blob: ede2461e79c51e8d3bd1ac04fcac4c7a974bbbbe [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * raid1.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
5 *
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
7 *
8 * RAID-1 management functions.
9 *
10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
11 *
Jan Engelhardt96de0e22007-10-19 23:21:04 +020012 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
14 *
NeilBrown191ea9b2005-06-21 17:17:23 -070015 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
16 * bitmapped intelligence in resync:
17 *
18 * - bitmap marked during normal i/o
19 * - bitmap used to skip nondirty blocks during sync
20 *
21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
22 * - persistent bitmap code
23 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * This program is free software; you can redistribute it and/or modify
25 * it under the terms of the GNU General Public License as published by
26 * the Free Software Foundation; either version 2, or (at your option)
27 * any later version.
28 *
29 * You should have received a copy of the GNU General Public License
30 * (for example /usr/src/linux/COPYING); if not, write to the Free
31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
32 */
33
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/slab.h>
Stephen Rothwell25570722008-10-15 09:09:21 +110035#include <linux/delay.h>
NeilBrownbff61972009-03-31 14:33:13 +110036#include <linux/blkdev.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040037#include <linux/module.h>
NeilBrownbff61972009-03-31 14:33:13 +110038#include <linux/seq_file.h>
Christian Dietrich8bda4702011-07-27 11:00:36 +100039#include <linux/ratelimit.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110040#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110041#include "raid1.h"
42#include "bitmap.h"
NeilBrown191ea9b2005-06-21 17:17:23 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * Number of guaranteed r1bios in case of extreme VM load:
46 */
47#define NR_RAID1_BIOS 256
48
NeilBrown34db0cd2011-10-11 16:50:01 +110049/* When there are this many requests queue to be written by
50 * the raid1 thread, we become 'congested' to provide back-pressure
51 * for writeback.
52 */
53static int max_queued_requests = 1024;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
NeilBrowne8096362011-10-11 16:49:05 +110055static void allow_barrier(struct r1conf *conf);
56static void lower_barrier(struct r1conf *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Al Virodd0fc662005-10-07 07:46:04 +010058static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059{
60 struct pool_info *pi = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +110061 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63 /* allocate a r1bio with room for raid_disks entries in the bios array */
Jens Axboe7eaceac2011-03-10 08:52:07 +010064 return kzalloc(size, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065}
66
67static void r1bio_pool_free(void *r1_bio, void *data)
68{
69 kfree(r1_bio);
70}
71
72#define RESYNC_BLOCK_SIZE (64*1024)
73//#define RESYNC_BLOCK_SIZE PAGE_SIZE
74#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
75#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
76#define RESYNC_WINDOW (2048*1024)
77
Al Virodd0fc662005-10-07 07:46:04 +010078static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079{
80 struct pool_info *pi = data;
81 struct page *page;
NeilBrown9f2c9d12011-10-11 16:48:43 +110082 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 struct bio *bio;
84 int i, j;
85
86 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
Jens Axboe7eaceac2011-03-10 08:52:07 +010087 if (!r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90 /*
91 * Allocate bios : 1 for reading, n-1 for writing
92 */
93 for (j = pi->raid_disks ; j-- ; ) {
NeilBrown67465572010-10-26 17:33:54 +110094 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 if (!bio)
96 goto out_free_bio;
97 r1_bio->bios[j] = bio;
98 }
99 /*
100 * Allocate RESYNC_PAGES data pages and attach them to
NeilBrownd11c1712006-01-06 00:20:26 -0800101 * the first bio.
102 * If this is a user-requested check/repair, allocate
103 * RESYNC_PAGES for each bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 */
NeilBrownd11c1712006-01-06 00:20:26 -0800105 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
106 j = pi->raid_disks;
107 else
108 j = 1;
109 while(j--) {
110 bio = r1_bio->bios[j];
111 for (i = 0; i < RESYNC_PAGES; i++) {
112 page = alloc_page(gfp_flags);
113 if (unlikely(!page))
114 goto out_free_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
NeilBrownd11c1712006-01-06 00:20:26 -0800116 bio->bi_io_vec[i].bv_page = page;
NeilBrown303a0e12009-04-06 14:40:38 +1000117 bio->bi_vcnt = i+1;
NeilBrownd11c1712006-01-06 00:20:26 -0800118 }
119 }
120 /* If not user-requests, copy the page pointers to all bios */
121 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
122 for (i=0; i<RESYNC_PAGES ; i++)
123 for (j=1; j<pi->raid_disks; j++)
124 r1_bio->bios[j]->bi_io_vec[i].bv_page =
125 r1_bio->bios[0]->bi_io_vec[i].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 }
127
128 r1_bio->master_bio = NULL;
129
130 return r1_bio;
131
132out_free_pages:
NeilBrown303a0e12009-04-06 14:40:38 +1000133 for (j=0 ; j < pi->raid_disks; j++)
134 for (i=0; i < r1_bio->bios[j]->bi_vcnt ; i++)
135 put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800136 j = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137out_free_bio:
138 while ( ++j < pi->raid_disks )
139 bio_put(r1_bio->bios[j]);
140 r1bio_pool_free(r1_bio, data);
141 return NULL;
142}
143
144static void r1buf_pool_free(void *__r1_bio, void *data)
145{
146 struct pool_info *pi = data;
NeilBrownd11c1712006-01-06 00:20:26 -0800147 int i,j;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100148 struct r1bio *r1bio = __r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
NeilBrownd11c1712006-01-06 00:20:26 -0800150 for (i = 0; i < RESYNC_PAGES; i++)
151 for (j = pi->raid_disks; j-- ;) {
152 if (j == 0 ||
153 r1bio->bios[j]->bi_io_vec[i].bv_page !=
154 r1bio->bios[0]->bi_io_vec[i].bv_page)
NeilBrown1345b1d2006-01-06 00:20:40 -0800155 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 for (i=0 ; i < pi->raid_disks; i++)
158 bio_put(r1bio->bios[i]);
159
160 r1bio_pool_free(r1bio, data);
161}
162
NeilBrowne8096362011-10-11 16:49:05 +1100163static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
165 int i;
166
167 for (i = 0; i < conf->raid_disks; i++) {
168 struct bio **bio = r1_bio->bios + i;
NeilBrown4367af52011-07-28 11:31:49 +1000169 if (!BIO_SPECIAL(*bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 bio_put(*bio);
171 *bio = NULL;
172 }
173}
174
NeilBrown9f2c9d12011-10-11 16:48:43 +1100175static void free_r1bio(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
NeilBrowne8096362011-10-11 16:49:05 +1100177 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 put_all_bios(conf, r1_bio);
180 mempool_free(r1_bio, conf->r1bio_pool);
181}
182
NeilBrown9f2c9d12011-10-11 16:48:43 +1100183static void put_buf(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
NeilBrowne8096362011-10-11 16:49:05 +1100185 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown3e198f72006-01-06 00:20:21 -0800186 int i;
187
188 for (i=0; i<conf->raid_disks; i++) {
189 struct bio *bio = r1_bio->bios[i];
190 if (bio->bi_end_io)
191 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 mempool_free(r1_bio, conf->r1buf_pool);
195
NeilBrown17999be2006-01-06 00:20:12 -0800196 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
NeilBrown9f2c9d12011-10-11 16:48:43 +1100199static void reschedule_retry(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
201 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +1100202 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +1100203 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205 spin_lock_irqsave(&conf->device_lock, flags);
206 list_add(&r1_bio->retry_list, &conf->retry_list);
NeilBrownddaf22a2006-01-06 00:20:19 -0800207 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 spin_unlock_irqrestore(&conf->device_lock, flags);
209
NeilBrown17999be2006-01-06 00:20:12 -0800210 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 md_wakeup_thread(mddev->thread);
212}
213
214/*
215 * raid_end_bio_io() is called when we have finished servicing a mirrored
216 * operation and are ready to return a success/failure code to the buffer
217 * cache layer.
218 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100219static void call_bio_endio(struct r1bio *r1_bio)
NeilBrownd2eb35a2011-07-28 11:31:48 +1000220{
221 struct bio *bio = r1_bio->master_bio;
222 int done;
NeilBrowne8096362011-10-11 16:49:05 +1100223 struct r1conf *conf = r1_bio->mddev->private;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000224
225 if (bio->bi_phys_segments) {
226 unsigned long flags;
227 spin_lock_irqsave(&conf->device_lock, flags);
228 bio->bi_phys_segments--;
229 done = (bio->bi_phys_segments == 0);
230 spin_unlock_irqrestore(&conf->device_lock, flags);
231 } else
232 done = 1;
233
234 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
235 clear_bit(BIO_UPTODATE, &bio->bi_flags);
236 if (done) {
237 bio_endio(bio, 0);
238 /*
239 * Wake up any possible resync thread that waits for the device
240 * to go idle.
241 */
242 allow_barrier(conf);
243 }
244}
245
NeilBrown9f2c9d12011-10-11 16:48:43 +1100246static void raid_end_bio_io(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
248 struct bio *bio = r1_bio->master_bio;
249
NeilBrown4b6d2872005-09-09 16:23:47 -0700250 /* if nobody has done the final endio yet, do it now */
251 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
NeilBrown36a4e1f2011-10-07 14:23:17 +1100252 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
253 (bio_data_dir(bio) == WRITE) ? "write" : "read",
254 (unsigned long long) bio->bi_sector,
255 (unsigned long long) bio->bi_sector +
256 (bio->bi_size >> 9) - 1);
NeilBrown4b6d2872005-09-09 16:23:47 -0700257
NeilBrownd2eb35a2011-07-28 11:31:48 +1000258 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 free_r1bio(r1_bio);
261}
262
263/*
264 * Update disk head position estimator based on IRQ completion info.
265 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100266static inline void update_head_pos(int disk, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
NeilBrowne8096362011-10-11 16:49:05 +1100268 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270 conf->mirrors[disk].head_position =
271 r1_bio->sector + (r1_bio->sectors);
272}
273
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100274/*
275 * Find the disk number which triggered given bio
276 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100277static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100278{
279 int mirror;
280 int raid_disks = r1_bio->mddev->raid_disks;
281
282 for (mirror = 0; mirror < raid_disks; mirror++)
283 if (r1_bio->bios[mirror] == bio)
284 break;
285
286 BUG_ON(mirror == raid_disks);
287 update_head_pos(mirror, r1_bio);
288
289 return mirror;
290}
291
NeilBrown6712ecf2007-09-27 12:47:43 +0200292static void raid1_end_read_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
294 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100295 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 int mirror;
NeilBrowne8096362011-10-11 16:49:05 +1100297 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 mirror = r1_bio->read_disk;
300 /*
301 * this branch is our 'one mirror IO has finished' event handler:
302 */
NeilBrownddaf22a2006-01-06 00:20:19 -0800303 update_head_pos(mirror, r1_bio);
304
NeilBrowndd00a992007-05-10 03:15:50 -0700305 if (uptodate)
306 set_bit(R1BIO_Uptodate, &r1_bio->state);
307 else {
308 /* If all other devices have failed, we want to return
309 * the error upwards rather than fail the last device.
