We scan the hash table for stripes which can be handled now.
4416 * During the scan, completed stripes are saved for us by the interrupt
4417 * handler, so that they will not have to wait for our next wakeup.
4418 */
NeilBrown6ed30032008-02-06 01:40:00 -08004419static void raid5d(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420{
4421 struct stripe_head *sh;
NeilBrown070ec552009-06-16 16:54:21 +10004422 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004423 int handled;
4424
Dan Williams45b42332007-07-09 11:56:43 -07004425 pr_debug("+++ raid5d active\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004426
4427 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004428
4429 handled = 0;
4430 spin_lock_irq(&conf->device_lock);
4431 while (1) {
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004432 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433
NeilBrownae3c20c2006-07-10 04:44:17 -07004434 if (conf->seq_flush != conf->seq_write) {
NeilBrown72626682005-09-09 16:23:54 -07004435 int seq = conf->seq_flush;
NeilBrown700e4322005-11-28 13:44:10 -08004436 spin_unlock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07004437 bitmap_unplug(mddev->bitmap);
NeilBrown700e4322005-11-28 13:44:10 -08004438 spin_lock_irq(&conf->device_lock);
NeilBrown72626682005-09-09 16:23:54 -07004439 conf->seq_write = seq;
4440 activate_bit_delay(conf);
4441 }
4442
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08004443 while ((bio = remove_bio_from_retry(conf))) {
4444 int ok;
4445 spin_unlock_irq(&conf->device_lock);
4446 ok = retry_aligned_read(conf, bio);
4447 spin_lock_irq(&conf->device_lock);
4448 if (!ok)
4449 break;
4450 handled++;
4451 }
4452
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004453 sh = __get_priority_stripe(conf);
4454
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004455 if (!sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004457 spin_unlock_irq(&conf->device_lock);
4458
4459 handled++;
Dan Williams417b8d42009-10-16 16:25:22 +11004460 handle_stripe(sh);
4461 release_stripe(sh);
4462 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004463
4464 spin_lock_irq(&conf->device_lock);
4465 }
Dan Williams45b42332007-07-09 11:56:43 -07004466 pr_debug("%d stripes handled\n", handled);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004467
4468 spin_unlock_irq(&conf->device_lock);
4469
Dan Williamsc9f21aa2008-07-23 12:05:51 -07004470 async_tx_issue_pending_all();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004471 unplug_slaves(mddev);
4472
Dan Williams45b42332007-07-09 11:56:43 -07004473 pr_debug("--- raid5d inactive\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004474}
4475
NeilBrown3f294f42005-11-08 21:39:25 -08004476static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004477raid5_show_stripe_cache_size(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004478{
NeilBrown070ec552009-06-16 16:54:21 +10004479 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004480 if (conf)
4481 return sprintf(page, "%d\n", conf->max_nr_stripes);
4482 else
4483 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004484}
4485
4486static ssize_t
NeilBrown007583c2005-11-08 21:39:30 -08004487raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len)
NeilBrown3f294f42005-11-08 21:39:25 -08004488{
NeilBrown070ec552009-06-16 16:54:21 +10004489 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004490 unsigned long new;
Dan Williamsb5470dc2008-06-27 21:44:04 -07004491 int err;
4492
NeilBrown3f294f42005-11-08 21:39:25 -08004493 if (len >= PAGE_SIZE)
4494 return -EINVAL;
NeilBrown96de1e62005-11-08 21:39:39 -08004495 if (!conf)
4496 return -ENODEV;
NeilBrown3f294f42005-11-08 21:39:25 -08004497
Dan Williams4ef197d82008-04-28 02:15:54 -07004498 if (strict_strtoul(page, 10, &new))
NeilBrown3f294f42005-11-08 21:39:25 -08004499 return -EINVAL;
4500 if (new <= 16 || new > 32768)
4501 return -EINVAL;
4502 while (new < conf->max_nr_stripes) {
4503 if (drop_one_stripe(conf))
4504 conf->max_nr_stripes--;
4505 else
4506 break;
4507 }
Dan Williamsb5470dc2008-06-27 21:44:04 -07004508 err = md_allow_write(mddev);
4509 if (err)
4510 return err;
NeilBrown3f294f42005-11-08 21:39:25 -08004511 while (new > conf->max_nr_stripes) {
4512 if (grow_one_stripe(conf))
4513 conf->max_nr_stripes++;
4514 else break;
4515 }
4516 return len;
4517}
NeilBrown007583c2005-11-08 21:39:30 -08004518
NeilBrown96de1e62005-11-08 21:39:39 -08004519static struct md_sysfs_entry
4520raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR,
4521 raid5_show_stripe_cache_size,
4522 raid5_store_stripe_cache_size);
NeilBrown3f294f42005-11-08 21:39:25 -08004523
4524static ssize_t
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004525raid5_show_preread_threshold(mddev_t *mddev, char *page)
4526{
NeilBrown070ec552009-06-16 16:54:21 +10004527 raid5_conf_t *conf = mddev->private;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004528 if (conf)
4529 return sprintf(page, "%d\n", conf->bypass_threshold);
4530 else
4531 return 0;
4532}
4533
4534static ssize_t
4535raid5_store_preread_threshold(mddev_t *mddev, const char *page, size_t len)
4536{
NeilBrown070ec552009-06-16 16:54:21 +10004537 raid5_conf_t *conf = mddev->private;
Dan Williams4ef197d82008-04-28 02:15:54 -07004538 unsigned long new;
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004539 if (len >= PAGE_SIZE)
4540 return -EINVAL;
4541 if (!conf)
4542 return -ENODEV;
4543
Dan Williams4ef197d82008-04-28 02:15:54 -07004544 if (strict_strtoul(page, 10, &new))
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004545 return -EINVAL;
Dan Williams4ef197d82008-04-28 02:15:54 -07004546 if (new > conf->max_nr_stripes)
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004547 return -EINVAL;
4548 conf->bypass_threshold = new;
4549 return len;
4550}
4551
4552static struct md_sysfs_entry
4553raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold,
4554 S_IRUGO | S_IWUSR,
4555 raid5_show_preread_threshold,
4556 raid5_store_preread_threshold);
4557
4558static ssize_t
NeilBrown96de1e62005-11-08 21:39:39 -08004559stripe_cache_active_show(mddev_t *mddev, char *page)
NeilBrown3f294f42005-11-08 21:39:25 -08004560{
NeilBrown070ec552009-06-16 16:54:21 +10004561 raid5_conf_t *conf = mddev->private;
NeilBrown96de1e62005-11-08 21:39:39 -08004562 if (conf)
4563 return sprintf(page, "%d\n", atomic_read(&conf->active_stripes));
4564 else
4565 return 0;
NeilBrown3f294f42005-11-08 21:39:25 -08004566}
4567
NeilBrown96de1e62005-11-08 21:39:39 -08004568static struct md_sysfs_entry
4569raid5_stripecache_active = __ATTR_RO(stripe_cache_active);
NeilBrown3f294f42005-11-08 21:39:25 -08004570
NeilBrown007583c2005-11-08 21:39:30 -08004571static struct attribute *raid5_attrs[] = {
NeilBrown3f294f42005-11-08 21:39:25 -08004572 &raid5_stripecache_size.attr,
4573 &raid5_stripecache_active.attr,
Dan Williams8b3e6cd2008-04-28 02:15:53 -07004574 &raid5_preread_bypass_threshold.attr,
NeilBrown3f294f42005-11-08 21:39:25 -08004575 NULL,
4576};
NeilBrown007583c2005-11-08 21:39:30 -08004577static struct attribute_group raid5_attrs_group = {
4578 .name = NULL,
4579 .attrs = raid5_attrs,
NeilBrown3f294f42005-11-08 21:39:25 -08004580};
4581
Dan Williams80c3a6c2009-03-17 18:10:40 -07004582static sector_t
4583raid5_size(mddev_t *mddev, sector_t sectors, int raid_disks)
4584{
NeilBrown070ec552009-06-16 16:54:21 +10004585 raid5_conf_t *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07004586
4587 if (!sectors)
4588 sectors = mddev->dev_sectors;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004589 if (!raid_disks)
NeilBrown7ec05472009-03-31 15:10:36 +11004590 /* size is defined by the smallest of previous and new size */
NeilBrown5e5e3e72009-10-16 16:35:30 +11004591 raid_disks = min(conf->raid_disks, conf->previous_raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004592
Andre Noll9d8f0362009-06-18 08:45:01 +10004593 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Andre Noll664e7c42009-06-18 08:45:27 +10004594 sectors &= ~((sector_t)mddev->new_chunk_sectors - 1);
Dan Williams80c3a6c2009-03-17 18:10:40 -07004595 return sectors * (raid_disks - conf->max_degraded);
4596}
4597
Dan Williams36d1c642009-07-14 11:48:22 -07004598static void raid5_free_percpu(raid5_conf_t *conf)
4599{
4600 struct raid5_percpu *percpu;
4601 unsigned long cpu;
4602
4603 if (!conf->percpu)
4604 return;
4605
4606 get_online_cpus();
4607 for_each_possible_cpu(cpu) {
4608 percpu = per_cpu_ptr(conf->percpu, cpu);