310 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 */
NeilBrowndd00a992007-05-10 03:15:50 -0700312 unsigned long flags;
313 spin_lock_irqsave(&conf->device_lock, flags);
314 if (r1_bio->mddev->degraded == conf->raid_disks ||
315 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
316 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)))
317 uptodate = 1;
318 spin_unlock_irqrestore(&conf->device_lock, flags);
319 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
NeilBrowndd00a992007-05-10 03:15:50 -0700321 if (uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 raid_end_bio_io(r1_bio);
NeilBrowndd00a992007-05-10 03:15:50 -0700323 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 /*
325 * oops, read error:
326 */
327 char b[BDEVNAME_SIZE];
Christian Dietrich8bda4702011-07-27 11:00:36 +1000328 printk_ratelimited(
329 KERN_ERR "md/raid1:%s: %s: "
330 "rescheduling sector %llu\n",
331 mdname(conf->mddev),
332 bdevname(conf->mirrors[mirror].rdev->bdev,
333 b),
334 (unsigned long long)r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000335 set_bit(R1BIO_ReadError, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 reschedule_retry(r1_bio);
337 }
338
339 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
NeilBrown9f2c9d12011-10-11 16:48:43 +1100342static void close_write(struct r1bio *r1_bio)
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000343{
344 /* it really is the end of this request */
345 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
346 /* free extra copy of the data pages */
347 int i = r1_bio->behind_page_count;
348 while (i--)
349 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
350 kfree(r1_bio->behind_bvecs);
351 r1_bio->behind_bvecs = NULL;
352 }
353 /* clear the bitmap if all writes complete successfully */
354 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
355 r1_bio->sectors,
356 !test_bit(R1BIO_Degraded, &r1_bio->state),
357 test_bit(R1BIO_BehindIO, &r1_bio->state));
358 md_write_end(r1_bio->mddev);
359}
360
NeilBrown9f2c9d12011-10-11 16:48:43 +1100361static void r1_bio_write_done(struct r1bio *r1_bio)
NeilBrown4e780642010-10-19 12:54:01 +1100362{
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000363 if (!atomic_dec_and_test(&r1_bio->remaining))
364 return;
365
366 if (test_bit(R1BIO_WriteError, &r1_bio->state))
367 reschedule_retry(r1_bio);
368 else {
369 close_write(r1_bio);
NeilBrown4367af52011-07-28 11:31:49 +1000370 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
371 reschedule_retry(r1_bio);
372 else
373 raid_end_bio_io(r1_bio);
NeilBrown4e780642010-10-19 12:54:01 +1100374 }
375}
376
NeilBrown6712ecf2007-09-27 12:47:43 +0200377static void raid1_end_write_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
379 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100380 struct r1bio *r1_bio = bio->bi_private;
NeilBrowna9701a32005-11-08 21:39:34 -0800381 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrowne8096362011-10-11 16:49:05 +1100382 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800383 struct bio *to_put = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100385 mirror = find_bio_disk(r1_bio, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Tejun Heoe9c74692010-09-03 11:56:18 +0200387 /*
388 * 'one mirror IO has finished' event handler:
389 */
Tejun Heoe9c74692010-09-03 11:56:18 +0200390 if (!uptodate) {
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000391 set_bit(WriteErrorSeen,
392 &conf->mirrors[mirror].rdev->flags);
393 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +1000394 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 /*
Tejun Heoe9c74692010-09-03 11:56:18 +0200396 * Set R1BIO_Uptodate in our master bio, so that we
397 * will return a good error code for to the higher
398 * levels even if IO on some other mirrored buffer
399 * fails.
400 *
401 * The 'master' represents the composite IO operation
402 * to user-side. So if something waits for IO, then it
403 * will wait for the 'master' bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 */
NeilBrown4367af52011-07-28 11:31:49 +1000405 sector_t first_bad;
406 int bad_sectors;
407
NeilBrowncd5ff9a2011-07-28 11:32:41 +1000408 r1_bio->bios[mirror] = NULL;
409 to_put = bio;
Tejun Heoe9c74692010-09-03 11:56:18 +0200410 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
NeilBrown4367af52011-07-28 11:31:49 +1000412 /* Maybe we can clear some bad blocks. */
413 if (is_badblock(conf->mirrors[mirror].rdev,
414 r1_bio->sector, r1_bio->sectors,
415 &first_bad, &bad_sectors)) {
416 r1_bio->bios[mirror] = IO_MADE_GOOD;
417 set_bit(R1BIO_MadeGood, &r1_bio->state);
418 }
419 }
420
Tejun Heoe9c74692010-09-03 11:56:18 +0200421 if (behind) {
422 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
423 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700424
Tejun Heoe9c74692010-09-03 11:56:18 +0200425 /*
426 * In behind mode, we ACK the master bio once the I/O
427 * has safely reached all non-writemostly
428 * disks. Setting the Returned bit ensures that this
429 * gets done only once -- we don't ever want to return
430 * -EIO here, instead we'll wait
431 */
432 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
433 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
434 /* Maybe we can return now */
435 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
436 struct bio *mbio = r1_bio->master_bio;
NeilBrown36a4e1f2011-10-07 14:23:17 +1100437 pr_debug("raid1: behind end write sectors"
438 " %llu-%llu\n",
439 (unsigned long long) mbio->bi_sector,
440 (unsigned long long) mbio->bi_sector +
441 (mbio->bi_size >> 9) - 1);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000442 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700443 }
444 }
445 }
NeilBrown4367af52011-07-28 11:31:49 +1000446 if (r1_bio->bios[mirror] == NULL)
447 rdev_dec_pending(conf->mirrors[mirror].rdev,
448 conf->mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 * Let's see if all mirrored write operations have finished
452 * already.
453 */
NeilBrownaf6d7b72011-05-11 14:51:19 +1000454 r1_bio_write_done(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700455
NeilBrown04b857f2006-03-09 17:33:46 -0800456 if (to_put)
457 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
460
461/*
462 * This routine returns the disk from which the requested read should
463 * be done. There is a per-array 'next expected sequential IO' sector
464 * number - if this matches on the next IO then we use the last disk.
465 * There is also a per-disk 'last know head position' sector that is
466 * maintained from IRQ contexts, both the normal and the resync IO
467 * completion handlers update this position correctly. If there is no
468 * perfect sequential match then we pick the disk whose head is closest.
469 *
470 * If there are 2 mirrors in the same 2 devices, performance degrades
471 * because position is mirror, not device based.
472 *
473 * The rdev for the device selected will have nr_pending incremented.
474 */
NeilBrowne8096362011-10-11 16:49:05 +1100475static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000477 const sector_t this_sector = r1_bio->sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000478 int sectors;
479 int best_good_sectors;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000480 int start_disk;
NeilBrown76073052011-05-11 14:34:56 +1000481 int best_disk;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000482 int i;
NeilBrown76073052011-05-11 14:34:56 +1000483 sector_t best_dist;
NeilBrown3cb03002011-10-11 16:45:26 +1100484 struct md_rdev *rdev;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000485 int choose_first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487 rcu_read_lock();
488 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700489 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 * device if no resync is going on, or below the resync window.
491 * We take the first readable disk when above the resync window.
492 */
493 retry:
NeilBrownd2eb35a2011-07-28 11:31:48 +1000494 sectors = r1_bio->sectors;
NeilBrown76073052011-05-11 14:34:56 +1000495 best_disk = -1;
496 best_dist = MaxSector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000497 best_good_sectors = 0;
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (conf->mddev->recovery_cp < MaxSector &&
500 (this_sector + sectors >= conf->next_resync)) {
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000501 choose_first = 1;
502 start_disk = 0;
503 } else {
504 choose_first = 0;
505 start_disk = conf->last_used;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 }
507
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000508 for (i = 0 ; i < conf->raid_disks ; i++) {
NeilBrown76073052011-05-11 14:34:56 +1000509 sector_t dist;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000510 sector_t first_bad;
511 int bad_sectors;
512
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000513 int disk = start_disk + i;
514 if (disk >= conf->raid_disks)
515 disk -= conf->raid_disks;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700516
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000517 rdev = rcu_dereference(conf->mirrors[disk].rdev);
518 if (r1_bio->bios[disk] == IO_BLOCKED
519 || rdev == NULL
NeilBrown76073052011-05-11 14:34:56 +1000520 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000521 continue;
NeilBrown76073052011-05-11 14:34:56 +1000522 if (!test_bit(In_sync, &rdev->flags) &&
523 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 continue;
NeilBrown76073052011-05-11 14:34:56 +1000525 if (test_bit(WriteMostly, &rdev->flags)) {
526 /* Don't balance among write-mostly, just
527 * use the first as a last resort */
528 if (best_disk < 0)
529 best_disk = disk;
530 continue;
531 }
532 /* This is a reasonable device to use. It might
533 * even be best.
534 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000535 if (is_badblock(rdev, this_sector, sectors,
536 &first_bad, &bad_sectors)) {
537 if (best_dist < MaxSector)
538 /* already have a better device */
539 continue;
540 if (first_bad <= this_sector) {
541 /* cannot read here. If this is the 'primary'
542 * device, then we must not read beyond
543 * bad_sectors from another device..
544 */
545 bad_sectors -= (this_sector - first_bad);
546 if (choose_first && sectors > bad_sectors)
547 sectors = bad_sectors;
548 if (best_good_sectors > sectors)
549 best_good_sectors = sectors;
550
551 } else {
552 sector_t good_sectors = first_bad - this_sector;
553 if (good_sectors > best_good_sectors) {
554 best_good_sectors = good_sectors;
555 best_disk = disk;
556 }
557 if (choose_first)
558 break;
559 }
560 continue;
561 } else
562 best_good_sectors = sectors;
563
NeilBrown76073052011-05-11 14:34:56 +1000564 dist = abs(this_sector - conf->mirrors[disk].head_position);
565 if (choose_first
566 /* Don't change to another disk for sequential reads */
567 || conf->next_seq_sect == this_sector
568 || dist == 0
569 /* If device is idle, use it */
570 || atomic_read(&rdev->nr_pending) == 0) {
571 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 break;
573 }
NeilBrown76073052011-05-11 14:34:56 +1000574 if (dist < best_dist) {
575 best_dist = dist;
576 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
NeilBrown76073052011-05-11 14:34:56 +1000580 if (best_disk >= 0) {
581 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700582 if (!rdev)
583 goto retry;
584 atomic_inc(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000585 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 /* cannot risk returning a device that failed
587 * before we inc'ed nr_pending
588 */
NeilBrown03c902e2006-01-06 00:20:46 -0800589 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 goto retry;
591 }
NeilBrownd2eb35a2011-07-28 11:31:48 +1000592 sectors = best_good_sectors;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700593 conf->next_seq_sect = this_sector + sectors;
NeilBrown76073052011-05-11 14:34:56 +1000594 conf->last_used = best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
596 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000597 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
NeilBrown76073052011-05-11 14:34:56 +1000599 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
NeilBrownfd01b882011-10-11 16:47:53 +1100602int md_raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700603{
NeilBrowne8096362011-10-11 16:49:05 +1100604 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700605 int i, ret = 0;
606
NeilBrown34db0cd2011-10-11 16:50:01 +1100607 if ((bits & (1 << BDI_async_congested)) &&
608 conf->pending_count >= max_queued_requests)
609 return 1;
610
NeilBrown0d129222006-10-03 01:15:54 -0700611 rcu_read_lock();
612 for (i = 0; i < mddev->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100613 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700614 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200615 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700616
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500617 BUG_ON(!q);
618
NeilBrown0d129222006-10-03 01:15:54 -0700619 /* Note the '|| 1' - when read_balance prefers
620 * non-congested targets, it can be removed
621 */
Alexander Beregalov91a9e992009-04-07 01:35:56 +0400622 if ((bits & (1<<BDI_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700623 ret |= bdi_congested(&q->backing_dev_info, bits);
624 else
625 ret &= bdi_congested(&q->backing_dev_info, bits);
626 }
627 }
628 rcu_read_unlock();
629 return ret;
630}
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500631EXPORT_SYMBOL_GPL(md_raid1_congested);
NeilBrown0d129222006-10-03 01:15:54 -0700632
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500633static int raid1_congested(void *data, int bits)
634{
NeilBrownfd01b882011-10-11 16:47:53 +1100635 struct mddev *mddev = data;
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500636
637 return mddev_congested(mddev, bits) ||
638 md_raid1_congested(mddev, bits);
639}
NeilBrown0d129222006-10-03 01:15:54 -0700640
NeilBrowne8096362011-10-11 16:49:05 +1100641static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800642{
643 /* Any writes that have been queued but are awaiting
644 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800645 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800646 spin_lock_irq(&conf->device_lock);
647
648 if (conf->pending_bio_list.head) {
649 struct bio *bio;
650 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100651 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800652 spin_unlock_irq(&conf->device_lock);
653 /* flush any pending bitmap writes to
654 * disk before proceeding w/ I/O */
655 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100656 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800657
658 while (bio) { /* submit pending writes */
659 struct bio *next = bio->bi_next;
660 bio->bi_next = NULL;
661 generic_make_request(bio);
662 bio = next;
663 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800664 } else
665 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100666}
667
NeilBrown17999be2006-01-06 00:20:12 -0800668/* Barriers....
669 * Sometimes we need to suspend IO while we do something else,
670 * either some resync/recovery, or reconfigure the array.
671 * To do this we raise a 'barrier'.
672 * The 'barrier' is a counter that can be raised multiple times
673 * to count how many activities are happening which preclude
674 * normal IO.
675 * We can only raise the barrier if there is no pending IO.
676 * i.e. if nr_pending == 0.
677 * We choose only to raise the barrier if no-one is waiting for the
678 * barrier to go down. This means that as soon as an IO request
679 * is ready, no other operations which require a barrier will start
680 * until the IO request has had a chance.
681 *
682 * So: regular IO calls 'wait_barrier'. When that returns there
683 * is no backgroup IO happening, It must arrange to call
684 * allow_barrier when it has finished its IO.