4609 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004610 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004611 }
4612#ifdef CONFIG_HOTPLUG_CPU
4613 unregister_cpu_notifier(&conf->cpu_notify);
4614#endif
4615 put_online_cpus();
4616
4617 free_percpu(conf->percpu);
4618}
4619
Dan Williams95fc17a2009-07-31 12:39:15 +10004620static void free_conf(raid5_conf_t *conf)
4621{
4622 shrink_stripes(conf);
Dan Williams36d1c642009-07-14 11:48:22 -07004623 raid5_free_percpu(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10004624 kfree(conf->disks);
4625 kfree(conf->stripe_hashtbl);
4626 kfree(conf);
4627}
4628
Dan Williams36d1c642009-07-14 11:48:22 -07004629#ifdef CONFIG_HOTPLUG_CPU
4630static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action,
4631 void *hcpu)
4632{
4633 raid5_conf_t *conf = container_of(nfb, raid5_conf_t, cpu_notify);
4634 long cpu = (long)hcpu;
4635 struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu);
4636
4637 switch (action) {
4638 case CPU_UP_PREPARE:
4639 case CPU_UP_PREPARE_FROZEN:
Dan Williamsd6f38f32009-07-14 11:50:52 -07004640 if (conf->level == 6 && !percpu->spare_page)
Dan Williams36d1c642009-07-14 11:48:22 -07004641 percpu->spare_page = alloc_page(GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004642 if (!percpu->scribble)
4643 percpu->scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
4644
4645 if (!percpu->scribble ||
4646 (conf->level == 6 && !percpu->spare_page)) {
4647 safe_put_page(percpu->spare_page);
4648 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004649 pr_err("%s: failed memory allocation for cpu%ld\n",
4650 __func__, cpu);
4651 return NOTIFY_BAD;
4652 }
4653 break;
4654 case CPU_DEAD:
4655 case CPU_DEAD_FROZEN:
4656 safe_put_page(percpu->spare_page);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004657 kfree(percpu->scribble);
Dan Williams36d1c642009-07-14 11:48:22 -07004658 percpu->spare_page = NULL;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004659 percpu->scribble = NULL;
Dan Williams36d1c642009-07-14 11:48:22 -07004660 break;
4661 default:
4662 break;
4663 }
4664 return NOTIFY_OK;
4665}
4666#endif
4667
4668static int raid5_alloc_percpu(raid5_conf_t *conf)
4669{
4670 unsigned long cpu;
4671 struct page *spare_page;
Tejun Heoa29d8b82010-02-02 14:39:15 +09004672 struct raid5_percpu __percpu *allcpus;
Dan Williamsd6f38f32009-07-14 11:50:52 -07004673 void *scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004674 int err;
4675
Dan Williams36d1c642009-07-14 11:48:22 -07004676 allcpus = alloc_percpu(struct raid5_percpu);
4677 if (!allcpus)
4678 return -ENOMEM;
4679 conf->percpu = allcpus;
4680
4681 get_online_cpus();
4682 err = 0;
4683 for_each_present_cpu(cpu) {
Dan Williamsd6f38f32009-07-14 11:50:52 -07004684 if (conf->level == 6) {
4685 spare_page = alloc_page(GFP_KERNEL);
4686 if (!spare_page) {
4687 err = -ENOMEM;
4688 break;
4689 }
4690 per_cpu_ptr(conf->percpu, cpu)->spare_page = spare_page;
4691 }
NeilBrown5e5e3e72009-10-16 16:35:30 +11004692 scribble = kmalloc(conf->scribble_len, GFP_KERNEL);
Dan Williamsd6f38f32009-07-14 11:50:52 -07004693 if (!scribble) {
Dan Williams36d1c642009-07-14 11:48:22 -07004694 err = -ENOMEM;
4695 break;
4696 }
Dan Williamsd6f38f32009-07-14 11:50:52 -07004697 per_cpu_ptr(conf->percpu, cpu)->scribble = scribble;
Dan Williams36d1c642009-07-14 11:48:22 -07004698 }
4699#ifdef CONFIG_HOTPLUG_CPU
4700 conf->cpu_notify.notifier_call = raid456_cpu_notify;
4701 conf->cpu_notify.priority = 0;
4702 if (err == 0)
4703 err = register_cpu_notifier(&conf->cpu_notify);
4704#endif
4705 put_online_cpus();
4706
4707 return err;
4708}
4709
NeilBrown91adb562009-03-31 14:39:39 +11004710static raid5_conf_t *setup_conf(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004711{
4712 raid5_conf_t *conf;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004713 int raid_disk, memory, max_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004714 mdk_rdev_t *rdev;
4715 struct disk_info *disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716
NeilBrown91adb562009-03-31 14:39:39 +11004717 if (mddev->new_level != 5
4718 && mddev->new_level != 4
4719 && mddev->new_level != 6) {
NeilBrown0c55e022010-05-03 14:09:02 +10004720 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004721 mdname(mddev), mddev->new_level);
4722 return ERR_PTR(-EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004723 }
NeilBrown91adb562009-03-31 14:39:39 +11004724 if ((mddev->new_level == 5
4725 && !algorithm_valid_raid5(mddev->new_layout)) ||
4726 (mddev->new_level == 6
4727 && !algorithm_valid_raid6(mddev->new_layout))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004728 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
NeilBrown91adb562009-03-31 14:39:39 +11004729 mdname(mddev), mddev->new_layout);
4730 return ERR_PTR(-EIO);
4731 }
4732 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
NeilBrown0c55e022010-05-03 14:09:02 +10004733 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
NeilBrown91adb562009-03-31 14:39:39 +11004734 mdname(mddev), mddev->raid_disks);
4735 return ERR_PTR(-EINVAL);
NeilBrown99c0fb52009-03-31 14:39:38 +11004736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004737
Andre Noll664e7c42009-06-18 08:45:27 +10004738 if (!mddev->new_chunk_sectors ||
4739 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
4740 !is_power_of_2(mddev->new_chunk_sectors)) {
NeilBrown0c55e022010-05-03 14:09:02 +10004741 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
4742 mdname(mddev), mddev->new_chunk_sectors << 9);
NeilBrown91adb562009-03-31 14:39:39 +11004743 return ERR_PTR(-EINVAL);
NeilBrown4bbf3772008-10-13 11:55:12 +11004744 }
4745
NeilBrown91adb562009-03-31 14:39:39 +11004746 conf = kzalloc(sizeof(raid5_conf_t), GFP_KERNEL);
4747 if (conf == NULL)
4748 goto abort;
Dan Williamsf5efd452009-10-16 15:55:38 +11004749 spin_lock_init(&conf->device_lock);
4750 init_waitqueue_head(&conf->wait_for_stripe);
4751 init_waitqueue_head(&conf->wait_for_overlap);
4752 INIT_LIST_HEAD(&conf->handle_list);
4753 INIT_LIST_HEAD(&conf->hold_list);
4754 INIT_LIST_HEAD(&conf->delayed_list);
4755 INIT_LIST_HEAD(&conf->bitmap_list);
4756 INIT_LIST_HEAD(&conf->inactive_list);
4757 atomic_set(&conf->active_stripes, 0);
4758 atomic_set(&conf->preread_active_stripes, 0);
4759 atomic_set(&conf->active_aligned_reads, 0);
4760 conf->bypass_threshold = BYPASS_THRESHOLD;
NeilBrown91adb562009-03-31 14:39:39 +11004761
4762 conf->raid_disks = mddev->raid_disks;
4763 if (mddev->reshape_position == MaxSector)
4764 conf->previous_raid_disks = mddev->raid_disks;
4765 else
4766 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004767 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
4768 conf->scribble_len = scribble_len(max_disks);
NeilBrown91adb562009-03-31 14:39:39 +11004769
NeilBrown5e5e3e72009-10-16 16:35:30 +11004770 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
NeilBrown91adb562009-03-31 14:39:39 +11004771 GFP_KERNEL);
4772 if (!conf->disks)
4773 goto abort;
4774
4775 conf->mddev = mddev;
4776
4777 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
4778 goto abort;
4779
Dan Williams36d1c642009-07-14 11:48:22 -07004780 conf->level = mddev->new_level;
4781 if (raid5_alloc_percpu(conf) != 0)
4782 goto abort;
4783
NeilBrown0c55e022010-05-03 14:09:02 +10004784 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
NeilBrown91adb562009-03-31 14:39:39 +11004785
4786 list_for_each_entry(rdev, &mddev->disks, same_set) {
4787 raid_disk = rdev->raid_disk;
NeilBrown5e5e3e72009-10-16 16:35:30 +11004788 if (raid_disk >= max_disks
NeilBrown91adb562009-03-31 14:39:39 +11004789 || raid_disk < 0)
4790 continue;
4791 disk = conf->disks + raid_disk;
4792
4793 disk->rdev = rdev;
4794
4795 if (test_bit(In_sync, &rdev->flags)) {
4796 char b[BDEVNAME_SIZE];
NeilBrown0c55e022010-05-03 14:09:02 +10004797 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
4798 " disk %d\n",
4799 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
NeilBrown91adb562009-03-31 14:39:39 +11004800 } else
4801 /* Cannot rely on bitmap to complete recovery */
4802 conf->fullsync = 1;
4803 }
4804
Andre Noll09c9e5f2009-06-18 08:45:55 +10004805 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown91adb562009-03-31 14:39:39 +11004806 conf->level = mddev->new_level;
4807 if (conf->level == 6)
4808 conf->max_degraded = 2;
4809 else
4810 conf->max_degraded = 1;
4811 conf->algorithm = mddev->new_layout;
4812 conf->max_nr_stripes = NR_STRIPES;