685 * backgroup IO calls must call raise_barrier. Once that returns
686 * there is no normal IO happeing. It must arrange to call
687 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 */
689#define RESYNC_DEPTH 32
690
NeilBrowne8096362011-10-11 16:49:05 +1100691static void raise_barrier(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
693 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800694
695 /* Wait until no block IO is waiting */
696 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
NeilBrownc3b328a2011-04-18 18:25:43 +1000697 conf->resync_lock, );
NeilBrown17999be2006-01-06 00:20:12 -0800698
699 /* block any new IO from starting */
700 conf->barrier++;
701
NeilBrown046abee2010-10-26 15:46:20 +1100702 /* Now wait for all pending IO to complete */
NeilBrown17999be2006-01-06 00:20:12 -0800703 wait_event_lock_irq(conf->wait_barrier,
704 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
NeilBrownc3b328a2011-04-18 18:25:43 +1000705 conf->resync_lock, );
NeilBrown17999be2006-01-06 00:20:12 -0800706
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 spin_unlock_irq(&conf->resync_lock);
708}
709
NeilBrowne8096362011-10-11 16:49:05 +1100710static void lower_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800711{
712 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100713 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800714 spin_lock_irqsave(&conf->resync_lock, flags);
715 conf->barrier--;
716 spin_unlock_irqrestore(&conf->resync_lock, flags);
717 wake_up(&conf->wait_barrier);
718}
719
NeilBrowne8096362011-10-11 16:49:05 +1100720static void wait_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800721{
722 spin_lock_irq(&conf->resync_lock);
723 if (conf->barrier) {
724 conf->nr_waiting++;
725 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
726 conf->resync_lock,
NeilBrownc3b328a2011-04-18 18:25:43 +1000727 );
NeilBrown17999be2006-01-06 00:20:12 -0800728 conf->nr_waiting--;
729 }
730 conf->nr_pending++;
731 spin_unlock_irq(&conf->resync_lock);
732}
733
NeilBrowne8096362011-10-11 16:49:05 +1100734static void allow_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800735{
736 unsigned long flags;
737 spin_lock_irqsave(&conf->resync_lock, flags);
738 conf->nr_pending--;
739 spin_unlock_irqrestore(&conf->resync_lock, flags);
740 wake_up(&conf->wait_barrier);
741}
742
NeilBrowne8096362011-10-11 16:49:05 +1100743static void freeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800744{
745 /* stop syncio and normal IO and wait for everything to
746 * go quite.
747 * We increment barrier and nr_waiting, and then
NeilBrown1c830532008-03-04 14:29:35 -0800748 * wait until nr_pending match nr_queued+1
749 * This is called in the context of one normal IO request
750 * that has failed. Thus any sync request that might be pending
751 * will be blocked by nr_pending, and we need to wait for
752 * pending IO requests to complete or be queued for re-try.
753 * Thus the number queued (nr_queued) plus this request (1)
754 * must match the number of pending IOs (nr_pending) before
755 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800756 */
757 spin_lock_irq(&conf->resync_lock);
758 conf->barrier++;
759 conf->nr_waiting++;
760 wait_event_lock_irq(conf->wait_barrier,
NeilBrown1c830532008-03-04 14:29:35 -0800761 conf->nr_pending == conf->nr_queued+1,
NeilBrownddaf22a2006-01-06 00:20:19 -0800762 conf->resync_lock,
NeilBrownc3b328a2011-04-18 18:25:43 +1000763 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -0800764 spin_unlock_irq(&conf->resync_lock);
765}
NeilBrowne8096362011-10-11 16:49:05 +1100766static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800767{
768 /* reverse the effect of the freeze */
769 spin_lock_irq(&conf->resync_lock);
770 conf->barrier--;
771 conf->nr_waiting--;
772 wake_up(&conf->wait_barrier);
773 spin_unlock_irq(&conf->resync_lock);
774}
775
NeilBrown17999be2006-01-06 00:20:12 -0800776
NeilBrown4e780642010-10-19 12:54:01 +1100777/* duplicate the data pages for behind I/O
NeilBrown4e780642010-10-19 12:54:01 +1100778 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100779static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
NeilBrown4b6d2872005-09-09 16:23:47 -0700780{
781 int i;
782 struct bio_vec *bvec;
NeilBrown2ca68f52011-07-28 11:32:10 +1000783 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
NeilBrown4b6d2872005-09-09 16:23:47 -0700784 GFP_NOIO);
NeilBrown2ca68f52011-07-28 11:32:10 +1000785 if (unlikely(!bvecs))
NeilBrownaf6d7b72011-05-11 14:51:19 +1000786 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700787
NeilBrown4b6d2872005-09-09 16:23:47 -0700788 bio_for_each_segment(bvec, bio, i) {
NeilBrown2ca68f52011-07-28 11:32:10 +1000789 bvecs[i] = *bvec;
790 bvecs[i].bv_page = alloc_page(GFP_NOIO);
791 if (unlikely(!bvecs[i].bv_page))
NeilBrown4b6d2872005-09-09 16:23:47 -0700792 goto do_sync_io;
NeilBrown2ca68f52011-07-28 11:32:10 +1000793 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
794 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
795 kunmap(bvecs[i].bv_page);
NeilBrown4b6d2872005-09-09 16:23:47 -0700796 kunmap(bvec->bv_page);
797 }
NeilBrown2ca68f52011-07-28 11:32:10 +1000798 r1_bio->behind_bvecs = bvecs;
NeilBrownaf6d7b72011-05-11 14:51:19 +1000799 r1_bio->behind_page_count = bio->bi_vcnt;
800 set_bit(R1BIO_BehindIO, &r1_bio->state);
801 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700802
803do_sync_io:
NeilBrownaf6d7b72011-05-11 14:51:19 +1000804 for (i = 0; i < bio->bi_vcnt; i++)
NeilBrown2ca68f52011-07-28 11:32:10 +1000805 if (bvecs[i].bv_page)
806 put_page(bvecs[i].bv_page);
807 kfree(bvecs);
NeilBrown36a4e1f2011-10-07 14:23:17 +1100808 pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
NeilBrown4b6d2872005-09-09 16:23:47 -0700809}
810
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700811static void make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
NeilBrowne8096362011-10-11 16:49:05 +1100813 struct r1conf *conf = mddev->private;
NeilBrown0f6d02d2011-10-11 16:48:46 +1100814 struct mirror_info *mirror;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100815 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 struct bio *read_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +1000817 int i, disks;
NeilBrown84255d12008-05-23 13:04:32 -0700818 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -0700819 unsigned long flags;
Jens Axboea3623572005-11-01 09:26:16 +0100820 const int rw = bio_data_dir(bio);
NeilBrown2c7d46e2010-08-18 16:16:05 +1000821 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
Tejun Heoe9c74692010-09-03 11:56:18 +0200822 const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
NeilBrown3cb03002011-10-11 16:45:26 +1100823 struct md_rdev *blocked_rdev;
NeilBrownc3b328a2011-04-18 18:25:43 +1000824 int plugged;
NeilBrown1f68f0c2011-07-28 11:31:48 +1000825 int first_clone;
826 int sectors_handled;
827 int max_sectors;
NeilBrown191ea9b2005-06-21 17:17:23 -0700828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 /*
830 * Register the new request and wait if the reconstruction
831 * thread has put up a bar for new requests.
832 * Continue immediately if no resync is active currently.
833 */
NeilBrown62de6082006-05-01 12:15:47 -0700834
NeilBrown3d310eb2005-06-21 17:17:26 -0700835 md_write_start(mddev, bio); /* wait on superblock update early */
836
NeilBrown6eef4b22009-12-14 12:49:51 +1100837 if (bio_data_dir(bio) == WRITE &&
838 bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo &&
839 bio->bi_sector < mddev->suspend_hi) {
840 /* As the suspend_* range is controlled by
841 * userspace, we want an interruptible
842 * wait.
843 */
844 DEFINE_WAIT(w);
845 for (;;) {
846 flush_signals(current);
847 prepare_to_wait(&conf->wait_barrier,
848 &w, TASK_INTERRUPTIBLE);
849 if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo ||
850 bio->bi_sector >= mddev->suspend_hi)
851 break;
852 schedule();
853 }
854 finish_wait(&conf->wait_barrier, &w);
855 }
NeilBrown62de6082006-05-01 12:15:47 -0700856
NeilBrown17999be2006-01-06 00:20:12 -0800857 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
NeilBrown84255d12008-05-23 13:04:32 -0700859 bitmap = mddev->bitmap;
860
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 /*
862 * make_request() can abort the operation when READA is being
863 * used and no empty request is available.
864 *
865 */
866 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
867
868 r1_bio->master_bio = bio;
869 r1_bio->sectors = bio->bi_size >> 9;
NeilBrown191ea9b2005-06-21 17:17:23 -0700870 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 r1_bio->mddev = mddev;
872 r1_bio->sector = bio->bi_sector;
873
NeilBrownd2eb35a2011-07-28 11:31:48 +1000874 /* We might need to issue multiple reads to different
875 * devices if there are bad blocks around, so we keep
876 * track of the number of reads in bio->bi_phys_segments.
877 * If this is 0, there is only one r1_bio and no locking
878 * will be needed when requests complete. If it is
879 * non-zero, then it is the number of not-completed requests.
880 */
881 bio->bi_phys_segments = 0;
882 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
883
Jens Axboea3623572005-11-01 09:26:16 +0100884 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 /*
886 * read balancing logic:
887 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000888 int rdisk;
889
890read_again:
891 rdisk = read_balance(conf, r1_bio, &max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
893 if (rdisk < 0) {
894 /* couldn't find anywhere to read from */
895 raid_end_bio_io(r1_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200896 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
898 mirror = conf->mirrors + rdisk;
899
NeilBrowne5551902010-03-31 11:21:44 +1100900 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
901 bitmap) {
902 /* Reading from a write-mostly device must
903 * take care not to over-take any writes
904 * that are 'behind'
905 */
906 wait_event(bitmap->behind_wait,
907 atomic_read(&bitmap->behind_writes) == 0);
908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 r1_bio->read_disk = rdisk;
910
NeilBrowna167f662010-10-26 18:31:13 +1100911 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000912 md_trim_bio(read_bio, r1_bio->sector - bio->bi_sector,
913 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915 r1_bio->bios[rdisk] = read_bio;
916
917 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
918 read_bio->bi_bdev = mirror->rdev->bdev;
919 read_bio->bi_end_io = raid1_end_read_request;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200920 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 read_bio->bi_private = r1_bio;
922
NeilBrownd2eb35a2011-07-28 11:31:48 +1000923 if (max_sectors < r1_bio->sectors) {
924 /* could not read all from this device, so we will
925 * need another r1_bio.
926 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000927
928 sectors_handled = (r1_bio->sector + max_sectors
929 - bio->bi_sector);
930 r1_bio->sectors = max_sectors;
931 spin_lock_irq(&conf->device_lock);
932 if (bio->bi_phys_segments == 0)
933 bio->bi_phys_segments = 2;
934 else
935 bio->bi_phys_segments++;
936 spin_unlock_irq(&conf->device_lock);
937 /* Cannot call generic_make_request directly
938 * as that will be queued in __make_request
939 * and subsequent mempool_alloc might block waiting
940 * for it. So hand bio over to raid1d.
941 */
942 reschedule_retry(r1_bio);
943
944 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
945
946 r1_bio->master_bio = bio;
947 r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled;
948 r1_bio->state = 0;
949 r1_bio->mddev = mddev;
950 r1_bio->sector = bio->bi_sector + sectors_handled;
951 goto read_again;
952 } else
953 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200954 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
956
957 /*
958 * WRITE:
959 */
NeilBrown34db0cd2011-10-11 16:50:01 +1100960 if (conf->pending_count >= max_queued_requests) {
961 md_wakeup_thread(mddev->thread);
962 wait_event(conf->wait_barrier,
963 conf->pending_count < max_queued_requests);
964 }
NeilBrown1f68f0c2011-07-28 11:31:48 +1000965 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 * inc refcount on their rdev. Record them by setting
967 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +1000968 * If there are known/acknowledged bad blocks on any device on
969 * which we have seen a write error, we want to avoid writing those
970 * blocks.
971 * This potentially requires several writes to write around
972 * the bad blocks. Each set of writes gets it's own r1bio
973 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 */
NeilBrownc3b328a2011-04-18 18:25:43 +1000975 plugged = mddev_check_plugged(mddev);
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 disks = conf->raid_disks;
Dan Williams6bfe0b42008-04-30 00:52:32 -0700978 retry_write:
979 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +1000981 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100983 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -0700984 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
985 atomic_inc(&rdev->nr_pending);
986 blocked_rdev = rdev;
987 break;
988 }
NeilBrown1f68f0c2011-07-28 11:31:48 +1000989 r1_bio->bios[i] = NULL;
990 if (!rdev || test_bit(Faulty, &rdev->flags)) {
991 set_bit(R1BIO_Degraded, &r1_bio->state);
992 continue;
993 }
994
995 atomic_inc(&rdev->nr_pending);
996 if (test_bit(WriteErrorSeen, &rdev->flags)) {
997 sector_t first_bad;
998 int bad_sectors;
999 int is_bad;
1000
1001 is_bad = is_badblock(rdev, r1_bio->sector,
1002 max_sectors,
1003 &first_bad, &bad_sectors);
1004 if (is_bad < 0) {
1005 /* mustn't write here until the bad block is
1006 * acknowledged*/
1007 set_bit(BlockedBadBlocks, &rdev->flags);
1008 blocked_rdev = rdev;
1009 break;
NeilBrown964147d2010-05-18 15:27:13 +10001010 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001011 if (is_bad && first_bad <= r1_bio->sector) {
1012 /* Cannot write here at all */
1013 bad_sectors -= (r1_bio->sector - first_bad);
1014 if (bad_sectors < max_sectors)
1015 /* mustn't write more than bad_sectors
1016 * to other devices yet
1017 */
1018 max_sectors = bad_sectors;
1019 rdev_dec_pending(rdev, mddev);
1020 /* We don't set R1BIO_Degraded as that
1021 * only applies if the disk is
1022 * missing, so it might be re-added,
1023 * and we want to know to recover this
1024 * chunk.
1025 * In this case the device is here,
1026 * and the fact that this chunk is not
1027 * in-sync is recorded in the bad
1028 * block log
1029 */
1030 continue;
1031 }
1032 if (is_bad) {
1033 int good_sectors = first_bad - r1_bio->sector;
1034 if (good_sectors < max_sectors)
1035 max_sectors = good_sectors;
1036 }
1037 }
1038 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040 rcu_read_unlock();
1041
Dan Williams6bfe0b42008-04-30 00:52:32 -07001042 if (unlikely(blocked_rdev)) {
1043 /* Wait for this device to become unblocked */
1044 int j;
1045
1046 for (j = 0; j < i; j++)
1047 if (r1_bio->bios[j])
1048 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001049 r1_bio->state = 0;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001050 allow_barrier(conf);
1051 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1052 wait_barrier(conf);
1053 goto retry_write;
1054 }
1055
NeilBrown1f68f0c2011-07-28 11:31:48 +10001056 if (max_sectors < r1_bio->sectors) {
1057 /* We are splitting this write into multiple parts, so
1058 * we need to prepare for allocating another r1_bio.
1059 */
1060 r1_bio->sectors = max_sectors;
1061 spin_lock_irq(&conf->device_lock);
1062 if (bio->bi_phys_segments == 0)
1063 bio->bi_phys_segments = 2;
1064 else
1065 bio->bi_phys_segments++;
1066 spin_unlock_irq(&conf->device_lock);
NeilBrown191ea9b2005-06-21 17:17:23 -07001067 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001068 sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector;
NeilBrown4b6d2872005-09-09 16:23:47 -07001069
NeilBrown4e780642010-10-19 12:54:01 +11001070 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001071 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001072
NeilBrown1f68f0c2011-07-28 11:31:48 +10001073 first_clone = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 for (i = 0; i < disks; i++) {
1075 struct bio *mbio;
1076 if (!r1_bio->bios[i])
1077 continue;
1078
NeilBrowna167f662010-10-26 18:31:13 +11001079 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001080 md_trim_bio(mbio, r1_bio->sector - bio->bi_sector, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
NeilBrown1f68f0c2011-07-28 11:31:48 +10001082 if (first_clone) {
1083 /* do behind I/O ?
1084 * Not if there are too many, or cannot
1085 * allocate memory, or a reader on WriteMostly
1086 * is waiting for behind writes to flush */
1087 if (bitmap &&
1088 (atomic_read(&bitmap->behind_writes)
1089 < mddev->bitmap_info.max_write_behind) &&
1090 !waitqueue_active(&bitmap->behind_wait))
1091 alloc_behind_pages(mbio, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
NeilBrown1f68f0c2011-07-28 11:31:48 +10001093 bitmap_startwrite(bitmap, r1_bio->sector,
1094 r1_bio->sectors,
1095 test_bit(R1BIO_BehindIO,
1096 &r1_bio->state));
1097 first_clone = 0;
1098 }
NeilBrown2ca68f52011-07-28 11:32:10 +10001099 if (r1_bio->behind_bvecs) {
NeilBrown4b6d2872005-09-09 16:23:47 -07001100 struct bio_vec *bvec;
1101 int j;
1102
1103 /* Yes, I really want the '__' version so that
1104 * we clear any unused pointer in the io_vec, rather
1105 * than leave them unchanged. This is important
1106 * because when we come to free the pages, we won't
NeilBrown046abee2010-10-26 15:46:20 +11001107 * know the original bi_idx, so we just free
NeilBrown4b6d2872005-09-09 16:23:47 -07001108 * them all
1109 */
1110 __bio_for_each_segment(bvec, mbio, j, 0)
NeilBrown2ca68f52011-07-28 11:32:10 +10001111 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
NeilBrown4b6d2872005-09-09 16:23:47 -07001112 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1113 atomic_inc(&r1_bio->behind_remaining);
1114 }
1115
NeilBrown1f68f0c2011-07-28 11:31:48 +10001116 r1_bio->bios[i] = mbio;
1117
1118 mbio->bi_sector = (r1_bio->sector +
1119 conf->mirrors[i].rdev->data_offset);
1120 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1121 mbio->bi_end_io = raid1_end_write_request;
1122 mbio->bi_rw = WRITE | do_flush_fua | do_sync;
1123 mbio->bi_private = r1_bio;
1124
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 atomic_inc(&r1_bio->remaining);
NeilBrown4e780642010-10-19 12:54:01 +11001126 spin_lock_irqsave(&conf->device_lock, flags);
1127 bio_list_add(&conf->pending_bio_list, mbio);
NeilBrown34db0cd2011-10-11 16:50:01 +11001128 conf->pending_count++;
NeilBrown4e780642010-10-19 12:54:01 +11001129 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 }
NeilBrown079fa162011-09-10 17:21:23 +10001131 /* Mustn't call r1_bio_write_done before this next test,
1132 * as it could result in the bio being freed.
1133 */
NeilBrown1f68f0c2011-07-28 11:31:48 +10001134 if (sectors_handled < (bio->bi_size >> 9)) {
NeilBrown079fa162011-09-10 17:21:23 +10001135 r1_bio_write_done(r1_bio);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001136 /* We need another r1_bio. It has already been counted
1137 * in bio->bi_phys_segments
1138 */
1139 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1140 r1_bio->master_bio = bio;
1141 r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled;
1142 r1_bio->state = 0;
1143 r1_bio->mddev = mddev;
1144 r1_bio->sector = bio->bi_sector + sectors_handled;
1145 goto retry_write;
1146 }
1147
NeilBrown079fa162011-09-10 17:21:23 +10001148 r1_bio_write_done(r1_bio);
1149
1150 /* In case raid1d snuck in to freeze_array */
1151 wake_up(&conf->wait_barrier);
1152
NeilBrownc3b328a2011-04-18 18:25:43 +10001153 if (do_sync || !bitmap || !plugged)
Lars Ellenberge3881a62007-01-10 23:15:37 -08001154 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
NeilBrownfd01b882011-10-11 16:47:53 +11001157static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
NeilBrowne8096362011-10-11 16:49:05 +11001159 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 int i;
1161
1162 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001163 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001164 rcu_read_lock();
1165 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001166 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001168 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1169 }
1170 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 seq_printf(seq, "]");
1172}
1173
1174
NeilBrownfd01b882011-10-11 16:47:53 +11001175static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
1177 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001178 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 /*
1181 * If it is not operational, then we have already marked it as dead
1182 * else if it is the last working disks, ignore the error, let the
1183 * next level up know.
1184 * else mark the drive as failed
1185 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001186 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001187 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 /*
1189 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001190 * normal single drive.
1191 * However don't try a recovery from this drive as
1192 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 */
NeilBrown53890422011-07-27 11:00:36 +10001194 conf->recovery_disabled = mddev->recovery_disabled;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001196 }
NeilBrownde393cd2011-07-28 11:31:48 +10001197 set_bit(Blocked, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001198 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1199 unsigned long flags;
1200 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001202 set_bit(Faulty, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001203 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 /*
1205 * if recovery is running, make sure it aborts.
1206 */
NeilBrowndfc70642008-05-23 13:04:39 -07001207 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrowndd00a992007-05-10 03:15:50 -07001208 } else
1209 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001210 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Joe Perches067032b2011-01-14 09:14:33 +11001211 printk(KERN_ALERT
1212 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1213 "md/raid1:%s: Operation continuing on %d devices.\n",
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001214 mdname(mddev), bdevname(rdev->bdev, b),
1215 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216}
1217
NeilBrowne8096362011-10-11 16:49:05 +11001218static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
1220 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001222 printk(KERN_DEBUG "RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if (!conf) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001224 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 return;
1226 }
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001227 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 conf->raid_disks);
1229
NeilBrownddac7c72006-08-31 21:27:36 -07001230 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 for (i = 0; i < conf->raid_disks; i++) {
1232 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001233 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001234 if (rdev)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001235 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001236 i, !test_bit(In_sync, &rdev->flags),
1237 !test_bit(Faulty, &rdev->flags),
1238 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 }
NeilBrownddac7c72006-08-31 21:27:36 -07001240 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241}
1242
NeilBrowne8096362011-10-11 16:49:05 +11001243static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244{
NeilBrown17999be2006-01-06 00:20:12 -08001245 wait_barrier(conf);
1246 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
1248 mempool_destroy(conf->r1buf_pool);
1249 conf->r1buf_pool = NULL;
1250}
1251
NeilBrownfd01b882011-10-11 16:47:53 +11001252static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253{
1254 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001255 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001256 int count = 0;
1257 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 /*
1260 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001261 * and mark them readable.
1262 * Called under mddev lock, so rcu protection not needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 */
1264 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001265 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrownddac7c72006-08-31 21:27:36 -07001266 if (rdev
1267 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001268 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001269 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001270 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 }
1272 }
NeilBrown6b965622010-08-18 11:56:59 +10001273 spin_lock_irqsave(&conf->device_lock, flags);
1274 mddev->degraded -= count;
1275 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
1277 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001278 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279}
1280
1281
NeilBrownfd01b882011-10-11 16:47:53 +11001282static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283{
NeilBrowne8096362011-10-11 16:49:05 +11001284 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001285 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001286 int mirror = 0;
NeilBrown0f6d02d2011-10-11 16:48:46 +11001287 struct mirror_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001288 int first = 0;
1289 int last = mddev->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
NeilBrown53890422011-07-27 11:00:36 +10001291 if (mddev->recovery_disabled == conf->recovery_disabled)
1292 return -EBUSY;
1293
Neil Brown6c2fce22008-06-28 08:31:31 +10001294 if (rdev->raid_disk >= 0)
1295 first = last = rdev->raid_disk;
1296
1297 for (mirror = first; mirror <= last; mirror++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 if ( !(p=conf->mirrors+mirror)->rdev) {
1299
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10001300 disk_stack_limits(mddev->gendisk, rdev->bdev,
1301 rdev->data_offset << 9);
NeilBrown627a2d32010-03-08 16:44:38 +11001302 /* as we don't honour merge_bvec_fn, we must
1303 * never risk violating it, so limit
1304 * ->max_segments to one lying with a single
1305 * page, as a one page request is never in
1306 * violation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 */
NeilBrown627a2d32010-03-08 16:44:38 +11001308 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
1309 blk_queue_max_segments(mddev->queue, 1);
1310 blk_queue_segment_boundary(mddev->queue,
1311 PAGE_CACHE_SIZE - 1);
1312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
1314 p->head_position = 0;
1315 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001316 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001317 /* As all devices are equivalent, we don't need a full recovery
1318 * if this was recently any drive of the array
1319 */
1320 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001321 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001322 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 break;
1324 }
Andre Nollac5e7112009-08-03 10:59:47 +10001325 md_integrity_add_rdev(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001327 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328}
1329
NeilBrownfd01b882011-10-11 16:47:53 +11001330static int raid1_remove_disk(struct mddev *mddev, int number)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
NeilBrowne8096362011-10-11 16:49:05 +11001332 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 int err = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001334 struct md_rdev *rdev;
NeilBrown0f6d02d2011-10-11 16:48:46 +11001335 struct mirror_info *p = conf->mirrors+ number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 print_conf(conf);
1338 rdev = p->rdev;
1339 if (rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001340 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 atomic_read(&rdev->nr_pending)) {
1342 err = -EBUSY;
1343 goto abort;
1344 }
NeilBrown046abee2010-10-26 15:46:20 +11001345 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001346 * is not possible.