NeilBrownfef9c612009-03-31 15:16:46 +11004813 conf->reshape_progress = mddev->reshape_position;
NeilBrowne183eae2009-03-31 15:20:22 +11004814 if (conf->reshape_progress != MaxSector) {
Andre Noll09c9e5f2009-06-18 08:45:55 +10004815 conf->prev_chunk_sectors = mddev->chunk_sectors;
NeilBrowne183eae2009-03-31 15:20:22 +11004816 conf->prev_algo = mddev->layout;
4817 }
NeilBrown91adb562009-03-31 14:39:39 +11004818
4819 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
NeilBrown5e5e3e72009-10-16 16:35:30 +11004820 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
NeilBrown91adb562009-03-31 14:39:39 +11004821 if (grow_stripes(conf, conf->max_nr_stripes)) {
4822 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004823 "md/raid:%s: couldn't allocate %dkB for buffers\n",
4824 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004825 goto abort;
4826 } else
NeilBrown0c55e022010-05-03 14:09:02 +10004827 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
4828 mdname(mddev), memory);
NeilBrown91adb562009-03-31 14:39:39 +11004829
NeilBrown0da3c612009-09-23 18:09:45 +10004830 conf->thread = md_register_thread(raid5d, mddev, NULL);
NeilBrown91adb562009-03-31 14:39:39 +11004831 if (!conf->thread) {
4832 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10004833 "md/raid:%s: couldn't allocate thread.\n",
NeilBrown91adb562009-03-31 14:39:39 +11004834 mdname(mddev));
4835 goto abort;
4836 }
4837
4838 return conf;
4839
4840 abort:
4841 if (conf) {
Dan Williams95fc17a2009-07-31 12:39:15 +10004842 free_conf(conf);
NeilBrown91adb562009-03-31 14:39:39 +11004843 return ERR_PTR(-EIO);
4844 } else
4845 return ERR_PTR(-ENOMEM);
4846}
4847
NeilBrownc148ffd2009-11-13 17:47:00 +11004848
4849static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded)
4850{
4851 switch (algo) {
4852 case ALGORITHM_PARITY_0:
4853 if (raid_disk < max_degraded)
4854 return 1;
4855 break;
4856 case ALGORITHM_PARITY_N:
4857 if (raid_disk >= raid_disks - max_degraded)
4858 return 1;
4859 break;
4860 case ALGORITHM_PARITY_0_6:
4861 if (raid_disk == 0 ||
4862 raid_disk == raid_disks - 1)
4863 return 1;
4864 break;
4865 case ALGORITHM_LEFT_ASYMMETRIC_6:
4866 case ALGORITHM_RIGHT_ASYMMETRIC_6:
4867 case ALGORITHM_LEFT_SYMMETRIC_6:
4868 case ALGORITHM_RIGHT_SYMMETRIC_6:
4869 if (raid_disk == raid_disks - 1)
4870 return 1;
4871 }
4872 return 0;
4873}
4874
NeilBrown91adb562009-03-31 14:39:39 +11004875static int run(mddev_t *mddev)
4876{
4877 raid5_conf_t *conf;
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10004878 int working_disks = 0, chunk_size;
NeilBrownc148ffd2009-11-13 17:47:00 +11004879 int dirty_parity_disks = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004880 mdk_rdev_t *rdev;
NeilBrownc148ffd2009-11-13 17:47:00 +11004881 sector_t reshape_offset = 0;
NeilBrown91adb562009-03-31 14:39:39 +11004882
Andre Noll8c6ac862009-06-18 08:48:06 +10004883 if (mddev->recovery_cp != MaxSector)
NeilBrown0c55e022010-05-03 14:09:02 +10004884 printk(KERN_NOTICE "md/raid:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10004885 " -- starting background reconstruction\n",
4886 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004887 if (mddev->reshape_position != MaxSector) {
4888 /* Check that we can continue the reshape.
4889 * Currently only disks can change, it must
4890 * increase, and we must be past the point where
4891 * a stripe over-writes itself
4892 */
4893 sector_t here_new, here_old;
4894 int old_disks;
Andre Noll18b00332009-03-31 15:00:56 +11004895 int max_degraded = (mddev->level == 6 ? 2 : 1);
NeilBrownf6705572006-03-27 01:18:11 -08004896
NeilBrown88ce4932009-03-31 15:24:23 +11004897 if (mddev->new_level != mddev->level) {
NeilBrown0c55e022010-05-03 14:09:02 +10004898 printk(KERN_ERR "md/raid:%s: unsupported reshape "
NeilBrownf4168852007-02-28 20:11:53 -08004899 "required - aborting.\n",
NeilBrownf6705572006-03-27 01:18:11 -08004900 mdname(mddev));
4901 return -EINVAL;
4902 }
NeilBrownf6705572006-03-27 01:18:11 -08004903 old_disks = mddev->raid_disks - mddev->delta_disks;
4904 /* reshape_position must be on a new-stripe boundary, and one
NeilBrownf4168852007-02-28 20:11:53 -08004905 * further up in new geometry must map after here in old
4906 * geometry.
NeilBrownf6705572006-03-27 01:18:11 -08004907 */
4908 here_new = mddev->reshape_position;
Andre Noll664e7c42009-06-18 08:45:27 +10004909 if (sector_div(here_new, mddev->new_chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004910 (mddev->raid_disks - max_degraded))) {
NeilBrown0c55e022010-05-03 14:09:02 +10004911 printk(KERN_ERR "md/raid:%s: reshape_position not "
4912 "on a stripe boundary\n", mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004913 return -EINVAL;
4914 }
NeilBrownc148ffd2009-11-13 17:47:00 +11004915 reshape_offset = here_new * mddev->new_chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08004916 /* here_new is the stripe we will write to */
4917 here_old = mddev->reshape_position;
Andre Noll9d8f0362009-06-18 08:45:01 +10004918 sector_div(here_old, mddev->chunk_sectors *
NeilBrownf4168852007-02-28 20:11:53 -08004919 (old_disks-max_degraded));
4920 /* here_old is the first stripe that we might need to read
4921 * from */
NeilBrown67ac6012009-08-13 10:06:24 +10004922 if (mddev->delta_disks == 0) {
4923 /* We cannot be sure it is safe to start an in-place
4924 * reshape. It is only safe if user-space if monitoring
4925 * and taking constant backups.
4926 * mdadm always starts a situation like this in
4927 * readonly mode so it can take control before
4928 * allowing any writes. So just check for that.
4929 */
4930 if ((here_new * mddev->new_chunk_sectors !=
4931 here_old * mddev->chunk_sectors) ||
4932 mddev->ro == 0) {
NeilBrown0c55e022010-05-03 14:09:02 +10004933 printk(KERN_ERR "md/raid:%s: in-place reshape must be started"
4934 " in read-only mode - aborting\n",
4935 mdname(mddev));
NeilBrown67ac6012009-08-13 10:06:24 +10004936 return -EINVAL;
4937 }
4938 } else if (mddev->delta_disks < 0
4939 ? (here_new * mddev->new_chunk_sectors <=
4940 here_old * mddev->chunk_sectors)
4941 : (here_new * mddev->new_chunk_sectors >=
4942 here_old * mddev->chunk_sectors)) {
NeilBrownf6705572006-03-27 01:18:11 -08004943 /* Reading from the same stripe as writing to - bad */
NeilBrown0c55e022010-05-03 14:09:02 +10004944 printk(KERN_ERR "md/raid:%s: reshape_position too early for "
4945 "auto-recovery - aborting.\n",
4946 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004947 return -EINVAL;
4948 }
NeilBrown0c55e022010-05-03 14:09:02 +10004949 printk(KERN_INFO "md/raid:%s: reshape will continue\n",
4950 mdname(mddev));
NeilBrownf6705572006-03-27 01:18:11 -08004951 /* OK, we should be able to continue; */
NeilBrownf6705572006-03-27 01:18:11 -08004952 } else {
NeilBrown91adb562009-03-31 14:39:39 +11004953 BUG_ON(mddev->level != mddev->new_level);
4954 BUG_ON(mddev->layout != mddev->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10004955 BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors);
NeilBrown91adb562009-03-31 14:39:39 +11004956 BUG_ON(mddev->delta_disks != 0);
NeilBrownf6705572006-03-27 01:18:11 -08004957 }
4958
NeilBrown245f46c2009-03-31 14:39:39 +11004959 if (mddev->private == NULL)
4960 conf = setup_conf(mddev);
4961 else
4962 conf = mddev->private;
4963
NeilBrown91adb562009-03-31 14:39:39 +11004964 if (IS_ERR(conf))
4965 return PTR_ERR(conf);
NeilBrown9ffae0c2006-01-06 00:20:32 -08004966
NeilBrown91adb562009-03-31 14:39:39 +11004967 mddev->thread = conf->thread;
4968 conf->thread = NULL;
4969 mddev->private = conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004970
Linus Torvalds1da177e2005-04-16 15:20:36 -07004971 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07004972 * 0 for a fully functional array, 1 or 2 for a degraded array.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004973 */
NeilBrownc148ffd2009-11-13 17:47:00 +11004974 list_for_each_entry(rdev, &mddev->disks, same_set) {
4975 if (rdev->raid_disk < 0)
4976 continue;
4977 if (test_bit(In_sync, &rdev->flags))
NeilBrown91adb562009-03-31 14:39:39 +11004978 working_disks++;
NeilBrownc148ffd2009-11-13 17:47:00 +11004979 /* This disc is not fully in-sync. However if it
4980 * just stored parity (beyond the recovery_offset),
4981 * when we don't need to be concerned about the
4982 * array being dirty.
4983 * When reshape goes 'backwards', we never have
4984 * partially completed devices, so we only need
4985 * to worry about reshape going forwards.