1347 */
1348 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001349 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001350 mddev->degraded < conf->raid_disks) {
1351 err = -EBUSY;
1352 goto abort;
1353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001355 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 if (atomic_read(&rdev->nr_pending)) {
1357 /* lost the race, try later */
1358 err = -EBUSY;
1359 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +10001360 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
Martin K. Petersena91a2782011-03-17 11:11:05 +01001362 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364abort:
1365
1366 print_conf(conf);
1367 return err;
1368}
1369
1370
NeilBrown6712ecf2007-09-27 12:47:43 +02001371static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001373 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
NeilBrown0fc280f2011-10-07 14:22:55 +11001375 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 /*
1378 * we have read a block, now it needs to be re-written,
1379 * or re-read if the read failed.
1380 * We don't do much here, just schedule handling by raid1d
1381 */
NeilBrown69382e82006-01-06 00:20:22 -08001382 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001384
1385 if (atomic_dec_and_test(&r1_bio->remaining))
1386 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
NeilBrown6712ecf2007-09-27 12:47:43 +02001389static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390{
1391 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +11001392 struct r1bio *r1_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001393 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001394 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 int mirror=0;
NeilBrown4367af52011-07-28 11:31:49 +10001396 sector_t first_bad;
1397 int bad_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001399 mirror = find_bio_disk(r1_bio, bio);
1400
NeilBrown6b1117d2006-03-31 02:31:57 -08001401 if (!uptodate) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001402 sector_t sync_blocks = 0;
NeilBrown6b1117d2006-03-31 02:31:57 -08001403 sector_t s = r1_bio->sector;
1404 long sectors_to_go = r1_bio->sectors;
1405 /* make sure these bits doesn't get cleared. */
1406 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001407 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001408 &sync_blocks, 1);
1409 s += sync_blocks;
1410 sectors_to_go -= sync_blocks;
1411 } while (sectors_to_go > 0);
NeilBrownd8f05d22011-07-28 11:33:00 +10001412 set_bit(WriteErrorSeen,
1413 &conf->mirrors[mirror].rdev->flags);
1414 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +10001415 } else if (is_badblock(conf->mirrors[mirror].rdev,
1416 r1_bio->sector,
1417 r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001418 &first_bad, &bad_sectors) &&
1419 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1420 r1_bio->sector,
1421 r1_bio->sectors,
1422 &first_bad, &bad_sectors)
1423 )
NeilBrown4367af52011-07-28 11:31:49 +10001424 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001427 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001428 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1429 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001430 reschedule_retry(r1_bio);
1431 else {
1432 put_buf(r1_bio);
1433 md_done_sync(mddev, s, uptodate);
1434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436}
1437
NeilBrown3cb03002011-10-11 16:45:26 +11001438static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001439 int sectors, struct page *page, int rw)
1440{
1441 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
1442 /* success */
1443 return 1;
1444 if (rw == WRITE)
1445 set_bit(WriteErrorSeen, &rdev->flags);
1446 /* need to record an error - either for the block or the device */
1447 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1448 md_error(rdev->mddev, rdev);
1449 return 0;
1450}
1451
NeilBrown9f2c9d12011-10-11 16:48:43 +11001452static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001453{
1454 /* Try some synchronous reads of other devices to get
1455 * good data, much like with normal read errors. Only
1456 * read into the pages we already have so we don't
1457 * need to re-issue the read request.
1458 * We don't need to freeze the array, because being in an
1459 * active sync request, there is no normal IO, and
1460 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001461 * We don't need to check is_badblock() again as we
1462 * made sure that anything with a bad block in range
1463 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001464 */
NeilBrownfd01b882011-10-11 16:47:53 +11001465 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001466 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001467 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1468 sector_t sect = r1_bio->sector;
1469 int sectors = r1_bio->sectors;
1470 int idx = 0;
1471
1472 while(sectors) {
1473 int s = sectors;
1474 int d = r1_bio->read_disk;
1475 int success = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001476 struct md_rdev *rdev;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001477 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10001478
1479 if (s > (PAGE_SIZE>>9))
1480 s = PAGE_SIZE >> 9;
1481 do {
1482 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1483 /* No rcu protection needed here devices
1484 * can only be removed when no resync is
1485 * active, and resync is currently active
1486 */
1487 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001488 if (sync_page_io(rdev, sect, s<<9,
NeilBrowna68e5872011-05-11 14:40:44 +10001489 bio->bi_io_vec[idx].bv_page,
1490 READ, false)) {
1491 success = 1;
1492 break;
1493 }
1494 }
1495 d++;
1496 if (d == conf->raid_disks)
1497 d = 0;
1498 } while (!success && d != r1_bio->read_disk);
1499
NeilBrown78d7f5f2011-05-11 14:48:56 +10001500 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10001501 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10001502 int abort = 0;
1503 /* Cannot read from anywhere, this block is lost.
1504 * Record a bad block on each device. If that doesn't
1505 * work just disable and interrupt the recovery.
1506 * Don't fail devices as that won't really help.
1507 */
NeilBrowna68e5872011-05-11 14:40:44 +10001508 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1509 " for block %llu\n",
1510 mdname(mddev),
1511 bdevname(bio->bi_bdev, b),
1512 (unsigned long long)r1_bio->sector);
NeilBrown3a9f28a2011-07-28 11:33:42 +10001513 for (d = 0; d < conf->raid_disks; d++) {
1514 rdev = conf->mirrors[d].rdev;
1515 if (!rdev || test_bit(Faulty, &rdev->flags))
1516 continue;
1517 if (!rdev_set_badblocks(rdev, sect, s, 0))
1518 abort = 1;
1519 }
1520 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11001521 conf->recovery_disabled =
1522 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10001523 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1524 md_done_sync(mddev, r1_bio->sectors, 0);
1525 put_buf(r1_bio);
1526 return 0;
1527 }
1528 /* Try next page */
1529 sectors -= s;
1530 sect += s;
1531 idx++;
1532 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001533 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001534
1535 start = d;
1536 /* write it back and re-read */
1537 while (d != r1_bio->read_disk) {
1538 if (d == 0)
1539 d = conf->raid_disks;
1540 d--;
1541 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1542 continue;
1543 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001544 if (r1_sync_page_io(rdev, sect, s,
1545 bio->bi_io_vec[idx].bv_page,
1546 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001547 r1_bio->bios[d]->bi_end_io = NULL;
1548 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001549 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001550 }
1551 d = start;
1552 while (d != r1_bio->read_disk) {
1553 if (d == 0)
1554 d = conf->raid_disks;
1555 d--;
1556 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1557 continue;
1558 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001559 if (r1_sync_page_io(rdev, sect, s,
1560 bio->bi_io_vec[idx].bv_page,
1561 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001562 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001563 }
NeilBrowna68e5872011-05-11 14:40:44 +10001564 sectors -= s;
1565 sect += s;
1566 idx ++;
1567 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001568 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrown7ca78d52011-05-11 14:50:37 +10001569 set_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrowna68e5872011-05-11 14:40:44 +10001570 return 1;
1571}
1572
NeilBrown9f2c9d12011-10-11 16:48:43 +11001573static int process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001574{
1575 /* We have read all readable devices. If we haven't
1576 * got the block, then there is no hope left.
1577 * If we have, then we want to do a comparison
1578 * and skip the write if everything is the same.
1579 * If any blocks failed to read, then we need to
1580 * attempt an over-write
1581 */
NeilBrownfd01b882011-10-11 16:47:53 +11001582 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001583 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001584 int primary;
1585 int i;
1586
NeilBrown78d7f5f2011-05-11 14:48:56 +10001587 for (primary = 0; primary < conf->raid_disks; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10001588 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1589 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1590 r1_bio->bios[primary]->bi_end_io = NULL;
1591 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1592 break;
1593 }
1594 r1_bio->read_disk = primary;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001595 for (i = 0; i < conf->raid_disks; i++) {
1596 int j;
1597 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1598 struct bio *pbio = r1_bio->bios[primary];
1599 struct bio *sbio = r1_bio->bios[i];
1600 int size;
NeilBrowna68e5872011-05-11 14:40:44 +10001601
NeilBrown78d7f5f2011-05-11 14:48:56 +10001602 if (r1_bio->bios[i]->bi_end_io != end_sync_read)
1603 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001604
NeilBrown78d7f5f2011-05-11 14:48:56 +10001605 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1606 for (j = vcnt; j-- ; ) {
1607 struct page *p, *s;
1608 p = pbio->bi_io_vec[j].bv_page;
1609 s = sbio->bi_io_vec[j].bv_page;
1610 if (memcmp(page_address(p),
1611 page_address(s),
1612 PAGE_SIZE))
1613 break;
NeilBrowna68e5872011-05-11 14:40:44 +10001614 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001615 } else
1616 j = 0;
1617 if (j >= 0)
1618 mddev->resync_mismatches += r1_bio->sectors;
1619 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1620 && test_bit(BIO_UPTODATE, &sbio->bi_flags))) {
1621 /* No need to write to this device. */
1622 sbio->bi_end_io = NULL;
1623 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1624 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001625 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001626 /* fixup the bio for reuse */
1627 sbio->bi_vcnt = vcnt;
1628 sbio->bi_size = r1_bio->sectors << 9;
1629 sbio->bi_idx = 0;
1630 sbio->bi_phys_segments = 0;
1631 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1632 sbio->bi_flags |= 1 << BIO_UPTODATE;
1633 sbio->bi_next = NULL;
1634 sbio->bi_sector = r1_bio->sector +
1635 conf->mirrors[i].rdev->data_offset;
1636 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1637 size = sbio->bi_size;
1638 for (j = 0; j < vcnt ; j++) {
1639 struct bio_vec *bi;
1640 bi = &sbio->bi_io_vec[j];
1641 bi->bv_offset = 0;
1642 if (size > PAGE_SIZE)
1643 bi->bv_len = PAGE_SIZE;
1644 else
1645 bi->bv_len = size;
1646 size -= PAGE_SIZE;
1647 memcpy(page_address(bi->bv_page),
1648 page_address(pbio->bi_io_vec[j].bv_page),
1649 PAGE_SIZE);
1650 }
1651 }
NeilBrowna68e5872011-05-11 14:40:44 +10001652 return 0;
1653}
1654
NeilBrown9f2c9d12011-10-11 16:48:43 +11001655static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
NeilBrowne8096362011-10-11 16:49:05 +11001657 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 int i;
1659 int disks = conf->raid_disks;
1660 struct bio *bio, *wbio;
1661
1662 bio = r1_bio->bios[r1_bio->read_disk];
1663
NeilBrowna68e5872011-05-11 14:40:44 +10001664 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
1665 /* ouch - failed to read all of that. */
1666 if (!fix_sync_read_error(r1_bio))
1667 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10001668
1669 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1670 if (process_checks(r1_bio) < 0)
1671 return;
NeilBrownd11c1712006-01-06 00:20:26 -08001672 /*
1673 * schedule writes
1674 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 atomic_set(&r1_bio->remaining, 1);
1676 for (i = 0; i < disks ; i++) {
1677 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08001678 if (wbio->bi_end_io == NULL ||
1679 (wbio->bi_end_io == end_sync_read &&
1680 (i == r1_bio->read_disk ||
1681 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 continue;
1683
NeilBrown3e198f72006-01-06 00:20:21 -08001684 wbio->bi_rw = WRITE;
1685 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 atomic_inc(&r1_bio->remaining);
1687 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
NeilBrown191ea9b2005-06-21 17:17:23 -07001688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 generic_make_request(wbio);
1690 }
1691
1692 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001693 /* if we're here, all write(s) have completed, so clean up */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 md_done_sync(mddev, r1_bio->sectors, 1);
1695 put_buf(r1_bio);
1696 }
1697}
1698
1699/*
1700 * This is a kernel thread which:
1701 *
1702 * 1. Retries failed read operations on working mirrors.
1703 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10001704 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 */
1706
NeilBrowne8096362011-10-11 16:49:05 +11001707static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07001708 sector_t sect, int sectors)
1709{
NeilBrownfd01b882011-10-11 16:47:53 +11001710 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07001711 while(sectors) {
1712 int s = sectors;
1713 int d = read_disk;
1714 int success = 0;
1715 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11001716 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07001717
1718 if (s > (PAGE_SIZE>>9))
1719 s = PAGE_SIZE >> 9;
1720
1721 do {
1722 /* Note: no rcu protection needed here
1723 * as this is synchronous in the raid1d thread
1724 * which is the thread that might remove
1725 * a device. If raid1d ever becomes multi-threaded....
1726 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001727 sector_t first_bad;
1728 int bad_sectors;
1729
NeilBrown867868f2006-10-03 01:15:51 -07001730 rdev = conf->mirrors[d].rdev;
1731 if (rdev &&
1732 test_bit(In_sync, &rdev->flags) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10001733 is_badblock(rdev, sect, s,
1734 &first_bad, &bad_sectors) == 0 &&
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11001735 sync_page_io(rdev, sect, s<<9,
1736 conf->tmppage, READ, false))
NeilBrown867868f2006-10-03 01:15:51 -07001737 success = 1;
1738 else {
1739 d++;
1740 if (d == conf->raid_disks)
1741 d = 0;
1742 }
1743 } while (!success && d != read_disk);
1744
1745 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001746 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11001747 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001748 if (!rdev_set_badblocks(rdev, sect, s, 0))
1749 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07001750 break;
1751 }
1752 /* write it back and re-read */
1753 start = d;
1754 while (d != read_disk) {
1755 if (d==0)
1756 d = conf->raid_disks;
1757 d--;
1758 rdev = conf->mirrors[d].rdev;
1759 if (rdev &&
NeilBrownd8f05d22011-07-28 11:33:00 +10001760 test_bit(In_sync, &rdev->flags))
1761 r1_sync_page_io(rdev, sect, s,
1762 conf->tmppage, WRITE);
NeilBrown867868f2006-10-03 01:15:51 -07001763 }
1764 d = start;
1765 while (d != read_disk) {
1766 char b[BDEVNAME_SIZE];
1767 if (d==0)
1768 d = conf->raid_disks;
1769 d--;
1770 rdev = conf->mirrors[d].rdev;
1771 if (rdev &&
1772 test_bit(In_sync, &rdev->flags)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001773 if (r1_sync_page_io(rdev, sect, s,
1774 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07001775 atomic_add(s, &rdev->corrected_errors);
1776 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001777 "md/raid1:%s: read error corrected "
NeilBrown867868f2006-10-03 01:15:51 -07001778 "(%d sectors at %llu on %s)\n",
1779 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07001780 (unsigned long long)(sect +
1781 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07001782 bdevname(rdev->bdev, b));
1783 }
1784 }
1785 }
1786 sectors -= s;
1787 sect += s;
1788 }
1789}
1790
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001791static void bi_complete(struct bio *bio, int error)
1792{
1793 complete((struct completion *)bio->bi_private);
1794}
1795
1796static int submit_bio_wait(int rw, struct bio *bio)
1797{
1798 struct completion event;
1799 rw |= REQ_SYNC;
1800
1801 init_completion(&event);
1802 bio->bi_private = &event;
1803 bio->bi_end_io = bi_complete;
1804 submit_bio(rw, bio);
1805 wait_for_completion(&event);
1806
1807 return test_bit(BIO_UPTODATE, &bio->bi_flags);
1808}
1809
NeilBrown9f2c9d12011-10-11 16:48:43 +11001810static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001811{
NeilBrownfd01b882011-10-11 16:47:53 +11001812 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001813 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11001814 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a2011-07-28 11:32:41 +10001815 int vcnt, idx;
1816 struct bio_vec *vec;
1817
1818 /* bio has the data to be written to device 'i' where
1819 * we just recently had a write error.
1820 * We repeatedly clone the bio and trim down to one block,
1821 * then try the write. Where the write fails we record
1822 * a bad block.
1823 * It is conceivable that the bio doesn't exactly align with
1824 * blocks. We must handle this somehow.
1825 *
1826 * We currently own a reference on the rdev.
1827 */
1828
1829 int block_sectors;
1830 sector_t sector;
1831 int sectors;
1832 int sect_to_write = r1_bio->sectors;
1833 int ok = 1;
1834
1835 if (rdev->badblocks.shift < 0)
1836 return 0;
1837
1838 block_sectors = 1 << rdev->badblocks.shift;
1839 sector = r1_bio->sector;
1840 sectors = ((sector + block_sectors)
1841 & ~(sector_t)(block_sectors - 1))
1842 - sector;
1843
1844 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
1845 vcnt = r1_bio->behind_page_count;
1846 vec = r1_bio->behind_bvecs;
1847 idx = 0;
1848 while (vec[idx].bv_page == NULL)
1849 idx++;
1850 } else {
1851 vcnt = r1_bio->master_bio->bi_vcnt;
1852 vec = r1_bio->master_bio->bi_io_vec;
1853 idx = r1_bio->master_bio->bi_idx;
1854 }
1855 while (sect_to_write) {
1856 struct bio *wbio;
1857 if (sectors > sect_to_write)
1858 sectors = sect_to_write;
1859 /* Write at 'sector' for 'sectors'*/
1860
1861 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
1862 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
1863 wbio->bi_sector = r1_bio->sector;
1864 wbio->bi_rw = WRITE;
1865 wbio->bi_vcnt = vcnt;
1866 wbio->bi_size = r1_bio->sectors << 9;
1867 wbio->bi_idx = idx;
1868
1869 md_trim_bio(wbio, sector - r1_bio->sector, sectors);
1870 wbio->bi_sector += rdev->data_offset;
1871 wbio->bi_bdev = rdev->bdev;
1872 if (submit_bio_wait(WRITE, wbio) == 0)
1873 /* failure! */
1874 ok = rdev_set_badblocks(rdev, sector,
1875 sectors, 0)
1876 && ok;
1877
1878 bio_put(wbio);
1879 sect_to_write -= sectors;
1880 sector += sectors;
1881 sectors = block_sectors;
1882 }
1883 return ok;
1884}
1885
NeilBrowne8096362011-10-11 16:49:05 +11001886static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10001887{
1888 int m;
1889 int s = r1_bio->sectors;
1890 for (m = 0; m < conf->raid_disks ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001891 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10001892 struct bio *bio = r1_bio->bios[m];
1893 if (bio->bi_end_io == NULL)
1894 continue;
1895 if (test_bit(BIO_UPTODATE, &bio->bi_flags) &&
1896 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
1897 rdev_clear_badblocks(rdev, r1_bio->sector, s);
1898 }
1899 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
1900 test_bit(R1BIO_WriteError, &r1_bio->state)) {
1901 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
1902 md_error(conf->mddev, rdev);
1903 }
1904 }
1905 put_buf(r1_bio);
1906 md_done_sync(conf->mddev, s, 1);
1907}
1908
NeilBrowne8096362011-10-11 16:49:05 +11001909static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10001910{
1911 int m;
1912 for (m = 0; m < conf->raid_disks ; m++)
1913 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11001914 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10001915 rdev_clear_badblocks(rdev,
1916 r1_bio->sector,
1917 r1_bio->sectors);
1918 rdev_dec_pending(rdev, conf->mddev);
1919 } else if (r1_bio->bios[m] != NULL) {
1920 /* This drive got a write error. We need to
1921 * narrow down and record precise write
1922 * errors.
1923 */
1924 if (!narrow_write_error(r1_bio, m)) {
1925 md_error(conf->mddev,
1926 conf->mirrors[m].rdev);
1927 /* an I/O failed, we can't clear the bitmap */
1928 set_bit(R1BIO_Degraded, &r1_bio->state);
1929 }
1930 rdev_dec_pending(conf->mirrors[m].rdev,
1931 conf->mddev);
1932 }
1933 if (test_bit(R1BIO_WriteError, &r1_bio->state))
1934 close_write(r1_bio);
1935 raid_end_bio_io(r1_bio);
1936}
1937
NeilBrowne8096362011-10-11 16:49:05 +11001938static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10001939{
1940 int disk;
1941 int max_sectors;
NeilBrownfd01b882011-10-11 16:47:53 +11001942 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10001943 struct bio *bio;
1944 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001945 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10001946
1947 clear_bit(R1BIO_ReadError, &r1_bio->state);
1948 /* we got a read error. Maybe the drive is bad. Maybe just
1949 * the block and we can fix it.
1950 * We freeze all other IO, and try reading the block from
1951 * other devices. When we find one, we re-write
1952 * and check it that fixes the read error.
1953 * This is all done synchronously while the array is
1954 * frozen
1955 */
1956 if (mddev->ro == 0) {
1957 freeze_array(conf);
1958 fix_read_error(conf, r1_bio->read_disk,
1959 r1_bio->sector, r1_bio->sectors);
1960 unfreeze_array(conf);
1961 } else
1962 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1963
1964 bio = r1_bio->bios[r1_bio->read_disk];
1965 bdevname(bio->bi_bdev, b);
1966read_more:
1967 disk = read_balance(conf, r1_bio, &max_sectors);
1968 if (disk == -1) {
1969 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
1970 " read error for block %llu\n",
1971 mdname(mddev), b, (unsigned long long)r1_bio->sector);
1972 raid_end_bio_io(r1_bio);
1973 } else {
1974 const unsigned long do_sync
1975 = r1_bio->master_bio->bi_rw & REQ_SYNC;
1976 if (bio) {
1977 r1_bio->bios[r1_bio->read_disk] =
1978 mddev->ro ? IO_BLOCKED : NULL;
1979 bio_put(bio);
1980 }
1981 r1_bio->read_disk = disk;
1982 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
1983 md_trim_bio(bio, r1_bio->sector - bio->bi_sector, max_sectors);
1984 r1_bio->bios[r1_bio->read_disk] = bio;
1985 rdev = conf->mirrors[disk].rdev;
1986 printk_ratelimited(KERN_ERR
1987 "md/raid1:%s: redirecting sector %llu"
1988 " to other mirror: %s\n",
1989 mdname(mddev),
1990 (unsigned long long)r1_bio->sector,
1991 bdevname(rdev->bdev, b));
1992 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1993 bio->bi_bdev = rdev->bdev;
1994 bio->bi_end_io = raid1_end_read_request;
1995 bio->bi_rw = READ | do_sync;
1996 bio->bi_private = r1_bio;
1997 if (max_sectors < r1_bio->sectors) {
1998 /* Drat - have to split this up more */
1999 struct bio *mbio = r1_bio->master_bio;
2000 int sectors_handled = (r1_bio->sector + max_sectors
2001 - mbio->bi_sector);
2002 r1_bio->sectors = max_sectors;
2003 spin_lock_irq(&conf->device_lock);
2004 if (mbio->bi_phys_segments == 0)
2005 mbio->bi_phys_segments = 2;
2006 else
2007 mbio->bi_phys_segments++;
2008 spin_unlock_irq(&conf->device_lock);
2009 generic_make_request(bio);
2010 bio = NULL;
2011
2012 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2013
2014 r1_bio->master_bio = mbio;
2015 r1_bio->sectors = (mbio->bi_size >> 9)
2016 - sectors_handled;
2017 r1_bio->state = 0;
2018 set_bit(R1BIO_ReadError, &r1_bio->state);
2019 r1_bio->mddev = mddev;
2020 r1_bio->sector = mbio->bi_sector + sectors_handled;
2021
2022 goto read_more;
2023 } else
2024 generic_make_request(bio);
2025 }
2026}
2027
NeilBrownfd01b882011-10-11 16:47:53 +11002028static void raid1d(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
NeilBrown9f2c9d12011-10-11 16:48:43 +11002030 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002032 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002034 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
2036 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002037
2038 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002040
NeilBrownc3b328a2011-04-18 18:25:43 +10002041 if (atomic_read(&mddev->plug_cnt) == 0)
2042 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002045 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002046 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002048 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002049 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08002051 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 spin_unlock_irqrestore(&conf->device_lock, flags);
2053
2054 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002055 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002056 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002057 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002058 test_bit(R1BIO_WriteError, &r1_bio->state))
2059 handle_sync_write_finished(conf, r1_bio);
2060 else
NeilBrown4367af52011-07-28 11:31:49 +10002061 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a2011-07-28 11:32:41 +10002062 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002063 test_bit(R1BIO_WriteError, &r1_bio->state))
2064 handle_write_finished(conf, r1_bio);
2065 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2066 handle_read_error(conf, r1_bio);
2067 else
NeilBrownd2eb35a2011-07-28 11:31:48 +10002068 /* just a partial read to be scheduled from separate
2069 * context
2070 */
2071 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
NeilBrown62096bc2011-07-28 11:38:13 +10002072
NeilBrown1d9d5242009-10-16 15:55:32 +11002073 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002074 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2075 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002077 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078}
2079
2080
NeilBrowne8096362011-10-11 16:49:05 +11002081static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082{
2083 int buffs;
2084
2085 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002086 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2088 conf->poolinfo);
2089 if (!conf->r1buf_pool)
2090 return -ENOMEM;
2091 conf->next_resync = 0;
2092 return 0;
2093}
2094
2095/*
2096 * perform a "sync" on one "block"
2097 *
2098 * We need to make sure that no normal I/O request - particularly write
2099 * requests - conflict with active sync requests.