4986 */
4987 /* Hack because v0.91 doesn't store recovery_offset properly. */
4988 if (mddev->major_version == 0 &&
4989 mddev->minor_version > 90)
4990 rdev->recovery_offset = reshape_offset;
4991
NeilBrownc148ffd2009-11-13 17:47:00 +11004992 if (rdev->recovery_offset < reshape_offset) {
4993 /* We need to check old and new layout */
4994 if (!only_parity(rdev->raid_disk,
4995 conf->algorithm,
4996 conf->raid_disks,
4997 conf->max_degraded))
4998 continue;
4999 }
5000 if (!only_parity(rdev->raid_disk,
5001 conf->prev_algo,
5002 conf->previous_raid_disks,
5003 conf->max_degraded))
5004 continue;
5005 dirty_parity_disks++;
5006 }
NeilBrown91adb562009-03-31 14:39:39 +11005007
NeilBrown5e5e3e72009-10-16 16:35:30 +11005008 mddev->degraded = (max(conf->raid_disks, conf->previous_raid_disks)
5009 - working_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005010
NeilBrown16a53ec2006-06-26 00:27:38 -07005011 if (mddev->degraded > conf->max_degraded) {
NeilBrown0c55e022010-05-03 14:09:02 +10005012 printk(KERN_ERR "md/raid:%s: not enough operational devices"
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013 " (%d/%d failed)\n",
NeilBrown02c2de82006-10-03 01:15:47 -07005014 mdname(mddev), mddev->degraded, conf->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005015 goto abort;
5016 }
5017
NeilBrown91adb562009-03-31 14:39:39 +11005018 /* device size must be a multiple of chunk size */
Andre Noll9d8f0362009-06-18 08:45:01 +10005019 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
NeilBrown91adb562009-03-31 14:39:39 +11005020 mddev->resync_max_sectors = mddev->dev_sectors;
5021
NeilBrownc148ffd2009-11-13 17:47:00 +11005022 if (mddev->degraded > dirty_parity_disks &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07005023 mddev->recovery_cp != MaxSector) {
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005024 if (mddev->ok_start_degraded)
5025 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10005026 "md/raid:%s: starting dirty degraded array"
5027 " - data corruption possible.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005028 mdname(mddev));
5029 else {
5030 printk(KERN_ERR
NeilBrown0c55e022010-05-03 14:09:02 +10005031 "md/raid:%s: cannot start dirty degraded array.\n",
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005032 mdname(mddev));
5033 goto abort;
5034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005035 }
5036
Linus Torvalds1da177e2005-04-16 15:20:36 -07005037 if (mddev->degraded == 0)
NeilBrown0c55e022010-05-03 14:09:02 +10005038 printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d"
5039 " devices, algorithm %d\n", mdname(mddev), conf->level,
NeilBrowne183eae2009-03-31 15:20:22 +11005040 mddev->raid_disks-mddev->degraded, mddev->raid_disks,
5041 mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005042 else
NeilBrown0c55e022010-05-03 14:09:02 +10005043 printk(KERN_ALERT "md/raid:%s: raid level %d active with %d"
5044 " out of %d devices, algorithm %d\n",
5045 mdname(mddev), conf->level,
5046 mddev->raid_disks - mddev->degraded,
5047 mddev->raid_disks, mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005048
5049 print_raid5_conf(conf);
5050
NeilBrownfef9c612009-03-31 15:16:46 +11005051 if (conf->reshape_progress != MaxSector) {
NeilBrownfef9c612009-03-31 15:16:46 +11005052 conf->reshape_safe = conf->reshape_progress;
NeilBrownf6705572006-03-27 01:18:11 -08005053 atomic_set(&conf->reshape_stripes, 0);
5054 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5055 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5056 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5057 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5058 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005059 "reshape");
NeilBrownf6705572006-03-27 01:18:11 -08005060 }
5061
Linus Torvalds1da177e2005-04-16 15:20:36 -07005062 /* read-ahead size must cover two whole stripes, which is
NeilBrown16a53ec2006-06-26 00:27:38 -07005063 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
Linus Torvalds1da177e2005-04-16 15:20:36 -07005064 */
5065 {
NeilBrown16a53ec2006-06-26 00:27:38 -07005066 int data_disks = conf->previous_raid_disks - conf->max_degraded;
5067 int stripe = data_disks *
Andre Noll9d8f0362009-06-18 08:45:01 +10005068 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005069 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5070 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5071 }
5072
5073 /* Ok, everything is just fine now */
NeilBrowna64c8762010-04-14 17:15:37 +10005074 if (mddev->to_remove == &raid5_attrs_group)
5075 mddev->to_remove = NULL;
5076 else if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08005077 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10005078 "md/raid:%s: failed to create sysfs attributes.\n",
NeilBrown5e55e2f2007-03-26 21:32:14 -08005079 mdname(mddev));
NeilBrown7a5febe2005-05-16 21:53:16 -07005080
NeilBrown91adb562009-03-31 14:39:39 +11005081 mddev->queue->queue_lock = &conf->device_lock;
5082
NeilBrown7a5febe2005-05-16 21:53:16 -07005083 mddev->queue->unplug_fn = raid5_unplug_device;
NeilBrownf022b2f2006-10-03 01:15:56 -07005084 mddev->queue->backing_dev_info.congested_data = mddev;
NeilBrown041ae522007-03-26 21:32:14 -08005085 mddev->queue->backing_dev_info.congested_fn = raid5_congested;
NeilBrownf022b2f2006-10-03 01:15:56 -07005086
Dan Williams1f403622009-03-31 14:59:03 +11005087 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
NeilBrown7a5febe2005-05-16 21:53:16 -07005088
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08005089 blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec);
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10005090 chunk_size = mddev->chunk_sectors << 9;
5091 blk_queue_io_min(mddev->queue, chunk_size);
5092 blk_queue_io_opt(mddev->queue, chunk_size *
5093 (conf->raid_disks - conf->max_degraded));
5094
5095 list_for_each_entry(rdev, &mddev->disks, same_set)
5096 disk_stack_limits(mddev->gendisk, rdev->bdev,
5097 rdev->data_offset << 9);
Raz Ben-Jehuda(caro)23032a02006-12-10 02:20:45 -08005098
Linus Torvalds1da177e2005-04-16 15:20:36 -07005099 return 0;
5100abort:
NeilBrowne0cf8f02009-03-31 14:39:39 +11005101 md_unregister_thread(mddev->thread);
NeilBrown91adb562009-03-31 14:39:39 +11005102 mddev->thread = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005103 if (conf) {
5104 print_raid5_conf(conf);
Dan Williams95fc17a2009-07-31 12:39:15 +10005105 free_conf(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005106 }
5107 mddev->private = NULL;
NeilBrown0c55e022010-05-03 14:09:02 +10005108 printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005109 return -EIO;
5110}
5111
NeilBrown3f294f42005-11-08 21:39:25 -08005112static int stop(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005113{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005114 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005115
5116 md_unregister_thread(mddev->thread);
5117 mddev->thread = NULL;
NeilBrown041ae522007-03-26 21:32:14 -08005118 mddev->queue->backing_dev_info.congested_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005119 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
Dan Williams95fc17a2009-07-31 12:39:15 +10005120 free_conf(conf);
NeilBrowna64c8762010-04-14 17:15:37 +10005121 mddev->private = NULL;
5122 mddev->to_remove = &raid5_attrs_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005123 return 0;
5124}
5125
Dan Williams45b42332007-07-09 11:56:43 -07005126#ifdef DEBUG
NeilBrownd710e132008-10-13 11:55:12 +11005127static void print_sh(struct seq_file *seq, struct stripe_head *sh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005128{
5129 int i;
5130
NeilBrown16a53ec2006-06-26 00:27:38 -07005131 seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n",
5132 (unsigned long long)sh->sector, sh->pd_idx, sh->state);
5133 seq_printf(seq, "sh %llu, count %d.\n",
5134 (unsigned long long)sh->sector, atomic_read(&sh->count));
5135 seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector);
NeilBrown7ecaa1e2006-03-27 01:18:08 -08005136 for (i = 0; i < sh->disks; i++) {
NeilBrown16a53ec2006-06-26 00:27:38 -07005137 seq_printf(seq, "(cache%d: %p %ld) ",
5138 i, sh->dev[i].page, sh->dev[i].flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005139 }
NeilBrown16a53ec2006-06-26 00:27:38 -07005140 seq_printf(seq, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005141}
5142
NeilBrownd710e132008-10-13 11:55:12 +11005143static void printall(struct seq_file *seq, raid5_conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005144{
5145 struct stripe_head *sh;
NeilBrownfccddba2006-01-06 00:20:33 -08005146 struct hlist_node *hn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005147 int i;
5148
5149 spin_lock_irq(&conf->device_lock);
5150 for (i = 0; i < NR_HASH; i++) {
NeilBrownfccddba2006-01-06 00:20:33 -08005151 hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005152 if (sh->raid_conf != conf)
5153 continue;
NeilBrown16a53ec2006-06-26 00:27:38 -07005154 print_sh(seq, sh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005155 }
5156 }
5157 spin_unlock_irq(&conf->device_lock);
5158}
5159#endif
5160
NeilBrownd710e132008-10-13 11:55:12 +11005161static void status(struct seq_file *seq, mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005162{
H Hartley Sweeten7b928132010-03-08 16:02:40 +11005163 raid5_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005164 int i;
5165
Andre Noll9d8f0362009-06-18 08:45:01 +10005166 seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level,
5167 mddev->chunk_sectors / 2, mddev->layout);
NeilBrown02c2de82006-10-03 01:15:47 -07005168 seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005169 for (i = 0; i < conf->raid_disks; i++)
5170 seq_printf (seq, "%s",
5171 conf->disks[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08005172 test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005173 seq_printf (seq, "]");
Dan Williams45b42332007-07-09 11:56:43 -07005174#ifdef DEBUG
NeilBrown16a53ec2006-06-26 00:27:38 -07005175 seq_printf (seq, "\n");
5176 printall(seq, conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005177#endif
5178}
5179
5180static void print_raid5_conf (raid5_conf_t *conf)
5181{
5182 int i;
5183 struct disk_info *tmp;
5184
NeilBrown0c55e022010-05-03 14:09:02 +10005185 printk(KERN_DEBUG "RAID conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005186 if (!conf) {
5187 printk("(conf==NULL)\n");
5188 return;
5189 }
NeilBrown0c55e022010-05-03 14:09:02 +10005190 printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level,
5191 conf->raid_disks,
5192 conf->raid_disks - conf->mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005193
5194 for (i = 0; i < conf->raid_disks; i++) {
5195 char b[BDEVNAME_SIZE];
5196 tmp = conf->disks + i;
5197 if (tmp->rdev)
NeilBrown0c55e022010-05-03 14:09:02 +10005198 printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n",
5199 i, !test_bit(Faulty, &tmp->rdev->flags),
5200 bdevname(tmp->rdev->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005201 }
5202}
5203
5204static int raid5_spare_active(mddev_t *mddev)
5205{
5206 int i;
5207 raid5_conf_t *conf = mddev->private;
5208 struct disk_info *tmp;
5209
5210 for (i = 0; i < conf->raid_disks; i++) {
5211 tmp = conf->disks + i;
5212 if (tmp->rdev
NeilBrownb2d444d2005-11-08 21:39:31 -08005213 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07005214 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
5215 unsigned long flags;
5216 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005217 mddev->degraded--;
NeilBrownc04be0a2006-10-03 01:15:53 -07005218 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005219 }
5220 }
5221 print_raid5_conf(conf);
5222 return 0;
5223}
5224
5225static int raid5_remove_disk(mddev_t *mddev, int number)
5226{
5227 raid5_conf_t *conf = mddev->private;
5228 int err = 0;
5229 mdk_rdev_t *rdev;
5230 struct disk_info *p = conf->disks + number;
5231
5232 print_raid5_conf(conf);
5233 rdev = p->rdev;
5234 if (rdev) {
NeilBrownec32a2b2009-03-31 15:17:38 +11005235 if (number >= conf->raid_disks &&
5236 conf->reshape_progress == MaxSector)
5237 clear_bit(In_sync, &rdev->flags);
5238
NeilBrownb2d444d2005-11-08 21:39:31 -08005239 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07005240 atomic_read(&rdev->nr_pending)) {
5241 err = -EBUSY;
5242 goto abort;
5243 }
NeilBrowndfc70642008-05-23 13:04:39 -07005244 /* Only remove non-faulty devices if recovery
5245 * isn't possible.
5246 */
5247 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrownec32a2b2009-03-31 15:17:38 +11005248 mddev->degraded <= conf->max_degraded &&
5249 number < conf->raid_disks) {
NeilBrowndfc70642008-05-23 13:04:39 -07005250 err = -EBUSY;
5251 goto abort;
5252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005253 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07005254 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005255 if (atomic_read(&rdev->nr_pending)) {
5256 /* lost the race, try later */
5257 err = -EBUSY;
5258 p->rdev = rdev;
5259 }
5260 }
5261abort:
5262
5263 print_raid5_conf(conf);
5264 return err;
5265}
5266
5267static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
5268{
5269 raid5_conf_t *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10005270 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005271 int disk;
5272 struct disk_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10005273 int first = 0;
5274 int last = conf->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005275
NeilBrown16a53ec2006-06-26 00:27:38 -07005276 if (mddev->degraded > conf->max_degraded)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005277 /* no point adding a device */
Neil Brown199050e2008-06-28 08:31:33 +10005278 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005279
Neil Brown6c2fce22008-06-28 08:31:31 +10005280 if (rdev->raid_disk >= 0)
5281 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005282
5283 /*
NeilBrown16a53ec2006-06-26 00:27:38 -07005284 * find the disk ... but prefer rdev->saved_raid_disk
5285 * if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005286 */
NeilBrown16a53ec2006-06-26 00:27:38 -07005287 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10005288 rdev->saved_raid_disk >= first &&
NeilBrown16a53ec2006-06-26 00:27:38 -07005289 conf->disks[rdev->saved_raid_disk].rdev == NULL)
5290 disk = rdev->saved_raid_disk;
5291 else
Neil Brown6c2fce22008-06-28 08:31:31 +10005292 disk = first;
5293 for ( ; disk <= last ; disk++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005294 if ((p=conf->disks + disk)->rdev == NULL) {
NeilBrownb2d444d2005-11-08 21:39:31 -08005295 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005296 rdev->raid_disk = disk;
Neil Brown199050e2008-06-28 08:31:33 +10005297 err = 0;
NeilBrown72626682005-09-09 16:23:54 -07005298 if (rdev->saved_raid_disk != disk)
5299 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08005300 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005301 break;
5302 }
5303 print_raid5_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10005304 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005305}
5306
5307static int raid5_resize(mddev_t *mddev, sector_t sectors)
5308{
5309 /* no resync is happening, and there is enough space
5310 * on all devices, so we can resize.
5311 * We need to make sure resync covers any new space.
5312 * If the array is shrinking we should possibly wait until
5313 * any io in the removed space completes, but it hardly seems
5314 * worth it.
5315 */
Andre Noll9d8f0362009-06-18 08:45:01 +10005316 sectors &= ~((sector_t)mddev->chunk_sectors - 1);
Dan Williams1f403622009-03-31 14:59:03 +11005317 md_set_array_sectors(mddev, raid5_size(mddev, sectors,
5318 mddev->raid_disks));
Dan Williamsb522adc2009-03-31 15:00:31 +11005319 if (mddev->array_sectors >
5320 raid5_size(mddev, sectors, mddev->raid_disks))
5321 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10005322 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005323 revalidate_disk(mddev->gendisk);
Andre Noll58c0fed2009-03-31 14:33:13 +11005324 if (sectors > mddev->dev_sectors && mddev->recovery_cp == MaxSector) {
5325 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005326 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5327 }
Andre Noll58c0fed2009-03-31 14:33:13 +11005328 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07005329 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005330 return 0;
5331}
5332
NeilBrown01ee22b2009-06-18 08:47:20 +10005333static int check_stripe_cache(mddev_t *mddev)
5334{
5335 /* Can only proceed if there are plenty of stripe_heads.
5336 * We need a minimum of one full stripe,, and for sensible progress
5337 * it is best to have about 4 times that.
5338 * If we require 4 times, then the default 256 4K stripe_heads will
5339 * allow for chunk sizes up to 256K, which is probably OK.
5340 * If the chunk size is greater, user-space should request more
5341 * stripe_heads first.
5342 */
5343 raid5_conf_t *conf = mddev->private;
5344 if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4
5345 > conf->max_nr_stripes ||
5346 ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4
5347 > conf->max_nr_stripes) {
NeilBrown0c55e022010-05-03 14:09:02 +10005348 printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n",
5349 mdname(mddev),
NeilBrown01ee22b2009-06-18 08:47:20 +10005350 ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9)
5351 / STRIPE_SIZE)*4);
5352 return 0;
5353 }
5354 return 1;
5355}
5356
NeilBrown50ac1682009-06-18 08:47:55 +10005357static int check_reshape(mddev_t *mddev)
NeilBrown29269552006-03-27 01:18:10 -08005358{
NeilBrown070ec552009-06-16 16:54:21 +10005359 raid5_conf_t *conf = mddev->private;
NeilBrown29269552006-03-27 01:18:10 -08005360
NeilBrown88ce4932009-03-31 15:24:23 +11005361 if (mddev->delta_disks == 0 &&
5362 mddev->new_layout == mddev->layout &&
Andre Noll664e7c42009-06-18 08:45:27 +10005363 mddev->new_chunk_sectors == mddev->chunk_sectors)
NeilBrown50ac1682009-06-18 08:47:55 +10005364 return 0; /* nothing to do */
NeilBrowndba034e2008-08-05 15:54:13 +10005365 if (mddev->bitmap)
5366 /* Cannot grow a bitmap yet */
5367 return -EBUSY;
NeilBrownec32a2b2009-03-31 15:17:38 +11005368 if (mddev->degraded > conf->max_degraded)
5369 return -EINVAL;
5370 if (mddev->delta_disks < 0) {
5371 /* We might be able to shrink, but the devices must
5372 * be made bigger first.
5373 * For raid6, 4 is the minimum size.
5374 * Otherwise 2 is the minimum
5375 */
5376 int min = 2;
5377 if (mddev->level == 6)
5378 min = 4;
5379 if (mddev->raid_disks + mddev->delta_disks < min)
5380 return -EINVAL;
5381 }
NeilBrown29269552006-03-27 01:18:10 -08005382
NeilBrown01ee22b2009-06-18 08:47:20 +10005383 if (!check_stripe_cache(mddev))
NeilBrown29269552006-03-27 01:18:10 -08005384 return -ENOSPC;
NeilBrown29269552006-03-27 01:18:10 -08005385
NeilBrownec32a2b2009-03-31 15:17:38 +11005386 return resize_stripes(conf, conf->raid_disks + mddev->delta_disks);
NeilBrown63c70c42006-03-27 01:18:13 -08005387}
5388
5389static int raid5_start_reshape(mddev_t *mddev)
5390{
NeilBrown070ec552009-06-16 16:54:21 +10005391 raid5_conf_t *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08005392 mdk_rdev_t *rdev;
NeilBrown63c70c42006-03-27 01:18:13 -08005393 int spares = 0;
5394 int added_devices = 0;
NeilBrownc04be0a2006-10-03 01:15:53 -07005395 unsigned long flags;
NeilBrown63c70c42006-03-27 01:18:13 -08005396
NeilBrownf4168852007-02-28 20:11:53 -08005397 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown63c70c42006-03-27 01:18:13 -08005398 return -EBUSY;
5399
NeilBrown01ee22b2009-06-18 08:47:20 +10005400 if (!check_stripe_cache(mddev))
5401 return -ENOSPC;
5402
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005403 list_for_each_entry(rdev, &mddev->disks, same_set)
NeilBrown29269552006-03-27 01:18:10 -08005404 if (rdev->raid_disk < 0 &&
5405 !test_bit(Faulty, &rdev->flags))
5406 spares++;
NeilBrown63c70c42006-03-27 01:18:13 -08005407
NeilBrownf4168852007-02-28 20:11:53 -08005408 if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded)
NeilBrown29269552006-03-27 01:18:10 -08005409 /* Not enough devices even to make a degraded array
5410 * of that size
5411 */
5412 return -EINVAL;
5413
NeilBrownec32a2b2009-03-31 15:17:38 +11005414 /* Refuse to reduce size of the array. Any reductions in
5415 * array size must be through explicit setting of array_size
5416 * attribute.