2100 *
2101 * This is achieved by tracking pending requests and a 'barrier' concept
2102 * that can be installed to exclude normal IO requests.
2103 */
2104
NeilBrownfd01b882011-10-11 16:47:53 +11002105static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106{
NeilBrowne8096362011-10-11 16:49:05 +11002107 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002108 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 struct bio *bio;
2110 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002111 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002113 int wonly = -1;
2114 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002115 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002116 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002117 int good_sectors = RESYNC_SECTORS;
2118 int min_bad = 0; /* number of sectors that are bad in all devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
2120 if (!conf->r1buf_pool)
2121 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002122 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
Andre Noll58c0fed2009-03-31 14:33:13 +11002124 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002126 /* If we aborted, we need to abort the
2127 * sync on the 'current' bitmap chunk (there will
2128 * only be one in raid1 resync.
2129 * We can find the current addess in mddev->curr_resync
2130 */
NeilBrown6a806c52005-07-15 03:56:35 -07002131 if (mddev->curr_resync < max_sector) /* aborted */
2132 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07002133 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002134 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002135 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002136
2137 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 close_sync(conf);
2139 return 0;
2140 }
2141
NeilBrown07d84d102006-06-26 00:27:56 -07002142 if (mddev->bitmap == NULL &&
2143 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002144 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002145 conf->fullsync == 0) {
2146 *skipped = 1;
2147 return max_sector - sector_nr;
2148 }
NeilBrown6394cca2006-08-27 01:23:50 -07002149 /* before building a request, check if we can skip these blocks..
2150 * This call the bitmap_start_sync doesn't actually record anything
2151 */
NeilBrowne3b97032005-08-04 12:53:34 -07002152 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de4852005-11-08 21:39:38 -08002153 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002154 /* We can skip this block, and probably several more */
2155 *skipped = 1;
2156 return sync_blocks;
2157 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 /*
NeilBrown17999be2006-01-06 00:20:12 -08002159 * If there is non-resync activity waiting for a turn,
2160 * and resync is going fast enough,
2161 * then let it though before starting on this new sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 */
NeilBrown17999be2006-01-06 00:20:12 -08002163 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 msleep_interruptible(1000);
NeilBrown17999be2006-01-06 00:20:12 -08002165
NeilBrownb47490c2008-02-06 01:39:50 -08002166 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
NeilBrown1c4588e2010-10-26 17:41:22 +11002167 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown17999be2006-01-06 00:20:12 -08002168 raise_barrier(conf);
2169
2170 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
NeilBrown3e198f72006-01-06 00:20:21 -08002172 rcu_read_lock();
2173 /*
2174 * If we get a correctably read error during resync or recovery,
2175 * we might want to read from a different device. So we
2176 * flag all drives that could conceivably be read from for READ,
2177 * and any others (which will be non-In_sync devices) for WRITE.
2178 * If a read fails, we try reading from something else for which READ
2179 * is OK.
2180 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 r1_bio->mddev = mddev;
2183 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002184 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
2187 for (i=0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002188 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 bio = r1_bio->bios[i];
2190
2191 /* take from bio_init */
2192 bio->bi_next = NULL;
NeilBrowndb8d9d32010-10-07 12:00:50 +11002193 bio->bi_flags &= ~(BIO_POOL_MASK-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 bio->bi_flags |= 1 << BIO_UPTODATE;
NeilBrown802ba062006-12-13 00:34:13 -08002195 bio->bi_rw = READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 bio->bi_vcnt = 0;
2197 bio->bi_idx = 0;
2198 bio->bi_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 bio->bi_size = 0;
2200 bio->bi_end_io = NULL;
2201 bio->bi_private = NULL;
2202
NeilBrown3e198f72006-01-06 00:20:21 -08002203 rdev = rcu_dereference(conf->mirrors[i].rdev);
2204 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002205 test_bit(Faulty, &rdev->flags)) {
NeilBrowne3b97032005-08-04 12:53:34 -07002206 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002207 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 bio->bi_rw = WRITE;
2209 bio->bi_end_io = end_sync_write;
2210 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002211 } else {
2212 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002213 sector_t first_bad = MaxSector;
2214 int bad_sectors;
2215
2216 if (is_badblock(rdev, sector_nr, good_sectors,
2217 &first_bad, &bad_sectors)) {
2218 if (first_bad > sector_nr)
2219 good_sectors = first_bad - sector_nr;
2220 else {
2221 bad_sectors -= (sector_nr - first_bad);
2222 if (min_bad == 0 ||
2223 min_bad > bad_sectors)
2224 min_bad = bad_sectors;
2225 }
NeilBrown3e198f72006-01-06 00:20:21 -08002226 }
NeilBrown06f60382011-07-28 11:31:48 +10002227 if (sector_nr < first_bad) {
2228 if (test_bit(WriteMostly, &rdev->flags)) {
2229 if (wonly < 0)
2230 wonly = i;
2231 } else {
2232 if (disk < 0)
2233 disk = i;
2234 }
2235 bio->bi_rw = READ;
2236 bio->bi_end_io = end_sync_read;
2237 read_targets++;
2238 }
NeilBrown3e198f72006-01-06 00:20:21 -08002239 }
NeilBrown06f60382011-07-28 11:31:48 +10002240 if (bio->bi_end_io) {
2241 atomic_inc(&rdev->nr_pending);
2242 bio->bi_sector = sector_nr + rdev->data_offset;
2243 bio->bi_bdev = rdev->bdev;
2244 bio->bi_private = r1_bio;
2245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 }
NeilBrown3e198f72006-01-06 00:20:21 -08002247 rcu_read_unlock();
2248 if (disk < 0)
2249 disk = wonly;
2250 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002251
NeilBrown06f60382011-07-28 11:31:48 +10002252 if (read_targets == 0 && min_bad > 0) {
2253 /* These sectors are bad on all InSync devices, so we
2254 * need to mark them bad on all write targets
2255 */
2256 int ok = 1;
2257 for (i = 0 ; i < conf->raid_disks ; i++)
2258 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
NeilBrown3cb03002011-10-11 16:45:26 +11002259 struct md_rdev *rdev =
NeilBrown06f60382011-07-28 11:31:48 +10002260 rcu_dereference(conf->mirrors[i].rdev);
2261 ok = rdev_set_badblocks(rdev, sector_nr,
2262 min_bad, 0
2263 ) && ok;
2264 }
2265 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2266 *skipped = 1;
2267 put_buf(r1_bio);
2268
2269 if (!ok) {
2270 /* Cannot record the badblocks, so need to
2271 * abort the resync.
2272 * If there are multiple read targets, could just
2273 * fail the really bad ones ???
2274 */
2275 conf->recovery_disabled = mddev->recovery_disabled;
2276 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2277 return 0;
2278 } else
2279 return min_bad;
2280
2281 }
2282 if (min_bad > 0 && min_bad < good_sectors) {
2283 /* only resync enough to reach the next bad->good
2284 * transition */
2285 good_sectors = min_bad;
2286 }
2287
NeilBrown3e198f72006-01-06 00:20:21 -08002288 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2289 /* extra read targets are also write targets */
2290 write_targets += read_targets-1;
2291
2292 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 /* There is nowhere to write, so all non-sync
2294 * drives must be failed - so we are finished
2295 */
NeilBrown57afd892005-06-21 17:17:13 -07002296 sector_t rv = max_sector - sector_nr;
2297 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 return rv;
2300 }
2301
NeilBrownc6207272008-02-06 01:39:52 -08002302 if (max_sector > mddev->resync_max)
2303 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002304 if (max_sector > sector_nr + good_sectors)
2305 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002307 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 do {
2309 struct page *page;
2310 int len = PAGE_SIZE;
2311 if (sector_nr + (len>>9) > max_sector)
2312 len = (max_sector - sector_nr) << 9;
2313 if (len == 0)
2314 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002315 if (sync_blocks == 0) {
2316 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de4852005-11-08 21:39:38 -08002317 &sync_blocks, still_degraded) &&
2318 !conf->fullsync &&
2319 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002320 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002321 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown7571ae82010-10-07 11:54:46 +11002322 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002323 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002324 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 for (i=0 ; i < conf->raid_disks; i++) {
2327 bio = r1_bio->bios[i];
2328 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08002329 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 if (bio_add_page(bio, page, len, 0) == 0) {
2331 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08002332 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 while (i > 0) {
2334 i--;
2335 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07002336 if (bio->bi_end_io==NULL)
2337 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 /* remove last page from this bio */
2339 bio->bi_vcnt--;
2340 bio->bi_size -= len;
2341 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
2342 }
2343 goto bio_full;
2344 }
2345 }
2346 }
2347 nr_sectors += len>>9;
2348 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002349 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2351 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 r1_bio->sectors = nr_sectors;
2353
NeilBrownd11c1712006-01-06 00:20:26 -08002354 /* For a user-requested sync, we read all readable devices and do a
2355 * compare
2356 */
2357 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2358 atomic_set(&r1_bio->remaining, read_targets);
2359 for (i=0; i<conf->raid_disks; i++) {
2360 bio = r1_bio->bios[i];
2361 if (bio->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07002362 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002363 generic_make_request(bio);
2364 }
2365 }
2366 } else {
2367 atomic_set(&r1_bio->remaining, 1);
2368 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07002369 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002370 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
NeilBrownd11c1712006-01-06 00:20:26 -08002372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 return nr_sectors;
2374}
2375
NeilBrownfd01b882011-10-11 16:47:53 +11002376static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002377{
2378 if (sectors)
2379 return sectors;
2380
2381 return mddev->dev_sectors;
2382}
2383
NeilBrowne8096362011-10-11 16:49:05 +11002384static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385{
NeilBrowne8096362011-10-11 16:49:05 +11002386 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002387 int i;
NeilBrown0f6d02d2011-10-11 16:48:46 +11002388 struct mirror_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002389 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002390 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391
NeilBrowne8096362011-10-11 16:49:05 +11002392 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002394 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
NeilBrown9ffae0c2006-01-06 00:20:32 -08002396 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 GFP_KERNEL);
2398 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002399 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400
NeilBrownddaf22a2006-01-06 00:20:19 -08002401 conf->tmppage = alloc_page(GFP_KERNEL);
2402 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002403 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002404
NeilBrown709ae482009-12-14 12:49:51 +11002405 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002407 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 conf->poolinfo->raid_disks = mddev->raid_disks;
2409 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2410 r1bio_pool_free,
2411 conf->poolinfo);
2412 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11002413 goto abort;
2414
NeilBrowned9bfdf2009-10-16 15:55:44 +11002415 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
Neil Browne7e72bf2008-05-14 16:05:54 -07002417 spin_lock_init(&conf->device_lock);
Cheng Renquan159ec1f2009-01-09 08:31:08 +11002418 list_for_each_entry(rdev, &mddev->disks, same_set) {
NeilBrown709ae482009-12-14 12:49:51 +11002419 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 if (disk_idx >= mddev->raid_disks
2421 || disk_idx < 0)
2422 continue;
2423 disk = conf->mirrors + disk_idx;
2424
2425 disk->rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
2427 disk->head_position = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 }
2429 conf->raid_disks = mddev->raid_disks;
2430 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 INIT_LIST_HEAD(&conf->retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432
2433 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002434 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435
NeilBrown191ea9b2005-06-21 17:17:23 -07002436 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11002437 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11002438 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07002439
NeilBrown709ae482009-12-14 12:49:51 +11002440 conf->last_used = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 for (i = 0; i < conf->raid_disks; i++) {
2442
2443 disk = conf->mirrors + i;
2444
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002445 if (!disk->rdev ||
2446 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 disk->head_position = 0;
NeilBrown918f0232007-08-22 14:01:52 -07002448 if (disk->rdev)
2449 conf->fullsync = 1;
NeilBrown709ae482009-12-14 12:49:51 +11002450 } else if (conf->last_used < 0)
2451 /*
2452 * The first working device is used as a
2453 * starting point to read balancing.