5417 */
5418 if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks)
5419 < mddev->array_sectors) {
NeilBrown0c55e022010-05-03 14:09:02 +10005420 printk(KERN_ERR "md/raid:%s: array size must be reduced "
NeilBrownec32a2b2009-03-31 15:17:38 +11005421 "before number of disks\n", mdname(mddev));
5422 return -EINVAL;
5423 }
5424
NeilBrownf6705572006-03-27 01:18:11 -08005425 atomic_set(&conf->reshape_stripes, 0);
NeilBrown29269552006-03-27 01:18:10 -08005426 spin_lock_irq(&conf->device_lock);
5427 conf->previous_raid_disks = conf->raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08005428 conf->raid_disks += mddev->delta_disks;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005429 conf->prev_chunk_sectors = conf->chunk_sectors;
5430 conf->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown88ce4932009-03-31 15:24:23 +11005431 conf->prev_algo = conf->algorithm;
5432 conf->algorithm = mddev->new_layout;
NeilBrownfef9c612009-03-31 15:16:46 +11005433 if (mddev->delta_disks < 0)
5434 conf->reshape_progress = raid5_size(mddev, 0, 0);
5435 else
5436 conf->reshape_progress = 0;
5437 conf->reshape_safe = conf->reshape_progress;
NeilBrown86b42c72009-03-31 15:19:03 +11005438 conf->generation++;
NeilBrown29269552006-03-27 01:18:10 -08005439 spin_unlock_irq(&conf->device_lock);
5440
5441 /* Add some new drives, as many as will fit.
5442 * We know there are enough to make the newly sized array work.
5443 */
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005444 list_for_each_entry(rdev, &mddev->disks, same_set)
NeilBrown29269552006-03-27 01:18:10 -08005445 if (rdev->raid_disk < 0 &&
5446 !test_bit(Faulty, &rdev->flags)) {
Neil Brown199050e2008-06-28 08:31:33 +10005447 if (raid5_add_disk(mddev, rdev) == 0) {
NeilBrown29269552006-03-27 01:18:10 -08005448 char nm[20];
NeilBrown9eb07c22010-02-09 12:31:47 +11005449 if (rdev->raid_disk >= conf->previous_raid_disks) {
NeilBrown7ef90142009-11-13 17:40:51 +11005450 set_bit(In_sync, &rdev->flags);
NeilBrown9eb07c22010-02-09 12:31:47 +11005451 added_devices++;
5452 } else
NeilBrown7ef90142009-11-13 17:40:51 +11005453 rdev->recovery_offset = 0;
NeilBrown29269552006-03-27 01:18:10 -08005454 sprintf(nm, "rd%d", rdev->raid_disk);
NeilBrown5e55e2f2007-03-26 21:32:14 -08005455 if (sysfs_create_link(&mddev->kobj,
5456 &rdev->kobj, nm))
5457 printk(KERN_WARNING
NeilBrown0c55e022010-05-03 14:09:02 +10005458 "md/raid:%s: failed to create "
5459 " link %s\n",
5460 mdname(mddev), nm);
NeilBrown29269552006-03-27 01:18:10 -08005461 } else
5462 break;
5463 }
5464
NeilBrown9eb07c22010-02-09 12:31:47 +11005465 /* When a reshape changes the number of devices, ->degraded
5466 * is measured against the large of the pre and post number of
5467 * devices.*/
NeilBrownec32a2b2009-03-31 15:17:38 +11005468 if (mddev->delta_disks > 0) {
5469 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown9eb07c22010-02-09 12:31:47 +11005470 mddev->degraded += (conf->raid_disks - conf->previous_raid_disks)
NeilBrownec32a2b2009-03-31 15:17:38 +11005471 - added_devices;
5472 spin_unlock_irqrestore(&conf->device_lock, flags);
5473 }
NeilBrown63c70c42006-03-27 01:18:13 -08005474 mddev->raid_disks = conf->raid_disks;
NeilBrowne5164022009-08-03 10:59:57 +10005475 mddev->reshape_position = conf->reshape_progress;
NeilBrown850b2b42006-10-03 01:15:46 -07005476 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownf6705572006-03-27 01:18:11 -08005477
NeilBrown29269552006-03-27 01:18:10 -08005478 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5479 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5480 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5481 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5482 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10005483 "reshape");
NeilBrown29269552006-03-27 01:18:10 -08005484 if (!mddev->sync_thread) {
5485 mddev->recovery = 0;
5486 spin_lock_irq(&conf->device_lock);
5487 mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005488 conf->reshape_progress = MaxSector;
NeilBrown29269552006-03-27 01:18:10 -08005489 spin_unlock_irq(&conf->device_lock);
5490 return -EAGAIN;
5491 }
NeilBrownc8f517c2009-03-31 15:28:40 +11005492 conf->reshape_checkpoint = jiffies;
NeilBrown29269552006-03-27 01:18:10 -08005493 md_wakeup_thread(mddev->sync_thread);
5494 md_new_event(mddev);
5495 return 0;
5496}
NeilBrown29269552006-03-27 01:18:10 -08005497
NeilBrownec32a2b2009-03-31 15:17:38 +11005498/* This is called from the reshape thread and should make any
5499 * changes needed in 'conf'
5500 */
NeilBrown29269552006-03-27 01:18:10 -08005501static void end_reshape(raid5_conf_t *conf)
5502{
NeilBrown29269552006-03-27 01:18:10 -08005503
NeilBrownf6705572006-03-27 01:18:11 -08005504 if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
Dan Williams80c3a6c2009-03-17 18:10:40 -07005505
NeilBrownf6705572006-03-27 01:18:11 -08005506 spin_lock_irq(&conf->device_lock);
NeilBrowncea9c222009-03-31 15:15:05 +11005507 conf->previous_raid_disks = conf->raid_disks;
NeilBrownfef9c612009-03-31 15:16:46 +11005508 conf->reshape_progress = MaxSector;
NeilBrownf6705572006-03-27 01:18:11 -08005509 spin_unlock_irq(&conf->device_lock);
NeilBrownb0f9ec02009-03-31 15:27:18 +11005510 wake_up(&conf->wait_for_overlap);
NeilBrown16a53ec2006-06-26 00:27:38 -07005511
5512 /* read-ahead size must cover two whole stripes, which is
5513 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
5514 */
5515 {
NeilBrowncea9c222009-03-31 15:15:05 +11005516 int data_disks = conf->raid_disks - conf->max_degraded;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005517 int stripe = data_disks * ((conf->chunk_sectors << 9)
NeilBrowncea9c222009-03-31 15:15:05 +11005518 / PAGE_SIZE);
NeilBrown16a53ec2006-06-26 00:27:38 -07005519 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
5520 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
5521 }
NeilBrown29269552006-03-27 01:18:10 -08005522 }
NeilBrown29269552006-03-27 01:18:10 -08005523}
5524
NeilBrownec32a2b2009-03-31 15:17:38 +11005525/* This is called from the raid5d thread with mddev_lock held.
5526 * It makes config changes to the device.
5527 */
NeilBrowncea9c222009-03-31 15:15:05 +11005528static void raid5_finish_reshape(mddev_t *mddev)
5529{
NeilBrown070ec552009-06-16 16:54:21 +10005530 raid5_conf_t *conf = mddev->private;
NeilBrowncea9c222009-03-31 15:15:05 +11005531
5532 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5533
NeilBrownec32a2b2009-03-31 15:17:38 +11005534 if (mddev->delta_disks > 0) {
5535 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5536 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10005537 revalidate_disk(mddev->gendisk);
NeilBrownec32a2b2009-03-31 15:17:38 +11005538 } else {
5539 int d;
NeilBrownec32a2b2009-03-31 15:17:38 +11005540 mddev->degraded = conf->raid_disks;
5541 for (d = 0; d < conf->raid_disks ; d++)
5542 if (conf->disks[d].rdev &&
5543 test_bit(In_sync,
5544 &conf->disks[d].rdev->flags))
5545 mddev->degraded--;
5546 for (d = conf->raid_disks ;
5547 d < conf->raid_disks - mddev->delta_disks;
NeilBrown1a67dde2009-08-13 10:41:49 +10005548 d++) {
5549 mdk_rdev_t *rdev = conf->disks[d].rdev;
5550 if (rdev && raid5_remove_disk(mddev, d) == 0) {
5551 char nm[20];
5552 sprintf(nm, "rd%d", rdev->raid_disk);
5553 sysfs_remove_link(&mddev->kobj, nm);
5554 rdev->raid_disk = -1;
5555 }
5556 }
NeilBrowncea9c222009-03-31 15:15:05 +11005557 }
NeilBrown88ce4932009-03-31 15:24:23 +11005558 mddev->layout = conf->algorithm;
Andre Noll09c9e5f2009-06-18 08:45:55 +10005559 mddev->chunk_sectors = conf->chunk_sectors;
NeilBrownec32a2b2009-03-31 15:17:38 +11005560 mddev->reshape_position = MaxSector;
5561 mddev->delta_disks = 0;
NeilBrowncea9c222009-03-31 15:15:05 +11005562 }
5563}
5564
NeilBrown72626682005-09-09 16:23:54 -07005565static void raid5_quiesce(mddev_t *mddev, int state)
5566{
NeilBrown070ec552009-06-16 16:54:21 +10005567 raid5_conf_t *conf = mddev->private;
NeilBrown72626682005-09-09 16:23:54 -07005568
5569 switch(state) {
NeilBrowne464eaf2006-03-27 01:18:14 -08005570 case 2: /* resume for a suspend */
5571 wake_up(&conf->wait_for_overlap);
5572 break;
5573
NeilBrown72626682005-09-09 16:23:54 -07005574 case 1: /* stop all writes */
5575 spin_lock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005576 /* '2' tells resync/reshape to pause so that all
5577 * active stripes can drain
5578 */
5579 conf->quiesce = 2;
NeilBrown72626682005-09-09 16:23:54 -07005580 wait_event_lock_irq(conf->wait_for_stripe,
Raz Ben-Jehuda(caro)46031f92006-12-10 02:20:47 -08005581 atomic_read(&conf->active_stripes) == 0 &&
5582 atomic_read(&conf->active_aligned_reads) == 0,
NeilBrown72626682005-09-09 16:23:54 -07005583 conf->device_lock, /* nothing */);
NeilBrown64bd6602009-08-03 10:59:58 +10005584 conf->quiesce = 1;
NeilBrown72626682005-09-09 16:23:54 -07005585 spin_unlock_irq(&conf->device_lock);
NeilBrown64bd6602009-08-03 10:59:58 +10005586 /* allow reshape to continue */
5587 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005588 break;
5589
5590 case 0: /* re-enable writes */
5591 spin_lock_irq(&conf->device_lock);
5592 conf->quiesce = 0;
5593 wake_up(&conf->wait_for_stripe);
NeilBrowne464eaf2006-03-27 01:18:14 -08005594 wake_up(&conf->wait_for_overlap);
NeilBrown72626682005-09-09 16:23:54 -07005595 spin_unlock_irq(&conf->device_lock);
5596 break;
5597 }
NeilBrown72626682005-09-09 16:23:54 -07005598}
NeilBrownb15c2e52006-01-06 00:20:16 -08005599
NeilBrownd562b0c2009-03-31 14:39:39 +11005600
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005601static void *raid45_takeover_raid0(mddev_t *mddev, int level)
Trela Maciej54071b32010-03-08 16:02:42 +11005602{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005603 struct raid0_private_data *raid0_priv = mddev->private;
Trela Maciej54071b32010-03-08 16:02:42 +11005604
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005605 /* for raid0 takeover only one zone is supported */
5606 if (raid0_priv->nr_strip_zones > 1) {
NeilBrown0c55e022010-05-03 14:09:02 +10005607 printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n",
5608 mdname(mddev));
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005609 return ERR_PTR(-EINVAL);
5610 }
5611
5612 mddev->new_level = level;
Trela Maciej54071b32010-03-08 16:02:42 +11005613 mddev->new_layout = ALGORITHM_PARITY_N;
5614 mddev->new_chunk_sectors = mddev->chunk_sectors;
5615 mddev->raid_disks += 1;
5616 mddev->delta_disks = 1;
5617 /* make sure it will be not marked as dirty */
5618 mddev->recovery_cp = MaxSector;
5619
5620 return setup_conf(mddev);
5621}
5622
5623
NeilBrownd562b0c2009-03-31 14:39:39 +11005624static void *raid5_takeover_raid1(mddev_t *mddev)
5625{
5626 int chunksect;
5627
5628 if (mddev->raid_disks != 2 ||
5629 mddev->degraded > 1)
5630 return ERR_PTR(-EINVAL);
5631
5632 /* Should check if there are write-behind devices? */
5633
5634 chunksect = 64*2; /* 64K by default */
5635
5636 /* The array must be an exact multiple of chunksize */
5637 while (chunksect && (mddev->array_sectors & (chunksect-1)))
5638 chunksect >>= 1;
5639
5640 if ((chunksect<<9) < STRIPE_SIZE)
5641 /* array size does not allow a suitable chunk size */
5642 return ERR_PTR(-EINVAL);
5643
5644 mddev->new_level = 5;
5645 mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC;
Andre Noll664e7c42009-06-18 08:45:27 +10005646 mddev->new_chunk_sectors = chunksect;
NeilBrownd562b0c2009-03-31 14:39:39 +11005647
5648 return setup_conf(mddev);
5649}
5650
NeilBrownfc9739c2009-03-31 14:57:20 +11005651static void *raid5_takeover_raid6(mddev_t *mddev)
5652{
5653 int new_layout;
5654
5655 switch (mddev->layout) {
5656 case ALGORITHM_LEFT_ASYMMETRIC_6:
5657 new_layout = ALGORITHM_LEFT_ASYMMETRIC;
5658 break;
5659 case ALGORITHM_RIGHT_ASYMMETRIC_6:
5660 new_layout = ALGORITHM_RIGHT_ASYMMETRIC;
5661 break;
5662 case ALGORITHM_LEFT_SYMMETRIC_6:
5663 new_layout = ALGORITHM_LEFT_SYMMETRIC;
5664 break;
5665 case ALGORITHM_RIGHT_SYMMETRIC_6:
5666 new_layout = ALGORITHM_RIGHT_SYMMETRIC;
5667 break;
5668 case ALGORITHM_PARITY_0_6:
5669 new_layout = ALGORITHM_PARITY_0;
5670 break;
5671 case ALGORITHM_PARITY_N:
5672 new_layout = ALGORITHM_PARITY_N;
5673 break;
5674 default:
5675 return ERR_PTR(-EINVAL);
5676 }
5677 mddev->new_level = 5;
5678 mddev->new_layout = new_layout;
5679 mddev->delta_disks = -1;
5680 mddev->raid_disks -= 1;
5681 return setup_conf(mddev);
5682}
5683
NeilBrownd562b0c2009-03-31 14:39:39 +11005684
NeilBrown50ac1682009-06-18 08:47:55 +10005685static int raid5_check_reshape(mddev_t *mddev)
NeilBrownb3546032009-03-31 14:56:41 +11005686{
NeilBrown88ce4932009-03-31 15:24:23 +11005687 /* For a 2-drive array, the layout and chunk size can be changed
5688 * immediately as not restriping is needed.
5689 * For larger arrays we record the new value - after validation
5690 * to be used by a reshape pass.
NeilBrownb3546032009-03-31 14:56:41 +11005691 */
NeilBrown070ec552009-06-16 16:54:21 +10005692 raid5_conf_t *conf = mddev->private;
NeilBrown597a7112009-06-18 08:47:42 +10005693 int new_chunk = mddev->new_chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11005694
NeilBrown597a7112009-06-18 08:47:42 +10005695 if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout))
NeilBrownb3546032009-03-31 14:56:41 +11005696 return -EINVAL;
5697 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005698 if (!is_power_of_2(new_chunk))
NeilBrownb3546032009-03-31 14:56:41 +11005699 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005700 if (new_chunk < (PAGE_SIZE>>9))
NeilBrownb3546032009-03-31 14:56:41 +11005701 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005702 if (mddev->array_sectors & (new_chunk-1))
NeilBrownb3546032009-03-31 14:56:41 +11005703 /* not factor of array size */
5704 return -EINVAL;
5705 }
5706
5707 /* They look valid */
5708
NeilBrown88ce4932009-03-31 15:24:23 +11005709 if (mddev->raid_disks == 2) {
NeilBrown597a7112009-06-18 08:47:42 +10005710 /* can make the change immediately */
5711 if (mddev->new_layout >= 0) {
5712 conf->algorithm = mddev->new_layout;
5713 mddev->layout = mddev->new_layout;
NeilBrown88ce4932009-03-31 15:24:23 +11005714 }
5715 if (new_chunk > 0) {
NeilBrown597a7112009-06-18 08:47:42 +10005716 conf->chunk_sectors = new_chunk ;
5717 mddev->chunk_sectors = new_chunk;
NeilBrown88ce4932009-03-31 15:24:23 +11005718 }
5719 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5720 md_wakeup_thread(mddev->thread);
NeilBrownb3546032009-03-31 14:56:41 +11005721 }
NeilBrown50ac1682009-06-18 08:47:55 +10005722 return check_reshape(mddev);
NeilBrown88ce4932009-03-31 15:24:23 +11005723}
5724
NeilBrown50ac1682009-06-18 08:47:55 +10005725static int raid6_check_reshape(mddev_t *mddev)
NeilBrown88ce4932009-03-31 15:24:23 +11005726{
NeilBrown597a7112009-06-18 08:47:42 +10005727 int new_chunk = mddev->new_chunk_sectors;
NeilBrown50ac1682009-06-18 08:47:55 +10005728
NeilBrown597a7112009-06-18 08:47:42 +10005729 if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout))
NeilBrown88ce4932009-03-31 15:24:23 +11005730 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005731 if (new_chunk > 0) {
Andre Noll0ba459d2009-06-18 08:46:10 +10005732 if (!is_power_of_2(new_chunk))
NeilBrown88ce4932009-03-31 15:24:23 +11005733 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005734 if (new_chunk < (PAGE_SIZE >> 9))
NeilBrown88ce4932009-03-31 15:24:23 +11005735 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10005736 if (mddev->array_sectors & (new_chunk-1))
NeilBrown88ce4932009-03-31 15:24:23 +11005737 /* not factor of array size */
5738 return -EINVAL;
NeilBrownb3546032009-03-31 14:56:41 +11005739 }
NeilBrown88ce4932009-03-31 15:24:23 +11005740
5741 /* They look valid */
NeilBrown50ac1682009-06-18 08:47:55 +10005742 return check_reshape(mddev);
NeilBrownb3546032009-03-31 14:56:41 +11005743}
5744
NeilBrownd562b0c2009-03-31 14:39:39 +11005745static void *raid5_takeover(mddev_t *mddev)
5746{
5747 /* raid5 can take over:
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005748 * raid0 - if there is only one strip zone - make it a raid4 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005749 * raid1 - if there are two drives. We need to know the chunk size
5750 * raid4 - trivial - just use a raid4 layout.