2454 */
2455 conf->last_used = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 }
NeilBrown709ae482009-12-14 12:49:51 +11002457
2458 err = -EIO;
2459 if (conf->last_used < 0) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002460 printk(KERN_ERR "md/raid1:%s: no operational mirrors\n",
NeilBrown709ae482009-12-14 12:49:51 +11002461 mdname(mddev));
2462 goto abort;
NeilBrown11ce99e2006-10-03 01:15:52 -07002463 }
NeilBrown709ae482009-12-14 12:49:51 +11002464 err = -ENOMEM;
2465 conf->thread = md_register_thread(raid1d, mddev, NULL);
2466 if (!conf->thread) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002467 printk(KERN_ERR
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002468 "md/raid1:%s: couldn't allocate thread\n",
NeilBrown191ea9b2005-06-21 17:17:23 -07002469 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002470 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002472
NeilBrown709ae482009-12-14 12:49:51 +11002473 return conf;
2474
2475 abort:
2476 if (conf) {
2477 if (conf->r1bio_pool)
2478 mempool_destroy(conf->r1bio_pool);
2479 kfree(conf->mirrors);
2480 safe_put_page(conf->tmppage);
2481 kfree(conf->poolinfo);
2482 kfree(conf);
2483 }
2484 return ERR_PTR(err);
2485}
2486
NeilBrownfd01b882011-10-11 16:47:53 +11002487static int run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002488{
NeilBrowne8096362011-10-11 16:49:05 +11002489 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002490 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11002491 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002492
2493 if (mddev->level != 1) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002494 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
NeilBrown709ae482009-12-14 12:49:51 +11002495 mdname(mddev), mddev->level);
2496 return -EIO;
2497 }
2498 if (mddev->reshape_position != MaxSector) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002499 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
NeilBrown709ae482009-12-14 12:49:51 +11002500 mdname(mddev));
2501 return -EIO;
2502 }
2503 /*
2504 * copy the already verified devices into our private RAID1
2505 * bookkeeping area. [whatever we allocate in run(),
2506 * should be freed in stop()]
2507 */
2508 if (mddev->private == NULL)
2509 conf = setup_conf(mddev);
2510 else
2511 conf = mddev->private;
2512
2513 if (IS_ERR(conf))
2514 return PTR_ERR(conf);
2515
NeilBrown709ae482009-12-14 12:49:51 +11002516 list_for_each_entry(rdev, &mddev->disks, same_set) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002517 if (!mddev->gendisk)
2518 continue;
NeilBrown709ae482009-12-14 12:49:51 +11002519 disk_stack_limits(mddev->gendisk, rdev->bdev,
2520 rdev->data_offset << 9);
2521 /* as we don't honour merge_bvec_fn, we must never risk
NeilBrown627a2d32010-03-08 16:44:38 +11002522 * violating it, so limit ->max_segments to 1 lying within
2523 * a single page, as a one page request is never in violation.
NeilBrown709ae482009-12-14 12:49:51 +11002524 */
NeilBrown627a2d32010-03-08 16:44:38 +11002525 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
2526 blk_queue_max_segments(mddev->queue, 1);
2527 blk_queue_segment_boundary(mddev->queue,
2528 PAGE_CACHE_SIZE - 1);
2529 }
NeilBrown709ae482009-12-14 12:49:51 +11002530 }
2531
2532 mddev->degraded = 0;
2533 for (i=0; i < conf->raid_disks; i++)
2534 if (conf->mirrors[i].rdev == NULL ||
2535 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2536 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2537 mddev->degraded++;
2538
2539 if (conf->raid_disks - mddev->degraded == 1)
2540 mddev->recovery_cp = MaxSector;
2541
Andre Noll8c6ac862009-06-18 08:48:06 +10002542 if (mddev->recovery_cp != MaxSector)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002543 printk(KERN_NOTICE "md/raid1:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10002544 " -- starting background reconstruction\n",
2545 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002547 "md/raid1:%s: active with %d out of %d mirrors\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 mdname(mddev), mddev->raid_disks - mddev->degraded,
2549 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 /*
2552 * Ok, everything is just fine now
2553 */
NeilBrown709ae482009-12-14 12:49:51 +11002554 mddev->thread = conf->thread;
2555 conf->thread = NULL;
2556 mddev->private = conf;
2557
Dan Williams1f403622009-03-31 14:59:03 +11002558 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002560 if (mddev->queue) {
2561 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2562 mddev->queue->backing_dev_info.congested_data = mddev;
2563 }
Martin K. Petersena91a2782011-03-17 11:11:05 +01002564 return md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565}
2566
NeilBrownfd01b882011-10-11 16:47:53 +11002567static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568{
NeilBrowne8096362011-10-11 16:49:05 +11002569 struct r1conf *conf = mddev->private;
NeilBrown4b6d2872005-09-09 16:23:47 -07002570 struct bitmap *bitmap = mddev->bitmap;
NeilBrown4b6d2872005-09-09 16:23:47 -07002571
2572 /* wait for behind writes to complete */
NeilBrowne5551902010-03-31 11:21:44 +11002573 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002574 printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n",
2575 mdname(mddev));
NeilBrown4b6d2872005-09-09 16:23:47 -07002576 /* need to kick something here to make sure I/O goes? */
NeilBrowne5551902010-03-31 11:21:44 +11002577 wait_event(bitmap->behind_wait,
2578 atomic_read(&bitmap->behind_writes) == 0);
NeilBrown4b6d2872005-09-09 16:23:47 -07002579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
NeilBrown409c57f2009-03-31 14:39:39 +11002581 raise_barrier(conf);
2582 lower_barrier(conf);
2583
NeilBrown01f96c02011-09-21 15:30:20 +10002584 md_unregister_thread(&mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 if (conf->r1bio_pool)
2586 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002587 kfree(conf->mirrors);
2588 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 kfree(conf);
2590 mddev->private = NULL;
2591 return 0;
2592}
2593
NeilBrownfd01b882011-10-11 16:47:53 +11002594static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595{
2596 /* no resync is happening, and there is enough space
2597 * on all devices, so we can resize.
2598 * We need to make sure resync covers any new space.
2599 * If the array is shrinking we should possibly wait until
2600 * any io in the removed space completes, but it hardly seems
2601 * worth it.
2602 */
Dan Williams1f403622009-03-31 14:59:03 +11002603 md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0));
Dan Williamsb522adc2009-03-31 15:00:31 +11002604 if (mddev->array_sectors > raid1_size(mddev, sectors, 0))
2605 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10002606 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10002607 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11002608 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10002609 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11002610 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2612 }
Dan Williamsb522adc2009-03-31 15:00:31 +11002613 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002614 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 return 0;
2616}
2617
NeilBrownfd01b882011-10-11 16:47:53 +11002618static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619{
2620 /* We need to:
2621 * 1/ resize the r1bio_pool
2622 * 2/ resize conf->mirrors
2623 *
2624 * We allocate a new r1bio_pool if we can.
2625 * Then raise a device barrier and wait until all IO stops.
2626 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07002627 *
2628 * At the same time, we "pack" the devices so that all the missing
2629 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 */
2631 mempool_t *newpool, *oldpool;
2632 struct pool_info *newpoolinfo;
NeilBrown0f6d02d2011-10-11 16:48:46 +11002633 struct mirror_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11002634 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08002635 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07002636 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07002637 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
NeilBrown63c70c42006-03-27 01:18:13 -08002639 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10002640 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08002641 mddev->layout != mddev->new_layout ||
2642 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10002643 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08002644 mddev->new_layout = mddev->layout;
2645 mddev->new_level = mddev->level;
2646 return -EINVAL;
2647 }
2648
Dan Williamsb5470dc2008-06-27 21:44:04 -07002649 err = md_allow_write(mddev);
2650 if (err)
2651 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08002652
NeilBrown63c70c42006-03-27 01:18:13 -08002653 raid_disks = mddev->raid_disks + mddev->delta_disks;
2654
NeilBrown6ea9c072005-06-21 17:17:09 -07002655 if (raid_disks < conf->raid_disks) {
2656 cnt=0;
2657 for (d= 0; d < conf->raid_disks; d++)
2658 if (conf->mirrors[d].rdev)
2659 cnt++;
2660 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07002662 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
2664 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2665 if (!newpoolinfo)
2666 return -ENOMEM;
2667 newpoolinfo->mddev = mddev;
2668 newpoolinfo->raid_disks = raid_disks;
2669
2670 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2671 r1bio_pool_free, newpoolinfo);
2672 if (!newpool) {
2673 kfree(newpoolinfo);
2674 return -ENOMEM;
2675 }
NeilBrown9ffae0c2006-01-06 00:20:32 -08002676 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 if (!newmirrors) {
2678 kfree(newpoolinfo);
2679 mempool_destroy(newpool);
2680 return -ENOMEM;
2681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682
NeilBrown17999be2006-01-06 00:20:12 -08002683 raise_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684
2685 /* ok, everything is stopped */
2686 oldpool = conf->r1bio_pool;
2687 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07002688
NeilBrowna88aa782007-08-22 14:01:53 -07002689 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002690 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07002691 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10002692 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07002693 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10002694 sysfs_unlink_rdev(mddev, rdev);
2695 if (sysfs_link_rdev(mddev, rdev))
NeilBrowna88aa782007-08-22 14:01:53 -07002696 printk(KERN_WARNING
Namhyung Kim36fad852011-07-27 11:00:36 +10002697 "md/raid1:%s: cannot register rd%d\n",
2698 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07002699 }
NeilBrowna88aa782007-08-22 14:01:53 -07002700 if (rdev)
2701 newmirrors[d2++].rdev = rdev;
2702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 kfree(conf->mirrors);
2704 conf->mirrors = newmirrors;
2705 kfree(conf->poolinfo);
2706 conf->poolinfo = newpoolinfo;
2707
NeilBrownc04be0a2006-10-03 01:15:53 -07002708 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07002710 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08002712 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713
NeilBrown6ea9c072005-06-21 17:17:09 -07002714 conf->last_used = 0; /* just make sure it is in-range */
NeilBrown17999be2006-01-06 00:20:12 -08002715 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716
2717 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2718 md_wakeup_thread(mddev->thread);
2719
2720 mempool_destroy(oldpool);
2721 return 0;
2722}
2723
NeilBrownfd01b882011-10-11 16:47:53 +11002724static void raid1_quiesce(struct mddev *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07002725{
NeilBrowne8096362011-10-11 16:49:05 +11002726 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07002727
2728 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11002729 case 2: /* wake for suspend */
2730 wake_up(&conf->wait_barrier);
2731 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002732 case 1:
NeilBrown17999be2006-01-06 00:20:12 -08002733 raise_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002734 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002735 case 0:
NeilBrown17999be2006-01-06 00:20:12 -08002736 lower_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002737 break;
2738 }
NeilBrown36fa3062005-09-09 16:23:45 -07002739}
2740
NeilBrownfd01b882011-10-11 16:47:53 +11002741static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002742{
2743 /* raid1 can take over:
2744 * raid5 with 2 devices, any layout or chunk size
2745 */
2746 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11002747 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002748 mddev->new_level = 1;
2749 mddev->new_layout = 0;
2750 mddev->new_chunk_sectors = 0;
2751 conf = setup_conf(mddev);
2752 if (!IS_ERR(conf))
2753 conf->barrier = 1;
2754 return conf;
2755 }
2756 return ERR_PTR(-EINVAL);
2757}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758
NeilBrown84fc4b52011-10-11 16:49:58 +11002759static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760{
2761 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08002762 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 .owner = THIS_MODULE,
2764 .make_request = make_request,
2765 .run = run,
2766 .stop = stop,
2767 .status = status,
2768 .error_handler = error,
2769 .hot_add_disk = raid1_add_disk,
2770 .hot_remove_disk= raid1_remove_disk,
2771 .spare_active = raid1_spare_active,
2772 .sync_request = sync_request,
2773 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07002774 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08002775 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07002776 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11002777 .takeover = raid1_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778};
2779
2780static int __init raid_init(void)
2781{
NeilBrown2604b702006-01-06 00:20:36 -08002782 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783}
2784
2785static void raid_exit(void)
2786{
NeilBrown2604b702006-01-06 00:20:36 -08002787 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788}
2789
2790module_init(raid_init);
2791module_exit(raid_exit);
2792MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11002793MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002795MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08002796MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11002797
2798module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);