5751 * raid6 - Providing it is a *_6 layout
NeilBrownd562b0c2009-03-31 14:39:39 +11005752 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005753 if (mddev->level == 0)
5754 return raid45_takeover_raid0(mddev, 5);
NeilBrownd562b0c2009-03-31 14:39:39 +11005755 if (mddev->level == 1)
5756 return raid5_takeover_raid1(mddev);
NeilBrowne9d47582009-03-31 14:57:09 +11005757 if (mddev->level == 4) {
5758 mddev->new_layout = ALGORITHM_PARITY_N;
5759 mddev->new_level = 5;
5760 return setup_conf(mddev);
5761 }
NeilBrownfc9739c2009-03-31 14:57:20 +11005762 if (mddev->level == 6)
5763 return raid5_takeover_raid6(mddev);
NeilBrownd562b0c2009-03-31 14:39:39 +11005764
5765 return ERR_PTR(-EINVAL);
5766}
5767
NeilBrowna78d38a2010-03-22 16:53:49 +11005768static void *raid4_takeover(mddev_t *mddev)
5769{
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005770 /* raid4 can take over:
5771 * raid0 - if there is only one strip zone
5772 * raid5 - if layout is right
NeilBrowna78d38a2010-03-22 16:53:49 +11005773 */
Dan Williamsf1b29bc2010-05-01 18:09:05 -07005774 if (mddev->level == 0)
5775 return raid45_takeover_raid0(mddev, 4);
NeilBrowna78d38a2010-03-22 16:53:49 +11005776 if (mddev->level == 5 &&
5777 mddev->layout == ALGORITHM_PARITY_N) {
5778 mddev->new_layout = 0;
5779 mddev->new_level = 4;
5780 return setup_conf(mddev);
5781 }
5782 return ERR_PTR(-EINVAL);
5783}
NeilBrownd562b0c2009-03-31 14:39:39 +11005784
NeilBrown245f46c2009-03-31 14:39:39 +11005785static struct mdk_personality raid5_personality;
5786
5787static void *raid6_takeover(mddev_t *mddev)
5788{
5789 /* Currently can only take over a raid5. We map the
5790 * personality to an equivalent raid6 personality
5791 * with the Q block at the end.
5792 */
5793 int new_layout;
5794
5795 if (mddev->pers != &raid5_personality)
5796 return ERR_PTR(-EINVAL);
5797 if (mddev->degraded > 1)
5798 return ERR_PTR(-EINVAL);
5799 if (mddev->raid_disks > 253)
5800 return ERR_PTR(-EINVAL);
5801 if (mddev->raid_disks < 3)
5802 return ERR_PTR(-EINVAL);
5803
5804 switch (mddev->layout) {
5805 case ALGORITHM_LEFT_ASYMMETRIC:
5806 new_layout = ALGORITHM_LEFT_ASYMMETRIC_6;
5807 break;
5808 case ALGORITHM_RIGHT_ASYMMETRIC:
5809 new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
5810 break;
5811 case ALGORITHM_LEFT_SYMMETRIC:
5812 new_layout = ALGORITHM_LEFT_SYMMETRIC_6;
5813 break;
5814 case ALGORITHM_RIGHT_SYMMETRIC:
5815 new_layout = ALGORITHM_RIGHT_SYMMETRIC_6;
5816 break;
5817 case ALGORITHM_PARITY_0:
5818 new_layout = ALGORITHM_PARITY_0_6;
5819 break;
5820 case ALGORITHM_PARITY_N:
5821 new_layout = ALGORITHM_PARITY_N;
5822 break;
5823 default:
5824 return ERR_PTR(-EINVAL);
5825 }
5826 mddev->new_level = 6;
5827 mddev->new_layout = new_layout;
5828 mddev->delta_disks = 1;
5829 mddev->raid_disks += 1;
5830 return setup_conf(mddev);
5831}
5832
5833
NeilBrown16a53ec2006-06-26 00:27:38 -07005834static struct mdk_personality raid6_personality =
5835{
5836 .name = "raid6",
5837 .level = 6,
5838 .owner = THIS_MODULE,
5839 .make_request = make_request,
5840 .run = run,
5841 .stop = stop,
5842 .status = status,
5843 .error_handler = error,
5844 .hot_add_disk = raid5_add_disk,
5845 .hot_remove_disk= raid5_remove_disk,
5846 .spare_active = raid5_spare_active,
5847 .sync_request = sync_request,
5848 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005849 .size = raid5_size,
NeilBrown50ac1682009-06-18 08:47:55 +10005850 .check_reshape = raid6_check_reshape,
NeilBrownf4168852007-02-28 20:11:53 -08005851 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005852 .finish_reshape = raid5_finish_reshape,
NeilBrown16a53ec2006-06-26 00:27:38 -07005853 .quiesce = raid5_quiesce,
NeilBrown245f46c2009-03-31 14:39:39 +11005854 .takeover = raid6_takeover,
NeilBrown16a53ec2006-06-26 00:27:38 -07005855};
NeilBrown2604b702006-01-06 00:20:36 -08005856static struct mdk_personality raid5_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005857{
5858 .name = "raid5",
NeilBrown2604b702006-01-06 00:20:36 -08005859 .level = 5,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005860 .owner = THIS_MODULE,
5861 .make_request = make_request,
5862 .run = run,
5863 .stop = stop,
5864 .status = status,
5865 .error_handler = error,
5866 .hot_add_disk = raid5_add_disk,
5867 .hot_remove_disk= raid5_remove_disk,
5868 .spare_active = raid5_spare_active,
5869 .sync_request = sync_request,
5870 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005871 .size = raid5_size,
NeilBrown63c70c42006-03-27 01:18:13 -08005872 .check_reshape = raid5_check_reshape,
5873 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005874 .finish_reshape = raid5_finish_reshape,
NeilBrown72626682005-09-09 16:23:54 -07005875 .quiesce = raid5_quiesce,
NeilBrownd562b0c2009-03-31 14:39:39 +11005876 .takeover = raid5_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07005877};
5878
NeilBrown2604b702006-01-06 00:20:36 -08005879static struct mdk_personality raid4_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07005880{
NeilBrown2604b702006-01-06 00:20:36 -08005881 .name = "raid4",
5882 .level = 4,
5883 .owner = THIS_MODULE,
5884 .make_request = make_request,
5885 .run = run,
5886 .stop = stop,
5887 .status = status,
5888 .error_handler = error,
5889 .hot_add_disk = raid5_add_disk,
5890 .hot_remove_disk= raid5_remove_disk,
5891 .spare_active = raid5_spare_active,
5892 .sync_request = sync_request,
5893 .resize = raid5_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07005894 .size = raid5_size,
NeilBrown3d378902007-03-26 21:32:13 -08005895 .check_reshape = raid5_check_reshape,
5896 .start_reshape = raid5_start_reshape,
NeilBrowncea9c222009-03-31 15:15:05 +11005897 .finish_reshape = raid5_finish_reshape,
NeilBrown2604b702006-01-06 00:20:36 -08005898 .quiesce = raid5_quiesce,
NeilBrowna78d38a2010-03-22 16:53:49 +11005899 .takeover = raid4_takeover,
NeilBrown2604b702006-01-06 00:20:36 -08005900};
5901
5902static int __init raid5_init(void)
5903{
NeilBrown16a53ec2006-06-26 00:27:38 -07005904 register_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005905 register_md_personality(&raid5_personality);
5906 register_md_personality(&raid4_personality);
5907 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005908}
5909
NeilBrown2604b702006-01-06 00:20:36 -08005910static void raid5_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005911{
NeilBrown16a53ec2006-06-26 00:27:38 -07005912 unregister_md_personality(&raid6_personality);
NeilBrown2604b702006-01-06 00:20:36 -08005913 unregister_md_personality(&raid5_personality);
5914 unregister_md_personality(&raid4_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005915}
5916
5917module_init(raid5_init);
5918module_exit(raid5_exit);
5919MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11005920MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005921MODULE_ALIAS("md-personality-4"); /* RAID5 */
NeilBrownd9d166c2006-01-06 00:20:51 -08005922MODULE_ALIAS("md-raid5");
5923MODULE_ALIAS("md-raid4");
NeilBrown2604b702006-01-06 00:20:36 -08005924MODULE_ALIAS("md-level-5");
5925MODULE_ALIAS("md-level-4");
NeilBrown16a53ec2006-06-26 00:27:38 -07005926MODULE_ALIAS("md-personality-8"); /* RAID6 */
5927MODULE_ALIAS("md-raid6");
5928MODULE_ALIAS("md-level-6");
5929
5930/* This used to be two separate modules, they were: */
5931MODULE_ALIAS("raid5");
5932MODULE_ALIAS("raid6");